141 files changed, 19272 insertions, 9989 deletions

diff --git a/src/AL/alctypes.h b/src/AL/alctypes.h
index 181384b0..a7c7858f 100644
--- a/src/AL/alctypes.h
+++ b/src/AL/alctypes.h
@@ -123,6 +123,7 @@ typedef void ALCvoid;
  */
 #define ALC_DEFAULT_DEVICE_SPECIFIER             0x1004
 #define ALC_DEVICE_SPECIFIER                     0x1005
+#define ALC_ALL_DEVICES_SPECIFIER                0x1013
 #define ALC_EXTENSIONS                           0x1006
 
 #define ALC_MAJOR_VERSION                        0x1000
diff --git a/src/client/cl_avi.c b/src/client/cl_avi.c
index c9680522..7f673fa8 100644
--- a/src/client/cl_avi.c
+++ b/src/client/cl_avi.c
@@ -368,8 +368,13 @@ qboolean CL_OpenAVIForWriting( const char *fileName )
   else
     afd.motionJpeg = qfalse;
 
-  afd.cBuffer = Z_Malloc( afd.width * afd.height * 4 );
-  afd.eBuffer = Z_Malloc( afd.width * afd.height * 4 );
+  // Buffers only need to store RGB pixels.
+  // Allocate a bit more space for the capture buffer to account for possible
+  // padding at the end of pixel lines, and padding for alignment
+  #define MAX_PACK_LEN 16
+  afd.cBuffer = Z_Malloc((afd.width * 3 + MAX_PACK_LEN - 1) * afd.height + MAX_PACK_LEN - 1);
+  // raw avi files have pixel lines start on 4-byte boundaries
+  afd.eBuffer = Z_Malloc(PAD(afd.width * 3, AVI_LINE_PADDING) * afd.height);
 
   afd.a.rate = dma.speed;
   afd.a.format = WAV_FORMAT_PCM;
@@ -467,7 +472,7 @@ void CL_WriteAVIVideoFrame( const byte *imageBuffer, int size )
 {
   int   chunkOffset = afd.fileSize - afd.moviOffset - 8;
   int   chunkSize = 8 + size;
-  int   paddingSize = PAD( size, 2 ) - size;
+  int   paddingSize = PADLEN(size, 2);
   byte  padding[ 4 ] = { 0 };
 
   if( !afd.fileOpen )
@@ -541,7 +546,7 @@ void CL_WriteAVIAudioFrame( const byte *pcmBuffer, int size )
   {
     int   chunkOffset = afd.fileSize - afd.moviOffset - 8;
     int   chunkSize = 8 + bytesInBuffer;
-    int   paddingSize = PAD( bytesInBuffer, 2 ) - bytesInBuffer;
+    int   paddingSize = PADLEN(bytesInBuffer, 2);
     byte  padding[ 4 ] = { 0 };
 
     bufIndex = 0;
diff --git a/src/client/cl_cgame.c b/src/client/cl_cgame.c
index 9f983cd8..babc5241 100644
--- a/src/client/cl_cgame.c
+++ b/src/client/cl_cgame.c
@@ -428,7 +428,7 @@ intptr_t CL_CgameSystemCalls( intptr_t *args ) {
 		Cvar_Update( VMA(1) );
 		return 0;
 	case CG_CVAR_SET:
-		Cvar_Set( VMA(1), VMA(2) );
+		Cvar_SetSafe( VMA(1), VMA(2) );
 		return 0;
 	case CG_CVAR_VARIABLESTRINGBUFFER:
 		Cvar_VariableStringBuffer( VMA(1), VMA(2), args[3] );
@@ -466,7 +466,7 @@ intptr_t CL_CgameSystemCalls( intptr_t *args ) {
 		CL_AddCgameCommand( VMA(1) );
 		return 0;
 	case CG_REMOVECOMMAND:
-		Cmd_RemoveCommand( VMA(1) );
+		Cmd_RemoveCommandSafe( VMA(1) );
 		return 0;
 	case CG_SENDCLIENTCOMMAND:
 		CL_AddReliableCommand(VMA(1), qfalse);
diff --git a/src/client/cl_cin.c b/src/client/cl_cin.c
index 6d03d40a..213b7a40 100644
--- a/src/client/cl_cin.c
+++ b/src/client/cl_cin.c
@@ -1500,6 +1500,66 @@ void CIN_SetLooping(int handle, qboolean loop) {
 
 /*
 ==================
+CIN_ResampleCinematic
+
+Resample cinematic to 256x256 and store in buf2
+==================
+*/
+void CIN_ResampleCinematic(int handle, int *buf2) {
+	int ix, iy, *buf3, xm, ym, ll;
+	byte	*buf;
+
+	buf = cinTable[handle].buf;
+
+	xm = cinTable[handle].CIN_WIDTH/256;
+	ym = cinTable[handle].CIN_HEIGHT/256;
+	ll = 8;
+	if (cinTable[handle].CIN_WIDTH==512) {
+		ll = 9;
+	}
+
+	buf3 = (int*)buf;
+	if (xm==2 && ym==2) {
+		byte *bc2, *bc3;
+		int	ic, iiy;
+
+		bc2 = (byte *)buf2;
+		bc3 = (byte *)buf3;
+		for (iy = 0; iy<256; iy++) {
+			iiy = iy<<12;
+			for (ix = 0; ix<2048; ix+=8) {
+				for(ic = ix;ic<(ix+4);ic++) {
+					*bc2=(bc3[iiy+ic]+bc3[iiy+4+ic]+bc3[iiy+2048+ic]+bc3[iiy+2048+4+ic])>>2;
+					bc2++;
+				}
+			}
+		}
+	} else if (xm==2 && ym==1) {
+		byte *bc2, *bc3;
+		int	ic, iiy;
+
+		bc2 = (byte *)buf2;
+		bc3 = (byte *)buf3;
+		for (iy = 0; iy<256; iy++) {
+			iiy = iy<<11;
+			for (ix = 0; ix<2048; ix+=8) {
+				for(ic = ix;ic<(ix+4);ic++) {
+					*bc2=(bc3[iiy+ic]+bc3[iiy+4+ic])>>1;
+					bc2++;
+				}
+			}
+		}
+	} else {
+		for (iy = 0; iy<256; iy++) {
+			for (ix = 0; ix<256; ix++) {
+					buf2[(iy<<8)+ix] = buf3[((iy*ym)<<ll) + (ix*xm)];
+			}
+		}
+	}
+}
+
+/*
+==================
 SCR_DrawCinematic
 
 ==================
@@ -1522,54 +1582,12 @@ void CIN_DrawCinematic (int handle) {
 	SCR_AdjustFrom640( &x, &y, &w, &h );
 
 	if (cinTable[handle].dirty && (cinTable[handle].CIN_WIDTH != cinTable[handle].drawX || cinTable[handle].CIN_HEIGHT != cinTable[handle].drawY)) {
-		int ix, iy, *buf2, *buf3, xm, ym, ll;
-                
-		xm = cinTable[handle].CIN_WIDTH/256;
-		ym = cinTable[handle].CIN_HEIGHT/256;
-                ll = 8;
-                if (cinTable[handle].CIN_WIDTH==512) {
-                    ll = 9;
-                }
-                
-		buf3 = (int*)buf;
+		int *buf2;
+
 		buf2 = Hunk_AllocateTempMemory( 256*256*4 );
-                if (xm==2 && ym==2) {
-                    byte *bc2, *bc3;
-                    int	ic, iiy;
-                    
-                    bc2 = (byte *)buf2;
-                    bc3 = (byte *)buf3;
-                    for (iy = 0; iy<256; iy++) {
-                            iiy = iy<<12;
-                            for (ix = 0; ix<2048; ix+=8) {
-                                for(ic = ix;ic<(ix+4);ic++) {
-                                    *bc2=(bc3[iiy+ic]+bc3[iiy+4+ic]+bc3[iiy+2048+ic]+bc3[iiy+2048+4+ic])>>2;
-                                    bc2++;
-                                }
-                            }
-                    }
-                } else if (xm==2 && ym==1) {
-                    byte *bc2, *bc3;
-                    int	ic, iiy;
-                    
-                    bc2 = (byte *)buf2;
-                    bc3 = (byte *)buf3;
-                    for (iy = 0; iy<256; iy++) {
-                            iiy = iy<<11;
-                            for (ix = 0; ix<2048; ix+=8) {
-                                for(ic = ix;ic<(ix+4);ic++) {
-                                    *bc2=(bc3[iiy+ic]+bc3[iiy+4+ic])>>1;
-                                    bc2++;
-                                }
-                            }
-                    }
-                } else {
-                    for (iy = 0; iy<256; iy++) {
-                            for (ix = 0; ix<256; ix++) {
-                                    buf2[(iy<<8)+ix] = buf3[((iy*ym)<<ll) + (ix*xm)];
-                            }
-                    }
-                }
+
+		CIN_ResampleCinematic(handle, buf2);
+
 		re.DrawStretchRaw( x, y, w, h, 256, 256, (byte *)buf2, handle, qtrue);
 		cinTable[handle].dirty = qfalse;
 		Hunk_FreeTempMemory(buf2);
@@ -1649,7 +1667,25 @@ void CIN_UploadCinematic(int handle) {
 				}
 			}
 		}
-		re.UploadCinematic( 256, 256, 256, 256, cinTable[handle].buf, handle, cinTable[handle].dirty);
+
+		// Resample the video if needed
+		if (cinTable[handle].dirty && (cinTable[handle].CIN_WIDTH != cinTable[handle].drawX || cinTable[handle].CIN_HEIGHT != cinTable[handle].drawY))  {
+			int *buf2;
+
+			buf2 = Hunk_AllocateTempMemory( 256*256*4 );
+
+			CIN_ResampleCinematic(handle, buf2);
+
+			re.UploadCinematic( cinTable[handle].CIN_WIDTH, cinTable[handle].CIN_HEIGHT, 256, 256, (byte *)buf2, handle, qtrue);
+			cinTable[handle].dirty = qfalse;
+			Hunk_FreeTempMemory(buf2);
+		} else {
+			// Upload video at normal resolution
+			re.UploadCinematic( cinTable[handle].CIN_WIDTH, cinTable[handle].CIN_HEIGHT, cinTable[handle].drawX, cinTable[handle].drawY,
+					cinTable[handle].buf, handle, cinTable[handle].dirty);
+			cinTable[handle].dirty = qfalse;
+		}
+
 		if (cl_inGameVideo->integer == 0 && cinTable[handle].playonwalls == 1) {
 			cinTable[handle].playonwalls--;
 		}
diff --git a/src/client/cl_console.c b/src/client/cl_console.c
index 88ced07a..c7d01bc7 100644
--- a/src/client/cl_console.c
+++ b/src/client/cl_console.c
@@ -240,7 +240,7 @@ Cmd_CompleteTxtName
 */
 void Cmd_CompleteTxtName( char *args, int argNum ) {
 	if( argNum == 2 ) {
-		Field_CompleteFilename( "", "txt", qfalse );
+		Field_CompleteFilename( "", "txt", qfalse, qtrue );
 	}
 }
 
@@ -297,9 +297,9 @@ If no console is visible, the text will appear at the top of the game window
 ================
 */
 void CL_ConsolePrint( char *txt ) {
-	int		y;
-	int		c, l;
-	int		color;
+	int		y, l;
+	unsigned char	c;
+	unsigned short	color;
 	qboolean skipnotify = qfalse;		// NERVE - SMF
 
 	// TTimo - prefix for text that shows up in console but not in notify
@@ -336,7 +336,7 @@ void CL_ConsolePrint( char *txt ) {
 
 	color = ColorIndex(COLOR_WHITE);
 
-	while ( (c = *txt) != 0 ) {
+	while ( (c = *((unsigned char *) txt)) != 0 ) {
 		if ( Q_IsColorString( txt ) ) {
 			color = ColorIndex( *(txt+1) );
 			txt += 2;
@@ -373,10 +373,8 @@ void CL_ConsolePrint( char *txt ) {
 			y = con.current % con.totallines;
 			con.text[y*con.linewidth+con.x] = (color << 8) | c;
 			con.x++;
-			if (con.x >= con.linewidth) {
+			if(con.x >= con.linewidth)
 				Con_Linefeed(skipnotify);
-				con.x = 0;
-			}
 			break;
 		}
 	}
diff --git a/src/client/cl_input.c b/src/client/cl_input.c
index 2f063161..ee5b478e 100644
--- a/src/client/cl_input.c
+++ b/src/client/cl_input.c
@@ -417,15 +417,19 @@ void CL_JoystickMove( usercmd_t *cmd ) {
 	}
 
 	if ( !in_strafe.active ) {
-		cl.viewangles[YAW] += anglespeed * cl_yawspeed->value * cl.joystickAxis[AXIS_SIDE];
+		cl.viewangles[YAW] += anglespeed * j_yaw->value * cl.joystickAxis[j_yaw_axis->integer];
+		cmd->rightmove = ClampChar( cmd->rightmove + (int) (j_side->value * cl.joystickAxis[j_side_axis->integer]) );
 	} else {
-		cmd->rightmove = ClampChar( cmd->rightmove + cl.joystickAxis[AXIS_SIDE] );
+		cl.viewangles[YAW] += anglespeed * j_side->value * cl.joystickAxis[j_side_axis->integer];
+		cmd->rightmove = ClampChar( cmd->rightmove + (int) (j_yaw->value * cl.joystickAxis[j_yaw_axis->integer]) );
 	}
 
 	if ( in_mlooking ) {
-		cl.viewangles[PITCH] += anglespeed * cl_pitchspeed->value * cl.joystickAxis[AXIS_FORWARD];
+		cl.viewangles[PITCH] += anglespeed * j_forward->value * cl.joystickAxis[j_forward_axis->integer];
+		cmd->forwardmove = ClampChar( cmd->forwardmove + (int) (j_pitch->value * cl.joystickAxis[j_pitch_axis->integer]) );
 	} else {
-		cmd->forwardmove = ClampChar( cmd->forwardmove + cl.joystickAxis[AXIS_FORWARD] );
+		cl.viewangles[PITCH] += anglespeed * j_pitch->value * cl.joystickAxis[j_pitch_axis->integer];
+		cmd->forwardmove = ClampChar( cmd->forwardmove + (int) (j_forward->value * cl.joystickAxis[j_forward_axis->integer]) );
 	}
 
 	cmd->upmove = ClampChar( cmd->upmove + cl.joystickAxis[AXIS_UP] );
diff --git a/src/client/cl_keys.c b/src/client/cl_keys.c
index c2cda2d6..71569fe5 100644
--- a/src/client/cl_keys.c
+++ b/src/client/cl_keys.c
@@ -1127,7 +1127,7 @@ void CL_KeyDownEvent( int key, unsigned time )
 {
 	keys[key].down = qtrue;
 	keys[key].repeats++;
-	if( keys[key].repeats == 1 )
+	if( keys[key].repeats == 1 && key != K_SCROLLOCK && key != K_KP_NUMLOCK && key != K_CAPSLOCK )
 		anykeydown++;
 
 	if( keys[K_ALT].down && key == K_ENTER )
@@ -1212,7 +1212,9 @@ void CL_KeyUpEvent( int key, unsigned time )
 {
 	keys[key].repeats = 0;
 	keys[key].down = qfalse;
-	anykeydown--;
+	if (key != K_SCROLLOCK && key != K_KP_NUMLOCK && key != K_CAPSLOCK)
+		anykeydown--;
+
 	if (anykeydown < 0) {
 		anykeydown = 0;
 	}
@@ -1293,6 +1295,9 @@ void Key_ClearStates (void)
 	anykeydown = 0;
 
 	for ( i=0 ; i < MAX_KEYS ; i++ ) {
+		if (i == K_SCROLLOCK || i == K_KP_NUMLOCK || i == K_CAPSLOCK)
+			continue;
+
 		if ( keys[i].down ) {
 			CL_KeyEvent( i, qfalse, 0 );
 
diff --git a/src/client/cl_main.c b/src/client/cl_main.c
index c7ae2614..5166804f 100644
--- a/src/client/cl_main.c
+++ b/src/client/cl_main.c
@@ -84,6 +84,15 @@ cvar_t	*m_forward;
 cvar_t	*m_side;
 cvar_t	*m_filter;
 
+cvar_t	*j_pitch;
+cvar_t	*j_yaw;
+cvar_t	*j_forward;
+cvar_t	*j_side;
+cvar_t	*j_pitch_axis;
+cvar_t	*j_yaw_axis;
+cvar_t	*j_forward_axis;
+cvar_t	*j_side_axis;
+
 cvar_t	*cl_activeAction;
 
 cvar_t	*cl_motdString;
@@ -102,6 +111,8 @@ cvar_t	*cl_guidServerUniq;
 
 cvar_t	*cl_consoleKeys;
 
+cvar_t  *cl_gamename;
+
 clientActive_t		cl;
 clientConnection_t	clc;
 clientStatic_t		cls;
@@ -129,6 +140,7 @@ int serverStatusCount;
 	void hA3Dg_ExportRenderGeom (refexport_t *incoming_re);
 #endif
 
+extern void GLimp_Minimize(void);
 extern void SV_BotFrame( int time );
 void CL_CheckForResend( void );
 void CL_ShowIP_f(void);
@@ -676,14 +688,14 @@ void CL_Record_f( void ) {
 	if ( Cmd_Argc() == 2 ) {
 		s = Cmd_Argv(1);
 		Q_strncpyz( demoName, s, sizeof( demoName ) );
-		Com_sprintf (name, sizeof(name), "demos/%s.dm_%d", demoName, PROTOCOL_VERSION );
+		Com_sprintf (name, sizeof(name), "demos/%s.%s%d", demoName, DEMOEXT, PROTOCOL_VERSION );
 	} else {
 		int		number;
 
 		// scan for a free demo name
 		for ( number = 0 ; number <= 9999 ; number++ ) {
 			CL_DemoFilename( number, demoName );
-			Com_sprintf (name, sizeof(name), "demos/%s.dm_%d", demoName, PROTOCOL_VERSION );
+			Com_sprintf (name, sizeof(name), "demos/%s.%s%d", demoName, DEMOEXT, PROTOCOL_VERSION );
 
 			if (!FS_FileExists(name))
 				break;	// file doesn't exist
@@ -933,9 +945,22 @@ static void CL_WalkDemoExt(char *arg, char *name, int *demofile)
 {
 	int i = 0;
 	*demofile = 0;
+
+	Com_sprintf (name, MAX_OSPATH, "demos/%s.%s%d", arg, DEMOEXT, PROTOCOL_VERSION);
+
+	FS_FOpenFileRead( name, demofile, qtrue );
+
+	if (*demofile)
+	{
+		Com_Printf("Demo file: %s\n", name);
+		return;
+	}
+
+	Com_Printf("Not found: %s\n", name);
+
 	while(demo_protocols[i])
 	{
-		Com_sprintf (name, MAX_OSPATH, "demos/%s.dm_%d", arg, demo_protocols[i]);
+		Com_sprintf (name, MAX_OSPATH, "demos/%s.%s%d", arg, DEMOEXT, demo_protocols[i]);
 		FS_FOpenFileRead( name, demofile, qtrue );
 		if (*demofile)
 		{
@@ -959,8 +984,8 @@ static void CL_CompleteDemoName( char *args, int argNum )
 	{
 		char demoExt[ 16 ];
 
-		Com_sprintf( demoExt, sizeof( demoExt ), ".dm_%d", PROTOCOL_VERSION );
-		Field_CompleteFilename( "demos", demoExt, qtrue );
+		Com_sprintf(demoExt, sizeof(demoExt), ".%s%d", DEMOEXT, PROTOCOL_VERSION);
+		Field_CompleteFilename( "demos", demoExt, qtrue, qtrue );
 	}
 }
 
@@ -992,34 +1017,41 @@ void CL_PlayDemo_f( void ) {
 	
 	CL_Disconnect( qtrue );
 
-	// check for an extension .dm_?? (?? is protocol)
-	ext_test = arg + strlen(arg) - 6;
-	if ((strlen(arg) > 6) && (ext_test[0] == '.') &&
-		((ext_test[1] == 'd') || (ext_test[1] == 'D')) &&
-		((ext_test[2] == 'm') || (ext_test[2] == 'M')) &&
-		(ext_test[3] == '_'))
+	// check for an extension .DEMOEXT_?? (?? is protocol)
+	ext_test = Q_strrchr(arg, '.');
+	
+	if(ext_test && !Q_stricmpn(ext_test + 1, DEMOEXT, ARRAY_LEN(DEMOEXT) - 1))
 	{
-		protocol = atoi(ext_test+4);
-		i=0;
-		while(demo_protocols[i])
+		protocol = atoi(ext_test + ARRAY_LEN(DEMOEXT));
+
+		for(i = 0; demo_protocols[i]; i++)
 		{
-			if (demo_protocols[i] == protocol)
+			if(demo_protocols[i] == protocol)
 				break;
-			i++;
 		}
-		if (demo_protocols[i])
+
+		if(demo_protocols[i] || protocol == PROTOCOL_VERSION)
 		{
-			Com_sprintf (name, sizeof(name), "demos/%s", arg);
-			FS_FOpenFileRead( name, &clc.demofile, qtrue );
-		} else {
+			Com_sprintf(name, sizeof(name), "demos/%s", arg);
+			FS_FOpenFileRead(name, &clc.demofile, qtrue);
+		}
+		else
+		{
+			int len;
+
 			Com_Printf("Protocol %d not supported for demos\n", protocol);
-			Q_strncpyz(retry, arg, sizeof(retry));
-			retry[strlen(retry)-6] = 0;
-			CL_WalkDemoExt( retry, name, &clc.demofile );
+			len = ext_test - arg;
+
+			if(len >= ARRAY_LEN(retry))
+				len = ARRAY_LEN(retry) - 1;
+
+			Q_strncpyz(retry, arg, len + 1);
+			retry[len] = '\0';
+			CL_WalkDemoExt(retry, name, &clc.demofile);
 		}
-	} else {
-		CL_WalkDemoExt( arg, name, &clc.demofile );
 	}
+	else
+		CL_WalkDemoExt(arg, name, &clc.demofile);
 	
 	if (!clc.demofile) {
 		Com_Error( ERR_DROP, "couldn't open %s", name);
@@ -2421,7 +2453,7 @@ CL_ServersResponsePacket
 ===================
 */
 void CL_ServersResponsePacket( const netadr_t* from, msg_t *msg, qboolean extended ) {
-	int				i, count, total;
+	int				i, j, count, total;
 	netadr_t addresses[MAX_SERVERSPERPACKET];
 	int				numservers;
 	byte*			buffptr;
@@ -2538,6 +2570,18 @@ void CL_ServersResponsePacket( const netadr_t* from, msg_t *msg, qboolean extend
 		// build net address
 		serverInfo_t *server = &cls.globalServers[count];
 
+		// Tequila: It's possible to have sent many master server requests. Then
+		// we may receive many times the same addresses from the master server.
+		// We just avoid to add a server if it is still in the global servers list.
+		for (j = 0; j < count; j++)
+		{
+			if (NET_CompareAdr(cls.globalServers[j].adr, addresses[i]))
+				break;
+		}
+
+		if (j < count)
+			continue;
+
 		CL_InitServerInfo( server, &addresses[i] );
 		Q_strncpyz( server->label, label, sizeof( server->label ) );
 		// advance to next slot
@@ -3363,6 +3407,7 @@ void CL_Init( void ) {
 	// offset for the power function (for style 1, ignored otherwise)
 	// this should be set to the max rate value
 	cl_mouseAccelOffset = Cvar_Get( "cl_mouseAccelOffset", "5", CVAR_ARCHIVE );
+	Cvar_CheckRange(cl_mouseAccelOffset, 0.001f, 50000.0f, qfalse);
 
 	cl_showMouseRate = Cvar_Get ("cl_showmouserate", "0", 0);
 
@@ -3394,6 +3439,15 @@ void CL_Init( void ) {
 	m_filter = Cvar_Get ("m_filter", "0", CVAR_ARCHIVE);
 #endif
 
+	j_pitch =        Cvar_Get ("j_pitch",        "0.022", CVAR_ARCHIVE);
+	j_yaw =          Cvar_Get ("j_yaw",          "-0.022", CVAR_ARCHIVE);
+	j_forward =      Cvar_Get ("j_forward",      "-0.25", CVAR_ARCHIVE);
+	j_side =         Cvar_Get ("j_side",         "0.25", CVAR_ARCHIVE);
+	j_pitch_axis =   Cvar_Get ("j_pitch_axis",   "3", CVAR_ARCHIVE);
+	j_yaw_axis =     Cvar_Get ("j_yaw_axis",     "4", CVAR_ARCHIVE);
+	j_forward_axis = Cvar_Get ("j_forward_axis", "1", CVAR_ARCHIVE);
+	j_side_axis =    Cvar_Get ("j_side_axis",    "0", CVAR_ARCHIVE);
+
 	cl_motdString = Cvar_Get( "cl_motdString", "", CVAR_ROM );
 
 	Cvar_Get( "cl_maxPing", "800", CVAR_ARCHIVE );
@@ -3405,6 +3459,8 @@ void CL_Init( void ) {
 	// ~ and `, as keys and characters
 	cl_consoleKeys = Cvar_Get( "cl_consoleKeys", "~ ` 0x7e 0x60", CVAR_ARCHIVE);
 
+	cl_gamename = Cvar_Get("cl_gamename", GAMENAME_FOR_MASTER, CVAR_TEMP);
+
 	// userinfo
 	Cvar_Get ("name", Sys_GetCurrentUser( ), CVAR_USERINFO | CVAR_ARCHIVE );
 	Cvar_Get ("rate", "25000", CVAR_USERINFO | CVAR_ARCHIVE );
@@ -3481,6 +3537,7 @@ void CL_Init( void ) {
 	Cmd_AddCommand ("model", CL_SetModel_f );
 	Cmd_AddCommand ("video", CL_Video_f );
 	Cmd_AddCommand ("stopvideo", CL_StopVideo_f );
+	Cmd_AddCommand("minimize", GLimp_Minimize);
 	CL_InitRef();
 
 	SCR_Init ();
@@ -3944,9 +4001,9 @@ void CL_GlobalServers_f( void ) {
 	int			count, i, masterNum;
 	char		command[1024], *masteraddress;
 	
-	if ((count = Cmd_Argc()) < 3 || (masterNum = atoi(Cmd_Argv(1))) < 0 || masterNum > 4)
+	if ((count = Cmd_Argc()) < 3 || (masterNum = atoi(Cmd_Argv(1))) < 0 || masterNum > MAX_MASTER_SERVERS - 1)
 	{
-		Com_Printf( "usage: globalservers <master# 0-4> <protocol> [keywords]\n");
+		Com_Printf("usage: globalservers <master# 0-%d> <protocol> [keywords]\n", MAX_MASTER_SERVERS - 1);
 		return;	
 	}
 
@@ -3977,10 +4034,27 @@ void CL_GlobalServers_f( void ) {
 	cls.numglobalservers = -1;
 	cls.pingUpdateSource = AS_GLOBAL;
 
-	// TODO: test if we only have an IPv6 connection. If it's the case,
-	//       request IPv6 servers only by appending " ipv6" to the command
-	Com_sprintf( command, sizeof(command), "getserversExt "
-		GAMENAME_FOR_MASTER " %s", Cmd_Argv(2) );
+	// Use the extended query for IPv6 masters
+	if (to.type == NA_IP6 || to.type == NA_MULTICAST6)
+	{
+		int v4enabled = Cvar_VariableIntegerValue("net_enabled") & NET_ENABLEV4;
+		
+		if(v4enabled)
+		{
+			Com_sprintf(command, sizeof(command), "getserversExt %s %s ipv6",
+				cl_gamename->string, Cmd_Argv(2));
+		}
+		else
+		{
+			Com_sprintf(command, sizeof(command), "getserversExt %s %s",
+				cl_gamename->string, Cmd_Argv(2));
+		}
+
+		// TODO: test if we only have an IPv6 connection. If it's the case,
+		//       request IPv6 servers only by appending " ipv6" to the command
+	}
+	else
+		Com_sprintf(command, sizeof(command), "getservers %s", Cmd_Argv(2));
 
 	for (i=3; i < count; i++)
 	{
@@ -4004,9 +4078,9 @@ void CL_GetPing( int n, char *buf, int buflen, int *pingtime )
 	int		time;
 	int		maxPing;
 
-	if (!cl_pinglist[n].adr.port)
+	if (n < 0 || n >= MAX_PINGREQUESTS || !cl_pinglist[n].adr.port)
 	{
-		// empty slot
+		// empty or invalid slot
 		buf[0]    = '\0';
 		*pingtime = 0;
 		return;
@@ -4038,29 +4112,14 @@ void CL_GetPing( int n, char *buf, int buflen, int *pingtime )
 
 /*
 ==================
-CL_UpdateServerInfo
-==================
-*/
-void CL_UpdateServerInfo( int n )
-{
-	if (!cl_pinglist[n].adr.port)
-	{
-		return;
-	}
-
-	CL_SetServerInfoByAddress(cl_pinglist[n].adr, cl_pinglist[n].info, cl_pinglist[n].time );
-}
-
-/*
-==================
 CL_GetPingInfo
 ==================
 */
 void CL_GetPingInfo( int n, char *buf, int buflen )
 {
-	if (!cl_pinglist[n].adr.port)
+	if (n < 0 || n >= MAX_PINGREQUESTS || !cl_pinglist[n].adr.port)
 	{
-		// empty slot
+		// empty or invalid slot
 		if (buflen)
 			buf[0] = '\0';
 		return;
diff --git a/src/client/cl_parse.c b/src/client/cl_parse.c
index 64404f78..42465f1b 100644
--- a/src/client/cl_parse.c
+++ b/src/client/cl_parse.c
@@ -413,13 +413,13 @@ void CL_SystemInfoChanged( void ) {
 		else
 		{
 			// If this cvar may not be modified by a server discard the value.
-			if(!(cvar_flags & (CVAR_SYSTEMINFO | CVAR_SERVER_CREATED)))
+			if(!(cvar_flags & (CVAR_SYSTEMINFO | CVAR_SERVER_CREATED | CVAR_USER_CREATED)))
 			{
 				Com_Printf(S_COLOR_YELLOW "WARNING: server is not allowed to set %s=%s\n", key, value);
 				continue;
 			}
 
-			Cvar_Set(key, value);
+			Cvar_SetSafe(key, value);
 		}
 	}
 	// if game folder should not be set and it is set at the client side
diff --git a/src/client/cl_ui.c b/src/client/cl_ui.c
index 6b748cf7..edf5ef35 100644
--- a/src/client/cl_ui.c
+++ b/src/client/cl_ui.c
@@ -724,7 +724,7 @@ intptr_t CL_UISystemCalls( intptr_t *args ) {
 		return 0;
 
 	case UI_CVAR_SET:
-		Cvar_Set( VMA(1), VMA(2) );
+		Cvar_SetSafe( VMA(1), VMA(2) );
 		return 0;
 
 	case UI_CVAR_VARIABLEVALUE:
@@ -735,7 +735,7 @@ intptr_t CL_UISystemCalls( intptr_t *args ) {
 		return 0;
 
 	case UI_CVAR_SETVALUE:
-		Cvar_SetValue( VMA(1), VMF(2) );
+		Cvar_SetValueSafe( VMA(1), VMF(2) );
 		return 0;
 
 	case UI_CVAR_RESET:
diff --git a/src/client/client.h b/src/client/client.h
index 0afdc686..26af5072 100644
--- a/src/client/client.h
+++ b/src/client/client.h
@@ -397,6 +397,15 @@ extern	cvar_t	*m_forward;
 extern	cvar_t	*m_side;
 extern	cvar_t	*m_filter;
 
+extern	cvar_t	*j_pitch;
+extern	cvar_t	*j_yaw;
+extern	cvar_t	*j_forward;
+extern	cvar_t	*j_side;
+extern	cvar_t	*j_pitch_axis;
+extern	cvar_t	*j_yaw_axis;
+extern	cvar_t	*j_forward_axis;
+extern	cvar_t	*j_side_axis;
+
 extern	cvar_t	*cl_timedemo;
 extern	cvar_t	*cl_aviFrameRate;
 extern	cvar_t	*cl_aviMotionJpeg;
diff --git a/src/client/libmumblelink.c b/src/client/libmumblelink.c
index 10e90011..25ca6786 100644
--- a/src/client/libmumblelink.c
+++ b/src/client/libmumblelink.c
@@ -26,6 +26,9 @@
 #define uint32_t UINT32
 #else
 #include <unistd.h>
+#ifdef __sun
+#define _POSIX_C_SOURCE 199309L
+#endif
 #include <sys/mman.h>
 #include <sys/types.h>
 #include <sys/stat.h>
diff --git a/src/client/qal.h b/src/client/qal.h
index f999dc4f..7665e6ef 100644
--- a/src/client/qal.h
+++ b/src/client/qal.h
@@ -48,6 +48,13 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 #endif
 #endif
 
+/* Hack to enable compiling both on OpenAL SDK and OpenAL-soft. */
+#ifndef ALC_ENUMERATE_ALL_EXT
+#  define ALC_ENUMERATE_ALL_EXT 1
+#  define ALC_DEFAULT_ALL_DEVICES_SPECIFIER        0x1012
+#  define ALC_ALL_DEVICES_SPECIFIER                0x1013
+#endif
+
 #ifdef USE_OPENAL_DLOPEN
 extern LPALENABLE qalEnable;
 extern LPALDISABLE qalDisable;
diff --git a/src/client/snd_codec_ogg.c b/src/client/snd_codec_ogg.c
index eb754ca0..0ca2f689 100644
--- a/src/client/snd_codec_ogg.c
+++ b/src/client/snd_codec_ogg.c
@@ -32,6 +32,7 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 
 // includes for the OGG codec
 #include <errno.h>
+#define OV_EXCLUDE_STATIC_CALLBACKS
 #include <vorbis/vorbisfile.h>
 
 // The OGG codec can return the samples in a number of different formats,
diff --git a/src/client/snd_main.c b/src/client/snd_main.c
index 1215e954..22213cc9 100644
--- a/src/client/snd_main.c
+++ b/src/client/snd_main.c
@@ -425,12 +425,9 @@ void S_Play_f( void ) {
 
 	i = 1;
 	while ( i<Cmd_Argc() ) {
-		if ( !Q_strrchr(Cmd_Argv(i), '.') ) {
-			Com_sprintf( name, sizeof(name), "%s.wav", Cmd_Argv(1) );
-		} else {
-			Q_strncpyz( name, Cmd_Argv(i), sizeof(name) );
-		}
+		Q_strncpyz( name, Cmd_Argv(i), sizeof(name) );
 		h = si.RegisterSound( name, qfalse );
+
 		if( h ) {
 			si.StartLocalSound( h, CHAN_LOCAL_SOUND );
 		}
diff --git a/src/client/snd_openal.c b/src/client/snd_openal.c
index 6f48713f..9b60aa71 100644
--- a/src/client/snd_openal.c
+++ b/src/client/snd_openal.c
@@ -42,7 +42,13 @@ cvar_t *s_alRolloff;
 cvar_t *s_alGraceDistance;
 cvar_t *s_alDriver;
 cvar_t *s_alDevice;
+cvar_t *s_alInputDevice;
 cvar_t *s_alAvailableDevices;
+cvar_t *s_alAvailableInputDevices;
+
+static qboolean enumeration_ext = qfalse;
+static qboolean enumeration_all_ext = qfalse;
+static qboolean capture_ext = qfalse;
 
 /*
 =================
@@ -630,7 +636,7 @@ static void S_AL_ScaleGain(src_t *chksrc, vec3_t origin)
 		
 		scaleFactor *= chksrc->curGain;
 		
-		if(chksrc->scaleGain != scaleFactor);
+		if(chksrc->scaleGain != scaleFactor)
 		{
 			chksrc->scaleGain = scaleFactor;
 			S_AL_Gain(chksrc->alSource, chksrc->scaleGain);
@@ -2286,22 +2292,34 @@ void S_AL_MasterGain( float gain )
 S_AL_SoundInfo
 =================
 */
-static
-void S_AL_SoundInfo( void )
+static void S_AL_SoundInfo(void)
 {
 	Com_Printf( "OpenAL info:\n" );
-	Com_Printf( "  Vendor:     %s\n", qalGetString( AL_VENDOR ) );
-	Com_Printf( "  Version:    %s\n", qalGetString( AL_VERSION ) );
-	Com_Printf( "  Renderer:   %s\n", qalGetString( AL_RENDERER ) );
-	Com_Printf( "  AL Extensions: %s\n", qalGetString( AL_EXTENSIONS ) );
+	Com_Printf( "  Vendor:         %s\n", qalGetString( AL_VENDOR ) );
+	Com_Printf( "  Version:        %s\n", qalGetString( AL_VERSION ) );
+	Com_Printf( "  Renderer:       %s\n", qalGetString( AL_RENDERER ) );
+	Com_Printf( "  AL Extensions:  %s\n", qalGetString( AL_EXTENSIONS ) );
 	Com_Printf( "  ALC Extensions: %s\n", qalcGetString( alDevice, ALC_EXTENSIONS ) );
-	if(qalcIsExtensionPresent(NULL, "ALC_ENUMERATION_EXT"))
+
+	if(enumeration_all_ext)
+		Com_Printf("  Device:         %s\n", qalcGetString(alDevice, ALC_ALL_DEVICES_SPECIFIER));
+	else if(enumeration_ext)
+		Com_Printf("  Device:         %s\n", qalcGetString(alDevice, ALC_DEVICE_SPECIFIER));
+
+	if(enumeration_all_ext || enumeration_ext)
+		Com_Printf("  Available Devices:\n%s", s_alAvailableDevices->string);
+
+#ifdef USE_VOIP
+	if(capture_ext)
 	{
-		Com_Printf("  Device:     %s\n", qalcGetString(alDevice, ALC_DEVICE_SPECIFIER));
-		Com_Printf("Available Devices:\n%s", s_alAvailableDevices->string);
+		Com_Printf("  Input Device:   %s\n", qalcGetString(alCaptureDevice, ALC_CAPTURE_DEVICE_SPECIFIER));
+		Com_Printf("  Available Input Devices:\n%s", s_alAvailableInputDevices->string);
 	}
+#endif
 }
 
+
+
 /*
 =================
 S_AL_Shutdown
@@ -2350,6 +2368,7 @@ qboolean S_AL_Init( soundInterface_t *si )
 {
 #ifdef USE_OPENAL
 	const char* device = NULL;
+	const char* inputdevice = NULL;
 	int i;
 
 	if( !si ) {
@@ -2375,6 +2394,7 @@ qboolean S_AL_Init( soundInterface_t *si )
 
 	s_alDriver = Cvar_Get( "s_alDriver", ALDRIVER_DEFAULT, CVAR_ARCHIVE | CVAR_LATCH );
 
+	s_alInputDevice = Cvar_Get( "s_alInputDevice", "", CVAR_ARCHIVE | CVAR_LATCH );
 	s_alDevice = Cvar_Get("s_alDevice", "", CVAR_ARCHIVE | CVAR_LATCH);
 
 	// Load QAL
@@ -2388,17 +2408,35 @@ qboolean S_AL_Init( soundInterface_t *si )
 	if(device && !*device)
 		device = NULL;
 
-	// Device enumeration support (extension is implemented reasonably only on Windows right now).
-	if(qalcIsExtensionPresent(NULL, "ALC_ENUMERATION_EXT"))
+	inputdevice = s_alInputDevice->string;
+	if(inputdevice && !*inputdevice)
+		inputdevice = NULL;
+
+
+	// Device enumeration support
+	enumeration_all_ext = qalcIsExtensionPresent(NULL, "ALC_ENUMERATE_ALL_EXT");
+	enumeration_ext = qalcIsExtensionPresent(NULL, "ALC_ENUMERATION_EXT");
+
+	if(enumeration_ext || enumeration_all_ext)
 	{
-		char devicenames[1024] = "";
+		char devicenames[16384] = "";
 		const char *devicelist;
 		const char *defaultdevice;
 		int curlen;
-		
+
 		// get all available devices + the default device name.
-		devicelist = qalcGetString(NULL, ALC_DEVICE_SPECIFIER);
-		defaultdevice = qalcGetString(NULL, ALC_DEFAULT_DEVICE_SPECIFIER);
+		if(enumeration_ext)
+		{
+			devicelist = qalcGetString(NULL, ALC_ALL_DEVICES_SPECIFIER);
+			defaultdevice = qalcGetString(NULL, ALC_DEFAULT_ALL_DEVICES_SPECIFIER);
+		}
+		else
+		{
+			// We don't have ALC_ENUMERATE_ALL_EXT but normal enumeration.
+			devicelist = qalcGetString(NULL, ALC_DEVICE_SPECIFIER);
+			defaultdevice = qalcGetString(NULL, ALC_DEFAULT_DEVICE_SPECIFIER);
+			enumeration_ext = qtrue;
+		}
 
 #ifdef _WIN32
 		// check whether the default device is generic hardware. If it is, change to
@@ -2487,15 +2525,40 @@ qboolean S_AL_Init( soundInterface_t *si )
 		}
 		else
 		{
+			char inputdevicenames[16384] = "";
+			const char *inputdevicelist;
+			const char *defaultinputdevice;
+			int curlen;
+
+			capture_ext = qtrue;
+
+			// get all available input devices + the default input device name.
+			inputdevicelist = qalcGetString(NULL, ALC_CAPTURE_DEVICE_SPECIFIER);
+			defaultinputdevice = qalcGetString(NULL, ALC_CAPTURE_DEFAULT_DEVICE_SPECIFIER);
+
+			// dump a list of available devices to a cvar for the user to see.
+			while((curlen = strlen(inputdevicelist)))
+			{
+				Q_strcat(inputdevicenames, sizeof(inputdevicenames), inputdevicelist);
+				Q_strcat(inputdevicenames, sizeof(inputdevicenames), "\n");
+				inputdevicelist += curlen + 1;
+			}
+
+			s_alAvailableInputDevices = Cvar_Get("s_alAvailableInputDevices", inputdevicenames, CVAR_ROM | CVAR_NORESTART);
+
 			// !!! FIXME: 8000Hz is what Speex narrowband mode needs, but we
 			// !!! FIXME:  should probably open the capture device after
 			// !!! FIXME:  initializing Speex so we can change to wideband
 			// !!! FIXME:  if we like.
-			Com_Printf("OpenAL default capture device is '%s'\n",
-			           qalcGetString(NULL, ALC_CAPTURE_DEFAULT_DEVICE_SPECIFIER));
-			alCaptureDevice = qalcCaptureOpenDevice(NULL, 8000, AL_FORMAT_MONO16, 4096);
+			Com_Printf("OpenAL default capture device is '%s'\n", defaultinputdevice);
+			alCaptureDevice = qalcCaptureOpenDevice(inputdevice, 8000, AL_FORMAT_MONO16, 4096);
+			if( !alCaptureDevice && inputdevice )
+			{
+				Com_Printf( "Failed to open OpenAL Input device '%s', trying default.\n", inputdevice );
+				alCaptureDevice = qalcCaptureOpenDevice(NULL, 8000, AL_FORMAT_MONO16, 4096);
+			}
 			Com_Printf( "OpenAL capture device %s.\n",
-			            (alCaptureDevice == NULL) ? "failed to open" : "opened");
+				    (alCaptureDevice == NULL) ? "failed to open" : "opened");
 		}
 	}
 #endif
diff --git a/src/game/bg_lib.h b/src/game/bg_lib.h
index 13739a0c..aafdd8fd 100644
--- a/src/game/bg_lib.h
+++ b/src/game/bg_lib.h
@@ -109,6 +109,9 @@ int     atoi( const char *string );
 int     _atoi( const char **stringPtr );
 long    strtol( const char *nptr, char **endptr, int base );
 
+int Q_vsnprintf( char *buffer, size_t length, const char *fmt, va_list argptr );
+int Q_snprintf( char *buffer, size_t length, const char *fmt, ... ) __attribute__ ((format (printf, 3, 4)));
+
 int     sscanf( const char *buffer, const char *fmt, ... );
 
