OpenGL2: Speedup for SSAO & blur shaders, fix sunlight normals in lightall.

4 files changed, 57 insertions, 54 deletions

diff --git a/src/renderergl2/glsl/depthblur_fp.glsl b/src/renderergl2/glsl/depthblur_fp.glsl
index 15f7be27..60a261c5 100644
--- a/src/renderergl2/glsl/depthblur_fp.glsl
+++ b/src/renderergl2/glsl/depthblur_fp.glsl
@@ -11,22 +11,23 @@ float gauss[4] = float[4](0.40, 0.24, 0.054, 0.0044);
 
 float getLinearDepth(sampler2D depthMap, const vec2 tex, const float zFarDivZNear)
 {
-		float sampleZDivW = texture2D(depthMap, tex).r;
-		return 1.0 / mix(zFarDivZNear, 1.0, sampleZDivW);
+	float sampleZDivW = texture2D(depthMap, tex).r;
+	return 1.0 / mix(zFarDivZNear, 1.0, sampleZDivW);
 }
 
-vec4 depthGaussian1D(sampler2D imageMap, sampler2D depthMap, vec2 tex, float zFarDivZNear, float zFar)
+vec4 depthGaussian1D(sampler2D imageMap, sampler2D depthMap, vec2 tex, float zFarDivZNear, float zFar, vec2 scale)
 {
-	vec2 scale = u_ViewInfo.zw;
+	float depthCenter = getLinearDepth(depthMap, tex, zFarDivZNear);
+	//scale /= zFarDivZNear * depthCenter;
+	//int blurSteps = int(float(BLUR_SIZE) / (zFarDivZNear * depthCenter));
 
 #if defined(USE_HORIZONTAL_BLUR)
-    vec2 direction = vec2(1.0, 0.0) * scale;
+    vec2 direction = vec2(scale.x, 0.0);
 #else // if defined(USE_VERTICAL_BLUR)
-	vec2 direction = vec2(0.0, 1.0) * scale;
+	vec2 direction = vec2(0.0, scale.y);
 #endif
-	
-	float depthCenter = zFar * getLinearDepth(depthMap, tex, zFarDivZNear);
-	vec2 centerSlope = vec2(dFdx(depthCenter), dFdy(depthCenter)) / vec2(dFdx(tex.x), dFdy(tex.y));
+
+	vec2 slope = vec2(dFdx(depthCenter), dFdy(depthCenter)) / vec2(dFdx(tex.x), dFdy(tex.y));
 
 #if defined(USE_GAUSS)
 	vec4 result = texture2D(imageMap, tex) * gauss[0];
@@ -36,33 +37,32 @@ vec4 depthGaussian1D(sampler2D imageMap, sampler2D depthMap, vec2 tex, float zFa
 	float total = 1.0;
 #endif
 
+	float zLimit = 5.0 / zFar;
 	int i, j;
 	for (i = 0; i < 2; i++)
 	{
 		for (j = 1; j < BLUR_SIZE; j++)
 		{
 			vec2 offset = direction * j;
-			float depthSample = zFar * getLinearDepth(depthMap, tex + offset, zFarDivZNear);
-			float depthExpected = depthCenter + dot(centerSlope, offset);
-			if(abs(depthSample - depthExpected) < 5.0)
-			{
+			float depthSample = getLinearDepth(depthMap, tex + offset, zFarDivZNear);
+			float depthExpected = depthCenter + dot(slope, offset);
+			float useSample = float(abs(depthSample - depthExpected) < zLimit);
 #if defined(USE_GAUSS)
-				result += texture2D(imageMap, tex + offset) * gauss[j];
-				total += gauss[j];
+			result += texture2D(imageMap, tex + offset) * (gauss[j] * useSample);
+			total += gauss[j] * useSample;
 #else
-				result += texture2D(imageMap, tex + offset);
-				total += 1.0;
+			result += texture2D(imageMap, tex + offset) * useSample;
+			total += useSample;
 #endif
-			}
 		}
-		
+
 		direction = -direction;
-	}	
-		
+	}
+
 	return result / total;
 }
 
