Optimize the parity generation.

diff --git a/hamming32.c b/hamming32.c
index fb53e3b2e9dfa48c0f908ba304092f56b44ccc9a..ea6605850680708b124120ddcf05bc4a57c248ff 100644 (file)
--- a/hamming32.c
+++ b/hamming32.c
@@ -14,14 +14,24 @@ unsigned char permutation_table[CODE_BITS] = {
        0, 5, 4, 31, 3, 30, 29, 28, 2, 27, 26, 25, 24, 23, 22, 21, 1, 20, 19, 18, 17, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6
 };
 
-unsigned find_parity_32(unsigned data)
+unsigned find_parity_32(unsigned x)
 {
-       data = (data >> 16) ^ data;
-       data = (data >> 8) ^ data;
-       data = (data >> 4) ^ data;
-       data = (data >> 2) ^ data;
-       data = (data >> 1) ^ data;
-       return (data & 1);
+#if 0
+       /* 
+        * This variant seems to be slightly faster, but depends on
+        * fast hardware multiplication.
+        */
+       x = x ^ (x >> 1);
+       x = (x ^ (x >> 2)) & 0x11111111;
+       x = x * 0x11111111;
+       return (x >> 28) & 1;
+#else
+       x ^= x >> 16;
+       x ^= x >> 8;
+       x ^= x >> 4;
+       x &= 0xf;
+       return (0x6996 >> x) & 1;
+#endif
 }
 
 unsigned generate_parity(unsigned data)