 // Memory functions
diff --git a/src/jpeg-6b/README b/src/jpeg-6b/README
deleted file mode 100644
index 86cc2066..00000000
--- a/src/jpeg-6b/README
+++ /dev/null
@@ -1,385 +0,0 @@
-The Independent JPEG Group's JPEG software
-==========================================
-
-README for release 6b of 27-Mar-1998
-====================================
-
-This distribution contains the sixth public release of the Independent JPEG
-Group's free JPEG software.  You are welcome to redistribute this software and
-to use it for any purpose, subject to the conditions under LEGAL ISSUES, below.
-
-Serious users of this software (particularly those incorporating it into
-larger programs) should contact IJG at jpeg-info@uunet.uu.net to be added to
-our electronic mailing list.  Mailing list members are notified of updates
-and have a chance to participate in technical discussions, etc.
-
-This software is the work of Tom Lane, Philip Gladstone, Jim Boucher,
-Lee Crocker, Julian Minguillon, Luis Ortiz, George Phillips, Davide Rossi,
-Guido Vollbeding, Ge' Weijers, and other members of the Independent JPEG
-Group.
-
-IJG is not affiliated with the official ISO JPEG standards committee.
-
-
-DOCUMENTATION ROADMAP
-=====================
-
-This file contains the following sections:
-
-OVERVIEW            General description of JPEG and the IJG software.
-LEGAL ISSUES        Copyright, lack of warranty, terms of distribution.
-REFERENCES          Where to learn more about JPEG.
-ARCHIVE LOCATIONS   Where to find newer versions of this software.
-RELATED SOFTWARE    Other stuff you should get.
-FILE FORMAT WARS    Software *not* to get.
-TO DO               Plans for future IJG releases.
-
-Other documentation files in the distribution are:
-
-User documentation:
-  install.doc       How to configure and install the IJG software.
-  usage.doc         Usage instructions for cjpeg, djpeg, jpegtran,
-                    rdjpgcom, and wrjpgcom.
-  *.1               Unix-style man pages for programs (same info as usage.doc).
-  wizard.doc        Advanced usage instructions for JPEG wizards only.
-  change.log        Version-to-version change highlights.
-Programmer and internal documentation:
-  libjpeg.doc       How to use the JPEG library in your own programs.
-  example.c         Sample code for calling the JPEG library.
-  structure.doc     Overview of the JPEG library's internal structure.
-  filelist.doc      Road map of IJG files.
-  coderules.doc     Coding style rules --- please read if you contribute code.
-
-Please read at least the files install.doc and usage.doc.  Useful information
-can also be found in the JPEG FAQ (Frequently Asked Questions) article.  See
-ARCHIVE LOCATIONS below to find out where to obtain the FAQ article.
-
-If you want to understand how the JPEG code works, we suggest reading one or
-more of the REFERENCES, then looking at the documentation files (in roughly
-the order listed) before diving into the code.
-
-
-OVERVIEW
-========
-
-This package contains C software to implement JPEG image compression and
-decompression.  JPEG (pronounced "jay-peg") is a standardized compression
-method for full-color and gray-scale images.  JPEG is intended for compressing
-"real-world" scenes; line drawings, cartoons and other non-realistic images
-are not its strong suit.  JPEG is lossy, meaning that the output image is not
-exactly identical to the input image.  Hence you must not use JPEG if you
-have to have identical output bits.  However, on typical photographic images,
-very good compression levels can be obtained with no visible change, and
-remarkably high compression levels are possible if you can tolerate a
-low-quality image.  For more details, see the references, or just experiment
-with various compression settings.
-
-This software implements JPEG baseline, extended-sequential, and progressive
-compression processes.  Provision is made for supporting all variants of these
-processes, although some uncommon parameter settings aren't implemented yet.
-For legal reasons, we are not distributing code for the arithmetic-coding
-variants of JPEG; see LEGAL ISSUES.  We have made no provision for supporting
-the hierarchical or lossless processes defined in the standard.
-
-We provide a set of library routines for reading and writing JPEG image files,
-plus two sample applications "cjpeg" and "djpeg", which use the library to
-perform conversion between JPEG and some other popular image file formats.
-The library is intended to be reused in other applications.
-
-In order to support file conversion and viewing software, we have included
-considerable functionality beyond the bare JPEG coding/decoding capability;
-for example, the color quantization modules are not strictly part of JPEG
-decoding, but they are essential for output to colormapped file formats or
-colormapped displays.  These extra functions can be compiled out of the
-library if not required for a particular application.  We have also included
-"jpegtran", a utility for lossless transcoding between different JPEG
-processes, and "rdjpgcom" and "wrjpgcom", two simple applications for
-inserting and extracting textual comments in JFIF files.
-
-The emphasis in designing this software has been on achieving portability and
-flexibility, while also making it fast enough to be useful.  In particular,
-the software is not intended to be read as a tutorial on JPEG.  (See the
-REFERENCES section for introductory material.)  Rather, it is intended to
-be reliable, portable, industrial-strength code.  We do not claim to have
-achieved that goal in every aspect of the software, but we strive for it.
-
-We welcome the use of this software as a component of commercial products.
-No royalty is required, but we do ask for an acknowledgement in product
-documentation, as described under LEGAL ISSUES.
-
-
-LEGAL ISSUES
-============
-
-In plain English:
-
-1. We don't promise that this software works.  (But if you find any bugs,
-   please let us know!)
-2. You can use this software for whatever you want.  You don't have to pay us.
-3. You may not pretend that you wrote this software.  If you use it in a
-   program, you must acknowledge somewhere in your documentation that
-   you've used the IJG code.
-
-In legalese:
-
-The authors make NO WARRANTY or representation, either express or implied,
-with respect to this software, its quality, accuracy, merchantability, or
-fitness for a particular purpose.  This software is provided "AS IS", and you,
-its user, assume the entire risk as to its quality and accuracy.
-
-This software is copyright (C) 1991-1998, Thomas G. Lane.
-All Rights Reserved except as specified below.
-
-Permission is hereby granted to use, copy, modify, and distribute this
-software (or portions thereof) for any purpose, without fee, subject to these
-conditions:
-(1) If any part of the source code for this software is distributed, then this
-README file must be included, with this copyright and no-warranty notice
-unaltered; and any additions, deletions, or changes to the original files
-must be clearly indicated in accompanying documentation.
-(2) If only executable code is distributed, then the accompanying
-documentation must state that "this software is based in part on the work of
-the Independent JPEG Group".
-(3) Permission for use of this software is granted only if the user accepts
-full responsibility for any undesirable consequences; the authors accept
-NO LIABILITY for damages of any kind.
-
-These conditions apply to any software derived from or based on the IJG code,
-not just to the unmodified library.  If you use our work, you ought to
-acknowledge us.
-
-Permission is NOT granted for the use of any IJG author's name or company name
-in advertising or publicity relating to this software or products derived from
-it.  This software may be referred to only as "the Independent JPEG Group's
-software".
-
-We specifically permit and encourage the use of this software as the basis of
-commercial products, provided that all warranty or liability claims are
-assumed by the product vendor.
-
-
-ansi2knr.c is included in this distribution by permission of L. Peter Deutsch,
-sole proprietor of its copyright holder, Aladdin Enterprises of Menlo Park, CA.
-ansi2knr.c is NOT covered by the above copyright and conditions, but instead
-by the usual distribution terms of the Free Software Foundation; principally,
-that you must include source code if you redistribute it.  (See the file
-ansi2knr.c for full details.)  However, since ansi2knr.c is not needed as part
-of any program generated from the IJG code, this does not limit you more than
-the foregoing paragraphs do.
-
-The Unix configuration script "configure" was produced with GNU Autoconf.
-It is copyright by the Free Software Foundation but is freely distributable.
-The same holds for its supporting scripts (config.guess, config.sub,
-ltconfig, ltmain.sh).  Another support script, install-sh, is copyright
-by M.I.T. but is also freely distributable.
-
-It appears that the arithmetic coding option of the JPEG spec is covered by
-patents owned by IBM, AT&T, and Mitsubishi.  Hence arithmetic coding cannot
-legally be used without obtaining one or more licenses.  For this reason,
-support for arithmetic coding has been removed from the free JPEG software.
-(Since arithmetic coding provides only a marginal gain over the unpatented
-Huffman mode, it is unlikely that very many implementations will support it.)
-So far as we are aware, there are no patent restrictions on the remaining
-code.
-
-The IJG distribution formerly included code to read and write GIF files.
-To avoid entanglement with the Unisys LZW patent, GIF reading support has
-been removed altogether, and the GIF writer has been simplified to produce
-"uncompressed GIFs".  This technique does not use the LZW algorithm; the
-resulting GIF files are larger than usual, but are readable by all standard
-GIF decoders.
-
-We are required to state that
-    "The Graphics Interchange Format(c) is the Copyright property of
-    CompuServe Incorporated.  GIF(sm) is a Service Mark property of
-    CompuServe Incorporated."
-
-
-REFERENCES
-==========
-
-We highly recommend reading one or more of these references before trying to
-understand the innards of the JPEG software.
-
-The best short technical introduction to the JPEG compression algorithm is
-	Wallace, Gregory K.  "The JPEG Still Picture Compression Standard",
-	Communications of the ACM, April 1991 (vol. 34 no. 4), pp. 30-44.
-(Adjacent articles in that issue discuss MPEG motion picture compression,
-applications of JPEG, and related topics.)  If you don't have the CACM issue
-handy, a PostScript file containing a revised version of Wallace's article is
-available at ftp://ftp.uu.net/graphics/jpeg/wallace.ps.gz.  The file (actually
-a preprint for an article that appeared in IEEE Trans. Consumer Electronics)
-omits the sample images that appeared in CACM, but it includes corrections
-and some added material.  Note: the Wallace article is copyright ACM and IEEE,
-and it may not be used for commercial purposes.
-
-A somewhat less technical, more leisurely introduction to JPEG can be found in
-"The Data Compression Book" by Mark Nelson and Jean-loup Gailly, published by
-M&T Books (New York), 2nd ed. 1996, ISBN 1-55851-434-1.  This book provides
-good explanations and example C code for a multitude of compression methods
-including JPEG.  It is an excellent source if you are comfortable reading C
-code but don't know much about data compression in general.  The book's JPEG
-sample code is far from industrial-strength, but when you are ready to look
-at a full implementation, you've got one here...
-
-The best full description of JPEG is the textbook "JPEG Still Image Data
-Compression Standard" by William B. Pennebaker and Joan L. Mitchell, published
-by Van Nostrand Reinhold, 1993, ISBN 0-442-01272-1.  Price US$59.95, 638 pp.
-The book includes the complete text of the ISO JPEG standards (DIS 10918-1
-and draft DIS 10918-2).  This is by far the most complete exposition of JPEG
-in existence, and we highly recommend it.
-
-The JPEG standard itself is not available electronically; you must order a
-paper copy through ISO or ITU.  (Unless you feel a need to own a certified
-official copy, we recommend buying the Pennebaker and Mitchell book instead;
-it's much cheaper and includes a great deal of useful explanatory material.)
-In the USA, copies of the standard may be ordered from ANSI Sales at (212)
-642-4900, or from Global Engineering Documents at (800) 854-7179.  (ANSI
-doesn't take credit card orders, but Global does.)  It's not cheap: as of
-1992, ANSI was charging $95 for Part 1 and $47 for Part 2, plus 7%
-shipping/handling.  The standard is divided into two parts, Part 1 being the
-actual specification, while Part 2 covers compliance testing methods.  Part 1
-is titled "Digital Compression and Coding of Continuous-tone Still Images,
-Part 1: Requirements and guidelines" and has document numbers ISO/IEC IS
-10918-1, ITU-T T.81.  Part 2 is titled "Digital Compression and Coding of
-Continuous-tone Still Images, Part 2: Compliance testing" and has document
-numbers ISO/IEC IS 10918-2, ITU-T T.83.
-
-Some extensions to the original JPEG standard are defined in JPEG Part 3,
-a newer ISO standard numbered ISO/IEC IS 10918-3 and ITU-T T.84.  IJG
-currently does not support any Part 3 extensions.
-
-The JPEG standard does not specify all details of an interchangeable file
-format.  For the omitted details we follow the "JFIF" conventions, revision
-1.02.  A copy of the JFIF spec is available from:
-	Literature Department
-	C-Cube Microsystems, Inc.
-	1778 McCarthy Blvd.
-	Milpitas, CA 95035
-	phone (408) 944-6300,  fax (408) 944-6314
-A PostScript version of this document is available by FTP at
-ftp://ftp.uu.net/graphics/jpeg/jfif.ps.gz.  There is also a plain text
-version at ftp://ftp.uu.net/graphics/jpeg/jfif.txt.gz, but it is missing
-the figures.
-
-The TIFF 6.0 file format specification can be obtained by FTP from
-ftp://ftp.sgi.com/graphics/tiff/TIFF6.ps.gz.  The JPEG incorporation scheme
-found in the TIFF 6.0 spec of 3-June-92 has a number of serious problems.
-IJG does not recommend use of the TIFF 6.0 design (TIFF Compression tag 6).
-Instead, we recommend the JPEG design proposed by TIFF Technical Note #2
-(Compression tag 7).  Copies of this Note can be obtained from ftp.sgi.com or
-from ftp://ftp.uu.net/graphics/jpeg/.  It is expected that the next revision
-of the TIFF spec will replace the 6.0 JPEG design with the Note's design.
-Although IJG's own code does not support TIFF/JPEG, the free libtiff library
-uses our library to implement TIFF/JPEG per the Note.  libtiff is available
-from ftp://ftp.sgi.com/graphics/tiff/.
-
-
-ARCHIVE LOCATIONS
-=================
-
-The "official" archive site for this software is ftp.uu.net (Internet
-address 192.48.96.9).  The most recent released version can always be found
-there in directory graphics/jpeg.  This particular version will be archived
-as ftp://ftp.uu.net/graphics/jpeg/jpegsrc.v6b.tar.gz.  If you don't have
-direct Internet access, UUNET's archives are also available via UUCP; contact
-help@uunet.uu.net for information on retrieving files that way.
-
-Numerous Internet sites maintain copies of the UUNET files.  However, only
-ftp.uu.net is guaranteed to have the latest official version.
-
-You can also obtain this software in DOS-compatible "zip" archive format from
-the SimTel archives (ftp://ftp.simtel.net/pub/simtelnet/msdos/graphics/), or
-on CompuServe in the Graphics Support forum (GO CIS:GRAPHSUP), library 12
-"JPEG Tools".  Again, these versions may sometimes lag behind the ftp.uu.net
-release.
-
-The JPEG FAQ (Frequently Asked Questions) article is a useful source of
-general information about JPEG.  It is updated constantly and therefore is
-not included in this distribution.  The FAQ is posted every two weeks to
-Usenet newsgroups comp.graphics.misc, news.answers, and other groups.
-It is available on the World Wide Web at http://www.faqs.org/faqs/jpeg-faq/
-and other news.answers archive sites, including the official news.answers
-archive at rtfm.mit.edu: ftp://rtfm.mit.edu/pub/usenet/news.answers/jpeg-faq/.
-If you don't have Web or FTP access, send e-mail to mail-server@rtfm.mit.edu
-with body
-	send usenet/news.answers/jpeg-faq/part1
-	send usenet/news.answers/jpeg-faq/part2
-
-
-RELATED SOFTWARE
-================
-
-Numerous viewing and image manipulation programs now support JPEG.  (Quite a
-few of them use this library to do so.)  The JPEG FAQ described above lists
-some of the more popular free and shareware viewers, and tells where to
-obtain them on Internet.
-
-If you are on a Unix machine, we highly recommend Jef Poskanzer's free
-PBMPLUS software, which provides many useful operations on PPM-format image
-files.  In particular, it can convert PPM images to and from a wide range of
-other formats, thus making cjpeg/djpeg considerably more useful.  The latest
-version is distributed by the NetPBM group, and is available from numerous
-sites, notably ftp://wuarchive.wustl.edu/graphics/graphics/packages/NetPBM/.
-Unfortunately PBMPLUS/NETPBM is not nearly as portable as the IJG software is;
-you are likely to have difficulty making it work on any non-Unix machine.
-
-A different free JPEG implementation, written by the PVRG group at Stanford,
-is available from ftp://havefun.stanford.edu/pub/jpeg/.  This program
-is designed for research and experimentation rather than production use;
-it is slower, harder to use, and less portable than the IJG code, but it
-is easier to read and modify.  Also, the PVRG code supports lossless JPEG,
-which we do not.  (On the other hand, it doesn't do progressive JPEG.)
-
-
-FILE FORMAT WARS
-================
-
-Some JPEG programs produce files that are not compatible with our library.
-The root of the problem is that the ISO JPEG committee failed to specify a
-concrete file format.  Some vendors "filled in the blanks" on their own,
-creating proprietary formats that no one else could read.  (For example, none
-of the early commercial JPEG implementations for the Macintosh were able to
-exchange compressed files.)
-
-The file format we have adopted is called JFIF (see REFERENCES).  This format
-has been agreed to by a number of major commercial JPEG vendors, and it has
-become the de facto standard.  JFIF is a minimal or "low end" representation.
-We recommend the use of TIFF/JPEG (TIFF revision 6.0 as modified by TIFF
-Technical Note #2) for "high end" applications that need to record a lot of
-additional data about an image.  TIFF/JPEG is fairly new and not yet widely
-supported, unfortunately.
-
-The upcoming JPEG Part 3 standard defines a file format called SPIFF.
-SPIFF is interoperable with JFIF, in the sense that most JFIF decoders should
-be able to read the most common variant of SPIFF.  SPIFF has some technical
-advantages over JFIF, but its major claim to fame is simply that it is an
-official standard rather than an informal one.  At this point it is unclear
-whether SPIFF will supersede JFIF or whether JFIF will remain the de-facto
-standard.  IJG intends to support SPIFF once the standard is frozen, but we
-have not decided whether it should become our default output format or not.
-(In any case, our decoder will remain capable of reading JFIF indefinitely.)
-
-Various proprietary file formats incorporating JPEG compression also exist.
-We have little or no sympathy for the existence of these formats.  Indeed,
-one of the original reasons for developing this free software was to help
-force convergence on common, open format standards for JPEG files.  Don't
-use a proprietary file format!
-
-
-TO DO
-=====
-
-The major thrust for v7 will probably be improvement of visual quality.
-The current method for scaling the quantization tables is known not to be
-very good at low Q values.  We also intend to investigate block boundary
-smoothing, "poor man's variable quantization", and other means of improving
-quality-vs-file-size performance without sacrificing compatibility.
-
-In future versions, we are considering supporting some of the upcoming JPEG
-Part 3 extensions --- principally, variable quantization and the SPIFF file
-format.
-
-As always, speeding things up is of great interest.
-
-Please send bug reports, offers of help, etc. to jpeg-info@uunet.uu.net.
diff --git a/src/jpeg-6b/ioq3-changes.diff b/src/jpeg-6b/ioq3-changes.diff
deleted file mode 100644
index 1ce424bc..00000000
--- a/src/jpeg-6b/ioq3-changes.diff
+++ /dev/null
@@ -1,916 +0,0 @@
-diff -u -w /home/tma/sources/jpeg-6b/jcdctmgr.c ./jcdctmgr.c
---- /home/tma/sources/jpeg-6b/jcdctmgr.c	1996-01-13 19:15:12.000000000 +0000
-+++ ./jcdctmgr.c	2008-08-22 00:07:09.000000000 +0100
-@@ -57,7 +57,6 @@
-   int ci, qtblno, i;
-   jpeg_component_info *compptr;
-   JQUANT_TBL * qtbl;
--  DCTELEM * dtbl;
- 
-   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
-        ci++, compptr++) {
-@@ -168,6 +167,8 @@
- }
- 
- 
-+/* code/jpeg-6b/jcdctmgr.c:184: warning: ‘forward_DCT’ defined but not used */
-+#if 0
- /*
-  * Perform forward DCT on one or more blocks of a component.
-  *
-@@ -262,6 +263,7 @@
-     }
-   }
- }
-+#endif
- 
- 
- #ifdef DCT_FLOAT_SUPPORTED
-diff -u -w /home/tma/sources/jpeg-6b/jcmainct.c ./jcmainct.c
---- /home/tma/sources/jpeg-6b/jcmainct.c	1996-01-06 23:24:59.000000000 +0000
-+++ ./jcmainct.c	2008-08-22 00:10:21.000000000 +0100
-@@ -68,32 +68,32 @@
- METHODDEF(void)
- start_pass_main (j_compress_ptr cinfo, J_BUF_MODE pass_mode)
- {
--  my_main_ptr main = (my_main_ptr) cinfo->main;
-+  my_main_ptr jmain = (my_main_ptr) cinfo->main;
- 
-   /* Do nothing in raw-data mode. */
-   if (cinfo->raw_data_in)
-     return;
- 
--  main->cur_iMCU_row = 0;	/* initialize counters */
--  main->rowgroup_ctr = 0;
--  main->suspended = FALSE;
--  main->pass_mode = pass_mode;	/* save mode for use by process_data */
-+  jmain->cur_iMCU_row = 0;	/* initialize counters */
-+  jmain->rowgroup_ctr = 0;
-+  jmain->suspended = FALSE;
-+  jmain->pass_mode = pass_mode;	/* save mode for use by process_data */
- 
-   switch (pass_mode) {
-   case JBUF_PASS_THRU:
- #ifdef FULL_MAIN_BUFFER_SUPPORTED
--    if (main->whole_image[0] != NULL)
-+    if (jmain->whole_image[0] != NULL)
-       ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
- #endif
--    main->pub.process_data = process_data_simple_main;
-+    jmain->pub.process_data = process_data_simple_main;
-     break;
- #ifdef FULL_MAIN_BUFFER_SUPPORTED
-   case JBUF_SAVE_SOURCE:
-   case JBUF_CRANK_DEST:
-   case JBUF_SAVE_AND_PASS:
--    if (main->whole_image[0] == NULL)
-+    if (jmain->whole_image[0] == NULL)
-       ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
--    main->pub.process_data = process_data_buffer_main;
-+    jmain->pub.process_data = process_data_buffer_main;
-     break;
- #endif
-   default:
-@@ -114,46 +114,46 @@
- 			  JSAMPARRAY input_buf, JDIMENSION *in_row_ctr,
- 			  JDIMENSION in_rows_avail)
- {
--  my_main_ptr main = (my_main_ptr) cinfo->main;
-+  my_main_ptr jmain = (my_main_ptr) cinfo->main;
- 
--  while (main->cur_iMCU_row < cinfo->total_iMCU_rows) {
--    /* Read input data if we haven't filled the main buffer yet */
--    if (main->rowgroup_ctr < DCTSIZE)
-+  while (jmain->cur_iMCU_row < cinfo->total_iMCU_rows) {
-+    /* Read input data if we haven't filled the jmain buffer yet */
-+    if (jmain->rowgroup_ctr < DCTSIZE)
-       (*cinfo->prep->pre_process_data) (cinfo,
- 					input_buf, in_row_ctr, in_rows_avail,
--					main->buffer, &main->rowgroup_ctr,
-+					jmain->buffer, &jmain->rowgroup_ctr,
- 					(JDIMENSION) DCTSIZE);
- 
-     /* If we don't have a full iMCU row buffered, return to application for
-      * more data.  Note that preprocessor will always pad to fill the iMCU row
-      * at the bottom of the image.
-      */
--    if (main->rowgroup_ctr != DCTSIZE)
-+    if (jmain->rowgroup_ctr != DCTSIZE)
-       return;
- 
-     /* Send the completed row to the compressor */
--    if (! (*cinfo->coef->compress_data) (cinfo, main->buffer)) {
-+    if (! (*cinfo->coef->compress_data) (cinfo, jmain->buffer)) {
-       /* If compressor did not consume the whole row, then we must need to
-        * suspend processing and return to the application.  In this situation
-        * we pretend we didn't yet consume the last input row; otherwise, if
-        * it happened to be the last row of the image, the application would
-        * think we were done.
-        */
--      if (! main->suspended) {
-+      if (! jmain->suspended) {
- 	(*in_row_ctr)--;
--	main->suspended = TRUE;
-+	jmain->suspended = TRUE;
-       }
-       return;
-     }
-     /* We did finish the row.  Undo our little suspension hack if a previous
--     * call suspended; then mark the main buffer empty.
-+     * call suspended; then mark the jmain buffer empty.
-      */
--    if (main->suspended) {
-+    if (jmain->suspended) {
-       (*in_row_ctr)++;
--      main->suspended = FALSE;
-+      jmain->suspended = FALSE;
-     }
--    main->rowgroup_ctr = 0;
--    main->cur_iMCU_row++;
-+    jmain->rowgroup_ctr = 0;
-+    jmain->cur_iMCU_row++;
-   }
- }
- 
-@@ -170,25 +170,25 @@
- 			  JSAMPARRAY input_buf, JDIMENSION *in_row_ctr,
- 			  JDIMENSION in_rows_avail)
- {
--  my_main_ptr main = (my_main_ptr) cinfo->main;
-+  my_main_ptr jmain = (my_main_ptr) cinfo->main;
-   int ci;
-   jpeg_component_info *compptr;
--  boolean writing = (main->pass_mode != JBUF_CRANK_DEST);
-+  boolean writing = (jmain->pass_mode != JBUF_CRANK_DEST);
- 
--  while (main->cur_iMCU_row < cinfo->total_iMCU_rows) {
-+  while (jmain->cur_iMCU_row < cinfo->total_iMCU_rows) {
-     /* Realign the virtual buffers if at the start of an iMCU row. */
--    if (main->rowgroup_ctr == 0) {
-+    if (jmain->rowgroup_ctr == 0) {
-       for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
- 	   ci++, compptr++) {
--	main->buffer[ci] = (*cinfo->mem->access_virt_sarray)
--	  ((j_common_ptr) cinfo, main->whole_image[ci],
--	   main->cur_iMCU_row * (compptr->v_samp_factor * DCTSIZE),
-+	jmain->buffer[ci] = (*cinfo->mem->access_virt_sarray)
-+	  ((j_common_ptr) cinfo, jmain->whole_image[ci],
-+	   jmain->cur_iMCU_row * (compptr->v_samp_factor * DCTSIZE),
- 	   (JDIMENSION) (compptr->v_samp_factor * DCTSIZE), writing);
-       }
-       /* In a read pass, pretend we just read some source data. */
-       if (! writing) {
- 	*in_row_ctr += cinfo->max_v_samp_factor * DCTSIZE;
--	main->rowgroup_ctr = DCTSIZE;
-+	jmain->rowgroup_ctr = DCTSIZE;
-       }
-     }
- 
-@@ -197,40 +197,40 @@
-     if (writing) {
-       (*cinfo->prep->pre_process_data) (cinfo,
- 					input_buf, in_row_ctr, in_rows_avail,
--					main->buffer, &main->rowgroup_ctr,
-+					jmain->buffer, &jmain->rowgroup_ctr,
- 					(JDIMENSION) DCTSIZE);
-       /* Return to application if we need more data to fill the iMCU row. */
--      if (main->rowgroup_ctr < DCTSIZE)
-+      if (jmain->rowgroup_ctr < DCTSIZE)
- 	return;
-     }
- 
-     /* Emit data, unless this is a sink-only pass. */
--    if (main->pass_mode != JBUF_SAVE_SOURCE) {
--      if (! (*cinfo->coef->compress_data) (cinfo, main->buffer)) {
-+    if (jmain->pass_mode != JBUF_SAVE_SOURCE) {
-+      if (! (*cinfo->coef->compress_data) (cinfo, jmain->buffer)) {
- 	/* If compressor did not consume the whole row, then we must need to
- 	 * suspend processing and return to the application.  In this situation
- 	 * we pretend we didn't yet consume the last input row; otherwise, if
- 	 * it happened to be the last row of the image, the application would
- 	 * think we were done.
- 	 */
--	if (! main->suspended) {
-+	if (! jmain->suspended) {
- 	  (*in_row_ctr)--;
--	  main->suspended = TRUE;
-+	  jmain->suspended = TRUE;
- 	}
- 	return;
-       }
-       /* We did finish the row.  Undo our little suspension hack if a previous
--       * call suspended; then mark the main buffer empty.
-+       * call suspended; then mark the jmain buffer empty.
-        */
--      if (main->suspended) {
-+      if (jmain->suspended) {
- 	(*in_row_ctr)++;
--	main->suspended = FALSE;
-+	jmain->suspended = FALSE;
-       }
-     }
- 
-     /* If get here, we are done with this iMCU row.  Mark buffer empty. */
--    main->rowgroup_ctr = 0;
--    main->cur_iMCU_row++;
-+    jmain->rowgroup_ctr = 0;
-+    jmain->cur_iMCU_row++;
-   }
- }
- 
-@@ -238,21 +238,21 @@
- 
- 
- /*
-- * Initialize main buffer controller.
-+ * Initialize jmain buffer controller.
-  */
- 
- GLOBAL(void)
- jinit_c_main_controller (j_compress_ptr cinfo, boolean need_full_buffer)
- {
--  my_main_ptr main;
-+  my_main_ptr jmain;
-   int ci;
-   jpeg_component_info *compptr;
- 
--  main = (my_main_ptr)
-+  jmain = (my_main_ptr)
-     (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
- 				SIZEOF(my_main_controller));
--  cinfo->main = (struct jpeg_c_main_controller *) main;
--  main->pub.start_pass = start_pass_main;
-+  cinfo->main = (struct jpeg_c_main_controller *) jmain;
-+  jmain->pub.start_pass = start_pass_main;
- 
-   /* We don't need to create a buffer in raw-data mode. */
-   if (cinfo->raw_data_in)
-@@ -267,7 +267,7 @@
-     /* Note we pad the bottom to a multiple of the iMCU height */
-     for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
- 	 ci++, compptr++) {
--      main->whole_image[ci] = (*cinfo->mem->request_virt_sarray)
-+      jmain->whole_image[ci] = (*cinfo->mem->request_virt_sarray)
- 	((j_common_ptr) cinfo, JPOOL_IMAGE, FALSE,
- 	 compptr->width_in_blocks * DCTSIZE,
- 	 (JDIMENSION) jround_up((long) compptr->height_in_blocks,
-@@ -279,12 +279,12 @@
- #endif
-   } else {
- #ifdef FULL_MAIN_BUFFER_SUPPORTED
--    main->whole_image[0] = NULL; /* flag for no virtual arrays */
-+    jmain->whole_image[0] = NULL; /* flag for no virtual arrays */
- #endif
-     /* Allocate a strip buffer for each component */
-     for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
- 	 ci++, compptr++) {
--      main->buffer[ci] = (*cinfo->mem->alloc_sarray)
-+      jmain->buffer[ci] = (*cinfo->mem->alloc_sarray)
- 	((j_common_ptr) cinfo, JPOOL_IMAGE,
- 	 compptr->width_in_blocks * DCTSIZE,
- 	 (JDIMENSION) (compptr->v_samp_factor * DCTSIZE));
-diff -u -w /home/tma/sources/jpeg-6b/jdatasrc.c ./jdatasrc.c
---- /home/tma/sources/jpeg-6b/jdatasrc.c	1996-01-06 23:26:42.000000000 +0000
-+++ ./jdatasrc.c	2008-08-22 00:00:59.000000000 +0100
-@@ -19,13 +19,17 @@
- #include "jpeglib.h"
- #include "jerror.h"
- 
-+#ifndef MIN
-+#define MIN(a, b) ((a)<(b)?(a):(b))
-+#endif
- 
- /* Expanded data source object for stdio input */
- 
- typedef struct {
-   struct jpeg_source_mgr pub;	/* public fields */
- 
--  FILE * infile;		/* source stream */
-+  unsigned char *inbuf;		/* source stream */
-+  size_t inbufbytes;
-   JOCTET * buffer;		/* start of buffer */
-   boolean start_of_file;	/* have we gotten any data yet? */
- } my_source_mgr;
-@@ -90,18 +94,19 @@
- fill_input_buffer (j_decompress_ptr cinfo)
- {
-   my_src_ptr src = (my_src_ptr) cinfo->src;
--  size_t nbytes;
--
--  nbytes = JFREAD(src->infile, src->buffer, INPUT_BUF_SIZE);
-+  size_t nbytes = MIN(src->inbufbytes, INPUT_BUF_SIZE);
- 
-   if (nbytes <= 0) {
--    if (src->start_of_file)	/* Treat empty input file as fatal error */
--      ERREXIT(cinfo, JERR_INPUT_EMPTY);
-     WARNMS(cinfo, JWRN_JPEG_EOF);
-     /* Insert a fake EOI marker */
-     src->buffer[0] = (JOCTET) 0xFF;
-     src->buffer[1] = (JOCTET) JPEG_EOI;
-     nbytes = 2;
-+  } else {
-+    memcpy( src->buffer, src->inbuf, nbytes);
-+
-+    src->inbuf += nbytes;
-+    src->inbufbytes -= nbytes;
-   }
- 
-   src->pub.next_input_byte = src->buffer;
-@@ -179,7 +184,7 @@
-  */
- 
- GLOBAL(void)
--jpeg_stdio_src (j_decompress_ptr cinfo, FILE * infile)
-+jpeg_mem_src (j_decompress_ptr cinfo, unsigned char *inbuf, size_t size)
- {
-   my_src_ptr src;
- 
-@@ -206,7 +211,8 @@
-   src->pub.skip_input_data = skip_input_data;
-   src->pub.resync_to_restart = jpeg_resync_to_restart; /* use default method */
-   src->pub.term_source = term_source;
--  src->infile = infile;
-+  src->inbuf = inbuf;
-+  src->inbufbytes = size;
-   src->pub.bytes_in_buffer = 0; /* forces fill_input_buffer on first read */
-   src->pub.next_input_byte = NULL; /* until buffer loaded */
- }
-diff -u -w /home/tma/sources/jpeg-6b/jdmainct.c ./jdmainct.c
---- /home/tma/sources/jpeg-6b/jdmainct.c	1996-01-06 23:27:17.000000000 +0000
-+++ ./jdmainct.c	2008-08-22 00:13:48.000000000 +0100
-@@ -159,7 +159,7 @@
-  * This is done only once, not once per pass.
-  */
- {
--  my_main_ptr main = (my_main_ptr) cinfo->main;
-+  my_main_ptr jmain = (my_main_ptr) cinfo->main;
-   int ci, rgroup;
-   int M = cinfo->min_DCT_scaled_size;
-   jpeg_component_info *compptr;
-@@ -168,10 +168,10 @@
-   /* Get top-level space for component array pointers.
-    * We alloc both arrays with one call to save a few cycles.
-    */
--  main->xbuffer[0] = (JSAMPIMAGE)
-+  jmain->xbuffer[0] = (JSAMPIMAGE)
-     (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
- 				cinfo->num_components * 2 * SIZEOF(JSAMPARRAY));
--  main->xbuffer[1] = main->xbuffer[0] + cinfo->num_components;
-+  jmain->xbuffer[1] = jmain->xbuffer[0] + cinfo->num_components;
- 
-   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
-        ci++, compptr++) {
-@@ -184,9 +184,9 @@
-       (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
- 				  2 * (rgroup * (M + 4)) * SIZEOF(JSAMPROW));
-     xbuf += rgroup;		/* want one row group at negative offsets */
--    main->xbuffer[0][ci] = xbuf;
-+    jmain->xbuffer[0][ci] = xbuf;
-     xbuf += rgroup * (M + 4);
--    main->xbuffer[1][ci] = xbuf;
-+    jmain->xbuffer[1][ci] = xbuf;
-   }
- }
- 
-@@ -194,13 +194,13 @@
- LOCAL(void)
- make_funny_pointers (j_decompress_ptr cinfo)
- /* Create the funny pointer lists discussed in the comments above.
-- * The actual workspace is already allocated (in main->buffer),
-+ * The actual workspace is already allocated (in jmain->buffer),
-  * and the space for the pointer lists is allocated too.
-  * This routine just fills in the curiously ordered lists.
-  * This will be repeated at the beginning of each pass.
-  */
- {
--  my_main_ptr main = (my_main_ptr) cinfo->main;
-+  my_main_ptr jmain = (my_main_ptr) cinfo->main;
-   int ci, i, rgroup;
-   int M = cinfo->min_DCT_scaled_size;
-   jpeg_component_info *compptr;
-@@ -210,10 +210,10 @@
-        ci++, compptr++) {
-     rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
-       cinfo->min_DCT_scaled_size; /* height of a row group of component */
--    xbuf0 = main->xbuffer[0][ci];
--    xbuf1 = main->xbuffer[1][ci];
-+    xbuf0 = jmain->xbuffer[0][ci];
-+    xbuf1 = jmain->xbuffer[1][ci];
-     /* First copy the workspace pointers as-is */
--    buf = main->buffer[ci];
-+    buf = jmain->buffer[ci];
-     for (i = 0; i < rgroup * (M + 2); i++) {
-       xbuf0[i] = xbuf1[i] = buf[i];
-     }
-@@ -240,7 +240,7 @@
-  * This changes the pointer list state from top-of-image to the normal state.
-  */
- {
--  my_main_ptr main = (my_main_ptr) cinfo->main;
-+  my_main_ptr jmain = (my_main_ptr) cinfo->main;
-   int ci, i, rgroup;
-   int M = cinfo->min_DCT_scaled_size;
-   jpeg_component_info *compptr;
-@@ -250,8 +250,8 @@
-        ci++, compptr++) {
-     rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
-       cinfo->min_DCT_scaled_size; /* height of a row group of component */
--    xbuf0 = main->xbuffer[0][ci];
--    xbuf1 = main->xbuffer[1][ci];
-+    xbuf0 = jmain->xbuffer[0][ci];
-+    xbuf1 = jmain->xbuffer[1][ci];
-     for (i = 0; i < rgroup; i++) {
-       xbuf0[i - rgroup] = xbuf0[rgroup*(M+1) + i];
-       xbuf1[i - rgroup] = xbuf1[rgroup*(M+1) + i];
-@@ -269,7 +269,7 @@
-  * Also sets rowgroups_avail to indicate number of nondummy row groups in row.
-  */
- {
--  my_main_ptr main = (my_main_ptr) cinfo->main;
-+  my_main_ptr jmain = (my_main_ptr) cinfo->main;
-   int ci, i, rgroup, iMCUheight, rows_left;
-   jpeg_component_info *compptr;
-   JSAMPARRAY xbuf;
-@@ -286,12 +286,12 @@
-      * so we need only do it once.
-      */
-     if (ci == 0) {
--      main->rowgroups_avail = (JDIMENSION) ((rows_left-1) / rgroup + 1);
-+      jmain->rowgroups_avail = (JDIMENSION) ((rows_left-1) / rgroup + 1);
-     }
-     /* Duplicate the last real sample row rgroup*2 times; this pads out the
-      * last partial rowgroup and ensures at least one full rowgroup of context.
-      */
--    xbuf = main->xbuffer[main->whichptr][ci];
-+    xbuf = jmain->xbuffer[jmain->whichptr][ci];
-     for (i = 0; i < rgroup * 2; i++) {
-       xbuf[rows_left + i] = xbuf[rows_left-1];
-     }
-@@ -306,27 +306,27 @@
- METHODDEF(void)
- start_pass_main (j_decompress_ptr cinfo, J_BUF_MODE pass_mode)
- {
--  my_main_ptr main = (my_main_ptr) cinfo->main;
-+  my_main_ptr jmain = (my_main_ptr) cinfo->main;
- 
-   switch (pass_mode) {
-   case JBUF_PASS_THRU:
-     if (cinfo->upsample->need_context_rows) {
--      main->pub.process_data = process_data_context_main;
-+      jmain->pub.process_data = process_data_context_main;
-       make_funny_pointers(cinfo); /* Create the xbuffer[] lists */
--      main->whichptr = 0;	/* Read first iMCU row into xbuffer[0] */
--      main->context_state = CTX_PREPARE_FOR_IMCU;
--      main->iMCU_row_ctr = 0;
-+      jmain->whichptr = 0;	/* Read first iMCU row into xbuffer[0] */
-+      jmain->context_state = CTX_PREPARE_FOR_IMCU;
-+      jmain->iMCU_row_ctr = 0;
-     } else {
-       /* Simple case with no context needed */
--      main->pub.process_data = process_data_simple_main;
-+      jmain->pub.process_data = process_data_simple_main;
-     }
--    main->buffer_full = FALSE;	/* Mark buffer empty */
--    main->rowgroup_ctr = 0;
-+    jmain->buffer_full = FALSE;	/* Mark buffer empty */
-+    jmain->rowgroup_ctr = 0;
-     break;
- #ifdef QUANT_2PASS_SUPPORTED
-   case JBUF_CRANK_DEST:
-     /* For last pass of 2-pass quantization, just crank the postprocessor */
--    main->pub.process_data = process_data_crank_post;
-+    jmain->pub.process_data = process_data_crank_post;
-     break;
- #endif
-   default:
-@@ -346,14 +346,14 @@
- 			  JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
- 			  JDIMENSION out_rows_avail)
- {
--  my_main_ptr main = (my_main_ptr) cinfo->main;
-+  my_main_ptr jmain = (my_main_ptr) cinfo->main;
-   JDIMENSION rowgroups_avail;
- 
--  /* Read input data if we haven't filled the main buffer yet */
--  if (! main->buffer_full) {
--    if (! (*cinfo->coef->decompress_data) (cinfo, main->buffer))
-+  /* Read input data if we haven't filled the jmain buffer yet */
-+  if (! jmain->buffer_full) {
-+    if (! (*cinfo->coef->decompress_data) (cinfo, jmain->buffer))
-       return;			/* suspension forced, can do nothing more */
--    main->buffer_full = TRUE;	/* OK, we have an iMCU row to work with */
-+    jmain->buffer_full = TRUE;	/* OK, we have an iMCU row to work with */
-   }
- 
-   /* There are always min_DCT_scaled_size row groups in an iMCU row. */
-@@ -364,14 +364,14 @@
-    */
- 
-   /* Feed the postprocessor */
--  (*cinfo->post->post_process_data) (cinfo, main->buffer,
--				     &main->rowgroup_ctr, rowgroups_avail,
-+  (*cinfo->post->post_process_data) (cinfo, jmain->buffer,
-+				     &jmain->rowgroup_ctr, rowgroups_avail,
- 				     output_buf, out_row_ctr, out_rows_avail);
- 
-   /* Has postprocessor consumed all the data yet? If so, mark buffer empty */
--  if (main->rowgroup_ctr >= rowgroups_avail) {
--    main->buffer_full = FALSE;
--    main->rowgroup_ctr = 0;
-+  if (jmain->rowgroup_ctr >= rowgroups_avail) {
-+    jmain->buffer_full = FALSE;
-+    jmain->rowgroup_ctr = 0;
-   }
- }
- 
-@@ -386,15 +386,15 @@
- 			   JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
- 			   JDIMENSION out_rows_avail)
- {
--  my_main_ptr main = (my_main_ptr) cinfo->main;
-+  my_main_ptr jmain = (my_main_ptr) cinfo->main;
- 
--  /* Read input data if we haven't filled the main buffer yet */
--  if (! main->buffer_full) {
-+  /* Read input data if we haven't filled the jmain buffer yet */
-+  if (! jmain->buffer_full) {
-     if (! (*cinfo->coef->decompress_data) (cinfo,
--					   main->xbuffer[main->whichptr]))
-+					   jmain->xbuffer[jmain->whichptr]))
-       return;			/* suspension forced, can do nothing more */
--    main->buffer_full = TRUE;	/* OK, we have an iMCU row to work with */
--    main->iMCU_row_ctr++;	/* count rows received */
-+    jmain->buffer_full = TRUE;	/* OK, we have an iMCU row to work with */
-+    jmain->iMCU_row_ctr++;	/* count rows received */
-   }
- 
-   /* Postprocessor typically will not swallow all the input data it is handed
-@@ -402,47 +402,47 @@
-    * to exit and restart.  This switch lets us keep track of how far we got.
-    * Note that each case falls through to the next on successful completion.
-    */
--  switch (main->context_state) {
-+  switch (jmain->context_state) {
-   case CTX_POSTPONED_ROW:
-     /* Call postprocessor using previously set pointers for postponed row */
--    (*cinfo->post->post_process_data) (cinfo, main->xbuffer[main->whichptr],
--			&main->rowgroup_ctr, main->rowgroups_avail,
-+    (*cinfo->post->post_process_data) (cinfo, jmain->xbuffer[jmain->whichptr],
-+			&jmain->rowgroup_ctr, jmain->rowgroups_avail,
- 			output_buf, out_row_ctr, out_rows_avail);
--    if (main->rowgroup_ctr < main->rowgroups_avail)
-+    if (jmain->rowgroup_ctr < jmain->rowgroups_avail)
-       return;			/* Need to suspend */
--    main->context_state = CTX_PREPARE_FOR_IMCU;
-+    jmain->context_state = CTX_PREPARE_FOR_IMCU;
-     if (*out_row_ctr >= out_rows_avail)
-       return;			/* Postprocessor exactly filled output buf */
-     /*FALLTHROUGH*/
-   case CTX_PREPARE_FOR_IMCU:
-     /* Prepare to process first M-1 row groups of this iMCU row */
--    main->rowgroup_ctr = 0;
--    main->rowgroups_avail = (JDIMENSION) (cinfo->min_DCT_scaled_size - 1);
-+    jmain->rowgroup_ctr = 0;
-+    jmain->rowgroups_avail = (JDIMENSION) (cinfo->min_DCT_scaled_size - 1);
-     /* Check for bottom of image: if so, tweak pointers to "duplicate"
-      * the last sample row, and adjust rowgroups_avail to ignore padding rows.
-      */
--    if (main->iMCU_row_ctr == cinfo->total_iMCU_rows)
-+    if (jmain->iMCU_row_ctr == cinfo->total_iMCU_rows)
-       set_bottom_pointers(cinfo);
--    main->context_state = CTX_PROCESS_IMCU;
-+    jmain->context_state = CTX_PROCESS_IMCU;
-     /*FALLTHROUGH*/
-   case CTX_PROCESS_IMCU:
-     /* Call postprocessor using previously set pointers */
--    (*cinfo->post->post_process_data) (cinfo, main->xbuffer[main->whichptr],
--			&main->rowgroup_ctr, main->rowgroups_avail,
-+    (*cinfo->post->post_process_data) (cinfo, jmain->xbuffer[jmain->whichptr],
-+			&jmain->rowgroup_ctr, jmain->rowgroups_avail,
- 			output_buf, out_row_ctr, out_rows_avail);
--    if (main->rowgroup_ctr < main->rowgroups_avail)
-+    if (jmain->rowgroup_ctr < jmain->rowgroups_avail)
-       return;			/* Need to suspend */
-     /* After the first iMCU, change wraparound pointers to normal state */
--    if (main->iMCU_row_ctr == 1)
-+    if (jmain->iMCU_row_ctr == 1)
-       set_wraparound_pointers(cinfo);
-     /* Prepare to load new iMCU row using other xbuffer list */
--    main->whichptr ^= 1;	/* 0=>1 or 1=>0 */
--    main->buffer_full = FALSE;
-+    jmain->whichptr ^= 1;	/* 0=>1 or 1=>0 */
-+    jmain->buffer_full = FALSE;
-     /* Still need to process last row group of this iMCU row, */
-     /* which is saved at index M+1 of the other xbuffer */
--    main->rowgroup_ctr = (JDIMENSION) (cinfo->min_DCT_scaled_size + 1);
--    main->rowgroups_avail = (JDIMENSION) (cinfo->min_DCT_scaled_size + 2);
--    main->context_state = CTX_POSTPONED_ROW;
-+    jmain->rowgroup_ctr = (JDIMENSION) (cinfo->min_DCT_scaled_size + 1);
-+    jmain->rowgroups_avail = (JDIMENSION) (cinfo->min_DCT_scaled_size + 2);
-+    jmain->context_state = CTX_POSTPONED_ROW;
-   }
- }
- 
-@@ -469,21 +469,21 @@
- 
- 
- /*
-- * Initialize main buffer controller.
-+ * Initialize jmain buffer controller.
-  */
- 
- GLOBAL(void)
- jinit_d_main_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
- {
--  my_main_ptr main;
-+  my_main_ptr jmain;
-   int ci, rgroup, ngroups;
-   jpeg_component_info *compptr;
- 
--  main = (my_main_ptr)
-+  jmain = (my_main_ptr)
-     (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
- 				SIZEOF(my_main_controller));
--  cinfo->main = (struct jpeg_d_main_controller *) main;
--  main->pub.start_pass = start_pass_main;
-+  cinfo->main = (struct jpeg_d_main_controller *) jmain;
-+  jmain->pub.start_pass = start_pass_main;
- 
-   if (need_full_buffer)		/* shouldn't happen */
-     ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
-@@ -504,7 +504,7 @@
-        ci++, compptr++) {
-     rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
-       cinfo->min_DCT_scaled_size; /* height of a row group of component */
--    main->buffer[ci] = (*cinfo->mem->alloc_sarray)
-+    jmain->buffer[ci] = (*cinfo->mem->alloc_sarray)
- 			((j_common_ptr) cinfo, JPOOL_IMAGE,
- 			 compptr->width_in_blocks * compptr->DCT_scaled_size,
- 			 (JDIMENSION) (rgroup * ngroups));
-diff -u -w /home/tma/sources/jpeg-6b/jerror.c ./jerror.c
---- /home/tma/sources/jpeg-6b/jerror.c	1998-02-22 01:03:15.000000000 +0000
-+++ ./jerror.c	2008-08-21 23:58:36.000000000 +0100
-@@ -18,6 +18,8 @@
-  * These routines are used by both the compression and decompression code.
-  */
- 
-+#include "../renderer/tr_local.h"
-+
- /* this is not a core library module, so it doesn't define JPEG_INTERNALS */
- #include "jinclude.h"
- #include "jpeglib.h"
-@@ -69,13 +71,15 @@
- METHODDEF(void)
- error_exit (j_common_ptr cinfo)
- {
--  /* Always display the message */
--  (*cinfo->err->output_message) (cinfo);
-+  char buffer[JMSG_LENGTH_MAX];
-+
-+  /* Create the message */
-+  (*cinfo->err->format_message) (cinfo, buffer);
- 
-   /* Let the memory manager delete any temp files before we die */
-   jpeg_destroy(cinfo);
- 
--  exit(EXIT_FAILURE);
-+  ri.Error( ERR_FATAL, "%s\n", buffer );
- }
- 
- 
-@@ -108,7 +112,7 @@
- 	     MB_OK | MB_ICONERROR);
- #else
-   /* Send it to stderr, adding a newline */
--  fprintf(stderr, "%s\n", buffer);
-+  ri.Printf(PRINT_ALL, "%s\n", buffer);
- #endif
- }
- 
-diff -u -w /home/tma/sources/jpeg-6b/jinclude.h ./jinclude.h
---- /home/tma/sources/jpeg-6b/jinclude.h	1994-04-01 21:29:31.000000000 +0100
-+++ ./jinclude.h	2008-08-21 23:58:36.000000000 +0100
-@@ -15,9 +15,34 @@
-  */
- 
- 
-+#ifdef _MSC_VER
-+
-+#pragma warning(disable : 4018)     // signed/unsigned mismatch
-+#pragma warning(disable : 4032)
-+#pragma warning(disable : 4051)
-+#pragma warning(disable : 4057)		// slightly different base types
-+#pragma warning(disable : 4100)		// unreferenced formal parameter
-+#pragma warning(disable : 4115)
-+#pragma warning(disable : 4125)		// decimal digit terminates octal escape sequence
-+#pragma warning(disable : 4127)		// conditional expression is constant
-+#pragma warning(disable : 4136)
-+#pragma warning(disable : 4152)		// nonstandard extension, function/data pointer conversion in expression
-+#pragma warning(disable : 4201)
-+#pragma warning(disable : 4214)
-+#pragma warning(disable : 4244)
-+#pragma warning(disable : 4305)		// truncation from const double to float
-+#pragma warning(disable : 4310)		// cast truncates constant value
-+#pragma warning(disable:  4505) 	// unreferenced local function has been removed
-+#pragma warning(disable : 4514)
-+#pragma warning(disable : 4702)		// unreachable code
-+#pragma warning(disable : 4711)		// selected for automatic inline expansion
-+#pragma warning(disable : 4220)		// varargs matches remaining parameters
-+#pragma warning(disable : 4761)		// integral size mismatch 
-+#endif
-+
- /* Include auto-config file to find out which system include files we need. */
- 
--#include "jconfig.h"		/* auto configuration options */
-+#include "../jpeg-6b/jconfig.h"		/* auto configuration options */
- #define JCONFIG_INCLUDED	/* so that jpeglib.h doesn't do it again */
- 
- /*
-diff -u -w /home/tma/sources/jpeg-6b/jmemnobs.c ./jmemnobs.c
---- /home/tma/sources/jpeg-6b/jmemnobs.c	1996-01-06 23:31:18.000000000 +0000
-+++ ./jmemnobs.c	2008-08-21 23:58:36.000000000 +0100
-@@ -8,39 +8,35 @@
-  * This file provides a really simple implementation of the system-
-  * dependent portion of the JPEG memory manager.  This implementation
-  * assumes that no backing-store files are needed: all required space
-- * can be obtained from malloc().
-+ * can be obtained from ri.Malloc().
-  * This is very portable in the sense that it'll compile on almost anything,
-  * but you'd better have lots of main memory (or virtual memory) if you want
-  * to process big images.
-  * Note that the max_memory_to_use option is ignored by this implementation.
-  */
- 
-+#include "../renderer/tr_local.h"
-+
- #define JPEG_INTERNALS
- #include "jinclude.h"
- #include "jpeglib.h"
- #include "jmemsys.h"		/* import the system-dependent declarations */
- 
--#ifndef HAVE_STDLIB_H		/* <stdlib.h> should declare malloc(),free() */
--extern void * malloc JPP((size_t size));
--extern void free JPP((void *ptr));
--#endif
--
--
- /*
-  * Memory allocation and freeing are controlled by the regular library
-- * routines malloc() and free().
-+ * routines ri.Malloc() and ri.Free().
-  */
- 
- GLOBAL(void *)
- jpeg_get_small (j_common_ptr cinfo, size_t sizeofobject)
- {
--  return (void *) malloc(sizeofobject);
-+  return (void *) ri.Malloc(sizeofobject);
- }
- 
- GLOBAL(void)
- jpeg_free_small (j_common_ptr cinfo, void * object, size_t sizeofobject)
- {
--  free(object);
-+  ri.Free(object);
- }
- 
- 
-@@ -54,13 +50,13 @@
- GLOBAL(void FAR *)
- jpeg_get_large (j_common_ptr cinfo, size_t sizeofobject)
- {
--  return (void FAR *) malloc(sizeofobject);
-+  return (void FAR *) ri.Malloc(sizeofobject);
- }
- 
- GLOBAL(void)
- jpeg_free_large (j_common_ptr cinfo, void FAR * object, size_t sizeofobject)
- {
--  free(object);
-+  ri.Free(object);
- }
- 
- 
-diff -u -w /home/tma/sources/jpeg-6b/jmorecfg.h ./jmorecfg.h
---- /home/tma/sources/jpeg-6b/jmorecfg.h	1997-08-10 00:58:56.000000000 +0100
-+++ ./jmorecfg.h	2008-08-21 23:58:36.000000000 +0100
-@@ -157,7 +157,8 @@
- 
- /* INT32 must hold at least signed 32-bit values. */
- 
--#ifndef XMD_H			/* X11/xmd.h correctly defines INT32 */
-+/* MinGW basetsd.h defines INT32 - don't redefine it */
-+#if !(defined __MINGW32__ && defined _BASETSD_H)
- typedef long INT32;
- #endif
- 
-@@ -210,8 +211,10 @@
-  */
- 
- #ifdef NEED_FAR_POINTERS
-+#undef FAR
- #define FAR  far
- #else
-+#undef FAR
- #define FAR
- #endif
- 
-@@ -223,9 +226,7 @@
-  * Defining HAVE_BOOLEAN before including jpeglib.h should make it work.
-  */
- 
--#ifndef HAVE_BOOLEAN
--typedef int boolean;
--#endif
-+typedef unsigned char boolean;
- #ifndef FALSE			/* in case these macros already exist */
- #define FALSE	0		/* values of boolean */
- #endif
-@@ -260,8 +261,8 @@
- 
- /* Capability options common to encoder and decoder: */
- 
--#define DCT_ISLOW_SUPPORTED	/* slow but accurate integer algorithm */
--#define DCT_IFAST_SUPPORTED	/* faster, less accurate integer method */
-+#undef DCT_ISLOW_SUPPORTED	/* slow but accurate integer algorithm */
-+#undef DCT_IFAST_SUPPORTED	/* faster, less accurate integer method */
- #define DCT_FLOAT_SUPPORTED	/* floating-point: accurate, fast on fast HW */
- 
- /* Encoder capability options: */
-@@ -283,15 +284,15 @@
- /* Decoder capability options: */
- 
- #undef  D_ARITH_CODING_SUPPORTED    /* Arithmetic coding back end? */
--#define D_MULTISCAN_FILES_SUPPORTED /* Multiple-scan JPEG files? */
--#define D_PROGRESSIVE_SUPPORTED	    /* Progressive JPEG? (Requires MULTISCAN)*/
--#define SAVE_MARKERS_SUPPORTED	    /* jpeg_save_markers() needed? */
--#define BLOCK_SMOOTHING_SUPPORTED   /* Block smoothing? (Progressive only) */
--#define IDCT_SCALING_SUPPORTED	    /* Output rescaling via IDCT? */
-+#undef D_MULTISCAN_FILES_SUPPORTED /* Multiple-scan JPEG files? */
-+#undef D_PROGRESSIVE_SUPPORTED	    /* Progressive JPEG? (Requires MULTISCAN)*/
-+#undef SAVE_MARKERS_SUPPORTED	    /* jpeg_save_markers() needed? */
-+#undef BLOCK_SMOOTHING_SUPPORTED   /* Block smoothing? (Progressive only) */
-+#undef IDCT_SCALING_SUPPORTED	    /* Output rescaling via IDCT? */
- #undef  UPSAMPLE_SCALING_SUPPORTED  /* Output rescaling at upsample stage? */
--#define UPSAMPLE_MERGING_SUPPORTED  /* Fast path for sloppy upsampling? */
--#define QUANT_1PASS_SUPPORTED	    /* 1-pass color quantization? */
--#define QUANT_2PASS_SUPPORTED	    /* 2-pass color quantization? */
-+#undef UPSAMPLE_MERGING_SUPPORTED  /* Fast path for sloppy upsampling? */
-+#undef QUANT_1PASS_SUPPORTED	    /* 1-pass color quantization? */
-+#undef QUANT_2PASS_SUPPORTED	    /* 2-pass color quantization? */
- 
- /* more capability options later, no doubt */
- 
-@@ -314,7 +315,7 @@
- #define RGB_RED		0	/* Offset of Red in an RGB scanline element */
- #define RGB_GREEN	1	/* Offset of Green */
- #define RGB_BLUE	2	/* Offset of Blue */
--#define RGB_PIXELSIZE	3	/* JSAMPLEs per RGB scanline element */
-+#define RGB_PIXELSIZE	4	/* JSAMPLEs per RGB scanline element */
- 
- 
- /* Definitions for speed-related optimizations. */
-diff -u -w /home/tma/sources/jpeg-6b/jpeglib.h ./jpeglib.h
---- /home/tma/sources/jpeg-6b/jpeglib.h	1998-02-21 19:48:14.000000000 +0000
-+++ ./jpeglib.h	2008-08-22 00:01:58.000000000 +0100
-@@ -21,9 +21,9 @@
-  */
- 
- #ifndef JCONFIG_INCLUDED	/* in case jinclude.h already did */
--#include "jconfig.h"		/* widely used configuration options */
-+#include "../jpeg-6b/jconfig.h"		/* widely used configuration options */
- #endif
--#include "jmorecfg.h"		/* seldom changed options */
-+#include "../jpeg-6b/jmorecfg.h"		/* seldom changed options */
- 
- 
- /* Version ID for the JPEG library.
-@@ -835,7 +835,7 @@
- #define jpeg_destroy_compress	jDestCompress
- #define jpeg_destroy_decompress	jDestDecompress
- #define jpeg_stdio_dest		jStdDest
--#define jpeg_stdio_src		jStdSrc
-+#define jpeg_mem_src		jMemSrc
- #define jpeg_set_defaults	jSetDefaults
- #define jpeg_set_colorspace	jSetColorspace
- #define jpeg_default_colorspace	jDefColorspace
-@@ -908,7 +908,7 @@
- /* Standard data source and destination managers: stdio streams. */
- /* Caller is responsible for opening the file before and closing after. */
- EXTERN(void) jpeg_stdio_dest JPP((j_compress_ptr cinfo, FILE * outfile));
--EXTERN(void) jpeg_stdio_src JPP((j_decompress_ptr cinfo, FILE * infile));
-+EXTERN(void) jpeg_mem_src JPP((j_decompress_ptr cinfo, unsigned char *inbuf, size_t size));
- 
- /* Default parameter setup for compression */
- EXTERN(void) jpeg_set_defaults JPP((j_compress_ptr cinfo));
-@@ -1089,8 +1089,8 @@
-  */
- 
- #ifdef JPEG_INTERNALS
--#include "jpegint.h"		/* fetch private declarations */
--#include "jerror.h"		/* fetch error codes too */
-+#include "../jpeg-6b/jpegint.h"		/* fetch private declarations */
-+#include "../jpeg-6b/jerror.h"		/* fetch error codes too */
- #endif
- 
- #endif /* JPEGLIB_H */
diff --git a/src/jpeg-6b/jchuff.c b/src/jpeg-6b/jchuff.c
deleted file mode 100644
index f2352505..00000000
--- a/src/jpeg-6b/jchuff.c
+++ /dev/null
@@ -1,909 +0,0 @@
-/*
- * jchuff.c
- *
- * Copyright (C) 1991-1997, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This file contains Huffman entropy encoding routines.
- *
- * Much of the complexity here has to do with supporting output suspension.
- * If the data destination module demands suspension, we want to be able to
- * back up to the start of the current MCU.  To do this, we copy state
- * variables into local working storage, and update them back to the
- * permanent JPEG objects only upon successful completion of an MCU.
- */
-
-#define JPEG_INTERNALS
-#include "jinclude.h"
-#include "jpeglib.h"
-#include "jchuff.h"		/* Declarations shared with jcphuff.c */
-
-
-/* Expanded entropy encoder object for Huffman encoding.
- *
- * The savable_state subrecord contains fields that change within an MCU,
- * but must not be updated permanently until we complete the MCU.
- */
-
-typedef struct {
-  INT32 put_buffer;		/* current bit-accumulation buffer */
-  int put_bits;			/* # of bits now in it */
-  int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
-} savable_state;
-
-/* This macro is to work around compilers with missing or broken
- * structure assignment.  You'll need to fix this code if you have
- * such a compiler and you change MAX_COMPS_IN_SCAN.
- */
-
-#ifndef NO_STRUCT_ASSIGN
-#define ASSIGN_STATE(dest,src)  ((dest) = (src))
-#else
-#if MAX_COMPS_IN_SCAN == 4
-#define ASSIGN_STATE(dest,src)  \
-	((dest).put_buffer = (src).put_buffer, \
-	 (dest).put_bits = (src).put_bits, \
-	 (dest).last_dc_val[0] = (src).last_dc_val[0], \
-	 (dest).last_dc_val[1] = (src).last_dc_val[1], \
-	 (dest).last_dc_val[2] = (src).last_dc_val[2], \
-	 (dest).last_dc_val[3] = (src).last_dc_val[3])
-#endif
-#endif
-
-
-typedef struct {
-  struct jpeg_entropy_encoder pub; /* public fields */
-
-  savable_state saved;		/* Bit buffer & DC state at start of MCU */
-
-  /* These fields are NOT loaded into local working state. */
-  unsigned int restarts_to_go;	/* MCUs left in this restart interval */
-  int next_restart_num;		/* next restart number to write (0-7) */
-
-  /* Pointers to derived tables (these workspaces have image lifespan) */
-  c_derived_tbl * dc_derived_tbls[NUM_HUFF_TBLS];
-  c_derived_tbl * ac_derived_tbls[NUM_HUFF_TBLS];
-
-#ifdef ENTROPY_OPT_SUPPORTED	/* Statistics tables for optimization */
-  long * dc_count_ptrs[NUM_HUFF_TBLS];
-  long * ac_count_ptrs[NUM_HUFF_TBLS];
-#endif
-} huff_entropy_encoder;
-
-typedef huff_entropy_encoder * huff_entropy_ptr;
-
-/* Working state while writing an MCU.
- * This struct contains all the fields that are needed by subroutines.
- */
-
-typedef struct {
-  JOCTET * next_output_byte;	/* => next byte to write in buffer */
-  size_t free_in_buffer;	/* # of byte spaces remaining in buffer */
-  savable_state cur;		/* Current bit buffer & DC state */
-  j_compress_ptr cinfo;		/* dump_buffer needs access to this */
-} working_state;
-
-
-/* Forward declarations */
-METHODDEF(boolean) encode_mcu_huff JPP((j_compress_ptr cinfo,
-					JBLOCKROW *MCU_data));
-METHODDEF(void) finish_pass_huff JPP((j_compress_ptr cinfo));
-#ifdef ENTROPY_OPT_SUPPORTED
-METHODDEF(boolean) encode_mcu_gather JPP((j_compress_ptr cinfo,
-					  JBLOCKROW *MCU_data));
-METHODDEF(void) finish_pass_gather JPP((j_compress_ptr cinfo));
-#endif
-
-
-/*
- * Initialize for a Huffman-compressed scan.
- * If gather_statistics is TRUE, we do not output anything during the scan,
- * just count the Huffman symbols used and generate Huffman code tables.
- */
-
-METHODDEF(void)
-start_pass_huff (j_compress_ptr cinfo, boolean gather_statistics)
-{
-  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
-  int ci, dctbl, actbl;
-  jpeg_component_info * compptr;
-
-  if (gather_statistics) {
-#ifdef ENTROPY_OPT_SUPPORTED
-    entropy->pub.encode_mcu = encode_mcu_gather;
-    entropy->pub.finish_pass = finish_pass_gather;
-#else
-    ERREXIT(cinfo, JERR_NOT_COMPILED);
-#endif
-  } else {
-    entropy->pub.encode_mcu = encode_mcu_huff;
-    entropy->pub.finish_pass = finish_pass_huff;
-  }
-
-  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
-    compptr = cinfo->cur_comp_info[ci];
-    dctbl = compptr->dc_tbl_no;
-    actbl = compptr->ac_tbl_no;
-    if (gather_statistics) {
-#ifdef ENTROPY_OPT_SUPPORTED
-      /* Check for invalid table indexes */
-      /* (make_c_derived_tbl does this in the other path) */
-      if (dctbl < 0 || dctbl >= NUM_HUFF_TBLS)
-	ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, dctbl);
-      if (actbl < 0 || actbl >= NUM_HUFF_TBLS)
-	ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, actbl);
-      /* Allocate and zero the statistics tables */
-      /* Note that jpeg_gen_optimal_table expects 257 entries in each table! */
-      if (entropy->dc_count_ptrs[dctbl] == NULL)
-	entropy->dc_count_ptrs[dctbl] = (long *)
-	  (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				      257 * SIZEOF(long));
-      MEMZERO(entropy->dc_count_ptrs[dctbl], 257 * SIZEOF(long));
-      if (entropy->ac_count_ptrs[actbl] == NULL)
-	entropy->ac_count_ptrs[actbl] = (long *)
-	  (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				      257 * SIZEOF(long));
-      MEMZERO(entropy->ac_count_ptrs[actbl], 257 * SIZEOF(long));
-#endif
-    } else {
-      /* Compute derived values for Huffman tables */
-      /* We may do this more than once for a table, but it's not expensive */
-      jpeg_make_c_derived_tbl(cinfo, TRUE, dctbl,
-			      & entropy->dc_derived_tbls[dctbl]);
-      jpeg_make_c_derived_tbl(cinfo, FALSE, actbl,
-			      & entropy->ac_derived_tbls[actbl]);
-    }
-    /* Initialize DC predictions to 0 */
-    entropy->saved.last_dc_val[ci] = 0;
-  }
-
-  /* Initialize bit buffer to empty */
-  entropy->saved.put_buffer = 0;
-  entropy->saved.put_bits = 0;
-
-  /* Initialize restart stuff */
-  entropy->restarts_to_go = cinfo->restart_interval;
-  entropy->next_restart_num = 0;
-}
-
-
-/*
- * Compute the derived values for a Huffman table.
- * This routine also performs some validation checks on the table.
- *
- * Note this is also used by jcphuff.c.
- */
-
-GLOBAL(void)
-jpeg_make_c_derived_tbl (j_compress_ptr cinfo, boolean isDC, int tblno,
-			 c_derived_tbl ** pdtbl)
-{
-  JHUFF_TBL *htbl;
-  c_derived_tbl *dtbl;
-  int p, i, l, lastp, si, maxsymbol;
-  char huffsize[257];
-  unsigned int huffcode[257];
-  unsigned int code;
-
-  /* Note that huffsize[] and huffcode[] are filled in code-length order,
-   * paralleling the order of the symbols themselves in htbl->huffval[].
-   */
-
-  /* Find the input Huffman table */
-  if (tblno < 0 || tblno >= NUM_HUFF_TBLS)
-    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
-  htbl =
-    isDC ? cinfo->dc_huff_tbl_ptrs[tblno] : cinfo->ac_huff_tbl_ptrs[tblno];
-  if (htbl == NULL)
-    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
-
-  /* Allocate a workspace if we haven't already done so. */
-  if (*pdtbl == NULL)
-    *pdtbl = (c_derived_tbl *)
-      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				  SIZEOF(c_derived_tbl));
-  dtbl = *pdtbl;
-  
-  /* Figure C.1: make table of Huffman code length for each symbol */
-
-  p = 0;
-  for (l = 1; l <= 16; l++) {
-    i = (int) htbl->bits[l];
-    if (i < 0 || p + i > 256)	/* protect against table overrun */
-      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
-    while (i--)
-      huffsize[p++] = (char) l;
-  }
-  huffsize[p] = 0;
-  lastp = p;
-  
-  /* Figure C.2: generate the codes themselves */
-  /* We also validate that the counts represent a legal Huffman code tree. */
-
-  code = 0;
-  si = huffsize[0];
-  p = 0;
-  while (huffsize[p]) {
-    while (((int) huffsize[p]) == si) {
-      huffcode[p++] = code;
-      code++;
-    }
-    /* code is now 1 more than the last code used for codelength si; but
-     * it must still fit in si bits, since no code is allowed to be all ones.
-     */
-    if (((INT32) code) >= (((INT32) 1) << si))
-      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
-    code <<= 1;
-    si++;
-  }
-  
-  /* Figure C.3: generate encoding tables */
-  /* These are code and size indexed by symbol value */
-
-  /* Set all codeless symbols to have code length 0;
-   * this lets us detect duplicate VAL entries here, and later
-   * allows emit_bits to detect any attempt to emit such symbols.
-   */
-  MEMZERO(dtbl->ehufsi, SIZEOF(dtbl->ehufsi));
-
-  /* This is also a convenient place to check for out-of-range
-   * and duplicated VAL entries.  We allow 0..255 for AC symbols
-   * but only 0..15 for DC.  (We could constrain them further
-   * based on data depth and mode, but this seems enough.)
-   */
-  maxsymbol = isDC ? 15 : 255;
-
-  for (p = 0; p < lastp; p++) {
-    i = htbl->huffval[p];
-    if (i < 0 || i > maxsymbol || dtbl->ehufsi[i])
-      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
-    dtbl->ehufco[i] = huffcode[p];
-    dtbl->ehufsi[i] = huffsize[p];
-  }
-}
-
-
-/* Outputting bytes to the file */
-
-/* Emit a byte, taking 'action' if must suspend. */
-#define emit_byte(state,val,action)  \
-	{ *(state)->next_output_byte++ = (JOCTET) (val);  \
-	  if (--(state)->free_in_buffer == 0)  \
-	    if (! dump_buffer(state))  \
-	      { action; } }
-
-
-LOCAL(boolean)
-dump_buffer (working_state * state)
-/* Empty the output buffer; return TRUE if successful, FALSE if must suspend */
-{
-  struct jpeg_destination_mgr * dest = state->cinfo->dest;
-
-  if (! (*dest->empty_output_buffer) (state->cinfo))
-    return FALSE;
-  /* After a successful buffer dump, must reset buffer pointers */
-  state->next_output_byte = dest->next_output_byte;
-  state->free_in_buffer = dest->free_in_buffer;
-  return TRUE;
-}
-
-
-/* Outputting bits to the file */
-
-/* Only the right 24 bits of put_buffer are used; the valid bits are
- * left-justified in this part.  At most 16 bits can be passed to emit_bits
- * in one call, and we never retain more than 7 bits in put_buffer
- * between calls, so 24 bits are sufficient.
- */
-
-INLINE
-LOCAL(boolean)
-emit_bits (working_state * state, unsigned int code, int size)
-/* Emit some bits; return TRUE if successful, FALSE if must suspend */
-{
-  /* This routine is heavily used, so it's worth coding tightly. */
-  register INT32 put_buffer = (INT32) code;
-  register int put_bits = state->cur.put_bits;
-
-  /* if size is 0, caller used an invalid Huffman table entry */
-  if (size == 0)
-    ERREXIT(state->cinfo, JERR_HUFF_MISSING_CODE);
-
-  put_buffer &= (((INT32) 1)<<size) - 1; /* mask off any extra bits in code */
-  
-  put_bits += size;		/* new number of bits in buffer */
-  
-  put_buffer <<= 24 - put_bits; /* align incoming bits */
-
-  put_buffer |= state->cur.put_buffer; /* and merge with old buffer contents */
-  
-  while (put_bits >= 8) {
-    int c = (int) ((put_buffer >> 16) & 0xFF);
-    
-    emit_byte(state, c, return FALSE);
-    if (c == 0xFF) {		/* need to stuff a zero byte? */
-      emit_byte(state, 0, return FALSE);
-    }
-    put_buffer <<= 8;
-    put_bits -= 8;
-  }
-
-  state->cur.put_buffer = put_buffer; /* update state variables */
-  state->cur.put_bits = put_bits;
-
-  return TRUE;
-}
-
-
-LOCAL(boolean)
-flush_bits (working_state * state)
-{
-  if (! emit_bits(state, 0x7F, 7)) /* fill any partial byte with ones */
-    return FALSE;
-  state->cur.put_buffer = 0;	/* and reset bit-buffer to empty */
-  state->cur.put_bits = 0;
-  return TRUE;
-}
-
-
-/* Encode a single block's worth of coefficients */
-
-LOCAL(boolean)
-encode_one_block (working_state * state, JCOEFPTR block, int last_dc_val,
-		  c_derived_tbl *dctbl, c_derived_tbl *actbl)
-{
-  register int temp, temp2;
-  register int nbits;
-  register int k, r, i;
-  
-  /* Encode the DC coefficient difference per section F.1.2.1 */
-  
-  temp = temp2 = block[0] - last_dc_val;
-
-  if (temp < 0) {
-    temp = -temp;		/* temp is abs value of input */
-    /* For a negative input, want temp2 = bitwise complement of abs(input) */
-    /* This code assumes we are on a two's complement machine */
-    temp2--;
-  }
-  
-  /* Find the number of bits needed for the magnitude of the coefficient */
-  nbits = 0;
-  while (temp) {
-    nbits++;
-    temp >>= 1;
-  }
-  /* Check for out-of-range coefficient values.
-   * Since we're encoding a difference, the range limit is twice as much.
-   */
-  if (nbits > MAX_COEF_BITS+1)
-    ERREXIT(state->cinfo, JERR_BAD_DCT_COEF);
-  
-  /* Emit the Huffman-coded symbol for the number of bits */
-  if (! emit_bits(state, dctbl->ehufco[nbits], dctbl->ehufsi[nbits]))
-    return FALSE;
-
-  /* Emit that number of bits of the value, if positive, */
-  /* or the complement of its magnitude, if negative. */
-  if (nbits)			/* emit_bits rejects calls with size 0 */
-    if (! emit_bits(state, (unsigned int) temp2, nbits))
-      return FALSE;
-
-  /* Encode the AC coefficients per section F.1.2.2 */
-  
-  r = 0;			/* r = run length of zeros */
-  
-  for (k = 1; k < DCTSIZE2; k++) {
-    if ((temp = block[jpeg_natural_order[k]]) == 0) {
-      r++;
-    } else {
-      /* if run length > 15, must emit special run-length-16 codes (0xF0) */
-      while (r > 15) {
-	if (! emit_bits(state, actbl->ehufco[0xF0], actbl->ehufsi[0xF0]))
-	  return FALSE;
-	r -= 16;
-      }
-
-      temp2 = temp;
-      if (temp < 0) {
-	temp = -temp;		/* temp is abs value of input */
-	/* This code assumes we are on a two's complement machine */
-	temp2--;
-      }
-      
-      /* Find the number of bits needed for the magnitude of the coefficient */
-      nbits = 1;		/* there must be at least one 1 bit */
-      while ((temp >>= 1))
-	nbits++;
-      /* Check for out-of-range coefficient values */
-      if (nbits > MAX_COEF_BITS)
-	ERREXIT(state->cinfo, JERR_BAD_DCT_COEF);
-      
-      /* Emit Huffman symbol for run length / number of bits */
-      i = (r << 4) + nbits;
-      if (! emit_bits(state, actbl->ehufco[i], actbl->ehufsi[i]))
-	return FALSE;
-
-      /* Emit that number of bits of the value, if positive, */
-      /* or the complement of its magnitude, if negative. */
-      if (! emit_bits(state, (unsigned int) temp2, nbits))
-	return FALSE;
-      
-      r = 0;
-    }
-  }
-
-  /* If the last coef(s) were zero, emit an end-of-block code */
-  if (r > 0)
-    if (! emit_bits(state, actbl->ehufco[0], actbl->ehufsi[0]))
-      return FALSE;
-
-  return TRUE;
-}
-
-
-/*
- * Emit a restart marker & resynchronize predictions.
- */
-
-LOCAL(boolean)
-emit_restart (working_state * state, int restart_num)
-{
-  int ci;
-
-  if (! flush_bits(state))
-    return FALSE;
-
-  emit_byte(state, 0xFF, return FALSE);
-  emit_byte(state, JPEG_RST0 + restart_num, return FALSE);
-
-  /* Re-initialize DC predictions to 0 */
-  for (ci = 0; ci < state->cinfo->comps_in_scan; ci++)
-    state->cur.last_dc_val[ci] = 0;
-
-  /* The restart counter is not updated until we successfully write the MCU. */
-
-  return TRUE;
-}
-
-
-/*
- * Encode and output one MCU's worth of Huffman-compressed coefficients.
- */
-
-METHODDEF(boolean)
-encode_mcu_huff (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
-{
-  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
-  working_state state;
-  int blkn, ci;
-  jpeg_component_info * compptr;
-
-  /* Load up working state */
-  state.next_output_byte = cinfo->dest->next_output_byte;
-  state.free_in_buffer = cinfo->dest->free_in_buffer;
-  ASSIGN_STATE(state.cur, entropy->saved);
-  state.cinfo = cinfo;
-
-  /* Emit restart marker if needed */
-  if (cinfo->restart_interval) {
-    if (entropy->restarts_to_go == 0)
-      if (! emit_restart(&state, entropy->next_restart_num))
-	return FALSE;
-  }
-
-  /* Encode the MCU data blocks */
-  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
-    ci = cinfo->MCU_membership[blkn];
-    compptr = cinfo->cur_comp_info[ci];
-    if (! encode_one_block(&state,
-			   MCU_data[blkn][0], state.cur.last_dc_val[ci],
-			   entropy->dc_derived_tbls[compptr->dc_tbl_no],
-			   entropy->ac_derived_tbls[compptr->ac_tbl_no]))
-      return FALSE;
-    /* Update last_dc_val */
-    state.cur.last_dc_val[ci] = MCU_data[blkn][0][0];
-  }
-
-  /* Completed MCU, so update state */
-  cinfo->dest->next_output_byte = state.next_output_byte;
-  cinfo->dest->free_in_buffer = state.free_in_buffer;
-  ASSIGN_STATE(entropy->saved, state.cur);
-
-  /* Update restart-interval state too */
-  if (cinfo->restart_interval) {
-    if (entropy->restarts_to_go == 0) {
-      entropy->restarts_to_go = cinfo->restart_interval;
-      entropy->next_restart_num++;
-      entropy->next_restart_num &= 7;
-    }
-    entropy->restarts_to_go--;
-  }
-
-  return TRUE;
-}
-
-
-/*
- * Finish up at the end of a Huffman-compressed scan.
- */
-
-METHODDEF(void)
-finish_pass_huff (j_compress_ptr cinfo)
-{
-  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
-  working_state state;
-
-  /* Load up working state ... flush_bits needs it */
-  state.next_output_byte = cinfo->dest->next_output_byte;
-  state.free_in_buffer = cinfo->dest->free_in_buffer;
-  ASSIGN_STATE(state.cur, entropy->saved);
-  state.cinfo = cinfo;
-
-  /* Flush out the last data */
-  if (! flush_bits(&state))
-    ERREXIT(cinfo, JERR_CANT_SUSPEND);
-
-  /* Update state */
-  cinfo->dest->next_output_byte = state.next_output_byte;
-  cinfo->dest->free_in_buffer = state.free_in_buffer;
-  ASSIGN_STATE(entropy->saved, state.cur);
-}
-
-
-/*
- * Huffman coding optimization.
- *
- * We first scan the supplied data and count the number of uses of each symbol
- * that is to be Huffman-coded. (This process MUST agree with the code above.)
- * Then we build a Huffman coding tree for the observed counts.
- * Symbols which are not needed at all for the particular image are not
- * assigned any code, which saves space in the DHT marker as well as in
- * the compressed data.
- */
-
-#ifdef ENTROPY_OPT_SUPPORTED
-
-
-/* Process a single block's worth of coefficients */
-
-LOCAL(void)
-htest_one_block (j_compress_ptr cinfo, JCOEFPTR block, int last_dc_val,
-		 long dc_counts[], long ac_counts[])
-{
-  register int temp;
-  register int nbits;
-  register int k, r;
-  
-  /* Encode the DC coefficient difference per section F.1.2.1 */
-  
-  temp = block[0] - last_dc_val;
-  if (temp < 0)
-    temp = -temp;
-  
-  /* Find the number of bits needed for the magnitude of the coefficient */
-  nbits = 0;
-  while (temp) {
-    nbits++;
-    temp >>= 1;
-  }
-  /* Check for out-of-range coefficient values.
-   * Since we're encoding a difference, the range limit is twice as much.
-   */
-  if (nbits > MAX_COEF_BITS+1)
-    ERREXIT(cinfo, JERR_BAD_DCT_COEF);
-
-  /* Count the Huffman symbol for the number of bits */
-  dc_counts[nbits]++;
-  
-  /* Encode the AC coefficients per section F.1.2.2 */
-  
-  r = 0;			/* r = run length of zeros */
-  
-  for (k = 1; k < DCTSIZE2; k++) {
-    if ((temp = block[jpeg_natural_order[k]]) == 0) {
-      r++;
-    } else {
-      /* if run length > 15, must emit special run-length-16 codes (0xF0) */
-      while (r > 15) {
-	ac_counts[0xF0]++;
-	r -= 16;
-      }
-      
-      /* Find the number of bits needed for the magnitude of the coefficient */
-      if (temp < 0)
-	temp = -temp;
-      
-      /* Find the number of bits needed for the magnitude of the coefficient */
-      nbits = 1;		/* there must be at least one 1 bit */
-      while ((temp >>= 1))
-	nbits++;
-      /* Check for out-of-range coefficient values */
-      if (nbits > MAX_COEF_BITS)
-	ERREXIT(cinfo, JERR_BAD_DCT_COEF);
-      
-      /* Count Huffman symbol for run length / number of bits */
-      ac_counts[(r << 4) + nbits]++;
-      
-      r = 0;
-    }
-  }
-
-  /* If the last coef(s) were zero, emit an end-of-block code */
-  if (r > 0)
-    ac_counts[0]++;
-}
-
-
-/*
- * Trial-encode one MCU's worth of Huffman-compressed coefficients.
- * No data is actually output, so no suspension return is possible.
- */
-
-METHODDEF(boolean)
-encode_mcu_gather (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
-{
-  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
-  int blkn, ci;
-  jpeg_component_info * compptr;
-
-  /* Take care of restart intervals if needed */
-  if (cinfo->restart_interval) {
-    if (entropy->restarts_to_go == 0) {
-      /* Re-initialize DC predictions to 0 */
-      for (ci = 0; ci < cinfo->comps_in_scan; ci++)
-	entropy->saved.last_dc_val[ci] = 0;
-      /* Update restart state */
-      entropy->restarts_to_go = cinfo->restart_interval;
-    }
-    entropy->restarts_to_go--;
-  }
-
-  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
-    ci = cinfo->MCU_membership[blkn];
-    compptr = cinfo->cur_comp_info[ci];
-    htest_one_block(cinfo, MCU_data[blkn][0], entropy->saved.last_dc_val[ci],
-		    entropy->dc_count_ptrs[compptr->dc_tbl_no],
-		    entropy->ac_count_ptrs[compptr->ac_tbl_no]);
-    entropy->saved.last_dc_val[ci] = MCU_data[blkn][0][0];
-  }
-
-  return TRUE;
-}
-
-
-/*
- * Generate the best Huffman code table for the given counts, fill htbl.
- * Note this is also used by jcphuff.c.
- *
- * The JPEG standard requires that no symbol be assigned a codeword of all
- * one bits (so that padding bits added at the end of a compressed segment
- * can't look like a valid code).  Because of the canonical ordering of
- * codewords, this just means that there must be an unused slot in the
- * longest codeword length category.  Section K.2 of the JPEG spec suggests
- * reserving such a slot by pretending that symbol 256 is a valid symbol
- * with count 1.  In theory that's not optimal; giving it count zero but
- * including it in the symbol set anyway should give a better Huffman code.
- * But the theoretically better code actually seems to come out worse in
- * practice, because it produces more all-ones bytes (which incur stuffed
- * zero bytes in the final file).  In any case the difference is tiny.
- *
- * The JPEG standard requires Huffman codes to be no more than 16 bits long.
- * If some symbols have a very small but nonzero probability, the Huffman tree
- * must be adjusted to meet the code length restriction.  We currently use
- * the adjustment method suggested in JPEG section K.2.  This method is *not*
- * optimal; it may not choose the best possible limited-length code.  But
- * typically only very-low-frequency symbols will be given less-than-optimal
- * lengths, so the code is almost optimal.  Experimental comparisons against
- * an optimal limited-length-code algorithm indicate that the difference is
- * microscopic --- usually less than a hundredth of a percent of total size.
- * So the extra complexity of an optimal algorithm doesn't seem worthwhile.
- */
-
-GLOBAL(void)
-jpeg_gen_optimal_table (j_compress_ptr cinfo, JHUFF_TBL * htbl, long freq[])
-{
-#define MAX_CLEN 32		/* assumed maximum initial code length */
-  UINT8 bits[MAX_CLEN+1];	/* bits[k] = # of symbols with code length k */
-  int codesize[257];		/* codesize[k] = code length of symbol k */
-  int others[257];		/* next symbol in current branch of tree */
-  int c1, c2;
-  int p, i, j;
-  long v;
-
-  /* This algorithm is explained in section K.2 of the JPEG standard */
-
-  MEMZERO(bits, SIZEOF(bits));
-  MEMZERO(codesize, SIZEOF(codesize));
-  for (i = 0; i < 257; i++)
-    others[i] = -1;		/* init links to empty */
-  
-  freq[256] = 1;		/* make sure 256 has a nonzero count */
-  /* Including the pseudo-symbol 256 in the Huffman procedure guarantees
-   * that no real symbol is given code-value of all ones, because 256
-   * will be placed last in the largest codeword category.
-   */
-
-  /* Huffman's basic algorithm to assign optimal code lengths to symbols */
-
-  for (;;) {
-    /* Find the smallest nonzero frequency, set c1 = its symbol */
-    /* In case of ties, take the larger symbol number */
-    c1 = -1;
-    v = 1000000000L;
-    for (i = 0; i <= 256; i++) {
-      if (freq[i] && freq[i] <= v) {
-	v = freq[i];
-	c1 = i;
-      }
-    }
-
-    /* Find the next smallest nonzero frequency, set c2 = its symbol */
-    /* In case of ties, take the larger symbol number */
-    c2 = -1;
-    v = 1000000000L;
-    for (i = 0; i <= 256; i++) {
-      if (freq[i] && freq[i] <= v && i != c1) {
-	v = freq[i];
-	c2 = i;
-      }
-    }
-
-    /* Done if we've merged everything into one frequency */
-    if (c2 < 0)
-      break;
-    
-    /* Else merge the two counts/trees */
-    freq[c1] += freq[c2];
-    freq[c2] = 0;
-
-    /* Increment the codesize of everything in c1's tree branch */
-    codesize[c1]++;
-    while (others[c1] >= 0) {
-      c1 = others[c1];
-      codesize[c1]++;
-    }
-    
-    others[c1] = c2;		/* chain c2 onto c1's tree branch */
-    
-    /* Increment the codesize of everything in c2's tree branch */
-    codesize[c2]++;
-    while (others[c2] >= 0) {
-      c2 = others[c2];
-      codesize[c2]++;
-    }
-  }
-
-  /* Now count the number of symbols of each code length */
-  for (i = 0; i <= 256; i++) {
-    if (codesize[i]) {
-      /* The JPEG standard seems to think that this can't happen, */
-      /* but I'm paranoid... */
-      if (codesize[i] > MAX_CLEN)
-	ERREXIT(cinfo, JERR_HUFF_CLEN_OVERFLOW);
-
-      bits[codesize[i]]++;
-    }
-  }
-
-  /* JPEG doesn't allow symbols with code lengths over 16 bits, so if the pure
-   * Huffman procedure assigned any such lengths, we must adjust the coding.
-   * Here is what the JPEG spec says about how this next bit works:
-   * Since symbols are paired for the longest Huffman code, the symbols are
-   * removed from this length category two at a time.  The prefix for the pair
-   * (which is one bit shorter) is allocated to one of the pair; then,
-   * skipping the BITS entry for that prefix length, a code word from the next
-   * shortest nonzero BITS entry is converted into a prefix for two code words
-   * one bit longer.
-   */
-  
-  for (i = MAX_CLEN; i > 16; i--) {
-    while (bits[i] > 0) {
-      j = i - 2;		/* find length of new prefix to be used */
-      while (bits[j] == 0)
-	j--;
-      
-      bits[i] -= 2;		/* remove two symbols */
-      bits[i-1]++;		/* one goes in this length */
-      bits[j+1] += 2;		/* two new symbols in this length */
-      bits[j]--;		/* symbol of this length is now a prefix */
-    }
-  }
-
-  /* Remove the count for the pseudo-symbol 256 from the largest codelength */
-  while (bits[i] == 0)		/* find largest codelength still in use */
-    i--;
-  bits[i]--;
-  
-  /* Return final symbol counts (only for lengths 0..16) */
-  MEMCOPY(htbl->bits, bits, SIZEOF(htbl->bits));
-  
-  /* Return a list of the symbols sorted by code length */
-  /* It's not real clear to me why we don't need to consider the codelength
-   * changes made above, but the JPEG spec seems to think this works.
-   */
-  p = 0;
-  for (i = 1; i <= MAX_CLEN; i++) {
-    for (j = 0; j <= 255; j++) {
-      if (codesize[j] == i) {
-	htbl->huffval[p] = (UINT8) j;
-	p++;
-      }
-    }
-  }
-
-  /* Set sent_table FALSE so updated table will be written to JPEG file. */
-  htbl->sent_table = FALSE;
-}
-
-
-/*
- * Finish up a statistics-gathering pass and create the new Huffman tables.
- */
-
-METHODDEF(void)
-finish_pass_gather (j_compress_ptr cinfo)
-{
-  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
-  int ci, dctbl, actbl;
-  jpeg_component_info * compptr;
-  JHUFF_TBL **htblptr;
-  boolean did_dc[NUM_HUFF_TBLS];
-  boolean did_ac[NUM_HUFF_TBLS];
-
-  /* It's important not to apply jpeg_gen_optimal_table more than once
-   * per table, because it clobbers the input frequency counts!
-   */
-  MEMZERO(did_dc, SIZEOF(did_dc));
-  MEMZERO(did_ac, SIZEOF(did_ac));
-
-  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
-    compptr = cinfo->cur_comp_info[ci];
-    dctbl = compptr->dc_tbl_no;
-    actbl = compptr->ac_tbl_no;
-    if (! did_dc[dctbl]) {
-      htblptr = & cinfo->dc_huff_tbl_ptrs[dctbl];
-      if (*htblptr == NULL)
-	*htblptr = jpeg_alloc_huff_table((j_common_ptr) cinfo);
-      jpeg_gen_optimal_table(cinfo, *htblptr, entropy->dc_count_ptrs[dctbl]);
-      did_dc[dctbl] = TRUE;
-    }
-    if (! did_ac[actbl]) {
-      htblptr = & cinfo->ac_huff_tbl_ptrs[actbl];
-      if (*htblptr == NULL)
-	*htblptr = jpeg_alloc_huff_table((j_common_ptr) cinfo);
-      jpeg_gen_optimal_table(cinfo, *htblptr, entropy->ac_count_ptrs[actbl]);
-      did_ac[actbl] = TRUE;
-    }
-  }
-}
-
-
-#endif /* ENTROPY_OPT_SUPPORTED */
-
-
-/*
- * Module initialization routine for Huffman entropy encoding.
- */
-
-GLOBAL(void)
-jinit_huff_encoder (j_compress_ptr cinfo)
-{
-  huff_entropy_ptr entropy;
-  int i;
-
-  entropy = (huff_entropy_ptr)
-    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				SIZEOF(huff_entropy_encoder));
-  cinfo->entropy = (struct jpeg_entropy_encoder *) entropy;
-  entropy->pub.start_pass = start_pass_huff;
-
-  /* Mark tables unallocated */
-  for (i = 0; i < NUM_HUFF_TBLS; i++) {
-    entropy->dc_derived_tbls[i] = entropy->ac_derived_tbls[i] = NULL;
-#ifdef ENTROPY_OPT_SUPPORTED
-    entropy->dc_count_ptrs[i] = entropy->ac_count_ptrs[i] = NULL;
-#endif
-  }
-}
diff --git a/src/jpeg-6b/jchuff.h b/src/jpeg-6b/jchuff.h
deleted file mode 100644
index a9599fc1..00000000
--- a/src/jpeg-6b/jchuff.h
+++ /dev/null
@@ -1,47 +0,0 @@
-/*
- * jchuff.h
- *
- * Copyright (C) 1991-1997, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This file contains declarations for Huffman entropy encoding routines
- * that are shared between the sequential encoder (jchuff.c) and the
- * progressive encoder (jcphuff.c).  No other modules need to see these.
- */
-
-/* The legal range of a DCT coefficient is
- *  -1024 .. +1023  for 8-bit data;
- * -16384 .. +16383 for 12-bit data.
- * Hence the magnitude should always fit in 10 or 14 bits respectively.
- */
-
-#if BITS_IN_JSAMPLE == 8
-#define MAX_COEF_BITS 10
-#else
-#define MAX_COEF_BITS 14
-#endif
-
-/* Derived data constructed for each Huffman table */
-
-typedef struct {
-  unsigned int ehufco[256];	/* code for each symbol */
-  char ehufsi[256];		/* length of code for each symbol */
-  /* If no code has been allocated for a symbol S, ehufsi[S] contains 0 */
-} c_derived_tbl;
-
-/* Short forms of external names for systems with brain-damaged linkers. */
-
-#ifdef NEED_SHORT_EXTERNAL_NAMES
-#define jpeg_make_c_derived_tbl	jMkCDerived
-#define jpeg_gen_optimal_table	jGenOptTbl
-#endif /* NEED_SHORT_EXTERNAL_NAMES */
-
-/* Expand a Huffman table definition into the derived format */
-EXTERN(void) jpeg_make_c_derived_tbl
-	JPP((j_compress_ptr cinfo, boolean isDC, int tblno,
-	     c_derived_tbl ** pdtbl));
-
-/* Generate an optimal table definition given the specified counts */
-EXTERN(void) jpeg_gen_optimal_table
-	JPP((j_compress_ptr cinfo, JHUFF_TBL * htbl, long freq[]));
diff --git a/src/jpeg-6b/jconfig.h b/src/jpeg-6b/jconfig.h
deleted file mode 100644
index 361586ee..00000000
--- a/src/jpeg-6b/jconfig.h
+++ /dev/null
@@ -1,40 +0,0 @@
-/* jconfig.wat --- jconfig.h for Watcom C/C++ on MS-DOS or OS/2. */
-/* see jconfig.doc for explanations */
-
-#define HAVE_PROTOTYPES
-#define HAVE_UNSIGNED_CHAR
-#define HAVE_UNSIGNED_SHORT
-/* #define void char */
-/* #define const */
-#define CHAR_IS_UNSIGNED
-#define HAVE_STDDEF_H
-#undef NEED_BSD_STRINGS
-#undef NEED_SYS_TYPES_H
-#undef NEED_FAR_POINTERS	/* Watcom uses flat 32-bit addressing */
-#undef NEED_SHORT_EXTERNAL_NAMES
-#undef INCOMPLETE_TYPES_BROKEN
-
-#define JDCT_DEFAULT  JDCT_FLOAT
-#define JDCT_FASTEST  JDCT_FLOAT
-
-#ifdef JPEG_INTERNALS
-
-#undef RIGHT_SHIFT_IS_UNSIGNED
-
-#endif /* JPEG_INTERNALS */
-
-#ifdef JPEG_CJPEG_DJPEG
-
-#define BMP_SUPPORTED		/* BMP image file format */
-#define GIF_SUPPORTED		/* GIF image file format */
-#define PPM_SUPPORTED		/* PBMPLUS PPM/PGM image file format */
-#undef RLE_SUPPORTED		/* Utah RLE image file format */
-#define TARGA_SUPPORTED		/* Targa image file format */
-
-#undef TWO_FILE_COMMANDLINE	/* optional */
-#define USE_SETMODE		/* Needed to make one-file style work in Watcom */
-#undef NEED_SIGNAL_CATCHER	/* Define this if you use jmemname.c */
-#undef DONT_USE_B_MODE
-#undef PROGRESS_REPORT		/* optional */
-
-#endif /* JPEG_CJPEG_DJPEG */
diff --git a/src/jpeg-6b/jcphuff.c b/src/jpeg-6b/jcphuff.c
deleted file mode 100644
index 07f9178b..00000000
--- a/src/jpeg-6b/jcphuff.c
+++ /dev/null
@@ -1,833 +0,0 @@
-/*
- * jcphuff.c
- *
- * Copyright (C) 1995-1997, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This file contains Huffman entropy encoding routines for progressive JPEG.
- *
- * We do not support output suspension in this module, since the library
- * currently does not allow multiple-scan files to be written with output
- * suspension.
- */
-
-#define JPEG_INTERNALS
-#include "jinclude.h"
-#include "jpeglib.h"
-#include "jchuff.h"		/* Declarations shared with jchuff.c */
-
-#ifdef C_PROGRESSIVE_SUPPORTED
-
-/* Expanded entropy encoder object for progressive Huffman encoding. */
-
-typedef struct {
-  struct jpeg_entropy_encoder pub; /* public fields */
-
-  /* Mode flag: TRUE for optimization, FALSE for actual data output */
-  boolean gather_statistics;
-
-  /* Bit-level coding status.
-   * next_output_byte/free_in_buffer are local copies of cinfo->dest fields.
-   */
-  JOCTET * next_output_byte;	/* => next byte to write in buffer */
-  size_t free_in_buffer;	/* # of byte spaces remaining in buffer */
-  INT32 put_buffer;		/* current bit-accumulation buffer */
-  int put_bits;			/* # of bits now in it */
-  j_compress_ptr cinfo;		/* link to cinfo (needed for dump_buffer) */
-
-  /* Coding status for DC components */
-  int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
-
-  /* Coding status for AC components */
-  int ac_tbl_no;		/* the table number of the single component */
-  unsigned int EOBRUN;		/* run length of EOBs */
-  unsigned int BE;		/* # of buffered correction bits before MCU */
-  char * bit_buffer;		/* buffer for correction bits (1 per char) */
-  /* packing correction bits tightly would save some space but cost time... */
-
-  unsigned int restarts_to_go;	/* MCUs left in this restart interval */
-  int next_restart_num;		/* next restart number to write (0-7) */
-
-  /* Pointers to derived tables (these workspaces have image lifespan).
-   * Since any one scan codes only DC or only AC, we only need one set
-   * of tables, not one for DC and one for AC.
-   */
-  c_derived_tbl * derived_tbls[NUM_HUFF_TBLS];
-
-  /* Statistics tables for optimization; again, one set is enough */
-  long * count_ptrs[NUM_HUFF_TBLS];
-} phuff_entropy_encoder;
-
-typedef phuff_entropy_encoder * phuff_entropy_ptr;
-
-/* MAX_CORR_BITS is the number of bits the AC refinement correction-bit
- * buffer can hold.  Larger sizes may slightly improve compression, but
- * 1000 is already well into the realm of overkill.
- * The minimum safe size is 64 bits.
- */
-
-#define MAX_CORR_BITS  1000	/* Max # of correction bits I can buffer */
-
-/* IRIGHT_SHIFT is like RIGHT_SHIFT, but works on int rather than INT32.
- * We assume that int right shift is unsigned if INT32 right shift is,
- * which should be safe.
- */
-
-#ifdef RIGHT_SHIFT_IS_UNSIGNED
-#define ISHIFT_TEMPS	int ishift_temp;
-#define IRIGHT_SHIFT(x,shft)  \
-	((ishift_temp = (x)) < 0 ? \
-	 (ishift_temp >> (shft)) | ((~0) << (16-(shft))) : \
-	 (ishift_temp >> (shft)))
-#else
-#define ISHIFT_TEMPS
-#define IRIGHT_SHIFT(x,shft)	((x) >> (shft))
-#endif
-
-/* Forward declarations */
-METHODDEF(boolean) encode_mcu_DC_first JPP((j_compress_ptr cinfo,
-					    JBLOCKROW *MCU_data));
-METHODDEF(boolean) encode_mcu_AC_first JPP((j_compress_ptr cinfo,
-					    JBLOCKROW *MCU_data));
-METHODDEF(boolean) encode_mcu_DC_refine JPP((j_compress_ptr cinfo,
-					     JBLOCKROW *MCU_data));
-METHODDEF(boolean) encode_mcu_AC_refine JPP((j_compress_ptr cinfo,
-					     JBLOCKROW *MCU_data));
-METHODDEF(void) finish_pass_phuff JPP((j_compress_ptr cinfo));
-METHODDEF(void) finish_pass_gather_phuff JPP((j_compress_ptr cinfo));
-
-
-/*
- * Initialize for a Huffman-compressed scan using progressive JPEG.
- */
-
-METHODDEF(void)
-start_pass_phuff (j_compress_ptr cinfo, boolean gather_statistics)
-{  
-  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
-  boolean is_DC_band;
-  int ci, tbl;
-  jpeg_component_info * compptr;
-
-  entropy->cinfo = cinfo;
-  entropy->gather_statistics = gather_statistics;
-
-  is_DC_band = (cinfo->Ss == 0);
-
-  /* We assume jcmaster.c already validated the scan parameters. */
-
-  /* Select execution routines */
-  if (cinfo->Ah == 0) {
-    if (is_DC_band)
-      entropy->pub.encode_mcu = encode_mcu_DC_first;
-    else
-      entropy->pub.encode_mcu = encode_mcu_AC_first;
-  } else {
-    if (is_DC_band)
-      entropy->pub.encode_mcu = encode_mcu_DC_refine;
-    else {
-      entropy->pub.encode_mcu = encode_mcu_AC_refine;
-      /* AC refinement needs a correction bit buffer */
-      if (entropy->bit_buffer == NULL)
-	entropy->bit_buffer = (char *)
-	  (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				      MAX_CORR_BITS * SIZEOF(char));
-    }
-  }
-  if (gather_statistics)
-    entropy->pub.finish_pass = finish_pass_gather_phuff;
-  else
-    entropy->pub.finish_pass = finish_pass_phuff;
-
-  /* Only DC coefficients may be interleaved, so cinfo->comps_in_scan = 1
-   * for AC coefficients.
-   */
-  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
-    compptr = cinfo->cur_comp_info[ci];
-    /* Initialize DC predictions to 0 */
-    entropy->last_dc_val[ci] = 0;
-    /* Get table index */
-    if (is_DC_band) {
-      if (cinfo->Ah != 0)	/* DC refinement needs no table */
-	continue;
-      tbl = compptr->dc_tbl_no;
-    } else {
-      entropy->ac_tbl_no = tbl = compptr->ac_tbl_no;
-    }
-    if (gather_statistics) {
-      /* Check for invalid table index */
-      /* (make_c_derived_tbl does this in the other path) */
-      if (tbl < 0 || tbl >= NUM_HUFF_TBLS)
-        ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tbl);
-      /* Allocate and zero the statistics tables */
-      /* Note that jpeg_gen_optimal_table expects 257 entries in each table! */
-      if (entropy->count_ptrs[tbl] == NULL)
-	entropy->count_ptrs[tbl] = (long *)
-	  (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				      257 * SIZEOF(long));
-      MEMZERO(entropy->count_ptrs[tbl], 257 * SIZEOF(long));
-    } else {
-      /* Compute derived values for Huffman table */
-      /* We may do this more than once for a table, but it's not expensive */
-      jpeg_make_c_derived_tbl(cinfo, is_DC_band, tbl,
-			      & entropy->derived_tbls[tbl]);
-    }
-  }
-
-  /* Initialize AC stuff */
-  entropy->EOBRUN = 0;
-  entropy->BE = 0;
-
-  /* Initialize bit buffer to empty */
-  entropy->put_buffer = 0;
-  entropy->put_bits = 0;
-
-  /* Initialize restart stuff */
-  entropy->restarts_to_go = cinfo->restart_interval;
-  entropy->next_restart_num = 0;
-}
-
-
-/* Outputting bytes to the file.
- * NB: these must be called only when actually outputting,
- * that is, entropy->gather_statistics == FALSE.
- */
-
-/* Emit a byte */
-#define emit_byte(entropy,val)  \
-	{ *(entropy)->next_output_byte++ = (JOCTET) (val);  \
-	  if (--(entropy)->free_in_buffer == 0)  \
-	    dump_buffer(entropy); }
-
-
-LOCAL(void)
-dump_buffer (phuff_entropy_ptr entropy)
-/* Empty the output buffer; we do not support suspension in this module. */
-{
-  struct jpeg_destination_mgr * dest = entropy->cinfo->dest;
-
-  if (! (*dest->empty_output_buffer) (entropy->cinfo))
-    ERREXIT(entropy->cinfo, JERR_CANT_SUSPEND);
-  /* After a successful buffer dump, must reset buffer pointers */
-  entropy->next_output_byte = dest->next_output_byte;
-  entropy->free_in_buffer = dest->free_in_buffer;
-}
-
-
-/* Outputting bits to the file */
-
-/* Only the right 24 bits of put_buffer are used; the valid bits are
- * left-justified in this part.  At most 16 bits can be passed to emit_bits
- * in one call, and we never retain more than 7 bits in put_buffer
- * between calls, so 24 bits are sufficient.
- */
-
-INLINE
-LOCAL(void)
-emit_bits (phuff_entropy_ptr entropy, unsigned int code, int size)
-/* Emit some bits, unless we are in gather mode */
-{
-  /* This routine is heavily used, so it's worth coding tightly. */
-  register INT32 put_buffer = (INT32) code;
-  register int put_bits = entropy->put_bits;
-
-  /* if size is 0, caller used an invalid Huffman table entry */
-  if (size == 0)
-    ERREXIT(entropy->cinfo, JERR_HUFF_MISSING_CODE);
-
-  if (entropy->gather_statistics)
-    return;			/* do nothing if we're only getting stats */
-
-  put_buffer &= (((INT32) 1)<<size) - 1; /* mask off any extra bits in code */
-  
-  put_bits += size;		/* new number of bits in buffer */
-  
-  put_buffer <<= 24 - put_bits; /* align incoming bits */
-
-  put_buffer |= entropy->put_buffer; /* and merge with old buffer contents */
-
-  while (put_bits >= 8) {
-    int c = (int) ((put_buffer >> 16) & 0xFF);
-    
-    emit_byte(entropy, c);
-    if (c == 0xFF) {		/* need to stuff a zero byte? */
-      emit_byte(entropy, 0);
-    }
-    put_buffer <<= 8;
-    put_bits -= 8;
-  }
-
-  entropy->put_buffer = put_buffer; /* update variables */
-  entropy->put_bits = put_bits;
-}
-
-
-LOCAL(void)
-flush_bits (phuff_entropy_ptr entropy)
-{
-  emit_bits(entropy, 0x7F, 7); /* fill any partial byte with ones */
-  entropy->put_buffer = 0;     /* and reset bit-buffer to empty */
-  entropy->put_bits = 0;
-}
-
-
-/*
- * Emit (or just count) a Huffman symbol.
- */
-
-INLINE
-LOCAL(void)
-emit_symbol (phuff_entropy_ptr entropy, int tbl_no, int symbol)
-{
-  if (entropy->gather_statistics)
-    entropy->count_ptrs[tbl_no][symbol]++;
-  else {
-    c_derived_tbl * tbl = entropy->derived_tbls[tbl_no];
-    emit_bits(entropy, tbl->ehufco[symbol], tbl->ehufsi[symbol]);
-  }
-}
-
-
-/*
- * Emit bits from a correction bit buffer.
- */
-
-LOCAL(void)
-emit_buffered_bits (phuff_entropy_ptr entropy, char * bufstart,
-		    unsigned int nbits)
-{
-  if (entropy->gather_statistics)
-    return;			/* no real work */
-
-  while (nbits > 0) {
-    emit_bits(entropy, (unsigned int) (*bufstart), 1);
-    bufstart++;
-    nbits--;
-  }
-}
-
-
-/*
- * Emit any pending EOBRUN symbol.
- */
-
-LOCAL(void)
-emit_eobrun (phuff_entropy_ptr entropy)
-{
-  register int temp, nbits;
-
-  if (entropy->EOBRUN > 0) {	/* if there is any pending EOBRUN */
-    temp = entropy->EOBRUN;
-    nbits = 0;
-    while ((temp >>= 1))
-      nbits++;
-    /* safety check: shouldn't happen given limited correction-bit buffer */
-    if (nbits > 14)
-      ERREXIT(entropy->cinfo, JERR_HUFF_MISSING_CODE);
-
-    emit_symbol(entropy, entropy->ac_tbl_no, nbits << 4);
-    if (nbits)
-      emit_bits(entropy, entropy->EOBRUN, nbits);
-
-    entropy->EOBRUN = 0;
-
-    /* Emit any buffered correction bits */
-    emit_buffered_bits(entropy, entropy->bit_buffer, entropy->BE);
-    entropy->BE = 0;
-  }
-}
-
-
-/*
- * Emit a restart marker & resynchronize predictions.
- */
-
-LOCAL(void)
-emit_restart (phuff_entropy_ptr entropy, int restart_num)
-{
-  int ci;
-
-  emit_eobrun(entropy);
-
-  if (! entropy->gather_statistics) {
-    flush_bits(entropy);
-    emit_byte(entropy, 0xFF);
-    emit_byte(entropy, JPEG_RST0 + restart_num);
-  }
-
-  if (entropy->cinfo->Ss == 0) {
-    /* Re-initialize DC predictions to 0 */
-    for (ci = 0; ci < entropy->cinfo->comps_in_scan; ci++)
-      entropy->last_dc_val[ci] = 0;
-  } else {
-    /* Re-initialize all AC-related fields to 0 */
-    entropy->EOBRUN = 0;
-    entropy->BE = 0;
-  }
-}
-
-
-/*
- * MCU encoding for DC initial scan (either spectral selection,
- * or first pass of successive approximation).
- */
-
-METHODDEF(boolean)
-encode_mcu_DC_first (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
-{
-  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
-  register int temp, temp2;
-  register int nbits;
-  int blkn, ci;
-  int Al = cinfo->Al;
-  JBLOCKROW block;
-  jpeg_component_info * compptr;
-  ISHIFT_TEMPS
-
-  entropy->next_output_byte = cinfo->dest->next_output_byte;
-  entropy->free_in_buffer = cinfo->dest->free_in_buffer;
-
-  /* Emit restart marker if needed */
-  if (cinfo->restart_interval)
-    if (entropy->restarts_to_go == 0)
-      emit_restart(entropy, entropy->next_restart_num);
-
-  /* Encode the MCU data blocks */
-  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
-    block = MCU_data[blkn];
-    ci = cinfo->MCU_membership[blkn];
-    compptr = cinfo->cur_comp_info[ci];
-
-    /* Compute the DC value after the required point transform by Al.
-     * This is simply an arithmetic right shift.
-     */
-    temp2 = IRIGHT_SHIFT((int) ((*block)[0]), Al);
-
-    /* DC differences are figured on the point-transformed values. */
-    temp = temp2 - entropy->last_dc_val[ci];
-    entropy->last_dc_val[ci] = temp2;
-
-    /* Encode the DC coefficient difference per section G.1.2.1 */
-    temp2 = temp;
-    if (temp < 0) {
-      temp = -temp;		/* temp is abs value of input */
-      /* For a negative input, want temp2 = bitwise complement of abs(input) */
-      /* This code assumes we are on a two's complement machine */
-      temp2--;
-    }
-    
-    /* Find the number of bits needed for the magnitude of the coefficient */
-    nbits = 0;
-    while (temp) {
-      nbits++;
-      temp >>= 1;
-    }
-    /* Check for out-of-range coefficient values.
-     * Since we're encoding a difference, the range limit is twice as much.
-     */
-    if (nbits > MAX_COEF_BITS+1)
-      ERREXIT(cinfo, JERR_BAD_DCT_COEF);
-    
-    /* Count/emit the Huffman-coded symbol for the number of bits */
-    emit_symbol(entropy, compptr->dc_tbl_no, nbits);
-    
-    /* Emit that number of bits of the value, if positive, */
-    /* or the complement of its magnitude, if negative. */
-    if (nbits)			/* emit_bits rejects calls with size 0 */
-      emit_bits(entropy, (unsigned int) temp2, nbits);
-  }
-
-  cinfo->dest->next_output_byte = entropy->next_output_byte;
-  cinfo->dest->free_in_buffer = entropy->free_in_buffer;
-
-  /* Update restart-interval state too */
-  if (cinfo->restart_interval) {
-    if (entropy->restarts_to_go == 0) {
-      entropy->restarts_to_go = cinfo->restart_interval;
-      entropy->next_restart_num++;
-      entropy->next_restart_num &= 7;
-    }
-    entropy->restarts_to_go--;
-  }
-
-  return TRUE;
-}
-
-
-/*
- * MCU encoding for AC initial scan (either spectral selection,
- * or first pass of successive approximation).
- */
-
-METHODDEF(boolean)
-encode_mcu_AC_first (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
-{
-  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
-  register int temp, temp2;
-  register int nbits;
-  register int r, k;
-  int Se = cinfo->Se;
-  int Al = cinfo->Al;
-  JBLOCKROW block;
-
-  entropy->next_output_byte = cinfo->dest->next_output_byte;
-  entropy->free_in_buffer = cinfo->dest->free_in_buffer;
-
-  /* Emit restart marker if needed */
-  if (cinfo->restart_interval)
-    if (entropy->restarts_to_go == 0)
-      emit_restart(entropy, entropy->next_restart_num);
-
-  /* Encode the MCU data block */
-  block = MCU_data[0];
-
-  /* Encode the AC coefficients per section G.1.2.2, fig. G.3 */
-  
-  r = 0;			/* r = run length of zeros */
-   
-  for (k = cinfo->Ss; k <= Se; k++) {
-    if ((temp = (*block)[jpeg_natural_order[k]]) == 0) {
-      r++;
-      continue;
-    }
-    /* We must apply the point transform by Al.  For AC coefficients this
-     * is an integer division with rounding towards 0.  To do this portably
-     * in C, we shift after obtaining the absolute value; so the code is
-     * interwoven with finding the abs value (temp) and output bits (temp2).
-     */
-    if (temp < 0) {
-      temp = -temp;		/* temp is abs value of input */
-      temp >>= Al;		/* apply the point transform */
-      /* For a negative coef, want temp2 = bitwise complement of abs(coef) */
-      temp2 = ~temp;
-    } else {
-      temp >>= Al;		/* apply the point transform */
-      temp2 = temp;
-    }
-    /* Watch out for case that nonzero coef is zero after point transform */
-    if (temp == 0) {
-      r++;
-      continue;
-    }
-
-    /* Emit any pending EOBRUN */
-    if (entropy->EOBRUN > 0)
-      emit_eobrun(entropy);
-    /* if run length > 15, must emit special run-length-16 codes (0xF0) */
-    while (r > 15) {
-      emit_symbol(entropy, entropy->ac_tbl_no, 0xF0);
-      r -= 16;
-    }
-
-    /* Find the number of bits needed for the magnitude of the coefficient */
-    nbits = 1;			/* there must be at least one 1 bit */
-    while ((temp >>= 1))
-      nbits++;
-    /* Check for out-of-range coefficient values */
-    if (nbits > MAX_COEF_BITS)
-      ERREXIT(cinfo, JERR_BAD_DCT_COEF);
-
-    /* Count/emit Huffman symbol for run length / number of bits */
-    emit_symbol(entropy, entropy->ac_tbl_no, (r << 4) + nbits);
-
-    /* Emit that number of bits of the value, if positive, */
-    /* or the complement of its magnitude, if negative. */
-    emit_bits(entropy, (unsigned int) temp2, nbits);
-
-    r = 0;			/* reset zero run length */
-  }
-
-  if (r > 0) {			/* If there are trailing zeroes, */
-    entropy->EOBRUN++;		/* count an EOB */
-    if (entropy->EOBRUN == 0x7FFF)
-      emit_eobrun(entropy);	/* force it out to avoid overflow */
-  }
-
-  cinfo->dest->next_output_byte = entropy->next_output_byte;
-  cinfo->dest->free_in_buffer = entropy->free_in_buffer;
-
-  /* Update restart-interval state too */
-  if (cinfo->restart_interval) {
-    if (entropy->restarts_to_go == 0) {
-      entropy->restarts_to_go = cinfo->restart_interval;
-      entropy->next_restart_num++;
-      entropy->next_restart_num &= 7;
-    }
-    entropy->restarts_to_go--;
-  }
-
-  return TRUE;
-}
-
-
-/*
- * MCU encoding for DC successive approximation refinement scan.
- * Note: we assume such scans can be multi-component, although the spec
- * is not very clear on the point.
- */
-
-METHODDEF(boolean)
-encode_mcu_DC_refine (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
-{
-  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
-  register int temp;
-  int blkn;
-  int Al = cinfo->Al;
-  JBLOCKROW block;
-
-  entropy->next_output_byte = cinfo->dest->next_output_byte;
-  entropy->free_in_buffer = cinfo->dest->free_in_buffer;
-
-  /* Emit restart marker if needed */
-  if (cinfo->restart_interval)
-    if (entropy->restarts_to_go == 0)
-      emit_restart(entropy, entropy->next_restart_num);
-
-  /* Encode the MCU data blocks */
-  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
-    block = MCU_data[blkn];
-
-    /* We simply emit the Al'th bit of the DC coefficient value. */
-    temp = (*block)[0];
-    emit_bits(entropy, (unsigned int) (temp >> Al), 1);
-  }
-
-  cinfo->dest->next_output_byte = entropy->next_output_byte;
-  cinfo->dest->free_in_buffer = entropy->free_in_buffer;
-
-  /* Update restart-interval state too */
-  if (cinfo->restart_interval) {
-    if (entropy->restarts_to_go == 0) {
-      entropy->restarts_to_go = cinfo->restart_interval;
-      entropy->next_restart_num++;
-      entropy->next_restart_num &= 7;
-    }
-    entropy->restarts_to_go--;
-  }
-
-  return TRUE;
-}
-
-
-/*
- * MCU encoding for AC successive approximation refinement scan.
- */
-
-METHODDEF(boolean)
-encode_mcu_AC_refine (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
-{
-  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
-  register int temp;
-  register int r, k;
-  int EOB;
-  char *BR_buffer;
-  unsigned int BR;
-  int Se = cinfo->Se;
-  int Al = cinfo->Al;
-  JBLOCKROW block;
-  int absvalues[DCTSIZE2];
-
-  entropy->next_output_byte = cinfo->dest->next_output_byte;
-  entropy->free_in_buffer = cinfo->dest->free_in_buffer;
-
-  /* Emit restart marker if needed */
-  if (cinfo->restart_interval)
-    if (entropy->restarts_to_go == 0)
-      emit_restart(entropy, entropy->next_restart_num);
-
-  /* Encode the MCU data block */
-  block = MCU_data[0];
-
-  /* It is convenient to make a pre-pass to determine the transformed
-   * coefficients' absolute values and the EOB position.
-   */
-  EOB = 0;
-  for (k = cinfo->Ss; k <= Se; k++) {
-    temp = (*block)[jpeg_natural_order[k]];
-    /* We must apply the point transform by Al.  For AC coefficients this
-     * is an integer division with rounding towards 0.  To do this portably
-     * in C, we shift after obtaining the absolute value.
-     */
-    if (temp < 0)
-      temp = -temp;		/* temp is abs value of input */
-    temp >>= Al;		/* apply the point transform */
-    absvalues[k] = temp;	/* save abs value for main pass */
-    if (temp == 1)
-      EOB = k;			/* EOB = index of last newly-nonzero coef */
-  }
-
-  /* Encode the AC coefficients per section G.1.2.3, fig. G.7 */
-  
-  r = 0;			/* r = run length of zeros */
-  BR = 0;			/* BR = count of buffered bits added now */
-  BR_buffer = entropy->bit_buffer + entropy->BE; /* Append bits to buffer */
-
-  for (k = cinfo->Ss; k <= Se; k++) {
-    if ((temp = absvalues[k]) == 0) {
-      r++;
-      continue;
-    }
-
-    /* Emit any required ZRLs, but not if they can be folded into EOB */
-    while (r > 15 && k <= EOB) {
-      /* emit any pending EOBRUN and the BE correction bits */
-      emit_eobrun(entropy);
-      /* Emit ZRL */
-      emit_symbol(entropy, entropy->ac_tbl_no, 0xF0);
-      r -= 16;
-      /* Emit buffered correction bits that must be associated with ZRL */
-      emit_buffered_bits(entropy, BR_buffer, BR);
-      BR_buffer = entropy->bit_buffer; /* BE bits are gone now */
-      BR = 0;
-    }
-
-    /* If the coef was previously nonzero, it only needs a correction bit.
-     * NOTE: a straight translation of the spec's figure G.7 would suggest
-     * that we also need to test r > 15.  But if r > 15, we can only get here
-     * if k > EOB, which implies that this coefficient is not 1.
-     */
-    if (temp > 1) {
-      /* The correction bit is the next bit of the absolute value. */
-      BR_buffer[BR++] = (char) (temp & 1);
-      continue;
-    }
-
-    /* Emit any pending EOBRUN and the BE correction bits */
-    emit_eobrun(entropy);
-
-    /* Count/emit Huffman symbol for run length / number of bits */
-    emit_symbol(entropy, entropy->ac_tbl_no, (r << 4) + 1);
-
-    /* Emit output bit for newly-nonzero coef */
-    temp = ((*block)[jpeg_natural_order[k]] < 0) ? 0 : 1;
-    emit_bits(entropy, (unsigned int) temp, 1);
-
-    /* Emit buffered correction bits that must be associated with this code */
-    emit_buffered_bits(entropy, BR_buffer, BR);
-    BR_buffer = entropy->bit_buffer; /* BE bits are gone now */
-    BR = 0;
-    r = 0;			/* reset zero run length */
-  }
-
-  if (r > 0 || BR > 0) {	/* If there are trailing zeroes, */
-    entropy->EOBRUN++;		/* count an EOB */
-    entropy->BE += BR;		/* concat my correction bits to older ones */
-    /* We force out the EOB if we risk either:
-     * 1. overflow of the EOB counter;
-     * 2. overflow of the correction bit buffer during the next MCU.
-     */
-    if (entropy->EOBRUN == 0x7FFF || entropy->BE > (MAX_CORR_BITS-DCTSIZE2+1))
-      emit_eobrun(entropy);
-  }
-
-  cinfo->dest->next_output_byte = entropy->next_output_byte;
-  cinfo->dest->free_in_buffer = entropy->free_in_buffer;
-
-  /* Update restart-interval state too */
-  if (cinfo->restart_interval) {
-    if (entropy->restarts_to_go == 0) {
-      entropy->restarts_to_go = cinfo->restart_interval;
-      entropy->next_restart_num++;
-      entropy->next_restart_num &= 7;
-    }
-    entropy->restarts_to_go--;
-  }
-
-  return TRUE;
-}
-
-
-/*
- * Finish up at the end of a Huffman-compressed progressive scan.
- */
-
-METHODDEF(void)
-finish_pass_phuff (j_compress_ptr cinfo)
-{   
-  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
-
-  entropy->next_output_byte = cinfo->dest->next_output_byte;
-  entropy->free_in_buffer = cinfo->dest->free_in_buffer;
-
-  /* Flush out any buffered data */
-  emit_eobrun(entropy);
-  flush_bits(entropy);
-
-  cinfo->dest->next_output_byte = entropy->next_output_byte;
-  cinfo->dest->free_in_buffer = entropy->free_in_buffer;
-}
-
-
-/*
- * Finish up a statistics-gathering pass and create the new Huffman tables.
- */
-
-METHODDEF(void)
-finish_pass_gather_phuff (j_compress_ptr cinfo)
-{
-  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
-  boolean is_DC_band;
-  int ci, tbl;
-  jpeg_component_info * compptr;
-  JHUFF_TBL **htblptr;
-  boolean did[NUM_HUFF_TBLS];
-
-  /* Flush out buffered data (all we care about is counting the EOB symbol) */
-  emit_eobrun(entropy);
-
-  is_DC_band = (cinfo->Ss == 0);
-
-  /* It's important not to apply jpeg_gen_optimal_table more than once
-   * per table, because it clobbers the input frequency counts!
-   */
-  MEMZERO(did, SIZEOF(did));
-
-  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
-    compptr = cinfo->cur_comp_info[ci];
-    if (is_DC_band) {
-      if (cinfo->Ah != 0)	/* DC refinement needs no table */
-	continue;
-      tbl = compptr->dc_tbl_no;
-    } else {
-      tbl = compptr->ac_tbl_no;
-    }
-    if (! did[tbl]) {
-      if (is_DC_band)
-        htblptr = & cinfo->dc_huff_tbl_ptrs[tbl];
-      else
-        htblptr = & cinfo->ac_huff_tbl_ptrs[tbl];
-      if (*htblptr == NULL)
-        *htblptr = jpeg_alloc_huff_table((j_common_ptr) cinfo);
-      jpeg_gen_optimal_table(cinfo, *htblptr, entropy->count_ptrs[tbl]);
-      did[tbl] = TRUE;
-    }
-  }
-}
-
-
-/*
- * Module initialization routine for progressive Huffman entropy encoding.
- */
-
-GLOBAL(void)
-jinit_phuff_encoder (j_compress_ptr cinfo)
-{
-  phuff_entropy_ptr entropy;
-  int i;
-
-  entropy = (phuff_entropy_ptr)
-    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				SIZEOF(phuff_entropy_encoder));
-  cinfo->entropy = (struct jpeg_entropy_encoder *) entropy;
-  entropy->pub.start_pass = start_pass_phuff;
-
-  /* Mark tables unallocated */
-  for (i = 0; i < NUM_HUFF_TBLS; i++) {
-    entropy->derived_tbls[i] = NULL;
-    entropy->count_ptrs[i] = NULL;
-  }
-  entropy->bit_buffer = NULL;	/* needed only in AC refinement scan */
-}
-
-#endif /* C_PROGRESSIVE_SUPPORTED */
diff --git a/src/jpeg-6b/jdct.h b/src/jpeg-6b/jdct.h
deleted file mode 100644
index 04192a26..00000000
--- a/src/jpeg-6b/jdct.h
+++ /dev/null
@@ -1,176 +0,0 @@
-/*
- * jdct.h
- *
- * Copyright (C) 1994-1996, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This include file contains common declarations for the forward and
- * inverse DCT modules.  These declarations are private to the DCT managers
- * (jcdctmgr.c, jddctmgr.c) and the individual DCT algorithms.
- * The individual DCT algorithms are kept in separate files to ease 
- * machine-dependent tuning (e.g., assembly coding).
- */
-
-
-/*
- * A forward DCT routine is given a pointer to a work area of type DCTELEM[];
- * the DCT is to be performed in-place in that buffer.  Type DCTELEM is int
- * for 8-bit samples, INT32 for 12-bit samples.  (NOTE: Floating-point DCT
- * implementations use an array of type FAST_FLOAT, instead.)
- * The DCT inputs are expected to be signed (range +-CENTERJSAMPLE).
- * The DCT outputs are returned scaled up by a factor of 8; they therefore
- * have a range of +-8K for 8-bit data, +-128K for 12-bit data.  This
- * convention improves accuracy in integer implementations and saves some
- * work in floating-point ones.
- * Quantization of the output coefficients is done by jcdctmgr.c.
- */
-
-#if BITS_IN_JSAMPLE == 8
-typedef int DCTELEM;		/* 16 or 32 bits is fine */
-#else
-typedef INT32 DCTELEM;		/* must have 32 bits */
-#endif
-
-typedef JMETHOD(void, forward_DCT_method_ptr, (DCTELEM * data));
-typedef JMETHOD(void, float_DCT_method_ptr, (FAST_FLOAT * data));
-
-
-/*
- * An inverse DCT routine is given a pointer to the input JBLOCK and a pointer
- * to an output sample array.  The routine must dequantize the input data as
- * well as perform the IDCT; for dequantization, it uses the multiplier table
- * pointed to by compptr->dct_table.  The output data is to be placed into the
- * sample array starting at a specified column.  (Any row offset needed will
- * be applied to the array pointer before it is passed to the IDCT code.)
- * Note that the number of samples emitted by the IDCT routine is
- * DCT_scaled_size * DCT_scaled_size.
- */
-
-/* typedef inverse_DCT_method_ptr is declared in jpegint.h */
-
-/*
- * Each IDCT routine has its own ideas about the best dct_table element type.
- */
-
-typedef MULTIPLIER ISLOW_MULT_TYPE; /* short or int, whichever is faster */
-#if BITS_IN_JSAMPLE == 8
-typedef MULTIPLIER IFAST_MULT_TYPE; /* 16 bits is OK, use short if faster */
-#define IFAST_SCALE_BITS  2	/* fractional bits in scale factors */
-#else
-typedef INT32 IFAST_MULT_TYPE;	/* need 32 bits for scaled quantizers */
-#define IFAST_SCALE_BITS  13	/* fractional bits in scale factors */
-#endif
-typedef FAST_FLOAT FLOAT_MULT_TYPE; /* preferred floating type */
-
-
-/*
- * Each IDCT routine is responsible for range-limiting its results and
- * converting them to unsigned form (0..MAXJSAMPLE).  The raw outputs could
- * be quite far out of range if the input data is corrupt, so a bulletproof
- * range-limiting step is required.  We use a mask-and-table-lookup method
- * to do the combined operations quickly.  See the comments with
- * prepare_range_limit_table (in jdmaster.c) for more info.
- */
-
-#define IDCT_range_limit(cinfo)  ((cinfo)->sample_range_limit + CENTERJSAMPLE)
-
-#define RANGE_MASK  (MAXJSAMPLE * 4 + 3) /* 2 bits wider than legal samples */
-
-
-/* Short forms of external names for systems with brain-damaged linkers. */
-
-#ifdef NEED_SHORT_EXTERNAL_NAMES
-#define jpeg_fdct_islow		jFDislow
-#define jpeg_fdct_ifast		jFDifast
-#define jpeg_fdct_float		jFDfloat
-#define jpeg_idct_islow		jRDislow
-#define jpeg_idct_ifast		jRDifast
-#define jpeg_idct_float		jRDfloat
-#define jpeg_idct_4x4		jRD4x4
-#define jpeg_idct_2x2		jRD2x2
-#define jpeg_idct_1x1		jRD1x1
-#endif /* NEED_SHORT_EXTERNAL_NAMES */
-
-/* Extern declarations for the forward and inverse DCT routines. */
-
-EXTERN(void) jpeg_fdct_islow JPP((DCTELEM * data));
-EXTERN(void) jpeg_fdct_ifast JPP((DCTELEM * data));
-EXTERN(void) jpeg_fdct_float JPP((FAST_FLOAT * data));
-
-EXTERN(void) jpeg_idct_islow
-    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
-	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
-EXTERN(void) jpeg_idct_ifast
-    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
-	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
-EXTERN(void) jpeg_idct_float
-    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
-	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
-EXTERN(void) jpeg_idct_4x4
-    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
-	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
-EXTERN(void) jpeg_idct_2x2
-    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
-	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
-EXTERN(void) jpeg_idct_1x1
-    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
-	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
-
-
-/*
- * Macros for handling fixed-point arithmetic; these are used by many
- * but not all of the DCT/IDCT modules.
- *
- * All values are expected to be of type INT32.
- * Fractional constants are scaled left by CONST_BITS bits.
- * CONST_BITS is defined within each module using these macros,
- * and may differ from one module to the next.
- */
-
-#define ONE	((INT32) 1)
-#define CONST_SCALE (ONE << CONST_BITS)
-
-/* Convert a positive real constant to an integer scaled by CONST_SCALE.
- * Caution: some C compilers fail to reduce "FIX(constant)" at compile time,
- * thus causing a lot of useless floating-point operations at run time.
- */
-
-#define FIX(x)	((INT32) ((x) * CONST_SCALE + 0.5))
-
-/* Descale and correctly round an INT32 value that's scaled by N bits.
- * We assume RIGHT_SHIFT rounds towards minus infinity, so adding
- * the fudge factor is correct for either sign of X.
- */
-
-#define DESCALE(x,n)  RIGHT_SHIFT((x) + (ONE << ((n)-1)), n)
-
-/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
- * This macro is used only when the two inputs will actually be no more than
- * 16 bits wide, so that a 16x16->32 bit multiply can be used instead of a
- * full 32x32 multiply.  This provides a useful speedup on many machines.
- * Unfortunately there is no way to specify a 16x16->32 multiply portably
- * in C, but some C compilers will do the right thing if you provide the
- * correct combination of casts.
- */
-
-#ifdef SHORTxSHORT_32		/* may work if 'int' is 32 bits */
-#define MULTIPLY16C16(var,const)  (((INT16) (var)) * ((INT16) (const)))
-#endif
-#ifdef SHORTxLCONST_32		/* known to work with Microsoft C 6.0 */
-#define MULTIPLY16C16(var,const)  (((INT16) (var)) * ((INT32) (const)))
-#endif
-
-#ifndef MULTIPLY16C16		/* default definition */
-#define MULTIPLY16C16(var,const)  ((var) * (const))
-#endif
-
-/* Same except both inputs are variables. */
-
-#ifdef SHORTxSHORT_32		/* may work if 'int' is 32 bits */
-#define MULTIPLY16V16(var1,var2)  (((INT16) (var1)) * ((INT16) (var2)))
-#endif
-
-#ifndef MULTIPLY16V16		/* default definition */
-#define MULTIPLY16V16(var1,var2)  ((var1) * (var2))
-#endif
diff --git a/src/jpeg-6b/jdhuff.c b/src/jpeg-6b/jdhuff.c
deleted file mode 100644
index b5ba39f7..00000000
--- a/src/jpeg-6b/jdhuff.c
+++ /dev/null
@@ -1,651 +0,0 @@
-/*
- * jdhuff.c
- *
- * Copyright (C) 1991-1997, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This file contains Huffman entropy decoding routines.
- *
- * Much of the complexity here has to do with supporting input suspension.
- * If the data source module demands suspension, we want to be able to back
- * up to the start of the current MCU.  To do this, we copy state variables
- * into local working storage, and update them back to the permanent
- * storage only upon successful completion of an MCU.
- */
-
-#define JPEG_INTERNALS
-#include "jinclude.h"
-#include "jpeglib.h"
-#include "jdhuff.h"		/* Declarations shared with jdphuff.c */
-
-
-/*
- * Expanded entropy decoder object for Huffman decoding.
- *
- * The savable_state subrecord contains fields that change within an MCU,
- * but must not be updated permanently until we complete the MCU.
- */
-
-typedef struct {
-  int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
-} savable_state;
-
-/* This macro is to work around compilers with missing or broken
- * structure assignment.  You'll need to fix this code if you have
- * such a compiler and you change MAX_COMPS_IN_SCAN.
- */
-
-#ifndef NO_STRUCT_ASSIGN
-#define ASSIGN_STATE(dest,src)  ((dest) = (src))
-#else
-#if MAX_COMPS_IN_SCAN == 4
-#define ASSIGN_STATE(dest,src)  \
-	((dest).last_dc_val[0] = (src).last_dc_val[0], \
-	 (dest).last_dc_val[1] = (src).last_dc_val[1], \
-	 (dest).last_dc_val[2] = (src).last_dc_val[2], \
-	 (dest).last_dc_val[3] = (src).last_dc_val[3])
-#endif
-#endif
-
-
-typedef struct {
-  struct jpeg_entropy_decoder pub; /* public fields */
-
-  /* These fields are loaded into local variables at start of each MCU.
-   * In case of suspension, we exit WITHOUT updating them.
-   */
-  bitread_perm_state bitstate;	/* Bit buffer at start of MCU */
-  savable_state saved;		/* Other state at start of MCU */
-
-  /* These fields are NOT loaded into local working state. */
-  unsigned int restarts_to_go;	/* MCUs left in this restart interval */
-
-  /* Pointers to derived tables (these workspaces have image lifespan) */
-  d_derived_tbl * dc_derived_tbls[NUM_HUFF_TBLS];
-  d_derived_tbl * ac_derived_tbls[NUM_HUFF_TBLS];
-
-  /* Precalculated info set up by start_pass for use in decode_mcu: */
-
-  /* Pointers to derived tables to be used for each block within an MCU */
-  d_derived_tbl * dc_cur_tbls[D_MAX_BLOCKS_IN_MCU];
-  d_derived_tbl * ac_cur_tbls[D_MAX_BLOCKS_IN_MCU];
-  /* Whether we care about the DC and AC coefficient values for each block */
-  boolean dc_needed[D_MAX_BLOCKS_IN_MCU];
-  boolean ac_needed[D_MAX_BLOCKS_IN_MCU];
-} huff_entropy_decoder;
-
-typedef huff_entropy_decoder * huff_entropy_ptr;
-
-
-/*
- * Initialize for a Huffman-compressed scan.
- */
-
-METHODDEF(void)
-start_pass_huff_decoder (j_decompress_ptr cinfo)
-{
-  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
-  int ci, blkn, dctbl, actbl;
-  jpeg_component_info * compptr;
-
-  /* Check that the scan parameters Ss, Se, Ah/Al are OK for sequential JPEG.
-   * This ought to be an error condition, but we make it a warning because
-   * there are some baseline files out there with all zeroes in these bytes.
-   */
-  if (cinfo->Ss != 0 || cinfo->Se != DCTSIZE2-1 ||
-      cinfo->Ah != 0 || cinfo->Al != 0)
-    WARNMS(cinfo, JWRN_NOT_SEQUENTIAL);
-
-  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
-    compptr = cinfo->cur_comp_info[ci];
-    dctbl = compptr->dc_tbl_no;
-    actbl = compptr->ac_tbl_no;
-    /* Compute derived values for Huffman tables */
-    /* We may do this more than once for a table, but it's not expensive */
-    jpeg_make_d_derived_tbl(cinfo, TRUE, dctbl,
-			    & entropy->dc_derived_tbls[dctbl]);
-    jpeg_make_d_derived_tbl(cinfo, FALSE, actbl,
-			    & entropy->ac_derived_tbls[actbl]);
-    /* Initialize DC predictions to 0 */
-    entropy->saved.last_dc_val[ci] = 0;
-  }
-
-  /* Precalculate decoding info for each block in an MCU of this scan */
-  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
-    ci = cinfo->MCU_membership[blkn];
-    compptr = cinfo->cur_comp_info[ci];
-    /* Precalculate which table to use for each block */
-    entropy->dc_cur_tbls[blkn] = entropy->dc_derived_tbls[compptr->dc_tbl_no];
-    entropy->ac_cur_tbls[blkn] = entropy->ac_derived_tbls[compptr->ac_tbl_no];
-    /* Decide whether we really care about the coefficient values */
-    if (compptr->component_needed) {
-      entropy->dc_needed[blkn] = TRUE;
-      /* we don't need the ACs if producing a 1/8th-size image */
-      entropy->ac_needed[blkn] = (compptr->DCT_scaled_size > 1);
-    } else {
-      entropy->dc_needed[blkn] = entropy->ac_needed[blkn] = FALSE;
-    }
-  }
-
-  /* Initialize bitread state variables */
-  entropy->bitstate.bits_left = 0;
-  entropy->bitstate.get_buffer = 0; /* unnecessary, but keeps Purify quiet */
-  entropy->pub.insufficient_data = FALSE;
-
-  /* Initialize restart counter */
-  entropy->restarts_to_go = cinfo->restart_interval;
-}
-
-
-/*
- * Compute the derived values for a Huffman table.
- * This routine also performs some validation checks on the table.
- *
- * Note this is also used by jdphuff.c.
- */
-
-GLOBAL(void)
-jpeg_make_d_derived_tbl (j_decompress_ptr cinfo, boolean isDC, int tblno,
-			 d_derived_tbl ** pdtbl)
-{
-  JHUFF_TBL *htbl;
-  d_derived_tbl *dtbl;
-  int p, i, l, si, numsymbols;
-  int lookbits, ctr;
-  char huffsize[257];
-  unsigned int huffcode[257];
-  unsigned int code;
-
-  /* Note that huffsize[] and huffcode[] are filled in code-length order,
-   * paralleling the order of the symbols themselves in htbl->huffval[].
-   */
-
-  /* Find the input Huffman table */
-  if (tblno < 0 || tblno >= NUM_HUFF_TBLS)
-    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
-  htbl =
-    isDC ? cinfo->dc_huff_tbl_ptrs[tblno] : cinfo->ac_huff_tbl_ptrs[tblno];
-  if (htbl == NULL)
-    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
-
-  /* Allocate a workspace if we haven't already done so. */
-  if (*pdtbl == NULL)
-    *pdtbl = (d_derived_tbl *)
-      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				  SIZEOF(d_derived_tbl));
-  dtbl = *pdtbl;
-  dtbl->pub = htbl;		/* fill in back link */
-  
-  /* Figure C.1: make table of Huffman code length for each symbol */
-
-  p = 0;
-  for (l = 1; l <= 16; l++) {
-    i = (int) htbl->bits[l];
-    if (i < 0 || p + i > 256)	/* protect against table overrun */
-      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
-    while (i--)
-      huffsize[p++] = (char) l;
-  }
-  huffsize[p] = 0;
-  numsymbols = p;
-  
-  /* Figure C.2: generate the codes themselves */
-  /* We also validate that the counts represent a legal Huffman code tree. */
-  
-  code = 0;
-  si = huffsize[0];
-  p = 0;
-  while (huffsize[p]) {
-    while (((int) huffsize[p]) == si) {
-      huffcode[p++] = code;
-      code++;
-    }
-    /* code is now 1 more than the last code used for codelength si; but
-     * it must still fit in si bits, since no code is allowed to be all ones.
-     */
-    if (((INT32) code) >= (((INT32) 1) << si))
-      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
-    code <<= 1;
-    si++;
-  }
-
-  /* Figure F.15: generate decoding tables for bit-sequential decoding */
-
-  p = 0;
-  for (l = 1; l <= 16; l++) {
-    if (htbl->bits[l]) {
-      /* valoffset[l] = huffval[] index of 1st symbol of code length l,
-       * minus the minimum code of length l
-       */
-      dtbl->valoffset[l] = (INT32) p - (INT32) huffcode[p];
-      p += htbl->bits[l];
-      dtbl->maxcode[l] = huffcode[p-1]; /* maximum code of length l */
-    } else {
-      dtbl->maxcode[l] = -1;	/* -1 if no codes of this length */
-    }
-  }
-  dtbl->maxcode[17] = 0xFFFFFL; /* ensures jpeg_huff_decode terminates */
-
-  /* Compute lookahead tables to speed up decoding.
-   * First we set all the table entries to 0, indicating "too long";
-   * then we iterate through the Huffman codes that are short enough and
-   * fill in all the entries that correspond to bit sequences starting
-   * with that code.
-   */
-
-  MEMZERO(dtbl->look_nbits, SIZEOF(dtbl->look_nbits));
-
-  p = 0;
-  for (l = 1; l <= HUFF_LOOKAHEAD; l++) {
-    for (i = 1; i <= (int) htbl->bits[l]; i++, p++) {
-      /* l = current code's length, p = its index in huffcode[] & huffval[]. */
-      /* Generate left-justified code followed by all possible bit sequences */
-      lookbits = huffcode[p] << (HUFF_LOOKAHEAD-l);
-      for (ctr = 1 << (HUFF_LOOKAHEAD-l); ctr > 0; ctr--) {
-	dtbl->look_nbits[lookbits] = l;
-	dtbl->look_sym[lookbits] = htbl->huffval[p];
-	lookbits++;
-      }
-    }
-  }
-
-  /* Validate symbols as being reasonable.
-   * For AC tables, we make no check, but accept all byte values 0..255.
-   * For DC tables, we require the symbols to be in range 0..15.
-   * (Tighter bounds could be applied depending on the data depth and mode,
-   * but this is sufficient to ensure safe decoding.)
-   */
-  if (isDC) {
-    for (i = 0; i < numsymbols; i++) {
-      int sym = htbl->huffval[i];
-      if (sym < 0 || sym > 15)
-	ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
-    }
-  }
-}
-
-
-/*
- * Out-of-line code for bit fetching (shared with jdphuff.c).
- * See jdhuff.h for info about usage.
- * Note: current values of get_buffer and bits_left are passed as parameters,
- * but are returned in the corresponding fields of the state struct.
- *
- * On most machines MIN_GET_BITS should be 25 to allow the full 32-bit width
- * of get_buffer to be used.  (On machines with wider words, an even larger
- * buffer could be used.)  However, on some machines 32-bit shifts are
- * quite slow and take time proportional to the number of places shifted.
- * (This is true with most PC compilers, for instance.)  In this case it may
- * be a win to set MIN_GET_BITS to the minimum value of 15.  This reduces the
- * average shift distance at the cost of more calls to jpeg_fill_bit_buffer.
- */
-
-#ifdef SLOW_SHIFT_32
-#define MIN_GET_BITS  15	/* minimum allowable value */
-#else
-#define MIN_GET_BITS  (BIT_BUF_SIZE-7)
-#endif
-
-
-GLOBAL(boolean)
-jpeg_fill_bit_buffer (bitread_working_state * state,
-		      register bit_buf_type get_buffer, register int bits_left,
-		      int nbits)
-/* Load up the bit buffer to a depth of at least nbits */
-{
-  /* Copy heavily used state fields into locals (hopefully registers) */
-  register const JOCTET * next_input_byte = state->next_input_byte;
-  register size_t bytes_in_buffer = state->bytes_in_buffer;
-  j_decompress_ptr cinfo = state->cinfo;
-
-  /* Attempt to load at least MIN_GET_BITS bits into get_buffer. */
-  /* (It is assumed that no request will be for more than that many bits.) */
-  /* We fail to do so only if we hit a marker or are forced to suspend. */
-
-  if (cinfo->unread_marker == 0) {	/* cannot advance past a marker */
-    while (bits_left < MIN_GET_BITS) {
-      register int c;
-
-      /* Attempt to read a byte */
-      if (bytes_in_buffer == 0) {
-	if (! (*cinfo->src->fill_input_buffer) (cinfo))
-	  return FALSE;
-	next_input_byte = cinfo->src->next_input_byte;
-	bytes_in_buffer = cinfo->src->bytes_in_buffer;
-      }
-      bytes_in_buffer--;
-      c = GETJOCTET(*next_input_byte++);
-
-      /* If it's 0xFF, check and discard stuffed zero byte */
-      if (c == 0xFF) {
-	/* Loop here to discard any padding FF's on terminating marker,
-	 * so that we can save a valid unread_marker value.  NOTE: we will
-	 * accept multiple FF's followed by a 0 as meaning a single FF data
-	 * byte.  This data pattern is not valid according to the standard.
-	 */
-	do {
-	  if (bytes_in_buffer == 0) {
-	    if (! (*cinfo->src->fill_input_buffer) (cinfo))
-	      return FALSE;
-	    next_input_byte = cinfo->src->next_input_byte;
-	    bytes_in_buffer = cinfo->src->bytes_in_buffer;
-	  }
-	  bytes_in_buffer--;
-	  c = GETJOCTET(*next_input_byte++);
-	} while (c == 0xFF);
-
-	if (c == 0) {
-	  /* Found FF/00, which represents an FF data byte */
-	  c = 0xFF;
-	} else {
-	  /* Oops, it's actually a marker indicating end of compressed data.
-	   * Save the marker code for later use.
-	   * Fine point: it might appear that we should save the marker into
-	   * bitread working state, not straight into permanent state.  But
-	   * once we have hit a marker, we cannot need to suspend within the
-	   * current MCU, because we will read no more bytes from the data
-	   * source.  So it is OK to update permanent state right away.
-	   */
-	  cinfo->unread_marker = c;
-	  /* See if we need to insert some fake zero bits. */
-	  goto no_more_bytes;
-	}
-      }
-
-      /* OK, load c into get_buffer */
-      get_buffer = (get_buffer << 8) | c;
-      bits_left += 8;
-    } /* end while */
-  } else {
-  no_more_bytes:
-    /* We get here if we've read the marker that terminates the compressed
-     * data segment.  There should be enough bits in the buffer register
-     * to satisfy the request; if so, no problem.
-     */
-    if (nbits > bits_left) {
-      /* Uh-oh.  Report corrupted data to user and stuff zeroes into
-       * the data stream, so that we can produce some kind of image.
-       * We use a nonvolatile flag to ensure that only one warning message
-       * appears per data segment.
-       */
-      if (! cinfo->entropy->insufficient_data) {
-	WARNMS(cinfo, JWRN_HIT_MARKER);
-	cinfo->entropy->insufficient_data = TRUE;
-      }
-      /* Fill the buffer with zero bits */
-      get_buffer <<= MIN_GET_BITS - bits_left;
-      bits_left = MIN_GET_BITS;
-    }
-  }
-
-  /* Unload the local registers */
-  state->next_input_byte = next_input_byte;
-  state->bytes_in_buffer = bytes_in_buffer;
-  state->get_buffer = get_buffer;
-  state->bits_left = bits_left;
-
-  return TRUE;
-}
-
-
-/*
- * Out-of-line code for Huffman code decoding.
- * See jdhuff.h for info about usage.
- */
-
-GLOBAL(int)
-jpeg_huff_decode (bitread_working_state * state,
-		  register bit_buf_type get_buffer, register int bits_left,
-		  d_derived_tbl * htbl, int min_bits)
-{
-  register int l = min_bits;
-  register INT32 code;
-
-  /* HUFF_DECODE has determined that the code is at least min_bits */
-  /* bits long, so fetch that many bits in one swoop. */
-
-  CHECK_BIT_BUFFER(*state, l, return -1);
-  code = GET_BITS(l);
-
-  /* Collect the rest of the Huffman code one bit at a time. */
-  /* This is per Figure F.16 in the JPEG spec. */
-
-  while (code > htbl->maxcode[l]) {
-    code <<= 1;
-    CHECK_BIT_BUFFER(*state, 1, return -1);
-    code |= GET_BITS(1);
-    l++;
-  }
-
-  /* Unload the local registers */
-  state->get_buffer = get_buffer;
-  state->bits_left = bits_left;
-
-  /* With garbage input we may reach the sentinel value l = 17. */
-
-  if (l > 16) {
-    WARNMS(state->cinfo, JWRN_HUFF_BAD_CODE);
-    return 0;			/* fake a zero as the safest result */
-  }
-
-  return htbl->pub->huffval[ (int) (code + htbl->valoffset[l]) ];
-}
-
-
-/*
- * Figure F.12: extend sign bit.
- * On some machines, a shift and add will be faster than a table lookup.
- */
-
-#ifdef AVOID_TABLES
-
-#define HUFF_EXTEND(x,s)  ((x) < (1<<((s)-1)) ? (x) + (((-1)<<(s)) + 1) : (x))
-
-#else
-
-#define HUFF_EXTEND(x,s)  ((x) < extend_test[s] ? (x) + extend_offset[s] : (x))
-
-static const int extend_test[16] =   /* entry n is 2**(n-1) */
-  { 0, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080,
-    0x0100, 0x0200, 0x0400, 0x0800, 0x1000, 0x2000, 0x4000 };
-
-static const int extend_offset[16] = /* entry n is (-1 << n) + 1 */
-  { 0, ((-1)<<1) + 1, ((-1)<<2) + 1, ((-1)<<3) + 1, ((-1)<<4) + 1,
-    ((-1)<<5) + 1, ((-1)<<6) + 1, ((-1)<<7) + 1, ((-1)<<8) + 1,
-    ((-1)<<9) + 1, ((-1)<<10) + 1, ((-1)<<11) + 1, ((-1)<<12) + 1,
-    ((-1)<<13) + 1, ((-1)<<14) + 1, ((-1)<<15) + 1 };
-
-#endif /* AVOID_TABLES */
-
-
-/*
- * Check for a restart marker & resynchronize decoder.
- * Returns FALSE if must suspend.
- */
-
-LOCAL(boolean)
-process_restart (j_decompress_ptr cinfo)
-{
-  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
-  int ci;
-
-  /* Throw away any unused bits remaining in bit buffer; */
-  /* include any full bytes in next_marker's count of discarded bytes */
-  cinfo->marker->discarded_bytes += entropy->bitstate.bits_left / 8;
-  entropy->bitstate.bits_left = 0;
-
-  /* Advance past the RSTn marker */
-  if (! (*cinfo->marker->read_restart_marker) (cinfo))
-    return FALSE;
-
-  /* Re-initialize DC predictions to 0 */
-  for (ci = 0; ci < cinfo->comps_in_scan; ci++)
-    entropy->saved.last_dc_val[ci] = 0;
-
-  /* Reset restart counter */
-  entropy->restarts_to_go = cinfo->restart_interval;
-
-  /* Reset out-of-data flag, unless read_restart_marker left us smack up
-   * against a marker.  In that case we will end up treating the next data
-   * segment as empty, and we can avoid producing bogus output pixels by
-   * leaving the flag set.
-   */
-  if (cinfo->unread_marker == 0)
-    entropy->pub.insufficient_data = FALSE;
-
-  return TRUE;
-}
-
-
-/*
- * Decode and return one MCU's worth of Huffman-compressed coefficients.
- * The coefficients are reordered from zigzag order into natural array order,
- * but are not dequantized.
- *
- * The i'th block of the MCU is stored into the block pointed to by
- * MCU_data[i].  WE ASSUME THIS AREA HAS BEEN ZEROED BY THE CALLER.
- * (Wholesale zeroing is usually a little faster than retail...)
- *
- * Returns FALSE if data source requested suspension.  In that case no
- * changes have been made to permanent state.  (Exception: some output
- * coefficients may already have been assigned.  This is harmless for
- * this module, since we'll just re-assign them on the next call.)
- */
-
-METHODDEF(boolean)
-decode_mcu (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
-{
-  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
-  int blkn;
-  BITREAD_STATE_VARS;
-  savable_state state;
-
-  /* Process restart marker if needed; may have to suspend */
-  if (cinfo->restart_interval) {
-    if (entropy->restarts_to_go == 0)
-      if (! process_restart(cinfo))
-	return FALSE;
-  }
-
-  /* If we've run out of data, just leave the MCU set to zeroes.
-   * This way, we return uniform gray for the remainder of the segment.
-   */
-  if (! entropy->pub.insufficient_data) {
-
-    /* Load up working state */
-    BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
-    ASSIGN_STATE(state, entropy->saved);
-
-    /* Outer loop handles each block in the MCU */
-
-    for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
-      JBLOCKROW block = MCU_data[blkn];
-      d_derived_tbl * dctbl = entropy->dc_cur_tbls[blkn];
-      d_derived_tbl * actbl = entropy->ac_cur_tbls[blkn];
-      register int s, k, r;
-
-      /* Decode a single block's worth of coefficients */
-
-      /* Section F.2.2.1: decode the DC coefficient difference */
-      HUFF_DECODE(s, br_state, dctbl, return FALSE, label1);
-      if (s) {
-	CHECK_BIT_BUFFER(br_state, s, return FALSE);
-	r = GET_BITS(s);
-	s = HUFF_EXTEND(r, s);
-      }
-
-      if (entropy->dc_needed[blkn]) {
-	/* Convert DC difference to actual value, update last_dc_val */
-	int ci = cinfo->MCU_membership[blkn];
-	s += state.last_dc_val[ci];
-	state.last_dc_val[ci] = s;
-	/* Output the DC coefficient (assumes jpeg_natural_order[0] = 0) */
-	(*block)[0] = (JCOEF) s;
-      }
-
-      if (entropy->ac_needed[blkn]) {
-
-	/* Section F.2.2.2: decode the AC coefficients */
-	/* Since zeroes are skipped, output area must be cleared beforehand */
-	for (k = 1; k < DCTSIZE2; k++) {
-	  HUFF_DECODE(s, br_state, actbl, return FALSE, label2);
-      
-	  r = s >> 4;
-	  s &= 15;
-      
-	  if (s) {
-	    k += r;
-	    CHECK_BIT_BUFFER(br_state, s, return FALSE);
-	    r = GET_BITS(s);
-	    s = HUFF_EXTEND(r, s);
-	    /* Output coefficient in natural (dezigzagged) order.
-	     * Note: the extra entries in jpeg_natural_order[] will save us
-	     * if k >= DCTSIZE2, which could happen if the data is corrupted.
-	     */
-	    (*block)[jpeg_natural_order[k]] = (JCOEF) s;
-	  } else {
-	    if (r != 15)
-	      break;
-	    k += 15;
-	  }
-	}
-
-      } else {
-
-	/* Section F.2.2.2: decode the AC coefficients */
-	/* In this path we just discard the values */
-	for (k = 1; k < DCTSIZE2; k++) {
-	  HUFF_DECODE(s, br_state, actbl, return FALSE, label3);
-      
-	  r = s >> 4;
-	  s &= 15;
-      
-	  if (s) {
-	    k += r;
-	    CHECK_BIT_BUFFER(br_state, s, return FALSE);
-	    DROP_BITS(s);
-	  } else {
-	    if (r != 15)
-	      break;
-	    k += 15;
-	  }
-	}
-
-      }
-    }
-
-    /* Completed MCU, so update state */
-    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
-    ASSIGN_STATE(entropy->saved, state);
-  }
-
-  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
-
-  return TRUE;
-}
-
-
-/*
- * Module initialization routine for Huffman entropy decoding.
- */
-
-GLOBAL(void)
-jinit_huff_decoder (j_decompress_ptr cinfo)
-{
-  huff_entropy_ptr entropy;
-  int i;
-
-  entropy = (huff_entropy_ptr)
-    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				SIZEOF(huff_entropy_decoder));
-  cinfo->entropy = (struct jpeg_entropy_decoder *) entropy;
-  entropy->pub.start_pass = start_pass_huff_decoder;
-  entropy->pub.decode_mcu = decode_mcu;
-
-  /* Mark tables unallocated */
-  for (i = 0; i < NUM_HUFF_TBLS; i++) {
-    entropy->dc_derived_tbls[i] = entropy->ac_derived_tbls[i] = NULL;
-  }
-}
diff --git a/src/jpeg-6b/jdhuff.h b/src/jpeg-6b/jdhuff.h
deleted file mode 100644
index ae19b6ca..00000000
--- a/src/jpeg-6b/jdhuff.h
+++ /dev/null
@@ -1,201 +0,0 @@
-/*
- * jdhuff.h
- *
- * Copyright (C) 1991-1997, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This file contains declarations for Huffman entropy decoding routines
- * that are shared between the sequential decoder (jdhuff.c) and the
- * progressive decoder (jdphuff.c).  No other modules need to see these.
- */
-
-/* Short forms of external names for systems with brain-damaged linkers. */
-
-#ifdef NEED_SHORT_EXTERNAL_NAMES
-#define jpeg_make_d_derived_tbl	jMkDDerived
-#define jpeg_fill_bit_buffer	jFilBitBuf
-#define jpeg_huff_decode	jHufDecode
-#endif /* NEED_SHORT_EXTERNAL_NAMES */
-
-
-/* Derived data constructed for each Huffman table */
-
-#define HUFF_LOOKAHEAD	8	/* # of bits of lookahead */
-
-typedef struct {
-  /* Basic tables: (element [0] of each array is unused) */
-  INT32 maxcode[18];		/* largest code of length k (-1 if none) */
-  /* (maxcode[17] is a sentinel to ensure jpeg_huff_decode terminates) */
-  INT32 valoffset[17];		/* huffval[] offset for codes of length k */
-  /* valoffset[k] = huffval[] index of 1st symbol of code length k, less
-   * the smallest code of length k; so given a code of length k, the
-   * corresponding symbol is huffval[code + valoffset[k]]
-   */
-
-  /* Link to public Huffman table (needed only in jpeg_huff_decode) */
-  JHUFF_TBL *pub;
-
-  /* Lookahead tables: indexed by the next HUFF_LOOKAHEAD bits of
-   * the input data stream.  If the next Huffman code is no more
-   * than HUFF_LOOKAHEAD bits long, we can obtain its length and
-   * the corresponding symbol directly from these tables.
-   */
-  int look_nbits[1<<HUFF_LOOKAHEAD]; /* # bits, or 0 if too long */
-  UINT8 look_sym[1<<HUFF_LOOKAHEAD]; /* symbol, or unused */
-} d_derived_tbl;
-
-/* Expand a Huffman table definition into the derived format */
-EXTERN(void) jpeg_make_d_derived_tbl
-	JPP((j_decompress_ptr cinfo, boolean isDC, int tblno,
-	     d_derived_tbl ** pdtbl));
-
-
-/*
- * Fetching the next N bits from the input stream is a time-critical operation
- * for the Huffman decoders.  We implement it with a combination of inline
- * macros and out-of-line subroutines.  Note that N (the number of bits
- * demanded at one time) never exceeds 15 for JPEG use.
- *
- * We read source bytes into get_buffer and dole out bits as needed.
- * If get_buffer already contains enough bits, they are fetched in-line
- * by the macros CHECK_BIT_BUFFER and GET_BITS.  When there aren't enough
- * bits, jpeg_fill_bit_buffer is called; it will attempt to fill get_buffer
- * as full as possible (not just to the number of bits needed; this
- * prefetching reduces the overhead cost of calling jpeg_fill_bit_buffer).
- * Note that jpeg_fill_bit_buffer may return FALSE to indicate suspension.
- * On TRUE return, jpeg_fill_bit_buffer guarantees that get_buffer contains
- * at least the requested number of bits --- dummy zeroes are inserted if
- * necessary.
- */
-
-typedef INT32 bit_buf_type;	/* type of bit-extraction buffer */
-#define BIT_BUF_SIZE  32	/* size of buffer in bits */
-
-/* If long is > 32 bits on your machine, and shifting/masking longs is
- * reasonably fast, making bit_buf_type be long and setting BIT_BUF_SIZE
- * appropriately should be a win.  Unfortunately we can't define the size
- * with something like  #define BIT_BUF_SIZE (sizeof(bit_buf_type)*8)
- * because not all machines measure sizeof in 8-bit bytes.
- */
-
-typedef struct {		/* Bitreading state saved across MCUs */
-  bit_buf_type get_buffer;	/* current bit-extraction buffer */
-  int bits_left;		/* # of unused bits in it */
-} bitread_perm_state;
-
-typedef struct {		/* Bitreading working state within an MCU */
-  /* Current data source location */
-  /* We need a copy, rather than munging the original, in case of suspension */
-  const JOCTET * next_input_byte; /* => next byte to read from source */
-  size_t bytes_in_buffer;	/* # of bytes remaining in source buffer */
-  /* Bit input buffer --- note these values are kept in register variables,
-   * not in this struct, inside the inner loops.
-   */
-  bit_buf_type get_buffer;	/* current bit-extraction buffer */
-  int bits_left;		/* # of unused bits in it */
-  /* Pointer needed by jpeg_fill_bit_buffer. */
-  j_decompress_ptr cinfo;	/* back link to decompress master record */
-} bitread_working_state;
-
-/* Macros to declare and load/save bitread local variables. */
-#define BITREAD_STATE_VARS  \
-	register bit_buf_type get_buffer;  \
-	register int bits_left;  \
-	bitread_working_state br_state
-
-#define BITREAD_LOAD_STATE(cinfop,permstate)  \
-	br_state.cinfo = cinfop; \
-	br_state.next_input_byte = cinfop->src->next_input_byte; \
-	br_state.bytes_in_buffer = cinfop->src->bytes_in_buffer; \
-	get_buffer = permstate.get_buffer; \
-	bits_left = permstate.bits_left;
-
-#define BITREAD_SAVE_STATE(cinfop,permstate)  \
-	cinfop->src->next_input_byte = br_state.next_input_byte; \
-	cinfop->src->bytes_in_buffer = br_state.bytes_in_buffer; \
-	permstate.get_buffer = get_buffer; \
-	permstate.bits_left = bits_left
-
-/*
- * These macros provide the in-line portion of bit fetching.
- * Use CHECK_BIT_BUFFER to ensure there are N bits in get_buffer
- * before using GET_BITS, PEEK_BITS, or DROP_BITS.
- * The variables get_buffer and bits_left are assumed to be locals,
- * but the state struct might not be (jpeg_huff_decode needs this).
- *	CHECK_BIT_BUFFER(state,n,action);
- *		Ensure there are N bits in get_buffer; if suspend, take action.
- *      val = GET_BITS(n);
- *		Fetch next N bits.
- *      val = PEEK_BITS(n);
- *		Fetch next N bits without removing them from the buffer.
- *	DROP_BITS(n);
- *		Discard next N bits.
- * The value N should be a simple variable, not an expression, because it
- * is evaluated multiple times.
- */
-
-#define CHECK_BIT_BUFFER(state,nbits,action) \
-	{ if (bits_left < (nbits)) {  \
-	    if (! jpeg_fill_bit_buffer(&(state),get_buffer,bits_left,nbits))  \
-	      { action; }  \
-	    get_buffer = (state).get_buffer; bits_left = (state).bits_left; } }
-
-#define GET_BITS(nbits) \
-	(((int) (get_buffer >> (bits_left -= (nbits)))) & ((1<<(nbits))-1))
-
-#define PEEK_BITS(nbits) \
-	(((int) (get_buffer >> (bits_left -  (nbits)))) & ((1<<(nbits))-1))
-
-#define DROP_BITS(nbits) \
-	(bits_left -= (nbits))
-
-/* Load up the bit buffer to a depth of at least nbits */
-EXTERN(boolean) jpeg_fill_bit_buffer
-	JPP((bitread_working_state * state, register bit_buf_type get_buffer,
-	     register int bits_left, int nbits));
-
-
-/*
- * Code for extracting next Huffman-coded symbol from input bit stream.
- * Again, this is time-critical and we make the main paths be macros.
- *
- * We use a lookahead table to process codes of up to HUFF_LOOKAHEAD bits
- * without looping.  Usually, more than 95% of the Huffman codes will be 8
- * or fewer bits long.  The few overlength codes are handled with a loop,
- * which need not be inline code.
- *
- * Notes about the HUFF_DECODE macro:
- * 1. Near the end of the data segment, we may fail to get enough bits
- *    for a lookahead.  In that case, we do it the hard way.
- * 2. If the lookahead table contains no entry, the next code must be
- *    more than HUFF_LOOKAHEAD bits long.
- * 3. jpeg_huff_decode returns -1 if forced to suspend.
- */
-
-#define HUFF_DECODE(result,state,htbl,failaction,slowlabel) \
-{ register int nb, look; \
-  if (bits_left < HUFF_LOOKAHEAD) { \
-    if (! jpeg_fill_bit_buffer(&state,get_buffer,bits_left, 0)) {failaction;} \
-    get_buffer = state.get_buffer; bits_left = state.bits_left; \
-    if (bits_left < HUFF_LOOKAHEAD) { \
-      nb = 1; goto slowlabel; \
-    } \
-  } \
-  look = PEEK_BITS(HUFF_LOOKAHEAD); \
-  if ((nb = htbl->look_nbits[look]) != 0) { \
-    DROP_BITS(nb); \
-    result = htbl->look_sym[look]; \
-  } else { \
-    nb = HUFF_LOOKAHEAD+1; \
-slowlabel: \
-    if ((result=jpeg_huff_decode(&state,get_buffer,bits_left,htbl,nb)) < 0) \
-	{ failaction; } \
-    get_buffer = state.get_buffer; bits_left = state.bits_left; \
-  } \
-}
-
-/* Out-of-line case for Huffman code fetching */
-EXTERN(int) jpeg_huff_decode
-	JPP((bitread_working_state * state, register bit_buf_type get_buffer,
-	     register int bits_left, d_derived_tbl * htbl, int min_bits));
diff --git a/src/jpeg-6b/jdinput.c b/src/jpeg-6b/jdinput.c
deleted file mode 100644
index 0c2ac8f1..00000000
--- a/src/jpeg-6b/jdinput.c
+++ /dev/null
@@ -1,381 +0,0 @@
-/*
- * jdinput.c
- *
- * Copyright (C) 1991-1997, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This file contains input control logic for the JPEG decompressor.
- * These routines are concerned with controlling the decompressor's input
- * processing (marker reading and coefficient decoding).  The actual input
- * reading is done in jdmarker.c, jdhuff.c, and jdphuff.c.
- */
-
-#define JPEG_INTERNALS
-#include "jinclude.h"
-#include "jpeglib.h"
-
-
-/* Private state */
-
-typedef struct {
-  struct jpeg_input_controller pub; /* public fields */
-
-  boolean inheaders;		/* TRUE until first SOS is reached */
-} my_input_controller;
-
-typedef my_input_controller * my_inputctl_ptr;
-
-
-/* Forward declarations */
-METHODDEF(int) consume_markers JPP((j_decompress_ptr cinfo));
-
-
-/*
- * Routines to calculate various quantities related to the size of the image.
- */
-
-LOCAL(void)
-initial_setup (j_decompress_ptr cinfo)
-/* Called once, when first SOS marker is reached */
-{
-  int ci;
-  jpeg_component_info *compptr;
-
-  /* Make sure image isn't bigger than I can handle */
-  if ((long) cinfo->image_height > (long) JPEG_MAX_DIMENSION ||
-      (long) cinfo->image_width > (long) JPEG_MAX_DIMENSION)
-    ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int) JPEG_MAX_DIMENSION);
-
-  /* For now, precision must match compiled-in value... */
-  if (cinfo->data_precision != BITS_IN_JSAMPLE)
-    ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);
-
-  /* Check that number of components won't exceed internal array sizes */
-  if (cinfo->num_components > MAX_COMPONENTS)
-    ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->num_components,
-	     MAX_COMPONENTS);
-
-  /* Compute maximum sampling factors; check factor validity */
-  cinfo->max_h_samp_factor = 1;
-  cinfo->max_v_samp_factor = 1;
-  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
-       ci++, compptr++) {
-    if (compptr->h_samp_factor<=0 || compptr->h_samp_factor>MAX_SAMP_FACTOR ||
-	compptr->v_samp_factor<=0 || compptr->v_samp_factor>MAX_SAMP_FACTOR)
-      ERREXIT(cinfo, JERR_BAD_SAMPLING);
-    cinfo->max_h_samp_factor = MAX(cinfo->max_h_samp_factor,
-				   compptr->h_samp_factor);
-    cinfo->max_v_samp_factor = MAX(cinfo->max_v_samp_factor,
-				   compptr->v_samp_factor);
-  }
-
-  /* We initialize DCT_scaled_size and min_DCT_scaled_size to DCTSIZE.
-   * In the full decompressor, this will be overridden by jdmaster.c;
-   * but in the transcoder, jdmaster.c is not used, so we must do it here.
-   */
-  cinfo->min_DCT_scaled_size = DCTSIZE;
-
-  /* Compute dimensions of components */
-  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
-       ci++, compptr++) {
-    compptr->DCT_scaled_size = DCTSIZE;
-    /* Size in DCT blocks */
-    compptr->width_in_blocks = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * (long) compptr->h_samp_factor,
-		    (long) (cinfo->max_h_samp_factor * DCTSIZE));
-    compptr->height_in_blocks = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * (long) compptr->v_samp_factor,
-		    (long) (cinfo->max_v_samp_factor * DCTSIZE));
-    /* downsampled_width and downsampled_height will also be overridden by
-     * jdmaster.c if we are doing full decompression.  The transcoder library
-     * doesn't use these values, but the calling application might.
-     */
-    /* Size in samples */
-    compptr->downsampled_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * (long) compptr->h_samp_factor,
-		    (long) cinfo->max_h_samp_factor);
-    compptr->downsampled_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * (long) compptr->v_samp_factor,
-		    (long) cinfo->max_v_samp_factor);
-    /* Mark component needed, until color conversion says otherwise */
-    compptr->component_needed = TRUE;
-    /* Mark no quantization table yet saved for component */
-    compptr->quant_table = NULL;
-  }
-
-  /* Compute number of fully interleaved MCU rows. */
-  cinfo->total_iMCU_rows = (JDIMENSION)
-    jdiv_round_up((long) cinfo->image_height,
-		  (long) (cinfo->max_v_samp_factor*DCTSIZE));
-
-  /* Decide whether file contains multiple scans */
-  if (cinfo->comps_in_scan < cinfo->num_components || cinfo->progressive_mode)
-    cinfo->inputctl->has_multiple_scans = TRUE;
-  else
-    cinfo->inputctl->has_multiple_scans = FALSE;
-}
-
-
-LOCAL(void)
-per_scan_setup (j_decompress_ptr cinfo)
-/* Do computations that are needed before processing a JPEG scan */
-/* cinfo->comps_in_scan and cinfo->cur_comp_info[] were set from SOS marker */
-{
-  int ci, mcublks, tmp;
-  jpeg_component_info *compptr;
-  
-  if (cinfo->comps_in_scan == 1) {
-    
-    /* Noninterleaved (single-component) scan */
-    compptr = cinfo->cur_comp_info[0];
-    
-    /* Overall image size in MCUs */
-    cinfo->MCUs_per_row = compptr->width_in_blocks;
-    cinfo->MCU_rows_in_scan = compptr->height_in_blocks;
-    
-    /* For noninterleaved scan, always one block per MCU */
-    compptr->MCU_width = 1;
-    compptr->MCU_height = 1;
-    compptr->MCU_blocks = 1;
-    compptr->MCU_sample_width = compptr->DCT_scaled_size;
-    compptr->last_col_width = 1;
-    /* For noninterleaved scans, it is convenient to define last_row_height
-     * as the number of block rows present in the last iMCU row.
-     */
-    tmp = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
-    if (tmp == 0) tmp = compptr->v_samp_factor;
-    compptr->last_row_height = tmp;
-    
-    /* Prepare array describing MCU composition */
-    cinfo->blocks_in_MCU = 1;
-    cinfo->MCU_membership[0] = 0;
-    
-  } else {
-    
-    /* Interleaved (multi-component) scan */
-    if (cinfo->comps_in_scan <= 0 || cinfo->comps_in_scan > MAX_COMPS_IN_SCAN)
-      ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->comps_in_scan,
-	       MAX_COMPS_IN_SCAN);
-    
-    /* Overall image size in MCUs */
-    cinfo->MCUs_per_row = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width,
-		    (long) (cinfo->max_h_samp_factor*DCTSIZE));
-    cinfo->MCU_rows_in_scan = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height,
-		    (long) (cinfo->max_v_samp_factor*DCTSIZE));
-    
-    cinfo->blocks_in_MCU = 0;
-    
-    for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
-      compptr = cinfo->cur_comp_info[ci];
-      /* Sampling factors give # of blocks of component in each MCU */
-      compptr->MCU_width = compptr->h_samp_factor;
-      compptr->MCU_height = compptr->v_samp_factor;
-      compptr->MCU_blocks = compptr->MCU_width * compptr->MCU_height;
-      compptr->MCU_sample_width = compptr->MCU_width * compptr->DCT_scaled_size;
-      /* Figure number of non-dummy blocks in last MCU column & row */
-      tmp = (int) (compptr->width_in_blocks % compptr->MCU_width);
-      if (tmp == 0) tmp = compptr->MCU_width;
-      compptr->last_col_width = tmp;
-      tmp = (int) (compptr->height_in_blocks % compptr->MCU_height);
-      if (tmp == 0) tmp = compptr->MCU_height;
-      compptr->last_row_height = tmp;
-      /* Prepare array describing MCU composition */
-      mcublks = compptr->MCU_blocks;
-      if (cinfo->blocks_in_MCU + mcublks > D_MAX_BLOCKS_IN_MCU)
-	ERREXIT(cinfo, JERR_BAD_MCU_SIZE);
-      while (mcublks-- > 0) {
-	cinfo->MCU_membership[cinfo->blocks_in_MCU++] = ci;
-      }
-    }
-    
-  }
-}
-
-
-/*
- * Save away a copy of the Q-table referenced by each component present
- * in the current scan, unless already saved during a prior scan.
- *
- * In a multiple-scan JPEG file, the encoder could assign different components
- * the same Q-table slot number, but change table definitions between scans
- * so that each component uses a different Q-table.  (The IJG encoder is not
- * currently capable of doing this, but other encoders might.)  Since we want
- * to be able to dequantize all the components at the end of the file, this
- * means that we have to save away the table actually used for each component.
- * We do this by copying the table at the start of the first scan containing
- * the component.
- * The JPEG spec prohibits the encoder from changing the contents of a Q-table
- * slot between scans of a component using that slot.  If the encoder does so
- * anyway, this decoder will simply use the Q-table values that were current
- * at the start of the first scan for the component.
- *
- * The decompressor output side looks only at the saved quant tables,
- * not at the current Q-table slots.
- */
-
-LOCAL(void)
-latch_quant_tables (j_decompress_ptr cinfo)
-{
-  int ci, qtblno;
-  jpeg_component_info *compptr;
-  JQUANT_TBL * qtbl;
-
-  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
-    compptr = cinfo->cur_comp_info[ci];
-    /* No work if we already saved Q-table for this component */
-    if (compptr->quant_table != NULL)
-      continue;
-    /* Make sure specified quantization table is present */
-    qtblno = compptr->quant_tbl_no;
-    if (qtblno < 0 || qtblno >= NUM_QUANT_TBLS ||
-	cinfo->quant_tbl_ptrs[qtblno] == NULL)
-      ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, qtblno);
-    /* OK, save away the quantization table */
-    qtbl = (JQUANT_TBL *)
-      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				  SIZEOF(JQUANT_TBL));
-    MEMCOPY(qtbl, cinfo->quant_tbl_ptrs[qtblno], SIZEOF(JQUANT_TBL));
-    compptr->quant_table = qtbl;
-  }
-}
-
-
-/*
- * Initialize the input modules to read a scan of compressed data.
- * The first call to this is done by jdmaster.c after initializing
- * the entire decompressor (during jpeg_start_decompress).
- * Subsequent calls come from consume_markers, below.
- */
-
-METHODDEF(void)
-start_input_pass (j_decompress_ptr cinfo)
-{
-  per_scan_setup(cinfo);
-  latch_quant_tables(cinfo);
-  (*cinfo->entropy->start_pass) (cinfo);
-  (*cinfo->coef->start_input_pass) (cinfo);
-  cinfo->inputctl->consume_input = cinfo->coef->consume_data;
-}
-
-
-/*
- * Finish up after inputting a compressed-data scan.
- * This is called by the coefficient controller after it's read all
- * the expected data of the scan.
- */
-
-METHODDEF(void)
-finish_input_pass (j_decompress_ptr cinfo)
-{
-  cinfo->inputctl->consume_input = consume_markers;
-}
-
-
-/*
- * Read JPEG markers before, between, or after compressed-data scans.
- * Change state as necessary when a new scan is reached.
- * Return value is JPEG_SUSPENDED, JPEG_REACHED_SOS, or JPEG_REACHED_EOI.
- *
- * The consume_input method pointer points either here or to the
- * coefficient controller's consume_data routine, depending on whether
- * we are reading a compressed data segment or inter-segment markers.
- */
-
-METHODDEF(int)
-consume_markers (j_decompress_ptr cinfo)
-{
-  my_inputctl_ptr inputctl = (my_inputctl_ptr) cinfo->inputctl;
-  int val;
-
-  if (inputctl->pub.eoi_reached) /* After hitting EOI, read no further */
-    return JPEG_REACHED_EOI;
-
-  val = (*cinfo->marker->read_markers) (cinfo);
-
-  switch (val) {
-  case JPEG_REACHED_SOS:	/* Found SOS */
-    if (inputctl->inheaders) {	/* 1st SOS */
-      initial_setup(cinfo);
-      inputctl->inheaders = FALSE;
-      /* Note: start_input_pass must be called by jdmaster.c
-       * before any more input can be consumed.  jdapimin.c is
-       * responsible for enforcing this sequencing.
-       */
-    } else {			/* 2nd or later SOS marker */
-      if (! inputctl->pub.has_multiple_scans)
-	ERREXIT(cinfo, JERR_EOI_EXPECTED); /* Oops, I wasn't expecting this! */
-      start_input_pass(cinfo);
-    }
-    break;
-  case JPEG_REACHED_EOI:	/* Found EOI */
-    inputctl->pub.eoi_reached = TRUE;
-    if (inputctl->inheaders) {	/* Tables-only datastream, apparently */
-      if (cinfo->marker->saw_SOF)
-	ERREXIT(cinfo, JERR_SOF_NO_SOS);
-    } else {
-      /* Prevent infinite loop in coef ctlr's decompress_data routine
-       * if user set output_scan_number larger than number of scans.
-       */
-      if (cinfo->output_scan_number > cinfo->input_scan_number)
-	cinfo->output_scan_number = cinfo->input_scan_number;
-    }
-    break;
-  case JPEG_SUSPENDED:
-    break;
-  }
-
-  return val;
-}
-
-
-/*
- * Reset state to begin a fresh datastream.
- */
-
-METHODDEF(void)
-reset_input_controller (j_decompress_ptr cinfo)
-{
-  my_inputctl_ptr inputctl = (my_inputctl_ptr) cinfo->inputctl;
-
-  inputctl->pub.consume_input = consume_markers;
-  inputctl->pub.has_multiple_scans = FALSE; /* "unknown" would be better */
-  inputctl->pub.eoi_reached = FALSE;
-  inputctl->inheaders = TRUE;
-  /* Reset other modules */
-  (*cinfo->err->reset_error_mgr) ((j_common_ptr) cinfo);
-  (*cinfo->marker->reset_marker_reader) (cinfo);
-  /* Reset progression state -- would be cleaner if entropy decoder did this */
-  cinfo->coef_bits = NULL;
-}
-
-
-/*
- * Initialize the input controller module.
- * This is called only once, when the decompression object is created.
- */
-
-GLOBAL(void)
-jinit_input_controller (j_decompress_ptr cinfo)
-{
-  my_inputctl_ptr inputctl;
-
-  /* Create subobject in permanent pool */
-  inputctl = (my_inputctl_ptr)
-    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
-				SIZEOF(my_input_controller));
-  cinfo->inputctl = (struct jpeg_input_controller *) inputctl;
-  /* Initialize method pointers */
-  inputctl->pub.consume_input = consume_markers;
-  inputctl->pub.reset_input_controller = reset_input_controller;
-  inputctl->pub.start_input_pass = start_input_pass;
-  inputctl->pub.finish_input_pass = finish_input_pass;
-  /* Initialize state: can't use reset_input_controller since we don't
-   * want to try to reset other modules yet.
-   */
-  inputctl->pub.has_multiple_scans = FALSE; /* "unknown" would be better */
-  inputctl->pub.eoi_reached = FALSE;
-  inputctl->inheaders = TRUE;
-}
diff --git a/src/jpeg-6b/jdphuff.c b/src/jpeg-6b/jdphuff.c
deleted file mode 100644
index 22678099..00000000
--- a/src/jpeg-6b/jdphuff.c
+++ /dev/null
@@ -1,668 +0,0 @@
-/*
- * jdphuff.c
- *
- * Copyright (C) 1995-1997, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This file contains Huffman entropy decoding routines for progressive JPEG.
- *
- * Much of the complexity here has to do with supporting input suspension.
- * If the data source module demands suspension, we want to be able to back
- * up to the start of the current MCU.  To do this, we copy state variables
- * into local working storage, and update them back to the permanent
- * storage only upon successful completion of an MCU.
- */
-
-#define JPEG_INTERNALS
-#include "jinclude.h"
-#include "jpeglib.h"
-#include "jdhuff.h"		/* Declarations shared with jdhuff.c */
-
-
-#ifdef D_PROGRESSIVE_SUPPORTED
-
-/*
- * Expanded entropy decoder object for progressive Huffman decoding.
- *
- * The savable_state subrecord contains fields that change within an MCU,
- * but must not be updated permanently until we complete the MCU.
- */
-
-typedef struct {
-  unsigned int EOBRUN;			/* remaining EOBs in EOBRUN */
-  int last_dc_val[MAX_COMPS_IN_SCAN];	/* last DC coef for each component */
-} savable_state;
-
-/* This macro is to work around compilers with missing or broken
- * structure assignment.  You'll need to fix this code if you have
- * such a compiler and you change MAX_COMPS_IN_SCAN.
- */
-
-#ifndef NO_STRUCT_ASSIGN
-#define ASSIGN_STATE(dest,src)  ((dest) = (src))
-#else
-#if MAX_COMPS_IN_SCAN == 4
-#define ASSIGN_STATE(dest,src)  \
-	((dest).EOBRUN = (src).EOBRUN, \
-	 (dest).last_dc_val[0] = (src).last_dc_val[0], \
-	 (dest).last_dc_val[1] = (src).last_dc_val[1], \
-	 (dest).last_dc_val[2] = (src).last_dc_val[2], \
-	 (dest).last_dc_val[3] = (src).last_dc_val[3])
-#endif
-#endif
-
-
-typedef struct {
-  struct jpeg_entropy_decoder pub; /* public fields */
-
-  /* These fields are loaded into local variables at start of each MCU.
-   * In case of suspension, we exit WITHOUT updating them.
-   */
-  bitread_perm_state bitstate;	/* Bit buffer at start of MCU */
-  savable_state saved;		/* Other state at start of MCU */
-
-  /* These fields are NOT loaded into local working state. */
-  unsigned int restarts_to_go;	/* MCUs left in this restart interval */
-
-  /* Pointers to derived tables (these workspaces have image lifespan) */
-  d_derived_tbl * derived_tbls[NUM_HUFF_TBLS];
-
-  d_derived_tbl * ac_derived_tbl; /* active table during an AC scan */
-} phuff_entropy_decoder;
-
-typedef phuff_entropy_decoder * phuff_entropy_ptr;
-
-/* Forward declarations */
-METHODDEF(boolean) decode_mcu_DC_first JPP((j_decompress_ptr cinfo,
-					    JBLOCKROW *MCU_data));
-METHODDEF(boolean) decode_mcu_AC_first JPP((j_decompress_ptr cinfo,
-					    JBLOCKROW *MCU_data));
-METHODDEF(boolean) decode_mcu_DC_refine JPP((j_decompress_ptr cinfo,
-					     JBLOCKROW *MCU_data));
-METHODDEF(boolean) decode_mcu_AC_refine JPP((j_decompress_ptr cinfo,
-					     JBLOCKROW *MCU_data));
-
-
-/*
- * Initialize for a Huffman-compressed scan.
- */
-
-METHODDEF(void)
-start_pass_phuff_decoder (j_decompress_ptr cinfo)
-{
-  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
-  boolean is_DC_band, bad;
-  int ci, coefi, tbl;
-  int *coef_bit_ptr;
-  jpeg_component_info * compptr;
-
-  is_DC_band = (cinfo->Ss == 0);
-
-  /* Validate scan parameters */
-  bad = FALSE;
-  if (is_DC_band) {
-    if (cinfo->Se != 0)
-      bad = TRUE;
-  } else {
-    /* need not check Ss/Se < 0 since they came from unsigned bytes */
-    if (cinfo->Ss > cinfo->Se || cinfo->Se >= DCTSIZE2)
-      bad = TRUE;
-    /* AC scans may have only one component */
-    if (cinfo->comps_in_scan != 1)
-      bad = TRUE;
-  }
-  if (cinfo->Ah != 0) {
-    /* Successive approximation refinement scan: must have Al = Ah-1. */
-    if (cinfo->Al != cinfo->Ah-1)
-      bad = TRUE;
-  }
-  if (cinfo->Al > 13)		/* need not check for < 0 */
-    bad = TRUE;
-  /* Arguably the maximum Al value should be less than 13 for 8-bit precision,
-   * but the spec doesn't say so, and we try to be liberal about what we
-   * accept.  Note: large Al values could result in out-of-range DC
-   * coefficients during early scans, leading to bizarre displays due to
-   * overflows in the IDCT math.  But we won't crash.
-   */
-  if (bad)
-    ERREXIT4(cinfo, JERR_BAD_PROGRESSION,
-	     cinfo->Ss, cinfo->Se, cinfo->Ah, cinfo->Al);
-  /* Update progression status, and verify that scan order is legal.
-   * Note that inter-scan inconsistencies are treated as warnings
-   * not fatal errors ... not clear if this is right way to behave.
-   */
-  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
-    int cindex = cinfo->cur_comp_info[ci]->component_index;
-    coef_bit_ptr = & cinfo->coef_bits[cindex][0];
-    if (!is_DC_band && coef_bit_ptr[0] < 0) /* AC without prior DC scan */
-      WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, 0);
-    for (coefi = cinfo->Ss; coefi <= cinfo->Se; coefi++) {
-      int expected = (coef_bit_ptr[coefi] < 0) ? 0 : coef_bit_ptr[coefi];
-      if (cinfo->Ah != expected)
-	WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, coefi);
-      coef_bit_ptr[coefi] = cinfo->Al;
-    }
-  }
-
-  /* Select MCU decoding routine */
-  if (cinfo->Ah == 0) {
-    if (is_DC_band)
-      entropy->pub.decode_mcu = decode_mcu_DC_first;
-    else
-      entropy->pub.decode_mcu = decode_mcu_AC_first;
-  } else {
-    if (is_DC_band)
-      entropy->pub.decode_mcu = decode_mcu_DC_refine;
-    else
-      entropy->pub.decode_mcu = decode_mcu_AC_refine;
-  }
-
-  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
-    compptr = cinfo->cur_comp_info[ci];
-    /* Make sure requested tables are present, and compute derived tables.
-     * We may build same derived table more than once, but it's not expensive.
-     */
-    if (is_DC_band) {
-      if (cinfo->Ah == 0) {	/* DC refinement needs no table */
-	tbl = compptr->dc_tbl_no;
-	jpeg_make_d_derived_tbl(cinfo, TRUE, tbl,
-				& entropy->derived_tbls[tbl]);
-      }
-    } else {
-      tbl = compptr->ac_tbl_no;
-      jpeg_make_d_derived_tbl(cinfo, FALSE, tbl,
-			      & entropy->derived_tbls[tbl]);
-      /* remember the single active table */
-      entropy->ac_derived_tbl = entropy->derived_tbls[tbl];
-    }
-    /* Initialize DC predictions to 0 */
-    entropy->saved.last_dc_val[ci] = 0;
-  }
-
-  /* Initialize bitread state variables */
-  entropy->bitstate.bits_left = 0;
-  entropy->bitstate.get_buffer = 0; /* unnecessary, but keeps Purify quiet */
-  entropy->pub.insufficient_data = FALSE;
-
-  /* Initialize private state variables */
-  entropy->saved.EOBRUN = 0;
-
-  /* Initialize restart counter */
-  entropy->restarts_to_go = cinfo->restart_interval;
-}
-
-
-/*
- * Figure F.12: extend sign bit.
- * On some machines, a shift and add will be faster than a table lookup.
- */
-
-#ifdef AVOID_TABLES
-
-#define HUFF_EXTEND(x,s)  ((x) < (1<<((s)-1)) ? (x) + (((-1)<<(s)) + 1) : (x))
-
-#else
-
-#define HUFF_EXTEND(x,s)  ((x) < extend_test[s] ? (x) + extend_offset[s] : (x))
-
-static const int extend_test[16] =   /* entry n is 2**(n-1) */
-  { 0, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080,
-    0x0100, 0x0200, 0x0400, 0x0800, 0x1000, 0x2000, 0x4000 };
-
-static const int extend_offset[16] = /* entry n is (-1 << n) + 1 */
-  { 0, ((-1)<<1) + 1, ((-1)<<2) + 1, ((-1)<<3) + 1, ((-1)<<4) + 1,
-    ((-1)<<5) + 1, ((-1)<<6) + 1, ((-1)<<7) + 1, ((-1)<<8) + 1,
-    ((-1)<<9) + 1, ((-1)<<10) + 1, ((-1)<<11) + 1, ((-1)<<12) + 1,
-    ((-1)<<13) + 1, ((-1)<<14) + 1, ((-1)<<15) + 1 };
-
-#endif /* AVOID_TABLES */
-
-
-/*
- * Check for a restart marker & resynchronize decoder.
- * Returns FALSE if must suspend.
- */
-
-LOCAL(boolean)
-process_restart (j_decompress_ptr cinfo)
-{
-  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
-  int ci;
-
-  /* Throw away any unused bits remaining in bit buffer; */
-  /* include any full bytes in next_marker's count of discarded bytes */
-  cinfo->marker->discarded_bytes += entropy->bitstate.bits_left / 8;
-  entropy->bitstate.bits_left = 0;
-
-  /* Advance past the RSTn marker */
-  if (! (*cinfo->marker->read_restart_marker) (cinfo))
-    return FALSE;
-
-  /* Re-initialize DC predictions to 0 */
-  for (ci = 0; ci < cinfo->comps_in_scan; ci++)
-    entropy->saved.last_dc_val[ci] = 0;
-  /* Re-init EOB run count, too */
-  entropy->saved.EOBRUN = 0;
-
-  /* Reset restart counter */
-  entropy->restarts_to_go = cinfo->restart_interval;
-
-  /* Reset out-of-data flag, unless read_restart_marker left us smack up
-   * against a marker.  In that case we will end up treating the next data
-   * segment as empty, and we can avoid producing bogus output pixels by
-   * leaving the flag set.
-   */
-  if (cinfo->unread_marker == 0)
-    entropy->pub.insufficient_data = FALSE;
-
-  return TRUE;
-}
-
-
-/*
- * Huffman MCU decoding.
- * Each of these routines decodes and returns one MCU's worth of
- * Huffman-compressed coefficients. 
- * The coefficients are reordered from zigzag order into natural array order,
- * but are not dequantized.
- *
- * The i'th block of the MCU is stored into the block pointed to by
- * MCU_data[i].  WE ASSUME THIS AREA IS INITIALLY ZEROED BY THE CALLER.
- *
- * We return FALSE if data source requested suspension.  In that case no
- * changes have been made to permanent state.  (Exception: some output
- * coefficients may already have been assigned.  This is harmless for
- * spectral selection, since we'll just re-assign them on the next call.
- * Successive approximation AC refinement has to be more careful, however.)
- */
-
-/*
- * MCU decoding for DC initial scan (either spectral selection,
- * or first pass of successive approximation).
- */
-
-METHODDEF(boolean)
-decode_mcu_DC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
-{   
-  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
-  int Al = cinfo->Al;
-  register int s, r;
-  int blkn, ci;
-  JBLOCKROW block;
-  BITREAD_STATE_VARS;
-  savable_state state;
-  d_derived_tbl * tbl;
-  jpeg_component_info * compptr;
-
-  /* Process restart marker if needed; may have to suspend */
-  if (cinfo->restart_interval) {
-    if (entropy->restarts_to_go == 0)
-      if (! process_restart(cinfo))
-	return FALSE;
-  }
-
-  /* If we've run out of data, just leave the MCU set to zeroes.
-   * This way, we return uniform gray for the remainder of the segment.
-   */
-  if (! entropy->pub.insufficient_data) {
-
-    /* Load up working state */
-    BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
-    ASSIGN_STATE(state, entropy->saved);
-
-    /* Outer loop handles each block in the MCU */
-
-    for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
-      block = MCU_data[blkn];
-      ci = cinfo->MCU_membership[blkn];
-      compptr = cinfo->cur_comp_info[ci];
-      tbl = entropy->derived_tbls[compptr->dc_tbl_no];
-
-      /* Decode a single block's worth of coefficients */
-
-      /* Section F.2.2.1: decode the DC coefficient difference */
-      HUFF_DECODE(s, br_state, tbl, return FALSE, label1);
-      if (s) {
-	CHECK_BIT_BUFFER(br_state, s, return FALSE);
-	r = GET_BITS(s);
-	s = HUFF_EXTEND(r, s);
-      }
-
-      /* Convert DC difference to actual value, update last_dc_val */
-      s += state.last_dc_val[ci];
-      state.last_dc_val[ci] = s;
-      /* Scale and output the coefficient (assumes jpeg_natural_order[0]=0) */
-      (*block)[0] = (JCOEF) (s << Al);
-    }
-
-    /* Completed MCU, so update state */
-    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
-    ASSIGN_STATE(entropy->saved, state);
-  }
-
-  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
-
-  return TRUE;
-}
-
-
-/*
- * MCU decoding for AC initial scan (either spectral selection,
- * or first pass of successive approximation).
- */
-
-METHODDEF(boolean)
-decode_mcu_AC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
-{   
-  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
-  int Se = cinfo->Se;
-  int Al = cinfo->Al;
-  register int s, k, r;
-  unsigned int EOBRUN;
-  JBLOCKROW block;
-  BITREAD_STATE_VARS;
-  d_derived_tbl * tbl;
-
-  /* Process restart marker if needed; may have to suspend */
-  if (cinfo->restart_interval) {
-    if (entropy->restarts_to_go == 0)
-      if (! process_restart(cinfo))
-	return FALSE;
-  }
-
-  /* If we've run out of data, just leave the MCU set to zeroes.
-   * This way, we return uniform gray for the remainder of the segment.
-   */
-  if (! entropy->pub.insufficient_data) {
-
-    /* Load up working state.
-     * We can avoid loading/saving bitread state if in an EOB run.
-     */
-    EOBRUN = entropy->saved.EOBRUN;	/* only part of saved state we need */
-
-    /* There is always only one block per MCU */
-
-    if (EOBRUN > 0)		/* if it's a band of zeroes... */
-      EOBRUN--;			/* ...process it now (we do nothing) */
-    else {
-      BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
-      block = MCU_data[0];
-      tbl = entropy->ac_derived_tbl;
-
-      for (k = cinfo->Ss; k <= Se; k++) {
-	HUFF_DECODE(s, br_state, tbl, return FALSE, label2);
-	r = s >> 4;
-	s &= 15;
-	if (s) {
-	  k += r;
-	  CHECK_BIT_BUFFER(br_state, s, return FALSE);
-	  r = GET_BITS(s);
-	  s = HUFF_EXTEND(r, s);
-	  /* Scale and output coefficient in natural (dezigzagged) order */
-	  (*block)[jpeg_natural_order[k]] = (JCOEF) (s << Al);
-	} else {
-	  if (r == 15) {	/* ZRL */
-	    k += 15;		/* skip 15 zeroes in band */
-	  } else {		/* EOBr, run length is 2^r + appended bits */
-	    EOBRUN = 1 << r;
-	    if (r) {		/* EOBr, r > 0 */
-	      CHECK_BIT_BUFFER(br_state, r, return FALSE);
-	      r = GET_BITS(r);
-	      EOBRUN += r;
-	    }
-	    EOBRUN--;		/* this band is processed at this moment */
-	    break;		/* force end-of-band */
-	  }
-	}
-      }
-
-      BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
-    }
-
-    /* Completed MCU, so update state */
-    entropy->saved.EOBRUN = EOBRUN;	/* only part of saved state we need */
-  }
-
-  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
-
-  return TRUE;
-}
-
-
-/*
- * MCU decoding for DC successive approximation refinement scan.
- * Note: we assume such scans can be multi-component, although the spec
- * is not very clear on the point.
- */
-
-METHODDEF(boolean)
-decode_mcu_DC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
-{   
-  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
-  int p1 = 1 << cinfo->Al;	/* 1 in the bit position being coded */
-  int blkn;
-  JBLOCKROW block;
-  BITREAD_STATE_VARS;
-
-  /* Process restart marker if needed; may have to suspend */
-  if (cinfo->restart_interval) {
-    if (entropy->restarts_to_go == 0)
-      if (! process_restart(cinfo))
-	return FALSE;
-  }
-
-  /* Not worth the cycles to check insufficient_data here,
-   * since we will not change the data anyway if we read zeroes.
-   */
-
-  /* Load up working state */
-  BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
-
-  /* Outer loop handles each block in the MCU */
-
-  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
-    block = MCU_data[blkn];
-
-    /* Encoded data is simply the next bit of the two's-complement DC value */
-    CHECK_BIT_BUFFER(br_state, 1, return FALSE);
-    if (GET_BITS(1))
-      (*block)[0] |= p1;
-    /* Note: since we use |=, repeating the assignment later is safe */
-  }
-
-  /* Completed MCU, so update state */
-  BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
-
-  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
-
-  return TRUE;
-}
-
-
-/*
- * MCU decoding for AC successive approximation refinement scan.
- */
-
-METHODDEF(boolean)
-decode_mcu_AC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
-{   
-  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
-  int Se = cinfo->Se;
-  int p1 = 1 << cinfo->Al;	/* 1 in the bit position being coded */
-  int m1 = (-1) << cinfo->Al;	/* -1 in the bit position being coded */
-  register int s, k, r;
-  unsigned int EOBRUN;
-  JBLOCKROW block;
-  JCOEFPTR thiscoef;
-  BITREAD_STATE_VARS;
-  d_derived_tbl * tbl;
-  int num_newnz;
-  int newnz_pos[DCTSIZE2];
-
-  /* Process restart marker if needed; may have to suspend */
-  if (cinfo->restart_interval) {
-    if (entropy->restarts_to_go == 0)
-      if (! process_restart(cinfo))
-	return FALSE;
-  }
-
-  /* If we've run out of data, don't modify the MCU.
-   */
-  if (! entropy->pub.insufficient_data) {
-
-    /* Load up working state */
-    BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
-    EOBRUN = entropy->saved.EOBRUN; /* only part of saved state we need */
-
-    /* There is always only one block per MCU */
-    block = MCU_data[0];
-    tbl = entropy->ac_derived_tbl;
-
-    /* If we are forced to suspend, we must undo the assignments to any newly
-     * nonzero coefficients in the block, because otherwise we'd get confused
-     * next time about which coefficients were already nonzero.
-     * But we need not undo addition of bits to already-nonzero coefficients;
-     * instead, we can test the current bit to see if we already did it.
-     */
-    num_newnz = 0;
-
-    /* initialize coefficient loop counter to start of band */
-    k = cinfo->Ss;
-
-    if (EOBRUN == 0) {
-      for (; k <= Se; k++) {
-	HUFF_DECODE(s, br_state, tbl, goto undoit, label3);
-	r = s >> 4;
-	s &= 15;
-	if (s) {
-	  if (s != 1)		/* size of new coef should always be 1 */
-	    WARNMS(cinfo, JWRN_HUFF_BAD_CODE);
-	  CHECK_BIT_BUFFER(br_state, 1, goto undoit);
-	  if (GET_BITS(1))
-	    s = p1;		/* newly nonzero coef is positive */
-	  else
-	    s = m1;		/* newly nonzero coef is negative */
-	} else {
-	  if (r != 15) {
-	    EOBRUN = 1 << r;	/* EOBr, run length is 2^r + appended bits */
-	    if (r) {
-	      CHECK_BIT_BUFFER(br_state, r, goto undoit);
-	      r = GET_BITS(r);
-	      EOBRUN += r;
-	    }
-	    break;		/* rest of block is handled by EOB logic */
-	  }
-	  /* note s = 0 for processing ZRL */
-	}
-	/* Advance over already-nonzero coefs and r still-zero coefs,
-	 * appending correction bits to the nonzeroes.  A correction bit is 1
-	 * if the absolute value of the coefficient must be increased.
-	 */
-	do {
-	  thiscoef = *block + jpeg_natural_order[k];
-	  if (*thiscoef != 0) {
-	    CHECK_BIT_BUFFER(br_state, 1, goto undoit);
-	    if (GET_BITS(1)) {
-	      if ((*thiscoef & p1) == 0) { /* do nothing if already set it */
-		if (*thiscoef >= 0)
-		  *thiscoef += p1;
-		else
-		  *thiscoef += m1;
-	      }
-	    }
-	  } else {
-	    if (--r < 0)
-	      break;		/* reached target zero coefficient */
-	  }
-	  k++;
-	} while (k <= Se);
-	if (s) {
-	  int pos = jpeg_natural_order[k];
-	  /* Output newly nonzero coefficient */
-	  (*block)[pos] = (JCOEF) s;
-	  /* Remember its position in case we have to suspend */
-	  newnz_pos[num_newnz++] = pos;
-	}
-      }
-    }
-
-    if (EOBRUN > 0) {
-      /* Scan any remaining coefficient positions after the end-of-band
-       * (the last newly nonzero coefficient, if any).  Append a correction
-       * bit to each already-nonzero coefficient.  A correction bit is 1
-       * if the absolute value of the coefficient must be increased.
-       */
-      for (; k <= Se; k++) {
-	thiscoef = *block + jpeg_natural_order[k];
-	if (*thiscoef != 0) {
-	  CHECK_BIT_BUFFER(br_state, 1, goto undoit);
-	  if (GET_BITS(1)) {
-	    if ((*thiscoef & p1) == 0) { /* do nothing if already changed it */
-	      if (*thiscoef >= 0)
-		*thiscoef += p1;
-	      else
-		*thiscoef += m1;
-	    }
-	  }
-	}
-      }
-      /* Count one block completed in EOB run */
-      EOBRUN--;
-    }
-
-    /* Completed MCU, so update state */
-    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
-    entropy->saved.EOBRUN = EOBRUN; /* only part of saved state we need */
-  }
-
-  /* Account for restart interval (no-op if not using restarts) */
-  entropy->restarts_to_go--;
-
-  return TRUE;
-
-undoit:
-  /* Re-zero any output coefficients that we made newly nonzero */
-  while (num_newnz > 0)
-    (*block)[newnz_pos[--num_newnz]] = 0;
-
-  return FALSE;
-}
-
-
-/*
- * Module initialization routine for progressive Huffman entropy decoding.
- */
-
-GLOBAL(void)
-jinit_phuff_decoder (j_decompress_ptr cinfo)
-{
-  phuff_entropy_ptr entropy;
-  int *coef_bit_ptr;
-  int ci, i;
-
-  entropy = (phuff_entropy_ptr)
-    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				SIZEOF(phuff_entropy_decoder));
-  cinfo->entropy = (struct jpeg_entropy_decoder *) entropy;
-  entropy->pub.start_pass = start_pass_phuff_decoder;
-
-  /* Mark derived tables unallocated */
-  for (i = 0; i < NUM_HUFF_TBLS; i++) {
-    entropy->derived_tbls[i] = NULL;
-  }
-
-  /* Create progression status table */
-  cinfo->coef_bits = (int (*)[DCTSIZE2])
-    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-				cinfo->num_components*DCTSIZE2*SIZEOF(int));
-  coef_bit_ptr = & cinfo->coef_bits[0][0];
-  for (ci = 0; ci < cinfo->num_components; ci++) 
-    for (i = 0; i < DCTSIZE2; i++)
-      *coef_bit_ptr++ = -1;
-}
-
-#endif /* D_PROGRESSIVE_SUPPORTED */
diff --git a/src/jpeg-6b/jfdctint.c b/src/jpeg-6b/jfdctint.c
deleted file mode 100644
index 0a78b64a..00000000
--- a/src/jpeg-6b/jfdctint.c
+++ /dev/null
@@ -1,283 +0,0 @@
-/*
- * jfdctint.c
- *
- * Copyright (C) 1991-1996, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This file contains a slow-but-accurate integer implementation of the
- * forward DCT (Discrete Cosine Transform).
- *
- * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT
- * on each column.  Direct algorithms are also available, but they are
- * much more complex and seem not to be any faster when reduced to code.
- *
- * This implementation is based on an algorithm described in
- *   C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
- *   Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
- *   Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
- * The primary algorithm described there uses 11 multiplies and 29 adds.
- * We use their alternate method with 12 multiplies and 32 adds.
- * The advantage of this method is that no data path contains more than one
- * multiplication; this allows a very simple and accurate implementation in
- * scaled fixed-point arithmetic, with a minimal number of shifts.
- */
-
-#define JPEG_INTERNALS
-#include "jinclude.h"
-#include "jpeglib.h"
-#include "jdct.h"		/* Private declarations for DCT subsystem */
-
-#ifdef DCT_ISLOW_SUPPORTED
-
-
-/*
- * This module is specialized to the case DCTSIZE = 8.
- */
-
-#if DCTSIZE != 8
-  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
-#endif
-
-
-/*
- * The poop on this scaling stuff is as follows:
- *
- * Each 1-D DCT step produces outputs which are a factor of sqrt(N)
- * larger than the true DCT outputs.  The final outputs are therefore
- * a factor of N larger than desired; since N=8 this can be cured by
- * a simple right shift at the end of the algorithm.  The advantage of
- * this arrangement is that we save two multiplications per 1-D DCT,
- * because the y0 and y4 outputs need not be divided by sqrt(N).
- * In the IJG code, this factor of 8 is removed by the quantization step
- * (in jcdctmgr.c), NOT in this module.
- *
- * We have to do addition and subtraction of the integer inputs, which
- * is no problem, and multiplication by fractional constants, which is
- * a problem to do in integer arithmetic.  We multiply all the constants
- * by CONST_SCALE and convert them to integer constants (thus retaining
- * CONST_BITS bits of precision in the constants).  After doing a
- * multiplication we have to divide the product by CONST_SCALE, with proper
- * rounding, to produce the correct output.  This division can be done
- * cheaply as a right shift of CONST_BITS bits.  We postpone shifting
- * as long as possible so that partial sums can be added together with
- * full fractional precision.
- *
- * The outputs of the first pass are scaled up by PASS1_BITS bits so that
- * they are represented to better-than-integral precision.  These outputs
- * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
- * with the recommended scaling.  (For 12-bit sample data, the intermediate
- * array is INT32 anyway.)
- *
- * To avoid overflow of the 32-bit intermediate results in pass 2, we must
- * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26.  Error analysis
- * shows that the values given below are the most effective.
- */
-
-#if BITS_IN_JSAMPLE == 8
-#define CONST_BITS  13
-#define PASS1_BITS  2
-#else
-#define CONST_BITS  13
-#define PASS1_BITS  1		/* lose a little precision to avoid overflow */
-#endif
-
-/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
- * causing a lot of useless floating-point operations at run time.
- * To get around this we use the following pre-calculated constants.
- * If you change CONST_BITS you may want to add appropriate values.
- * (With a reasonable C compiler, you can just rely on the FIX() macro...)
- */
-
-#if CONST_BITS == 13
-#define FIX_0_298631336  ((INT32)  2446)	/* FIX(0.298631336) */
-#define FIX_0_390180644  ((INT32)  3196)	/* FIX(0.390180644) */
-#define FIX_0_541196100  ((INT32)  4433)	/* FIX(0.541196100) */
-#define FIX_0_765366865  ((INT32)  6270)	/* FIX(0.765366865) */
-#define FIX_0_899976223  ((INT32)  7373)	/* FIX(0.899976223) */
-#define FIX_1_175875602  ((INT32)  9633)	/* FIX(1.175875602) */
-#define FIX_1_501321110  ((INT32)  12299)	/* FIX(1.501321110) */
-#define FIX_1_847759065  ((INT32)  15137)	/* FIX(1.847759065) */
-#define FIX_1_961570560  ((INT32)  16069)	/* FIX(1.961570560) */
-#define FIX_2_053119869  ((INT32)  16819)	/* FIX(2.053119869) */
-#define FIX_2_562915447  ((INT32)  20995)	/* FIX(2.562915447) */
-#define FIX_3_072711026  ((INT32)  25172)	/* FIX(3.072711026) */
-#else
-#define FIX_0_298631336  FIX(0.298631336)
-#define FIX_0_390180644  FIX(0.390180644)
-#define FIX_0_541196100  FIX(0.541196100)
-#define FIX_0_765366865  FIX(0.765366865)
-#define FIX_0_899976223  FIX(0.899976223)
-#define FIX_1_175875602  FIX(1.175875602)
-#define FIX_1_501321110  FIX(1.501321110)
-#define FIX_1_847759065  FIX(1.847759065)
-#define FIX_1_961570560  FIX(1.961570560)
-#define FIX_2_053119869  FIX(2.053119869)
-#define FIX_2_562915447  FIX(2.562915447)
-#define FIX_3_072711026  FIX(3.072711026)
-#endif
-
-
-/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
- * For 8-bit samples with the recommended scaling, all the variable
- * and constant values involved are no more than 16 bits wide, so a
- * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
- * For 12-bit samples, a full 32-bit multiplication will be needed.
- */
-
-#if BITS_IN_JSAMPLE == 8
-#define MULTIPLY(var,const)  MULTIPLY16C16(var,const)
-#else
-#define MULTIPLY(var,const)  ((var) * (const))
-#endif
-
-
-/*
- * Perform the forward DCT on one block of samples.
- */
-
-GLOBAL(void)
-jpeg_fdct_islow (DCTELEM * data)
-{
-  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
-  INT32 tmp10, tmp11, tmp12, tmp13;
-  INT32 z1, z2, z3, z4, z5;
-  DCTELEM *dataptr;
-  int ctr;
-  SHIFT_TEMPS
-
-  /* Pass 1: process rows. */
-  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
-  /* furthermore, we scale the results by 2**PASS1_BITS. */
-
-  dataptr = data;
-  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
-    tmp0 = dataptr[0] + dataptr[7];
-    tmp7 = dataptr[0] - dataptr[7];
-    tmp1 = dataptr[1] + dataptr[6];
-    tmp6 = dataptr[1] - dataptr[6];
-    tmp2 = dataptr[2] + dataptr[5];
-    tmp5 = dataptr[2] - dataptr[5];
-    tmp3 = dataptr[3] + dataptr[4];
-    tmp4 = dataptr[3] - dataptr[4];
-    
-    /* Even part per LL&M figure 1 --- note that published figure is faulty;
-     * rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
-     */
-    
-    tmp10 = tmp0 + tmp3;
-    tmp13 = tmp0 - tmp3;
-    tmp11 = tmp1 + tmp2;
-    tmp12 = tmp1 - tmp2;
-    
-    dataptr[0] = (DCTELEM) ((tmp10 + tmp11) << PASS1_BITS);
-    dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << PASS1_BITS);
-    
-    z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
-    dataptr[2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865),
-				   CONST_BITS-PASS1_BITS);
-    dataptr[6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065),
-				   CONST_BITS-PASS1_BITS);
-    
-    /* Odd part per figure 8 --- note paper omits factor of sqrt(2).
-     * cK represents cos(K*pi/16).
-     * i0..i3 in the paper are tmp4..tmp7 here.
-     */
-    
-    z1 = tmp4 + tmp7;
-    z2 = tmp5 + tmp6;
-    z3 = tmp4 + tmp6;
-    z4 = tmp5 + tmp7;
-    z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
-    
-    tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
-    tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
-    tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
-    tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
-    z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
-    z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
-    z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
-    z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
-    
-    z3 += z5;
-    z4 += z5;
-    
-    dataptr[7] = (DCTELEM) DESCALE(tmp4 + z1 + z3, CONST_BITS-PASS1_BITS);
-    dataptr[5] = (DCTELEM) DESCALE(tmp5 + z2 + z4, CONST_BITS-PASS1_BITS);
-    dataptr[3] = (DCTELEM) DESCALE(tmp6 + z2 + z3, CONST_BITS-PASS1_BITS);
-    dataptr[1] = (DCTELEM) DESCALE(tmp7 + z1 + z4, CONST_BITS-PASS1_BITS);
-    
-    dataptr += DCTSIZE;		/* advance pointer to next row */
-  }
-
-  /* Pass 2: process columns.
-   * We remove the PASS1_BITS scaling, but leave the results scaled up
-   * by an overall factor of 8.
-   */
-
-  dataptr = data;
-  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
-    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
-    tmp7 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
-    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
-    tmp6 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
-    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
-    tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
-    tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
-    tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
-    
-    /* Even part per LL&M figure 1 --- note that published figure is faulty;
-     * rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
-     */
-    
-    tmp10 = tmp0 + tmp3;
-    tmp13 = tmp0 - tmp3;
-    tmp11 = tmp1 + tmp2;
-    tmp12 = tmp1 - tmp2;
-    
-    dataptr[DCTSIZE*0] = (DCTELEM) DESCALE(tmp10 + tmp11, PASS1_BITS);
-    dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp10 - tmp11, PASS1_BITS);
-    
-    z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
-    dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865),
-					   CONST_BITS+PASS1_BITS);
-    dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065),
-					   CONST_BITS+PASS1_BITS);
-    
-    /* Odd part per figure 8 --- note paper omits factor of sqrt(2).
-     * cK represents cos(K*pi/16).
-     * i0..i3 in the paper are tmp4..tmp7 here.
-     */
-    
-    z1 = tmp4 + tmp7;
-    z2 = tmp5 + tmp6;
-    z3 = tmp4 + tmp6;
-    z4 = tmp5 + tmp7;
-    z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
-    
-    tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
-    tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
-    tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
-    tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
-    z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
-    z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
-    z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
-    z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
-    
-    z3 += z5;
-    z4 += z5;
-    
-    dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp4 + z1 + z3,
-					   CONST_BITS+PASS1_BITS);
-    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp5 + z2 + z4,
-					   CONST_BITS+PASS1_BITS);
-    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp6 + z2 + z3,
-					   CONST_BITS+PASS1_BITS);
-    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp7 + z1 + z4,
-					   CONST_BITS+PASS1_BITS);
-    
-    dataptr++;			/* advance pointer to next column */
-  }
-}
-
-#endif /* DCT_ISLOW_SUPPORTED */
diff --git a/src/jpeg-6b/jidctint.c b/src/jpeg-6b/jidctint.c
deleted file mode 100644
index a72b3207..00000000
--- a/src/jpeg-6b/jidctint.c
+++ /dev/null
@@ -1,389 +0,0 @@
-/*
- * jidctint.c
- *
- * Copyright (C) 1991-1998, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This file contains a slow-but-accurate integer implementation of the
- * inverse DCT (Discrete Cosine Transform).  In the IJG code, this routine
- * must also perform dequantization of the input coefficients.
- *
- * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
- * on each row (or vice versa, but it's more convenient to emit a row at
- * a time).  Direct algorithms are also available, but they are much more
- * complex and seem not to be any faster when reduced to code.
- *
- * This implementation is based on an algorithm described in
- *   C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
- *   Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
- *   Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
- * The primary algorithm described there uses 11 multiplies and 29 adds.
- * We use their alternate method with 12 multiplies and 32 adds.
- * The advantage of this method is that no data path contains more than one
- * multiplication; this allows a very simple and accurate implementation in
- * scaled fixed-point arithmetic, with a minimal number of shifts.
- */
-
-#define JPEG_INTERNALS
-#include "jinclude.h"
-#include "jpeglib.h"
-#include "jdct.h"		/* Private declarations for DCT subsystem */
-
-#ifdef DCT_ISLOW_SUPPORTED
-
-
-/*
- * This module is specialized to the case DCTSIZE = 8.
- */
-
-#if DCTSIZE != 8
-  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
-#endif
-
-
-/*
- * The poop on this scaling stuff is as follows:
- *
- * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
- * larger than the true IDCT outputs.  The final outputs are therefore
- * a factor of N larger than desired; since N=8 this can be cured by
- * a simple right shift at the end of the algorithm.  The advantage of
- * this arrangement is that we save two multiplications per 1-D IDCT,
- * because the y0 and y4 inputs need not be divided by sqrt(N).
- *
- * We have to do addition and subtraction of the integer inputs, which
- * is no problem, and multiplication by fractional constants, which is
- * a problem to do in integer arithmetic.  We multiply all the constants
- * by CONST_SCALE and convert them to integer constants (thus retaining
- * CONST_BITS bits of precision in the constants).  After doing a
- * multiplication we have to divide the product by CONST_SCALE, with proper
- * rounding, to produce the correct output.  This division can be done
- * cheaply as a right shift of CONST_BITS bits.  We postpone shifting
- * as long as possible so that partial sums can be added together with
- * full fractional precision.
- *
- * The outputs of the first pass are scaled up by PASS1_BITS bits so that
- * they are represented to better-than-integral precision.  These outputs
- * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
- * with the recommended scaling.  (To scale up 12-bit sample data further, an
- * intermediate INT32 array would be needed.)
- *
- * To avoid overflow of the 32-bit intermediate results in pass 2, we must
- * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26.  Error analysis
- * shows that the values given below are the most effective.
- */
-
-#if BITS_IN_JSAMPLE == 8
-#define CONST_BITS  13
-#define PASS1_BITS  2
-#else
-#define CONST_BITS  13
-#define PASS1_BITS  1		/* lose a little precision to avoid overflow */
-#endif
-
-/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
- * causing a lot of useless floating-point operations at run time.
- * To get around this we use the following pre-calculated constants.
- * If you change CONST_BITS you may want to add appropriate values.
- * (With a reasonable C compiler, you can just rely on the FIX() macro...)
- */
-
-#if CONST_BITS == 13
-#define FIX_0_298631336  ((INT32)  2446)	/* FIX(0.298631336) */
-#define FIX_0_390180644  ((INT32)  3196)	/* FIX(0.390180644) */
-#define FIX_0_541196100  ((INT32)  4433)	/* FIX(0.541196100) */
-#define FIX_0_765366865  ((INT32)  6270)	/* FIX(0.765366865) */
-#define FIX_0_899976223  ((INT32)  7373)	/* FIX(0.899976223) */
-#define FIX_1_175875602  ((INT32)  9633)	/* FIX(1.175875602) */
-#define FIX_1_501321110  ((INT32)  12299)	/* FIX(1.501321110) */
-#define FIX_1_847759065  ((INT32)  15137)	/* FIX(1.847759065) */
-#define FIX_1_961570560  ((INT32)  16069)	/* FIX(1.961570560) */
-#define FIX_2_053119869  ((INT32)  16819)	/* FIX(2.053119869) */
-#define FIX_2_562915447  ((INT32)  20995)	/* FIX(2.562915447) */
-#define FIX_3_072711026  ((INT32)  25172)	/* FIX(3.072711026) */
-#else
-#define FIX_0_298631336  FIX(0.298631336)
-#define FIX_0_390180644  FIX(0.390180644)
-#define FIX_0_541196100  FIX(0.541196100)
-#define FIX_0_765366865  FIX(0.765366865)
-#define FIX_0_899976223  FIX(0.899976223)
-#define FIX_1_175875602  FIX(1.175875602)
-#define FIX_1_501321110  FIX(1.501321110)
-#define FIX_1_847759065  FIX(1.847759065)
-#define FIX_1_961570560  FIX(1.961570560)
-#define FIX_2_053119869  FIX(2.053119869)
-#define FIX_2_562915447  FIX(2.562915447)
-#define FIX_3_072711026  FIX(3.072711026)
-#endif
-
-
-/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
- * For 8-bit samples with the recommended scaling, all the variable
- * and constant values involved are no more than 16 bits wide, so a
- * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
- * For 12-bit samples, a full 32-bit multiplication will be needed.
- */
-
-#if BITS_IN_JSAMPLE == 8
-#define MULTIPLY(var,const)  MULTIPLY16C16(var,const)
-#else
-#define MULTIPLY(var,const)  ((var) * (const))
-#endif
-
-
-/* Dequantize a coefficient by multiplying it by the multiplier-table
- * entry; produce an int result.  In this module, both inputs and result
- * are 16 bits or less, so either int or short multiply will work.
- */
-
-#define DEQUANTIZE(coef,quantval)  (((ISLOW_MULT_TYPE) (coef)) * (quantval))
-
-
-/*
- * Perform dequantization and inverse DCT on one block of coefficients.
- */
-
-GLOBAL(void)
-jpeg_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
-		 JCOEFPTR coef_block,
-		 JSAMPARRAY output_buf, JDIMENSION output_col)
-{
-  INT32 tmp0, tmp1, tmp2, tmp3;
-  INT32 tmp10, tmp11, tmp12, tmp13;
-  INT32 z1, z2, z3, z4, z5;
-  JCOEFPTR inptr;
-  ISLOW_MULT_TYPE * quantptr;
-  int * wsptr;
-  JSAMPROW outptr;
-  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
-  int ctr;
-  int workspace[DCTSIZE2];	/* buffers data between passes */
-  SHIFT_TEMPS
-
-  /* Pass 1: process columns from input, store into work array. */
-  /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
-  /* furthermore, we scale the results by 2**PASS1_BITS. */
-
-  inptr = coef_block;
-  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
-  wsptr = workspace;
-  for (ctr = DCTSIZE; ctr > 0; ctr--) {
-    /* Due to quantization, we will usually find that many of the input
-     * coefficients are zero, especially the AC terms.  We can exploit this
-     * by short-circuiting the IDCT calculation for any column in which all
-     * the AC terms are zero.  In that case each output is equal to the
-     * DC coefficient (with scale factor as needed).
-     * With typical images and quantization tables, half or more of the
-     * column DCT calculations can be simplified this way.
-     */
-    
-    if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
-	inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
-	inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
-	inptr[DCTSIZE*7] == 0) {
-      /* AC terms all zero */
-      int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
-      
-      wsptr[DCTSIZE*0] = dcval;
-      wsptr[DCTSIZE*1] = dcval;
-      wsptr[DCTSIZE*2] = dcval;
-      wsptr[DCTSIZE*3] = dcval;
-      wsptr[DCTSIZE*4] = dcval;
-      wsptr[DCTSIZE*5] = dcval;
-      wsptr[DCTSIZE*6] = dcval;
-      wsptr[DCTSIZE*7] = dcval;
-      
-      inptr++;			/* advance pointers to next column */
-      quantptr++;
-      wsptr++;
-      continue;
-    }
-    
-    /* Even part: reverse the even part of the forward DCT. */
-    /* The rotator is sqrt(2)*c(-6). */
-    
-    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
-    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
-    
-    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
-    tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
-    tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
-    
-    z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
-    z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
-
-    tmp0 = (z2 + z3) << CONST_BITS;
-    tmp1 = (z2 - z3) << CONST_BITS;
-    
-    tmp10 = tmp0 + tmp3;
-    tmp13 = tmp0 - tmp3;
-    tmp11 = tmp1 + tmp2;
-    tmp12 = tmp1 - tmp2;
-    
-    /* Odd part per figure 8; the matrix is unitary and hence its
-     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
-     */
-    
-    tmp0 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
-    tmp1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
-    tmp2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
-    tmp3 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
-    
-    z1 = tmp0 + tmp3;
-    z2 = tmp1 + tmp2;
-    z3 = tmp0 + tmp2;
-    z4 = tmp1 + tmp3;
-    z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
-    
-    tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
-    tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
-    tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
-    tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
-    z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
-    z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
-    z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
-    z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
-    
-    z3 += z5;
-    z4 += z5;
-    
-    tmp0 += z1 + z3;
-    tmp1 += z2 + z4;
-    tmp2 += z2 + z3;
-    tmp3 += z1 + z4;
-    
-    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
-    
-    wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
-    wsptr[DCTSIZE*7] = (int) DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
-    wsptr[DCTSIZE*1] = (int) DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
-    wsptr[DCTSIZE*6] = (int) DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
-    wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
-    wsptr[DCTSIZE*5] = (int) DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
-    wsptr[DCTSIZE*3] = (int) DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
-    wsptr[DCTSIZE*4] = (int) DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
-    
-    inptr++;			/* advance pointers to next column */
-    quantptr++;
-    wsptr++;
-  }
-  
-  /* Pass 2: process rows from work array, store into output array. */
-  /* Note that we must descale the results by a factor of 8 == 2**3, */
-  /* and also undo the PASS1_BITS scaling. */
-
-  wsptr = workspace;
-  for (ctr = 0; ctr < DCTSIZE; ctr++) {
-    outptr = output_buf[ctr] + output_col;
-    /* Rows of zeroes can be exploited in the same way as we did with columns.
-     * However, the column calculation has created many nonzero AC terms, so
-     * the simplification applies less often (typically 5% to 10% of the time).
-     * On machines with very fast multiplication, it's possible that the
-     * test takes more time than it's worth.  In that case this section
-     * may be commented out.
-     */
-    
-#ifndef NO_ZERO_ROW_TEST
-    if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && wsptr[4] == 0 &&
-	wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
-      /* AC terms all zero */
-      JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
-				  & RANGE_MASK];
-      
-      outptr[0] = dcval;
-      outptr[1] = dcval;
-      outptr[2] = dcval;
-      outptr[3] = dcval;
-      outptr[4] = dcval;
-      outptr[5] = dcval;
-      outptr[6] = dcval;
-      outptr[7] = dcval;
-
-      wsptr += DCTSIZE;		/* advance pointer to next row */
-      continue;
-    }
-#endif
-    
-    /* Even part: reverse the even part of the forward DCT. */
-    /* The rotator is sqrt(2)*c(-6). */
-    
-    z2 = (INT32) wsptr[2];
-    z3 = (INT32) wsptr[6];
-    
-    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
-    tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
-    tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
-    
-    tmp0 = ((INT32) wsptr[0] + (INT32) wsptr[4]) << CONST_BITS;
-    tmp1 = ((INT32) wsptr[0] - (INT32) wsptr[4]) << CONST_BITS;
-    
-    tmp10 = tmp0 + tmp3;
-    tmp13 = tmp0 - tmp3;
-    tmp11 = tmp1 + tmp2;
-    tmp12 = tmp1 - tmp2;
-    
-    /* Odd part per figure 8; the matrix is unitary and hence its
-     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
-     */
-    
-    tmp0 = (INT32) wsptr[7];
-    tmp1 = (INT32) wsptr[5];
-    tmp2 = (INT32) wsptr[3];
-    tmp3 = (INT32) wsptr[1];
-    
-    z1 = tmp0 + tmp3;
-    z2 = tmp1 + tmp2;
-    z3 = tmp0 + tmp2;
-    z4 = tmp1 + tmp3;
-    z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
-    
-    tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
-    tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
-    tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
-    tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
-    z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
-    z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
-    z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
-    z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
-    
-    z3 += z5;
-    z4 += z5;
-    
-    tmp0 += z1 + z3;
-    tmp1 += z2 + z4;
-    tmp2 += z2 + z3;
-    tmp3 += z1 + z4;
-    
-    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
-    
-    outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp3,
-					  CONST_BITS+PASS1_BITS+3)
-			    & RANGE_MASK];
-    outptr[7] = range_limit[(int) DESCALE(tmp10 - tmp3,
-					  CONST_BITS+PASS1_BITS+3)
-			    & RANGE_MASK];
-    outptr[1] = range_limit[(int) DESCALE(tmp11 + tmp2,
-					  CONST_BITS+PASS1_BITS+3)
-			    & RANGE_MASK];
-    outptr[6] = range_limit[(int) DESCALE(tmp11 - tmp2,
-					  CONST_BITS+PASS1_BITS+3)
-			    & RANGE_MASK];
-    outptr[2] = range_limit[(int) DESCALE(tmp12 + tmp1,
-					  CONST_BITS+PASS1_BITS+3)
-			    & RANGE_MASK];
-    outptr[5] = range_limit[(int) DESCALE(tmp12 - tmp1,
-					  CONST_BITS+PASS1_BITS+3)
-			    & RANGE_MASK];
-    outptr[3] = range_limit[(int) DESCALE(tmp13 + tmp0,
-					  CONST_BITS+PASS1_BITS+3)
-			    & RANGE_MASK];
-    outptr[4] = range_limit[(int) DESCALE(tmp13 - tmp0,
-					  CONST_BITS+PASS1_BITS+3)
-			    & RANGE_MASK];
-    
-    wsptr += DCTSIZE;		/* advance pointer to next row */
-  }
-}
-
-#endif /* DCT_ISLOW_SUPPORTED */
diff --git a/src/jpeg-6b/jidctred.c b/src/jpeg-6b/jidctred.c
deleted file mode 100644
index 421f3c7c..00000000
--- a/src/jpeg-6b/jidctred.c
+++ /dev/null
@@ -1,398 +0,0 @@
-/*
- * jidctred.c
- *
- * Copyright (C) 1994-1998, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This file contains inverse-DCT routines that produce reduced-size output:
- * either 4x4, 2x2, or 1x1 pixels from an 8x8 DCT block.
- *
- * The implementation is based on the Loeffler, Ligtenberg and Moschytz (LL&M)
- * algorithm used in jidctint.c.  We simply replace each 8-to-8 1-D IDCT step
- * with an 8-to-4 step that produces the four averages of two adjacent outputs
- * (or an 8-to-2 step producing two averages of four outputs, for 2x2 output).
- * These steps were derived by computing the corresponding values at the end
- * of the normal LL&M code, then simplifying as much as possible.
- *
- * 1x1 is trivial: just take the DC coefficient divided by 8.
- *
- * See jidctint.c for additional comments.
- */
-
-#define JPEG_INTERNALS
-#include "jinclude.h"
-#include "jpeglib.h"
-#include "jdct.h"		/* Private declarations for DCT subsystem */
-
-#ifdef IDCT_SCALING_SUPPORTED
-
-
-/*
- * This module is specialized to the case DCTSIZE = 8.
- */
-
-#if DCTSIZE != 8
-  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
-#endif
-
-
-/* Scaling is the same as in jidctint.c. */
-
-#if BITS_IN_JSAMPLE == 8
-#define CONST_BITS  13
-#define PASS1_BITS  2
-#else
-#define CONST_BITS  13
-#define PASS1_BITS  1		/* lose a little precision to avoid overflow */
-#endif
-
-/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
- * causing a lot of useless floating-point operations at run time.
- * To get around this we use the following pre-calculated constants.
- * If you change CONST_BITS you may want to add appropriate values.
- * (With a reasonable C compiler, you can just rely on the FIX() macro...)
- */
-
-#if CONST_BITS == 13
-#define FIX_0_211164243  ((INT32)  1730)	/* FIX(0.211164243) */
-#define FIX_0_509795579  ((INT32)  4176)	/* FIX(0.509795579) */
-#define FIX_0_601344887  ((INT32)  4926)	/* FIX(0.601344887) */
-#define FIX_0_720959822  ((INT32)  5906)	/* FIX(0.720959822) */
-#define FIX_0_765366865  ((INT32)  6270)	/* FIX(0.765366865) */
-#define FIX_0_850430095  ((INT32)  6967)	/* FIX(0.850430095) */
-#define FIX_0_899976223  ((INT32)  7373)	/* FIX(0.899976223) */
-#define FIX_1_061594337  ((INT32)  8697)	/* FIX(1.061594337) */
-#define FIX_1_272758580  ((INT32)  10426)	/* FIX(1.272758580) */
-#define FIX_1_451774981  ((INT32)  11893)	/* FIX(1.451774981) */
-#define FIX_1_847759065  ((INT32)  15137)	/* FIX(1.847759065) */
-#define FIX_2_172734803  ((INT32)  17799)	/* FIX(2.172734803) */
-#define FIX_2_562915447  ((INT32)  20995)	/* FIX(2.562915447) */
-#define FIX_3_624509785  ((INT32)  29692)	/* FIX(3.624509785) */
-#else
-#define FIX_0_211164243  FIX(0.211164243)
-#define FIX_0_509795579  FIX(0.509795579)
-#define FIX_0_601344887  FIX(0.601344887)
-#define FIX_0_720959822  FIX(0.720959822)
-#define FIX_0_765366865  FIX(0.765366865)
-#define FIX_0_850430095  FIX(0.850430095)
-#define FIX_0_899976223  FIX(0.899976223)
-#define FIX_1_061594337  FIX(1.061594337)
-#define FIX_1_272758580  FIX(1.272758580)
-#define FIX_1_451774981  FIX(1.451774981)
-#define FIX_1_847759065  FIX(1.847759065)
-#define FIX_2_172734803  FIX(2.172734803)
-#define FIX_2_562915447  FIX(2.562915447)
-#define FIX_3_624509785  FIX(3.624509785)
-#endif
-
-
-/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
- * For 8-bit samples with the recommended scaling, all the variable
- * and constant values involved are no more than 16 bits wide, so a
- * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
- * For 12-bit samples, a full 32-bit multiplication will be needed.
- */
-
-#if BITS_IN_JSAMPLE == 8
-#define MULTIPLY(var,const)  MULTIPLY16C16(var,const)
-#else
-#define MULTIPLY(var,const)  ((var) * (const))
-#endif
-
-
-/* Dequantize a coefficient by multiplying it by the multiplier-table
- * entry; produce an int result.  In this module, both inputs and result
- * are 16 bits or less, so either int or short multiply will work.
- */
-
-#define DEQUANTIZE(coef,quantval)  (((ISLOW_MULT_TYPE) (coef)) * (quantval))
-
-
-/*
- * Perform dequantization and inverse DCT on one block of coefficients,
- * producing a reduced-size 4x4 output block.
- */
-
-GLOBAL(void)
-jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
-	       JCOEFPTR coef_block,
-	       JSAMPARRAY output_buf, JDIMENSION output_col)
-{
-  INT32 tmp0, tmp2, tmp10, tmp12;
-  INT32 z1, z2, z3, z4;
-  JCOEFPTR inptr;
-  ISLOW_MULT_TYPE * quantptr;
-  int * wsptr;
-  JSAMPROW outptr;
-  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
-  int ctr;
-  int workspace[DCTSIZE*4];	/* buffers data between passes */
-  SHIFT_TEMPS
-
-  /* Pass 1: process columns from input, store into work array. */
-
-  inptr = coef_block;
-  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
-  wsptr = workspace;
-  for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) {
-    /* Don't bother to process column 4, because second pass won't use it */
-    if (ctr == DCTSIZE-4)
-      continue;
-    if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
-	inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*5] == 0 &&
-	inptr[DCTSIZE*6] == 0 && inptr[DCTSIZE*7] == 0) {
-      /* AC terms all zero; we need not examine term 4 for 4x4 output */
-      int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
-      
-      wsptr[DCTSIZE*0] = dcval;
-      wsptr[DCTSIZE*1] = dcval;
-      wsptr[DCTSIZE*2] = dcval;
-      wsptr[DCTSIZE*3] = dcval;
-      
-      continue;
-    }
-    
-    /* Even part */
-    
-    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
-    tmp0 <<= (CONST_BITS+1);
-    
-    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
-    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
-
-    tmp2 = MULTIPLY(z2, FIX_1_847759065) + MULTIPLY(z3, - FIX_0_765366865);
-    
-    tmp10 = tmp0 + tmp2;
-    tmp12 = tmp0 - tmp2;
-    
-    /* Odd part */
-    
-    z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
-    z2 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
-    z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
-    z4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
-    
-    tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */
-	 + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */
-	 + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */
-	 + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */
-    
-    tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */
-	 + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */
-	 + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */
-	 + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */
-
-    /* Final output stage */
-    
-    wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp2, CONST_BITS-PASS1_BITS+1);
-    wsptr[DCTSIZE*3] = (int) DESCALE(tmp10 - tmp2, CONST_BITS-PASS1_BITS+1);
-    wsptr[DCTSIZE*1] = (int) DESCALE(tmp12 + tmp0, CONST_BITS-PASS1_BITS+1);
-    wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 - tmp0, CONST_BITS-PASS1_BITS+1);
-  }
-  
-  /* Pass 2: process 4 rows from work array, store into output array. */
-
-  wsptr = workspace;
-  for (ctr = 0; ctr < 4; ctr++) {
-    outptr = output_buf[ctr] + output_col;
-    /* It's not clear whether a zero row test is worthwhile here ... */
-
-#ifndef NO_ZERO_ROW_TEST
-    if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 &&
-	wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
-      /* AC terms all zero */
-      JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
-				  & RANGE_MASK];
-      
-      outptr[0] = dcval;
-      outptr[1] = dcval;
-      outptr[2] = dcval;
-      outptr[3] = dcval;
-      
-      wsptr += DCTSIZE;		/* advance pointer to next row */
-      continue;
-    }
-#endif
-    
-    /* Even part */
-    
-    tmp0 = ((INT32) wsptr[0]) << (CONST_BITS+1);
-    
-    tmp2 = MULTIPLY((INT32) wsptr[2], FIX_1_847759065)
-	 + MULTIPLY((INT32) wsptr[6], - FIX_0_765366865);
-    
-    tmp10 = tmp0 + tmp2;
-    tmp12 = tmp0 - tmp2;
-    
-    /* Odd part */
-    
-    z1 = (INT32) wsptr[7];
-    z2 = (INT32) wsptr[5];
-    z3 = (INT32) wsptr[3];
-    z4 = (INT32) wsptr[1];
-    
-    tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */
-	 + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */
-	 + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */
-	 + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */
-    
-    tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */
-	 + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */
-	 + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */
-	 + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */
-
-    /* Final output stage */
-    
-    outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp2,
-					  CONST_BITS+PASS1_BITS+3+1)
-			    & RANGE_MASK];
-    outptr[3] = range_limit[(int) DESCALE(tmp10 - tmp2,
-					  CONST_BITS+PASS1_BITS+3+1)
-			    & RANGE_MASK];
-    outptr[1] = range_limit[(int) DESCALE(tmp12 + tmp0,
-					  CONST_BITS+PASS1_BITS+3+1)
-			    & RANGE_MASK];
-    outptr[2] = range_limit[(int) DESCALE(tmp12 - tmp0,
-					  CONST_BITS+PASS1_BITS+3+1)
-			    & RANGE_MASK];
-    
-    wsptr += DCTSIZE;		/* advance pointer to next row */
-  }
-}
-
-
-/*
- * Perform dequantization and inverse DCT on one block of coefficients,
- * producing a reduced-size 2x2 output block.
- */
-
-GLOBAL(void)
-jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
-	       JCOEFPTR coef_block,
-	       JSAMPARRAY output_buf, JDIMENSION output_col)
-{
-  INT32 tmp0, tmp10, z1;
-  JCOEFPTR inptr;
-  ISLOW_MULT_TYPE * quantptr;
-  int * wsptr;
-  JSAMPROW outptr;
-  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
-  int ctr;
-  int workspace[DCTSIZE*2];	/* buffers data between passes */
-  SHIFT_TEMPS
-
-  /* Pass 1: process columns from input, store into work array. */
-
-  inptr = coef_block;
-  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
-  wsptr = workspace;
-  for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) {
-    /* Don't bother to process columns 2,4,6 */
-    if (ctr == DCTSIZE-2 || ctr == DCTSIZE-4 || ctr == DCTSIZE-6)
-      continue;
-    if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*3] == 0 &&
-	inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*7] == 0) {
-      /* AC terms all zero; we need not examine terms 2,4,6 for 2x2 output */
-      int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
-      
-      wsptr[DCTSIZE*0] = dcval;
-      wsptr[DCTSIZE*1] = dcval;
-      
-      continue;
-    }
-    
-    /* Even part */
-    
-    z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
-    tmp10 = z1 << (CONST_BITS+2);
-    
-    /* Odd part */
-
-    z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
-    tmp0 = MULTIPLY(z1, - FIX_0_720959822); /* sqrt(2) * (c7-c5+c3-c1) */
-    z1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
-    tmp0 += MULTIPLY(z1, FIX_0_850430095); /* sqrt(2) * (-c1+c3+c5+c7) */
-    z1 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
-    tmp0 += MULTIPLY(z1, - FIX_1_272758580); /* sqrt(2) * (-c1+c3-c5-c7) */
-    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
-    tmp0 += MULTIPLY(z1, FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */
-
-    /* Final output stage */
-    
-    wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp0, CONST_BITS-PASS1_BITS+2);
-    wsptr[DCTSIZE*1] = (int) DESCALE(tmp10 - tmp0, CONST_BITS-PASS1_BITS+2);
-  }
-  
-  /* Pass 2: process 2 rows from work array, store into output array. */
-
-  wsptr = workspace;
-  for (ctr = 0; ctr < 2; ctr++) {
-    outptr = output_buf[ctr] + output_col;
-    /* It's not clear whether a zero row test is worthwhile here ... */
-
-#ifndef NO_ZERO_ROW_TEST
-    if (wsptr[1] == 0 && wsptr[3] == 0 && wsptr[5] == 0 && wsptr[7] == 0) {
-      /* AC terms all zero */
-      JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
-				  & RANGE_MASK];
-      
-      outptr[0] = dcval;
-      outptr[1] = dcval;
-      
-      wsptr += DCTSIZE;		/* advance pointer to next row */
-      continue;
-    }
-#endif
-    
-    /* Even part */
-    
-    tmp10 = ((INT32) wsptr[0]) << (CONST_BITS+2);
-    
-    /* Odd part */
-
-    tmp0 = MULTIPLY((INT32) wsptr[7], - FIX_0_720959822) /* sqrt(2) * (c7-c5+c3-c1) */
-	 + MULTIPLY((INT32) wsptr[5], FIX_0_850430095) /* sqrt(2) * (-c1+c3+c5+c7) */
-	 + MULTIPLY((INT32) wsptr[3], - FIX_1_272758580) /* sqrt(2) * (-c1+c3-c5-c7) */
-	 + MULTIPLY((INT32) wsptr[1], FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */
-
-    /* Final output stage */
-    
-    outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp0,
-					  CONST_BITS+PASS1_BITS+3+2)
-			    & RANGE_MASK];
-    outptr[1] = range_limit[(int) DESCALE(tmp10 - tmp0,
-					  CONST_BITS+PASS1_BITS+3+2)
-			    & RANGE_MASK];
-    
-    wsptr += DCTSIZE;		/* advance pointer to next row */
-  }
-}
-
-
-/*
- * Perform dequantization and inverse DCT on one block of coefficients,
- * producing a reduced-size 1x1 output block.
- */
-
-GLOBAL(void)
-jpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
-	       JCOEFPTR coef_block,
-	       JSAMPARRAY output_buf, JDIMENSION output_col)
-{
-  int dcval;
-  ISLOW_MULT_TYPE * quantptr;
-  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
-  SHIFT_TEMPS
-
-  /* We hardly need an inverse DCT routine for this: just take the
-   * average pixel value, which is one-eighth of the DC coefficient.
-   */
-  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
-  dcval = DEQUANTIZE(coef_block[0], quantptr[0]);
-  dcval = (int) DESCALE((INT32) dcval, 3);
-
-  output_buf[0][output_col] = range_limit[dcval & RANGE_MASK];
-}
-
-#endif /* IDCT_SCALING_SUPPORTED */
diff --git a/src/jpeg-6b/jload.c b/src/jpeg-6b/jload.c
deleted file mode 100644
index 29e750b3..00000000
--- a/src/jpeg-6b/jload.c
+++ /dev/null
@@ -1,145 +0,0 @@
-
-#include "../qcommon/q_shared.h"
-#include "../qcommon/qcommon.h"
-
-/*
- * Include file for users of JPEG library.
- * You will need to have included system headers that define at least
- * the typedefs FILE and size_t before you can include jpeglib.h.
- * (stdio.h is sufficient on ANSI-conforming systems.)
- * You may also wish to include "jerror.h".
- */
-
-#include "jpeglib.h"
-
-
-int LoadJPG( const char *filename, unsigned char **pic, int *width, int *height ) {
-  /* This struct contains the JPEG decompression parameters and pointers to
-   * working space (which is allocated as needed by the JPEG library).
-   */
-  struct jpeg_decompress_struct cinfo;
-  /* We use our private extension JPEG error handler.
-   * Note that this struct must live as long as the main JPEG parameter
-   * struct, to avoid dangling-pointer problems.
-   */
-  /* This struct represents a JPEG error handler.  It is declared separately
-   * because applications often want to supply a specialized error handler
-   * (see the second half of this file for an example).  But here we just
-   * take the easy way out and use the standard error handler, which will
-   * print a message on stderr and call exit() if compression fails.
-   * Note that this struct must live as long as the main JPEG parameter
-   * struct, to avoid dangling-pointer problems.
-   */
-  struct jpeg_error_mgr jerr;
-  /* More stuff */
-  fileHandle_t infile;		/* source file */
-  JSAMPARRAY buffer;		/* Output row buffer */
-  int row_stride;		/* physical row width in output buffer */
-  unsigned char *out;
-
-  /* In this example we want to open the input file before doing anything else,
-   * so that the setjmp() error recovery below can assume the file is open.
-   * VERY IMPORTANT: use "b" option to fopen() if you are on a machine that
-   * requires it in order to read binary files.
-   */
-
-  FS_FOpenFileRead( filename, &infile, qfalse );
-  if (infile == 0) {
-    return 0;
-  }
-
-  /* Step 1: allocate and initialize JPEG decompression object */
-
-  /* We have to set up the error handler first, in case the initialization
-   * step fails.  (Unlikely, but it could happen if you are out of memory.)
-   * This routine fills in the contents of struct jerr, and returns jerr's
-   * address which we place into the link field in cinfo.
-   */
-  cinfo.err = jpeg_std_error(&jerr);
-
-  /* Now we can initialize the JPEG decompression object. */
-  jpeg_create_decompress(&cinfo);
-
-  /* Step 2: specify data source (eg, a file) */
-
-  jpeg_stdio_src(&cinfo, infile);
-
-  /* Step 3: read file parameters with jpeg_read_header() */
-
-  (void) jpeg_read_header(&cinfo, TRUE);
-  /* We can ignore the return value from jpeg_read_header since
-   *   (a) suspension is not possible with the stdio data source, and
-   *   (b) we passed TRUE to reject a tables-only JPEG file as an error.
-   * See libjpeg.doc for more info.
-   */
-
-  /* Step 4: set parameters for decompression */
-
-  /* In this example, we don't need to change any of the defaults set by
-   * jpeg_read_header(), so we do nothing here.
-   */
-
-  /* Step 5: Start decompressor */
-
-  (void) jpeg_start_decompress(&cinfo);
-  /* We can ignore the return value since suspension is not possible
-   * with the stdio data source.
-   */
-
-  /* We may need to do some setup of our own at this point before reading
-   * the data.  After jpeg_start_decompress() we have the correct scaled
-   * output image dimensions available, as well as the output colormap
-   * if we asked for color quantization.
-   * In this example, we need to make an output work buffer of the right size.
-   */ 
-  /* JSAMPLEs per row in output buffer */
-  row_stride = cinfo.output_width * cinfo.output_components;
-
-  out = Z_Malloc(cinfo.output_width*cinfo.output_height*cinfo.output_components);
-
-  *pic = out;
-  *width = cinfo.output_width;
-  *height = cinfo.output_height;
-
-  /* Step 6: while (scan lines remain to be read) */
-  /*           jpeg_read_scanlines(...); */
-
-  /* Here we use the library's state variable cinfo.output_scanline as the
-   * loop counter, so that we don't have to keep track ourselves.
-   */
-  while (cinfo.output_scanline < cinfo.output_height) {
-    /* jpeg_read_scanlines expects an array of pointers to scanlines.
-     * Here the array is only one element long, but you could ask for
-     * more than one scanline at a time if that's more convenient.
-     */
-	buffer = (JSAMPARRAY)out+(row_stride*cinfo.output_scanline);
-    (void) jpeg_read_scanlines(&cinfo, buffer, 1);
-  }
-
-  /* Step 7: Finish decompression */
-
-  (void) jpeg_finish_decompress(&cinfo);
-  /* We can ignore the return value since suspension is not possible
-   * with the stdio data source.
-   */
-
-  /* Step 8: Release JPEG decompression object */
-
-  /* This is an important step since it will release a good deal of memory. */
-  jpeg_destroy_decompress(&cinfo);
-
-  /* After finish_decompress, we can close the input file.
-   * Here we postpone it until after no more JPEG errors are possible,
-   * so as to simplify the setjmp error logic above.  (Actually, I don't
-   * think that jpeg_destroy can do an error exit, but why assume anything...)
-   */
-  FS_FCloseFile(infile);
-
-  /* At this point you may want to check to see whether any corrupt-data
-   * warnings occurred (test whether jerr.pub.num_warnings is nonzero).
-   */
-
-  /* And we're done! */
-  return 1;
-}
-
diff --git a/src/jpeg-6b/jmemansi.c b/src/jpeg-6b/jmemansi.c
deleted file mode 100644
index 2d93e496..00000000
--- a/src/jpeg-6b/jmemansi.c
+++ /dev/null
@@ -1,167 +0,0 @@
-/*
- * jmemansi.c
- *
- * Copyright (C) 1992-1996, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This file provides a simple generic implementation of the system-
- * dependent portion of the JPEG memory manager.  This implementation
- * assumes that you have the ANSI-standard library routine tmpfile().
- * Also, the problem of determining the amount of memory available
- * is shoved onto the user.
- */
-
-#define JPEG_INTERNALS
-#include "jinclude.h"
-#include "jpeglib.h"
-#include "jmemsys.h"		/* import the system-dependent declarations */
-
-#ifndef HAVE_STDLIB_H		/* <stdlib.h> should declare malloc(),free() */
-extern void * malloc JPP((size_t size));
-extern void free JPP((void *ptr));
-#endif
-
-#ifndef SEEK_SET		/* pre-ANSI systems may not define this; */
-#define SEEK_SET  0		/* if not, assume 0 is correct */
-#endif
-
-
-/*
- * Memory allocation and freeing are controlled by the regular library
- * routines malloc() and free().
- */
-
-GLOBAL(void *)
-jpeg_get_small (j_common_ptr cinfo, size_t sizeofobject)
-{
-  return (void *) malloc(sizeofobject);
-}
-
-GLOBAL(void)
-jpeg_free_small (j_common_ptr cinfo, void * object, size_t sizeofobject)
-{
-  free(object);
-}
-
-
-/*
- * "Large" objects are treated the same as "small" ones.
- * NB: although we include FAR keywords in the routine declarations,
- * this file won't actually work in 80x86 small/medium model; at least,
- * you probably won't be able to process useful-size images in only 64KB.
- */
-
-GLOBAL(void FAR *)
-jpeg_get_large (j_common_ptr cinfo, size_t sizeofobject)
-{
-  return (void FAR *) malloc(sizeofobject);
-}
-
-GLOBAL(void)
-jpeg_free_large (j_common_ptr cinfo, void FAR * object, size_t sizeofobject)
-{
-  free(object);
-}
-
-
-/*
- * This routine computes the total memory space available for allocation.
- * It's impossible to do this in a portable way; our current solution is
- * to make the user tell us (with a default value set at compile time).
- * If you can actually get the available space, it's a good idea to subtract
- * a slop factor of 5% or so.
- */
-
-#ifndef DEFAULT_MAX_MEM		/* so can override from makefile */
-#define DEFAULT_MAX_MEM		1000000L /* default: one megabyte */
-#endif
-
-GLOBAL(long)
-jpeg_mem_available (j_common_ptr cinfo, long min_bytes_needed,
-		    long max_bytes_needed, long already_allocated)
-{
-  return cinfo->mem->max_memory_to_use - already_allocated;
-}
-
-
-/*
- * Backing store (temporary file) management.
- * Backing store objects are only used when the value returned by
- * jpeg_mem_available is less than the total space needed.  You can dispense
- * with these routines if you have plenty of virtual memory; see jmemnobs.c.
- */
-
-
-METHODDEF(void)
-read_backing_store (j_common_ptr cinfo, backing_store_ptr info,
-		    void FAR * buffer_address,
-		    long file_offset, long byte_count)
-{
-  if (fseek(info->temp_file, file_offset, SEEK_SET))
-    ERREXIT(cinfo, JERR_TFILE_SEEK);
-  if (JFREAD(info->temp_file, buffer_address, byte_count)
-      != (size_t) byte_count)
-    ERREXIT(cinfo, JERR_TFILE_READ);
-}
-
-
-METHODDEF(void)
-write_backing_store (j_common_ptr cinfo, backing_store_ptr info,
-		     void FAR * buffer_address,
-		     long file_offset, long byte_count)
-{
-  if (fseek(info->temp_file, file_offset, SEEK_SET))
-    ERREXIT(cinfo, JERR_TFILE_SEEK);
-  if (JFWRITE(info->temp_file, buffer_address, byte_count)
-      != (size_t) byte_count)
-    ERREXIT(cinfo, JERR_TFILE_WRITE);
-}
-
-
-METHODDEF(void)
-close_backing_store (j_common_ptr cinfo, backing_store_ptr info)
-{
-  fclose(info->temp_file);
-  /* Since this implementation uses tmpfile() to create the file,
-   * no explicit file deletion is needed.
-   */
-}
-
-
-/*
- * Initial opening of a backing-store object.
- *
- * This version uses tmpfile(), which constructs a suitable file name
- * behind the scenes.  We don't have to use info->temp_name[] at all;
- * indeed, we can't even find out the actual name of the temp file.
- */
-
-GLOBAL(void)
-jpeg_open_backing_store (j_common_ptr cinfo, backing_store_ptr info,
-			 long total_bytes_needed)
-{
-  if ((info->temp_file = tmpfile()) == NULL)
-    ERREXITS(cinfo, JERR_TFILE_CREATE, "");
-  info->read_backing_store = read_backing_store;
-  info->write_backing_store = write_backing_store;
-  info->close_backing_store = close_backing_store;
-}
-
-
-/*
- * These routines take care of any system-dependent initialization and
- * cleanup required.
- */
-
-GLOBAL(long)
-jpeg_mem_init (j_common_ptr cinfo)
-{
-  return DEFAULT_MAX_MEM;	/* default for max_memory_to_use */
-}
-
-GLOBAL(void)
-jpeg_mem_term (j_common_ptr cinfo)
-{
-  /* no work */
-}
diff --git a/src/jpeg-6b/jmemdos.c b/src/jpeg-6b/jmemdos.c
deleted file mode 100644
index 60b45c69..00000000
--- a/src/jpeg-6b/jmemdos.c
+++ /dev/null
@@ -1,638 +0,0 @@
-/*
- * jmemdos.c
- *
- * Copyright (C) 1992-1997, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This file provides an MS-DOS-compatible implementation of the system-
- * dependent portion of the JPEG memory manager.  Temporary data can be
- * stored in extended or expanded memory as well as in regular DOS files.
- *
- * If you use this file, you must be sure that NEED_FAR_POINTERS is defined
- * if you compile in a small-data memory model; it should NOT be defined if
- * you use a large-data memory model.  This file is not recommended if you
- * are using a flat-memory-space 386 environment such as DJGCC or Watcom C.
- * Also, this code will NOT work if struct fields are aligned on greater than
- * 2-byte boundaries.
- *
- * Based on code contributed by Ge' Weijers.
- */
-
-/*
- * If you have both extended and expanded memory, you may want to change the
- * order in which they are tried in jopen_backing_store.  On a 286 machine
- * expanded memory is usually faster, since extended memory access involves
- * an expensive protected-mode-and-back switch.  On 386 and better, extended
- * memory is usually faster.  As distributed, the code tries extended memory
- * first (what? not everyone has a 386? :-).
- *
- * You can disable use of extended/expanded memory entirely by altering these
- * definitions or overriding them from the Makefile (eg, -DEMS_SUPPORTED=0).
- */
-
-#ifndef XMS_SUPPORTED
-#define XMS_SUPPORTED  1
-#endif
-#ifndef EMS_SUPPORTED
-#define EMS_SUPPORTED  1
-#endif
-
-
-#define JPEG_INTERNALS
-#include "jinclude.h"
-#include "jpeglib.h"
-#include "jmemsys.h"		/* import the system-dependent declarations */
-
-#ifndef HAVE_STDLIB_H		/* <stdlib.h> should declare these */
-extern void * malloc JPP((size_t size));
-extern void free JPP((void *ptr));
-extern char * getenv JPP((const char * name));
-#endif
-
-#ifdef NEED_FAR_POINTERS
-
-#ifdef __TURBOC__
-/* These definitions work for Borland C (Turbo C) */
-#include <alloc.h>		/* need farmalloc(), farfree() */
-#define far_malloc(x)	farmalloc(x)
-#define far_free(x)	farfree(x)
-#else
-/* These definitions work for Microsoft C and compatible compilers */
-#include <malloc.h>		/* need _fmalloc(), _ffree() */
-#define far_malloc(x)	_fmalloc(x)
-#define far_free(x)	_ffree(x)
-#endif
-
-#else /* not NEED_FAR_POINTERS */
-
-#define far_malloc(x)	malloc(x)
-#define far_free(x)	free(x)
-
-#endif /* NEED_FAR_POINTERS */
-
-#ifdef DONT_USE_B_MODE		/* define mode parameters for fopen() */
-#define READ_BINARY	"r"
-#else
-#define READ_BINARY	"rb"
-#endif
-
-#ifndef USE_MSDOS_MEMMGR	/* make sure user got configuration right */
-  You forgot to define USE_MSDOS_MEMMGR in jconfig.h. /* deliberate syntax error */
-#endif
-
-#if MAX_ALLOC_CHUNK >= 65535L	/* make sure jconfig.h got this right */
-  MAX_ALLOC_CHUNK should be less than 64K. /* deliberate syntax error */
-#endif
-
-
-/*
- * Declarations for assembly-language support routines (see jmemdosa.asm).
- *
- * The functions are declared "far" as are all their pointer arguments;
- * this ensures the assembly source code will work regardless of the
- * compiler memory model.  We assume "short" is 16 bits, "long" is 32.
- */
-
-typedef void far * XMSDRIVER;	/* actually a pointer to code */
-typedef struct {		/* registers for calling XMS driver */
-	unsigned short ax, dx, bx;
-	void far * ds_si;
-      } XMScontext;
-typedef struct {		/* registers for calling EMS driver */
-	unsigned short ax, dx, bx;
-	void far * ds_si;
-      } EMScontext;
-
-extern short far jdos_open JPP((short far * handle, char far * filename));
-extern short far jdos_close JPP((short handle));
-extern short far jdos_seek JPP((short handle, long offset));
-extern short far jdos_read JPP((short handle, void far * buffer,
-				unsigned short count));
-extern short far jdos_write JPP((short handle, void far * buffer,
-				 unsigned short count));
-extern void far jxms_getdriver JPP((XMSDRIVER far *));
-extern void far jxms_calldriver JPP((XMSDRIVER, XMScontext far *));
-extern short far jems_available JPP((void));
-extern void far jems_calldriver JPP((EMScontext far *));
-
-
-/*
- * Selection of a file name for a temporary file.
- * This is highly system-dependent, and you may want to customize it.
- */
-
-static int next_file_num;	/* to distinguish among several temp files */
-
-LOCAL(void)
-select_file_name (char * fname)
-{
-  const char * env;
-  char * ptr;
-  FILE * tfile;
-
-  /* Keep generating file names till we find one that's not in use */
-  for (;;) {
-    /* Get temp directory name from environment TMP or TEMP variable;
-     * if none, use "."
-     */
-    if ((env = (const char *) getenv("TMP")) == NULL)
-      if ((env = (const char *) getenv("TEMP")) == NULL)
-	env = ".";
-    if (*env == '\0')		/* null string means "." */
-      env = ".";
-    ptr = fname;		/* copy name to fname */
-    while (*env != '\0')
-      *ptr++ = *env++;
-    if (ptr[-1] != '\\' && ptr[-1] != '/')
-      *ptr++ = '\\';		/* append backslash if not in env variable */
-    /* Append a suitable file name */
-    next_file_num++;		/* advance counter */
-    sprintf(ptr, "JPG%03d.TMP", next_file_num);
-    /* Probe to see if file name is already in use */
-    if ((tfile = fopen(fname, READ_BINARY)) == NULL)
-      break;
-    fclose(tfile);		/* oops, it's there; close tfile & try again */
-  }
-}
-
-
-/*
- * Near-memory allocation and freeing are controlled by the regular library
- * routines malloc() and free().
- */
-
-GLOBAL(void *)
-jpeg_get_small (j_common_ptr cinfo, size_t sizeofobject)
-{
-  return (void *) malloc(sizeofobject);
-}
-
-GLOBAL(void)
-jpeg_free_small (j_common_ptr cinfo, void * object, size_t sizeofobject)
-{
-  free(object);
-}
-
-
-/*
- * "Large" objects are allocated in far memory, if possible
- */
-
-GLOBAL(void FAR *)
-jpeg_get_large (j_common_ptr cinfo, size_t sizeofobject)
-{
-  return (void FAR *) far_malloc(sizeofobject);
-}
-
-GLOBAL(void)
-jpeg_free_large (j_common_ptr cinfo, void FAR * object, size_t sizeofobject)
-{
-  far_free(object);
-}
-
-
-/*
- * This routine computes the total memory space available for allocation.
- * It's impossible to do this in a portable way; our current solution is
- * to make the user tell us (with a default value set at compile time).
- * If you can actually get the available space, it's a good idea to subtract
- * a slop factor of 5% or so.
- */
-
-#ifndef DEFAULT_MAX_MEM		/* so can override from makefile */
-#define DEFAULT_MAX_MEM		300000L /* for total usage about 450K */
-#endif
-
-GLOBAL(long)
-jpeg_mem_available (j_common_ptr cinfo, long min_bytes_needed,
-		    long max_bytes_needed, long already_allocated)
-{
-  return cinfo->mem->max_memory_to_use - already_allocated;
-}
-
-
-/*
- * Backing store (temporary file) management.
- * Backing store objects are only used when the value returned by
- * jpeg_mem_available is less than the total space needed.  You can dispense
- * with these routines if you have plenty of virtual memory; see jmemnobs.c.
- */
-
-/*
- * For MS-DOS we support three types of backing storage:
- *   1. Conventional DOS files.  We access these by direct DOS calls rather
- *      than via the stdio package.  This provides a bit better performance,
- *      but the real reason is that the buffers to be read or written are FAR.
- *      The stdio library for small-data memory models can't cope with that.
- *   2. Extended memory, accessed per the XMS V2.0 specification.
- *   3. Expanded memory, accessed per the LIM/EMS 4.0 specification.
- * You'll need copies of those specs to make sense of the related code.
- * The specs are available by Internet FTP from the SIMTEL archives 
- * (oak.oakland.edu and its various mirror sites).  See files
- * pub/msdos/microsoft/xms20.arc and pub/msdos/info/limems41.zip.
- */
-
-
-/*
- * Access methods for a DOS file.
- */
-
-
-METHODDEF(void)
-read_file_store (j_common_ptr cinfo, backing_store_ptr info,
-		 void FAR * buffer_address,
-		 long file_offset, long byte_count)
-{
-  if (jdos_seek(info->handle.file_handle, file_offset))
-    ERREXIT(cinfo, JERR_TFILE_SEEK);
-  /* Since MAX_ALLOC_CHUNK is less than 64K, byte_count will be too. */
-  if (byte_count > 65535L)	/* safety check */
-    ERREXIT(cinfo, JERR_BAD_ALLOC_CHUNK);
-  if (jdos_read(info->handle.file_handle, buffer_address,
-		(unsigned short) byte_count))
-    ERREXIT(cinfo, JERR_TFILE_READ);
-}
-
-
-METHODDEF(void)
-write_file_store (j_common_ptr cinfo, backing_store_ptr info,
-		  void FAR * buffer_address,
-		  long file_offset, long byte_count)
-{
-  if (jdos_seek(info->handle.file_handle, file_offset))
-    ERREXIT(cinfo, JERR_TFILE_SEEK);
-  /* Since MAX_ALLOC_CHUNK is less than 64K, byte_count will be too. */
-  if (byte_count > 65535L)	/* safety check */
-    ERREXIT(cinfo, JERR_BAD_ALLOC_CHUNK);
-  if (jdos_write(info->handle.file_handle, buffer_address,
-		 (unsigned short) byte_count))
-    ERREXIT(cinfo, JERR_TFILE_WRITE);
-}
-
-
-METHODDEF(void)
-close_file_store (j_common_ptr cinfo, backing_store_ptr info)
-{
-  jdos_close(info->handle.file_handle);	/* close the file */
-  remove(info->temp_name);	/* delete the file */
-/* If your system doesn't have remove(), try unlink() instead.
- * remove() is the ANSI-standard name for this function, but
- * unlink() was more common in pre-ANSI systems.
- */
-  TRACEMSS(cinfo, 1, JTRC_TFILE_CLOSE, info->temp_name);
-}
-
-
-LOCAL(boolean)
-open_file_store (j_common_ptr cinfo, backing_store_ptr info,
-		 long total_bytes_needed)
-{
-  short handle;
-
-  select_file_name(info->temp_name);
-  if (jdos_open((short far *) & handle, (char far *) info->temp_name)) {
-    /* might as well exit since jpeg_open_backing_store will fail anyway */
-    ERREXITS(cinfo, JERR_TFILE_CREATE, info->temp_name);
-    return FALSE;
-  }
-  info->handle.file_handle = handle;
-  info->read_backing_store = read_file_store;
-  info->write_backing_store = write_file_store;
-  info->close_backing_store = close_file_store;
-  TRACEMSS(cinfo, 1, JTRC_TFILE_OPEN, info->temp_name);
-  return TRUE;			/* succeeded */
-}
-
-
-/*
- * Access methods for extended memory.
- */
-
-#if XMS_SUPPORTED
-
-static XMSDRIVER xms_driver;	/* saved address of XMS driver */
-
-typedef union {			/* either long offset or real-mode pointer */
-	long offset;
-	void far * ptr;
-      } XMSPTR;
-
-typedef struct {		/* XMS move specification structure */
-	long length;
-	XMSH src_handle;
-	XMSPTR src;
-	XMSH dst_handle;
-	XMSPTR dst;
-      } XMSspec;
-
-#define ODD(X)	(((X) & 1L) != 0)
-
-
-METHODDEF(void)
-read_xms_store (j_common_ptr cinfo, backing_store_ptr info,
-		void FAR * buffer_address,
-		long file_offset, long byte_count)
-{
-  XMScontext ctx;
-  XMSspec spec;
-  char endbuffer[2];
-
-  /* The XMS driver can't cope with an odd length, so handle the last byte
-   * specially if byte_count is odd.  We don't expect this to be common.
-   */
-
-  spec.length = byte_count & (~ 1L);
-  spec.src_handle = info->handle.xms_handle;
-  spec.src.offset = file_offset;
-  spec.dst_handle = 0;
-  spec.dst.ptr = buffer_address;
-  
-  ctx.ds_si = (void far *) & spec;
-  ctx.ax = 0x0b00;		/* EMB move */
-  jxms_calldriver(xms_driver, (XMScontext far *) & ctx);
-  if (ctx.ax != 1)
-    ERREXIT(cinfo, JERR_XMS_READ);
-
-  if (ODD(byte_count)) {
-    read_xms_store(cinfo, info, (void FAR *) endbuffer,
-		   file_offset + byte_count - 1L, 2L);
-    ((char FAR *) buffer_address)[byte_count - 1L] = endbuffer[0];
-  }
-}
-
-
-METHODDEF(void)
-write_xms_store (j_common_ptr cinfo, backing_store_ptr info,
-		 void FAR * buffer_address,
-		 long file_offset, long byte_count)
-{
-  XMScontext ctx;
-  XMSspec spec;
-  char endbuffer[2];
-
-  /* The XMS driver can't cope with an odd length, so handle the last byte
-   * specially if byte_count is odd.  We don't expect this to be common.
-   */
-
-  spec.length = byte_count & (~ 1L);
-  spec.src_handle = 0;
-  spec.src.ptr = buffer_address;
-  spec.dst_handle = info->handle.xms_handle;
-  spec.dst.offset = file_offset;
-
-  ctx.ds_si = (void far *) & spec;
-  ctx.ax = 0x0b00;		/* EMB move */
-  jxms_calldriver(xms_driver, (XMScontext far *) & ctx);
-  if (ctx.ax != 1)
-    ERREXIT(cinfo, JERR_XMS_WRITE);
-
-  if (ODD(byte_count)) {
-    read_xms_store(cinfo, info, (void FAR *) endbuffer,
-		   file_offset + byte_count - 1L, 2L);
-    endbuffer[0] = ((char FAR *) buffer_address)[byte_count - 1L];
-    write_xms_store(cinfo, info, (void FAR *) endbuffer,
-		    file_offset + byte_count - 1L, 2L);
-  }
-}
-
-
-METHODDEF(void)
-close_xms_store (j_common_ptr cinfo, backing_store_ptr info)
-{
-  XMScontext ctx;
-
-  ctx.dx = info->handle.xms_handle;
-  ctx.ax = 0x0a00;
-  jxms_calldriver(xms_driver, (XMScontext far *) & ctx);
-  TRACEMS1(cinfo, 1, JTRC_XMS_CLOSE, info->handle.xms_handle);
-  /* we ignore any error return from the driver */
-}
-
-
-LOCAL(boolean)
-open_xms_store (j_common_ptr cinfo, backing_store_ptr info,
-		long total_bytes_needed)
-{
-  XMScontext ctx;
-
-  /* Get address of XMS driver */
-  jxms_getdriver((XMSDRIVER far *) & xms_driver);
-  if (xms_driver == NULL)
-    return FALSE;		/* no driver to be had */
-
-  /* Get version number, must be >= 2.00 */
-  ctx.ax = 0x0000;
-  jxms_calldriver(xms_driver, (XMScontext far *) & ctx);
-  if (ctx.ax < (unsigned short) 0x0200)
-    return FALSE;
-
-  /* Try to get space (expressed in kilobytes) */
-  ctx.dx = (unsigned short) ((total_bytes_needed + 1023L) >> 10);
-  ctx.ax = 0x0900;
-  jxms_calldriver(xms_driver, (XMScontext far *) & ctx);
-  if (ctx.ax != 1)
-    return FALSE;
-
-  /* Succeeded, save the handle and away we go */
-  info->handle.xms_handle = ctx.dx;
-  info->read_backing_store = read_xms_store;
-  info->write_backing_store = write_xms_store;
-  info->close_backing_store = close_xms_store;
-  TRACEMS1(cinfo, 1, JTRC_XMS_OPEN, ctx.dx);
-  return TRUE;			/* succeeded */
-}
-
-#endif /* XMS_SUPPORTED */
-
-
-/*
- * Access methods for expanded memory.
- */
-
-#if EMS_SUPPORTED
-
-/* The EMS move specification structure requires word and long fields aligned
- * at odd byte boundaries.  Some compilers will align struct fields at even
- * byte boundaries.  While it's usually possible to force byte alignment,
- * that causes an overall performance penalty and may pose problems in merging
- * JPEG into a larger application.  Instead we accept some rather dirty code
- * here.  Note this code would fail if the hardware did not allow odd-byte
- * word & long accesses, but all 80x86 CPUs do.
- */
-
-typedef void far * EMSPTR;
-
-typedef union {			/* EMS move specification structure */
-	long length;		/* It's easy to access first 4 bytes */
-	char bytes[18];		/* Misaligned fields in here! */
-      } EMSspec;
-
-/* Macros for accessing misaligned fields */
-#define FIELD_AT(spec,offset,type)  (*((type *) &(spec.bytes[offset])))
-#define SRC_TYPE(spec)		FIELD_AT(spec,4,char)
-#define SRC_HANDLE(spec)	FIELD_AT(spec,5,EMSH)
-#define SRC_OFFSET(spec)	FIELD_AT(spec,7,unsigned short)
-#define SRC_PAGE(spec)		FIELD_AT(spec,9,unsigned short)
-#define SRC_PTR(spec)		FIELD_AT(spec,7,EMSPTR)
-#define DST_TYPE(spec)		FIELD_AT(spec,11,char)
-#define DST_HANDLE(spec)	FIELD_AT(spec,12,EMSH)
-#define DST_OFFSET(spec)	FIELD_AT(spec,14,unsigned short)
-#define DST_PAGE(spec)		FIELD_AT(spec,16,unsigned short)
-#define DST_PTR(spec)		FIELD_AT(spec,14,EMSPTR)
-
-#define EMSPAGESIZE	16384L	/* gospel, see the EMS specs */
-
-#define HIBYTE(W)  (((W) >> 8) & 0xFF)
-#define LOBYTE(W)  ((W) & 0xFF)
-
-
-METHODDEF(void)
-read_ems_store (j_common_ptr cinfo, backing_store_ptr info,
-		void FAR * buffer_address,
-		long file_offset, long byte_count)
-{
-  EMScontext ctx;
-  EMSspec spec;
-
-  spec.length = byte_count;
-  SRC_TYPE(spec) = 1;
-  SRC_HANDLE(spec) = info->handle.ems_handle;
-  SRC_PAGE(spec)   = (unsigned short) (file_offset / EMSPAGESIZE);
-  SRC_OFFSET(spec) = (unsigned short) (file_offset % EMSPAGESIZE);
-  DST_TYPE(spec) = 0;
-  DST_HANDLE(spec) = 0;
-  DST_PTR(spec)    = buffer_address;
-  
-  ctx.ds_si = (void far *) & spec;
-  ctx.ax = 0x5700;		/* move memory region */
-  jems_calldriver((EMScontext far *) & ctx);
-  if (HIBYTE(ctx.ax) != 0)
-    ERREXIT(cinfo, JERR_EMS_READ);
-}
-
-
-METHODDEF(void)
-write_ems_store (j_common_ptr cinfo, backing_store_ptr info,
-		 void FAR * buffer_address,
-		 long file_offset, long byte_count)
-{
-  EMScontext ctx;
-  EMSspec spec;
-
-  spec.length = byte_count;
-  SRC_TYPE(spec) = 0;
-  SRC_HANDLE(spec) = 0;
-  SRC_PTR(spec)    = buffer_address;
-  DST_TYPE(spec) = 1;
-  DST_HANDLE(spec) = info->handle.ems_handle;
-  DST_PAGE(spec)   = (unsigned short) (file_offset / EMSPAGESIZE);
-  DST_OFFSET(spec) = (unsigned short) (file_offset % EMSPAGESIZE);
-  
-  ctx.ds_si = (void far *) & spec;
-  ctx.ax = 0x5700;		/* move memory region */
-  jems_calldriver((EMScontext far *) & ctx);
-  if (HIBYTE(ctx.ax) != 0)
-    ERREXIT(cinfo, JERR_EMS_WRITE);
-}
-
-
-METHODDEF(void)
-close_ems_store (j_common_ptr cinfo, backing_store_ptr info)
-{
-  EMScontext ctx;
-
-  ctx.ax = 0x4500;
-  ctx.dx = info->handle.ems_handle;
-  jems_calldriver((EMScontext far *) & ctx);
-  TRACEMS1(cinfo, 1, JTRC_EMS_CLOSE, info->handle.ems_handle);
-  /* we ignore any error return from the driver */
-}
-
-
-LOCAL(boolean)
-open_ems_store (j_common_ptr cinfo, backing_store_ptr info,
-		long total_bytes_needed)
-{
-  EMScontext ctx;
-
-  /* Is EMS driver there? */
-  if (! jems_available())
-    return FALSE;
-
-  /* Get status, make sure EMS is OK */
-  ctx.ax = 0x4000;
-  jems_calldriver((EMScontext far *) & ctx);
-  if (HIBYTE(ctx.ax) != 0)
-    return FALSE;
-
-  /* Get version, must be >= 4.0 */
-  ctx.ax = 0x4600;
-  jems_calldriver((EMScontext far *) & ctx);
-  if (HIBYTE(ctx.ax) != 0 || LOBYTE(ctx.ax) < 0x40)
-    return FALSE;
-
-  /* Try to allocate requested space */
-  ctx.ax = 0x4300;
-  ctx.bx = (unsigned short) ((total_bytes_needed + EMSPAGESIZE-1L) / EMSPAGESIZE);
-  jems_calldriver((EMScontext far *) & ctx);
-  if (HIBYTE(ctx.ax) != 0)
-    return FALSE;
-
-  /* Succeeded, save the handle and away we go */
-  info->handle.ems_handle = ctx.dx;
-  info->read_backing_store = read_ems_store;
-  info->write_backing_store = write_ems_store;
-  info->close_backing_store = close_ems_store;
-  TRACEMS1(cinfo, 1, JTRC_EMS_OPEN, ctx.dx);
-  return TRUE;			/* succeeded */
-}
-
-#endif /* EMS_SUPPORTED */
-
-
-/*
- * Initial opening of a backing-store object.
- */
-
-GLOBAL(void)
-jpeg_open_backing_store (j_common_ptr cinfo, backing_store_ptr info,
-			 long total_bytes_needed)
-{
-  /* Try extended memory, then expanded memory, then regular file. */
-#if XMS_SUPPORTED
-  if (open_xms_store(cinfo, info, total_bytes_needed))
-    return;
-#endif
-#if EMS_SUPPORTED
-  if (open_ems_store(cinfo, info, total_bytes_needed))
-    return;
-#endif
-  if (open_file_store(cinfo, info, total_bytes_needed))
-    return;
-  ERREXITS(cinfo, JERR_TFILE_CREATE, "");
-}
-
-
-/*
- * These routines take care of any system-dependent initialization and
- * cleanup required.
- */
-
-GLOBAL(long)
-jpeg_mem_init (j_common_ptr cinfo)
-{
-  next_file_num = 0;		/* initialize temp file name generator */
-  return DEFAULT_MAX_MEM;	/* default for max_memory_to_use */
-}
-
-GLOBAL(void)
-jpeg_mem_term (j_common_ptr cinfo)
-{
-  /* Microsoft C, at least in v6.00A, will not successfully reclaim freed
-   * blocks of size > 32Kbytes unless we give it a kick in the rear, like so:
-   */
-#ifdef NEED_FHEAPMIN
-  _fheapmin();
-#endif
-}
diff --git a/src/jpeg-6b/jmemname.c b/src/jpeg-6b/jmemname.c
deleted file mode 100644
index ed96dee1..00000000
--- a/src/jpeg-6b/jmemname.c
+++ /dev/null
@@ -1,276 +0,0 @@
-/*
- * jmemname.c
- *
- * Copyright (C) 1992-1997, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This file provides a generic implementation of the system-dependent
- * portion of the JPEG memory manager.  This implementation assumes that
- * you must explicitly construct a name for each temp file.
- * Also, the problem of determining the amount of memory available
- * is shoved onto the user.
- */
-
-#define JPEG_INTERNALS
-#include "jinclude.h"
-#include "jpeglib.h"
-#include "jmemsys.h"		/* import the system-dependent declarations */
-
-#ifndef HAVE_STDLIB_H		/* <stdlib.h> should declare malloc(),free() */
-extern void * malloc JPP((size_t size));
-extern void free JPP((void *ptr));
-#endif
-
-#ifndef SEEK_SET		/* pre-ANSI systems may not define this; */
-#define SEEK_SET  0		/* if not, assume 0 is correct */
-#endif
-
-#ifdef DONT_USE_B_MODE		/* define mode parameters for fopen() */
-#define READ_BINARY	"r"
-#define RW_BINARY	"w+"
-#else
-#ifdef VMS			/* VMS is very nonstandard */
-#define READ_BINARY	"rb", "ctx=stm"
-#define RW_BINARY	"w+b", "ctx=stm"
-#else				/* standard ANSI-compliant case */
-#define READ_BINARY	"rb"
-#define RW_BINARY	"w+b"
-#endif
-#endif
-
-
-/*
- * Selection of a file name for a temporary file.
- * This is system-dependent!
- *
- * The code as given is suitable for most Unix systems, and it is easily
- * modified for most non-Unix systems.  Some notes:
- *  1.  The temp file is created in the directory named by TEMP_DIRECTORY.
- *      The default value is /usr/tmp, which is the conventional place for
- *      creating large temp files on Unix.  On other systems you'll probably
- *      want to change the file location.  You can do this by editing the
- *      #define, or (preferred) by defining TEMP_DIRECTORY in jconfig.h.
- *
- *  2.  If you need to change the file name as well as its location,
- *      you can override the TEMP_FILE_NAME macro.  (Note that this is
- *      actually a printf format string; it must contain %s and %d.)
- *      Few people should need to do this.
- *
- *  3.  mktemp() is used to ensure that multiple processes running
- *      simultaneously won't select the same file names.  If your system
- *      doesn't have mktemp(), define NO_MKTEMP to do it the hard way.
- *      (If you don't have <errno.h>, also define NO_ERRNO_H.)
- *
- *  4.  You probably want to define NEED_SIGNAL_CATCHER so that cjpeg.c/djpeg.c
- *      will cause the temp files to be removed if you stop the program early.
- */
-
-#ifndef TEMP_DIRECTORY		/* can override from jconfig.h or Makefile */
-#define TEMP_DIRECTORY  "/usr/tmp/" /* recommended setting for Unix */
-#endif
-
-static int next_file_num;	/* to distinguish among several temp files */
-
-#ifdef NO_MKTEMP
-
-#ifndef TEMP_FILE_NAME		/* can override from jconfig.h or Makefile */
-#define TEMP_FILE_NAME  "%sJPG%03d.TMP"
-#endif
-
-#ifndef NO_ERRNO_H
-#include <errno.h>		/* to define ENOENT */
-#endif
-
-/* ANSI C specifies that errno is a macro, but on older systems it's more
- * likely to be a plain int variable.  And not all versions of errno.h
- * bother to declare it, so we have to in order to be most portable.  Thus:
- */
-#ifndef errno
-extern int errno;
-#endif
-
-
-LOCAL(void)
-select_file_name (char * fname)
-{
-  FILE * tfile;
-
-  /* Keep generating file names till we find one that's not in use */
-  for (;;) {
-    next_file_num++;		/* advance counter */
-    sprintf(fname, TEMP_FILE_NAME, TEMP_DIRECTORY, next_file_num);
-    if ((tfile = fopen(fname, READ_BINARY)) == NULL) {
-      /* fopen could have failed for a reason other than the file not
-       * being there; for example, file there but unreadable.
-       * If <errno.h> isn't available, then we cannot test the cause.
-       */
-#ifdef ENOENT
-      if (errno != ENOENT)
-	continue;
-#endif
-      break;
-    }
-    fclose(tfile);		/* oops, it's there; close tfile & try again */
-  }
-}
-
-#else /* ! NO_MKTEMP */
-
-/* Note that mktemp() requires the initial filename to end in six X's */
-#ifndef TEMP_FILE_NAME		/* can override from jconfig.h or Makefile */
-#define TEMP_FILE_NAME  "%sJPG%dXXXXXX"
-#endif
-
-LOCAL(void)
-select_file_name (char * fname)
-{
-  next_file_num++;		/* advance counter */
-  sprintf(fname, TEMP_FILE_NAME, TEMP_DIRECTORY, next_file_num);
-  mktemp(fname);		/* make sure file name is unique */
-  /* mktemp replaces the trailing XXXXXX with a unique string of characters */
-}
-
-#endif /* NO_MKTEMP */
-
-
-/*
- * Memory allocation and freeing are controlled by the regular library
- * routines malloc() and free().
- */
-
-GLOBAL(void *)
-jpeg_get_small (j_common_ptr cinfo, size_t sizeofobject)
-{
-  return (void *) malloc(sizeofobject);
-}
-
-GLOBAL(void)
-jpeg_free_small (j_common_ptr cinfo, void * object, size_t sizeofobject)
-{
-  free(object);
-}
-
-
-/*
- * "Large" objects are treated the same as "small" ones.
- * NB: although we include FAR keywords in the routine declarations,
- * this file won't actually work in 80x86 small/medium model; at least,
- * you probably won't be able to process useful-size images in only 64KB.
- */
-
-GLOBAL(void FAR *)
-jpeg_get_large (j_common_ptr cinfo, size_t sizeofobject)
-{
-  return (void FAR *) malloc(sizeofobject);
-}
-
-GLOBAL(void)
-jpeg_free_large (j_common_ptr cinfo, void FAR * object, size_t sizeofobject)
-{
-  free(object);
-}
-
-
-/*
- * This routine computes the total memory space available for allocation.
- * It's impossible to do this in a portable way; our current solution is
- * to make the user tell us (with a default value set at compile time).
- * If you can actually get the available space, it's a good idea to subtract
- * a slop factor of 5% or so.
- */
-
-#ifndef DEFAULT_MAX_MEM		/* so can override from makefile */
-#define DEFAULT_MAX_MEM		1000000L /* default: one megabyte */
-#endif
-
-GLOBAL(long)
-jpeg_mem_available (j_common_ptr cinfo, long min_bytes_needed,
-		    long max_bytes_needed, long already_allocated)
-{
-  return cinfo->mem->max_memory_to_use - already_allocated;
-}
-
-
-/*
- * Backing store (temporary file) management.
- * Backing store objects are only used when the value returned by
- * jpeg_mem_available is less than the total space needed.  You can dispense
- * with these routines if you have plenty of virtual memory; see jmemnobs.c.
- */
-
-
-METHODDEF(void)
-read_backing_store (j_common_ptr cinfo, backing_store_ptr info,
-		    void FAR * buffer_address,
-		    long file_offset, long byte_count)
-{
-  if (fseek(info->temp_file, file_offset, SEEK_SET))
-    ERREXIT(cinfo, JERR_TFILE_SEEK);
-  if (JFREAD(info->temp_file, buffer_address, byte_count)
-      != (size_t) byte_count)
-    ERREXIT(cinfo, JERR_TFILE_READ);
-}
-
-
-METHODDEF(void)
-write_backing_store (j_common_ptr cinfo, backing_store_ptr info,
-		     void FAR * buffer_address,
-		     long file_offset, long byte_count)
-{
-  if (fseek(info->temp_file, file_offset, SEEK_SET))
-    ERREXIT(cinfo, JERR_TFILE_SEEK);
-  if (JFWRITE(info->temp_file, buffer_address, byte_count)
-      != (size_t) byte_count)
-    ERREXIT(cinfo, JERR_TFILE_WRITE);
-}
-
-
-METHODDEF(void)
-close_backing_store (j_common_ptr cinfo, backing_store_ptr info)
-{
-  fclose(info->temp_file);	/* close the file */
-  unlink(info->temp_name);	/* delete the file */
-/* If your system doesn't have unlink(), use remove() instead.
- * remove() is the ANSI-standard name for this function, but if
- * your system was ANSI you'd be using jmemansi.c, right?
- */
-  TRACEMSS(cinfo, 1, JTRC_TFILE_CLOSE, info->temp_name);
-}
-
-
-/*
- * Initial opening of a backing-store object.
- */
-
-GLOBAL(void)
-jpeg_open_backing_store (j_common_ptr cinfo, backing_store_ptr info,
-			 long total_bytes_needed)
-{
-  select_file_name(info->temp_name);
-  if ((info->temp_file = fopen(info->temp_name, RW_BINARY)) == NULL)
-    ERREXITS(cinfo, JERR_TFILE_CREATE, info->temp_name);
-  info->read_backing_store = read_backing_store;
-  info->write_backing_store = write_backing_store;
-  info->close_backing_store = close_backing_store;
-  TRACEMSS(cinfo, 1, JTRC_TFILE_OPEN, info->temp_name);
-}
-
-
-/*
- * These routines take care of any system-dependent initialization and
- * cleanup required.
- */
-
-GLOBAL(long)
-jpeg_mem_init (j_common_ptr cinfo)
-{
-  next_file_num = 0;		/* initialize temp file name generator */
-  return DEFAULT_MAX_MEM;	/* default for max_memory_to_use */
-}
-
-GLOBAL(void)
-jpeg_mem_term (j_common_ptr cinfo)
-{
-  /* no work */
-}
diff --git a/src/jpeg-6b/jpegtran.c b/src/jpeg-6b/jpegtran.c
deleted file mode 100644
index 20ef111b..00000000
--- a/src/jpeg-6b/jpegtran.c
+++ /dev/null
@@ -1,504 +0,0 @@
-/*
- * jpegtran.c
- *
- * Copyright (C) 1995-1997, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
- *
- * This file contains a command-line user interface for JPEG transcoding.
- * It is very similar to cjpeg.c, but provides lossless transcoding between
- * different JPEG file formats.  It also provides some lossless and sort-of-
- * lossless transformations of JPEG data.
- */
-
-#include "cdjpeg.h"		/* Common decls for cjpeg/djpeg applications */
-#include "transupp.h"		/* Support routines for jpegtran */
-#include "jversion.h"		/* for version message */
-
-#ifdef USE_CCOMMAND		/* command-line reader for Macintosh */
-#ifdef __MWERKS__
-#include <SIOUX.h>              /* Metrowerks needs this */
-#include <console.h>		/* ... and this */
-#endif
-#ifdef THINK_C
-#include <console.h>		/* Think declares it here */
-#endif
-#endif
-
-
-/*
- * Argument-parsing code.
- * The switch parser is designed to be useful with DOS-style command line
- * syntax, ie, intermixed switches and file names, where only the switches
- * to the left of a given file name affect processing of that file.
- * The main program in this file doesn't actually use this capability...
- */
-
-
-static const char * progname;	/* program name for error messages */
-static char * outfilename;	/* for -outfile switch */
-static JCOPY_OPTION copyoption;	/* -copy switch */
-static jpeg_transform_info transformoption; /* image transformation options */
-
-
-LOCAL(void)
-usage (void)
-/* complain about bad command line */
-{
-  fprintf(stderr, "usage: %s [switches] ", progname);
-#ifdef TWO_FILE_COMMANDLINE
-  fprintf(stderr, "inputfile outputfile\n");
-#else
-  fprintf(stderr, "[inputfile]\n");
-#endif
-
-  fprintf(stderr, "Switches (names may be abbreviated):\n");
-  fprintf(stderr, "  -copy none     Copy no extra markers from source file\n");
-  fprintf(stderr, "  -copy comments Copy only comment markers (default)\n");
-  fprintf(stderr, "  -copy all      Copy all extra markers\n");
-#ifdef ENTROPY_OPT_SUPPORTED
-  fprintf(stderr, "  -optimize      Optimize Huffman table (smaller file, but slow compression)\n");
-#endif
-#ifdef C_PROGRESSIVE_SUPPORTED
-  fprintf(stderr, "  -progressive   Create progressive JPEG file\n");
-#endif
-#if TRANSFORMS_SUPPORTED
-  fprintf(stderr, "Switches for modifying the image:\n");
-  fprintf(stderr, "  -grayscale     Reduce to grayscale (omit color data)\n");
-  fprintf(stderr, "  -flip [horizontal|vertical]  Mirror image (left-right or top-bottom)\n");
-  fprintf(stderr, "  -rotate [90|180|270]         Rotate image (degrees clockwise)\n");
-  fprintf(stderr, "  -transpose     Transpose image\n");
-  fprintf(stderr, "  -transverse    Transverse transpose image\n");
-  fprintf(stderr, "  -trim          Drop non-transformable edge blocks\n");
-#endif /* TRANSFORMS_SUPPORTED */
-  fprintf(stderr, "Switches for advanced users:\n");
-  fprintf(stderr, "  -restart N     Set restart interval in rows, or in blocks with B\n");
-  fprintf(stderr, "  -maxmemory N   Maximum memory to use (in kbytes)\n");
-  fprintf(stderr, "  -outfile name  Specify name for output file\n");
-  fprintf(stderr, "  -verbose  or  -debug   Emit debug output\n");
-  fprintf(stderr, "Switches for wizards:\n");
-#ifdef C_ARITH_CODING_SUPPORTED
-  fprintf(stderr, "  -arithmetic    Use arithmetic coding\n");
-#endif
-#ifdef C_MULTISCAN_FILES_SUPPORTED
-  fprintf(stderr, "  -scans file    Create multi-scan JPEG per script file\n");
-#endif
-  exit(EXIT_FAILURE);
-}
-
-
-LOCAL(void)
-select_transform (JXFORM_CODE transform)
-/* Silly little routine to detect multiple transform options,
- * which we can't handle.
- */
-{
-#if TRANSFORMS_SUPPORTED
-  if (transformoption.transform == JXFORM_NONE ||
-      transformoption.transform == transform) {
-    transformoption.transform = transform;
-  } else {
-    fprintf(stderr, "%s: can only do one image transformation at a time\n",
-	    progname);
-    usage();
-  }
-#else
-  fprintf(stderr, "%s: sorry, image transformation was not compiled\n",
-	  progname);
-  exit(EXIT_FAILURE);
-#endif
-}
-
-
-LOCAL(int)
-parse_switches (j_compress_ptr cinfo, int argc, char **argv,
-		int last_file_arg_seen, boolean for_real)
-/* Parse optional switches.
- * Returns argv[] index of first file-name argument (== argc if none).
- * Any file names with indexes <= last_file_arg_seen are ignored;
- * they have presumably been processed in a previous iteration.
- * (Pass 0 for last_file_arg_seen on the first or only iteration.)
- * for_real is FALSE on the first (dummy) pass; we may skip any expensive
- * processing.
- */
-{
-  int argn;
-  char * arg;
-  boolean simple_progressive;
-  char * scansarg = NULL;	/* saves -scans parm if any */
-
-  /* Set up default JPEG parameters. */
-  simple_progressive = FALSE;
-  outfilename = NULL;
-  copyoption = JCOPYOPT_DEFAULT;
-  transformoption.transform = JXFORM_NONE;
-  transformoption.trim = FALSE;
-  transformoption.force_grayscale = FALSE;
-  cinfo->err->trace_level = 0;
-
-  /* Scan command line options, adjust parameters */
-
-  for (argn = 1; argn < argc; argn++) {
-    arg = argv[argn];
-    if (*arg != '-') {
-      /* Not a switch, must be a file name argument */
-      if (argn <= last_file_arg_seen) {
-	outfilename = NULL;	/* -outfile applies to just one input file */
-	continue;		/* ignore this name if previously processed */
-      }
-      break;			/* else done parsing switches */
-    }
-    arg++;			/* advance past switch marker character */
-
-    if (keymatch(arg, "arithmetic", 1)) {
-      /* Use arithmetic coding. */
-#ifdef C_ARITH_CODING_SUPPORTED
-      cinfo->arith_code = TRUE;
-#else
-      fprintf(stderr, "%s: sorry, arithmetic coding not supported\n",
-	      progname);
-      exit(EXIT_FAILURE);
-#endif
-
-    } else if (keymatch(arg, "copy", 1)) {
-      /* Select which extra markers to copy. */
-      if (++argn >= argc)	/* advance to next argument */
-	usage();
-      if (keymatch(argv[argn], "none", 1)) {
-	copyoption = JCOPYOPT_NONE;
-      } else if (keymatch(argv[argn], "comments", 1)) {
-	copyoption = JCOPYOPT_COMMENTS;
-      } else if (keymatch(argv[argn], "all", 1)) {
-	copyoption = JCOPYOPT_ALL;
-      } else
-	usage();
-
-    } else if (keymatch(arg, "debug", 1) || keymatch(arg, "verbose", 1)) {
-      /* Enable debug printouts. */
-      /* On first -d, print version identification */
-      static boolean printed_version = FALSE;
-
-      if (! printed_version) {
-	fprintf(stderr, "Independent JPEG Group's JPEGTRAN, version %s\n%s\n",
-		JVERSION, JCOPYRIGHT);
-	printed_version = TRUE;
-      }
-      cinfo->err->trace_level++;
-
-    } else if (keymatch(arg, "flip", 1)) {
-      /* Mirror left-right or top-bottom. */
-      if (++argn >= argc)	/* advance to next argument */
-	usage();
-      if (keymatch(argv[argn], "horizontal", 1))
-	select_transform(JXFORM_FLIP_H);
-      else if (keymatch(argv[argn], "vertical", 1))
-	select_transform(JXFORM_FLIP_V);
-      else
-	usage();
-
-    } else if (keymatch(arg, "grayscale", 1) || keymatch(arg, "greyscale",1)) {
-      /* Force to grayscale. */
-#if TRANSFORMS_SUPPORTED
-      transformoption.force_grayscale = TRUE;
-#else
-      select_transform(JXFORM_NONE);	/* force an error */
-#endif
-
-    } else if (keymatch(arg, "maxmemory", 3)) {
-      /* Maximum memory in Kb (or Mb with 'm'). */
-      long lval;
-      char ch = 'x';
-
-      if (++argn >= argc)	/* advance to next argument */
-	usage();
-      if (sscanf(argv[argn], "%ld%c", &lval, &ch) < 1)
-	usage();
-      if (ch == 'm' || ch == 'M')
-	lval *= 1000L;
-      cinfo->mem->max_memory_to_use = lval * 1000L;
-
-    } else if (keymatch(arg, "optimize", 1) || keymatch(arg, "optimise", 1)) {
-      /* Enable entropy parm optimization. */
-#ifdef ENTROPY_OPT_SUPPORTED
-      cinfo->optimize_coding = TRUE;
-#else
-      fprintf(stderr, "%s: sorry, entropy optimization was not compiled\n",
-	      progname);
-      exit(EXIT_FAILURE);
-#endif
-
-    } else if (keymatch(arg, "outfile", 4)) {
-      /* Set output file name. */
-      if (++argn >= argc)	/* advance to next argument */
-	usage();
-      outfilename = argv[argn];	/* save it away for later use */
-
-    } else if (keymatch(arg, "progressive", 1)) {
-      /* Select simple progressive mode. */
-#ifdef C_PROGRESSIVE_SUPPORTED
-      simple_progressive = TRUE;
-      /* We must postpone execution until num_components is known. */
-#else
-      fprintf(stderr, "%s: sorry, progressive output was not compiled\n",
-	      progname);
-      exit(EXIT_FAILURE);
-#endif
-
-    } else if (keymatch(arg, "restart", 1)) {
-      /* Restart interval in MCU rows (or in MCUs with 'b'). */
-      long lval;
-      char ch = 'x';
-
-      if (++argn >= argc)	/* advance to next argument */
-	usage();
-      if (sscanf(argv[argn], "%ld%c", &lval, &ch) < 1)
-	usage();
-      if (lval < 0 || lval > 65535L)
-	usage();
-      if (ch == 'b' || ch == 'B') {
-	cinfo->restart_interval = (unsigned int) lval;
-	cinfo->restart_in_rows = 0; /* else prior '-restart n' overrides me */
-      } else {
-	cinfo->restart_in_rows = (int) lval;
-	/* restart_interval will be computed during startup */
-      }
-
-    } else if (keymatch(arg, "rotate", 2)) {
-      /* Rotate 90, 180, or 270 degrees (measured clockwise). */
-      if (++argn >= argc)	/* advance to next argument */
-	usage();
-      if (keymatch(argv[argn], "90", 2))
-	select_transform(JXFORM_ROT_90);
-      else if (keymatch(argv[argn], "180", 3))
-	select_transform(JXFORM_ROT_180);
-      else if (keymatch(argv[argn], "270", 3))
-	select_transform(JXFORM_ROT_270);
-      else
-	usage();
-
-    } else if (keymatch(arg, "scans", 1)) {
-      /* Set scan script. */
-#ifdef C_MULTISCAN_FILES_SUPPORTED
-      if (++argn >= argc)	/* advance to next argument */
-	usage();
-      scansarg = argv[argn];
-      /* We must postpone reading the file in case -progressive appears. */
-#else
-      fprintf(stderr, "%s: sorry, multi-scan output was not compiled\n",
-	      progname);
-      exit(EXIT_FAILURE);
-#endif
-
-    } else if (keymatch(arg, "transpose", 1)) {
-      /* Transpose (across UL-to-LR axis). */
-      select_transform(JXFORM_TRANSPOSE);
-
-    } else if (keymatch(arg, "transverse", 6)) {
-      /* Transverse transpose (across UR-to-LL axis). */
-      select_transform(JXFORM_TRANSVERSE);
-
-    } else if (keymatch(arg, "trim", 3)) {
-      /* Trim off any partial edge MCUs that the transform can't handle. */
-      transformoption.trim = TRUE;
-
-    } else {
-      usage();			/* bogus switch */
-    }
-  }
-
-  /* Post-switch-scanning cleanup */
-
-  if (for_real) {
-
-#ifdef C_PROGRESSIVE_SUPPORTED
-    if (simple_progressive)	/* process -progressive; -scans can override */
-      jpeg_simple_progression(cinfo);
-#endif
-
-#ifdef C_MULTISCAN_FILES_SUPPORTED
-    if (scansarg != NULL)	/* process -scans if it was present */
-      if (! read_scan_script(cinfo, scansarg))
-	usage();
-#endif
-  }
-
-  return argn;			/* return index of next arg (file name) */
-}
-
-
-/*
- * The main program.
- */
-
-int
-main (int argc, char **argv)
-{
-  struct jpeg_decompress_struct srcinfo;
-  struct jpeg_compress_struct dstinfo;
-  struct jpeg_error_mgr jsrcerr, jdsterr;
-#ifdef PROGRESS_REPORT
-  struct cdjpeg_progress_mgr progress;
-#endif
-  jvirt_barray_ptr * src_coef_arrays;
-  jvirt_barray_ptr * dst_coef_arrays;
-  int file_index;
-  FILE * input_file;
-  FILE * output_file;
-
-  /* On Mac, fetch a command line. */
-#ifdef USE_CCOMMAND
-  argc = ccommand(&argv);
-#endif
-
-  progname = argv[0];
-  if (progname == NULL || progname[0] == 0)
-    progname = "jpegtran";	/* in case C library doesn't provide it */
-
-  /* Initialize the JPEG decompression object with default error handling. */
-  srcinfo.err = jpeg_std_error(&jsrcerr);
-  jpeg_create_decompress(&srcinfo);
-  /* Initialize the JPEG compression object with default error handling. */
-  dstinfo.err = jpeg_std_error(&jdsterr);
-  jpeg_create_compress(&dstinfo);
-
-  /* Now safe to enable signal catcher.
-   * Note: we assume only the decompression object will have virtual arrays.
-   */
-#ifdef NEED_SIGNAL_CATCHER
-  enable_signal_catcher((j_common_ptr) &srcinfo);
-#endif
-
-  /* Scan command line to find file names.
-   * It is convenient to use just one switch-parsing routine, but the switch
-   * values read here are mostly ignored; we will rescan the switches after
-   * opening the input file.  Also note that most of the switches affect the
-   * destination JPEG object, so we parse into that and then copy over what
-   * needs to affects the source too.
-   */
-
-  file_index = parse_switches(&dstinfo, argc, argv, 0, FALSE);
-  jsrcerr.trace_level = jdsterr.trace_level;
-  srcinfo.mem->max_memory_to_use = dstinfo.mem->max_memory_to_use;
-
-#ifdef TWO_FILE_COMMANDLINE
-  /* Must have either -outfile switch or explicit output file name */
-  if (outfilename == NULL) {
-    if (file_index != argc-2) {
-      fprintf(stderr, "%s: must name one input and one output file\n",
-	      progname);
-      usage();
-    }
-    outfilename = argv[file_index+1];
-  } else {
-    if (file_index != argc-1) {
-      fprintf(stderr, "%s: must name one input and one output file\n",
-	      progname);
-      usage();
-    }
-  }
-#else
-  /* Unix style: expect zero or one file name */
-  if (file_index < argc-1) {
-    fprintf(stderr, "%s: only one input file\n", progname);
-    usage();
-  }
-#endif /* TWO_FILE_COMMANDLINE */
-
-  /* Open the input file. */
-  if (file_index < argc) {
-    if ((input_file = fopen(argv[file_index], READ_BINARY)) == NULL) {
-      fprintf(stderr, "%s: can't open %s\n", progname, argv[file_index]);
-      exit(EXIT_FAILURE);
-    }
-  } else {
-    /* default input file is stdin */
-    input_file = read_stdin();
-  }
-
-  /* Open the output file. */
-  if (outfilename != NULL) {
-    if ((output_file = fopen(outfilename, WRITE_BINARY)) == NULL) {
-      fprintf(stderr, "%s: can't open %s\n", progname, outfilename);
-      exit(EXIT_FAILURE);
-    }
-  } else {
-    /* default output file is stdout */
-    output_file = write_stdout();
-  }
-
-#ifdef PROGRESS_REPORT
-  start_progress_monitor((j_common_ptr) &dstinfo, &progress);
-#endif
-
-  /* Specify data source for decompression */
-  jpeg_stdio_src(&srcinfo, input_file);
-
-  /* Enable saving of extra markers that we want to copy */
-  jcopy_markers_setup(&srcinfo, copyoption);
-
-  /* Read file header */
-  (void) jpeg_read_header(&srcinfo, TRUE);
-
-  /* Any space needed by a transform option must be requested before
-   * jpeg_read_coefficients so that memory allocation will be done right.
-   */
-#if TRANSFORMS_SUPPORTED
-  jtransform_request_workspace(&srcinfo, &transformoption);
-#endif
-
-  /* Read source file as DCT coefficients */
-  src_coef_arrays = jpeg_read_coefficients(&srcinfo);
-
-  /* Initialize destination compression parameters from source values */
-  jpeg_copy_critical_parameters(&srcinfo, &dstinfo);
-
-  /* Adjust destination parameters if required by transform options;
-   * also find out which set of coefficient arrays will hold the output.
-   */
-#if TRANSFORMS_SUPPORTED
-  dst_coef_arrays = jtransform_adjust_parameters(&srcinfo, &dstinfo,
-						 src_coef_arrays,
-						 &transformoption);
-#else
-  dst_coef_arrays = src_coef_arrays;
-#endif
-
-  /* Adjust default compression parameters by re-parsing the options */
-  file_index = parse_switches(&dstinfo, argc, argv, 0, TRUE);
-
-  /* Specify data destination for compression */
-  jpeg_stdio_dest(&dstinfo, output_file);
-
-  /* Start compressor (note no image data is actually written here) */
-  jpeg_write_coefficients(&dstinfo, dst_coef_arrays);
-
-  /* Copy to the output file any extra markers that we want to preserve */
-  jcopy_markers_execute(&srcinfo, &dstinfo, copyoption);
-
-  /* Execute image transformation, if any */
-#if TRANSFORMS_SUPPORTED
-  jtransform_execute_transformation(&srcinfo, &dstinfo,
-				    src_coef_arrays,
-				    &transformoption);
-#endif
-
-  /* Finish compression and release memory */
-  jpeg_finish_compress(&dstinfo);
-  jpeg_destroy_compress(&dstinfo);
-  (void) jpeg_finish_decompress(&srcinfo);
-  jpeg_destroy_decompress(&srcinfo);
-
-  /* Close files, if we opened them */
-  if (input_file != stdin)
-    fclose(input_file);
-  if (output_file != stdout)
-    fclose(output_file);
-
-#ifdef PROGRESS_REPORT
-  end_progress_monitor((j_common_ptr) &dstinfo);
-#endif
-
-  /* All done. */
-  exit(jsrcerr.num_warnings + jdsterr.num_warnings ?EXIT_WARNING:EXIT_SUCCESS);
-  return 0;			/* suppress no-return-value warnings */
-}
diff --git a/src/jpeg-8c/README b/src/jpeg-8c/README
new file mode 100644
index 00000000..451265d7
--- /dev/null
+++ b/src/jpeg-8c/README
@@ -0,0 +1,326 @@
+The Independent JPEG Group's JPEG software
+==========================================
+
+README for release 8c of 16-Jan-2011
+====================================
+
+This distribution contains the eighth public release of the Independent JPEG
+Group's free JPEG software.  You are welcome to redistribute this software and
+to use it for any purpose, subject to the conditions under LEGAL ISSUES, below.
+
+This software is the work of Tom Lane, Guido Vollbeding, Philip Gladstone,
+Bill Allombert, Jim Boucher, Lee Crocker, Bob Friesenhahn, Ben Jackson,
+Julian Minguillon, Luis Ortiz, George Phillips, Davide Rossi, Ge' Weijers,
+and other members of the Independent JPEG Group.
+
+IJG is not affiliated with the official ISO JPEG standards committee.
+
+
+DOCUMENTATION ROADMAP
+=====================
+
+This file contains the following sections:
+
+OVERVIEW            General description of JPEG and the IJG software.
+LEGAL ISSUES        Copyright, lack of warranty, terms of distribution.
+REFERENCES          Where to learn more about JPEG.
+ARCHIVE LOCATIONS   Where to find newer versions of this software.
+ACKNOWLEDGMENTS     Special thanks.
+FILE FORMAT WARS    Software *not* to get.
+TO DO               Plans for future IJG releases.
+
+Other documentation files in the distribution are:
+
+User documentation:
+  install.txt       How to configure and install the IJG software.
+  usage.txt         Usage instructions for cjpeg, djpeg, jpegtran,
+                    rdjpgcom, and wrjpgcom.
+  *.1               Unix-style man pages for programs (same info as usage.txt).
+  wizard.txt        Advanced usage instructions for JPEG wizards only.
+  change.log        Version-to-version change highlights.
+Programmer and internal documentation:
+  libjpeg.txt       How to use the JPEG library in your own programs.
+  example.c         Sample code for calling the JPEG library.
+  structure.txt     Overview of the JPEG library's internal structure.
+  filelist.txt      Road map of IJG files.
+  coderules.txt     Coding style rules --- please read if you contribute code.
+
+Please read at least the files install.txt and usage.txt.  Some information
+can also be found in the JPEG FAQ (Frequently Asked Questions) article.  See
+ARCHIVE LOCATIONS below to find out where to obtain the FAQ article.
+
+If you want to understand how the JPEG code works, we suggest reading one or
+more of the REFERENCES, then looking at the documentation files (in roughly
+the order listed) before diving into the code.
+
+
+OVERVIEW
+========
+
+This package contains C software to implement JPEG image encoding, decoding,
+and transcoding.  JPEG (pronounced "jay-peg") is a standardized compression
+method for full-color and gray-scale images.
+
+This software implements JPEG baseline, extended-sequential, and progressive
+compression processes.  Provision is made for supporting all variants of these
+processes, although some uncommon parameter settings aren't implemented yet.
+We have made no provision for supporting the hierarchical or lossless
+processes defined in the standard.
+
+We provide a set of library routines for reading and writing JPEG image files,
+plus two sample applications "cjpeg" and "djpeg", which use the library to
+perform conversion between JPEG and some other popular image file formats.
+The library is intended to be reused in other applications.
+
+In order to support file conversion and viewing software, we have included
+considerable functionality beyond the bare JPEG coding/decoding capability;
+for example, the color quantization modules are not strictly part of JPEG
+decoding, but they are essential for output to colormapped file formats or
+colormapped displays.  These extra functions can be compiled out of the
+library if not required for a particular application.
+
+We have also included "jpegtran", a utility for lossless transcoding between
+different JPEG processes, and "rdjpgcom" and "wrjpgcom", two simple
+applications for inserting and extracting textual comments in JFIF files.
+
+The emphasis in designing this software has been on achieving portability and
+flexibility, while also making it fast enough to be useful.  In particular,
+the software is not intended to be read as a tutorial on JPEG.  (See the
+REFERENCES section for introductory material.)  Rather, it is intended to
+be reliable, portable, industrial-strength code.  We do not claim to have
+achieved that goal in every aspect of the software, but we strive for it.
+
+We welcome the use of this software as a component of commercial products.
+No royalty is required, but we do ask for an acknowledgement in product
+documentation, as described under LEGAL ISSUES.
+
+
+LEGAL ISSUES
+============
+
+In plain English:
+
+1. We don't promise that this software works.  (But if you find any bugs,
+   please let us know!)
+2. You can use this software for whatever you want.  You don't have to pay us.
+3. You may not pretend that you wrote this software.  If you use it in a
+   program, you must acknowledge somewhere in your documentation that
+   you've used the IJG code.
+
+In legalese:
+
+The authors make NO WARRANTY or representation, either express or implied,
+with respect to this software, its quality, accuracy, merchantability, or
+fitness for a particular purpose.  This software is provided "AS IS", and you,
+its user, assume the entire risk as to its quality and accuracy.
+
+This software is copyright (C) 1991-2011, Thomas G. Lane, Guido Vollbeding.
+All Rights Reserved except as specified below.
+
+Permission is hereby granted to use, copy, modify, and distribute this
+software (or portions thereof) for any purpose, without fee, subject to these
+conditions:
+(1) If any part of the source code for this software is distributed, then this
+README file must be included, with this copyright and no-warranty notice
+unaltered; and any additions, deletions, or changes to the original files
+must be clearly indicated in accompanying documentation.
+(2) If only executable code is distributed, then the accompanying
+documentation must state that "this software is based in part on the work of
+the Independent JPEG Group".
+(3) Permission for use of this software is granted only if the user accepts
+full responsibility for any undesirable consequences; the authors accept
+NO LIABILITY for damages of any kind.
+
+These conditions apply to any software derived from or based on the IJG code,
+not just to the unmodified library.  If you use our work, you ought to
+acknowledge us.
+
+Permission is NOT granted for the use of any IJG author's name or company name
+in advertising or publicity relating to this software or products derived from
+it.  This software may be referred to only as "the Independent JPEG Group's
+software".
+
+We specifically permit and encourage the use of this software as the basis of
+commercial products, provided that all warranty or liability claims are
+assumed by the product vendor.
+
+
+ansi2knr.c is included in this distribution by permission of L. Peter Deutsch,
+sole proprietor of its copyright holder, Aladdin Enterprises of Menlo Park, CA.
+ansi2knr.c is NOT covered by the above copyright and conditions, but instead
+by the usual distribution terms of the Free Software Foundation; principally,
+that you must include source code if you redistribute it.  (See the file
+ansi2knr.c for full details.)  However, since ansi2knr.c is not needed as part
+of any program generated from the IJG code, this does not limit you more than
+the foregoing paragraphs do.
+
+The Unix configuration script "configure" was produced with GNU Autoconf.
+It is copyright by the Free Software Foundation but is freely distributable.
+The same holds for its supporting scripts (config.guess, config.sub,
+ltmain.sh).  Another support script, install-sh, is copyright by X Consortium
+but is also freely distributable.
+
+The IJG distribution formerly included code to read and write GIF files.
+To avoid entanglement with the Unisys LZW patent, GIF reading support has
+been removed altogether, and the GIF writer has been simplified to produce
+"uncompressed GIFs".  This technique does not use the LZW algorithm; the
+resulting GIF files are larger than usual, but are readable by all standard
+GIF decoders.
+
+We are required to state that
+    "The Graphics Interchange Format(c) is the Copyright property of
+    CompuServe Incorporated.  GIF(sm) is a Service Mark property of
+    CompuServe Incorporated."
+
+
+REFERENCES
+==========
+
+We recommend reading one or more of these references before trying to
+understand the innards of the JPEG software.
+
+The best short technical introduction to the JPEG compression algorithm is
+	Wallace, Gregory K.  "The JPEG Still Picture Compression Standard",
+	Communications of the ACM, April 1991 (vol. 34 no. 4), pp. 30-44.
+(Adjacent articles in that issue discuss MPEG motion picture compression,
+applications of JPEG, and related topics.)  If you don't have the CACM issue
+handy, a PostScript file containing a revised version of Wallace's article is
+available at http://www.ijg.org/files/wallace.ps.gz.  The file (actually
+a preprint for an article that appeared in IEEE Trans. Consumer Electronics)
+omits the sample images that appeared in CACM, but it includes corrections
+and some added material.  Note: the Wallace article is copyright ACM and IEEE,
+and it may not be used for commercial purposes.
+
+A somewhat less technical, more leisurely introduction to JPEG can be found in
+"The Data Compression Book" by Mark Nelson and Jean-loup Gailly, published by
+M&T Books (New York), 2nd ed. 1996, ISBN 1-55851-434-1.  This book provides
+good explanations and example C code for a multitude of compression methods
+including JPEG.  It is an excellent source if you are comfortable reading C
+code but don't know much about data compression in general.  The book's JPEG
+sample code is far from industrial-strength, but when you are ready to look
+at a full implementation, you've got one here...
+
+The best currently available description of JPEG is the textbook "JPEG Still
+Image Data Compression Standard" by William B. Pennebaker and Joan L.
+Mitchell, published by Van Nostrand Reinhold, 1993, ISBN 0-442-01272-1.
+Price US$59.95, 638 pp.  The book includes the complete text of the ISO JPEG
+standards (DIS 10918-1 and draft DIS 10918-2).
+Although this is by far the most detailed and comprehensive exposition of
+JPEG publicly available, we point out that it is still missing an explanation
+of the most essential properties and algorithms of the underlying DCT
+technology.
+If you think that you know about DCT-based JPEG after reading this book,
+then you are in delusion.  The real fundamentals and corresponding potential
+of DCT-based JPEG are not publicly known so far, and that is the reason for
+all the mistaken developments taking place in the image coding domain.
+
+The original JPEG standard is divided into two parts, Part 1 being the actual
+specification, while Part 2 covers compliance testing methods.  Part 1 is
+titled "Digital Compression and Coding of Continuous-tone Still Images,
+Part 1: Requirements and guidelines" and has document numbers ISO/IEC IS
+10918-1, ITU-T T.81.  Part 2 is titled "Digital Compression and Coding of
+Continuous-tone Still Images, Part 2: Compliance testing" and has document
+numbers ISO/IEC IS 10918-2, ITU-T T.83.
+IJG JPEG 8 introduces an implementation of the JPEG SmartScale extension
+which is specified in a contributed document at ITU and ISO with title "ITU-T
+JPEG-Plus Proposal for Extending ITU-T T.81 for Advanced Image Coding", April
+2006, Geneva, Switzerland.  The latest version of the document is Revision 3.
+
+The JPEG standard does not specify all details of an interchangeable file
+format.  For the omitted details we follow the "JFIF" conventions, revision
+1.02.  JFIF 1.02 has been adopted as an Ecma International Technical Report
+and thus received a formal publication status.  It is available as a free
+download in PDF format from
+http://www.ecma-international.org/publications/techreports/E-TR-098.htm.
+A PostScript version of the JFIF document is available at
+http://www.ijg.org/files/jfif.ps.gz.  There is also a plain text version at
+http://www.ijg.org/files/jfif.txt.gz, but it is missing the figures.
+
+The TIFF 6.0 file format specification can be obtained by FTP from
+ftp://ftp.sgi.com/graphics/tiff/TIFF6.ps.gz.  The JPEG incorporation scheme
+found in the TIFF 6.0 spec of 3-June-92 has a number of serious problems.
+IJG does not recommend use of the TIFF 6.0 design (TIFF Compression tag 6).
+Instead, we recommend the JPEG design proposed by TIFF Technical Note #2
+(Compression tag 7).  Copies of this Note can be obtained from
+http://www.ijg.org/files/.  It is expected that the next revision
+of the TIFF spec will replace the 6.0 JPEG design with the Note's design.
+Although IJG's own code does not support TIFF/JPEG, the free libtiff library
+uses our library to implement TIFF/JPEG per the Note.
+
+
+ARCHIVE LOCATIONS
+=================
+
+The "official" archive site for this software is www.ijg.org.
+The most recent released version can always be found there in
+directory "files".  This particular version will be archived as
+http://www.ijg.org/files/jpegsrc.v8c.tar.gz, and in Windows-compatible
+"zip" archive format as http://www.ijg.org/files/jpegsr8c.zip.
+
+The JPEG FAQ (Frequently Asked Questions) article is a source of some
+general information about JPEG.
+It is available on the World Wide Web at http://www.faqs.org/faqs/jpeg-faq/
+and other news.answers archive sites, including the official news.answers
+archive at rtfm.mit.edu: ftp://rtfm.mit.edu/pub/usenet/news.answers/jpeg-faq/.
+If you don't have Web or FTP access, send e-mail to mail-server@rtfm.mit.edu
+with body
+	send usenet/news.answers/jpeg-faq/part1
+	send usenet/news.answers/jpeg-faq/part2
+
+
+ACKNOWLEDGMENTS
+===============
+
+Thank to Juergen Bruder for providing me with a copy of the common DCT
+algorithm article, only to find out that I had come to the same result
+in a more direct and comprehensible way with a more generative approach.
+
+Thank to Istvan Sebestyen and Joan L. Mitchell for inviting me to the
+ITU JPEG (Study Group 16) meeting in Geneva, Switzerland.
+
+Thank to Thomas Wiegand and Gary Sullivan for inviting me to the
+Joint Video Team (MPEG & ITU) meeting in Geneva, Switzerland.
+
+Thank to John Korejwa and Massimo Ballerini for inviting me to
+fruitful consultations in Boston, MA and Milan, Italy.
+
+Thank to Hendrik Elstner, Roland Fassauer, Simone Zuck, Guenther
+Maier-Gerber, Walter Stoeber, Fred Schmitz, and Norbert Braunagel
+for corresponding business development.
+
+Thank to Nico Zschach and Dirk Stelling of the technical support team
+at the Digital Images company in Halle for providing me with extra
+equipment for configuration tests.
+
+Thank to Richard F. Lyon (then of Foveon Inc.) for fruitful
+communication about JPEG configuration in Sigma Photo Pro software.
+
+Thank to Andrew Finkenstadt for hosting the ijg.org site.
+
+Last but not least special thank to Thomas G. Lane for the original
+design and development of this singular software package.
+
+
+FILE FORMAT WARS
+================
+
+The ISO JPEG standards committee actually promotes different formats like
+"JPEG 2000" or "JPEG XR" which are incompatible with original DCT-based
+JPEG and which are based on faulty technologies.  IJG therefore does not
+and will not support such momentary mistakes (see REFERENCES).
+We have little or no sympathy for the promotion of these formats.  Indeed,
+one of the original reasons for developing this free software was to help
+force convergence on common, interoperable format standards for JPEG files.
+Don't use an incompatible file format!
+(In any case, our decoder will remain capable of reading existing JPEG
+image files indefinitely.)
+
+
+TO DO
+=====
+
+Version 8 is the first release of a new generation JPEG standard
+to overcome the limitations of the original JPEG specification.
+More features are being prepared for coming releases...
+
+Please send bug reports, offers of help, etc. to jpeg-info@uc.ag.
diff --git a/src/jpeg-8c/jaricom.c b/src/jpeg-8c/jaricom.c
new file mode 100644
index 00000000..f43e2ea7
--- /dev/null
+++ b/src/jpeg-8c/jaricom.c
@@ -0,0 +1,153 @@
+/*
+ * jaricom.c
+ *
+ * Developed 1997-2009 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains probability estimation tables for common use in
+ * arithmetic entropy encoding and decoding routines.
+ *
+ * This data represents Table D.2 in the JPEG spec (ISO/IEC IS 10918-1
+ * and CCITT Recommendation ITU-T T.81) and Table 24 in the JBIG spec
+ * (ISO/IEC IS 11544 and CCITT Recommendation ITU-T T.82).
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+/* The following #define specifies the packing of the four components
+ * into the compact INT32 representation.
+ * Note that this formula must match the actual arithmetic encoder
+ * and decoder implementation.  The implementation has to be changed
+ * if this formula is changed.
+ * The current organization is leaned on Markus Kuhn's JBIG
+ * implementation (jbig_tab.c).
+ */
+
+#define V(i,a,b,c,d) (((INT32)a << 16) | ((INT32)c << 8) | ((INT32)d << 7) | b)
+
+const INT32 jpeg_aritab[113+1] = {
+/*
+ * Index, Qe_Value, Next_Index_LPS, Next_Index_MPS, Switch_MPS
+ */
+  V(   0, 0x5a1d,   1,   1, 1 ),
+  V(   1, 0x2586,  14,   2, 0 ),
+  V(   2, 0x1114,  16,   3, 0 ),
+  V(   3, 0x080b,  18,   4, 0 ),
+  V(   4, 0x03d8,  20,   5, 0 ),
+  V(   5, 0x01da,  23,   6, 0 ),
+  V(   6, 0x00e5,  25,   7, 0 ),
+  V(   7, 0x006f,  28,   8, 0 ),
+  V(   8, 0x0036,  30,   9, 0 ),
+  V(   9, 0x001a,  33,  10, 0 ),
+  V(  10, 0x000d,  35,  11, 0 ),
+  V(  11, 0x0006,   9,  12, 0 ),
+  V(  12, 0x0003,  10,  13, 0 ),
+  V(  13, 0x0001,  12,  13, 0 ),
+  V(  14, 0x5a7f,  15,  15, 1 ),
+  V(  15, 0x3f25,  36,  16, 0 ),
+  V(  16, 0x2cf2,  38,  17, 0 ),
+  V(  17, 0x207c,  39,  18, 0 ),
+  V(  18, 0x17b9,  40,  19, 0 ),
+  V(  19, 0x1182,  42,  20, 0 ),
+  V(  20, 0x0cef,  43,  21, 0 ),
+  V(  21, 0x09a1,  45,  22, 0 ),
+  V(  22, 0x072f,  46,  23, 0 ),
+  V(  23, 0x055c,  48,  24, 0 ),
+  V(  24, 0x0406,  49,  25, 0 ),
+  V(  25, 0x0303,  51,  26, 0 ),
+  V(  26, 0x0240,  52,  27, 0 ),
+  V(  27, 0x01b1,  54,  28, 0 ),
+  V(  28, 0x0144,  56,  29, 0 ),
+  V(  29, 0x00f5,  57,  30, 0 ),
+  V(  30, 0x00b7,  59,  31, 0 ),
+  V(  31, 0x008a,  60,  32, 0 ),
+  V(  32, 0x0068,  62,  33, 0 ),
+  V(  33, 0x004e,  63,  34, 0 ),
+  V(  34, 0x003b,  32,  35, 0 ),
+  V(  35, 0x002c,  33,   9, 0 ),
+  V(  36, 0x5ae1,  37,  37, 1 ),
+  V(  37, 0x484c,  64,  38, 0 ),
+  V(  38, 0x3a0d,  65,  39, 0 ),
+  V(  39, 0x2ef1,  67,  40, 0 ),
+  V(  40, 0x261f,  68,  41, 0 ),
+  V(  41, 0x1f33,  69,  42, 0 ),
+  V(  42, 0x19a8,  70,  43, 0 ),
+  V(  43, 0x1518,  72,  44, 0 ),
+  V(  44, 0x1177,  73,  45, 0 ),
+  V(  45, 0x0e74,  74,  46, 0 ),
+  V(  46, 0x0bfb,  75,  47, 0 ),
+  V(  47, 0x09f8,  77,  48, 0 ),
+  V(  48, 0x0861,  78,  49, 0 ),
+  V(  49, 0x0706,  79,  50, 0 ),
+  V(  50, 0x05cd,  48,  51, 0 ),
+  V(  51, 0x04de,  50,  52, 0 ),
+  V(  52, 0x040f,  50,  53, 0 ),
+  V(  53, 0x0363,  51,  54, 0 ),
+  V(  54, 0x02d4,  52,  55, 0 ),
+  V(  55, 0x025c,  53,  56, 0 ),
+  V(  56, 0x01f8,  54,  57, 0 ),
+  V(  57, 0x01a4,  55,  58, 0 ),
+  V(  58, 0x0160,  56,  59, 0 ),
+  V(  59, 0x0125,  57,  60, 0 ),
+  V(  60, 0x00f6,  58,  61, 0 ),
+  V(  61, 0x00cb,  59,  62, 0 ),
+  V(  62, 0x00ab,  61,  63, 0 ),
+  V(  63, 0x008f,  61,  32, 0 ),
+  V(  64, 0x5b12,  65,  65, 1 ),
+  V(  65, 0x4d04,  80,  66, 0 ),
+  V(  66, 0x412c,  81,  67, 0 ),
+  V(  67, 0x37d8,  82,  68, 0 ),
+  V(  68, 0x2fe8,  83,  69, 0 ),
+  V(  69, 0x293c,  84,  70, 0 ),
+  V(  70, 0x2379,  86,  71, 0 ),
+  V(  71, 0x1edf,  87,  72, 0 ),
+  V(  72, 0x1aa9,  87,  73, 0 ),
+  V(  73, 0x174e,  72,  74, 0 ),
+  V(  74, 0x1424,  72,  75, 0 ),
+  V(  75, 0x119c,  74,  76, 0 ),
+  V(  76, 0x0f6b,  74,  77, 0 ),
+  V(  77, 0x0d51,  75,  78, 0 ),
+  V(  78, 0x0bb6,  77,  79, 0 ),
+  V(  79, 0x0a40,  77,  48, 0 ),
+  V(  80, 0x5832,  80,  81, 1 ),
+  V(  81, 0x4d1c,  88,  82, 0 ),
+  V(  82, 0x438e,  89,  83, 0 ),
+  V(  83, 0x3bdd,  90,  84, 0 ),
+  V(  84, 0x34ee,  91,  85, 0 ),
+  V(  85, 0x2eae,  92,  86, 0 ),
+  V(  86, 0x299a,  93,  87, 0 ),
+  V(  87, 0x2516,  86,  71, 0 ),
+  V(  88, 0x5570,  88,  89, 1 ),
+  V(  89, 0x4ca9,  95,  90, 0 ),
+  V(  90, 0x44d9,  96,  91, 0 ),
+  V(  91, 0x3e22,  97,  92, 0 ),
+  V(  92, 0x3824,  99,  93, 0 ),
+  V(  93, 0x32b4,  99,  94, 0 ),
+  V(  94, 0x2e17,  93,  86, 0 ),
+  V(  95, 0x56a8,  95,  96, 1 ),
+  V(  96, 0x4f46, 101,  97, 0 ),
+  V(  97, 0x47e5, 102,  98, 0 ),
+  V(  98, 0x41cf, 103,  99, 0 ),
+  V(  99, 0x3c3d, 104, 100, 0 ),
+  V( 100, 0x375e,  99,  93, 0 ),
+  V( 101, 0x5231, 105, 102, 0 ),
+  V( 102, 0x4c0f, 106, 103, 0 ),
+  V( 103, 0x4639, 107, 104, 0 ),
+  V( 104, 0x415e, 103,  99, 0 ),
+  V( 105, 0x5627, 105, 106, 1 ),
+  V( 106, 0x50e7, 108, 107, 0 ),
+  V( 107, 0x4b85, 109, 103, 0 ),
+  V( 108, 0x5597, 110, 109, 0 ),
+  V( 109, 0x504f, 111, 107, 0 ),
+  V( 110, 0x5a10, 110, 111, 1 ),
+  V( 111, 0x5522, 112, 109, 0 ),
+  V( 112, 0x59eb, 112, 111, 1 ),
+/*
+ * This last entry is used for fixed probability estimate of 0.5
+ * as recommended in Section 10.3 Table 5 of ITU-T Rec. T.851.
+ */
+  V( 113, 0x5a1d, 113, 113, 0 )
+};
diff --git a/src/jpeg-6b/jcapimin.c b/src/jpeg-8c/jcapimin.c
index 54fb8c58..639ce86f 100644
--- a/src/jpeg-6b/jcapimin.c
+++ b/src/jpeg-8c/jcapimin.c
@@ -2,6 +2,7 @@
  * jcapimin.c
  *
  * Copyright (C) 1994-1998, Thomas G. Lane.
+ * Modified 2003-2010 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -63,14 +64,21 @@ jpeg_CreateCompress (j_compress_ptr cinfo, int version, size_t structsize)
 
