mirror of
https://github.com/mpv-player/mpv
synced 2024-12-28 10:02:17 +00:00
runtime cpu detection
git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@3394 b3059339-0415-0410-9bf9-f77b7e298cf2
This commit is contained in:
michael
parent
5830ab523a
commit
1ec539d353
425
libvo/aclib.c
425
libvo/aclib.c
@ -1,12 +1,14 @@
|
||||
#include "../config.h"
|
||||
|
||||
#ifdef USE_FASTMEMCPY
|
||||
|
||||
/*
|
||||
aclib - advanced C library ;)
|
||||
This file contains functions which improve and expand standard C-library
|
||||
see aclib_template.c ... this file only contains runtime cpu detection and config options stuff
|
||||
runtime cpu detection by michael niedermayer (michaelni@gmx.at) is under GPL
|
||||
*/
|
||||
|
||||
#include <stddef.h>
|
||||
#include "../cpudetect.h"
|
||||
|
||||
#define BLOCK_SIZE 4096
|
||||
#define CONFUSION_FACTOR 0
|
||||
@ -14,353 +16,106 @@
|
||||
|
||||
//#define STATISTICS
|
||||
|
||||
#ifndef HAVE_SSE2
|
||||
#ifdef ARCH_X86
|
||||
#define CAN_COMPILE_X86_ASM
|
||||
#endif
|
||||
|
||||
//Note: we have MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one
|
||||
//Plain C versions
|
||||
//#if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
|
||||
//#define COMPILE_C
|
||||
//#endif
|
||||
|
||||
#ifdef CAN_COMPILE_X86_ASM
|
||||
|
||||
#if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
|
||||
#define COMPILE_MMX
|
||||
#endif
|
||||
|
||||
#if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
|
||||
#define COMPILE_MMX2
|
||||
#endif
|
||||
|
||||
#if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
|
||||
#define COMPILE_3DNOW
|
||||
#endif
|
||||
|
||||
#undef HAVE_MMX
|
||||
#undef HAVE_MMX2
|
||||
#undef HAVE_3DNOW
|
||||
#undef ARCH_X86
|
||||
/*
|
||||
P3 processor has only one SSE decoder so can execute only 1 sse insn per
|
||||
cpu clock, but it has 3 mmx decoders (include load/store unit)
|
||||
and executes 3 mmx insns per cpu clock.
|
||||
P4 processor has some chances, but after reading:
|
||||
http://www.emulators.com/pentium4.htm
|
||||
I have doubts. Anyway SSE2 version of this code can be written better.
|
||||
#ifdef COMPILE_C
|
||||
#undef HAVE_MMX
|
||||
#undef HAVE_MMX2
|
||||
#undef HAVE_3DNOW
|
||||
#undef ARCH_X86
|
||||
#define RENAME(a) a ## _C
|
||||
#include "aclib_template.c"
|
||||
#endif
|
||||
*/
|
||||
#undef HAVE_SSE
|
||||
//MMX versions
|
||||
#ifdef COMPILE_MMX
|
||||
#undef RENAME
|
||||
#define HAVE_MMX
|
||||
#undef HAVE_MMX2
|
||||
#undef HAVE_3DNOW
|
||||
#define ARCH_X86
|
||||
#define RENAME(a) a ## _MMX
|
||||
#include "aclib_template.c"
|
||||
#endif
|
||||
|
||||
|
||||
/*
|
||||
This part of code was taken by me 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@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
|
||||
//MMX2 versions
|
