Re: [PATCH] x86-64: memset optimization

[Stephen Hemminger <shemminger@linux-foundation.org>]

On Sat, 18 Aug 2007 11:46:24 +0200
Andi Kleen <ak@suse.de> wrote:

> On Saturday 18 August 2007 01:34:46 Stephen Hemminger wrote:
> > Optimize uses of memset with small constant offsets.
> > This will generate smaller code, and avoid the slow rep/string instructions.
> > Code copied from i386 with a little cleanup.
> 
> 
> Newer gcc should do all this on its own.  That is why I intentionally
> didn't implement it on 64bit.
> 
> On what compiler version did you see smaller code?
> 
> -Andi
> 

The problem is that on x86-64 you are overriding memset() so the builtin
version doesn't kick in.  You allow gcc to inline memcpy but not memset.

What about adding code similar to memcpy() stuff.

--- a/include/asm-x86_64/string.h	2007-08-18 07:37:58.000000000 -0700
+++ b/include/asm-x86_64/string.h	2007-08-18 07:44:31.000000000 -0700
@@ -43,8 +43,13 @@ extern void *__memcpy(void *to, const vo
 	   __ret; }) 
 #endif
 
-#define __HAVE_ARCH_MEMSET
-void *memset(void *s, int c, size_t n);
+#define __HAVE_ARCH_MEMSET 1
+#if (__GNUC__ == 4 && __GNUC_MINOR__ >= 3) || __GNUC__ > 4
+extern void memset(void *s, int c, size_t n);
+#else
+#define memset(s, c, n) __builtin_memset((s),(c),(n))
+#endif
+
 
 #define __HAVE_ARCH_MEMMOVE
 void * memmove(void * dest,const void *src,size_t count);
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/