Even faster, yet again.

3 files changed, 24 insertions, 1 deletions

diff --git a/Makefile b/Makefile
index d85c7c7..3d16add 100644
--- a/Makefile
+++ b/Makefile
@@ -14,6 +14,7 @@ SRC := src/bigtab.c \
        src/new.c \
        src/new32.c \
        src/new32v2.c \
+       src/new32v3.c \
        src/null.c \
        src/ref.c
        
diff --git a/src/main.c b/src/main.c
index 4961879..04e0bad 100644
--- a/src/main.c
+++ b/src/main.c
@@ -20,6 +20,7 @@ int ref2(int i, int x, int y);
 int new(int i, int x, int y);
 int new32(int i, int x, int y);
 int new32v2(int i, int x, int y);
+int new32v3(int i, int x, int y);
 void bigtab_init(void);
 int bigtab(int i, int x, int y);
 
@@ -38,7 +39,8 @@ static struct solfunc solfuncs[ ] = {
 	{"bigtab", bigtab},
 	{"new", new},
 	{"new32", new32},
-	{"new32v2", new32v2}
+	{"new32v2", new32v2},
+	{"new32v3", new32v3}
 };
 
 #define PASSES 5
diff --git a/src/new32v3.c b/src/new32v3.c
new file mode 100644
index 0000000..d64459d
--- /dev/null
+++ b/src/new32v3.c
@@ -0,0 +1,20 @@
+#include <stdint.h>
+
+static const unsigned int divtab[730] = {
+	[1] = 0xffff + 1l,
+	[3] = 0x5555 + 1l,
+	[9] = 0x1c71 + 1l,
+	[27] = 0x097b + 1l,
+	[81] = 0x0329 + 1l,
+	[243] = 0x010d + 1l,
+	[729] = 0x0059 + 1l
+};
+
+#define FASTDIV(n, d) (((n) * divtab[d]) >> 16)
+
+int new32v3(int i, int x, int y)
+{
+	int A;
+	A = i + x * y;
+	return A - 3 * y * (FASTDIV(A, 3 * y) - FASTDIV(i, 3 * y));
+}