 void main()
-{		
-	gl_FragColor = depthGaussian1D(u_ScreenImageMap, u_ScreenDepthMap, var_ScreenTex, u_ViewInfo.x, u_ViewInfo.y);
+{
+	gl_FragColor = depthGaussian1D(u_ScreenImageMap, u_ScreenDepthMap, var_ScreenTex, u_ViewInfo.x, u_ViewInfo.y, u_ViewInfo.zw);
 }
diff --git a/src/renderergl2/glsl/lightall_fp.glsl b/src/renderergl2/glsl/lightall_fp.glsl
index 8edbebda..f16298a3 100644
--- a/src/renderergl2/glsl/lightall_fp.glsl
+++ b/src/renderergl2/glsl/lightall_fp.glsl
@@ -292,7 +292,7 @@ void main()
 	float shadowValue = texture2D(u_ShadowMap, shadowTex).r;
 
 	// surfaces not facing the light are always shadowed
-	shadowValue *= clamp(dot(var_Normal.xyz, var_PrimaryLightDir.xyz), 0.0, 1.0);
+	shadowValue *= clamp(dot(N, var_PrimaryLightDir.xyz), 0.0, 1.0);
 
     #if defined(SHADOWMAP_MODULATE)
 	lightColor *= shadowValue * (1.0 - u_PrimaryLightAmbient.r) + u_PrimaryLightAmbient.r;
@@ -330,7 +330,7 @@ void main()
 
   #if defined(USE_PBR)
 	// diffuse rgb is base color
-	// specular red is smoothness
+	// specular red is gloss
 	// specular green is metallicness
 	float gloss = specular.r;
 	specular.rgb = specular.g * diffuse.rgb + vec3(0.04 - 0.04 * specular.g);
diff --git a/src/renderergl2/glsl/ssao_fp.glsl b/src/renderergl2/glsl/ssao_fp.glsl
index 6263284c..84f18cb7 100644
--- a/src/renderergl2/glsl/ssao_fp.glsl
+++ b/src/renderergl2/glsl/ssao_fp.glsl
@@ -1,6 +1,6 @@
 uniform sampler2D u_ScreenDepthMap;
 
-uniform vec4   u_ViewInfo; // zfar / znear, zfar
+uniform vec4   u_ViewInfo; // zfar / znear, zfar, 1/width, 1/height
 
 varying vec2   var_ScreenTex;
 
@@ -11,6 +11,7 @@ vec2(0.5784913, -0.002528916), vec2(0.192888, 0.4064181),
 vec2(-0.6335801, -0.5247476),  vec2(-0.5579782, 0.7491854),
 vec2(0.7320465, 0.6317794)
 );
+#define NUM_SAMPLES 3
 
 // Input: It uses texture coords as the random number seed.
 // Output: Random number: [0,1), that is between 0.0 and 0.999999... inclusive.
@@ -39,48 +40,47 @@ mat2 randomRotation( const vec2 p )
 
 float getLinearDepth(sampler2D depthMap, const vec2 tex, const float zFarDivZNear)
 {
-		float sampleZDivW = texture2D(depthMap, tex).r;
-		return 1.0 / mix(zFarDivZNear, 1.0, sampleZDivW);
+	float sampleZDivW = texture2D(depthMap, tex).r;
+	return 1.0 / mix(zFarDivZNear, 1.0, sampleZDivW);
 }
 
-float ambientOcclusion(sampler2D depthMap, const vec2 tex, const float zFarDivZNear, const float zFar)
+float ambientOcclusion(sampler2D depthMap, const vec2 tex, const float zFarDivZNear, const float zFar, const vec2 scale)
 {
 	float result = 0;
 
-	float sampleZ = zFar * getLinearDepth(depthMap, tex, zFarDivZNear);
+	float sampleZ = getLinearDepth(depthMap, tex, zFarDivZNear);
+	float scaleZ = zFarDivZNear * sampleZ;
 
