[PATCH] optimize hweight64 for x86_64

Based on patch from David Rientjes <rientjes@google.com>, but
changed by AK.

Optimizes the 64-bit hamming weight for x86_64 processors assuming they
have fast multiplication.  Uses five fewer bitops than the generic
hweight64.  Benchmark on one EMT64 showed ~25% speedup with 2^24
consecutive calls.

Define a new ARCH_HAS_FAST_MULTIPLIER that can be set by other
architectures that can also multiply fast.
Signed-off-by: NAndi Kleen <ak@suse.de>

include/asm-x86_64/bitops.h

@@ -399,6 +399,8 @@ static __inline__ int fls(int x)
 	return r+1;
 }
 
+#define ARCH_HAS_FAST_MULTIPLIER 1
+
 #include <asm-generic/bitops/hweight.h>
 
 #endif /* __KERNEL__ */

lib/hweight.c

 #include <linux/module.h>
 #include <asm/types.h>
+#include <asm/bitops.h>
 
 /**
  * hweightN - returns the hamming weight of a N-bit word
@@ -40,14 +41,19 @@ unsigned long hweight64(__u64 w)
 #if BITS_PER_LONG == 32
 	return hweight32((unsigned int)(w >> 32)) + hweight32((unsigned int)w);
 #elif BITS_PER_LONG == 64
+#ifdef ARCH_HAS_FAST_MULTIPLIER
+	w -= (w >> 1) & 0x5555555555555555ul;
+	w =  (w & 0x3333333333333333ul) + ((w >> 2) & 0x3333333333333333ul);
+	w =  (w + (w >> 4)) & 0x0f0f0f0f0f0f0f0ful;
+	return (w * 0x0101010101010101ul) >> 56;
+#else
 	__u64 res = w - ((w >> 1) & 0x5555555555555555ul);
 	res = (res & 0x3333333333333333ul) + ((res >> 2) & 0x3333333333333333ul);
 	res = (res + (res >> 4)) & 0x0F0F0F0F0F0F0F0Ful;
 	res = res + (res >> 8);
 	res = res + (res >> 16);
 	return (res + (res >> 32)) & 0x00000000000000FFul;
-#else
-#error BITS_PER_LONG not defined
+#endif
 #endif
 }
 EXPORT_SYMBOL(hweight64);