||||
#ifdef COMPILE_MMX2
|
||||
#undef RENAME
|
||||
#define HAVE_MMX
|
||||
#define HAVE_MMX2
|
||||
#undef HAVE_3DNOW
|
||||
#define ARCH_X86
|
||||
#define RENAME(a) a ## _MMX2
|
||||
#include "aclib_template.c"
|
||||
#endif
|
||||
|
||||
|
||||
#undef HAVE_K6_2PLUS
|
||||
#if !defined( HAVE_MMX2) && defined( HAVE_3DNOW)
|
||||
#define HAVE_K6_2PLUS
|
||||
//3DNOW versions
|
||||
#ifdef COMPILE_3DNOW
|
||||
#undef RENAME
|
||||
#define HAVE_MMX
|
||||
#undef HAVE_MMX2
|
||||
#define HAVE_3DNOW
|
||||
#define ARCH_X86
|
||||
#define RENAME(a) a ## _3DNow
|
||||
#include "aclib_template.c"
|
||||
#endif
|
||||
|
||||
/* for small memory blocks (<256 bytes) this version is faster */
|
||||
#define small_memcpy(to,from,n)\
|
||||
{\
|
||||
register unsigned long int dummy;\
|
||||
__asm__ __volatile__(\
|
||||
"rep; movsb"\
|
||||
:"=&D"(to), "=&S"(from), "=&c"(dummy)\
|
||||
/* 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");\
|
||||
}
|
||||
#endif // CAN_COMPILE_X86_ASM
|
||||
|
||||
#ifdef HAVE_SSE
|
||||
#define MMREG_SIZE 16
|
||||
#else
|
||||
#define MMREG_SIZE 64 //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
|
||||
|
||||
#ifdef HAVE_MMX1
|
||||
#define MIN_LEN 0x800 /* 2K blocks */
|
||||
#else
|
||||
#define MIN_LEN 0x40 /* 64-byte blocks */
|
||||
#endif
|
||||
|
||||
void * fast_memcpy(void * to, const void * from, size_t len)
|
||||
inline void * fast_memcpy(void * to, const void * from, size_t len)
|
||||
{
|
||||
void *retval;
|
||||
size_t i;
|
||||
retval = to;
|
||||
#ifdef STATISTICS
|
||||
{
|
||||
static int freq[33];
|
||||
static int t=0;
|
||||
int i;
|
||||
for(i=0; len>(1<<i); i++);
|
||||
freq[i]++;
|
||||
t++;
|
||||
if(1024*1024*1024 % t == 0)
|
||||
for(i=0; i<32; i++)
|
||||
printf("freq < %8d %4d\n", 1<<i, freq[i]);
|
||||
}
|
||||
#endif
|
||||
#ifndef HAVE_MMX1
|
||||
/* PREFETCH has effect even for MOVSB instruction ;) */
|
||||
__asm__ __volatile__ (
|
||||
PREFETCH" (%0)\n"
|
||||
PREFETCH" 64(%0)\n"
|
||||
PREFETCH" 128(%0)\n"
|
||||
PREFETCH" 192(%0)\n"
|
||||
PREFETCH" 256(%0)\n"
|
||||
: : "r" (from) );
|
||||
#endif
|
||||
if(len >= MIN_LEN)
|
||||
{
|
||||
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;
|
||||
/*
|
||||
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"
|
||||
"movups 32(%0), %%xmm2\n"
|
||||
"movups 48(%0), %%xmm3\n"
|
||||
"movntps %%xmm0, (%1)\n"
|
||||
"movntps %%xmm1, 16(%1)\n"
|
||||
"movntps %%xmm2, 32(%1)\n"
|
||||
"movntps %%xmm3, 48(%1)\n"
|
||||
:: "r" (from), "r" (to) : "memory");
|
||||
((const unsigned char *)from)+=64;
|
||||
((unsigned char *)to)+=64;
|
||||
}
|
||||
#ifdef RUNTIME_CPUDETECT
|
||||
#ifdef CAN_COMPILE_X86_ASM
|
||||
// ordered per speed fasterst first
|
||||
if(gCpuCaps.hasMMX2)
|
||||
fast_memcpy_MMX2(to, from, len);
|
||||
