Re: [Libvisual-devel] Next release cycle plans!

[Vitaly V. B. <vit...@uk...>]

On 11 Feb 2005 20:44:43 -0300
Duilio Javier Protti <dp...@fc...> wrote:

> > 	Libvisual:
> > 		Setup our own memcpy/memset wrappers and accelerate
> > 			these with mmx/sse etc etc.
> 
> I see on the Linux kernel (include/asm-i386/string.h):
> 
> static __inline__ void *__memcpy3d(void *to, const void *from, size_t
> len)
> {
> 	if (len < 512)
> 		return __constant_memcpy(to, from, len);
> 	return _mmx_memcpy(to, from, len);
> }
> 
> so it looks like MMX pays well on buffers > 512 bytes long
That's kernel :) 

Glibc has it's own mem* functions, compiler can compile it's
own inlined mem* versions. Compiler is the most interesting player
here especially if data size is a constant.

mmx/sse memcpy and memset can be useful while working with a pretty
large framebuffers.

-- 
Vitaly
GPG Key ID: F95A23B9