   cinfo->comp_info = NULL;
 
-  for (i = 0; i < NUM_QUANT_TBLS; i++)
+  for (i = 0; i < NUM_QUANT_TBLS; i++) {
     cinfo->quant_tbl_ptrs[i] = NULL;
+    cinfo->q_scale_factor[i] = 100;
+  }
 
   for (i = 0; i < NUM_HUFF_TBLS; i++) {
     cinfo->dc_huff_tbl_ptrs[i] = NULL;
     cinfo->ac_huff_tbl_ptrs[i] = NULL;
   }
 
+  /* Must do it here for emit_dqt in case jpeg_write_tables is used */
+  cinfo->block_size = DCTSIZE;
+  cinfo->natural_order = jpeg_natural_order;
+  cinfo->lim_Se = DCTSIZE2-1;
+
   cinfo->script_space = NULL;
 
   cinfo->input_gamma = 1.0;	/* in case application forgets */
diff --git a/src/jpeg-6b/jcapistd.c b/src/jpeg-8c/jcapistd.c
index c0320b1b..c0320b1b 100644
--- a/src/jpeg-6b/jcapistd.c
+++ b/src/jpeg-8c/jcapistd.c
diff --git a/src/jpeg-8c/jcarith.c b/src/jpeg-8c/jcarith.c
new file mode 100644
index 00000000..0b7ea55d
--- /dev/null
+++ b/src/jpeg-8c/jcarith.c
@@ -0,0 +1,934 @@
+/*
+ * jcarith.c
+ *
+ * Developed 1997-2009 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains portable arithmetic entropy encoding routines for JPEG
+ * (implementing the ISO/IEC IS 10918-1 and CCITT Recommendation ITU-T T.81).
+ *
+ * Both sequential and progressive modes are supported in this single module.
+ *
+ * Suspension is not currently supported in this module.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Expanded entropy encoder object for arithmetic encoding. */
+
+typedef struct {
+  struct jpeg_entropy_encoder pub; /* public fields */
+
+  INT32 c; /* C register, base of coding interval, layout as in sec. D.1.3 */
+  INT32 a;               /* A register, normalized size of coding interval */
+  INT32 sc;        /* counter for stacked 0xFF values which might overflow */
+  INT32 zc;          /* counter for pending 0x00 output values which might *
+                          * be discarded at the end ("Pacman" termination) */
+  int ct;  /* bit shift counter, determines when next byte will be written */
+  int buffer;                /* buffer for most recent output byte != 0xFF */
+
+  int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
+  int dc_context[MAX_COMPS_IN_SCAN]; /* context index for DC conditioning */
+
+  unsigned int restarts_to_go;	/* MCUs left in this restart interval */
+  int next_restart_num;		/* next restart number to write (0-7) */
+
+  /* Pointers to statistics areas (these workspaces have image lifespan) */
+  unsigned char * dc_stats[NUM_ARITH_TBLS];
+  unsigned char * ac_stats[NUM_ARITH_TBLS];
+
+  /* Statistics bin for coding with fixed probability 0.5 */
+  unsigned char fixed_bin[4];
+} arith_entropy_encoder;
+
+typedef arith_entropy_encoder * arith_entropy_ptr;
+
+/* The following two definitions specify the allocation chunk size
+ * for the statistics area.
+ * According to sections F.1.4.4.1.3 and F.1.4.4.2, we need at least
+ * 49 statistics bins for DC, and 245 statistics bins for AC coding.
+ *
+ * We use a compact representation with 1 byte per statistics bin,
+ * thus the numbers directly represent byte sizes.
+ * This 1 byte per statistics bin contains the meaning of the MPS
+ * (more probable symbol) in the highest bit (mask 0x80), and the
+ * index into the probability estimation state machine table
+ * in the lower bits (mask 0x7F).
+ */
+
+#define DC_STAT_BINS 64
+#define AC_STAT_BINS 256
+
+/* NOTE: Uncomment the following #define if you want to use the
+ * given formula for calculating the AC conditioning parameter Kx
+ * for spectral selection progressive coding in section G.1.3.2
+ * of the spec (Kx = Kmin + SRL (8 + Se - Kmin) 4).
+ * Although the spec and P&M authors claim that this "has proven
+ * to give good results for 8 bit precision samples", I'm not
+ * convinced yet that this is really beneficial.
+ * Early tests gave only very marginal compression enhancements
+ * (a few - around 5 or so - bytes even for very large files),
+ * which would turn out rather negative if we'd suppress the
+ * DAC (Define Arithmetic Conditioning) marker segments for
+ * the default parameters in the future.
+ * Note that currently the marker writing module emits 12-byte
+ * DAC segments for a full-component scan in a color image.
+ * This is not worth worrying about IMHO. However, since the
+ * spec defines the default values to be used if the tables
+ * are omitted (unlike Huffman tables, which are required
+ * anyway), one might optimize this behaviour in the future,
+ * and then it would be disadvantageous to use custom tables if
+ * they don't provide sufficient gain to exceed the DAC size.
+ *
+ * On the other hand, I'd consider it as a reasonable result
+ * that the conditioning has no significant influence on the
+ * compression performance. This means that the basic
+ * statistical model is already rather stable.
+ *
+ * Thus, at the moment, we use the default conditioning values
+ * anyway, and do not use the custom formula.
+ *
+#define CALCULATE_SPECTRAL_CONDITIONING
+ */
+
+/* IRIGHT_SHIFT is like RIGHT_SHIFT, but works on int rather than INT32.
+ * We assume that int right shift is unsigned if INT32 right shift is,
+ * which should be safe.
+ */
+
+#ifdef RIGHT_SHIFT_IS_UNSIGNED
+#define ISHIFT_TEMPS	int ishift_temp;
+#define IRIGHT_SHIFT(x,shft)  \
+	((ishift_temp = (x)) < 0 ? \
+	 (ishift_temp >> (shft)) | ((~0) << (16-(shft))) : \
+	 (ishift_temp >> (shft)))
+#else
+#define ISHIFT_TEMPS
+#define IRIGHT_SHIFT(x,shft)	((x) >> (shft))
+#endif
+
+
+LOCAL(void)
+emit_byte (int val, j_compress_ptr cinfo)
+/* Write next output byte; we do not support suspension in this module. */
+{
+  struct jpeg_destination_mgr * dest = cinfo->dest;
+
+  *dest->next_output_byte++ = (JOCTET) val;
+  if (--dest->free_in_buffer == 0)
+    if (! (*dest->empty_output_buffer) (cinfo))
+      ERREXIT(cinfo, JERR_CANT_SUSPEND);
+}
+
+
+/*
+ * Finish up at the end of an arithmetic-compressed scan.
+ */
+
+METHODDEF(void)
+finish_pass (j_compress_ptr cinfo)
+{
+  arith_entropy_ptr e = (arith_entropy_ptr) cinfo->entropy;
+  INT32 temp;
+
+  /* Section D.1.8: Termination of encoding */
+
+  /* Find the e->c in the coding interval with the largest
+   * number of trailing zero bits */
+  if ((temp = (e->a - 1 + e->c) & 0xFFFF0000L) < e->c)
+    e->c = temp + 0x8000L;
+  else
+    e->c = temp;
+  /* Send remaining bytes to output */
+  e->c <<= e->ct;
+  if (e->c & 0xF8000000L) {
+    /* One final overflow has to be handled */
+    if (e->buffer >= 0) {
+      if (e->zc)
+	do emit_byte(0x00, cinfo);
+	while (--e->zc);
+      emit_byte(e->buffer + 1, cinfo);
+      if (e->buffer + 1 == 0xFF)
+	emit_byte(0x00, cinfo);
+    }
+    e->zc += e->sc;  /* carry-over converts stacked 0xFF bytes to 0x00 */
+    e->sc = 0;
+  } else {
+    if (e->buffer == 0)
+      ++e->zc;
+    else if (e->buffer >= 0) {
+      if (e->zc)
+	do emit_byte(0x00, cinfo);
+	while (--e->zc);
+      emit_byte(e->buffer, cinfo);
+    }
+    if (e->sc) {
+      if (e->zc)
+	do emit_byte(0x00, cinfo);
+	while (--e->zc);
+      do {
+	emit_byte(0xFF, cinfo);
+	emit_byte(0x00, cinfo);
+      } while (--e->sc);
+    }
+  }
+  /* Output final bytes only if they are not 0x00 */
+  if (e->c & 0x7FFF800L) {
+    if (e->zc)  /* output final pending zero bytes */
+      do emit_byte(0x00, cinfo);
+      while (--e->zc);
+    emit_byte((e->c >> 19) & 0xFF, cinfo);
+    if (((e->c >> 19) & 0xFF) == 0xFF)
+      emit_byte(0x00, cinfo);
+    if (e->c & 0x7F800L) {
+      emit_byte((e->c >> 11) & 0xFF, cinfo);
+      if (((e->c >> 11) & 0xFF) == 0xFF)
+	emit_byte(0x00, cinfo);
+    }
+  }
+}
+
+
+/*
+ * The core arithmetic encoding routine (common in JPEG and JBIG).
+ * This needs to go as fast as possible.
+ * Machine-dependent optimization facilities
+ * are not utilized in this portable implementation.
+ * However, this code should be fairly efficient and
+ * may be a good base for further optimizations anyway.
+ *
+ * Parameter 'val' to be encoded may be 0 or 1 (binary decision).
+ *
+ * Note: I've added full "Pacman" termination support to the
+ * byte output routines, which is equivalent to the optional
+ * Discard_final_zeros procedure (Figure D.15) in the spec.
+ * Thus, we always produce the shortest possible output
+ * stream compliant to the spec (no trailing zero bytes,
+ * except for FF stuffing).
+ *
+ * I've also introduced a new scheme for accessing
+ * the probability estimation state machine table,
+ * derived from Markus Kuhn's JBIG implementation.
+ */
+
+LOCAL(void)
+arith_encode (j_compress_ptr cinfo, unsigned char *st, int val) 
+{
+  register arith_entropy_ptr e = (arith_entropy_ptr) cinfo->entropy;
+  register unsigned char nl, nm;
+  register INT32 qe, temp;
+  register int sv;
+
+  /* Fetch values from our compact representation of Table D.2:
+   * Qe values and probability estimation state machine
+   */
+  sv = *st;
+  qe = jpeg_aritab[sv & 0x7F];	/* => Qe_Value */
+  nl = qe & 0xFF; qe >>= 8;	/* Next_Index_LPS + Switch_MPS */
+  nm = qe & 0xFF; qe >>= 8;	/* Next_Index_MPS */
+
+  /* Encode & estimation procedures per sections D.1.4 & D.1.5 */
+  e->a -= qe;
+  if (val != (sv >> 7)) {
+    /* Encode the less probable symbol */
+    if (e->a >= qe) {
+      /* If the interval size (qe) for the less probable symbol (LPS)
+       * is larger than the interval size for the MPS, then exchange
+       * the two symbols for coding efficiency, otherwise code the LPS
+       * as usual: */
+      e->c += e->a;
+      e->a = qe;
+    }
+    *st = (sv & 0x80) ^ nl;	/* Estimate_after_LPS */
+  } else {
+    /* Encode the more probable symbol */
+    if (e->a >= 0x8000L)
+      return;  /* A >= 0x8000 -> ready, no renormalization required */
+    if (e->a < qe) {
+      /* If the interval size (qe) for the less probable symbol (LPS)
+       * is larger than the interval size for the MPS, then exchange
+       * the two symbols for coding efficiency: */
+      e->c += e->a;
+      e->a = qe;
+    }
+    *st = (sv & 0x80) ^ nm;	/* Estimate_after_MPS */
+  }
+
+  /* Renormalization & data output per section D.1.6 */
+  do {
+    e->a <<= 1;
+    e->c <<= 1;
+    if (--e->ct == 0) {
+      /* Another byte is ready for output */
+      temp = e->c >> 19;
+      if (temp > 0xFF) {
+	/* Handle overflow over all stacked 0xFF bytes */
+	if (e->buffer >= 0) {
+	  if (e->zc)
+	    do emit_byte(0x00, cinfo);
+	    while (--e->zc);
+	  emit_byte(e->buffer + 1, cinfo);
+	  if (e->buffer + 1 == 0xFF)
+	    emit_byte(0x00, cinfo);
+	}
+	e->zc += e->sc;  /* carry-over converts stacked 0xFF bytes to 0x00 */
+	e->sc = 0;
+	/* Note: The 3 spacer bits in the C register guarantee
+	 * that the new buffer byte can't be 0xFF here
+	 * (see page 160 in the P&M JPEG book). */
+	e->buffer = temp & 0xFF;  /* new output byte, might overflow later */
+      } else if (temp == 0xFF) {
+	++e->sc;  /* stack 0xFF byte (which might overflow later) */
+      } else {
+	/* Output all stacked 0xFF bytes, they will not overflow any more */
+	if (e->buffer == 0)
+	  ++e->zc;
+	else if (e->buffer >= 0) {
+	  if (e->zc)
+	    do emit_byte(0x00, cinfo);
+	    while (--e->zc);
+	  emit_byte(e->buffer, cinfo);
+	}
+	if (e->sc) {
+	  if (e->zc)
+	    do emit_byte(0x00, cinfo);
+	    while (--e->zc);
+	  do {
+	    emit_byte(0xFF, cinfo);
+	    emit_byte(0x00, cinfo);
+	  } while (--e->sc);
+	}
+	e->buffer = temp & 0xFF;  /* new output byte (can still overflow) */
+      }
+      e->c &= 0x7FFFFL;
+      e->ct += 8;
+    }
+  } while (e->a < 0x8000L);
+}
+
+
+/*
+ * Emit a restart marker & resynchronize predictions.
+ */
+
+LOCAL(void)
+emit_restart (j_compress_ptr cinfo, int restart_num)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  int ci;
+  jpeg_component_info * compptr;
+
+  finish_pass(cinfo);
+
+  emit_byte(0xFF, cinfo);
+  emit_byte(JPEG_RST0 + restart_num, cinfo);
+
+  /* Re-initialize statistics areas */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    /* DC needs no table for refinement scan */
+    if (cinfo->Ss == 0 && cinfo->Ah == 0) {
+      MEMZERO(entropy->dc_stats[compptr->dc_tbl_no], DC_STAT_BINS);
+      /* Reset DC predictions to 0 */
+      entropy->last_dc_val[ci] = 0;
+      entropy->dc_context[ci] = 0;
+    }
+    /* AC needs no table when not present */
+    if (cinfo->Se) {
+      MEMZERO(entropy->ac_stats[compptr->ac_tbl_no], AC_STAT_BINS);
+    }
+  }
+
+  /* Reset arithmetic encoding variables */
+  entropy->c = 0;
+  entropy->a = 0x10000L;
+  entropy->sc = 0;
+  entropy->zc = 0;
+  entropy->ct = 11;
+  entropy->buffer = -1;  /* empty */
+}
+
+
+/*
+ * MCU encoding for DC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+encode_mcu_DC_first (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  JBLOCKROW block;
+  unsigned char *st;
+  int blkn, ci, tbl;
+  int v, v2, m;
+  ISHIFT_TEMPS
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+    ci = cinfo->MCU_membership[blkn];
+    tbl = cinfo->cur_comp_info[ci]->dc_tbl_no;
+
+    /* Compute the DC value after the required point transform by Al.
+     * This is simply an arithmetic right shift.
+     */
+    m = IRIGHT_SHIFT((int) ((*block)[0]), cinfo->Al);
+
+    /* Sections F.1.4.1 & F.1.4.4.1: Encoding of DC coefficients */
+
+    /* Table F.4: Point to statistics bin S0 for DC coefficient coding */
+    st = entropy->dc_stats[tbl] + entropy->dc_context[ci];
+
+    /* Figure F.4: Encode_DC_DIFF */
+    if ((v = m - entropy->last_dc_val[ci]) == 0) {
+      arith_encode(cinfo, st, 0);
+      entropy->dc_context[ci] = 0;	/* zero diff category */
+    } else {
+      entropy->last_dc_val[ci] = m;
+      arith_encode(cinfo, st, 1);
+      /* Figure F.6: Encoding nonzero value v */
+      /* Figure F.7: Encoding the sign of v */
+      if (v > 0) {
+	arith_encode(cinfo, st + 1, 0);	/* Table F.4: SS = S0 + 1 */
+	st += 2;			/* Table F.4: SP = S0 + 2 */
+	entropy->dc_context[ci] = 4;	/* small positive diff category */
+      } else {
+	v = -v;
+	arith_encode(cinfo, st + 1, 1);	/* Table F.4: SS = S0 + 1 */
+	st += 3;			/* Table F.4: SN = S0 + 3 */
+	entropy->dc_context[ci] = 8;	/* small negative diff category */
+      }
+      /* Figure F.8: Encoding the magnitude category of v */
+      m = 0;
+      if (v -= 1) {
+	arith_encode(cinfo, st, 1);
+	m = 1;
+	v2 = v;
+	st = entropy->dc_stats[tbl] + 20; /* Table F.4: X1 = 20 */
+	while (v2 >>= 1) {
+	  arith_encode(cinfo, st, 1);
+	  m <<= 1;
+	  st += 1;
+	}
+      }
+      arith_encode(cinfo, st, 0);
+      /* Section F.1.4.4.1.2: Establish dc_context conditioning category */
+      if (m < (int) ((1L << cinfo->arith_dc_L[tbl]) >> 1))
+	entropy->dc_context[ci] = 0;	/* zero diff category */
+      else if (m > (int) ((1L << cinfo->arith_dc_U[tbl]) >> 1))
+	entropy->dc_context[ci] += 8;	/* large diff category */
+      /* Figure F.9: Encoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	arith_encode(cinfo, st, (m & v) ? 1 : 0);
+    }
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for AC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+encode_mcu_AC_first (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  JBLOCKROW block;
+  unsigned char *st;
+  int tbl, k, ke;
+  int v, v2, m;
+  const int * natural_order;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  natural_order = cinfo->natural_order;
+
+  /* Encode the MCU data block */
+  block = MCU_data[0];
+  tbl = cinfo->cur_comp_info[0]->ac_tbl_no;
+
+  /* Sections F.1.4.2 & F.1.4.4.2: Encoding of AC coefficients */
+
+  /* Establish EOB (end-of-block) index */
+  for (ke = cinfo->Se; ke > 0; ke--)
+    /* We must apply the point transform by Al.  For AC coefficients this
+     * is an integer division with rounding towards 0.  To do this portably
+     * in C, we shift after obtaining the absolute value.
+     */
+    if ((v = (*block)[natural_order[ke]]) >= 0) {
+      if (v >>= cinfo->Al) break;
+    } else {
+      v = -v;
+      if (v >>= cinfo->Al) break;
+    }
+
+  /* Figure F.5: Encode_AC_Coefficients */
+  for (k = cinfo->Ss; k <= ke; k++) {
+    st = entropy->ac_stats[tbl] + 3 * (k - 1);
+    arith_encode(cinfo, st, 0);		/* EOB decision */
+    for (;;) {
+      if ((v = (*block)[natural_order[k]]) >= 0) {
+	if (v >>= cinfo->Al) {
+	  arith_encode(cinfo, st + 1, 1);
+	  arith_encode(cinfo, entropy->fixed_bin, 0);
+	  break;
+	}
+      } else {
+	v = -v;
+	if (v >>= cinfo->Al) {
+	  arith_encode(cinfo, st + 1, 1);
+	  arith_encode(cinfo, entropy->fixed_bin, 1);
+	  break;
+	}
+      }
+      arith_encode(cinfo, st + 1, 0); st += 3; k++;
+    }
+    st += 2;
+    /* Figure F.8: Encoding the magnitude category of v */
+    m = 0;
+    if (v -= 1) {
+      arith_encode(cinfo, st, 1);
+      m = 1;
+      v2 = v;
+      if (v2 >>= 1) {
+	arith_encode(cinfo, st, 1);
+	m <<= 1;
+	st = entropy->ac_stats[tbl] +
+	     (k <= cinfo->arith_ac_K[tbl] ? 189 : 217);
+	while (v2 >>= 1) {
+	  arith_encode(cinfo, st, 1);
+	  m <<= 1;
+	  st += 1;
+	}
+      }
+    }
+    arith_encode(cinfo, st, 0);
+    /* Figure F.9: Encoding the magnitude bit pattern of v */
+    st += 14;
+    while (m >>= 1)
+      arith_encode(cinfo, st, (m & v) ? 1 : 0);
+  }
+  /* Encode EOB decision only if k <= cinfo->Se */
+  if (k <= cinfo->Se) {
+    st = entropy->ac_stats[tbl] + 3 * (k - 1);
+    arith_encode(cinfo, st, 1);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for DC successive approximation refinement scan.
+ */
+
+METHODDEF(boolean)
+encode_mcu_DC_refine (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  unsigned char *st;
+  int Al, blkn;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  st = entropy->fixed_bin;	/* use fixed probability estimation */
+  Al = cinfo->Al;
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    /* We simply emit the Al'th bit of the DC coefficient value. */
+    arith_encode(cinfo, st, (MCU_data[blkn][0][0] >> Al) & 1);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for AC successive approximation refinement scan.
+ */
+
+METHODDEF(boolean)
+encode_mcu_AC_refine (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  JBLOCKROW block;
+  unsigned char *st;
+  int tbl, k, ke, kex;
+  int v;
+  const int * natural_order;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  natural_order = cinfo->natural_order;
+
+  /* Encode the MCU data block */
+  block = MCU_data[0];
+  tbl = cinfo->cur_comp_info[0]->ac_tbl_no;
+
+  /* Section G.1.3.3: Encoding of AC coefficients */
+
+  /* Establish EOB (end-of-block) index */
+  for (ke = cinfo->Se; ke > 0; ke--)
+    /* We must apply the point transform by Al.  For AC coefficients this
+     * is an integer division with rounding towards 0.  To do this portably
+     * in C, we shift after obtaining the absolute value.
+     */
+    if ((v = (*block)[natural_order[ke]]) >= 0) {
+      if (v >>= cinfo->Al) break;
+    } else {
+      v = -v;
+      if (v >>= cinfo->Al) break;
+    }
+
+  /* Establish EOBx (previous stage end-of-block) index */
+  for (kex = ke; kex > 0; kex--)
+    if ((v = (*block)[natural_order[kex]]) >= 0) {
+      if (v >>= cinfo->Ah) break;
+    } else {
+      v = -v;
+      if (v >>= cinfo->Ah) break;
+    }
+
+  /* Figure G.10: Encode_AC_Coefficients_SA */
+  for (k = cinfo->Ss; k <= ke; k++) {
+    st = entropy->ac_stats[tbl] + 3 * (k - 1);
+    if (k > kex)
+      arith_encode(cinfo, st, 0);	/* EOB decision */
+    for (;;) {
+      if ((v = (*block)[natural_order[k]]) >= 0) {
+	if (v >>= cinfo->Al) {
+	  if (v >> 1)			/* previously nonzero coef */
+	    arith_encode(cinfo, st + 2, (v & 1));
+	  else {			/* newly nonzero coef */
+	    arith_encode(cinfo, st + 1, 1);
+	    arith_encode(cinfo, entropy->fixed_bin, 0);
+	  }
+	  break;
+	}
+      } else {
+	v = -v;
+	if (v >>= cinfo->Al) {
+	  if (v >> 1)			/* previously nonzero coef */
+	    arith_encode(cinfo, st + 2, (v & 1));
+	  else {			/* newly nonzero coef */
+	    arith_encode(cinfo, st + 1, 1);
+	    arith_encode(cinfo, entropy->fixed_bin, 1);
+	  }
+	  break;
+	}
+      }
+      arith_encode(cinfo, st + 1, 0); st += 3; k++;
+    }
+  }
+  /* Encode EOB decision only if k <= cinfo->Se */
+  if (k <= cinfo->Se) {
+    st = entropy->ac_stats[tbl] + 3 * (k - 1);
+    arith_encode(cinfo, st, 1);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * Encode and output one MCU's worth of arithmetic-compressed coefficients.
+ */
+
+METHODDEF(boolean)
+encode_mcu (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  jpeg_component_info * compptr;
+  JBLOCKROW block;
+  unsigned char *st;
+  int blkn, ci, tbl, k, ke;
+  int v, v2, m;
+  const int * natural_order;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  natural_order = cinfo->natural_order;
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+    ci = cinfo->MCU_membership[blkn];
+    compptr = cinfo->cur_comp_info[ci];
+
+    /* Sections F.1.4.1 & F.1.4.4.1: Encoding of DC coefficients */
+
+    tbl = compptr->dc_tbl_no;
+
+    /* Table F.4: Point to statistics bin S0 for DC coefficient coding */
+    st = entropy->dc_stats[tbl] + entropy->dc_context[ci];
+
+    /* Figure F.4: Encode_DC_DIFF */
+    if ((v = (*block)[0] - entropy->last_dc_val[ci]) == 0) {
+      arith_encode(cinfo, st, 0);
+      entropy->dc_context[ci] = 0;	/* zero diff category */
+    } else {
+      entropy->last_dc_val[ci] = (*block)[0];
+      arith_encode(cinfo, st, 1);
+      /* Figure F.6: Encoding nonzero value v */
+      /* Figure F.7: Encoding the sign of v */
+      if (v > 0) {
+	arith_encode(cinfo, st + 1, 0);	/* Table F.4: SS = S0 + 1 */
+	st += 2;			/* Table F.4: SP = S0 + 2 */
+	entropy->dc_context[ci] = 4;	/* small positive diff category */
+      } else {
+	v = -v;
+	arith_encode(cinfo, st + 1, 1);	/* Table F.4: SS = S0 + 1 */
+	st += 3;			/* Table F.4: SN = S0 + 3 */
+	entropy->dc_context[ci] = 8;	/* small negative diff category */
+      }
+      /* Figure F.8: Encoding the magnitude category of v */
+      m = 0;
+      if (v -= 1) {
+	arith_encode(cinfo, st, 1);
+	m = 1;
+	v2 = v;
+	st = entropy->dc_stats[tbl] + 20; /* Table F.4: X1 = 20 */
+	while (v2 >>= 1) {
+	  arith_encode(cinfo, st, 1);
+	  m <<= 1;
+	  st += 1;
+	}
+      }
+      arith_encode(cinfo, st, 0);
+      /* Section F.1.4.4.1.2: Establish dc_context conditioning category */
+      if (m < (int) ((1L << cinfo->arith_dc_L[tbl]) >> 1))
+	entropy->dc_context[ci] = 0;	/* zero diff category */
+      else if (m > (int) ((1L << cinfo->arith_dc_U[tbl]) >> 1))
+	entropy->dc_context[ci] += 8;	/* large diff category */
+      /* Figure F.9: Encoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	arith_encode(cinfo, st, (m & v) ? 1 : 0);
+    }
+
+    /* Sections F.1.4.2 & F.1.4.4.2: Encoding of AC coefficients */
+
+    tbl = compptr->ac_tbl_no;
+
+    /* Establish EOB (end-of-block) index */
+    for (ke = cinfo->lim_Se; ke > 0; ke--)
+      if ((*block)[natural_order[ke]]) break;
+
+    /* Figure F.5: Encode_AC_Coefficients */
+    for (k = 1; k <= ke; k++) {
+      st = entropy->ac_stats[tbl] + 3 * (k - 1);
+      arith_encode(cinfo, st, 0);	/* EOB decision */
+      while ((v = (*block)[natural_order[k]]) == 0) {
+	arith_encode(cinfo, st + 1, 0); st += 3; k++;
+      }
+      arith_encode(cinfo, st + 1, 1);
+      /* Figure F.6: Encoding nonzero value v */
+      /* Figure F.7: Encoding the sign of v */
+      if (v > 0) {
+	arith_encode(cinfo, entropy->fixed_bin, 0);
+      } else {
+	v = -v;
+	arith_encode(cinfo, entropy->fixed_bin, 1);
+      }
+      st += 2;
+      /* Figure F.8: Encoding the magnitude category of v */
+      m = 0;
+      if (v -= 1) {
+	arith_encode(cinfo, st, 1);
+	m = 1;
+	v2 = v;
+	if (v2 >>= 1) {
+	  arith_encode(cinfo, st, 1);
+	  m <<= 1;
+	  st = entropy->ac_stats[tbl] +
+	       (k <= cinfo->arith_ac_K[tbl] ? 189 : 217);
+	  while (v2 >>= 1) {
+	    arith_encode(cinfo, st, 1);
+	    m <<= 1;
+	    st += 1;
+	  }
+	}
+      }
+      arith_encode(cinfo, st, 0);
+      /* Figure F.9: Encoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	arith_encode(cinfo, st, (m & v) ? 1 : 0);
+    }
+    /* Encode EOB decision only if k <= cinfo->lim_Se */
+    if (k <= cinfo->lim_Se) {
+      st = entropy->ac_stats[tbl] + 3 * (k - 1);
+      arith_encode(cinfo, st, 1);
+    }
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * Initialize for an arithmetic-compressed scan.
+ */
+
+METHODDEF(void)
+start_pass (j_compress_ptr cinfo, boolean gather_statistics)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  int ci, tbl;
+  jpeg_component_info * compptr;
+
+  if (gather_statistics)
+    /* Make sure to avoid that in the master control logic!
+     * We are fully adaptive here and need no extra
+     * statistics gathering pass!
+     */
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+
+  /* We assume jcmaster.c already validated the progressive scan parameters. */
+
+  /* Select execution routines */
+  if (cinfo->progressive_mode) {
+    if (cinfo->Ah == 0) {
+      if (cinfo->Ss == 0)
+	entropy->pub.encode_mcu = encode_mcu_DC_first;
+      else
+	entropy->pub.encode_mcu = encode_mcu_AC_first;
+    } else {
+      if (cinfo->Ss == 0)
+	entropy->pub.encode_mcu = encode_mcu_DC_refine;
+      else
+	entropy->pub.encode_mcu = encode_mcu_AC_refine;
+    }
+  } else
+    entropy->pub.encode_mcu = encode_mcu;
+
+  /* Allocate & initialize requested statistics areas */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    /* DC needs no table for refinement scan */
+    if (cinfo->Ss == 0 && cinfo->Ah == 0) {
+      tbl = compptr->dc_tbl_no;
+      if (tbl < 0 || tbl >= NUM_ARITH_TBLS)
+	ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl);
+      if (entropy->dc_stats[tbl] == NULL)
+	entropy->dc_stats[tbl] = (unsigned char *) (*cinfo->mem->alloc_small)
+	  ((j_common_ptr) cinfo, JPOOL_IMAGE, DC_STAT_BINS);
+      MEMZERO(entropy->dc_stats[tbl], DC_STAT_BINS);
+      /* Initialize DC predictions to 0 */
+      entropy->last_dc_val[ci] = 0;
+      entropy->dc_context[ci] = 0;
+    }
+    /* AC needs no table when not present */
+    if (cinfo->Se) {
+      tbl = compptr->ac_tbl_no;
+      if (tbl < 0 || tbl >= NUM_ARITH_TBLS)
+	ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl);
+      if (entropy->ac_stats[tbl] == NULL)
+	entropy->ac_stats[tbl] = (unsigned char *) (*cinfo->mem->alloc_small)
+	  ((j_common_ptr) cinfo, JPOOL_IMAGE, AC_STAT_BINS);
+      MEMZERO(entropy->ac_stats[tbl], AC_STAT_BINS);
+#ifdef CALCULATE_SPECTRAL_CONDITIONING
+      if (cinfo->progressive_mode)
+	/* Section G.1.3.2: Set appropriate arithmetic conditioning value Kx */
+	cinfo->arith_ac_K[tbl] = cinfo->Ss + ((8 + cinfo->Se - cinfo->Ss) >> 4);
+#endif
+    }
+  }
+
+  /* Initialize arithmetic encoding variables */
+  entropy->c = 0;
+  entropy->a = 0x10000L;
+  entropy->sc = 0;
+  entropy->zc = 0;
+  entropy->ct = 11;
+  entropy->buffer = -1;  /* empty */
+
+  /* Initialize restart stuff */
+  entropy->restarts_to_go = cinfo->restart_interval;
+  entropy->next_restart_num = 0;
+}
+
+
+/*
+ * Module initialization routine for arithmetic entropy encoding.
+ */
+
+GLOBAL(void)
+jinit_arith_encoder (j_compress_ptr cinfo)
+{
+  arith_entropy_ptr entropy;
+  int i;
+
+  entropy = (arith_entropy_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				SIZEOF(arith_entropy_encoder));
+  cinfo->entropy = (struct jpeg_entropy_encoder *) entropy;
+  entropy->pub.start_pass = start_pass;
+  entropy->pub.finish_pass = finish_pass;
+
+  /* Mark tables unallocated */
+  for (i = 0; i < NUM_ARITH_TBLS; i++) {
+    entropy->dc_stats[i] = NULL;
+    entropy->ac_stats[i] = NULL;
+  }
+
+  /* Initialize index for fixed probability estimation */
+  entropy->fixed_bin[0] = 113;
+}
diff --git a/src/jpeg-6b/jccoefct.c b/src/jpeg-8c/jccoefct.c
index 1963ddb6..d775313b 100644
--- a/src/jpeg-6b/jccoefct.c
+++ b/src/jpeg-8c/jccoefct.c
@@ -149,6 +149,7 @@ compress_data (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
   int blkn, bi, ci, yindex, yoffset, blockcnt;
   JDIMENSION ypos, xpos;
   jpeg_component_info *compptr;
+  forward_DCT_ptr forward_DCT;
 
   /* Loop to write as much as one whole iMCU row */
   for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
@@ -167,17 +168,19 @@ compress_data (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
       blkn = 0;
       for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
 	compptr = cinfo->cur_comp_info[ci];
+	forward_DCT = cinfo->fdct->forward_DCT[compptr->component_index];
 	blockcnt = (MCU_col_num < last_MCU_col) ? compptr->MCU_width
 						: compptr->last_col_width;
 	xpos = MCU_col_num * compptr->MCU_sample_width;
-	ypos = yoffset * DCTSIZE; /* ypos == (yoffset+yindex) * DCTSIZE */
+	ypos = yoffset * compptr->DCT_v_scaled_size;
+	/* ypos == (yoffset+yindex) * DCTSIZE */
 	for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
 	  if (coef->iMCU_row_num < last_iMCU_row ||
 	      yoffset+yindex < compptr->last_row_height) {
-	    (*cinfo->fdct->forward_DCT) (cinfo, compptr,
-					 input_buf[compptr->component_index],
-					 coef->MCU_buffer[blkn],
-					 ypos, xpos, (JDIMENSION) blockcnt);
+	    (*forward_DCT) (cinfo, compptr,
+			    input_buf[compptr->component_index],
+			    coef->MCU_buffer[blkn],
+			    ypos, xpos, (JDIMENSION) blockcnt);
 	    if (blockcnt < compptr->MCU_width) {
 	      /* Create some dummy blocks at the right edge of the image. */
 	      jzero_far((void FAR *) coef->MCU_buffer[blkn + blockcnt],
@@ -195,7 +198,7 @@ compress_data (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
 	    }
 	  }
 	  blkn += compptr->MCU_width;
-	  ypos += DCTSIZE;
+	  ypos += compptr->DCT_v_scaled_size;
 	}
       }
       /* Try to write the MCU.  In event of a suspension failure, we will
@@ -252,6 +255,7 @@ compress_first_pass (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
   jpeg_component_info *compptr;
   JBLOCKARRAY buffer;
   JBLOCKROW thisblockrow, lastblockrow;
+  forward_DCT_ptr forward_DCT;
 
   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
        ci++, compptr++) {
@@ -274,15 +278,15 @@ compress_first_pass (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
     ndummy = (int) (blocks_across % h_samp_factor);
     if (ndummy > 0)
       ndummy = h_samp_factor - ndummy;
+    forward_DCT = cinfo->fdct->forward_DCT[ci];
     /* Perform DCT for all non-dummy blocks in this iMCU row.  Each call
      * on forward_DCT processes a complete horizontal row of DCT blocks.
      */
     for (block_row = 0; block_row < block_rows; block_row++) {
       thisblockrow = buffer[block_row];
-      (*cinfo->fdct->forward_DCT) (cinfo, compptr,
-				   input_buf[ci], thisblockrow,
-				   (JDIMENSION) (block_row * DCTSIZE),
-				   (JDIMENSION) 0, blocks_across);
+      (*forward_DCT) (cinfo, compptr, input_buf[ci], thisblockrow,
+		      (JDIMENSION) (block_row * compptr->DCT_v_scaled_size),
+		      (JDIMENSION) 0, blocks_across);
       if (ndummy > 0) {
 	/* Create dummy blocks at the right edge of the image. */
 	thisblockrow += blocks_across; /* => first dummy block */
diff --git a/src/jpeg-6b/jccolor.c b/src/jpeg-8c/jccolor.c
index 0a8a4b5d..0a8a4b5d 100644
--- a/src/jpeg-6b/jccolor.c
+++ b/src/jpeg-8c/jccolor.c
diff --git a/src/jpeg-6b/jcdctmgr.c b/src/jpeg-8c/jcdctmgr.c
index 94d5cda4..0bbdbb68 100644
--- a/src/jpeg-6b/jcdctmgr.c
+++ b/src/jpeg-8c/jcdctmgr.c
@@ -23,7 +23,7 @@ typedef struct {
   struct jpeg_forward_dct pub;	/* public fields */
 
   /* Pointer to the DCT routine actually in use */
-  forward_DCT_method_ptr do_dct;
+  forward_DCT_method_ptr do_dct[MAX_COMPONENTS];
 
   /* The actual post-DCT divisors --- not identical to the quant table
    * entries, because of scaling (especially for an unnormalized DCT).
@@ -33,7 +33,7 @@ typedef struct {
 
 #ifdef DCT_FLOAT_SUPPORTED
   /* Same as above for the floating-point case. */
-  float_DCT_method_ptr do_float_dct;
+  float_DCT_method_ptr do_float_dct[MAX_COMPONENTS];
   FAST_FLOAT * float_divisors[NUM_QUANT_TBLS];
 #endif
 } my_fdct_controller;
@@ -41,6 +41,132 @@ typedef struct {
 typedef my_fdct_controller * my_fdct_ptr;
 
 
+/* The current scaled-DCT routines require ISLOW-style divisor tables,
+ * so be sure to compile that code if either ISLOW or SCALING is requested.
+ */
+#ifdef DCT_ISLOW_SUPPORTED
+#define PROVIDE_ISLOW_TABLES
+#else
+#ifdef DCT_SCALING_SUPPORTED
+#define PROVIDE_ISLOW_TABLES
+#endif
+#endif
+
+
+/*
+ * Perform forward DCT on one or more blocks of a component.
+ *
+ * The input samples are taken from the sample_data[] array starting at
+ * position start_row/start_col, and moving to the right for any additional
+ * blocks. The quantized coefficients are returned in coef_blocks[].
+ */
+
+METHODDEF(void)
+forward_DCT (j_compress_ptr cinfo, jpeg_component_info * compptr,
+	     JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
+	     JDIMENSION start_row, JDIMENSION start_col,
+	     JDIMENSION num_blocks)
+/* This version is used for integer DCT implementations. */
+{
+  /* This routine is heavily used, so it's worth coding it tightly. */
+  my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
+  forward_DCT_method_ptr do_dct = fdct->do_dct[compptr->component_index];
+  DCTELEM * divisors = fdct->divisors[compptr->quant_tbl_no];
+  DCTELEM workspace[DCTSIZE2];	/* work area for FDCT subroutine */
+  JDIMENSION bi;
+
+  sample_data += start_row;	/* fold in the vertical offset once */
+
+  for (bi = 0; bi < num_blocks; bi++, start_col += compptr->DCT_h_scaled_size) {
+    /* Perform the DCT */
+    (*do_dct) (workspace, sample_data, start_col);
+
+    /* Quantize/descale the coefficients, and store into coef_blocks[] */
+    { register DCTELEM temp, qval;
+      register int i;
+      register JCOEFPTR output_ptr = coef_blocks[bi];
+
+      for (i = 0; i < DCTSIZE2; i++) {
+	qval = divisors[i];
+	temp = workspace[i];
+	/* Divide the coefficient value by qval, ensuring proper rounding.
+	 * Since C does not specify the direction of rounding for negative
+	 * quotients, we have to force the dividend positive for portability.
+	 *
+	 * In most files, at least half of the output values will be zero
+	 * (at default quantization settings, more like three-quarters...)
+	 * so we should ensure that this case is fast.  On many machines,
+	 * a comparison is enough cheaper than a divide to make a special test
+	 * a win.  Since both inputs will be nonnegative, we need only test
+	 * for a < b to discover whether a/b is 0.
+	 * If your machine's division is fast enough, define FAST_DIVIDE.
+	 */
+#ifdef FAST_DIVIDE
+#define DIVIDE_BY(a,b)	a /= b
+#else
+#define DIVIDE_BY(a,b)	if (a >= b) a /= b; else a = 0
+#endif
+	if (temp < 0) {
+	  temp = -temp;
+	  temp += qval>>1;	/* for rounding */
+	  DIVIDE_BY(temp, qval);
+	  temp = -temp;
+	} else {
+	  temp += qval>>1;	/* for rounding */
+	  DIVIDE_BY(temp, qval);
+	}
+	output_ptr[i] = (JCOEF) temp;
+      }
+    }
+  }
+}
+
+
+#ifdef DCT_FLOAT_SUPPORTED
+
+METHODDEF(void)
+forward_DCT_float (j_compress_ptr cinfo, jpeg_component_info * compptr,
+		   JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
+		   JDIMENSION start_row, JDIMENSION start_col,
+		   JDIMENSION num_blocks)
+/* This version is used for floating-point DCT implementations. */
+{
+  /* This routine is heavily used, so it's worth coding it tightly. */
+  my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
+  float_DCT_method_ptr do_dct = fdct->do_float_dct[compptr->component_index];
+  FAST_FLOAT * divisors = fdct->float_divisors[compptr->quant_tbl_no];
+  FAST_FLOAT workspace[DCTSIZE2]; /* work area for FDCT subroutine */
+  JDIMENSION bi;
+
+  sample_data += start_row;	/* fold in the vertical offset once */
+
+  for (bi = 0; bi < num_blocks; bi++, start_col += compptr->DCT_h_scaled_size) {
+    /* Perform the DCT */
+    (*do_dct) (workspace, sample_data, start_col);
+
+    /* Quantize/descale the coefficients, and store into coef_blocks[] */
+    { register FAST_FLOAT temp;
+      register int i;
+      register JCOEFPTR output_ptr = coef_blocks[bi];
+
+      for (i = 0; i < DCTSIZE2; i++) {
+	/* Apply the quantization and scaling factor */
+	temp = workspace[i] * divisors[i];
+	/* Round to nearest integer.
+	 * Since C does not specify the direction of rounding for negative
+	 * quotients, we have to force the dividend positive for portability.
+	 * The maximum coefficient size is +-16K (for 12-bit data), so this
+	 * code should work for either 16-bit or 32-bit ints.
+	 */
+	output_ptr[i] = (JCOEF) ((int) (temp + (FAST_FLOAT) 16384.5) - 16384);
+      }
+    }
+  }
+}
+
+#endif /* DCT_FLOAT_SUPPORTED */
+
+
 /*
  * Initialize for a processing pass.
  * Verify that all referenced Q-tables are present, and set up
@@ -56,10 +182,170 @@ start_pass_fdctmgr (j_compress_ptr cinfo)
   my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
   int ci, qtblno, i;
   jpeg_component_info *compptr;
+  int method = 0;
   JQUANT_TBL * qtbl;
+  DCTELEM * dtbl;
 
   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
        ci++, compptr++) {
+    /* Select the proper DCT routine for this component's scaling */
+    switch ((compptr->DCT_h_scaled_size << 8) + compptr->DCT_v_scaled_size) {
+#ifdef DCT_SCALING_SUPPORTED
+    case ((1 << 8) + 1):
+      fdct->do_dct[ci] = jpeg_fdct_1x1;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((2 << 8) + 2):
+      fdct->do_dct[ci] = jpeg_fdct_2x2;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((3 << 8) + 3):
+      fdct->do_dct[ci] = jpeg_fdct_3x3;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((4 << 8) + 4):
+      fdct->do_dct[ci] = jpeg_fdct_4x4;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((5 << 8) + 5):
+      fdct->do_dct[ci] = jpeg_fdct_5x5;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((6 << 8) + 6):
+      fdct->do_dct[ci] = jpeg_fdct_6x6;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((7 << 8) + 7):
+      fdct->do_dct[ci] = jpeg_fdct_7x7;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((9 << 8) + 9):
+      fdct->do_dct[ci] = jpeg_fdct_9x9;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((10 << 8) + 10):
+      fdct->do_dct[ci] = jpeg_fdct_10x10;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((11 << 8) + 11):
+      fdct->do_dct[ci] = jpeg_fdct_11x11;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((12 << 8) + 12):
+      fdct->do_dct[ci] = jpeg_fdct_12x12;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((13 << 8) + 13):
+      fdct->do_dct[ci] = jpeg_fdct_13x13;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((14 << 8) + 14):
+      fdct->do_dct[ci] = jpeg_fdct_14x14;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((15 << 8) + 15):
+      fdct->do_dct[ci] = jpeg_fdct_15x15;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((16 << 8) + 16):
+      fdct->do_dct[ci] = jpeg_fdct_16x16;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((16 << 8) + 8):
+      fdct->do_dct[ci] = jpeg_fdct_16x8;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((14 << 8) + 7):
+      fdct->do_dct[ci] = jpeg_fdct_14x7;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((12 << 8) + 6):
+      fdct->do_dct[ci] = jpeg_fdct_12x6;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((10 << 8) + 5):
+      fdct->do_dct[ci] = jpeg_fdct_10x5;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((8 << 8) + 4):
+      fdct->do_dct[ci] = jpeg_fdct_8x4;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((6 << 8) + 3):
+      fdct->do_dct[ci] = jpeg_fdct_6x3;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((4 << 8) + 2):
+      fdct->do_dct[ci] = jpeg_fdct_4x2;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((2 << 8) + 1):
+      fdct->do_dct[ci] = jpeg_fdct_2x1;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((8 << 8) + 16):
+      fdct->do_dct[ci] = jpeg_fdct_8x16;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((7 << 8) + 14):
+      fdct->do_dct[ci] = jpeg_fdct_7x14;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((6 << 8) + 12):
+      fdct->do_dct[ci] = jpeg_fdct_6x12;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((5 << 8) + 10):
+      fdct->do_dct[ci] = jpeg_fdct_5x10;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((4 << 8) + 8):
+      fdct->do_dct[ci] = jpeg_fdct_4x8;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((3 << 8) + 6):
+      fdct->do_dct[ci] = jpeg_fdct_3x6;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((2 << 8) + 4):
+      fdct->do_dct[ci] = jpeg_fdct_2x4;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+    case ((1 << 8) + 2):
+      fdct->do_dct[ci] = jpeg_fdct_1x2;
+      method = JDCT_ISLOW;	/* jfdctint uses islow-style table */
+      break;
+#endif
+    case ((DCTSIZE << 8) + DCTSIZE):
+      switch (cinfo->dct_method) {
+#ifdef DCT_ISLOW_SUPPORTED
+      case JDCT_ISLOW:
+	fdct->do_dct[ci] = jpeg_fdct_islow;
+	method = JDCT_ISLOW;
+	break;
+#endif
+#ifdef DCT_IFAST_SUPPORTED
+      case JDCT_IFAST:
+	fdct->do_dct[ci] = jpeg_fdct_ifast;
+	method = JDCT_IFAST;
+	break;
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+      case JDCT_FLOAT:
+	fdct->do_float_dct[ci] = jpeg_fdct_float;
+	method = JDCT_FLOAT;
+	break;
+#endif
+      default:
+	ERREXIT(cinfo, JERR_NOT_COMPILED);
+	break;
+      }
+      break;
+    default:
+      ERREXIT2(cinfo, JERR_BAD_DCTSIZE,
+	       compptr->DCT_h_scaled_size, compptr->DCT_v_scaled_size);
+      break;
+    }
     qtblno = compptr->quant_tbl_no;
     /* Make sure specified quantization table is present */
     if (qtblno < 0 || qtblno >= NUM_QUANT_TBLS ||
@@ -68,8 +354,8 @@ start_pass_fdctmgr (j_compress_ptr cinfo)
     qtbl = cinfo->quant_tbl_ptrs[qtblno];
     /* Compute divisors for this quant table */
     /* We may do this more than once for same table, but it's not a big deal */
-    switch (cinfo->dct_method) {
-#ifdef DCT_ISLOW_SUPPORTED
+    switch (method) {
+#ifdef PROVIDE_ISLOW_TABLES
     case JDCT_ISLOW:
       /* For LL&M IDCT method, divisors are equal to raw quantization
        * coefficients multiplied by 8 (to counteract scaling).
@@ -83,6 +369,7 @@ start_pass_fdctmgr (j_compress_ptr cinfo)
       for (i = 0; i < DCTSIZE2; i++) {
 	dtbl[i] = ((DCTELEM) qtbl->quantval[i]) << 3;
       }
+      fdct->pub.forward_DCT[ci] = forward_DCT;
       break;
 #endif
 #ifdef DCT_IFAST_SUPPORTED
@@ -121,6 +408,7 @@ start_pass_fdctmgr (j_compress_ptr cinfo)
 		    CONST_BITS-3);
 	}
       }
+      fdct->pub.forward_DCT[ci] = forward_DCT;
       break;
 #endif
 #ifdef DCT_FLOAT_SUPPORTED
@@ -157,6 +445,7 @@ start_pass_fdctmgr (j_compress_ptr cinfo)
 	  }
 	}
       }
+      fdct->pub.forward_DCT[ci] = forward_DCT_float;
       break;
 #endif
     default:
@@ -167,178 +456,6 @@ start_pass_fdctmgr (j_compress_ptr cinfo)
 }
 
 
-/* code/jpeg-6b/jcdctmgr.c:184: warning: ‘forward_DCT’ defined but not used */
-#if 0
-/*
- * Perform forward DCT on one or more blocks of a component.
- *
- * The input samples are taken from the sample_data[] array starting at
- * position start_row/start_col, and moving to the right for any additional
- * blocks. The quantized coefficients are returned in coef_blocks[].
- */
-
-METHODDEF(void)
-forward_DCT (j_compress_ptr cinfo, jpeg_component_info * compptr,
-	     JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
-	     JDIMENSION start_row, JDIMENSION start_col,
-	     JDIMENSION num_blocks)
-/* This version is used for integer DCT implementations. */
-{
-  /* This routine is heavily used, so it's worth coding it tightly. */
-  my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
-  forward_DCT_method_ptr do_dct = fdct->do_dct;
-  DCTELEM * divisors = fdct->divisors[compptr->quant_tbl_no];
-  DCTELEM workspace[DCTSIZE2];	/* work area for FDCT subroutine */
-  JDIMENSION bi;
-
-  sample_data += start_row;	/* fold in the vertical offset once */
-
-  for (bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE) {
-    /* Load data into workspace, applying unsigned->signed conversion */
-    { register DCTELEM *workspaceptr;
-      register JSAMPROW elemptr;
-      register int elemr;
-
-      workspaceptr = workspace;
-      for (elemr = 0; elemr < DCTSIZE; elemr++) {
-	elemptr = sample_data[elemr] + start_col;
-#if DCTSIZE == 8		/* unroll the inner loop */
-	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-#else
-	{ register int elemc;
-	  for (elemc = DCTSIZE; elemc > 0; elemc--) {
-	    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
-	  }
-	}
-#endif
-      }
-    }
-
-    /* Perform the DCT */
-    (*do_dct) (workspace);
-
-    /* Quantize/descale the coefficients, and store into coef_blocks[] */
-    { register DCTELEM temp, qval;
-      register int i;
-      register JCOEFPTR output_ptr = coef_blocks[bi];
-
-      for (i = 0; i < DCTSIZE2; i++) {
-	qval = divisors[i];
-	temp = workspace[i];
-	/* Divide the coefficient value by qval, ensuring proper rounding.
-	 * Since C does not specify the direction of rounding for negative
-	 * quotients, we have to force the dividend positive for portability.
-	 *
-	 * In most files, at least half of the output values will be zero
-	 * (at default quantization settings, more like three-quarters...)
-	 * so we should ensure that this case is fast.  On many machines,
-	 * a comparison is enough cheaper than a divide to make a special test
-	 * a win.  Since both inputs will be nonnegative, we need only test
-	 * for a < b to discover whether a/b is 0.
-	 * If your machine's division is fast enough, define FAST_DIVIDE.
-	 */
-#ifdef FAST_DIVIDE
-#define DIVIDE_BY(a,b)	a /= b
-#else
-#define DIVIDE_BY(a,b)	if (a >= b) a /= b; else a = 0
-#endif
-	if (temp < 0) {
-	  temp = -temp;
-	  temp += qval>>1;	/* for rounding */
-	  DIVIDE_BY(temp, qval);
-	  temp = -temp;
-	} else {
-	  temp += qval>>1;	/* for rounding */
-	  DIVIDE_BY(temp, qval);
-	}
-	output_ptr[i] = (JCOEF) temp;
-      }
-    }
-  }
-}
-#endif
-
-
-#ifdef DCT_FLOAT_SUPPORTED
-
-METHODDEF(void)
-forward_DCT_float (j_compress_ptr cinfo, jpeg_component_info * compptr,
-		   JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
-		   JDIMENSION start_row, JDIMENSION start_col,
-		   JDIMENSION num_blocks)
-/* This version is used for floating-point DCT implementations. */
-{
-  /* This routine is heavily used, so it's worth coding it tightly. */
-  my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
-  float_DCT_method_ptr do_dct = fdct->do_float_dct;
-  FAST_FLOAT * divisors = fdct->float_divisors[compptr->quant_tbl_no];
-  FAST_FLOAT workspace[DCTSIZE2]; /* work area for FDCT subroutine */
-  JDIMENSION bi;
-
-  sample_data += start_row;	/* fold in the vertical offset once */
-
-  for (bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE) {
-    /* Load data into workspace, applying unsigned->signed conversion */
-    { register FAST_FLOAT *workspaceptr;
-      register JSAMPROW elemptr;
-      register int elemr;
-
-      workspaceptr = workspace;
-      for (elemr = 0; elemr < DCTSIZE; elemr++) {
-	elemptr = sample_data[elemr] + start_col;
-#if DCTSIZE == 8		/* unroll the inner loop */
-	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-#else
-	{ register int elemc;
-	  for (elemc = DCTSIZE; elemc > 0; elemc--) {
-	    *workspaceptr++ = (FAST_FLOAT)
-	      (GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
-	  }
-	}
-#endif
-      }
-    }
-
-    /* Perform the DCT */
-    (*do_dct) (workspace);
-
-    /* Quantize/descale the coefficients, and store into coef_blocks[] */
-    { register FAST_FLOAT temp;
-      register int i;
-      register JCOEFPTR output_ptr = coef_blocks[bi];
-
-      for (i = 0; i < DCTSIZE2; i++) {
-	/* Apply the quantization and scaling factor */
-	temp = workspace[i] * divisors[i];
-	/* Round to nearest integer.
-	 * Since C does not specify the direction of rounding for negative
-	 * quotients, we have to force the dividend positive for portability.
-	 * The maximum coefficient size is +-16K (for 12-bit data), so this
-	 * code should work for either 16-bit or 32-bit ints.
-	 */
-	output_ptr[i] = (JCOEF) ((int) (temp + (FAST_FLOAT) 16384.5) - 16384);
-      }
-    }
-  }
-}
-
-#endif /* DCT_FLOAT_SUPPORTED */
-
-
 /*
  * Initialize FDCT manager.
  */
@@ -355,30 +472,6 @@ jinit_forward_dct (j_compress_ptr cinfo)
   cinfo->fdct = (struct jpeg_forward_dct *) fdct;
   fdct->pub.start_pass = start_pass_fdctmgr;
 
-  switch (cinfo->dct_method) {
-#ifdef DCT_ISLOW_SUPPORTED
-  case JDCT_ISLOW:
-    fdct->pub.forward_DCT = forward_DCT;
-    fdct->do_dct = jpeg_fdct_islow;
-    break;
-#endif
-#ifdef DCT_IFAST_SUPPORTED
-  case JDCT_IFAST:
-    fdct->pub.forward_DCT = forward_DCT;
-    fdct->do_dct = jpeg_fdct_ifast;
-    break;
-#endif
-#ifdef DCT_FLOAT_SUPPORTED
-  case JDCT_FLOAT:
-    fdct->pub.forward_DCT = forward_DCT_float;
-    fdct->do_float_dct = jpeg_fdct_float;
-    break;
-#endif
-  default:
-    ERREXIT(cinfo, JERR_NOT_COMPILED);
-    break;
-  }
-
   /* Mark divisor tables unallocated */
   for (i = 0; i < NUM_QUANT_TBLS; i++) {
     fdct->divisors[i] = NULL;
diff --git a/src/jpeg-8c/jchuff.c b/src/jpeg-8c/jchuff.c
new file mode 100644
index 00000000..257d7aa1
--- /dev/null
+++ b/src/jpeg-8c/jchuff.c
@@ -0,0 +1,1576 @@
+/*
+ * jchuff.c
+ *
+ * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Modified 2006-2009 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains Huffman entropy encoding routines.
+ * Both sequential and progressive modes are supported in this single module.
+ *
+ * Much of the complexity here has to do with supporting output suspension.
+ * If the data destination module demands suspension, we want to be able to
+ * back up to the start of the current MCU.  To do this, we copy state
+ * variables into local working storage, and update them back to the
+ * permanent JPEG objects only upon successful completion of an MCU.
+ *
+ * We do not support output suspension for the progressive JPEG mode, since
+ * the library currently does not allow multiple-scan files to be written
+ * with output suspension.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* The legal range of a DCT coefficient is
+ *  -1024 .. +1023  for 8-bit data;
+ * -16384 .. +16383 for 12-bit data.
+ * Hence the magnitude should always fit in 10 or 14 bits respectively.
+ */
+
+#if BITS_IN_JSAMPLE == 8
+#define MAX_COEF_BITS 10
+#else
+#define MAX_COEF_BITS 14
+#endif
+
+/* Derived data constructed for each Huffman table */
+
+typedef struct {
+  unsigned int ehufco[256];	/* code for each symbol */
+  char ehufsi[256];		/* length of code for each symbol */
+  /* If no code has been allocated for a symbol S, ehufsi[S] contains 0 */
+} c_derived_tbl;
+
+
+/* Expanded entropy encoder object for Huffman encoding.
+ *
+ * The savable_state subrecord contains fields that change within an MCU,
+ * but must not be updated permanently until we complete the MCU.
+ */
+
+typedef struct {
+  INT32 put_buffer;		/* current bit-accumulation buffer */
+  int put_bits;			/* # of bits now in it */
+  int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
+} savable_state;
+
+/* This macro is to work around compilers with missing or broken
+ * structure assignment.  You'll need to fix this code if you have
+ * such a compiler and you change MAX_COMPS_IN_SCAN.
+ */
+
+#ifndef NO_STRUCT_ASSIGN
+#define ASSIGN_STATE(dest,src)  ((dest) = (src))
+#else
+#if MAX_COMPS_IN_SCAN == 4
+#define ASSIGN_STATE(dest,src)  \
+	((dest).put_buffer = (src).put_buffer, \
+	 (dest).put_bits = (src).put_bits, \
+	 (dest).last_dc_val[0] = (src).last_dc_val[0], \
+	 (dest).last_dc_val[1] = (src).last_dc_val[1], \
+	 (dest).last_dc_val[2] = (src).last_dc_val[2], \
+	 (dest).last_dc_val[3] = (src).last_dc_val[3])
+#endif
+#endif
+
+
+typedef struct {
+  struct jpeg_entropy_encoder pub; /* public fields */
+
+  savable_state saved;		/* Bit buffer & DC state at start of MCU */
+
+  /* These fields are NOT loaded into local working state. */
+  unsigned int restarts_to_go;	/* MCUs left in this restart interval */
+  int next_restart_num;		/* next restart number to write (0-7) */
+
+  /* Pointers to derived tables (these workspaces have image lifespan) */
+  c_derived_tbl * dc_derived_tbls[NUM_HUFF_TBLS];
+  c_derived_tbl * ac_derived_tbls[NUM_HUFF_TBLS];
+
+  /* Statistics tables for optimization */
+  long * dc_count_ptrs[NUM_HUFF_TBLS];
+  long * ac_count_ptrs[NUM_HUFF_TBLS];
+
+  /* Following fields used only in progressive mode */
+
+  /* Mode flag: TRUE for optimization, FALSE for actual data output */
+  boolean gather_statistics;
+
+  /* next_output_byte/free_in_buffer are local copies of cinfo->dest fields.
+   */
+  JOCTET * next_output_byte;	/* => next byte to write in buffer */
+  size_t free_in_buffer;	/* # of byte spaces remaining in buffer */
+  j_compress_ptr cinfo;		/* link to cinfo (needed for dump_buffer) */
+
+  /* Coding status for AC components */
+  int ac_tbl_no;		/* the table number of the single component */
+  unsigned int EOBRUN;		/* run length of EOBs */
+  unsigned int BE;		/* # of buffered correction bits before MCU */
+  char * bit_buffer;		/* buffer for correction bits (1 per char) */
+  /* packing correction bits tightly would save some space but cost time... */
+} huff_entropy_encoder;
+
+typedef huff_entropy_encoder * huff_entropy_ptr;
+
+/* Working state while writing an MCU (sequential mode).
+ * This struct contains all the fields that are needed by subroutines.
+ */
+
+typedef struct {
+  JOCTET * next_output_byte;	/* => next byte to write in buffer */
+  size_t free_in_buffer;	/* # of byte spaces remaining in buffer */
+  savable_state cur;		/* Current bit buffer & DC state */
+  j_compress_ptr cinfo;		/* dump_buffer needs access to this */
+} working_state;
+
+/* MAX_CORR_BITS is the number of bits the AC refinement correction-bit
+ * buffer can hold.  Larger sizes may slightly improve compression, but
+ * 1000 is already well into the realm of overkill.
+ * The minimum safe size is 64 bits.
+ */
+
+#define MAX_CORR_BITS  1000	/* Max # of correction bits I can buffer */
+
+/* IRIGHT_SHIFT is like RIGHT_SHIFT, but works on int rather than INT32.
+ * We assume that int right shift is unsigned if INT32 right shift is,
+ * which should be safe.
+ */
+
+#ifdef RIGHT_SHIFT_IS_UNSIGNED
+#define ISHIFT_TEMPS	int ishift_temp;
+#define IRIGHT_SHIFT(x,shft)  \
+	((ishift_temp = (x)) < 0 ? \
+	 (ishift_temp >> (shft)) | ((~0) << (16-(shft))) : \
+	 (ishift_temp >> (shft)))
+#else
+#define ISHIFT_TEMPS
+#define IRIGHT_SHIFT(x,shft)	((x) >> (shft))
+#endif
+
+
+/*
+ * Compute the derived values for a Huffman table.
+ * This routine also performs some validation checks on the table.
+ */
+
+LOCAL(void)
+jpeg_make_c_derived_tbl (j_compress_ptr cinfo, boolean isDC, int tblno,
+			 c_derived_tbl ** pdtbl)
+{
+  JHUFF_TBL *htbl;
+  c_derived_tbl *dtbl;
+  int p, i, l, lastp, si, maxsymbol;
+  char huffsize[257];
+  unsigned int huffcode[257];
+  unsigned int code;
+
+  /* Note that huffsize[] and huffcode[] are filled in code-length order,
+   * paralleling the order of the symbols themselves in htbl->huffval[].
+   */
+
+  /* Find the input Huffman table */
+  if (tblno < 0 || tblno >= NUM_HUFF_TBLS)
+    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
+  htbl =
+    isDC ? cinfo->dc_huff_tbl_ptrs[tblno] : cinfo->ac_huff_tbl_ptrs[tblno];
+  if (htbl == NULL)
+    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
+
+  /* Allocate a workspace if we haven't already done so. */
+  if (*pdtbl == NULL)
+    *pdtbl = (c_derived_tbl *)
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				  SIZEOF(c_derived_tbl));
+  dtbl = *pdtbl;
+  
+  /* Figure C.1: make table of Huffman code length for each symbol */
+
+  p = 0;
+  for (l = 1; l <= 16; l++) {
+    i = (int) htbl->bits[l];
+    if (i < 0 || p + i > 256)	/* protect against table overrun */
+      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
+    while (i--)
+      huffsize[p++] = (char) l;
+  }
+  huffsize[p] = 0;
+  lastp = p;
+  
+  /* Figure C.2: generate the codes themselves */
+  /* We also validate that the counts represent a legal Huffman code tree. */
+
+  code = 0;
+  si = huffsize[0];
+  p = 0;
+  while (huffsize[p]) {
+    while (((int) huffsize[p]) == si) {
+      huffcode[p++] = code;
+      code++;
+    }
+    /* code is now 1 more than the last code used for codelength si; but
+     * it must still fit in si bits, since no code is allowed to be all ones.
+     */
+    if (((INT32) code) >= (((INT32) 1) << si))
+      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
+    code <<= 1;
+    si++;
+  }
+  
+  /* Figure C.3: generate encoding tables */
+  /* These are code and size indexed by symbol value */
+
+  /* Set all codeless symbols to have code length 0;
+   * this lets us detect duplicate VAL entries here, and later
+   * allows emit_bits to detect any attempt to emit such symbols.
+   */
+  MEMZERO(dtbl->ehufsi, SIZEOF(dtbl->ehufsi));
+
+  /* This is also a convenient place to check for out-of-range
+   * and duplicated VAL entries.  We allow 0..255 for AC symbols
+   * but only 0..15 for DC.  (We could constrain them further
+   * based on data depth and mode, but this seems enough.)
+   */
+  maxsymbol = isDC ? 15 : 255;
+
+  for (p = 0; p < lastp; p++) {
+    i = htbl->huffval[p];
+    if (i < 0 || i > maxsymbol || dtbl->ehufsi[i])
+      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
+    dtbl->ehufco[i] = huffcode[p];
+    dtbl->ehufsi[i] = huffsize[p];
+  }
+}
+
+
+/* Outputting bytes to the file.
+ * NB: these must be called only when actually outputting,
+ * that is, entropy->gather_statistics == FALSE.
+ */
+
+/* Emit a byte, taking 'action' if must suspend. */
+#define emit_byte_s(state,val,action)  \
+	{ *(state)->next_output_byte++ = (JOCTET) (val);  \
+	  if (--(state)->free_in_buffer == 0)  \
+	    if (! dump_buffer_s(state))  \
+	      { action; } }
+
+/* Emit a byte */
+#define emit_byte_e(entropy,val)  \
+	{ *(entropy)->next_output_byte++ = (JOCTET) (val);  \
+	  if (--(entropy)->free_in_buffer == 0)  \
+	    dump_buffer_e(entropy); }
+
+
+LOCAL(boolean)
+dump_buffer_s (working_state * state)
+/* Empty the output buffer; return TRUE if successful, FALSE if must suspend */
+{
+  struct jpeg_destination_mgr * dest = state->cinfo->dest;
+
+  if (! (*dest->empty_output_buffer) (state->cinfo))
+    return FALSE;
+  /* After a successful buffer dump, must reset buffer pointers */
+  state->next_output_byte = dest->next_output_byte;
+  state->free_in_buffer = dest->free_in_buffer;
+  return TRUE;
+}
+
+
+LOCAL(void)
+dump_buffer_e (huff_entropy_ptr entropy)
+/* Empty the output buffer; we do not support suspension in this case. */
+{
+  struct jpeg_destination_mgr * dest = entropy->cinfo->dest;
+
+  if (! (*dest->empty_output_buffer) (entropy->cinfo))
+    ERREXIT(entropy->cinfo, JERR_CANT_SUSPEND);
+  /* After a successful buffer dump, must reset buffer pointers */
+  entropy->next_output_byte = dest->next_output_byte;
+  entropy->free_in_buffer = dest->free_in_buffer;
+}
+
+
+/* Outputting bits to the file */
+
+/* Only the right 24 bits of put_buffer are used; the valid bits are
+ * left-justified in this part.  At most 16 bits can be passed to emit_bits
+ * in one call, and we never retain more than 7 bits in put_buffer
+ * between calls, so 24 bits are sufficient.
+ */
+
+INLINE
+LOCAL(boolean)
+emit_bits_s (working_state * state, unsigned int code, int size)
+/* Emit some bits; return TRUE if successful, FALSE if must suspend */
+{
+  /* This routine is heavily used, so it's worth coding tightly. */
+  register INT32 put_buffer = (INT32) code;
+  register int put_bits = state->cur.put_bits;
+
+  /* if size is 0, caller used an invalid Huffman table entry */
+  if (size == 0)
+    ERREXIT(state->cinfo, JERR_HUFF_MISSING_CODE);
+
+  put_buffer &= (((INT32) 1)<<size) - 1; /* mask off any extra bits in code */
+  
+  put_bits += size;		/* new number of bits in buffer */
+  
+  put_buffer <<= 24 - put_bits; /* align incoming bits */
+
+  put_buffer |= state->cur.put_buffer; /* and merge with old buffer contents */
+  
+  while (put_bits >= 8) {
+    int c = (int) ((put_buffer >> 16) & 0xFF);
+    
+    emit_byte_s(state, c, return FALSE);
+    if (c == 0xFF) {		/* need to stuff a zero byte? */
+      emit_byte_s(state, 0, return FALSE);
+    }
+    put_buffer <<= 8;
+    put_bits -= 8;
+  }
+
+  state->cur.put_buffer = put_buffer; /* update state variables */
+  state->cur.put_bits = put_bits;
+
+  return TRUE;
+}
+
+
+INLINE
+LOCAL(void)
+emit_bits_e (huff_entropy_ptr entropy, unsigned int code, int size)
+/* Emit some bits, unless we are in gather mode */
+{
+  /* This routine is heavily used, so it's worth coding tightly. */
+  register INT32 put_buffer = (INT32) code;
+  register int put_bits = entropy->saved.put_bits;
+
+  /* if size is 0, caller used an invalid Huffman table entry */
+  if (size == 0)
+    ERREXIT(entropy->cinfo, JERR_HUFF_MISSING_CODE);
+
+  if (entropy->gather_statistics)
+    return;			/* do nothing if we're only getting stats */
+
+  put_buffer &= (((INT32) 1)<<size) - 1; /* mask off any extra bits in code */
+  
+  put_bits += size;		/* new number of bits in buffer */
+
+  put_buffer <<= 24 - put_bits; /* align incoming bits */
+
+  /* and merge with old buffer contents */
+  put_buffer |= entropy->saved.put_buffer;
+
+  while (put_bits >= 8) {
+    int c = (int) ((put_buffer >> 16) & 0xFF);
+
+    emit_byte_e(entropy, c);
+    if (c == 0xFF) {		/* need to stuff a zero byte? */
+      emit_byte_e(entropy, 0);
+    }
+    put_buffer <<= 8;
+    put_bits -= 8;
+  }
+
+  entropy->saved.put_buffer = put_buffer; /* update variables */
+  entropy->saved.put_bits = put_bits;
+}
+
+
+LOCAL(boolean)
+flush_bits_s (working_state * state)
+{
+  if (! emit_bits_s(state, 0x7F, 7)) /* fill any partial byte with ones */
+    return FALSE;
+  state->cur.put_buffer = 0;	     /* and reset bit-buffer to empty */
+  state->cur.put_bits = 0;
+  return TRUE;
+}
+
+
+LOCAL(void)
+flush_bits_e (huff_entropy_ptr entropy)
+{
+  emit_bits_e(entropy, 0x7F, 7); /* fill any partial byte with ones */
+  entropy->saved.put_buffer = 0; /* and reset bit-buffer to empty */
+  entropy->saved.put_bits = 0;
+}
+
+
+/*
+ * Emit (or just count) a Huffman symbol.
+ */
+
+INLINE
+LOCAL(void)
+emit_dc_symbol (huff_entropy_ptr entropy, int tbl_no, int symbol)
+{
+  if (entropy->gather_statistics)
+    entropy->dc_count_ptrs[tbl_no][symbol]++;
+  else {
+    c_derived_tbl * tbl = entropy->dc_derived_tbls[tbl_no];
+    emit_bits_e(entropy, tbl->ehufco[symbol], tbl->ehufsi[symbol]);
+  }
+}
+
+
+INLINE
+LOCAL(void)
+emit_ac_symbol (huff_entropy_ptr entropy, int tbl_no, int symbol)
+{
+  if (entropy->gather_statistics)
+    entropy->ac_count_ptrs[tbl_no][symbol]++;
+  else {
+    c_derived_tbl * tbl = entropy->ac_derived_tbls[tbl_no];
+    emit_bits_e(entropy, tbl->ehufco[symbol], tbl->ehufsi[symbol]);
+  }
+}
+
+
+/*
+ * Emit bits from a correction bit buffer.
+ */
+
+LOCAL(void)
+emit_buffered_bits (huff_entropy_ptr entropy, char * bufstart,
+		    unsigned int nbits)
+{
+  if (entropy->gather_statistics)
+    return;			/* no real work */
+
+  while (nbits > 0) {
+    emit_bits_e(entropy, (unsigned int) (*bufstart), 1);
+    bufstart++;
+    nbits--;
+  }
+}
+
+
+/*
+ * Emit any pending EOBRUN symbol.
+ */
+
+LOCAL(void)
+emit_eobrun (huff_entropy_ptr entropy)
+{
+  register int temp, nbits;
+
+  if (entropy->EOBRUN > 0) {	/* if there is any pending EOBRUN */
+    temp = entropy->EOBRUN;
+    nbits = 0;
+    while ((temp >>= 1))
+      nbits++;
+    /* safety check: shouldn't happen given limited correction-bit buffer */
+    if (nbits > 14)
+      ERREXIT(entropy->cinfo, JERR_HUFF_MISSING_CODE);
+
+    emit_ac_symbol(entropy, entropy->ac_tbl_no, nbits << 4);
+    if (nbits)
+      emit_bits_e(entropy, entropy->EOBRUN, nbits);
+
+    entropy->EOBRUN = 0;
+
+    /* Emit any buffered correction bits */
+    emit_buffered_bits(entropy, entropy->bit_buffer, entropy->BE);
+    entropy->BE = 0;
+  }
+}
+
+
+/*
+ * Emit a restart marker & resynchronize predictions.
+ */
+
+LOCAL(boolean)
+emit_restart_s (working_state * state, int restart_num)
+{
+  int ci;
+
+  if (! flush_bits_s(state))
+    return FALSE;
+
+  emit_byte_s(state, 0xFF, return FALSE);
+  emit_byte_s(state, JPEG_RST0 + restart_num, return FALSE);
+
+  /* Re-initialize DC predictions to 0 */
+  for (ci = 0; ci < state->cinfo->comps_in_scan; ci++)
+    state->cur.last_dc_val[ci] = 0;
+
+  /* The restart counter is not updated until we successfully write the MCU. */
+
+  return TRUE;
+}
+
+
+LOCAL(void)
+emit_restart_e (huff_entropy_ptr entropy, int restart_num)
+{
+  int ci;
+
+  emit_eobrun(entropy);
+
+  if (! entropy->gather_statistics) {
+    flush_bits_e(entropy);
+    emit_byte_e(entropy, 0xFF);
+    emit_byte_e(entropy, JPEG_RST0 + restart_num);
+  }
+
+  if (entropy->cinfo->Ss == 0) {
+    /* Re-initialize DC predictions to 0 */
+    for (ci = 0; ci < entropy->cinfo->comps_in_scan; ci++)
+      entropy->saved.last_dc_val[ci] = 0;
+  } else {
+    /* Re-initialize all AC-related fields to 0 */
+    entropy->EOBRUN = 0;
+    entropy->BE = 0;
+  }
+}
+
+
+/*
+ * MCU encoding for DC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+encode_mcu_DC_first (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  register int temp, temp2;
+  register int nbits;
+  int blkn, ci;
+  int Al = cinfo->Al;
+  JBLOCKROW block;
+  jpeg_component_info * compptr;
+  ISHIFT_TEMPS
+
+  entropy->next_output_byte = cinfo->dest->next_output_byte;
+  entropy->free_in_buffer = cinfo->dest->free_in_buffer;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval)
+    if (entropy->restarts_to_go == 0)
+      emit_restart_e(entropy, entropy->next_restart_num);
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+    ci = cinfo->MCU_membership[blkn];
+    compptr = cinfo->cur_comp_info[ci];
+
+    /* Compute the DC value after the required point transform by Al.
+     * This is simply an arithmetic right shift.
+     */
+    temp2 = IRIGHT_SHIFT((int) ((*block)[0]), Al);
+
+    /* DC differences are figured on the point-transformed values. */
+    temp = temp2 - entropy->saved.last_dc_val[ci];
+    entropy->saved.last_dc_val[ci] = temp2;
+
+    /* Encode the DC coefficient difference per section G.1.2.1 */
+    temp2 = temp;
+    if (temp < 0) {
+      temp = -temp;		/* temp is abs value of input */
+      /* For a negative input, want temp2 = bitwise complement of abs(input) */
+      /* This code assumes we are on a two's complement machine */
+      temp2--;
+    }
+    
+    /* Find the number of bits needed for the magnitude of the coefficient */
+    nbits = 0;
+    while (temp) {
+      nbits++;
+      temp >>= 1;
+    }
+    /* Check for out-of-range coefficient values.
+     * Since we're encoding a difference, the range limit is twice as much.
+     */
+    if (nbits > MAX_COEF_BITS+1)
+      ERREXIT(cinfo, JERR_BAD_DCT_COEF);
+    
+    /* Count/emit the Huffman-coded symbol for the number of bits */
+    emit_dc_symbol(entropy, compptr->dc_tbl_no, nbits);
+    
+    /* Emit that number of bits of the value, if positive, */
+    /* or the complement of its magnitude, if negative. */
+    if (nbits)			/* emit_bits rejects calls with size 0 */
+      emit_bits_e(entropy, (unsigned int) temp2, nbits);
+  }
+
+  cinfo->dest->next_output_byte = entropy->next_output_byte;
+  cinfo->dest->free_in_buffer = entropy->free_in_buffer;
+
+  /* Update restart-interval state too */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for AC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+encode_mcu_AC_first (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  register int temp, temp2;
+  register int nbits;
+  register int r, k;
+  int Se, Al;
+  const int * natural_order;
+  JBLOCKROW block;
+
+  entropy->next_output_byte = cinfo->dest->next_output_byte;
+  entropy->free_in_buffer = cinfo->dest->free_in_buffer;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval)
+    if (entropy->restarts_to_go == 0)
+      emit_restart_e(entropy, entropy->next_restart_num);
+
+  Se = cinfo->Se;
+  Al = cinfo->Al;
+  natural_order = cinfo->natural_order;
+
+  /* Encode the MCU data block */
+  block = MCU_data[0];
+
+  /* Encode the AC coefficients per section G.1.2.2, fig. G.3 */
+  
+  r = 0;			/* r = run length of zeros */
+   
+  for (k = cinfo->Ss; k <= Se; k++) {
+    if ((temp = (*block)[natural_order[k]]) == 0) {
+      r++;
+      continue;
+    }
+    /* We must apply the point transform by Al.  For AC coefficients this
+     * is an integer division with rounding towards 0.  To do this portably
+     * in C, we shift after obtaining the absolute value; so the code is
+     * interwoven with finding the abs value (temp) and output bits (temp2).
+     */
+    if (temp < 0) {
+      temp = -temp;		/* temp is abs value of input */
+      temp >>= Al;		/* apply the point transform */
+      /* For a negative coef, want temp2 = bitwise complement of abs(coef) */
+      temp2 = ~temp;
+    } else {
+      temp >>= Al;		/* apply the point transform */
+      temp2 = temp;
+    }
+    /* Watch out for case that nonzero coef is zero after point transform */
+    if (temp == 0) {
+      r++;
+      continue;
+    }
+
+    /* Emit any pending EOBRUN */
+    if (entropy->EOBRUN > 0)
+      emit_eobrun(entropy);
+    /* if run length > 15, must emit special run-length-16 codes (0xF0) */
+    while (r > 15) {
+      emit_ac_symbol(entropy, entropy->ac_tbl_no, 0xF0);
+      r -= 16;
+    }
+
+    /* Find the number of bits needed for the magnitude of the coefficient */
+    nbits = 1;			/* there must be at least one 1 bit */
+    while ((temp >>= 1))
+      nbits++;
+    /* Check for out-of-range coefficient values */
+    if (nbits > MAX_COEF_BITS)
+      ERREXIT(cinfo, JERR_BAD_DCT_COEF);
+
+    /* Count/emit Huffman symbol for run length / number of bits */
+    emit_ac_symbol(entropy, entropy->ac_tbl_no, (r << 4) + nbits);
+
+    /* Emit that number of bits of the value, if positive, */
+    /* or the complement of its magnitude, if negative. */
+    emit_bits_e(entropy, (unsigned int) temp2, nbits);
+
+    r = 0;			/* reset zero run length */
+  }
+
+  if (r > 0) {			/* If there are trailing zeroes, */
+    entropy->EOBRUN++;		/* count an EOB */
+    if (entropy->EOBRUN == 0x7FFF)
+      emit_eobrun(entropy);	/* force it out to avoid overflow */
+  }
+
+  cinfo->dest->next_output_byte = entropy->next_output_byte;
+  cinfo->dest->free_in_buffer = entropy->free_in_buffer;
+
+  /* Update restart-interval state too */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for DC successive approximation refinement scan.
+ * Note: we assume such scans can be multi-component, although the spec
+ * is not very clear on the point.
+ */
+
+METHODDEF(boolean)
+encode_mcu_DC_refine (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  register int temp;
+  int blkn;
+  int Al = cinfo->Al;
+  JBLOCKROW block;
+
+  entropy->next_output_byte = cinfo->dest->next_output_byte;
+  entropy->free_in_buffer = cinfo->dest->free_in_buffer;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval)
+    if (entropy->restarts_to_go == 0)
+      emit_restart_e(entropy, entropy->next_restart_num);
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+
+    /* We simply emit the Al'th bit of the DC coefficient value. */
+    temp = (*block)[0];
+    emit_bits_e(entropy, (unsigned int) (temp >> Al), 1);
+  }
+
+  cinfo->dest->next_output_byte = entropy->next_output_byte;
+  cinfo->dest->free_in_buffer = entropy->free_in_buffer;
+
+  /* Update restart-interval state too */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for AC successive approximation refinement scan.
+ */
+
+METHODDEF(boolean)
+encode_mcu_AC_refine (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  register int temp;
+  register int r, k;
+  int EOB;
+  char *BR_buffer;
+  unsigned int BR;
+  int Se, Al;
+  const int * natural_order;
+  JBLOCKROW block;
+  int absvalues[DCTSIZE2];
+
+  entropy->next_output_byte = cinfo->dest->next_output_byte;
+  entropy->free_in_buffer = cinfo->dest->free_in_buffer;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval)
+    if (entropy->restarts_to_go == 0)
+      emit_restart_e(entropy, entropy->next_restart_num);
+
+  Se = cinfo->Se;
+  Al = cinfo->Al;
+  natural_order = cinfo->natural_order;
+
+  /* Encode the MCU data block */
+  block = MCU_data[0];
+
+  /* It is convenient to make a pre-pass to determine the transformed
+   * coefficients' absolute values and the EOB position.
+   */
+  EOB = 0;
+  for (k = cinfo->Ss; k <= Se; k++) {
+    temp = (*block)[natural_order[k]];
+    /* We must apply the point transform by Al.  For AC coefficients this
+     * is an integer division with rounding towards 0.  To do this portably
+     * in C, we shift after obtaining the absolute value.
+     */
+    if (temp < 0)
+      temp = -temp;		/* temp is abs value of input */
+    temp >>= Al;		/* apply the point transform */
+    absvalues[k] = temp;	/* save abs value for main pass */
+    if (temp == 1)
+      EOB = k;			/* EOB = index of last newly-nonzero coef */
+  }
+
+  /* Encode the AC coefficients per section G.1.2.3, fig. G.7 */
+  
+  r = 0;			/* r = run length of zeros */
+  BR = 0;			/* BR = count of buffered bits added now */
+  BR_buffer = entropy->bit_buffer + entropy->BE; /* Append bits to buffer */
+
+  for (k = cinfo->Ss; k <= Se; k++) {
+    if ((temp = absvalues[k]) == 0) {
+      r++;
+      continue;
+    }
+
+    /* Emit any required ZRLs, but not if they can be folded into EOB */
+    while (r > 15 && k <= EOB) {
+      /* emit any pending EOBRUN and the BE correction bits */
+      emit_eobrun(entropy);
+      /* Emit ZRL */
+      emit_ac_symbol(entropy, entropy->ac_tbl_no, 0xF0);
+      r -= 16;
+      /* Emit buffered correction bits that must be associated with ZRL */
+      emit_buffered_bits(entropy, BR_buffer, BR);
+      BR_buffer = entropy->bit_buffer; /* BE bits are gone now */
+      BR = 0;
+    }
+
+    /* If the coef was previously nonzero, it only needs a correction bit.
+     * NOTE: a straight translation of the spec's figure G.7 would suggest
+     * that we also need to test r > 15.  But if r > 15, we can only get here
+     * if k > EOB, which implies that this coefficient is not 1.
+     */
+    if (temp > 1) {
+      /* The correction bit is the next bit of the absolute value. */
+      BR_buffer[BR++] = (char) (temp & 1);
+      continue;
+    }
+
+    /* Emit any pending EOBRUN and the BE correction bits */
+    emit_eobrun(entropy);
+
+    /* Count/emit Huffman symbol for run length / number of bits */
+    emit_ac_symbol(entropy, entropy->ac_tbl_no, (r << 4) + 1);
+
+    /* Emit output bit for newly-nonzero coef */
+    temp = ((*block)[natural_order[k]] < 0) ? 0 : 1;
+    emit_bits_e(entropy, (unsigned int) temp, 1);
+
+    /* Emit buffered correction bits that must be associated with this code */
+    emit_buffered_bits(entropy, BR_buffer, BR);
+    BR_buffer = entropy->bit_buffer; /* BE bits are gone now */
+    BR = 0;
+    r = 0;			/* reset zero run length */
+  }
+
+  if (r > 0 || BR > 0) {	/* If there are trailing zeroes, */
+    entropy->EOBRUN++;		/* count an EOB */
+    entropy->BE += BR;		/* concat my correction bits to older ones */
+    /* We force out the EOB if we risk either:
+     * 1. overflow of the EOB counter;
+     * 2. overflow of the correction bit buffer during the next MCU.
+     */
+    if (entropy->EOBRUN == 0x7FFF || entropy->BE > (MAX_CORR_BITS-DCTSIZE2+1))
+      emit_eobrun(entropy);
+  }
+
+  cinfo->dest->next_output_byte = entropy->next_output_byte;
+  cinfo->dest->free_in_buffer = entropy->free_in_buffer;
+
+  /* Update restart-interval state too */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  return TRUE;
+}
+
+
+/* Encode a single block's worth of coefficients */
+
+LOCAL(boolean)
+encode_one_block (working_state * state, JCOEFPTR block, int last_dc_val,
+		  c_derived_tbl *dctbl, c_derived_tbl *actbl)
+{
+  register int temp, temp2;
+  register int nbits;
+  register int k, r, i;
+  int Se = state->cinfo->lim_Se;
+  const int * natural_order = state->cinfo->natural_order;
+
+  /* Encode the DC coefficient difference per section F.1.2.1 */
+
+  temp = temp2 = block[0] - last_dc_val;
+
+  if (temp < 0) {
+    temp = -temp;		/* temp is abs value of input */
+    /* For a negative input, want temp2 = bitwise complement of abs(input) */
+    /* This code assumes we are on a two's complement machine */
+    temp2--;
+  }
+
+  /* Find the number of bits needed for the magnitude of the coefficient */
+  nbits = 0;
+  while (temp) {
+    nbits++;
+    temp >>= 1;
+  }
+  /* Check for out-of-range coefficient values.
+   * Since we're encoding a difference, the range limit is twice as much.
+   */
+  if (nbits > MAX_COEF_BITS+1)
+    ERREXIT(state->cinfo, JERR_BAD_DCT_COEF);
+
+  /* Emit the Huffman-coded symbol for the number of bits */
+  if (! emit_bits_s(state, dctbl->ehufco[nbits], dctbl->ehufsi[nbits]))
+    return FALSE;
+
+  /* Emit that number of bits of the value, if positive, */
+  /* or the complement of its magnitude, if negative. */
+  if (nbits)			/* emit_bits rejects calls with size 0 */
+    if (! emit_bits_s(state, (unsigned int) temp2, nbits))
+      return FALSE;
+
+  /* Encode the AC coefficients per section F.1.2.2 */
+
+  r = 0;			/* r = run length of zeros */
+
+  for (k = 1; k <= Se; k++) {
+    if ((temp = block[natural_order[k]]) == 0) {
+      r++;
+    } else {
+      /* if run length > 15, must emit special run-length-16 codes (0xF0) */
+      while (r > 15) {
+	if (! emit_bits_s(state, actbl->ehufco[0xF0], actbl->ehufsi[0xF0]))
+	  return FALSE;
+	r -= 16;
+      }
+
+      temp2 = temp;
+      if (temp < 0) {
+	temp = -temp;		/* temp is abs value of input */
+	/* This code assumes we are on a two's complement machine */
+	temp2--;
+      }
+
+      /* Find the number of bits needed for the magnitude of the coefficient */
+      nbits = 1;		/* there must be at least one 1 bit */
+      while ((temp >>= 1))
+	nbits++;
+      /* Check for out-of-range coefficient values */
+      if (nbits > MAX_COEF_BITS)
+	ERREXIT(state->cinfo, JERR_BAD_DCT_COEF);
+
+      /* Emit Huffman symbol for run length / number of bits */
+      i = (r << 4) + nbits;
+      if (! emit_bits_s(state, actbl->ehufco[i], actbl->ehufsi[i]))
+	return FALSE;
+
+      /* Emit that number of bits of the value, if positive, */
+      /* or the complement of its magnitude, if negative. */
+      if (! emit_bits_s(state, (unsigned int) temp2, nbits))
+	return FALSE;
+
+      r = 0;
+    }
+  }
+
+  /* If the last coef(s) were zero, emit an end-of-block code */
+  if (r > 0)
+    if (! emit_bits_s(state, actbl->ehufco[0], actbl->ehufsi[0]))
+      return FALSE;
+
+  return TRUE;
+}
+
+
+/*
+ * Encode and output one MCU's worth of Huffman-compressed coefficients.
+ */
+
+METHODDEF(boolean)
+encode_mcu_huff (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  working_state state;
+  int blkn, ci;
+  jpeg_component_info * compptr;
+
+  /* Load up working state */
+  state.next_output_byte = cinfo->dest->next_output_byte;
+  state.free_in_buffer = cinfo->dest->free_in_buffer;
+  ASSIGN_STATE(state.cur, entropy->saved);
+  state.cinfo = cinfo;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      if (! emit_restart_s(&state, entropy->next_restart_num))
+	return FALSE;
+  }
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    ci = cinfo->MCU_membership[blkn];
+    compptr = cinfo->cur_comp_info[ci];
+    if (! encode_one_block(&state,
+			   MCU_data[blkn][0], state.cur.last_dc_val[ci],
+			   entropy->dc_derived_tbls[compptr->dc_tbl_no],
+			   entropy->ac_derived_tbls[compptr->ac_tbl_no]))
+      return FALSE;
+    /* Update last_dc_val */
+    state.cur.last_dc_val[ci] = MCU_data[blkn][0][0];
+  }
+
+  /* Completed MCU, so update state */
+  cinfo->dest->next_output_byte = state.next_output_byte;
+  cinfo->dest->free_in_buffer = state.free_in_buffer;
+  ASSIGN_STATE(entropy->saved, state.cur);
+
+  /* Update restart-interval state too */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * Finish up at the end of a Huffman-compressed scan.
+ */
+
+METHODDEF(void)
+finish_pass_huff (j_compress_ptr cinfo)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  working_state state;
+
+  if (cinfo->progressive_mode) {
+    entropy->next_output_byte = cinfo->dest->next_output_byte;
+    entropy->free_in_buffer = cinfo->dest->free_in_buffer;
+
+    /* Flush out any buffered data */
+    emit_eobrun(entropy);
+    flush_bits_e(entropy);
+
+    cinfo->dest->next_output_byte = entropy->next_output_byte;
+    cinfo->dest->free_in_buffer = entropy->free_in_buffer;
+  } else {
+    /* Load up working state ... flush_bits needs it */
+    state.next_output_byte = cinfo->dest->next_output_byte;
+    state.free_in_buffer = cinfo->dest->free_in_buffer;
+    ASSIGN_STATE(state.cur, entropy->saved);
+    state.cinfo = cinfo;
+
+    /* Flush out the last data */
+    if (! flush_bits_s(&state))
+      ERREXIT(cinfo, JERR_CANT_SUSPEND);
+
+    /* Update state */
+    cinfo->dest->next_output_byte = state.next_output_byte;
+    cinfo->dest->free_in_buffer = state.free_in_buffer;
+    ASSIGN_STATE(entropy->saved, state.cur);
+  }
+}
+
+
+/*
+ * Huffman coding optimization.
+ *
+ * We first scan the supplied data and count the number of uses of each symbol
+ * that is to be Huffman-coded. (This process MUST agree with the code above.)
+ * Then we build a Huffman coding tree for the observed counts.
+ * Symbols which are not needed at all for the particular image are not
+ * assigned any code, which saves space in the DHT marker as well as in
+ * the compressed data.
+ */
+
+
+/* Process a single block's worth of coefficients */
+
+LOCAL(void)
+htest_one_block (j_compress_ptr cinfo, JCOEFPTR block, int last_dc_val,
+		 long dc_counts[], long ac_counts[])
+{
+  register int temp;
+  register int nbits;
+  register int k, r;
+  int Se = cinfo->lim_Se;
+  const int * natural_order = cinfo->natural_order;
+  
+  /* Encode the DC coefficient difference per section F.1.2.1 */
+  
+  temp = block[0] - last_dc_val;
+  if (temp < 0)
+    temp = -temp;
+  
+  /* Find the number of bits needed for the magnitude of the coefficient */
+  nbits = 0;
+  while (temp) {
+    nbits++;
+    temp >>= 1;
+  }
+  /* Check for out-of-range coefficient values.
+   * Since we're encoding a difference, the range limit is twice as much.
+   */
+  if (nbits > MAX_COEF_BITS+1)
+    ERREXIT(cinfo, JERR_BAD_DCT_COEF);
+
+  /* Count the Huffman symbol for the number of bits */
+  dc_counts[nbits]++;
+  
+  /* Encode the AC coefficients per section F.1.2.2 */
+  
+  r = 0;			/* r = run length of zeros */
+  
+  for (k = 1; k <= Se; k++) {
+    if ((temp = block[natural_order[k]]) == 0) {
+      r++;
+    } else {
+      /* if run length > 15, must emit special run-length-16 codes (0xF0) */
+      while (r > 15) {
+	ac_counts[0xF0]++;
+	r -= 16;
+      }
+      
+      /* Find the number of bits needed for the magnitude of the coefficient */
+      if (temp < 0)
+	temp = -temp;
+      
+      /* Find the number of bits needed for the magnitude of the coefficient */
+      nbits = 1;		/* there must be at least one 1 bit */
+      while ((temp >>= 1))
+	nbits++;
+      /* Check for out-of-range coefficient values */
+      if (nbits > MAX_COEF_BITS)
+	ERREXIT(cinfo, JERR_BAD_DCT_COEF);
+      
+      /* Count Huffman symbol for run length / number of bits */
+      ac_counts[(r << 4) + nbits]++;
+      
+      r = 0;
+    }
+  }
+
+  /* If the last coef(s) were zero, emit an end-of-block code */
+  if (r > 0)
+    ac_counts[0]++;
+}
+
+
+/*
+ * Trial-encode one MCU's worth of Huffman-compressed coefficients.
+ * No data is actually output, so no suspension return is possible.
+ */
+
+METHODDEF(boolean)
+encode_mcu_gather (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int blkn, ci;
+  jpeg_component_info * compptr;
+
+  /* Take care of restart intervals if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      /* Re-initialize DC predictions to 0 */
+      for (ci = 0; ci < cinfo->comps_in_scan; ci++)
+	entropy->saved.last_dc_val[ci] = 0;
+      /* Update restart state */
+      entropy->restarts_to_go = cinfo->restart_interval;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    ci = cinfo->MCU_membership[blkn];
+    compptr = cinfo->cur_comp_info[ci];
+    htest_one_block(cinfo, MCU_data[blkn][0], entropy->saved.last_dc_val[ci],
+		    entropy->dc_count_ptrs[compptr->dc_tbl_no],
+		    entropy->ac_count_ptrs[compptr->ac_tbl_no]);
+    entropy->saved.last_dc_val[ci] = MCU_data[blkn][0][0];
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * Generate the best Huffman code table for the given counts, fill htbl.
+ *
+ * The JPEG standard requires that no symbol be assigned a codeword of all
+ * one bits (so that padding bits added at the end of a compressed segment
+ * can't look like a valid code).  Because of the canonical ordering of
+ * codewords, this just means that there must be an unused slot in the
+ * longest codeword length category.  Section K.2 of the JPEG spec suggests
+ * reserving such a slot by pretending that symbol 256 is a valid symbol
+ * with count 1.  In theory that's not optimal; giving it count zero but
+ * including it in the symbol set anyway should give a better Huffman code.
+ * But the theoretically better code actually seems to come out worse in
+ * practice, because it produces more all-ones bytes (which incur stuffed
+ * zero bytes in the final file).  In any case the difference is tiny.
+ *
+ * The JPEG standard requires Huffman codes to be no more than 16 bits long.
+ * If some symbols have a very small but nonzero probability, the Huffman tree
+ * must be adjusted to meet the code length restriction.  We currently use
+ * the adjustment method suggested in JPEG section K.2.  This method is *not*
+ * optimal; it may not choose the best possible limited-length code.  But
+ * typically only very-low-frequency symbols will be given less-than-optimal
+ * lengths, so the code is almost optimal.  Experimental comparisons against
+ * an optimal limited-length-code algorithm indicate that the difference is
+ * microscopic --- usually less than a hundredth of a percent of total size.
+ * So the extra complexity of an optimal algorithm doesn't seem worthwhile.
+ */
+
+LOCAL(void)
+jpeg_gen_optimal_table (j_compress_ptr cinfo, JHUFF_TBL * htbl, long freq[])
+{
+#define MAX_CLEN 32		/* assumed maximum initial code length */
+  UINT8 bits[MAX_CLEN+1];	/* bits[k] = # of symbols with code length k */
+  int codesize[257];		/* codesize[k] = code length of symbol k */
+  int others[257];		/* next symbol in current branch of tree */
+  int c1, c2;
+  int p, i, j;
+  long v;
+
+  /* This algorithm is explained in section K.2 of the JPEG standard */
+
+  MEMZERO(bits, SIZEOF(bits));
+  MEMZERO(codesize, SIZEOF(codesize));
+  for (i = 0; i < 257; i++)
+    others[i] = -1;		/* init links to empty */
+  
+  freq[256] = 1;		/* make sure 256 has a nonzero count */
+  /* Including the pseudo-symbol 256 in the Huffman procedure guarantees
+   * that no real symbol is given code-value of all ones, because 256
+   * will be placed last in the largest codeword category.
+   */
+
+  /* Huffman's basic algorithm to assign optimal code lengths to symbols */
+
+  for (;;) {
+    /* Find the smallest nonzero frequency, set c1 = its symbol */
+    /* In case of ties, take the larger symbol number */
+    c1 = -1;
+    v = 1000000000L;
+    for (i = 0; i <= 256; i++) {
+      if (freq[i] && freq[i] <= v) {
+	v = freq[i];
+	c1 = i;
+      }
+    }
+
+    /* Find the next smallest nonzero frequency, set c2 = its symbol */
+    /* In case of ties, take the larger symbol number */
+    c2 = -1;
+    v = 1000000000L;
+    for (i = 0; i <= 256; i++) {
+      if (freq[i] && freq[i] <= v && i != c1) {
+	v = freq[i];
+	c2 = i;
+      }
+    }
+
+    /* Done if we've merged everything into one frequency */
+    if (c2 < 0)
+      break;
+    
+    /* Else merge the two counts/trees */
+    freq[c1] += freq[c2];
+    freq[c2] = 0;
+
+    /* Increment the codesize of everything in c1's tree branch */
+    codesize[c1]++;
+    while (others[c1] >= 0) {
+      c1 = others[c1];
+      codesize[c1]++;
+    }
+    
+    others[c1] = c2;		/* chain c2 onto c1's tree branch */
+    
+    /* Increment the codesize of everything in c2's tree branch */
+    codesize[c2]++;
+    while (others[c2] >= 0) {
+      c2 = others[c2];
+      codesize[c2]++;
+    }
+  }
+
+  /* Now count the number of symbols of each code length */
+  for (i = 0; i <= 256; i++) {
+    if (codesize[i]) {
+      /* The JPEG standard seems to think that this can't happen, */
+      /* but I'm paranoid... */
+      if (codesize[i] > MAX_CLEN)
+	ERREXIT(cinfo, JERR_HUFF_CLEN_OVERFLOW);
+
+      bits[codesize[i]]++;
+    }
+  }
+
+  /* JPEG doesn't allow symbols with code lengths over 16 bits, so if the pure
+   * Huffman procedure assigned any such lengths, we must adjust the coding.
+   * Here is what the JPEG spec says about how this next bit works:
+   * Since symbols are paired for the longest Huffman code, the symbols are
+   * removed from this length category two at a time.  The prefix for the pair
+   * (which is one bit shorter) is allocated to one of the pair; then,
+   * skipping the BITS entry for that prefix length, a code word from the next
+   * shortest nonzero BITS entry is converted into a prefix for two code words
+   * one bit longer.
+   */
+  
+  for (i = MAX_CLEN; i > 16; i--) {
+    while (bits[i] > 0) {
+      j = i - 2;		/* find length of new prefix to be used */
+      while (bits[j] == 0)
+	j--;
+      
+      bits[i] -= 2;		/* remove two symbols */
+      bits[i-1]++;		/* one goes in this length */
+      bits[j+1] += 2;		/* two new symbols in this length */
+      bits[j]--;		/* symbol of this length is now a prefix */
+    }
+  }
+
+  /* Remove the count for the pseudo-symbol 256 from the largest codelength */
+  while (bits[i] == 0)		/* find largest codelength still in use */
+    i--;
+  bits[i]--;
+  
+  /* Return final symbol counts (only for lengths 0..16) */
+  MEMCOPY(htbl->bits, bits, SIZEOF(htbl->bits));
+  
+  /* Return a list of the symbols sorted by code length */
+  /* It's not real clear to me why we don't need to consider the codelength
+   * changes made above, but the JPEG spec seems to think this works.
+   */
+  p = 0;
+  for (i = 1; i <= MAX_CLEN; i++) {
+    for (j = 0; j <= 255; j++) {
+      if (codesize[j] == i) {
+	htbl->huffval[p] = (UINT8) j;
+	p++;
+      }
+    }
+  }
+
+  /* Set sent_table FALSE so updated table will be written to JPEG file. */
+  htbl->sent_table = FALSE;
+}
+
+
+/*
+ * Finish up a statistics-gathering pass and create the new Huffman tables.
+ */
+
+METHODDEF(void)
+finish_pass_gather (j_compress_ptr cinfo)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int ci, tbl;
+  jpeg_component_info * compptr;
+  JHUFF_TBL **htblptr;
+  boolean did_dc[NUM_HUFF_TBLS];
+  boolean did_ac[NUM_HUFF_TBLS];
+
+  /* It's important not to apply jpeg_gen_optimal_table more than once
+   * per table, because it clobbers the input frequency counts!
+   */
+  if (cinfo->progressive_mode)
+    /* Flush out buffered data (all we care about is counting the EOB symbol) */
+    emit_eobrun(entropy);
+
+  MEMZERO(did_dc, SIZEOF(did_dc));
+  MEMZERO(did_ac, SIZEOF(did_ac));
+
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    /* DC needs no table for refinement scan */
+    if (cinfo->Ss == 0 && cinfo->Ah == 0) {
+      tbl = compptr->dc_tbl_no;
+      if (! did_dc[tbl]) {
+	htblptr = & cinfo->dc_huff_tbl_ptrs[tbl];
+	if (*htblptr == NULL)
+	  *htblptr = jpeg_alloc_huff_table((j_common_ptr) cinfo);
+	jpeg_gen_optimal_table(cinfo, *htblptr, entropy->dc_count_ptrs[tbl]);
+	did_dc[tbl] = TRUE;
+      }
+    }
+    /* AC needs no table when not present */
+    if (cinfo->Se) {
+      tbl = compptr->ac_tbl_no;
+      if (! did_ac[tbl]) {
+	htblptr = & cinfo->ac_huff_tbl_ptrs[tbl];
+	if (*htblptr == NULL)
+	  *htblptr = jpeg_alloc_huff_table((j_common_ptr) cinfo);
+	jpeg_gen_optimal_table(cinfo, *htblptr, entropy->ac_count_ptrs[tbl]);
+	did_ac[tbl] = TRUE;
+      }
+    }
+  }
+}
+
+
+/*
+ * Initialize for a Huffman-compressed scan.
+ * If gather_statistics is TRUE, we do not output anything during the scan,
+ * just count the Huffman symbols used and generate Huffman code tables.
+ */
+
+METHODDEF(void)
+start_pass_huff (j_compress_ptr cinfo, boolean gather_statistics)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int ci, tbl;
+  jpeg_component_info * compptr;
+
+  if (gather_statistics)
+    entropy->pub.finish_pass = finish_pass_gather;
+  else
+    entropy->pub.finish_pass = finish_pass_huff;
+
+  if (cinfo->progressive_mode) {
+    entropy->cinfo = cinfo;
+    entropy->gather_statistics = gather_statistics;
+
+    /* We assume jcmaster.c already validated the scan parameters. */
+
+    /* Select execution routine */
+    if (cinfo->Ah == 0) {
+      if (cinfo->Ss == 0)
+	entropy->pub.encode_mcu = encode_mcu_DC_first;
+      else
+	entropy->pub.encode_mcu = encode_mcu_AC_first;
+    } else {
+      if (cinfo->Ss == 0)
+	entropy->pub.encode_mcu = encode_mcu_DC_refine;
+      else {
+	entropy->pub.encode_mcu = encode_mcu_AC_refine;
+	/* AC refinement needs a correction bit buffer */
+	if (entropy->bit_buffer == NULL)
+	  entropy->bit_buffer = (char *)
+	    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+					MAX_CORR_BITS * SIZEOF(char));
+      }
+    }
+
+    /* Initialize AC stuff */
+    entropy->ac_tbl_no = cinfo->cur_comp_info[0]->ac_tbl_no;
+    entropy->EOBRUN = 0;
+    entropy->BE = 0;
+  } else {
+    if (gather_statistics)
+      entropy->pub.encode_mcu = encode_mcu_gather;
+    else
+      entropy->pub.encode_mcu = encode_mcu_huff;
+  }
+
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    /* DC needs no table for refinement scan */
+    if (cinfo->Ss == 0 && cinfo->Ah == 0) {
+      tbl = compptr->dc_tbl_no;
+      if (gather_statistics) {
+	/* Check for invalid table index */
+	/* (make_c_derived_tbl does this in the other path) */
+	if (tbl < 0 || tbl >= NUM_HUFF_TBLS)
+	  ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tbl);
+	/* Allocate and zero the statistics tables */
+	/* Note that jpeg_gen_optimal_table expects 257 entries in each table! */
+	if (entropy->dc_count_ptrs[tbl] == NULL)
+	  entropy->dc_count_ptrs[tbl] = (long *)
+	    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+					257 * SIZEOF(long));
+	MEMZERO(entropy->dc_count_ptrs[tbl], 257 * SIZEOF(long));
+      } else {
+	/* Compute derived values for Huffman tables */
+	/* We may do this more than once for a table, but it's not expensive */
+	jpeg_make_c_derived_tbl(cinfo, TRUE, tbl,
+				& entropy->dc_derived_tbls[tbl]);
+      }
+      /* Initialize DC predictions to 0 */
+      entropy->saved.last_dc_val[ci] = 0;
+    }
+    /* AC needs no table when not present */
+    if (cinfo->Se) {
+      tbl = compptr->ac_tbl_no;
+      if (gather_statistics) {
+	if (tbl < 0 || tbl >= NUM_HUFF_TBLS)
+	  ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tbl);
+	if (entropy->ac_count_ptrs[tbl] == NULL)
+	  entropy->ac_count_ptrs[tbl] = (long *)
+	    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+					257 * SIZEOF(long));
+	MEMZERO(entropy->ac_count_ptrs[tbl], 257 * SIZEOF(long));
+      } else {
+	jpeg_make_c_derived_tbl(cinfo, FALSE, tbl,
+				& entropy->ac_derived_tbls[tbl]);
+      }
+    }
+  }
+
+  /* Initialize bit buffer to empty */
+  entropy->saved.put_buffer = 0;
+  entropy->saved.put_bits = 0;
+
+  /* Initialize restart stuff */
+  entropy->restarts_to_go = cinfo->restart_interval;
+  entropy->next_restart_num = 0;
+}
+
+
+/*
+ * Module initialization routine for Huffman entropy encoding.
+ */
+
+GLOBAL(void)
+jinit_huff_encoder (j_compress_ptr cinfo)
+{
+  huff_entropy_ptr entropy;
+  int i;
+
+  entropy = (huff_entropy_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				SIZEOF(huff_entropy_encoder));
+  cinfo->entropy = (struct jpeg_entropy_encoder *) entropy;
+  entropy->pub.start_pass = start_pass_huff;
+
+  /* Mark tables unallocated */
+  for (i = 0; i < NUM_HUFF_TBLS; i++) {
+    entropy->dc_derived_tbls[i] = entropy->ac_derived_tbls[i] = NULL;
+    entropy->dc_count_ptrs[i] = entropy->ac_count_ptrs[i] = NULL;
+  }
+
+  if (cinfo->progressive_mode)
+    entropy->bit_buffer = NULL;	/* needed only in AC refinement scan */
+}
diff --git a/src/jpeg-6b/jcinit.c b/src/jpeg-8c/jcinit.c
index 5efffe33..0ba310f2 100644
--- a/src/jpeg-6b/jcinit.c
+++ b/src/jpeg-8c/jcinit.c
@@ -41,17 +41,10 @@ jinit_compress_master (j_compress_ptr cinfo)
   /* Forward DCT */
   jinit_forward_dct(cinfo);
   /* Entropy encoding: either Huffman or arithmetic coding. */
-  if (cinfo->arith_code) {
-    ERREXIT(cinfo, JERR_ARITH_NOTIMPL);
-  } else {
-    if (cinfo->progressive_mode) {
-#ifdef C_PROGRESSIVE_SUPPORTED
-      jinit_phuff_encoder(cinfo);
-#else
-      ERREXIT(cinfo, JERR_NOT_COMPILED);
-#endif
-    } else
-      jinit_huff_encoder(cinfo);
+  if (cinfo->arith_code)
+    jinit_arith_encoder(cinfo);
+  else {
+    jinit_huff_encoder(cinfo);
   }
 