else if(gCpuCaps.has3DNow)
|
||||
fast_memcpy_3DNow(to, from, len);
|
||||
else if(gCpuCaps.hasMMX)
|
||||
fast_memcpy_MMX(to, from, len);
|
||||
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"
|
||||
"movaps 32(%0), %%xmm2\n"
|
||||
"movaps 48(%0), %%xmm3\n"
|
||||
"movntps %%xmm0, (%1)\n"
|
||||
"movntps %%xmm1, 16(%1)\n"
|
||||
"movntps %%xmm2, 32(%1)\n"
|
||||
"movntps %%xmm3, 48(%1)\n"
|
||||
:: "r" (from), "r" (to) : "memory");
|
||||
((const unsigned char *)from)+=64;
|
||||
((unsigned char *)to)+=64;
|
||||
}
|
||||
#endif //CAN_COMPILE_X86_ASM
|
||||
memcpy(to, from, len); // prior to mmx we use the standart memcpy
|
||||
#else
|
||||
// Align destination at BLOCK_SIZE boundary
|
||||
for(; ((int)to & (BLOCK_SIZE-1)) && 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"
|
||||
"movq 32(%0), %%mm4\n"
|
||||
"movq 40(%0), %%mm5\n"
|
||||
"movq 48(%0), %%mm6\n"
|
||||
"movq 56(%0), %%mm7\n"
|
||||
MOVNTQ" %%mm0, (%1)\n"
|
||||
MOVNTQ" %%mm1, 8(%1)\n"
|
||||
MOVNTQ" %%mm2, 16(%1)\n"
|
||||
MOVNTQ" %%mm3, 24(%1)\n"
|
||||
MOVNTQ" %%mm4, 32(%1)\n"
|
||||
MOVNTQ" %%mm5, 40(%1)\n"
|
||||
MOVNTQ" %%mm6, 48(%1)\n"
|
||||
MOVNTQ" %%mm7, 56(%1)\n"
|
||||
:: "r" (from), "r" (to) : "memory");
|
||||
((const unsigned char *)from)+=64;
|
||||
((unsigned char *)to)+=64;
|
||||
}
|
||||
|
||||
// printf(" %d %d\n", (int)from&1023, (int)to&1023);
|
||||
// Pure Assembly cuz gcc is a bit unpredictable ;)
|
||||
if(i>=BLOCK_SIZE/64)
|
||||
asm volatile(
|
||||
"xorl %%eax, %%eax \n\t"
|
||||
".balign 16 \n\t"
|
||||
"1: \n\t"
|
||||
"movl (%0, %%eax), %%ebx \n\t"
|
||||
"movl 32(%0, %%eax), %%ebx \n\t"
|
||||
"movl 64(%0, %%eax), %%ebx \n\t"
|
||||
"movl 96(%0, %%eax), %%ebx \n\t"
|
||||
"addl $128, %%eax \n\t"
|
||||
"cmpl %3, %%eax \n\t"
|
||||
" jb 1b \n\t"
|
||||
|
||||
"xorl %%eax, %%eax \n\t"
|
||||
|
||||
".balign 16 \n\t"
|
||||
"2: \n\t"
|
||||
"movq (%0, %%eax), %%mm0\n"
|
||||
"movq 8(%0, %%eax), %%mm1\n"
|
||||
"movq 16(%0, %%eax), %%mm2\n"
|
||||
"movq 24(%0, %%eax), %%mm3\n"
|
||||
"movq 32(%0, %%eax), %%mm4\n"
|
||||
"movq 40(%0, %%eax), %%mm5\n"
|
||||
"movq 48(%0, %%eax), %%mm6\n"
|
||||
"movq 56(%0, %%eax), %%mm7\n"
|
||||
MOVNTQ" %%mm0, (%1, %%eax)\n"
|
||||
MOVNTQ" %%mm1, 8(%1, %%eax)\n"
|
||||
MOVNTQ" %%mm2, 16(%1, %%eax)\n"
|
||||
MOVNTQ" %%mm3, 24(%1, %%eax)\n"
|
||||
MOVNTQ" %%mm4, 32(%1, %%eax)\n"
|
||||
MOVNTQ" %%mm5, 40(%1, %%eax)\n"
|
||||
MOVNTQ" %%mm6, 48(%1, %%eax)\n"
|
||||
MOVNTQ" %%mm7, 56(%1, %%eax)\n"
|
||||
"addl $64, %%eax \n\t"
|
||||
"cmpl %3, %%eax \n\t"
|
||||
"jb 2b \n\t"
|
||||
|
||||
#if CONFUSION_FACTOR > 0
|
||||
// a few percent speedup on out of order executing CPUs
|
||||
"movl %5, %%eax \n\t"
|
||||
"2: \n\t"
|
||||
"movl (%0), %%ebx \n\t"
|
||||
"movl (%0), %%ebx \n\t"
|
||||
"movl (%0), %%ebx \n\t"
|
||||
