new32v2, an even faster solution.

3 files changed, 25 insertions, 1 deletions

diff --git a/Makefile b/Makefile
index f3400a6..d85c7c7 100644
--- a/Makefile
+++ b/Makefile
@@ -13,6 +13,7 @@ SRC := src/bigtab.c \
        src/main.c \
        src/new.c \
        src/new32.c \
+       src/new32v2.c \
        src/null.c \
        src/ref.c
        
diff --git a/src/main.c b/src/main.c
index 8d82af5..4961879 100644
--- a/src/main.c
+++ b/src/main.c
@@ -19,6 +19,7 @@ int ref1(int i, int x, int y);
 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);
 void bigtab_init(void);
 int bigtab(int i, int x, int y);
 
@@ -36,7 +37,8 @@ static struct solfunc solfuncs[ ] = {
 	{"ref2", ref2},
 	{"bigtab", bigtab},
 	{"new", new},
-	{"new32", new32}
+	{"new32", new32},
+	{"new32v2", new32v2}
 };
 
 #define PASSES 5
diff --git a/src/new32v2.c b/src/new32v2.c
new file mode 100644
index 0000000..a755d6d
--- /dev/null
+++ b/src/new32v2.c
@@ -0,0 +1,21 @@
+#include <stdint.h>
+
+static const unsigned int divtab[244] = {
+	[1] = 0xffff + 1l,
+	[3] = 0x5555 + 1l,
+	[9] = 0x1c71 + 1l,
+	[27] = 0x097b + 1l,
+	[81] = 0x0329 + 1l,
+	[243] = 0x010d + 1l
+};
+
+#define FASTDIV(n, d) (((n) * (divtab[d])) >> 16)
+
+int new32v2(int i, int x, int y)
+{
+	int A;
+
+	A = FASTDIV(i, y);
+
+	return i + (x - 3 * FASTDIV(A + x, 3) + 3 * FASTDIV(A, 3)) * y;
+}