   /* Need a full-image coefficient buffer in any multi-pass mode. */
diff --git a/src/jpeg-6b/jcmainct.c b/src/jpeg-8c/jcmainct.c
index 03a61a7e..b9f525b2 100644
--- a/src/jpeg-6b/jcmainct.c
+++ b/src/jpeg-8c/jcmainct.c
@@ -68,32 +68,32 @@ METHODDEF(void) process_data_buffer_main
 METHODDEF(void)
 start_pass_main (j_compress_ptr cinfo, J_BUF_MODE pass_mode)
 {
-  my_main_ptr jmain = (my_main_ptr) cinfo->main;
+  my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
 
   /* Do nothing in raw-data mode. */
   if (cinfo->raw_data_in)
     return;
 
-  jmain->cur_iMCU_row = 0;	/* initialize counters */
-  jmain->rowgroup_ctr = 0;
-  jmain->suspended = FALSE;
-  jmain->pass_mode = pass_mode;	/* save mode for use by process_data */
+  main_ptr->cur_iMCU_row = 0;	/* initialize counters */
+  main_ptr->rowgroup_ctr = 0;
+  main_ptr->suspended = FALSE;
+  main_ptr->pass_mode = pass_mode;	/* save mode for use by process_data */
 
   switch (pass_mode) {
   case JBUF_PASS_THRU:
 #ifdef FULL_MAIN_BUFFER_SUPPORTED
-    if (jmain->whole_image[0] != NULL)
+    if (main_ptr->whole_image[0] != NULL)
       ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
 #endif
-    jmain->pub.process_data = process_data_simple_main;
+    main_ptr->pub.process_data = process_data_simple_main;
     break;
 #ifdef FULL_MAIN_BUFFER_SUPPORTED
   case JBUF_SAVE_SOURCE:
   case JBUF_CRANK_DEST:
   case JBUF_SAVE_AND_PASS:
-    if (jmain->whole_image[0] == NULL)
+    if (main_ptr->whole_image[0] == NULL)
       ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
-    jmain->pub.process_data = process_data_buffer_main;
+    main_ptr->pub.process_data = process_data_buffer_main;
     break;
 #endif
   default:
@@ -114,46 +114,46 @@ process_data_simple_main (j_compress_ptr cinfo,
 			  JSAMPARRAY input_buf, JDIMENSION *in_row_ctr,
 			  JDIMENSION in_rows_avail)
 {
-  my_main_ptr jmain = (my_main_ptr) cinfo->main;
+  my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
 
-  while (jmain->cur_iMCU_row < cinfo->total_iMCU_rows) {
-    /* Read input data if we haven't filled the jmain buffer yet */
-    if (jmain->rowgroup_ctr < DCTSIZE)
+  while (main_ptr->cur_iMCU_row < cinfo->total_iMCU_rows) {
+    /* Read input data if we haven't filled the main buffer yet */
+    if (main_ptr->rowgroup_ctr < (JDIMENSION) cinfo->min_DCT_v_scaled_size)
       (*cinfo->prep->pre_process_data) (cinfo,
 					input_buf, in_row_ctr, in_rows_avail,
-					jmain->buffer, &jmain->rowgroup_ctr,
-					(JDIMENSION) DCTSIZE);
+					main_ptr->buffer, &main_ptr->rowgroup_ctr,
+					(JDIMENSION) cinfo->min_DCT_v_scaled_size);
 
     /* If we don't have a full iMCU row buffered, return to application for
      * more data.  Note that preprocessor will always pad to fill the iMCU row
      * at the bottom of the image.
      */
-    if (jmain->rowgroup_ctr != DCTSIZE)
+    if (main_ptr->rowgroup_ctr != (JDIMENSION) cinfo->min_DCT_v_scaled_size)
       return;
 
     /* Send the completed row to the compressor */
-    if (! (*cinfo->coef->compress_data) (cinfo, jmain->buffer)) {
+    if (! (*cinfo->coef->compress_data) (cinfo, main_ptr->buffer)) {
       /* If compressor did not consume the whole row, then we must need to
        * suspend processing and return to the application.  In this situation
        * we pretend we didn't yet consume the last input row; otherwise, if
        * it happened to be the last row of the image, the application would
        * think we were done.
        */
-      if (! jmain->suspended) {
+      if (! main_ptr->suspended) {
 	(*in_row_ctr)--;
-	jmain->suspended = TRUE;
+	main_ptr->suspended = TRUE;
       }
       return;
     }
     /* We did finish the row.  Undo our little suspension hack if a previous
-     * call suspended; then mark the jmain buffer empty.
+     * call suspended; then mark the main buffer empty.
      */
-    if (jmain->suspended) {
+    if (main_ptr->suspended) {
       (*in_row_ctr)++;
-      jmain->suspended = FALSE;
+      main_ptr->suspended = FALSE;
     }
-    jmain->rowgroup_ctr = 0;
-    jmain->cur_iMCU_row++;
+    main_ptr->rowgroup_ctr = 0;
+    main_ptr->cur_iMCU_row++;
   }
 }
 
@@ -170,25 +170,25 @@ process_data_buffer_main (j_compress_ptr cinfo,
 			  JSAMPARRAY input_buf, JDIMENSION *in_row_ctr,
 			  JDIMENSION in_rows_avail)
 {
-  my_main_ptr jmain = (my_main_ptr) cinfo->main;
+  my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
   int ci;
   jpeg_component_info *compptr;
-  boolean writing = (jmain->pass_mode != JBUF_CRANK_DEST);
+  boolean writing = (main_ptr->pass_mode != JBUF_CRANK_DEST);
 
-  while (jmain->cur_iMCU_row < cinfo->total_iMCU_rows) {
+  while (main_ptr->cur_iMCU_row < cinfo->total_iMCU_rows) {
     /* Realign the virtual buffers if at the start of an iMCU row. */
-    if (jmain->rowgroup_ctr == 0) {
+    if (main_ptr->rowgroup_ctr == 0) {
       for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
 	   ci++, compptr++) {
-	jmain->buffer[ci] = (*cinfo->mem->access_virt_sarray)
-	  ((j_common_ptr) cinfo, jmain->whole_image[ci],
-	   jmain->cur_iMCU_row * (compptr->v_samp_factor * DCTSIZE),
+	main_ptr->buffer[ci] = (*cinfo->mem->access_virt_sarray)
+	  ((j_common_ptr) cinfo, main_ptr->whole_image[ci],
+	   main_ptr->cur_iMCU_row * (compptr->v_samp_factor * DCTSIZE),
 	   (JDIMENSION) (compptr->v_samp_factor * DCTSIZE), writing);
       }
       /* In a read pass, pretend we just read some source data. */
       if (! writing) {
 	*in_row_ctr += cinfo->max_v_samp_factor * DCTSIZE;
-	jmain->rowgroup_ctr = DCTSIZE;
+	main_ptr->rowgroup_ctr = DCTSIZE;
       }
     }
 
@@ -197,40 +197,40 @@ process_data_buffer_main (j_compress_ptr cinfo,
     if (writing) {
       (*cinfo->prep->pre_process_data) (cinfo,
 					input_buf, in_row_ctr, in_rows_avail,
-					jmain->buffer, &jmain->rowgroup_ctr,
+					main_ptr->buffer, &main_ptr->rowgroup_ctr,
 					(JDIMENSION) DCTSIZE);
       /* Return to application if we need more data to fill the iMCU row. */
-      if (jmain->rowgroup_ctr < DCTSIZE)
+      if (main_ptr->rowgroup_ctr < DCTSIZE)
 	return;
     }
 
     /* Emit data, unless this is a sink-only pass. */
-    if (jmain->pass_mode != JBUF_SAVE_SOURCE) {
-      if (! (*cinfo->coef->compress_data) (cinfo, jmain->buffer)) {
+    if (main_ptr->pass_mode != JBUF_SAVE_SOURCE) {
+      if (! (*cinfo->coef->compress_data) (cinfo, main_ptr->buffer)) {
 	/* If compressor did not consume the whole row, then we must need to
 	 * suspend processing and return to the application.  In this situation
 	 * we pretend we didn't yet consume the last input row; otherwise, if
 	 * it happened to be the last row of the image, the application would
 	 * think we were done.
 	 */
-	if (! jmain->suspended) {
+	if (! main_ptr->suspended) {
 	  (*in_row_ctr)--;
-	  jmain->suspended = TRUE;
+	  main_ptr->suspended = TRUE;
 	}
 	return;
       }
       /* We did finish the row.  Undo our little suspension hack if a previous
-       * call suspended; then mark the jmain buffer empty.
+       * call suspended; then mark the main buffer empty.
        */
-      if (jmain->suspended) {
+      if (main_ptr->suspended) {
 	(*in_row_ctr)++;
-	jmain->suspended = FALSE;
+	main_ptr->suspended = FALSE;
       }
     }
 
     /* If get here, we are done with this iMCU row.  Mark buffer empty. */
-    jmain->rowgroup_ctr = 0;
-    jmain->cur_iMCU_row++;
+    main_ptr->rowgroup_ctr = 0;
+    main_ptr->cur_iMCU_row++;
   }
 }
 
@@ -238,21 +238,21 @@ process_data_buffer_main (j_compress_ptr cinfo,
 
 
 /*
- * Initialize jmain buffer controller.
+ * Initialize main buffer controller.
  */
 
 GLOBAL(void)
 jinit_c_main_controller (j_compress_ptr cinfo, boolean need_full_buffer)
 {
-  my_main_ptr jmain;
+  my_main_ptr main_ptr;
   int ci;
   jpeg_component_info *compptr;
 
-  jmain = (my_main_ptr)
+  main_ptr = (my_main_ptr)
     (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
 				SIZEOF(my_main_controller));
-  cinfo->main = (struct jpeg_c_main_controller *) jmain;
-  jmain->pub.start_pass = start_pass_main;
+  cinfo->main = (struct jpeg_c_main_controller *) main_ptr;
+  main_ptr->pub.start_pass = start_pass_main;
 
   /* We don't need to create a buffer in raw-data mode. */
   if (cinfo->raw_data_in)
@@ -267,27 +267,27 @@ jinit_c_main_controller (j_compress_ptr cinfo, boolean need_full_buffer)
     /* Note we pad the bottom to a multiple of the iMCU height */
     for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
 	 ci++, compptr++) {
-      jmain->whole_image[ci] = (*cinfo->mem->request_virt_sarray)
+      main_ptr->whole_image[ci] = (*cinfo->mem->request_virt_sarray)
 	((j_common_ptr) cinfo, JPOOL_IMAGE, FALSE,
-	 compptr->width_in_blocks * DCTSIZE,
+	 compptr->width_in_blocks * compptr->DCT_h_scaled_size,
 	 (JDIMENSION) jround_up((long) compptr->height_in_blocks,
 				(long) compptr->v_samp_factor) * DCTSIZE,
-	 (JDIMENSION) (compptr->v_samp_factor * DCTSIZE));
+	 (JDIMENSION) (compptr->v_samp_factor * compptr->DCT_v_scaled_size));
     }
 #else
     ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
 #endif
   } else {
 #ifdef FULL_MAIN_BUFFER_SUPPORTED
-    jmain->whole_image[0] = NULL; /* flag for no virtual arrays */
+    main_ptr->whole_image[0] = NULL; /* flag for no virtual arrays */
 #endif
     /* Allocate a strip buffer for each component */
     for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
 	 ci++, compptr++) {
-      jmain->buffer[ci] = (*cinfo->mem->alloc_sarray)
+      main_ptr->buffer[ci] = (*cinfo->mem->alloc_sarray)
 	((j_common_ptr) cinfo, JPOOL_IMAGE,
-	 compptr->width_in_blocks * DCTSIZE,
-	 (JDIMENSION) (compptr->v_samp_factor * DCTSIZE));
+	 compptr->width_in_blocks * compptr->DCT_h_scaled_size,
+	 (JDIMENSION) (compptr->v_samp_factor * compptr->DCT_v_scaled_size));
     }
   }
 }
diff --git a/src/jpeg-6b/jcmarker.c b/src/jpeg-8c/jcmarker.c
index 3d1e6c6d..606c19af 100644
--- a/src/jpeg-6b/jcmarker.c
+++ b/src/jpeg-8c/jcmarker.c
@@ -2,6 +2,7 @@
  * jcmarker.c
  *
  * Copyright (C) 1991-1998, Thomas G. Lane.
+ * Modified 2003-2010 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -153,21 +154,22 @@ emit_dqt (j_compress_ptr cinfo, int index)
     ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, index);
 
   prec = 0;
-  for (i = 0; i < DCTSIZE2; i++) {
-    if (qtbl->quantval[i] > 255)
+  for (i = 0; i <= cinfo->lim_Se; i++) {
+    if (qtbl->quantval[cinfo->natural_order[i]] > 255)
       prec = 1;
   }
 
   if (! qtbl->sent_table) {
     emit_marker(cinfo, M_DQT);
 
-    emit_2bytes(cinfo, prec ? DCTSIZE2*2 + 1 + 2 : DCTSIZE2 + 1 + 2);
+    emit_2bytes(cinfo,
+      prec ? cinfo->lim_Se * 2 + 2 + 1 + 2 : cinfo->lim_Se + 1 + 1 + 2);
 
     emit_byte(cinfo, index + (prec<<4));
 
-    for (i = 0; i < DCTSIZE2; i++) {
+    for (i = 0; i <= cinfo->lim_Se; i++) {
       /* The table entries must be emitted in zigzag order. */
-      unsigned int qval = qtbl->quantval[jpeg_natural_order[i]];
+      unsigned int qval = qtbl->quantval[cinfo->natural_order[i]];
       if (prec)
 	emit_byte(cinfo, (int) (qval >> 8));
       emit_byte(cinfo, (int) (qval & 0xFF));
@@ -229,32 +231,38 @@ emit_dac (j_compress_ptr cinfo)
   char ac_in_use[NUM_ARITH_TBLS];
   int length, i;
   jpeg_component_info *compptr;
-  
+
   for (i = 0; i < NUM_ARITH_TBLS; i++)
     dc_in_use[i] = ac_in_use[i] = 0;
-  
+
   for (i = 0; i < cinfo->comps_in_scan; i++) {
     compptr = cinfo->cur_comp_info[i];
-    dc_in_use[compptr->dc_tbl_no] = 1;
-    ac_in_use[compptr->ac_tbl_no] = 1;
+    /* DC needs no table for refinement scan */
+    if (cinfo->Ss == 0 && cinfo->Ah == 0)
+      dc_in_use[compptr->dc_tbl_no] = 1;
+    /* AC needs no table when not present */
+    if (cinfo->Se)
+      ac_in_use[compptr->ac_tbl_no] = 1;
   }
-  
+
   length = 0;
   for (i = 0; i < NUM_ARITH_TBLS; i++)
     length += dc_in_use[i] + ac_in_use[i];
-  
-  emit_marker(cinfo, M_DAC);
-  
-  emit_2bytes(cinfo, length*2 + 2);
-  
-  for (i = 0; i < NUM_ARITH_TBLS; i++) {
-    if (dc_in_use[i]) {
-      emit_byte(cinfo, i);
-      emit_byte(cinfo, cinfo->arith_dc_L[i] + (cinfo->arith_dc_U[i]<<4));
-    }
-    if (ac_in_use[i]) {
-      emit_byte(cinfo, i + 0x10);
-      emit_byte(cinfo, cinfo->arith_ac_K[i]);
+
+  if (length) {
+    emit_marker(cinfo, M_DAC);
+
+    emit_2bytes(cinfo, length*2 + 2);
+
+    for (i = 0; i < NUM_ARITH_TBLS; i++) {
+      if (dc_in_use[i]) {
+	emit_byte(cinfo, i);
+	emit_byte(cinfo, cinfo->arith_dc_L[i] + (cinfo->arith_dc_U[i]<<4));
+      }
+      if (ac_in_use[i]) {
+	emit_byte(cinfo, i + 0x10);
+	emit_byte(cinfo, cinfo->arith_ac_K[i]);
+      }
     }
   }
 #endif /* C_ARITH_CODING_SUPPORTED */
@@ -285,13 +293,13 @@ emit_sof (j_compress_ptr cinfo, JPEG_MARKER code)
   emit_2bytes(cinfo, 3 * cinfo->num_components + 2 + 5 + 1); /* length */
 
   /* Make sure image isn't bigger than SOF field can handle */
-  if ((long) cinfo->image_height > 65535L ||
-      (long) cinfo->image_width > 65535L)
+  if ((long) cinfo->jpeg_height > 65535L ||
+      (long) cinfo->jpeg_width > 65535L)
     ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int) 65535);
 
   emit_byte(cinfo, cinfo->data_precision);
-  emit_2bytes(cinfo, (int) cinfo->image_height);
-  emit_2bytes(cinfo, (int) cinfo->image_width);
+  emit_2bytes(cinfo, (int) cinfo->jpeg_height);
+  emit_2bytes(cinfo, (int) cinfo->jpeg_width);
 
   emit_byte(cinfo, cinfo->num_components);
 
@@ -320,22 +328,16 @@ emit_sos (j_compress_ptr cinfo)
   for (i = 0; i < cinfo->comps_in_scan; i++) {
     compptr = cinfo->cur_comp_info[i];
     emit_byte(cinfo, compptr->component_id);
-    td = compptr->dc_tbl_no;
-    ta = compptr->ac_tbl_no;
-    if (cinfo->progressive_mode) {
-      /* Progressive mode: only DC or only AC tables are used in one scan;
-       * furthermore, Huffman coding of DC refinement uses no table at all.
-       * We emit 0 for unused field(s); this is recommended by the P&M text
-       * but does not seem to be specified in the standard.
-       */
-      if (cinfo->Ss == 0) {
-	ta = 0;			/* DC scan */
-	if (cinfo->Ah != 0 && !cinfo->arith_code)
-	  td = 0;		/* no DC table either */
-      } else {
-	td = 0;			/* AC scan */
-      }
-    }
+
+    /* We emit 0 for unused field(s); this is recommended by the P&M text
+     * but does not seem to be specified in the standard.
+     */
+
+    /* DC needs no table for refinement scan */
+    td = cinfo->Ss == 0 && cinfo->Ah == 0 ? compptr->dc_tbl_no : 0;
+    /* AC needs no table when not present */
+    ta = cinfo->Se ? compptr->ac_tbl_no : 0;
+
     emit_byte(cinfo, (td << 4) + ta);
   }
 
@@ -346,6 +348,22 @@ emit_sos (j_compress_ptr cinfo)
 
 
 LOCAL(void)
+emit_pseudo_sos (j_compress_ptr cinfo)
+/* Emit a pseudo SOS marker */
+{
+  emit_marker(cinfo, M_SOS);
+  
+  emit_2bytes(cinfo, 2 + 1 + 3); /* length */
+  
+  emit_byte(cinfo, 0); /* Ns */
+
+  emit_byte(cinfo, 0); /* Ss */
+  emit_byte(cinfo, cinfo->block_size * cinfo->block_size - 1); /* Se */
+  emit_byte(cinfo, 0); /* Ah/Al */
+}
+
+
+LOCAL(void)
 emit_jfif_app0 (j_compress_ptr cinfo)
 /* Emit a JFIF-compliant APP0 marker */
 {
@@ -484,7 +502,7 @@ write_file_header (j_compress_ptr cinfo)
 
 /*
  * Write frame header.
- * This consists of DQT and SOFn markers.
+ * This consists of DQT and SOFn markers, and a conditional pseudo SOS marker.
  * Note that we do not emit the SOF until we have emitted the DQT(s).
  * This avoids compatibility problems with incorrect implementations that
  * try to error-check the quant table numbers as soon as they see the SOF.
@@ -511,7 +529,7 @@ write_frame_header (j_compress_ptr cinfo)
    * Note we assume that Huffman table numbers won't be changed later.
    */
   if (cinfo->arith_code || cinfo->progressive_mode ||
-      cinfo->data_precision != 8) {
+      cinfo->data_precision != 8 || cinfo->block_size != DCTSIZE) {
     is_baseline = FALSE;
   } else {
     is_baseline = TRUE;
@@ -529,7 +547,10 @@ write_frame_header (j_compress_ptr cinfo)
 
   /* Emit the proper SOF marker */
   if (cinfo->arith_code) {
-    emit_sof(cinfo, M_SOF9);	/* SOF code for arithmetic coding */
+    if (cinfo->progressive_mode)
+      emit_sof(cinfo, M_SOF10); /* SOF code for progressive arithmetic */
+    else
+      emit_sof(cinfo, M_SOF9);  /* SOF code for sequential arithmetic */
   } else {
     if (cinfo->progressive_mode)
       emit_sof(cinfo, M_SOF2);	/* SOF code for progressive Huffman */
@@ -538,6 +559,10 @@ write_frame_header (j_compress_ptr cinfo)
     else
       emit_sof(cinfo, M_SOF1);	/* SOF code for non-baseline Huffman file */
   }
+
+  /* Check to emit pseudo SOS marker */
+  if (cinfo->progressive_mode && cinfo->block_size != DCTSIZE)
+    emit_pseudo_sos(cinfo);
 }
 
 
@@ -566,19 +591,12 @@ write_scan_header (j_compress_ptr cinfo)
      */
     for (i = 0; i < cinfo->comps_in_scan; i++) {
       compptr = cinfo->cur_comp_info[i];
-      if (cinfo->progressive_mode) {
-	/* Progressive mode: only DC or only AC tables are used in one scan */
-	if (cinfo->Ss == 0) {
-	  if (cinfo->Ah == 0)	/* DC needs no table for refinement scan */
-	    emit_dht(cinfo, compptr->dc_tbl_no, FALSE);
-	} else {
-	  emit_dht(cinfo, compptr->ac_tbl_no, TRUE);
-	}
-      } else {
-	/* Sequential mode: need both DC and AC tables */
+      /* DC needs no table for refinement scan */
+      if (cinfo->Ss == 0 && cinfo->Ah == 0)
 	emit_dht(cinfo, compptr->dc_tbl_no, FALSE);
+      /* AC needs no table when not present */
+      if (cinfo->Se)
 	emit_dht(cinfo, compptr->ac_tbl_no, TRUE);
-      }
     }
   }
 
diff --git a/src/jpeg-6b/jcmaster.c b/src/jpeg-8c/jcmaster.c
index aab4020b..caf80a53 100644
--- a/src/jpeg-6b/jcmaster.c
+++ b/src/jpeg-8c/jcmaster.c
@@ -2,6 +2,7 @@
  * jcmaster.c
  *
  * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Modified 2003-2011 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -42,23 +43,220 @@ typedef my_comp_master * my_master_ptr;
  * Support routines that do various essential calculations.
  */
 
+/*
+ * Compute JPEG image dimensions and related values.
+ * NOTE: this is exported for possible use by application.
+ * Hence it mustn't do anything that can't be done twice.
+ */
+
+GLOBAL(void)
+jpeg_calc_jpeg_dimensions (j_compress_ptr cinfo)
+/* Do computations that are needed before master selection phase */
+{
+#ifdef DCT_SCALING_SUPPORTED
+
+  /* Sanity check on input image dimensions to prevent overflow in
+   * following calculation.
+   * We do check jpeg_width and jpeg_height in initial_setup below,
+   * but image_width and image_height can come from arbitrary data,
+   * and we need some space for multiplication by block_size.
+   */
+  if (((long) cinfo->image_width >> 24) || ((long) cinfo->image_height >> 24))
+    ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int) JPEG_MAX_DIMENSION);
+
+  /* Compute actual JPEG image dimensions and DCT scaling choices. */
+  if (cinfo->scale_num >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/1 scaling */
+    cinfo->jpeg_width = cinfo->image_width * cinfo->block_size;
+    cinfo->jpeg_height = cinfo->image_height * cinfo->block_size;
+    cinfo->min_DCT_h_scaled_size = 1;
+    cinfo->min_DCT_v_scaled_size = 1;
+  } else if (cinfo->scale_num * 2 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/2 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 2L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 2L);
+    cinfo->min_DCT_h_scaled_size = 2;
+    cinfo->min_DCT_v_scaled_size = 2;
+  } else if (cinfo->scale_num * 3 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/3 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 3L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 3L);
+    cinfo->min_DCT_h_scaled_size = 3;
+    cinfo->min_DCT_v_scaled_size = 3;
+  } else if (cinfo->scale_num * 4 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/4 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 4L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 4L);
+    cinfo->min_DCT_h_scaled_size = 4;
+    cinfo->min_DCT_v_scaled_size = 4;
+  } else if (cinfo->scale_num * 5 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/5 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 5L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 5L);
+    cinfo->min_DCT_h_scaled_size = 5;
+    cinfo->min_DCT_v_scaled_size = 5;
+  } else if (cinfo->scale_num * 6 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/6 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 6L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 6L);
+    cinfo->min_DCT_h_scaled_size = 6;
+    cinfo->min_DCT_v_scaled_size = 6;
+  } else if (cinfo->scale_num * 7 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/7 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 7L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 7L);
+    cinfo->min_DCT_h_scaled_size = 7;
+    cinfo->min_DCT_v_scaled_size = 7;
+  } else if (cinfo->scale_num * 8 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/8 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 8L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 8L);
+    cinfo->min_DCT_h_scaled_size = 8;
+    cinfo->min_DCT_v_scaled_size = 8;
+  } else if (cinfo->scale_num * 9 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/9 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 9L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 9L);
+    cinfo->min_DCT_h_scaled_size = 9;
+    cinfo->min_DCT_v_scaled_size = 9;
+  } else if (cinfo->scale_num * 10 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/10 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 10L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 10L);
+    cinfo->min_DCT_h_scaled_size = 10;
+    cinfo->min_DCT_v_scaled_size = 10;
+  } else if (cinfo->scale_num * 11 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/11 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 11L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 11L);
+    cinfo->min_DCT_h_scaled_size = 11;
+    cinfo->min_DCT_v_scaled_size = 11;
+  } else if (cinfo->scale_num * 12 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/12 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 12L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 12L);
+    cinfo->min_DCT_h_scaled_size = 12;
+    cinfo->min_DCT_v_scaled_size = 12;
+  } else if (cinfo->scale_num * 13 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/13 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 13L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 13L);
+    cinfo->min_DCT_h_scaled_size = 13;
+    cinfo->min_DCT_v_scaled_size = 13;
+  } else if (cinfo->scale_num * 14 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/14 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 14L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 14L);
+    cinfo->min_DCT_h_scaled_size = 14;
+    cinfo->min_DCT_v_scaled_size = 14;
+  } else if (cinfo->scale_num * 15 >= cinfo->scale_denom * cinfo->block_size) {
+    /* Provide block_size/15 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 15L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 15L);
+    cinfo->min_DCT_h_scaled_size = 15;
+    cinfo->min_DCT_v_scaled_size = 15;
+  } else {
+    /* Provide block_size/16 scaling */
+    cinfo->jpeg_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * cinfo->block_size, 16L);
+    cinfo->jpeg_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * cinfo->block_size, 16L);
+    cinfo->min_DCT_h_scaled_size = 16;
+    cinfo->min_DCT_v_scaled_size = 16;
+  }
+
+#else /* !DCT_SCALING_SUPPORTED */
+
+  /* Hardwire it to "no scaling" */
+  cinfo->jpeg_width = cinfo->image_width;
+  cinfo->jpeg_height = cinfo->image_height;
+  cinfo->min_DCT_h_scaled_size = DCTSIZE;
+  cinfo->min_DCT_v_scaled_size = DCTSIZE;
+
+#endif /* DCT_SCALING_SUPPORTED */
+}
+
+
+LOCAL(void)
+jpeg_calc_trans_dimensions (j_compress_ptr cinfo)
+{
+  if (cinfo->min_DCT_h_scaled_size != cinfo->min_DCT_v_scaled_size)
+    ERREXIT2(cinfo, JERR_BAD_DCTSIZE,
+	     cinfo->min_DCT_h_scaled_size, cinfo->min_DCT_v_scaled_size);
+
+  cinfo->block_size = cinfo->min_DCT_h_scaled_size;
+}
+
+
 LOCAL(void)
-initial_setup (j_compress_ptr cinfo)
+initial_setup (j_compress_ptr cinfo, boolean transcode_only)
 /* Do computations that are needed before master selection phase */
 {
-  int ci;
+  int ci, ssize;
   jpeg_component_info *compptr;
   long samplesperrow;
   JDIMENSION jd_samplesperrow;
 
+  if (transcode_only)
+    jpeg_calc_trans_dimensions(cinfo);
+  else
+    jpeg_calc_jpeg_dimensions(cinfo);
+
+  /* Sanity check on block_size */
+  if (cinfo->block_size < 1 || cinfo->block_size > 16)
+    ERREXIT2(cinfo, JERR_BAD_DCTSIZE, cinfo->block_size, cinfo->block_size);
+
+  /* Derive natural_order from block_size */
+  switch (cinfo->block_size) {
+  case 2: cinfo->natural_order = jpeg_natural_order2; break;
+  case 3: cinfo->natural_order = jpeg_natural_order3; break;
+  case 4: cinfo->natural_order = jpeg_natural_order4; break;
+  case 5: cinfo->natural_order = jpeg_natural_order5; break;
+  case 6: cinfo->natural_order = jpeg_natural_order6; break;
+  case 7: cinfo->natural_order = jpeg_natural_order7; break;
+  default: cinfo->natural_order = jpeg_natural_order; break;
+  }
+
+  /* Derive lim_Se from block_size */
+  cinfo->lim_Se = cinfo->block_size < DCTSIZE ?
+    cinfo->block_size * cinfo->block_size - 1 : DCTSIZE2-1;
+
   /* Sanity check on image dimensions */
-  if (cinfo->image_height <= 0 || cinfo->image_width <= 0
-      || cinfo->num_components <= 0 || cinfo->input_components <= 0)
+  if (cinfo->jpeg_height <= 0 || cinfo->jpeg_width <= 0 ||
+      cinfo->num_components <= 0 || cinfo->input_components <= 0)
     ERREXIT(cinfo, JERR_EMPTY_IMAGE);
 
   /* Make sure image isn't bigger than I can handle */
-  if ((long) cinfo->image_height > (long) JPEG_MAX_DIMENSION ||
-      (long) cinfo->image_width > (long) JPEG_MAX_DIMENSION)
+  if ((long) cinfo->jpeg_height > (long) JPEG_MAX_DIMENSION ||
+      (long) cinfo->jpeg_width > (long) JPEG_MAX_DIMENSION)
     ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int) JPEG_MAX_DIMENSION);
 
   /* Width of an input scanline must be representable as JDIMENSION. */
@@ -95,22 +293,52 @@ initial_setup (j_compress_ptr cinfo)
        ci++, compptr++) {
     /* Fill in the correct component_index value; don't rely on application */
     compptr->component_index = ci;
-    /* For compression, we never do DCT scaling. */
-    compptr->DCT_scaled_size = DCTSIZE;
+    /* In selecting the actual DCT scaling for each component, we try to
+     * scale down the chroma components via DCT scaling rather than downsampling.
+     * This saves time if the downsampler gets to use 1:1 scaling.
+     * Note this code adapts subsampling ratios which are powers of 2.
+     */
+    ssize = 1;
+#ifdef DCT_SCALING_SUPPORTED
+    while (cinfo->min_DCT_h_scaled_size * ssize <=
+	   (cinfo->do_fancy_downsampling ? DCTSIZE : DCTSIZE / 2) &&
+	   (cinfo->max_h_samp_factor % (compptr->h_samp_factor * ssize * 2)) == 0) {
+      ssize = ssize * 2;
+    }
+#endif
+    compptr->DCT_h_scaled_size = cinfo->min_DCT_h_scaled_size * ssize;
+    ssize = 1;
+#ifdef DCT_SCALING_SUPPORTED
+    while (cinfo->min_DCT_v_scaled_size * ssize <=
+	   (cinfo->do_fancy_downsampling ? DCTSIZE : DCTSIZE / 2) &&
+	   (cinfo->max_v_samp_factor % (compptr->v_samp_factor * ssize * 2)) == 0) {
+      ssize = ssize * 2;
+    }
+#endif
+    compptr->DCT_v_scaled_size = cinfo->min_DCT_v_scaled_size * ssize;
+
+    /* We don't support DCT ratios larger than 2. */
+    if (compptr->DCT_h_scaled_size > compptr->DCT_v_scaled_size * 2)
+	compptr->DCT_h_scaled_size = compptr->DCT_v_scaled_size * 2;
+    else if (compptr->DCT_v_scaled_size > compptr->DCT_h_scaled_size * 2)
+	compptr->DCT_v_scaled_size = compptr->DCT_h_scaled_size * 2;
+
     /* Size in DCT blocks */
     compptr->width_in_blocks = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * (long) compptr->h_samp_factor,
-		    (long) (cinfo->max_h_samp_factor * DCTSIZE));
+      jdiv_round_up((long) cinfo->jpeg_width * (long) compptr->h_samp_factor,
+		    (long) (cinfo->max_h_samp_factor * cinfo->block_size));
     compptr->height_in_blocks = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * (long) compptr->v_samp_factor,
-		    (long) (cinfo->max_v_samp_factor * DCTSIZE));
+      jdiv_round_up((long) cinfo->jpeg_height * (long) compptr->v_samp_factor,
+		    (long) (cinfo->max_v_samp_factor * cinfo->block_size));
     /* Size in samples */
     compptr->downsampled_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width * (long) compptr->h_samp_factor,
-		    (long) cinfo->max_h_samp_factor);
+      jdiv_round_up((long) cinfo->jpeg_width *
+		    (long) (compptr->h_samp_factor * compptr->DCT_h_scaled_size),
+		    (long) (cinfo->max_h_samp_factor * cinfo->block_size));
     compptr->downsampled_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height * (long) compptr->v_samp_factor,
-		    (long) cinfo->max_v_samp_factor);
+      jdiv_round_up((long) cinfo->jpeg_height *
+		    (long) (compptr->v_samp_factor * compptr->DCT_v_scaled_size),
+		    (long) (cinfo->max_v_samp_factor * cinfo->block_size));
     /* Mark component needed (this flag isn't actually used for compression) */
     compptr->component_needed = TRUE;
   }
@@ -119,8 +347,8 @@ initial_setup (j_compress_ptr cinfo)
    * main controller will call coefficient controller).
    */
   cinfo->total_iMCU_rows = (JDIMENSION)
-    jdiv_round_up((long) cinfo->image_height,
-		  (long) (cinfo->max_v_samp_factor*DCTSIZE));
+    jdiv_round_up((long) cinfo->jpeg_height,
+		  (long) (cinfo->max_v_samp_factor * cinfo->block_size));
 }
 
 
@@ -260,6 +488,39 @@ validate_script (j_compress_ptr cinfo)
   }
 }
 
+
+LOCAL(void)
+reduce_script (j_compress_ptr cinfo)
+/* Adapt scan script for use with reduced block size;
+ * assume that script has been validated before.
+ */
+{
+  jpeg_scan_info * scanptr;
+  int idxout, idxin;
+
+  /* Circumvent const declaration for this function */
+  scanptr = (jpeg_scan_info *) cinfo->scan_info;
+  idxout = 0;
+
+  for (idxin = 0; idxin < cinfo->num_scans; idxin++) {
+    /* After skipping, idxout becomes smaller than idxin */
+    if (idxin != idxout)
+      /* Copy rest of data;
+       * note we stay in given chunk of allocated memory.
+       */
+      scanptr[idxout] = scanptr[idxin];
+    if (scanptr[idxout].Ss > cinfo->lim_Se)
+      /* Entire scan out of range - skip this entry */
+      continue;
+    if (scanptr[idxout].Se > cinfo->lim_Se)
+      /* Limit scan to end of block */
+      scanptr[idxout].Se = cinfo->lim_Se;
+    idxout++;
+  }
+
+  cinfo->num_scans = idxout;
+}
+
 #endif /* C_MULTISCAN_FILES_SUPPORTED */
 
 
@@ -280,10 +541,13 @@ select_scan_parameters (j_compress_ptr cinfo)
       cinfo->cur_comp_info[ci] =
 	&cinfo->comp_info[scanptr->component_index[ci]];
     }
-    cinfo->Ss = scanptr->Ss;
-    cinfo->Se = scanptr->Se;
-    cinfo->Ah = scanptr->Ah;
-    cinfo->Al = scanptr->Al;
+    if (cinfo->progressive_mode) {
+      cinfo->Ss = scanptr->Ss;
+      cinfo->Se = scanptr->Se;
+      cinfo->Ah = scanptr->Ah;
+      cinfo->Al = scanptr->Al;
+      return;
+    }
   }
   else
 #endif
@@ -296,11 +560,11 @@ select_scan_parameters (j_compress_ptr cinfo)
     for (ci = 0; ci < cinfo->num_components; ci++) {
       cinfo->cur_comp_info[ci] = &cinfo->comp_info[ci];
     }
-    cinfo->Ss = 0;
-    cinfo->Se = DCTSIZE2-1;
-    cinfo->Ah = 0;
-    cinfo->Al = 0;
   }
+  cinfo->Ss = 0;
+  cinfo->Se = cinfo->block_size * cinfo->block_size - 1;
+  cinfo->Ah = 0;
+  cinfo->Al = 0;
 }
 
 
@@ -325,7 +589,7 @@ per_scan_setup (j_compress_ptr cinfo)
     compptr->MCU_width = 1;
     compptr->MCU_height = 1;
     compptr->MCU_blocks = 1;
-    compptr->MCU_sample_width = DCTSIZE;
+    compptr->MCU_sample_width = compptr->DCT_h_scaled_size;
     compptr->last_col_width = 1;
     /* For noninterleaved scans, it is convenient to define last_row_height
      * as the number of block rows present in the last iMCU row.
@@ -347,11 +611,11 @@ per_scan_setup (j_compress_ptr cinfo)
     
     /* Overall image size in MCUs */
     cinfo->MCUs_per_row = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width,
-		    (long) (cinfo->max_h_samp_factor*DCTSIZE));
+      jdiv_round_up((long) cinfo->jpeg_width,
+		    (long) (cinfo->max_h_samp_factor * cinfo->block_size));
     cinfo->MCU_rows_in_scan = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height,
-		    (long) (cinfo->max_v_samp_factor*DCTSIZE));
+      jdiv_round_up((long) cinfo->jpeg_height,
+		    (long) (cinfo->max_v_samp_factor * cinfo->block_size));
     
     cinfo->blocks_in_MCU = 0;
     
@@ -361,7 +625,7 @@ per_scan_setup (j_compress_ptr cinfo)
       compptr->MCU_width = compptr->h_samp_factor;
       compptr->MCU_height = compptr->v_samp_factor;
       compptr->MCU_blocks = compptr->MCU_width * compptr->MCU_height;
-      compptr->MCU_sample_width = compptr->MCU_width * DCTSIZE;
+      compptr->MCU_sample_width = compptr->MCU_width * compptr->DCT_h_scaled_size;
       /* Figure number of non-dummy blocks in last MCU column & row */
       tmp = (int) (compptr->width_in_blocks % compptr->MCU_width);
       if (tmp == 0) tmp = compptr->MCU_width;
@@ -433,7 +697,7 @@ prepare_for_pass (j_compress_ptr cinfo)
     /* Do Huffman optimization for a scan after the first one. */
     select_scan_parameters(cinfo);
     per_scan_setup(cinfo);
-    if (cinfo->Ss != 0 || cinfo->Ah == 0 || cinfo->arith_code) {
+    if (cinfo->Ss != 0 || cinfo->Ah == 0) {
       (*cinfo->entropy->start_pass) (cinfo, TRUE);
       (*cinfo->coef->start_pass) (cinfo, JBUF_CRANK_DEST);
       master->pub.call_pass_startup = FALSE;
@@ -554,11 +818,13 @@ jinit_c_master_control (j_compress_ptr cinfo, boolean transcode_only)
   master->pub.is_last_pass = FALSE;
 
   /* Validate parameters, determine derived values */
-  initial_setup(cinfo);
+  initial_setup(cinfo, transcode_only);
 
   if (cinfo->scan_info != NULL) {
 #ifdef C_MULTISCAN_FILES_SUPPORTED
     validate_script(cinfo);
+    if (cinfo->block_size < DCTSIZE)
+      reduce_script(cinfo);
 #else
     ERREXIT(cinfo, JERR_NOT_COMPILED);
 #endif
@@ -567,8 +833,10 @@ jinit_c_master_control (j_compress_ptr cinfo, boolean transcode_only)
     cinfo->num_scans = 1;
   }
 
-  if (cinfo->progressive_mode)	/*  TEMPORARY HACK ??? */
-    cinfo->optimize_coding = TRUE; /* assume default tables no good for progressive mode */
+  if ((cinfo->progressive_mode || cinfo->block_size < DCTSIZE) &&
+      !cinfo->arith_code)			/*  TEMPORARY HACK ??? */
+    /* assume default tables no good for progressive or downscale mode */
+    cinfo->optimize_coding = TRUE;
 
   /* Initialize my private state */
   if (transcode_only) {
diff --git a/src/jpeg-6b/jcomapi.c b/src/jpeg-8c/jcomapi.c
index 9b1fa756..9b1fa756 100644
--- a/src/jpeg-6b/jcomapi.c
+++ b/src/jpeg-8c/jcomapi.c
diff --git a/src/jpeg-8c/jconfig.h b/src/jpeg-8c/jconfig.h
new file mode 100644
index 00000000..ee83f412
--- /dev/null
+++ b/src/jpeg-8c/jconfig.h
@@ -0,0 +1,56 @@
+/* jconfig.h.  Generated from jconfig.cfg by configure.  */
+/* jconfig.cfg --- source file edited by configure script */
+/* see jconfig.txt for explanations */
+
+/* Define this if you get warnings about undefined structures. */
+/* #undef INCOMPLETE_TYPES_BROKEN */
+
+#define HAVE_PROTOTYPES 1
+#define HAVE_UNSIGNED_CHAR 1
+#define HAVE_UNSIGNED_SHORT 1
+/* #undef void */
+/* #undef const */
+/* #undef CHAR_IS_UNSIGNED */
+
+#ifdef JPEG_INTERNALS
+
+#define HAVE_STDDEF_H 1
+#define HAVE_STDLIB_H 1
+#define HAVE_LOCALE_H 1
+/* #undef NEED_BSD_STRINGS */
+/* #undef NEED_SYS_TYPES_H */
+/* #undef NEED_FAR_POINTERS */
+/* #undef NEED_SHORT_EXTERNAL_NAMES */
+
+/* Define "boolean" as unsigned char, not int, on Windows systems. */
+#ifdef _WIN32
+#ifndef __RPCNDR_H__		/* don't conflict if rpcndr.h already read */
+typedef unsigned char boolean;
+#endif
+#define HAVE_BOOLEAN		/* prevent jmorecfg.h from redefining it */
+#endif
+
+/* #undef RIGHT_SHIFT_IS_UNSIGNED */
+#define INLINE __inline__
+/* These are for configuring the JPEG memory manager. */
+/* #undef DEFAULT_MAX_MEM */
+/* #undef NO_MKTEMP */
+
+#endif /* JPEG_INTERNALS */
+
+#ifdef JPEG_CJPEG_DJPEG
+
+#define BMP_SUPPORTED		/* BMP image file format */
+#define GIF_SUPPORTED		/* GIF image file format */
+#define PPM_SUPPORTED		/* PBMPLUS PPM/PGM image file format */
+/* #undef RLE_SUPPORTED */
+#define TARGA_SUPPORTED		/* Targa image file format */
+
+/* #undef TWO_FILE_COMMANDLINE */
+/* #undef NEED_SIGNAL_CATCHER */
+/* #undef DONT_USE_B_MODE */
+
+/* Define this if you want percent-done progress reports from cjpeg/djpeg. */
+/* #undef PROGRESS_REPORT */
+
+#endif /* JPEG_CJPEG_DJPEG */
diff --git a/src/jpeg-6b/jcparam.c b/src/jpeg-8c/jcparam.c
index 6fc48f53..c5e85dda 100644
--- a/src/jpeg-6b/jcparam.c
+++ b/src/jpeg-8c/jcparam.c
@@ -2,6 +2,7 @@
  * jcparam.c
  *
  * Copyright (C) 1991-1998, Thomas G. Lane.
+ * Modified 2003-2008 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -60,6 +61,47 @@ jpeg_add_quant_table (j_compress_ptr cinfo, int which_tbl,
 }
 
 
+/* These are the sample quantization tables given in JPEG spec section K.1.
+ * The spec says that the values given produce "good" quality, and
+ * when divided by 2, "very good" quality.
+ */
+static const unsigned int std_luminance_quant_tbl[DCTSIZE2] = {
+  16,  11,  10,  16,  24,  40,  51,  61,
+  12,  12,  14,  19,  26,  58,  60,  55,
+  14,  13,  16,  24,  40,  57,  69,  56,
+  14,  17,  22,  29,  51,  87,  80,  62,
+  18,  22,  37,  56,  68, 109, 103,  77,
+  24,  35,  55,  64,  81, 104, 113,  92,
+  49,  64,  78,  87, 103, 121, 120, 101,
+  72,  92,  95,  98, 112, 100, 103,  99
+};
+static const unsigned int std_chrominance_quant_tbl[DCTSIZE2] = {
+  17,  18,  24,  47,  99,  99,  99,  99,
+  18,  21,  26,  66,  99,  99,  99,  99,
+  24,  26,  56,  99,  99,  99,  99,  99,
+  47,  66,  99,  99,  99,  99,  99,  99,
+  99,  99,  99,  99,  99,  99,  99,  99,
+  99,  99,  99,  99,  99,  99,  99,  99,
+  99,  99,  99,  99,  99,  99,  99,  99,
+  99,  99,  99,  99,  99,  99,  99,  99
+};
+
+
+GLOBAL(void)
+jpeg_default_qtables (j_compress_ptr cinfo, boolean force_baseline)
+/* Set or change the 'quality' (quantization) setting, using default tables
+ * and straight percentage-scaling quality scales.
+ * This entry point allows different scalings for luminance and chrominance.
+ */
+{
+  /* Set up two quantization tables using the specified scaling */
+  jpeg_add_quant_table(cinfo, 0, std_luminance_quant_tbl,
+		       cinfo->q_scale_factor[0], force_baseline);
+  jpeg_add_quant_table(cinfo, 1, std_chrominance_quant_tbl,
+		       cinfo->q_scale_factor[1], force_baseline);
+}
+
+
 GLOBAL(void)
 jpeg_set_linear_quality (j_compress_ptr cinfo, int scale_factor,
 			 boolean force_baseline)
@@ -69,31 +111,6 @@ jpeg_set_linear_quality (j_compress_ptr cinfo, int scale_factor,
  * applications that insist on a linear percentage scaling.
  */
 {
-  /* These are the sample quantization tables given in JPEG spec section K.1.
-   * The spec says that the values given produce "good" quality, and
-   * when divided by 2, "very good" quality.
-   */
-  static const unsigned int std_luminance_quant_tbl[DCTSIZE2] = {
-    16,  11,  10,  16,  24,  40,  51,  61,
-    12,  12,  14,  19,  26,  58,  60,  55,
-    14,  13,  16,  24,  40,  57,  69,  56,
-    14,  17,  22,  29,  51,  87,  80,  62,
-    18,  22,  37,  56,  68, 109, 103,  77,
-    24,  35,  55,  64,  81, 104, 113,  92,
-    49,  64,  78,  87, 103, 121, 120, 101,
-    72,  92,  95,  98, 112, 100, 103,  99
-  };
-  static const unsigned int std_chrominance_quant_tbl[DCTSIZE2] = {
-    17,  18,  24,  47,  99,  99,  99,  99,
-    18,  21,  26,  66,  99,  99,  99,  99,
-    24,  26,  56,  99,  99,  99,  99,  99,
-    47,  66,  99,  99,  99,  99,  99,  99,
-    99,  99,  99,  99,  99,  99,  99,  99,
-    99,  99,  99,  99,  99,  99,  99,  99,
-    99,  99,  99,  99,  99,  99,  99,  99,
-    99,  99,  99,  99,  99,  99,  99,  99
-  };
-
   /* Set up two quantization tables using the specified scaling */
   jpeg_add_quant_table(cinfo, 0, std_luminance_quant_tbl,
 		       scale_factor, force_baseline);
@@ -284,6 +301,8 @@ jpeg_set_defaults (j_compress_ptr cinfo)
 
   /* Initialize everything not dependent on the color space */
 
+  cinfo->scale_num = 1;		/* 1:1 scaling */
+  cinfo->scale_denom = 1;
   cinfo->data_precision = BITS_IN_JSAMPLE;
   /* Set up two quantization tables using default quality of 75 */
   jpeg_set_quality(cinfo, 75, TRUE);
@@ -320,6 +339,9 @@ jpeg_set_defaults (j_compress_ptr cinfo)
   /* By default, use the simpler non-cosited sampling alignment */
   cinfo->CCIR601_sampling = FALSE;
 
+  /* By default, apply fancy downsampling */
+  cinfo->do_fancy_downsampling = TRUE;
+
   /* No input smoothing */
   cinfo->smoothing_factor = 0;
 
diff --git a/src/jpeg-6b/jcprepct.c b/src/jpeg-8c/jcprepct.c
index fa93333d..be44cc4b 100644
--- a/src/jpeg-6b/jcprepct.c
+++ b/src/jpeg-8c/jcprepct.c
@@ -173,10 +173,12 @@ pre_process_data (j_compress_ptr cinfo,
 	*out_row_group_ctr < out_row_groups_avail) {
       for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
 	   ci++, compptr++) {
+	numrows = (compptr->v_samp_factor * compptr->DCT_v_scaled_size) /
+		  cinfo->min_DCT_v_scaled_size;
 	expand_bottom_edge(output_buf[ci],
-			   compptr->width_in_blocks * DCTSIZE,
-			   (int) (*out_row_group_ctr * compptr->v_samp_factor),
-			   (int) (out_row_groups_avail * compptr->v_samp_factor));
+			   compptr->width_in_blocks * compptr->DCT_h_scaled_size,
+			   (int) (*out_row_group_ctr * numrows),
+			   (int) (out_row_groups_avail * numrows));
       }
       *out_row_group_ctr = out_row_groups_avail;
       break;			/* can exit outer loop without test */
@@ -288,7 +290,8 @@ create_context_buffer (j_compress_ptr cinfo)
      */
     true_buffer = (*cinfo->mem->alloc_sarray)
       ((j_common_ptr) cinfo, JPOOL_IMAGE,
-       (JDIMENSION) (((long) compptr->width_in_blocks * DCTSIZE *
+       (JDIMENSION) (((long) compptr->width_in_blocks *
+		      cinfo->min_DCT_h_scaled_size *
 		      cinfo->max_h_samp_factor) / compptr->h_samp_factor),
        (JDIMENSION) (3 * rgroup_height));
     /* Copy true buffer row pointers into the middle of the fake row array */
@@ -346,7 +349,8 @@ jinit_c_prep_controller (j_compress_ptr cinfo, boolean need_full_buffer)
 	 ci++, compptr++) {
       prep->color_buf[ci] = (*cinfo->mem->alloc_sarray)
 	((j_common_ptr) cinfo, JPOOL_IMAGE,
-	 (JDIMENSION) (((long) compptr->width_in_blocks * DCTSIZE *
+	 (JDIMENSION) (((long) compptr->width_in_blocks *
+			cinfo->min_DCT_h_scaled_size *
 			cinfo->max_h_samp_factor) / compptr->h_samp_factor),
 	 (JDIMENSION) cinfo->max_v_samp_factor);
     }
diff --git a/src/jpeg-6b/jcsample.c b/src/jpeg-8c/jcsample.c
index 212ec875..4d36f85f 100644
--- a/src/jpeg-6b/jcsample.c
+++ b/src/jpeg-8c/jcsample.c
@@ -62,6 +62,15 @@ typedef struct {
 
   /* Downsampling method pointers, one per component */
   downsample1_ptr methods[MAX_COMPONENTS];
+
+  /* Height of an output row group for each component. */
+  int rowgroup_height[MAX_COMPONENTS];
+
+  /* These arrays save pixel expansion factors so that int_downsample need not
+   * recompute them each time.  They are unused for other downsampling methods.
+   */
+  UINT8 h_expand[MAX_COMPONENTS];
+  UINT8 v_expand[MAX_COMPONENTS];
 } my_downsampler;
 
 typedef my_downsampler * my_downsample_ptr;
@@ -123,7 +132,8 @@ sep_downsample (j_compress_ptr cinfo,
   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
        ci++, compptr++) {
     in_ptr = input_buf[ci] + in_row_index;
-    out_ptr = output_buf[ci] + (out_row_group_index * compptr->v_samp_factor);
+    out_ptr = output_buf[ci] +
+	      (out_row_group_index * downsample->rowgroup_height[ci]);
     (*downsample->methods[ci]) (cinfo, compptr, in_ptr, out_ptr);
   }
 }
@@ -140,14 +150,15 @@ METHODDEF(void)
 int_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
 		JSAMPARRAY input_data, JSAMPARRAY output_data)
 {
+  my_downsample_ptr downsample = (my_downsample_ptr) cinfo->downsample;
   int inrow, outrow, h_expand, v_expand, numpix, numpix2, h, v;
   JDIMENSION outcol, outcol_h;	/* outcol_h == outcol*h_expand */
-  JDIMENSION output_cols = compptr->width_in_blocks * DCTSIZE;
+  JDIMENSION output_cols = compptr->width_in_blocks * compptr->DCT_h_scaled_size;
   JSAMPROW inptr, outptr;
   INT32 outvalue;
 
-  h_expand = cinfo->max_h_samp_factor / compptr->h_samp_factor;
-  v_expand = cinfo->max_v_samp_factor / compptr->v_samp_factor;
+  h_expand = downsample->h_expand[compptr->component_index];
+  v_expand = downsample->v_expand[compptr->component_index];
   numpix = h_expand * v_expand;
   numpix2 = numpix/2;
 
@@ -158,8 +169,8 @@ int_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
   expand_right_edge(input_data, cinfo->max_v_samp_factor,
 		    cinfo->image_width, output_cols * h_expand);
 
-  inrow = 0;
-  for (outrow = 0; outrow < compptr->v_samp_factor; outrow++) {
+  inrow = outrow = 0;
+  while (inrow < cinfo->max_v_samp_factor) {
     outptr = output_data[outrow];
     for (outcol = 0, outcol_h = 0; outcol < output_cols;
 	 outcol++, outcol_h += h_expand) {
@@ -173,6 +184,7 @@ int_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
       *outptr++ = (JSAMPLE) ((outvalue + numpix2) / numpix);
     }
     inrow += v_expand;
+    outrow++;
   }
 }
 
@@ -191,8 +203,8 @@ fullsize_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
   jcopy_sample_rows(input_data, 0, output_data, 0,
 		    cinfo->max_v_samp_factor, cinfo->image_width);
   /* Edge-expand */
-  expand_right_edge(output_data, cinfo->max_v_samp_factor,
-		    cinfo->image_width, compptr->width_in_blocks * DCTSIZE);
+  expand_right_edge(output_data, cinfo->max_v_samp_factor, cinfo->image_width,
+		    compptr->width_in_blocks * compptr->DCT_h_scaled_size);
 }
 
 
@@ -212,9 +224,9 @@ METHODDEF(void)
 h2v1_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
 		 JSAMPARRAY input_data, JSAMPARRAY output_data)
 {
-  int outrow;
+  int inrow;
   JDIMENSION outcol;
-  JDIMENSION output_cols = compptr->width_in_blocks * DCTSIZE;
+  JDIMENSION output_cols = compptr->width_in_blocks * compptr->DCT_h_scaled_size;
   register JSAMPROW inptr, outptr;
   register int bias;
 
@@ -225,9 +237,9 @@ h2v1_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
   expand_right_edge(input_data, cinfo->max_v_samp_factor,
 		    cinfo->image_width, output_cols * 2);
 
-  for (outrow = 0; outrow < compptr->v_samp_factor; outrow++) {
-    outptr = output_data[outrow];
-    inptr = input_data[outrow];
+  for (inrow = 0; inrow < cinfo->max_v_samp_factor; inrow++) {
+    outptr = output_data[inrow];
+    inptr = input_data[inrow];
     bias = 0;			/* bias = 0,1,0,1,... for successive samples */
     for (outcol = 0; outcol < output_cols; outcol++) {
       *outptr++ = (JSAMPLE) ((GETJSAMPLE(*inptr) + GETJSAMPLE(inptr[1])
@@ -251,7 +263,7 @@ h2v2_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
 {
   int inrow, outrow;
   JDIMENSION outcol;
-  JDIMENSION output_cols = compptr->width_in_blocks * DCTSIZE;
+  JDIMENSION output_cols = compptr->width_in_blocks * compptr->DCT_h_scaled_size;
   register JSAMPROW inptr0, inptr1, outptr;
   register int bias;
 
@@ -262,8 +274,8 @@ h2v2_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
   expand_right_edge(input_data, cinfo->max_v_samp_factor,
 		    cinfo->image_width, output_cols * 2);
 
-  inrow = 0;
-  for (outrow = 0; outrow < compptr->v_samp_factor; outrow++) {
+  inrow = outrow = 0;
+  while (inrow < cinfo->max_v_samp_factor) {
     outptr = output_data[outrow];
     inptr0 = input_data[inrow];
     inptr1 = input_data[inrow+1];
@@ -276,6 +288,7 @@ h2v2_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
       inptr0 += 2; inptr1 += 2;
     }
     inrow += 2;
+    outrow++;
   }
 }
 
@@ -294,7 +307,7 @@ h2v2_smooth_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
 {
   int inrow, outrow;
   JDIMENSION colctr;
-  JDIMENSION output_cols = compptr->width_in_blocks * DCTSIZE;
+  JDIMENSION output_cols = compptr->width_in_blocks * compptr->DCT_h_scaled_size;
   register JSAMPROW inptr0, inptr1, above_ptr, below_ptr, outptr;
   INT32 membersum, neighsum, memberscale, neighscale;
 
@@ -321,8 +334,8 @@ h2v2_smooth_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
   memberscale = 16384 - cinfo->smoothing_factor * 80; /* scaled (1-5*SF)/4 */
   neighscale = cinfo->smoothing_factor * 16; /* scaled SF/4 */
 
-  inrow = 0;
-  for (outrow = 0; outrow < compptr->v_samp_factor; outrow++) {
+  inrow = outrow = 0;
+  while (inrow < cinfo->max_v_samp_factor) {
     outptr = output_data[outrow];
     inptr0 = input_data[inrow];
     inptr1 = input_data[inrow+1];
@@ -378,6 +391,7 @@ h2v2_smooth_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
     *outptr = (JSAMPLE) ((membersum + 32768) >> 16);
 
     inrow += 2;
+    outrow++;
   }
 }
 
@@ -392,9 +406,9 @@ METHODDEF(void)
 fullsize_smooth_downsample (j_compress_ptr cinfo, jpeg_component_info *compptr,
 			    JSAMPARRAY input_data, JSAMPARRAY output_data)
 {
-  int outrow;
+  int inrow;
   JDIMENSION colctr;
-  JDIMENSION output_cols = compptr->width_in_blocks * DCTSIZE;
+  JDIMENSION output_cols = compptr->width_in_blocks * compptr->DCT_h_scaled_size;
   register JSAMPROW inptr, above_ptr, below_ptr, outptr;
   INT32 membersum, neighsum, memberscale, neighscale;
   int colsum, lastcolsum, nextcolsum;
@@ -415,11 +429,11 @@ fullsize_smooth_downsample (j_compress_ptr cinfo, jpeg_component_info *compptr,
   memberscale = 65536L - cinfo->smoothing_factor * 512L; /* scaled 1-8*SF */
   neighscale = cinfo->smoothing_factor * 64; /* scaled SF */
 
-  for (outrow = 0; outrow < compptr->v_samp_factor; outrow++) {
-    outptr = output_data[outrow];
-    inptr = input_data[outrow];
-    above_ptr = input_data[outrow-1];
-    below_ptr = input_data[outrow+1];
+  for (inrow = 0; inrow < cinfo->max_v_samp_factor; inrow++) {
+    outptr = output_data[inrow];
+    inptr = input_data[inrow];
+    above_ptr = input_data[inrow-1];
+    below_ptr = input_data[inrow+1];
 
     /* Special case for first column */
     colsum = GETJSAMPLE(*above_ptr++) + GETJSAMPLE(*below_ptr++) +
@@ -467,6 +481,7 @@ jinit_downsampler (j_compress_ptr cinfo)
   int ci;
   jpeg_component_info * compptr;
   boolean smoothok = TRUE;
+  int h_in_group, v_in_group, h_out_group, v_out_group;
 
   downsample = (my_downsample_ptr)
     (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
@@ -482,8 +497,17 @@ jinit_downsampler (j_compress_ptr cinfo)
   /* Verify we can handle the sampling factors, and set up method pointers */
   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
        ci++, compptr++) {
-    if (compptr->h_samp_factor == cinfo->max_h_samp_factor &&
-	compptr->v_samp_factor == cinfo->max_v_samp_factor) {
+    /* Compute size of an "output group" for DCT scaling.  This many samples
+     * are to be converted from max_h_samp_factor * max_v_samp_factor pixels.
+     */
+    h_out_group = (compptr->h_samp_factor * compptr->DCT_h_scaled_size) /
+		  cinfo->min_DCT_h_scaled_size;
+    v_out_group = (compptr->v_samp_factor * compptr->DCT_v_scaled_size) /
+		  cinfo->min_DCT_v_scaled_size;
+    h_in_group = cinfo->max_h_samp_factor;
+    v_in_group = cinfo->max_v_samp_factor;
+    downsample->rowgroup_height[ci] = v_out_group; /* save for use later */
+    if (h_in_group == h_out_group && v_in_group == v_out_group) {
 #ifdef INPUT_SMOOTHING_SUPPORTED
       if (cinfo->smoothing_factor) {
 	downsample->methods[ci] = fullsize_smooth_downsample;
@@ -491,12 +515,12 @@ jinit_downsampler (j_compress_ptr cinfo)
       } else
 #endif
 	downsample->methods[ci] = fullsize_downsample;
-    } else if (compptr->h_samp_factor * 2 == cinfo->max_h_samp_factor &&
-	       compptr->v_samp_factor == cinfo->max_v_samp_factor) {
+    } else if (h_in_group == h_out_group * 2 &&
+	       v_in_group == v_out_group) {
       smoothok = FALSE;
       downsample->methods[ci] = h2v1_downsample;
-    } else if (compptr->h_samp_factor * 2 == cinfo->max_h_samp_factor &&
-	       compptr->v_samp_factor * 2 == cinfo->max_v_samp_factor) {
+    } else if (h_in_group == h_out_group * 2 &&
+	       v_in_group == v_out_group * 2) {
 #ifdef INPUT_SMOOTHING_SUPPORTED
       if (cinfo->smoothing_factor) {
 	downsample->methods[ci] = h2v2_smooth_downsample;
@@ -504,10 +528,12 @@ jinit_downsampler (j_compress_ptr cinfo)
       } else
 #endif
 	downsample->methods[ci] = h2v2_downsample;
-    } else if ((cinfo->max_h_samp_factor % compptr->h_samp_factor) == 0 &&
-	       (cinfo->max_v_samp_factor % compptr->v_samp_factor) == 0) {
+    } else if ((h_in_group % h_out_group) == 0 &&
+	       (v_in_group % v_out_group) == 0) {
       smoothok = FALSE;
       downsample->methods[ci] = int_downsample;
+      downsample->h_expand[ci] = (UINT8) (h_in_group / h_out_group);
+      downsample->v_expand[ci] = (UINT8) (v_in_group / v_out_group);
     } else
       ERREXIT(cinfo, JERR_FRACT_SAMPLE_NOTIMPL);
   }
diff --git a/src/jpeg-6b/jctrans.c b/src/jpeg-8c/jctrans.c
index 0e6d7076..cee6b0f3 100644
--- a/src/jpeg-6b/jctrans.c
+++ b/src/jpeg-8c/jctrans.c
@@ -2,6 +2,7 @@
  * jctrans.c
  *
  * Copyright (C) 1995-1998, Thomas G. Lane.
+ * Modified 2000-2009 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -76,6 +77,10 @@ jpeg_copy_critical_parameters (j_decompress_ptr srcinfo,
   dstinfo->image_height = srcinfo->image_height;
   dstinfo->input_components = srcinfo->num_components;
   dstinfo->in_color_space = srcinfo->jpeg_color_space;
+  dstinfo->jpeg_width = srcinfo->output_width;
+  dstinfo->jpeg_height = srcinfo->output_height;
+  dstinfo->min_DCT_h_scaled_size = srcinfo->min_DCT_h_scaled_size;
+  dstinfo->min_DCT_v_scaled_size = srcinfo->min_DCT_v_scaled_size;
   /* Initialize all parameters to default values */
   jpeg_set_defaults(dstinfo);
   /* jpeg_set_defaults may choose wrong colorspace, eg YCbCr if input is RGB.
@@ -158,25 +163,14 @@ LOCAL(void)
 transencode_master_selection (j_compress_ptr cinfo,
 			      jvirt_barray_ptr * coef_arrays)
 {
-  /* Although we don't actually use input_components for transcoding,
-   * jcmaster.c's initial_setup will complain if input_components is 0.
-   */
-  cinfo->input_components = 1;
   /* Initialize master control (includes parameter checking/processing) */
   jinit_c_master_control(cinfo, TRUE /* transcode only */);
 
   /* Entropy encoding: either Huffman or arithmetic coding. */
-  if (cinfo->arith_code) {
-    ERREXIT(cinfo, JERR_ARITH_NOTIMPL);
-  } else {
-    if (cinfo->progressive_mode) {
-#ifdef C_PROGRESSIVE_SUPPORTED
-      jinit_phuff_encoder(cinfo);
-#else
-      ERREXIT(cinfo, JERR_NOT_COMPILED);
-#endif
-    } else
-      jinit_huff_encoder(cinfo);
+  if (cinfo->arith_code)
+    jinit_arith_encoder(cinfo);
+  else {
+    jinit_huff_encoder(cinfo);
   }
 
   /* We need a special coefficient buffer controller. */
diff --git a/src/jpeg-6b/jdapimin.c b/src/jpeg-8c/jdapimin.c
index cadb59fc..7f1ce4c0 100644
--- a/src/jpeg-6b/jdapimin.c
+++ b/src/jpeg-8c/jdapimin.c
@@ -2,6 +2,7 @@
  * jdapimin.c
  *
  * Copyright (C) 1994-1998, Thomas G. Lane.
+ * Modified 2009 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -185,8 +186,8 @@ default_decompress_parms (j_decompress_ptr cinfo)
   }
 
   /* Set defaults for other decompression parameters. */
-  cinfo->scale_num = 1;		/* 1:1 scaling */
-  cinfo->scale_denom = 1;
+  cinfo->scale_num = cinfo->block_size;		/* 1:1 scaling */
+  cinfo->scale_denom = cinfo->block_size;
   cinfo->output_gamma = 1.0;
   cinfo->buffered_image = FALSE;
   cinfo->raw_data_out = FALSE;
diff --git a/src/jpeg-6b/jdapistd.c b/src/jpeg-8c/jdapistd.c
index c8e3fa0c..9d745377 100644
--- a/src/jpeg-6b/jdapistd.c
+++ b/src/jpeg-8c/jdapistd.c
@@ -202,7 +202,7 @@ jpeg_read_raw_data (j_decompress_ptr cinfo, JSAMPIMAGE data,
   }
 
   /* Verify that at least one iMCU row can be returned. */
-  lines_per_iMCU_row = cinfo->max_v_samp_factor * cinfo->min_DCT_scaled_size;
+  lines_per_iMCU_row = cinfo->max_v_samp_factor * cinfo->min_DCT_v_scaled_size;
   if (max_lines < lines_per_iMCU_row)
     ERREXIT(cinfo, JERR_BUFFER_SIZE);
 
diff --git a/src/jpeg-8c/jdarith.c b/src/jpeg-8c/jdarith.c
new file mode 100644
index 00000000..c858b248
--- /dev/null
+++ b/src/jpeg-8c/jdarith.c
@@ -0,0 +1,772 @@
+/*
+ * jdarith.c
+ *
+ * Developed 1997-2009 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains portable arithmetic entropy decoding routines for JPEG
+ * (implementing the ISO/IEC IS 10918-1 and CCITT Recommendation ITU-T T.81).
+ *
+ * Both sequential and progressive modes are supported in this single module.
+ *
+ * Suspension is not currently supported in this module.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Expanded entropy decoder object for arithmetic decoding. */
+
+typedef struct {
+  struct jpeg_entropy_decoder pub; /* public fields */
+
+  INT32 c;       /* C register, base of coding interval + input bit buffer */
+  INT32 a;               /* A register, normalized size of coding interval */
+  int ct;     /* bit shift counter, # of bits left in bit buffer part of C */
+                                                         /* init: ct = -16 */
+                                                         /* run: ct = 0..7 */
+                                                         /* error: ct = -1 */
+  int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
+  int dc_context[MAX_COMPS_IN_SCAN]; /* context index for DC conditioning */
+
+  unsigned int restarts_to_go;	/* MCUs left in this restart interval */
+
+  /* Pointers to statistics areas (these workspaces have image lifespan) */
+  unsigned char * dc_stats[NUM_ARITH_TBLS];
+  unsigned char * ac_stats[NUM_ARITH_TBLS];
+
+  /* Statistics bin for coding with fixed probability 0.5 */
+  unsigned char fixed_bin[4];
+} arith_entropy_decoder;
+
+typedef arith_entropy_decoder * arith_entropy_ptr;
+
+/* The following two definitions specify the allocation chunk size
+ * for the statistics area.
+ * According to sections F.1.4.4.1.3 and F.1.4.4.2, we need at least
+ * 49 statistics bins for DC, and 245 statistics bins for AC coding.
+ *
+ * We use a compact representation with 1 byte per statistics bin,
+ * thus the numbers directly represent byte sizes.
+ * This 1 byte per statistics bin contains the meaning of the MPS
+ * (more probable symbol) in the highest bit (mask 0x80), and the
+ * index into the probability estimation state machine table
+ * in the lower bits (mask 0x7F).
+ */
+
+#define DC_STAT_BINS 64
+#define AC_STAT_BINS 256
+
+
+LOCAL(int)
+get_byte (j_decompress_ptr cinfo)
+/* Read next input byte; we do not support suspension in this module. */
+{
+  struct jpeg_source_mgr * src = cinfo->src;
+
+  if (src->bytes_in_buffer == 0)
+    if (! (*src->fill_input_buffer) (cinfo))
+      ERREXIT(cinfo, JERR_CANT_SUSPEND);
+  src->bytes_in_buffer--;
+  return GETJOCTET(*src->next_input_byte++);
+}
+
+
+/*
+ * The core arithmetic decoding routine (common in JPEG and JBIG).
+ * This needs to go as fast as possible.
+ * Machine-dependent optimization facilities
+ * are not utilized in this portable implementation.
+ * However, this code should be fairly efficient and
+ * may be a good base for further optimizations anyway.
+ *
+ * Return value is 0 or 1 (binary decision).
+ *
+ * Note: I've changed the handling of the code base & bit
+ * buffer register C compared to other implementations
+ * based on the standards layout & procedures.
+ * While it also contains both the actual base of the
+ * coding interval (16 bits) and the next-bits buffer,
+ * the cut-point between these two parts is floating
+ * (instead of fixed) with the bit shift counter CT.
+ * Thus, we also need only one (variable instead of
+ * fixed size) shift for the LPS/MPS decision, and
+ * we can get away with any renormalization update
+ * of C (except for new data insertion, of course).
+ *
+ * I've also introduced a new scheme for accessing
+ * the probability estimation state machine table,
+ * derived from Markus Kuhn's JBIG implementation.
+ */
+
+LOCAL(int)
+arith_decode (j_decompress_ptr cinfo, unsigned char *st)
+{
+  register arith_entropy_ptr e = (arith_entropy_ptr) cinfo->entropy;
+  register unsigned char nl, nm;
+  register INT32 qe, temp;
+  register int sv, data;
+
+  /* Renormalization & data input per section D.2.6 */
+  while (e->a < 0x8000L) {
+    if (--e->ct < 0) {
+      /* Need to fetch next data byte */
+      if (cinfo->unread_marker)
+	data = 0;		/* stuff zero data */
+      else {
+	data = get_byte(cinfo);	/* read next input byte */
+	if (data == 0xFF) {	/* zero stuff or marker code */
+	  do data = get_byte(cinfo);
+	  while (data == 0xFF);	/* swallow extra 0xFF bytes */
+	  if (data == 0)
+	    data = 0xFF;	/* discard stuffed zero byte */
+	  else {
+	    /* Note: Different from the Huffman decoder, hitting
+	     * a marker while processing the compressed data
+	     * segment is legal in arithmetic coding.
+	     * The convention is to supply zero data
+	     * then until decoding is complete.
+	     */
+	    cinfo->unread_marker = data;
+	    data = 0;
+	  }
+	}
+      }
+      e->c = (e->c << 8) | data; /* insert data into C register */
+      if ((e->ct += 8) < 0)	 /* update bit shift counter */
+	/* Need more initial bytes */
+	if (++e->ct == 0)
+	  /* Got 2 initial bytes -> re-init A and exit loop */
+	  e->a = 0x8000L; /* => e->a = 0x10000L after loop exit */
+    }
+    e->a <<= 1;
+  }
+
+  /* Fetch values from our compact representation of Table D.2:
+   * Qe values and probability estimation state machine
+   */
+  sv = *st;
+  qe = jpeg_aritab[sv & 0x7F];	/* => Qe_Value */
+  nl = qe & 0xFF; qe >>= 8;	/* Next_Index_LPS + Switch_MPS */
+  nm = qe & 0xFF; qe >>= 8;	/* Next_Index_MPS */
+
+  /* Decode & estimation procedures per sections D.2.4 & D.2.5 */
+  temp = e->a - qe;
+  e->a = temp;
+  temp <<= e->ct;
+  if (e->c >= temp) {
+    e->c -= temp;
+    /* Conditional LPS (less probable symbol) exchange */
+    if (e->a < qe) {
+      e->a = qe;
+      *st = (sv & 0x80) ^ nm;	/* Estimate_after_MPS */
+    } else {
+      e->a = qe;
+      *st = (sv & 0x80) ^ nl;	/* Estimate_after_LPS */
+      sv ^= 0x80;		/* Exchange LPS/MPS */
+    }
+  } else if (e->a < 0x8000L) {
+    /* Conditional MPS (more probable symbol) exchange */
+    if (e->a < qe) {
+      *st = (sv & 0x80) ^ nl;	/* Estimate_after_LPS */
+      sv ^= 0x80;		/* Exchange LPS/MPS */
+    } else {
+      *st = (sv & 0x80) ^ nm;	/* Estimate_after_MPS */
+    }
+  }
+
+  return sv >> 7;
+}
+
+
+/*
+ * Check for a restart marker & resynchronize decoder.
+ */
+
+LOCAL(void)
+process_restart (j_decompress_ptr cinfo)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  int ci;
+  jpeg_component_info * compptr;
+
+  /* Advance past the RSTn marker */
+  if (! (*cinfo->marker->read_restart_marker) (cinfo))
+    ERREXIT(cinfo, JERR_CANT_SUSPEND);
+
+  /* Re-initialize statistics areas */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    if (! cinfo->progressive_mode || (cinfo->Ss == 0 && cinfo->Ah == 0)) {
+      MEMZERO(entropy->dc_stats[compptr->dc_tbl_no], DC_STAT_BINS);
+      /* Reset DC predictions to 0 */
+      entropy->last_dc_val[ci] = 0;
+      entropy->dc_context[ci] = 0;
+    }
+    if ((! cinfo->progressive_mode && cinfo->lim_Se) ||
+	(cinfo->progressive_mode && cinfo->Ss)) {
+      MEMZERO(entropy->ac_stats[compptr->ac_tbl_no], AC_STAT_BINS);
+    }
+  }
+
+  /* Reset arithmetic decoding variables */
+  entropy->c = 0;
+  entropy->a = 0;
+  entropy->ct = -16;	/* force reading 2 initial bytes to fill C */
+
+  /* Reset restart counter */
+  entropy->restarts_to_go = cinfo->restart_interval;
+}
+
+
+/*
+ * Arithmetic MCU decoding.
+ * Each of these routines decodes and returns one MCU's worth of
+ * arithmetic-compressed coefficients.
+ * The coefficients are reordered from zigzag order into natural array order,
+ * but are not dequantized.
+ *
+ * The i'th block of the MCU is stored into the block pointed to by
+ * MCU_data[i].  WE ASSUME THIS AREA IS INITIALLY ZEROED BY THE CALLER.
+ */
+
+/*
+ * MCU decoding for DC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+decode_mcu_DC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  JBLOCKROW block;
+  unsigned char *st;
+  int blkn, ci, tbl, sign;
+  int v, m;
+
+  /* Process restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      process_restart(cinfo);
+    entropy->restarts_to_go--;
+  }
+
+  if (entropy->ct == -1) return TRUE;	/* if error do nothing */
+
+  /* Outer loop handles each block in the MCU */
+
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+    ci = cinfo->MCU_membership[blkn];
+    tbl = cinfo->cur_comp_info[ci]->dc_tbl_no;
+
+    /* Sections F.2.4.1 & F.1.4.4.1: Decoding of DC coefficients */
+
+    /* Table F.4: Point to statistics bin S0 for DC coefficient coding */
+    st = entropy->dc_stats[tbl] + entropy->dc_context[ci];
+
+    /* Figure F.19: Decode_DC_DIFF */
+    if (arith_decode(cinfo, st) == 0)
+      entropy->dc_context[ci] = 0;
+    else {
+      /* Figure F.21: Decoding nonzero value v */
+      /* Figure F.22: Decoding the sign of v */
+      sign = arith_decode(cinfo, st + 1);
+      st += 2; st += sign;
+      /* Figure F.23: Decoding the magnitude category of v */
+      if ((m = arith_decode(cinfo, st)) != 0) {
+	st = entropy->dc_stats[tbl] + 20;	/* Table F.4: X1 = 20 */
+	while (arith_decode(cinfo, st)) {
+	  if ((m <<= 1) == 0x8000) {
+	    WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
+	    entropy->ct = -1;			/* magnitude overflow */
+	    return TRUE;
+	  }
+	  st += 1;
+	}
+      }
+      /* Section F.1.4.4.1.2: Establish dc_context conditioning category */
+      if (m < (int) ((1L << cinfo->arith_dc_L[tbl]) >> 1))
+	entropy->dc_context[ci] = 0;		   /* zero diff category */
+      else if (m > (int) ((1L << cinfo->arith_dc_U[tbl]) >> 1))
+	entropy->dc_context[ci] = 12 + (sign * 4); /* large diff category */
+      else
+	entropy->dc_context[ci] = 4 + (sign * 4);  /* small diff category */
+      v = m;
+      /* Figure F.24: Decoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	if (arith_decode(cinfo, st)) v |= m;
+      v += 1; if (sign) v = -v;
+      entropy->last_dc_val[ci] += v;
+    }
+
+    /* Scale and output the DC coefficient (assumes jpeg_natural_order[0]=0) */
+    (*block)[0] = (JCOEF) (entropy->last_dc_val[ci] << cinfo->Al);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU decoding for AC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+decode_mcu_AC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  JBLOCKROW block;
+  unsigned char *st;
+  int tbl, sign, k;
+  int v, m;
+  const int * natural_order;
+
+  /* Process restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      process_restart(cinfo);
+    entropy->restarts_to_go--;
+  }
+
+  if (entropy->ct == -1) return TRUE;	/* if error do nothing */
+
+  natural_order = cinfo->natural_order;
+
+  /* There is always only one block per MCU */
+  block = MCU_data[0];
+  tbl = cinfo->cur_comp_info[0]->ac_tbl_no;
+
+  /* Sections F.2.4.2 & F.1.4.4.2: Decoding of AC coefficients */
+
+  /* Figure F.20: Decode_AC_coefficients */
+  for (k = cinfo->Ss; k <= cinfo->Se; k++) {
+    st = entropy->ac_stats[tbl] + 3 * (k - 1);
+    if (arith_decode(cinfo, st)) break;		/* EOB flag */
+    while (arith_decode(cinfo, st + 1) == 0) {
+      st += 3; k++;
+      if (k > cinfo->Se) {
+	WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
+	entropy->ct = -1;			/* spectral overflow */
+	return TRUE;
+      }
+    }
+    /* Figure F.21: Decoding nonzero value v */
+    /* Figure F.22: Decoding the sign of v */
+    sign = arith_decode(cinfo, entropy->fixed_bin);
+    st += 2;
+    /* Figure F.23: Decoding the magnitude category of v */
+    if ((m = arith_decode(cinfo, st)) != 0) {
+      if (arith_decode(cinfo, st)) {
+	m <<= 1;
+	st = entropy->ac_stats[tbl] +
+	     (k <= cinfo->arith_ac_K[tbl] ? 189 : 217);
+	while (arith_decode(cinfo, st)) {
+	  if ((m <<= 1) == 0x8000) {
+	    WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
+	    entropy->ct = -1;			/* magnitude overflow */
+	    return TRUE;
+	  }
+	  st += 1;
+	}
+      }
+    }
+    v = m;
+    /* Figure F.24: Decoding the magnitude bit pattern of v */
+    st += 14;
+    while (m >>= 1)
+      if (arith_decode(cinfo, st)) v |= m;
+    v += 1; if (sign) v = -v;
+    /* Scale and output coefficient in natural (dezigzagged) order */
+    (*block)[natural_order[k]] = (JCOEF) (v << cinfo->Al);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU decoding for DC successive approximation refinement scan.
+ */
+
+METHODDEF(boolean)
+decode_mcu_DC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  unsigned char *st;
+  int p1, blkn;
+
+  /* Process restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      process_restart(cinfo);
+    entropy->restarts_to_go--;
+  }
+
+  st = entropy->fixed_bin;	/* use fixed probability estimation */
+  p1 = 1 << cinfo->Al;		/* 1 in the bit position being coded */
+
+  /* Outer loop handles each block in the MCU */
+
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    /* Encoded data is simply the next bit of the two's-complement DC value */
+    if (arith_decode(cinfo, st))
+      MCU_data[blkn][0][0] |= p1;
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU decoding for AC successive approximation refinement scan.
+ */
+
+METHODDEF(boolean)
+decode_mcu_AC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  JBLOCKROW block;
+  JCOEFPTR thiscoef;
+  unsigned char *st;
+  int tbl, k, kex;
+  int p1, m1;
+  const int * natural_order;
+
+  /* Process restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      process_restart(cinfo);
+    entropy->restarts_to_go--;
+  }
+
+  if (entropy->ct == -1) return TRUE;	/* if error do nothing */
+
+  natural_order = cinfo->natural_order;
+
+  /* There is always only one block per MCU */
+  block = MCU_data[0];
+  tbl = cinfo->cur_comp_info[0]->ac_tbl_no;
+
+  p1 = 1 << cinfo->Al;		/* 1 in the bit position being coded */
+  m1 = (-1) << cinfo->Al;	/* -1 in the bit position being coded */
+
+  /* Establish EOBx (previous stage end-of-block) index */
+  for (kex = cinfo->Se; kex > 0; kex--)
+    if ((*block)[natural_order[kex]]) break;
+
+  for (k = cinfo->Ss; k <= cinfo->Se; k++) {
+    st = entropy->ac_stats[tbl] + 3 * (k - 1);
+    if (k > kex)
+      if (arith_decode(cinfo, st)) break;	/* EOB flag */
+    for (;;) {
+      thiscoef = *block + natural_order[k];
+      if (*thiscoef) {				/* previously nonzero coef */
+	if (arith_decode(cinfo, st + 2)) {
+	  if (*thiscoef < 0)
+	    *thiscoef += m1;
+	  else
+	    *thiscoef += p1;
+	}
+	break;
+      }
+      if (arith_decode(cinfo, st + 1)) {	/* newly nonzero coef */
+	if (arith_decode(cinfo, entropy->fixed_bin))
+	  *thiscoef = m1;
+	else
+	  *thiscoef = p1;
+	break;
+      }
+      st += 3; k++;
+      if (k > cinfo->Se) {
+	WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
+	entropy->ct = -1;			/* spectral overflow */
+	return TRUE;
+      }
+    }
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * Decode one MCU's worth of arithmetic-compressed coefficients.
+ */
+
+METHODDEF(boolean)
+decode_mcu (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  jpeg_component_info * compptr;
+  JBLOCKROW block;
+  unsigned char *st;
+  int blkn, ci, tbl, sign, k;
+  int v, m;
+  const int * natural_order;
+
+  /* Process restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      process_restart(cinfo);
+    entropy->restarts_to_go--;
+  }
+
+  if (entropy->ct == -1) return TRUE;	/* if error do nothing */
+
+  natural_order = cinfo->natural_order;
+
+  /* Outer loop handles each block in the MCU */
+
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+    ci = cinfo->MCU_membership[blkn];
+    compptr = cinfo->cur_comp_info[ci];
+
+    /* Sections F.2.4.1 & F.1.4.4.1: Decoding of DC coefficients */
+
+    tbl = compptr->dc_tbl_no;
+
+    /* Table F.4: Point to statistics bin S0 for DC coefficient coding */
+    st = entropy->dc_stats[tbl] + entropy->dc_context[ci];
+
+    /* Figure F.19: Decode_DC_DIFF */
+    if (arith_decode(cinfo, st) == 0)
+      entropy->dc_context[ci] = 0;
+    else {
+      /* Figure F.21: Decoding nonzero value v */
+      /* Figure F.22: Decoding the sign of v */
+      sign = arith_decode(cinfo, st + 1);
+      st += 2; st += sign;
+      /* Figure F.23: Decoding the magnitude category of v */
+      if ((m = arith_decode(cinfo, st)) != 0) {
+	st = entropy->dc_stats[tbl] + 20;	/* Table F.4: X1 = 20 */
+	while (arith_decode(cinfo, st)) {
+	  if ((m <<= 1) == 0x8000) {
+	    WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
+	    entropy->ct = -1;			/* magnitude overflow */
+	    return TRUE;
+	  }
+	  st += 1;
+	}
+      }
+      /* Section F.1.4.4.1.2: Establish dc_context conditioning category */
+      if (m < (int) ((1L << cinfo->arith_dc_L[tbl]) >> 1))
+	entropy->dc_context[ci] = 0;		   /* zero diff category */
+      else if (m > (int) ((1L << cinfo->arith_dc_U[tbl]) >> 1))
+	entropy->dc_context[ci] = 12 + (sign * 4); /* large diff category */
+      else
+	entropy->dc_context[ci] = 4 + (sign * 4);  /* small diff category */
+      v = m;
+      /* Figure F.24: Decoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	if (arith_decode(cinfo, st)) v |= m;
+      v += 1; if (sign) v = -v;
+      entropy->last_dc_val[ci] += v;
+    }
+
+    (*block)[0] = (JCOEF) entropy->last_dc_val[ci];
+
+    /* Sections F.2.4.2 & F.1.4.4.2: Decoding of AC coefficients */
+
+    tbl = compptr->ac_tbl_no;
+
+    /* Figure F.20: Decode_AC_coefficients */
+    for (k = 1; k <= cinfo->lim_Se; k++) {
+      st = entropy->ac_stats[tbl] + 3 * (k - 1);
+      if (arith_decode(cinfo, st)) break;	/* EOB flag */
+      while (arith_decode(cinfo, st + 1) == 0) {
+	st += 3; k++;
+	if (k > cinfo->lim_Se) {
+	  WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
+	  entropy->ct = -1;			/* spectral overflow */
+	  return TRUE;
+	}
+      }
+      /* Figure F.21: Decoding nonzero value v */
+      /* Figure F.22: Decoding the sign of v */
+      sign = arith_decode(cinfo, entropy->fixed_bin);
+      st += 2;
+      /* Figure F.23: Decoding the magnitude category of v */
+      if ((m = arith_decode(cinfo, st)) != 0) {
+	if (arith_decode(cinfo, st)) {
+	  m <<= 1;
+	  st = entropy->ac_stats[tbl] +
+	       (k <= cinfo->arith_ac_K[tbl] ? 189 : 217);
+	  while (arith_decode(cinfo, st)) {
+	    if ((m <<= 1) == 0x8000) {
+	      WARNMS(cinfo, JWRN_ARITH_BAD_CODE);
+	      entropy->ct = -1;			/* magnitude overflow */
+	      return TRUE;
+	    }
+	    st += 1;
+	  }
+	}
+      }
+      v = m;
+      /* Figure F.24: Decoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+	if (arith_decode(cinfo, st)) v |= m;
+      v += 1; if (sign) v = -v;
+      (*block)[natural_order[k]] = (JCOEF) v;
+    }
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * Initialize for an arithmetic-compressed scan.
+ */
+
+METHODDEF(void)
+start_pass (j_decompress_ptr cinfo)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr) cinfo->entropy;
+  int ci, tbl;
+  jpeg_component_info * compptr;
+
+  if (cinfo->progressive_mode) {
+    /* Validate progressive scan parameters */
+    if (cinfo->Ss == 0) {
+      if (cinfo->Se != 0)
+	goto bad;
+    } else {
+      /* need not check Ss/Se < 0 since they came from unsigned bytes */
+      if (cinfo->Se < cinfo->Ss || cinfo->Se > cinfo->lim_Se)
+	goto bad;
+      /* AC scans may have only one component */
+      if (cinfo->comps_in_scan != 1)
+	goto bad;
+    }
+    if (cinfo->Ah != 0) {
+      /* Successive approximation refinement scan: must have Al = Ah-1. */
+      if (cinfo->Ah-1 != cinfo->Al)
+	goto bad;
+    }
+    if (cinfo->Al > 13) {	/* need not check for < 0 */
+      bad:
+      ERREXIT4(cinfo, JERR_BAD_PROGRESSION,
+	       cinfo->Ss, cinfo->Se, cinfo->Ah, cinfo->Al);
+    }
+    /* Update progression status, and verify that scan order is legal.
+     * Note that inter-scan inconsistencies are treated as warnings
+     * not fatal errors ... not clear if this is right way to behave.
+     */
+    for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+      int coefi, cindex = cinfo->cur_comp_info[ci]->component_index;
+      int *coef_bit_ptr = & cinfo->coef_bits[cindex][0];
+      if (cinfo->Ss && coef_bit_ptr[0] < 0) /* AC without prior DC scan */
+	WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, 0);
+      for (coefi = cinfo->Ss; coefi <= cinfo->Se; coefi++) {
+	int expected = (coef_bit_ptr[coefi] < 0) ? 0 : coef_bit_ptr[coefi];
+	if (cinfo->Ah != expected)
+	  WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, coefi);
+	coef_bit_ptr[coefi] = cinfo->Al;
+      }
+    }
+    /* Select MCU decoding routine */
+    if (cinfo->Ah == 0) {
+      if (cinfo->Ss == 0)
+	entropy->pub.decode_mcu = decode_mcu_DC_first;
+      else
+	entropy->pub.decode_mcu = decode_mcu_AC_first;
+    } else {
+      if (cinfo->Ss == 0)
+	entropy->pub.decode_mcu = decode_mcu_DC_refine;
+      else
+	entropy->pub.decode_mcu = decode_mcu_AC_refine;
+    }
+  } else {
+    /* Check that the scan parameters Ss, Se, Ah/Al are OK for sequential JPEG.
+     * This ought to be an error condition, but we make it a warning.
+     */
+    if (cinfo->Ss != 0 || cinfo->Ah != 0 || cinfo->Al != 0 ||
+	(cinfo->Se < DCTSIZE2 && cinfo->Se != cinfo->lim_Se))
+      WARNMS(cinfo, JWRN_NOT_SEQUENTIAL);
+    /* Select MCU decoding routine */
+    entropy->pub.decode_mcu = decode_mcu;
+  }
+
+  /* Allocate & initialize requested statistics areas */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    if (! cinfo->progressive_mode || (cinfo->Ss == 0 && cinfo->Ah == 0)) {
+      tbl = compptr->dc_tbl_no;
+      if (tbl < 0 || tbl >= NUM_ARITH_TBLS)
+	ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl);
+      if (entropy->dc_stats[tbl] == NULL)
+	entropy->dc_stats[tbl] = (unsigned char *) (*cinfo->mem->alloc_small)
+	  ((j_common_ptr) cinfo, JPOOL_IMAGE, DC_STAT_BINS);
+      MEMZERO(entropy->dc_stats[tbl], DC_STAT_BINS);
+      /* Initialize DC predictions to 0 */
+      entropy->last_dc_val[ci] = 0;
+      entropy->dc_context[ci] = 0;
+    }
+    if ((! cinfo->progressive_mode && cinfo->lim_Se) ||
+	(cinfo->progressive_mode && cinfo->Ss)) {
+      tbl = compptr->ac_tbl_no;
+      if (tbl < 0 || tbl >= NUM_ARITH_TBLS)
+	ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl);
+      if (entropy->ac_stats[tbl] == NULL)
+	entropy->ac_stats[tbl] = (unsigned char *) (*cinfo->mem->alloc_small)
+	  ((j_common_ptr) cinfo, JPOOL_IMAGE, AC_STAT_BINS);
+      MEMZERO(entropy->ac_stats[tbl], AC_STAT_BINS);
+    }
+  }
+
+  /* Initialize arithmetic decoding variables */
+  entropy->c = 0;
+  entropy->a = 0;
+  entropy->ct = -16;	/* force reading 2 initial bytes to fill C */
+
+  /* Initialize restart counter */
+  entropy->restarts_to_go = cinfo->restart_interval;
+}
+
+
+/*
+ * Module initialization routine for arithmetic entropy decoding.
+ */
+
+GLOBAL(void)
+jinit_arith_decoder (j_decompress_ptr cinfo)
+{
+  arith_entropy_ptr entropy;
+  int i;
+
+  entropy = (arith_entropy_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				SIZEOF(arith_entropy_decoder));
+  cinfo->entropy = (struct jpeg_entropy_decoder *) entropy;
+  entropy->pub.start_pass = start_pass;
+
+  /* Mark tables unallocated */
+  for (i = 0; i < NUM_ARITH_TBLS; i++) {
+    entropy->dc_stats[i] = NULL;
+    entropy->ac_stats[i] = NULL;
+  }
+
+  /* Initialize index for fixed probability estimation */
+  entropy->fixed_bin[0] = 113;
+
+  if (cinfo->progressive_mode) {
+    /* Create progression status table */
+    int *coef_bit_ptr, ci;
+    cinfo->coef_bits = (int (*)[DCTSIZE2])
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				  cinfo->num_components*DCTSIZE2*SIZEOF(int));
+    coef_bit_ptr = & cinfo->coef_bits[0][0];
+    for (ci = 0; ci < cinfo->num_components; ci++) 
+      for (i = 0; i < DCTSIZE2; i++)
+	*coef_bit_ptr++ = -1;
+  }
+}
diff --git a/src/jpeg-6b/jdatadst.c b/src/jpeg-8c/jdatadst.c
index a8f6fb0e..472d5f32 100644
--- a/src/jpeg-6b/jdatadst.c
+++ b/src/jpeg-8c/jdatadst.c
@@ -2,13 +2,14 @@
  * jdatadst.c
  *
  * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2009 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
  * This file contains compression data destination routines for the case of
- * emitting JPEG data to a file (or any stdio stream).  While these routines
- * are sufficient for most applications, some will want to use a different
- * destination manager.
+ * emitting JPEG data to memory or to a file (or any stdio stream).
+ * While these routines are sufficient for most applications,
+ * some will want to use a different destination manager.
  * IMPORTANT: we assume that fwrite() will correctly transcribe an array of
  * JOCTETs into 8-bit-wide elements on external storage.  If char is wider
  * than 8 bits on your machine, you may need to do some tweaking.
@@ -19,6 +20,11 @@
 #include "jpeglib.h"
 #include "jerror.h"
 
+#ifndef HAVE_STDLIB_H		/* <stdlib.h> should declare malloc(),free() */
+extern void * malloc JPP((size_t size));
+extern void free JPP((void *ptr));
+#endif
+
 
 /* Expanded data destination object for stdio output */
 
@@ -34,6 +40,21 @@ typedef my_destination_mgr * my_dest_ptr;
 #define OUTPUT_BUF_SIZE  4096	/* choose an efficiently fwrite'able size */
 
 
+/* Expanded data destination object for memory output */
+
+typedef struct {
+  struct jpeg_destination_mgr pub; /* public fields */
+
+  unsigned char ** outbuffer;	/* target buffer */
+  unsigned long * outsize;
+  unsigned char * newbuffer;	/* newly allocated buffer */
+  JOCTET * buffer;		/* start of buffer */
+  size_t bufsize;
+} my_mem_destination_mgr;
+
+typedef my_mem_destination_mgr * my_mem_dest_ptr;
+
+
 /*
  * Initialize destination --- called by jpeg_start_compress
  * before any data is actually written.
@@ -53,6 +74,12 @@ init_destination (j_compress_ptr cinfo)
   dest->pub.free_in_buffer = OUTPUT_BUF_SIZE;
 }
 
+METHODDEF(void)
+init_mem_destination (j_compress_ptr cinfo)
+{
+  /* no work necessary here */
+}
+
 
 /*
  * Empty the output buffer --- called whenever buffer fills up.
@@ -92,6 +119,36 @@ empty_output_buffer (j_compress_ptr cinfo)
   return TRUE;
 }
 
+METHODDEF(boolean)
+empty_mem_output_buffer (j_compress_ptr cinfo)
+{
+  size_t nextsize;
+  JOCTET * nextbuffer;
+  my_mem_dest_ptr dest = (my_mem_dest_ptr) cinfo->dest;
+
+  /* Try to allocate new buffer with double size */
+  nextsize = dest->bufsize * 2;
+  nextbuffer = malloc(nextsize);
+
+  if (nextbuffer == NULL)
+    ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, 10);
+
+  MEMCOPY(nextbuffer, dest->buffer, dest->bufsize);
+
+  if (dest->newbuffer != NULL)
+    free(dest->newbuffer);
+
+  dest->newbuffer = nextbuffer;
+
+  dest->pub.next_output_byte = nextbuffer + dest->bufsize;
+  dest->pub.free_in_buffer = dest->bufsize;
+
+  dest->buffer = nextbuffer;
+  dest->bufsize = nextsize;
+
+  return TRUE;
+}
+
 
 /*
  * Terminate destination --- called by jpeg_finish_compress
@@ -119,6 +176,15 @@ term_destination (j_compress_ptr cinfo)
     ERREXIT(cinfo, JERR_FILE_WRITE);
 }
 
+METHODDEF(void)
+term_mem_destination (j_compress_ptr cinfo)
+{
+  my_mem_dest_ptr dest = (my_mem_dest_ptr) cinfo->dest;
+
+  *dest->outbuffer = dest->buffer;
+  *dest->outsize = dest->bufsize - dest->pub.free_in_buffer;
+}
+
 
 /*
  * Prepare for output to a stdio stream.
@@ -149,3 +215,53 @@ jpeg_stdio_dest (j_compress_ptr cinfo, FILE * outfile)
   dest->pub.term_destination = term_destination;
   dest->outfile = outfile;
 }
+
+
+/*
+ * Prepare for output to a memory buffer.
+ * The caller may supply an own initial buffer with appropriate size.
+ * Otherwise, or when the actual data output exceeds the given size,
+ * the library adapts the buffer size as necessary.
+ * The standard library functions malloc/free are used for allocating
+ * larger memory, so the buffer is available to the application after
+ * finishing compression, and then the application is responsible for
+ * freeing the requested memory.
+ */
+
+GLOBAL(void)
+jpeg_mem_dest (j_compress_ptr cinfo,
+	       unsigned char ** outbuffer, unsigned long * outsize)
+{
+  my_mem_dest_ptr dest;
+
+  if (outbuffer == NULL || outsize == NULL)	/* sanity check */
+    ERREXIT(cinfo, JERR_BUFFER_SIZE);
+
+  /* The destination object is made permanent so that multiple JPEG images
+   * can be written to the same buffer without re-executing jpeg_mem_dest.
+   */
+  if (cinfo->dest == NULL) {	/* first time for this JPEG object? */
+    cinfo->dest = (struct jpeg_destination_mgr *)
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
+				  SIZEOF(my_mem_destination_mgr));
+  }
+
+  dest = (my_mem_dest_ptr) cinfo->dest;
+  dest->pub.init_destination = init_mem_destination;
+  dest->pub.empty_output_buffer = empty_mem_output_buffer;
+  dest->pub.term_destination = term_mem_destination;
+  dest->outbuffer = outbuffer;
+  dest->outsize = outsize;
+  dest->newbuffer = NULL;
+
+  if (*outbuffer == NULL || *outsize == 0) {
+    /* Allocate initial buffer */
+    dest->newbuffer = *outbuffer = malloc(OUTPUT_BUF_SIZE);
+    if (dest->newbuffer == NULL)
+      ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, 10);
+    *outsize = OUTPUT_BUF_SIZE;
+  }
+
+  dest->pub.next_output_byte = dest->buffer = *outbuffer;
+  dest->pub.free_in_buffer = dest->bufsize = *outsize;
+}
diff --git a/src/jpeg-6b/jdatasrc.c b/src/jpeg-8c/jdatasrc.c
index 795f65ed..c8fe3daf 100644
--- a/src/jpeg-6b/jdatasrc.c
+++ b/src/jpeg-8c/jdatasrc.c
@@ -2,13 +2,14 @@
  * jdatasrc.c
  *
  * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2009-2010 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
  * This file contains decompression data source routines for the case of
- * reading JPEG data from a file (or any stdio stream).  While these routines
- * are sufficient for most applications, some will want to use a different
- * source manager.
+ * reading JPEG data from memory or from a file (or any stdio stream).
+ * While these routines are sufficient for most applications,
+ * some will want to use a different source manager.
  * IMPORTANT: we assume that fread() will correctly transcribe an array of
  * JOCTETs from 8-bit-wide elements on external storage.  If char is wider
  * than 8 bits on your machine, you may need to do some tweaking.
@@ -19,17 +20,13 @@
 #include "jpeglib.h"
 #include "jerror.h"
 
-#ifndef MIN
-#define MIN(a, b) ((a)<(b)?(a):(b))
-#endif
 
 /* Expanded data source object for stdio input */
 
 typedef struct {
   struct jpeg_source_mgr pub;	/* public fields */
 
-  unsigned char *inbuf;		/* source stream */
-  size_t inbufbytes;
+  FILE * infile;		/* source stream */
   JOCTET * buffer;		/* start of buffer */
   boolean start_of_file;	/* have we gotten any data yet? */
 } my_source_mgr;
@@ -56,6 +53,12 @@ init_source (j_decompress_ptr cinfo)
   src->start_of_file = TRUE;
 }
 
+METHODDEF(void)
+init_mem_source (j_decompress_ptr cinfo)
+{
+  /* no work necessary here */
+}
+
 
 /*
  * Fill the input buffer --- called whenever buffer is emptied.
@@ -94,19 +97,18 @@ METHODDEF(boolean)
 fill_input_buffer (j_decompress_ptr cinfo)
 {
   my_src_ptr src = (my_src_ptr) cinfo->src;
-  size_t nbytes = MIN(src->inbufbytes, INPUT_BUF_SIZE);
+  size_t nbytes;
+
+  nbytes = JFREAD(src->infile, src->buffer, INPUT_BUF_SIZE);
 
   if (nbytes <= 0) {
+    if (src->start_of_file)	/* Treat empty input file as fatal error */
+      ERREXIT(cinfo, JERR_INPUT_EMPTY);
     WARNMS(cinfo, JWRN_JPEG_EOF);
     /* Insert a fake EOI marker */
     src->buffer[0] = (JOCTET) 0xFF;
     src->buffer[1] = (JOCTET) JPEG_EOI;
     nbytes = 2;
-  } else {
-    memcpy( src->buffer, src->inbuf, nbytes);
-
-    src->inbuf += nbytes;
-    src->inbufbytes -= nbytes;
   }
 
   src->pub.next_input_byte = src->buffer;
@@ -116,6 +118,26 @@ fill_input_buffer (j_decompress_ptr cinfo)
   return TRUE;
 }
 
+METHODDEF(boolean)
+fill_mem_input_buffer (j_decompress_ptr cinfo)
+{
+  static JOCTET mybuffer[4];
+
+  /* The whole JPEG data is expected to reside in the supplied memory
+   * buffer, so any request for more data beyond the given buffer size
+   * is treated as an error.
+   */
+  WARNMS(cinfo, JWRN_JPEG_EOF);
+  /* Insert a fake EOI marker */
+  mybuffer[0] = (JOCTET) 0xFF;
+  mybuffer[1] = (JOCTET) JPEG_EOI;
+
+  cinfo->src->next_input_byte = mybuffer;
+  cinfo->src->bytes_in_buffer = 2;
+
+  return TRUE;
+}
+
 
 /*
  * Skip data --- used to skip over a potentially large amount of
@@ -132,22 +154,22 @@ fill_input_buffer (j_decompress_ptr cinfo)
 METHODDEF(void)
 skip_input_data (j_decompress_ptr cinfo, long num_bytes)
 {
-  my_src_ptr src = (my_src_ptr) cinfo->src;
+  struct jpeg_source_mgr * src = cinfo->src;
 
   /* Just a dumb implementation for now.  Could use fseek() except
    * it doesn't work on pipes.  Not clear that being smart is worth
    * any trouble anyway --- large skips are infrequent.
    */
   if (num_bytes > 0) {
-    while (num_bytes > (long) src->pub.bytes_in_buffer) {
-      num_bytes -= (long) src->pub.bytes_in_buffer;
-      (void) fill_input_buffer(cinfo);
+    while (num_bytes > (long) src->bytes_in_buffer) {
+      num_bytes -= (long) src->bytes_in_buffer;
+      (void) (*src->fill_input_buffer) (cinfo);
       /* note we assume that fill_input_buffer will never return FALSE,
        * so suspension need not be handled.
        */
     }
-    src->pub.next_input_byte += (size_t) num_bytes;
-    src->pub.bytes_in_buffer -= (size_t) num_bytes;
+    src->next_input_byte += (size_t) num_bytes;
+    src->bytes_in_buffer -= (size_t) num_bytes;
   }
 }
 
@@ -184,7 +206,7 @@ term_source (j_decompress_ptr cinfo)
  */
 
 GLOBAL(void)
-jpeg_mem_src (j_decompress_ptr cinfo, unsigned char *inbuf, size_t size)
+jpeg_stdio_src (j_decompress_ptr cinfo, FILE * infile)
 {
   my_src_ptr src;
 
@@ -211,8 +233,42 @@ jpeg_mem_src (j_decompress_ptr cinfo, unsigned char *inbuf, size_t size)
   src->pub.skip_input_data = skip_input_data;
   src->pub.resync_to_restart = jpeg_resync_to_restart; /* use default method */
   src->pub.term_source = term_source;
-  src->inbuf = inbuf;
-  src->inbufbytes = size;
+  src->infile = infile;
   src->pub.bytes_in_buffer = 0; /* forces fill_input_buffer on first read */
   src->pub.next_input_byte = NULL; /* until buffer loaded */
 }
+
+
+/*
+ * Prepare for input from a supplied memory buffer.
+ * The buffer must contain the whole JPEG data.
+ */
+
+GLOBAL(void)
+jpeg_mem_src (j_decompress_ptr cinfo,
+	      unsigned char * inbuffer, unsigned long insize)
+{
+  struct jpeg_source_mgr * src;
+
+  if (inbuffer == NULL || insize == 0)	/* Treat empty input as fatal error */
+    ERREXIT(cinfo, JERR_INPUT_EMPTY);
+
+  /* The source object is made permanent so that a series of JPEG images
+   * can be read from the same buffer by calling jpeg_mem_src only before
+   * the first one.
+   */
+  if (cinfo->src == NULL) {	/* first time for this JPEG object? */
+    cinfo->src = (struct jpeg_source_mgr *)
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
+				  SIZEOF(struct jpeg_source_mgr));
+  }
+
+  src = cinfo->src;
+  src->init_source = init_mem_source;
+  src->fill_input_buffer = fill_mem_input_buffer;
+  src->skip_input_data = skip_input_data;
+  src->resync_to_restart = jpeg_resync_to_restart; /* use default method */
+  src->term_source = term_source;
+  src->bytes_in_buffer = (size_t) insize;
+  src->next_input_byte = (JOCTET *) inbuffer;
+}
diff --git a/src/jpeg-6b/jdcoefct.c b/src/jpeg-8c/jdcoefct.c
index 4938d20f..462e92c6 100644
--- a/src/jpeg-6b/jdcoefct.c
+++ b/src/jpeg-8c/jdcoefct.c
@@ -187,7 +187,7 @@ decompress_onepass (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
 	useful_width = (MCU_col_num < last_MCU_col) ? compptr->MCU_width
 						    : compptr->last_col_width;
 	output_ptr = output_buf[compptr->component_index] +
-	  yoffset * compptr->DCT_scaled_size;
+	  yoffset * compptr->DCT_v_scaled_size;
 	start_col = MCU_col_num * compptr->MCU_sample_width;
 	for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
 	  if (cinfo->input_iMCU_row < last_iMCU_row ||
@@ -197,11 +197,11 @@ decompress_onepass (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
 	      (*inverse_DCT) (cinfo, compptr,
 			      (JCOEFPTR) coef->MCU_buffer[blkn+xindex],
 			      output_ptr, output_col);
-	      output_col += compptr->DCT_scaled_size;
+	      output_col += compptr->DCT_h_scaled_size;
 	    }
 	  }
 	  blkn += compptr->MCU_width;
-	  output_ptr += compptr->DCT_scaled_size;
+	  output_ptr += compptr->DCT_v_scaled_size;
 	}
       }
     }
@@ -362,9 +362,9 @@ decompress_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
 	(*inverse_DCT) (cinfo, compptr, (JCOEFPTR) buffer_ptr,
 			output_ptr, output_col);
 	buffer_ptr++;
-	output_col += compptr->DCT_scaled_size;
+	output_col += compptr->DCT_h_scaled_size;
       }
-      output_ptr += compptr->DCT_scaled_size;
+      output_ptr += compptr->DCT_v_scaled_size;
     }
   }
 
@@ -654,9 +654,9 @@ decompress_smooth_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
 	DC4 = DC5; DC5 = DC6;
 	DC7 = DC8; DC8 = DC9;
 	buffer_ptr++, prev_block_row++, next_block_row++;
-	output_col += compptr->DCT_scaled_size;
+	output_col += compptr->DCT_h_scaled_size;
       }
-      output_ptr += compptr->DCT_scaled_size;
+      output_ptr += compptr->DCT_v_scaled_size;
     }
   }
 