-	vec2 expectedSlope = vec2(dFdx(sampleZ), dFdy(sampleZ)) / vec2(dFdx(tex.x), dFdy(tex.y));
-	
-	if (length(expectedSlope) > 5000.0)
+	vec2 slope = vec2(dFdx(sampleZ), dFdy(sampleZ)) / vec2(dFdx(tex.x), dFdy(tex.y));
+
+	if (length(slope) * zFar > 5000.0)
 		return 1.0;
-	
-	vec2 offsetScale = vec2(3.0 / sampleZ);
-	
+
+	vec2 offsetScale = vec2(scale * 1024.0 / scaleZ);
+
 	mat2 rmat = randomRotation(tex);
-		
+
+	float invZFar = 1.0 / zFar;
+	float zLimit = 20.0 * invZFar;
 	int i;
-	for (i = 0; i < 3; i++)
+	for (i = 0; i < NUM_SAMPLES; i++)
 	{
 		vec2 offset = rmat * poissonDisc[i] * offsetScale;
-		float sampleZ2 = zFar * getLinearDepth(depthMap, tex + offset, zFarDivZNear);
-
-		if (abs(sampleZ - sampleZ2) > 20.0)
-			result += 1.0;
-		else
-		{
-			float expectedZ = sampleZ + dot(expectedSlope, offset);
-			result += step(expectedZ - 1.0, sampleZ2);
-		}
+		float sampleDiff = getLinearDepth(depthMap, tex + offset, zFarDivZNear) - sampleZ;
+
+		bool s1 = abs(sampleDiff) > zLimit;
+		bool s2 = sampleDiff + invZFar > dot(slope, offset);
+		result += float(s1 || s2);
 	}
-	
-	result *= 0.33333;
-	
+
+	result *= 1.0 / float(NUM_SAMPLES);
+
 	return result;
 }
 
 void main()
 {
-	float result = ambientOcclusion(u_ScreenDepthMap, var_ScreenTex, u_ViewInfo.x, u_ViewInfo.y);
-			
+	float result = ambientOcclusion(u_ScreenDepthMap, var_ScreenTex, u_ViewInfo.x, u_ViewInfo.y, u_ViewInfo.zw);
+
 	gl_FragColor = vec4(vec3(result), 1.0);
 }
diff --git a/src/renderergl2/tr_backend.c b/src/renderergl2/tr_backend.c
index e0715d04..420c785b 100644
--- a/src/renderergl2/tr_backend.c
+++ b/src/renderergl2/tr_backend.c
@@ -1099,8 +1099,8 @@ const void	*RB_DrawSurfs( const void *data ) {
 			vec4_t quadVerts[4];
 			vec2_t texCoords[4];
 
-			viewInfo[2] = 1.0f / (float)(tr.quarterImage[0]->width);
-			viewInfo[3] = 1.0f / (float)(tr.quarterImage[0]->height);
+			viewInfo[2] = 1.0f / ((float)(tr.quarterImage[0]->width)  * tan(backEnd.viewParms.fovX * M_PI / 360.0f) * 2.0f);
+			viewInfo[3] = 1.0f / ((float)(tr.quarterImage[0]->height) * tan(backEnd.viewParms.fovY * M_PI / 360.0f) * 2.0f);
 
 			FBO_Bind(tr.quarterFbo[0]);
 
@@ -1128,6 +1128,9 @@ const void	*RB_DrawSurfs( const void *data ) {
 			RB_InstantQuad2(quadVerts, texCoords); //, color, shaderProgram, invTexRes);
 
 
+			viewInfo[2] = 1.0f / (float)(tr.quarterImage[0]->width);
+			viewInfo[3] = 1.0f / (float)(tr.quarterImage[0]->height);
+
 			FBO_Bind(tr.quarterFbo[1]);
 
 			qglViewport(0, 0, tr.quarterFbo[1]->width, tr.quarterFbo[1]->height);