[Mplayer-users] [mplayer PATCH] Probably final fix of fastmemcpy
Nick Kurshev
nickols_k at mail.ru
Thu Apr 26 12:20:10 CEST 2001
Hello!
I want to suggest new version of fastmemcpy.
What is noticeable:
1. MMX only support was added. It really speedups memory copying no Pentium-MMX, Celeron-1, P2 at least
on 10%, but it has effect only if dest is aligned (see don't applied previous patch). For K6 it's disputable since
K6 has top slow MMX unit (sorry! I didn't tested it on k6), but lets K6 owners test it. I'm ready to add
-DHAVE_MMX0 for k6.
2. Fixed bug of small_memcpy:
I've tested it on my own project (it's really best for stress testing than mplayer).
Bug: small_memcpy didn't tell to compiler that edi, esi, ecx are clobbered after movsb execution.
3. New version of fastmemcpy is portable. (It's not significand for mplayer but may be in the future?)
memcpy must have type of 3rd argument as 'size_t' but not 'unsigned int'. Some gcc-related development
systems (like os2-emx) have 'typedef unsigned long size_t'.
4. Little improvements for readability only.
Best regards! Nick
Below full text of new fast_memcpy, but if you want:
#ifndef __MPLAYER_MEMCPY
#define __MPLAYER_MEMCPY 1
#include <stddef.h>
/*
This part of code was taken by from Linux-2.4.3 and slightly modified
for MMX, MMX2, SSE instruction set. I have done it since linux uses page aligned
blocks but mplayer uses weakly ordered data and original sources can not
speedup them. Only using PREFETCHNTA and MOVNTQ together have effect!
>From IA-32 Intel Architecture Software Developer's Manual Volume 1,
Order Number 245470:
"10.4.6. Cacheability Control, Prefetch, and Memory Ordering Instructions"
Data referenced by a program can be temporal (data will be used again) or
non-temporal (data will be referenced once and not reused in the immediate
future). To make efficient use of the processor's caches, it is generally
desirable to cache temporal data and not cache non-temporal data. Overloading
the processor's caches with non-temporal data is sometimes referred to as
"polluting the caches".
The non-temporal data is written to memory with Write-Combining semantics.
The PREFETCHh instructions permits a program to load data into the processor
at a suggested cache level, so that it is closer to the processors load and
store unit when it is needed. If the data is already present in a level of
the cache hierarchy that is closer to the processor, the PREFETCHh instruction
will not result in any data movement.
But we should you PREFETCHNTA: Non-temporal data fetch data into location
close to the processor, minimizing cache pollution.
The MOVNTQ (store quadword using non-temporal hint) instruction stores
packed integer data from an MMX register to memory, using a non-temporal hint.
The MOVNTPS (store packed single-precision floating-point values using
non-temporal hint) instruction stores packed floating-point data from an
XMM register to memory, using a non-temporal hint.
The SFENCE (Store Fence) instruction controls write ordering by creating a
fence for memory store operations. This instruction guarantees that the results
of every store instruction that precedes the store fence in program order is
globally visible before any store instruction that follows the fence. The
SFENCE instruction provides an efficient way of ensuring ordering between
procedures that produce weakly-ordered data and procedures that consume that
data.
If you have questions please contact with me: Nick Kurshev: nickols_k at mail.ru.
*/
// 3dnow memcpy support from kernel 2.4.2
// by Pontscho/fresh!mindworkz
#if defined( HAVE_MMX2 ) || defined( HAVE_3DNOW ) || defined( HAVE_MMX )
#undef HAVE_MMX1
#if defined(HAVE_MMX) && !defined(HAVE_MMX2) && !defined(HAVE_3DNOW) && !defined(HAVE_SSE)
/* means: mmx v.1. Note: Since we added alignment of destinition it speedups
of memory copying on PentMMX, Celeron-1 and P2 upto 12% versus
standard (non MMX-optimized) version.