diff --git a/src/jpeg-6b/jdcolor.c b/src/jpeg-8c/jdcolor.c
index 6c04dfe8..6c04dfe8 100644
--- a/src/jpeg-6b/jdcolor.c
+++ b/src/jpeg-8c/jdcolor.c
diff --git a/src/jpeg-8c/jdct.h b/src/jpeg-8c/jdct.h
new file mode 100644
index 00000000..360dec80
--- /dev/null
+++ b/src/jpeg-8c/jdct.h
@@ -0,0 +1,393 @@
+/*
+ * jdct.h
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This include file contains common declarations for the forward and
+ * inverse DCT modules.  These declarations are private to the DCT managers
+ * (jcdctmgr.c, jddctmgr.c) and the individual DCT algorithms.
+ * The individual DCT algorithms are kept in separate files to ease 
+ * machine-dependent tuning (e.g., assembly coding).
+ */
+
+
+/*
+ * A forward DCT routine is given a pointer to an input sample array and
+ * a pointer to a work area of type DCTELEM[]; the DCT is to be performed
+ * in-place in that buffer.  Type DCTELEM is int for 8-bit samples, INT32
+ * for 12-bit samples.  (NOTE: Floating-point DCT implementations use an
+ * array of type FAST_FLOAT, instead.)
+ * The input data is to be fetched from the sample array starting at a
+ * specified column.  (Any row offset needed will be applied to the array
+ * pointer before it is passed to the FDCT code.)
+ * Note that the number of samples fetched by the FDCT routine is
+ * DCT_h_scaled_size * DCT_v_scaled_size.
+ * The DCT outputs are returned scaled up by a factor of 8; they therefore
+ * have a range of +-8K for 8-bit data, +-128K for 12-bit data.  This
+ * convention improves accuracy in integer implementations and saves some
+ * work in floating-point ones.
+ * Quantization of the output coefficients is done by jcdctmgr.c.
+ */
+
+#if BITS_IN_JSAMPLE == 8
+typedef int DCTELEM;		/* 16 or 32 bits is fine */
+#else
+typedef INT32 DCTELEM;		/* must have 32 bits */
+#endif
+
+typedef JMETHOD(void, forward_DCT_method_ptr, (DCTELEM * data,
+					       JSAMPARRAY sample_data,
+					       JDIMENSION start_col));
+typedef JMETHOD(void, float_DCT_method_ptr, (FAST_FLOAT * data,
+					     JSAMPARRAY sample_data,
+					     JDIMENSION start_col));
+
+
+/*
+ * An inverse DCT routine is given a pointer to the input JBLOCK and a pointer
+ * to an output sample array.  The routine must dequantize the input data as
+ * well as perform the IDCT; for dequantization, it uses the multiplier table
+ * pointed to by compptr->dct_table.  The output data is to be placed into the
+ * sample array starting at a specified column.  (Any row offset needed will
+ * be applied to the array pointer before it is passed to the IDCT code.)
+ * Note that the number of samples emitted by the IDCT routine is
+ * DCT_h_scaled_size * DCT_v_scaled_size.
+ */
+
+/* typedef inverse_DCT_method_ptr is declared in jpegint.h */
+
+/*
+ * Each IDCT routine has its own ideas about the best dct_table element type.
+ */
+
+typedef MULTIPLIER ISLOW_MULT_TYPE; /* short or int, whichever is faster */
+#if BITS_IN_JSAMPLE == 8
+typedef MULTIPLIER IFAST_MULT_TYPE; /* 16 bits is OK, use short if faster */
+#define IFAST_SCALE_BITS  2	/* fractional bits in scale factors */
+#else
+typedef INT32 IFAST_MULT_TYPE;	/* need 32 bits for scaled quantizers */
+#define IFAST_SCALE_BITS  13	/* fractional bits in scale factors */
+#endif
+typedef FAST_FLOAT FLOAT_MULT_TYPE; /* preferred floating type */
+
+
+/*
+ * Each IDCT routine is responsible for range-limiting its results and
+ * converting them to unsigned form (0..MAXJSAMPLE).  The raw outputs could
+ * be quite far out of range if the input data is corrupt, so a bulletproof
+ * range-limiting step is required.  We use a mask-and-table-lookup method
+ * to do the combined operations quickly.  See the comments with
+ * prepare_range_limit_table (in jdmaster.c) for more info.
+ */
+
+#define IDCT_range_limit(cinfo)  ((cinfo)->sample_range_limit + CENTERJSAMPLE)
+
+#define RANGE_MASK  (MAXJSAMPLE * 4 + 3) /* 2 bits wider than legal samples */
+
+
+/* Short forms of external names for systems with brain-damaged linkers. */
+
+#ifdef NEED_SHORT_EXTERNAL_NAMES
+#define jpeg_fdct_islow		jFDislow
+#define jpeg_fdct_ifast		jFDifast
+#define jpeg_fdct_float		jFDfloat
+#define jpeg_fdct_7x7		jFD7x7
+#define jpeg_fdct_6x6		jFD6x6
+#define jpeg_fdct_5x5		jFD5x5
+#define jpeg_fdct_4x4		jFD4x4
+#define jpeg_fdct_3x3		jFD3x3
+#define jpeg_fdct_2x2		jFD2x2
+#define jpeg_fdct_1x1		jFD1x1
+#define jpeg_fdct_9x9		jFD9x9
+#define jpeg_fdct_10x10		jFD10x10
+#define jpeg_fdct_11x11		jFD11x11
+#define jpeg_fdct_12x12		jFD12x12
+#define jpeg_fdct_13x13		jFD13x13
+#define jpeg_fdct_14x14		jFD14x14
+#define jpeg_fdct_15x15		jFD15x15
+#define jpeg_fdct_16x16		jFD16x16
+#define jpeg_fdct_16x8		jFD16x8
+#define jpeg_fdct_14x7		jFD14x7
+#define jpeg_fdct_12x6		jFD12x6
+#define jpeg_fdct_10x5		jFD10x5
+#define jpeg_fdct_8x4		jFD8x4
+#define jpeg_fdct_6x3		jFD6x3
+#define jpeg_fdct_4x2		jFD4x2
+#define jpeg_fdct_2x1		jFD2x1
+#define jpeg_fdct_8x16		jFD8x16
+#define jpeg_fdct_7x14		jFD7x14
+#define jpeg_fdct_6x12		jFD6x12
+#define jpeg_fdct_5x10		jFD5x10
+#define jpeg_fdct_4x8		jFD4x8
+#define jpeg_fdct_3x6		jFD3x6
+#define jpeg_fdct_2x4		jFD2x4
+#define jpeg_fdct_1x2		jFD1x2
+#define jpeg_idct_islow		jRDislow
+#define jpeg_idct_ifast		jRDifast
+#define jpeg_idct_float		jRDfloat
+#define jpeg_idct_7x7		jRD7x7
+#define jpeg_idct_6x6		jRD6x6
+#define jpeg_idct_5x5		jRD5x5
+#define jpeg_idct_4x4		jRD4x4
+#define jpeg_idct_3x3		jRD3x3
+#define jpeg_idct_2x2		jRD2x2
+#define jpeg_idct_1x1		jRD1x1
+#define jpeg_idct_9x9		jRD9x9
+#define jpeg_idct_10x10		jRD10x10
+#define jpeg_idct_11x11		jRD11x11
+#define jpeg_idct_12x12		jRD12x12
+#define jpeg_idct_13x13		jRD13x13
+#define jpeg_idct_14x14		jRD14x14
+#define jpeg_idct_15x15		jRD15x15
+#define jpeg_idct_16x16		jRD16x16
+#define jpeg_idct_16x8		jRD16x8
+#define jpeg_idct_14x7		jRD14x7
+#define jpeg_idct_12x6		jRD12x6
+#define jpeg_idct_10x5		jRD10x5
+#define jpeg_idct_8x4		jRD8x4
+#define jpeg_idct_6x3		jRD6x3
+#define jpeg_idct_4x2		jRD4x2
+#define jpeg_idct_2x1		jRD2x1
+#define jpeg_idct_8x16		jRD8x16
+#define jpeg_idct_7x14		jRD7x14
+#define jpeg_idct_6x12		jRD6x12
+#define jpeg_idct_5x10		jRD5x10
+#define jpeg_idct_4x8		jRD4x8
+#define jpeg_idct_3x6		jRD3x8
+#define jpeg_idct_2x4		jRD2x4
+#define jpeg_idct_1x2		jRD1x2
+#endif /* NEED_SHORT_EXTERNAL_NAMES */
+
+/* Extern declarations for the forward and inverse DCT routines. */
+
+EXTERN(void) jpeg_fdct_islow
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_ifast
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_float
+    JPP((FAST_FLOAT * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_7x7
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_6x6
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_5x5
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_4x4
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_3x3
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_2x2
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_1x1
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_9x9
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_10x10
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_11x11
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_12x12
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_13x13
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_14x14
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_15x15
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_16x16
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_16x8
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_14x7
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_12x6
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_10x5
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_8x4
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_6x3
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_4x2
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_2x1
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_8x16
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_7x14
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_6x12
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_5x10
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_4x8
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_3x6
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_2x4
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+EXTERN(void) jpeg_fdct_1x2
+    JPP((DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col));
+
+EXTERN(void) jpeg_idct_islow
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_ifast
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_float
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_7x7
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_6x6
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_5x5
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_4x4
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_3x3
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_2x2
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_1x1
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_9x9
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_10x10
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_11x11
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_12x12
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_13x13
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_14x14
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_15x15
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_16x16
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_16x8
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_14x7
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_12x6
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_10x5
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_8x4
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_6x3
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_4x2
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_2x1
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_8x16
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_7x14
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_6x12
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_5x10
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_4x8
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_3x6
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_2x4
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+EXTERN(void) jpeg_idct_1x2
+    JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
+
+
+/*
+ * Macros for handling fixed-point arithmetic; these are used by many
+ * but not all of the DCT/IDCT modules.
+ *
+ * All values are expected to be of type INT32.
+ * Fractional constants are scaled left by CONST_BITS bits.
+ * CONST_BITS is defined within each module using these macros,
+ * and may differ from one module to the next.
+ */
+
+#define ONE	((INT32) 1)
+#define CONST_SCALE (ONE << CONST_BITS)
+
+/* Convert a positive real constant to an integer scaled by CONST_SCALE.
+ * Caution: some C compilers fail to reduce "FIX(constant)" at compile time,
+ * thus causing a lot of useless floating-point operations at run time.
+ */
+
+#define FIX(x)	((INT32) ((x) * CONST_SCALE + 0.5))
+
+/* Descale and correctly round an INT32 value that's scaled by N bits.
+ * We assume RIGHT_SHIFT rounds towards minus infinity, so adding
+ * the fudge factor is correct for either sign of X.
+ */
+
+#define DESCALE(x,n)  RIGHT_SHIFT((x) + (ONE << ((n)-1)), n)
+
+/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
+ * This macro is used only when the two inputs will actually be no more than
+ * 16 bits wide, so that a 16x16->32 bit multiply can be used instead of a
+ * full 32x32 multiply.  This provides a useful speedup on many machines.
+ * Unfortunately there is no way to specify a 16x16->32 multiply portably
+ * in C, but some C compilers will do the right thing if you provide the
+ * correct combination of casts.
+ */
+
+#ifdef SHORTxSHORT_32		/* may work if 'int' is 32 bits */
+#define MULTIPLY16C16(var,const)  (((INT16) (var)) * ((INT16) (const)))
+#endif
+#ifdef SHORTxLCONST_32		/* known to work with Microsoft C 6.0 */
+#define MULTIPLY16C16(var,const)  (((INT16) (var)) * ((INT32) (const)))
+#endif
+
+#ifndef MULTIPLY16C16		/* default definition */
+#define MULTIPLY16C16(var,const)  ((var) * (const))
+#endif
+
+/* Same except both inputs are variables. */
+
+#ifdef SHORTxSHORT_32		/* may work if 'int' is 32 bits */
+#define MULTIPLY16V16(var1,var2)  (((INT16) (var1)) * ((INT16) (var2)))
+#endif
+
+#ifndef MULTIPLY16V16		/* default definition */
+#define MULTIPLY16V16(var1,var2)  ((var1) * (var2))
+#endif
diff --git a/src/jpeg-6b/jddctmgr.c b/src/jpeg-8c/jddctmgr.c
index bbf8d0e9..0ded9d57 100644
--- a/src/jpeg-6b/jddctmgr.c
+++ b/src/jpeg-8c/jddctmgr.c
@@ -2,6 +2,7 @@
  * jddctmgr.c
  *
  * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2002-2010 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -98,22 +99,134 @@ start_pass (j_decompress_ptr cinfo)
   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
        ci++, compptr++) {
     /* Select the proper IDCT routine for this component's scaling */
-    switch (compptr->DCT_scaled_size) {
+    switch ((compptr->DCT_h_scaled_size << 8) + compptr->DCT_v_scaled_size) {
 #ifdef IDCT_SCALING_SUPPORTED
-    case 1:
+    case ((1 << 8) + 1):
       method_ptr = jpeg_idct_1x1;
-      method = JDCT_ISLOW;	/* jidctred uses islow-style table */
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
       break;
-    case 2:
+    case ((2 << 8) + 2):
       method_ptr = jpeg_idct_2x2;
-      method = JDCT_ISLOW;	/* jidctred uses islow-style table */
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
       break;
-    case 4:
+    case ((3 << 8) + 3):
+      method_ptr = jpeg_idct_3x3;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((4 << 8) + 4):
       method_ptr = jpeg_idct_4x4;
-      method = JDCT_ISLOW;	/* jidctred uses islow-style table */
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((5 << 8) + 5):
+      method_ptr = jpeg_idct_5x5;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((6 << 8) + 6):
+      method_ptr = jpeg_idct_6x6;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((7 << 8) + 7):
+      method_ptr = jpeg_idct_7x7;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((9 << 8) + 9):
+      method_ptr = jpeg_idct_9x9;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((10 << 8) + 10):
+      method_ptr = jpeg_idct_10x10;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((11 << 8) + 11):
+      method_ptr = jpeg_idct_11x11;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((12 << 8) + 12):
+      method_ptr = jpeg_idct_12x12;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((13 << 8) + 13):
+      method_ptr = jpeg_idct_13x13;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((14 << 8) + 14):
+      method_ptr = jpeg_idct_14x14;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((15 << 8) + 15):
+      method_ptr = jpeg_idct_15x15;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((16 << 8) + 16):
+      method_ptr = jpeg_idct_16x16;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((16 << 8) + 8):
+      method_ptr = jpeg_idct_16x8;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((14 << 8) + 7):
+      method_ptr = jpeg_idct_14x7;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((12 << 8) + 6):
+      method_ptr = jpeg_idct_12x6;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((10 << 8) + 5):
+      method_ptr = jpeg_idct_10x5;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((8 << 8) + 4):
+      method_ptr = jpeg_idct_8x4;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((6 << 8) + 3):
+      method_ptr = jpeg_idct_6x3;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((4 << 8) + 2):
+      method_ptr = jpeg_idct_4x2;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((2 << 8) + 1):
+      method_ptr = jpeg_idct_2x1;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((8 << 8) + 16):
+      method_ptr = jpeg_idct_8x16;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((7 << 8) + 14):
+      method_ptr = jpeg_idct_7x14;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((6 << 8) + 12):
+      method_ptr = jpeg_idct_6x12;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((5 << 8) + 10):
+      method_ptr = jpeg_idct_5x10;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((4 << 8) + 8):
+      method_ptr = jpeg_idct_4x8;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((3 << 8) + 6):
+      method_ptr = jpeg_idct_3x6;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((2 << 8) + 4):
+      method_ptr = jpeg_idct_2x4;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((1 << 8) + 2):
+      method_ptr = jpeg_idct_1x2;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
       break;
 #endif
-    case DCTSIZE:
+    case ((DCTSIZE << 8) + DCTSIZE):
       switch (cinfo->dct_method) {
 #ifdef DCT_ISLOW_SUPPORTED
       case JDCT_ISLOW:
@@ -139,7 +252,8 @@ start_pass (j_decompress_ptr cinfo)
       }
       break;
     default:
-      ERREXIT1(cinfo, JERR_BAD_DCTSIZE, compptr->DCT_scaled_size);
+      ERREXIT2(cinfo, JERR_BAD_DCTSIZE,
+	       compptr->DCT_h_scaled_size, compptr->DCT_v_scaled_size);
       break;
     }
     idct->pub.inverse_DCT[ci] = method_ptr;
@@ -211,6 +325,7 @@ start_pass (j_decompress_ptr cinfo)
 	 * coefficients scaled by scalefactor[row]*scalefactor[col], where
 	 *   scalefactor[0] = 1
 	 *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
+	 * We apply a further scale factor of 1/8.
 	 */
 	FLOAT_MULT_TYPE * fmtbl = (FLOAT_MULT_TYPE *) compptr->dct_table;
 	int row, col;
@@ -224,7 +339,7 @@ start_pass (j_decompress_ptr cinfo)
 	  for (col = 0; col < DCTSIZE; col++) {
 	    fmtbl[i] = (FLOAT_MULT_TYPE)
 	      ((double) qtbl->quantval[i] *
-	       aanscalefactor[row] * aanscalefactor[col]);
+	       aanscalefactor[row] * aanscalefactor[col] * 0.125);
 	    i++;
 	  }
 	}
diff --git a/src/jpeg-8c/jdhuff.c b/src/jpeg-8c/jdhuff.c
new file mode 100644
index 00000000..06f92fe4
--- /dev/null
+++ b/src/jpeg-8c/jdhuff.c
@@ -0,0 +1,1541 @@
+/*
+ * jdhuff.c
+ *
+ * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Modified 2006-2009 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains Huffman entropy decoding routines.
+ * Both sequential and progressive modes are supported in this single module.
+ *
+ * Much of the complexity here has to do with supporting input suspension.
+ * If the data source module demands suspension, we want to be able to back
+ * up to the start of the current MCU.  To do this, we copy state variables
+ * into local working storage, and update them back to the permanent
+ * storage only upon successful completion of an MCU.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Derived data constructed for each Huffman table */
+
+#define HUFF_LOOKAHEAD	8	/* # of bits of lookahead */
+
+typedef struct {
+  /* Basic tables: (element [0] of each array is unused) */
+  INT32 maxcode[18];		/* largest code of length k (-1 if none) */
+  /* (maxcode[17] is a sentinel to ensure jpeg_huff_decode terminates) */
+  INT32 valoffset[17];		/* huffval[] offset for codes of length k */
+  /* valoffset[k] = huffval[] index of 1st symbol of code length k, less
+   * the smallest code of length k; so given a code of length k, the
+   * corresponding symbol is huffval[code + valoffset[k]]
+   */
+
+  /* Link to public Huffman table (needed only in jpeg_huff_decode) */
+  JHUFF_TBL *pub;
+
+  /* Lookahead tables: indexed by the next HUFF_LOOKAHEAD bits of
+   * the input data stream.  If the next Huffman code is no more
+   * than HUFF_LOOKAHEAD bits long, we can obtain its length and
+   * the corresponding symbol directly from these tables.
+   */
+  int look_nbits[1<<HUFF_LOOKAHEAD]; /* # bits, or 0 if too long */
+  UINT8 look_sym[1<<HUFF_LOOKAHEAD]; /* symbol, or unused */
+} d_derived_tbl;
+
+
+/*
+ * Fetching the next N bits from the input stream is a time-critical operation
+ * for the Huffman decoders.  We implement it with a combination of inline
+ * macros and out-of-line subroutines.  Note that N (the number of bits
+ * demanded at one time) never exceeds 15 for JPEG use.
+ *
+ * We read source bytes into get_buffer and dole out bits as needed.
+ * If get_buffer already contains enough bits, they are fetched in-line
+ * by the macros CHECK_BIT_BUFFER and GET_BITS.  When there aren't enough
+ * bits, jpeg_fill_bit_buffer is called; it will attempt to fill get_buffer
+ * as full as possible (not just to the number of bits needed; this
+ * prefetching reduces the overhead cost of calling jpeg_fill_bit_buffer).
+ * Note that jpeg_fill_bit_buffer may return FALSE to indicate suspension.
+ * On TRUE return, jpeg_fill_bit_buffer guarantees that get_buffer contains
+ * at least the requested number of bits --- dummy zeroes are inserted if
+ * necessary.
+ */
+
+typedef INT32 bit_buf_type;	/* type of bit-extraction buffer */
+#define BIT_BUF_SIZE  32	/* size of buffer in bits */
+
+/* If long is > 32 bits on your machine, and shifting/masking longs is
+ * reasonably fast, making bit_buf_type be long and setting BIT_BUF_SIZE
+ * appropriately should be a win.  Unfortunately we can't define the size
+ * with something like  #define BIT_BUF_SIZE (sizeof(bit_buf_type)*8)
+ * because not all machines measure sizeof in 8-bit bytes.
+ */
+
+typedef struct {		/* Bitreading state saved across MCUs */
+  bit_buf_type get_buffer;	/* current bit-extraction buffer */
+  int bits_left;		/* # of unused bits in it */
+} bitread_perm_state;
+
+typedef struct {		/* Bitreading working state within an MCU */
+  /* Current data source location */
+  /* We need a copy, rather than munging the original, in case of suspension */
+  const JOCTET * next_input_byte; /* => next byte to read from source */
+  size_t bytes_in_buffer;	/* # of bytes remaining in source buffer */
+  /* Bit input buffer --- note these values are kept in register variables,
+   * not in this struct, inside the inner loops.
+   */
+  bit_buf_type get_buffer;	/* current bit-extraction buffer */
+  int bits_left;		/* # of unused bits in it */
+  /* Pointer needed by jpeg_fill_bit_buffer. */
+  j_decompress_ptr cinfo;	/* back link to decompress master record */
+} bitread_working_state;
+
+/* Macros to declare and load/save bitread local variables. */
+#define BITREAD_STATE_VARS  \
+	register bit_buf_type get_buffer;  \
+	register int bits_left;  \
+	bitread_working_state br_state
+
+#define BITREAD_LOAD_STATE(cinfop,permstate)  \
+	br_state.cinfo = cinfop; \
+	br_state.next_input_byte = cinfop->src->next_input_byte; \
+	br_state.bytes_in_buffer = cinfop->src->bytes_in_buffer; \
+	get_buffer = permstate.get_buffer; \
+	bits_left = permstate.bits_left;
+
+#define BITREAD_SAVE_STATE(cinfop,permstate)  \
+	cinfop->src->next_input_byte = br_state.next_input_byte; \
+	cinfop->src->bytes_in_buffer = br_state.bytes_in_buffer; \
+	permstate.get_buffer = get_buffer; \
+	permstate.bits_left = bits_left
+
+/*
+ * These macros provide the in-line portion of bit fetching.
+ * Use CHECK_BIT_BUFFER to ensure there are N bits in get_buffer
+ * before using GET_BITS, PEEK_BITS, or DROP_BITS.
+ * The variables get_buffer and bits_left are assumed to be locals,
+ * but the state struct might not be (jpeg_huff_decode needs this).
+ *	CHECK_BIT_BUFFER(state,n,action);
+ *		Ensure there are N bits in get_buffer; if suspend, take action.
+ *      val = GET_BITS(n);
+ *		Fetch next N bits.
+ *      val = PEEK_BITS(n);
+ *		Fetch next N bits without removing them from the buffer.
+ *	DROP_BITS(n);
+ *		Discard next N bits.
+ * The value N should be a simple variable, not an expression, because it
+ * is evaluated multiple times.
+ */
+
+#define CHECK_BIT_BUFFER(state,nbits,action) \
+	{ if (bits_left < (nbits)) {  \
+	    if (! jpeg_fill_bit_buffer(&(state),get_buffer,bits_left,nbits))  \
+	      { action; }  \
+	    get_buffer = (state).get_buffer; bits_left = (state).bits_left; } }
+
+#define GET_BITS(nbits) \
+	(((int) (get_buffer >> (bits_left -= (nbits)))) & BIT_MASK(nbits))
+
+#define PEEK_BITS(nbits) \
+	(((int) (get_buffer >> (bits_left -  (nbits)))) & BIT_MASK(nbits))
+
+#define DROP_BITS(nbits) \
+	(bits_left -= (nbits))
+
+
+/*
+ * Code for extracting next Huffman-coded symbol from input bit stream.
+ * Again, this is time-critical and we make the main paths be macros.
+ *
+ * We use a lookahead table to process codes of up to HUFF_LOOKAHEAD bits
+ * without looping.  Usually, more than 95% of the Huffman codes will be 8
+ * or fewer bits long.  The few overlength codes are handled with a loop,
+ * which need not be inline code.
+ *
+ * Notes about the HUFF_DECODE macro:
+ * 1. Near the end of the data segment, we may fail to get enough bits
+ *    for a lookahead.  In that case, we do it the hard way.
+ * 2. If the lookahead table contains no entry, the next code must be
+ *    more than HUFF_LOOKAHEAD bits long.
+ * 3. jpeg_huff_decode returns -1 if forced to suspend.
+ */
+
+#define HUFF_DECODE(result,state,htbl,failaction,slowlabel) \
+{ register int nb, look; \
+  if (bits_left < HUFF_LOOKAHEAD) { \
+    if (! jpeg_fill_bit_buffer(&state,get_buffer,bits_left, 0)) {failaction;} \
+    get_buffer = state.get_buffer; bits_left = state.bits_left; \
+    if (bits_left < HUFF_LOOKAHEAD) { \
+      nb = 1; goto slowlabel; \
+    } \
+  } \
+  look = PEEK_BITS(HUFF_LOOKAHEAD); \
+  if ((nb = htbl->look_nbits[look]) != 0) { \
+    DROP_BITS(nb); \
+    result = htbl->look_sym[look]; \
+  } else { \
+    nb = HUFF_LOOKAHEAD+1; \
+slowlabel: \
+    if ((result=jpeg_huff_decode(&state,get_buffer,bits_left,htbl,nb)) < 0) \
+	{ failaction; } \
+    get_buffer = state.get_buffer; bits_left = state.bits_left; \
+  } \
+}
+
+
+/*
+ * Expanded entropy decoder object for Huffman decoding.
+ *
+ * The savable_state subrecord contains fields that change within an MCU,
+ * but must not be updated permanently until we complete the MCU.
+ */
+
+typedef struct {
+  unsigned int EOBRUN;			/* remaining EOBs in EOBRUN */
+  int last_dc_val[MAX_COMPS_IN_SCAN];	/* last DC coef for each component */
+} savable_state;
+
+/* This macro is to work around compilers with missing or broken
+ * structure assignment.  You'll need to fix this code if you have
+ * such a compiler and you change MAX_COMPS_IN_SCAN.
+ */
+
+#ifndef NO_STRUCT_ASSIGN
+#define ASSIGN_STATE(dest,src)  ((dest) = (src))
+#else
+#if MAX_COMPS_IN_SCAN == 4
+#define ASSIGN_STATE(dest,src)  \
+	((dest).EOBRUN = (src).EOBRUN, \
+	 (dest).last_dc_val[0] = (src).last_dc_val[0], \
+	 (dest).last_dc_val[1] = (src).last_dc_val[1], \
+	 (dest).last_dc_val[2] = (src).last_dc_val[2], \
+	 (dest).last_dc_val[3] = (src).last_dc_val[3])
+#endif
+#endif
+
+
+typedef struct {
+  struct jpeg_entropy_decoder pub; /* public fields */
+
+  /* These fields are loaded into local variables at start of each MCU.
+   * In case of suspension, we exit WITHOUT updating them.
+   */
+  bitread_perm_state bitstate;	/* Bit buffer at start of MCU */
+  savable_state saved;		/* Other state at start of MCU */
+
+  /* These fields are NOT loaded into local working state. */
+  boolean insufficient_data;	/* set TRUE after emitting warning */
+  unsigned int restarts_to_go;	/* MCUs left in this restart interval */
+
+  /* Following two fields used only in progressive mode */
+
+  /* Pointers to derived tables (these workspaces have image lifespan) */
+  d_derived_tbl * derived_tbls[NUM_HUFF_TBLS];
+
+  d_derived_tbl * ac_derived_tbl; /* active table during an AC scan */
+
+  /* Following fields used only in sequential mode */
+
+  /* Pointers to derived tables (these workspaces have image lifespan) */
+  d_derived_tbl * dc_derived_tbls[NUM_HUFF_TBLS];
+  d_derived_tbl * ac_derived_tbls[NUM_HUFF_TBLS];
+
+  /* Precalculated info set up by start_pass for use in decode_mcu: */
+
+  /* Pointers to derived tables to be used for each block within an MCU */
+  d_derived_tbl * dc_cur_tbls[D_MAX_BLOCKS_IN_MCU];
+  d_derived_tbl * ac_cur_tbls[D_MAX_BLOCKS_IN_MCU];
+  /* Whether we care about the DC and AC coefficient values for each block */
+  int coef_limit[D_MAX_BLOCKS_IN_MCU];
+} huff_entropy_decoder;
+
+typedef huff_entropy_decoder * huff_entropy_ptr;
+
+
+static const int jpeg_zigzag_order[8][8] = {
+  {  0,  1,  5,  6, 14, 15, 27, 28 },
+  {  2,  4,  7, 13, 16, 26, 29, 42 },
+  {  3,  8, 12, 17, 25, 30, 41, 43 },
+  {  9, 11, 18, 24, 31, 40, 44, 53 },
+  { 10, 19, 23, 32, 39, 45, 52, 54 },
+  { 20, 22, 33, 38, 46, 51, 55, 60 },
+  { 21, 34, 37, 47, 50, 56, 59, 61 },
+  { 35, 36, 48, 49, 57, 58, 62, 63 }
+};
+
+static const int jpeg_zigzag_order7[7][7] = {
+  {  0,  1,  5,  6, 14, 15, 27 },
+  {  2,  4,  7, 13, 16, 26, 28 },
+  {  3,  8, 12, 17, 25, 29, 38 },
+  {  9, 11, 18, 24, 30, 37, 39 },
+  { 10, 19, 23, 31, 36, 40, 45 },
+  { 20, 22, 32, 35, 41, 44, 46 },
+  { 21, 33, 34, 42, 43, 47, 48 }
+};
+
+static const int jpeg_zigzag_order6[6][6] = {
+  {  0,  1,  5,  6, 14, 15 },
+  {  2,  4,  7, 13, 16, 25 },
+  {  3,  8, 12, 17, 24, 26 },
+  {  9, 11, 18, 23, 27, 32 },
+  { 10, 19, 22, 28, 31, 33 },
+  { 20, 21, 29, 30, 34, 35 }
+};
+
+static const int jpeg_zigzag_order5[5][5] = {
+  {  0,  1,  5,  6, 14 },
+  {  2,  4,  7, 13, 15 },
+  {  3,  8, 12, 16, 21 },
+  {  9, 11, 17, 20, 22 },
+  { 10, 18, 19, 23, 24 }
+};
+
+static const int jpeg_zigzag_order4[4][4] = {
+  { 0,  1,  5,  6 },
+  { 2,  4,  7, 12 },
+  { 3,  8, 11, 13 },
+  { 9, 10, 14, 15 }
+};
+
+static const int jpeg_zigzag_order3[3][3] = {
+  { 0, 1, 5 },
+  { 2, 4, 6 },
+  { 3, 7, 8 }
+};
+
+static const int jpeg_zigzag_order2[2][2] = {
+  { 0, 1 },
+  { 2, 3 }
+};
+
+
+/*
+ * Compute the derived values for a Huffman table.
+ * This routine also performs some validation checks on the table.
+ */
+
+LOCAL(void)
+jpeg_make_d_derived_tbl (j_decompress_ptr cinfo, boolean isDC, int tblno,
+			 d_derived_tbl ** pdtbl)
+{
+  JHUFF_TBL *htbl;
+  d_derived_tbl *dtbl;
+  int p, i, l, si, numsymbols;
+  int lookbits, ctr;
+  char huffsize[257];
+  unsigned int huffcode[257];
+  unsigned int code;
+
+  /* Note that huffsize[] and huffcode[] are filled in code-length order,
+   * paralleling the order of the symbols themselves in htbl->huffval[].
+   */
+
+  /* Find the input Huffman table */
+  if (tblno < 0 || tblno >= NUM_HUFF_TBLS)
+    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
+  htbl =
+    isDC ? cinfo->dc_huff_tbl_ptrs[tblno] : cinfo->ac_huff_tbl_ptrs[tblno];
+  if (htbl == NULL)
+    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
+
+  /* Allocate a workspace if we haven't already done so. */
+  if (*pdtbl == NULL)
+    *pdtbl = (d_derived_tbl *)
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				  SIZEOF(d_derived_tbl));
+  dtbl = *pdtbl;
+  dtbl->pub = htbl;		/* fill in back link */
+  
+  /* Figure C.1: make table of Huffman code length for each symbol */
+
+  p = 0;
+  for (l = 1; l <= 16; l++) {
+    i = (int) htbl->bits[l];
+    if (i < 0 || p + i > 256)	/* protect against table overrun */
+      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
+    while (i--)
+      huffsize[p++] = (char) l;
+  }
+  huffsize[p] = 0;
+  numsymbols = p;
+  
+  /* Figure C.2: generate the codes themselves */
+  /* We also validate that the counts represent a legal Huffman code tree. */
+  
+  code = 0;
+  si = huffsize[0];
+  p = 0;
+  while (huffsize[p]) {
+    while (((int) huffsize[p]) == si) {
+      huffcode[p++] = code;
+      code++;
+    }
+    /* code is now 1 more than the last code used for codelength si; but
+     * it must still fit in si bits, since no code is allowed to be all ones.
+     */
+    if (((INT32) code) >= (((INT32) 1) << si))
+      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
+    code <<= 1;
+    si++;
+  }
+
+  /* Figure F.15: generate decoding tables for bit-sequential decoding */
+
+  p = 0;
+  for (l = 1; l <= 16; l++) {
+    if (htbl->bits[l]) {
+      /* valoffset[l] = huffval[] index of 1st symbol of code length l,
+       * minus the minimum code of length l
+       */
+      dtbl->valoffset[l] = (INT32) p - (INT32) huffcode[p];
+      p += htbl->bits[l];
+      dtbl->maxcode[l] = huffcode[p-1]; /* maximum code of length l */
+    } else {
+      dtbl->maxcode[l] = -1;	/* -1 if no codes of this length */
+    }
+  }
+  dtbl->maxcode[17] = 0xFFFFFL; /* ensures jpeg_huff_decode terminates */
+
+  /* Compute lookahead tables to speed up decoding.
+   * First we set all the table entries to 0, indicating "too long";
+   * then we iterate through the Huffman codes that are short enough and
+   * fill in all the entries that correspond to bit sequences starting
+   * with that code.
+   */
+
+  MEMZERO(dtbl->look_nbits, SIZEOF(dtbl->look_nbits));
+
+  p = 0;
+  for (l = 1; l <= HUFF_LOOKAHEAD; l++) {
+    for (i = 1; i <= (int) htbl->bits[l]; i++, p++) {
+      /* l = current code's length, p = its index in huffcode[] & huffval[]. */
+      /* Generate left-justified code followed by all possible bit sequences */
+      lookbits = huffcode[p] << (HUFF_LOOKAHEAD-l);
+      for (ctr = 1 << (HUFF_LOOKAHEAD-l); ctr > 0; ctr--) {
+	dtbl->look_nbits[lookbits] = l;
+	dtbl->look_sym[lookbits] = htbl->huffval[p];
+	lookbits++;
+      }
+    }
+  }
+
+  /* Validate symbols as being reasonable.
+   * For AC tables, we make no check, but accept all byte values 0..255.
+   * For DC tables, we require the symbols to be in range 0..15.
+   * (Tighter bounds could be applied depending on the data depth and mode,
+   * but this is sufficient to ensure safe decoding.)
+   */
+  if (isDC) {
+    for (i = 0; i < numsymbols; i++) {
+      int sym = htbl->huffval[i];
+      if (sym < 0 || sym > 15)
+	ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
+    }
+  }
+}
+
+
+/*
+ * Out-of-line code for bit fetching.
+ * Note: current values of get_buffer and bits_left are passed as parameters,
+ * but are returned in the corresponding fields of the state struct.
+ *
+ * On most machines MIN_GET_BITS should be 25 to allow the full 32-bit width
+ * of get_buffer to be used.  (On machines with wider words, an even larger
+ * buffer could be used.)  However, on some machines 32-bit shifts are
+ * quite slow and take time proportional to the number of places shifted.
+ * (This is true with most PC compilers, for instance.)  In this case it may
+ * be a win to set MIN_GET_BITS to the minimum value of 15.  This reduces the
+ * average shift distance at the cost of more calls to jpeg_fill_bit_buffer.
+ */
+
+#ifdef SLOW_SHIFT_32
+#define MIN_GET_BITS  15	/* minimum allowable value */
+#else
+#define MIN_GET_BITS  (BIT_BUF_SIZE-7)
+#endif
+
+
+LOCAL(boolean)
+jpeg_fill_bit_buffer (bitread_working_state * state,
+		      register bit_buf_type get_buffer, register int bits_left,
+		      int nbits)
+/* Load up the bit buffer to a depth of at least nbits */
+{
+  /* Copy heavily used state fields into locals (hopefully registers) */
+  register const JOCTET * next_input_byte = state->next_input_byte;
+  register size_t bytes_in_buffer = state->bytes_in_buffer;
+  j_decompress_ptr cinfo = state->cinfo;
+
+  /* Attempt to load at least MIN_GET_BITS bits into get_buffer. */
+  /* (It is assumed that no request will be for more than that many bits.) */
+  /* We fail to do so only if we hit a marker or are forced to suspend. */
+
+  if (cinfo->unread_marker == 0) {	/* cannot advance past a marker */
+    while (bits_left < MIN_GET_BITS) {
+      register int c;
+
+      /* Attempt to read a byte */
+      if (bytes_in_buffer == 0) {
+	if (! (*cinfo->src->fill_input_buffer) (cinfo))
+	  return FALSE;
+	next_input_byte = cinfo->src->next_input_byte;
+	bytes_in_buffer = cinfo->src->bytes_in_buffer;
+      }
+      bytes_in_buffer--;
+      c = GETJOCTET(*next_input_byte++);
+
+      /* If it's 0xFF, check and discard stuffed zero byte */
+      if (c == 0xFF) {
+	/* Loop here to discard any padding FF's on terminating marker,
+	 * so that we can save a valid unread_marker value.  NOTE: we will
+	 * accept multiple FF's followed by a 0 as meaning a single FF data
+	 * byte.  This data pattern is not valid according to the standard.
+	 */
+	do {
+	  if (bytes_in_buffer == 0) {
+	    if (! (*cinfo->src->fill_input_buffer) (cinfo))
+	      return FALSE;
+	    next_input_byte = cinfo->src->next_input_byte;
+	    bytes_in_buffer = cinfo->src->bytes_in_buffer;
+	  }
+	  bytes_in_buffer--;
+	  c = GETJOCTET(*next_input_byte++);
+	} while (c == 0xFF);
+
+	if (c == 0) {
+	  /* Found FF/00, which represents an FF data byte */
+	  c = 0xFF;
+	} else {
+	  /* Oops, it's actually a marker indicating end of compressed data.
+	   * Save the marker code for later use.
+	   * Fine point: it might appear that we should save the marker into
+	   * bitread working state, not straight into permanent state.  But
+	   * once we have hit a marker, we cannot need to suspend within the
+	   * current MCU, because we will read no more bytes from the data
+	   * source.  So it is OK to update permanent state right away.
+	   */
+	  cinfo->unread_marker = c;
+	  /* See if we need to insert some fake zero bits. */
+	  goto no_more_bytes;
+	}
+      }
+
+      /* OK, load c into get_buffer */
+      get_buffer = (get_buffer << 8) | c;
+      bits_left += 8;
+    } /* end while */
+  } else {
+  no_more_bytes:
+    /* We get here if we've read the marker that terminates the compressed
+     * data segment.  There should be enough bits in the buffer register
+     * to satisfy the request; if so, no problem.
+     */
+    if (nbits > bits_left) {
+      /* Uh-oh.  Report corrupted data to user and stuff zeroes into
+       * the data stream, so that we can produce some kind of image.
+       * We use a nonvolatile flag to ensure that only one warning message
+       * appears per data segment.
+       */
+      if (! ((huff_entropy_ptr) cinfo->entropy)->insufficient_data) {
+	WARNMS(cinfo, JWRN_HIT_MARKER);
+	((huff_entropy_ptr) cinfo->entropy)->insufficient_data = TRUE;
+      }
+      /* Fill the buffer with zero bits */
+      get_buffer <<= MIN_GET_BITS - bits_left;
+      bits_left = MIN_GET_BITS;
+    }
+  }
+
+  /* Unload the local registers */
+  state->next_input_byte = next_input_byte;
+  state->bytes_in_buffer = bytes_in_buffer;
+  state->get_buffer = get_buffer;
+  state->bits_left = bits_left;
+
+  return TRUE;
+}
+
+
+/*
+ * Figure F.12: extend sign bit.
+ * On some machines, a shift and sub will be faster than a table lookup.
+ */
+
+#ifdef AVOID_TABLES
+
+#define BIT_MASK(nbits)   ((1<<(nbits))-1)
+#define HUFF_EXTEND(x,s)  ((x) < (1<<((s)-1)) ? (x) - ((1<<(s))-1) : (x))
+
+#else
+
+#define BIT_MASK(nbits)   bmask[nbits]
+#define HUFF_EXTEND(x,s)  ((x) <= bmask[(s) - 1] ? (x) - bmask[s] : (x))
+
+static const int bmask[16] =	/* bmask[n] is mask for n rightmost bits */
+  { 0, 0x0001, 0x0003, 0x0007, 0x000F, 0x001F, 0x003F, 0x007F, 0x00FF,
+    0x01FF, 0x03FF, 0x07FF, 0x0FFF, 0x1FFF, 0x3FFF, 0x7FFF };
+
+#endif /* AVOID_TABLES */
+
+
+/*
+ * Out-of-line code for Huffman code decoding.
+ */
+
+LOCAL(int)
+jpeg_huff_decode (bitread_working_state * state,
+		  register bit_buf_type get_buffer, register int bits_left,
+		  d_derived_tbl * htbl, int min_bits)
+{
+  register int l = min_bits;
+  register INT32 code;
+
+  /* HUFF_DECODE has determined that the code is at least min_bits */
+  /* bits long, so fetch that many bits in one swoop. */
+
+  CHECK_BIT_BUFFER(*state, l, return -1);
+  code = GET_BITS(l);
+
+  /* Collect the rest of the Huffman code one bit at a time. */
+  /* This is per Figure F.16 in the JPEG spec. */
+
+  while (code > htbl->maxcode[l]) {
+    code <<= 1;
+    CHECK_BIT_BUFFER(*state, 1, return -1);
+    code |= GET_BITS(1);
+    l++;
+  }
+
+  /* Unload the local registers */
+  state->get_buffer = get_buffer;
+  state->bits_left = bits_left;
+
+  /* With garbage input we may reach the sentinel value l = 17. */
+
+  if (l > 16) {
+    WARNMS(state->cinfo, JWRN_HUFF_BAD_CODE);
+    return 0;			/* fake a zero as the safest result */
+  }
+
+  return htbl->pub->huffval[ (int) (code + htbl->valoffset[l]) ];
+}
+
+
+/*
+ * Check for a restart marker & resynchronize decoder.
+ * Returns FALSE if must suspend.
+ */
+
+LOCAL(boolean)
+process_restart (j_decompress_ptr cinfo)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int ci;
+
+  /* Throw away any unused bits remaining in bit buffer; */
+  /* include any full bytes in next_marker's count of discarded bytes */
+  cinfo->marker->discarded_bytes += entropy->bitstate.bits_left / 8;
+  entropy->bitstate.bits_left = 0;
+
+  /* Advance past the RSTn marker */
+  if (! (*cinfo->marker->read_restart_marker) (cinfo))
+    return FALSE;
+
+  /* Re-initialize DC predictions to 0 */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++)
+    entropy->saved.last_dc_val[ci] = 0;
+  /* Re-init EOB run count, too */
+  entropy->saved.EOBRUN = 0;
+
+  /* Reset restart counter */
+  entropy->restarts_to_go = cinfo->restart_interval;
+
+  /* Reset out-of-data flag, unless read_restart_marker left us smack up
+   * against a marker.  In that case we will end up treating the next data
+   * segment as empty, and we can avoid producing bogus output pixels by
+   * leaving the flag set.
+   */
+  if (cinfo->unread_marker == 0)
+    entropy->insufficient_data = FALSE;
+
+  return TRUE;
+}
+
+
+/*
+ * Huffman MCU decoding.
+ * Each of these routines decodes and returns one MCU's worth of
+ * Huffman-compressed coefficients. 
+ * The coefficients are reordered from zigzag order into natural array order,
+ * but are not dequantized.
+ *
+ * The i'th block of the MCU is stored into the block pointed to by
+ * MCU_data[i].  WE ASSUME THIS AREA IS INITIALLY ZEROED BY THE CALLER.
+ * (Wholesale zeroing is usually a little faster than retail...)
+ *
+ * We return FALSE if data source requested suspension.  In that case no
+ * changes have been made to permanent state.  (Exception: some output
+ * coefficients may already have been assigned.  This is harmless for
+ * spectral selection, since we'll just re-assign them on the next call.
+ * Successive approximation AC refinement has to be more careful, however.)
+ */
+
+/*
+ * MCU decoding for DC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+decode_mcu_DC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{   
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int Al = cinfo->Al;
+  register int s, r;
+  int blkn, ci;
+  JBLOCKROW block;
+  BITREAD_STATE_VARS;
+  savable_state state;
+  d_derived_tbl * tbl;
+  jpeg_component_info * compptr;
+
+  /* Process restart marker if needed; may have to suspend */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      if (! process_restart(cinfo))
+	return FALSE;
+  }
+
+  /* If we've run out of data, just leave the MCU set to zeroes.
+   * This way, we return uniform gray for the remainder of the segment.
+   */
+  if (! entropy->insufficient_data) {
+
+    /* Load up working state */
+    BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+    ASSIGN_STATE(state, entropy->saved);
+
+    /* Outer loop handles each block in the MCU */
+
+    for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+      block = MCU_data[blkn];
+      ci = cinfo->MCU_membership[blkn];
+      compptr = cinfo->cur_comp_info[ci];
+      tbl = entropy->derived_tbls[compptr->dc_tbl_no];
+
+      /* Decode a single block's worth of coefficients */
+
+      /* Section F.2.2.1: decode the DC coefficient difference */
+      HUFF_DECODE(s, br_state, tbl, return FALSE, label1);
+      if (s) {
+	CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	r = GET_BITS(s);
+	s = HUFF_EXTEND(r, s);
+      }
+
+      /* Convert DC difference to actual value, update last_dc_val */
+      s += state.last_dc_val[ci];
+      state.last_dc_val[ci] = s;
+      /* Scale and output the coefficient (assumes jpeg_natural_order[0]=0) */
+      (*block)[0] = (JCOEF) (s << Al);
+    }
+
+    /* Completed MCU, so update state */
+    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+    ASSIGN_STATE(entropy->saved, state);
+  }
+
+  /* Account for restart interval (no-op if not using restarts) */
+  entropy->restarts_to_go--;
+
+  return TRUE;
+}
+
+
+/*
+ * MCU decoding for AC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+decode_mcu_AC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{   
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  register int s, k, r;
+  unsigned int EOBRUN;
+  int Se, Al;
+  const int * natural_order;
+  JBLOCKROW block;
+  BITREAD_STATE_VARS;
+  d_derived_tbl * tbl;
+
+  /* Process restart marker if needed; may have to suspend */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      if (! process_restart(cinfo))
+	return FALSE;
+  }
+
+  /* If we've run out of data, just leave the MCU set to zeroes.
+   * This way, we return uniform gray for the remainder of the segment.
+   */
+  if (! entropy->insufficient_data) {
+
+    Se = cinfo->Se;
+    Al = cinfo->Al;
+    natural_order = cinfo->natural_order;
+
+    /* Load up working state.
+     * We can avoid loading/saving bitread state if in an EOB run.
+     */
+    EOBRUN = entropy->saved.EOBRUN;	/* only part of saved state we need */
+
+    /* There is always only one block per MCU */
+
+    if (EOBRUN > 0)		/* if it's a band of zeroes... */
+      EOBRUN--;			/* ...process it now (we do nothing) */
+    else {
+      BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+      block = MCU_data[0];
+      tbl = entropy->ac_derived_tbl;
+
+      for (k = cinfo->Ss; k <= Se; k++) {
+	HUFF_DECODE(s, br_state, tbl, return FALSE, label2);
+	r = s >> 4;
+	s &= 15;
+	if (s) {
+	  k += r;
+	  CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	  r = GET_BITS(s);
+	  s = HUFF_EXTEND(r, s);
+	  /* Scale and output coefficient in natural (dezigzagged) order */
+	  (*block)[natural_order[k]] = (JCOEF) (s << Al);
+	} else {
+	  if (r == 15) {	/* ZRL */
+	    k += 15;		/* skip 15 zeroes in band */
+	  } else {		/* EOBr, run length is 2^r + appended bits */
+	    EOBRUN = 1 << r;
+	    if (r) {		/* EOBr, r > 0 */
+	      CHECK_BIT_BUFFER(br_state, r, return FALSE);
+	      r = GET_BITS(r);
+	      EOBRUN += r;
+	    }
+	    EOBRUN--;		/* this band is processed at this moment */
+	    break;		/* force end-of-band */
+	  }
+	}
+      }
+
+      BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+    }
+
+    /* Completed MCU, so update state */
+    entropy->saved.EOBRUN = EOBRUN;	/* only part of saved state we need */
+  }
+
+  /* Account for restart interval (no-op if not using restarts) */
+  entropy->restarts_to_go--;
+
+  return TRUE;
+}
+
+
+/*
+ * MCU decoding for DC successive approximation refinement scan.
+ * Note: we assume such scans can be multi-component, although the spec
+ * is not very clear on the point.
+ */
+
+METHODDEF(boolean)
+decode_mcu_DC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{   
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int p1 = 1 << cinfo->Al;	/* 1 in the bit position being coded */
+  int blkn;
+  JBLOCKROW block;
+  BITREAD_STATE_VARS;
+
+  /* Process restart marker if needed; may have to suspend */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      if (! process_restart(cinfo))
+	return FALSE;
+  }
+
+  /* Not worth the cycles to check insufficient_data here,
+   * since we will not change the data anyway if we read zeroes.
+   */
+
+  /* Load up working state */
+  BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+
+  /* Outer loop handles each block in the MCU */
+
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+
+    /* Encoded data is simply the next bit of the two's-complement DC value */
+    CHECK_BIT_BUFFER(br_state, 1, return FALSE);
+    if (GET_BITS(1))
+      (*block)[0] |= p1;
+    /* Note: since we use |=, repeating the assignment later is safe */
+  }
+
+  /* Completed MCU, so update state */
+  BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+
+  /* Account for restart interval (no-op if not using restarts) */
+  entropy->restarts_to_go--;
+
+  return TRUE;
+}
+
+
+/*
+ * MCU decoding for AC successive approximation refinement scan.
+ */
+
+METHODDEF(boolean)
+decode_mcu_AC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{   
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  register int s, k, r;
+  unsigned int EOBRUN;
+  int Se, p1, m1;
+  const int * natural_order;
+  JBLOCKROW block;
+  JCOEFPTR thiscoef;
+  BITREAD_STATE_VARS;
+  d_derived_tbl * tbl;
+  int num_newnz;
+  int newnz_pos[DCTSIZE2];
+
+  /* Process restart marker if needed; may have to suspend */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      if (! process_restart(cinfo))
+	return FALSE;
+  }
+
+  /* If we've run out of data, don't modify the MCU.
+   */
+  if (! entropy->insufficient_data) {
+
+    Se = cinfo->Se;
+    p1 = 1 << cinfo->Al;	/* 1 in the bit position being coded */
+    m1 = (-1) << cinfo->Al;	/* -1 in the bit position being coded */
+    natural_order = cinfo->natural_order;
+
+    /* Load up working state */
+    BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+    EOBRUN = entropy->saved.EOBRUN; /* only part of saved state we need */
+
+    /* There is always only one block per MCU */
+    block = MCU_data[0];
+    tbl = entropy->ac_derived_tbl;
+
+    /* If we are forced to suspend, we must undo the assignments to any newly
+     * nonzero coefficients in the block, because otherwise we'd get confused
+     * next time about which coefficients were already nonzero.
+     * But we need not undo addition of bits to already-nonzero coefficients;
+     * instead, we can test the current bit to see if we already did it.
+     */
+    num_newnz = 0;
+
+    /* initialize coefficient loop counter to start of band */
+    k = cinfo->Ss;
+
+    if (EOBRUN == 0) {
+      for (; k <= Se; k++) {
+	HUFF_DECODE(s, br_state, tbl, goto undoit, label3);
+	r = s >> 4;
+	s &= 15;
+	if (s) {
+	  if (s != 1)		/* size of new coef should always be 1 */
+	    WARNMS(cinfo, JWRN_HUFF_BAD_CODE);
+	  CHECK_BIT_BUFFER(br_state, 1, goto undoit);
+	  if (GET_BITS(1))
+	    s = p1;		/* newly nonzero coef is positive */
+	  else
+	    s = m1;		/* newly nonzero coef is negative */
+	} else {
+	  if (r != 15) {
+	    EOBRUN = 1 << r;	/* EOBr, run length is 2^r + appended bits */
+	    if (r) {
+	      CHECK_BIT_BUFFER(br_state, r, goto undoit);
+	      r = GET_BITS(r);
+	      EOBRUN += r;
+	    }
+	    break;		/* rest of block is handled by EOB logic */
+	  }
+	  /* note s = 0 for processing ZRL */
+	}
+	/* Advance over already-nonzero coefs and r still-zero coefs,
+	 * appending correction bits to the nonzeroes.  A correction bit is 1
+	 * if the absolute value of the coefficient must be increased.
+	 */
+	do {
+	  thiscoef = *block + natural_order[k];
+	  if (*thiscoef != 0) {
+	    CHECK_BIT_BUFFER(br_state, 1, goto undoit);
+	    if (GET_BITS(1)) {
+	      if ((*thiscoef & p1) == 0) { /* do nothing if already set it */
+		if (*thiscoef >= 0)
+		  *thiscoef += p1;
+		else
+		  *thiscoef += m1;
+	      }
+	    }
+	  } else {
+	    if (--r < 0)
+	      break;		/* reached target zero coefficient */
+	  }
+	  k++;
+	} while (k <= Se);
+	if (s) {
+	  int pos = natural_order[k];
+	  /* Output newly nonzero coefficient */
+	  (*block)[pos] = (JCOEF) s;
+	  /* Remember its position in case we have to suspend */
+	  newnz_pos[num_newnz++] = pos;
+	}
+      }
+    }
+
+    if (EOBRUN > 0) {
+      /* Scan any remaining coefficient positions after the end-of-band
+       * (the last newly nonzero coefficient, if any).  Append a correction
+       * bit to each already-nonzero coefficient.  A correction bit is 1
+       * if the absolute value of the coefficient must be increased.
+       */
+      for (; k <= Se; k++) {
+	thiscoef = *block + natural_order[k];
+	if (*thiscoef != 0) {
+	  CHECK_BIT_BUFFER(br_state, 1, goto undoit);
+	  if (GET_BITS(1)) {
+	    if ((*thiscoef & p1) == 0) { /* do nothing if already changed it */
+	      if (*thiscoef >= 0)
+		*thiscoef += p1;
+	      else
+		*thiscoef += m1;
+	    }
+	  }
+	}
+      }
+      /* Count one block completed in EOB run */
+      EOBRUN--;
+    }
+
+    /* Completed MCU, so update state */
+    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+    entropy->saved.EOBRUN = EOBRUN; /* only part of saved state we need */
+  }
+
+  /* Account for restart interval (no-op if not using restarts) */
+  entropy->restarts_to_go--;
+
+  return TRUE;
+
+undoit:
+  /* Re-zero any output coefficients that we made newly nonzero */
+  while (num_newnz > 0)
+    (*block)[newnz_pos[--num_newnz]] = 0;
+
+  return FALSE;
+}
+
+
+/*
+ * Decode one MCU's worth of Huffman-compressed coefficients,
+ * partial blocks.
+ */
+
+METHODDEF(boolean)
+decode_mcu_sub (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  const int * natural_order;
+  int Se, blkn;
+  BITREAD_STATE_VARS;
+  savable_state state;
+
+  /* Process restart marker if needed; may have to suspend */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      if (! process_restart(cinfo))
+	return FALSE;
+  }
+
+  /* If we've run out of data, just leave the MCU set to zeroes.
+   * This way, we return uniform gray for the remainder of the segment.
+   */
+  if (! entropy->insufficient_data) {
+
+    natural_order = cinfo->natural_order;
+    Se = cinfo->lim_Se;
+
+    /* Load up working state */
+    BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+    ASSIGN_STATE(state, entropy->saved);
+
+    /* Outer loop handles each block in the MCU */
+
+    for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+      JBLOCKROW block = MCU_data[blkn];
+      d_derived_tbl * htbl;
+      register int s, k, r;
+      int coef_limit, ci;
+
+      /* Decode a single block's worth of coefficients */
+
+      /* Section F.2.2.1: decode the DC coefficient difference */
+      htbl = entropy->dc_cur_tbls[blkn];
+      HUFF_DECODE(s, br_state, htbl, return FALSE, label1);
+
+      htbl = entropy->ac_cur_tbls[blkn];
+      k = 1;
+      coef_limit = entropy->coef_limit[blkn];
+      if (coef_limit) {
+	/* Convert DC difference to actual value, update last_dc_val */
+	if (s) {
+	  CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	  r = GET_BITS(s);
+	  s = HUFF_EXTEND(r, s);
+	}
+	ci = cinfo->MCU_membership[blkn];
+	s += state.last_dc_val[ci];
+	state.last_dc_val[ci] = s;
+	/* Output the DC coefficient */
+	(*block)[0] = (JCOEF) s;
+
+	/* Section F.2.2.2: decode the AC coefficients */
+	/* Since zeroes are skipped, output area must be cleared beforehand */
+	for (; k < coef_limit; k++) {
+	  HUFF_DECODE(s, br_state, htbl, return FALSE, label2);
+
+	  r = s >> 4;
+	  s &= 15;
+
+	  if (s) {
+	    k += r;
+	    CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	    r = GET_BITS(s);
+	    s = HUFF_EXTEND(r, s);
+	    /* Output coefficient in natural (dezigzagged) order.
+	     * Note: the extra entries in natural_order[] will save us
+	     * if k > Se, which could happen if the data is corrupted.
+	     */
+	    (*block)[natural_order[k]] = (JCOEF) s;
+	  } else {
+	    if (r != 15)
+	      goto EndOfBlock;
+	    k += 15;
+	  }
+	}
+      } else {
+	if (s) {
+	  CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	  DROP_BITS(s);
+	}
+      }
+
+      /* Section F.2.2.2: decode the AC coefficients */
+      /* In this path we just discard the values */
+      for (; k <= Se; k++) {
+	HUFF_DECODE(s, br_state, htbl, return FALSE, label3);
+
+	r = s >> 4;
+	s &= 15;
+
+	if (s) {
+	  k += r;
+	  CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	  DROP_BITS(s);
+	} else {
+	  if (r != 15)
+	    break;
+	  k += 15;
+	}
+      }
+
+      EndOfBlock: ;
+    }
+
+    /* Completed MCU, so update state */
+    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+    ASSIGN_STATE(entropy->saved, state);
+  }
+
+  /* Account for restart interval (no-op if not using restarts) */
+  entropy->restarts_to_go--;
+
+  return TRUE;
+}
+
+
+/*
+ * Decode one MCU's worth of Huffman-compressed coefficients,
+ * full-size blocks.
+ */
+
+METHODDEF(boolean)
+decode_mcu (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int blkn;
+  BITREAD_STATE_VARS;
+  savable_state state;
+
+  /* Process restart marker if needed; may have to suspend */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      if (! process_restart(cinfo))
+	return FALSE;
+  }
+
+  /* If we've run out of data, just leave the MCU set to zeroes.
+   * This way, we return uniform gray for the remainder of the segment.
+   */
+  if (! entropy->insufficient_data) {
+
+    /* Load up working state */
+    BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+    ASSIGN_STATE(state, entropy->saved);
+
+    /* Outer loop handles each block in the MCU */
+
+    for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+      JBLOCKROW block = MCU_data[blkn];
+      d_derived_tbl * htbl;
+      register int s, k, r;
+      int coef_limit, ci;
+
+      /* Decode a single block's worth of coefficients */
+
+      /* Section F.2.2.1: decode the DC coefficient difference */
+      htbl = entropy->dc_cur_tbls[blkn];
+      HUFF_DECODE(s, br_state, htbl, return FALSE, label1);
+
+      htbl = entropy->ac_cur_tbls[blkn];
+      k = 1;
+      coef_limit = entropy->coef_limit[blkn];
+      if (coef_limit) {
+	/* Convert DC difference to actual value, update last_dc_val */
+	if (s) {
+	  CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	  r = GET_BITS(s);
+	  s = HUFF_EXTEND(r, s);
+	}
+	ci = cinfo->MCU_membership[blkn];
+	s += state.last_dc_val[ci];
+	state.last_dc_val[ci] = s;
+	/* Output the DC coefficient */
+	(*block)[0] = (JCOEF) s;
+
+	/* Section F.2.2.2: decode the AC coefficients */
+	/* Since zeroes are skipped, output area must be cleared beforehand */
+	for (; k < coef_limit; k++) {
+	  HUFF_DECODE(s, br_state, htbl, return FALSE, label2);
+
+	  r = s >> 4;
+	  s &= 15;
+
+	  if (s) {
+	    k += r;
+	    CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	    r = GET_BITS(s);
+	    s = HUFF_EXTEND(r, s);
+	    /* Output coefficient in natural (dezigzagged) order.
+	     * Note: the extra entries in jpeg_natural_order[] will save us
+	     * if k >= DCTSIZE2, which could happen if the data is corrupted.
+	     */
+	    (*block)[jpeg_natural_order[k]] = (JCOEF) s;
+	  } else {
+	    if (r != 15)
+	      goto EndOfBlock;
+	    k += 15;
+	  }
+	}
+      } else {
+	if (s) {
+	  CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	  DROP_BITS(s);
+	}
+      }
+
+      /* Section F.2.2.2: decode the AC coefficients */
+      /* In this path we just discard the values */
+      for (; k < DCTSIZE2; k++) {
+	HUFF_DECODE(s, br_state, htbl, return FALSE, label3);
+
+	r = s >> 4;
+	s &= 15;
+
+	if (s) {
+	  k += r;
+	  CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	  DROP_BITS(s);
+	} else {
+	  if (r != 15)
+	    break;
+	  k += 15;
+	}
+      }
+
+      EndOfBlock: ;
+    }
+
+    /* Completed MCU, so update state */
+    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+    ASSIGN_STATE(entropy->saved, state);
+  }
+
+  /* Account for restart interval (no-op if not using restarts) */
+  entropy->restarts_to_go--;
+
+  return TRUE;
+}
+
+
+/*
+ * Initialize for a Huffman-compressed scan.
+ */
+
+METHODDEF(void)
+start_pass_huff_decoder (j_decompress_ptr cinfo)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int ci, blkn, tbl, i;
+  jpeg_component_info * compptr;
+
+  if (cinfo->progressive_mode) {
+    /* Validate progressive scan parameters */
+    if (cinfo->Ss == 0) {
+      if (cinfo->Se != 0)
+	goto bad;
+    } else {
+      /* need not check Ss/Se < 0 since they came from unsigned bytes */
+      if (cinfo->Se < cinfo->Ss || cinfo->Se > cinfo->lim_Se)
+	goto bad;
+      /* AC scans may have only one component */
+      if (cinfo->comps_in_scan != 1)
+	goto bad;
+    }
+    if (cinfo->Ah != 0) {
+      /* Successive approximation refinement scan: must have Al = Ah-1. */
+      if (cinfo->Ah-1 != cinfo->Al)
+	goto bad;
+    }
+    if (cinfo->Al > 13) {	/* need not check for < 0 */
+      /* Arguably the maximum Al value should be less than 13 for 8-bit precision,
+       * but the spec doesn't say so, and we try to be liberal about what we
+       * accept.  Note: large Al values could result in out-of-range DC
+       * coefficients during early scans, leading to bizarre displays due to
+       * overflows in the IDCT math.  But we won't crash.
+       */
+      bad:
+      ERREXIT4(cinfo, JERR_BAD_PROGRESSION,
+	       cinfo->Ss, cinfo->Se, cinfo->Ah, cinfo->Al);
+    }
+    /* Update progression status, and verify that scan order is legal.
+     * Note that inter-scan inconsistencies are treated as warnings
+     * not fatal errors ... not clear if this is right way to behave.
+     */
+    for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+      int coefi, cindex = cinfo->cur_comp_info[ci]->component_index;
+      int *coef_bit_ptr = & cinfo->coef_bits[cindex][0];
+      if (cinfo->Ss && coef_bit_ptr[0] < 0) /* AC without prior DC scan */
+	WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, 0);
+      for (coefi = cinfo->Ss; coefi <= cinfo->Se; coefi++) {
+	int expected = (coef_bit_ptr[coefi] < 0) ? 0 : coef_bit_ptr[coefi];
+	if (cinfo->Ah != expected)
+	  WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, coefi);
+	coef_bit_ptr[coefi] = cinfo->Al;
+      }
+    }
+
+    /* Select MCU decoding routine */
+    if (cinfo->Ah == 0) {
+      if (cinfo->Ss == 0)
+	entropy->pub.decode_mcu = decode_mcu_DC_first;
+      else
+	entropy->pub.decode_mcu = decode_mcu_AC_first;
+    } else {
+      if (cinfo->Ss == 0)
+	entropy->pub.decode_mcu = decode_mcu_DC_refine;
+      else
+	entropy->pub.decode_mcu = decode_mcu_AC_refine;
+    }
+
+    for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+      compptr = cinfo->cur_comp_info[ci];
+      /* Make sure requested tables are present, and compute derived tables.
+       * We may build same derived table more than once, but it's not expensive.
+       */
+      if (cinfo->Ss == 0) {
+	if (cinfo->Ah == 0) {	/* DC refinement needs no table */
+	  tbl = compptr->dc_tbl_no;
+	  jpeg_make_d_derived_tbl(cinfo, TRUE, tbl,
+				  & entropy->derived_tbls[tbl]);
+	}
+      } else {
+	tbl = compptr->ac_tbl_no;
+	jpeg_make_d_derived_tbl(cinfo, FALSE, tbl,
+				& entropy->derived_tbls[tbl]);
+	/* remember the single active table */
+	entropy->ac_derived_tbl = entropy->derived_tbls[tbl];
+      }
+      /* Initialize DC predictions to 0 */
+      entropy->saved.last_dc_val[ci] = 0;
+    }
+
+    /* Initialize private state variables */
+    entropy->saved.EOBRUN = 0;
+  } else {
+    /* Check that the scan parameters Ss, Se, Ah/Al are OK for sequential JPEG.
+     * This ought to be an error condition, but we make it a warning because
+     * there are some baseline files out there with all zeroes in these bytes.
+     */
+    if (cinfo->Ss != 0 || cinfo->Ah != 0 || cinfo->Al != 0 ||
+	((cinfo->is_baseline || cinfo->Se < DCTSIZE2) &&
+	cinfo->Se != cinfo->lim_Se))
+      WARNMS(cinfo, JWRN_NOT_SEQUENTIAL);
+
+    /* Select MCU decoding routine */
+    /* We retain the hard-coded case for full-size blocks.
+     * This is not necessary, but it appears that this version is slightly
+     * more performant in the given implementation.
+     * With an improved implementation we would prefer a single optimized
+     * function.
+     */
+    if (cinfo->lim_Se != DCTSIZE2-1)
+      entropy->pub.decode_mcu = decode_mcu_sub;
+    else
+      entropy->pub.decode_mcu = decode_mcu;
+
+    for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+      compptr = cinfo->cur_comp_info[ci];
+      /* Compute derived values for Huffman tables */
+      /* We may do this more than once for a table, but it's not expensive */
+      tbl = compptr->dc_tbl_no;
+      jpeg_make_d_derived_tbl(cinfo, TRUE, tbl,
+			      & entropy->dc_derived_tbls[tbl]);
+      if (cinfo->lim_Se) {	/* AC needs no table when not present */
+	tbl = compptr->ac_tbl_no;
+	jpeg_make_d_derived_tbl(cinfo, FALSE, tbl,
+				& entropy->ac_derived_tbls[tbl]);
+      }
+      /* Initialize DC predictions to 0 */
+      entropy->saved.last_dc_val[ci] = 0;
+    }
+
+    /* Precalculate decoding info for each block in an MCU of this scan */
+    for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+      ci = cinfo->MCU_membership[blkn];
+      compptr = cinfo->cur_comp_info[ci];
+      /* Precalculate which table to use for each block */
+      entropy->dc_cur_tbls[blkn] = entropy->dc_derived_tbls[compptr->dc_tbl_no];
+      entropy->ac_cur_tbls[blkn] = entropy->ac_derived_tbls[compptr->ac_tbl_no];
+      /* Decide whether we really care about the coefficient values */
+      if (compptr->component_needed) {
+	ci = compptr->DCT_v_scaled_size;
+	i = compptr->DCT_h_scaled_size;
+	switch (cinfo->lim_Se) {
+	case (1*1-1):
+	  entropy->coef_limit[blkn] = 1;
+	  break;
+	case (2*2-1):
+	  if (ci <= 0 || ci > 2) ci = 2;
+	  if (i <= 0 || i > 2) i = 2;
+	  entropy->coef_limit[blkn] = 1 + jpeg_zigzag_order2[ci - 1][i - 1];
+	  break;
+	case (3*3-1):
+	  if (ci <= 0 || ci > 3) ci = 3;
+	  if (i <= 0 || i > 3) i = 3;
+	  entropy->coef_limit[blkn] = 1 + jpeg_zigzag_order3[ci - 1][i - 1];
+	  break;
+	case (4*4-1):
+	  if (ci <= 0 || ci > 4) ci = 4;
+	  if (i <= 0 || i > 4) i = 4;
+	  entropy->coef_limit[blkn] = 1 + jpeg_zigzag_order4[ci - 1][i - 1];
+	  break;
+	case (5*5-1):
+	  if (ci <= 0 || ci > 5) ci = 5;
+	  if (i <= 0 || i > 5) i = 5;
+	  entropy->coef_limit[blkn] = 1 + jpeg_zigzag_order5[ci - 1][i - 1];
+	  break;
+	case (6*6-1):
+	  if (ci <= 0 || ci > 6) ci = 6;
+	  if (i <= 0 || i > 6) i = 6;
+	  entropy->coef_limit[blkn] = 1 + jpeg_zigzag_order6[ci - 1][i - 1];
+	  break;
+	case (7*7-1):
+	  if (ci <= 0 || ci > 7) ci = 7;
+	  if (i <= 0 || i > 7) i = 7;
+	  entropy->coef_limit[blkn] = 1 + jpeg_zigzag_order7[ci - 1][i - 1];
+	  break;
+	default:
+	  if (ci <= 0 || ci > 8) ci = 8;
+	  if (i <= 0 || i > 8) i = 8;
+	  entropy->coef_limit[blkn] = 1 + jpeg_zigzag_order[ci - 1][i - 1];
+	  break;
+	}
+      } else {
+	entropy->coef_limit[blkn] = 0;
+      }
+    }
+  }
+
+  /* Initialize bitread state variables */
+  entropy->bitstate.bits_left = 0;
+  entropy->bitstate.get_buffer = 0; /* unnecessary, but keeps Purify quiet */
+  entropy->insufficient_data = FALSE;
+
+  /* Initialize restart counter */
+  entropy->restarts_to_go = cinfo->restart_interval;
+}
+
+
+/*
+ * Module initialization routine for Huffman entropy decoding.
+ */
+
+GLOBAL(void)
+jinit_huff_decoder (j_decompress_ptr cinfo)
+{
+  huff_entropy_ptr entropy;
+  int i;
+
+  entropy = (huff_entropy_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				SIZEOF(huff_entropy_decoder));
+  cinfo->entropy = (struct jpeg_entropy_decoder *) entropy;
+  entropy->pub.start_pass = start_pass_huff_decoder;
+
+  if (cinfo->progressive_mode) {
+    /* Create progression status table */
+    int *coef_bit_ptr, ci;
+    cinfo->coef_bits = (int (*)[DCTSIZE2])
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				  cinfo->num_components*DCTSIZE2*SIZEOF(int));
+    coef_bit_ptr = & cinfo->coef_bits[0][0];
+    for (ci = 0; ci < cinfo->num_components; ci++)
+      for (i = 0; i < DCTSIZE2; i++)
+	*coef_bit_ptr++ = -1;
+
+    /* Mark derived tables unallocated */
+    for (i = 0; i < NUM_HUFF_TBLS; i++) {
+      entropy->derived_tbls[i] = NULL;
+    }
+  } else {
+    /* Mark tables unallocated */
+    for (i = 0; i < NUM_HUFF_TBLS; i++) {
+      entropy->dc_derived_tbls[i] = entropy->ac_derived_tbls[i] = NULL;
+    }
+  }
+}
diff --git a/src/jpeg-8c/jdinput.c b/src/jpeg-8c/jdinput.c
new file mode 100644
index 00000000..2c5c717b
--- /dev/null
+++ b/src/jpeg-8c/jdinput.c
@@ -0,0 +1,661 @@
+/*
+ * jdinput.c
+ *
+ * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Modified 2002-2009 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains input control logic for the JPEG decompressor.
+ * These routines are concerned with controlling the decompressor's input
+ * processing (marker reading and coefficient decoding).  The actual input
+ * reading is done in jdmarker.c, jdhuff.c, and jdarith.c.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Private state */
+
+typedef struct {
+  struct jpeg_input_controller pub; /* public fields */
+
+  int inheaders;		/* Nonzero until first SOS is reached */
+} my_input_controller;
+
+typedef my_input_controller * my_inputctl_ptr;
+
+
+/* Forward declarations */
+METHODDEF(int) consume_markers JPP((j_decompress_ptr cinfo));
+
+
+/*
+ * Routines to calculate various quantities related to the size of the image.
+ */
+
+
+/*
+ * Compute output image dimensions and related values.
+ * NOTE: this is exported for possible use by application.
+ * Hence it mustn't do anything that can't be done twice.
+ */
+
+GLOBAL(void)
+jpeg_core_output_dimensions (j_decompress_ptr cinfo)
+/* Do computations that are needed before master selection phase.
+ * This function is used for transcoding and full decompression.
+ */
+{
+#ifdef IDCT_SCALING_SUPPORTED
+  int ci;
+  jpeg_component_info *compptr;
+
+  /* Compute actual output image dimensions and DCT scaling choices. */
+  if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom) {
+    /* Provide 1/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 1;
+    cinfo->min_DCT_v_scaled_size = 1;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 2) {
+    /* Provide 2/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 2L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 2L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 2;
+    cinfo->min_DCT_v_scaled_size = 2;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 3) {
+    /* Provide 3/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 3L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 3L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 3;
+    cinfo->min_DCT_v_scaled_size = 3;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 4) {
+    /* Provide 4/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 4L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 4L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 4;
+    cinfo->min_DCT_v_scaled_size = 4;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 5) {
+    /* Provide 5/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 5L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 5L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 5;
+    cinfo->min_DCT_v_scaled_size = 5;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 6) {
+    /* Provide 6/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 6L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 6L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 6;
+    cinfo->min_DCT_v_scaled_size = 6;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 7) {
+    /* Provide 7/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 7L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 7L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 7;
+    cinfo->min_DCT_v_scaled_size = 7;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 8) {
+    /* Provide 8/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 8L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 8L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 8;
+    cinfo->min_DCT_v_scaled_size = 8;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 9) {
+    /* Provide 9/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 9L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 9L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 9;
+    cinfo->min_DCT_v_scaled_size = 9;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 10) {
+    /* Provide 10/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 10L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 10L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 10;
+    cinfo->min_DCT_v_scaled_size = 10;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 11) {
+    /* Provide 11/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 11L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 11L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 11;
+    cinfo->min_DCT_v_scaled_size = 11;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 12) {
+    /* Provide 12/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 12L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 12L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 12;
+    cinfo->min_DCT_v_scaled_size = 12;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 13) {
+    /* Provide 13/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 13L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 13L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 13;
+    cinfo->min_DCT_v_scaled_size = 13;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 14) {
+    /* Provide 14/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 14L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 14L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 14;
+    cinfo->min_DCT_v_scaled_size = 14;
+  } else if (cinfo->scale_num * cinfo->block_size <= cinfo->scale_denom * 15) {
+    /* Provide 15/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 15L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 15L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 15;
+    cinfo->min_DCT_v_scaled_size = 15;
+  } else {
+    /* Provide 16/block_size scaling */
+    cinfo->output_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * 16L, (long) cinfo->block_size);
+    cinfo->output_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * 16L, (long) cinfo->block_size);
+    cinfo->min_DCT_h_scaled_size = 16;
+    cinfo->min_DCT_v_scaled_size = 16;
+  }
+
+  /* Recompute dimensions of components */
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    compptr->DCT_h_scaled_size = cinfo->min_DCT_h_scaled_size;
+    compptr->DCT_v_scaled_size = cinfo->min_DCT_v_scaled_size;
+  }
+
+#else /* !IDCT_SCALING_SUPPORTED */
+
+  /* Hardwire it to "no scaling" */
+  cinfo->output_width = cinfo->image_width;
+  cinfo->output_height = cinfo->image_height;
+  /* jdinput.c has already initialized DCT_scaled_size,
+   * and has computed unscaled downsampled_width and downsampled_height.
+   */
+
+#endif /* IDCT_SCALING_SUPPORTED */
+}
+
+
+LOCAL(void)
+initial_setup (j_decompress_ptr cinfo)
+/* Called once, when first SOS marker is reached */
+{
+  int ci;
+  jpeg_component_info *compptr;
+
+  /* Make sure image isn't bigger than I can handle */
+  if ((long) cinfo->image_height > (long) JPEG_MAX_DIMENSION ||
+      (long) cinfo->image_width > (long) JPEG_MAX_DIMENSION)
+    ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int) JPEG_MAX_DIMENSION);
+
+  /* For now, precision must match compiled-in value... */
+  if (cinfo->data_precision != BITS_IN_JSAMPLE)
+    ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);
+
+  /* Check that number of components won't exceed internal array sizes */
+  if (cinfo->num_components > MAX_COMPONENTS)
+    ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->num_components,
+	     MAX_COMPONENTS);
+
+  /* Compute maximum sampling factors; check factor validity */
+  cinfo->max_h_samp_factor = 1;
+  cinfo->max_v_samp_factor = 1;
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    if (compptr->h_samp_factor<=0 || compptr->h_samp_factor>MAX_SAMP_FACTOR ||
+	compptr->v_samp_factor<=0 || compptr->v_samp_factor>MAX_SAMP_FACTOR)
+      ERREXIT(cinfo, JERR_BAD_SAMPLING);
+    cinfo->max_h_samp_factor = MAX(cinfo->max_h_samp_factor,
+				   compptr->h_samp_factor);
+    cinfo->max_v_samp_factor = MAX(cinfo->max_v_samp_factor,
+				   compptr->v_samp_factor);
+  }
+
+  /* Derive block_size, natural_order, and lim_Se */
+  if (cinfo->is_baseline || (cinfo->progressive_mode &&
+      cinfo->comps_in_scan)) { /* no pseudo SOS marker */
+    cinfo->block_size = DCTSIZE;
+    cinfo->natural_order = jpeg_natural_order;
+    cinfo->lim_Se = DCTSIZE2-1;
+  } else
+    switch (cinfo->Se) {
+    case (1*1-1):
+      cinfo->block_size = 1;
+      cinfo->natural_order = jpeg_natural_order; /* not needed */
+      cinfo->lim_Se = cinfo->Se;
+      break;
+    case (2*2-1):
+      cinfo->block_size = 2;
+      cinfo->natural_order = jpeg_natural_order2;
+      cinfo->lim_Se = cinfo->Se;
+      break;
+    case (3*3-1):
+      cinfo->block_size = 3;
+      cinfo->natural_order = jpeg_natural_order3;
+      cinfo->lim_Se = cinfo->Se;
+      break;
+    case (4*4-1):
+      cinfo->block_size = 4;
+      cinfo->natural_order = jpeg_natural_order4;
+      cinfo->lim_Se = cinfo->Se;
+      break;
+    case (5*5-1):
+      cinfo->block_size = 5;
+      cinfo->natural_order = jpeg_natural_order5;
+      cinfo->lim_Se = cinfo->Se;
+      break;
+    case (6*6-1):
+      cinfo->block_size = 6;
+      cinfo->natural_order = jpeg_natural_order6;
+      cinfo->lim_Se = cinfo->Se;
+      break;
+    case (7*7-1):
+      cinfo->block_size = 7;
+      cinfo->natural_order = jpeg_natural_order7;
+      cinfo->lim_Se = cinfo->Se;
+      break;
+    case (8*8-1):
+      cinfo->block_size = 8;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (9*9-1):
+      cinfo->block_size = 9;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (10*10-1):
+      cinfo->block_size = 10;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (11*11-1):
+      cinfo->block_size = 11;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (12*12-1):
+      cinfo->block_size = 12;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (13*13-1):
+      cinfo->block_size = 13;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (14*14-1):
+      cinfo->block_size = 14;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (15*15-1):
+      cinfo->block_size = 15;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    case (16*16-1):
+      cinfo->block_size = 16;
+      cinfo->natural_order = jpeg_natural_order;
+      cinfo->lim_Se = DCTSIZE2-1;
+      break;
+    default:
+      ERREXIT4(cinfo, JERR_BAD_PROGRESSION,
+	       cinfo->Ss, cinfo->Se, cinfo->Ah, cinfo->Al);
+      break;
+    }
+
+  /* We initialize DCT_scaled_size and min_DCT_scaled_size to block_size.
+   * In the full decompressor,
+   * this will be overridden by jpeg_calc_output_dimensions in jdmaster.c;
+   * but in the transcoder,
+   * jpeg_calc_output_dimensions is not used, so we must do it here.
+   */
+  cinfo->min_DCT_h_scaled_size = cinfo->block_size;
+  cinfo->min_DCT_v_scaled_size = cinfo->block_size;
+
+  /* Compute dimensions of components */
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    compptr->DCT_h_scaled_size = cinfo->block_size;
+    compptr->DCT_v_scaled_size = cinfo->block_size;
+    /* Size in DCT blocks */
+    compptr->width_in_blocks = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * (long) compptr->h_samp_factor,
+		    (long) (cinfo->max_h_samp_factor * cinfo->block_size));
+    compptr->height_in_blocks = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * (long) compptr->v_samp_factor,
+		    (long) (cinfo->max_v_samp_factor * cinfo->block_size));
+    /* downsampled_width and downsampled_height will also be overridden by
+     * jdmaster.c if we are doing full decompression.  The transcoder library
+     * doesn't use these values, but the calling application might.
+     */
+    /* Size in samples */
+    compptr->downsampled_width = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width * (long) compptr->h_samp_factor,
+		    (long) cinfo->max_h_samp_factor);
+    compptr->downsampled_height = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height * (long) compptr->v_samp_factor,
+		    (long) cinfo->max_v_samp_factor);
+    /* Mark component needed, until color conversion says otherwise */
+    compptr->component_needed = TRUE;
+    /* Mark no quantization table yet saved for component */
+    compptr->quant_table = NULL;
+  }
+
+  /* Compute number of fully interleaved MCU rows. */
+  cinfo->total_iMCU_rows = (JDIMENSION)
+    jdiv_round_up((long) cinfo->image_height,
+	          (long) (cinfo->max_v_samp_factor * cinfo->block_size));
+
+  /* Decide whether file contains multiple scans */
+  if (cinfo->comps_in_scan < cinfo->num_components || cinfo->progressive_mode)
+    cinfo->inputctl->has_multiple_scans = TRUE;
+  else
+    cinfo->inputctl->has_multiple_scans = FALSE;
+}
+
+
+LOCAL(void)
+per_scan_setup (j_decompress_ptr cinfo)
+/* Do computations that are needed before processing a JPEG scan */
+/* cinfo->comps_in_scan and cinfo->cur_comp_info[] were set from SOS marker */
+{
+  int ci, mcublks, tmp;
+  jpeg_component_info *compptr;
+  
+  if (cinfo->comps_in_scan == 1) {
+    
+    /* Noninterleaved (single-component) scan */
+    compptr = cinfo->cur_comp_info[0];
+    
+    /* Overall image size in MCUs */
+    cinfo->MCUs_per_row = compptr->width_in_blocks;
+    cinfo->MCU_rows_in_scan = compptr->height_in_blocks;
+    
+    /* For noninterleaved scan, always one block per MCU */
+    compptr->MCU_width = 1;
+    compptr->MCU_height = 1;
+    compptr->MCU_blocks = 1;
+    compptr->MCU_sample_width = compptr->DCT_h_scaled_size;
+    compptr->last_col_width = 1;
+    /* For noninterleaved scans, it is convenient to define last_row_height
+     * as the number of block rows present in the last iMCU row.
+     */
+    tmp = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
+    if (tmp == 0) tmp = compptr->v_samp_factor;
+    compptr->last_row_height = tmp;
+    
+    /* Prepare array describing MCU composition */
+    cinfo->blocks_in_MCU = 1;
+    cinfo->MCU_membership[0] = 0;
+    
+  } else {
+    
+    /* Interleaved (multi-component) scan */
+    if (cinfo->comps_in_scan <= 0 || cinfo->comps_in_scan > MAX_COMPS_IN_SCAN)
+      ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->comps_in_scan,
+	       MAX_COMPS_IN_SCAN);
+    
+    /* Overall image size in MCUs */
+    cinfo->MCUs_per_row = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_width,
+		    (long) (cinfo->max_h_samp_factor * cinfo->block_size));
+    cinfo->MCU_rows_in_scan = (JDIMENSION)
+      jdiv_round_up((long) cinfo->image_height,
+		    (long) (cinfo->max_v_samp_factor * cinfo->block_size));
+    
+    cinfo->blocks_in_MCU = 0;
+    
+    for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+      compptr = cinfo->cur_comp_info[ci];
+      /* Sampling factors give # of blocks of component in each MCU */
+      compptr->MCU_width = compptr->h_samp_factor;
+      compptr->MCU_height = compptr->v_samp_factor;
+      compptr->MCU_blocks = compptr->MCU_width * compptr->MCU_height;
+      compptr->MCU_sample_width = compptr->MCU_width * compptr->DCT_h_scaled_size;
+      /* Figure number of non-dummy blocks in last MCU column & row */
+      tmp = (int) (compptr->width_in_blocks % compptr->MCU_width);
+      if (tmp == 0) tmp = compptr->MCU_width;
+      compptr->last_col_width = tmp;
+      tmp = (int) (compptr->height_in_blocks % compptr->MCU_height);
+      if (tmp == 0) tmp = compptr->MCU_height;
+      compptr->last_row_height = tmp;
+      /* Prepare array describing MCU composition */
+      mcublks = compptr->MCU_blocks;
+      if (cinfo->blocks_in_MCU + mcublks > D_MAX_BLOCKS_IN_MCU)
+	ERREXIT(cinfo, JERR_BAD_MCU_SIZE);
+      while (mcublks-- > 0) {
+	cinfo->MCU_membership[cinfo->blocks_in_MCU++] = ci;
+      }
+    }
+    
+  }
+}
+
+
+/*
+ * Save away a copy of the Q-table referenced by each component present
+ * in the current scan, unless already saved during a prior scan.
+ *
+ * In a multiple-scan JPEG file, the encoder could assign different components
+ * the same Q-table slot number, but change table definitions between scans
+ * so that each component uses a different Q-table.  (The IJG encoder is not
+ * currently capable of doing this, but other encoders might.)  Since we want
+ * to be able to dequantize all the components at the end of the file, this
+ * means that we have to save away the table actually used for each component.
+ * We do this by copying the table at the start of the first scan containing
+ * the component.
+ * The JPEG spec prohibits the encoder from changing the contents of a Q-table
+ * slot between scans of a component using that slot.  If the encoder does so
+ * anyway, this decoder will simply use the Q-table values that were current
+ * at the start of the first scan for the component.
+ *
+ * The decompressor output side looks only at the saved quant tables,
+ * not at the current Q-table slots.
+ */
+
+LOCAL(void)
+latch_quant_tables (j_decompress_ptr cinfo)
+{
+  int ci, qtblno;
+  jpeg_component_info *compptr;
+  JQUANT_TBL * qtbl;
+
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    /* No work if we already saved Q-table for this component */
+    if (compptr->quant_table != NULL)
+      continue;
+    /* Make sure specified quantization table is present */
+    qtblno = compptr->quant_tbl_no;
+    if (qtblno < 0 || qtblno >= NUM_QUANT_TBLS ||
+	cinfo->quant_tbl_ptrs[qtblno] == NULL)
+      ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, qtblno);
+    /* OK, save away the quantization table */
+    qtbl = (JQUANT_TBL *)
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				  SIZEOF(JQUANT_TBL));
+    MEMCOPY(qtbl, cinfo->quant_tbl_ptrs[qtblno], SIZEOF(JQUANT_TBL));
+    compptr->quant_table = qtbl;
+  }
+}
+
+
+/*
+ * Initialize the input modules to read a scan of compressed data.
+ * The first call to this is done by jdmaster.c after initializing
+ * the entire decompressor (during jpeg_start_decompress).
+ * Subsequent calls come from consume_markers, below.
+ */
+
+METHODDEF(void)
+start_input_pass (j_decompress_ptr cinfo)
+{
+  per_scan_setup(cinfo);
+  latch_quant_tables(cinfo);
+  (*cinfo->entropy->start_pass) (cinfo);
+  (*cinfo->coef->start_input_pass) (cinfo);
+  cinfo->inputctl->consume_input = cinfo->coef->consume_data;
+}
+
+
+/*
+ * Finish up after inputting a compressed-data scan.
+ * This is called by the coefficient controller after it's read all
+ * the expected data of the scan.
+ */
+
+METHODDEF(void)
+finish_input_pass (j_decompress_ptr cinfo)
+{
+  cinfo->inputctl->consume_input = consume_markers;
+}
+
+
+/*
+ * Read JPEG markers before, between, or after compressed-data scans.
+ * Change state as necessary when a new scan is reached.
+ * Return value is JPEG_SUSPENDED, JPEG_REACHED_SOS, or JPEG_REACHED_EOI.
+ *
+ * The consume_input method pointer points either here or to the
+ * coefficient controller's consume_data routine, depending on whether
+ * we are reading a compressed data segment or inter-segment markers.
+ *
+ * Note: This function should NOT return a pseudo SOS marker (with zero
+ * component number) to the caller.  A pseudo marker received by
+ * read_markers is processed and then skipped for other markers.
+ */
+
+METHODDEF(int)
+consume_markers (j_decompress_ptr cinfo)
+{
+  my_inputctl_ptr inputctl = (my_inputctl_ptr) cinfo->inputctl;
+  int val;
+
+  if (inputctl->pub.eoi_reached) /* After hitting EOI, read no further */
+    return JPEG_REACHED_EOI;
+
+  for (;;) {			/* Loop to pass pseudo SOS marker */
+    val = (*cinfo->marker->read_markers) (cinfo);
+
+    switch (val) {
+    case JPEG_REACHED_SOS:	/* Found SOS */
+      if (inputctl->inheaders) { /* 1st SOS */
+	if (inputctl->inheaders == 1)
+	  initial_setup(cinfo);
+	if (cinfo->comps_in_scan == 0) { /* pseudo SOS marker */
+	  inputctl->inheaders = 2;
+	  break;
+	}
+	inputctl->inheaders = 0;
+	/* Note: start_input_pass must be called by jdmaster.c
+	 * before any more input can be consumed.  jdapimin.c is
+	 * responsible for enforcing this sequencing.
+	 */
+      } else {			/* 2nd or later SOS marker */
+	if (! inputctl->pub.has_multiple_scans)
+	  ERREXIT(cinfo, JERR_EOI_EXPECTED); /* Oops, I wasn't expecting this! */
+	if (cinfo->comps_in_scan == 0) /* unexpected pseudo SOS marker */
+	  break;
+	start_input_pass(cinfo);
+      }
+      return val;
+    case JPEG_REACHED_EOI:	/* Found EOI */
+      inputctl->pub.eoi_reached = TRUE;
+      if (inputctl->inheaders) { /* Tables-only datastream, apparently */
+	if (cinfo->marker->saw_SOF)
+	  ERREXIT(cinfo, JERR_SOF_NO_SOS);
+      } else {
+	/* Prevent infinite loop in coef ctlr's decompress_data routine
+	 * if user set output_scan_number larger than number of scans.
+	 */
+	if (cinfo->output_scan_number > cinfo->input_scan_number)
+	  cinfo->output_scan_number = cinfo->input_scan_number;
+      }
+      return val;
+    case JPEG_SUSPENDED:
+      return val;
+    default:
+      return val;
+    }
+  }
+}
+
+
+/*
+ * Reset state to begin a fresh datastream.
+ */
+
+METHODDEF(void)
+reset_input_controller (j_decompress_ptr cinfo)
+{
+  my_inputctl_ptr inputctl = (my_inputctl_ptr) cinfo->inputctl;
+
+  inputctl->pub.consume_input = consume_markers;
+  inputctl->pub.has_multiple_scans = FALSE; /* "unknown" would be better */
+  inputctl->pub.eoi_reached = FALSE;
+  inputctl->inheaders = 1;
+  /* Reset other modules */
+  (*cinfo->err->reset_error_mgr) ((j_common_ptr) cinfo);
+  (*cinfo->marker->reset_marker_reader) (cinfo);
+  /* Reset progression state -- would be cleaner if entropy decoder did this */
+  cinfo->coef_bits = NULL;
+}
+
+
+/*
+ * Initialize the input controller module.
+ * This is called only once, when the decompression object is created.
+ */
+
+GLOBAL(void)
+jinit_input_controller (j_decompress_ptr cinfo)
+{
+  my_inputctl_ptr inputctl;
+
+  /* Create subobject in permanent pool */
+  inputctl = (my_inputctl_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
+				SIZEOF(my_input_controller));
+  cinfo->inputctl = (struct jpeg_input_controller *) inputctl;
+  /* Initialize method pointers */
+  inputctl->pub.consume_input = consume_markers;
+  inputctl->pub.reset_input_controller = reset_input_controller;
+  inputctl->pub.start_input_pass = start_input_pass;
+  inputctl->pub.finish_input_pass = finish_input_pass;
+  /* Initialize state: can't use reset_input_controller since we don't
+   * want to try to reset other modules yet.
+   */
+  inputctl->pub.has_multiple_scans = FALSE; /* "unknown" would be better */
+  inputctl->pub.eoi_reached = FALSE;
+  inputctl->inheaders = 1;
+}
diff --git a/src/jpeg-6b/jdmainct.c b/src/jpeg-8c/jdmainct.c
index 2f265336..c04e8223 100644
--- a/src/jpeg-6b/jdmainct.c
+++ b/src/jpeg-8c/jdmainct.c
@@ -159,24 +159,24 @@ alloc_funny_pointers (j_decompress_ptr cinfo)
  * This is done only once, not once per pass.
  */
 {
-  my_main_ptr jmain = (my_main_ptr) cinfo->main;
+  my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
   int ci, rgroup;
-  int M = cinfo->min_DCT_scaled_size;
+  int M = cinfo->min_DCT_v_scaled_size;
   jpeg_component_info *compptr;
   JSAMPARRAY xbuf;
 
   /* Get top-level space for component array pointers.
    * We alloc both arrays with one call to save a few cycles.
    */
-  jmain->xbuffer[0] = (JSAMPIMAGE)
+  main_ptr->xbuffer[0] = (JSAMPIMAGE)
     (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
 				cinfo->num_components * 2 * SIZEOF(JSAMPARRAY));
-  jmain->xbuffer[1] = jmain->xbuffer[0] + cinfo->num_components;
+  main_ptr->xbuffer[1] = main_ptr->xbuffer[0] + cinfo->num_components;
 
   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
        ci++, compptr++) {
-    rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
-      cinfo->min_DCT_scaled_size; /* height of a row group of component */
+    rgroup = (compptr->v_samp_factor * compptr->DCT_v_scaled_size) /
+      cinfo->min_DCT_v_scaled_size; /* height of a row group of component */
     /* Get space for pointer lists --- M+4 row groups in each list.
      * We alloc both pointer lists with one call to save a few cycles.
      */
@@ -184,9 +184,9 @@ alloc_funny_pointers (j_decompress_ptr cinfo)
       (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
 				  2 * (rgroup * (M + 4)) * SIZEOF(JSAMPROW));
     xbuf += rgroup;		/* want one row group at negative offsets */
-    jmain->xbuffer[0][ci] = xbuf;
+    main_ptr->xbuffer[0][ci] = xbuf;
     xbuf += rgroup * (M + 4);
-    jmain->xbuffer[1][ci] = xbuf;
+    main_ptr->xbuffer[1][ci] = xbuf;
   }
 }
 
@@ -194,26 +194,26 @@ alloc_funny_pointers (j_decompress_ptr cinfo)
 LOCAL(void)
 make_funny_pointers (j_decompress_ptr cinfo)
 /* Create the funny pointer lists discussed in the comments above.
- * The actual workspace is already allocated (in jmain->buffer),
+ * The actual workspace is already allocated (in main_ptr->buffer),
  * and the space for the pointer lists is allocated too.
  * This routine just fills in the curiously ordered lists.
  * This will be repeated at the beginning of each pass.
  */
 {
-  my_main_ptr jmain = (my_main_ptr) cinfo->main;
+  my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
   int ci, i, rgroup;
-  int M = cinfo->min_DCT_scaled_size;
+  int M = cinfo->min_DCT_v_scaled_size;
   jpeg_component_info *compptr;
   JSAMPARRAY buf, xbuf0, xbuf1;
 
   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
        ci++, compptr++) {
-    rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
-      cinfo->min_DCT_scaled_size; /* height of a row group of component */
-    xbuf0 = jmain->xbuffer[0][ci];
-    xbuf1 = jmain->xbuffer[1][ci];
+    rgroup = (compptr->v_samp_factor * compptr->DCT_v_scaled_size) /
+      cinfo->min_DCT_v_scaled_size; /* height of a row group of component */
+    xbuf0 = main_ptr->xbuffer[0][ci];
+    xbuf1 = main_ptr->xbuffer[1][ci];
     /* First copy the workspace pointers as-is */
-    buf = jmain->buffer[ci];
+    buf = main_ptr->buffer[ci];
     for (i = 0; i < rgroup * (M + 2); i++) {
       xbuf0[i] = xbuf1[i] = buf[i];
     }
@@ -240,18 +240,18 @@ set_wraparound_pointers (j_decompress_ptr cinfo)
  * This changes the pointer list state from top-of-image to the normal state.
  */
 {
-  my_main_ptr jmain = (my_main_ptr) cinfo->main;
+  my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
   int ci, i, rgroup;
-  int M = cinfo->min_DCT_scaled_size;
+  int M = cinfo->min_DCT_v_scaled_size;
   jpeg_component_info *compptr;
   JSAMPARRAY xbuf0, xbuf1;
 
   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
        ci++, compptr++) {
-    rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
-      cinfo->min_DCT_scaled_size; /* height of a row group of component */
-    xbuf0 = jmain->xbuffer[0][ci];
-    xbuf1 = jmain->xbuffer[1][ci];
+    rgroup = (compptr->v_samp_factor * compptr->DCT_v_scaled_size) /
+      cinfo->min_DCT_v_scaled_size; /* height of a row group of component */
+    xbuf0 = main_ptr->xbuffer[0][ci];
+    xbuf1 = main_ptr->xbuffer[1][ci];
     for (i = 0; i < rgroup; i++) {
       xbuf0[i - rgroup] = xbuf0[rgroup*(M+1) + i];
       xbuf1[i - rgroup] = xbuf1[rgroup*(M+1) + i];
@@ -269,7 +269,7 @@ set_bottom_pointers (j_decompress_ptr cinfo)
  * Also sets rowgroups_avail to indicate number of nondummy row groups in row.
  */
 {
-  my_main_ptr jmain = (my_main_ptr) cinfo->main;
+  my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
   int ci, i, rgroup, iMCUheight, rows_left;
   jpeg_component_info *compptr;
   JSAMPARRAY xbuf;
@@ -277,8 +277,8 @@ set_bottom_pointers (j_decompress_ptr cinfo)
   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
        ci++, compptr++) {
     /* Count sample rows in one iMCU row and in one row group */
-    iMCUheight = compptr->v_samp_factor * compptr->DCT_scaled_size;
-    rgroup = iMCUheight / cinfo->min_DCT_scaled_size;
+    iMCUheight = compptr->v_samp_factor * compptr->DCT_v_scaled_size;
+    rgroup = iMCUheight / cinfo->min_DCT_v_scaled_size;
     /* Count nondummy sample rows remaining for this component */
     rows_left = (int) (compptr->downsampled_height % (JDIMENSION) iMCUheight);
     if (rows_left == 0) rows_left = iMCUheight;
@@ -286,12 +286,12 @@ set_bottom_pointers (j_decompress_ptr cinfo)
      * so we need only do it once.
      */
     if (ci == 0) {
-      jmain->rowgroups_avail = (JDIMENSION) ((rows_left-1) / rgroup + 1);
+      main_ptr->rowgroups_avail = (JDIMENSION) ((rows_left-1) / rgroup + 1);
     }
     /* Duplicate the last real sample row rgroup*2 times; this pads out the
      * last partial rowgroup and ensures at least one full rowgroup of context.
      */
-    xbuf = jmain->xbuffer[jmain->whichptr][ci];
+    xbuf = main_ptr->xbuffer[main_ptr->whichptr][ci];
     for (i = 0; i < rgroup * 2; i++) {
       xbuf[rows_left + i] = xbuf[rows_left-1];
     }
@@ -306,27 +306,27 @@ set_bottom_pointers (j_decompress_ptr cinfo)
 METHODDEF(void)
 start_pass_main (j_decompress_ptr cinfo, J_BUF_MODE pass_mode)
 {
-  my_main_ptr jmain = (my_main_ptr) cinfo->main;
+  my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
 
   switch (pass_mode) {
   case JBUF_PASS_THRU:
     if (cinfo->upsample->need_context_rows) {
-      jmain->pub.process_data = process_data_context_main;
+      main_ptr->pub.process_data = process_data_context_main;
       make_funny_pointers(cinfo); /* Create the xbuffer[] lists */
-      jmain->whichptr = 0;	/* Read first iMCU row into xbuffer[0] */
-      jmain->context_state = CTX_PREPARE_FOR_IMCU;
-      jmain->iMCU_row_ctr = 0;
+      main_ptr->whichptr = 0;	/* Read first iMCU row into xbuffer[0] */
+      main_ptr->context_state = CTX_PREPARE_FOR_IMCU;
+      main_ptr->iMCU_row_ctr = 0;
     } else {
       /* Simple case with no context needed */
-      jmain->pub.process_data = process_data_simple_main;
+      main_ptr->pub.process_data = process_data_simple_main;
     }
-    jmain->buffer_full = FALSE;	/* Mark buffer empty */
-    jmain->rowgroup_ctr = 0;
+    main_ptr->buffer_full = FALSE;	/* Mark buffer empty */
+    main_ptr->rowgroup_ctr = 0;
     break;
 #ifdef QUANT_2PASS_SUPPORTED
   case JBUF_CRANK_DEST:
     /* For last pass of 2-pass quantization, just crank the postprocessor */
-    jmain->pub.process_data = process_data_crank_post;
+    main_ptr->pub.process_data = process_data_crank_post;
     break;
 #endif
   default:
@@ -346,32 +346,32 @@ process_data_simple_main (j_decompress_ptr cinfo,
 			  JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
 			  JDIMENSION out_rows_avail)
 {
-  my_main_ptr jmain = (my_main_ptr) cinfo->main;
+  my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
   JDIMENSION rowgroups_avail;
 
-  /* Read input data if we haven't filled the jmain buffer yet */
-  if (! jmain->buffer_full) {
-    if (! (*cinfo->coef->decompress_data) (cinfo, jmain->buffer))
+  /* Read input data if we haven't filled the main buffer yet */
+  if (! main_ptr->buffer_full) {
+    if (! (*cinfo->coef->decompress_data) (cinfo, main_ptr->buffer))
       return;			/* suspension forced, can do nothing more */
-    jmain->buffer_full = TRUE;	/* OK, we have an iMCU row to work with */
+    main_ptr->buffer_full = TRUE;	/* OK, we have an iMCU row to work with */
   }
 
   /* There are always min_DCT_scaled_size row groups in an iMCU row. */
-  rowgroups_avail = (JDIMENSION) cinfo->min_DCT_scaled_size;
+  rowgroups_avail = (JDIMENSION) cinfo->min_DCT_v_scaled_size;
   /* Note: at the bottom of the image, we may pass extra garbage row groups
    * to the postprocessor.  The postprocessor has to check for bottom
    * of image anyway (at row resolution), so no point in us doing it too.
    */
 
   /* Feed the postprocessor */
-  (*cinfo->post->post_process_data) (cinfo, jmain->buffer,
-				     &jmain->rowgroup_ctr, rowgroups_avail,
+  (*cinfo->post->post_process_data) (cinfo, main_ptr->buffer,
+				     &main_ptr->rowgroup_ctr, rowgroups_avail,
 				     output_buf, out_row_ctr, out_rows_avail);
 
   /* Has postprocessor consumed all the data yet? If so, mark buffer empty */
-  if (jmain->rowgroup_ctr >= rowgroups_avail) {
-    jmain->buffer_full = FALSE;
-    jmain->rowgroup_ctr = 0;
+  if (main_ptr->rowgroup_ctr >= rowgroups_avail) {
+    main_ptr->buffer_full = FALSE;
+    main_ptr->rowgroup_ctr = 0;
   }
 }
 
@@ -386,15 +386,15 @@ process_data_context_main (j_decompress_ptr cinfo,
 			   JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
 			   JDIMENSION out_rows_avail)
 {
-  my_main_ptr jmain = (my_main_ptr) cinfo->main;
+  my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
 
-  /* Read input data if we haven't filled the jmain buffer yet */
-  if (! jmain->buffer_full) {
+  /* Read input data if we haven't filled the main buffer yet */
+  if (! main_ptr->buffer_full) {
     if (! (*cinfo->coef->decompress_data) (cinfo,
-					   jmain->xbuffer[jmain->whichptr]))
+					   main_ptr->xbuffer[main_ptr->whichptr]))
       return;			/* suspension forced, can do nothing more */
-    jmain->buffer_full = TRUE;	/* OK, we have an iMCU row to work with */
-    jmain->iMCU_row_ctr++;	/* count rows received */
+    main_ptr->buffer_full = TRUE;	/* OK, we have an iMCU row to work with */
+    main_ptr->iMCU_row_ctr++;	/* count rows received */
   }
 
   /* Postprocessor typically will not swallow all the input data it is handed
@@ -402,47 +402,47 @@ process_data_context_main (j_decompress_ptr cinfo,
    * to exit and restart.  This switch lets us keep track of how far we got.
    * Note that each case falls through to the next on successful completion.
    */
-  switch (jmain->context_state) {
+  switch (main_ptr->context_state) {
   case CTX_POSTPONED_ROW:
     /* Call postprocessor using previously set pointers for postponed row */
-    (*cinfo->post->post_process_data) (cinfo, jmain->xbuffer[jmain->whichptr],
-			&jmain->rowgroup_ctr, jmain->rowgroups_avail,
+    (*cinfo->post->post_process_data) (cinfo, main_ptr->xbuffer[main_ptr->whichptr],
+			&main_ptr->rowgroup_ctr, main_ptr->rowgroups_avail,
 			output_buf, out_row_ctr, out_rows_avail);
-    if (jmain->rowgroup_ctr < jmain->rowgroups_avail)
+    if (main_ptr->rowgroup_ctr < main_ptr->rowgroups_avail)
       return;			/* Need to suspend */
-    jmain->context_state = CTX_PREPARE_FOR_IMCU;
+    main_ptr->context_state = CTX_PREPARE_FOR_IMCU;
     if (*out_row_ctr >= out_rows_avail)
       return;			/* Postprocessor exactly filled output buf */
     /*FALLTHROUGH*/
   case CTX_PREPARE_FOR_IMCU:
     /* Prepare to process first M-1 row groups of this iMCU row */
-    jmain->rowgroup_ctr = 0;
-    jmain->rowgroups_avail = (JDIMENSION) (cinfo->min_DCT_scaled_size - 1);
+    main_ptr->rowgroup_ctr = 0;
+    main_ptr->rowgroups_avail = (JDIMENSION) (cinfo->min_DCT_v_scaled_size - 1);
     /* Check for bottom of image: if so, tweak pointers to "duplicate"
      * the last sample row, and adjust rowgroups_avail to ignore padding rows.
      */
-    if (jmain->iMCU_row_ctr == cinfo->total_iMCU_rows)
+    if (main_ptr->iMCU_row_ctr == cinfo->total_iMCU_rows)
       set_bottom_pointers(cinfo);
-    jmain->context_state = CTX_PROCESS_IMCU;
+    main_ptr->context_state = CTX_PROCESS_IMCU;
     /*FALLTHROUGH*/
   case CTX_PROCESS_IMCU:
     /* Call postprocessor using previously set pointers */
-    (*cinfo->post->post_process_data) (cinfo, jmain->xbuffer[jmain->whichptr],
-			&jmain->rowgroup_ctr, jmain->rowgroups_avail,
+    (*cinfo->post->post_process_data) (cinfo, main_ptr->xbuffer[main_ptr->whichptr],
+			&main_ptr->rowgroup_ctr, main_ptr->rowgroups_avail,
 			output_buf, out_row_ctr, out_rows_avail);
-    if (jmain->rowgroup_ctr < jmain->rowgroups_avail)
+    if (main_ptr->rowgroup_ctr < main_ptr->rowgroups_avail)
       return;			/* Need to suspend */
     /* After the first iMCU, change wraparound pointers to normal state */
-    if (jmain->iMCU_row_ctr == 1)
+    if (main_ptr->iMCU_row_ctr == 1)
       set_wraparound_pointers(cinfo);
     /* Prepare to load new iMCU row using other xbuffer list */
-    jmain->whichptr ^= 1;	/* 0=>1 or 1=>0 */
-    jmain->buffer_full = FALSE;
+    main_ptr->whichptr ^= 1;	/* 0=>1 or 1=>0 */
+    main_ptr->buffer_full = FALSE;
     /* Still need to process last row group of this iMCU row, */
     /* which is saved at index M+1 of the other xbuffer */
-    jmain->rowgroup_ctr = (JDIMENSION) (cinfo->min_DCT_scaled_size + 1);
-    jmain->rowgroups_avail = (JDIMENSION) (cinfo->min_DCT_scaled_size + 2);
-    jmain->context_state = CTX_POSTPONED_ROW;
+    main_ptr->rowgroup_ctr = (JDIMENSION) (cinfo->min_DCT_v_scaled_size + 1);
+    main_ptr->rowgroups_avail = (JDIMENSION) (cinfo->min_DCT_v_scaled_size + 2);
+    main_ptr->context_state = CTX_POSTPONED_ROW;
   }
 }
 
@@ -469,21 +469,21 @@ process_data_crank_post (j_decompress_ptr cinfo,
 
 
 /*
- * Initialize jmain buffer controller.
+ * Initialize main buffer controller.
  */
 
 GLOBAL(void)
 jinit_d_main_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
 {
-  my_main_ptr jmain;
+  my_main_ptr main_ptr;
   int ci, rgroup, ngroups;
   jpeg_component_info *compptr;
 
-  jmain = (my_main_ptr)
+  main_ptr = (my_main_ptr)
     (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
 				SIZEOF(my_main_controller));
-  cinfo->main = (struct jpeg_d_main_controller *) jmain;
-  jmain->pub.start_pass = start_pass_main;
+  cinfo->main = (struct jpeg_d_main_controller *) main_ptr;
+  main_ptr->pub.start_pass = start_pass_main;
 
   if (need_full_buffer)		/* shouldn't happen */
     ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
@@ -492,21 +492,21 @@ jinit_d_main_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
    * ngroups is the number of row groups we need.
    */
   if (cinfo->upsample->need_context_rows) {
-    if (cinfo->min_DCT_scaled_size < 2) /* unsupported, see comments above */
+    if (cinfo->min_DCT_v_scaled_size < 2) /* unsupported, see comments above */
       ERREXIT(cinfo, JERR_NOTIMPL);
     alloc_funny_pointers(cinfo); /* Alloc space for xbuffer[] lists */
-    ngroups = cinfo->min_DCT_scaled_size + 2;
+    ngroups = cinfo->min_DCT_v_scaled_size + 2;
   } else {
-    ngroups = cinfo->min_DCT_scaled_size;
+    ngroups = cinfo->min_DCT_v_scaled_size;
   }
 
   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
        ci++, compptr++) {
-    rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
-      cinfo->min_DCT_scaled_size; /* height of a row group of component */
-    jmain->buffer[ci] = (*cinfo->mem->alloc_sarray)
+    rgroup = (compptr->v_samp_factor * compptr->DCT_v_scaled_size) /
+      cinfo->min_DCT_v_scaled_size; /* height of a row group of component */
+    main_ptr->buffer[ci] = (*cinfo->mem->alloc_sarray)
 			((j_common_ptr) cinfo, JPOOL_IMAGE,
-			 compptr->width_in_blocks * compptr->DCT_scaled_size,
+			 compptr->width_in_blocks * compptr->DCT_h_scaled_size,
 			 (JDIMENSION) (rgroup * ngroups));
   }
 }
diff --git a/src/jpeg-6b/jdmarker.c b/src/jpeg-8c/jdmarker.c
index f4cca8cc..f2a9cc42 100644
--- a/src/jpeg-6b/jdmarker.c
+++ b/src/jpeg-8c/jdmarker.c
@@ -2,6 +2,7 @@
  * jdmarker.c
  *
  * Copyright (C) 1991-1998, Thomas G. Lane.
+ * Modified 2009 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -234,7 +235,8 @@ get_soi (j_decompress_ptr cinfo)
 
 
 LOCAL(boolean)
-get_sof (j_decompress_ptr cinfo, boolean is_prog, boolean is_arith)
+get_sof (j_decompress_ptr cinfo, boolean is_baseline, boolean is_prog,
+	 boolean is_arith)
 /* Process a SOFn marker */
 {
   INT32 length;
@@ -242,6 +244,7 @@ get_sof (j_decompress_ptr cinfo, boolean is_prog, boolean is_arith)
   jpeg_component_info * compptr;
   INPUT_VARS(cinfo);
 
+  cinfo->is_baseline = is_baseline;
   cinfo->progressive_mode = is_prog;
   cinfo->arith_code = is_arith;
 
@@ -315,7 +318,9 @@ get_sos (j_decompress_ptr cinfo)
 
   TRACEMS1(cinfo, 1, JTRC_SOS, n);
 
-  if (length != (n * 2 + 6) || n < 1 || n > MAX_COMPS_IN_SCAN)
+  if (length != (n * 2 + 6) || n > MAX_COMPS_IN_SCAN ||
+      (n == 0 && !cinfo->progressive_mode))
+      /* pseudo SOS marker only allowed in progressive mode */
     ERREXIT(cinfo, JERR_BAD_LENGTH);
 
   cinfo->comps_in_scan = n;
@@ -359,8 +364,8 @@ get_sos (j_decompress_ptr cinfo)
   /* Prepare to scan data & restart markers */
   cinfo->marker->next_restart_num = 0;
 
-  /* Count another SOS marker */
-  cinfo->input_scan_number++;
+  /* Count another (non-pseudo) SOS marker */
+  if (n) cinfo->input_scan_number++;
 
   INPUT_SYNC(cinfo);
   return TRUE;
@@ -490,16 +495,18 @@ LOCAL(boolean)
 get_dqt (j_decompress_ptr cinfo)
 /* Process a DQT marker */
 {
-  INT32 length;
-  int n, i, prec;
+  INT32 length, count, i;
+  int n, prec;
   unsigned int tmp;
   JQUANT_TBL *quant_ptr;
+  const int *natural_order;
   INPUT_VARS(cinfo);
 
   INPUT_2BYTES(cinfo, length, return FALSE);
   length -= 2;
 
   while (length > 0) {
+    length--;
     INPUT_BYTE(cinfo, n, return FALSE);
     prec = n >> 4;
     n &= 0x0F;
@@ -513,13 +520,43 @@ get_dqt (j_decompress_ptr cinfo)
       cinfo->quant_tbl_ptrs[n] = jpeg_alloc_quant_table((j_common_ptr) cinfo);
     quant_ptr = cinfo->quant_tbl_ptrs[n];
 
-    for (i = 0; i < DCTSIZE2; i++) {
+    if (prec) {
+      if (length < DCTSIZE2 * 2) {
+	/* Initialize full table for safety. */
+	for (i = 0; i < DCTSIZE2; i++) {
+	  quant_ptr->quantval[i] = 1;
+	}
+	count = length >> 1;
+      } else
+	count = DCTSIZE2;
+    } else {
+      if (length < DCTSIZE2) {
+	/* Initialize full table for safety. */
+	for (i = 0; i < DCTSIZE2; i++) {
+	  quant_ptr->quantval[i] = 1;
+	}
+	count = length;
+      } else
+	count = DCTSIZE2;
+    }
+
+    switch (count) {
+    case (2*2): natural_order = jpeg_natural_order2; break;
+    case (3*3): natural_order = jpeg_natural_order3; break;
+    case (4*4): natural_order = jpeg_natural_order4; break;
+    case (5*5): natural_order = jpeg_natural_order5; break;
+    case (6*6): natural_order = jpeg_natural_order6; break;
+    case (7*7): natural_order = jpeg_natural_order7; break;
+    default:    natural_order = jpeg_natural_order;  break;
+    }
+
+    for (i = 0; i < count; i++) {
       if (prec)
 	INPUT_2BYTES(cinfo, tmp, return FALSE);
       else
 	INPUT_BYTE(cinfo, tmp, return FALSE);
       /* We convert the zigzag-order table to natural array order. */
-      quant_ptr->quantval[jpeg_natural_order[i]] = (UINT16) tmp;
+      quant_ptr->quantval[natural_order[i]] = (UINT16) tmp;
     }
 
     if (cinfo->err->trace_level >= 2) {
@@ -532,8 +569,8 @@ get_dqt (j_decompress_ptr cinfo)
       }
     }
 
-    length -= DCTSIZE2+1;
-    if (prec) length -= DCTSIZE2;
+    length -= count;
+    if (prec) length -= count;
   }
 
   if (length != 0)
@@ -946,6 +983,11 @@ first_marker (j_decompress_ptr cinfo)
  *
  * Returns same codes as are defined for jpeg_consume_input:
  * JPEG_SUSPENDED, JPEG_REACHED_SOS, or JPEG_REACHED_EOI.
+ *
+ * Note: This function may return a pseudo SOS marker (with zero
+ * component number) for treat by input controller's consume_input.
+ * consume_input itself should filter out (skip) the pseudo marker
+ * after processing for the caller.
  */
 
 METHODDEF(int)
@@ -975,23 +1017,27 @@ read_markers (j_decompress_ptr cinfo)
       break;
 
     case M_SOF0:		/* Baseline */
+      if (! get_sof(cinfo, TRUE, FALSE, FALSE))
+	return JPEG_SUSPENDED;
+      break;
+
     case M_SOF1:		/* Extended sequential, Huffman */
-      if (! get_sof(cinfo, FALSE, FALSE))
+      if (! get_sof(cinfo, FALSE, FALSE, FALSE))
 	return JPEG_SUSPENDED;
       break;
 
     case M_SOF2:		/* Progressive, Huffman */
-      if (! get_sof(cinfo, TRUE, FALSE))
+      if (! get_sof(cinfo, FALSE, TRUE, FALSE))
 	return JPEG_SUSPENDED;
       break;
 
     case M_SOF9:		/* Extended sequential, arithmetic */
-      if (! get_sof(cinfo, FALSE, TRUE))
+      if (! get_sof(cinfo, FALSE, FALSE, TRUE))
 	return JPEG_SUSPENDED;
       break;
 
     case M_SOF10:		/* Progressive, arithmetic */
-      if (! get_sof(cinfo, TRUE, TRUE))
+      if (! get_sof(cinfo, FALSE, TRUE, TRUE))
 	return JPEG_SUSPENDED;
       break;
 
diff --git a/src/jpeg-6b/jdmaster.c b/src/jpeg-8c/jdmaster.c
index 2802c5b7..8c1146e4 100644
--- a/src/jpeg-6b/jdmaster.c
+++ b/src/jpeg-8c/jdmaster.c
@@ -2,6 +2,7 @@
  * jdmaster.c
  *
  * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Modified 2002-2009 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -61,9 +62,12 @@ use_merged_upsample (j_decompress_ptr cinfo)
       cinfo->comp_info[2].v_samp_factor != 1)
     return FALSE;
   /* furthermore, it doesn't work if we've scaled the IDCTs differently */
-  if (cinfo->comp_info[0].DCT_scaled_size != cinfo->min_DCT_scaled_size ||
-      cinfo->comp_info[1].DCT_scaled_size != cinfo->min_DCT_scaled_size ||
-      cinfo->comp_info[2].DCT_scaled_size != cinfo->min_DCT_scaled_size)
+  if (cinfo->comp_info[0].DCT_h_scaled_size != cinfo->min_DCT_h_scaled_size ||
+      cinfo->comp_info[1].DCT_h_scaled_size != cinfo->min_DCT_h_scaled_size ||
+      cinfo->comp_info[2].DCT_h_scaled_size != cinfo->min_DCT_h_scaled_size ||
+      cinfo->comp_info[0].DCT_v_scaled_size != cinfo->min_DCT_v_scaled_size ||
+      cinfo->comp_info[1].DCT_v_scaled_size != cinfo->min_DCT_v_scaled_size ||
+      cinfo->comp_info[2].DCT_v_scaled_size != cinfo->min_DCT_v_scaled_size)
     return FALSE;
   /* ??? also need to test for upsample-time rescaling, when & if supported */
   return TRUE;			/* by golly, it'll work... */
@@ -82,7 +86,9 @@ use_merged_upsample (j_decompress_ptr cinfo)
 
 GLOBAL(void)
 jpeg_calc_output_dimensions (j_decompress_ptr cinfo)
-/* Do computations that are needed before master selection phase */
+/* Do computations that are needed before master selection phase.
+ * This function is used for full decompression.
+ */
 {
 #ifdef IDCT_SCALING_SUPPORTED
   int ci;
@@ -93,52 +99,38 @@ jpeg_calc_output_dimensions (j_decompress_ptr cinfo)
   if (cinfo->global_state != DSTATE_READY)
     ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
 
+  /* Compute core output image dimensions and DCT scaling choices. */
+  jpeg_core_output_dimensions(cinfo);
+
 #ifdef IDCT_SCALING_SUPPORTED
 
-  /* Compute actual output image dimensions and DCT scaling choices. */
-  if (cinfo->scale_num * 8 <= cinfo->scale_denom) {
-    /* Provide 1/8 scaling */
-    cinfo->output_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width, 8L);
-    cinfo->output_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height, 8L);
-    cinfo->min_DCT_scaled_size = 1;
-  } else if (cinfo->scale_num * 4 <= cinfo->scale_denom) {
-    /* Provide 1/4 scaling */
-    cinfo->output_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width, 4L);
-    cinfo->output_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height, 4L);
-    cinfo->min_DCT_scaled_size = 2;
-  } else if (cinfo->scale_num * 2 <= cinfo->scale_denom) {
-    /* Provide 1/2 scaling */
-    cinfo->output_width = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_width, 2L);
-    cinfo->output_height = (JDIMENSION)
-      jdiv_round_up((long) cinfo->image_height, 2L);
-    cinfo->min_DCT_scaled_size = 4;
-  } else {
-    /* Provide 1/1 scaling */
-    cinfo->output_width = cinfo->image_width;
-    cinfo->output_height = cinfo->image_height;
-    cinfo->min_DCT_scaled_size = DCTSIZE;
-  }
   /* In selecting the actual DCT scaling for each component, we try to
    * scale up the chroma components via IDCT scaling rather than upsampling.
    * This saves time if the upsampler gets to use 1:1 scaling.
-   * Note this code assumes that the supported DCT scalings are powers of 2.
+   * Note this code adapts subsampling ratios which are powers of 2.
    */
   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
        ci++, compptr++) {
-    int ssize = cinfo->min_DCT_scaled_size;
-    while (ssize < DCTSIZE &&
-	   (compptr->h_samp_factor * ssize * 2 <=
-	    cinfo->max_h_samp_factor * cinfo->min_DCT_scaled_size) &&
-	   (compptr->v_samp_factor * ssize * 2 <=
-	    cinfo->max_v_samp_factor * cinfo->min_DCT_scaled_size)) {
+    int ssize = 1;
+    while (cinfo->min_DCT_h_scaled_size * ssize <=
+	   (cinfo->do_fancy_upsampling ? DCTSIZE : DCTSIZE / 2) &&
+	   (cinfo->max_h_samp_factor % (compptr->h_samp_factor * ssize * 2)) == 0) {
       ssize = ssize * 2;
     }
-    compptr->DCT_scaled_size = ssize;
+    compptr->DCT_h_scaled_size = cinfo->min_DCT_h_scaled_size * ssize;
+    ssize = 1;
+    while (cinfo->min_DCT_v_scaled_size * ssize <=
+	   (cinfo->do_fancy_upsampling ? DCTSIZE : DCTSIZE / 2) &&
+	   (cinfo->max_v_samp_factor % (compptr->v_samp_factor * ssize * 2)) == 0) {
+      ssize = ssize * 2;
+    }
+    compptr->DCT_v_scaled_size = cinfo->min_DCT_v_scaled_size * ssize;
+
+    /* We don't support IDCT ratios larger than 2. */
+    if (compptr->DCT_h_scaled_size > compptr->DCT_v_scaled_size * 2)
+	compptr->DCT_h_scaled_size = compptr->DCT_v_scaled_size * 2;
+    else if (compptr->DCT_v_scaled_size > compptr->DCT_h_scaled_size * 2)
+	compptr->DCT_v_scaled_size = compptr->DCT_h_scaled_size * 2;
   }
 
   /* Recompute downsampled dimensions of components;
@@ -149,23 +141,14 @@ jpeg_calc_output_dimensions (j_decompress_ptr cinfo)
     /* Size in samples, after IDCT scaling */
     compptr->downsampled_width = (JDIMENSION)
       jdiv_round_up((long) cinfo->image_width *
-		    (long) (compptr->h_samp_factor * compptr->DCT_scaled_size),
-		    (long) (cinfo->max_h_samp_factor * DCTSIZE));
+		    (long) (compptr->h_samp_factor * compptr->DCT_h_scaled_size),
+		    (long) (cinfo->max_h_samp_factor * cinfo->block_size));
     compptr->downsampled_height = (JDIMENSION)
       jdiv_round_up((long) cinfo->image_height *
-		    (long) (compptr->v_samp_factor * compptr->DCT_scaled_size),
-		    (long) (cinfo->max_v_samp_factor * DCTSIZE));
+		    (long) (compptr->v_samp_factor * compptr->DCT_v_scaled_size),
+		    (long) (cinfo->max_v_samp_factor * cinfo->block_size));
   }
 
-#else /* !IDCT_SCALING_SUPPORTED */
-
-  /* Hardwire it to "no scaling" */
-  cinfo->output_width = cinfo->image_width;
-  cinfo->output_height = cinfo->image_height;
-  /* jdinput.c has already initialized DCT_scaled_size to DCTSIZE,
-   * and has computed unscaled downsampled_width and downsampled_height.
-   */
-
 #endif /* IDCT_SCALING_SUPPORTED */
 
   /* Report number of components in selected colorspace. */
@@ -372,17 +355,10 @@ master_selection (j_decompress_ptr cinfo)
   /* Inverse DCT */
   jinit_inverse_dct(cinfo);
   /* Entropy decoding: either Huffman or arithmetic coding. */
-  if (cinfo->arith_code) {
-    ERREXIT(cinfo, JERR_ARITH_NOTIMPL);
-  } else {
-    if (cinfo->progressive_mode) {
-#ifdef D_PROGRESSIVE_SUPPORTED
-      jinit_phuff_decoder(cinfo);
-#else
-      ERREXIT(cinfo, JERR_NOT_COMPILED);
-#endif
-    } else
-      jinit_huff_decoder(cinfo);
+  if (cinfo->arith_code)
+    jinit_arith_decoder(cinfo);
+  else {
+    jinit_huff_decoder(cinfo);
   }
 
   /* Initialize principal buffer controllers. */
diff --git a/src/jpeg-6b/jdmerge.c b/src/jpeg-8c/jdmerge.c
index 37444468..37444468 100644
--- a/src/jpeg-6b/jdmerge.c
+++ b/src/jpeg-8c/jdmerge.c
diff --git a/src/jpeg-6b/jdpostct.c b/src/jpeg-8c/jdpostct.c
index 571563d7..571563d7 100644
--- a/src/jpeg-6b/jdpostct.c
+++ b/src/jpeg-8c/jdpostct.c
diff --git a/src/jpeg-6b/jdsample.c b/src/jpeg-8c/jdsample.c
index 80ffefb2..7bc8885b 100644
--- a/src/jpeg-6b/jdsample.c
+++ b/src/jpeg-8c/jdsample.c
@@ -2,13 +2,14 @@
  * jdsample.c
  *
  * Copyright (C) 1991-1996, Thomas G. Lane.
+ * Modified 2002-2008 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
  * This file contains upsampling routines.
  *
  * Upsampling input data is counted in "row groups".  A row group
- * is defined to be (v_samp_factor * DCT_scaled_size / min_DCT_scaled_size)
+ * is defined to be (v_samp_factor * DCT_v_scaled_size / min_DCT_v_scaled_size)
  * sample rows of each component.  Upsampling will normally produce
  * max_v_samp_factor pixel rows from each row group (but this could vary
  * if the upsampler is applying a scale factor of its own).
@@ -237,11 +238,11 @@ h2v1_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
   register JSAMPROW inptr, outptr;
   register JSAMPLE invalue;
   JSAMPROW outend;
-  int inrow;
+  int outrow;
 
-  for (inrow = 0; inrow < cinfo->max_v_samp_factor; inrow++) {
-    inptr = input_data[inrow];
-    outptr = output_data[inrow];
+  for (outrow = 0; outrow < cinfo->max_v_samp_factor; outrow++) {
+    inptr = input_data[outrow];
+    outptr = output_data[outrow];
     outend = outptr + cinfo->output_width;
     while (outptr < outend) {
       invalue = *inptr++;	/* don't need GETJSAMPLE() here */
@@ -286,112 +287,6 @@ h2v2_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
 
 
 /*
- * Fancy processing for the common case of 2:1 horizontal and 1:1 vertical.
- *
- * The upsampling algorithm is linear interpolation between pixel centers,
- * also known as a "triangle filter".  This is a good compromise between
- * speed and visual quality.  The centers of the output pixels are 1/4 and 3/4
- * of the way between input pixel centers.
- *
- * A note about the "bias" calculations: when rounding fractional values to
- * integer, we do not want to always round 0.5 up to the next integer.
- * If we did that, we'd introduce a noticeable bias towards larger values.
- * Instead, this code is arranged so that 0.5 will be rounded up or down at
- * alternate pixel locations (a simple ordered dither pattern).
- */
-
-METHODDEF(void)
-h2v1_fancy_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
-		     JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
-{
-  JSAMPARRAY output_data = *output_data_ptr;
-  register JSAMPROW inptr, outptr;
-  register int invalue;
-  register JDIMENSION colctr;
-  int inrow;
-
-  for (inrow = 0; inrow < cinfo->max_v_samp_factor; inrow++) {
-    inptr = input_data[inrow];
-    outptr = output_data[inrow];
-    /* Special case for first column */
-    invalue = GETJSAMPLE(*inptr++);
-    *outptr++ = (JSAMPLE) invalue;
-    *outptr++ = (JSAMPLE) ((invalue * 3 + GETJSAMPLE(*inptr) + 2) >> 2);
-
-    for (colctr = compptr->downsampled_width - 2; colctr > 0; colctr--) {
-      /* General case: 3/4 * nearer pixel + 1/4 * further pixel */
-      invalue = GETJSAMPLE(*inptr++) * 3;
-      *outptr++ = (JSAMPLE) ((invalue + GETJSAMPLE(inptr[-2]) + 1) >> 2);
-      *outptr++ = (JSAMPLE) ((invalue + GETJSAMPLE(*inptr) + 2) >> 2);
-    }
-
-    /* Special case for last column */
-    invalue = GETJSAMPLE(*inptr);
-    *outptr++ = (JSAMPLE) ((invalue * 3 + GETJSAMPLE(inptr[-1]) + 1) >> 2);
-    *outptr++ = (JSAMPLE) invalue;
-  }
-}
-
-
-/*
- * Fancy processing for the common case of 2:1 horizontal and 2:1 vertical.
- * Again a triangle filter; see comments for h2v1 case, above.
- *
- * It is OK for us to reference the adjacent input rows because we demanded
- * context from the main buffer controller (see initialization code).
- */
-
-METHODDEF(void)
-h2v2_fancy_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
-		     JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
-{
-  JSAMPARRAY output_data = *output_data_ptr;
-  register JSAMPROW inptr0, inptr1, outptr;
-#if BITS_IN_JSAMPLE == 8
-  register int thiscolsum, lastcolsum, nextcolsum;
-#else
-  register INT32 thiscolsum, lastcolsum, nextcolsum;
-#endif
-  register JDIMENSION colctr;
-  int inrow, outrow, v;
-
-  inrow = outrow = 0;
-  while (outrow < cinfo->max_v_samp_factor) {
-    for (v = 0; v < 2; v++) {
-      /* inptr0 points to nearest input row, inptr1 points to next nearest */
-      inptr0 = input_data[inrow];
-      if (v == 0)		/* next nearest is row above */
-	inptr1 = input_data[inrow-1];
-      else			/* next nearest is row below */
-	inptr1 = input_data[inrow+1];
-      outptr = output_data[outrow++];
-
-      /* Special case for first column */
-      thiscolsum = GETJSAMPLE(*inptr0++) * 3 + GETJSAMPLE(*inptr1++);
-      nextcolsum = GETJSAMPLE(*inptr0++) * 3 + GETJSAMPLE(*inptr1++);
-      *outptr++ = (JSAMPLE) ((thiscolsum * 4 + 8) >> 4);
-      *outptr++ = (JSAMPLE) ((thiscolsum * 3 + nextcolsum + 7) >> 4);
-      lastcolsum = thiscolsum; thiscolsum = nextcolsum;
-
-      for (colctr = compptr->downsampled_width - 2; colctr > 0; colctr--) {
-	/* General case: 3/4 * nearer pixel + 1/4 * further pixel in each */
-	/* dimension, thus 9/16, 3/16, 3/16, 1/16 overall */
-	nextcolsum = GETJSAMPLE(*inptr0++) * 3 + GETJSAMPLE(*inptr1++);
-	*outptr++ = (JSAMPLE) ((thiscolsum * 3 + lastcolsum + 8) >> 4);
-	*outptr++ = (JSAMPLE) ((thiscolsum * 3 + nextcolsum + 7) >> 4);
-	lastcolsum = thiscolsum; thiscolsum = nextcolsum;
-      }
-
-      /* Special case for last column */
-      *outptr++ = (JSAMPLE) ((thiscolsum * 3 + lastcolsum + 8) >> 4);
-      *outptr++ = (JSAMPLE) ((thiscolsum * 4 + 7) >> 4);
-    }
-    inrow++;
-  }
-}
-
-
-/*
  * Module initialization routine for upsampling.
  */
 
@@ -401,7 +296,7 @@ jinit_upsampler (j_decompress_ptr cinfo)
   my_upsample_ptr upsample;
   int ci;
   jpeg_component_info * compptr;
-  boolean need_buffer, do_fancy;
+  boolean need_buffer;
   int h_in_group, v_in_group, h_out_group, v_out_group;
 
   upsample = (my_upsample_ptr)
@@ -415,11 +310,6 @@ jinit_upsampler (j_decompress_ptr cinfo)
   if (cinfo->CCIR601_sampling)	/* this isn't supported */
     ERREXIT(cinfo, JERR_CCIR601_NOTIMPL);
 
-  /* jdmainct.c doesn't support context rows when min_DCT_scaled_size = 1,
-   * so don't ask for it.
-   */
-  do_fancy = cinfo->do_fancy_upsampling && cinfo->min_DCT_scaled_size > 1;
-
   /* Verify we can handle the sampling factors, select per-component methods,
    * and create storage as needed.
    */
@@ -428,10 +318,10 @@ jinit_upsampler (j_decompress_ptr cinfo)
     /* Compute size of an "input group" after IDCT scaling.  This many samples
      * are to be converted to max_h_samp_factor * max_v_samp_factor pixels.
      */
-    h_in_group = (compptr->h_samp_factor * compptr->DCT_scaled_size) /
-		 cinfo->min_DCT_scaled_size;
-    v_in_group = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
-		 cinfo->min_DCT_scaled_size;
+    h_in_group = (compptr->h_samp_factor * compptr->DCT_h_scaled_size) /
+		 cinfo->min_DCT_h_scaled_size;
+    v_in_group = (compptr->v_samp_factor * compptr->DCT_v_scaled_size) /
+		 cinfo->min_DCT_v_scaled_size;
     h_out_group = cinfo->max_h_samp_factor;
     v_out_group = cinfo->max_v_samp_factor;
     upsample->rowgroup_height[ci] = v_in_group; /* save for use later */
@@ -446,19 +336,12 @@ jinit_upsampler (j_decompress_ptr cinfo)
       need_buffer = FALSE;
     } else if (h_in_group * 2 == h_out_group &&
 	       v_in_group == v_out_group) {
-      /* Special cases for 2h1v upsampling */
-      if (do_fancy && compptr->downsampled_width > 2)
-	upsample->methods[ci] = h2v1_fancy_upsample;
-      else
-	upsample->methods[ci] = h2v1_upsample;
+      /* Special case for 2h1v upsampling */
+      upsample->methods[ci] = h2v1_upsample;
     } else if (h_in_group * 2 == h_out_group &&
 	       v_in_group * 2 == v_out_group) {
-      /* Special cases for 2h2v upsampling */
-      if (do_fancy && compptr->downsampled_width > 2) {
-	upsample->methods[ci] = h2v2_fancy_upsample;
-	upsample->pub.need_context_rows = TRUE;
-      } else
-	upsample->methods[ci] = h2v2_upsample;
+      /* Special case for 2h2v upsampling */
+      upsample->methods[ci] = h2v2_upsample;
     } else if ((h_out_group % h_in_group) == 0 &&
 	       (v_out_group % v_in_group) == 0) {
       /* Generic integral-factors upsampling method */
diff --git a/src/jpeg-6b/jdtrans.c b/src/jpeg-8c/jdtrans.c
index 6c0ab715..22dd47fb 100644
--- a/src/jpeg-6b/jdtrans.c
+++ b/src/jpeg-8c/jdtrans.c
@@ -2,6 +2,7 @@
  * jdtrans.c
  *
  * Copyright (C) 1995-1997, Thomas G. Lane.
+ * Modified 2000-2009 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -99,18 +100,14 @@ transdecode_master_selection (j_decompress_ptr cinfo)
   /* This is effectively a buffered-image operation. */
   cinfo->buffered_image = TRUE;
 
+  /* Compute output image dimensions and related values. */
+  jpeg_core_output_dimensions(cinfo);
+
   /* Entropy decoding: either Huffman or arithmetic coding. */
-  if (cinfo->arith_code) {
-    ERREXIT(cinfo, JERR_ARITH_NOTIMPL);
-  } else {
-    if (cinfo->progressive_mode) {
-#ifdef D_PROGRESSIVE_SUPPORTED
-      jinit_phuff_decoder(cinfo);
-#else
-      ERREXIT(cinfo, JERR_NOT_COMPILED);
-#endif
-    } else
-      jinit_huff_decoder(cinfo);
+  if (cinfo->arith_code)
+    jinit_arith_decoder(cinfo);
+  else {
+    jinit_huff_decoder(cinfo);
   }
 
   /* Always get a full-image coefficient buffer. */
diff --git a/src/jpeg-6b/jerror.c b/src/jpeg-8c/jerror.c
index 9df35d69..2860d3fa 100644
--- a/src/jpeg-6b/jerror.c
+++ b/src/jpeg-8c/jerror.c
@@ -18,14 +18,14 @@
  * These routines are used by both the compression and decompression code.
  */
 
-#include "../renderer/tr_local.h"
-
 /* this is not a core library module, so it doesn't define JPEG_INTERNALS */
 #include "jinclude.h"
 #include "jpeglib.h"
 #include "jversion.h"
 #include "jerror.h"
 
+#include <stdlib.h>
+
 #ifdef USE_WINDOWS_MESSAGEBOX
 #include <windows.h>
 #endif
@@ -71,15 +71,13 @@ const char * const jpeg_std_message_table[] = {
 METHODDEF(void)
 error_exit (j_common_ptr cinfo)
 {
-  char buffer[JMSG_LENGTH_MAX];
-
-  /* Create the message */
-  (*cinfo->err->format_message) (cinfo, buffer);
+  /* Always display the message */
+  (*cinfo->err->output_message) (cinfo);
 
   /* Let the memory manager delete any temp files before we die */
   jpeg_destroy(cinfo);
 
-  ri.Error( ERR_FATAL, "%s\n", buffer );
+  exit(EXIT_FAILURE);
 }
 
 
@@ -112,7 +110,7 @@ output_message (j_common_ptr cinfo)
 	     MB_OK | MB_ICONERROR);
 #else
   /* Send it to stderr, adding a newline */
-  ri.Printf(PRINT_ALL, "%s\n", buffer);
+  fprintf(stderr, "%s\n", buffer);
 #endif
 }
 
diff --git a/src/jpeg-6b/jerror.h b/src/jpeg-8c/jerror.h
index fc2fffea..1cfb2b19 100644
--- a/src/jpeg-6b/jerror.h
+++ b/src/jpeg-8c/jerror.h
@@ -2,6 +2,7 @@
  * jerror.h
  *
  * Copyright (C) 1994-1997, Thomas G. Lane.
+ * Modified 1997-2009 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -39,14 +40,15 @@ typedef enum {
 JMESSAGE(JMSG_NOMESSAGE, "Bogus message code %d") /* Must be first entry! */
 
 /* For maintenance convenience, list is alphabetical by message code name */
-JMESSAGE(JERR_ARITH_NOTIMPL,
-	 "Sorry, there are legal restrictions on arithmetic coding")
 JMESSAGE(JERR_BAD_ALIGN_TYPE, "ALIGN_TYPE is wrong, please fix")
 JMESSAGE(JERR_BAD_ALLOC_CHUNK, "MAX_ALLOC_CHUNK is wrong, please fix")
 JMESSAGE(JERR_BAD_BUFFER_MODE, "Bogus buffer control mode")
 JMESSAGE(JERR_BAD_COMPONENT_ID, "Invalid component ID %d in SOS")
+JMESSAGE(JERR_BAD_CROP_SPEC, "Invalid crop request")
 JMESSAGE(JERR_BAD_DCT_COEF, "DCT coefficient out of range")
-JMESSAGE(JERR_BAD_DCTSIZE, "IDCT output block size %d not supported")
+JMESSAGE(JERR_BAD_DCTSIZE, "DCT scaled block size %dx%d not supported")
+JMESSAGE(JERR_BAD_DROP_SAMPLING,
+	 "Component index %d: mismatching sampling ratio %d:%d, %d:%d, %c")
 JMESSAGE(JERR_BAD_HUFF_TABLE, "Bogus Huffman table definition")
 JMESSAGE(JERR_BAD_IN_COLORSPACE, "Bogus input colorspace")
 JMESSAGE(JERR_BAD_J_COLORSPACE, "Bogus JPEG colorspace")
@@ -93,6 +95,7 @@ JMESSAGE(JERR_MISSING_DATA, "Scan script does not transmit all data")
 JMESSAGE(JERR_MODE_CHANGE, "Invalid color quantization mode change")
 JMESSAGE(JERR_NOTIMPL, "Not implemented yet")
 JMESSAGE(JERR_NOT_COMPILED, "Requested feature was omitted at compile time")
+JMESSAGE(JERR_NO_ARITH_TABLE, "Arithmetic table 0x%02x was not defined")
 JMESSAGE(JERR_NO_BACKING_STORE, "Backing store not supported")
 JMESSAGE(JERR_NO_HUFF_TABLE, "Huffman table 0x%02x was not defined")
 JMESSAGE(JERR_NO_IMAGE, "JPEG datastream contains no image")
@@ -170,6 +173,7 @@ JMESSAGE(JTRC_UNKNOWN_IDS,
 JMESSAGE(JTRC_XMS_CLOSE, "Freed XMS handle %u")
 JMESSAGE(JTRC_XMS_OPEN, "Obtained XMS handle %u")
 JMESSAGE(JWRN_ADOBE_XFORM, "Unknown Adobe color transform code %d")
+JMESSAGE(JWRN_ARITH_BAD_CODE, "Corrupt JPEG data: bad arithmetic code")
 JMESSAGE(JWRN_BOGUS_PROGRESSION,
 	 "Inconsistent progression sequence for component %d coefficient %d")
 JMESSAGE(JWRN_EXTRANEOUS_DATA,
@@ -227,6 +231,15 @@ JMESSAGE(JWRN_TOO_MUCH_DATA, "Application transferred too many scanlines")
    (cinfo)->err->msg_parm.i[2] = (p3), \
    (cinfo)->err->msg_parm.i[3] = (p4), \
    (*(cinfo)->err->error_exit) ((j_common_ptr) (cinfo)))
+#define ERREXIT6(cinfo,code,p1,p2,p3,p4,p5,p6)  \
+  ((cinfo)->err->msg_code = (code), \
+   (cinfo)->err->msg_parm.i[0] = (p1), \
+   (cinfo)->err->msg_parm.i[1] = (p2), \
+   (cinfo)->err->msg_parm.i[2] = (p3), \
+   (cinfo)->err->msg_parm.i[3] = (p4), \
+   (cinfo)->err->msg_parm.i[4] = (p5), \
+   (cinfo)->err->msg_parm.i[5] = (p6), \
+   (*(cinfo)->err->error_exit) ((j_common_ptr) (cinfo)))
 #define ERREXITS(cinfo,code,str)  \
   ((cinfo)->err->msg_code = (code), \
    strncpy((cinfo)->err->msg_parm.s, (str), JMSG_STR_PARM_MAX), \
diff --git a/src/jpeg-6b/jfdctflt.c b/src/jpeg-8c/jfdctflt.c
index 79d7a007..74d0d862 100644
--- a/src/jpeg-6b/jfdctflt.c
+++ b/src/jpeg-8c/jfdctflt.c
@@ -2,6 +2,7 @@
  * jfdctflt.c
  *
  * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2003-2009 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -56,41 +57,46 @@
  */
 
 GLOBAL(void)
-jpeg_fdct_float (FAST_FLOAT * data)
+jpeg_fdct_float (FAST_FLOAT * data, JSAMPARRAY sample_data, JDIMENSION start_col)
 {
   FAST_FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
   FAST_FLOAT tmp10, tmp11, tmp12, tmp13;
   FAST_FLOAT z1, z2, z3, z4, z5, z11, z13;
   FAST_FLOAT *dataptr;
+  JSAMPROW elemptr;
   int ctr;
 
   /* Pass 1: process rows. */
 
   dataptr = data;
-  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
-    tmp0 = dataptr[0] + dataptr[7];
-    tmp7 = dataptr[0] - dataptr[7];
-    tmp1 = dataptr[1] + dataptr[6];
-    tmp6 = dataptr[1] - dataptr[6];
-    tmp2 = dataptr[2] + dataptr[5];
-    tmp5 = dataptr[2] - dataptr[5];
-    tmp3 = dataptr[3] + dataptr[4];
-    tmp4 = dataptr[3] - dataptr[4];
-    
+  for (ctr = 0; ctr < DCTSIZE; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Load data into workspace */
+    tmp0 = (FAST_FLOAT) (GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[7]));
+    tmp7 = (FAST_FLOAT) (GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[7]));
+    tmp1 = (FAST_FLOAT) (GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[6]));
+    tmp6 = (FAST_FLOAT) (GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[6]));
+    tmp2 = (FAST_FLOAT) (GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[5]));
+    tmp5 = (FAST_FLOAT) (GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[5]));
+    tmp3 = (FAST_FLOAT) (GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[4]));
+    tmp4 = (FAST_FLOAT) (GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[4]));
+
     /* Even part */
-    
+
     tmp10 = tmp0 + tmp3;	/* phase 2 */
     tmp13 = tmp0 - tmp3;
     tmp11 = tmp1 + tmp2;
     tmp12 = tmp1 - tmp2;
-    
-    dataptr[0] = tmp10 + tmp11; /* phase 3 */
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = tmp10 + tmp11 - 8 * CENTERJSAMPLE; /* phase 3 */
     dataptr[4] = tmp10 - tmp11;
-    
+
     z1 = (tmp12 + tmp13) * ((FAST_FLOAT) 0.707106781); /* c4 */
     dataptr[2] = tmp13 + z1;	/* phase 5 */
     dataptr[6] = tmp13 - z1;
-    
+
     /* Odd part */
 
     tmp10 = tmp4 + tmp5;	/* phase 2 */
@@ -126,21 +132,21 @@ jpeg_fdct_float (FAST_FLOAT * data)
     tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
     tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
     tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
-    
+
     /* Even part */
-    
+
     tmp10 = tmp0 + tmp3;	/* phase 2 */
     tmp13 = tmp0 - tmp3;
     tmp11 = tmp1 + tmp2;
     tmp12 = tmp1 - tmp2;
-    
+
     dataptr[DCTSIZE*0] = tmp10 + tmp11; /* phase 3 */
     dataptr[DCTSIZE*4] = tmp10 - tmp11;
-    
+
     z1 = (tmp12 + tmp13) * ((FAST_FLOAT) 0.707106781); /* c4 */
     dataptr[DCTSIZE*2] = tmp13 + z1; /* phase 5 */
     dataptr[DCTSIZE*6] = tmp13 - z1;
-    
+
     /* Odd part */
 
     tmp10 = tmp4 + tmp5;	/* phase 2 */
diff --git a/src/jpeg-6b/jfdctfst.c b/src/jpeg-8c/jfdctfst.c
index ccb378a3..8cad5f22 100644
--- a/src/jpeg-6b/jfdctfst.c
+++ b/src/jpeg-8c/jfdctfst.c
@@ -2,6 +2,7 @@
  * jfdctfst.c
  *
  * Copyright (C) 1994-1996, Thomas G. Lane.
+ * Modified 2003-2009 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -111,42 +112,47 @@
  */
 
 GLOBAL(void)
-jpeg_fdct_ifast (DCTELEM * data)
+jpeg_fdct_ifast (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
 {
   DCTELEM tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
   DCTELEM tmp10, tmp11, tmp12, tmp13;
   DCTELEM z1, z2, z3, z4, z5, z11, z13;
   DCTELEM *dataptr;
+  JSAMPROW elemptr;
   int ctr;
   SHIFT_TEMPS
 
   /* Pass 1: process rows. */
 
   dataptr = data;
-  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
-    tmp0 = dataptr[0] + dataptr[7];
-    tmp7 = dataptr[0] - dataptr[7];
-    tmp1 = dataptr[1] + dataptr[6];
-    tmp6 = dataptr[1] - dataptr[6];
-    tmp2 = dataptr[2] + dataptr[5];
-    tmp5 = dataptr[2] - dataptr[5];
-    tmp3 = dataptr[3] + dataptr[4];
-    tmp4 = dataptr[3] - dataptr[4];
-    
+  for (ctr = 0; ctr < DCTSIZE; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Load data into workspace */
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[7]);
+    tmp7 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[7]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[6]);
+    tmp6 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[6]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[5]);
+    tmp5 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[5]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[4]);
+    tmp4 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[4]);
+
     /* Even part */
-    
+
     tmp10 = tmp0 + tmp3;	/* phase 2 */
     tmp13 = tmp0 - tmp3;
     tmp11 = tmp1 + tmp2;
     tmp12 = tmp1 - tmp2;
-    
-    dataptr[0] = tmp10 + tmp11; /* phase 3 */
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = tmp10 + tmp11 - 8 * CENTERJSAMPLE; /* phase 3 */
     dataptr[4] = tmp10 - tmp11;
-    
+
     z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
     dataptr[2] = tmp13 + z1;	/* phase 5 */
     dataptr[6] = tmp13 - z1;
-    
+
     /* Odd part */
 
     tmp10 = tmp4 + tmp5;	/* phase 2 */
@@ -182,21 +188,21 @@ jpeg_fdct_ifast (DCTELEM * data)
     tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
     tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
     tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
-    
+
     /* Even part */
-    
+
     tmp10 = tmp0 + tmp3;	/* phase 2 */
     tmp13 = tmp0 - tmp3;
     tmp11 = tmp1 + tmp2;
     tmp12 = tmp1 - tmp2;
-    
+
     dataptr[DCTSIZE*0] = tmp10 + tmp11; /* phase 3 */
     dataptr[DCTSIZE*4] = tmp10 - tmp11;
-    
+
     z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
     dataptr[DCTSIZE*2] = tmp13 + z1; /* phase 5 */
     dataptr[DCTSIZE*6] = tmp13 - z1;
-    
+
     /* Odd part */
 
