Re: [PATCH] x86-64: memset optimization

[Andi Kleen <ak@suse.de>]

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


> 
> Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
> 
> --- a/include/asm-x86_64/string.h	2007-08-17 15:14:32.000000000 -0700
> +++ b/include/asm-x86_64/string.h	2007-08-17 15:36:30.000000000 -0700
> @@ -42,9 +42,51 @@ extern void *__memcpy(void *to, const vo
>  		 __ret = __builtin_memcpy((dst),(src),__len);	\
>  	   __ret; }) 
>  #endif
> -
>  #define __HAVE_ARCH_MEMSET
> -void *memset(void *s, int c, size_t n);
> +void *__memset(void *s, int c, size_t n);
> +
> +/* Optimize for cases of trivial memset's
> + * Compiler should optimize away all but the case used.
> + */
> +static __always_inline void *
> +__constant_c_and_count_memset(void *s, int c, size_t count)
> +{
> +	unsigned long pattern = 0x01010101UL * (unsigned char) c;
> +
> +	switch (count) {
> +	case 0:
> +		return s;
> +	case 1:
> +		*(unsigned char *)s = pattern;
> +		return s;
> +	case 2:
> +		*(unsigned short *)s = pattern;
> +		return s;
> +	case 3:
> +		*(unsigned short *)s = pattern;
> +		*(2+(unsigned char *)s) = pattern;
> +		return s;
> +	case 4:
> +		*(unsigned long *)s = pattern;
> +		return s;
> +	case 6:
> +		*(unsigned long *)s = pattern;
> +		*(2+(unsigned short *)s) = pattern;
> +		return s;
> +	case 8:
> +		*(unsigned long *)s = pattern;
> +		*(1+(unsigned long *)s) = pattern;
> +		return s;
> +	default:
> +		return __memset(s, c, count);
> +	}
> +}
> +#define memset(s, c, count)					\
> +	(__builtin_constant_p(c)				\
> +	 ? __constant_c_and_count_memset((s),(c),(count))	\
> +	 : __memset((s),(c),(count)))
> +
> +
>  
>  #define __HAVE_ARCH_MEMMOVE
>  void * memmove(void * dest,const void *src,size_t count);
> --- a/arch/x86_64/kernel/x8664_ksyms.c	2007-08-17 15:14:32.000000000 -0700
> +++ b/arch/x86_64/kernel/x8664_ksyms.c	2007-08-17 15:44:58.000000000 -0700
> @@ -48,10 +48,12 @@ EXPORT_SYMBOL(__read_lock_failed);
>  #undef memmove
>  
>  extern void * memset(void *,int,__kernel_size_t);
> +extern void * __memset(void *,int,__kernel_size_t);
>  extern void * memcpy(void *,const void *,__kernel_size_t);
>  extern void * __memcpy(void *,const void *,__kernel_size_t);
>  
>  EXPORT_SYMBOL(memset);
> +EXPORT_SYMBOL(__memset);
>  EXPORT_SYMBOL(memcpy);
>  EXPORT_SYMBOL(__memcpy);
>  
>