Add a divide-less version of mastur.

2 files changed, 22 insertions, 1 deletions

diff --git a/src/main.c b/src/main.c
index 6651ff3..4f7ba7f 100644
--- a/src/main.c
+++ b/src/main.c
@@ -22,6 +22,7 @@ int new32(int i, int x, int y);
 int new32v2(int i, int x, int y);
 int new32v3(int i, int x, int y);
 int mastur(int i, int x, int y);
+int mastur2(int i, int x, int y);
 void bigtab_init(void);
 int bigtab(int i, int x, int y);
 
@@ -42,7 +43,8 @@ static struct solfunc solfuncs[ ] = {
 	{"new32", new32},
 	{"new32v2", new32v2},
 	{"new32v3", new32v3},
-	{"mastur", mastur}
+	{"mastur", mastur},
+	{"mastur2", mastur2}
 };
 
 #define PASSES 5
diff --git a/src/mastur2.c b/src/mastur2.c
new file mode 100644
index 0000000..1be6d1d
--- /dev/null
+++ b/src/mastur2.c
@@ -0,0 +1,19 @@
+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)
+#define FASTMOD(n, d) ((n) - (d) * FASTDIV((n), (d)))
+
+static const int A[] = {0, 1, -1};
+static const int C[][3] = {{1, 1, -2}, {1, 1, -2}, {-2, 1, 1}};
+
+int mastur2(const int i, const int x, const int y)
+{
+	return i + A[x] * C[x][FASTMOD(FASTDIV(i, y), 3)] * y;
+}