     tmp10 = tmp4 + tmp5;	/* phase 2 */
diff --git a/src/jpeg-8c/jfdctint.c b/src/jpeg-8c/jfdctint.c
new file mode 100644
index 00000000..1dde58c4
--- /dev/null
+++ b/src/jpeg-8c/jfdctint.c
@@ -0,0 +1,4348 @@
+/*
+ * jfdctint.c
+ *
+ * Copyright (C) 1991-1996, Thomas G. Lane.
+ * Modification developed 2003-2009 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains a slow-but-accurate integer implementation of the
+ * forward DCT (Discrete Cosine Transform).
+ *
+ * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT
+ * on each column.  Direct algorithms are also available, but they are
+ * much more complex and seem not to be any faster when reduced to code.
+ *
+ * This implementation is based on an algorithm described in
+ *   C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
+ *   Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
+ *   Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
+ * The primary algorithm described there uses 11 multiplies and 29 adds.
+ * We use their alternate method with 12 multiplies and 32 adds.
+ * The advantage of this method is that no data path contains more than one
+ * multiplication; this allows a very simple and accurate implementation in
+ * scaled fixed-point arithmetic, with a minimal number of shifts.
+ *
+ * We also provide FDCT routines with various input sample block sizes for
+ * direct resolution reduction or enlargement and for direct resolving the
+ * common 2x1 and 1x2 subsampling cases without additional resampling: NxN
+ * (N=1...16), 2NxN, and Nx2N (N=1...8) pixels for one 8x8 output DCT block.
+ *
+ * For N<8 we fill the remaining block coefficients with zero.
+ * For N>8 we apply a partial N-point FDCT on the input samples, computing
+ * just the lower 8 frequency coefficients and discarding the rest.
+ *
+ * We must scale the output coefficients of the N-point FDCT appropriately
+ * to the standard 8-point FDCT level by 8/N per 1-D pass.  This scaling
+ * is folded into the constant multipliers (pass 2) and/or final/initial
+ * shifting.
+ *
+ * CAUTION: We rely on the FIX() macro except for the N=1,2,4,8 cases
+ * since there would be too many additional constants to pre-calculate.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jdct.h"		/* Private declarations for DCT subsystem */
+
+#ifdef DCT_ISLOW_SUPPORTED
+
+
+/*
+ * This module is specialized to the case DCTSIZE = 8.
+ */
+
+#if DCTSIZE != 8
+  Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
+#endif
+
+
+/*
+ * The poop on this scaling stuff is as follows:
+ *
+ * Each 1-D DCT step produces outputs which are a factor of sqrt(N)
+ * larger than the true DCT outputs.  The final outputs are therefore
+ * a factor of N larger than desired; since N=8 this can be cured by
+ * a simple right shift at the end of the algorithm.  The advantage of
+ * this arrangement is that we save two multiplications per 1-D DCT,
+ * because the y0 and y4 outputs need not be divided by sqrt(N).
+ * In the IJG code, this factor of 8 is removed by the quantization step
+ * (in jcdctmgr.c), NOT in this module.
+ *
+ * We have to do addition and subtraction of the integer inputs, which
+ * is no problem, and multiplication by fractional constants, which is
+ * a problem to do in integer arithmetic.  We multiply all the constants
+ * by CONST_SCALE and convert them to integer constants (thus retaining
+ * CONST_BITS bits of precision in the constants).  After doing a
+ * multiplication we have to divide the product by CONST_SCALE, with proper
+ * rounding, to produce the correct output.  This division can be done
+ * cheaply as a right shift of CONST_BITS bits.  We postpone shifting
+ * as long as possible so that partial sums can be added together with
+ * full fractional precision.
+ *
+ * The outputs of the first pass are scaled up by PASS1_BITS bits so that
+ * they are represented to better-than-integral precision.  These outputs
+ * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
+ * with the recommended scaling.  (For 12-bit sample data, the intermediate
+ * array is INT32 anyway.)
+ *
+ * To avoid overflow of the 32-bit intermediate results in pass 2, we must
+ * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26.  Error analysis
+ * shows that the values given below are the most effective.
+ */
+
+#if BITS_IN_JSAMPLE == 8
+#define CONST_BITS  13
+#define PASS1_BITS  2
+#else
+#define CONST_BITS  13
+#define PASS1_BITS  1		/* lose a little precision to avoid overflow */
+#endif
+
+/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
+ * causing a lot of useless floating-point operations at run time.
+ * To get around this we use the following pre-calculated constants.
+ * If you change CONST_BITS you may want to add appropriate values.
+ * (With a reasonable C compiler, you can just rely on the FIX() macro...)
+ */
+
+#if CONST_BITS == 13
+#define FIX_0_298631336  ((INT32)  2446)	/* FIX(0.298631336) */
+#define FIX_0_390180644  ((INT32)  3196)	/* FIX(0.390180644) */
+#define FIX_0_541196100  ((INT32)  4433)	/* FIX(0.541196100) */
+#define FIX_0_765366865  ((INT32)  6270)	/* FIX(0.765366865) */
+#define FIX_0_899976223  ((INT32)  7373)	/* FIX(0.899976223) */
+#define FIX_1_175875602  ((INT32)  9633)	/* FIX(1.175875602) */
+#define FIX_1_501321110  ((INT32)  12299)	/* FIX(1.501321110) */
+#define FIX_1_847759065  ((INT32)  15137)	/* FIX(1.847759065) */
+#define FIX_1_961570560  ((INT32)  16069)	/* FIX(1.961570560) */
+#define FIX_2_053119869  ((INT32)  16819)	/* FIX(2.053119869) */
+#define FIX_2_562915447  ((INT32)  20995)	/* FIX(2.562915447) */
+#define FIX_3_072711026  ((INT32)  25172)	/* FIX(3.072711026) */
+#else
+#define FIX_0_298631336  FIX(0.298631336)
+#define FIX_0_390180644  FIX(0.390180644)
+#define FIX_0_541196100  FIX(0.541196100)
+#define FIX_0_765366865  FIX(0.765366865)
+#define FIX_0_899976223  FIX(0.899976223)
+#define FIX_1_175875602  FIX(1.175875602)
+#define FIX_1_501321110  FIX(1.501321110)
+#define FIX_1_847759065  FIX(1.847759065)
+#define FIX_1_961570560  FIX(1.961570560)
+#define FIX_2_053119869  FIX(2.053119869)
+#define FIX_2_562915447  FIX(2.562915447)
+#define FIX_3_072711026  FIX(3.072711026)
+#endif
+
+
+/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
+ * For 8-bit samples with the recommended scaling, all the variable
+ * and constant values involved are no more than 16 bits wide, so a
+ * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
+ * For 12-bit samples, a full 32-bit multiplication will be needed.
+ */
+
+#if BITS_IN_JSAMPLE == 8
+#define MULTIPLY(var,const)  MULTIPLY16C16(var,const)
+#else
+#define MULTIPLY(var,const)  ((var) * (const))
+#endif
+
+
+/*
+ * Perform the forward DCT on one block of samples.
+ */
+
+GLOBAL(void)
+jpeg_fdct_islow (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3;
+  INT32 tmp10, tmp11, tmp12, tmp13;
+  INT32 z1;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+
+  dataptr = data;
+  for (ctr = 0; ctr < DCTSIZE; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part per LL&M figure 1 --- note that published figure is faulty;
+     * rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
+     */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[7]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[6]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[5]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[4]);
+
+    tmp10 = tmp0 + tmp3;
+    tmp12 = tmp0 - tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp13 = tmp1 - tmp2;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[7]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[6]);
+    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[5]);
+    tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[4]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM) ((tmp10 + tmp11 - 8 * CENTERJSAMPLE) << PASS1_BITS);
+    dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << PASS1_BITS);
+
+    z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
+    /* Add fudge factor here for final descale. */
+    z1 += ONE << (CONST_BITS-PASS1_BITS-1);
+    dataptr[2] = (DCTELEM) RIGHT_SHIFT(z1 + MULTIPLY(tmp12, FIX_0_765366865),
+				       CONST_BITS-PASS1_BITS);
+    dataptr[6] = (DCTELEM) RIGHT_SHIFT(z1 - MULTIPLY(tmp13, FIX_1_847759065),
+				       CONST_BITS-PASS1_BITS);
+
+    /* Odd part per figure 8 --- note paper omits factor of sqrt(2).
+     * cK represents sqrt(2) * cos(K*pi/16).
+     * i0..i3 in the paper are tmp0..tmp3 here.
+     */
+
+    tmp10 = tmp0 + tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp12 = tmp0 + tmp2;
+    tmp13 = tmp1 + tmp3;
+    z1 = MULTIPLY(tmp12 + tmp13, FIX_1_175875602); /*  c3 */
+    /* Add fudge factor here for final descale. */
+    z1 += ONE << (CONST_BITS-PASS1_BITS-1);
+
+    tmp0  = MULTIPLY(tmp0,    FIX_1_501321110);    /*  c1+c3-c5-c7 */
+    tmp1  = MULTIPLY(tmp1,    FIX_3_072711026);    /*  c1+c3+c5-c7 */
+    tmp2  = MULTIPLY(tmp2,    FIX_2_053119869);    /*  c1+c3-c5+c7 */
+    tmp3  = MULTIPLY(tmp3,    FIX_0_298631336);    /* -c1+c3+c5-c7 */
+    tmp10 = MULTIPLY(tmp10, - FIX_0_899976223);    /*  c7-c3 */
+    tmp11 = MULTIPLY(tmp11, - FIX_2_562915447);    /* -c1-c3 */
+    tmp12 = MULTIPLY(tmp12, - FIX_0_390180644);    /*  c5-c3 */
+    tmp13 = MULTIPLY(tmp13, - FIX_1_961570560);    /* -c3-c5 */
+
+    tmp12 += z1;
+    tmp13 += z1;
+
+    dataptr[1] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 + tmp10 + tmp12, CONST_BITS-PASS1_BITS);
+    dataptr[3] = (DCTELEM)
+      RIGHT_SHIFT(tmp1 + tmp11 + tmp13, CONST_BITS-PASS1_BITS);
+    dataptr[5] = (DCTELEM)
+      RIGHT_SHIFT(tmp2 + tmp11 + tmp12, CONST_BITS-PASS1_BITS);
+    dataptr[7] = (DCTELEM)
+      RIGHT_SHIFT(tmp3 + tmp10 + tmp13, CONST_BITS-PASS1_BITS);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   */
+
+  dataptr = data;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part per LL&M figure 1 --- note that published figure is faulty;
+     * rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
+     */
+
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
+    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
+    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
+    tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
+
+    /* Add fudge factor here for final descale. */
+    tmp10 = tmp0 + tmp3 + (ONE << (PASS1_BITS-1));
+    tmp12 = tmp0 - tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp13 = tmp1 - tmp2;
+
+    tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
+    tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
+    tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
+    tmp3 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
+
+    dataptr[DCTSIZE*0] = (DCTELEM) RIGHT_SHIFT(tmp10 + tmp11, PASS1_BITS);
+    dataptr[DCTSIZE*4] = (DCTELEM) RIGHT_SHIFT(tmp10 - tmp11, PASS1_BITS);
+
+    z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
+    /* Add fudge factor here for final descale. */
+    z1 += ONE << (CONST_BITS+PASS1_BITS-1);
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      RIGHT_SHIFT(z1 + MULTIPLY(tmp12, FIX_0_765366865), CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*6] = (DCTELEM)
+      RIGHT_SHIFT(z1 - MULTIPLY(tmp13, FIX_1_847759065), CONST_BITS+PASS1_BITS);
+
+    /* Odd part per figure 8 --- note paper omits factor of sqrt(2).
+     * cK represents sqrt(2) * cos(K*pi/16).
+     * i0..i3 in the paper are tmp0..tmp3 here.
+     */
+
+    tmp10 = tmp0 + tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp12 = tmp0 + tmp2;
+    tmp13 = tmp1 + tmp3;
+    z1 = MULTIPLY(tmp12 + tmp13, FIX_1_175875602); /*  c3 */
+    /* Add fudge factor here for final descale. */
+    z1 += ONE << (CONST_BITS+PASS1_BITS-1);
+
+    tmp0  = MULTIPLY(tmp0,    FIX_1_501321110);    /*  c1+c3-c5-c7 */
+    tmp1  = MULTIPLY(tmp1,    FIX_3_072711026);    /*  c1+c3+c5-c7 */
+    tmp2  = MULTIPLY(tmp2,    FIX_2_053119869);    /*  c1+c3-c5+c7 */
+    tmp3  = MULTIPLY(tmp3,    FIX_0_298631336);    /* -c1+c3+c5-c7 */
+    tmp10 = MULTIPLY(tmp10, - FIX_0_899976223);    /*  c7-c3 */
+    tmp11 = MULTIPLY(tmp11, - FIX_2_562915447);    /* -c1-c3 */
+    tmp12 = MULTIPLY(tmp12, - FIX_0_390180644);    /*  c5-c3 */
+    tmp13 = MULTIPLY(tmp13, - FIX_1_961570560);    /* -c3-c5 */
+
+    tmp12 += z1;
+    tmp13 += z1;
+
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 + tmp10 + tmp12, CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*3] = (DCTELEM)
+      RIGHT_SHIFT(tmp1 + tmp11 + tmp13, CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*5] = (DCTELEM)
+      RIGHT_SHIFT(tmp2 + tmp11 + tmp12, CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*7] = (DCTELEM)
+      RIGHT_SHIFT(tmp3 + tmp10 + tmp13, CONST_BITS+PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+#ifdef DCT_SCALING_SUPPORTED
+
+
+/*
+ * Perform the forward DCT on a 7x7 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_7x7 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3;
+  INT32 tmp10, tmp11, tmp12;
+  INT32 z1, z2, z3;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* cK represents sqrt(2) * cos(K*pi/14). */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 7; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[6]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[5]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[4]);
+    tmp3 = GETJSAMPLE(elemptr[3]);
+
+    tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[6]);
+    tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[5]);
+    tmp12 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[4]);
+
+    z1 = tmp0 + tmp2;
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((z1 + tmp1 + tmp3 - 7 * CENTERJSAMPLE) << PASS1_BITS);
+    tmp3 += tmp3;
+    z1 -= tmp3;
+    z1 -= tmp3;
+    z1 = MULTIPLY(z1, FIX(0.353553391));                /* (c2+c6-c4)/2 */
+    z2 = MULTIPLY(tmp0 - tmp2, FIX(0.920609002));       /* (c2+c4-c6)/2 */
+    z3 = MULTIPLY(tmp1 - tmp2, FIX(0.314692123));       /* c6 */
+    dataptr[2] = (DCTELEM) DESCALE(z1 + z2 + z3, CONST_BITS-PASS1_BITS);
+    z1 -= z2;
+    z2 = MULTIPLY(tmp0 - tmp1, FIX(0.881747734));       /* c4 */
+    dataptr[4] = (DCTELEM)
+      DESCALE(z2 + z3 - MULTIPLY(tmp1 - tmp3, FIX(0.707106781)), /* c2+c6-c4 */
+	      CONST_BITS-PASS1_BITS);
+    dataptr[6] = (DCTELEM) DESCALE(z1 + z2, CONST_BITS-PASS1_BITS);
+
+    /* Odd part */
+
+    tmp1 = MULTIPLY(tmp10 + tmp11, FIX(0.935414347));   /* (c3+c1-c5)/2 */
+    tmp2 = MULTIPLY(tmp10 - tmp11, FIX(0.170262339));   /* (c3+c5-c1)/2 */
+    tmp0 = tmp1 - tmp2;
+    tmp1 += tmp2;
+    tmp2 = MULTIPLY(tmp11 + tmp12, - FIX(1.378756276)); /* -c1 */
+    tmp1 += tmp2;
+    tmp3 = MULTIPLY(tmp10 + tmp12, FIX(0.613604268));   /* c5 */
+    tmp0 += tmp3;
+    tmp2 += tmp3 + MULTIPLY(tmp12, FIX(1.870828693));   /* c3+c1-c5 */
+
+    dataptr[1] = (DCTELEM) DESCALE(tmp0, CONST_BITS-PASS1_BITS);
+    dataptr[3] = (DCTELEM) DESCALE(tmp1, CONST_BITS-PASS1_BITS);
+    dataptr[5] = (DCTELEM) DESCALE(tmp2, CONST_BITS-PASS1_BITS);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by (8/7)**2 = 64/49, which we fold
+   * into the constant multipliers:
+   * cK now represents sqrt(2) * cos(K*pi/14) * 64/49.
+   */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 7; ctr++) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*6];
+    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*5];
+    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*4];
+    tmp3 = dataptr[DCTSIZE*3];
+
+    tmp10 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*6];
+    tmp11 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*5];
+    tmp12 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*4];
+
+    z1 = tmp0 + tmp2;
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(z1 + tmp1 + tmp3, FIX(1.306122449)), /* 64/49 */
+	      CONST_BITS+PASS1_BITS);
+    tmp3 += tmp3;
+    z1 -= tmp3;
+    z1 -= tmp3;
+    z1 = MULTIPLY(z1, FIX(0.461784020));                /* (c2+c6-c4)/2 */
+    z2 = MULTIPLY(tmp0 - tmp2, FIX(1.202428084));       /* (c2+c4-c6)/2 */
+    z3 = MULTIPLY(tmp1 - tmp2, FIX(0.411026446));       /* c6 */
+    dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + z2 + z3, CONST_BITS+PASS1_BITS);
+    z1 -= z2;
+    z2 = MULTIPLY(tmp0 - tmp1, FIX(1.151670509));       /* c4 */
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(z2 + z3 - MULTIPLY(tmp1 - tmp3, FIX(0.923568041)), /* c2+c6-c4 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 + z2, CONST_BITS+PASS1_BITS);
+
+    /* Odd part */
+
+    tmp1 = MULTIPLY(tmp10 + tmp11, FIX(1.221765677));   /* (c3+c1-c5)/2 */
+    tmp2 = MULTIPLY(tmp10 - tmp11, FIX(0.222383464));   /* (c3+c5-c1)/2 */
+    tmp0 = tmp1 - tmp2;
+    tmp1 += tmp2;
+    tmp2 = MULTIPLY(tmp11 + tmp12, - FIX(1.800824523)); /* -c1 */
+    tmp1 += tmp2;
+    tmp3 = MULTIPLY(tmp10 + tmp12, FIX(0.801442310));   /* c5 */
+    tmp0 += tmp3;
+    tmp2 += tmp3 + MULTIPLY(tmp12, FIX(2.443531355));   /* c3+c1-c5 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp0, CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp1, CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp2, CONST_BITS+PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 6x6 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_6x6 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2;
+  INT32 tmp10, tmp11, tmp12;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* cK represents sqrt(2) * cos(K*pi/12). */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 6; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[5]);
+    tmp11 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[4]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[3]);
+
+    tmp10 = tmp0 + tmp2;
+    tmp12 = tmp0 - tmp2;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[5]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[4]);
+    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[3]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp10 + tmp11 - 6 * CENTERJSAMPLE) << PASS1_BITS);
+    dataptr[2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp12, FIX(1.224744871)),                 /* c2 */
+	      CONST_BITS-PASS1_BITS);
+    dataptr[4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(0.707106781)), /* c4 */
+	      CONST_BITS-PASS1_BITS);
+
+    /* Odd part */
+
+    tmp10 = DESCALE(MULTIPLY(tmp0 + tmp2, FIX(0.366025404)),     /* c5 */
+		    CONST_BITS-PASS1_BITS);
+
+    dataptr[1] = (DCTELEM) (tmp10 + ((tmp0 + tmp1) << PASS1_BITS));
+    dataptr[3] = (DCTELEM) ((tmp0 - tmp1 - tmp2) << PASS1_BITS);
+    dataptr[5] = (DCTELEM) (tmp10 + ((tmp2 - tmp1) << PASS1_BITS));
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by (8/6)**2 = 16/9, which we fold
+   * into the constant multipliers:
+   * cK now represents sqrt(2) * cos(K*pi/12) * 16/9.
+   */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 6; ctr++) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*5];
+    tmp11 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*4];
+    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*3];
+
+    tmp10 = tmp0 + tmp2;
+    tmp12 = tmp0 - tmp2;
+
+    tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*5];
+    tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*4];
+    tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*3];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 + tmp11, FIX(1.777777778)),         /* 16/9 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp12, FIX(2.177324216)),                 /* c2 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(1.257078722)), /* c4 */
+	      CONST_BITS+PASS1_BITS);
+
+    /* Odd part */
+
+    tmp10 = MULTIPLY(tmp0 + tmp2, FIX(0.650711829));             /* c5 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp0 + tmp1, FIX(1.777777778)),   /* 16/9 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*3] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0 - tmp1 - tmp2, FIX(1.777777778)),    /* 16/9 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*5] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp2 - tmp1, FIX(1.777777778)),   /* 16/9 */
+	      CONST_BITS+PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 5x5 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_5x5 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2;
+  INT32 tmp10, tmp11;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* We scale the results further by 2 as part of output adaption */
+  /* scaling for different DCT size. */
+  /* cK represents sqrt(2) * cos(K*pi/10). */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 5; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[4]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[3]);
+    tmp2 = GETJSAMPLE(elemptr[2]);
+
+    tmp10 = tmp0 + tmp1;
+    tmp11 = tmp0 - tmp1;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[4]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[3]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp10 + tmp2 - 5 * CENTERJSAMPLE) << (PASS1_BITS+1));
+    tmp11 = MULTIPLY(tmp11, FIX(0.790569415));          /* (c2+c4)/2 */
+    tmp10 -= tmp2 << 2;
+    tmp10 = MULTIPLY(tmp10, FIX(0.353553391));          /* (c2-c4)/2 */
+    dataptr[2] = (DCTELEM) DESCALE(tmp11 + tmp10, CONST_BITS-PASS1_BITS-1);
+    dataptr[4] = (DCTELEM) DESCALE(tmp11 - tmp10, CONST_BITS-PASS1_BITS-1);
+
+    /* Odd part */
+
+    tmp10 = MULTIPLY(tmp0 + tmp1, FIX(0.831253876));    /* c3 */
+
+    dataptr[1] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp0, FIX(0.513743148)), /* c1-c3 */
+	      CONST_BITS-PASS1_BITS-1);
+    dataptr[3] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp1, FIX(2.176250899)), /* c1+c3 */
+	      CONST_BITS-PASS1_BITS-1);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by (8/5)**2 = 64/25, which we partially
+   * fold into the constant multipliers (other part was done in pass 1):
+   * cK now represents sqrt(2) * cos(K*pi/10) * 32/25.
+   */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 5; ctr++) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*4];
+    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*3];
+    tmp2 = dataptr[DCTSIZE*2];
+
+    tmp10 = tmp0 + tmp1;
+    tmp11 = tmp0 - tmp1;
+
+    tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*4];
+    tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*3];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 + tmp2, FIX(1.28)),        /* 32/25 */
+	      CONST_BITS+PASS1_BITS);
+    tmp11 = MULTIPLY(tmp11, FIX(1.011928851));          /* (c2+c4)/2 */
+    tmp10 -= tmp2 << 2;
+    tmp10 = MULTIPLY(tmp10, FIX(0.452548340));          /* (c2-c4)/2 */
+    dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(tmp11 + tmp10, CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp11 - tmp10, CONST_BITS+PASS1_BITS);
+
+    /* Odd part */
+
+    tmp10 = MULTIPLY(tmp0 + tmp1, FIX(1.064004961));    /* c3 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp0, FIX(0.657591230)), /* c1-c3 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*3] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp1, FIX(2.785601151)), /* c1+c3 */
+	      CONST_BITS+PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 4x4 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_4x4 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1;
+  INT32 tmp10, tmp11;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* We must also scale the output by (8/4)**2 = 2**2, which we add here. */
+  /* cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point FDCT]. */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 4; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[3]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[2]);
+
+    tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[3]);
+    tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[2]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp0 + tmp1 - 4 * CENTERJSAMPLE) << (PASS1_BITS+2));
+    dataptr[2] = (DCTELEM) ((tmp0 - tmp1) << (PASS1_BITS+2));
+
+    /* Odd part */
+
+    tmp0 = MULTIPLY(tmp10 + tmp11, FIX_0_541196100);       /* c6 */
+    /* Add fudge factor here for final descale. */
+    tmp0 += ONE << (CONST_BITS-PASS1_BITS-3);
+
+    dataptr[1] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 + MULTIPLY(tmp10, FIX_0_765366865), /* c2-c6 */
+		  CONST_BITS-PASS1_BITS-2);
+    dataptr[3] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 - MULTIPLY(tmp11, FIX_1_847759065), /* c2+c6 */
+		  CONST_BITS-PASS1_BITS-2);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 4; ctr++) {
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*3] + (ONE << (PASS1_BITS-1));
+    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*2];
+
+    tmp10 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*3];
+    tmp11 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*2];
+
+    dataptr[DCTSIZE*0] = (DCTELEM) RIGHT_SHIFT(tmp0 + tmp1, PASS1_BITS);
+    dataptr[DCTSIZE*2] = (DCTELEM) RIGHT_SHIFT(tmp0 - tmp1, PASS1_BITS);
+
+    /* Odd part */
+
+    tmp0 = MULTIPLY(tmp10 + tmp11, FIX_0_541196100);       /* c6 */
+    /* Add fudge factor here for final descale. */
+    tmp0 += ONE << (CONST_BITS+PASS1_BITS-1);
+
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 + MULTIPLY(tmp10, FIX_0_765366865), /* c2-c6 */
+		  CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*3] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 - MULTIPLY(tmp11, FIX_1_847759065), /* c2+c6 */
+		  CONST_BITS+PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 3x3 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_3x3 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* We scale the results further by 2**2 as part of output adaption */
+  /* scaling for different DCT size. */
+  /* cK represents sqrt(2) * cos(K*pi/6). */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 3; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[2]);
+    tmp1 = GETJSAMPLE(elemptr[1]);
+
+    tmp2 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[2]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp0 + tmp1 - 3 * CENTERJSAMPLE) << (PASS1_BITS+2));
+    dataptr[2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0 - tmp1 - tmp1, FIX(0.707106781)), /* c2 */
+	      CONST_BITS-PASS1_BITS-2);
+
+    /* Odd part */
+
+    dataptr[1] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp2, FIX(1.224744871)),               /* c1 */
+	      CONST_BITS-PASS1_BITS-2);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by (8/3)**2 = 64/9, which we partially
+   * fold into the constant multipliers (other part was done in pass 1):
+   * cK now represents sqrt(2) * cos(K*pi/6) * 16/9.
+   */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 3; ctr++) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*2];
+    tmp1 = dataptr[DCTSIZE*1];
+
+    tmp2 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*2];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0 + tmp1, FIX(1.777777778)),        /* 16/9 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0 - tmp1 - tmp1, FIX(1.257078722)), /* c2 */
+	      CONST_BITS+PASS1_BITS);
+
+    /* Odd part */
+
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp2, FIX(2.177324216)),               /* c1 */
+	      CONST_BITS+PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 2x2 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_2x2 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3;
+  JSAMPROW elemptr;
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT. */
+
+  /* Row 0 */
+  elemptr = sample_data[0] + start_col;
+
+  tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[1]);
+  tmp1 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[1]);
+
+  /* Row 1 */
+  elemptr = sample_data[1] + start_col;
+
+  tmp2 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[1]);
+  tmp3 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[1]);
+
+  /* Pass 2: process columns.
+   * We leave the results scaled up by an overall factor of 8.
+   * We must also scale the output by (8/2)**2 = 2**4.
+   */
+
+  /* Column 0 */
+  /* Apply unsigned->signed conversion */
+  data[DCTSIZE*0] = (DCTELEM) ((tmp0 + tmp2 - 4 * CENTERJSAMPLE) << 4);
+  data[DCTSIZE*1] = (DCTELEM) ((tmp0 - tmp2) << 4);
+
+  /* Column 1 */
+  data[DCTSIZE*0+1] = (DCTELEM) ((tmp1 + tmp3) << 4);
+  data[DCTSIZE*1+1] = (DCTELEM) ((tmp1 - tmp3) << 4);
+}
+
+
+/*
+ * Perform the forward DCT on a 1x1 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_1x1 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* We leave the result scaled up by an overall factor of 8. */
+  /* We must also scale the output by (8/1)**2 = 2**6. */
+  /* Apply unsigned->signed conversion */
+  data[0] = (DCTELEM)
+    ((GETJSAMPLE(sample_data[0][start_col]) - CENTERJSAMPLE) << 6);
+}
+
+
+/*
+ * Perform the forward DCT on a 9x9 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_9x9 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4;
+  INT32 tmp10, tmp11, tmp12, tmp13;
+  INT32 z1, z2;
+  DCTELEM workspace[8];
+  DCTELEM *dataptr;
+  DCTELEM *wsptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* we scale the results further by 2 as part of output adaption */
+  /* scaling for different DCT size. */
+  /* cK represents sqrt(2) * cos(K*pi/18). */
+
+  dataptr = data;
+  ctr = 0;
+  for (;;) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[8]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[7]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[6]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[5]);
+    tmp4 = GETJSAMPLE(elemptr[4]);
+
+    tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[8]);
+    tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[7]);
+    tmp12 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[6]);
+    tmp13 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[5]);
+
+    z1 = tmp0 + tmp2 + tmp3;
+    z2 = tmp1 + tmp4;
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM) ((z1 + z2 - 9 * CENTERJSAMPLE) << 1);
+    dataptr[6] = (DCTELEM)
+      DESCALE(MULTIPLY(z1 - z2 - z2, FIX(0.707106781)),  /* c6 */
+	      CONST_BITS-1);
+    z1 = MULTIPLY(tmp0 - tmp2, FIX(1.328926049));        /* c2 */
+    z2 = MULTIPLY(tmp1 - tmp4 - tmp4, FIX(0.707106781)); /* c6 */
+    dataptr[2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp2 - tmp3, FIX(1.083350441))    /* c4 */
+	      + z1 + z2, CONST_BITS-1);
+    dataptr[4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp3 - tmp0, FIX(0.245575608))    /* c8 */
+	      + z1 - z2, CONST_BITS-1);
+
+    /* Odd part */
+
+    dataptr[3] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp12 - tmp13, FIX(1.224744871)), /* c3 */
+	      CONST_BITS-1);
+
+    tmp11 = MULTIPLY(tmp11, FIX(1.224744871));        /* c3 */
+    tmp0 = MULTIPLY(tmp10 + tmp12, FIX(0.909038955)); /* c5 */
+    tmp1 = MULTIPLY(tmp10 + tmp13, FIX(0.483689525)); /* c7 */
+
+    dataptr[1] = (DCTELEM) DESCALE(tmp11 + tmp0 + tmp1, CONST_BITS-1);
+
+    tmp2 = MULTIPLY(tmp12 - tmp13, FIX(1.392728481)); /* c1 */
+
+    dataptr[5] = (DCTELEM) DESCALE(tmp0 - tmp11 - tmp2, CONST_BITS-1);
+    dataptr[7] = (DCTELEM) DESCALE(tmp1 - tmp11 + tmp2, CONST_BITS-1);
+
+    ctr++;
+
+    if (ctr != DCTSIZE) {
+      if (ctr == 9)
+	break;			/* Done. */
+      dataptr += DCTSIZE;	/* advance pointer to next row */
+    } else
+      dataptr = workspace;	/* switch pointer to extended workspace */
+  }
+
+  /* Pass 2: process columns.
+   * We leave the results scaled up by an overall factor of 8.
+   * We must also scale the output by (8/9)**2 = 64/81, which we partially
+   * fold into the constant multipliers and final/initial shifting:
+   * cK now represents sqrt(2) * cos(K*pi/18) * 128/81.
+   */
+
+  dataptr = data;
+  wsptr = workspace;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*0];
+    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*7];
+    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*6];
+    tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*5];
+    tmp4 = dataptr[DCTSIZE*4];
+
+    tmp10 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*0];
+    tmp11 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*7];
+    tmp12 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*6];
+    tmp13 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*5];
+
+    z1 = tmp0 + tmp2 + tmp3;
+    z2 = tmp1 + tmp4;
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(z1 + z2, FIX(1.580246914)),       /* 128/81 */
+	      CONST_BITS+2);
+    dataptr[DCTSIZE*6] = (DCTELEM)
+      DESCALE(MULTIPLY(z1 - z2 - z2, FIX(1.117403309)),  /* c6 */
+	      CONST_BITS+2);
+    z1 = MULTIPLY(tmp0 - tmp2, FIX(2.100031287));        /* c2 */
+    z2 = MULTIPLY(tmp1 - tmp4 - tmp4, FIX(1.117403309)); /* c6 */
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp2 - tmp3, FIX(1.711961190))    /* c4 */
+	      + z1 + z2, CONST_BITS+2);
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp3 - tmp0, FIX(0.388070096))    /* c8 */
+	      + z1 - z2, CONST_BITS+2);
+
+    /* Odd part */
+
+    dataptr[DCTSIZE*3] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp12 - tmp13, FIX(1.935399303)), /* c3 */
+	      CONST_BITS+2);
+
+    tmp11 = MULTIPLY(tmp11, FIX(1.935399303));        /* c3 */
+    tmp0 = MULTIPLY(tmp10 + tmp12, FIX(1.436506004)); /* c5 */
+    tmp1 = MULTIPLY(tmp10 + tmp13, FIX(0.764348879)); /* c7 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      DESCALE(tmp11 + tmp0 + tmp1, CONST_BITS+2);
+
+    tmp2 = MULTIPLY(tmp12 - tmp13, FIX(2.200854883)); /* c1 */
+
+    dataptr[DCTSIZE*5] = (DCTELEM)
+      DESCALE(tmp0 - tmp11 - tmp2, CONST_BITS+2);
+    dataptr[DCTSIZE*7] = (DCTELEM)
+      DESCALE(tmp1 - tmp11 + tmp2, CONST_BITS+2);
+
+    dataptr++;			/* advance pointer to next column */
+    wsptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 10x10 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_10x10 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
+  DCTELEM workspace[8*2];
+  DCTELEM *dataptr;
+  DCTELEM *wsptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* we scale the results further by 2 as part of output adaption */
+  /* scaling for different DCT size. */
+  /* cK represents sqrt(2) * cos(K*pi/20). */
+
+  dataptr = data;
+  ctr = 0;
+  for (;;) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[9]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[8]);
+    tmp12 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[7]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[6]);
+    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[5]);
+
+    tmp10 = tmp0 + tmp4;
+    tmp13 = tmp0 - tmp4;
+    tmp11 = tmp1 + tmp3;
+    tmp14 = tmp1 - tmp3;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[9]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[8]);
+    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[7]);
+    tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[6]);
+    tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[5]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp10 + tmp11 + tmp12 - 10 * CENTERJSAMPLE) << 1);
+    tmp12 += tmp12;
+    dataptr[4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.144122806)) - /* c4 */
+	      MULTIPLY(tmp11 - tmp12, FIX(0.437016024)),  /* c8 */
+	      CONST_BITS-1);
+    tmp10 = MULTIPLY(tmp13 + tmp14, FIX(0.831253876));    /* c6 */
+    dataptr[2] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp13, FIX(0.513743148)),  /* c2-c6 */
+	      CONST_BITS-1);
+    dataptr[6] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp14, FIX(2.176250899)),  /* c2+c6 */
+	      CONST_BITS-1);
+
+    /* Odd part */
+
+    tmp10 = tmp0 + tmp4;
+    tmp11 = tmp1 - tmp3;
+    dataptr[5] = (DCTELEM) ((tmp10 - tmp11 - tmp2) << 1);
+    tmp2 <<= CONST_BITS;
+    dataptr[1] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0, FIX(1.396802247)) +          /* c1 */
+	      MULTIPLY(tmp1, FIX(1.260073511)) + tmp2 +   /* c3 */
+	      MULTIPLY(tmp3, FIX(0.642039522)) +          /* c7 */
+	      MULTIPLY(tmp4, FIX(0.221231742)),           /* c9 */
+	      CONST_BITS-1);
+    tmp12 = MULTIPLY(tmp0 - tmp4, FIX(0.951056516)) -     /* (c3+c7)/2 */
+	    MULTIPLY(tmp1 + tmp3, FIX(0.587785252));      /* (c1-c9)/2 */
+    tmp13 = MULTIPLY(tmp10 + tmp11, FIX(0.309016994)) +   /* (c3-c7)/2 */
+	    (tmp11 << (CONST_BITS - 1)) - tmp2;
+    dataptr[3] = (DCTELEM) DESCALE(tmp12 + tmp13, CONST_BITS-1);
+    dataptr[7] = (DCTELEM) DESCALE(tmp12 - tmp13, CONST_BITS-1);
+
+    ctr++;
+
+    if (ctr != DCTSIZE) {
+      if (ctr == 10)
+	break;			/* Done. */
+      dataptr += DCTSIZE;	/* advance pointer to next row */
+    } else
+      dataptr = workspace;	/* switch pointer to extended workspace */
+  }
+
+  /* Pass 2: process columns.
+   * We leave the results scaled up by an overall factor of 8.
+   * We must also scale the output by (8/10)**2 = 16/25, which we partially
+   * fold into the constant multipliers and final/initial shifting:
+   * cK now represents sqrt(2) * cos(K*pi/20) * 32/25.
+   */
+
+  dataptr = data;
+  wsptr = workspace;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*1];
+    tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*0];
+    tmp12 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*7];
+    tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*6];
+    tmp4 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*5];
+
+    tmp10 = tmp0 + tmp4;
+    tmp13 = tmp0 - tmp4;
+    tmp11 = tmp1 + tmp3;
+    tmp14 = tmp1 - tmp3;
+
+    tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*1];
+    tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*0];
+    tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*7];
+    tmp3 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*6];
+    tmp4 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*5];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12, FIX(1.28)), /* 32/25 */
+	      CONST_BITS+2);
+    tmp12 += tmp12;
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.464477191)) - /* c4 */
+	      MULTIPLY(tmp11 - tmp12, FIX(0.559380511)),  /* c8 */
+	      CONST_BITS+2);
+    tmp10 = MULTIPLY(tmp13 + tmp14, FIX(1.064004961));    /* c6 */
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp13, FIX(0.657591230)),  /* c2-c6 */
+	      CONST_BITS+2);
+    dataptr[DCTSIZE*6] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp14, FIX(2.785601151)),  /* c2+c6 */
+	      CONST_BITS+2);
+
+    /* Odd part */
+
+    tmp10 = tmp0 + tmp4;
+    tmp11 = tmp1 - tmp3;
+    dataptr[DCTSIZE*5] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp11 - tmp2, FIX(1.28)),  /* 32/25 */
+	      CONST_BITS+2);
+    tmp2 = MULTIPLY(tmp2, FIX(1.28));                     /* 32/25 */
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0, FIX(1.787906876)) +          /* c1 */
+	      MULTIPLY(tmp1, FIX(1.612894094)) + tmp2 +   /* c3 */
+	      MULTIPLY(tmp3, FIX(0.821810588)) +          /* c7 */
+	      MULTIPLY(tmp4, FIX(0.283176630)),           /* c9 */
+	      CONST_BITS+2);
+    tmp12 = MULTIPLY(tmp0 - tmp4, FIX(1.217352341)) -     /* (c3+c7)/2 */
+	    MULTIPLY(tmp1 + tmp3, FIX(0.752365123));      /* (c1-c9)/2 */
+    tmp13 = MULTIPLY(tmp10 + tmp11, FIX(0.395541753)) +   /* (c3-c7)/2 */
+	    MULTIPLY(tmp11, FIX(0.64)) - tmp2;            /* 16/25 */
+    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp12 + tmp13, CONST_BITS+2);
+    dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp12 - tmp13, CONST_BITS+2);
+
+    dataptr++;			/* advance pointer to next column */
+    wsptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on an 11x11 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_11x11 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
+  INT32 z1, z2, z3;
+  DCTELEM workspace[8*3];
+  DCTELEM *dataptr;
+  DCTELEM *wsptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* we scale the results further by 2 as part of output adaption */
+  /* scaling for different DCT size. */
+  /* cK represents sqrt(2) * cos(K*pi/22). */
+
+  dataptr = data;
+  ctr = 0;
+  for (;;) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[10]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[9]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[8]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[7]);
+    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[6]);
+    tmp5 = GETJSAMPLE(elemptr[5]);
+
+    tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[10]);
+    tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[9]);
+    tmp12 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[8]);
+    tmp13 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[7]);
+    tmp14 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[6]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp0 + tmp1 + tmp2 + tmp3 + tmp4 + tmp5 - 11 * CENTERJSAMPLE) << 1);
+    tmp5 += tmp5;
+    tmp0 -= tmp5;
+    tmp1 -= tmp5;
+    tmp2 -= tmp5;
+    tmp3 -= tmp5;
+    tmp4 -= tmp5;
+    z1 = MULTIPLY(tmp0 + tmp3, FIX(1.356927976)) +       /* c2 */
+	 MULTIPLY(tmp2 + tmp4, FIX(0.201263574));        /* c10 */
+    z2 = MULTIPLY(tmp1 - tmp3, FIX(0.926112931));        /* c6 */
+    z3 = MULTIPLY(tmp0 - tmp1, FIX(1.189712156));        /* c4 */
+    dataptr[2] = (DCTELEM)
+      DESCALE(z1 + z2 - MULTIPLY(tmp3, FIX(1.018300590)) /* c2+c8-c6 */
+	      - MULTIPLY(tmp4, FIX(1.390975730)),        /* c4+c10 */
+	      CONST_BITS-1);
+    dataptr[4] = (DCTELEM)
+      DESCALE(z2 + z3 + MULTIPLY(tmp1, FIX(0.062335650)) /* c4-c6-c10 */
+	      - MULTIPLY(tmp2, FIX(1.356927976))         /* c2 */
+	      + MULTIPLY(tmp4, FIX(0.587485545)),        /* c8 */
+	      CONST_BITS-1);
+    dataptr[6] = (DCTELEM)
+      DESCALE(z1 + z3 - MULTIPLY(tmp0, FIX(1.620527200)) /* c2+c4-c6 */
+	      - MULTIPLY(tmp2, FIX(0.788749120)),        /* c8+c10 */
+	      CONST_BITS-1);
+
+    /* Odd part */
+
+    tmp1 = MULTIPLY(tmp10 + tmp11, FIX(1.286413905));    /* c3 */
+    tmp2 = MULTIPLY(tmp10 + tmp12, FIX(1.068791298));    /* c5 */
+    tmp3 = MULTIPLY(tmp10 + tmp13, FIX(0.764581576));    /* c7 */
+    tmp0 = tmp1 + tmp2 + tmp3 - MULTIPLY(tmp10, FIX(1.719967871)) /* c7+c5+c3-c1 */
+	   + MULTIPLY(tmp14, FIX(0.398430003));          /* c9 */
+    tmp4 = MULTIPLY(tmp11 + tmp12, - FIX(0.764581576));  /* -c7 */
+    tmp5 = MULTIPLY(tmp11 + tmp13, - FIX(1.399818907));  /* -c1 */
+    tmp1 += tmp4 + tmp5 + MULTIPLY(tmp11, FIX(1.276416582)) /* c9+c7+c1-c3 */
+	    - MULTIPLY(tmp14, FIX(1.068791298));         /* c5 */
+    tmp10 = MULTIPLY(tmp12 + tmp13, FIX(0.398430003));   /* c9 */
+    tmp2 += tmp4 + tmp10 - MULTIPLY(tmp12, FIX(1.989053629)) /* c9+c5+c3-c7 */
+	    + MULTIPLY(tmp14, FIX(1.399818907));         /* c1 */
+    tmp3 += tmp5 + tmp10 + MULTIPLY(tmp13, FIX(1.305598626)) /* c1+c5-c9-c7 */
+	    - MULTIPLY(tmp14, FIX(1.286413905));         /* c3 */
+
+    dataptr[1] = (DCTELEM) DESCALE(tmp0, CONST_BITS-1);
+    dataptr[3] = (DCTELEM) DESCALE(tmp1, CONST_BITS-1);
+    dataptr[5] = (DCTELEM) DESCALE(tmp2, CONST_BITS-1);
+    dataptr[7] = (DCTELEM) DESCALE(tmp3, CONST_BITS-1);
+
+    ctr++;
+
+    if (ctr != DCTSIZE) {
+      if (ctr == 11)
+	break;			/* Done. */
+      dataptr += DCTSIZE;	/* advance pointer to next row */
+    } else
+      dataptr = workspace;	/* switch pointer to extended workspace */
+  }
+
+  /* Pass 2: process columns.
+   * We leave the results scaled up by an overall factor of 8.
+   * We must also scale the output by (8/11)**2 = 64/121, which we partially
+   * fold into the constant multipliers and final/initial shifting:
+   * cK now represents sqrt(2) * cos(K*pi/22) * 128/121.
+   */
+
+  dataptr = data;
+  wsptr = workspace;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*2];
+    tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*1];
+    tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*0];
+    tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*7];
+    tmp4 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*6];
+    tmp5 = dataptr[DCTSIZE*5];
+
+    tmp10 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*2];
+    tmp11 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*1];
+    tmp12 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*0];
+    tmp13 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*7];
+    tmp14 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*6];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0 + tmp1 + tmp2 + tmp3 + tmp4 + tmp5,
+		       FIX(1.057851240)),                /* 128/121 */
+	      CONST_BITS+2);
+    tmp5 += tmp5;
+    tmp0 -= tmp5;
+    tmp1 -= tmp5;
+    tmp2 -= tmp5;
+    tmp3 -= tmp5;
+    tmp4 -= tmp5;
+    z1 = MULTIPLY(tmp0 + tmp3, FIX(1.435427942)) +       /* c2 */
+	 MULTIPLY(tmp2 + tmp4, FIX(0.212906922));        /* c10 */
+    z2 = MULTIPLY(tmp1 - tmp3, FIX(0.979689713));        /* c6 */
+    z3 = MULTIPLY(tmp0 - tmp1, FIX(1.258538479));        /* c4 */
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(z1 + z2 - MULTIPLY(tmp3, FIX(1.077210542)) /* c2+c8-c6 */
+	      - MULTIPLY(tmp4, FIX(1.471445400)),        /* c4+c10 */
+	      CONST_BITS+2);
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(z2 + z3 + MULTIPLY(tmp1, FIX(0.065941844)) /* c4-c6-c10 */
+	      - MULTIPLY(tmp2, FIX(1.435427942))         /* c2 */
+	      + MULTIPLY(tmp4, FIX(0.621472312)),        /* c8 */
+	      CONST_BITS+2);
+    dataptr[DCTSIZE*6] = (DCTELEM)
+      DESCALE(z1 + z3 - MULTIPLY(tmp0, FIX(1.714276708)) /* c2+c4-c6 */
+	      - MULTIPLY(tmp2, FIX(0.834379234)),        /* c8+c10 */
+	      CONST_BITS+2);
+
+    /* Odd part */
+
+    tmp1 = MULTIPLY(tmp10 + tmp11, FIX(1.360834544));    /* c3 */
+    tmp2 = MULTIPLY(tmp10 + tmp12, FIX(1.130622199));    /* c5 */
+    tmp3 = MULTIPLY(tmp10 + tmp13, FIX(0.808813568));    /* c7 */
+    tmp0 = tmp1 + tmp2 + tmp3 - MULTIPLY(tmp10, FIX(1.819470145)) /* c7+c5+c3-c1 */
+	   + MULTIPLY(tmp14, FIX(0.421479672));          /* c9 */
+    tmp4 = MULTIPLY(tmp11 + tmp12, - FIX(0.808813568));  /* -c7 */
+    tmp5 = MULTIPLY(tmp11 + tmp13, - FIX(1.480800167));  /* -c1 */
+    tmp1 += tmp4 + tmp5 + MULTIPLY(tmp11, FIX(1.350258864)) /* c9+c7+c1-c3 */
+	    - MULTIPLY(tmp14, FIX(1.130622199));         /* c5 */
+    tmp10 = MULTIPLY(tmp12 + tmp13, FIX(0.421479672));   /* c9 */
+    tmp2 += tmp4 + tmp10 - MULTIPLY(tmp12, FIX(2.104122847)) /* c9+c5+c3-c7 */
+	    + MULTIPLY(tmp14, FIX(1.480800167));         /* c1 */
+    tmp3 += tmp5 + tmp10 + MULTIPLY(tmp13, FIX(1.381129125)) /* c1+c5-c9-c7 */
+	    - MULTIPLY(tmp14, FIX(1.360834544));         /* c3 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp0, CONST_BITS+2);
+    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp1, CONST_BITS+2);
+    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp2, CONST_BITS+2);
+    dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp3, CONST_BITS+2);
+
+    dataptr++;			/* advance pointer to next column */
+    wsptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 12x12 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_12x12 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
+  DCTELEM workspace[8*4];
+  DCTELEM *dataptr;
+  DCTELEM *wsptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT. */
+  /* cK represents sqrt(2) * cos(K*pi/24). */
+
+  dataptr = data;
+  ctr = 0;
+  for (;;) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[11]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[10]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[9]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[8]);
+    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[7]);
+    tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[6]);
+
+    tmp10 = tmp0 + tmp5;
+    tmp13 = tmp0 - tmp5;
+    tmp11 = tmp1 + tmp4;
+    tmp14 = tmp1 - tmp4;
+    tmp12 = tmp2 + tmp3;
+    tmp15 = tmp2 - tmp3;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[11]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[10]);
+    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[9]);
+    tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[8]);
+    tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[7]);
+    tmp5 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[6]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM) (tmp10 + tmp11 + tmp12 - 12 * CENTERJSAMPLE);
+    dataptr[6] = (DCTELEM) (tmp13 - tmp14 - tmp15);
+    dataptr[4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.224744871)), /* c4 */
+	      CONST_BITS);
+    dataptr[2] = (DCTELEM)
+      DESCALE(tmp14 - tmp15 + MULTIPLY(tmp13 + tmp15, FIX(1.366025404)), /* c2 */
+	      CONST_BITS);
+
+    /* Odd part */
+
+    tmp10 = MULTIPLY(tmp1 + tmp4, FIX_0_541196100);    /* c9 */
+    tmp14 = tmp10 + MULTIPLY(tmp1, FIX_0_765366865);   /* c3-c9 */
+    tmp15 = tmp10 - MULTIPLY(tmp4, FIX_1_847759065);   /* c3+c9 */
+    tmp12 = MULTIPLY(tmp0 + tmp2, FIX(1.121971054));   /* c5 */
+    tmp13 = MULTIPLY(tmp0 + tmp3, FIX(0.860918669));   /* c7 */
+    tmp10 = tmp12 + tmp13 + tmp14 - MULTIPLY(tmp0, FIX(0.580774953)) /* c5+c7-c1 */
+	    + MULTIPLY(tmp5, FIX(0.184591911));        /* c11 */
+    tmp11 = MULTIPLY(tmp2 + tmp3, - FIX(0.184591911)); /* -c11 */
+    tmp12 += tmp11 - tmp15 - MULTIPLY(tmp2, FIX(2.339493912)) /* c1+c5-c11 */
+	    + MULTIPLY(tmp5, FIX(0.860918669));        /* c7 */
+    tmp13 += tmp11 - tmp14 + MULTIPLY(tmp3, FIX(0.725788011)) /* c1+c11-c7 */
+	    - MULTIPLY(tmp5, FIX(1.121971054));        /* c5 */
+    tmp11 = tmp15 + MULTIPLY(tmp0 - tmp3, FIX(1.306562965)) /* c3 */
+	    - MULTIPLY(tmp2 + tmp5, FIX_0_541196100);  /* c9 */
+
+    dataptr[1] = (DCTELEM) DESCALE(tmp10, CONST_BITS);
+    dataptr[3] = (DCTELEM) DESCALE(tmp11, CONST_BITS);
+    dataptr[5] = (DCTELEM) DESCALE(tmp12, CONST_BITS);
+    dataptr[7] = (DCTELEM) DESCALE(tmp13, CONST_BITS);
+
+    ctr++;
+
+    if (ctr != DCTSIZE) {
+      if (ctr == 12)
+	break;			/* Done. */
+      dataptr += DCTSIZE;	/* advance pointer to next row */
+    } else
+      dataptr = workspace;	/* switch pointer to extended workspace */
+  }
+
+  /* Pass 2: process columns.
+   * We leave the results scaled up by an overall factor of 8.
+   * We must also scale the output by (8/12)**2 = 4/9, which we partially
+   * fold into the constant multipliers and final shifting:
+   * cK now represents sqrt(2) * cos(K*pi/24) * 8/9.
+   */
+
+  dataptr = data;
+  wsptr = workspace;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*3];
+    tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*2];
+    tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*1];
+    tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*0];
+    tmp4 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*7];
+    tmp5 = dataptr[DCTSIZE*5] + dataptr[DCTSIZE*6];
+
+    tmp10 = tmp0 + tmp5;
+    tmp13 = tmp0 - tmp5;
+    tmp11 = tmp1 + tmp4;
+    tmp14 = tmp1 - tmp4;
+    tmp12 = tmp2 + tmp3;
+    tmp15 = tmp2 - tmp3;
+
+    tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*3];
+    tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*2];
+    tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*1];
+    tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*0];
+    tmp4 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*7];
+    tmp5 = dataptr[DCTSIZE*5] - dataptr[DCTSIZE*6];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12, FIX(0.888888889)), /* 8/9 */
+	      CONST_BITS+1);
+    dataptr[DCTSIZE*6] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp13 - tmp14 - tmp15, FIX(0.888888889)), /* 8/9 */
+	      CONST_BITS+1);
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.088662108)),         /* c4 */
+	      CONST_BITS+1);
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp14 - tmp15, FIX(0.888888889)) +        /* 8/9 */
+	      MULTIPLY(tmp13 + tmp15, FIX(1.214244803)),         /* c2 */
+	      CONST_BITS+1);
+
+    /* Odd part */
+
+    tmp10 = MULTIPLY(tmp1 + tmp4, FIX(0.481063200));   /* c9 */
+    tmp14 = tmp10 + MULTIPLY(tmp1, FIX(0.680326102));  /* c3-c9 */
+    tmp15 = tmp10 - MULTIPLY(tmp4, FIX(1.642452502));  /* c3+c9 */
+    tmp12 = MULTIPLY(tmp0 + tmp2, FIX(0.997307603));   /* c5 */
+    tmp13 = MULTIPLY(tmp0 + tmp3, FIX(0.765261039));   /* c7 */
+    tmp10 = tmp12 + tmp13 + tmp14 - MULTIPLY(tmp0, FIX(0.516244403)) /* c5+c7-c1 */
+	    + MULTIPLY(tmp5, FIX(0.164081699));        /* c11 */
+    tmp11 = MULTIPLY(tmp2 + tmp3, - FIX(0.164081699)); /* -c11 */
+    tmp12 += tmp11 - tmp15 - MULTIPLY(tmp2, FIX(2.079550144)) /* c1+c5-c11 */
+	    + MULTIPLY(tmp5, FIX(0.765261039));        /* c7 */
+    tmp13 += tmp11 - tmp14 + MULTIPLY(tmp3, FIX(0.645144899)) /* c1+c11-c7 */
+	    - MULTIPLY(tmp5, FIX(0.997307603));        /* c5 */
+    tmp11 = tmp15 + MULTIPLY(tmp0 - tmp3, FIX(1.161389302)) /* c3 */
+	    - MULTIPLY(tmp2 + tmp5, FIX(0.481063200)); /* c9 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp10, CONST_BITS+1);
+    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp11, CONST_BITS+1);
+    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp12, CONST_BITS+1);
+    dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp13, CONST_BITS+1);
+
+    dataptr++;			/* advance pointer to next column */
+    wsptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 13x13 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_13x13 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
+  INT32 z1, z2;
+  DCTELEM workspace[8*5];
+  DCTELEM *dataptr;
+  DCTELEM *wsptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT. */
+  /* cK represents sqrt(2) * cos(K*pi/26). */
+
+  dataptr = data;
+  ctr = 0;
+  for (;;) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[12]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[11]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[10]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[9]);
+    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[8]);
+    tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[7]);
+    tmp6 = GETJSAMPLE(elemptr[6]);
+
+    tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[12]);
+    tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[11]);
+    tmp12 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[10]);
+    tmp13 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[9]);
+    tmp14 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[8]);
+    tmp15 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[7]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      (tmp0 + tmp1 + tmp2 + tmp3 + tmp4 + tmp5 + tmp6 - 13 * CENTERJSAMPLE);
+    tmp6 += tmp6;
+    tmp0 -= tmp6;
+    tmp1 -= tmp6;
+    tmp2 -= tmp6;
+    tmp3 -= tmp6;
+    tmp4 -= tmp6;
+    tmp5 -= tmp6;
+    dataptr[2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0, FIX(1.373119086)) +   /* c2 */
+	      MULTIPLY(tmp1, FIX(1.058554052)) +   /* c6 */
+	      MULTIPLY(tmp2, FIX(0.501487041)) -   /* c10 */
+	      MULTIPLY(tmp3, FIX(0.170464608)) -   /* c12 */
+	      MULTIPLY(tmp4, FIX(0.803364869)) -   /* c8 */
+	      MULTIPLY(tmp5, FIX(1.252223920)),    /* c4 */
+	      CONST_BITS);
+    z1 = MULTIPLY(tmp0 - tmp2, FIX(1.155388986)) - /* (c4+c6)/2 */
+	 MULTIPLY(tmp3 - tmp4, FIX(0.435816023)) - /* (c2-c10)/2 */
+	 MULTIPLY(tmp1 - tmp5, FIX(0.316450131));  /* (c8-c12)/2 */
+    z2 = MULTIPLY(tmp0 + tmp2, FIX(0.096834934)) - /* (c4-c6)/2 */
+	 MULTIPLY(tmp3 + tmp4, FIX(0.937303064)) + /* (c2+c10)/2 */
+	 MULTIPLY(tmp1 + tmp5, FIX(0.486914739));  /* (c8+c12)/2 */
+
+    dataptr[4] = (DCTELEM) DESCALE(z1 + z2, CONST_BITS);
+    dataptr[6] = (DCTELEM) DESCALE(z1 - z2, CONST_BITS);
+
+    /* Odd part */
+
+    tmp1 = MULTIPLY(tmp10 + tmp11, FIX(1.322312651));   /* c3 */
+    tmp2 = MULTIPLY(tmp10 + tmp12, FIX(1.163874945));   /* c5 */
+    tmp3 = MULTIPLY(tmp10 + tmp13, FIX(0.937797057)) +  /* c7 */
+	   MULTIPLY(tmp14 + tmp15, FIX(0.338443458));   /* c11 */
+    tmp0 = tmp1 + tmp2 + tmp3 -
+	   MULTIPLY(tmp10, FIX(2.020082300)) +          /* c3+c5+c7-c1 */
+	   MULTIPLY(tmp14, FIX(0.318774355));           /* c9-c11 */
+    tmp4 = MULTIPLY(tmp14 - tmp15, FIX(0.937797057)) -  /* c7 */
+	   MULTIPLY(tmp11 + tmp12, FIX(0.338443458));   /* c11 */
+    tmp5 = MULTIPLY(tmp11 + tmp13, - FIX(1.163874945)); /* -c5 */
+    tmp1 += tmp4 + tmp5 +
+	    MULTIPLY(tmp11, FIX(0.837223564)) -         /* c5+c9+c11-c3 */
+	    MULTIPLY(tmp14, FIX(2.341699410));          /* c1+c7 */
+    tmp6 = MULTIPLY(tmp12 + tmp13, - FIX(0.657217813)); /* -c9 */
+    tmp2 += tmp4 + tmp6 -
+	    MULTIPLY(tmp12, FIX(1.572116027)) +         /* c1+c5-c9-c11 */
+	    MULTIPLY(tmp15, FIX(2.260109708));          /* c3+c7 */
+    tmp3 += tmp5 + tmp6 +
+	    MULTIPLY(tmp13, FIX(2.205608352)) -         /* c3+c5+c9-c7 */
+	    MULTIPLY(tmp15, FIX(1.742345811));          /* c1+c11 */
+
+    dataptr[1] = (DCTELEM) DESCALE(tmp0, CONST_BITS);
+    dataptr[3] = (DCTELEM) DESCALE(tmp1, CONST_BITS);
+    dataptr[5] = (DCTELEM) DESCALE(tmp2, CONST_BITS);
+    dataptr[7] = (DCTELEM) DESCALE(tmp3, CONST_BITS);
+
+    ctr++;
+
+    if (ctr != DCTSIZE) {
+      if (ctr == 13)
+	break;			/* Done. */
+      dataptr += DCTSIZE;	/* advance pointer to next row */
+    } else
+      dataptr = workspace;	/* switch pointer to extended workspace */
+  }
+
+  /* Pass 2: process columns.
+   * We leave the results scaled up by an overall factor of 8.
+   * We must also scale the output by (8/13)**2 = 64/169, which we partially
+   * fold into the constant multipliers and final shifting:
+   * cK now represents sqrt(2) * cos(K*pi/26) * 128/169.
+   */
+
+  dataptr = data;
+  wsptr = workspace;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*4];
+    tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*3];
+    tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*2];
+    tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*1];
+    tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*0];
+    tmp5 = dataptr[DCTSIZE*5] + dataptr[DCTSIZE*7];
+    tmp6 = dataptr[DCTSIZE*6];
+
+    tmp10 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*4];
+    tmp11 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*3];
+    tmp12 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*2];
+    tmp13 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*1];
+    tmp14 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*0];
+    tmp15 = dataptr[DCTSIZE*5] - dataptr[DCTSIZE*7];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0 + tmp1 + tmp2 + tmp3 + tmp4 + tmp5 + tmp6,
+		       FIX(0.757396450)),          /* 128/169 */
+	      CONST_BITS+1);
+    tmp6 += tmp6;
+    tmp0 -= tmp6;
+    tmp1 -= tmp6;
+    tmp2 -= tmp6;
+    tmp3 -= tmp6;
+    tmp4 -= tmp6;
+    tmp5 -= tmp6;
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0, FIX(1.039995521)) +   /* c2 */
+	      MULTIPLY(tmp1, FIX(0.801745081)) +   /* c6 */
+	      MULTIPLY(tmp2, FIX(0.379824504)) -   /* c10 */
+	      MULTIPLY(tmp3, FIX(0.129109289)) -   /* c12 */
+	      MULTIPLY(tmp4, FIX(0.608465700)) -   /* c8 */
+	      MULTIPLY(tmp5, FIX(0.948429952)),    /* c4 */
+	      CONST_BITS+1);
+    z1 = MULTIPLY(tmp0 - tmp2, FIX(0.875087516)) - /* (c4+c6)/2 */
+	 MULTIPLY(tmp3 - tmp4, FIX(0.330085509)) - /* (c2-c10)/2 */
+	 MULTIPLY(tmp1 - tmp5, FIX(0.239678205));  /* (c8-c12)/2 */
+    z2 = MULTIPLY(tmp0 + tmp2, FIX(0.073342435)) - /* (c4-c6)/2 */
+	 MULTIPLY(tmp3 + tmp4, FIX(0.709910013)) + /* (c2+c10)/2 */
+	 MULTIPLY(tmp1 + tmp5, FIX(0.368787494));  /* (c8+c12)/2 */
+
+    dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(z1 + z2, CONST_BITS+1);
+    dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 - z2, CONST_BITS+1);
+
+    /* Odd part */
+
+    tmp1 = MULTIPLY(tmp10 + tmp11, FIX(1.001514908));   /* c3 */
+    tmp2 = MULTIPLY(tmp10 + tmp12, FIX(0.881514751));   /* c5 */
+    tmp3 = MULTIPLY(tmp10 + tmp13, FIX(0.710284161)) +  /* c7 */
+	   MULTIPLY(tmp14 + tmp15, FIX(0.256335874));   /* c11 */
+    tmp0 = tmp1 + tmp2 + tmp3 -
+	   MULTIPLY(tmp10, FIX(1.530003162)) +          /* c3+c5+c7-c1 */
+	   MULTIPLY(tmp14, FIX(0.241438564));           /* c9-c11 */
+    tmp4 = MULTIPLY(tmp14 - tmp15, FIX(0.710284161)) -  /* c7 */
+	   MULTIPLY(tmp11 + tmp12, FIX(0.256335874));   /* c11 */
+    tmp5 = MULTIPLY(tmp11 + tmp13, - FIX(0.881514751)); /* -c5 */
+    tmp1 += tmp4 + tmp5 +
+	    MULTIPLY(tmp11, FIX(0.634110155)) -         /* c5+c9+c11-c3 */
+	    MULTIPLY(tmp14, FIX(1.773594819));          /* c1+c7 */
+    tmp6 = MULTIPLY(tmp12 + tmp13, - FIX(0.497774438)); /* -c9 */
+    tmp2 += tmp4 + tmp6 -
+	    MULTIPLY(tmp12, FIX(1.190715098)) +         /* c1+c5-c9-c11 */
+	    MULTIPLY(tmp15, FIX(1.711799069));          /* c3+c7 */
+    tmp3 += tmp5 + tmp6 +
+	    MULTIPLY(tmp13, FIX(1.670519935)) -         /* c3+c5+c9-c7 */
+	    MULTIPLY(tmp15, FIX(1.319646532));          /* c1+c11 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp0, CONST_BITS+1);
+    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp1, CONST_BITS+1);
+    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp2, CONST_BITS+1);
+    dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp3, CONST_BITS+1);
+
+    dataptr++;			/* advance pointer to next column */
+    wsptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 14x14 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_14x14 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
+  DCTELEM workspace[8*6];
+  DCTELEM *dataptr;
+  DCTELEM *wsptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT. */
+  /* cK represents sqrt(2) * cos(K*pi/28). */
+
+  dataptr = data;
+  ctr = 0;
+  for (;;) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[13]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[12]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[11]);
+    tmp13 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[10]);
+    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[9]);
+    tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[8]);
+    tmp6 = GETJSAMPLE(elemptr[6]) + GETJSAMPLE(elemptr[7]);
+
+    tmp10 = tmp0 + tmp6;
+    tmp14 = tmp0 - tmp6;
+    tmp11 = tmp1 + tmp5;
+    tmp15 = tmp1 - tmp5;
+    tmp12 = tmp2 + tmp4;
+    tmp16 = tmp2 - tmp4;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[13]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[12]);
+    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[11]);
+    tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[10]);
+    tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[9]);
+    tmp5 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[8]);
+    tmp6 = GETJSAMPLE(elemptr[6]) - GETJSAMPLE(elemptr[7]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      (tmp10 + tmp11 + tmp12 + tmp13 - 14 * CENTERJSAMPLE);
+    tmp13 += tmp13;
+    dataptr[4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp13, FIX(1.274162392)) + /* c4 */
+	      MULTIPLY(tmp11 - tmp13, FIX(0.314692123)) - /* c12 */
+	      MULTIPLY(tmp12 - tmp13, FIX(0.881747734)),  /* c8 */
+	      CONST_BITS);
+
+    tmp10 = MULTIPLY(tmp14 + tmp15, FIX(1.105676686));    /* c6 */
+
+    dataptr[2] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp14, FIX(0.273079590))   /* c2-c6 */
+	      + MULTIPLY(tmp16, FIX(0.613604268)),        /* c10 */
+	      CONST_BITS);
+    dataptr[6] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp15, FIX(1.719280954))   /* c6+c10 */
+	      - MULTIPLY(tmp16, FIX(1.378756276)),        /* c2 */
+	      CONST_BITS);
+
+    /* Odd part */
+
+    tmp10 = tmp1 + tmp2;
+    tmp11 = tmp5 - tmp4;
+    dataptr[7] = (DCTELEM) (tmp0 - tmp10 + tmp3 - tmp11 - tmp6);
+    tmp3 <<= CONST_BITS;
+    tmp10 = MULTIPLY(tmp10, - FIX(0.158341681));          /* -c13 */
+    tmp11 = MULTIPLY(tmp11, FIX(1.405321284));            /* c1 */
+    tmp10 += tmp11 - tmp3;
+    tmp11 = MULTIPLY(tmp0 + tmp2, FIX(1.197448846)) +     /* c5 */
+	    MULTIPLY(tmp4 + tmp6, FIX(0.752406978));      /* c9 */
+    dataptr[5] = (DCTELEM)
+      DESCALE(tmp10 + tmp11 - MULTIPLY(tmp2, FIX(2.373959773)) /* c3+c5-c13 */
+	      + MULTIPLY(tmp4, FIX(1.119999435)),         /* c1+c11-c9 */
+	      CONST_BITS);
+    tmp12 = MULTIPLY(tmp0 + tmp1, FIX(1.334852607)) +     /* c3 */
+	    MULTIPLY(tmp5 - tmp6, FIX(0.467085129));      /* c11 */
+    dataptr[3] = (DCTELEM)
+      DESCALE(tmp10 + tmp12 - MULTIPLY(tmp1, FIX(0.424103948)) /* c3-c9-c13 */
+	      - MULTIPLY(tmp5, FIX(3.069855259)),         /* c1+c5+c11 */
+	      CONST_BITS);
+    dataptr[1] = (DCTELEM)
+      DESCALE(tmp11 + tmp12 + tmp3 + tmp6 -
+	      MULTIPLY(tmp0 + tmp6, FIX(1.126980169)),    /* c3+c5-c1 */
+	      CONST_BITS);
+
+    ctr++;
+
+    if (ctr != DCTSIZE) {
+      if (ctr == 14)
+	break;			/* Done. */
+      dataptr += DCTSIZE;	/* advance pointer to next row */
+    } else
+      dataptr = workspace;	/* switch pointer to extended workspace */
+  }
+
+  /* Pass 2: process columns.
+   * We leave the results scaled up by an overall factor of 8.
+   * We must also scale the output by (8/14)**2 = 16/49, which we partially
+   * fold into the constant multipliers and final shifting:
+   * cK now represents sqrt(2) * cos(K*pi/28) * 32/49.
+   */
+
+  dataptr = data;
+  wsptr = workspace;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*5];
+    tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*4];
+    tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*3];
+    tmp13 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*2];
+    tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*1];
+    tmp5 = dataptr[DCTSIZE*5] + wsptr[DCTSIZE*0];
+    tmp6 = dataptr[DCTSIZE*6] + dataptr[DCTSIZE*7];
+
+    tmp10 = tmp0 + tmp6;
+    tmp14 = tmp0 - tmp6;
+    tmp11 = tmp1 + tmp5;
+    tmp15 = tmp1 - tmp5;
+    tmp12 = tmp2 + tmp4;
+    tmp16 = tmp2 - tmp4;
+
+    tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*5];
+    tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*4];
+    tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*3];
+    tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*2];
+    tmp4 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*1];
+    tmp5 = dataptr[DCTSIZE*5] - wsptr[DCTSIZE*0];
+    tmp6 = dataptr[DCTSIZE*6] - dataptr[DCTSIZE*7];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12 + tmp13,
+		       FIX(0.653061224)),                 /* 32/49 */
+	      CONST_BITS+1);
+    tmp13 += tmp13;
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp13, FIX(0.832106052)) + /* c4 */
+	      MULTIPLY(tmp11 - tmp13, FIX(0.205513223)) - /* c12 */
+	      MULTIPLY(tmp12 - tmp13, FIX(0.575835255)),  /* c8 */
+	      CONST_BITS+1);
+
+    tmp10 = MULTIPLY(tmp14 + tmp15, FIX(0.722074570));    /* c6 */
+
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp14, FIX(0.178337691))   /* c2-c6 */
+	      + MULTIPLY(tmp16, FIX(0.400721155)),        /* c10 */
+	      CONST_BITS+1);
+    dataptr[DCTSIZE*6] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp15, FIX(1.122795725))   /* c6+c10 */
+	      - MULTIPLY(tmp16, FIX(0.900412262)),        /* c2 */
+	      CONST_BITS+1);
+
+    /* Odd part */
+
+    tmp10 = tmp1 + tmp2;
+    tmp11 = tmp5 - tmp4;
+    dataptr[DCTSIZE*7] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0 - tmp10 + tmp3 - tmp11 - tmp6,
+		       FIX(0.653061224)),                 /* 32/49 */
+	      CONST_BITS+1);
+    tmp3  = MULTIPLY(tmp3 , FIX(0.653061224));            /* 32/49 */
+    tmp10 = MULTIPLY(tmp10, - FIX(0.103406812));          /* -c13 */
+    tmp11 = MULTIPLY(tmp11, FIX(0.917760839));            /* c1 */
+    tmp10 += tmp11 - tmp3;
+    tmp11 = MULTIPLY(tmp0 + tmp2, FIX(0.782007410)) +     /* c5 */
+	    MULTIPLY(tmp4 + tmp6, FIX(0.491367823));      /* c9 */
+    dataptr[DCTSIZE*5] = (DCTELEM)
+      DESCALE(tmp10 + tmp11 - MULTIPLY(tmp2, FIX(1.550341076)) /* c3+c5-c13 */
+	      + MULTIPLY(tmp4, FIX(0.731428202)),         /* c1+c11-c9 */
+	      CONST_BITS+1);
+    tmp12 = MULTIPLY(tmp0 + tmp1, FIX(0.871740478)) +     /* c3 */
+	    MULTIPLY(tmp5 - tmp6, FIX(0.305035186));      /* c11 */
+    dataptr[DCTSIZE*3] = (DCTELEM)
+      DESCALE(tmp10 + tmp12 - MULTIPLY(tmp1, FIX(0.276965844)) /* c3-c9-c13 */
+	      - MULTIPLY(tmp5, FIX(2.004803435)),         /* c1+c5+c11 */
+	      CONST_BITS+1);
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      DESCALE(tmp11 + tmp12 + tmp3
+	      - MULTIPLY(tmp0, FIX(0.735987049))          /* c3+c5-c1 */
+	      - MULTIPLY(tmp6, FIX(0.082925825)),         /* c9-c11-c13 */
+	      CONST_BITS+1);
+
+    dataptr++;			/* advance pointer to next column */
+    wsptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 15x15 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_15x15 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
+  INT32 z1, z2, z3;
+  DCTELEM workspace[8*7];
+  DCTELEM *dataptr;
+  DCTELEM *wsptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT. */
+  /* cK represents sqrt(2) * cos(K*pi/30). */
+
+  dataptr = data;
+  ctr = 0;
+  for (;;) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[14]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[13]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[12]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[11]);
+    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[10]);
+    tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[9]);
+    tmp6 = GETJSAMPLE(elemptr[6]) + GETJSAMPLE(elemptr[8]);
+    tmp7 = GETJSAMPLE(elemptr[7]);
+
+    tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[14]);
+    tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[13]);
+    tmp12 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[12]);
+    tmp13 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[11]);
+    tmp14 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[10]);
+    tmp15 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[9]);
+    tmp16 = GETJSAMPLE(elemptr[6]) - GETJSAMPLE(elemptr[8]);
+
+    z1 = tmp0 + tmp4 + tmp5;
+    z2 = tmp1 + tmp3 + tmp6;
+    z3 = tmp2 + tmp7;
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM) (z1 + z2 + z3 - 15 * CENTERJSAMPLE);
+    z3 += z3;
+    dataptr[6] = (DCTELEM)
+      DESCALE(MULTIPLY(z1 - z3, FIX(1.144122806)) - /* c6 */
+	      MULTIPLY(z2 - z3, FIX(0.437016024)),  /* c12 */
+	      CONST_BITS);
+    tmp2 += ((tmp1 + tmp4) >> 1) - tmp7 - tmp7;
+    z1 = MULTIPLY(tmp3 - tmp2, FIX(1.531135173)) -  /* c2+c14 */
+         MULTIPLY(tmp6 - tmp2, FIX(2.238241955));   /* c4+c8 */
+    z2 = MULTIPLY(tmp5 - tmp2, FIX(0.798468008)) -  /* c8-c14 */
+	 MULTIPLY(tmp0 - tmp2, FIX(0.091361227));   /* c2-c4 */
+    z3 = MULTIPLY(tmp0 - tmp3, FIX(1.383309603)) +  /* c2 */
+	 MULTIPLY(tmp6 - tmp5, FIX(0.946293579)) +  /* c8 */
+	 MULTIPLY(tmp1 - tmp4, FIX(0.790569415));   /* (c6+c12)/2 */
+
+    dataptr[2] = (DCTELEM) DESCALE(z1 + z3, CONST_BITS);
+    dataptr[4] = (DCTELEM) DESCALE(z2 + z3, CONST_BITS);
+
+    /* Odd part */
+
+    tmp2 = MULTIPLY(tmp10 - tmp12 - tmp13 + tmp15 + tmp16,
+		    FIX(1.224744871));                         /* c5 */
+    tmp1 = MULTIPLY(tmp10 - tmp14 - tmp15, FIX(1.344997024)) + /* c3 */
+	   MULTIPLY(tmp11 - tmp13 - tmp16, FIX(0.831253876));  /* c9 */
+    tmp12 = MULTIPLY(tmp12, FIX(1.224744871));                 /* c5 */
+    tmp4 = MULTIPLY(tmp10 - tmp16, FIX(1.406466353)) +         /* c1 */
+	   MULTIPLY(tmp11 + tmp14, FIX(1.344997024)) +         /* c3 */
+	   MULTIPLY(tmp13 + tmp15, FIX(0.575212477));          /* c11 */
+    tmp0 = MULTIPLY(tmp13, FIX(0.475753014)) -                 /* c7-c11 */
+	   MULTIPLY(tmp14, FIX(0.513743148)) +                 /* c3-c9 */
+	   MULTIPLY(tmp16, FIX(1.700497885)) + tmp4 + tmp12;   /* c1+c13 */
+    tmp3 = MULTIPLY(tmp10, - FIX(0.355500862)) -               /* -(c1-c7) */
+	   MULTIPLY(tmp11, FIX(2.176250899)) -                 /* c3+c9 */
+	   MULTIPLY(tmp15, FIX(0.869244010)) + tmp4 - tmp12;   /* c11+c13 */
+
+    dataptr[1] = (DCTELEM) DESCALE(tmp0, CONST_BITS);
+    dataptr[3] = (DCTELEM) DESCALE(tmp1, CONST_BITS);
+    dataptr[5] = (DCTELEM) DESCALE(tmp2, CONST_BITS);
+    dataptr[7] = (DCTELEM) DESCALE(tmp3, CONST_BITS);
+
+    ctr++;
+
+    if (ctr != DCTSIZE) {
+      if (ctr == 15)
+	break;			/* Done. */
+      dataptr += DCTSIZE;	/* advance pointer to next row */
+    } else
+      dataptr = workspace;	/* switch pointer to extended workspace */
+  }
+
+  /* Pass 2: process columns.
+   * We leave the results scaled up by an overall factor of 8.
+   * We must also scale the output by (8/15)**2 = 64/225, which we partially
+   * fold into the constant multipliers and final shifting:
+   * cK now represents sqrt(2) * cos(K*pi/30) * 256/225.
+   */
+
+  dataptr = data;
+  wsptr = workspace;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*6];
+    tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*5];
+    tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*4];
+    tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*3];
+    tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*2];
+    tmp5 = dataptr[DCTSIZE*5] + wsptr[DCTSIZE*1];
+    tmp6 = dataptr[DCTSIZE*6] + wsptr[DCTSIZE*0];
+    tmp7 = dataptr[DCTSIZE*7];
+
+    tmp10 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*6];
+    tmp11 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*5];
+    tmp12 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*4];
+    tmp13 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*3];
+    tmp14 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*2];
+    tmp15 = dataptr[DCTSIZE*5] - wsptr[DCTSIZE*1];
+    tmp16 = dataptr[DCTSIZE*6] - wsptr[DCTSIZE*0];
+
+    z1 = tmp0 + tmp4 + tmp5;
+    z2 = tmp1 + tmp3 + tmp6;
+    z3 = tmp2 + tmp7;
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(z1 + z2 + z3, FIX(1.137777778)), /* 256/225 */
+	      CONST_BITS+2);
+    z3 += z3;
+    dataptr[DCTSIZE*6] = (DCTELEM)
+      DESCALE(MULTIPLY(z1 - z3, FIX(1.301757503)) - /* c6 */
+	      MULTIPLY(z2 - z3, FIX(0.497227121)),  /* c12 */
+	      CONST_BITS+2);
+    tmp2 += ((tmp1 + tmp4) >> 1) - tmp7 - tmp7;
+    z1 = MULTIPLY(tmp3 - tmp2, FIX(1.742091575)) -  /* c2+c14 */
+         MULTIPLY(tmp6 - tmp2, FIX(2.546621957));   /* c4+c8 */
+    z2 = MULTIPLY(tmp5 - tmp2, FIX(0.908479156)) -  /* c8-c14 */
+	 MULTIPLY(tmp0 - tmp2, FIX(0.103948774));   /* c2-c4 */
+    z3 = MULTIPLY(tmp0 - tmp3, FIX(1.573898926)) +  /* c2 */
+	 MULTIPLY(tmp6 - tmp5, FIX(1.076671805)) +  /* c8 */
+	 MULTIPLY(tmp1 - tmp4, FIX(0.899492312));   /* (c6+c12)/2 */
+
+    dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + z3, CONST_BITS+2);
+    dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(z2 + z3, CONST_BITS+2);
+
+    /* Odd part */
+
+    tmp2 = MULTIPLY(tmp10 - tmp12 - tmp13 + tmp15 + tmp16,
+		    FIX(1.393487498));                         /* c5 */
+    tmp1 = MULTIPLY(tmp10 - tmp14 - tmp15, FIX(1.530307725)) + /* c3 */
+	   MULTIPLY(tmp11 - tmp13 - tmp16, FIX(0.945782187));  /* c9 */
+    tmp12 = MULTIPLY(tmp12, FIX(1.393487498));                 /* c5 */
+    tmp4 = MULTIPLY(tmp10 - tmp16, FIX(1.600246161)) +         /* c1 */
+	   MULTIPLY(tmp11 + tmp14, FIX(1.530307725)) +         /* c3 */
+	   MULTIPLY(tmp13 + tmp15, FIX(0.654463974));          /* c11 */
+    tmp0 = MULTIPLY(tmp13, FIX(0.541301207)) -                 /* c7-c11 */
+	   MULTIPLY(tmp14, FIX(0.584525538)) +                 /* c3-c9 */
+	   MULTIPLY(tmp16, FIX(1.934788705)) + tmp4 + tmp12;   /* c1+c13 */
+    tmp3 = MULTIPLY(tmp10, - FIX(0.404480980)) -               /* -(c1-c7) */
+	   MULTIPLY(tmp11, FIX(2.476089912)) -                 /* c3+c9 */
+	   MULTIPLY(tmp15, FIX(0.989006518)) + tmp4 - tmp12;   /* c11+c13 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp0, CONST_BITS+2);
+    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp1, CONST_BITS+2);
+    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp2, CONST_BITS+2);
+    dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp3, CONST_BITS+2);
+
+    dataptr++;			/* advance pointer to next column */
+    wsptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 16x16 sample block.
+ */
+
+GLOBAL(void)
+jpeg_fdct_16x16 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16, tmp17;
+  DCTELEM workspace[DCTSIZE2];
+  DCTELEM *dataptr;
+  DCTELEM *wsptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* cK represents sqrt(2) * cos(K*pi/32). */
+
+  dataptr = data;
+  ctr = 0;
+  for (;;) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[15]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[14]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[13]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[12]);
+    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[11]);
+    tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[10]);
+    tmp6 = GETJSAMPLE(elemptr[6]) + GETJSAMPLE(elemptr[9]);
+    tmp7 = GETJSAMPLE(elemptr[7]) + GETJSAMPLE(elemptr[8]);
+
+    tmp10 = tmp0 + tmp7;
+    tmp14 = tmp0 - tmp7;
+    tmp11 = tmp1 + tmp6;
+    tmp15 = tmp1 - tmp6;
+    tmp12 = tmp2 + tmp5;
+    tmp16 = tmp2 - tmp5;
+    tmp13 = tmp3 + tmp4;
+    tmp17 = tmp3 - tmp4;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[15]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[14]);
+    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[13]);
+    tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[12]);
+    tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[11]);
+    tmp5 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[10]);
+    tmp6 = GETJSAMPLE(elemptr[6]) - GETJSAMPLE(elemptr[9]);
+    tmp7 = GETJSAMPLE(elemptr[7]) - GETJSAMPLE(elemptr[8]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp10 + tmp11 + tmp12 + tmp13 - 16 * CENTERJSAMPLE) << PASS1_BITS);
+    dataptr[4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp13, FIX(1.306562965)) + /* c4[16] = c2[8] */
+	      MULTIPLY(tmp11 - tmp12, FIX_0_541196100),   /* c12[16] = c6[8] */
+	      CONST_BITS-PASS1_BITS);
+
+    tmp10 = MULTIPLY(tmp17 - tmp15, FIX(0.275899379)) +   /* c14[16] = c7[8] */
+	    MULTIPLY(tmp14 - tmp16, FIX(1.387039845));    /* c2[16] = c1[8] */
+
+    dataptr[2] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp15, FIX(1.451774982))   /* c6+c14 */
+	      + MULTIPLY(tmp16, FIX(2.172734804)),        /* c2+c10 */
+	      CONST_BITS-PASS1_BITS);
+    dataptr[6] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp14, FIX(0.211164243))   /* c2-c6 */
+	      - MULTIPLY(tmp17, FIX(1.061594338)),        /* c10+c14 */
+	      CONST_BITS-PASS1_BITS);
+
+    /* Odd part */
+
+    tmp11 = MULTIPLY(tmp0 + tmp1, FIX(1.353318001)) +         /* c3 */
+	    MULTIPLY(tmp6 - tmp7, FIX(0.410524528));          /* c13 */
+    tmp12 = MULTIPLY(tmp0 + tmp2, FIX(1.247225013)) +         /* c5 */
+	    MULTIPLY(tmp5 + tmp7, FIX(0.666655658));          /* c11 */
+    tmp13 = MULTIPLY(tmp0 + tmp3, FIX(1.093201867)) +         /* c7 */
+	    MULTIPLY(tmp4 - tmp7, FIX(0.897167586));          /* c9 */
+    tmp14 = MULTIPLY(tmp1 + tmp2, FIX(0.138617169)) +         /* c15 */
+	    MULTIPLY(tmp6 - tmp5, FIX(1.407403738));          /* c1 */
+    tmp15 = MULTIPLY(tmp1 + tmp3, - FIX(0.666655658)) +       /* -c11 */
+	    MULTIPLY(tmp4 + tmp6, - FIX(1.247225013));        /* -c5 */
+    tmp16 = MULTIPLY(tmp2 + tmp3, - FIX(1.353318001)) +       /* -c3 */
+	    MULTIPLY(tmp5 - tmp4, FIX(0.410524528));          /* c13 */
+    tmp10 = tmp11 + tmp12 + tmp13 -
+	    MULTIPLY(tmp0, FIX(2.286341144)) +                /* c7+c5+c3-c1 */
+	    MULTIPLY(tmp7, FIX(0.779653625));                 /* c15+c13-c11+c9 */
+    tmp11 += tmp14 + tmp15 + MULTIPLY(tmp1, FIX(0.071888074)) /* c9-c3-c15+c11 */
+	     - MULTIPLY(tmp6, FIX(1.663905119));              /* c7+c13+c1-c5 */
+    tmp12 += tmp14 + tmp16 - MULTIPLY(tmp2, FIX(1.125726048)) /* c7+c5+c15-c3 */
+	     + MULTIPLY(tmp5, FIX(1.227391138));              /* c9-c11+c1-c13 */
+    tmp13 += tmp15 + tmp16 + MULTIPLY(tmp3, FIX(1.065388962)) /* c15+c3+c11-c7 */
+	     + MULTIPLY(tmp4, FIX(2.167985692));              /* c1+c13+c5-c9 */
+
+    dataptr[1] = (DCTELEM) DESCALE(tmp10, CONST_BITS-PASS1_BITS);
+    dataptr[3] = (DCTELEM) DESCALE(tmp11, CONST_BITS-PASS1_BITS);
+    dataptr[5] = (DCTELEM) DESCALE(tmp12, CONST_BITS-PASS1_BITS);
+    dataptr[7] = (DCTELEM) DESCALE(tmp13, CONST_BITS-PASS1_BITS);
+
+    ctr++;
+
+    if (ctr != DCTSIZE) {
+      if (ctr == DCTSIZE * 2)
+	break;			/* Done. */
+      dataptr += DCTSIZE;	/* advance pointer to next row */
+    } else
+      dataptr = workspace;	/* switch pointer to extended workspace */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by (8/16)**2 = 1/2**2.
+   */
+
+  dataptr = data;
+  wsptr = workspace;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*7];
+    tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*6];
+    tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*5];
+    tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*4];
+    tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*3];
+    tmp5 = dataptr[DCTSIZE*5] + wsptr[DCTSIZE*2];
+    tmp6 = dataptr[DCTSIZE*6] + wsptr[DCTSIZE*1];
+    tmp7 = dataptr[DCTSIZE*7] + wsptr[DCTSIZE*0];
+
+    tmp10 = tmp0 + tmp7;
+    tmp14 = tmp0 - tmp7;
+    tmp11 = tmp1 + tmp6;
+    tmp15 = tmp1 - tmp6;
+    tmp12 = tmp2 + tmp5;
+    tmp16 = tmp2 - tmp5;
+    tmp13 = tmp3 + tmp4;
+    tmp17 = tmp3 - tmp4;
+
+    tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*7];
+    tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*6];
+    tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*5];
+    tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*4];
+    tmp4 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*3];
+    tmp5 = dataptr[DCTSIZE*5] - wsptr[DCTSIZE*2];
+    tmp6 = dataptr[DCTSIZE*6] - wsptr[DCTSIZE*1];
+    tmp7 = dataptr[DCTSIZE*7] - wsptr[DCTSIZE*0];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(tmp10 + tmp11 + tmp12 + tmp13, PASS1_BITS+2);
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp13, FIX(1.306562965)) + /* c4[16] = c2[8] */
+	      MULTIPLY(tmp11 - tmp12, FIX_0_541196100),   /* c12[16] = c6[8] */
+	      CONST_BITS+PASS1_BITS+2);
+
+    tmp10 = MULTIPLY(tmp17 - tmp15, FIX(0.275899379)) +   /* c14[16] = c7[8] */
+	    MULTIPLY(tmp14 - tmp16, FIX(1.387039845));    /* c2[16] = c1[8] */
+
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp15, FIX(1.451774982))   /* c6+c14 */
+	      + MULTIPLY(tmp16, FIX(2.172734804)),        /* c2+10 */
+	      CONST_BITS+PASS1_BITS+2);
+    dataptr[DCTSIZE*6] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp14, FIX(0.211164243))   /* c2-c6 */
+	      - MULTIPLY(tmp17, FIX(1.061594338)),        /* c10+c14 */
+	      CONST_BITS+PASS1_BITS+2);
+
+    /* Odd part */
+
+    tmp11 = MULTIPLY(tmp0 + tmp1, FIX(1.353318001)) +         /* c3 */
+	    MULTIPLY(tmp6 - tmp7, FIX(0.410524528));          /* c13 */
+    tmp12 = MULTIPLY(tmp0 + tmp2, FIX(1.247225013)) +         /* c5 */
+	    MULTIPLY(tmp5 + tmp7, FIX(0.666655658));          /* c11 */
+    tmp13 = MULTIPLY(tmp0 + tmp3, FIX(1.093201867)) +         /* c7 */
+	    MULTIPLY(tmp4 - tmp7, FIX(0.897167586));          /* c9 */
+    tmp14 = MULTIPLY(tmp1 + tmp2, FIX(0.138617169)) +         /* c15 */
+	    MULTIPLY(tmp6 - tmp5, FIX(1.407403738));          /* c1 */
+    tmp15 = MULTIPLY(tmp1 + tmp3, - FIX(0.666655658)) +       /* -c11 */
+	    MULTIPLY(tmp4 + tmp6, - FIX(1.247225013));        /* -c5 */
+    tmp16 = MULTIPLY(tmp2 + tmp3, - FIX(1.353318001)) +       /* -c3 */
+	    MULTIPLY(tmp5 - tmp4, FIX(0.410524528));          /* c13 */
+    tmp10 = tmp11 + tmp12 + tmp13 -
+	    MULTIPLY(tmp0, FIX(2.286341144)) +                /* c7+c5+c3-c1 */
+	    MULTIPLY(tmp7, FIX(0.779653625));                 /* c15+c13-c11+c9 */
+    tmp11 += tmp14 + tmp15 + MULTIPLY(tmp1, FIX(0.071888074)) /* c9-c3-c15+c11 */
+	     - MULTIPLY(tmp6, FIX(1.663905119));              /* c7+c13+c1-c5 */
+    tmp12 += tmp14 + tmp16 - MULTIPLY(tmp2, FIX(1.125726048)) /* c7+c5+c15-c3 */
+	     + MULTIPLY(tmp5, FIX(1.227391138));              /* c9-c11+c1-c13 */
+    tmp13 += tmp15 + tmp16 + MULTIPLY(tmp3, FIX(1.065388962)) /* c15+c3+c11-c7 */
+	     + MULTIPLY(tmp4, FIX(2.167985692));              /* c1+c13+c5-c9 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp10, CONST_BITS+PASS1_BITS+2);
+    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp11, CONST_BITS+PASS1_BITS+2);
+    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp12, CONST_BITS+PASS1_BITS+2);
+    dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp13, CONST_BITS+PASS1_BITS+2);
+
+    dataptr++;			/* advance pointer to next column */
+    wsptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 16x8 sample block.
+ *
+ * 16-point FDCT in pass 1 (rows), 8-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_16x8 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16, tmp17;
+  INT32 z1;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* 16-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/32). */
+
+  dataptr = data;
+  ctr = 0;
+  for (ctr = 0; ctr < DCTSIZE; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[15]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[14]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[13]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[12]);
+    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[11]);
+    tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[10]);
+    tmp6 = GETJSAMPLE(elemptr[6]) + GETJSAMPLE(elemptr[9]);
+    tmp7 = GETJSAMPLE(elemptr[7]) + GETJSAMPLE(elemptr[8]);
+
+    tmp10 = tmp0 + tmp7;
+    tmp14 = tmp0 - tmp7;
+    tmp11 = tmp1 + tmp6;
+    tmp15 = tmp1 - tmp6;
+    tmp12 = tmp2 + tmp5;
+    tmp16 = tmp2 - tmp5;
+    tmp13 = tmp3 + tmp4;
+    tmp17 = tmp3 - tmp4;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[15]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[14]);
+    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[13]);
+    tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[12]);
+    tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[11]);
+    tmp5 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[10]);
+    tmp6 = GETJSAMPLE(elemptr[6]) - GETJSAMPLE(elemptr[9]);
+    tmp7 = GETJSAMPLE(elemptr[7]) - GETJSAMPLE(elemptr[8]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp10 + tmp11 + tmp12 + tmp13 - 16 * CENTERJSAMPLE) << PASS1_BITS);
+    dataptr[4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp13, FIX(1.306562965)) + /* c4[16] = c2[8] */
+	      MULTIPLY(tmp11 - tmp12, FIX_0_541196100),   /* c12[16] = c6[8] */
+	      CONST_BITS-PASS1_BITS);
+
+    tmp10 = MULTIPLY(tmp17 - tmp15, FIX(0.275899379)) +   /* c14[16] = c7[8] */
+	    MULTIPLY(tmp14 - tmp16, FIX(1.387039845));    /* c2[16] = c1[8] */
+
+    dataptr[2] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp15, FIX(1.451774982))   /* c6+c14 */
+	      + MULTIPLY(tmp16, FIX(2.172734804)),        /* c2+c10 */
+	      CONST_BITS-PASS1_BITS);
+    dataptr[6] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp14, FIX(0.211164243))   /* c2-c6 */
+	      - MULTIPLY(tmp17, FIX(1.061594338)),        /* c10+c14 */
+	      CONST_BITS-PASS1_BITS);
+
+    /* Odd part */
+
+    tmp11 = MULTIPLY(tmp0 + tmp1, FIX(1.353318001)) +         /* c3 */
+	    MULTIPLY(tmp6 - tmp7, FIX(0.410524528));          /* c13 */
+    tmp12 = MULTIPLY(tmp0 + tmp2, FIX(1.247225013)) +         /* c5 */
+	    MULTIPLY(tmp5 + tmp7, FIX(0.666655658));          /* c11 */
+    tmp13 = MULTIPLY(tmp0 + tmp3, FIX(1.093201867)) +         /* c7 */
+	    MULTIPLY(tmp4 - tmp7, FIX(0.897167586));          /* c9 */
+    tmp14 = MULTIPLY(tmp1 + tmp2, FIX(0.138617169)) +         /* c15 */
+	    MULTIPLY(tmp6 - tmp5, FIX(1.407403738));          /* c1 */
+    tmp15 = MULTIPLY(tmp1 + tmp3, - FIX(0.666655658)) +       /* -c11 */
+	    MULTIPLY(tmp4 + tmp6, - FIX(1.247225013));        /* -c5 */
+    tmp16 = MULTIPLY(tmp2 + tmp3, - FIX(1.353318001)) +       /* -c3 */
+	    MULTIPLY(tmp5 - tmp4, FIX(0.410524528));          /* c13 */
+    tmp10 = tmp11 + tmp12 + tmp13 -
+	    MULTIPLY(tmp0, FIX(2.286341144)) +                /* c7+c5+c3-c1 */
+	    MULTIPLY(tmp7, FIX(0.779653625));                 /* c15+c13-c11+c9 */
+    tmp11 += tmp14 + tmp15 + MULTIPLY(tmp1, FIX(0.071888074)) /* c9-c3-c15+c11 */
+	     - MULTIPLY(tmp6, FIX(1.663905119));              /* c7+c13+c1-c5 */
+    tmp12 += tmp14 + tmp16 - MULTIPLY(tmp2, FIX(1.125726048)) /* c7+c5+c15-c3 */
+	     + MULTIPLY(tmp5, FIX(1.227391138));              /* c9-c11+c1-c13 */
+    tmp13 += tmp15 + tmp16 + MULTIPLY(tmp3, FIX(1.065388962)) /* c15+c3+c11-c7 */
+	     + MULTIPLY(tmp4, FIX(2.167985692));              /* c1+c13+c5-c9 */
+
+    dataptr[1] = (DCTELEM) DESCALE(tmp10, CONST_BITS-PASS1_BITS);
+    dataptr[3] = (DCTELEM) DESCALE(tmp11, CONST_BITS-PASS1_BITS);
+    dataptr[5] = (DCTELEM) DESCALE(tmp12, CONST_BITS-PASS1_BITS);
+    dataptr[7] = (DCTELEM) DESCALE(tmp13, CONST_BITS-PASS1_BITS);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by 8/16 = 1/2.
+   */
+
+  dataptr = data;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part per LL&M figure 1 --- note that published figure is faulty;
+     * rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
+     */
+
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
+    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
+    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
+    tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
+
+    tmp10 = tmp0 + tmp3;
+    tmp12 = tmp0 - tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp13 = tmp1 - tmp2;
+
+    tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
+    tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
+    tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
+    tmp3 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
+
+    dataptr[DCTSIZE*0] = (DCTELEM) DESCALE(tmp10 + tmp11, PASS1_BITS+1);
+    dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp10 - tmp11, PASS1_BITS+1);
+
+    z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
+    dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, FIX_0_765366865),
+					   CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 - MULTIPLY(tmp13, FIX_1_847759065),
+					   CONST_BITS+PASS1_BITS+1);
+
+    /* Odd part per figure 8 --- note paper omits factor of sqrt(2).
+     * 8-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
+     * i0..i3 in the paper are tmp0..tmp3 here.
+     */
+
+    tmp10 = tmp0 + tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp12 = tmp0 + tmp2;
+    tmp13 = tmp1 + tmp3;
+    z1 = MULTIPLY(tmp12 + tmp13, FIX_1_175875602); /*  c3 */
+
+    tmp0  = MULTIPLY(tmp0,    FIX_1_501321110);    /*  c1+c3-c5-c7 */
+    tmp1  = MULTIPLY(tmp1,    FIX_3_072711026);    /*  c1+c3+c5-c7 */
+    tmp2  = MULTIPLY(tmp2,    FIX_2_053119869);    /*  c1+c3-c5+c7 */
+    tmp3  = MULTIPLY(tmp3,    FIX_0_298631336);    /* -c1+c3+c5-c7 */
+    tmp10 = MULTIPLY(tmp10, - FIX_0_899976223);    /*  c7-c3 */
+    tmp11 = MULTIPLY(tmp11, - FIX_2_562915447);    /* -c1-c3 */
+    tmp12 = MULTIPLY(tmp12, - FIX_0_390180644);    /*  c5-c3 */
+    tmp13 = MULTIPLY(tmp13, - FIX_1_961570560);    /* -c3-c5 */
+
+    tmp12 += z1;
+    tmp13 += z1;
+
+    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp0 + tmp10 + tmp12,
+					   CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp1 + tmp11 + tmp13,
+					   CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp2 + tmp11 + tmp12,
+					   CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp3 + tmp10 + tmp13,
+					   CONST_BITS+PASS1_BITS+1);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 14x7 sample block.
+ *
+ * 14-point FDCT in pass 1 (rows), 7-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_14x7 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
+  INT32 z1, z2, z3;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Zero bottom row of output coefficient block. */
+  MEMZERO(&data[DCTSIZE*7], SIZEOF(DCTELEM) * DCTSIZE);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* 14-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/28). */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 7; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[13]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[12]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[11]);
+    tmp13 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[10]);
+    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[9]);
+    tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[8]);
+    tmp6 = GETJSAMPLE(elemptr[6]) + GETJSAMPLE(elemptr[7]);
+
+    tmp10 = tmp0 + tmp6;
+    tmp14 = tmp0 - tmp6;
+    tmp11 = tmp1 + tmp5;
+    tmp15 = tmp1 - tmp5;
+    tmp12 = tmp2 + tmp4;
+    tmp16 = tmp2 - tmp4;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[13]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[12]);
+    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[11]);
+    tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[10]);
+    tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[9]);
+    tmp5 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[8]);
+    tmp6 = GETJSAMPLE(elemptr[6]) - GETJSAMPLE(elemptr[7]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp10 + tmp11 + tmp12 + tmp13 - 14 * CENTERJSAMPLE) << PASS1_BITS);
+    tmp13 += tmp13;
+    dataptr[4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp13, FIX(1.274162392)) + /* c4 */
+	      MULTIPLY(tmp11 - tmp13, FIX(0.314692123)) - /* c12 */
+	      MULTIPLY(tmp12 - tmp13, FIX(0.881747734)),  /* c8 */
+	      CONST_BITS-PASS1_BITS);
+
+    tmp10 = MULTIPLY(tmp14 + tmp15, FIX(1.105676686));    /* c6 */
+
+    dataptr[2] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp14, FIX(0.273079590))   /* c2-c6 */
+	      + MULTIPLY(tmp16, FIX(0.613604268)),        /* c10 */
+	      CONST_BITS-PASS1_BITS);
+    dataptr[6] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp15, FIX(1.719280954))   /* c6+c10 */
+	      - MULTIPLY(tmp16, FIX(1.378756276)),        /* c2 */
+	      CONST_BITS-PASS1_BITS);
+
+    /* Odd part */
+
+    tmp10 = tmp1 + tmp2;
+    tmp11 = tmp5 - tmp4;
+    dataptr[7] = (DCTELEM) ((tmp0 - tmp10 + tmp3 - tmp11 - tmp6) << PASS1_BITS);
+    tmp3 <<= CONST_BITS;
+    tmp10 = MULTIPLY(tmp10, - FIX(0.158341681));          /* -c13 */
+    tmp11 = MULTIPLY(tmp11, FIX(1.405321284));            /* c1 */
+    tmp10 += tmp11 - tmp3;
+    tmp11 = MULTIPLY(tmp0 + tmp2, FIX(1.197448846)) +     /* c5 */
+	    MULTIPLY(tmp4 + tmp6, FIX(0.752406978));      /* c9 */
+    dataptr[5] = (DCTELEM)
+      DESCALE(tmp10 + tmp11 - MULTIPLY(tmp2, FIX(2.373959773)) /* c3+c5-c13 */
+	      + MULTIPLY(tmp4, FIX(1.119999435)),         /* c1+c11-c9 */
+	      CONST_BITS-PASS1_BITS);
+    tmp12 = MULTIPLY(tmp0 + tmp1, FIX(1.334852607)) +     /* c3 */
+	    MULTIPLY(tmp5 - tmp6, FIX(0.467085129));      /* c11 */
+    dataptr[3] = (DCTELEM)
+      DESCALE(tmp10 + tmp12 - MULTIPLY(tmp1, FIX(0.424103948)) /* c3-c9-c13 */
+	      - MULTIPLY(tmp5, FIX(3.069855259)),         /* c1+c5+c11 */
+	      CONST_BITS-PASS1_BITS);
+    dataptr[1] = (DCTELEM)
+      DESCALE(tmp11 + tmp12 + tmp3 + tmp6 -
+	      MULTIPLY(tmp0 + tmp6, FIX(1.126980169)),    /* c3+c5-c1 */
+	      CONST_BITS-PASS1_BITS);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by (8/14)*(8/7) = 32/49, which we
+   * partially fold into the constant multipliers and final shifting:
+   * 7-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/14) * 64/49.
+   */
+
+  dataptr = data;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*6];
+    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*5];
+    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*4];
+    tmp3 = dataptr[DCTSIZE*3];
+
+    tmp10 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*6];
+    tmp11 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*5];
+    tmp12 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*4];
+
+    z1 = tmp0 + tmp2;
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(z1 + tmp1 + tmp3, FIX(1.306122449)), /* 64/49 */
+	      CONST_BITS+PASS1_BITS+1);
+    tmp3 += tmp3;
+    z1 -= tmp3;
+    z1 -= tmp3;
+    z1 = MULTIPLY(z1, FIX(0.461784020));                /* (c2+c6-c4)/2 */
+    z2 = MULTIPLY(tmp0 - tmp2, FIX(1.202428084));       /* (c2+c4-c6)/2 */
+    z3 = MULTIPLY(tmp1 - tmp2, FIX(0.411026446));       /* c6 */
+    dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + z2 + z3, CONST_BITS+PASS1_BITS+1);
+    z1 -= z2;
+    z2 = MULTIPLY(tmp0 - tmp1, FIX(1.151670509));       /* c4 */
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(z2 + z3 - MULTIPLY(tmp1 - tmp3, FIX(0.923568041)), /* c2+c6-c4 */
+	      CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 + z2, CONST_BITS+PASS1_BITS+1);
+
+    /* Odd part */
+
+    tmp1 = MULTIPLY(tmp10 + tmp11, FIX(1.221765677));   /* (c3+c1-c5)/2 */
+    tmp2 = MULTIPLY(tmp10 - tmp11, FIX(0.222383464));   /* (c3+c5-c1)/2 */
+    tmp0 = tmp1 - tmp2;
+    tmp1 += tmp2;
+    tmp2 = MULTIPLY(tmp11 + tmp12, - FIX(1.800824523)); /* -c1 */
+    tmp1 += tmp2;
+    tmp3 = MULTIPLY(tmp10 + tmp12, FIX(0.801442310));   /* c5 */
+    tmp0 += tmp3;
+    tmp2 += tmp3 + MULTIPLY(tmp12, FIX(2.443531355));   /* c3+c1-c5 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp0, CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp1, CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp2, CONST_BITS+PASS1_BITS+1);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 12x6 sample block.
+ *
+ * 12-point FDCT in pass 1 (rows), 6-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_12x6 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Zero 2 bottom rows of output coefficient block. */
+  MEMZERO(&data[DCTSIZE*6], SIZEOF(DCTELEM) * DCTSIZE * 2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* 12-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/24). */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 6; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[11]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[10]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[9]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[8]);
+    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[7]);
+    tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[6]);
+
+    tmp10 = tmp0 + tmp5;
+    tmp13 = tmp0 - tmp5;
+    tmp11 = tmp1 + tmp4;
+    tmp14 = tmp1 - tmp4;
+    tmp12 = tmp2 + tmp3;
+    tmp15 = tmp2 - tmp3;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[11]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[10]);
+    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[9]);
+    tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[8]);
+    tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[7]);
+    tmp5 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[6]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp10 + tmp11 + tmp12 - 12 * CENTERJSAMPLE) << PASS1_BITS);
+    dataptr[6] = (DCTELEM) ((tmp13 - tmp14 - tmp15) << PASS1_BITS);
+    dataptr[4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.224744871)), /* c4 */
+	      CONST_BITS-PASS1_BITS);
+    dataptr[2] = (DCTELEM)
+      DESCALE(tmp14 - tmp15 + MULTIPLY(tmp13 + tmp15, FIX(1.366025404)), /* c2 */
+	      CONST_BITS-PASS1_BITS);
+
+    /* Odd part */
+
+    tmp10 = MULTIPLY(tmp1 + tmp4, FIX_0_541196100);    /* c9 */
+    tmp14 = tmp10 + MULTIPLY(tmp1, FIX_0_765366865);   /* c3-c9 */
+    tmp15 = tmp10 - MULTIPLY(tmp4, FIX_1_847759065);   /* c3+c9 */
+    tmp12 = MULTIPLY(tmp0 + tmp2, FIX(1.121971054));   /* c5 */
+    tmp13 = MULTIPLY(tmp0 + tmp3, FIX(0.860918669));   /* c7 */
+    tmp10 = tmp12 + tmp13 + tmp14 - MULTIPLY(tmp0, FIX(0.580774953)) /* c5+c7-c1 */
+	    + MULTIPLY(tmp5, FIX(0.184591911));        /* c11 */
+    tmp11 = MULTIPLY(tmp2 + tmp3, - FIX(0.184591911)); /* -c11 */
+    tmp12 += tmp11 - tmp15 - MULTIPLY(tmp2, FIX(2.339493912)) /* c1+c5-c11 */
+	    + MULTIPLY(tmp5, FIX(0.860918669));        /* c7 */
+    tmp13 += tmp11 - tmp14 + MULTIPLY(tmp3, FIX(0.725788011)) /* c1+c11-c7 */
+	    - MULTIPLY(tmp5, FIX(1.121971054));        /* c5 */
+    tmp11 = tmp15 + MULTIPLY(tmp0 - tmp3, FIX(1.306562965)) /* c3 */
+	    - MULTIPLY(tmp2 + tmp5, FIX_0_541196100);  /* c9 */
+
+    dataptr[1] = (DCTELEM) DESCALE(tmp10, CONST_BITS-PASS1_BITS);
+    dataptr[3] = (DCTELEM) DESCALE(tmp11, CONST_BITS-PASS1_BITS);
+    dataptr[5] = (DCTELEM) DESCALE(tmp12, CONST_BITS-PASS1_BITS);
+    dataptr[7] = (DCTELEM) DESCALE(tmp13, CONST_BITS-PASS1_BITS);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by (8/12)*(8/6) = 8/9, which we
+   * partially fold into the constant multipliers and final shifting:
+   * 6-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/12) * 16/9.
+   */
+
+  dataptr = data;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*5];
+    tmp11 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*4];
+    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*3];
+
+    tmp10 = tmp0 + tmp2;
+    tmp12 = tmp0 - tmp2;
+
+    tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*5];
+    tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*4];
+    tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*3];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 + tmp11, FIX(1.777777778)),         /* 16/9 */
+	      CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp12, FIX(2.177324216)),                 /* c2 */
+	      CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(1.257078722)), /* c4 */
+	      CONST_BITS+PASS1_BITS+1);
+
+    /* Odd part */
+
+    tmp10 = MULTIPLY(tmp0 + tmp2, FIX(0.650711829));             /* c5 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp0 + tmp1, FIX(1.777777778)),   /* 16/9 */
+	      CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*3] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0 - tmp1 - tmp2, FIX(1.777777778)),    /* 16/9 */
+	      CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*5] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp2 - tmp1, FIX(1.777777778)),   /* 16/9 */
+	      CONST_BITS+PASS1_BITS+1);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 10x5 sample block.
+ *
+ * 10-point FDCT in pass 1 (rows), 5-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_10x5 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Zero 3 bottom rows of output coefficient block. */
+  MEMZERO(&data[DCTSIZE*5], SIZEOF(DCTELEM) * DCTSIZE * 3);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* 10-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/20). */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 5; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[9]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[8]);
+    tmp12 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[7]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[6]);
+    tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[5]);
+
+    tmp10 = tmp0 + tmp4;
+    tmp13 = tmp0 - tmp4;
+    tmp11 = tmp1 + tmp3;
+    tmp14 = tmp1 - tmp3;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[9]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[8]);
+    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[7]);
+    tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[6]);
+    tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[5]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp10 + tmp11 + tmp12 - 10 * CENTERJSAMPLE) << PASS1_BITS);
+    tmp12 += tmp12;
+    dataptr[4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.144122806)) - /* c4 */
+	      MULTIPLY(tmp11 - tmp12, FIX(0.437016024)),  /* c8 */
+	      CONST_BITS-PASS1_BITS);
+    tmp10 = MULTIPLY(tmp13 + tmp14, FIX(0.831253876));    /* c6 */
+    dataptr[2] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp13, FIX(0.513743148)),  /* c2-c6 */
+	      CONST_BITS-PASS1_BITS);
+    dataptr[6] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp14, FIX(2.176250899)),  /* c2+c6 */
+	      CONST_BITS-PASS1_BITS);
+
+    /* Odd part */
+
+    tmp10 = tmp0 + tmp4;
+    tmp11 = tmp1 - tmp3;
+    dataptr[5] = (DCTELEM) ((tmp10 - tmp11 - tmp2) << PASS1_BITS);
+    tmp2 <<= CONST_BITS;
+    dataptr[1] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0, FIX(1.396802247)) +          /* c1 */
+	      MULTIPLY(tmp1, FIX(1.260073511)) + tmp2 +   /* c3 */
+	      MULTIPLY(tmp3, FIX(0.642039522)) +          /* c7 */
+	      MULTIPLY(tmp4, FIX(0.221231742)),           /* c9 */
+	      CONST_BITS-PASS1_BITS);
+    tmp12 = MULTIPLY(tmp0 - tmp4, FIX(0.951056516)) -     /* (c3+c7)/2 */
+	    MULTIPLY(tmp1 + tmp3, FIX(0.587785252));      /* (c1-c9)/2 */
+    tmp13 = MULTIPLY(tmp10 + tmp11, FIX(0.309016994)) +   /* (c3-c7)/2 */
+	    (tmp11 << (CONST_BITS - 1)) - tmp2;
+    dataptr[3] = (DCTELEM) DESCALE(tmp12 + tmp13, CONST_BITS-PASS1_BITS);
+    dataptr[7] = (DCTELEM) DESCALE(tmp12 - tmp13, CONST_BITS-PASS1_BITS);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by (8/10)*(8/5) = 32/25, which we
+   * fold into the constant multipliers:
+   * 5-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/10) * 32/25.
+   */
+
+  dataptr = data;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*4];
+    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*3];
+    tmp2 = dataptr[DCTSIZE*2];
+
+    tmp10 = tmp0 + tmp1;
+    tmp11 = tmp0 - tmp1;
+
+    tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*4];
+    tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*3];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 + tmp2, FIX(1.28)),        /* 32/25 */
+	      CONST_BITS+PASS1_BITS);
+    tmp11 = MULTIPLY(tmp11, FIX(1.011928851));          /* (c2+c4)/2 */
+    tmp10 -= tmp2 << 2;
+    tmp10 = MULTIPLY(tmp10, FIX(0.452548340));          /* (c2-c4)/2 */
+    dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(tmp11 + tmp10, CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp11 - tmp10, CONST_BITS+PASS1_BITS);
+
+    /* Odd part */
+
+    tmp10 = MULTIPLY(tmp0 + tmp1, FIX(1.064004961));    /* c3 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp0, FIX(0.657591230)), /* c1-c3 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*3] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp1, FIX(2.785601151)), /* c1+c3 */
+	      CONST_BITS+PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on an 8x4 sample block.
+ *
+ * 8-point FDCT in pass 1 (rows), 4-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_8x4 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3;
+  INT32 tmp10, tmp11, tmp12, tmp13;
+  INT32 z1;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Zero 4 bottom rows of output coefficient block. */
+  MEMZERO(&data[DCTSIZE*4], SIZEOF(DCTELEM) * DCTSIZE * 4);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* We must also scale the output by 8/4 = 2, which we add here. */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 4; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part per LL&M figure 1 --- note that published figure is faulty;
+     * rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
+     */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[7]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[6]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[5]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[4]);
+
+    tmp10 = tmp0 + tmp3;
+    tmp12 = tmp0 - tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp13 = tmp1 - tmp2;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[7]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[6]);
+    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[5]);
+    tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[4]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp10 + tmp11 - 8 * CENTERJSAMPLE) << (PASS1_BITS+1));
+    dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << (PASS1_BITS+1));
+
+    z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
+    /* Add fudge factor here for final descale. */
+    z1 += ONE << (CONST_BITS-PASS1_BITS-2);
+    dataptr[2] = (DCTELEM) RIGHT_SHIFT(z1 + MULTIPLY(tmp12, FIX_0_765366865),
+				       CONST_BITS-PASS1_BITS-1);
+    dataptr[6] = (DCTELEM) RIGHT_SHIFT(z1 - MULTIPLY(tmp13, FIX_1_847759065),
+				       CONST_BITS-PASS1_BITS-1);
+
+    /* Odd part per figure 8 --- note paper omits factor of sqrt(2).
+     * 8-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
+     * i0..i3 in the paper are tmp0..tmp3 here.
+     */
+
+    tmp10 = tmp0 + tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp12 = tmp0 + tmp2;
+    tmp13 = tmp1 + tmp3;
+    z1 = MULTIPLY(tmp12 + tmp13, FIX_1_175875602); /*  c3 */
+    /* Add fudge factor here for final descale. */
+    z1 += ONE << (CONST_BITS-PASS1_BITS-2);
+
+    tmp0  = MULTIPLY(tmp0,    FIX_1_501321110);    /*  c1+c3-c5-c7 */
+    tmp1  = MULTIPLY(tmp1,    FIX_3_072711026);    /*  c1+c3+c5-c7 */
+    tmp2  = MULTIPLY(tmp2,    FIX_2_053119869);    /*  c1+c3-c5+c7 */
+    tmp3  = MULTIPLY(tmp3,    FIX_0_298631336);    /* -c1+c3+c5-c7 */
+    tmp10 = MULTIPLY(tmp10, - FIX_0_899976223);    /*  c7-c3 */
+    tmp11 = MULTIPLY(tmp11, - FIX_2_562915447);    /* -c1-c3 */
+    tmp12 = MULTIPLY(tmp12, - FIX_0_390180644);    /*  c5-c3 */
+    tmp13 = MULTIPLY(tmp13, - FIX_1_961570560);    /* -c3-c5 */
+
+    tmp12 += z1;
+    tmp13 += z1;
+
+    dataptr[1] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 + tmp10 + tmp12, CONST_BITS-PASS1_BITS-1);
+    dataptr[3] = (DCTELEM)
+      RIGHT_SHIFT(tmp1 + tmp11 + tmp13, CONST_BITS-PASS1_BITS-1);
+    dataptr[5] = (DCTELEM)
+      RIGHT_SHIFT(tmp2 + tmp11 + tmp12, CONST_BITS-PASS1_BITS-1);
+    dataptr[7] = (DCTELEM)
+      RIGHT_SHIFT(tmp3 + tmp10 + tmp13, CONST_BITS-PASS1_BITS-1);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * 4-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
+   */
+
+  dataptr = data;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*3] + (ONE << (PASS1_BITS-1));
+    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*2];
+
+    tmp10 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*3];
+    tmp11 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*2];
+
+    dataptr[DCTSIZE*0] = (DCTELEM) RIGHT_SHIFT(tmp0 + tmp1, PASS1_BITS);
+    dataptr[DCTSIZE*2] = (DCTELEM) RIGHT_SHIFT(tmp0 - tmp1, PASS1_BITS);
+
+    /* Odd part */
+
+    tmp0 = MULTIPLY(tmp10 + tmp11, FIX_0_541196100);   /* c6 */
+    /* Add fudge factor here for final descale. */
+    tmp0 += ONE << (CONST_BITS+PASS1_BITS-1);
+
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 + MULTIPLY(tmp10, FIX_0_765366865), /* c2-c6 */
+		  CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*3] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 - MULTIPLY(tmp11, FIX_1_847759065), /* c2+c6 */
+		  CONST_BITS+PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 6x3 sample block.
+ *
+ * 6-point FDCT in pass 1 (rows), 3-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_6x3 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2;
+  INT32 tmp10, tmp11, tmp12;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* We scale the results further by 2 as part of output adaption */
+  /* scaling for different DCT size. */
+  /* 6-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/12). */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 3; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[5]);
+    tmp11 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[4]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[3]);
+
+    tmp10 = tmp0 + tmp2;
+    tmp12 = tmp0 - tmp2;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[5]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[4]);
+    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[3]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp10 + tmp11 - 6 * CENTERJSAMPLE) << (PASS1_BITS+1));
+    dataptr[2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp12, FIX(1.224744871)),                 /* c2 */
+	      CONST_BITS-PASS1_BITS-1);
+    dataptr[4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(0.707106781)), /* c4 */
+	      CONST_BITS-PASS1_BITS-1);
+
+    /* Odd part */
+
+    tmp10 = DESCALE(MULTIPLY(tmp0 + tmp2, FIX(0.366025404)),     /* c5 */
+		    CONST_BITS-PASS1_BITS-1);
+
+    dataptr[1] = (DCTELEM) (tmp10 + ((tmp0 + tmp1) << (PASS1_BITS+1)));
+    dataptr[3] = (DCTELEM) ((tmp0 - tmp1 - tmp2) << (PASS1_BITS+1));
+    dataptr[5] = (DCTELEM) (tmp10 + ((tmp2 - tmp1) << (PASS1_BITS+1)));
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by (8/6)*(8/3) = 32/9, which we partially
+   * fold into the constant multipliers (other part was done in pass 1):
+   * 3-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/6) * 16/9.
+   */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 6; ctr++) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*2];
+    tmp1 = dataptr[DCTSIZE*1];
+
+    tmp2 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*2];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0 + tmp1, FIX(1.777777778)),        /* 16/9 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0 - tmp1 - tmp1, FIX(1.257078722)), /* c2 */
+	      CONST_BITS+PASS1_BITS);
+
+    /* Odd part */
+
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp2, FIX(2.177324216)),               /* c1 */
+	      CONST_BITS+PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 4x2 sample block.
+ *
+ * 4-point FDCT in pass 1 (rows), 2-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_4x2 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1;
+  INT32 tmp10, tmp11;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* We must also scale the output by (8/4)*(8/2) = 2**3, which we add here. */
+  /* 4-point FDCT kernel, */
+  /* cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point FDCT]. */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 2; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[3]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[2]);
+
+    tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[3]);
+    tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[2]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp0 + tmp1 - 4 * CENTERJSAMPLE) << (PASS1_BITS+3));
+    dataptr[2] = (DCTELEM) ((tmp0 - tmp1) << (PASS1_BITS+3));
+
+    /* Odd part */
+
+    tmp0 = MULTIPLY(tmp10 + tmp11, FIX_0_541196100);       /* c6 */
+    /* Add fudge factor here for final descale. */
+    tmp0 += ONE << (CONST_BITS-PASS1_BITS-4);
+
+    dataptr[1] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 + MULTIPLY(tmp10, FIX_0_765366865), /* c2-c6 */
+		  CONST_BITS-PASS1_BITS-3);
+    dataptr[3] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 - MULTIPLY(tmp11, FIX_1_847759065), /* c2+c6 */
+		  CONST_BITS-PASS1_BITS-3);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 4; ctr++) {
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp0 = dataptr[DCTSIZE*0] + (ONE << (PASS1_BITS-1));
+    tmp1 = dataptr[DCTSIZE*1];
+
+    dataptr[DCTSIZE*0] = (DCTELEM) RIGHT_SHIFT(tmp0 + tmp1, PASS1_BITS);
+
+    /* Odd part */
+
+    dataptr[DCTSIZE*1] = (DCTELEM) RIGHT_SHIFT(tmp0 - tmp1, PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 2x1 sample block.
+ *
+ * 2-point FDCT in pass 1 (rows), 1-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_2x1 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1;
+  JSAMPROW elemptr;
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  elemptr = sample_data[0] + start_col;
+
+  tmp0 = GETJSAMPLE(elemptr[0]);
+  tmp1 = GETJSAMPLE(elemptr[1]);
+
+  /* We leave the results scaled up by an overall factor of 8.
+   * We must also scale the output by (8/2)*(8/1) = 2**5.
+   */
+
+  /* Even part */
+  /* Apply unsigned->signed conversion */
+  data[0] = (DCTELEM) ((tmp0 + tmp1 - 2 * CENTERJSAMPLE) << 5);
+
+  /* Odd part */
+  data[1] = (DCTELEM) ((tmp0 - tmp1) << 5);
+}
+
+
+/*
+ * Perform the forward DCT on an 8x16 sample block.
+ *
+ * 8-point FDCT in pass 1 (rows), 16-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_8x16 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16, tmp17;
+  INT32 z1;
+  DCTELEM workspace[DCTSIZE2];
+  DCTELEM *dataptr;
+  DCTELEM *wsptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+
+  dataptr = data;
+  ctr = 0;
+  for (;;) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part per LL&M figure 1 --- note that published figure is faulty;
+     * rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
+     */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[7]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[6]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[5]);
+    tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[4]);
+
+    tmp10 = tmp0 + tmp3;
+    tmp12 = tmp0 - tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp13 = tmp1 - tmp2;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[7]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[6]);
+    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[5]);
+    tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[4]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM) ((tmp10 + tmp11 - 8 * CENTERJSAMPLE) << PASS1_BITS);
+    dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << PASS1_BITS);
+
+    z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
+    dataptr[2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, FIX_0_765366865),
+				   CONST_BITS-PASS1_BITS);
+    dataptr[6] = (DCTELEM) DESCALE(z1 - MULTIPLY(tmp13, FIX_1_847759065),
+				   CONST_BITS-PASS1_BITS);
+
+    /* Odd part per figure 8 --- note paper omits factor of sqrt(2).
+     * 8-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
+     * i0..i3 in the paper are tmp0..tmp3 here.
+     */
+
+    tmp10 = tmp0 + tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp12 = tmp0 + tmp2;
+    tmp13 = tmp1 + tmp3;
+    z1 = MULTIPLY(tmp12 + tmp13, FIX_1_175875602); /*  c3 */
+
+    tmp0  = MULTIPLY(tmp0,    FIX_1_501321110);    /*  c1+c3-c5-c7 */
+    tmp1  = MULTIPLY(tmp1,    FIX_3_072711026);    /*  c1+c3+c5-c7 */
+    tmp2  = MULTIPLY(tmp2,    FIX_2_053119869);    /*  c1+c3-c5+c7 */
+    tmp3  = MULTIPLY(tmp3,    FIX_0_298631336);    /* -c1+c3+c5-c7 */
+    tmp10 = MULTIPLY(tmp10, - FIX_0_899976223);    /*  c7-c3 */
+    tmp11 = MULTIPLY(tmp11, - FIX_2_562915447);    /* -c1-c3 */
+    tmp12 = MULTIPLY(tmp12, - FIX_0_390180644);    /*  c5-c3 */
+    tmp13 = MULTIPLY(tmp13, - FIX_1_961570560);    /* -c3-c5 */
+
+    tmp12 += z1;
+    tmp13 += z1;
+
+    dataptr[1] = (DCTELEM) DESCALE(tmp0 + tmp10 + tmp12, CONST_BITS-PASS1_BITS);
+    dataptr[3] = (DCTELEM) DESCALE(tmp1 + tmp11 + tmp13, CONST_BITS-PASS1_BITS);
+    dataptr[5] = (DCTELEM) DESCALE(tmp2 + tmp11 + tmp12, CONST_BITS-PASS1_BITS);
+    dataptr[7] = (DCTELEM) DESCALE(tmp3 + tmp10 + tmp13, CONST_BITS-PASS1_BITS);
+
+    ctr++;
+
+    if (ctr != DCTSIZE) {
+      if (ctr == DCTSIZE * 2)
+	break;			/* Done. */
+      dataptr += DCTSIZE;	/* advance pointer to next row */
+    } else
+      dataptr = workspace;	/* switch pointer to extended workspace */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by 8/16 = 1/2.
+   * 16-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/32).
+   */
+
+  dataptr = data;
+  wsptr = workspace;
+  for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*7];
+    tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*6];
+    tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*5];
+    tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*4];
+    tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*3];
+    tmp5 = dataptr[DCTSIZE*5] + wsptr[DCTSIZE*2];
+    tmp6 = dataptr[DCTSIZE*6] + wsptr[DCTSIZE*1];
+    tmp7 = dataptr[DCTSIZE*7] + wsptr[DCTSIZE*0];
+
+    tmp10 = tmp0 + tmp7;
+    tmp14 = tmp0 - tmp7;
+    tmp11 = tmp1 + tmp6;
+    tmp15 = tmp1 - tmp6;
+    tmp12 = tmp2 + tmp5;
+    tmp16 = tmp2 - tmp5;
+    tmp13 = tmp3 + tmp4;
+    tmp17 = tmp3 - tmp4;
+
+    tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*7];
+    tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*6];
+    tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*5];
+    tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*4];
+    tmp4 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*3];
+    tmp5 = dataptr[DCTSIZE*5] - wsptr[DCTSIZE*2];
+    tmp6 = dataptr[DCTSIZE*6] - wsptr[DCTSIZE*1];
+    tmp7 = dataptr[DCTSIZE*7] - wsptr[DCTSIZE*0];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(tmp10 + tmp11 + tmp12 + tmp13, PASS1_BITS+1);
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp13, FIX(1.306562965)) + /* c4[16] = c2[8] */
+	      MULTIPLY(tmp11 - tmp12, FIX_0_541196100),   /* c12[16] = c6[8] */
+	      CONST_BITS+PASS1_BITS+1);
+
+    tmp10 = MULTIPLY(tmp17 - tmp15, FIX(0.275899379)) +   /* c14[16] = c7[8] */
+	    MULTIPLY(tmp14 - tmp16, FIX(1.387039845));    /* c2[16] = c1[8] */
+
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp15, FIX(1.451774982))   /* c6+c14 */
+	      + MULTIPLY(tmp16, FIX(2.172734804)),        /* c2+c10 */
+	      CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*6] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp14, FIX(0.211164243))   /* c2-c6 */
+	      - MULTIPLY(tmp17, FIX(1.061594338)),        /* c10+c14 */
+	      CONST_BITS+PASS1_BITS+1);
+
+    /* Odd part */
+
+    tmp11 = MULTIPLY(tmp0 + tmp1, FIX(1.353318001)) +         /* c3 */
+	    MULTIPLY(tmp6 - tmp7, FIX(0.410524528));          /* c13 */
+    tmp12 = MULTIPLY(tmp0 + tmp2, FIX(1.247225013)) +         /* c5 */
+	    MULTIPLY(tmp5 + tmp7, FIX(0.666655658));          /* c11 */
+    tmp13 = MULTIPLY(tmp0 + tmp3, FIX(1.093201867)) +         /* c7 */
+	    MULTIPLY(tmp4 - tmp7, FIX(0.897167586));          /* c9 */
+    tmp14 = MULTIPLY(tmp1 + tmp2, FIX(0.138617169)) +         /* c15 */
+	    MULTIPLY(tmp6 - tmp5, FIX(1.407403738));          /* c1 */
+    tmp15 = MULTIPLY(tmp1 + tmp3, - FIX(0.666655658)) +       /* -c11 */
+	    MULTIPLY(tmp4 + tmp6, - FIX(1.247225013));        /* -c5 */
+    tmp16 = MULTIPLY(tmp2 + tmp3, - FIX(1.353318001)) +       /* -c3 */
+	    MULTIPLY(tmp5 - tmp4, FIX(0.410524528));          /* c13 */
+    tmp10 = tmp11 + tmp12 + tmp13 -
+	    MULTIPLY(tmp0, FIX(2.286341144)) +                /* c7+c5+c3-c1 */
+	    MULTIPLY(tmp7, FIX(0.779653625));                 /* c15+c13-c11+c9 */
+    tmp11 += tmp14 + tmp15 + MULTIPLY(tmp1, FIX(0.071888074)) /* c9-c3-c15+c11 */
+	     - MULTIPLY(tmp6, FIX(1.663905119));              /* c7+c13+c1-c5 */
+    tmp12 += tmp14 + tmp16 - MULTIPLY(tmp2, FIX(1.125726048)) /* c7+c5+c15-c3 */
+	     + MULTIPLY(tmp5, FIX(1.227391138));              /* c9-c11+c1-c13 */
+    tmp13 += tmp15 + tmp16 + MULTIPLY(tmp3, FIX(1.065388962)) /* c15+c3+c11-c7 */
+	     + MULTIPLY(tmp4, FIX(2.167985692));              /* c1+c13+c5-c9 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp10, CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp11, CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp12, CONST_BITS+PASS1_BITS+1);
+    dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp13, CONST_BITS+PASS1_BITS+1);
+
+    dataptr++;			/* advance pointer to next column */
+    wsptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 7x14 sample block.
+ *
+ * 7-point FDCT in pass 1 (rows), 14-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_7x14 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
+  INT32 z1, z2, z3;
+  DCTELEM workspace[8*6];
+  DCTELEM *dataptr;
+  DCTELEM *wsptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* 7-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/14). */
+
+  dataptr = data;
+  ctr = 0;
+  for (;;) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[6]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[5]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[4]);
+    tmp3 = GETJSAMPLE(elemptr[3]);
+
+    tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[6]);
+    tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[5]);
+    tmp12 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[4]);
+
+    z1 = tmp0 + tmp2;
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((z1 + tmp1 + tmp3 - 7 * CENTERJSAMPLE) << PASS1_BITS);
+    tmp3 += tmp3;
+    z1 -= tmp3;
+    z1 -= tmp3;
+    z1 = MULTIPLY(z1, FIX(0.353553391));                /* (c2+c6-c4)/2 */
+    z2 = MULTIPLY(tmp0 - tmp2, FIX(0.920609002));       /* (c2+c4-c6)/2 */
+    z3 = MULTIPLY(tmp1 - tmp2, FIX(0.314692123));       /* c6 */
+    dataptr[2] = (DCTELEM) DESCALE(z1 + z2 + z3, CONST_BITS-PASS1_BITS);
+    z1 -= z2;
+    z2 = MULTIPLY(tmp0 - tmp1, FIX(0.881747734));       /* c4 */
+    dataptr[4] = (DCTELEM)
+      DESCALE(z2 + z3 - MULTIPLY(tmp1 - tmp3, FIX(0.707106781)), /* c2+c6-c4 */
+	      CONST_BITS-PASS1_BITS);
+    dataptr[6] = (DCTELEM) DESCALE(z1 + z2, CONST_BITS-PASS1_BITS);
+
+    /* Odd part */
+
+    tmp1 = MULTIPLY(tmp10 + tmp11, FIX(0.935414347));   /* (c3+c1-c5)/2 */
+    tmp2 = MULTIPLY(tmp10 - tmp11, FIX(0.170262339));   /* (c3+c5-c1)/2 */
+    tmp0 = tmp1 - tmp2;
+    tmp1 += tmp2;
+    tmp2 = MULTIPLY(tmp11 + tmp12, - FIX(1.378756276)); /* -c1 */
+    tmp1 += tmp2;
+    tmp3 = MULTIPLY(tmp10 + tmp12, FIX(0.613604268));   /* c5 */
+    tmp0 += tmp3;
+    tmp2 += tmp3 + MULTIPLY(tmp12, FIX(1.870828693));   /* c3+c1-c5 */
+
+    dataptr[1] = (DCTELEM) DESCALE(tmp0, CONST_BITS-PASS1_BITS);
+    dataptr[3] = (DCTELEM) DESCALE(tmp1, CONST_BITS-PASS1_BITS);
+    dataptr[5] = (DCTELEM) DESCALE(tmp2, CONST_BITS-PASS1_BITS);
+
+    ctr++;
+
+    if (ctr != DCTSIZE) {
+      if (ctr == 14)
+	break;			/* Done. */
+      dataptr += DCTSIZE;	/* advance pointer to next row */
+    } else
+      dataptr = workspace;	/* switch pointer to extended workspace */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by (8/7)*(8/14) = 32/49, which we
+   * fold into the constant multipliers:
+   * 14-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/28) * 32/49.
+   */
+
+  dataptr = data;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 7; ctr++) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*5];
+    tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*4];
+    tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*3];
+    tmp13 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*2];
+    tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*1];
+    tmp5 = dataptr[DCTSIZE*5] + wsptr[DCTSIZE*0];
+    tmp6 = dataptr[DCTSIZE*6] + dataptr[DCTSIZE*7];
+
+    tmp10 = tmp0 + tmp6;
+    tmp14 = tmp0 - tmp6;
+    tmp11 = tmp1 + tmp5;
+    tmp15 = tmp1 - tmp5;
+    tmp12 = tmp2 + tmp4;
+    tmp16 = tmp2 - tmp4;
+
+    tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*5];
+    tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*4];
+    tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*3];
+    tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*2];
+    tmp4 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*1];
+    tmp5 = dataptr[DCTSIZE*5] - wsptr[DCTSIZE*0];
+    tmp6 = dataptr[DCTSIZE*6] - dataptr[DCTSIZE*7];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12 + tmp13,
+		       FIX(0.653061224)),                 /* 32/49 */
+	      CONST_BITS+PASS1_BITS);
+    tmp13 += tmp13;
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp13, FIX(0.832106052)) + /* c4 */
+	      MULTIPLY(tmp11 - tmp13, FIX(0.205513223)) - /* c12 */
+	      MULTIPLY(tmp12 - tmp13, FIX(0.575835255)),  /* c8 */
+	      CONST_BITS+PASS1_BITS);
+
+    tmp10 = MULTIPLY(tmp14 + tmp15, FIX(0.722074570));    /* c6 */
+
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp14, FIX(0.178337691))   /* c2-c6 */
+	      + MULTIPLY(tmp16, FIX(0.400721155)),        /* c10 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*6] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp15, FIX(1.122795725))   /* c6+c10 */
+	      - MULTIPLY(tmp16, FIX(0.900412262)),        /* c2 */
+	      CONST_BITS+PASS1_BITS);
+
+    /* Odd part */
+
+    tmp10 = tmp1 + tmp2;
+    tmp11 = tmp5 - tmp4;
+    dataptr[DCTSIZE*7] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0 - tmp10 + tmp3 - tmp11 - tmp6,
+		       FIX(0.653061224)),                 /* 32/49 */
+	      CONST_BITS+PASS1_BITS);
+    tmp3  = MULTIPLY(tmp3 , FIX(0.653061224));            /* 32/49 */
+    tmp10 = MULTIPLY(tmp10, - FIX(0.103406812));          /* -c13 */
+    tmp11 = MULTIPLY(tmp11, FIX(0.917760839));            /* c1 */
+    tmp10 += tmp11 - tmp3;
+    tmp11 = MULTIPLY(tmp0 + tmp2, FIX(0.782007410)) +     /* c5 */
+	    MULTIPLY(tmp4 + tmp6, FIX(0.491367823));      /* c9 */
+    dataptr[DCTSIZE*5] = (DCTELEM)
+      DESCALE(tmp10 + tmp11 - MULTIPLY(tmp2, FIX(1.550341076)) /* c3+c5-c13 */
+	      + MULTIPLY(tmp4, FIX(0.731428202)),         /* c1+c11-c9 */
+	      CONST_BITS+PASS1_BITS);
+    tmp12 = MULTIPLY(tmp0 + tmp1, FIX(0.871740478)) +     /* c3 */
+	    MULTIPLY(tmp5 - tmp6, FIX(0.305035186));      /* c11 */
+    dataptr[DCTSIZE*3] = (DCTELEM)
+      DESCALE(tmp10 + tmp12 - MULTIPLY(tmp1, FIX(0.276965844)) /* c3-c9-c13 */
+	      - MULTIPLY(tmp5, FIX(2.004803435)),         /* c1+c5+c11 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      DESCALE(tmp11 + tmp12 + tmp3
+	      - MULTIPLY(tmp0, FIX(0.735987049))          /* c3+c5-c1 */
+	      - MULTIPLY(tmp6, FIX(0.082925825)),         /* c9-c11-c13 */
+	      CONST_BITS+PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+    wsptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 6x12 sample block.
+ *
+ * 6-point FDCT in pass 1 (rows), 12-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_6x12 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
+  DCTELEM workspace[8*4];
+  DCTELEM *dataptr;
+  DCTELEM *wsptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* 6-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/12). */
+
+  dataptr = data;
+  ctr = 0;
+  for (;;) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[5]);
+    tmp11 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[4]);
+    tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[3]);
+
+    tmp10 = tmp0 + tmp2;
+    tmp12 = tmp0 - tmp2;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[5]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[4]);
+    tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[3]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp10 + tmp11 - 6 * CENTERJSAMPLE) << PASS1_BITS);
+    dataptr[2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp12, FIX(1.224744871)),                 /* c2 */
+	      CONST_BITS-PASS1_BITS);
+    dataptr[4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(0.707106781)), /* c4 */
+	      CONST_BITS-PASS1_BITS);
+
+    /* Odd part */
+
+    tmp10 = DESCALE(MULTIPLY(tmp0 + tmp2, FIX(0.366025404)),     /* c5 */
+		    CONST_BITS-PASS1_BITS);
+
+    dataptr[1] = (DCTELEM) (tmp10 + ((tmp0 + tmp1) << PASS1_BITS));
+    dataptr[3] = (DCTELEM) ((tmp0 - tmp1 - tmp2) << PASS1_BITS);
+    dataptr[5] = (DCTELEM) (tmp10 + ((tmp2 - tmp1) << PASS1_BITS));
+
+    ctr++;
+
+    if (ctr != DCTSIZE) {
+      if (ctr == 12)
+	break;			/* Done. */
+      dataptr += DCTSIZE;	/* advance pointer to next row */
+    } else
+      dataptr = workspace;	/* switch pointer to extended workspace */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by (8/6)*(8/12) = 8/9, which we
+   * fold into the constant multipliers:
+   * 12-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/24) * 8/9.
+   */
+
+  dataptr = data;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 6; ctr++) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*3];
+    tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*2];
+    tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*1];
+    tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*0];
+    tmp4 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*7];
+    tmp5 = dataptr[DCTSIZE*5] + dataptr[DCTSIZE*6];
+
+    tmp10 = tmp0 + tmp5;
+    tmp13 = tmp0 - tmp5;
+    tmp11 = tmp1 + tmp4;
+    tmp14 = tmp1 - tmp4;
+    tmp12 = tmp2 + tmp3;
+    tmp15 = tmp2 - tmp3;
+
+    tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*3];
+    tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*2];
+    tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*1];
+    tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*0];
+    tmp4 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*7];
+    tmp5 = dataptr[DCTSIZE*5] - dataptr[DCTSIZE*6];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12, FIX(0.888888889)), /* 8/9 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*6] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp13 - tmp14 - tmp15, FIX(0.888888889)), /* 8/9 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.088662108)),         /* c4 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp14 - tmp15, FIX(0.888888889)) +        /* 8/9 */
+	      MULTIPLY(tmp13 + tmp15, FIX(1.214244803)),         /* c2 */
+	      CONST_BITS+PASS1_BITS);
+
+    /* Odd part */
+
+    tmp10 = MULTIPLY(tmp1 + tmp4, FIX(0.481063200));   /* c9 */
+    tmp14 = tmp10 + MULTIPLY(tmp1, FIX(0.680326102));  /* c3-c9 */
+    tmp15 = tmp10 - MULTIPLY(tmp4, FIX(1.642452502));  /* c3+c9 */
+    tmp12 = MULTIPLY(tmp0 + tmp2, FIX(0.997307603));   /* c5 */
+    tmp13 = MULTIPLY(tmp0 + tmp3, FIX(0.765261039));   /* c7 */
+    tmp10 = tmp12 + tmp13 + tmp14 - MULTIPLY(tmp0, FIX(0.516244403)) /* c5+c7-c1 */
+	    + MULTIPLY(tmp5, FIX(0.164081699));        /* c11 */
+    tmp11 = MULTIPLY(tmp2 + tmp3, - FIX(0.164081699)); /* -c11 */
+    tmp12 += tmp11 - tmp15 - MULTIPLY(tmp2, FIX(2.079550144)) /* c1+c5-c11 */
+	    + MULTIPLY(tmp5, FIX(0.765261039));        /* c7 */
+    tmp13 += tmp11 - tmp14 + MULTIPLY(tmp3, FIX(0.645144899)) /* c1+c11-c7 */
+	    - MULTIPLY(tmp5, FIX(0.997307603));        /* c5 */
+    tmp11 = tmp15 + MULTIPLY(tmp0 - tmp3, FIX(1.161389302)) /* c3 */
+	    - MULTIPLY(tmp2 + tmp5, FIX(0.481063200)); /* c9 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp10, CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp11, CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp12, CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp13, CONST_BITS+PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+    wsptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 5x10 sample block.
+ *
+ * 5-point FDCT in pass 1 (rows), 10-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_5x10 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4;
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
+  DCTELEM workspace[8*2];
+  DCTELEM *dataptr;
+  DCTELEM *wsptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* 5-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/10). */
+
+  dataptr = data;
+  ctr = 0;
+  for (;;) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[4]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[3]);
+    tmp2 = GETJSAMPLE(elemptr[2]);
+
+    tmp10 = tmp0 + tmp1;
+    tmp11 = tmp0 - tmp1;
+
+    tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[4]);
+    tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[3]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp10 + tmp2 - 5 * CENTERJSAMPLE) << PASS1_BITS);
+    tmp11 = MULTIPLY(tmp11, FIX(0.790569415));          /* (c2+c4)/2 */
+    tmp10 -= tmp2 << 2;
+    tmp10 = MULTIPLY(tmp10, FIX(0.353553391));          /* (c2-c4)/2 */
+    dataptr[2] = (DCTELEM) DESCALE(tmp11 + tmp10, CONST_BITS-PASS1_BITS);
+    dataptr[4] = (DCTELEM) DESCALE(tmp11 - tmp10, CONST_BITS-PASS1_BITS);
+
+    /* Odd part */
+
+    tmp10 = MULTIPLY(tmp0 + tmp1, FIX(0.831253876));    /* c3 */
+
+    dataptr[1] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp0, FIX(0.513743148)), /* c1-c3 */
+	      CONST_BITS-PASS1_BITS);
+    dataptr[3] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp1, FIX(2.176250899)), /* c1+c3 */
+	      CONST_BITS-PASS1_BITS);
+
+    ctr++;
+
+    if (ctr != DCTSIZE) {
+      if (ctr == 10)
+	break;			/* Done. */
+      dataptr += DCTSIZE;	/* advance pointer to next row */
+    } else
+      dataptr = workspace;	/* switch pointer to extended workspace */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by (8/5)*(8/10) = 32/25, which we
+   * fold into the constant multipliers:
+   * 10-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/20) * 32/25.
+   */
+
+  dataptr = data;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 5; ctr++) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*1];
+    tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*0];
+    tmp12 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*7];
+    tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*6];
+    tmp4 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*5];
+
+    tmp10 = tmp0 + tmp4;
+    tmp13 = tmp0 - tmp4;
+    tmp11 = tmp1 + tmp3;
+    tmp14 = tmp1 - tmp3;
+
+    tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*1];
+    tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*0];
+    tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*7];
+    tmp3 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*6];
+    tmp4 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*5];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12, FIX(1.28)), /* 32/25 */
+	      CONST_BITS+PASS1_BITS);
+    tmp12 += tmp12;
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.464477191)) - /* c4 */
+	      MULTIPLY(tmp11 - tmp12, FIX(0.559380511)),  /* c8 */
+	      CONST_BITS+PASS1_BITS);
+    tmp10 = MULTIPLY(tmp13 + tmp14, FIX(1.064004961));    /* c6 */
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp13, FIX(0.657591230)),  /* c2-c6 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*6] = (DCTELEM)
+      DESCALE(tmp10 - MULTIPLY(tmp14, FIX(2.785601151)),  /* c2+c6 */
+	      CONST_BITS+PASS1_BITS);
+
+    /* Odd part */
+
+    tmp10 = tmp0 + tmp4;
+    tmp11 = tmp1 - tmp3;
+    dataptr[DCTSIZE*5] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp11 - tmp2, FIX(1.28)),  /* 32/25 */
+	      CONST_BITS+PASS1_BITS);
+    tmp2 = MULTIPLY(tmp2, FIX(1.28));                     /* 32/25 */
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0, FIX(1.787906876)) +          /* c1 */
+	      MULTIPLY(tmp1, FIX(1.612894094)) + tmp2 +   /* c3 */
+	      MULTIPLY(tmp3, FIX(0.821810588)) +          /* c7 */
+	      MULTIPLY(tmp4, FIX(0.283176630)),           /* c9 */
+	      CONST_BITS+PASS1_BITS);
+    tmp12 = MULTIPLY(tmp0 - tmp4, FIX(1.217352341)) -     /* (c3+c7)/2 */
+	    MULTIPLY(tmp1 + tmp3, FIX(0.752365123));      /* (c1-c9)/2 */
+    tmp13 = MULTIPLY(tmp10 + tmp11, FIX(0.395541753)) +   /* (c3-c7)/2 */
+	    MULTIPLY(tmp11, FIX(0.64)) - tmp2;            /* 16/25 */
+    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp12 + tmp13, CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp12 - tmp13, CONST_BITS+PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+    wsptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 4x8 sample block.
+ *
+ * 4-point FDCT in pass 1 (rows), 8-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_4x8 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3;
+  INT32 tmp10, tmp11, tmp12, tmp13;
+  INT32 z1;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* We must also scale the output by 8/4 = 2, which we add here. */
+  /* 4-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/16). */
+
+  dataptr = data;
+  for (ctr = 0; ctr < DCTSIZE; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[3]);
+    tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[2]);
+
+    tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[3]);
+    tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[2]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp0 + tmp1 - 4 * CENTERJSAMPLE) << (PASS1_BITS+1));
+    dataptr[2] = (DCTELEM) ((tmp0 - tmp1) << (PASS1_BITS+1));
+
+    /* Odd part */
+
+    tmp0 = MULTIPLY(tmp10 + tmp11, FIX_0_541196100);       /* c6 */
+    /* Add fudge factor here for final descale. */
+    tmp0 += ONE << (CONST_BITS-PASS1_BITS-2);
+
+    dataptr[1] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 + MULTIPLY(tmp10, FIX_0_765366865), /* c2-c6 */
+		  CONST_BITS-PASS1_BITS-1);
+    dataptr[3] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 - MULTIPLY(tmp11, FIX_1_847759065), /* c2+c6 */
+		  CONST_BITS-PASS1_BITS-1);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 4; ctr++) {
+    /* Even part per LL&M figure 1 --- note that published figure is faulty;
+     * rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
+     */
+
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
+    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
+    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
+    tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
+
+    /* Add fudge factor here for final descale. */
+    tmp10 = tmp0 + tmp3 + (ONE << (PASS1_BITS-1));
+    tmp12 = tmp0 - tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp13 = tmp1 - tmp2;
+
+    tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
+    tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
+    tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
+    tmp3 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
+
+    dataptr[DCTSIZE*0] = (DCTELEM) RIGHT_SHIFT(tmp10 + tmp11, PASS1_BITS);
+    dataptr[DCTSIZE*4] = (DCTELEM) RIGHT_SHIFT(tmp10 - tmp11, PASS1_BITS);
+
+    z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
+    /* Add fudge factor here for final descale. */
+    z1 += ONE << (CONST_BITS+PASS1_BITS-1);
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      RIGHT_SHIFT(z1 + MULTIPLY(tmp12, FIX_0_765366865), CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*6] = (DCTELEM)
+      RIGHT_SHIFT(z1 - MULTIPLY(tmp13, FIX_1_847759065), CONST_BITS+PASS1_BITS);
+
+    /* Odd part per figure 8 --- note paper omits factor of sqrt(2).
+     * 8-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
+     * i0..i3 in the paper are tmp0..tmp3 here.
+     */
+
+    tmp10 = tmp0 + tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp12 = tmp0 + tmp2;
+    tmp13 = tmp1 + tmp3;
+    z1 = MULTIPLY(tmp12 + tmp13, FIX_1_175875602); /*  c3 */
+    /* Add fudge factor here for final descale. */
+    z1 += ONE << (CONST_BITS+PASS1_BITS-1);
+
+    tmp0  = MULTIPLY(tmp0,    FIX_1_501321110);    /*  c1+c3-c5-c7 */
+    tmp1  = MULTIPLY(tmp1,    FIX_3_072711026);    /*  c1+c3+c5-c7 */
+    tmp2  = MULTIPLY(tmp2,    FIX_2_053119869);    /*  c1+c3-c5+c7 */
+    tmp3  = MULTIPLY(tmp3,    FIX_0_298631336);    /* -c1+c3+c5-c7 */
+    tmp10 = MULTIPLY(tmp10, - FIX_0_899976223);    /*  c7-c3 */
+    tmp11 = MULTIPLY(tmp11, - FIX_2_562915447);    /* -c1-c3 */
+    tmp12 = MULTIPLY(tmp12, - FIX_0_390180644);    /*  c5-c3 */
+    tmp13 = MULTIPLY(tmp13, - FIX_1_961570560);    /* -c3-c5 */
+
+    tmp12 += z1;
+    tmp13 += z1;
+
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 + tmp10 + tmp12, CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*3] = (DCTELEM)
+      RIGHT_SHIFT(tmp1 + tmp11 + tmp13, CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*5] = (DCTELEM)
+      RIGHT_SHIFT(tmp2 + tmp11 + tmp12, CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*7] = (DCTELEM)
+      RIGHT_SHIFT(tmp3 + tmp10 + tmp13, CONST_BITS+PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 3x6 sample block.
+ *
+ * 3-point FDCT in pass 1 (rows), 6-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_3x6 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1, tmp2;
+  INT32 tmp10, tmp11, tmp12;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+  /* We scale the results further by 2 as part of output adaption */
+  /* scaling for different DCT size. */
+  /* 3-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/6). */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 6; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[2]);
+    tmp1 = GETJSAMPLE(elemptr[1]);
+
+    tmp2 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[2]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM)
+      ((tmp0 + tmp1 - 3 * CENTERJSAMPLE) << (PASS1_BITS+1));
+    dataptr[2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0 - tmp1 - tmp1, FIX(0.707106781)), /* c2 */
+	      CONST_BITS-PASS1_BITS-1);
+
+    /* Odd part */
+
+    dataptr[1] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp2, FIX(1.224744871)),               /* c1 */
+	      CONST_BITS-PASS1_BITS-1);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We remove the PASS1_BITS scaling, but leave the results scaled up
+   * by an overall factor of 8.
+   * We must also scale the output by (8/6)*(8/3) = 32/9, which we partially
+   * fold into the constant multipliers (other part was done in pass 1):
+   * 6-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/12) * 16/9.
+   */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 3; ctr++) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*5];
+    tmp11 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*4];
+    tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*3];
+
+    tmp10 = tmp0 + tmp2;
+    tmp12 = tmp0 - tmp2;
+
+    tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*5];
+    tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*4];
+    tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*3];
+
+    dataptr[DCTSIZE*0] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 + tmp11, FIX(1.777777778)),         /* 16/9 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*2] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp12, FIX(2.177324216)),                 /* c2 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*4] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(1.257078722)), /* c4 */
+	      CONST_BITS+PASS1_BITS);
+
+    /* Odd part */
+
+    tmp10 = MULTIPLY(tmp0 + tmp2, FIX(0.650711829));             /* c5 */
+
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp0 + tmp1, FIX(1.777777778)),   /* 16/9 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*3] = (DCTELEM)
+      DESCALE(MULTIPLY(tmp0 - tmp1 - tmp2, FIX(1.777777778)),    /* 16/9 */
+	      CONST_BITS+PASS1_BITS);
+    dataptr[DCTSIZE*5] = (DCTELEM)
+      DESCALE(tmp10 + MULTIPLY(tmp2 - tmp1, FIX(1.777777778)),   /* 16/9 */
+	      CONST_BITS+PASS1_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 2x4 sample block.
+ *
+ * 2-point FDCT in pass 1 (rows), 4-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_2x4 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1;
+  INT32 tmp10, tmp11;
+  DCTELEM *dataptr;
+  JSAMPROW elemptr;
+  int ctr;
+  SHIFT_TEMPS
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true DCT. */
+  /* We must also scale the output by (8/2)*(8/4) = 2**3, which we add here. */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 4; ctr++) {
+    elemptr = sample_data[ctr] + start_col;
+
+    /* Even part */
+
+    tmp0 = GETJSAMPLE(elemptr[0]);
+    tmp1 = GETJSAMPLE(elemptr[1]);
+
+    /* Apply unsigned->signed conversion */
+    dataptr[0] = (DCTELEM) ((tmp0 + tmp1 - 2 * CENTERJSAMPLE) << 3);
+
+    /* Odd part */
+
+    dataptr[1] = (DCTELEM) ((tmp0 - tmp1) << 3);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns.
+   * We leave the results scaled up by an overall factor of 8.
+   * 4-point FDCT kernel,
+   * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point FDCT].
+   */
+
+  dataptr = data;
+  for (ctr = 0; ctr < 2; ctr++) {
+    /* Even part */
+
+    tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*3];
+    tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*2];
+
+    tmp10 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*3];
+    tmp11 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*2];
+
+    dataptr[DCTSIZE*0] = (DCTELEM) (tmp0 + tmp1);
+    dataptr[DCTSIZE*2] = (DCTELEM) (tmp0 - tmp1);
+
+    /* Odd part */
+
+    tmp0 = MULTIPLY(tmp10 + tmp11, FIX_0_541196100);       /* c6 */
+    /* Add fudge factor here for final descale. */
+    tmp0 += ONE << (CONST_BITS-1);
+
+    dataptr[DCTSIZE*1] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 + MULTIPLY(tmp10, FIX_0_765366865), /* c2-c6 */
+		  CONST_BITS);
+    dataptr[DCTSIZE*3] = (DCTELEM)
+      RIGHT_SHIFT(tmp0 - MULTIPLY(tmp11, FIX_1_847759065), /* c2+c6 */
+		  CONST_BITS);
+
+    dataptr++;			/* advance pointer to next column */
+  }
+}
+
+
+/*
+ * Perform the forward DCT on a 1x2 sample block.
+ *
+ * 1-point FDCT in pass 1 (rows), 2-point in pass 2 (columns).
+ */
+
+GLOBAL(void)
+jpeg_fdct_1x2 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col)
+{
+  INT32 tmp0, tmp1;
+
+  /* Pre-zero output coefficient block. */
+  MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2);
+
+  tmp0 = GETJSAMPLE(sample_data[0][start_col]);
+  tmp1 = GETJSAMPLE(sample_data[1][start_col]);
+
+  /* We leave the results scaled up by an overall factor of 8.
+   * We must also scale the output by (8/1)*(8/2) = 2**5.
+   */
+
+  /* Even part */
+  /* Apply unsigned->signed conversion */
+  data[DCTSIZE*0] = (DCTELEM) ((tmp0 + tmp1 - 2 * CENTERJSAMPLE) << 5);
+
+  /* Odd part */
+  data[DCTSIZE*1] = (DCTELEM) ((tmp0 - tmp1) << 5);
+}
+
+#endif /* DCT_SCALING_SUPPORTED */
+#endif /* DCT_ISLOW_SUPPORTED */
diff --git a/src/jpeg-6b/jidctflt.c b/src/jpeg-8c/jidctflt.c
index 0188ce3d..23ae9d33 100644
--- a/src/jpeg-6b/jidctflt.c
+++ b/src/jpeg-8c/jidctflt.c
@@ -2,6 +2,7 @@
  * jidctflt.c
  *
  * Copyright (C) 1994-1998, Thomas G. Lane.
+ * Modified 2010 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -76,10 +77,9 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
   FLOAT_MULT_TYPE * quantptr;
   FAST_FLOAT * wsptr;
   JSAMPROW outptr;
-  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  JSAMPLE *range_limit = cinfo->sample_range_limit;
   int ctr;
   FAST_FLOAT workspace[DCTSIZE2]; /* buffers data between passes */
-  SHIFT_TEMPS
 
   /* Pass 1: process columns from input, store into work array. */
 
@@ -152,12 +152,12 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
     tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562); /* 2*c4 */
 
     z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
-    tmp10 = ((FAST_FLOAT) 1.082392200) * z12 - z5; /* 2*(c2-c6) */
-    tmp12 = ((FAST_FLOAT) -2.613125930) * z10 + z5; /* -2*(c2+c6) */
+    tmp10 = z5 - z12 * ((FAST_FLOAT) 1.082392200); /* 2*(c2-c6) */
+    tmp12 = z5 - z10 * ((FAST_FLOAT) 2.613125930); /* 2*(c2+c6) */
 
     tmp6 = tmp12 - tmp7;	/* phase 2 */
     tmp5 = tmp11 - tmp6;
-    tmp4 = tmp10 + tmp5;
+    tmp4 = tmp10 - tmp5;
 
     wsptr[DCTSIZE*0] = tmp0 + tmp7;
     wsptr[DCTSIZE*7] = tmp0 - tmp7;
@@ -165,8 +165,8 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
     wsptr[DCTSIZE*6] = tmp1 - tmp6;
     wsptr[DCTSIZE*2] = tmp2 + tmp5;
     wsptr[DCTSIZE*5] = tmp2 - tmp5;
-    wsptr[DCTSIZE*4] = tmp3 + tmp4;
-    wsptr[DCTSIZE*3] = tmp3 - tmp4;
+    wsptr[DCTSIZE*3] = tmp3 + tmp4;
+    wsptr[DCTSIZE*4] = tmp3 - tmp4;
 
     inptr++;			/* advance pointers to next column */
     quantptr++;
@@ -174,7 +174,6 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
   }
   
   /* Pass 2: process rows from work array, store into output array. */
-  /* Note that we must descale the results by a factor of 8 == 2**3. */
 
   wsptr = workspace;
   for (ctr = 0; ctr < DCTSIZE; ctr++) {
@@ -187,8 +186,10 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
     
     /* Even part */
 
-    tmp10 = wsptr[0] + wsptr[4];
-    tmp11 = wsptr[0] - wsptr[4];
+    /* Apply signed->unsigned and prepare float->int conversion */
+    z5 = wsptr[0] + ((FAST_FLOAT) CENTERJSAMPLE + (FAST_FLOAT) 0.5);
+    tmp10 = z5 + wsptr[4];
+    tmp11 = z5 - wsptr[4];
 
     tmp13 = wsptr[2] + wsptr[6];
     tmp12 = (wsptr[2] - wsptr[6]) * ((FAST_FLOAT) 1.414213562) - tmp13;
@@ -209,31 +210,23 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
     tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562);
 
     z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
-    tmp10 = ((FAST_FLOAT) 1.082392200) * z12 - z5; /* 2*(c2-c6) */
-    tmp12 = ((FAST_FLOAT) -2.613125930) * z10 + z5; /* -2*(c2+c6) */
+    tmp10 = z5 - z12 * ((FAST_FLOAT) 1.082392200); /* 2*(c2-c6) */
+    tmp12 = z5 - z10 * ((FAST_FLOAT) 2.613125930); /* 2*(c2+c6) */
 
     tmp6 = tmp12 - tmp7;
     tmp5 = tmp11 - tmp6;
-    tmp4 = tmp10 + tmp5;
-
-    /* Final output stage: scale down by a factor of 8 and range-limit */
-
-    outptr[0] = range_limit[(int) DESCALE((INT32) (tmp0 + tmp7), 3)
-			    & RANGE_MASK];
-    outptr[7] = range_limit[(int) DESCALE((INT32) (tmp0 - tmp7), 3)
-			    & RANGE_MASK];
-    outptr[1] = range_limit[(int) DESCALE((INT32) (tmp1 + tmp6), 3)
-			    & RANGE_MASK];
-    outptr[6] = range_limit[(int) DESCALE((INT32) (tmp1 - tmp6), 3)
-			    & RANGE_MASK];
-    outptr[2] = range_limit[(int) DESCALE((INT32) (tmp2 + tmp5), 3)
-			    & RANGE_MASK];
-    outptr[5] = range_limit[(int) DESCALE((INT32) (tmp2 - tmp5), 3)
-			    & RANGE_MASK];
-    outptr[4] = range_limit[(int) DESCALE((INT32) (tmp3 + tmp4), 3)
-			    & RANGE_MASK];
-    outptr[3] = range_limit[(int) DESCALE((INT32) (tmp3 - tmp4), 3)
-			    & RANGE_MASK];
+    tmp4 = tmp10 - tmp5;
+
+    /* Final output stage: float->int conversion and range-limit */
+
+    outptr[0] = range_limit[((int) (tmp0 + tmp7)) & RANGE_MASK];
+    outptr[7] = range_limit[((int) (tmp0 - tmp7)) & RANGE_MASK];
+    outptr[1] = range_limit[((int) (tmp1 + tmp6)) & RANGE_MASK];
+    outptr[6] = range_limit[((int) (tmp1 - tmp6)) & RANGE_MASK];
+    outptr[2] = range_limit[((int) (tmp2 + tmp5)) & RANGE_MASK];
+    outptr[5] = range_limit[((int) (tmp2 - tmp5)) & RANGE_MASK];
+    outptr[3] = range_limit[((int) (tmp3 + tmp4)) & RANGE_MASK];
+    outptr[4] = range_limit[((int) (tmp3 - tmp4)) & RANGE_MASK];
     