Note: on K6-2+ it speedups memory copying upto 25% and
on K7 and P3 about 500% (5 times). */
#define HAVE_MMX1
#endif
#undef HAVE_K6_2PLUS
#if !defined( HAVE_MMX2) && defined( HAVE_3DNOW)
#define HAVE_K6_2PLUS
#endif
/* for small memory blocks (<256 bytes) this version is faster */
#define small_memcpy(to,from,n)\
{\
__asm__ __volatile__(\
"rep; movsb"\
:"=D"(to), "=S"(from), "=c"(n)\
/* It's most portable way to notify compiler */\
/* that edi, esi and ecx are clobbered in asm block. */\
/* Thanks to A'rpi for hint!!! */\
:"0" (to), "1" (from),"2" (n)\
: "memory");\
}
#ifdef HAVE_SSE
#define MMREG_SIZE 16
#else
#define MMREG_SIZE 8
#endif
/* Small defines (for readability only) ;) */
#ifdef HAVE_K6_2PLUS
#define PREFETCH "prefetch"
/* On K6 femms is faster of emms. On K7 femms is directly mapped on emms. */
#define EMMS "femms"
#else
#define PREFETCH "prefetchnta"
#define EMMS "emms"
#endif
#ifdef HAVE_MMX2
#define MOVNTQ "movntq"
#else
#define MOVNTQ "movq"
#endif
inline static void * fast_memcpy(void * to, const void * from, size_t len)
{
void *retval;
int i;
retval = to;
if(len >= 0x200) /* 512-byte blocks */
{
register unsigned long int delta;
/* Align destinition to MMREG_SIZE -boundary */
delta = ((unsigned long int)to)&(MMREG_SIZE-1);
if(delta)
{
delta=MMREG_SIZE-delta;
len -= delta;
small_memcpy(to, from, delta);
}
i = len >> 6; /* len/64 */
len&=63;
#ifndef HAVE_MMX1
__asm__ __volatile__ (
PREFETCH" (%0)\n"
PREFETCH" 64(%0)\n"
PREFETCH" 128(%0)\n"
PREFETCH" 192(%0)\n"
PREFETCH" 256(%0)\n"
: : "r" (from) );
#endif
/*
This algorithm is top effective when the code consequently
reads and writes blocks which have size of cache line.
Size of cache line is processor-dependent.
It will, however, be a minimum of 32 bytes on any processors.
It would be better to have a number of instructions which
perform reading and writing to be multiple to a number of
processor's decoders, but it's not always possible.
*/
#ifdef HAVE_SSE /* Only P3 (may be Cyrix3) */
if(((unsigned long)from) & 15)
/* if SRC is misaligned */
for(; i>0; i--)
{
__asm__ __volatile__ (
PREFETCH" 320(%0)\n"
"movups (%0), %%xmm0\n"
"movups 16(%0), %%xmm1\n"
"movntps %%xmm0, (%1)\n"
"movntps %%xmm1, 16(%1)\n"
"movups 32(%0), %%xmm0\n"
"movups 48(%0), %%xmm1\n"
"movntps %%xmm0, 32(%1)\n"
"movntps %%xmm1, 48(%1)\n"
:: "r" (from), "r" (to) : "memory");
((const unsigned char *)from)+=64;
((unsigned char *)to)+=64;
}
else
/*
Only if SRC is aligned on 16-byte boundary.
It allows to use movaps instead of movups, which required data
to be aligned or a general-protection exception (#GP) is generated.
*/
for(; i>0; i--)
{
__asm__ __volatile__ (
PREFETCH" 320(%0)\n"
"movaps (%0), %%xmm0\n"
"movaps 16(%0), %%xmm1\n"
"movntps %%xmm0, (%1)\n"
"movntps %%xmm1, 16(%1)\n"
"movaps 32(%0), %%xmm0\n"
"movaps 48(%0), %%xmm1\n"
"movntps %%xmm0, 32(%1)\n"
"movntps %%xmm1, 48(%1)\n"
:: "r" (from), "r" (to) : "memory");
((const unsigned char *)from)+=64;
((unsigned char *)to)+=64;
}
#else
for(; i>0; i--)
{
__asm__ __volatile__ (
#ifndef HAVE_MMX1
PREFETCH" 320(%0)\n"
#endif
"movq (%0), %%mm0\n"
"movq 8(%0), %%mm1\n"
"movq 16(%0), %%mm2\n"
"movq 24(%0), %%mm3\n"
MOVNTQ" %%mm0, (%1)\n"
MOVNTQ" %%mm1, 8(%1)\n"
MOVNTQ" %%mm2, 16(%1)\n"
MOVNTQ" %%mm3, 24(%1)\n"
"movq 32(%0), %%mm0\n"
"movq 40(%0), %%mm1\n"
"movq 48(%0), %%mm2\n"
"movq 56(%0), %%mm3\n"
MOVNTQ" %%mm0, 32(%1)\n"
MOVNTQ" %%mm1, 40(%1)\n"
MOVNTQ" %%mm2, 48(%1)\n"
MOVNTQ" %%mm3, 56(%1)\n"
:: "r" (from), "r" (to) : "memory");
((const unsigned char *)from)+=64;
((unsigned char *)to)+=64;
}
#endif /* Have SSE */
#ifdef HAVE_MMX2
/* since movntq is weakly-ordered, a "sfence"
* is needed to become ordered again. */
__asm__ __volatile__ ("sfence":::"memory");
#endif
#ifndef HAVE_SSE
/* enables to use FPU */
__asm__ __volatile__ (EMMS:::"memory");
#endif
}
/*
* Now do the tail of the block
*/
if(len) small_memcpy(to, from, len);
return retval;
}
#define memcpy(a,b,c) fast_memcpy(a,b,c)
#undef small_memcpy
#endif
#endif
_______________________________________________
Mplayer-users mailing list
Mplayer-users at lists.sourceforge.net
http://lists.sourceforge.net/lists/listinfo/mplayer-users
More information about the MPlayer-users
mailing list