     wsptr += DCTSIZE;		/* advance pointer to next row */
   }
diff --git a/src/jpeg-6b/jidctfst.c b/src/jpeg-8c/jidctfst.c
index dba4216f..dba4216f 100644
--- a/src/jpeg-6b/jidctfst.c
+++ b/src/jpeg-8c/jidctfst.c
diff --git a/src/jpeg-8c/jidctint.c b/src/jpeg-8c/jidctint.c
new file mode 100644
index 00000000..dcdf7ce4
--- /dev/null
+++ b/src/jpeg-8c/jidctint.c
@@ -0,0 +1,5137 @@
+/*
+ * jidctint.c
+ *
+ * Copyright (C) 1991-1998, Thomas G. Lane.
+ * Modification developed 2002-2009 by Guido Vollbeding.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains a slow-but-accurate integer implementation of the
+ * inverse DCT (Discrete Cosine Transform).  In the IJG code, this routine
+ * must also perform dequantization of the input coefficients.
+ *
+ * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
+ * on each row (or vice versa, but it's more convenient to emit a row at
+ * a time).  Direct algorithms are also available, but they are much more
+ * complex and seem not to be any faster when reduced to code.
+ *
+ * This implementation is based on an algorithm described in
+ *   C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
+ *   Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
+ *   Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
+ * The primary algorithm described there uses 11 multiplies and 29 adds.
+ * We use their alternate method with 12 multiplies and 32 adds.
+ * The advantage of this method is that no data path contains more than one
+ * multiplication; this allows a very simple and accurate implementation in
+ * scaled fixed-point arithmetic, with a minimal number of shifts.
+ *
+ * We also provide IDCT routines with various output sample block sizes for
+ * direct resolution reduction or enlargement and for direct resolving the
+ * common 2x1 and 1x2 subsampling cases without additional resampling: NxN
+ * (N=1...16), 2NxN, and Nx2N (N=1...8) pixels for one 8x8 input DCT block.
+ *
+ * For N<8 we simply take the corresponding low-frequency coefficients of
+ * the 8x8 input DCT block and apply an NxN point IDCT on the sub-block
+ * to yield the downscaled outputs.
+ * This can be seen as direct low-pass downsampling from the DCT domain
+ * point of view rather than the usual spatial domain point of view,
+ * yielding significant computational savings and results at least
+ * as good as common bilinear (averaging) spatial downsampling.
+ *
+ * For N>8 we apply a partial NxN IDCT on the 8 input coefficients as
+ * lower frequencies and higher frequencies assumed to be zero.
+ * It turns out that the computational effort is similar to the 8x8 IDCT
+ * regarding the output size.
+ * Furthermore, the scaling and descaling is the same for all IDCT sizes.
+ *
+ * CAUTION: We rely on the FIX() macro except for the N=1,2,4,8 cases
+ * since there would be too many additional constants to pre-calculate.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jdct.h"		/* Private declarations for DCT subsystem */
+
+#ifdef DCT_ISLOW_SUPPORTED
+
+
+/*
+ * This module is specialized to the case DCTSIZE = 8.
+ */
+
+#if DCTSIZE != 8
+  Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
+#endif
+
+
+/*
+ * The poop on this scaling stuff is as follows:
+ *
+ * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
+ * larger than the true IDCT outputs.  The final outputs are therefore
+ * a factor of N larger than desired; since N=8 this can be cured by
+ * a simple right shift at the end of the algorithm.  The advantage of
+ * this arrangement is that we save two multiplications per 1-D IDCT,
+ * because the y0 and y4 inputs need not be divided by sqrt(N).
+ *
+ * We have to do addition and subtraction of the integer inputs, which
+ * is no problem, and multiplication by fractional constants, which is
+ * a problem to do in integer arithmetic.  We multiply all the constants
+ * by CONST_SCALE and convert them to integer constants (thus retaining
+ * CONST_BITS bits of precision in the constants).  After doing a
+ * multiplication we have to divide the product by CONST_SCALE, with proper
+ * rounding, to produce the correct output.  This division can be done
+ * cheaply as a right shift of CONST_BITS bits.  We postpone shifting
+ * as long as possible so that partial sums can be added together with
+ * full fractional precision.
+ *
+ * The outputs of the first pass are scaled up by PASS1_BITS bits so that
+ * they are represented to better-than-integral precision.  These outputs
+ * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
+ * with the recommended scaling.  (To scale up 12-bit sample data further, an
+ * intermediate INT32 array would be needed.)
+ *
+ * To avoid overflow of the 32-bit intermediate results in pass 2, we must
+ * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26.  Error analysis
+ * shows that the values given below are the most effective.
+ */
+
+#if BITS_IN_JSAMPLE == 8
+#define CONST_BITS  13
+#define PASS1_BITS  2
+#else
+#define CONST_BITS  13
+#define PASS1_BITS  1		/* lose a little precision to avoid overflow */
+#endif
+
+/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
+ * causing a lot of useless floating-point operations at run time.
+ * To get around this we use the following pre-calculated constants.
+ * If you change CONST_BITS you may want to add appropriate values.
+ * (With a reasonable C compiler, you can just rely on the FIX() macro...)
+ */
+
+#if CONST_BITS == 13
+#define FIX_0_298631336  ((INT32)  2446)	/* FIX(0.298631336) */
+#define FIX_0_390180644  ((INT32)  3196)	/* FIX(0.390180644) */
+#define FIX_0_541196100  ((INT32)  4433)	/* FIX(0.541196100) */
+#define FIX_0_765366865  ((INT32)  6270)	/* FIX(0.765366865) */
+#define FIX_0_899976223  ((INT32)  7373)	/* FIX(0.899976223) */
+#define FIX_1_175875602  ((INT32)  9633)	/* FIX(1.175875602) */
+#define FIX_1_501321110  ((INT32)  12299)	/* FIX(1.501321110) */
+#define FIX_1_847759065  ((INT32)  15137)	/* FIX(1.847759065) */
+#define FIX_1_961570560  ((INT32)  16069)	/* FIX(1.961570560) */
+#define FIX_2_053119869  ((INT32)  16819)	/* FIX(2.053119869) */
+#define FIX_2_562915447  ((INT32)  20995)	/* FIX(2.562915447) */
+#define FIX_3_072711026  ((INT32)  25172)	/* FIX(3.072711026) */
+#else
+#define FIX_0_298631336  FIX(0.298631336)
+#define FIX_0_390180644  FIX(0.390180644)
+#define FIX_0_541196100  FIX(0.541196100)
+#define FIX_0_765366865  FIX(0.765366865)
+#define FIX_0_899976223  FIX(0.899976223)
+#define FIX_1_175875602  FIX(1.175875602)
+#define FIX_1_501321110  FIX(1.501321110)
+#define FIX_1_847759065  FIX(1.847759065)
+#define FIX_1_961570560  FIX(1.961570560)
+#define FIX_2_053119869  FIX(2.053119869)
+#define FIX_2_562915447  FIX(2.562915447)
+#define FIX_3_072711026  FIX(3.072711026)
+#endif
+
+
+/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
+ * For 8-bit samples with the recommended scaling, all the variable
+ * and constant values involved are no more than 16 bits wide, so a
+ * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
+ * For 12-bit samples, a full 32-bit multiplication will be needed.
+ */
+
+#if BITS_IN_JSAMPLE == 8
+#define MULTIPLY(var,const)  MULTIPLY16C16(var,const)
+#else
+#define MULTIPLY(var,const)  ((var) * (const))
+#endif
+
+
+/* Dequantize a coefficient by multiplying it by the multiplier-table
+ * entry; produce an int result.  In this module, both inputs and result
+ * are 16 bits or less, so either int or short multiply will work.
+ */
+
+#define DEQUANTIZE(coef,quantval)  (((ISLOW_MULT_TYPE) (coef)) * (quantval))
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients.
+ */
+
+GLOBAL(void)
+jpeg_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		 JCOEFPTR coef_block,
+		 JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3;
+  INT32 tmp10, tmp11, tmp12, tmp13;
+  INT32 z1, z2, z3;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[DCTSIZE2];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+  /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = DCTSIZE; ctr > 0; ctr--) {
+    /* Due to quantization, we will usually find that many of the input
+     * coefficients are zero, especially the AC terms.  We can exploit this
+     * by short-circuiting the IDCT calculation for any column in which all
+     * the AC terms are zero.  In that case each output is equal to the
+     * DC coefficient (with scale factor as needed).
+     * With typical images and quantization tables, half or more of the
+     * column DCT calculations can be simplified this way.
+     */
+
+    if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
+	inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
+	inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
+	inptr[DCTSIZE*7] == 0) {
+      /* AC terms all zero */
+      int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
+
+      wsptr[DCTSIZE*0] = dcval;
+      wsptr[DCTSIZE*1] = dcval;
+      wsptr[DCTSIZE*2] = dcval;
+      wsptr[DCTSIZE*3] = dcval;
+      wsptr[DCTSIZE*4] = dcval;
+      wsptr[DCTSIZE*5] = dcval;
+      wsptr[DCTSIZE*6] = dcval;
+      wsptr[DCTSIZE*7] = dcval;
+
+      inptr++;			/* advance pointers to next column */
+      quantptr++;
+      wsptr++;
+      continue;
+    }
+
+    /* Even part: reverse the even part of the forward DCT. */
+    /* The rotator is sqrt(2)*c(-6). */
+    
+    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
+    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);
+    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);
+
+    z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z2 <<= CONST_BITS;
+    z3 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    z2 += ONE << (CONST_BITS-PASS1_BITS-1);
+
+    tmp0 = z2 + z3;
+    tmp1 = z2 - z3;
+
+    tmp10 = tmp0 + tmp2;
+    tmp13 = tmp0 - tmp2;
+    tmp11 = tmp1 + tmp3;
+    tmp12 = tmp1 - tmp3;
+
+    /* Odd part per figure 8; the matrix is unitary and hence its
+     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
+     */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+    tmp1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    tmp2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    tmp3 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    
+    z2 = tmp0 + tmp2;
+    z3 = tmp1 + tmp3;
+
+    z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* sqrt(2) * c3 */
+    z2 = MULTIPLY(z2, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
+    z3 = MULTIPLY(z3, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
+    z2 += z1;
+    z3 += z1;
+
+    z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
+    tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
+    tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
+    tmp0 += z1 + z2;
+    tmp3 += z1 + z3;
+
+    z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
+    tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
+    tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
+    tmp1 += z1 + z3;
+    tmp2 += z1 + z2;
+
+    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
+
+    wsptr[DCTSIZE*0] = (int) RIGHT_SHIFT(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
+    wsptr[DCTSIZE*7] = (int) RIGHT_SHIFT(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
+    wsptr[DCTSIZE*1] = (int) RIGHT_SHIFT(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
+    wsptr[DCTSIZE*6] = (int) RIGHT_SHIFT(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
+    wsptr[DCTSIZE*2] = (int) RIGHT_SHIFT(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
+    wsptr[DCTSIZE*5] = (int) RIGHT_SHIFT(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
+    wsptr[DCTSIZE*3] = (int) RIGHT_SHIFT(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[DCTSIZE*4] = (int) RIGHT_SHIFT(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
+    
+    inptr++;			/* advance pointers to next column */
+    quantptr++;
+    wsptr++;
+  }
+
+  /* Pass 2: process rows from work array, store into output array. */
+  /* Note that we must descale the results by a factor of 8 == 2**3, */
+  /* and also undo the PASS1_BITS scaling. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < DCTSIZE; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+    /* Rows of zeroes can be exploited in the same way as we did with columns.
+     * However, the column calculation has created many nonzero AC terms, so
+     * the simplification applies less often (typically 5% to 10% of the time).
+     * On machines with very fast multiplication, it's possible that the
+     * test takes more time than it's worth.  In that case this section
+     * may be commented out.
+     */
+
+#ifndef NO_ZERO_ROW_TEST
+    if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && wsptr[4] == 0 &&
+	wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
+      /* AC terms all zero */
+      JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
+				  & RANGE_MASK];
+
+      outptr[0] = dcval;
+      outptr[1] = dcval;
+      outptr[2] = dcval;
+      outptr[3] = dcval;
+      outptr[4] = dcval;
+      outptr[5] = dcval;
+      outptr[6] = dcval;
+      outptr[7] = dcval;
+
+      wsptr += DCTSIZE;		/* advance pointer to next row */
+      continue;
+    }
+#endif
+
+    /* Even part: reverse the even part of the forward DCT. */
+    /* The rotator is sqrt(2)*c(-6). */
+    
+    z2 = (INT32) wsptr[2];
+    z3 = (INT32) wsptr[6];
+
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
+    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);
+    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);
+
+    /* Add fudge factor here for final descale. */
+    z2 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    z3 = (INT32) wsptr[4];
+
+    tmp0 = (z2 + z3) << CONST_BITS;
+    tmp1 = (z2 - z3) << CONST_BITS;
+    
+    tmp10 = tmp0 + tmp2;
+    tmp13 = tmp0 - tmp2;
+    tmp11 = tmp1 + tmp3;
+    tmp12 = tmp1 - tmp3;
+
+    /* Odd part per figure 8; the matrix is unitary and hence its
+     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
+     */
+
+    tmp0 = (INT32) wsptr[7];
+    tmp1 = (INT32) wsptr[5];
+    tmp2 = (INT32) wsptr[3];
+    tmp3 = (INT32) wsptr[1];
+
+    z2 = tmp0 + tmp2;
+    z3 = tmp1 + tmp3;
+
+    z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* sqrt(2) * c3 */
+    z2 = MULTIPLY(z2, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
+    z3 = MULTIPLY(z3, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
+    z2 += z1;
+    z3 += z1;
+
+    z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
+    tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
+    tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
+    tmp0 += z1 + z2;
+    tmp3 += z1 + z3;
+
+    z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
+    tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
+    tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
+    tmp1 += z1 + z3;
+    tmp2 += z1 + z2;
+
+    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp3,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp3,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += DCTSIZE;		/* advance pointer to next row */
+  }
+}
+
+#ifdef IDCT_SCALING_SUPPORTED
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 7x7 output block.
+ *
+ * Optimized algorithm with 12 multiplications in the 1-D kernel.
+ * cK represents sqrt(2) * cos(K*pi/14).
+ */
+
+GLOBAL(void)
+jpeg_idct_7x7 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12, tmp13;
+  INT32 z1, z2, z3;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[7*7];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 7; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp13 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp13 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    tmp13 += ONE << (CONST_BITS-PASS1_BITS-1);
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734));     /* c4 */
+    tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123));     /* c6 */
+    tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
+    tmp0 = z1 + z3;
+    z2 -= tmp0;
+    tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
+    tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536));  /* c2-c4-c6 */
+    tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249));  /* c2+c4+c6 */
+    tmp13 += MULTIPLY(z2, FIX(1.414213562));         /* c0 */
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+
+    tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347));      /* (c3+c1-c5)/2 */
+    tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339));      /* (c3+c5-c1)/2 */
+    tmp0 = tmp1 - tmp2;
+    tmp1 += tmp2;
+    tmp2 = MULTIPLY(z2 + z3, - FIX(1.378756276));    /* -c1 */
+    tmp1 += tmp2;
+    z2 = MULTIPLY(z1 + z3, FIX(0.613604268));        /* c5 */
+    tmp0 += z2;
+    tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693));     /* c3+c1-c5 */
+
+    /* Final output stage */
+
+    wsptr[7*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[7*6] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[7*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
+    wsptr[7*5] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
+    wsptr[7*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
+    wsptr[7*4] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
+    wsptr[7*3] = (int) RIGHT_SHIFT(tmp13, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 7 rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 7; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp13 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp13 <<= CONST_BITS;
+
+    z1 = (INT32) wsptr[2];
+    z2 = (INT32) wsptr[4];
+    z3 = (INT32) wsptr[6];
+
+    tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734));     /* c4 */
+    tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123));     /* c6 */
+    tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
+    tmp0 = z1 + z3;
+    z2 -= tmp0;
+    tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
+    tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536));  /* c2-c4-c6 */
+    tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249));  /* c2+c4+c6 */
+    tmp13 += MULTIPLY(z2, FIX(1.414213562));         /* c0 */
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+
+    tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347));      /* (c3+c1-c5)/2 */
+    tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339));      /* (c3+c5-c1)/2 */
+    tmp0 = tmp1 - tmp2;
+    tmp1 += tmp2;
+    tmp2 = MULTIPLY(z2 + z3, - FIX(1.378756276));    /* -c1 */
+    tmp1 += tmp2;
+    z2 = MULTIPLY(z1 + z3, FIX(0.613604268));        /* c5 */
+    tmp0 += z2;
+    tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693));     /* c3+c1-c5 */
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 7;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a reduced-size 6x6 output block.
+ *
+ * Optimized algorithm with 3 multiplications in the 1-D kernel.
+ * cK represents sqrt(2) * cos(K*pi/12).
+ */
+
+GLOBAL(void)
+jpeg_idct_6x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
+  INT32 z1, z2, z3;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[6*6];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp0 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
+    tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    tmp10 = MULTIPLY(tmp2, FIX(0.707106781));   /* c4 */
+    tmp1 = tmp0 + tmp10;
+    tmp11 = RIGHT_SHIFT(tmp0 - tmp10 - tmp10, CONST_BITS-PASS1_BITS);
+    tmp10 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    tmp0 = MULTIPLY(tmp10, FIX(1.224744871));   /* c2 */
+    tmp10 = tmp1 + tmp0;
+    tmp12 = tmp1 - tmp0;
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
+    tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
+    tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
+    tmp1 = (z1 - z2 - z3) << PASS1_BITS;
+
+    /* Final output stage */
+
+    wsptr[6*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[6*5] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[6*1] = (int) (tmp11 + tmp1);
+    wsptr[6*4] = (int) (tmp11 - tmp1);
+    wsptr[6*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
+    wsptr[6*3] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 6 rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 6; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp0 <<= CONST_BITS;
+    tmp2 = (INT32) wsptr[4];
+    tmp10 = MULTIPLY(tmp2, FIX(0.707106781));   /* c4 */
+    tmp1 = tmp0 + tmp10;
+    tmp11 = tmp0 - tmp10 - tmp10;
+    tmp10 = (INT32) wsptr[2];
+    tmp0 = MULTIPLY(tmp10, FIX(1.224744871));   /* c2 */
+    tmp10 = tmp1 + tmp0;
+    tmp12 = tmp1 - tmp0;
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+    tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
+    tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
+    tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
+    tmp1 = (z1 - z2 - z3) << CONST_BITS;
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 6;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a reduced-size 5x5 output block.
+ *
+ * Optimized algorithm with 5 multiplications in the 1-D kernel.
+ * cK represents sqrt(2) * cos(K*pi/10).
+ */
+
+GLOBAL(void)
+jpeg_idct_5x5 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp1, tmp10, tmp11, tmp12;
+  INT32 z1, z2, z3;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[5*5];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 5; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp12 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp12 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    tmp12 += ONE << (CONST_BITS-PASS1_BITS-1);
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    tmp1 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */
+    z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */
+    z3 = tmp12 + z2;
+    tmp10 = z3 + z1;
+    tmp11 = z3 - z1;
+    tmp12 -= z2 << 2;
+
+    /* Odd part */
+
+    z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+
+    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));     /* c3 */
+    tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148));   /* c1-c3 */
+    tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899));   /* c1+c3 */
+
+    /* Final output stage */
+
+    wsptr[5*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[5*4] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[5*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
+    wsptr[5*3] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
+    wsptr[5*2] = (int) RIGHT_SHIFT(tmp12, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 5 rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 5; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp12 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp12 <<= CONST_BITS;
+    tmp0 = (INT32) wsptr[2];
+    tmp1 = (INT32) wsptr[4];
+    z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */
+    z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */
+    z3 = tmp12 + z2;
+    tmp10 = z3 + z1;
+    tmp11 = z3 - z1;
+    tmp12 -= z2 << 2;
+
+    /* Odd part */
+
+    z2 = (INT32) wsptr[1];
+    z3 = (INT32) wsptr[3];
+
+    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));     /* c3 */
+    tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148));   /* c1-c3 */
+    tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899));   /* c1+c3 */
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 5;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a reduced-size 4x4 output block.
+ *
+ * Optimized algorithm with 3 multiplications in the 1-D kernel.
+ * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
+ */
+
+GLOBAL(void)
+jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp2, tmp10, tmp12;
+  INT32 z1, z2, z3;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[4*4];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 4; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    
+    tmp10 = (tmp0 + tmp2) << PASS1_BITS;
+    tmp12 = (tmp0 - tmp2) << PASS1_BITS;
+
+    /* Odd part */
+    /* Same rotation as in the even part of the 8x8 LL&M IDCT */
+
+    z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);               /* c6 */
+    /* Add fudge factor here for final descale. */
+    z1 += ONE << (CONST_BITS-PASS1_BITS-1);
+    tmp0 = RIGHT_SHIFT(z1 + MULTIPLY(z2, FIX_0_765366865), /* c2-c6 */
+		       CONST_BITS-PASS1_BITS);
+    tmp2 = RIGHT_SHIFT(z1 - MULTIPLY(z3, FIX_1_847759065), /* c2+c6 */
+		       CONST_BITS-PASS1_BITS);
+
+    /* Final output stage */
+
+    wsptr[4*0] = (int) (tmp10 + tmp0);
+    wsptr[4*3] = (int) (tmp10 - tmp0);
+    wsptr[4*1] = (int) (tmp12 + tmp2);
+    wsptr[4*2] = (int) (tmp12 - tmp2);
+  }
+
+  /* Pass 2: process 4 rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 4; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp2 = (INT32) wsptr[2];
+
+    tmp10 = (tmp0 + tmp2) << CONST_BITS;
+    tmp12 = (tmp0 - tmp2) << CONST_BITS;
+
+    /* Odd part */
+    /* Same rotation as in the even part of the 8x8 LL&M IDCT */
+
+    z2 = (INT32) wsptr[1];
+    z3 = (INT32) wsptr[3];
+
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);   /* c6 */
+    tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
+    tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 4;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a reduced-size 3x3 output block.
+ *
+ * Optimized algorithm with 2 multiplications in the 1-D kernel.
+ * cK represents sqrt(2) * cos(K*pi/6).
+ */
+
+GLOBAL(void)
+jpeg_idct_3x3 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp2, tmp10, tmp12;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[3*3];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 3; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp0 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
+    tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
+    tmp10 = tmp0 + tmp12;
+    tmp2 = tmp0 - tmp12 - tmp12;
+
+    /* Odd part */
+
+    tmp12 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
+
+    /* Final output stage */
+
+    wsptr[3*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[3*2] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[3*1] = (int) RIGHT_SHIFT(tmp2, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 3 rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 3; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp0 <<= CONST_BITS;
+    tmp2 = (INT32) wsptr[2];
+    tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
+    tmp10 = tmp0 + tmp12;
+    tmp2 = tmp0 - tmp12 - tmp12;
+
+    /* Odd part */
+
+    tmp12 = (INT32) wsptr[1];
+    tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 3;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a reduced-size 2x2 output block.
+ *
+ * Multiplication-less algorithm.
+ */
+
+GLOBAL(void)
+jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+  ISLOW_MULT_TYPE * quantptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input. */
+
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+
+  /* Column 0 */
+  tmp4 = DEQUANTIZE(coef_block[DCTSIZE*0], quantptr[DCTSIZE*0]);
+  tmp5 = DEQUANTIZE(coef_block[DCTSIZE*1], quantptr[DCTSIZE*1]);
+  /* Add fudge factor here for final descale. */
+  tmp4 += ONE << 2;
+
+  tmp0 = tmp4 + tmp5;
+  tmp2 = tmp4 - tmp5;
+
+  /* Column 1 */
+  tmp4 = DEQUANTIZE(coef_block[DCTSIZE*0+1], quantptr[DCTSIZE*0+1]);
+  tmp5 = DEQUANTIZE(coef_block[DCTSIZE*1+1], quantptr[DCTSIZE*1+1]);
+
+  tmp1 = tmp4 + tmp5;
+  tmp3 = tmp4 - tmp5;
+
+  /* Pass 2: process 2 rows, store into output array. */
+
+  /* Row 0 */
+  outptr = output_buf[0] + output_col;
+
+  outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp0 + tmp1, 3) & RANGE_MASK];
+  outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp0 - tmp1, 3) & RANGE_MASK];
+
+  /* Row 1 */
+  outptr = output_buf[1] + output_col;
+
+  outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp2 + tmp3, 3) & RANGE_MASK];
+  outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp2 - tmp3, 3) & RANGE_MASK];
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a reduced-size 1x1 output block.
+ *
+ * We hardly need an inverse DCT routine for this: just take the
+ * average pixel value, which is one-eighth of the DC coefficient.
+ */
+
+GLOBAL(void)
+jpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  int dcval;
+  ISLOW_MULT_TYPE * quantptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  SHIFT_TEMPS
+
+  /* 1x1 is trivial: just take the DC coefficient divided by 8. */
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  dcval = DEQUANTIZE(coef_block[0], quantptr[0]);
+  dcval = (int) DESCALE((INT32) dcval, 3);
+
+  output_buf[0][output_col] = range_limit[dcval & RANGE_MASK];
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 9x9 output block.
+ *
+ * Optimized algorithm with 10 multiplications in the 1-D kernel.
+ * cK represents sqrt(2) * cos(K*pi/18).
+ */
+
+GLOBAL(void)
+jpeg_idct_9x9 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13, tmp14;
+  INT32 z1, z2, z3, z4;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[8*9];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp0 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    tmp3 = MULTIPLY(z3, FIX(0.707106781));      /* c6 */
+    tmp1 = tmp0 + tmp3;
+    tmp2 = tmp0 - tmp3 - tmp3;
+
+    tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */
+    tmp11 = tmp2 + tmp0;
+    tmp14 = tmp2 - tmp0 - tmp0;
+
+    tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */
+    tmp2 = MULTIPLY(z1, FIX(1.083350441));      /* c4 */
+    tmp3 = MULTIPLY(z2, FIX(0.245575608));      /* c8 */
+
+    tmp10 = tmp1 + tmp0 - tmp3;
+    tmp12 = tmp1 - tmp0 + tmp2;
+    tmp13 = tmp1 - tmp2 + tmp3;
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+
+    z2 = MULTIPLY(z2, - FIX(1.224744871));           /* -c3 */
+
+    tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955));      /* c5 */
+    tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525));      /* c7 */
+    tmp0 = tmp2 + tmp3 - z2;
+    tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481));      /* c1 */
+    tmp2 += z2 - tmp1;
+    tmp3 += z2 + tmp1;
+    tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */
+
+    /* Final output stage */
+
+    wsptr[8*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[8*8] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[8*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
+    wsptr[8*7] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
+    wsptr[8*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
+    wsptr[8*6] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
+    wsptr[8*3] = (int) RIGHT_SHIFT(tmp13 + tmp3, CONST_BITS-PASS1_BITS);
+    wsptr[8*5] = (int) RIGHT_SHIFT(tmp13 - tmp3, CONST_BITS-PASS1_BITS);
+    wsptr[8*4] = (int) RIGHT_SHIFT(tmp14, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 9 rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 9; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp0 <<= CONST_BITS;
+
+    z1 = (INT32) wsptr[2];
+    z2 = (INT32) wsptr[4];
+    z3 = (INT32) wsptr[6];
+
+    tmp3 = MULTIPLY(z3, FIX(0.707106781));      /* c6 */
+    tmp1 = tmp0 + tmp3;
+    tmp2 = tmp0 - tmp3 - tmp3;
+
+    tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */
+    tmp11 = tmp2 + tmp0;
+    tmp14 = tmp2 - tmp0 - tmp0;
+
+    tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */
+    tmp2 = MULTIPLY(z1, FIX(1.083350441));      /* c4 */
+    tmp3 = MULTIPLY(z2, FIX(0.245575608));      /* c8 */
+
+    tmp10 = tmp1 + tmp0 - tmp3;
+    tmp12 = tmp1 - tmp0 + tmp2;
+    tmp13 = tmp1 - tmp2 + tmp3;
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+    z4 = (INT32) wsptr[7];
+
+    z2 = MULTIPLY(z2, - FIX(1.224744871));           /* -c3 */
+
+    tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955));      /* c5 */
+    tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525));      /* c7 */
+    tmp0 = tmp2 + tmp3 - z2;
+    tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481));      /* c1 */
+    tmp2 += z2 - tmp1;
+    tmp3 += z2 + tmp1;
+    tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp3,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp3,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp14,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 8;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 10x10 output block.
+ *
+ * Optimized algorithm with 12 multiplications in the 1-D kernel.
+ * cK represents sqrt(2) * cos(K*pi/20).
+ */
+
+GLOBAL(void)
+jpeg_idct_10x10 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		 JCOEFPTR coef_block,
+		 JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24;
+  INT32 z1, z2, z3, z4, z5;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[8*10];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    z3 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    z3 += ONE << (CONST_BITS-PASS1_BITS-1);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z1 = MULTIPLY(z4, FIX(1.144122806));         /* c4 */
+    z2 = MULTIPLY(z4, FIX(0.437016024));         /* c8 */
+    tmp10 = z3 + z1;
+    tmp11 = z3 - z2;
+
+    tmp22 = RIGHT_SHIFT(z3 - ((z1 - z2) << 1),   /* c0 = (c4-c8)*2 */
+			CONST_BITS-PASS1_BITS);
+
+    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));    /* c6 */
+    tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
+    tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
+
+    tmp20 = tmp10 + tmp12;
+    tmp24 = tmp10 - tmp12;
+    tmp21 = tmp11 + tmp13;
+    tmp23 = tmp11 - tmp13;
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+
+    tmp11 = z2 + z4;
+    tmp13 = z2 - z4;
+
+    tmp12 = MULTIPLY(tmp13, FIX(0.309016994));        /* (c3-c7)/2 */
+    z5 = z3 << CONST_BITS;
+
+    z2 = MULTIPLY(tmp11, FIX(0.951056516));           /* (c3+c7)/2 */
+    z4 = z5 + tmp12;
+
+    tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
+    tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
+
+    z2 = MULTIPLY(tmp11, FIX(0.587785252));           /* (c1-c9)/2 */
+    z4 = z5 - tmp12 - (tmp13 << (CONST_BITS - 1));
+
+    tmp12 = (z1 - tmp13 - z3) << PASS1_BITS;
+
+    tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
+    tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
+
+    /* Final output stage */
+
+    wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*9] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*8] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*2] = (int) (tmp22 + tmp12);
+    wsptr[8*7] = (int) (tmp22 - tmp12);
+    wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[8*6] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[8*5] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 10 rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 10; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    z3 <<= CONST_BITS;
+    z4 = (INT32) wsptr[4];
+    z1 = MULTIPLY(z4, FIX(1.144122806));         /* c4 */
+    z2 = MULTIPLY(z4, FIX(0.437016024));         /* c8 */
+    tmp10 = z3 + z1;
+    tmp11 = z3 - z2;
+
+    tmp22 = z3 - ((z1 - z2) << 1);               /* c0 = (c4-c8)*2 */
+
+    z2 = (INT32) wsptr[2];
+    z3 = (INT32) wsptr[6];
+
+    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));    /* c6 */
+    tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
+    tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
+
+    tmp20 = tmp10 + tmp12;
+    tmp24 = tmp10 - tmp12;
+    tmp21 = tmp11 + tmp13;
+    tmp23 = tmp11 - tmp13;
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+    z3 <<= CONST_BITS;
+    z4 = (INT32) wsptr[7];
+
+    tmp11 = z2 + z4;
+    tmp13 = z2 - z4;
+
+    tmp12 = MULTIPLY(tmp13, FIX(0.309016994));        /* (c3-c7)/2 */
+
+    z2 = MULTIPLY(tmp11, FIX(0.951056516));           /* (c3+c7)/2 */
+    z4 = z3 + tmp12;
+
+    tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
+    tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
+
+    z2 = MULTIPLY(tmp11, FIX(0.587785252));           /* (c1-c9)/2 */
+    z4 = z3 - tmp12 - (tmp13 << (CONST_BITS - 1));
+
+    tmp12 = ((z1 - tmp13) << CONST_BITS) - z3;
+
+    tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
+    tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 8;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 11x11 output block.
+ *
+ * Optimized algorithm with 24 multiplications in the 1-D kernel.
+ * cK represents sqrt(2) * cos(K*pi/22).
+ */
+
+GLOBAL(void)
+jpeg_idct_11x11 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		 JCOEFPTR coef_block,
+		 JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
+  INT32 z1, z2, z3, z4;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[8*11];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp10 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp10 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    tmp10 += ONE << (CONST_BITS-PASS1_BITS-1);
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132));     /* c2+c4 */
+    tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045));     /* c2-c6 */
+    z4 = z1 + z3;
+    tmp24 = MULTIPLY(z4, - FIX(1.155664402));        /* -(c2-c10) */
+    z4 -= z2;
+    tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976));  /* c2 */
+    tmp21 = tmp20 + tmp23 + tmp25 -
+	    MULTIPLY(z2, FIX(1.821790775));          /* c2+c4+c10-c6 */
+    tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */
+    tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */
+    tmp24 += tmp25;
+    tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120));  /* c8+c10 */
+    tmp24 += MULTIPLY(z2, FIX(1.944413522)) -        /* c2+c8 */
+	     MULTIPLY(z1, FIX(1.390975730));         /* c4+c10 */
+    tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562));  /* c0 */
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+
+    tmp11 = z1 + z2;
+    tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */
+    tmp11 = MULTIPLY(tmp11, FIX(0.887983902));           /* c3-c9 */
+    tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295));         /* c5-c9 */
+    tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */
+    tmp10 = tmp11 + tmp12 + tmp13 -
+	    MULTIPLY(z1, FIX(0.923107866));              /* c7+c5+c3-c1-2*c9 */
+    z1    = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */
+    tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588));        /* c1+c7+3*c9-c3 */
+    tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623));        /* c3+c5-c7-c9 */
+    z1    = MULTIPLY(z2 + z4, - FIX(1.798248910));       /* -(c1+c9) */
+    tmp11 += z1;
+    tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632));        /* c1+c5+c9-c7 */
+    tmp14 += MULTIPLY(z2, - FIX(1.467221301)) +          /* -(c5+c9) */
+	     MULTIPLY(z3, FIX(1.001388905)) -            /* c1-c9 */
+	     MULTIPLY(z4, FIX(1.684843907));             /* c3+c9 */
+
+    /* Final output stage */
+
+    wsptr[8*0]  = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*10] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*1]  = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*9]  = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*2]  = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*8]  = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*3]  = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[8*7]  = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[8*4]  = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[8*6]  = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[8*5]  = (int) RIGHT_SHIFT(tmp25, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 11 rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 11; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp10 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp10 <<= CONST_BITS;
+
+    z1 = (INT32) wsptr[2];
+    z2 = (INT32) wsptr[4];
+    z3 = (INT32) wsptr[6];
+
+    tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132));     /* c2+c4 */
+    tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045));     /* c2-c6 */
+    z4 = z1 + z3;
+    tmp24 = MULTIPLY(z4, - FIX(1.155664402));        /* -(c2-c10) */
+    z4 -= z2;
+    tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976));  /* c2 */
+    tmp21 = tmp20 + tmp23 + tmp25 -
+	    MULTIPLY(z2, FIX(1.821790775));          /* c2+c4+c10-c6 */
+    tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */
+    tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */
+    tmp24 += tmp25;
+    tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120));  /* c8+c10 */
+    tmp24 += MULTIPLY(z2, FIX(1.944413522)) -        /* c2+c8 */
+	     MULTIPLY(z1, FIX(1.390975730));         /* c4+c10 */
+    tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562));  /* c0 */
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+    z4 = (INT32) wsptr[7];
+
+    tmp11 = z1 + z2;
+    tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */
+    tmp11 = MULTIPLY(tmp11, FIX(0.887983902));           /* c3-c9 */
+    tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295));         /* c5-c9 */
+    tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */
+    tmp10 = tmp11 + tmp12 + tmp13 -
+	    MULTIPLY(z1, FIX(0.923107866));              /* c7+c5+c3-c1-2*c9 */
+    z1    = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */
+    tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588));        /* c1+c7+3*c9-c3 */
+    tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623));        /* c3+c5-c7-c9 */
+    z1    = MULTIPLY(z2 + z4, - FIX(1.798248910));       /* -(c1+c9) */
+    tmp11 += z1;
+    tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632));        /* c1+c5+c9-c7 */
+    tmp14 += MULTIPLY(z2, - FIX(1.467221301)) +          /* -(c5+c9) */
+	     MULTIPLY(z3, FIX(1.001388905)) -            /* c1-c9 */
+	     MULTIPLY(z4, FIX(1.684843907));             /* c3+c9 */
+
+    /* Final output stage */
+
+    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+
+    wsptr += 8;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 12x12 output block.
+ *
+ * Optimized algorithm with 15 multiplications in the 1-D kernel.
+ * cK represents sqrt(2) * cos(K*pi/24).
+ */
+
+GLOBAL(void)
+jpeg_idct_12x12 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		 JCOEFPTR coef_block,
+		 JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
+  INT32 z1, z2, z3, z4;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[8*12];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    z3 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    z3 += ONE << (CONST_BITS-PASS1_BITS-1);
+
+    z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
+
+    tmp10 = z3 + z4;
+    tmp11 = z3 - z4;
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
+    z1 <<= CONST_BITS;
+    z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+    z2 <<= CONST_BITS;
+
+    tmp12 = z1 - z2;
+
+    tmp21 = z3 + tmp12;
+    tmp24 = z3 - tmp12;
+
+    tmp12 = z4 + z2;
+
+    tmp20 = tmp10 + tmp12;
+    tmp25 = tmp10 - tmp12;
+
+    tmp12 = z4 - z1 - z2;
+
+    tmp22 = tmp11 + tmp12;
+    tmp23 = tmp11 - tmp12;
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+
+    tmp11 = MULTIPLY(z2, FIX(1.306562965));                  /* c3 */
+    tmp14 = MULTIPLY(z2, - FIX_0_541196100);                 /* -c9 */
+
+    tmp10 = z1 + z3;
+    tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669));          /* c7 */
+    tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384));       /* c5-c7 */
+    tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716));  /* c1-c5 */
+    tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580));           /* -(c7+c11) */
+    tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
+    tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
+    tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) -        /* c7-c11 */
+	     MULTIPLY(z4, FIX(1.982889723));                 /* c5+c7 */
+
+    z1 -= z4;
+    z2 -= z3;
+    z3 = MULTIPLY(z1 + z2, FIX_0_541196100);                 /* c9 */
+    tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865);              /* c3-c9 */
+    tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065);              /* c3+c9 */
+
+    /* Final output stage */
+
+    wsptr[8*0]  = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*11] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*1]  = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*10] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*2]  = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*9]  = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*3]  = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[8*8]  = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[8*4]  = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[8*7]  = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[8*5]  = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
+    wsptr[8*6]  = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 12 rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 12; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    z3 <<= CONST_BITS;
+
+    z4 = (INT32) wsptr[4];
+    z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
+
+    tmp10 = z3 + z4;
+    tmp11 = z3 - z4;
+
+    z1 = (INT32) wsptr[2];
+    z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
+    z1 <<= CONST_BITS;
+    z2 = (INT32) wsptr[6];
+    z2 <<= CONST_BITS;
+
+    tmp12 = z1 - z2;
+
+    tmp21 = z3 + tmp12;
+    tmp24 = z3 - tmp12;
+
+    tmp12 = z4 + z2;
+
+    tmp20 = tmp10 + tmp12;
+    tmp25 = tmp10 - tmp12;
+
+    tmp12 = z4 - z1 - z2;
+
+    tmp22 = tmp11 + tmp12;
+    tmp23 = tmp11 - tmp12;
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+    z4 = (INT32) wsptr[7];
+
+    tmp11 = MULTIPLY(z2, FIX(1.306562965));                  /* c3 */
+    tmp14 = MULTIPLY(z2, - FIX_0_541196100);                 /* -c9 */
+
+    tmp10 = z1 + z3;
+    tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669));          /* c7 */
+    tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384));       /* c5-c7 */
+    tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716));  /* c1-c5 */
+    tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580));           /* -(c7+c11) */
+    tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
+    tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
+    tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) -        /* c7-c11 */
+	     MULTIPLY(z4, FIX(1.982889723));                 /* c5+c7 */
+
+    z1 -= z4;
+    z2 -= z3;
+    z3 = MULTIPLY(z1 + z2, FIX_0_541196100);                 /* c9 */
+    tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865);              /* c3-c9 */
+    tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065);              /* c3+c9 */
+
+    /* Final output stage */
+
+    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+
+    wsptr += 8;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 13x13 output block.
+ *
+ * Optimized algorithm with 29 multiplications in the 1-D kernel.
+ * cK represents sqrt(2) * cos(K*pi/26).
+ */
+
+GLOBAL(void)
+jpeg_idct_13x13 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		 JCOEFPTR coef_block,
+		 JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
+  INT32 z1, z2, z3, z4;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[8*13];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    z1 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    z1 += ONE << (CONST_BITS-PASS1_BITS-1);
+
+    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    tmp10 = z3 + z4;
+    tmp11 = z3 - z4;
+
+    tmp12 = MULTIPLY(tmp10, FIX(1.155388986));                /* (c4+c6)/2 */
+    tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1;           /* (c4-c6)/2 */
+
+    tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13;   /* c2 */
+    tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13;   /* c10 */
+
+    tmp12 = MULTIPLY(tmp10, FIX(0.316450131));                /* (c8-c12)/2 */
+    tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1;           /* (c8+c12)/2 */
+
+    tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13;   /* c6 */
+    tmp25 = MULTIPLY(z2, - FIX(1.252223920)) + tmp12 + tmp13; /* c4 */
+
+    tmp12 = MULTIPLY(tmp10, FIX(0.435816023));                /* (c2-c10)/2 */
+    tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1;           /* (c2+c10)/2 */
+
+    tmp23 = MULTIPLY(z2, - FIX(0.170464608)) - tmp12 - tmp13; /* c12 */
+    tmp24 = MULTIPLY(z2, - FIX(0.803364869)) + tmp12 - tmp13; /* c8 */
+
+    tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1;      /* c0 */
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+
+    tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651));     /* c3 */
+    tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945));     /* c5 */
+    tmp15 = z1 + z4;
+    tmp13 = MULTIPLY(tmp15, FIX(0.937797057));       /* c7 */
+    tmp10 = tmp11 + tmp12 + tmp13 -
+	    MULTIPLY(z1, FIX(2.020082300));          /* c7+c5+c3-c1 */
+    tmp14 = MULTIPLY(z2 + z3, - FIX(0.338443458));   /* -c11 */
+    tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */
+    tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */
+    tmp14 = MULTIPLY(z2 + z4, - FIX(1.163874945));   /* -c5 */
+    tmp11 += tmp14;
+    tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */
+    tmp14 = MULTIPLY(z3 + z4, - FIX(0.657217813));   /* -c9 */
+    tmp12 += tmp14;
+    tmp13 += tmp14;
+    tmp15 = MULTIPLY(tmp15, FIX(0.338443458));       /* c11 */
+    tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */
+	    MULTIPLY(z2, FIX(0.466105296));          /* c1-c7 */
+    z1    = MULTIPLY(z3 - z2, FIX(0.937797057));     /* c7 */
+    tmp14 += z1;
+    tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) -   /* c3-c7 */
+	     MULTIPLY(z4, FIX(1.742345811));         /* c1+c11 */
+
+    /* Final output stage */
+
+    wsptr[8*0]  = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*12] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*1]  = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*11] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*2]  = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*10] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*3]  = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[8*9]  = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[8*4]  = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[8*8]  = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[8*5]  = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
+    wsptr[8*7]  = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
+    wsptr[8*6]  = (int) RIGHT_SHIFT(tmp26, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 13 rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 13; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    z1 <<= CONST_BITS;
+
+    z2 = (INT32) wsptr[2];
+    z3 = (INT32) wsptr[4];
+    z4 = (INT32) wsptr[6];
+
+    tmp10 = z3 + z4;
+    tmp11 = z3 - z4;
+
+    tmp12 = MULTIPLY(tmp10, FIX(1.155388986));                /* (c4+c6)/2 */
+    tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1;           /* (c4-c6)/2 */
+
+    tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13;   /* c2 */
+    tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13;   /* c10 */
+
+    tmp12 = MULTIPLY(tmp10, FIX(0.316450131));                /* (c8-c12)/2 */
+    tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1;           /* (c8+c12)/2 */
+
+    tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13;   /* c6 */
+    tmp25 = MULTIPLY(z2, - FIX(1.252223920)) + tmp12 + tmp13; /* c4 */
+
+    tmp12 = MULTIPLY(tmp10, FIX(0.435816023));                /* (c2-c10)/2 */
+    tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1;           /* (c2+c10)/2 */
+
+    tmp23 = MULTIPLY(z2, - FIX(0.170464608)) - tmp12 - tmp13; /* c12 */
+    tmp24 = MULTIPLY(z2, - FIX(0.803364869)) + tmp12 - tmp13; /* c8 */
+
+    tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1;      /* c0 */
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+    z4 = (INT32) wsptr[7];
+
+    tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651));     /* c3 */
+    tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945));     /* c5 */
+    tmp15 = z1 + z4;
+    tmp13 = MULTIPLY(tmp15, FIX(0.937797057));       /* c7 */
+    tmp10 = tmp11 + tmp12 + tmp13 -
+	    MULTIPLY(z1, FIX(2.020082300));          /* c7+c5+c3-c1 */
+    tmp14 = MULTIPLY(z2 + z3, - FIX(0.338443458));   /* -c11 */
+    tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */
+    tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */
+    tmp14 = MULTIPLY(z2 + z4, - FIX(1.163874945));   /* -c5 */
+    tmp11 += tmp14;
+    tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */
+    tmp14 = MULTIPLY(z3 + z4, - FIX(0.657217813));   /* -c9 */
+    tmp12 += tmp14;
+    tmp13 += tmp14;
+    tmp15 = MULTIPLY(tmp15, FIX(0.338443458));       /* c11 */
+    tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */
+	    MULTIPLY(z2, FIX(0.466105296));          /* c1-c7 */
+    z1    = MULTIPLY(z3 - z2, FIX(0.937797057));     /* c7 */
+    tmp14 += z1;
+    tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) -   /* c3-c7 */
+	     MULTIPLY(z4, FIX(1.742345811));         /* c1+c11 */
+
+    /* Final output stage */
+
+    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp26,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+
+    wsptr += 8;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 14x14 output block.
+ *
+ * Optimized algorithm with 20 multiplications in the 1-D kernel.
+ * cK represents sqrt(2) * cos(K*pi/28).
+ */
+
+GLOBAL(void)
+jpeg_idct_14x14 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		 JCOEFPTR coef_block,
+		 JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
+  INT32 z1, z2, z3, z4;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[8*14];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    z1 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    z1 += ONE << (CONST_BITS-PASS1_BITS-1);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z2 = MULTIPLY(z4, FIX(1.274162392));         /* c4 */
+    z3 = MULTIPLY(z4, FIX(0.314692123));         /* c12 */
+    z4 = MULTIPLY(z4, FIX(0.881747734));         /* c8 */
+
+    tmp10 = z1 + z2;
+    tmp11 = z1 + z3;
+    tmp12 = z1 - z4;
+
+    tmp23 = RIGHT_SHIFT(z1 - ((z2 + z3 - z4) << 1), /* c0 = (c4+c12-c8)*2 */
+			CONST_BITS-PASS1_BITS);
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    z3 = MULTIPLY(z1 + z2, FIX(1.105676686));    /* c6 */
+
+    tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
+    tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
+    tmp15 = MULTIPLY(z1, FIX(0.613604268)) -     /* c10 */
+	    MULTIPLY(z2, FIX(1.378756276));      /* c2 */
+
+    tmp20 = tmp10 + tmp13;
+    tmp26 = tmp10 - tmp13;
+    tmp21 = tmp11 + tmp14;
+    tmp25 = tmp11 - tmp14;
+    tmp22 = tmp12 + tmp15;
+    tmp24 = tmp12 - tmp15;
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+    tmp13 = z4 << CONST_BITS;
+
+    tmp14 = z1 + z3;
+    tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607));           /* c3 */
+    tmp12 = MULTIPLY(tmp14, FIX(1.197448846));             /* c5 */
+    tmp10 = tmp11 + tmp12 + tmp13 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
+    tmp14 = MULTIPLY(tmp14, FIX(0.752406978));             /* c9 */
+    tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426));        /* c9+c11-c13 */
+    z1    -= z2;
+    tmp15 = MULTIPLY(z1, FIX(0.467085129)) - tmp13;        /* c11 */
+    tmp16 += tmp15;
+    z1    += z4;
+    z4    = MULTIPLY(z2 + z3, - FIX(0.158341681)) - tmp13; /* -c13 */
+    tmp11 += z4 - MULTIPLY(z2, FIX(0.424103948));          /* c3-c9-c13 */
+    tmp12 += z4 - MULTIPLY(z3, FIX(2.373959773));          /* c3+c5-c13 */
+    z4    = MULTIPLY(z3 - z2, FIX(1.405321284));           /* c1 */
+    tmp14 += z4 + tmp13 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
+    tmp15 += z4 + MULTIPLY(z2, FIX(0.674957567));          /* c1+c11-c5 */
+
+    tmp13 = (z1 - z3) << PASS1_BITS;
+
+    /* Final output stage */
+
+    wsptr[8*0]  = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*13] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*1]  = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*12] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*2]  = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*11] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*3]  = (int) (tmp23 + tmp13);
+    wsptr[8*10] = (int) (tmp23 - tmp13);
+    wsptr[8*4]  = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[8*9]  = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[8*5]  = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
+    wsptr[8*8]  = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
+    wsptr[8*6]  = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
+    wsptr[8*7]  = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 14 rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 14; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    z1 <<= CONST_BITS;
+    z4 = (INT32) wsptr[4];
+    z2 = MULTIPLY(z4, FIX(1.274162392));         /* c4 */
+    z3 = MULTIPLY(z4, FIX(0.314692123));         /* c12 */
+    z4 = MULTIPLY(z4, FIX(0.881747734));         /* c8 */
+
+    tmp10 = z1 + z2;
+    tmp11 = z1 + z3;
+    tmp12 = z1 - z4;
+
+    tmp23 = z1 - ((z2 + z3 - z4) << 1);          /* c0 = (c4+c12-c8)*2 */
+
+    z1 = (INT32) wsptr[2];
+    z2 = (INT32) wsptr[6];
+
+    z3 = MULTIPLY(z1 + z2, FIX(1.105676686));    /* c6 */
+
+    tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
+    tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
+    tmp15 = MULTIPLY(z1, FIX(0.613604268)) -     /* c10 */
+	    MULTIPLY(z2, FIX(1.378756276));      /* c2 */
+
+    tmp20 = tmp10 + tmp13;
+    tmp26 = tmp10 - tmp13;
+    tmp21 = tmp11 + tmp14;
+    tmp25 = tmp11 - tmp14;
+    tmp22 = tmp12 + tmp15;
+    tmp24 = tmp12 - tmp15;
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+    z4 = (INT32) wsptr[7];
+    z4 <<= CONST_BITS;
+
+    tmp14 = z1 + z3;
+    tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607));           /* c3 */
+    tmp12 = MULTIPLY(tmp14, FIX(1.197448846));             /* c5 */
+    tmp10 = tmp11 + tmp12 + z4 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
+    tmp14 = MULTIPLY(tmp14, FIX(0.752406978));             /* c9 */
+    tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426));        /* c9+c11-c13 */
+    z1    -= z2;
+    tmp15 = MULTIPLY(z1, FIX(0.467085129)) - z4;           /* c11 */
+    tmp16 += tmp15;
+    tmp13 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - z4;    /* -c13 */
+    tmp11 += tmp13 - MULTIPLY(z2, FIX(0.424103948));       /* c3-c9-c13 */
+    tmp12 += tmp13 - MULTIPLY(z3, FIX(2.373959773));       /* c3+c5-c13 */
+    tmp13 = MULTIPLY(z3 - z2, FIX(1.405321284));           /* c1 */
+    tmp14 += tmp13 + z4 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
+    tmp15 += tmp13 + MULTIPLY(z2, FIX(0.674957567));       /* c1+c11-c5 */
+
+    tmp13 = ((z1 - z3) << CONST_BITS) + z4;
+
+    /* Final output stage */
+
+    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+
+    wsptr += 8;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 15x15 output block.
+ *
+ * Optimized algorithm with 22 multiplications in the 1-D kernel.
+ * cK represents sqrt(2) * cos(K*pi/30).
+ */
+
+GLOBAL(void)
+jpeg_idct_15x15 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		 JCOEFPTR coef_block,
+		 JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
+  INT32 z1, z2, z3, z4;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[8*15];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    z1 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    z1 += ONE << (CONST_BITS-PASS1_BITS-1);
+
+    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */
+    tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */
+
+    tmp12 = z1 - tmp10;
+    tmp13 = z1 + tmp11;
+    z1 -= (tmp11 - tmp10) << 1;             /* c0 = (c6-c12)*2 */
+
+    z4 = z2 - z3;
+    z3 += z2;
+    tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */
+    tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */
+    z2 = MULTIPLY(z2, FIX(1.439773946));    /* c4+c14 */
+
+    tmp20 = tmp13 + tmp10 + tmp11;
+    tmp23 = tmp12 - tmp10 + tmp11 + z2;
+
+    tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */
+    tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */
+
+    tmp25 = tmp13 - tmp10 - tmp11;
+    tmp26 = tmp12 + tmp10 - tmp11 - z2;
+
+    tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */
+    tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */
+
+    tmp21 = tmp12 + tmp10 + tmp11;
+    tmp24 = tmp13 - tmp10 + tmp11;
+    tmp11 += tmp11;
+    tmp22 = z1 + tmp11;                     /* c10 = c6-c12 */
+    tmp27 = z1 - tmp11 - tmp11;             /* c0 = (c6-c12)*2 */
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    z3 = MULTIPLY(z4, FIX(1.224744871));                    /* c5 */
+    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+
+    tmp13 = z2 - z4;
+    tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876));         /* c9 */
+    tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148));         /* c3-c9 */
+    tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899));      /* c3+c9 */
+
+    tmp13 = MULTIPLY(z2, - FIX(0.831253876));               /* -c9 */
+    tmp15 = MULTIPLY(z2, - FIX(1.344997024));               /* -c3 */
+    z2 = z1 - z4;
+    tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353));            /* c1 */
+
+    tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */
+    tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */
+    tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3;            /* c5 */
+    z2 = MULTIPLY(z1 + z4, FIX(0.575212477));               /* c11 */
+    tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3;      /* c7-c11 */
+    tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3;      /* c11+c13 */
+
+    /* Final output stage */
+
+    wsptr[8*0]  = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*14] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*1]  = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*13] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*2]  = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*12] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*3]  = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[8*11] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[8*4]  = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[8*10] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[8*5]  = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
+    wsptr[8*9]  = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
+    wsptr[8*6]  = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
+    wsptr[8*8]  = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
+    wsptr[8*7]  = (int) RIGHT_SHIFT(tmp27, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 15 rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 15; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    z1 <<= CONST_BITS;
+
+    z2 = (INT32) wsptr[2];
+    z3 = (INT32) wsptr[4];
+    z4 = (INT32) wsptr[6];
+
+    tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */
+    tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */
+
+    tmp12 = z1 - tmp10;
+    tmp13 = z1 + tmp11;
+    z1 -= (tmp11 - tmp10) << 1;             /* c0 = (c6-c12)*2 */
+
+    z4 = z2 - z3;
+    z3 += z2;
+    tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */
+    tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */
+    z2 = MULTIPLY(z2, FIX(1.439773946));    /* c4+c14 */
+
+    tmp20 = tmp13 + tmp10 + tmp11;
+    tmp23 = tmp12 - tmp10 + tmp11 + z2;
+
+    tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */
+    tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */
+
+    tmp25 = tmp13 - tmp10 - tmp11;
+    tmp26 = tmp12 + tmp10 - tmp11 - z2;
+
+    tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */
+    tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */
+
+    tmp21 = tmp12 + tmp10 + tmp11;
+    tmp24 = tmp13 - tmp10 + tmp11;
+    tmp11 += tmp11;
+    tmp22 = z1 + tmp11;                     /* c10 = c6-c12 */
+    tmp27 = z1 - tmp11 - tmp11;             /* c0 = (c6-c12)*2 */
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z4 = (INT32) wsptr[5];
+    z3 = MULTIPLY(z4, FIX(1.224744871));                    /* c5 */
+    z4 = (INT32) wsptr[7];
+
+    tmp13 = z2 - z4;
+    tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876));         /* c9 */
+    tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148));         /* c3-c9 */
+    tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899));      /* c3+c9 */
+
+    tmp13 = MULTIPLY(z2, - FIX(0.831253876));               /* -c9 */
+    tmp15 = MULTIPLY(z2, - FIX(1.344997024));               /* -c3 */
+    z2 = z1 - z4;
+    tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353));            /* c1 */
+
+    tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */
+    tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */
+    tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3;            /* c5 */
+    z2 = MULTIPLY(z1 + z4, FIX(0.575212477));               /* c11 */
+    tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3;      /* c7-c11 */
+    tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3;      /* c11+c13 */
+
+    /* Final output stage */
+
+    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp27,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+
+    wsptr += 8;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 16x16 output block.
+ *
+ * Optimized algorithm with 28 multiplications in the 1-D kernel.
+ * cK represents sqrt(2) * cos(K*pi/32).
+ */
+
+GLOBAL(void)
+jpeg_idct_16x16 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		 JCOEFPTR coef_block,
+		 JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
+  INT32 z1, z2, z3, z4;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[8*16];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp0 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    tmp0 += 1 << (CONST_BITS-PASS1_BITS-1);
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    tmp1 = MULTIPLY(z1, FIX(1.306562965));      /* c4[16] = c2[8] */
+    tmp2 = MULTIPLY(z1, FIX_0_541196100);       /* c12[16] = c6[8] */
+
+    tmp10 = tmp0 + tmp1;
+    tmp11 = tmp0 - tmp1;
+    tmp12 = tmp0 + tmp2;
+    tmp13 = tmp0 - tmp2;
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+    z3 = z1 - z2;
+    z4 = MULTIPLY(z3, FIX(0.275899379));        /* c14[16] = c7[8] */
+    z3 = MULTIPLY(z3, FIX(1.387039845));        /* c2[16] = c1[8] */
+
+    tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447);  /* (c6+c2)[16] = (c3+c1)[8] */
+    tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223);  /* (c6-c14)[16] = (c3-c7)[8] */
+    tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
+    tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
+
+    tmp20 = tmp10 + tmp0;
+    tmp27 = tmp10 - tmp0;
+    tmp21 = tmp12 + tmp1;
+    tmp26 = tmp12 - tmp1;
+    tmp22 = tmp13 + tmp2;
+    tmp25 = tmp13 - tmp2;
+    tmp23 = tmp11 + tmp3;
+    tmp24 = tmp11 - tmp3;
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+
+    tmp11 = z1 + z3;
+
+    tmp1  = MULTIPLY(z1 + z2, FIX(1.353318001));   /* c3 */
+    tmp2  = MULTIPLY(tmp11,   FIX(1.247225013));   /* c5 */
+    tmp3  = MULTIPLY(z1 + z4, FIX(1.093201867));   /* c7 */
+    tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586));   /* c9 */
+    tmp11 = MULTIPLY(tmp11,   FIX(0.666655658));   /* c11 */
+    tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528));   /* c13 */
+    tmp0  = tmp1 + tmp2 + tmp3 -
+	    MULTIPLY(z1, FIX(2.286341144));        /* c7+c5+c3-c1 */
+    tmp13 = tmp10 + tmp11 + tmp12 -
+	    MULTIPLY(z1, FIX(1.835730603));        /* c9+c11+c13-c15 */
+    z1    = MULTIPLY(z2 + z3, FIX(0.138617169));   /* c15 */
+    tmp1  += z1 + MULTIPLY(z2, FIX(0.071888074));  /* c9+c11-c3-c15 */
+    tmp2  += z1 - MULTIPLY(z3, FIX(1.125726048));  /* c5+c7+c15-c3 */
+    z1    = MULTIPLY(z3 - z2, FIX(1.407403738));   /* c1 */
+    tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282));  /* c1+c11-c9-c13 */
+    tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411));  /* c1+c5+c13-c7 */
+    z2    += z4;
+    z1    = MULTIPLY(z2, - FIX(0.666655658));      /* -c11 */
+    tmp1  += z1;
+    tmp3  += z1 + MULTIPLY(z4, FIX(1.065388962));  /* c3+c11+c15-c7 */
+    z2    = MULTIPLY(z2, - FIX(1.247225013));      /* -c5 */
+    tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809));  /* c1+c5+c9-c13 */
+    tmp12 += z2;
+    z2    = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
+    tmp2  += z2;
+    tmp3  += z2;
+    z2    = MULTIPLY(z4 - z3, FIX(0.410524528));   /* c13 */
+    tmp10 += z2;
+    tmp11 += z2;
+
+    /* Final output stage */
+
+    wsptr[8*0]  = (int) RIGHT_SHIFT(tmp20 + tmp0,  CONST_BITS-PASS1_BITS);
+    wsptr[8*15] = (int) RIGHT_SHIFT(tmp20 - tmp0,  CONST_BITS-PASS1_BITS);
+    wsptr[8*1]  = (int) RIGHT_SHIFT(tmp21 + tmp1,  CONST_BITS-PASS1_BITS);
+    wsptr[8*14] = (int) RIGHT_SHIFT(tmp21 - tmp1,  CONST_BITS-PASS1_BITS);
+    wsptr[8*2]  = (int) RIGHT_SHIFT(tmp22 + tmp2,  CONST_BITS-PASS1_BITS);
+    wsptr[8*13] = (int) RIGHT_SHIFT(tmp22 - tmp2,  CONST_BITS-PASS1_BITS);
+    wsptr[8*3]  = (int) RIGHT_SHIFT(tmp23 + tmp3,  CONST_BITS-PASS1_BITS);
+    wsptr[8*12] = (int) RIGHT_SHIFT(tmp23 - tmp3,  CONST_BITS-PASS1_BITS);
+    wsptr[8*4]  = (int) RIGHT_SHIFT(tmp24 + tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*11] = (int) RIGHT_SHIFT(tmp24 - tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*5]  = (int) RIGHT_SHIFT(tmp25 + tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*10] = (int) RIGHT_SHIFT(tmp25 - tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*6]  = (int) RIGHT_SHIFT(tmp26 + tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*9]  = (int) RIGHT_SHIFT(tmp26 - tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*7]  = (int) RIGHT_SHIFT(tmp27 + tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[8*8]  = (int) RIGHT_SHIFT(tmp27 - tmp13, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 16 rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 16; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp0 <<= CONST_BITS;
+
+    z1 = (INT32) wsptr[4];
+    tmp1 = MULTIPLY(z1, FIX(1.306562965));      /* c4[16] = c2[8] */
+    tmp2 = MULTIPLY(z1, FIX_0_541196100);       /* c12[16] = c6[8] */
+
+    tmp10 = tmp0 + tmp1;
+    tmp11 = tmp0 - tmp1;
+    tmp12 = tmp0 + tmp2;
+    tmp13 = tmp0 - tmp2;
+
+    z1 = (INT32) wsptr[2];
+    z2 = (INT32) wsptr[6];
+    z3 = z1 - z2;
+    z4 = MULTIPLY(z3, FIX(0.275899379));        /* c14[16] = c7[8] */
+    z3 = MULTIPLY(z3, FIX(1.387039845));        /* c2[16] = c1[8] */
+
+    tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447);  /* (c6+c2)[16] = (c3+c1)[8] */
+    tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223);  /* (c6-c14)[16] = (c3-c7)[8] */
+    tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
+    tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
+
+    tmp20 = tmp10 + tmp0;
+    tmp27 = tmp10 - tmp0;
+    tmp21 = tmp12 + tmp1;
+    tmp26 = tmp12 - tmp1;
+    tmp22 = tmp13 + tmp2;
+    tmp25 = tmp13 - tmp2;
+    tmp23 = tmp11 + tmp3;
+    tmp24 = tmp11 - tmp3;
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+    z4 = (INT32) wsptr[7];
+
+    tmp11 = z1 + z3;
+
+    tmp1  = MULTIPLY(z1 + z2, FIX(1.353318001));   /* c3 */
+    tmp2  = MULTIPLY(tmp11,   FIX(1.247225013));   /* c5 */
+    tmp3  = MULTIPLY(z1 + z4, FIX(1.093201867));   /* c7 */
+    tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586));   /* c9 */
+    tmp11 = MULTIPLY(tmp11,   FIX(0.666655658));   /* c11 */
+    tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528));   /* c13 */
+    tmp0  = tmp1 + tmp2 + tmp3 -
+	    MULTIPLY(z1, FIX(2.286341144));        /* c7+c5+c3-c1 */
+    tmp13 = tmp10 + tmp11 + tmp12 -
+	    MULTIPLY(z1, FIX(1.835730603));        /* c9+c11+c13-c15 */
+    z1    = MULTIPLY(z2 + z3, FIX(0.138617169));   /* c15 */
+    tmp1  += z1 + MULTIPLY(z2, FIX(0.071888074));  /* c9+c11-c3-c15 */
+    tmp2  += z1 - MULTIPLY(z3, FIX(1.125726048));  /* c5+c7+c15-c3 */
+    z1    = MULTIPLY(z3 - z2, FIX(1.407403738));   /* c1 */
+    tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282));  /* c1+c11-c9-c13 */
+    tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411));  /* c1+c5+c13-c7 */
+    z2    += z4;
+    z1    = MULTIPLY(z2, - FIX(0.666655658));      /* -c11 */
+    tmp1  += z1;
+    tmp3  += z1 + MULTIPLY(z4, FIX(1.065388962));  /* c3+c11+c15-c7 */
+    z2    = MULTIPLY(z2, - FIX(1.247225013));      /* -c5 */
+    tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809));  /* c1+c5+c9-c13 */
+    tmp12 += z2;
+    z2    = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
+    tmp2  += z2;
+    tmp3  += z2;
+    z2    = MULTIPLY(z4 - z3, FIX(0.410524528));   /* c13 */
+    tmp10 += z2;
+    tmp11 += z2;
+
+    /* Final output stage */
+
+    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp0,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[15] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp0,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp1,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp1,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp2,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp2,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp3,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp3,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp27 + tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp27 - tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+
+    wsptr += 8;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 16x8 output block.
+ *
+ * 8-point IDCT in pass 1 (columns), 16-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_16x8 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		JCOEFPTR coef_block,
+		JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
+  INT32 z1, z2, z3, z4;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[8*8];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+  /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = DCTSIZE; ctr > 0; ctr--) {
+    /* Due to quantization, we will usually find that many of the input
+     * coefficients are zero, especially the AC terms.  We can exploit this
+     * by short-circuiting the IDCT calculation for any column in which all
+     * the AC terms are zero.  In that case each output is equal to the
+     * DC coefficient (with scale factor as needed).
+     * With typical images and quantization tables, half or more of the
+     * column DCT calculations can be simplified this way.
+     */
+    
+    if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
+	inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
+	inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
+	inptr[DCTSIZE*7] == 0) {
+      /* AC terms all zero */
+      int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
+      
+      wsptr[DCTSIZE*0] = dcval;
+      wsptr[DCTSIZE*1] = dcval;
+      wsptr[DCTSIZE*2] = dcval;
+      wsptr[DCTSIZE*3] = dcval;
+      wsptr[DCTSIZE*4] = dcval;
+      wsptr[DCTSIZE*5] = dcval;
+      wsptr[DCTSIZE*6] = dcval;
+      wsptr[DCTSIZE*7] = dcval;
+      
+      inptr++;			/* advance pointers to next column */
+      quantptr++;
+      wsptr++;
+      continue;
+    }
+    
+    /* Even part: reverse the even part of the forward DCT. */
+    /* The rotator is sqrt(2)*c(-6). */
+    
+    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+    
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
+    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);
+    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);
+    
+    z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z2 <<= CONST_BITS;
+    z3 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    z2 += ONE << (CONST_BITS-PASS1_BITS-1);
+
+    tmp0 = z2 + z3;
+    tmp1 = z2 - z3;
+    
+    tmp10 = tmp0 + tmp2;
+    tmp13 = tmp0 - tmp2;
+    tmp11 = tmp1 + tmp3;
+    tmp12 = tmp1 - tmp3;
+    
+    /* Odd part per figure 8; the matrix is unitary and hence its
+     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
+     */
+    
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+    tmp1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    tmp2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    tmp3 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    
+    z2 = tmp0 + tmp2;
+    z3 = tmp1 + tmp3;
+
+    z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* sqrt(2) * c3 */
+    z2 = MULTIPLY(z2, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
+    z3 = MULTIPLY(z3, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
+    z2 += z1;
+    z3 += z1;
+
+    z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
+    tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
+    tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
+    tmp0 += z1 + z2;
+    tmp3 += z1 + z3;
+
+    z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
+    tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
+    tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
+    tmp1 += z1 + z3;
+    tmp2 += z1 + z2;
+    
+    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
+    
+    wsptr[DCTSIZE*0] = (int) RIGHT_SHIFT(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
+    wsptr[DCTSIZE*7] = (int) RIGHT_SHIFT(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
+    wsptr[DCTSIZE*1] = (int) RIGHT_SHIFT(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
+    wsptr[DCTSIZE*6] = (int) RIGHT_SHIFT(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
+    wsptr[DCTSIZE*2] = (int) RIGHT_SHIFT(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
+    wsptr[DCTSIZE*5] = (int) RIGHT_SHIFT(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
+    wsptr[DCTSIZE*3] = (int) RIGHT_SHIFT(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[DCTSIZE*4] = (int) RIGHT_SHIFT(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
+    
+    inptr++;			/* advance pointers to next column */
+    quantptr++;
+    wsptr++;
+  }
+
+  /* Pass 2: process 8 rows from work array, store into output array.
+   * 16-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/32).
+   */
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp0 <<= CONST_BITS;
+
+    z1 = (INT32) wsptr[4];
+    tmp1 = MULTIPLY(z1, FIX(1.306562965));      /* c4[16] = c2[8] */
+    tmp2 = MULTIPLY(z1, FIX_0_541196100);       /* c12[16] = c6[8] */
+
+    tmp10 = tmp0 + tmp1;
+    tmp11 = tmp0 - tmp1;
+    tmp12 = tmp0 + tmp2;
+    tmp13 = tmp0 - tmp2;
+
+    z1 = (INT32) wsptr[2];
+    z2 = (INT32) wsptr[6];
+    z3 = z1 - z2;
+    z4 = MULTIPLY(z3, FIX(0.275899379));        /* c14[16] = c7[8] */
+    z3 = MULTIPLY(z3, FIX(1.387039845));        /* c2[16] = c1[8] */
+
+    tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447);  /* (c6+c2)[16] = (c3+c1)[8] */
+    tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223);  /* (c6-c14)[16] = (c3-c7)[8] */
+    tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
+    tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
+
+    tmp20 = tmp10 + tmp0;
+    tmp27 = tmp10 - tmp0;
+    tmp21 = tmp12 + tmp1;
+    tmp26 = tmp12 - tmp1;
+    tmp22 = tmp13 + tmp2;
+    tmp25 = tmp13 - tmp2;
+    tmp23 = tmp11 + tmp3;
+    tmp24 = tmp11 - tmp3;
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+    z4 = (INT32) wsptr[7];
+
+    tmp11 = z1 + z3;
+
+    tmp1  = MULTIPLY(z1 + z2, FIX(1.353318001));   /* c3 */
+    tmp2  = MULTIPLY(tmp11,   FIX(1.247225013));   /* c5 */
+    tmp3  = MULTIPLY(z1 + z4, FIX(1.093201867));   /* c7 */
+    tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586));   /* c9 */
+    tmp11 = MULTIPLY(tmp11,   FIX(0.666655658));   /* c11 */
+    tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528));   /* c13 */
+    tmp0  = tmp1 + tmp2 + tmp3 -
+	    MULTIPLY(z1, FIX(2.286341144));        /* c7+c5+c3-c1 */
+    tmp13 = tmp10 + tmp11 + tmp12 -
+	    MULTIPLY(z1, FIX(1.835730603));        /* c9+c11+c13-c15 */
+    z1    = MULTIPLY(z2 + z3, FIX(0.138617169));   /* c15 */
+    tmp1  += z1 + MULTIPLY(z2, FIX(0.071888074));  /* c9+c11-c3-c15 */
+    tmp2  += z1 - MULTIPLY(z3, FIX(1.125726048));  /* c5+c7+c15-c3 */
+    z1    = MULTIPLY(z3 - z2, FIX(1.407403738));   /* c1 */
+    tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282));  /* c1+c11-c9-c13 */
+    tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411));  /* c1+c5+c13-c7 */
+    z2    += z4;
+    z1    = MULTIPLY(z2, - FIX(0.666655658));      /* -c11 */
+    tmp1  += z1;
+    tmp3  += z1 + MULTIPLY(z4, FIX(1.065388962));  /* c3+c11+c15-c7 */
+    z2    = MULTIPLY(z2, - FIX(1.247225013));      /* -c5 */
+    tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809));  /* c1+c5+c9-c13 */
+    tmp12 += z2;
+    z2    = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
+    tmp2  += z2;
+    tmp3  += z2;
+    z2    = MULTIPLY(z4 - z3, FIX(0.410524528));   /* c13 */
+    tmp10 += z2;
+    tmp11 += z2;
+
+    /* Final output stage */
+
+    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp0,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[15] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp0,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp1,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp1,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp2,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp2,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp3,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp3,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp27 + tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp27 - tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+
+    wsptr += 8;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 14x7 output block.
+ *
+ * 7-point IDCT in pass 1 (columns), 14-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_14x7 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		JCOEFPTR coef_block,
+		JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
+  INT32 z1, z2, z3, z4;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[8*7];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array.
+   * 7-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/14).
+   */
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp23 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp23 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    tmp23 += ONE << (CONST_BITS-PASS1_BITS-1);
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    tmp20 = MULTIPLY(z2 - z3, FIX(0.881747734));       /* c4 */
+    tmp22 = MULTIPLY(z1 - z2, FIX(0.314692123));       /* c6 */
+    tmp21 = tmp20 + tmp22 + tmp23 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
+    tmp10 = z1 + z3;
+    z2 -= tmp10;
+    tmp10 = MULTIPLY(tmp10, FIX(1.274162392)) + tmp23; /* c2 */
+    tmp20 += tmp10 - MULTIPLY(z3, FIX(0.077722536));   /* c2-c4-c6 */
+    tmp22 += tmp10 - MULTIPLY(z1, FIX(2.470602249));   /* c2+c4+c6 */
+    tmp23 += MULTIPLY(z2, FIX(1.414213562));           /* c0 */
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+
+    tmp11 = MULTIPLY(z1 + z2, FIX(0.935414347));       /* (c3+c1-c5)/2 */
+    tmp12 = MULTIPLY(z1 - z2, FIX(0.170262339));       /* (c3+c5-c1)/2 */
+    tmp10 = tmp11 - tmp12;
+    tmp11 += tmp12;
+    tmp12 = MULTIPLY(z2 + z3, - FIX(1.378756276));     /* -c1 */
+    tmp11 += tmp12;
+    z2 = MULTIPLY(z1 + z3, FIX(0.613604268));          /* c5 */
+    tmp10 += z2;
+    tmp12 += z2 + MULTIPLY(z3, FIX(1.870828693));      /* c3+c1-c5 */
+
+    /* Final output stage */
+
+    wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*6] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*5] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*4] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*3] = (int) RIGHT_SHIFT(tmp23, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 7 rows from work array, store into output array.
+   * 14-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/28).
+   */
+  wsptr = workspace;
+  for (ctr = 0; ctr < 7; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    z1 <<= CONST_BITS;
+    z4 = (INT32) wsptr[4];
+    z2 = MULTIPLY(z4, FIX(1.274162392));         /* c4 */
+    z3 = MULTIPLY(z4, FIX(0.314692123));         /* c12 */
+    z4 = MULTIPLY(z4, FIX(0.881747734));         /* c8 */
+
+    tmp10 = z1 + z2;
+    tmp11 = z1 + z3;
+    tmp12 = z1 - z4;
+
+    tmp23 = z1 - ((z2 + z3 - z4) << 1);          /* c0 = (c4+c12-c8)*2 */
+
+    z1 = (INT32) wsptr[2];
+    z2 = (INT32) wsptr[6];
+
+    z3 = MULTIPLY(z1 + z2, FIX(1.105676686));    /* c6 */
+
+    tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
+    tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
+    tmp15 = MULTIPLY(z1, FIX(0.613604268)) -     /* c10 */
+	    MULTIPLY(z2, FIX(1.378756276));      /* c2 */
+
+    tmp20 = tmp10 + tmp13;
+    tmp26 = tmp10 - tmp13;
+    tmp21 = tmp11 + tmp14;
+    tmp25 = tmp11 - tmp14;
+    tmp22 = tmp12 + tmp15;
+    tmp24 = tmp12 - tmp15;
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+    z4 = (INT32) wsptr[7];
+    z4 <<= CONST_BITS;
+
+    tmp14 = z1 + z3;
+    tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607));           /* c3 */
+    tmp12 = MULTIPLY(tmp14, FIX(1.197448846));             /* c5 */
+    tmp10 = tmp11 + tmp12 + z4 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
+    tmp14 = MULTIPLY(tmp14, FIX(0.752406978));             /* c9 */
+    tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426));        /* c9+c11-c13 */
+    z1    -= z2;
+    tmp15 = MULTIPLY(z1, FIX(0.467085129)) - z4;           /* c11 */
+    tmp16 += tmp15;
+    tmp13 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - z4;    /* -c13 */
+    tmp11 += tmp13 - MULTIPLY(z2, FIX(0.424103948));       /* c3-c9-c13 */
+    tmp12 += tmp13 - MULTIPLY(z3, FIX(2.373959773));       /* c3+c5-c13 */
+    tmp13 = MULTIPLY(z3 - z2, FIX(1.405321284));           /* c1 */
+    tmp14 += tmp13 + z4 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
+    tmp15 += tmp13 + MULTIPLY(z2, FIX(0.674957567));       /* c1+c11-c5 */
+
+    tmp13 = ((z1 - z3) << CONST_BITS) + z4;
+
+    /* Final output stage */
+
+    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+
+    wsptr += 8;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 12x6 output block.
+ *
+ * 6-point IDCT in pass 1 (columns), 12-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_12x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		JCOEFPTR coef_block,
+		JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
+  INT32 z1, z2, z3, z4;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[8*6];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array.
+   * 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
+   */
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp10 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp10 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    tmp10 += ONE << (CONST_BITS-PASS1_BITS-1);
+    tmp12 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    tmp20 = MULTIPLY(tmp12, FIX(0.707106781));   /* c4 */
+    tmp11 = tmp10 + tmp20;
+    tmp21 = RIGHT_SHIFT(tmp10 - tmp20 - tmp20, CONST_BITS-PASS1_BITS);
+    tmp20 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    tmp10 = MULTIPLY(tmp20, FIX(1.224744871));   /* c2 */
+    tmp20 = tmp11 + tmp10;
+    tmp22 = tmp11 - tmp10;
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    tmp11 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
+    tmp10 = tmp11 + ((z1 + z2) << CONST_BITS);
+    tmp12 = tmp11 + ((z3 - z2) << CONST_BITS);
+    tmp11 = (z1 - z2 - z3) << PASS1_BITS;
+
+    /* Final output stage */
+
+    wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*5] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*1] = (int) (tmp21 + tmp11);
+    wsptr[8*4] = (int) (tmp21 - tmp11);
+    wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*3] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 6 rows from work array, store into output array.
+   * 12-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/24).
+   */
+  wsptr = workspace;
+  for (ctr = 0; ctr < 6; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    z3 <<= CONST_BITS;
+
+    z4 = (INT32) wsptr[4];
+    z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
+
+    tmp10 = z3 + z4;
+    tmp11 = z3 - z4;
+
+    z1 = (INT32) wsptr[2];
+    z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
+    z1 <<= CONST_BITS;
+    z2 = (INT32) wsptr[6];
+    z2 <<= CONST_BITS;
+
+    tmp12 = z1 - z2;
+
+    tmp21 = z3 + tmp12;
+    tmp24 = z3 - tmp12;
+
+    tmp12 = z4 + z2;
+
+    tmp20 = tmp10 + tmp12;
+    tmp25 = tmp10 - tmp12;
+
+    tmp12 = z4 - z1 - z2;
+
+    tmp22 = tmp11 + tmp12;
+    tmp23 = tmp11 - tmp12;
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+    z4 = (INT32) wsptr[7];
+
+    tmp11 = MULTIPLY(z2, FIX(1.306562965));                  /* c3 */
+    tmp14 = MULTIPLY(z2, - FIX_0_541196100);                 /* -c9 */
+
+    tmp10 = z1 + z3;
+    tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669));          /* c7 */
+    tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384));       /* c5-c7 */
+    tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716));  /* c1-c5 */
+    tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580));           /* -(c7+c11) */
+    tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
+    tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
+    tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) -        /* c7-c11 */
+	     MULTIPLY(z4, FIX(1.982889723));                 /* c5+c7 */
+
+    z1 -= z4;
+    z2 -= z3;
+    z3 = MULTIPLY(z1 + z2, FIX_0_541196100);                 /* c9 */
+    tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865);              /* c3-c9 */
+    tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065);              /* c3+c9 */
+
+    /* Final output stage */
+
+    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
+					       CONST_BITS+PASS1_BITS+3)
+			     & RANGE_MASK];
+
+    wsptr += 8;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 10x5 output block.
+ *
+ * 5-point IDCT in pass 1 (columns), 10-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_10x5 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		JCOEFPTR coef_block,
+		JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24;
+  INT32 z1, z2, z3, z4;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[8*5];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array.
+   * 5-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/10).
+   */
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp12 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp12 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    tmp12 += ONE << (CONST_BITS-PASS1_BITS-1);
+    tmp13 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    tmp14 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z1 = MULTIPLY(tmp13 + tmp14, FIX(0.790569415)); /* (c2+c4)/2 */
+    z2 = MULTIPLY(tmp13 - tmp14, FIX(0.353553391)); /* (c2-c4)/2 */
+    z3 = tmp12 + z2;
+    tmp10 = z3 + z1;
+    tmp11 = z3 - z1;
+    tmp12 -= z2 << 2;
+
+    /* Odd part */
+
+    z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+
+    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));       /* c3 */
+    tmp13 = z1 + MULTIPLY(z2, FIX(0.513743148));    /* c1-c3 */
+    tmp14 = z1 - MULTIPLY(z3, FIX(2.176250899));    /* c1+c3 */
+
+    /* Final output stage */
+
+    wsptr[8*0] = (int) RIGHT_SHIFT(tmp10 + tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[8*4] = (int) RIGHT_SHIFT(tmp10 - tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[8*1] = (int) RIGHT_SHIFT(tmp11 + tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[8*3] = (int) RIGHT_SHIFT(tmp11 - tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[8*2] = (int) RIGHT_SHIFT(tmp12, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 5 rows from work array, store into output array.
+   * 10-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/20).
+   */
+  wsptr = workspace;
+  for (ctr = 0; ctr < 5; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    z3 <<= CONST_BITS;
+    z4 = (INT32) wsptr[4];
+    z1 = MULTIPLY(z4, FIX(1.144122806));         /* c4 */
+    z2 = MULTIPLY(z4, FIX(0.437016024));         /* c8 */
+    tmp10 = z3 + z1;
+    tmp11 = z3 - z2;
+
+    tmp22 = z3 - ((z1 - z2) << 1);               /* c0 = (c4-c8)*2 */
+
+    z2 = (INT32) wsptr[2];
+    z3 = (INT32) wsptr[6];
+
+    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));    /* c6 */
+    tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
+    tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
+
+    tmp20 = tmp10 + tmp12;
+    tmp24 = tmp10 - tmp12;
+    tmp21 = tmp11 + tmp13;
+    tmp23 = tmp11 - tmp13;
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+    z3 <<= CONST_BITS;
+    z4 = (INT32) wsptr[7];
+
+    tmp11 = z2 + z4;
+    tmp13 = z2 - z4;
+
+    tmp12 = MULTIPLY(tmp13, FIX(0.309016994));        /* (c3-c7)/2 */
+
+    z2 = MULTIPLY(tmp11, FIX(0.951056516));           /* (c3+c7)/2 */
+    z4 = z3 + tmp12;
+
+    tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
+    tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
+
+    z2 = MULTIPLY(tmp11, FIX(0.587785252));           /* (c1-c9)/2 */
+    z4 = z3 - tmp12 - (tmp13 << (CONST_BITS - 1));
+
+    tmp12 = ((z1 - tmp13) << CONST_BITS) - z3;
+
+    tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
+    tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 8;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 8x4 output block.
+ *
+ * 4-point IDCT in pass 1 (columns), 8-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_8x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3;
+  INT32 tmp10, tmp11, tmp12, tmp13;
+  INT32 z1, z2, z3;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[8*4];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array.
+   * 4-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
+   */
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+
+    tmp10 = (tmp0 + tmp2) << PASS1_BITS;
+    tmp12 = (tmp0 - tmp2) << PASS1_BITS;
+
+    /* Odd part */
+    /* Same rotation as in the even part of the 8x8 LL&M IDCT */
+
+    z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);               /* c6 */
+    /* Add fudge factor here for final descale. */
+    z1 += ONE << (CONST_BITS-PASS1_BITS-1);
+    tmp0 = RIGHT_SHIFT(z1 + MULTIPLY(z2, FIX_0_765366865), /* c2-c6 */
+		       CONST_BITS-PASS1_BITS);
+    tmp2 = RIGHT_SHIFT(z1 - MULTIPLY(z3, FIX_1_847759065), /* c2+c6 */
+		       CONST_BITS-PASS1_BITS);
+
+    /* Final output stage */
+
+    wsptr[8*0] = (int) (tmp10 + tmp0);
+    wsptr[8*3] = (int) (tmp10 - tmp0);
+    wsptr[8*1] = (int) (tmp12 + tmp2);
+    wsptr[8*2] = (int) (tmp12 - tmp2);
+  }
+
+  /* Pass 2: process rows from work array, store into output array. */
+  /* Note that we must descale the results by a factor of 8 == 2**3, */
+  /* and also undo the PASS1_BITS scaling. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 4; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part: reverse the even part of the forward DCT. */
+    /* The rotator is sqrt(2)*c(-6). */
+
+    z2 = (INT32) wsptr[2];
+    z3 = (INT32) wsptr[6];
+    
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
+    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);
+    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);
+    
+    /* Add fudge factor here for final descale. */
+    z2 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    z3 = (INT32) wsptr[4];
+    
+    tmp0 = (z2 + z3) << CONST_BITS;
+    tmp1 = (z2 - z3) << CONST_BITS;
+    
+    tmp10 = tmp0 + tmp2;
+    tmp13 = tmp0 - tmp2;
+    tmp11 = tmp1 + tmp3;
+    tmp12 = tmp1 - tmp3;
+
+    /* Odd part per figure 8; the matrix is unitary and hence its
+     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
+     */
+
+    tmp0 = (INT32) wsptr[7];
+    tmp1 = (INT32) wsptr[5];
+    tmp2 = (INT32) wsptr[3];
+    tmp3 = (INT32) wsptr[1];
+
+    z2 = tmp0 + tmp2;
+    z3 = tmp1 + tmp3;
+
+    z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* sqrt(2) * c3 */
+    z2 = MULTIPLY(z2, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
+    z3 = MULTIPLY(z3, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
+    z2 += z1;
+    z3 += z1;
+
+    z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
+    tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
+    tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
+    tmp0 += z1 + z2;
+    tmp3 += z1 + z3;
+
+    z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
+    tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
+    tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
+    tmp1 += z1 + z3;
+    tmp2 += z1 + z2;
+
+    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp3,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp3,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += DCTSIZE;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a reduced-size 6x3 output block.
+ *
+ * 3-point IDCT in pass 1 (columns), 6-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_6x3 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
+  INT32 z1, z2, z3;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[6*3];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array.
+   * 3-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/6).
+   */
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp0 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
+    tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
+    tmp10 = tmp0 + tmp12;
+    tmp2 = tmp0 - tmp12 - tmp12;
+
+    /* Odd part */
+
+    tmp12 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
+
+    /* Final output stage */
+
+    wsptr[6*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[6*2] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[6*1] = (int) RIGHT_SHIFT(tmp2, CONST_BITS-PASS1_BITS);
+  }
+  
+  /* Pass 2: process 3 rows from work array, store into output array.
+   * 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
+   */
+  wsptr = workspace;
+  for (ctr = 0; ctr < 3; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp0 <<= CONST_BITS;
+    tmp2 = (INT32) wsptr[4];
+    tmp10 = MULTIPLY(tmp2, FIX(0.707106781));   /* c4 */
+    tmp1 = tmp0 + tmp10;
+    tmp11 = tmp0 - tmp10 - tmp10;
+    tmp10 = (INT32) wsptr[2];
+    tmp0 = MULTIPLY(tmp10, FIX(1.224744871));   /* c2 */
+    tmp10 = tmp1 + tmp0;
+    tmp12 = tmp1 - tmp0;
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+    tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
+    tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
+    tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
+    tmp1 = (z1 - z2 - z3) << CONST_BITS;
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 6;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 4x2 output block.
+ *
+ * 2-point IDCT in pass 1 (columns), 4-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_4x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp2, tmp10, tmp12;
+  INT32 z1, z2, z3;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  INT32 * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  INT32 workspace[4*2];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 4; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp10 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+
+    /* Odd part */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+
+    /* Final output stage */
+
+    wsptr[4*0] = tmp10 + tmp0;
+    wsptr[4*1] = tmp10 - tmp0;
+  }
+
+  /* Pass 2: process 2 rows from work array, store into output array.
+   * 4-point IDCT kernel,
+   * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
+   */
+  wsptr = workspace;
+  for (ctr = 0; ctr < 2; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp0 = wsptr[0] + (ONE << 2);
+    tmp2 = wsptr[2];
+
+    tmp10 = (tmp0 + tmp2) << CONST_BITS;
+    tmp12 = (tmp0 - tmp2) << CONST_BITS;
+
+    /* Odd part */
+    /* Same rotation as in the even part of the 8x8 LL&M IDCT */
+
+    z2 = wsptr[1];
+    z3 = wsptr[3];
+
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);   /* c6 */
+    tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
+    tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
+					      CONST_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
+					      CONST_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
+					      CONST_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
+					      CONST_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 4;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 2x1 output block.
+ *
+ * 1-point IDCT in pass 1 (columns), 2-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_2x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp10;
+  ISLOW_MULT_TYPE * quantptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  SHIFT_TEMPS
+
+  /* Pass 1: empty. */
+
+  /* Pass 2: process 1 row from input, store into output array. */
+
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  outptr = output_buf[0] + output_col;
+
+  /* Even part */
+
+  tmp10 = DEQUANTIZE(coef_block[0], quantptr[0]);
+  /* Add fudge factor here for final descale. */
+  tmp10 += ONE << 2;
+
+  /* Odd part */
+
+  tmp0 = DEQUANTIZE(coef_block[1], quantptr[1]);
+
+  /* Final output stage */
+
+  outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0, 3) & RANGE_MASK];
+  outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0, 3) & RANGE_MASK];
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 8x16 output block.
+ *
+ * 16-point IDCT in pass 1 (columns), 8-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_8x16 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		JCOEFPTR coef_block,
+		JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
+  INT32 z1, z2, z3, z4;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[8*16];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array.
+   * 16-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/32).
+   */
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp0 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    tmp1 = MULTIPLY(z1, FIX(1.306562965));      /* c4[16] = c2[8] */
+    tmp2 = MULTIPLY(z1, FIX_0_541196100);       /* c12[16] = c6[8] */
+
+    tmp10 = tmp0 + tmp1;
+    tmp11 = tmp0 - tmp1;
+    tmp12 = tmp0 + tmp2;
+    tmp13 = tmp0 - tmp2;
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+    z3 = z1 - z2;
+    z4 = MULTIPLY(z3, FIX(0.275899379));        /* c14[16] = c7[8] */
+    z3 = MULTIPLY(z3, FIX(1.387039845));        /* c2[16] = c1[8] */
+
+    tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447);  /* (c6+c2)[16] = (c3+c1)[8] */
+    tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223);  /* (c6-c14)[16] = (c3-c7)[8] */
+    tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
+    tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
+
+    tmp20 = tmp10 + tmp0;
+    tmp27 = tmp10 - tmp0;
+    tmp21 = tmp12 + tmp1;
+    tmp26 = tmp12 - tmp1;
+    tmp22 = tmp13 + tmp2;
+    tmp25 = tmp13 - tmp2;
+    tmp23 = tmp11 + tmp3;
+    tmp24 = tmp11 - tmp3;
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+
+    tmp11 = z1 + z3;
+
+    tmp1  = MULTIPLY(z1 + z2, FIX(1.353318001));   /* c3 */
+    tmp2  = MULTIPLY(tmp11,   FIX(1.247225013));   /* c5 */
+    tmp3  = MULTIPLY(z1 + z4, FIX(1.093201867));   /* c7 */
+    tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586));   /* c9 */
+    tmp11 = MULTIPLY(tmp11,   FIX(0.666655658));   /* c11 */
+    tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528));   /* c13 */
+    tmp0  = tmp1 + tmp2 + tmp3 -
+	    MULTIPLY(z1, FIX(2.286341144));        /* c7+c5+c3-c1 */
+    tmp13 = tmp10 + tmp11 + tmp12 -
+	    MULTIPLY(z1, FIX(1.835730603));        /* c9+c11+c13-c15 */
+    z1    = MULTIPLY(z2 + z3, FIX(0.138617169));   /* c15 */
+    tmp1  += z1 + MULTIPLY(z2, FIX(0.071888074));  /* c9+c11-c3-c15 */
+    tmp2  += z1 - MULTIPLY(z3, FIX(1.125726048));  /* c5+c7+c15-c3 */
+    z1    = MULTIPLY(z3 - z2, FIX(1.407403738));   /* c1 */
+    tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282));  /* c1+c11-c9-c13 */
+    tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411));  /* c1+c5+c13-c7 */
+    z2    += z4;
+    z1    = MULTIPLY(z2, - FIX(0.666655658));      /* -c11 */
+    tmp1  += z1;
+    tmp3  += z1 + MULTIPLY(z4, FIX(1.065388962));  /* c3+c11+c15-c7 */
+    z2    = MULTIPLY(z2, - FIX(1.247225013));      /* -c5 */
+    tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809));  /* c1+c5+c9-c13 */
+    tmp12 += z2;
+    z2    = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
+    tmp2  += z2;
+    tmp3  += z2;
+    z2    = MULTIPLY(z4 - z3, FIX(0.410524528));   /* c13 */
+    tmp10 += z2;
+    tmp11 += z2;
+
+    /* Final output stage */
+
+    wsptr[8*0]  = (int) RIGHT_SHIFT(tmp20 + tmp0,  CONST_BITS-PASS1_BITS);
+    wsptr[8*15] = (int) RIGHT_SHIFT(tmp20 - tmp0,  CONST_BITS-PASS1_BITS);
+    wsptr[8*1]  = (int) RIGHT_SHIFT(tmp21 + tmp1,  CONST_BITS-PASS1_BITS);
+    wsptr[8*14] = (int) RIGHT_SHIFT(tmp21 - tmp1,  CONST_BITS-PASS1_BITS);
+    wsptr[8*2]  = (int) RIGHT_SHIFT(tmp22 + tmp2,  CONST_BITS-PASS1_BITS);
+    wsptr[8*13] = (int) RIGHT_SHIFT(tmp22 - tmp2,  CONST_BITS-PASS1_BITS);
+    wsptr[8*3]  = (int) RIGHT_SHIFT(tmp23 + tmp3,  CONST_BITS-PASS1_BITS);
+    wsptr[8*12] = (int) RIGHT_SHIFT(tmp23 - tmp3,  CONST_BITS-PASS1_BITS);
+    wsptr[8*4]  = (int) RIGHT_SHIFT(tmp24 + tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*11] = (int) RIGHT_SHIFT(tmp24 - tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[8*5]  = (int) RIGHT_SHIFT(tmp25 + tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*10] = (int) RIGHT_SHIFT(tmp25 - tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[8*6]  = (int) RIGHT_SHIFT(tmp26 + tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*9]  = (int) RIGHT_SHIFT(tmp26 - tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[8*7]  = (int) RIGHT_SHIFT(tmp27 + tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[8*8]  = (int) RIGHT_SHIFT(tmp27 - tmp13, CONST_BITS-PASS1_BITS);
+  }
+  
+  /* Pass 2: process rows from work array, store into output array. */
+  /* Note that we must descale the results by a factor of 8 == 2**3, */
+  /* and also undo the PASS1_BITS scaling. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 16; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+    
+    /* Even part: reverse the even part of the forward DCT. */
+    /* The rotator is sqrt(2)*c(-6). */
+    
+    z2 = (INT32) wsptr[2];
+    z3 = (INT32) wsptr[6];
+    
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
+    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);
+    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);
+    
+    /* Add fudge factor here for final descale. */
+    z2 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    z3 = (INT32) wsptr[4];
+    
+    tmp0 = (z2 + z3) << CONST_BITS;
+    tmp1 = (z2 - z3) << CONST_BITS;
+    
+    tmp10 = tmp0 + tmp2;
+    tmp13 = tmp0 - tmp2;
+    tmp11 = tmp1 + tmp3;
+    tmp12 = tmp1 - tmp3;
+    
+    /* Odd part per figure 8; the matrix is unitary and hence its
+     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
+     */
+    
+    tmp0 = (INT32) wsptr[7];
+    tmp1 = (INT32) wsptr[5];
+    tmp2 = (INT32) wsptr[3];
+    tmp3 = (INT32) wsptr[1];
+    
+    z2 = tmp0 + tmp2;
+    z3 = tmp1 + tmp3;
+
+    z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* sqrt(2) * c3 */
+    z2 = MULTIPLY(z2, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
+    z3 = MULTIPLY(z3, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
+    z2 += z1;
+    z3 += z1;
+
+    z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
+    tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
+    tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
+    tmp0 += z1 + z2;
+    tmp3 += z1 + z3;
+
+    z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
+    tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
+    tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
+    tmp1 += z1 + z3;
+    tmp2 += z1 + z2;
+    
+    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
+    
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp3,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp3,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp1,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    
+    wsptr += DCTSIZE;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 7x14 output block.
+ *
+ * 14-point IDCT in pass 1 (columns), 7-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_7x14 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		JCOEFPTR coef_block,
+		JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
+  INT32 z1, z2, z3, z4;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[7*14];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array.
+   * 14-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/28).
+   */
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 7; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    z1 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    z1 += ONE << (CONST_BITS-PASS1_BITS-1);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z2 = MULTIPLY(z4, FIX(1.274162392));         /* c4 */
+    z3 = MULTIPLY(z4, FIX(0.314692123));         /* c12 */
+    z4 = MULTIPLY(z4, FIX(0.881747734));         /* c8 */
+
+    tmp10 = z1 + z2;
+    tmp11 = z1 + z3;
+    tmp12 = z1 - z4;
+
+    tmp23 = RIGHT_SHIFT(z1 - ((z2 + z3 - z4) << 1), /* c0 = (c4+c12-c8)*2 */
+			CONST_BITS-PASS1_BITS);
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    z3 = MULTIPLY(z1 + z2, FIX(1.105676686));    /* c6 */
+
+    tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
+    tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
+    tmp15 = MULTIPLY(z1, FIX(0.613604268)) -     /* c10 */
+	    MULTIPLY(z2, FIX(1.378756276));      /* c2 */
+
+    tmp20 = tmp10 + tmp13;
+    tmp26 = tmp10 - tmp13;
+    tmp21 = tmp11 + tmp14;
+    tmp25 = tmp11 - tmp14;
+    tmp22 = tmp12 + tmp15;
+    tmp24 = tmp12 - tmp15;
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+    tmp13 = z4 << CONST_BITS;
+
+    tmp14 = z1 + z3;
+    tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607));           /* c3 */
+    tmp12 = MULTIPLY(tmp14, FIX(1.197448846));             /* c5 */
+    tmp10 = tmp11 + tmp12 + tmp13 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
+    tmp14 = MULTIPLY(tmp14, FIX(0.752406978));             /* c9 */
+    tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426));        /* c9+c11-c13 */
+    z1    -= z2;
+    tmp15 = MULTIPLY(z1, FIX(0.467085129)) - tmp13;        /* c11 */
+    tmp16 += tmp15;
+    z1    += z4;
+    z4    = MULTIPLY(z2 + z3, - FIX(0.158341681)) - tmp13; /* -c13 */
+    tmp11 += z4 - MULTIPLY(z2, FIX(0.424103948));          /* c3-c9-c13 */
+    tmp12 += z4 - MULTIPLY(z3, FIX(2.373959773));          /* c3+c5-c13 */
+    z4    = MULTIPLY(z3 - z2, FIX(1.405321284));           /* c1 */
+    tmp14 += z4 + tmp13 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
+    tmp15 += z4 + MULTIPLY(z2, FIX(0.674957567));          /* c1+c11-c5 */
+
+    tmp13 = (z1 - z3) << PASS1_BITS;
+
+    /* Final output stage */
+
+    wsptr[7*0]  = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[7*13] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[7*1]  = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[7*12] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[7*2]  = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[7*11] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[7*3]  = (int) (tmp23 + tmp13);
+    wsptr[7*10] = (int) (tmp23 - tmp13);
+    wsptr[7*4]  = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[7*9]  = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[7*5]  = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
+    wsptr[7*8]  = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
+    wsptr[7*6]  = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
+    wsptr[7*7]  = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 14 rows from work array, store into output array.
+   * 7-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/14).
+   */
+  wsptr = workspace;
+  for (ctr = 0; ctr < 14; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp23 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp23 <<= CONST_BITS;
+
+    z1 = (INT32) wsptr[2];
+    z2 = (INT32) wsptr[4];
+    z3 = (INT32) wsptr[6];
+
+    tmp20 = MULTIPLY(z2 - z3, FIX(0.881747734));       /* c4 */
+    tmp22 = MULTIPLY(z1 - z2, FIX(0.314692123));       /* c6 */
+    tmp21 = tmp20 + tmp22 + tmp23 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
+    tmp10 = z1 + z3;
+    z2 -= tmp10;
+    tmp10 = MULTIPLY(tmp10, FIX(1.274162392)) + tmp23; /* c2 */
+    tmp20 += tmp10 - MULTIPLY(z3, FIX(0.077722536));   /* c2-c4-c6 */
+    tmp22 += tmp10 - MULTIPLY(z1, FIX(2.470602249));   /* c2+c4+c6 */
+    tmp23 += MULTIPLY(z2, FIX(1.414213562));           /* c0 */
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+
+    tmp11 = MULTIPLY(z1 + z2, FIX(0.935414347));       /* (c3+c1-c5)/2 */
+    tmp12 = MULTIPLY(z1 - z2, FIX(0.170262339));       /* (c3+c5-c1)/2 */
+    tmp10 = tmp11 - tmp12;
+    tmp11 += tmp12;
+    tmp12 = MULTIPLY(z2 + z3, - FIX(1.378756276));     /* -c1 */
+    tmp11 += tmp12;
+    z2 = MULTIPLY(z1 + z3, FIX(0.613604268));          /* c5 */
+    tmp10 += z2;
+    tmp12 += z2 + MULTIPLY(z3, FIX(1.870828693));      /* c3+c1-c5 */
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 7;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 6x12 output block.
+ *
+ * 12-point IDCT in pass 1 (columns), 6-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_6x12 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		JCOEFPTR coef_block,
+		JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
+  INT32 z1, z2, z3, z4;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[6*12];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array.
+   * 12-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/24).
+   */
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    z3 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    z3 += ONE << (CONST_BITS-PASS1_BITS-1);
+
+    z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
+
+    tmp10 = z3 + z4;
+    tmp11 = z3 - z4;
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
+    z1 <<= CONST_BITS;
+    z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+    z2 <<= CONST_BITS;
+
+    tmp12 = z1 - z2;
+
+    tmp21 = z3 + tmp12;
+    tmp24 = z3 - tmp12;
+
+    tmp12 = z4 + z2;
+
+    tmp20 = tmp10 + tmp12;
+    tmp25 = tmp10 - tmp12;
+
+    tmp12 = z4 - z1 - z2;
+
+    tmp22 = tmp11 + tmp12;
+    tmp23 = tmp11 - tmp12;
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+
+    tmp11 = MULTIPLY(z2, FIX(1.306562965));                  /* c3 */
+    tmp14 = MULTIPLY(z2, - FIX_0_541196100);                 /* -c9 */
+
+    tmp10 = z1 + z3;
+    tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669));          /* c7 */
+    tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384));       /* c5-c7 */
+    tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716));  /* c1-c5 */
+    tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580));           /* -(c7+c11) */
+    tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
+    tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
+    tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) -        /* c7-c11 */
+	     MULTIPLY(z4, FIX(1.982889723));                 /* c5+c7 */
+
+    z1 -= z4;
+    z2 -= z3;
+    z3 = MULTIPLY(z1 + z2, FIX_0_541196100);                 /* c9 */
+    tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865);              /* c3-c9 */
+    tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065);              /* c3+c9 */
+
+    /* Final output stage */
+
+    wsptr[6*0]  = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[6*11] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[6*1]  = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[6*10] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[6*2]  = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[6*9]  = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
+    wsptr[6*3]  = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[6*8]  = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[6*4]  = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[6*7]  = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[6*5]  = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
+    wsptr[6*6]  = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 12 rows from work array, store into output array.
+   * 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
+   */
+  wsptr = workspace;
+  for (ctr = 0; ctr < 12; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp10 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp10 <<= CONST_BITS;
+    tmp12 = (INT32) wsptr[4];
+    tmp20 = MULTIPLY(tmp12, FIX(0.707106781));   /* c4 */
+    tmp11 = tmp10 + tmp20;
+    tmp21 = tmp10 - tmp20 - tmp20;
+    tmp20 = (INT32) wsptr[2];
+    tmp10 = MULTIPLY(tmp20, FIX(1.224744871));   /* c2 */
+    tmp20 = tmp11 + tmp10;
+    tmp22 = tmp11 - tmp10;
+
+    /* Odd part */
+
+    z1 = (INT32) wsptr[1];
+    z2 = (INT32) wsptr[3];
+    z3 = (INT32) wsptr[5];
+    tmp11 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
+    tmp10 = tmp11 + ((z1 + z2) << CONST_BITS);
+    tmp12 = tmp11 + ((z3 - z2) << CONST_BITS);
+    tmp11 = (z1 - z2 - z3) << CONST_BITS;
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 6;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 5x10 output block.
+ *
+ * 10-point IDCT in pass 1 (columns), 5-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_5x10 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+		JCOEFPTR coef_block,
+		JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
+  INT32 tmp20, tmp21, tmp22, tmp23, tmp24;
+  INT32 z1, z2, z3, z4, z5;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[5*10];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array.
+   * 10-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/20).
+   */
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 5; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    z3 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    z3 += ONE << (CONST_BITS-PASS1_BITS-1);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z1 = MULTIPLY(z4, FIX(1.144122806));         /* c4 */
+    z2 = MULTIPLY(z4, FIX(0.437016024));         /* c8 */
+    tmp10 = z3 + z1;
+    tmp11 = z3 - z2;
+
+    tmp22 = RIGHT_SHIFT(z3 - ((z1 - z2) << 1),   /* c0 = (c4-c8)*2 */
+			CONST_BITS-PASS1_BITS);
+
+    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+
+    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));    /* c6 */
+    tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
+    tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
+
+    tmp20 = tmp10 + tmp12;
+    tmp24 = tmp10 - tmp12;
+    tmp21 = tmp11 + tmp13;
+    tmp23 = tmp11 - tmp13;
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+
+    tmp11 = z2 + z4;
+    tmp13 = z2 - z4;
+
+    tmp12 = MULTIPLY(tmp13, FIX(0.309016994));        /* (c3-c7)/2 */
+    z5 = z3 << CONST_BITS;
+
+    z2 = MULTIPLY(tmp11, FIX(0.951056516));           /* (c3+c7)/2 */
+    z4 = z5 + tmp12;
+
+    tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
+    tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
+
+    z2 = MULTIPLY(tmp11, FIX(0.587785252));           /* (c1-c9)/2 */
+    z4 = z5 - tmp12 - (tmp13 << (CONST_BITS - 1));
+
+    tmp12 = (z1 - tmp13 - z3) << PASS1_BITS;
+
+    tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
+    tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
+
+    /* Final output stage */
+
+    wsptr[5*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[5*9] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
+    wsptr[5*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[5*8] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
+    wsptr[5*2] = (int) (tmp22 + tmp12);
+    wsptr[5*7] = (int) (tmp22 - tmp12);
+    wsptr[5*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[5*6] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
+    wsptr[5*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
+    wsptr[5*5] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 10 rows from work array, store into output array.
+   * 5-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/10).
+   */
+  wsptr = workspace;
+  for (ctr = 0; ctr < 10; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp12 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp12 <<= CONST_BITS;
+    tmp13 = (INT32) wsptr[2];
+    tmp14 = (INT32) wsptr[4];
+    z1 = MULTIPLY(tmp13 + tmp14, FIX(0.790569415)); /* (c2+c4)/2 */
+    z2 = MULTIPLY(tmp13 - tmp14, FIX(0.353553391)); /* (c2-c4)/2 */
+    z3 = tmp12 + z2;
+    tmp10 = z3 + z1;
+    tmp11 = z3 - z1;
+    tmp12 -= z2 << 2;
+
+    /* Odd part */
+
+    z2 = (INT32) wsptr[1];
+    z3 = (INT32) wsptr[3];
+
+    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));       /* c3 */
+    tmp13 = z1 + MULTIPLY(z2, FIX(0.513743148));    /* c1-c3 */
+    tmp14 = z1 - MULTIPLY(z3, FIX(2.176250899));    /* c1+c3 */
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp13,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp13,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp14,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp14,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 5;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 4x8 output block.
+ *
+ * 8-point IDCT in pass 1 (columns), 4-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_4x8 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3;
+  INT32 tmp10, tmp11, tmp12, tmp13;
+  INT32 z1, z2, z3;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[4*8];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array. */
+  /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 4; ctr > 0; ctr--) {
+    /* Due to quantization, we will usually find that many of the input
+     * coefficients are zero, especially the AC terms.  We can exploit this
+     * by short-circuiting the IDCT calculation for any column in which all
+     * the AC terms are zero.  In that case each output is equal to the
+     * DC coefficient (with scale factor as needed).
+     * With typical images and quantization tables, half or more of the
+     * column DCT calculations can be simplified this way.
+     */
+
+    if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
+	inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
+	inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
+	inptr[DCTSIZE*7] == 0) {
+      /* AC terms all zero */
+      int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
+
+      wsptr[4*0] = dcval;
+      wsptr[4*1] = dcval;
+      wsptr[4*2] = dcval;
+      wsptr[4*3] = dcval;
+      wsptr[4*4] = dcval;
+      wsptr[4*5] = dcval;
+      wsptr[4*6] = dcval;
+      wsptr[4*7] = dcval;
+
+      inptr++;			/* advance pointers to next column */
+      quantptr++;
+      wsptr++;
+      continue;
+    }
+
+    /* Even part: reverse the even part of the forward DCT. */
+    /* The rotator is sqrt(2)*c(-6). */
+
+    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+    
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
+    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);
+    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);
+    
+    z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    z2 <<= CONST_BITS;
+    z3 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    z2 += ONE << (CONST_BITS-PASS1_BITS-1);
+
+    tmp0 = z2 + z3;
+    tmp1 = z2 - z3;
+    
+    tmp10 = tmp0 + tmp2;
+    tmp13 = tmp0 - tmp2;
+    tmp11 = tmp1 + tmp3;
+    tmp12 = tmp1 - tmp3;
+
+    /* Odd part per figure 8; the matrix is unitary and hence its
+     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
+     */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+    tmp1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    tmp2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    tmp3 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+
+    z2 = tmp0 + tmp2;
+    z3 = tmp1 + tmp3;
+
+    z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* sqrt(2) * c3 */
+    z2 = MULTIPLY(z2, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
+    z3 = MULTIPLY(z3, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
+    z2 += z1;
+    z3 += z1;
+
+    z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
+    tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
+    tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
+    tmp0 += z1 + z2;
+    tmp3 += z1 + z3;
+
+    z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
+    tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
+    tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
+    tmp1 += z1 + z3;
+    tmp2 += z1 + z2;
+
+    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
+
+    wsptr[4*0] = (int) RIGHT_SHIFT(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
+    wsptr[4*7] = (int) RIGHT_SHIFT(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
+    wsptr[4*1] = (int) RIGHT_SHIFT(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
+    wsptr[4*6] = (int) RIGHT_SHIFT(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
+    wsptr[4*2] = (int) RIGHT_SHIFT(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
+    wsptr[4*5] = (int) RIGHT_SHIFT(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
+    wsptr[4*3] = (int) RIGHT_SHIFT(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[4*4] = (int) RIGHT_SHIFT(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
+
+    inptr++;			/* advance pointers to next column */
+    quantptr++;
+    wsptr++;
+  }
+
+  /* Pass 2: process 8 rows from work array, store into output array.
+   * 4-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
+   */
+  wsptr = workspace;
+  for (ctr = 0; ctr < 8; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp2 = (INT32) wsptr[2];
+
+    tmp10 = (tmp0 + tmp2) << CONST_BITS;
+    tmp12 = (tmp0 - tmp2) << CONST_BITS;
+
+    /* Odd part */
+    /* Same rotation as in the even part of the 8x8 LL&M IDCT */
+
+    z2 = (INT32) wsptr[1];
+    z3 = (INT32) wsptr[3];
+
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);   /* c6 */
+    tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
+    tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    
+    wsptr += 4;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a reduced-size 3x6 output block.
+ *
+ * 6-point IDCT in pass 1 (columns), 3-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_3x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
+  INT32 z1, z2, z3;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  int * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  int workspace[3*6];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array.
+   * 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
+   */
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 3; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp0 <<= CONST_BITS;
+    /* Add fudge factor here for final descale. */
+    tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
+    tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
+    tmp10 = MULTIPLY(tmp2, FIX(0.707106781));   /* c4 */
+    tmp1 = tmp0 + tmp10;
+    tmp11 = RIGHT_SHIFT(tmp0 - tmp10 - tmp10, CONST_BITS-PASS1_BITS);
+    tmp10 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+    tmp0 = MULTIPLY(tmp10, FIX(1.224744871));   /* c2 */
+    tmp10 = tmp1 + tmp0;
+    tmp12 = tmp1 - tmp0;
+
+    /* Odd part */
+
+    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
+    tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
+    tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
+    tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
+    tmp1 = (z1 - z2 - z3) << PASS1_BITS;
+
+    /* Final output stage */
+
+    wsptr[3*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[3*5] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
+    wsptr[3*1] = (int) (tmp11 + tmp1);
+    wsptr[3*4] = (int) (tmp11 - tmp1);
+    wsptr[3*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
+    wsptr[3*3] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
+  }
+
+  /* Pass 2: process 6 rows from work array, store into output array.
+   * 3-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/6).
+   */
+  wsptr = workspace;
+  for (ctr = 0; ctr < 6; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
+    tmp0 <<= CONST_BITS;
+    tmp2 = (INT32) wsptr[2];
+    tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
+    tmp10 = tmp0 + tmp12;
+    tmp2 = tmp0 - tmp12 - tmp12;
+
+    /* Odd part */
+
+    tmp12 = (INT32) wsptr[1];
+    tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp2,
+					      CONST_BITS+PASS1_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 3;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 2x4 output block.
+ *
+ * 4-point IDCT in pass 1 (columns), 2-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_2x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp2, tmp10, tmp12;
+  INT32 z1, z2, z3;
+  JCOEFPTR inptr;
+  ISLOW_MULT_TYPE * quantptr;
+  INT32 * wsptr;
+  JSAMPROW outptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  int ctr;
+  INT32 workspace[2*4];	/* buffers data between passes */
+  SHIFT_TEMPS
+
+  /* Pass 1: process columns from input, store into work array.
+   * 4-point IDCT kernel,
+   * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
+   */
+  inptr = coef_block;
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+  wsptr = workspace;
+  for (ctr = 0; ctr < 2; ctr++, inptr++, quantptr++, wsptr++) {
+    /* Even part */
+
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+    tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
+
+    tmp10 = (tmp0 + tmp2) << CONST_BITS;
+    tmp12 = (tmp0 - tmp2) << CONST_BITS;
+
+    /* Odd part */
+    /* Same rotation as in the even part of the 8x8 LL&M IDCT */
+
+    z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
+    z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
+
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);   /* c6 */
+    tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
+    tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
+
+    /* Final output stage */
+
+    wsptr[2*0] = tmp10 + tmp0;
+    wsptr[2*3] = tmp10 - tmp0;
+    wsptr[2*1] = tmp12 + tmp2;
+    wsptr[2*2] = tmp12 - tmp2;
+  }
+
+  /* Pass 2: process 4 rows from work array, store into output array. */
+
+  wsptr = workspace;
+  for (ctr = 0; ctr < 4; ctr++) {
+    outptr = output_buf[ctr] + output_col;
+
+    /* Even part */
+
+    /* Add fudge factor here for final descale. */
+    tmp10 = wsptr[0] + (ONE << (CONST_BITS+2));
+
+    /* Odd part */
+
+    tmp0 = wsptr[1];
+
+    /* Final output stage */
+
+    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS+3)
+			    & RANGE_MASK];
+    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS+3)
+			    & RANGE_MASK];
+
+    wsptr += 2;		/* advance pointer to next row */
+  }
+}
+
+
+/*
+ * Perform dequantization and inverse DCT on one block of coefficients,
+ * producing a 1x2 output block.
+ *
+ * 2-point IDCT in pass 1 (columns), 1-point in pass 2 (rows).
+ */
+
+GLOBAL(void)
+jpeg_idct_1x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
+	       JCOEFPTR coef_block,
+	       JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+  INT32 tmp0, tmp10;
+  ISLOW_MULT_TYPE * quantptr;
+  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  SHIFT_TEMPS
+
+  /* Process 1 column from input, store into output array. */
+
+  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
+
+  /* Even part */
+    
+  tmp10 = DEQUANTIZE(coef_block[DCTSIZE*0], quantptr[DCTSIZE*0]);
+  /* Add fudge factor here for final descale. */
+  tmp10 += ONE << 2;
+
+  /* Odd part */
+
+  tmp0 = DEQUANTIZE(coef_block[DCTSIZE*1], quantptr[DCTSIZE*1]);
+
+  /* Final output stage */
+
+  output_buf[0][output_col] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0, 3)
+					  & RANGE_MASK];
+  output_buf[1][output_col] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0, 3)
+					  & RANGE_MASK];
+}
+
+#endif /* IDCT_SCALING_SUPPORTED */
+#endif /* DCT_ISLOW_SUPPORTED */
diff --git a/src/jpeg-6b/jinclude.h b/src/jpeg-8c/jinclude.h
index 33d2cd5b..0a4f1514 100644
--- a/src/jpeg-6b/jinclude.h
+++ b/src/jpeg-8c/jinclude.h
@@ -14,38 +14,10 @@
  * JPEG library.  Most applications need only include jpeglib.h.
  */
 
-#ifdef __WIN64__
-#include "basetsd.h"
-#endif
-
-#ifdef _MSC_VER
-
-#pragma warning(disable : 4018)     // signed/unsigned mismatch
-#pragma warning(disable : 4032)
-#pragma warning(disable : 4051)
-#pragma warning(disable : 4057)		// slightly different base types
-#pragma warning(disable : 4100)		// unreferenced formal parameter
-#pragma warning(disable : 4115)
-#pragma warning(disable : 4125)		// decimal digit terminates octal escape sequence
-#pragma warning(disable : 4127)		// conditional expression is constant
-#pragma warning(disable : 4136)
-#pragma warning(disable : 4152)		// nonstandard extension, function/data pointer conversion in expression
-#pragma warning(disable : 4201)
-#pragma warning(disable : 4214)
-#pragma warning(disable : 4244)
-#pragma warning(disable : 4305)		// truncation from const double to float
-#pragma warning(disable : 4310)		// cast truncates constant value
-#pragma warning(disable:  4505) 	// unreferenced local function has been removed
-#pragma warning(disable : 4514)
-#pragma warning(disable : 4702)		// unreachable code
-#pragma warning(disable : 4711)		// selected for automatic inline expansion
-#pragma warning(disable : 4220)		// varargs matches remaining parameters
-#pragma warning(disable : 4761)		// integral size mismatch 
-#endif
 
 /* Include auto-config file to find out which system include files we need. */
 
-#include "../jpeg-6b/jconfig.h"		/* auto configuration options */
+#include "jconfig.h"		/* auto configuration options */
 #define JCONFIG_INCLUDED	/* so that jpeglib.h doesn't do it again */
 
 /*
diff --git a/src/jpeg-6b/jmemmgr.c b/src/jpeg-8c/jmemmgr.c
index d801b322..d801b322 100644
--- a/src/jpeg-6b/jmemmgr.c
+++ b/src/jpeg-8c/jmemmgr.c
diff --git a/src/jpeg-6b/jmemnobs.c b/src/jpeg-8c/jmemnobs.c
index cbd60814..eb8c3377 100644
--- a/src/jpeg-6b/jmemnobs.c
+++ b/src/jpeg-8c/jmemnobs.c
@@ -8,35 +8,39 @@
  * This file provides a really simple implementation of the system-
  * dependent portion of the JPEG memory manager.  This implementation
  * assumes that no backing-store files are needed: all required space
- * can be obtained from ri.Malloc().
+ * can be obtained from malloc().
  * This is very portable in the sense that it'll compile on almost anything,
  * but you'd better have lots of main memory (or virtual memory) if you want
  * to process big images.
  * Note that the max_memory_to_use option is ignored by this implementation.
  */
 
-#include "../renderer/tr_local.h"
-
 #define JPEG_INTERNALS
 #include "jinclude.h"
 #include "jpeglib.h"
 #include "jmemsys.h"		/* import the system-dependent declarations */
 
+#ifndef HAVE_STDLIB_H		/* <stdlib.h> should declare malloc(),free() */
+extern void * malloc JPP((size_t size));
+extern void free JPP((void *ptr));
+#endif
+
+
 /*
  * Memory allocation and freeing are controlled by the regular library
- * routines ri.Malloc() and ri.Free().
+ * routines malloc() and free().
  */
 
 GLOBAL(void *)
 jpeg_get_small (j_common_ptr cinfo, size_t sizeofobject)
 {
-  return (void *) ri.Malloc(sizeofobject);
+  return (void *) malloc(sizeofobject);
 }
 
 GLOBAL(void)
 jpeg_free_small (j_common_ptr cinfo, void * object, size_t sizeofobject)
 {
-  ri.Free(object);
+  free(object);
 }
 
 
@@ -50,13 +54,13 @@ jpeg_free_small (j_common_ptr cinfo, void * object, size_t sizeofobject)
 GLOBAL(void FAR *)
 jpeg_get_large (j_common_ptr cinfo, size_t sizeofobject)
 {
-  return (void FAR *) ri.Malloc(sizeofobject);
+  return (void FAR *) malloc(sizeofobject);
 }
 
 GLOBAL(void)
 jpeg_free_large (j_common_ptr cinfo, void FAR * object, size_t sizeofobject)
 {
-  ri.Free(object);
+  free(object);
 }
 
 
diff --git a/src/jpeg-6b/jmemsys.h b/src/jpeg-8c/jmemsys.h
index 6c3c6d34..6c3c6d34 100644
--- a/src/jpeg-6b/jmemsys.h
+++ b/src/jpeg-8c/jmemsys.h
diff --git a/src/jpeg-6b/jmorecfg.h b/src/jpeg-8c/jmorecfg.h
index d677127d..928d052c 100644
--- a/src/jpeg-6b/jmorecfg.h
+++ b/src/jpeg-8c/jmorecfg.h
@@ -2,6 +2,7 @@
  * jmorecfg.h
  *
  * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Modified 1997-2009 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -157,12 +158,15 @@ typedef short INT16;
 
 /* INT32 must hold at least signed 32-bit values. */
 
-/* MinGW basetsd.h defines INT32 - don't redefine it */
-#ifndef __WIN64
-#if !(defined __MINGW32__ && (defined _BASETSD_H || defined _BASETSD_H_))
+#ifndef XMD_H			/* X11/xmd.h correctly defines INT32 */
+#ifndef _BASETSD_H_		/* Microsoft defines it in basetsd.h */
+#ifndef _BASETSD_H		/* MinGW is slightly different */
+#ifndef QGLOBAL_H		/* Qt defines it in qglobal.h */
 typedef long INT32;
 #endif
 #endif
+#endif
+#endif
 
 /* Datatype used for image dimensions.  The JPEG standard only supports
  * images up to 64K*64K due to 16-bit fields in SOF markers.  Therefore
@@ -212,13 +216,13 @@ typedef unsigned int JDIMENSION;
  * explicit coding is needed; see uses of the NEED_FAR_POINTERS symbol.
  */
 
+#ifndef FAR
 #ifdef NEED_FAR_POINTERS
-#undef FAR
 #define FAR  far
 #else
-#undef FAR
 #define FAR
 #endif
+#endif
 
 
 /*
@@ -228,7 +232,9 @@ typedef unsigned int JDIMENSION;
  * Defining HAVE_BOOLEAN before including jpeglib.h should make it work.
  */
 
-typedef unsigned char boolean;
+#ifndef HAVE_BOOLEAN
+typedef int boolean;
+#endif
 #ifndef FALSE			/* in case these macros already exist */
 #define FALSE	0		/* values of boolean */
 #endif
@@ -259,19 +265,18 @@ typedef unsigned char boolean;
  * (You may HAVE to do that if your compiler doesn't like null source files.)
  */
 
-/* Arithmetic coding is unsupported for legal reasons.  Complaints to IBM. */
-
 /* Capability options common to encoder and decoder: */
 
-#undef DCT_ISLOW_SUPPORTED	/* slow but accurate integer algorithm */
-#undef DCT_IFAST_SUPPORTED	/* faster, less accurate integer method */
+#define DCT_ISLOW_SUPPORTED	/* slow but accurate integer algorithm */
+#define DCT_IFAST_SUPPORTED	/* faster, less accurate integer method */
 #define DCT_FLOAT_SUPPORTED	/* floating-point: accurate, fast on fast HW */
 
 /* Encoder capability options: */
 
-#undef  C_ARITH_CODING_SUPPORTED    /* Arithmetic coding back end? */
+#define C_ARITH_CODING_SUPPORTED    /* Arithmetic coding back end? */
 #define C_MULTISCAN_FILES_SUPPORTED /* Multiple-scan JPEG files? */
 #define C_PROGRESSIVE_SUPPORTED	    /* Progressive JPEG? (Requires MULTISCAN)*/
+#define DCT_SCALING_SUPPORTED	    /* Input rescaling via DCT? (Requires DCT_ISLOW)*/
 #define ENTROPY_OPT_SUPPORTED	    /* Optimization of entropy coding parms? */
 /* Note: if you selected 12-bit data precision, it is dangerous to turn off
  * ENTROPY_OPT_SUPPORTED.  The standard Huffman tables are only good for 8-bit
@@ -285,16 +290,16 @@ typedef unsigned char boolean;
 
 /* Decoder capability options: */
 
-#undef  D_ARITH_CODING_SUPPORTED    /* Arithmetic coding back end? */
-#undef D_MULTISCAN_FILES_SUPPORTED /* Multiple-scan JPEG files? */
-#undef D_PROGRESSIVE_SUPPORTED	    /* Progressive JPEG? (Requires MULTISCAN)*/
-#undef SAVE_MARKERS_SUPPORTED	    /* jpeg_save_markers() needed? */
-#undef BLOCK_SMOOTHING_SUPPORTED   /* Block smoothing? (Progressive only) */
-#undef IDCT_SCALING_SUPPORTED	    /* Output rescaling via IDCT? */
+#define D_ARITH_CODING_SUPPORTED    /* Arithmetic coding back end? */
+#define D_MULTISCAN_FILES_SUPPORTED /* Multiple-scan JPEG files? */
+#define D_PROGRESSIVE_SUPPORTED	    /* Progressive JPEG? (Requires MULTISCAN)*/
+#define IDCT_SCALING_SUPPORTED	    /* Output rescaling via IDCT? */
+#define SAVE_MARKERS_SUPPORTED	    /* jpeg_save_markers() needed? */
+#define BLOCK_SMOOTHING_SUPPORTED   /* Block smoothing? (Progressive only) */
 #undef  UPSAMPLE_SCALING_SUPPORTED  /* Output rescaling at upsample stage? */
-#undef UPSAMPLE_MERGING_SUPPORTED  /* Fast path for sloppy upsampling? */
-#undef QUANT_1PASS_SUPPORTED	    /* 1-pass color quantization? */
-#undef QUANT_2PASS_SUPPORTED	    /* 2-pass color quantization? */
+#define UPSAMPLE_MERGING_SUPPORTED  /* Fast path for sloppy upsampling? */
+#define QUANT_1PASS_SUPPORTED	    /* 1-pass color quantization? */
+#define QUANT_2PASS_SUPPORTED	    /* 2-pass color quantization? */
 
 /* more capability options later, no doubt */
 
@@ -317,7 +322,7 @@ typedef unsigned char boolean;
 #define RGB_RED		0	/* Offset of Red in an RGB scanline element */
 #define RGB_GREEN	1	/* Offset of Green */
 #define RGB_BLUE	2	/* Offset of Blue */
-#define RGB_PIXELSIZE	4	/* JSAMPLEs per RGB scanline element */
+#define RGB_PIXELSIZE	3	/* JSAMPLEs per RGB scanline element */
 
 
 /* Definitions for speed-related optimizations. */
diff --git a/src/jpeg-6b/jpegint.h b/src/jpeg-8c/jpegint.h
index 95b00d40..0c27a4e4 100644
--- a/src/jpeg-6b/jpegint.h
+++ b/src/jpeg-8c/jpegint.h
@@ -2,6 +2,7 @@
  * jpegint.h
  *
  * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Modified 1997-2009 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -99,14 +100,16 @@ struct jpeg_downsampler {
 };
 
 /* Forward DCT (also controls coefficient quantization) */
+typedef JMETHOD(void, forward_DCT_ptr,
+		(j_compress_ptr cinfo, jpeg_component_info * compptr,
+		 JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
+		 JDIMENSION start_row, JDIMENSION start_col,
+		 JDIMENSION num_blocks));
+
 struct jpeg_forward_dct {
   JMETHOD(void, start_pass, (j_compress_ptr cinfo));
-  /* perhaps this should be an array??? */
-  JMETHOD(void, forward_DCT, (j_compress_ptr cinfo,
-			      jpeg_component_info * compptr,
-			      JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
-			      JDIMENSION start_row, JDIMENSION start_col,
-			      JDIMENSION num_blocks));
+  /* It is useful to allow each component to have a separate FDCT method. */
+  forward_DCT_ptr forward_DCT[MAX_COMPONENTS];
 };
 
 /* Entropy encoding */
@@ -210,10 +213,6 @@ struct jpeg_entropy_decoder {
   JMETHOD(void, start_pass, (j_decompress_ptr cinfo));
   JMETHOD(boolean, decode_mcu, (j_decompress_ptr cinfo,
 				JBLOCKROW *MCU_data));
-
-  /* This is here to share code between baseline and progressive decoders; */
-  /* other modules probably should not use it */
-  boolean insufficient_data;	/* set TRUE after emitting warning */
 };
 
 /* Inverse DCT (also performs dequantization) */
@@ -303,7 +302,7 @@ struct jpeg_color_quantizer {
 #define jinit_downsampler	jIDownsampler
 #define jinit_forward_dct	jIFDCT
 #define jinit_huff_encoder	jIHEncoder
-#define jinit_phuff_encoder	jIPHEncoder
+#define jinit_arith_encoder	jIAEncoder
 #define jinit_marker_writer	jIMWriter
 #define jinit_master_decompress	jIDMaster
 #define jinit_d_main_controller	jIDMainC
@@ -312,7 +311,7 @@ struct jpeg_color_quantizer {
 #define jinit_input_controller	jIInCtlr
 #define jinit_marker_reader	jIMReader
 #define jinit_huff_decoder	jIHDecoder
-#define jinit_phuff_decoder	jIPHDecoder
+#define jinit_arith_decoder	jIADecoder
 #define jinit_inverse_dct	jIIDCT
 #define jinit_upsampler		jIUpsampler
 #define jinit_color_deconverter	jIDColor
@@ -327,6 +326,13 @@ struct jpeg_color_quantizer {
 #define jzero_far		jZeroFar
 #define jpeg_zigzag_order	jZIGTable
 #define jpeg_natural_order	jZAGTable
+#define jpeg_natural_order7	jZAGTable7
+#define jpeg_natural_order6	jZAGTable6
+#define jpeg_natural_order5	jZAGTable5
+#define jpeg_natural_order4	jZAGTable4
+#define jpeg_natural_order3	jZAGTable3
+#define jpeg_natural_order2	jZAGTable2
+#define jpeg_aritab		jAriTab
 #endif /* NEED_SHORT_EXTERNAL_NAMES */
 
 
@@ -344,7 +350,7 @@ EXTERN(void) jinit_color_converter JPP((j_compress_ptr cinfo));
 EXTERN(void) jinit_downsampler JPP((j_compress_ptr cinfo));
 EXTERN(void) jinit_forward_dct JPP((j_compress_ptr cinfo));
 EXTERN(void) jinit_huff_encoder JPP((j_compress_ptr cinfo));
-EXTERN(void) jinit_phuff_encoder JPP((j_compress_ptr cinfo));
+EXTERN(void) jinit_arith_encoder JPP((j_compress_ptr cinfo));
 EXTERN(void) jinit_marker_writer JPP((j_compress_ptr cinfo));
 /* Decompression module initialization routines */
 EXTERN(void) jinit_master_decompress JPP((j_decompress_ptr cinfo));
@@ -357,7 +363,7 @@ EXTERN(void) jinit_d_post_controller JPP((j_decompress_ptr cinfo,
 EXTERN(void) jinit_input_controller JPP((j_decompress_ptr cinfo));
 EXTERN(void) jinit_marker_reader JPP((j_decompress_ptr cinfo));
 EXTERN(void) jinit_huff_decoder JPP((j_decompress_ptr cinfo));
-EXTERN(void) jinit_phuff_decoder JPP((j_decompress_ptr cinfo));
+EXTERN(void) jinit_arith_decoder JPP((j_decompress_ptr cinfo));
 EXTERN(void) jinit_inverse_dct JPP((j_decompress_ptr cinfo));
 EXTERN(void) jinit_upsampler JPP((j_decompress_ptr cinfo));
 EXTERN(void) jinit_color_deconverter JPP((j_decompress_ptr cinfo));
@@ -381,6 +387,15 @@ EXTERN(void) jzero_far JPP((void FAR * target, size_t bytestozero));
 extern const int jpeg_zigzag_order[]; /* natural coef order to zigzag order */
 #endif
 extern const int jpeg_natural_order[]; /* zigzag coef order to natural order */
+extern const int jpeg_natural_order7[]; /* zz to natural order for 7x7 block */
+extern const int jpeg_natural_order6[]; /* zz to natural order for 6x6 block */
+extern const int jpeg_natural_order5[]; /* zz to natural order for 5x5 block */
+extern const int jpeg_natural_order4[]; /* zz to natural order for 4x4 block */
+extern const int jpeg_natural_order3[]; /* zz to natural order for 3x3 block */
+extern const int jpeg_natural_order2[]; /* zz to natural order for 2x2 block */
+
+/* Arithmetic coding probability estimation tables in jaricom.c */
+extern const INT32 jpeg_aritab[];
 
 /* Suppress undefined-structure complaints if necessary. */
 
diff --git a/src/jpeg-6b/jpeglib.h b/src/jpeg-8c/jpeglib.h
index 8ff305d2..1eb1fac0 100644
--- a/src/jpeg-6b/jpeglib.h
+++ b/src/jpeg-8c/jpeglib.h
@@ -2,6 +2,7 @@
  * jpeglib.h
  *
  * Copyright (C) 1991-1998, Thomas G. Lane.
+ * Modified 2002-2010 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -21,16 +22,24 @@
  */
 
 #ifndef JCONFIG_INCLUDED	/* in case jinclude.h already did */
-#include "../jpeg-6b/jconfig.h"		/* widely used configuration options */
+#include "jconfig.h"		/* widely used configuration options */
 #endif
-#include "../jpeg-6b/jmorecfg.h"		/* seldom changed options */
+#include "jmorecfg.h"		/* seldom changed options */
 
 
-/* Version ID for the JPEG library.
- * Might be useful for tests like "#if JPEG_LIB_VERSION >= 60".
+#ifdef __cplusplus
+#ifndef DONT_USE_EXTERN_C
+extern "C" {
+#endif
+#endif
+
+/* Version IDs for the JPEG library.
+ * Might be useful for tests like "#if JPEG_LIB_VERSION >= 80".
  */
 
-#define JPEG_LIB_VERSION  62	/* Version 6b */
+#define JPEG_LIB_VERSION        80	/* Compatibility version 8.0 */
+#define JPEG_LIB_VERSION_MAJOR  8
+#define JPEG_LIB_VERSION_MINOR  3
 
 
 /* Various constants determining the sizes of things.
@@ -138,18 +147,18 @@ typedef struct {
    */
   JDIMENSION width_in_blocks;
   JDIMENSION height_in_blocks;
-  /* Size of a DCT block in samples.  Always DCTSIZE for compression.
-   * For decompression this is the size of the output from one DCT block,
-   * reflecting any scaling we choose to apply during the IDCT step.
-   * Values of 1,2,4,8 are likely to be supported.  Note that different
-   * components may receive different IDCT scalings.
+  /* Size of a DCT block in samples,
+   * reflecting any scaling we choose to apply during the DCT step.
+   * Values from 1 to 16 are supported.
+   * Note that different components may receive different DCT scalings.
    */
-  int DCT_scaled_size;
+  int DCT_h_scaled_size;
+  int DCT_v_scaled_size;
   /* The downsampled dimensions are the component's actual, unpadded number
-   * of samples at the main buffer (preprocessing/compression interface), thus
-   * downsampled_width = ceil(image_width * Hi/Hmax)
-   * and similarly for height.  For decompression, IDCT scaling is included, so
-   * downsampled_width = ceil(image_width * Hi/Hmax * DCT_scaled_size/DCTSIZE)
+   * of samples at the main buffer (preprocessing/compression interface);
+   * DCT scaling is included, so
+   * downsampled_width = ceil(image_width * Hi/Hmax * DCT_h_scaled_size/DCTSIZE)
+   * and similarly for height.
    */
   JDIMENSION downsampled_width;	 /* actual width in samples */
   JDIMENSION downsampled_height; /* actual height in samples */
@@ -164,7 +173,7 @@ typedef struct {
   int MCU_width;		/* number of blocks per MCU, horizontally */
   int MCU_height;		/* number of blocks per MCU, vertically */
   int MCU_blocks;		/* MCU_width * MCU_height */
-  int MCU_sample_width;		/* MCU width in samples, MCU_width*DCT_scaled_size */
+  int MCU_sample_width;	/* MCU width in samples: MCU_width * DCT_h_scaled_size */
   int last_col_width;		/* # of non-dummy blocks across in last MCU */
   int last_row_height;		/* # of non-dummy blocks down in last MCU */
 
@@ -291,6 +300,17 @@ struct jpeg_compress_struct {
    * helper routines to simplify changing parameters.
    */
 
+  unsigned int scale_num, scale_denom; /* fraction by which to scale image */
+
+  JDIMENSION jpeg_width;	/* scaled JPEG image width */
+  JDIMENSION jpeg_height;	/* scaled JPEG image height */
+  /* Dimensions of actual JPEG image that will be written to file,
+   * derived from input dimensions by scaling factors above.
+   * These fields are computed by jpeg_start_compress().
+   * You can also use jpeg_calc_jpeg_dimensions() to determine these values
+   * in advance of calling jpeg_start_compress().
+   */
+
   int data_precision;		/* bits of precision in image data */
 
   int num_components;		/* # of color components in JPEG image */
@@ -298,14 +318,17 @@ struct jpeg_compress_struct {
 
   jpeg_component_info * comp_info;
   /* comp_info[i] describes component that appears i'th in SOF */
-  
+
   JQUANT_TBL * quant_tbl_ptrs[NUM_QUANT_TBLS];
-  /* ptrs to coefficient quantization tables, or NULL if not defined */
-  
+  int q_scale_factor[NUM_QUANT_TBLS];
+  /* ptrs to coefficient quantization tables, or NULL if not defined,
+   * and corresponding scale factors (percentage, initialized 100).
+   */
+
   JHUFF_TBL * dc_huff_tbl_ptrs[NUM_HUFF_TBLS];
   JHUFF_TBL * ac_huff_tbl_ptrs[NUM_HUFF_TBLS];
   /* ptrs to Huffman coding tables, or NULL if not defined */
-  
+
   UINT8 arith_dc_L[NUM_ARITH_TBLS]; /* L values for DC arith-coding tables */
   UINT8 arith_dc_U[NUM_ARITH_TBLS]; /* U values for DC arith-coding tables */
   UINT8 arith_ac_K[NUM_ARITH_TBLS]; /* Kx values for AC arith-coding tables */
@@ -321,6 +344,7 @@ struct jpeg_compress_struct {
   boolean arith_code;		/* TRUE=arithmetic coding, FALSE=Huffman */
   boolean optimize_coding;	/* TRUE=optimize entropy encoding parms */
   boolean CCIR601_sampling;	/* TRUE=first samples are cosited */
+  boolean do_fancy_downsampling; /* TRUE=apply fancy downsampling */
   int smoothing_factor;		/* 1..100, or 0 for no input smoothing */
   J_DCT_METHOD dct_method;	/* DCT algorithm selector */
 
@@ -364,6 +388,9 @@ struct jpeg_compress_struct {
   int max_h_samp_factor;	/* largest h_samp_factor */
   int max_v_samp_factor;	/* largest v_samp_factor */
 
+  int min_DCT_h_scaled_size;	/* smallest DCT_h_scaled_size of any component */
+  int min_DCT_v_scaled_size;	/* smallest DCT_v_scaled_size of any component */
+
   JDIMENSION total_iMCU_rows;	/* # of iMCU rows to be input to coef ctlr */
   /* The coefficient controller receives data in units of MCU rows as defined
    * for fully interleaved scans (whether the JPEG file is interleaved or not).
@@ -389,6 +416,10 @@ struct jpeg_compress_struct {
 
   int Ss, Se, Ah, Al;		/* progressive JPEG parameters for scan */
 
+  int block_size;		/* the basic DCT block size: 1..16 */
+  const int * natural_order;	/* natural-order position array */
+  int lim_Se;			/* min( Se, DCTSIZE2-1 ) */
+
   /*
    * Links to compression subobjects (methods and private variables of modules)
    */
@@ -535,6 +566,7 @@ struct jpeg_decompress_struct {
   jpeg_component_info * comp_info;
   /* comp_info[i] describes component that appears i'th in SOF */
 
+  boolean is_baseline;		/* TRUE if Baseline SOF0 encountered */
   boolean progressive_mode;	/* TRUE if SOFn specifies progressive mode */
   boolean arith_code;		/* TRUE=arithmetic coding, FALSE=Huffman */
 
@@ -575,7 +607,8 @@ struct jpeg_decompress_struct {
   int max_h_samp_factor;	/* largest h_samp_factor */
   int max_v_samp_factor;	/* largest v_samp_factor */
 
-  int min_DCT_scaled_size;	/* smallest DCT_scaled_size of any component */
+  int min_DCT_h_scaled_size;	/* smallest DCT_h_scaled_size of any component */
+  int min_DCT_v_scaled_size;	/* smallest DCT_v_scaled_size of any component */
 
   JDIMENSION total_iMCU_rows;	/* # of iMCU rows in image */
   /* The coefficient controller's input and output progress is measured in
@@ -583,7 +616,7 @@ struct jpeg_decompress_struct {
    * in fully interleaved JPEG scans, but are used whether the scan is
    * interleaved or not.  We define an iMCU row as v_samp_factor DCT block
    * rows of each component.  Therefore, the IDCT output contains
-   * v_samp_factor*DCT_scaled_size sample rows of a component per iMCU row.
+   * v_samp_factor*DCT_v_scaled_size sample rows of a component per iMCU row.
    */
 
   JSAMPLE * sample_range_limit; /* table for fast range-limiting */
@@ -607,6 +640,12 @@ struct jpeg_decompress_struct {
 
   int Ss, Se, Ah, Al;		/* progressive JPEG parameters for scan */
 
+  /* These fields are derived from Se of first SOS marker.
+   */
+  int block_size;		/* the basic DCT block size: 1..16 */
+  const int * natural_order; /* natural-order position array for entropy decode */
+  int lim_Se;			/* min( Se, DCTSIZE2-1 ) for entropy decode */
+
   /* This field is shared between entropy decoder and marker parser.
    * It is either zero or the code of a JPEG marker that has been
    * read from the data source, but has not yet been processed.
@@ -835,12 +874,15 @@ typedef JMETHOD(boolean, jpeg_marker_parser_method, (j_decompress_ptr cinfo));
 #define jpeg_destroy_compress	jDestCompress
 #define jpeg_destroy_decompress	jDestDecompress
 #define jpeg_stdio_dest		jStdDest
+#define jpeg_stdio_src		jStdSrc
+#define jpeg_mem_dest		jMemDest
 #define jpeg_mem_src		jMemSrc
 #define jpeg_set_defaults	jSetDefaults
 #define jpeg_set_colorspace	jSetColorspace
 #define jpeg_default_colorspace	jDefColorspace
 #define jpeg_set_quality	jSetQuality
 #define jpeg_set_linear_quality	jSetLQuality
+#define jpeg_default_qtables	jDefQTables
 #define jpeg_add_quant_table	jAddQuantTable
 #define jpeg_quality_scaling	jQualityScaling
 #define jpeg_simple_progression	jSimProgress
@@ -850,6 +892,7 @@ typedef JMETHOD(boolean, jpeg_marker_parser_method, (j_decompress_ptr cinfo));
 #define jpeg_start_compress	jStrtCompress
 #define jpeg_write_scanlines	jWrtScanlines
 #define jpeg_finish_compress	jFinCompress
+#define jpeg_calc_jpeg_dimensions	jCjpegDimensions
 #define jpeg_write_raw_data	jWrtRawData
 #define jpeg_write_marker	jWrtMarker
 #define jpeg_write_m_header	jWrtMHeader
@@ -866,6 +909,7 @@ typedef JMETHOD(boolean, jpeg_marker_parser_method, (j_decompress_ptr cinfo));
 #define jpeg_input_complete	jInComplete
 #define jpeg_new_colormap	jNewCMap
 #define jpeg_consume_input	jConsumeInput
+#define jpeg_core_output_dimensions	jCoreDimensions
 #define jpeg_calc_output_dimensions	jCalcDimensions
 #define jpeg_save_markers	jSaveMarkers
 #define jpeg_set_marker_processor	jSetMarker
@@ -908,7 +952,15 @@ EXTERN(void) jpeg_destroy_decompress JPP((j_decompress_ptr cinfo));
 /* Standard data source and destination managers: stdio streams. */
 /* Caller is responsible for opening the file before and closing after. */
 EXTERN(void) jpeg_stdio_dest JPP((j_compress_ptr cinfo, FILE * outfile));
-EXTERN(void) jpeg_mem_src JPP((j_decompress_ptr cinfo, unsigned char *inbuf, size_t size));
+EXTERN(void) jpeg_stdio_src JPP((j_decompress_ptr cinfo, FILE * infile));
+
+/* Data source and destination managers: memory buffers. */
+EXTERN(void) jpeg_mem_dest JPP((j_compress_ptr cinfo,
+			       unsigned char ** outbuffer,
+			       unsigned long * outsize));
+EXTERN(void) jpeg_mem_src JPP((j_decompress_ptr cinfo,
+			      unsigned char * inbuffer,
+			      unsigned long insize));
 
 /* Default parameter setup for compression */
 EXTERN(void) jpeg_set_defaults JPP((j_compress_ptr cinfo));
@@ -921,6 +973,8 @@ EXTERN(void) jpeg_set_quality JPP((j_compress_ptr cinfo, int quality,
 EXTERN(void) jpeg_set_linear_quality JPP((j_compress_ptr cinfo,
 					  int scale_factor,
 					  boolean force_baseline));
+EXTERN(void) jpeg_default_qtables JPP((j_compress_ptr cinfo,
+				       boolean force_baseline));
 EXTERN(void) jpeg_add_quant_table JPP((j_compress_ptr cinfo, int which_tbl,
 				       const unsigned int *basic_table,
 				       int scale_factor,
@@ -940,12 +994,15 @@ EXTERN(JDIMENSION) jpeg_write_scanlines JPP((j_compress_ptr cinfo,
 					     JDIMENSION num_lines));
 EXTERN(void) jpeg_finish_compress JPP((j_compress_ptr cinfo));
 
+/* Precalculate JPEG dimensions for current compression parameters. */
+EXTERN(void) jpeg_calc_jpeg_dimensions JPP((j_compress_ptr cinfo));
+
 /* Replaces jpeg_write_scanlines when writing raw downsampled data. */
 EXTERN(JDIMENSION) jpeg_write_raw_data JPP((j_compress_ptr cinfo,
 					    JSAMPIMAGE data,
 					    JDIMENSION num_lines));
 
-/* Write a special marker.  See libjpeg.doc concerning safe usage. */
+/* Write a special marker.  See libjpeg.txt concerning safe usage. */
 EXTERN(void) jpeg_write_marker
 	JPP((j_compress_ptr cinfo, int marker,
 	     const JOCTET * dataptr, unsigned int datalen));
@@ -999,6 +1056,7 @@ EXTERN(int) jpeg_consume_input JPP((j_decompress_ptr cinfo));
 #define JPEG_SCAN_COMPLETED	4 /* Completed last iMCU row of a scan */
 
 /* Precalculate output dimensions for current decompression parameters. */
+EXTERN(void) jpeg_core_output_dimensions JPP((j_decompress_ptr cinfo));
 EXTERN(void) jpeg_calc_output_dimensions JPP((j_decompress_ptr cinfo));
 
 /* Control saving of COM and APPn markers into marker_list. */
@@ -1089,8 +1147,14 @@ struct jpeg_color_quantizer { long dummy; };
  */
 
 #ifdef JPEG_INTERNALS
-#include "../jpeg-6b/jpegint.h"		/* fetch private declarations */
-#include "../jpeg-6b/jerror.h"		/* fetch error codes too */
+#include "jpegint.h"		/* fetch private declarations */
+#include "jerror.h"		/* fetch error codes too */
+#endif
+
+#ifdef __cplusplus
+#ifndef DONT_USE_EXTERN_C
+}
+#endif
 #endif
 
 #endif /* JPEGLIB_H */
diff --git a/src/jpeg-6b/jquant1.c b/src/jpeg-8c/jquant1.c
index b2f96aa1..b2f96aa1 100644
--- a/src/jpeg-6b/jquant1.c
+++ b/src/jpeg-8c/jquant1.c
diff --git a/src/jpeg-6b/jquant2.c b/src/jpeg-8c/jquant2.c
index af601e33..af601e33 100644
--- a/src/jpeg-6b/jquant2.c
+++ b/src/jpeg-8c/jquant2.c
diff --git a/src/jpeg-6b/jutils.c b/src/jpeg-8c/jutils.c
index d18a9555..04351797 100644
--- a/src/jpeg-6b/jutils.c
+++ b/src/jpeg-8c/jutils.c
@@ -2,6 +2,7 @@
  * jutils.c
  *
  * Copyright (C) 1991-1996, Thomas G. Lane.
+ * Modified 2009 by Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -63,6 +64,57 @@ const int jpeg_natural_order[DCTSIZE2+16] = {
  63, 63, 63, 63, 63, 63, 63, 63
 };
 
+const int jpeg_natural_order7[7*7+16] = {
+  0,  1,  8, 16,  9,  2,  3, 10,
+ 17, 24, 32, 25, 18, 11,  4,  5,
+ 12, 19, 26, 33, 40, 48, 41, 34,
+ 27, 20, 13,  6, 14, 21, 28, 35,
+ 42, 49, 50, 43, 36, 29, 22, 30,
+ 37, 44, 51, 52, 45, 38, 46, 53,
+ 54,
+ 63, 63, 63, 63, 63, 63, 63, 63, /* extra entries for safety in decoder */
+ 63, 63, 63, 63, 63, 63, 63, 63
+};
+
+const int jpeg_natural_order6[6*6+16] = {
+  0,  1,  8, 16,  9,  2,  3, 10,
+ 17, 24, 32, 25, 18, 11,  4,  5,
+ 12, 19, 26, 33, 40, 41, 34, 27,
+ 20, 13, 21, 28, 35, 42, 43, 36,
+ 29, 37, 44, 45,
+ 63, 63, 63, 63, 63, 63, 63, 63, /* extra entries for safety in decoder */
+ 63, 63, 63, 63, 63, 63, 63, 63
+};
+
+const int jpeg_natural_order5[5*5+16] = {
+  0,  1,  8, 16,  9,  2,  3, 10,
+ 17, 24, 32, 25, 18, 11,  4, 12,
+ 19, 26, 33, 34, 27, 20, 28, 35,
+ 36,
+ 63, 63, 63, 63, 63, 63, 63, 63, /* extra entries for safety in decoder */
+ 63, 63, 63, 63, 63, 63, 63, 63
+};
+
+const int jpeg_natural_order4[4*4+16] = {
+  0,  1,  8, 16,  9,  2,  3, 10,
+ 17, 24, 25, 18, 11, 19, 26, 27,
+ 63, 63, 63, 63, 63, 63, 63, 63, /* extra entries for safety in decoder */
+ 63, 63, 63, 63, 63, 63, 63, 63
+};
+
+const int jpeg_natural_order3[3*3+16] = {
+  0,  1,  8, 16,  9,  2, 10, 17,
+ 18,
+ 63, 63, 63, 63, 63, 63, 63, 63, /* extra entries for safety in decoder */
+ 63, 63, 63, 63, 63, 63, 63, 63
+};
+
+const int jpeg_natural_order2[2*2+16] = {
+  0,  1,  8,  9,
+ 63, 63, 63, 63, 63, 63, 63, 63, /* extra entries for safety in decoder */
+ 63, 63, 63, 63, 63, 63, 63, 63
+};
+
 
 /*
  * Arithmetic utilities
diff --git a/src/jpeg-6b/jversion.h b/src/jpeg-8c/jversion.h
index 6472c58d..e868538c 100644
--- a/src/jpeg-6b/jversion.h
+++ b/src/jpeg-8c/jversion.h
@@ -1,7 +1,7 @@
 /*
  * jversion.h
  *
- * Copyright (C) 1991-1998, Thomas G. Lane.
+ * Copyright (C) 1991-2011, Thomas G. Lane, Guido Vollbeding.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -9,6 +9,6 @@
  */
 
 
-#define JVERSION	"6b  27-Mar-1998"
+#define JVERSION	"8c  16-Jan-2011"
 
-#define JCOPYRIGHT	"Copyright (C) 1998, Thomas G. Lane"
+#define JCOPYRIGHT	"Copyright (C) 2011, Thomas G. Lane, Guido Vollbeding"
diff --git a/src/null/mac_net.c b/src/null/mac_net.c
index e0551cb3..cca208bf 100644
--- a/src/null/mac_net.c
+++ b/src/null/mac_net.c
@@ -53,14 +53,3 @@ Sys_SendPacket
 */
 void Sys_SendPacket( int length, void *data, netadr_t to ) {
 }
-
-/*
-==================
-Sys_GetPacket
-
-Never called by the game logic, just the system event queing
-==================
-*/
-qboolean	Sys_GetPacket ( netadr_t *net_from, msg_t *net_message ) {
-	return false;
-}
diff --git a/src/null/null_net.c b/src/null/null_net.c
index d8392490..9562eacf 100644
--- a/src/null/null_net.c
+++ b/src/null/null_net.c
@@ -52,14 +52,3 @@ Sys_SendPacket
 */
 void Sys_SendPacket( int length, void *data, netadr_t to ) {
 }
-
-/*
-==================
-Sys_GetPacket
-
-Never called by the game logic, just the system event queing
-==================
-*/
-qboolean	Sys_GetPacket ( netadr_t *net_from, msg_t *net_message ) {
-	return false;
-}
diff --git a/src/qcommon/cmd.c b/src/qcommon/cmd.c
index 7ce50ccc..8ea761e8 100644
--- a/src/qcommon/cmd.c
+++ b/src/qcommon/cmd.c
@@ -470,6 +470,31 @@ char *Cmd_Cmd(void)
 }
 
 /*
+   Replace command separators with space to prevent interpretation
+   This is a hack to protect buggy qvms
+   https://bugzilla.icculus.org/show_bug.cgi?id=3593
+   https://bugzilla.icculus.org/show_bug.cgi?id=4769
+*/
+
+void Cmd_Args_Sanitize(void)
+{
+	int i;
+
+	for(i = 1; i < cmd.argc; i++)
+	{
+		char *c = cmd.argv[i];
+		
+		if(strlen(c) > MAX_CVAR_VALUE_STRING - 1)
+			c[MAX_CVAR_VALUE_STRING - 1] = '\0';
+		
+		while ((c = strpbrk(c, "\n\r;"))) {
+			*c = ' ';
+			++c;
+		}
+	}
+}
+
+/*
 ============
 Cmd_TokenizeString
 
@@ -604,6 +629,20 @@ void Cmd_TokenizeStringIgnoreQuotes( const char *text_in ) {
 
 /*
 ============
+Cmd_FindCommand
+============
+*/
+cmd_function_t *Cmd_FindCommand( const char *cmd_name )
+{
+	cmd_function_t *cmd;
+	for( cmd = cmd_functions; cmd; cmd = cmd->next )
+		if( !Q_stricmp( cmd_name, cmd->name ) )
+			return cmd;
+	return NULL;
+}
+
+/*
+============
 Cmd_AddCommand
 ============
 */
@@ -611,14 +650,12 @@ void	Cmd_AddCommand( const char *cmd_name, xcommand_t function ) {
 	cmd_function_t	*cmd;
 	
 	// fail if the command already exists
-	for ( cmd = cmd_functions ; cmd ; cmd=cmd->next ) {
-		if ( !strcmp( cmd_name, cmd->name ) ) {
-			// allow completion-only commands to be silently doubled
-			if ( function != NULL ) {
-				Com_Printf ("Cmd_AddCommand: %s already defined\n", cmd_name);
-			}
-			return;
-		}
+	if( Cmd_FindCommand( cmd_name ) )
+	{
+		// allow completion-only commands to be silently doubled
+		if( function != NULL )
+			Com_Printf( "Cmd_AddCommand: %s already defined\n", cmd_name );
+		return;
 	}
 
 	// use a small malloc to avoid zone fragmentation
@@ -672,6 +709,28 @@ void	Cmd_RemoveCommand( const char *cmd_name ) {
 	}
 }
 
+/*
+============
+Cmd_RemoveCommandSafe
+
+Only remove commands with no associated function
+============
+*/
+void Cmd_RemoveCommandSafe( const char *cmd_name )
+{
+	cmd_function_t *cmd = Cmd_FindCommand( cmd_name );
+
+	if( !cmd )
+		return;
+	if( cmd->function )
+	{
+		Com_Error( ERR_DROP, "Restricted source tried to remove "
+			"system command \"%s\"\n", cmd_name );
+		return;
+	}
+
+	Cmd_RemoveCommand( cmd_name );
+}
 
 /*
 ============
@@ -798,7 +857,7 @@ Cmd_CompleteCfgName
 */
 void Cmd_CompleteCfgName( char *args, int argNum ) {
 	if( argNum == 2 ) {
-		Field_CompleteFilename( "", "cfg", qfalse );
+		Field_CompleteFilename( "", "cfg", qfalse, qtrue );
 	}
 }
 
diff --git a/src/qcommon/common.c b/src/qcommon/common.c
index 38902e92..1cf54d07 100644
--- a/src/qcommon/common.c
+++ b/src/qcommon/common.c
@@ -51,6 +51,7 @@ jmp_buf abortframe;		// an ERR_DROP occured, exit the entire frame
 
 
 FILE *debuglogfile;
+static fileHandle_t pipefile;
 static fileHandle_t logfile;
 fileHandle_t	com_journalFile;			// events are written here
 fileHandle_t	com_journalDataFile;		// config files are written here
@@ -60,7 +61,6 @@ cvar_t	*com_developer;
 cvar_t	*com_dedicated;
 cvar_t	*com_timescale;
 cvar_t	*com_fixedtime;
-cvar_t	*com_dropsim;		// 0.0 to 1.0, simulated packet drops
 cvar_t	*com_journal;
 cvar_t	*com_maxfps;
 cvar_t	*com_altivec;
@@ -68,6 +68,7 @@ cvar_t	*com_timedemo;
 cvar_t	*com_sv_running;
 cvar_t	*com_cl_running;
 cvar_t	*com_logfile;		// 1 = buffer log, 2 = flush after each print
+cvar_t	*com_pipefile;
 cvar_t	*com_showtrace;
 cvar_t	*com_version;
 cvar_t	*com_blood;
@@ -83,6 +84,8 @@ cvar_t	*com_maxfpsUnfocused;
 cvar_t	*com_minimized;
 cvar_t	*com_maxfpsMinimized;
 cvar_t	*com_abnormalExit;
+cvar_t  *com_homepath;
+cvar_t	*com_busyWait;
 
 // com_speeds times
 int		time_game;
@@ -90,7 +93,6 @@ int		time_frontend;		// renderer frontend time
 int		time_backend;		// renderer backend time
 
 int			com_frameTime;
-int			com_frameMsec;
 int			com_frameNumber;
 
 qboolean	com_errorEntered = qfalse;
@@ -289,9 +291,9 @@ void QDECL Com_Error( int code, const char *fmt, ... ) {
 		Cvar_Set("com_errorMessage", com_errorMessage);
 
 	if (code == ERR_DISCONNECT || code == ERR_SERVERDISCONNECT) {
+		VM_Forced_Unload_Start();
 		SV_Shutdown( "Server disconnected" );
 		CL_Disconnect( qtrue );
-		VM_Forced_Unload_Start();
 		CL_FlushMemory( );
 		VM_Forced_Unload_Done();
 		// make sure we can get at our local stuff
@@ -300,32 +302,36 @@ void QDECL Com_Error( int code, const char *fmt, ... ) {
 		longjmp (abortframe, -1);
 	} else if (code == ERR_DROP) {
 		Com_Printf ("********************\nERROR: %s\n********************\n", com_errorMessage);
+		VM_Forced_Unload_Start();
 		SV_Shutdown (va("Server crashed: %s",  com_errorMessage));
 		CL_Disconnect( qtrue );
-		VM_Forced_Unload_Start();
 		CL_FlushMemory( );
 		VM_Forced_Unload_Done();
 		FS_PureServerSetLoadedPaks("", "");
 		com_errorEntered = qfalse;
 		longjmp (abortframe, -1);
 	} else if ( code == ERR_NEED_CD ) {
+		VM_Forced_Unload_Start();
 		SV_Shutdown( "Server didn't have CD" );
 		if ( com_cl_running && com_cl_running->integer ) {
 			CL_Disconnect( qtrue );
-			VM_Forced_Unload_Start();
 			CL_FlushMemory( );
 			VM_Forced_Unload_Done();
 			CL_CDDialog();
 		} else {
 			Com_Printf("Server didn't have CD\n" );
+			VM_Forced_Unload_Done();
 		}
+
 		FS_PureServerSetLoadedPaks("", "");
 
 		com_errorEntered = qfalse;
 		longjmp (abortframe, -1);
 	} else {
+		VM_Forced_Unload_Start();
 		CL_Shutdown (va("Client fatal crashed: %s", com_errorMessage));
 		SV_Shutdown (va("Server fatal crashed: %s", com_errorMessage));
+		VM_Forced_Unload_Done();
 	}
 
 	Com_Shutdown ();
@@ -446,7 +452,6 @@ be after execing the config and default.
 void Com_StartupVariable( const char *match ) {
 	int		i;
 	char	*s;
-	cvar_t	*cv;
 
 	for (i=0 ; i < com_numConsoleLines ; i++) {
 		Cmd_TokenizeString( com_consoleLines[i] );
@@ -455,11 +460,13 @@ void Com_StartupVariable( const char *match ) {
 		}
 
 		s = Cmd_Argv(1);
-		if ( !match || !strcmp( s, match ) ) {
-			Cvar_Set( s, Cmd_Argv(2) );
-			cv = Cvar_Get( s, "", 0 );
-			cv->flags |= CVAR_USER_CREATED;
-//			com_consoleLines[i] = 0;
+		
+		if(!match || !strcmp(s, match))
+		{
+			if(Cvar_Flags(s) == CVAR_NONEXISTENT)
+				Cvar_Get(s, Cmd_Argv(2), CVAR_USER_CREATED);
+			else
+				Cvar_Set(s, Cmd_Argv(2));
 		}
 	}
 }
@@ -504,8 +511,8 @@ qboolean Com_AddStartupCommands( void ) {
 //============================================================================
 
 void Info_Print( const char *s ) {
-	char	key[512];
-	char	value[512];
+	char	key[BIG_INFO_KEY];
+	char	value[BIG_INFO_VALUE];
 	char	*o;
 	int		l;
 
@@ -525,7 +532,7 @@ void Info_Print( const char *s ) {
 		}
 		else
 			*o = 0;
-		Com_Printf ("%s", key);
+		Com_Printf ("%s ", key);
 
 		if (!*s)
 		{
@@ -1946,7 +1953,6 @@ EVENT LOOP
 static sysEvent_t  eventQueue[ MAX_QUEUED_EVENTS ];
 static int         eventHead = 0;
 static int         eventTail = 0;
-static byte        sys_packetReceived[ MAX_MSGLEN ];
 
 /*
 ================
@@ -1999,8 +2005,6 @@ sysEvent_t Com_GetSystemEvent( void )
 {
 	sysEvent_t  ev;
 	char        *s;
-	msg_t       netmsg;
-	netadr_t    adr;
 
 	// return if we have data
 	if ( eventHead > eventTail )
@@ -2022,21 +2026,6 @@ sysEvent_t Com_GetSystemEvent( void )
 		Com_QueueEvent( 0, SE_CONSOLE, 0, 0, len, b );
 	}
 
-	// check for network packets
-	MSG_Init( &netmsg, sys_packetReceived, sizeof( sys_packetReceived ) );
-	if ( Sys_GetPacket ( &adr, &netmsg ) )
-	{
-		netadr_t  *buf;
-		int       len;
-
-		// copy out to a seperate buffer for qeueing
-		len = sizeof( netadr_t ) + netmsg.cursize;
-		buf = Z_Malloc( len );
-		*buf = adr;
-		memcpy( buf+1, netmsg.data, netmsg.cursize );
-		Com_QueueEvent( 0, SE_PACKET, 0, 0, len, buf );
-	}
-
 	// return if we have data
 	if ( eventHead > eventTail )
 	{
@@ -2196,7 +2185,6 @@ int Com_EventLoop( void ) {
 	MSG_Init( &buf, bufData, sizeof( bufData ) );
 
 	while ( 1 ) {
-		NET_FlushPacketQueue();
 		ev = Com_GetEvent();
 
 		// if no more events are available
@@ -2217,57 +2205,26 @@ int Com_EventLoop( void ) {
 		}
 
 
-		switch ( ev.evType ) {
-		default:
-			Com_Error( ERR_FATAL, "Com_EventLoop: bad event type %i", ev.evType );
-			break;
-        case SE_NONE:
-            break;
-		case SE_KEY:
-			CL_KeyEvent( ev.evValue, ev.evValue2, ev.evTime );
+		switch(ev.evType)
+		{
+			case SE_KEY:
+				CL_KeyEvent( ev.evValue, ev.evValue2, ev.evTime );
 			break;
-		case SE_CHAR:
-			CL_CharEvent( ev.evValue );
+			case SE_CHAR:
+				CL_CharEvent( ev.evValue );
 			break;
-		case SE_MOUSE:
-			CL_MouseEvent( ev.evValue, ev.evValue2, ev.evTime );
+			case SE_MOUSE:
+				CL_MouseEvent( ev.evValue, ev.evValue2, ev.evTime );
 			break;
-		case SE_JOYSTICK_AXIS:
-			CL_JoystickEvent( ev.evValue, ev.evValue2, ev.evTime );
+			case SE_JOYSTICK_AXIS:
+				CL_JoystickEvent( ev.evValue, ev.evValue2, ev.evTime );
 			break;
-		case SE_CONSOLE:
-			Cbuf_AddText( (char *)ev.evPtr );
-			Cbuf_AddText( "\n" );
+			case SE_CONSOLE:
+				Cbuf_AddText( (char *)ev.evPtr );
+				Cbuf_AddText( "\n" );
 			break;
-		case SE_PACKET:
-			// this cvar allows simulation of connections that
-			// drop a lot of packets.  Note that loopback connections
-			// don't go through here at all.
-			if ( com_dropsim->value > 0 ) {
-				static int seed;
-
-				if ( Q_random( &seed ) < com_dropsim->value ) {
-					break;		// drop this packet
-				}
-			}
-
-			evFrom = *(netadr_t *)ev.evPtr;
-			buf.cursize = ev.evPtrLength - sizeof( evFrom );
-
-			// we must copy the contents of the message out, because
-			// the event buffers are only large enough to hold the
-			// exact payload, but channel messages need to be large
-			// enough to hold fragment reassembly
-			if ( (unsigned)buf.cursize > buf.maxsize ) {
-				Com_Printf("Com_EventLoop: oversize packet\n");
-				continue;
-			}
-			Com_Memcpy( buf.data, (byte *)((netadr_t *)ev.evPtr + 1), buf.cursize );
-			if ( com_sv_running->integer ) {
-				Com_RunAndTimeServerPacket( &evFrom, &buf );
-			} else {
-				CL_PacketEvent( evFrom, &buf );
-			}
+			default:
+				Com_Error( ERR_FATAL, "Com_EventLoop: bad event type %i", ev.evType );
 			break;
 		}
 
@@ -2518,7 +2475,6 @@ void Com_Init( char *commandLine ) {
 
 	// Clear queues
 	Com_Memset( &eventQueue[ 0 ], 0, MAX_QUEUED_EVENTS * sizeof( sysEvent_t ) );
-	Com_Memset( &sys_packetReceived[ 0 ], 0, MAX_MSGLEN * sizeof( byte ) );
 
 	// initialize the weak pseudo-random number generator for use later.
 	Com_InitRand();
@@ -2543,11 +2499,14 @@ void Com_Init( char *commandLine ) {
 	Cmd_Init ();
 
 	// get the developer cvar set as early as possible
-	Com_StartupVariable( "developer" );
+	com_developer = Cvar_Get("developer", "0", CVAR_TEMP);
 
 	// done early so bind command exists
 	CL_InitKeyCommands();
 
+	com_homepath = Cvar_Get("com_homepath", "", CVAR_INIT);
+	
+	// Com_StartupVariable(
 	FS_InitFilesystem ();
 
 	Com_InitJournaling();
@@ -2593,13 +2552,11 @@ void Com_Init( char *commandLine ) {
 	com_maxfps = Cvar_Get ("com_maxfps", "85", CVAR_ARCHIVE);
 	com_blood = Cvar_Get ("com_blood", "1", CVAR_ARCHIVE);
 
-	com_developer = Cvar_Get ("developer", "0", CVAR_TEMP );
 	com_logfile = Cvar_Get ("logfile", "0", CVAR_TEMP );
 
 	com_timescale = Cvar_Get ("timescale", "1", CVAR_CHEAT | CVAR_SYSTEMINFO );
 	com_fixedtime = Cvar_Get ("fixedtime", "0", CVAR_CHEAT);
 	com_showtrace = Cvar_Get ("com_showtrace", "0", CVAR_CHEAT);
-	com_dropsim = Cvar_Get ("com_dropsim", "0", CVAR_CHEAT);
 	com_speeds = Cvar_Get ("com_speeds", "0", 0);
 	com_timedemo = Cvar_Get ("timedemo", "0", CVAR_CHEAT);
 	com_cameraMode = Cvar_Get ("com_cameraMode", "0", CVAR_CHEAT);
@@ -2618,6 +2575,7 @@ void Com_Init( char *commandLine ) {
 	com_minimized = Cvar_Get( "com_minimized", "0", CVAR_ROM );
 	com_maxfpsMinimized = Cvar_Get( "com_maxfpsMinimized", "0", CVAR_ARCHIVE );
 	com_abnormalExit = Cvar_Get( "com_abnormalExit", "0", CVAR_ROM );
+	com_busyWait = Cvar_Get("com_busyWait", "0", CVAR_ARCHIVE);
 
 	s = va("%s %s %s", Q3_VERSION, PLATFORM_STRING, __DATE__ );
 	com_version = Cvar_Get ("version", s, CVAR_ROM | CVAR_SERVERINFO );
@@ -2677,9 +2635,36 @@ void Com_Init( char *commandLine ) {
 	Com_Printf ("Altivec support is %s\n", com_altivec->integer ? "enabled" : "disabled");
 #endif
 
+	com_pipefile = Cvar_Get( "com_pipefile", "", CVAR_ARCHIVE|CVAR_LATCH );
+	if( com_pipefile->string[0] )
+	{
+		pipefile = FS_FCreateOpenPipeFile( com_pipefile->string );
+	}
+
 	Com_Printf ("--- Common Initialization Complete ---\n");
 }
 
+/*
+===============
+Com_ReadFromPipe
+
+Read whatever is in com_pipefile, if anything, and execute it
+===============
+*/
+void Com_ReadFromPipe( void )
+{
+	char buffer[MAX_STRING_CHARS] = {""};
+	qboolean read;
+
+	if( !pipefile )
+		return;
+
+	read = FS_Read( buffer, sizeof( buffer ), pipefile );
+	if( read )
+		Cbuf_ExecuteText( EXEC_APPEND, buffer );
+}
+
+
 //==================================================================
 
 void Com_WriteConfigToFile( const char *filename ) {
@@ -2801,19 +2786,16 @@ Com_Frame
 void Com_Frame( void ) {
 
 	int		msec, minMsec;
-	static int	lastTime;
-	int key;
+	int		timeVal;
+	static int	lastTime = 0, bias = 0;
  
 	int		timeBeforeFirstEvents;
-	int           timeBeforeServer;
-	int           timeBeforeEvents;
-	int           timeBeforeClient;
-	int           timeAfter;
+	int		timeBeforeServer;
+	int		timeBeforeEvents;
+	int		timeBeforeClient;
+	int		timeAfter;
   
 
-
-
-
 	if ( setjmp (abortframe) ) {
 		return;			// an ERR_DROP was thrown
 	}
@@ -2824,10 +2806,6 @@ void Com_Frame( void ) {
 	timeBeforeClient = 0;
 	timeAfter = 0;
 
-
-	// old net chan encryption key
-	key = 0x87243987;
-
 	// write config file if anything changed
 	Com_WriteConfiguration(); 
 
@@ -2838,37 +2816,59 @@ void Com_Frame( void ) {
 		timeBeforeFirstEvents = Sys_Milliseconds ();
 	}
 
-	// we may want to spin here if things are going too fast
-	if ( !com_dedicated->integer && !com_timedemo->integer ) {
-		if( com_minimized->integer && com_maxfpsMinimized->integer > 0 ) {
-			minMsec = 1000 / com_maxfpsMinimized->integer;
-		} else if( com_unfocused->integer && com_maxfpsUnfocused->integer > 0 ) {
-			minMsec = 1000 / com_maxfpsUnfocused->integer;
-		} else if( com_maxfps->integer > 0 ) {
-			minMsec = 1000 / com_maxfps->integer;
-		} else {
-			minMsec = 1;
+	// Figure out how much time we have
+	if(!com_timedemo->integer)
+	{
+		if(com_dedicated->integer)
+			minMsec = SV_FrameMsec();
+		else
+		{
+			if(com_minimized->integer && com_maxfpsMinimized->integer > 0)
+				minMsec = 1000 / com_maxfpsMinimized->integer;
+			else if(com_unfocused->integer && com_maxfpsUnfocused->integer > 0)
+				minMsec = 1000 / com_maxfpsUnfocused->integer;
+			else if(com_maxfps->integer > 0)
+				minMsec = 1000 / com_maxfps->integer;
+			else
+				minMsec = 1;
+			
+			timeVal = com_frameTime - lastTime;
+			bias += timeVal - minMsec;
+			
+			if(bias > minMsec)
+				bias = minMsec;
+			
+			// Adjust minMsec if previous frame took too long to render so
+			// that framerate is stable at the requested value.
+			minMsec -= bias;
 		}
-	} else {
-		minMsec = 1;
 	}
+	else
+		minMsec = 1;
 
-	msec = minMsec;
-	do {
-		int timeRemaining = minMsec - msec;
+	timeVal = 0;
+	do
+	{
+		// Busy sleep the last millisecond for better timeout precision
+		if(com_busyWait->integer || timeVal < 2)
+			NET_Sleep(0);
+		else
+			NET_Sleep(timeVal - 1);
 
-		// The existing Sys_Sleep implementations aren't really
-		// precise enough to be of use beyond 100fps
-		// FIXME: implement a more precise sleep (RDTSC or something)
-		if( timeRemaining >= 10 )
-			Sys_Sleep( timeRemaining );
+		msec = Sys_Milliseconds() - com_frameTime;
+		
+		if(msec >= minMsec)
+			timeVal = 0;
+		else
+			timeVal = minMsec - msec;
+
+	} while(timeVal > 0);
+	
+	lastTime = com_frameTime;
+	com_frameTime = Com_EventLoop();
+	
+	msec = com_frameTime - lastTime;
 
-		com_frameTime = Com_EventLoop();
-		if ( lastTime > com_frameTime ) {
-			lastTime = com_frameTime;		// possible on first frame
-		}
-		msec = com_frameTime - lastTime;
-	} while ( msec < minMsec );
 	Cbuf_Execute ();
 
 	if (com_altivec->modified)
@@ -2877,11 +2877,8 @@ void Com_Frame( void ) {
 		com_altivec->modified = qfalse;
 	}
 
-	lastTime = com_frameTime;
-
 	// mess with msec if needed
-	com_frameMsec = msec;
-	msec = Com_ModifyMsec( msec );
+	msec = Com_ModifyMsec(msec);
 
 	//
 	// server side
@@ -2941,6 +2938,9 @@ void Com_Frame( void ) {
 	}
 #endif
 
+
+	NET_FlushPacketQueue();
+
 	//
 	// report timing information
 	//
@@ -2974,8 +2974,7 @@ void Com_Frame( void ) {
 		c_pointcontents = 0;
 	}
 
-	// old net chan encryption key
-	key = lastTime * 0x87243987;
+	Com_ReadFromPipe( );
 
 	com_frameNumber++;
 }
@@ -2996,6 +2995,11 @@ void Com_Shutdown (void) {
 		com_journalFile = 0;
 	}
 
+	if( pipefile ) {
+		FS_FCloseFile( pipefile );
+		FS_HomeRemove( com_pipefile->string );
+	}
+
 }
 
 //------------------------------------------------------------------------
@@ -3183,15 +3187,15 @@ Field_CompleteFilename
 ===============
 */
 void Field_CompleteFilename( const char *dir,
-		const char *ext, qboolean stripExt )
+		const char *ext, qboolean stripExt, qboolean allowNonPureFilesOnDisk )
 {
 	matchCount = 0;
 	shortestMatch[ 0 ] = 0;
 
-	FS_FilenameCompletion( dir, ext, stripExt, FindMatches );
+	FS_FilenameCompletion( dir, ext, stripExt, FindMatches, allowNonPureFilesOnDisk );
 
 	if( !Field_Complete( ) )
-		FS_FilenameCompletion( dir, ext, stripExt, PrintMatches );
+		FS_FilenameCompletion( dir, ext, stripExt, PrintMatches, allowNonPureFilesOnDisk );
 }
 
 /*
diff --git a/src/qcommon/cvar.c b/src/qcommon/cvar.c
index b0a21506..8db30e6a 100644
--- a/src/qcommon/cvar.c
+++ b/src/qcommon/cvar.c
@@ -364,6 +364,18 @@ cvar_t *Cvar_Get( const char *var_name, const char *var_value, int flags ) {
 				flags &= ~CVAR_VM_CREATED;
 		}
 		
+		// Make sure servers cannot mark engine-added variables as SERVER_CREATED
+		if(var->flags & CVAR_SERVER_CREATED)
+		{
+			if(!(flags & CVAR_SERVER_CREATED))
+				var->flags &= ~CVAR_SERVER_CREATED;
+		}
+		else
+		{
+			if(flags & CVAR_SERVER_CREATED)
+				flags &= ~CVAR_SERVER_CREATED;
+		}
+		
 		var->flags |= flags;
 
 		// only allow one non-empty reset string without a warning
@@ -613,6 +625,28 @@ void Cvar_Set( const char *var_name, const char *value) {
 
 /*
 ============
+Cvar_SetSafe
+============
+*/
+void Cvar_SetSafe( const char *var_name, const char *value )
+{
+	int flags = Cvar_Flags( var_name );
+
+	if( flags != CVAR_NONEXISTENT && flags & CVAR_PROTECTED )
+	{
+		if( value )
+			Com_Error( ERR_DROP, "Restricted source tried to set "
+				"\"%s\" to \"%s\"\n", var_name, value );
+		else
+			Com_Error( ERR_DROP, "Restricted source tried to "
+				"modify \"%s\"\n", var_name );
+		return;
+	}
+	Cvar_Set( var_name, value );
+}
+
+/*
+============
 Cvar_SetLatched
 ============
 */
@@ -636,6 +670,21 @@ void Cvar_SetValue( const char *var_name, float value) {
 	Cvar_Set (var_name, val);
 }
 
+/*
+============
+Cvar_SetValueSafe
+============
+*/
+void Cvar_SetValueSafe( const char *var_name, float value )
+{
+	char val[32];
+
+	if( Q_isintegral( value ) )
+		Com_sprintf( val, sizeof(val), "%i", (int)value );
+	else
+		Com_sprintf( val, sizeof(val), "%f", value );
+	Cvar_SetSafe( var_name, val );
+}
 
 /*
 ============
@@ -1211,9 +1260,9 @@ void	Cvar_Update( vmCvar_t *vmCvar ) {
 	}
 	vmCvar->modificationCount = cv->modificationCount;
 	if ( strlen(cv->string)+1 > MAX_CVAR_VALUE_STRING ) 
-	  Com_Error( ERR_DROP, "Cvar_Update: src %s length %zd exceeds MAX_CVAR_VALUE_STRING",
+	  Com_Error( ERR_DROP, "Cvar_Update: src %s length %u exceeds MAX_CVAR_VALUE_STRING",
 		     cv->string, 
-		     strlen(cv->string));
+		     (unsigned int) strlen(cv->string));
 	Q_strncpyz( vmCvar->string, cv->string,  MAX_CVAR_VALUE_STRING ); 
 
 	vmCvar->value = cv->value;
diff --git a/src/qcommon/files.c b/src/qcommon/files.c
index 302ce0fe..62617223 100644
--- a/src/qcommon/files.c
+++ b/src/qcommon/files.c
@@ -173,21 +173,6 @@ or configs will never get loaded from disk!
 
 */
 
-// every time a new demo pk3 file is built, this checksum must be updated.
-// the easiest way to get it is to just run the game and see what it spits out
-#define	DEMO_PAK0_CHECKSUM	2985612116u
-static const unsigned pak_checksums[] = {
-	1566731103u,
-	298122907u,
-	412165236u,
-	2991495316u,
-	1197932710u,
-	4087071573u,
-	3709064859u,
-	908855077u,
-	977125798u
-};
-
 // if this is defined, the executable positively won't work with any paks other
 // than the demo pak, even if productid is present.  This is only used for our
 // last demo release to prevent the mac and linux users from using the demo
@@ -326,7 +311,7 @@ qboolean FS_PakIsPure( pack_t *pack ) {
 		for ( i = 0 ; i < fs_numServerPaks ; i++ ) {
 			// FIXME: also use hashed file names
 			// NOTE TTimo: a pk3 with same checksum but different name would be validated too
-			//   I don't see this as allowing for any exploit, it would only happen if the client does manips of it's file names 'not a bug'
+			//   I don't see this as allowing for any exploit, it would only happen if the client does manips of its file names 'not a bug'
 			if ( pack->checksum == fs_serverPaks[i] ) {
 				return qtrue;		// on the aproved list
 			}
@@ -538,57 +523,6 @@ static void FS_CheckFilenameIsNotExecutable( const char *filename,
 	}
 }
 
-
-/*
-=================
-FS_CopyFile
-
-Copy a fully specified file from one place to another
-=================
-*/
-static void FS_CopyFile( char *fromOSPath, char *toOSPath ) {
-	FILE	*f;
-	int		len;
-	byte	*buf;
-
-	Com_Printf( "copy %s to %s\n", fromOSPath, toOSPath );
-
-	FS_CheckFilenameIsNotExecutable( toOSPath, __func__ );
-
-	if (strstr(fromOSPath, "journal.dat") || strstr(fromOSPath, "journaldata.dat")) {
-		Com_Printf( "Ignoring journal files\n");
-		return;
-	}
-
-	f = fopen( fromOSPath, "rb" );
-	if ( !f ) {
-		return;
-	}
-	fseek (f, 0, SEEK_END);
-	len = ftell (f);
-	fseek (f, 0, SEEK_SET);
-
-	// we are using direct malloc instead of Z_Malloc here, so it
-	// probably won't work on a mac... Its only for developers anyway...
-	buf = malloc( len );
-	if (fread( buf, 1, len, f ) != len)
-		Com_Error( ERR_FATAL, "Short read in FS_Copyfiles()\n" );
-	fclose( f );
-
-	if( FS_CreatePath( toOSPath ) ) {
-		return;
-	}
-
-	f = fopen( toOSPath, "wb" );
-	if ( !f ) {
-		return;
-	}
-	if (fwrite( buf, 1, len, f ) != len)
-		Com_Error( ERR_FATAL, "Short write in FS_Copyfiles()\n" );
-	fclose( f );
-	free( buf );
-}
-
 /*
 ===========
 FS_Remove
@@ -800,11 +734,7 @@ void FS_SV_Rename( const char *from, const char *to ) {
 
 	FS_CheckFilenameIsNotExecutable( to_ospath, __func__ );
 
-	if (rename( from_ospath, to_ospath )) {
-		// Failed, try copying it and deleting the original
-		FS_CopyFile ( from_ospath, to_ospath );
-		FS_Remove ( from_ospath );
-	}
+	rename(from_ospath, to_ospath);
 }
 
 
@@ -834,11 +764,7 @@ void FS_Rename( const char *from, const char *to ) {
 
 	FS_CheckFilenameIsNotExecutable( to_ospath, __func__ );
 
-	if (rename( from_ospath, to_ospath )) {
-		// Failed, try copying it and deleting the original
-		FS_CopyFile ( from_ospath, to_ospath );
-		FS_Remove ( from_ospath );
-	}
+	rename(from_ospath, to_ospath);
 }
 
 /*
@@ -959,6 +885,52 @@ fileHandle_t FS_FOpenFileAppend( const char *filename ) {
 
 /*
 ===========
+FS_FCreateOpenPipeFile
+
+===========
+*/
+fileHandle_t FS_FCreateOpenPipeFile( const char *filename ) {
+	char	    		*ospath;
+	FILE					*fifo;
+	fileHandle_t	f;
+
+	if ( !fs_searchpaths ) {
+		Com_Error( ERR_FATAL, "Filesystem call made without initialization\n" );
+	}
+
+	f = FS_HandleForFile();
+	fsh[f].zipFile = qfalse;
+
+	Q_strncpyz( fsh[f].name, filename, sizeof( fsh[f].name ) );
+
+	// don't let sound stutter
+	S_ClearSoundBuffer();
+
+	ospath = FS_BuildOSPath( fs_homepath->string, fs_gamedir, filename );
+
+	if ( fs_debug->integer ) {
+		Com_Printf( "FS_FCreateOpenPipeFile: %s\n", ospath );
+	}
+
+	FS_CheckFilenameIsNotExecutable( ospath, __func__ );
+
+	fifo = Sys_Mkfifo( ospath );
+	if( fifo ) {
+		fsh[f].handleFiles.file.o = fifo;
+		fsh[f].handleSync = qfalse;
+	}
+	else
+	{
+		Com_Printf( S_COLOR_YELLOW "WARNING: Could not create new com_pipefile at %s. "
+			"com_pipefile will not be used.\n", ospath );
+		f = 0;
+	}
+
+	return f;
+}
+
+/*
+===========
 FS_FilenameCompare
 
 Ignore case and seprator char distinctions
@@ -995,6 +967,59 @@ qboolean FS_FilenameCompare( const char *s1, const char *s2 ) {
 
 /*
 ===========
+FS_IsExt
+
+Return qtrue if ext matches file extension filename
+===========
+*/
+
+qboolean FS_IsExt(const char *filename, const char *ext, int namelen)
+{
+	int extlen;
+
+	extlen = strlen(ext);
+
+	if(extlen > namelen)
+		return qfalse;
+
+	filename += namelen - extlen;
+
+	return !Q_stricmp(filename, ext);
+}
+
+/*
+===========
+FS_IsDemoExt
+
+Return qtrue if filename has a demo extension
+===========
+*/
+
+qboolean FS_IsDemoExt(const char *filename, int namelen)
+{
+	char *ext_test;
+	int index, protocol;
+
+	ext_test = Q_strrchr(filename, '.');
+	if(ext_test && !Q_stricmpn(ext_test + 1, DEMOEXT, ARRAY_LEN(DEMOEXT) - 1))
+	{
+		protocol = atoi(ext_test + ARRAY_LEN(DEMOEXT));
+
+		if(protocol == PROTOCOL_VERSION)
+			return qtrue;
+
+		for(index = 0; demo_protocols[index]; index++)
+		{
+			if(demo_protocols[index] == protocol)
+			return qtrue;
+		}
+	}
+
+	return qfalse;
+}
+
+/*
+===========
 FS_FOpenFileRead
 
 Finds the file in the search path.
@@ -1014,7 +1039,6 @@ int FS_FOpenFileRead( const char *filename, fileHandle_t *file, qboolean uniqueF
 	long			hash;
 	FILE			*temp;
 	int				l;
-	char demoExt[16];
 
 	hash = 0;
 
@@ -1061,7 +1085,6 @@ int FS_FOpenFileRead( const char *filename, fileHandle_t *file, qboolean uniqueF
 		Com_Error( ERR_FATAL, "FS_FOpenFileRead: NULL 'filename' parameter passed\n" );
 	}
 
-	Com_sprintf (demoExt, sizeof(demoExt), ".dm_%d",PROTOCOL_VERSION );
 	// qpaths are not supposed to have a leading slash
 	if ( filename[0] == '/' || filename[0] == '\\' ) {
 		filename++;
@@ -1106,16 +1129,19 @@ int FS_FOpenFileRead( const char *filename, fileHandle_t *file, qboolean uniqueF
 					// shaders, txt, arena files  by themselves do not count as a reference as 
 					// these are loaded from all pk3s 
 					// from every pk3 file.. 
-					l = strlen( filename );
-					if ( !(pak->referenced & FS_GENERAL_REF)) {
-						if ( Q_stricmp(filename + l - 7, ".shader") != 0 &&
-							Q_stricmp(filename + l - 4, ".txt") != 0 &&
-							Q_stricmp(filename + l - 4, ".cfg") != 0 &&
-							Q_stricmp(filename + l - 7, ".config") != 0 &&
-							strstr(filename, "levelshots") == NULL &&
-							Q_stricmp(filename + l - 4, ".bot") != 0 &&
-							Q_stricmp(filename + l - 6, ".arena") != 0 &&
-							Q_stricmp(filename + l - 5, ".menu") != 0) {
+					l = strlen(filename);
+					
+					if (!(pak->referenced & FS_GENERAL_REF))
+					{
+						if(!FS_IsExt(filename, ".shader", l) &&
+						   !FS_IsExt(filename, ".txt", l) &&
+						   !FS_IsExt(filename, ".cfg", l) &&
+						   !FS_IsExt(filename, ".config", l) &&
+						   !FS_IsExt(filename, ".bot", l) &&
+						   !FS_IsExt(filename, ".arena", l) &&
+						   !FS_IsExt(filename, ".menu", l) &&
+						   !strstr(filename, "levelshots"))
+						{
 							pak->referenced |= FS_GENERAL_REF;
 						}
 					}
@@ -1166,13 +1192,14 @@ int FS_FOpenFileRead( const char *filename, fileHandle_t *file, qboolean uniqueF
 			//   this test can make the search fail although the file is in the directory
 			// I had the problem on https://zerowing.idsoftware.com/bugzilla/show_bug.cgi?id=8
 			// turned out I used FS_FileExists instead
-			if ( fs_numServerPaks ) {
-
-				if ( Q_stricmp( filename + l - 4, ".cfg" )		// for config files
-					&& Q_stricmp( filename + l - 5, ".menu" )	// menu files
-					&& Q_stricmp( filename + l - 5, ".game" )	// menu files
-					&& Q_stricmp( filename + l - strlen(demoExt), demoExt )	// menu files
-					&& Q_stricmp( filename + l - 4, ".dat" ) ) {	// for journal files
+			if(fs_numServerPaks)
+			{
+				if(!FS_IsExt(filename, ".cfg", l) &&		// for config files
+				   !FS_IsExt(filename, ".menu", l) &&		// menu files
+				   !FS_IsExt(filename, ".game", l) &&		// menu files
+				   !FS_IsExt(filename, ".cfg", l) &&		// for journal files
+				   !FS_IsDemoExt(filename, l))			// demos
+				{
 					continue;
 				}
 			}
@@ -1206,6 +1233,32 @@ int FS_FOpenFileRead( const char *filename, fileHandle_t *file, qboolean uniqueF
 }
 
 
+char *FS_FindDll( const char *filename ) {
+	searchpath_t	*search;
+	directory_t		*dir;
+
+	if ( !fs_searchpaths ) {
+		Com_Error( ERR_FATAL, "Filesystem call made without initialization\n" );
+	}
+
+	for ( search = fs_searchpaths ; search ; search = search->next ) {
+		if ( search->dir ) {
+			FILE *f;
+			char *netpath;
+
+			dir = search->dir;
+			netpath = FS_BuildOSPath( dir->path, dir->gamedir, filename );
+			f = fopen( netpath, "rb" );
+			if (f) {
+				fclose( f );
+				return netpath;
+			}
+		}
+	}
+
+	return NULL;
+}
+
 /*
 =================
 FS_Read
@@ -1881,7 +1934,7 @@ Returns a uniqued list of files that match the given criteria
 from all search paths
 ===============
 */
-char **FS_ListFilteredFiles( const char *path, const char *extension, char *filter, int *numfiles ) {
+char **FS_ListFilteredFiles( const char *path, const char *extension, char *filter, int *numfiles, qboolean allowNonPureFilesOnDisk ) {
 	int				nfiles;
 	char			**listCopy;
 	char			*list[MAX_FOUND_FILES];
@@ -1977,7 +2030,7 @@ char **FS_ListFilteredFiles( const char *path, const char *extension, char *filt
 			char	*name;
 
 			// don't scan directories for files if we are pure or restricted
-			if ( fs_numServerPaks ) {
+			if ( fs_numServerPaks && !allowNonPureFilesOnDisk ) {
 		        continue;
 		    } else {
 				netpath = FS_BuildOSPath( search->dir->path, search->dir->gamedir, path );
@@ -2014,7 +2067,7 @@ FS_ListFiles
 =================
 */
 char **FS_ListFiles( const char *path, const char *extension, int *numfiles ) {
-	return FS_ListFilteredFiles( path, extension, NULL, numfiles );
+	return FS_ListFilteredFiles( path, extension, NULL, numfiles, qfalse );
 }
 
 /*
@@ -2393,7 +2446,7 @@ void FS_NewDir_f( void ) {
 
 	Com_Printf( "---------------\n" );
 
-	dirnames = FS_ListFilteredFiles( "", "", filter, &ndirs );
+	dirnames = FS_ListFilteredFiles( "", "", filter, &ndirs, qfalse );
 
 	FS_SortFileList(dirnames, ndirs);
 
@@ -2605,25 +2658,6 @@ void FS_AddGameDirectory( const char *path, const char *dir ) {
 
 /*
 ================
-FS_idPak
-================
-*/
-qboolean FS_idPak( char *pak, char *base ) {
-	int i;
-
-	for (i = 0; i < NUM_ID_PAKS; i++) {
-		if ( !FS_FilenameCompare(pak, va("%s/pak%d", base, i)) ) {
-			break;
-		}
-	}
-	if (i < NUM_ID_PAKS) {
-		return qtrue;
-	}
-	return qfalse;
-}
-
-/*
-================
 FS_CheckDirTraversal
 
 Check whether the string contains stuff like "../" to prevent directory traversal bugs
@@ -2682,11 +2716,6 @@ qboolean FS_ComparePaks( char *neededpaks, int len, qboolean dlstring ) {
 		badchecksum = qfalse;
 		havepak = qfalse;
 
-		// never autodownload any of the tremulous paks
-		if ( FS_idPak(fs_serverReferencedPakNames[i], BASEGAME) ) {
-			continue;
-		}
-
 		// Make sure the server cannot make us write to non-quake3 directories.
 		if(FS_CheckDirTraversal(fs_serverReferencedPakNames[i]))
 		{
@@ -2863,13 +2892,13 @@ static void FS_Startup( const char *gameName )
 	fs_packFiles = 0;
 
 	fs_debug = Cvar_Get( "fs_debug", "0", 0 );
-	fs_basepath = Cvar_Get ("fs_basepath", Sys_DefaultInstallPath(), CVAR_INIT );
+	fs_basepath = Cvar_Get ("fs_basepath", Sys_DefaultInstallPath(), CVAR_INIT|CVAR_PROTECTED );
 	fs_basegame = Cvar_Get ("fs_basegame", "", CVAR_INIT );
 	homePath = Sys_DefaultHomePath();
 	if (!homePath || !homePath[0]) {
 		homePath = fs_basepath->string;
 	}
-	fs_homepath = Cvar_Get ("fs_homepath", homePath, CVAR_INIT );
+	fs_homepath = Cvar_Get ("fs_homepath", homePath, CVAR_INIT|CVAR_PROTECTED );
 	fs_gamedirvar = Cvar_Get ("fs_game", "", CVAR_INIT|CVAR_SYSTEMINFO );
 
 	// add search path elements in reverse priority order
@@ -2879,7 +2908,7 @@ static void FS_Startup( const char *gameName )
 	// fs_homepath is somewhat particular to *nix systems, only add if relevant
 	
 	#ifdef MACOS_X
-	fs_apppath = Cvar_Get ("fs_apppath", Sys_DefaultAppPath(), CVAR_INIT );
+	fs_apppath = Cvar_Get ("fs_apppath", Sys_DefaultAppPath(), CVAR_INIT|CVAR_PROTECTED );
 	// Make MacOSX also include the base path included with the .app bundle
 	if (fs_apppath->string[0])
 		FS_AddGameDirectory(fs_apppath->string, gameName);
@@ -2937,7 +2966,6 @@ static void FS_Startup( const char *gameName )
 	Com_Printf( "%d files in pk3 files\n", fs_packFiles );
 }
 
-
 /*
 =====================
 FS_GamePureChecksum
@@ -3256,7 +3284,7 @@ void FS_PureServerSetReferencedPaks( const char *pakSums, const char *pakNames )
 		fs_serverReferencedPaks[i] = atoi( Cmd_Argv( i ) );
 	}
 
-	for (i = 0 ; i < sizeof(fs_serverReferencedPakNames) / sizeof(*fs_serverReferencedPakNames); i++)
+	for (i = 0 ; i < ARRAY_LEN(fs_serverReferencedPakNames); i++)
 	{
 		if(fs_serverReferencedPakNames[i])
 			Z_Free(fs_serverReferencedPakNames[i]);
@@ -3465,13 +3493,13 @@ void	FS_Flush( fileHandle_t f ) {
 }
 
 void	FS_FilenameCompletion( const char *dir, const char *ext,
-		qboolean stripExt, void(*callback)(const char *s) ) {
+		qboolean stripExt, void(*callback)(const char *s), qboolean allowNonPureFilesOnDisk ) {
 	char	**filenames;
 	int		nfiles;
 	int		i;
 	char	filename[ MAX_STRING_CHARS ];
 
-	filenames = FS_ListFilteredFiles( dir, ext, NULL, &nfiles );
+	filenames = FS_ListFilteredFiles( dir, ext, NULL, &nfiles, allowNonPureFilesOnDisk );
 
 	FS_SortFileList( filenames, nfiles );
 
@@ -3487,3 +3515,11 @@ void	FS_FilenameCompletion( const char *dir, const char *ext,
 	}
 	FS_FreeFileList( filenames );
 }
+
+const char *FS_GetCurrentGameDir(void)
+{
+	if(fs_gamedirvar->string[0])
+		return fs_gamedirvar->string;
+
+	return BASEGAME;
+}
diff --git a/src/qcommon/msg.c b/src/qcommon/msg.c
index ba956bc8..05a7aa9f 100644
--- a/src/qcommon/msg.c
+++ b/src/qcommon/msg.c
@@ -885,7 +885,7 @@ void MSG_WriteDeltaEntity( msg_t *msg, struct entityState_s *from, struct entity
 	float		fullFloat;
 	int			*fromF, *toF;
 
-	numFields = sizeof(entityStateFields)/sizeof(entityStateFields[0]);
+	numFields = ARRAY_LEN( entityStateFields );
 
 	// all fields should be 32 bits to avoid any compiler packing issues
 	// the "number" field is not part of the field list
@@ -1030,7 +1030,7 @@ void MSG_ReadDeltaEntity( msg_t *msg, entityState_t *from, entityState_t *to,
 		return;
 	}
 
-	numFields = sizeof(entityStateFields)/sizeof(entityStateFields[0]);
+	numFields = ARRAY_LEN( entityStateFields );
 	lc = MSG_ReadByte(msg);
 
 	if ( lc > numFields || lc < 0 ) {
@@ -1203,7 +1203,7 @@ void MSG_WriteDeltaPlayerstate( msg_t *msg, struct playerState_s *from, struct p
 
 	c = msg->cursize;
 
-	numFields = sizeof( playerStateFields ) / sizeof( playerStateFields[0] );
+	numFields = ARRAY_LEN( playerStateFields );
 
 	lc = 0;
 	for ( i = 0, field = playerStateFields ; i < numFields ; i++, field++ ) {
@@ -1353,7 +1353,7 @@ void MSG_ReadDeltaPlayerstate (msg_t *msg, playerState_t *from, playerState_t *t
 		print = 0;
 	}
 
-	numFields = sizeof( playerStateFields ) / sizeof( playerStateFields[0] );
+	numFields = ARRAY_LEN( playerStateFields );
 	lc = MSG_ReadByte(msg);
 
 	if ( lc > numFields || lc < 0 ) {
diff --git a/src/qcommon/net_ip.c b/src/qcommon/net_ip.c
index c1e527c5..e3a95e4a 100644
--- a/src/qcommon/net_ip.c
+++ b/src/qcommon/net_ip.c
@@ -51,6 +51,7 @@ typedef unsigned short sa_family_t;
 #	define EADDRNOTAVAIL	WSAEADDRNOTAVAIL
 #	define EAFNOSUPPORT		WSAEAFNOSUPPORT
 #	define ECONNRESET			WSAECONNRESET
+typedef u_long	ioctlarg_t;
 #	define socketError		WSAGetLastError( )
 
 static WSADATA	winsockdata;
@@ -86,6 +87,7 @@ typedef int SOCKET;
 #	define SOCKET_ERROR			-1
 #	define closesocket			close
 #	define ioctlsocket			ioctl
+typedef int	ioctlarg_t;
 #	define socketError			errno
 
 #endif
@@ -108,6 +110,8 @@ static cvar_t	*net_port6;
 static cvar_t	*net_mcast6addr;
 static cvar_t	*net_mcast6iface;
 
+static cvar_t	*net_dropsim;
+
 static struct sockaddr	socksRelayAddr;
 
 static SOCKET	ip_socket = INVALID_SOCKET;
@@ -202,7 +206,7 @@ char *NET_ErrorString( void ) {
 		default: return "NO ERROR";
 	}
 #else
-	return strerror (errno);
+	return strerror(socketError);
 #endif
 }
 
@@ -521,16 +525,17 @@ qboolean	NET_IsLocalAddress( netadr_t adr ) {
 
 /*
 ==================
-Sys_GetPacket
+NET_GetPacket
 
-Never called by the game logic, just the system event queing
+Receive one packet
 ==================
 */
 #ifdef _DEBUG
 int	recvfromCount;
 #endif
 
-qboolean Sys_GetPacket( netadr_t *net_from, msg_t *net_message ) {
+qboolean NET_GetPacket(netadr_t *net_from, msg_t *net_message, fd_set *fdr)
+{
 	int 	ret;
 	struct sockaddr_storage from;
 	socklen_t	fromlen;
@@ -540,7 +545,7 @@ qboolean Sys_GetPacket( netadr_t *net_from, msg_t *net_message ) {
 	recvfromCount++;		// performance check
 #endif
 	
-	if(ip_socket != INVALID_SOCKET)
+	if(ip_socket != INVALID_SOCKET && FD_ISSET(ip_socket, fdr))
 	{
 		fromlen = sizeof(from);
 		ret = recvfrom( ip_socket, (void *)net_message->data, net_message->maxsize, 0, (struct sockaddr *) &from, &fromlen );
@@ -574,7 +579,7 @@ qboolean Sys_GetPacket( netadr_t *net_from, msg_t *net_message ) {
 				net_message->readcount = 0;
 			}
 		
-			if( ret == net_message->maxsize ) {
+			if( ret >= net_message->maxsize ) {
 				Com_Printf( "Oversize packet from %s\n", NET_AdrToString (*net_from) );
 				return qfalse;
 			}
@@ -584,7 +589,7 @@ qboolean Sys_GetPacket( netadr_t *net_from, msg_t *net_message ) {
 		}
 	}
 	
-	if(ip6_socket != INVALID_SOCKET)
+	if(ip6_socket != INVALID_SOCKET && FD_ISSET(ip6_socket, fdr))
 	{
 		fromlen = sizeof(from);
 		ret = recvfrom(ip6_socket, (void *)net_message->data, net_message->maxsize, 0, (struct sockaddr *) &from, &fromlen);
@@ -601,7 +606,7 @@ qboolean Sys_GetPacket( netadr_t *net_from, msg_t *net_message ) {
 			SockadrToNetadr((struct sockaddr *) &from, net_from);
 			net_message->readcount = 0;
 		
-			if(ret == net_message->maxsize)
+			if(ret >= net_message->maxsize)
 			{
 				Com_Printf( "Oversize packet from %s\n", NET_AdrToString (*net_from) );
 				return qfalse;
@@ -612,7 +617,7 @@ qboolean Sys_GetPacket( netadr_t *net_from, msg_t *net_message ) {
 		}
 	}
 
-	if(multicast6_socket != INVALID_SOCKET && multicast6_socket != ip6_socket)
+	if(multicast6_socket != INVALID_SOCKET && multicast6_socket != ip6_socket && FD_ISSET(multicast6_socket, fdr))
 	{
 		fromlen = sizeof(from);
 		ret = recvfrom(multicast6_socket, (void *)net_message->data, net_message->maxsize, 0, (struct sockaddr *) &from, &fromlen);
@@ -629,7 +634,7 @@ qboolean Sys_GetPacket( netadr_t *net_from, msg_t *net_message ) {
 			SockadrToNetadr((struct sockaddr *) &from, net_from);
 			net_message->readcount = 0;
 		
-			if(ret == net_message->maxsize)
+			if(ret >= net_message->maxsize)
 			{
 				Com_Printf( "Oversize packet from %s\n", NET_AdrToString (*net_from) );
 				return qfalse;
@@ -825,7 +830,7 @@ NET_IPSocket
 int NET_IPSocket( char *net_interface, int port, int *err ) {
 	SOCKET				newsocket;
 	struct sockaddr_in	address;
-	u_long				_true = 1;
+	ioctlarg_t			_true = 1;
 	int					i = 1;
 
 	*err = 0;
@@ -893,7 +898,7 @@ NET_IP6Socket
 int NET_IP6Socket( char *net_interface, int port, struct sockaddr_in6 *bindto, int *err ) {
 	SOCKET				newsocket;
 	struct sockaddr_in6	address;
-	u_long				_true = 1;
+	ioctlarg_t			_true = 1;
 
 	*err = 0;
 
@@ -1496,6 +1501,8 @@ static qboolean NET_GetCvars( void ) {
 	modified += net_socksPassword->modified;
 	net_socksPassword->modified = qfalse;
 
+	net_dropsim = Cvar_Get("net_dropsim", "", CVAR_TEMP);
+
 	return modified ? qtrue : qfalse;
 }
 
@@ -1624,6 +1631,42 @@ void NET_Shutdown( void ) {
 #endif
 }
 
+/*
+====================
+NET_Event
+
+Called from NET_Sleep which uses select() to determine which sockets have seen action.
+====================
+*/
+
+void NET_Event(fd_set *fdr)
+{
+	byte bufData[MAX_MSGLEN + 1];
+	netadr_t from;
+	msg_t netmsg;
+	
+	while(1)
+	{
+		MSG_Init(&netmsg, bufData, sizeof(bufData));
+
+		if(NET_GetPacket(&from, &netmsg, fdr))
+		{
+			if(net_dropsim->value > 0.0f && net_dropsim->value <= 100.0f)
+			{
+				// com_dropsim->value percent of incoming packets get dropped.
+				if(rand() < (int) (((double) RAND_MAX) / 100.0 * (double) net_dropsim->value))
+					continue;          // drop this packet
+                        }
+
+			if(com_sv_running->integer)
+				Com_RunAndTimeServerPacket(&from, &netmsg);
+			else
+				CL_PacketEvent(from, &netmsg);
+		}
+		else
+			break;
+	}
+}
 
 /*
 ====================
@@ -1632,31 +1675,23 @@ NET_Sleep
 Sleeps msec or until something happens on the network
 ====================
 */
-void NET_Sleep( int msec ) {
+void NET_Sleep(int msec)
+{
 	struct timeval timeout;
-	fd_set	fdset;
-	int highestfd = -1;
-
-	if (!com_dedicated->integer)
-		return; // we're not a server, just run full speed
-
-	if (ip_socket == INVALID_SOCKET && ip6_socket == INVALID_SOCKET)
-		return;
-
-	if (msec < 0 )
-		return;
+	fd_set fdr;
+	int highestfd = -1, retval;
 
-	FD_ZERO(&fdset);
+	FD_ZERO(&fdr);
 
 	if(ip_socket != INVALID_SOCKET)
 	{
-		FD_SET(ip_socket, &fdset);
+		FD_SET(ip_socket, &fdr);
 
 		highestfd = ip_socket;
 	}
 	if(ip6_socket != INVALID_SOCKET)
 	{
-		FD_SET(ip6_socket, &fdset);
+		FD_SET(ip6_socket, &fdr);
 		
 		if(ip6_socket > highestfd)
 			highestfd = ip6_socket;
@@ -1664,9 +1699,23 @@ void NET_Sleep( int msec ) {
 
 	timeout.tv_sec = msec/1000;
 	timeout.tv_usec = (msec%1000)*1000;
-	select(highestfd + 1, &fdset, NULL, NULL, &timeout);
-}
+	
+#ifdef _WIN32
+	if(highestfd < 0)
+	{
+		// windows ain't happy when select is called without valid FDs
+		SleepEx(msec, 0);
+		return;
+	}
+#endif
 
+	retval = select(highestfd + 1, &fdr, NULL, NULL, &timeout);
+	
+	if(retval < 0)
+		Com_Printf("Warning: select() syscall failed: %s\n", NET_ErrorString());
+	else if(retval > 0)
+		NET_Event(&fdr);
+}
 
 /*
 ====================
diff --git a/src/qcommon/q_platform.h b/src/qcommon/q_platform.h
index 4d4d2856..3e9619ee 100644
--- a/src/qcommon/q_platform.h
+++ b/src/qcommon/q_platform.h
@@ -155,11 +155,16 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 
 //================================================================= LINUX ===
 
-#ifdef __linux__
+#if defined(__linux__) || defined(__FreeBSD_kernel__)
 
 #include <endian.h>
 
+#if defined(__linux__)
 #define OS_STRING "linux"
+#else
+#define OS_STRING "kFreeBSD"
+#endif
+
 #define ID_INLINE inline
 #define PATH_SEP '/'
 
diff --git a/src/qcommon/q_shared.c b/src/qcommon/q_shared.c
index ce28f953..8b13aba2 100644
--- a/src/qcommon/q_shared.c
+++ b/src/qcommon/q_shared.c
@@ -706,10 +706,6 @@ char* Q_strrchr( const char* string, int c )
 
 qboolean Q_isanumber( const char *s )
 {
-#ifdef Q3_VM
-	//FIXME: implement
-	return qfalse;
-#else
 	char *p;
 	double d;
 
@@ -719,7 +715,6 @@ qboolean Q_isanumber( const char *s )
 	d = strtod( s, &p );
 
 	return *p == '\0';
-#endif
 }
 
 qboolean Q_isintegral( float f )
@@ -727,6 +722,39 @@ qboolean Q_isintegral( float f )
 	return (int)f == f;
 }
 
+#ifdef _MSC_VER
+/*
+=============
+Q_vsnprintf
+ 
+Special wrapper function for Microsoft's broken _vsnprintf() function.
+MinGW comes with its own snprintf() which is not broken.
+=============
+*/
+
+int Q_vsnprintf(char *str, size_t size, const char *format, va_list ap)
+{
+	int retval;
+	
+	retval = _vsnprintf(str, size, format, ap);
+
+	if(retval < 0 || retval == size)
+	{
+		// Microsoft doesn't adhere to the C99 standard of vsnprintf,
+		// which states that the return value must be the number of
+		// bytes written if the output string had sufficient length.
+		//
+		// Obviously we cannot determine that value from Microsoft's
+		// implementation, so we have no choice but to return size.
+		
+		str[size - 1] = '\0';
+		return size;
+	}
+	
+	return retval;
+}
+#endif
+
 /*
 =============
 Q_strncpyz
@@ -954,28 +982,18 @@ void Q_ParseNewlines( char *dest, const char *src, int destsize )
   *dest++ = '\0';
 }
 
-void QDECL Com_sprintf( char *dest, int size, const char *fmt, ...) {
+void QDECL Com_sprintf(char *dest, int size, const char *fmt, ...)
+{
 	int		len;
 	va_list		argptr;
-	char	bigbuffer[32000];	// big, but small enough to fit in PPC stack
 
 	va_start (argptr,fmt);
-	len = Q_vsnprintf (bigbuffer, sizeof(bigbuffer), fmt,argptr);
+	len = Q_vsnprintf(dest, size, fmt, argptr);
 	va_end (argptr);
-	if ( len >= sizeof( bigbuffer ) ) {
-		Com_Error( ERR_FATAL, "Com_sprintf: overflowed bigbuffer" );
-	}
-	if (len >= size) {
-		Com_Printf ("Com_sprintf: overflow of %i in %i\n", len, size);
-#ifdef	_DEBUG
-		__asm {
-			int 3;
-		}
-#endif
-	}
-	Q_strncpyz (dest, bigbuffer, size );
-}
 
+	if(len >= size)
+		Com_Printf("Com_sprintf: Output length %d too short, require %d bytes.\n", size, len);
+}
 
 /*
 ============
@@ -1317,6 +1335,7 @@ void Info_SetValueForKey( char *s, const char *key, const char *value ) {
 Info_SetValueForKey_Big
 
 Changes or adds a key/value pair
+Includes and retains zero-length values
 ==================
 */
 void Info_SetValueForKey_Big( char *s, const char *key, const char *value ) {
@@ -1337,7 +1356,7 @@ void Info_SetValueForKey_Big( char *s, const char *key, const char *value ) {
 	}
 
 	Info_RemoveKey_Big (s, key);
-	if (!value || !strlen(value))
+	if (!value)
 		return;
 
 	Com_sprintf (newi, sizeof(newi), "\\%s\\%s", key, value);
diff --git a/src/qcommon/q_shared.h b/src/qcommon/q_shared.h
index 3f2cb9d5..1cedf6a8 100644
--- a/src/qcommon/q_shared.h
+++ b/src/qcommon/q_shared.h
@@ -38,8 +38,13 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 #define Q3_VERSION                 PRODUCT_NAME " " PRODUCT_VERSION
 
 #define GAMENAME_FOR_MASTER       "Tremulous"
+#define HEARTBEAT_FOR_MASTER      GAMENAME_FOR_MASTER
+#define FLATLINE_FOR_MASTER       GAMENAME_FOR_MASTER "dead"
 
 #define MAX_TEAMNAME 32
+#define MAX_MASTER_SERVERS      5 // number of supported master servers
+
+#define DEMOEXT	"dm_"			// standard demo extension
 
 #ifdef _MSC_VER
 
@@ -118,16 +123,6 @@ typedef int intptr_t;
 #include <ctype.h>
 #include <limits.h>
 
-// vsnprintf is ISO/IEC 9899:1999
-// abstracting this to make it portable
-#ifdef _WIN32
-  #define Q_vsnprintf _vsnprintf
-  #define Q_snprintf _snprintf
-#else
-  #define Q_vsnprintf vsnprintf
-  #define Q_snprintf snprintf
-#endif
-
 #ifdef _MSC_VER
   #include <io.h>
 
@@ -139,8 +134,15 @@ typedef int intptr_t;
   typedef unsigned __int32 uint32_t;
   typedef unsigned __int16 uint16_t;
   typedef unsigned __int8 uint8_t;
+
+  // vsnprintf is ISO/IEC 9899:1999
+  // abstracting this to make it portable
+  int Q_vsnprintf(char *str, size_t size, const char *format, va_list ap);
 #else
   #include <stdint.h>
+
+  #define Q_vsnprintf vsnprintf
+  #define Q_snprintf snprintf
 #endif
 
 #endif
@@ -165,12 +167,13 @@ typedef int		sfxHandle_t;
 typedef int		fileHandle_t;
 typedef int		clipHandle_t;
 
-#define PAD(x,y) (((x)+(y)-1) & ~((y)-1))
+#define PAD(x,y)	(((x)+(y)-1) & ~((y)-1))
+#define PADLEN(x,y)	(PAD((x), (y)) - (x))
 
 #ifdef __GNUC__
-#define ALIGN(x) __attribute__((aligned(x)))
+#define QALIGN(x) __attribute__((aligned(x)))
 #else
-#define ALIGN(x)
+#define QALIGN(x)
 #endif
 
 #ifndef NULL
@@ -184,7 +187,7 @@ typedef int		clipHandle_t;
 #define	MAX_QINT			0x7fffffff
 #define	MIN_QINT			(-MAX_QINT-1)
 
-#define ARRAY_LEN(x)			(sizeof(x) / sizeof(*x))
+#define ARRAY_LEN(x)			(sizeof(x) / sizeof(*(x)))
 
 
 // angle indexes
@@ -886,6 +889,7 @@ default values.
 
 #define CVAR_SERVER_CREATED	0x0800	// cvar was created by a server the client connected to.
 #define CVAR_VM_CREATED		0x1000	// cvar was created exclusively in one of the VMs.
+#define CVAR_PROTECTED		0x2000	// prevent modifying this var from VMs or the server
 #define CVAR_NONEXISTENT	0xFFFFFFFF	// Cvar doesn't exist.
 
 // nothing outside the Cvar_*() functions should modify these fields!
@@ -1414,4 +1418,7 @@ typedef struct
 #define DLP_PROMPTED  0x20 // prompt has been processed by client
 #define DLP_STALE     0x40 // prompt is not being shown by UI VM
 
+#define LERP( a, b, w ) ( ( a ) * ( 1.0f - ( w ) ) + ( b ) * ( w ) )
+#define LUMA( red, green, blue ) ( 0.2126f * ( red ) + 0.7152f * ( green ) + 0.0722f * ( blue ) )
+
 #endif	// __Q_SHARED_H
diff --git a/src/qcommon/qcommon.h b/src/qcommon/qcommon.h
index 09132159..d09ee299 100644
--- a/src/qcommon/qcommon.h
+++ b/src/qcommon/qcommon.h
@@ -416,6 +416,9 @@ void	Cmd_RemoveCommand( const char *cmd_name );
 
 typedef void (*completionFunc_t)( char *args, int argNum );
 
+// don't allow VMs to remove system commands
+void	Cmd_RemoveCommandSafe( const char *cmd_name );
+
 void	Cmd_CommandCompletion( void(*callback)(const char *s) );
 // callback with each valid string
 void Cmd_SetCommandCompletionFunc( const char *command,
@@ -431,6 +434,7 @@ char	*Cmd_ArgsFrom( int arg );
 void	Cmd_ArgsBuffer( char *buffer, int bufferLength );
 void	Cmd_LiteralArgsBuffer( char *buffer, int bufferLength );
 char	*Cmd_Cmd (void);
+void	Cmd_Args_Sanitize( void );
 // The functions that execute commands get their parameters with these
 // functions. Cmd_Argv () will return an empty string, not a NULL
 // if arg > argc, so string operations are allways safe.
@@ -489,11 +493,15 @@ void	Cvar_Update( vmCvar_t *vmCvar );
 void 	Cvar_Set( const char *var_name, const char *value );
 // will create the variable with no flags if it doesn't exist
 
+void	Cvar_SetSafe( const char *var_name, const char *value );
+// sometimes we set variables from an untrusted source: fail if flags & CVAR_PROTECTED
+
 void Cvar_SetLatched( const char *var_name, const char *value);
 // don't set the cvar immediately
 
 void	Cvar_SetValue( const char *var_name, float value );
-// expands value to a string and calls Cvar_Set
+void	Cvar_SetValueSafe( const char *var_name, float value );
+// expands value to a string and calls Cvar_Set/Cvar_SetSafe
 
 float	Cvar_VariableValue( const char *var_name );
 int		Cvar_VariableIntegerValue( const char *var_name );
@@ -561,8 +569,6 @@ issues.
 #define FS_UI_REF		0x02
 #define FS_CGAME_REF	0x04
 #define FS_QAGAME_REF	0x08
-// number of id paks that will never be autodownloaded from baseq3
-#define NUM_ID_PAKS		9
 
 #define	MAX_FILE_HANDLES	64
 
@@ -595,6 +601,9 @@ void	FS_FreeFileList( char **list );
 qboolean FS_FileExists( const char *file );
 
 qboolean FS_CreatePath (char *OSPath);
+
+char *FS_FindDll( const char *filename );
+
 char   *FS_BuildOSPath( const char *base, const char *game, const char *qpath );
 qboolean FS_CompareZipChecksum(const char *zipfile);
 
@@ -605,6 +614,7 @@ int		FS_GetModList(  char *listbuf, int bufsize );
 
 fileHandle_t	FS_FOpenFileWrite( const char *qpath );
 fileHandle_t	FS_FOpenFileAppend( const char *filename );
+fileHandle_t	FS_FCreateOpenPipeFile( const char *filename );
 // will properly create any needed paths and deal with seperater character issues
 
 fileHandle_t FS_SV_FOpenFileWrite( const char *filename );
@@ -692,7 +702,6 @@ void FS_PureServerSetLoadedPaks( const char *pakSums, const char *pakNames );
 // sole exception of .cfg files.
 
 qboolean FS_CheckDirTraversal(const char *checkdir);
-qboolean FS_idPak( char *pak, char *base );
 qboolean FS_ComparePaks( char *neededpaks, int len, qboolean dlstring );
 
 void FS_Rename( const char *from, const char *to );
@@ -701,7 +710,9 @@ void FS_Remove( const char *osPath );
 void FS_HomeRemove( const char *homePath );
 
 void	FS_FilenameCompletion( const char *dir, const char *ext,
-		qboolean stripExt, void(*callback)(const char *s) );
+		qboolean stripExt, void(*callback)(const char *s), qboolean allowNonPureFilesOnDisk );
+
+const char *FS_GetCurrentGameDir(void);
 
 /*
 ==============================================================
@@ -723,7 +734,7 @@ void Field_Clear( field_t *edit );
 void Field_AutoComplete( field_t *edit );
 void Field_CompleteKeyname( void );
 void Field_CompleteFilename( const char *dir,
-		const char *ext, qboolean stripExt );
+		const char *ext, qboolean stripExt, qboolean allowNonPureFilesOnDisk );
 void Field_CompleteCommand( char *cmd,
 		qboolean doCommands, qboolean doCvars );
 
@@ -754,13 +765,12 @@ typedef enum
 
 typedef enum {
 	// SE_NONE must be zero
-	SE_NONE = 0,	// evTime is still valid
-	SE_KEY,		// evValue is a key code, evValue2 is the down flag
-	SE_CHAR,	// evValue is an ascii char
-	SE_MOUSE,	// evValue and evValue2 are reletive signed x / y moves
+	SE_NONE = 0,		// evTime is still valid
+	SE_KEY,			// evValue is a key code, evValue2 is the down flag
+	SE_CHAR,		// evValue is an ascii char
+	SE_MOUSE,		// evValue and evValue2 are reletive signed x / y moves
 	SE_JOYSTICK_AXIS,	// evValue is an axis number and evValue2 is the current state (-127 to 127)
-	SE_CONSOLE,	// evPtr is a char*
-	SE_PACKET	// evPtr is a netadr_t followed by data bytes to evPtrLength
+	SE_CONSOLE		// evPtr is a char*
 } sysEventType_t;
 
 typedef struct {
@@ -793,6 +803,7 @@ int			Com_Filter(char *filter, char *name, int casesensitive);
 int			Com_FilterPath(char *filter, char *name, int casesensitive);
 int			Com_RealTime(qtime_t *qtime);
 qboolean	Com_SafeMode( void );
+void		Com_RunAndTimeServerPacket(netadr_t *evFrom, msg_t *buf);
 
 void		Com_StartupVariable( const char *match );
 // checks for and removes command line "+set var arg" constructs
@@ -817,6 +828,7 @@ extern	cvar_t	*com_maxfpsUnfocused;
 extern	cvar_t	*com_minimized;
 extern	cvar_t	*com_maxfpsMinimized;
 extern	cvar_t	*com_altivec;
+extern	cvar_t	*com_homepath;
 
 // both client and server must agree to pause
 extern	cvar_t	*cl_paused;
@@ -831,7 +843,6 @@ extern	int		time_frontend;
 extern	int		time_backend;		// renderer backend time
 
 extern	int		com_frameTime;
-extern	int		com_frameMsec;
 
 extern	qboolean	com_errorEntered;
 
@@ -976,6 +987,9 @@ void S_ClearSoundBuffer( void );
 
 void SCR_DebugGraph (float value, int color);	// FIXME: move logging to common?
 
+// AVI files have the start of pixel lines 4 byte-aligned
+#define AVI_LINE_PADDING 4
+
 //
 // server interface
 //
@@ -983,6 +997,7 @@ void SV_Init( void );
 void SV_Shutdown( char *finalmsg );
 void SV_Frame( int msec );
 void SV_PacketEvent( netadr_t from, msg_t *msg );
+int SV_FrameMsec(void);
 qboolean SV_GameCommand( void );
 
 
@@ -1012,7 +1027,7 @@ typedef enum {
 void	Sys_Init (void);
 
 // general development dll loading for virtual machine testing
-void	* QDECL Sys_LoadDll( const char *name, char *fqpath , intptr_t (QDECL **entryPoint)(int, ...),
+void	* QDECL Sys_LoadDll( const char *name, intptr_t (QDECL **entryPoint)(int, ...),
 				  intptr_t (QDECL *systemcalls)(intptr_t, ...) );
 void	Sys_UnloadDll( void *dllHandle );
 
@@ -1053,7 +1068,6 @@ cpuFeatures_t Sys_GetProcessorFeatures( void );
 void	Sys_SetErrorText( const char *text );
 
 void	Sys_SendPacket( int length, const void *data, netadr_t to );
-qboolean Sys_GetPacket( netadr_t *net_from, msg_t *net_message );
 
 qboolean	Sys_StringToAdr( const char *s, netadr_t *a, netadrtype_t family );
 //Does NOT parse port numbers, only base addresses.
@@ -1062,6 +1076,7 @@ qboolean	Sys_IsLANAddress (netadr_t adr);
 void		Sys_ShowIP(void);
 
 qboolean Sys_Mkdir( const char *path );
+FILE	*Sys_Mkfifo( const char *ospath );
 char	*Sys_Cwd( void );
 void	Sys_SetDefaultInstallPath(const char *path);
 char	*Sys_DefaultInstallPath(void);
diff --git a/src/qcommon/unzip.c b/src/qcommon/unzip.c
index d0b3d2a5..b307e98d 100644
--- a/src/qcommon/unzip.c
+++ b/src/qcommon/unzip.c
@@ -35,23 +35,10 @@ woven in by Terry Thorsen 1/2003.
  */
 
 
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
+#include "../qcommon/q_shared.h"
+#include "../qcommon/qcommon.h"
 #include "unzip.h"
 
-#ifdef STDC
-#  include <stddef.h>
-#  include <string.h>
-#  include <stdlib.h>
-#endif
-#ifdef NO_ERRNO_H
-    extern int errno;
-#else
-#   include <errno.h>
-#endif
-
-
 #ifndef local
 #  define local static
 #endif
@@ -74,10 +61,10 @@ woven in by Terry Thorsen 1/2003.
 #endif
 
 #ifndef ALLOC
-# define ALLOC(size) (malloc(size))
+# define ALLOC(size) (Z_Malloc(size))
 #endif
 #ifndef TRYFREE
-# define TRYFREE(p) {if (p) free(p);}
+# define TRYFREE(p) {if (p) Z_Free(p);}
 #endif
 
 #define SIZECENTRALDIRITEM (0x2e)
diff --git a/src/qcommon/vm.c b/src/qcommon/vm.c
index 14fa32e9..225119fb 100644
--- a/src/qcommon/vm.c
+++ b/src/qcommon/vm.c
@@ -307,7 +307,7 @@ Dlls will call this directly
 
  rcg010206 The horror; the horror.
 
-  The syscall mechanism relies on stack manipulation to get it's args.
+  The syscall mechanism relies on stack manipulation to get its args.
    This is likely due to C's inability to pass "..." parameters to
    a function in one clean chunk. On PowerPC Linux, these parameters
    are not necessarily passed on the stack, so while (&arg[0] == arg)
@@ -347,7 +347,7 @@ intptr_t QDECL VM_DllSyscall( intptr_t arg, ... ) {
   args[0] = arg;
   
   va_start(ap, arg);
-  for (i = 1; i < sizeof (args) / sizeof (args[i]); i++)
+  for (i = 1; i < ARRAY_LEN (args); i++)
     args[i] = va_arg(ap, intptr_t);
   va_end(ap);
   
@@ -385,8 +385,8 @@ vmHeader_t *VM_LoadQVM( vm_t *vm, qboolean alloc ) {
 		return NULL;
 	}
 
-    // show where the qvm was loaded from
-    Cmd_ExecuteString( va( "which %s\n", filename ) );
+	// show where the qvm was loaded from
+	Cmd_ExecuteString( va( "which %s\n", filename ) );
 
 	if( LittleLong( header.h->vmMagic ) == VM_MAGIC_VER2 ) {
 		Com_Printf( "...which has vmMagic VM_MAGIC_VER2\n" );
@@ -560,7 +560,7 @@ vm_t *VM_Create( const char *module, intptr_t (*systemCalls)(intptr_t *),
 	if ( interpret == VMI_NATIVE ) {
 		// try to load as a system dll
 		Com_Printf( "Loading dll file %s.\n", vm->name );
-		vm->dllHandle = Sys_LoadDll( module, vm->fqpath , &vm->entryPoint, VM_DllSyscall );
+		vm->dllHandle = Sys_LoadDll( module, &vm->entryPoint, VM_DllSyscall );
 		if ( vm->dllHandle ) {
 			return vm;
 		}
@@ -757,7 +757,7 @@ intptr_t	QDECL VM_Call( vm_t *vm, int callnum, ... ) {
 		int args[10];
 		va_list ap;
 		va_start(ap, callnum);
-		for (i = 0; i < sizeof (args) / sizeof (args[i]); i++) {
+		for (i = 0; i < ARRAY_LEN(args); i++) {
 			args[i] = va_arg(ap, int);
 		}
 		va_end(ap);
@@ -782,7 +782,7 @@ intptr_t	QDECL VM_Call( vm_t *vm, int callnum, ... ) {
 
 		a.callnum = callnum;
 		va_start(ap, callnum);
-		for (i = 0; i < sizeof (a.args) / sizeof (a.args[0]); i++) {
+		for (i = 0; i < ARRAY_LEN(a.args); i++) {
 			a.args[i] = va_arg(ap, int);
 		}
 		va_end(ap);
diff --git a/src/qcommon/vm_local.h b/src/qcommon/vm_local.h
index e9c3f7cc..80430be0 100644
--- a/src/qcommon/vm_local.h
+++ b/src/qcommon/vm_local.h
@@ -169,8 +169,6 @@ struct vm_s {
 	int			breakFunction;		// increment breakCount on function entry to this
 	int			breakCount;
 
-	char		fqpath[MAX_QPATH+1] ;
-
 	byte		*jumpTableTargets;
 	int			numJumpTableTargets;
 };
diff --git a/src/qcommon/vm_powerpc.c b/src/qcommon/vm_powerpc.c
index 6b7404dc..831e5672 100644
--- a/src/qcommon/vm_powerpc.c
+++ b/src/qcommon/vm_powerpc.c
@@ -690,7 +690,7 @@ static const long int gpr_list[] = {
 	r7, r8, r9, r10,
 };
 static const long int gpr_vstart = 8; /* position of first volatile register */
-static const long int gpr_total = sizeof( gpr_list ) / sizeof( gpr_list[0] );
+static const long int gpr_total = ARRAY_LEN( gpr_list );
 
 static const long int fpr_list[] = {
 	/* static registers, normally none is used */
@@ -704,7 +704,7 @@ static const long int fpr_list[] = {
 	f12, f13,
 };
 static const long int fpr_vstart = 8;
-static const long int fpr_total = sizeof( fpr_list ) / sizeof( fpr_list[0] );
+static const long int fpr_total = ARRAY_LEN( fpr_list );
 
 /*
  * prepare some dummy structures and emit init code
@@ -1837,7 +1837,7 @@ PPC_ComputeCode( vm_t *vm )
 	unsigned char *dataAndCode = mmap( NULL, codeLength,
 		PROT_READ|PROT_WRITE, MAP_SHARED|MAP_ANONYMOUS, -1, 0 );
 
-	if ( ! dataAndCode )
+	if (dataAndCode == MAP_FAILED)
 		DIE( "Not enough memory" );
 
 	ppc_instruction_t *codeNow, *codeBegin;
diff --git a/src/qcommon/vm_powerpc_asm.c b/src/qcommon/vm_powerpc_asm.c
index da6a8434..58d65a00 100644
--- a/src/qcommon/vm_powerpc_asm.c
+++ b/src/qcommon/vm_powerpc_asm.c
@@ -39,6 +39,7 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  *
  */
 
+#include "vm_local.h"
 #include "vm_powerpc_asm.h"
 
 #include <string.h>
@@ -389,8 +390,7 @@ static const struct powerpc_operand powerpc_operands[] =
 
 };
 
-static const unsigned int num_powerpc_operands =
-	(sizeof (powerpc_operands) / sizeof (powerpc_operands[0]));
+static const unsigned int num_powerpc_operands = ARRAY_LEN (powerpc_operands);
 
 /* The functions used to insert and extract complicated operands.  */
 
@@ -1004,6 +1004,3 @@ static const struct powerpc_opcode powerpc_opcodes[] = {
 { "fsub",    A(63,20,0), AFRC_MASK,	PPCCOM,		{ FRT, FRA, FRB } },
 { "fneg",    XRC(63,40,0), XRA_MASK,	COM,		{ FRT, FRB } },
 };
-
-static const int powerpc_num_opcodes =
-	sizeof (powerpc_opcodes) / sizeof (powerpc_opcodes[0]);
diff --git a/src/qcommon/vm_sparc.c b/src/qcommon/vm_sparc.c
index 69bbdc5f..370ee200 100644
--- a/src/qcommon/vm_sparc.c
+++ b/src/qcommon/vm_sparc.c
@@ -259,7 +259,7 @@ static const struct sparc_opcode sparc_opcodes[] = {
 	{ "fbg",	BFCC(0,6), { ARG_DISP22 }, },
 	{ "fble",	BFCC(0,13), { ARG_DISP22 }, },
 };
-#define SPARC_NUM_OPCODES (sizeof(sparc_opcodes) / sizeof(sparc_opcodes[0]))
+#define SPARC_NUM_OPCODES (ARRAY_LEN(sparc_opcodes))
 
 #define RS1(X)			(((X) & 0x1f) << 14)
 #define RS2(X)			(((X) & 0x1f) << 0)
@@ -320,7 +320,7 @@ static unsigned int sparc_assemble(enum sparc_iname iname, const int argc, const
 
 #define IN(inst, args...) \
 ({	const int argv[] = { args }; \
-	const int argc = sizeof(argv) / sizeof(argv[0]); \
+	const int argc = ARRAY_LEN(argv); \
 	sparc_assemble(inst, argc, argv); \
 })
 
@@ -1427,7 +1427,7 @@ static void sparc_compute_code(vm_t *vm, struct func_info * const fp)
 
 	data_and_code = mmap(NULL, code_length, PROT_READ | PROT_WRITE,
 			     MAP_SHARED | MAP_ANONYMOUS, -1, 0);
-	if (!data_and_code)
+	if (data_and_code == MAP_FAILED)
 		DIE("Not enough memory");
 
 	code_now = code_begin = (unsigned int *)
diff --git a/src/qcommon/vm_x86.c b/src/qcommon/vm_x86.c
index 41806f2a..f25be998 100644
--- a/src/qcommon/vm_x86.c
+++ b/src/qcommon/vm_x86.c
@@ -54,6 +54,7 @@ static void VM_Destroy_Compiled(vm_t* self);
 
 */
 
+#define VMFREE_BUFFERS() do {Z_Free(buf); Z_Free(jused);} while(0)
 static	byte	*buf = NULL;
 static	byte	*jused = NULL;
 static	int		compiledOfs = 0;
@@ -290,6 +291,7 @@ static int Hex( int c ) {
 		return c - '0';
 	}
 
+	VMFREE_BUFFERS();
 	Com_Error( ERR_DROP, "Hex: bad char '%c'", c );
 
 	return 0;
@@ -409,6 +411,7 @@ qboolean EmitMovEBXEDI(vm_t *vm, int andit) {
 #define JUSED(x) \
 	do { \
 		if (x < 0 || x >= jusedSize) { \
+		        VMFREE_BUFFERS(); \
 			Com_Error( ERR_DROP, \
 					"VM_CompileX86: jump target out of range at offset %d", pc ); \
 		} \
@@ -430,7 +433,7 @@ void VM_Compile( vm_t *vm, vmHeader_t *header ) {
 
 	// allocate a very large temp buffer, we will shrink it later
 	maxLength = header->codeLength * 8;
-	buf = Z_Malloc( maxLength );
+	buf = Z_Malloc(maxLength);
 	jused = Z_Malloc(jusedSize);
 	
 	Com_Memset(jused, 0, jusedSize);
@@ -455,16 +458,21 @@ void VM_Compile( vm_t *vm, vmHeader_t *header ) {
 
 	LastCommand = LAST_COMMAND_NONE;
 
-	while ( instruction < header->instructionCount ) {
-		if ( compiledOfs > maxLength - 16 ) {
-			Com_Error( ERR_FATAL, "VM_CompileX86: maxLength exceeded" );
+	while(instruction < header->instructionCount)
+	{
+		if(compiledOfs > maxLength - 16)
+		{
+	        	VMFREE_BUFFERS();
+			Com_Error(ERR_DROP, "VM_CompileX86: maxLength exceeded");
 		}
 
 		vm->instructionPointers[ instruction ] = compiledOfs;
 		instruction++;
 
-		if ( pc > header->codeLength ) {
-			Com_Error( ERR_FATAL, "VM_CompileX86: pc > header->codeLength" );
+		if(pc > header->codeLength)
+		{
+		        VMFREE_BUFFERS();
+			Com_Error(ERR_DROP, "VM_CompileX86: pc > header->codeLength");
 		}
 
 		op = code[ pc ];
@@ -1076,7 +1084,8 @@ void VM_Compile( vm_t *vm, vmHeader_t *header ) {
 			Emit4( (int)vm->instructionPointers );
 			break;
 		default:
-			Com_Error( ERR_DROP, "VM_CompileX86: bad opcode %i at offset %i", op, pc );
+		        VMFREE_BUFFERS();
+			Com_Error(ERR_DROP, "VM_CompileX86: bad opcode %i at offset %i", op, pc);
 		}
 		pop0 = pop1;
 		pop1 = op;
@@ -1087,13 +1096,13 @@ void VM_Compile( vm_t *vm, vmHeader_t *header ) {
 	vm->codeLength = compiledOfs;
 #ifdef VM_X86_MMAP
 	vm->codeBase = mmap(NULL, compiledOfs, PROT_WRITE, MAP_SHARED|MAP_ANONYMOUS, -1, 0);
-	if(vm->codeBase == (void*)-1)
-		Com_Error(ERR_DROP, "VM_CompileX86: can't mmap memory");
+	if(vm->codeBase == MAP_FAILED)
+		Com_Error(ERR_FATAL, "VM_CompileX86: can't mmap memory");
 #elif _WIN32
 	// allocate memory with EXECUTE permissions under windows.
 	vm->codeBase = VirtualAlloc(NULL, compiledOfs, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
 	if(!vm->codeBase)
-		Com_Error(ERR_DROP, "VM_CompileX86: VirtualAlloc failed");
+		Com_Error(ERR_FATAL, "VM_CompileX86: VirtualAlloc failed");
 #else
 	vm->codeBase = malloc(compiledOfs);
 #endif
@@ -1102,14 +1111,14 @@ void VM_Compile( vm_t *vm, vmHeader_t *header ) {
 
 #ifdef VM_X86_MMAP
 	if(mprotect(vm->codeBase, compiledOfs, PROT_READ|PROT_EXEC))
-		Com_Error(ERR_DROP, "VM_CompileX86: mprotect failed");
+		Com_Error(ERR_FATAL, "VM_CompileX86: mprotect failed");
 #elif _WIN32
 	{
 		DWORD oldProtect = 0;
 		
 		// remove write permissions.
 		if(!VirtualProtect(vm->codeBase, compiledOfs, PAGE_EXECUTE_READ, &oldProtect))
-			Com_Error(ERR_DROP, "VM_CompileX86: VirtualProtect failed");
+			Com_Error(ERR_FATAL, "VM_CompileX86: VirtualProtect failed");
 	}
 #endif
 
diff --git a/src/qcommon/vm_x86_64.c b/src/qcommon/vm_x86_64.c
index 13c39172..6b335447 100644
--- a/src/qcommon/vm_x86_64.c
+++ b/src/qcommon/vm_x86_64.c
@@ -55,6 +55,8 @@ static FILE* qdasmout;
 #define Dfprintf(args...)
 #endif
 
+#define VM_FREEBUFFERS(vm) do {assembler_init(0); VM_Destroy_Compiled(vm);} while(0)
+
 void assembler_set_output(char* buf);
 size_t assembler_get_code_size(void);
 void assembler_init(int pass);
@@ -252,6 +254,7 @@ void emit(const char* fmt, ...)
 
 #define CHECK_INSTR(nr) \
 	do { if(nr < 0 || nr >= header->instructionCount) { \
+		VM_FREEBUFFERS(vm); \
 		Com_Error( ERR_DROP, \
 			"%s: jump target 0x%x out of range at offset %d", __func__, nr, pc ); \
 	} } while(0)
@@ -430,6 +433,8 @@ void VM_Compile( vm_t *vm, vmHeader_t *header ) {
 
 	// const optimization
 	unsigned got_const = 0, const_value = 0;
+	
+	vm->codeBase = NULL;
 
 	gettimeofday(&tvstart, NULL);
 
@@ -442,15 +447,17 @@ void VM_Compile( vm_t *vm, vmHeader_t *header ) {
 
 		#ifdef VM_X86_64_MMAP
 			vm->codeBase = mmap(NULL, compiledOfs, PROT_WRITE, MAP_SHARED|MAP_ANONYMOUS, -1, 0);
-			if(vm->codeBase == (void*)-1)
-				Com_Error(ERR_DROP, "VM_CompileX86: can't mmap memory");
+			if(vm->codeBase == MAP_FAILED)
+				Com_Error(ERR_FATAL, "VM_CompileX86_64: can't mmap memory");
 		#elif __WIN64__
 			// allocate memory with write permissions under windows.
 			vm->codeBase = VirtualAlloc(NULL, compiledOfs, MEM_RESERVE|MEM_COMMIT, PAGE_READWRITE);
 			if(!vm->codeBase)
-				Com_Error(ERR_DROP, "VM_CompileX86: VirtualAlloc failed");
+				Com_Error(ERR_FATAL, "VM_CompileX86_64: VirtualAlloc failed");
 		#else
 			vm->codeBase = malloc(compiledOfs);
+			if(!vm_codeBase)
+				Com_Error(ERR_FATAL, "VM_CompileX86_64: Failed to allocate memory");
 		#endif
 
 		assembler_set_output((char*)vm->codeBase);
@@ -931,7 +938,9 @@ emit_do_syscall:
 
 	}
 
-	if(got_const) {
+	if(got_const)
+	{
+		VM_FREEBUFFERS(vm);
 		Com_Error(ERR_DROP, "leftover const\n");
 	}
 
@@ -944,14 +953,14 @@ emit_do_syscall:
 
 	#ifdef VM_X86_64_MMAP
 		if(mprotect(vm->codeBase, compiledOfs, PROT_READ|PROT_EXEC))
-			Com_Error(ERR_DROP, "VM_CompileX86: mprotect failed");
+			Com_Error(ERR_FATAL, "VM_CompileX86_64: mprotect failed");
 	#elif __WIN64__
 		{
 			DWORD oldProtect = 0;
 			
 			// remove write permissions; give exec permision
 			if(!VirtualProtect(vm->codeBase, compiledOfs, PAGE_EXECUTE_READ, &oldProtect))
-				Com_Error(ERR_DROP, "VM_CompileX86: VirtualProtect failed");
+				Com_Error(ERR_FATAL, "VM_CompileX86_64: VirtualProtect failed");
 		}
 	#endif
 
@@ -988,11 +997,16 @@ emit_do_syscall:
 
 void VM_Destroy_Compiled(vm_t* self)
 {
-#ifdef _WIN32
-	VirtualFree(self->codeBase, 0, MEM_RELEASE);
+	if(self && self->codeBase)
+	{
+#ifdef VM_X86_64_MMAP
+		munmap(self->codeBase, self->codeLength);
+#elif __WIN64__
+		VirtualFree(self->codeBase, 0, MEM_RELEASE);
 #else
-	munmap(self->codeBase, self->codeLength);
+		free(self->codeBase);
 #endif
+	}
 }
 
 /*
diff --git a/src/qcommon/vm_x86_64_assembler.c b/src/qcommon/vm_x86_64_assembler.c
index 99d2898e..06695d8d 100644
--- a/src/qcommon/vm_x86_64_assembler.c
+++ b/src/qcommon/vm_x86_64_assembler.c
@@ -22,6 +22,9 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 ===========================================================================
 */
 
+#define _ISOC99_SOURCE
+
+#include "vm_local.h"
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -240,9 +243,15 @@ static void hash_add_label(const char* label, unsigned address)
 {
 	struct hashentry* h;
 	unsigned i = hashkey(label, -1U);
-	i %= sizeof(labelhash)/sizeof(labelhash[0]);
-	h = malloc(sizeof(struct hashentry));
-	h->label = strdup(label);
+	int labellen;
+	
+	i %= ARRAY_LEN(labelhash);
+	h = Z_Malloc(sizeof(struct hashentry));
+	
+	labellen = strlen(label) + 1;
+	h->label = Z_Malloc(labellen);
+	memcpy(h->label, label, labellen);
+	
 	h->address = address;
 	h->next = labelhash[i];
 	labelhash[i] = h;
@@ -252,7 +261,7 @@ static unsigned lookup_label(const char* label)
 {
 	struct hashentry* h;
 	unsigned i = hashkey(label, -1U);
-	i %= sizeof(labelhash)/sizeof(labelhash[0]);
+	i %= ARRAY_LEN(labelhash);
 	for(h = labelhash[i]; h; h = h->next )
 	{
 		if(!strcmp(h->label, label))
@@ -268,15 +277,15 @@ static void labelhash_free(void)
 	struct hashentry* h;
 	unsigned i;
 	unsigned z = 0, min = -1U, max = 0, t = 0;
-	for ( i = 0; i < sizeof(labelhash)/sizeof(labelhash[0]); ++i)
+	for ( i = 0; i < ARRAY_LEN(labelhash); ++i)
 	{
 		unsigned n = 0;
 		h = labelhash[i];
 		while(h)
 		{
 			struct hashentry* next = h->next;
-			free(h->label);
-			free(h);
+			Z_Free(h->label);
+			Z_Free(h);
 			h = next;
 			++n;
 		}
@@ -286,7 +295,7 @@ static void labelhash_free(void)
 		min = MIN(min, n);
 		max = MAX(max, n);
 	}
-	printf("total %u, hsize %"PRIu64", zero %u, min %u, max %u\n", t, sizeof(labelhash)/sizeof(labelhash[0]), z, min, max);
+	printf("total %u, hsize %"PRIu64", zero %u, min %u, max %u\n", t, ARRAY_LEN(labelhash), z, min, max);
 	memset(labelhash, 0, sizeof(labelhash));
 }
 
@@ -1008,7 +1017,7 @@ static op_t* getop(const char* n)
 #else
 	unsigned m, t, b;
 	int r;
-	t = sizeof(ops)/sizeof(ops[0])-1;
+	t = ARRAY_LEN(ops)-1;
 	b = 0;
 
 	while(b <= t)
@@ -1299,7 +1308,7 @@ void assembler_init(int pass)
 	if(!ops_sorted)
 	{
 		ops_sorted = 1;
-		qsort(ops, sizeof(ops)/sizeof(ops[0])-1, sizeof(ops[0]), opsort);
+		qsort(ops, ARRAY_LEN(ops)-1, sizeof(ops[0]), opsort);
 	}
 }
 
diff --git a/src/renderer/tr_bsp.c b/src/renderer/tr_bsp.c
index caf62033..563bb36d 100644
--- a/src/renderer/tr_bsp.c
+++ b/src/renderer/tr_bsp.c
@@ -156,9 +156,6 @@ static	void R_LoadLightmaps( lump_t *l ) {
 		//FIXME: HACK: maps with only one lightmap turn up fullbright for some reason.
 		//this avoids this, but isn't the correct solution.
 		tr.numLightmaps++;
-	} else if ( tr.numLightmaps >= MAX_LIGHTMAPS ) { // 20051020 misantropia
-		ri.Printf( PRINT_WARNING, "WARNING: number of lightmaps > MAX_LIGHTMAPS\n" );
-		tr.numLightmaps = MAX_LIGHTMAPS;
 	}
 
 	// if we are in r_vertexLight mode, we don't need the lightmaps at all
@@ -166,6 +163,7 @@ static	void R_LoadLightmaps( lump_t *l ) {
 		return;
 	}
 
+	tr.lightmaps = ri.Hunk_Alloc( tr.numLightmaps * sizeof(image_t *), h_low );
 	for ( i = 0 ; i < tr.numLightmaps ; i++ ) {
 		// expand the 24 bit on-disk to 32 bit
 		buf_p = buf + i * LIGHTMAP_SIZE*LIGHTMAP_SIZE * 3;
diff --git a/src/renderer/tr_cmds.c b/src/renderer/tr_cmds.c
index a9e88918..833c0f07 100644
--- a/src/renderer/tr_cmds.c
+++ b/src/renderer/tr_cmds.c
@@ -145,7 +145,7 @@ void R_IssueRenderCommands( qboolean runPerformanceCounters ) {
 	}
 
 	// at this point, the back end thread is idle, so it is ok
-	// to look at it's performance counters
+	// to look at its performance counters
 	if ( runPerformanceCounters ) {
 		R_PerformanceCounters();
 	}
diff --git a/src/renderer/tr_flares.c b/src/renderer/tr_flares.c
index 7a5a8174..afef22a4 100644
--- a/src/renderer/tr_flares.c
+++ b/src/renderer/tr_flares.c
@@ -35,7 +35,7 @@ constant, irrespective of distance, but the intensity should be proportional to
 projected area of the light source.
 
 A surface that has been flagged as having a light flare will calculate the depth
-buffer value that it's midpoint should have when the surface is added.
+buffer value that its midpoint should have when the surface is added.
 
 After all opaque surfaces have been rendered, the depth buffer is read back for
 each flare in view.  If the point has not been obscured by a closer surface, the
diff --git a/src/renderer/tr_image.c b/src/renderer/tr_image.c
index 0cb18759..1fa8042d 100644
--- a/src/renderer/tr_image.c
+++ b/src/renderer/tr_image.c
@@ -561,6 +561,27 @@ static void Upload32( unsigned *data,
 	scan = ((byte *)data);
 	samples = 3;
 
+	if( r_greyscale->integer )
+	{
+		for ( i = 0; i < c; i++ )
+		{
+			byte luma = LUMA(scan[i*4], scan[i*4 + 1], scan[i*4 + 2]);
+			scan[i*4] = luma;
+			scan[i*4 + 1] = luma;
+			scan[i*4 + 2] = luma;
+		}
+	}
+	else if( r_greyscale->value )
+	{
+		for ( i = 0; i < c; i++ )
+		{
+			float luma = LUMA(scan[i*4], scan[i*4 + 1], scan[i*4 + 2]);
+			scan[i*4] = LERP(scan[i*4], luma, r_greyscale->value);
+			scan[i*4 + 1] = LERP(scan[i*4 + 1], luma, r_greyscale->value);
+			scan[i*4 + 2] = LERP(scan[i*4 + 2], luma, r_greyscale->value);
+		}
+	}
+
 	if(lightMap)
 	{
 		if(r_greyscale->integer)
diff --git a/src/renderer/tr_image_jpg.c b/src/renderer/tr_image_jpg.c
index baf2a6fd..7390e143 100644
--- a/src/renderer/tr_image_jpg.c
+++ b/src/renderer/tr_image_jpg.c
@@ -31,10 +31,43 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  * You may also wish to include "jerror.h".
  */
 
-#define JPEG_INTERNALS
-#include "../jpeg-6b/jpeglib.h"
+#ifdef USE_INTERNAL_JPEG
+#  define JPEG_INTERNALS
+#endif
 
-void R_LoadJPG( const char *filename, unsigned char **pic, int *width, int *height ) {
+#include <jpeglib.h>
+
+#ifndef USE_INTERNAL_JPEG
+#  if JPEG_LIB_VERSION < 80
+#    error Need system libjpeg >= 80
+#  endif
+#endif
+
+static void R_JPGErrorExit(j_common_ptr cinfo)
+{
+  char buffer[JMSG_LENGTH_MAX];
+  
+  (*cinfo->err->format_message) (cinfo, buffer);
+  
+  /* Let the memory manager delete any temp files before we die */
+  jpeg_destroy(cinfo);
+  
+  ri.Error(ERR_FATAL, "%s\n", buffer);
+}
+
+static void R_JPGOutputMessage(j_common_ptr cinfo)
+{
+  char buffer[JMSG_LENGTH_MAX];
+  
+  /* Create the message */
+  (*cinfo->err->format_message) (cinfo, buffer);
+  
+  /* Send it to stderr, adding a newline */
+  ri.Printf(PRINT_ALL, "%s\n", buffer);
+}
+
+void R_LoadJPG(const char *filename, unsigned char **pic, int *width, int *height)
+{
   /* This struct contains the JPEG decompression parameters and pointers to
    * working space (which is allocated as needed by the JPEG library).
    */
@@ -54,9 +87,10 @@ void R_LoadJPG( const char *filename, unsigned char **pic, int *width, int *heig
   struct jpeg_error_mgr jerr;
   /* More stuff */
   JSAMPARRAY buffer;		/* Output row buffer */
-  unsigned row_stride;		/* physical row width in output buffer */
-  unsigned pixelcount, memcount;
-  unsigned char *out;
+  unsigned int row_stride;	/* physical row width in output buffer */
+  unsigned int pixelcount, memcount;
+  unsigned int sindex, dindex;
+  byte *out;
   int len;
 	union {
 		byte *b;
@@ -83,6 +117,8 @@ void R_LoadJPG( const char *filename, unsigned char **pic, int *width, int *heig
    * address which we place into the link field in cinfo.
    */
   cinfo.err = jpeg_std_error(&jerr);
+  cinfo.err->error_exit = R_JPGErrorExit;
+  cinfo.err->output_message = R_JPGOutputMessage;
 
   /* Now we can initialize the JPEG decompression object. */
   jpeg_create_decompress(&cinfo);
@@ -102,9 +138,11 @@ void R_LoadJPG( const char *filename, unsigned char **pic, int *width, int *heig
 
   /* Step 4: set parameters for decompression */
 
-  /* In this example, we don't need to change any of the defaults set by
-   * jpeg_read_header(), so we do nothing here.
+  /*
+   * Make sure it always converts images to RGB color space. This will
+   * automatically convert 8-bit greyscale images to RGB as well.
    */
+  cinfo.out_color_space = JCS_RGB;
 
   /* Step 5: Start decompressor */
 
@@ -125,9 +163,14 @@ void R_LoadJPG( const char *filename, unsigned char **pic, int *width, int *heig
 
   if(!cinfo.output_width || !cinfo.output_height
       || ((pixelcount * 4) / cinfo.output_width) / 4 != cinfo.output_height
-      || pixelcount > 0x1FFFFFFF || cinfo.output_components > 4) // 4*1FFFFFFF == 0x7FFFFFFC < 0x7FFFFFFF
+      || pixelcount > 0x1FFFFFFF || cinfo.output_components != 3
+    )
   {
-    ri.Error (ERR_DROP, "LoadJPG: %s has an invalid image size: %dx%d*4=%d, components: %d\n", filename,
+    // Free the memory to make sure we don't leak memory
+    ri.FS_FreeFile (fbuffer.v);
+    jpeg_destroy_decompress(&cinfo);
+  
+    ri.Error(ERR_DROP, "LoadJPG: %s has an invalid image format: %dx%d*4=%d, components: %d\n", filename,
 		    cinfo.output_width, cinfo.output_height, pixelcount * 4, cinfo.output_components);
   }
 
@@ -157,42 +200,23 @@ void R_LoadJPG( const char *filename, unsigned char **pic, int *width, int *heig
   
   buf = out;
 
-  // If we are processing an 8-bit JPEG (greyscale), we'll have to convert
-  // the greyscale values to RGBA.
-  if(cinfo.output_components == 1)
-  {
-  	int sindex = pixelcount, dindex = memcount;
-	unsigned char greyshade;
-
-	// Only pixelcount number of bytes have been written.
-	// Expand the color values over the rest of the buffer, starting
-	// from the end.
-	do
-	{
-		greyshade = buf[--sindex];
-
-		buf[--dindex] = 255;
-		buf[--dindex] = greyshade;
-		buf[--dindex] = greyshade;
-		buf[--dindex] = greyshade;
-	} while(sindex);
-  }
-  else
-  {
-	// clear all the alphas to 255
-	int	i;
+  // Expand from RGB to RGBA
+  sindex = pixelcount * cinfo.output_components;
+  dindex = memcount;
 
-	for ( i = 3 ; i < memcount ; i+=4 )
-	{
-		buf[i] = 255;
-	}
-  }
+  do
+  {	
+    buf[--dindex] = 255;
+    buf[--dindex] = buf[--sindex];
+    buf[--dindex] = buf[--sindex];
+    buf[--dindex] = buf[--sindex];
+  } while(sindex);
 
   *pic = out;
 
   /* Step 7: Finish decompression */
 
-  (void) jpeg_finish_decompress(&cinfo);
+  jpeg_finish_decompress(&cinfo);
   /* We can ignore the return value since suspension is not possible
    * with the stdio data source.
    */
@@ -270,7 +294,15 @@ init_destination (j_compress_ptr cinfo)
 static boolean
 empty_output_buffer (j_compress_ptr cinfo)
 {
-  return TRUE;
+  my_dest_ptr dest = (my_dest_ptr) cinfo->dest;
+  
+  jpeg_destroy_compress(cinfo);
+  
+  // Make crash fatal or we would probably leak memory.
+  ri.Error(ERR_FATAL, "Output buffer for encoded JPEG image has insufficient size of %d bytes\n",
+           dest->size);
+
+  return FALSE;
 }
 
 /*
@@ -282,14 +314,8 @@ empty_output_buffer (j_compress_ptr cinfo)
  * for error exit.
  */
 
-static int hackSize;
-
-static void
-term_destination (j_compress_ptr cinfo)
+static void term_destination(j_compress_ptr cinfo)
 {
-  my_dest_ptr dest = (my_dest_ptr) cinfo->dest;
-  size_t datacount = dest->size - dest->pub.free_in_buffer;
-  hackSize = datacount;
 }
 
 
@@ -324,143 +350,40 @@ jpegDest (j_compress_ptr cinfo, byte* outfile, int size)
   dest->size = size;
 }
 
-void SaveJPG(char * filename, int quality, int image_width, int image_height, unsigned char *image_buffer) {
-  /* This struct contains the JPEG compression parameters and pointers to
-   * working space (which is allocated as needed by the JPEG library).
-   * It is possible to have several such structures, representing multiple
-   * compression/decompression processes, in existence at once.  We refer
-   * to any one struct (and its associated working data) as a "JPEG object".
-   */
-  struct jpeg_compress_struct cinfo;
-  /* This struct represents a JPEG error handler.  It is declared separately
-   * because applications often want to supply a specialized error handler
-   * (see the second half of this file for an example).  But here we just
-   * take the easy way out and use the standard error handler, which will
-   * print a message on stderr and call exit() if compression fails.
-   * Note that this struct must live as long as the main JPEG parameter
-   * struct, to avoid dangling-pointer problems.
-   */
-  struct jpeg_error_mgr jerr;
-  /* More stuff */
-  JSAMPROW row_pointer[1];	/* pointer to JSAMPLE row[s] */
-  int row_stride;		/* physical row width in image buffer */
-  unsigned char *out;
-
-  /* Step 1: allocate and initialize JPEG compression object */
-
-  /* We have to set up the error handler first, in case the initialization
-   * step fails.  (Unlikely, but it could happen if you are out of memory.)
-   * This routine fills in the contents of struct jerr, and returns jerr's
-   * address which we place into the link field in cinfo.
-   */
-  cinfo.err = jpeg_std_error(&jerr);
-  /* Now we can initialize the JPEG compression object. */
-  jpeg_create_compress(&cinfo);
-
-  /* Step 2: specify data destination (eg, a file) */
-  /* Note: steps 2 and 3 can be done in either order. */
-
-  /* Here we use the library-supplied code to send compressed data to a
-   * stdio stream.  You can also write your own code to do something else.
-   * VERY IMPORTANT: use "b" option to fopen() if you are on a machine that
-   * requires it in order to write binary files.
-   */
-  out = ri.Hunk_AllocateTempMemory(image_width*image_height*4);
-  jpegDest(&cinfo, out, image_width*image_height*4);
-
-  /* Step 3: set parameters for compression */
-
-  /* First we supply a description of the input image.
-   * Four fields of the cinfo struct must be filled in:
-   */
-  cinfo.image_width = image_width; 	/* image width and height, in pixels */
-  cinfo.image_height = image_height;
-  cinfo.input_components = 4;		/* # of color components per pixel */
-  cinfo.in_color_space = JCS_RGB; 	/* colorspace of input image */
-  /* Now use the library's routine to set default compression parameters.
-   * (You must set at least cinfo.in_color_space before calling this,
-   * since the defaults depend on the source color space.)
-   */
-  jpeg_set_defaults(&cinfo);
-  /* Now you can set any non-default parameters you wish to.
-   * Here we just illustrate the use of quality (quantization table) scaling:
-   */
-  jpeg_set_quality(&cinfo, quality, TRUE /* limit to baseline-JPEG values */);
-  /* If quality is set high, disable chroma subsampling */
-  if (quality >= 85) {
-    cinfo.comp_info[0].h_samp_factor = 1;
-    cinfo.comp_info[0].v_samp_factor = 1;
-  }
-
-  /* Step 4: Start compressor */
-
-  /* TRUE ensures that we will write a complete interchange-JPEG file.
-   * Pass TRUE unless you are very sure of what you're doing.
-   */
-  jpeg_start_compress(&cinfo, TRUE);
-
-  /* Step 5: while (scan lines remain to be written) */
-  /*           jpeg_write_scanlines(...); */
-
-  /* Here we use the library's state variable cinfo.next_scanline as the
-   * loop counter, so that we don't have to keep track ourselves.
-   * To keep things simple, we pass one scanline per call; you can pass
-   * more if you wish, though.
-   */
-  row_stride = image_width * 4;	/* JSAMPLEs per row in image_buffer */
-
-  while (cinfo.next_scanline < cinfo.image_height) {
-    /* jpeg_write_scanlines expects an array of pointers to scanlines.
-     * Here the array is only one element long, but you could pass
-     * more than one scanline at a time if that's more convenient.
-     */
-    row_pointer[0] = & image_buffer[((cinfo.image_height-1)*row_stride)-cinfo.next_scanline * row_stride];
-    (void) jpeg_write_scanlines(&cinfo, row_pointer, 1);
-  }
-
-  /* Step 6: Finish compression */
-
-  jpeg_finish_compress(&cinfo);
-  /* After finish_compress, we can close the output file. */
-  ri.FS_WriteFile( filename, out, hackSize );
-
-  ri.Hunk_FreeTempMemory(out);
-
-  /* Step 7: release JPEG compression object */
-
-  /* This is an important step since it will release a good deal of memory. */
-  jpeg_destroy_compress(&cinfo);
-
-  /* And we're done! */
-}
-
 /*
 =================
 SaveJPGToBuffer
+
+Encodes JPEG from image in image_buffer and writes to buffer.
+Expects RGB input data
 =================
 */
-int SaveJPGToBuffer( byte *buffer, int quality,
-    int image_width, int image_height,
-    byte *image_buffer )
+size_t RE_SaveJPGToBuffer(byte *buffer, size_t bufSize, int quality,
+    int image_width, int image_height, byte *image_buffer, int padding)
 {
   struct jpeg_compress_struct cinfo;
   struct jpeg_error_mgr jerr;
   JSAMPROW row_pointer[1];	/* pointer to JSAMPLE row[s] */
+  my_dest_ptr dest;
   int row_stride;		/* physical row width in image buffer */
+  size_t outcount;
 
   /* Step 1: allocate and initialize JPEG compression object */
   cinfo.err = jpeg_std_error(&jerr);
+  cinfo.err->error_exit = R_JPGErrorExit;
+  cinfo.err->output_message = R_JPGOutputMessage;
+
   /* Now we can initialize the JPEG compression object. */
   jpeg_create_compress(&cinfo);
 
   /* Step 2: specify data destination (eg, a file) */
   /* Note: steps 2 and 3 can be done in either order. */
-  jpegDest(&cinfo, buffer, image_width*image_height*4);
+  jpegDest(&cinfo, buffer, bufSize);
 
   /* Step 3: set parameters for compression */
   cinfo.image_width = image_width; 	/* image width and height, in pixels */
   cinfo.image_height = image_height;
-  cinfo.input_components = 4;		/* # of color components per pixel */
+  cinfo.input_components = 3;		/* # of color components per pixel */
   cinfo.in_color_space = JCS_RGB; 	/* colorspace of input image */
 
   jpeg_set_defaults(&cinfo);
@@ -476,23 +399,40 @@ int SaveJPGToBuffer( byte *buffer, int quality,
 
   /* Step 5: while (scan lines remain to be written) */
   /*           jpeg_write_scanlines(...); */
-  row_stride = image_width * 4;	/* JSAMPLEs per row in image_buffer */
-
+  row_stride = image_width * cinfo.input_components + padding; /* JSAMPLEs per row in image_buffer */
+  
   while (cinfo.next_scanline < cinfo.image_height) {
     /* jpeg_write_scanlines expects an array of pointers to scanlines.
      * Here the array is only one element long, but you could pass
      * more than one scanline at a time if that's more convenient.
      */
-    row_pointer[0] = & image_buffer[((cinfo.image_height-1)*row_stride)-cinfo.next_scanline * row_stride];
+    row_pointer[0] = &image_buffer[((cinfo.image_height-1)*row_stride)-cinfo.next_scanline * row_stride];
     (void) jpeg_write_scanlines(&cinfo, row_pointer, 1);
   }
 
   /* Step 6: Finish compression */
   jpeg_finish_compress(&cinfo);
-
+  
+  dest = (my_dest_ptr) cinfo.dest;
+  outcount = dest->size - dest->pub.free_in_buffer;
+ 
   /* Step 7: release JPEG compression object */
   jpeg_destroy_compress(&cinfo);
 
   /* And we're done! */
-  return hackSize;
+  return outcount;
+}
+
+void RE_SaveJPG(char * filename, int quality, int image_width, int image_height, byte *image_buffer, int padding)
+{
+  byte *out;
+  size_t bufSize;
+
+  bufSize = image_width * image_height * 3;
+  out = ri.Hunk_AllocateTempMemory(bufSize);
+
+  bufSize = RE_SaveJPGToBuffer(out, bufSize, quality, image_width, image_height, image_buffer, padding);
+  ri.FS_WriteFile(filename, out, bufSize);
+
+  ri.Hunk_FreeTempMemory(out);
 }
diff --git a/src/renderer/tr_init.c b/src/renderer/tr_init.c
index 89fcf3e3..10ea222c 100644
--- a/src/renderer/tr_init.c
+++ b/src/renderer/tr_init.c
@@ -152,6 +152,9 @@ cvar_t	*r_saveFontData;
 
 cvar_t	*r_marksOnTriangleMeshes;
 
+cvar_t	*r_aviMotionJpegQuality;
+cvar_t	*r_screenshotJpegQuality;
+
 cvar_t	*r_maxpolys;
 int		max_polys;
 cvar_t	*r_maxpolyverts;
@@ -362,15 +365,62 @@ FIXME: the statics don't get a reinit between fs_game changes
 
 /* 
 ================== 
+RB_ReadPixels
+
+Reads an image but takes care of alignment issues for reading RGB images.
+
+Reads a minimum offset for where the RGB data starts in the image from
+integer stored at pointer offset. When the function has returned the actual
+offset was written back to address offset. This address will always have an
+alignment of packAlign to ensure efficient copying.
+
+Stores the length of padding after a line of pixels to address padlen
+
+Return value must be freed with ri.Hunk_FreeTempMemory()
+================== 
+*/  
+
+byte *RB_ReadPixels(int x, int y, int width, int height, size_t *offset, int *padlen)
+{
+	byte *buffer, *bufstart;
+	int padwidth, linelen;
+	GLint packAlign;
+	
+	qglGetIntegerv(GL_PACK_ALIGNMENT, &packAlign);
+	
+	linelen = width * 3;
+	padwidth = PAD(linelen, packAlign);
+	
+	// Allocate a few more bytes so that we can choose an alignment we like
+	buffer = ri.Hunk_AllocateTempMemory(padwidth * height + *offset + packAlign - 1);
+	
+	bufstart = (byte *) PAD((intptr_t) buffer + *offset, packAlign);
+	qglReadPixels(x, y, width, height, GL_RGB, GL_UNSIGNED_BYTE, bufstart);
+	
+	*offset = bufstart - buffer;
+	*padlen = padwidth - linelen;
+	
+	return buffer;
+}
+
+/* 
+================== 
 RB_TakeScreenshot
 ================== 
 */  
-void RB_TakeScreenshot( int x, int y, int width, int height, char *fileName ) {
-	byte		*buffer;
-	int			i, c, temp;
+void RB_TakeScreenshot(int x, int y, int width, int height, char *fileName)
+{
+	byte *allbuf, *buffer;
+	byte *srcptr, *destptr;
+	byte *endline, *endmem;
+	byte temp;
+	
+	int linelen, padlen;
+	size_t offset = 18, memcount;
 		
-	buffer = ri.Hunk_AllocateTempMemory(glConfig.vidWidth*glConfig.vidHeight*3+18);
-
+	allbuf = RB_ReadPixels(x, y, width, height, &offset, &padlen);
+	buffer = allbuf + offset - 18;
+	
 	Com_Memset (buffer, 0, 18);
 	buffer[2] = 2;		// uncompressed type
 	buffer[12] = width & 255;
@@ -379,47 +429,62 @@ void RB_TakeScreenshot( int x, int y, int width, int height, char *fileName ) {
 	buffer[15] = height >> 8;
 	buffer[16] = 24;	// pixel size
 
-	qglReadPixels( x, y, width, height, GL_RGB, GL_UNSIGNED_BYTE, buffer+18 ); 
+	// swap rgb to bgr and remove padding from line endings
+	linelen = width * 3;
+	
+	srcptr = destptr = allbuf + offset;
+	endmem = srcptr + (linelen + padlen) * height;
+	
+	while(srcptr < endmem)
+	{
+		endline = srcptr + linelen;
 
-	// swap rgb to bgr
-	c = 18 + width * height * 3;
-	for (i=18 ; i<c ; i+=3) {
-		temp = buffer[i];
-		buffer[i] = buffer[i+2];
-		buffer[i+2] = temp;
+		while(srcptr < endline)
+		{
+			temp = srcptr[0];
+			*destptr++ = srcptr[2];
+			*destptr++ = srcptr[1];
+			*destptr++ = temp;
+			
+			srcptr += 3;
+		}
+		
+		// Skip the pad
+		srcptr += padlen;
 	}
 
+	memcount = linelen * height;
+
 	// gamma correct
-	if ( glConfig.deviceSupportsGamma ) {
-		R_GammaCorrect( buffer + 18, glConfig.vidWidth * glConfig.vidHeight * 3 );
-	}
+	if(glConfig.deviceSupportsGamma)
+		R_GammaCorrect(allbuf + offset, memcount);
 
-	ri.FS_WriteFile( fileName, buffer, c );
+	ri.FS_WriteFile(fileName, buffer, memcount + 18);
 
-	ri.Hunk_FreeTempMemory( buffer );
+	ri.Hunk_FreeTempMemory(allbuf);
 }
 
 /* 
 ================== 
 RB_TakeScreenshotJPEG
 ================== 
-*/  
-void RB_TakeScreenshotJPEG( int x, int y, int width, int height, char *fileName ) {
-	byte		*buffer;
+*/
 
-	buffer = ri.Hunk_AllocateTempMemory(glConfig.vidWidth*glConfig.vidHeight*4);
+void RB_TakeScreenshotJPEG(int x, int y, int width, int height, char *fileName)
+{
+	byte *buffer;
+	size_t offset = 0, memcount;
+	int padlen;
 
-	qglReadPixels( x, y, width, height, GL_RGBA, GL_UNSIGNED_BYTE, buffer ); 
+	buffer = RB_ReadPixels(x, y, width, height, &offset, &padlen);
+	memcount = (width * 3 + padlen) * height;
 
 	// gamma correct
-	if ( glConfig.deviceSupportsGamma ) {
-		R_GammaCorrect( buffer, glConfig.vidWidth * glConfig.vidHeight * 4 );
-	}
-
-	ri.FS_WriteFile( fileName, buffer, 1 );		// create path
-	SaveJPG( fileName, 90, glConfig.vidWidth, glConfig.vidHeight, buffer);
+	if(glConfig.deviceSupportsGamma)
+		R_GammaCorrect(buffer + offset, memcount);
 
-	ri.Hunk_FreeTempMemory( buffer );
+	RE_SaveJPG(fileName, r_screenshotJpegQuality->integer, width, height, buffer + offset, padlen);
+	ri.Hunk_FreeTempMemory(buffer);
 }
 
 /*
@@ -525,8 +590,10 @@ the menu system, sampled down from full screen distorted images
 void R_LevelShot( void ) {
 	char		checkname[MAX_OSPATH];
 	byte		*buffer;
-	byte		*source;
+	byte		*source, *allsource;
 	byte		*src, *dst;
+	size_t			offset = 0;
+	int			padlen;
 	int			x, y;
 	int			r, g, b;
 	float		xScale, yScale;
@@ -534,17 +601,16 @@ void R_LevelShot( void ) {
 
 	Com_sprintf(checkname, sizeof(checkname), "levelshots/%s.tga", tr.world->baseName);
 
-	source = ri.Hunk_AllocateTempMemory( glConfig.vidWidth * glConfig.vidHeight * 3 );
+	allsource = RB_ReadPixels(0, 0, glConfig.vidWidth, glConfig.vidHeight, &offset, &padlen);
+	source = allsource + offset;
 
-	buffer = ri.Hunk_AllocateTempMemory( 128 * 128*3 + 18);
+	buffer = ri.Hunk_AllocateTempMemory(128 * 128*3 + 18);
 	Com_Memset (buffer, 0, 18);
 	buffer[2] = 2;		// uncompressed type
 	buffer[12] = 128;
 	buffer[14] = 128;
 	buffer[16] = 24;	// pixel size
 
-	qglReadPixels( 0, 0, glConfig.vidWidth, glConfig.vidHeight, GL_RGB, GL_UNSIGNED_BYTE, source ); 
-
 	// resample from source
 	xScale = glConfig.vidWidth / 512.0f;
 	yScale = glConfig.vidHeight / 384.0f;
@@ -553,7 +619,8 @@ void R_LevelShot( void ) {
 			r = g = b = 0;
 			for ( yy = 0 ; yy < 3 ; yy++ ) {
 				for ( xx = 0 ; xx < 4 ; xx++ ) {
-					src = source + 3 * ( glConfig.vidWidth * (int)( (y*3+yy)*yScale ) + (int)( (x*4+xx)*xScale ) );
+					src = source + (3 * glConfig.vidWidth + padlen) * (int)((y*3 + yy) * yScale) +
+						3 * (int) ((x*4 + xx) * xScale);
 					r += src[0];
 					g += src[1];
 					b += src[2];
@@ -573,8 +640,8 @@ void R_LevelShot( void ) {
 
 	ri.FS_WriteFile( checkname, buffer, 128 * 128*3 + 18 );
 
-	ri.Hunk_FreeTempMemory( buffer );
-	ri.Hunk_FreeTempMemory( source );
+	ri.Hunk_FreeTempMemory(buffer);
+	ri.Hunk_FreeTempMemory(allsource);
 
 	ri.Printf( PRINT_ALL, "Wrote %s\n", checkname );
 }
@@ -707,36 +774,70 @@ RB_TakeVideoFrameCmd
 const void *RB_TakeVideoFrameCmd( const void *data )
 {
 	const videoFrameCommand_t	*cmd;
-	int												frameSize;
-	int												i;
+	byte				*cBuf;
+	size_t				memcount, linelen;
+	int				padwidth, avipadwidth, padlen, avipadlen;
+	GLint packAlign;
 	
 	cmd = (const videoFrameCommand_t *)data;
 	
-	qglReadPixels( 0, 0, cmd->width, cmd->height, GL_RGBA,
-			GL_UNSIGNED_BYTE, cmd->captureBuffer );
+	qglGetIntegerv(GL_PACK_ALIGNMENT, &packAlign);
+
+	linelen = cmd->width * 3;
+
+	// Alignment stuff for glReadPixels
+	padwidth = PAD(linelen, packAlign);
+	padlen = padwidth - linelen;
+	// AVI line padding
+	avipadwidth = PAD(linelen, AVI_LINE_PADDING);
+	avipadlen = avipadwidth - linelen;
+
+	cBuf = (byte *) PAD((intptr_t) cmd->captureBuffer, packAlign);
+		
+	qglReadPixels(0, 0, cmd->width, cmd->height, GL_RGB,
+		GL_UNSIGNED_BYTE, cBuf);
+
+	memcount = padwidth * cmd->height;
 
 	// gamma correct
-	if( glConfig.deviceSupportsGamma )
-		R_GammaCorrect( cmd->captureBuffer, cmd->width * cmd->height * 4 );
+	if(glConfig.deviceSupportsGamma)
+		R_GammaCorrect(cBuf, memcount);
 
-	if( cmd->motionJpeg )
+	if(cmd->motionJpeg)
 	{
-		frameSize = SaveJPGToBuffer( cmd->encodeBuffer, 90,
-				cmd->width, cmd->height, cmd->captureBuffer );
-		ri.CL_WriteAVIVideoFrame( cmd->encodeBuffer, frameSize );
+		memcount = RE_SaveJPGToBuffer(cmd->encodeBuffer, linelen * cmd->height,
+			r_aviMotionJpegQuality->integer,
+			cmd->width, cmd->height, cBuf, padlen);
+		ri.CL_WriteAVIVideoFrame(cmd->encodeBuffer, memcount);
 	}
 	else
 	{
-		frameSize = cmd->width * cmd->height;
-
-		for( i = 0; i < frameSize; i++)    // Pack to 24bpp and swap R and B
+		byte *lineend, *memend;
+		byte *srcptr, *destptr;
+	
+		srcptr = cBuf;
+		destptr = cmd->encodeBuffer;
+		memend = srcptr + memcount;
+		
+		// swap R and B and remove line paddings
+		while(srcptr < memend)
 		{
-			cmd->encodeBuffer[ i*3 ]     = cmd->captureBuffer[ i*4 + 2 ];
-			cmd->encodeBuffer[ i*3 + 1 ] = cmd->captureBuffer[ i*4 + 1 ];
-			cmd->encodeBuffer[ i*3 + 2 ] = cmd->captureBuffer[ i*4 ];
+			lineend = srcptr + linelen;
+			while(srcptr < lineend)
+			{
+				*destptr++ = srcptr[2];
+				*destptr++ = srcptr[1];
+				*destptr++ = srcptr[0];
+				srcptr += 3;
+			}
+			
+			Com_Memset(destptr, '\0', avipadlen);
+			destptr += avipadlen;
+			
+			srcptr += padlen;
 		}
-
-		ri.CL_WriteAVIVideoFrame( cmd->encodeBuffer, frameSize * 3 );
+		
+		ri.CL_WriteAVIVideoFrame(cmd->encodeBuffer, avipadwidth * cmd->height);
 	}
 
 	return (const void *)(cmd + 1);	
@@ -932,6 +1033,7 @@ void R_Register( void )
 	r_stereoEnabled = ri.Cvar_Get( "r_stereoEnabled", "0", CVAR_ARCHIVE | CVAR_LATCH);
 	r_ignoreFastPath = ri.Cvar_Get( "r_ignoreFastPath", "1", CVAR_ARCHIVE | CVAR_LATCH );
 	r_greyscale = ri.Cvar_Get("r_greyscale", "0", CVAR_ARCHIVE | CVAR_LATCH);
+	ri.Cvar_CheckRange(r_greyscale, 0, 1, qfalse);
 
 	//
 	// temporary latched variables that can only change over a restart
@@ -1024,6 +1126,9 @@ void R_Register( void )
 
 	r_marksOnTriangleMeshes = ri.Cvar_Get("r_marksOnTriangleMeshes", "0", CVAR_ARCHIVE);
 
+	r_aviMotionJpegQuality = ri.Cvar_Get("r_aviMotionJpegQuality", "90", CVAR_ARCHIVE);
+	r_screenshotJpegQuality = ri.Cvar_Get("r_screenshotJpegQuality", "90", CVAR_ARCHIVE);
+
 	r_maxpolys = ri.Cvar_Get( "r_maxpolys", va("%d", MAX_POLYS), 0);
 	r_maxpolyverts = ri.Cvar_Get( "r_maxpolyverts", va("%d", MAX_POLYVERTS), 0);
 
diff --git a/src/renderer/tr_local.h b/src/renderer/tr_local.h
index ad32c260..a1978c3a 100644
--- a/src/renderer/tr_local.h
+++ b/src/renderer/tr_local.h
@@ -780,7 +780,6 @@ void		R_Modellist_f (void);
 extern	refimport_t		ri;
 
 #define	MAX_DRAWIMAGES			2048
-#define	MAX_LIGHTMAPS			256
 #define	MAX_SKINS				1024
 
 
@@ -917,7 +916,7 @@ typedef struct {
 	shader_t				*sunShader;
 
 	int						numLightmaps;
-	image_t					*lightmaps[MAX_LIGHTMAPS];
+	image_t					**lightmaps;
 
 	trRefEntity_t			*currentEntity;
 	trRefEntity_t			worldEntity;		// point currentEntity at this when rendering world
@@ -1062,7 +1061,6 @@ extern	cvar_t	*r_colorMipLevels;				// development aid to see texture mip usage
 extern	cvar_t	*r_picmip;						// controls picmip values
 extern	cvar_t	*r_finish;
 extern	cvar_t	*r_drawBuffer;
-extern  cvar_t  *r_glDriver;
 extern	cvar_t	*r_swapInterval;
 extern	cvar_t	*r_textureMode;
 extern	cvar_t	*r_offsetFactor;
@@ -1115,8 +1113,6 @@ extern	cvar_t	*r_saveFontData;
 
 extern cvar_t	*r_marksOnTriangleMeshes;
 
-extern	cvar_t	*r_GLlibCoolDownMsec;
-
 //====================================================================
 
 float R_NoiseGet4f( float x, float y, float z, float t );
@@ -1302,19 +1298,19 @@ typedef struct stageVars
 
 typedef struct shaderCommands_s 
 {
-	glIndex_t	indexes[SHADER_MAX_INDEXES] ALIGN(16);
-	vec4_t		xyz[SHADER_MAX_VERTEXES] ALIGN(16);
-	vec4_t		normal[SHADER_MAX_VERTEXES] ALIGN(16);
-	vec2_t		texCoords[SHADER_MAX_VERTEXES][2] ALIGN(16);
-	color4ub_t	vertexColors[SHADER_MAX_VERTEXES] ALIGN(16);
-	int			vertexDlightBits[SHADER_MAX_VERTEXES] ALIGN(16);
+	glIndex_t	indexes[SHADER_MAX_INDEXES] QALIGN(16);
+	vec4_t		xyz[SHADER_MAX_VERTEXES] QALIGN(16);
+	vec4_t		normal[SHADER_MAX_VERTEXES] QALIGN(16);
+	vec2_t		texCoords[SHADER_MAX_VERTEXES][2] QALIGN(16);
+	color4ub_t	vertexColors[SHADER_MAX_VERTEXES] QALIGN(16);
+	int			vertexDlightBits[SHADER_MAX_VERTEXES] QALIGN(16);
 
-	stageVars_t	svars ALIGN(16);
+	stageVars_t	svars QALIGN(16);
 
-	color4ub_t	constantColor255[SHADER_MAX_VERTEXES] ALIGN(16);
+	color4ub_t	constantColor255[SHADER_MAX_VERTEXES] QALIGN(16);
 
 	shader_t	*shader;
-  float   shaderTime;
+	float		shaderTime;
 	int			fogNum;
 
 	int			dlightBits;	// or together of all vertexDlightBits
@@ -1699,10 +1695,10 @@ void RE_StretchPic ( float x, float y, float w, float h,
 					  float s1, float t1, float s2, float t2, qhandle_t hShader );
 void RE_BeginFrame( stereoFrame_t stereoFrame );
 void RE_EndFrame( int *frontEndMsec, int *backEndMsec );
-void SaveJPG(char * filename, int quality, int image_width, int image_height, unsigned char *image_buffer);
-int SaveJPGToBuffer( byte *buffer, int quality,
-		int image_width, int image_height,
-		byte *image_buffer );
+void RE_SaveJPG(char * filename, int quality, int image_width, int image_height,
+                unsigned char *image_buffer, int padding);
+size_t RE_SaveJPGToBuffer(byte *buffer, size_t bufSize, int quality,
+		          int image_width, int image_height, byte *image_buffer, int padding);
 void RE_TakeVideoFrame( int width, int height,
 		byte *captureBuffer, byte *encodeBuffer, qboolean motionJpeg );
 
diff --git a/src/renderer/tr_noise.c b/src/renderer/tr_noise.c
index a9707a4a..b4c4082d 100644
--- a/src/renderer/tr_noise.c
+++ b/src/renderer/tr_noise.c
@@ -32,8 +32,6 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 static float s_noise_table[NOISE_SIZE];
 static int s_noise_perm[NOISE_SIZE];
 
-#define LERP( a, b, w ) ( a * ( 1.0f - w ) + b * w )
-
 static float GetNoiseValue( int x, int y, int z, int t )
 {
 	int index = INDEX( ( int ) x, ( int ) y, ( int ) z, ( int ) t );
diff --git a/src/renderer/tr_shade.c b/src/renderer/tr_shade.c
index d4375b5e..93ec88ba 100644
--- a/src/renderer/tr_shade.c
+++ b/src/renderer/tr_shade.c
@@ -460,9 +460,18 @@ static void ProjectDlightTexture_altivec( void ) {
 		{
 			float luminance;
 			
-			luminance = (dl->color[0] * 255.0f + dl->color[1] * 255.0f + dl->color[2] * 255.0f) / 3;
+			luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
 			floatColor[0] = floatColor[1] = floatColor[2] = luminance;
 		}
+		else if(r_greyscale->value)
+		{
+			float luminance;
+			
+			luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
+			floatColor[0] = LERP(dl->color[0] * 255.0f, luminance, r_greyscale->value);
+			floatColor[1] = LERP(dl->color[1] * 255.0f, luminance, r_greyscale->value);
+			floatColor[2] = LERP(dl->color[2] * 255.0f, luminance, r_greyscale->value);
+		}
 		else
 		{
 			floatColor[0] = dl->color[0] * 255.0f;
@@ -613,10 +622,19 @@ static void ProjectDlightTexture_scalar( void ) {
 		if(r_greyscale->integer)
 		{
 			float luminance;
-			
-			luminance = (dl->color[0] * 255.0f + dl->color[1] * 255.0f + dl->color[2] * 255.0f) / 3;
+
+			luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
 			floatColor[0] = floatColor[1] = floatColor[2] = luminance;
 		}
+		else if(r_greyscale->value)
+		{
+			float luminance;
+			
+			luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
+			floatColor[0] = LERP(dl->color[0] * 255.0f, luminance, r_greyscale->value);
+			floatColor[1] = LERP(dl->color[1] * 255.0f, luminance, r_greyscale->value);
+			floatColor[2] = LERP(dl->color[2] * 255.0f, luminance, r_greyscale->value);
+		}
 		else
 		{
 			floatColor[0] = dl->color[0] * 255.0f;
@@ -965,11 +983,22 @@ static void ComputeColors( shaderStage_t *pStage )
 	if(r_greyscale->integer)
 	{
 		int scale;
+		for(i = 0; i < tess.numVertexes; i++)
+		{
+			scale = LUMA(tess.svars.colors[i][0], tess.svars.colors[i][1], tess.svars.colors[i][2]);
+ 			tess.svars.colors[i][0] = tess.svars.colors[i][1] = tess.svars.colors[i][2] = scale;
+		}
+	}
+	else if(r_greyscale->value)
+	{
+		float scale;
 		
 		for(i = 0; i < tess.numVertexes; i++)
 		{
-			scale = (tess.svars.colors[i][0] + tess.svars.colors[i][1] + tess.svars.colors[i][2]) / 3;
-			tess.svars.colors[i][0] = tess.svars.colors[i][1] = tess.svars.colors[i][2] = scale;
+			scale = LUMA(tess.svars.colors[i][0], tess.svars.colors[i][1], tess.svars.colors[i][2]);
+			tess.svars.colors[i][0] = LERP(tess.svars.colors[i][0], scale, r_greyscale->value);
+			tess.svars.colors[i][1] = LERP(tess.svars.colors[i][1], scale, r_greyscale->value);
+			tess.svars.colors[i][2] = LERP(tess.svars.colors[i][2], scale, r_greyscale->value);
 		}
 	}
 }
diff --git a/src/renderer/tr_shader.c b/src/renderer/tr_shader.c
index bfcd47d3..22e340c6 100644
--- a/src/renderer/tr_shader.c
+++ b/src/renderer/tr_shader.c
@@ -31,7 +31,6 @@ static char *s_shaderText;
 static	shaderStage_t	stages[MAX_SHADER_STAGES];		
 static	shader_t		shader;
 static	texModInfo_t	texMods[MAX_SHADER_STAGES][TR_MAX_TEXMODS];
-static	qboolean		deferLoad;
 
 #define FILE_HASH_SIZE		1024
 static	shader_t*		hashTable[FILE_HASH_SIZE];
@@ -1035,7 +1034,7 @@ static qboolean ParseStage( shaderStage_t *stage, char **text )
 	}
 
 	// decide which agens we can skip
-	if ( stage->alphaGen == CGEN_IDENTITY ) {
+	if ( stage->alphaGen == AGEN_IDENTITY ) {
 		if ( stage->rgbGen == CGEN_IDENTITY
 			|| stage->rgbGen == CGEN_LIGHTING_DIFFUSE ) {
 			stage->alphaGen = AGEN_SKIP;
@@ -1372,7 +1371,7 @@ surfaceparm <name>
 */
 static void ParseSurfaceParm( char **text ) {
 	char	*token;
-	int		numInfoParms = sizeof(infoParms) / sizeof(infoParms[0]);
+	int		numInfoParms = ARRAY_LEN( infoParms );
 	int		i;
 
 	token = COM_ParseExt( text, qfalse );
@@ -1825,7 +1824,7 @@ static qboolean CollapseMultitexture( void ) {
 			return qfalse;
 		}
 	}
-	if ( stages[0].alphaGen == CGEN_WAVEFORM )
+	if ( stages[0].alphaGen == AGEN_WAVEFORM )
 	{
 		if ( memcmp( &stages[0].alphaWave,
 					 &stages[1].alphaWave,
@@ -3071,8 +3070,6 @@ void R_InitShaders( void ) {
 
 	Com_Memset(hashTable, 0, sizeof(hashTable));
 
-	deferLoad = qfalse;
-
 	CreateInternalShaders();
 
 	ScanAndLoadShaderFiles();
diff --git a/src/renderer/tr_surface.c b/src/renderer/tr_surface.c
index cc7f0f3a..f14501f7 100644
--- a/src/renderer/tr_surface.c
+++ b/src/renderer/tr_surface.c
@@ -616,10 +616,10 @@ static void LerpMeshVertexes_altivec(md3Surface_t *surf, float backlerp)
 {
 	short	*oldXyz, *newXyz, *oldNormals, *newNormals;
 	float	*outXyz, *outNormal;
-	float	oldXyzScale ALIGN(16);
-	float   newXyzScale ALIGN(16);
-	float	oldNormalScale ALIGN(16);
-	float newNormalScale ALIGN(16);
+	float	oldXyzScale QALIGN(16);
+	float   newXyzScale QALIGN(16);
+	float	oldNormalScale QALIGN(16);
+	float newNormalScale QALIGN(16);
 	int		vertNum;
 	unsigned lat, lng;
 	int		numVerts;
diff --git a/src/renderer/tr_world.c b/src/renderer/tr_world.c
index 97c9c87e..feebb2ef 100644
--- a/src/renderer/tr_world.c
+++ b/src/renderer/tr_world.c
@@ -332,6 +332,7 @@ void R_AddBrushModelSurfaces ( trRefEntity_t *ent ) {
 		return;
 	}
 	
+	R_SetupEntityLighting( &tr.refdef, ent );
 	R_DlightBmodel( bmodel );
 
 	for ( i = 0 ; i < bmodel->numSurfaces ; i++ ) {
diff --git a/src/sdl/sdl_glimp.c b/src/sdl/sdl_glimp.c
index f065eb2d..ad30fdef 100644
--- a/src/sdl/sdl_glimp.c
+++ b/src/sdl/sdl_glimp.c
@@ -105,6 +105,19 @@ void GLimp_Shutdown( void )
 
 /*
 ===============
+GLimp_Minimize
+
+Minimize the game so that user is back at the desktop
+===============
+*/
+void GLimp_Minimize(void)
+{
+	SDL_WM_IconifyWindow();
+}
+
+
+/*
+===============
 GLimp_LogComment
 ===============
 */
diff --git a/src/sdl/sdl_input.c b/src/sdl/sdl_input.c
index 0d1164fb..8de34078 100644
--- a/src/sdl/sdl_input.c
+++ b/src/sdl/sdl_input.c
@@ -35,8 +35,6 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 #include "../client/client.h"
 #include "../sys/sys_local.h"
 
-#define ARRAYLEN(x) (sizeof(x)/sizeof(x[0]))
-
 #ifdef MACOS_X
 // Mouse acceleration needs to be disabled
 #define MACOS_X_ACCELERATION_HACK
@@ -70,6 +68,7 @@ static cvar_t *in_joystick          = NULL;
 static cvar_t *in_joystickDebug     = NULL;
 static cvar_t *in_joystickThreshold = NULL;
 static cvar_t *in_joystickNo        = NULL;
+static cvar_t *in_joystickUseAnalog = NULL;
 
 static int vidRestartTime = 0;
 
@@ -612,6 +611,7 @@ struct
 {
 	qboolean buttons[16];  // !!! FIXME: these might be too many.
 	unsigned int oldaxes;
+	int oldaaxes[16];
 	unsigned int oldhats;
 } stick_state;
 
@@ -625,6 +625,7 @@ static void IN_InitJoystick( void )
 {
 	int i = 0;
 	int total = 0;
+	char buf[16384] = "";
 
 	if (stick != NULL)
 		SDL_JoystickClose(stick);
@@ -632,11 +633,6 @@ static void IN_InitJoystick( void )
 	stick = NULL;
 	memset(&stick_state, '\0', sizeof (stick_state));
 
-	if( !in_joystick->integer ) {
-		Com_DPrintf( "Joystick is not active.\n" );
-		return;
-	}
-
 	if (!SDL_WasInit(SDL_INIT_JOYSTICK))
 	{
 		Com_DPrintf("Calling SDL_Init(SDL_INIT_JOYSTICK)...\n");
@@ -650,13 +646,28 @@ static void IN_InitJoystick( void )
 
 	total = SDL_NumJoysticks();
 	Com_DPrintf("%d possible joysticks\n", total);
+
+	// Print list and build cvar to allow ui to select joystick.
 	for (i = 0; i < total; i++)
-		Com_DPrintf("[%d] %s\n", i, SDL_JoystickName(i));
+	{
+		Q_strcat(buf, sizeof(buf), SDL_JoystickName(i));
+		Q_strcat(buf, sizeof(buf), "\n");
+	}
+
+	Cvar_Get( "in_availableJoysticks", buf, CVAR_ROM );
+
+	if( !in_joystick->integer ) {
+		Com_DPrintf( "Joystick is not active.\n" );
+		SDL_QuitSubSystem(SDL_INIT_JOYSTICK);
+		return;
+	}
 
 	in_joystickNo = Cvar_Get( "in_joystickNo", "0", CVAR_ARCHIVE );
 	if( in_joystickNo->integer < 0 || in_joystickNo->integer >= total )
 		Cvar_Set( "in_joystickNo", "0" );
 
+	in_joystickUseAnalog = Cvar_Get( "in_joystickUseAnalog", "0", CVAR_ARCHIVE );
+
 	stick = SDL_JoystickOpen( in_joystickNo->integer );
 
 	if (stick == NULL) {
@@ -665,11 +676,12 @@ static void IN_InitJoystick( void )
 	}
 
 	Com_DPrintf( "Joystick %d opened\n", in_joystickNo->integer );
-	Com_DPrintf( "Name:    %s\n", SDL_JoystickName(in_joystickNo->integer) );
-	Com_DPrintf( "Axes:    %d\n", SDL_JoystickNumAxes(stick) );
-	Com_DPrintf( "Hats:    %d\n", SDL_JoystickNumHats(stick) );
-	Com_DPrintf( "Buttons: %d\n", SDL_JoystickNumButtons(stick) );
-	Com_DPrintf( "Balls: %d\n", SDL_JoystickNumBalls(stick) );
+	Com_DPrintf( "Name:       %s\n", SDL_JoystickName(in_joystickNo->integer) );
+	Com_DPrintf( "Axes:       %d\n", SDL_JoystickNumAxes(stick) );
+	Com_DPrintf( "Hats:       %d\n", SDL_JoystickNumHats(stick) );
+	Com_DPrintf( "Buttons:    %d\n", SDL_JoystickNumButtons(stick) );
+	Com_DPrintf( "Balls:      %d\n", SDL_JoystickNumBalls(stick) );
+	Com_DPrintf( "Use Analog: %s\n", in_joystickUseAnalog->integer ? "Yes" : "No" );
 
 	SDL_JoystickEventState(SDL_QUERY);
 }
@@ -697,7 +709,7 @@ IN_JoyMove
 */
 static void IN_JoyMove( void )
 {
-	qboolean joy_pressed[ARRAYLEN(joy_keys)];
+	qboolean joy_pressed[ARRAY_LEN(joy_keys)];
 	unsigned int axes = 0;
 	unsigned int hats = 0;
 	int total = 0;
@@ -740,8 +752,8 @@ static void IN_JoyMove( void )
 	total = SDL_JoystickNumButtons(stick);
 	if (total > 0)
 	{
-		if (total > ARRAYLEN(stick_state.buttons))
-			total = ARRAYLEN(stick_state.buttons);
+		if (total > ARRAY_LEN(stick_state.buttons))
+			total = ARRAY_LEN(stick_state.buttons);
 		for (i = 0; i < total; i++)
 		{
 			qboolean pressed = (SDL_JoystickGetButton(stick, i) != 0);
@@ -850,11 +862,27 @@ static void IN_JoyMove( void )
 		for (i = 0; i < total; i++)
 		{
 			Sint16 axis = SDL_JoystickGetAxis(stick, i);
-			float f = ( (float) axis ) / 32767.0f;
-			if( f < -in_joystickThreshold->value ) {
-				axes |= ( 1 << ( i * 2 ) );
-			} else if( f > in_joystickThreshold->value ) {
-				axes |= ( 1 << ( ( i * 2 ) + 1 ) );
+
+			if (in_joystickUseAnalog->integer)
+			{
+				float f = ( (float) abs(axis) ) / 32767.0f;
+				
+				if( f < in_joystickThreshold->value ) axis = 0;
+
+				if ( axis != stick_state.oldaaxes[i] )
+				{
+					Com_QueueEvent( 0, SE_JOYSTICK_AXIS, i, axis, 0, NULL );
+					stick_state.oldaaxes[i] = axis;
+				}
+			}
+			else
+			{
+				float f = ( (float) axis ) / 32767.0f;
+				if( f < -in_joystickThreshold->value ) {
+					axes |= ( 1 << ( i * 2 ) );
+				} else if( f > in_joystickThreshold->value ) {
+					axes |= ( 1 << ( ( i * 2 ) + 1 ) );
+				}
 			}
 		}
 	}
@@ -1037,6 +1065,20 @@ void IN_Frame( void )
 
 /*
 ===============
+IN_InitKeyLockStates
+===============
+*/
+void IN_InitKeyLockStates( void )
+{
+	unsigned char *keystate = SDL_GetKeyState(NULL);
+
+	keys[K_SCROLLOCK].down = keystate[SDLK_SCROLLOCK];
+	keys[K_KP_NUMLOCK].down = keystate[SDLK_NUMLOCK];
+	keys[K_CAPSLOCK].down = keystate[SDLK_CAPSLOCK];
+}
+
+/*
+===============
 IN_Init
 ===============
 */
@@ -1060,7 +1102,7 @@ void IN_Init( void )
 
 	in_joystick = Cvar_Get( "in_joystick", "0", CVAR_ARCHIVE|CVAR_LATCH );
 	in_joystickDebug = Cvar_Get( "in_joystickDebug", "0", CVAR_TEMP );
-	in_joystickThreshold = Cvar_Get( "in_joystickThreshold", "0.15", CVAR_ARCHIVE );
+	in_joystickThreshold = Cvar_Get( "joy_threshold", "0.15", CVAR_ARCHIVE );
 
 #ifdef MACOS_X_ACCELERATION_HACK
 	in_disablemacosxmouseaccel = Cvar_Get( "in_disablemacosxmouseaccel", "1", CVAR_ARCHIVE );
@@ -1085,6 +1127,8 @@ void IN_Init( void )
 	Cvar_SetValue( "com_unfocused",	!( appState & SDL_APPINPUTFOCUS ) );
 	Cvar_SetValue( "com_minimized", !( appState & SDL_APPACTIVE ) );
 
+	IN_InitKeyLockStates( );
+
 	IN_InitJoystick( );
 	Com_DPrintf( "------------------------------------\n" );
 }
diff --git a/src/sdl/sdl_snd.c b/src/sdl/sdl_snd.c
index 24510329..6ab90ae0 100644
--- a/src/sdl/sdl_snd.c
+++ b/src/sdl/sdl_snd.c
@@ -100,8 +100,7 @@ static struct
 	{ AUDIO_S16MSB, "AUDIO_S16MSB" }
 };
 
-static int formatToStringTableSize =
-  sizeof( formatToStringTable ) / sizeof( formatToStringTable[ 0 ] );
+static int formatToStringTableSize = ARRAY_LEN( formatToStringTable );
 
 /*
 ===============
diff --git a/src/server/server.h b/src/server/server.h
index ad9fcde4..0475f0b9 100644
--- a/src/server/server.h
+++ b/src/server/server.h
@@ -220,10 +220,6 @@ typedef struct {
 	qboolean	connected;
 } challenge_t;
 
-
-#define	MAX_MASTERS	8				// max recipients for heartbeat packets
-
-
 // this structure will be cleared only when the game dll changes
 typedef struct {
 	qboolean	initialized;				// sv_init has completed
@@ -249,8 +245,6 @@ extern	serverStatic_t	svs;				// persistant server info across maps
 extern	server_t		sv;					// cleared each map
 extern	vm_t			*gvm;				// game virtual machine
 
-#define	MAX_MASTER_SERVERS	5
-
 extern	cvar_t	*sv_fps;
 extern	cvar_t	*sv_timeout;
 extern	cvar_t	*sv_zombietime;
@@ -276,6 +270,8 @@ extern	cvar_t	*sv_maxPing;
 extern	cvar_t	*sv_pure;
 extern	cvar_t	*sv_lanForceRate;
 extern	cvar_t	*sv_dequeuePeriod;
+extern	cvar_t	*sv_heartbeat;
+extern	cvar_t	*sv_flatline;
 
 #ifdef USE_VOIP
 extern	cvar_t	*sv_voip;
@@ -295,13 +291,11 @@ void SV_AddOperatorCommands (void);
 void SV_RemoveOperatorCommands (void);
 
 
-void SV_MasterHeartbeat (void);
 void SV_MasterShutdown (void);
 void SV_MasterGameStat( const char *data );
 
 
 
-
 //
 // sv_init.c
 //
diff --git a/src/server/sv_ccmds.c b/src/server/sv_ccmds.c
index 6768bb9c..8e76fa1f 100644
--- a/src/server/sv_ccmds.c
+++ b/src/server/sv_ccmds.c
@@ -205,9 +205,15 @@ static void SV_MapRestart_f( void ) {
 			continue;
 		}
 
-		client->state = CS_ACTIVE;
-
-		SV_ClientEnterWorld( client, &client->lastUsercmd );
+		if(client->state == CS_ACTIVE)
+			SV_ClientEnterWorld(client, &client->lastUsercmd);
+		else
+		{
+			// If we don't reset client->lastUsercmd and are restarting during map load,
+			// the client will hang because we'll use the last Usercmd from the previous map,
+			// which is wrong obviously.
+			SV_ClientEnterWorld(client, NULL);
+		}
 	}	
 
 	// run another frame to allow things to look at all the players
@@ -251,7 +257,7 @@ Examine or change the serverinfo string
 */
 static void SV_Systeminfo_f( void ) {
 	Com_Printf ("System info settings:\n");
-	Info_Print ( Cvar_InfoString( CVAR_SYSTEMINFO ) );
+	Info_Print ( Cvar_InfoString_Big( CVAR_SYSTEMINFO ) );
 }
 
 
@@ -273,7 +279,7 @@ SV_CompleteMapName
 */
 static void SV_CompleteMapName( char *args, int argNum ) {
 	if( argNum == 2 ) {
-		Field_CompleteFilename( "maps", "bsp", qtrue );
+		Field_CompleteFilename( "maps", "bsp", qtrue, qfalse );
 	}
 }
 
diff --git a/src/server/sv_client.c b/src/server/sv_client.c
index 815bf4db..a79fc2bc 100644
--- a/src/server/sv_client.c
+++ b/src/server/sv_client.c
@@ -505,7 +505,11 @@ void SV_ClientEnterWorld( client_t *client, usercmd_t *cmd ) {
 
 	client->deltaMessage = -1;
 	client->nextSnapshotTime = svs.time;	// generate a snapshot immediately
-	client->lastUsercmd = *cmd;
+
+	if(cmd)
+		memcpy(&client->lastUsercmd, cmd, sizeof(client->lastUsercmd));
+	else
+		memset(&client->lastUsercmd, '\0', sizeof(client->lastUsercmd));
 
 	// call the game begin function
 	VM_Call( gvm, GAME_CLIENT_BEGIN, client - svs.clients );
@@ -633,7 +637,7 @@ void SV_WriteDownloadToClient( client_t *cl , msg_t *msg )
 	int curindex;
 	int rate;
 	int blockspersnap;
-	int idPack = 0, missionPack = 0, unreferenced = 1;
+	int unreferenced = 1;
 	char errorMessage[1024];
 	char pakbuf[MAX_QPATH], *pakptr;
 	int numRefPaks;
@@ -641,7 +645,8 @@ void SV_WriteDownloadToClient( client_t *cl , msg_t *msg )
 	if (!*cl->downloadName)
 		return;	// Nothing being downloaded
 
-	if (!cl->download) {
+	if(!cl->download)
+	{
  		// Chop off filename extension.
 		Com_sprintf(pakbuf, sizeof(pakbuf), "%s", cl->downloadName);
 		pakptr = Q_strrchr(pakbuf, '.');
@@ -665,12 +670,6 @@ void SV_WriteDownloadToClient( client_t *cl , msg_t *msg )
 					if(!FS_FilenameCompare(Cmd_Argv(curindex), pakbuf))
 					{
 						unreferenced = 0;
-
-						// now that we know the file is referenced,
-						// check whether it's legal to download it.
-						missionPack = FS_idPak(pakbuf, "missionpack");
-						idPack = missionPack || FS_idPak(pakbuf, BASEGAME);
-
 						break;
 					}
 				}
@@ -682,7 +681,7 @@ void SV_WriteDownloadToClient( client_t *cl , msg_t *msg )
 		// We open the file here
 		if ( !(sv_allowDownload->integer & DLF_ENABLE) ||
 			(sv_allowDownload->integer & DLF_NO_UDP) ||
-			idPack || unreferenced ||
+			unreferenced ||
 			( cl->downloadSize = FS_SV_FOpenFileRead( cl->downloadName, &cl->download ) ) < 0 ) {
 			// cannot auto-download file
 			if(unreferenced)
@@ -690,16 +689,6 @@ void SV_WriteDownloadToClient( client_t *cl , msg_t *msg )
 				Com_Printf("clientDownload: %d : \"%s\" is not referenced and cannot be downloaded.\n", (int) (cl - svs.clients), cl->downloadName);
 				Com_sprintf(errorMessage, sizeof(errorMessage), "File \"%s\" is not referenced and cannot be downloaded.", cl->downloadName);
 			}
-			else if (idPack) {
-				Com_Printf("clientDownload: %d : \"%s\" cannot download id pk3 files\n", (int) (cl - svs.clients), cl->downloadName);
-				if (missionPack) {
-					Com_sprintf(errorMessage, sizeof(errorMessage), "Cannot autodownload Team Arena file \"%s\"\n"
-									"The Team Arena mission pack can be found in your local game store.", cl->downloadName);
-				}
-				else {
-					Com_sprintf(errorMessage, sizeof(errorMessage), "Cannot autodownload id pk3 file \"%s\"", cl->downloadName);
-				}
-			}
 			else if ( !(sv_allowDownload->integer & DLF_ENABLE) ||
 				(sv_allowDownload->integer & DLF_NO_UDP) ) {
 
@@ -1272,7 +1261,8 @@ void SV_ExecuteClientCommand( client_t *cl, const char *s, qboolean clientOK ) {
 
 	if (clientOK) {
 		// pass unknown strings to the game
-		if (!u->name && sv.state == SS_GAME && cl->state == CS_ACTIVE) {
+		if (!u->name && sv.state == SS_GAME && (cl->state == CS_ACTIVE || cl->state == CS_PRIMED)) {
+			Cmd_Args_Sanitize();
 			VM_Call( gvm, GAME_CLIENT_COMMAND, cl - svs.clients );
 		}
 	}
@@ -1560,7 +1550,7 @@ void SV_UserVoip( client_t *cl, msg_t *msg ) {
 
 		// Transmit this packet to the client.
 		// !!! FIXME: I don't like this queueing system.
-		if (client->queuedVoipPackets >= (sizeof (client->voipPacket) / sizeof (client->voipPacket[0]))) {
+		if (client->queuedVoipPackets >= ARRAY_LEN(client->voipPacket)) {
 			Com_Printf("Too many VoIP packets queued for client #%d\n", i);
 			continue;  // no room for another packet right now.
 		}
diff --git a/src/server/sv_game.c b/src/server/sv_game.c
index efd6265e..e6c98f9c 100644
--- a/src/server/sv_game.c
+++ b/src/server/sv_game.c
@@ -216,7 +216,7 @@ void SV_AdjustAreaPortalState( sharedEntity_t *ent, qboolean open ) {
 
 /*
 ==================
-SV_GameAreaEntities
+SV_EntityContact
 ==================
 */
 qboolean	SV_EntityContact( vec3_t mins, vec3_t maxs, const sharedEntity_t *gEnt, traceType_t type ) {
@@ -311,7 +311,7 @@ intptr_t SV_GameSystemCalls( intptr_t *args ) {
 		Cvar_Update( VMA(1) );
 		return 0;
 	case G_CVAR_SET:
-		Cvar_Set( (const char *)VMA(1), (const char *)VMA(2) );
+		Cvar_SetSafe( (const char *)VMA(1), (const char *)VMA(2) );
 		return 0;
 	case G_CVAR_VARIABLE_INTEGER_VALUE:
 		return Cvar_VariableIntegerValue( (const char *)VMA(1) );
diff --git a/src/server/sv_init.c b/src/server/sv_init.c
index 59e03464..e84ff0f4 100644
--- a/src/server/sv_init.c
+++ b/src/server/sv_init.c
@@ -618,13 +618,15 @@ SV_Init
 Only called at main exe startup, not for each game
 ===============
 */
-void SV_Init (void) {
+void SV_Init (void)
+{
+	int index;
+
 	SV_AddOperatorCommands ();
 
 	// serverinfo vars
 	Cvar_Get ("timelimit", "0", CVAR_SERVERINFO);
 	Cvar_Get ("sv_keywords", "", CVAR_SERVERINFO);
-	Cvar_Get ("protocol", va("%i", PROTOCOL_VERSION), CVAR_SERVERINFO | CVAR_ROM);
 	sv_mapname = Cvar_Get ("mapname", "nomap", CVAR_SERVERINFO | CVAR_ROM);
 	sv_privateClients = Cvar_Get ("sv_privateClients", "0", CVAR_SERVERINFO);
 	sv_hostname = Cvar_Get ("sv_hostname", "noname", CVAR_SERVERINFO | CVAR_ARCHIVE );
@@ -657,11 +659,11 @@ void SV_Init (void) {
 
 	sv_allowDownload = Cvar_Get ("sv_allowDownload", "0", CVAR_SERVERINFO);
 	Cvar_Get ("sv_dlURL", "http://downloads.tremulous.net", CVAR_SERVERINFO | CVAR_ARCHIVE);
-	sv_master[0] = Cvar_Get ("sv_master1", MASTER_SERVER_NAME, 0 );
-	sv_master[1] = Cvar_Get ("sv_master2", "", CVAR_ARCHIVE );
-	sv_master[2] = Cvar_Get ("sv_master3", "", CVAR_ARCHIVE );
-	sv_master[3] = Cvar_Get ("sv_master4", "", CVAR_ARCHIVE );
-	sv_master[4] = Cvar_Get ("sv_master5", "", CVAR_ARCHIVE );
+	
+	sv_master[0] = Cvar_Get("sv_master1", MASTER_SERVER_NAME, 0);
+	for(index = 1; index < MAX_MASTER_SERVERS; index++)
+		sv_master[index] = Cvar_Get(va("sv_master%d", index + 1), "", CVAR_ARCHIVE);
+
 	sv_reconnectlimit = Cvar_Get ("sv_reconnectlimit", "3", 0);
 	sv_showloss = Cvar_Get ("sv_showloss", "0", 0);
 	sv_padPackets = Cvar_Get ("sv_padPackets", "0", 0);
@@ -669,6 +671,8 @@ void SV_Init (void) {
 	sv_mapChecksum = Cvar_Get ("sv_mapChecksum", "", CVAR_ROM);
 	sv_lanForceRate = Cvar_Get ("sv_lanForceRate", "1", CVAR_ARCHIVE );
 	sv_dequeuePeriod = Cvar_Get ("sv_dequeuePeriod", "500", CVAR_ARCHIVE );
+	sv_heartbeat = Cvar_Get("sv_heartbeat", HEARTBEAT_FOR_MASTER, CVAR_INIT);
+	sv_flatline = Cvar_Get("sv_flatline", FLATLINE_FOR_MASTER, CVAR_INIT);
 }
 
 
diff --git a/src/server/sv_main.c b/src/server/sv_main.c
index fa009b62..014c3a96 100644
--- a/src/server/sv_main.c
+++ b/src/server/sv_main.c
@@ -31,17 +31,17 @@ serverStatic_t	svs;				// persistant server info
 server_t		sv;					// local server
 vm_t			*gvm = NULL;				// game virtual machine
 
-cvar_t	*sv_fps;				// time rate for running non-clients
+cvar_t	*sv_fps = NULL;			// time rate for running non-clients
 cvar_t	*sv_timeout;			// seconds without any message
 cvar_t	*sv_zombietime;			// seconds to sink messages after disconnect
 cvar_t	*sv_rconPassword;		// password for remote server commands
-cvar_t	*sv_privatePassword;	// password for the privateClient slots
+cvar_t	*sv_privatePassword;		// password for the privateClient slots
 cvar_t	*sv_allowDownload;
 cvar_t	*sv_maxclients;
 
 cvar_t	*sv_privateClients;		// number of clients reserved for password
 cvar_t	*sv_hostname;
-cvar_t	*sv_master[MAX_MASTER_SERVERS];		// master server ip address
+cvar_t	*sv_master[MAX_MASTER_SERVERS];	// master server ip address
 cvar_t	*sv_reconnectlimit;		// minimum seconds between connect messages
 cvar_t	*sv_showloss;			// report when usercmds are lost
 cvar_t	*sv_padPackets;			// add nop bytes to messages
@@ -56,6 +56,9 @@ cvar_t	*sv_maxPing;
 cvar_t	*sv_pure;
 cvar_t	*sv_lanForceRate; // dedicated 1 (LAN) server forces local client rates to 99999 (bug #491)
 cvar_t	*sv_dequeuePeriod;
+cvar_t  *sv_heartbeat;			// Heartbeat string that is sent to the master
+cvar_t  *sv_flatline;			// If the master server supports it we can send a flatline
+					// when server is killed
 
 /*
 =============================================================================
@@ -221,8 +224,8 @@ but not on every player enter or exit.
 ================
 */
 #define	HEARTBEAT_MSEC	300*1000
-#define	HEARTBEAT_GAME	"Tremulous"
-void SV_MasterHeartbeat( void ) {
+void SV_MasterHeartbeat(const char *message)
+{
 	static netadr_t	adr[MAX_MASTER_SERVERS][2]; // [2] for v4 and v6 address for the same address string.
 	int			i;
 	int			res;
@@ -303,9 +306,9 @@ void SV_MasterHeartbeat( void ) {
 		// ever incompatably changes
 
 		if(adr[i][0].type != NA_BAD)
-			NET_OutOfBandPrint( NS_SERVER, adr[i][0], "heartbeat %s\n", HEARTBEAT_GAME );
+			NET_OutOfBandPrint( NS_SERVER, adr[i][0], "heartbeat %s\n", message);
 		if(adr[i][1].type != NA_BAD)
-			NET_OutOfBandPrint( NS_SERVER, adr[i][1], "heartbeat %s\n", HEARTBEAT_GAME );
+			NET_OutOfBandPrint( NS_SERVER, adr[i][1], "heartbeat %s\n", message);
 	}
 }
 
@@ -319,11 +322,11 @@ Informs all masters that this server is going down
 void SV_MasterShutdown( void ) {
 	// send a hearbeat right now
 	svs.nextHeartbeatTime = -9999;
-	SV_MasterHeartbeat();
+	SV_MasterHeartbeat(sv_flatline->string);
 
 	// send it again to minimize chance of drops
 	svs.nextHeartbeatTime = -9999;
-	SV_MasterHeartbeat();
+	SV_MasterHeartbeat(sv_flatline->string);
 
 	// when the master tries to poll the server, it won't respond, so
 	// it will be removed from the list
@@ -460,7 +463,8 @@ static leakyBucket_t *SVC_BucketForAddress( netadr_t address, int burst, int per
 		interval = now - bucket->lastTime;
 
 		// Reclaim expired buckets
-		if ( bucket->lastTime > 0 && interval > ( burst * period ) ) {
+		if ( bucket->lastTime > 0 && ( interval > ( burst * period ) ||
+					interval < 0 ) ) {
 			if ( bucket->prev != NULL ) {
 				bucket->prev->next = bucket->next;
 			} else {
@@ -1003,6 +1007,29 @@ static qboolean SV_CheckPaused( void ) {
 
 /*
 ==================
+SV_FrameMsec
+Return time in millseconds until processing of the next server frame.
+==================
+*/
+int SV_FrameMsec()
+{
+	if(sv_fps)
+	{
+		int frameMsec;
+		
+		frameMsec = 1000.0f / sv_fps->value;
+		
+		if(frameMsec < sv.timeResidual)
+			return 0;
+		else
+			return frameMsec - sv.timeResidual;
+	}
+	else
+		return 1;
+}
+
+/*
+==================
 SV_Frame
 
 Player movement occurs as a result of packet events, which
@@ -1051,13 +1078,6 @@ void SV_Frame( int msec ) {
 
 	sv.timeResidual += msec;
 
-	if ( com_dedicated->integer && sv.timeResidual < frameMsec ) {
-		// NET_Sleep will give the OS time slices until either get a packet
-		// or time enough for a server frame has gone by
-		NET_Sleep(frameMsec - sv.timeResidual);
-		return;
-	}
-
 	// if time is about to hit the 32nd bit, kick all clients
 	// and clear sv.time, rather
 	// than checking for negative time wraparound everywhere.
@@ -1120,7 +1140,7 @@ void SV_Frame( int msec ) {
 	SV_SendClientMessages();
 
 	// send a heartbeat to the master if needed
-	SV_MasterHeartbeat();
+	SV_MasterHeartbeat(sv_heartbeat->string);
 }
 
 //============================================================================
diff --git a/src/server/sv_snapshot.c b/src/server/sv_snapshot.c
index bbd1fc01..cfc56958 100644
--- a/src/server/sv_snapshot.c
+++ b/src/server/sv_snapshot.c
@@ -351,7 +351,7 @@ static void SV_AddEntitiesVisibleFromPoint( vec3_t origin, clientSnapshot_t *fra
 		// entities can be flagged to be sent to a given mask of clients
 		if ( ent->r.svFlags & SVF_CLIENTMASK ) {
 			if (frame->ps.clientNum >= 32)
-				Com_Error( ERR_DROP, "SVF_CLIENTMASK: cientNum > 32\n" );
+				Com_Error( ERR_DROP, "SVF_CLIENTMASK: clientNum >= 32\n" );
 			if (~ent->r.singleClient & (1 << frame->ps.clientNum))
 				continue;
 		}
@@ -556,7 +556,7 @@ static int SV_RateMsec( client_t *client, int messageSize ) {
 			rate = sv_minRate->integer;
 	}
 
-	rateMsec = ( messageSize + HEADER_RATE_BYTES ) * 1000 / rate * com_timescale->value;
+	rateMsec = ( messageSize + HEADER_RATE_BYTES ) * 1000 / ((int) (rate * com_timescale->value));
 
 	return rateMsec;
 }
@@ -585,7 +585,7 @@ void SV_SendMessageToClient( msg_t *msg, client_t *client ) {
 	// TTimo - https://zerowing.idsoftware.com/bugzilla/show_bug.cgi?id=491
 	// added sv_lanForceRate check
 	if ( client->netchan.remoteAddress.type == NA_LOOPBACK || (sv_lanForceRate->integer && Sys_IsLANAddress (client->netchan.remoteAddress)) ) {
-		client->nextSnapshotTime = svs.time + (1000.0 / sv_fps->integer * com_timescale->value);
+		client->nextSnapshotTime = svs.time + ((int) (1000.0 / sv_fps->integer * com_timescale->value));
 		return;
 	}
 	
@@ -600,15 +600,15 @@ void SV_SendMessageToClient( msg_t *msg, client_t *client ) {
 		client->rateDelayed = qtrue;
 	}
 
-	client->nextSnapshotTime = svs.time + rateMsec * com_timescale->value;
+	client->nextSnapshotTime = svs.time + ((int) (rateMsec * com_timescale->value));
 
 	// don't pile up empty snapshots while connecting
 	if ( client->state != CS_ACTIVE ) {
 		// a gigantic connection message may have already put the nextSnapshotTime
 		// more than a second away, so don't shorten it
 		// do shorten if client is downloading
-		if (!*client->downloadName && client->nextSnapshotTime < svs.time + 1000 * com_timescale->value)
-			client->nextSnapshotTime = svs.time + 1000 * com_timescale->value;
+		if (!*client->downloadName && client->nextSnapshotTime < svs.time + ((int) (1000.0 * com_timescale->value)))
+			client->nextSnapshotTime = svs.time + ((int) (1000 * com_timescale->value));
 	}
 }
 
diff --git a/src/sys/sys_local.h b/src/sys/sys_local.h
index 9f4de9cf..8f99e7c5 100644
--- a/src/sys/sys_local.h
+++ b/src/sys/sys_local.h
@@ -52,6 +52,7 @@ char *Sys_StripAppBundle( char *pwd );
 void Sys_GLimpSafeInit( void );
 void Sys_GLimpInit( void );
 void Sys_PlatformInit( void );
+void Sys_PlatformExit( void );
 void Sys_SigHandler( int signal );
 void Sys_ErrorDialog( const char *error );
 void Sys_AnsiColorPrint( const char *msg );
diff --git a/src/sys/sys_main.c b/src/sys/sys_main.c
index 53fc57fb..46a795eb 100644
--- a/src/sys/sys_main.c
+++ b/src/sys/sys_main.c
@@ -208,6 +208,8 @@ static void Sys_Exit( int exitCode )
 		remove( Sys_PIDFileName( ) );
 	}
 
+	Sys_PlatformExit( );
+
 	exit( exitCode );
 }
 
@@ -408,34 +410,6 @@ void Sys_UnloadDll( void *dllHandle )
 
 /*
 =================
-Sys_TryLibraryLoad
-=================
-*/
-static void* Sys_TryLibraryLoad(const char* base, const char* gamedir, const char* fname, char* fqpath )
-{
-	void* libHandle;
-	char* fn;
-
-	*fqpath = 0;
-
-	fn = FS_BuildOSPath( base, gamedir, fname );
-	Com_Printf( "Sys_LoadDll(%s)... \n", fn );
-
-	libHandle = Sys_LoadLibrary(fn);
-
-	if(!libHandle) {
-		Com_Printf( "Sys_LoadDll(%s) failed:\n\"%s\"\n", fn, Sys_LibraryError() );
-		return NULL;
-	}
-
-	Com_Printf ( "Sys_LoadDll(%s): succeeded ...\n", fn );
-	Q_strncpyz ( fqpath , fn , MAX_QPATH ) ;
-
-	return libHandle;
-}
-
-/*
-=================
 Sys_LoadDll
 
 Used to load a development dll instead of a virtual machine
@@ -443,33 +417,31 @@ Used to load a development dll instead of a virtual machine
 #2 look in fs_basepath
 =================
 */
-void *Sys_LoadDll( const char *name, char *fqpath ,
+void *Sys_LoadDll( const char *name,
 	intptr_t (**entryPoint)(int, ...),
 	intptr_t (*systemcalls)(intptr_t, ...) )
 {
 	void  *libHandle;
 	void  (*dllEntry)( intptr_t (*syscallptr)(intptr_t, ...) );
 	char  fname[MAX_OSPATH];
-	char  *basepath;
-	char  *homepath;
-	char  *gamedir;
+	char  *netpath;
 
 	assert( name );
 
-	Q_snprintf (fname, sizeof(fname), "%s" ARCH_STRING DLL_EXT, name);
+	Com_sprintf(fname, sizeof(fname), "%s" ARCH_STRING DLL_EXT, name);
 
-	// TODO: use fs_searchpaths from files.c
-	basepath = Cvar_VariableString( "fs_basepath" );
-	homepath = Cvar_VariableString( "fs_homepath" );
-	gamedir = Cvar_VariableString( "fs_game" );
+	netpath = FS_FindDll(fname);
 
-	libHandle = Sys_TryLibraryLoad(homepath, gamedir, fname, fqpath);
+	if(!netpath) {
+		Com_Printf( "Sys_LoadDll(%s) could not find it\n", fname );
+		return NULL;
+	}
 
-	if(!libHandle && basepath)
-		libHandle = Sys_TryLibraryLoad(basepath, gamedir, fname, fqpath);
+	Com_Printf( "Loading DLL file: %s\n", netpath);
+	libHandle = Sys_LoadLibrary(netpath);
 
 	if(!libHandle) {
-		Com_Printf ( "Sys_LoadDll(%s) failed to load library\n", name );
+		Com_Printf( "Sys_LoadDll(%s) failed:\n\"%s\"\n", netpath, Sys_LibraryError() );
 		return NULL;
 	}
 
diff --git a/src/sys/sys_unix.c b/src/sys/sys_unix.c
index 5ab0f8c7..abbe6237 100644
--- a/src/sys/sys_unix.c
+++ b/src/sys/sys_unix.c
@@ -56,7 +56,7 @@ char *Sys_DefaultHomePath(void)
 	{
 		if( ( p = getenv( "HOME" ) ) != NULL )
 		{
-			Q_strncpyz( homePath, p, sizeof( homePath ) );
+			Com_sprintf(homePath, sizeof(homePath), "%s%c", p, PATH_SEP);
 #ifdef MACOS_X
 			Q_strcat( homePath, sizeof( homePath ),
 					"/Library/Application Support/Tremulous" );
@@ -242,6 +242,36 @@ qboolean Sys_Mkdir( const char *path )
 
 /*
 ==================
+Sys_Mkfifo
+==================
+*/
+FILE *Sys_Mkfifo( const char *ospath )
+{
+	FILE	*fifo;
+	int	result;
+	int	fn;
+	struct	stat buf;
+
+	// if file already exists AND is a pipefile, remove it
+	if( !stat( ospath, &buf ) && S_ISFIFO( buf.st_mode ) )
+		FS_Remove( ospath );
+
+	result = mkfifo( ospath, 0600 );
+	if( result != 0 )
+		return NULL;
+
+	fifo = fopen( ospath, "w+" );
+	if( fifo )
+	{
+		fn = fileno( fifo );
+		fcntl( fn, F_SETFL, O_NONBLOCK );
+	}
+
+	return fifo;
+}
+
+/*
+==================
 Sys_Cwd
 ==================
 */
@@ -499,7 +529,7 @@ void Sys_ErrorDialog( const char *error )
 	unsigned int size;
 	int f = -1;
 	const char *homepath = Cvar_VariableString( "fs_homepath" );
-	const char *gamedir = Cvar_VariableString( "fs_gamedir" );
+	const char *gamedir = Cvar_VariableString( "fs_game" );
 	const char *fileName = "crashlog.txt";
 	char *ospath = FS_BuildOSPath( homepath, gamedir, fileName );
 
@@ -742,6 +772,17 @@ void Sys_PlatformInit( void )
 
 /*
 ==============
+Sys_PlatformExit
+
+Unix specific deinitialisation
+==============
+*/
+void Sys_PlatformExit( void )
+{
+}
+
+/*
+==============
 Sys_SetEnv
 
 set/unset environment variables (empty value removes it)
diff --git a/src/sys/sys_win32.c b/src/sys/sys_win32.c
index ad6f18eb..8ca1bf62 100644
--- a/src/sys/sys_win32.c
+++ b/src/sys/sys_win32.c
@@ -43,6 +43,10 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 // Used to determine where to store user-specific files
 static char homePath[ MAX_OSPATH ] = { 0 };
 
+#ifndef DEDICATED
+static UINT timerResolution = 0;
+#endif
+
 #ifdef __WIN64__
 void Sys_SnapVector( float *v )
 {
@@ -313,6 +317,17 @@ qboolean Sys_Mkdir( const char *path )
 }
 
 /*
+==================
+Sys_Mkfifo
+Noop on windows because named pipes do not function the same way
+==================
+*/
+FILE *Sys_Mkfifo( const char *ospath )
+{
+	return NULL;
+}
+
+/*
 ==============
 Sys_Cwd
 ==============
@@ -699,6 +714,8 @@ Windows specific initialisation
 void Sys_PlatformInit( void )
 {
 #ifndef DEDICATED
+	TIMECAPS ptc;
+	
 	const char *SDL_VIDEODRIVER = getenv( "SDL_VIDEODRIVER" );
 
 	if( SDL_VIDEODRIVER )
@@ -709,6 +726,36 @@ void Sys_PlatformInit( void )
 	}
 	else
 		SDL_VIDEODRIVER_externallySet = qfalse;
+
+	if(timeGetDevCaps(&ptc, sizeof(ptc)) == MMSYSERR_NOERROR)
+	{
+		timerResolution = ptc.wPeriodMin;
+
+		if(timerResolution > 1)
+		{
+			Com_Printf("Warning: Minimum supported timer resolution is %ums "
+				"on this system, recommended resolution 1ms\n", timerResolution);
+		}
+		
+		timeBeginPeriod(timerResolution);				
+	}
+	else
+		timerResolution = 0;
+#endif
+}
+
+/*
+==============
+Sys_PlatformExit
+
+Windows specific initialisation
+==============
+*/
+void Sys_PlatformExit( void )
+{
+#ifndef DEDICATED
+	if(timerResolution)
+		timeEndPeriod(timerResolution);
 #endif
 }
 
@@ -721,7 +768,10 @@ set/unset environment variables (empty value removes it)
 */
 void Sys_SetEnv(const char *name, const char *value)
 {
-	_putenv(va("%s=%s", name, value));
+	if(value)
+		_putenv(va("%s=%s", name, value));
+	else
+		_putenv(va("%s=", name));
 }
 
 /*
diff --git a/src/tools/lcc/cpp/tokens.c b/src/tools/lcc/cpp/tokens.c
index 90ea47f2..ad9f2f7c 100644
--- a/src/tools/lcc/cpp/tokens.c
+++ b/src/tools/lcc/cpp/tokens.c
@@ -267,7 +267,7 @@ peektokens(Tokenrow *trp, char *str)
 	if (str)
 		fprintf(stderr, "%s ", str);
 	if (tp<trp->bp || tp>trp->lp)
-		fprintf(stderr, "(tp offset %d) ", tp-trp->bp);
+		fprintf(stderr, "(tp offset %ld) ", (long int) (tp - trp->bp));
 	for (tp=trp->bp; tp<trp->lp && tp<trp->bp+32; tp++) {
 		if (tp->type!=NL) {
 			int c = tp->t[tp->len];
diff --git a/src/tools/lcc/doc/4.html b/src/tools/lcc/doc/4.html
index a2e1213f..c36f280b 100644
--- a/src/tools/lcc/doc/4.html
+++ b/src/tools/lcc/doc/4.html
@@ -523,7 +523,7 @@ signed integer. Conversions that widen unsigned integers zero-extend; those that
 signed integers sign-extend.</p>
 
 <p>The front end composes conversions between types <em>T</em><sub>1</sub> and <em>T</em><sub>2</sub>
-by widening <em>T</em><sub>1</sub> to it's &quot;supertype&quot;, if necessary, converting
+by widening <em>T</em><sub>1</sub> to its &quot;supertype&quot;, if necessary, converting
 that result to <em>T</em><sub>2</sub>'s supertype, then narrowing the result to <em>T</em><sub>2</sub>,
 if necessary. The following table lists the supertypes; omitted entries are their own
 supertypes.</p>