"movl (%0), %%ebx \n\t"
|
||||
"decl %%eax \n\t"
|
||||
" jnz 2b \n\t"
|
||||
#endif
|
||||
|
||||
"xorl %%eax, %%eax \n\t"
|
||||
"addl %3, %0 \n\t"
|
||||
"addl %3, %1 \n\t"
|
||||
"subl %4, %2 \n\t"
|
||||
"cmpl %4, %2 \n\t"
|
||||
" jae 1b \n\t"
|
||||
: "+r" (from), "+r" (to), "+r" (i)
|
||||
: "r" (BLOCK_SIZE), "i" (BLOCK_SIZE/64), "i" (CONFUSION_FACTOR)
|
||||
: "%eax", "%ebx"
|
||||
);
|
||||
|
||||
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"
|
||||
"movq 32(%0), %%mm4\n"
|
||||
"movq 40(%0), %%mm5\n"
|
||||
"movq 48(%0), %%mm6\n"
|
||||
"movq 56(%0), %%mm7\n"
|
||||
MOVNTQ" %%mm0, (%1)\n"
|
||||
MOVNTQ" %%mm1, 8(%1)\n"
|
||||
MOVNTQ" %%mm2, 16(%1)\n"
|
||||
MOVNTQ" %%mm3, 24(%1)\n"
|
||||
MOVNTQ" %%mm4, 32(%1)\n"
|
||||
MOVNTQ" %%mm5, 40(%1)\n"
|
||||
MOVNTQ" %%mm6, 48(%1)\n"
|
||||
MOVNTQ" %%mm7, 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");
|
||||
fast_memcpy_MMX2(to, from, len);
|
||||
#elif defined (HAVE_3DNOW)
|
||||
fast_memcpy_3DNow(to, from, len);
|
||||
#elif defined (HAVE_MMX)
|
||||
fast_memcpy_MMX(to, from, len);
|
||||
#else
|
||||
memcpy(to, from, len); // prior to mmx we use the standart memcpy
|
||||
#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;
|
||||
|
||||
#endif //!RUNTIME_CPUDETECT
|
||||
}
|
||||
|
||||
|
||||
#endif /* #if defined( HAVE_MMX2 ) || defined( HAVE_3DNOW ) || defined( HAVE_MMX ) */
|
||||
#endif /* USE_FASTMEMCPY */
|
||||
#endif /* use fastmemcpy */
|
||||
@ -1,19 +1,8 @@
|
||||
#include "../config.h"
|
||||
|
||||
#ifdef USE_FASTMEMCPY
|
||||
/*
|
||||
aclib - advanced C library ;)
|
||||
This file contains functions which improve and expand standard C-library
|
||||
*/
|
||||
|
||||
#include <stddef.h>
|
||||
|
||||
#define BLOCK_SIZE 4096
|
||||
#define CONFUSION_FACTOR 0
|
||||
//Feel free to fine-tune the above 2, it might be possible to get some speedup with them :)
|
||||
|
||||
//#define STATISTICS
|
||||
|
||||
#ifndef HAVE_SSE2
|
||||
/*
|
||||
P3 processor has only one SSE decoder so can execute only 1 sse insn per
|
||||
@ -74,7 +63,6 @@ If you have questions please contact with me: Nick Kurshev: nickols_k@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)
|
||||
@ -106,6 +94,7 @@ __asm__ __volatile__(\
|
||||
: "memory");\
|
||||
}
|
||||
|
||||
#undef MMREG_SIZE
|
||||
#ifdef HAVE_SSE
|
||||
#define MMREG_SIZE 16
|
||||
#else
|
||||
@ -113,6 +102,8 @@ __asm__ __volatile__(\
|
||||
#endif
|
||||
|
||||
/* Small defines (for readability only) ;) */
|
||||
#undef PREFETCH
|
||||
#undef EMMS
|
||||
#ifdef HAVE_K6_2PLUS
|
||||
#define PREFETCH "prefetch"
|
||||
/* On K6 femms is faster of emms. On K7 femms is directly mapped on emms. */
|
||||
@ -122,19 +113,21 @@ __asm__ __volatile__(\
|
||||
#define EMMS "emms"
|
||||
#endif
|
||||
|
||||
#undef MOVNTQ
|
||||
#ifdef HAVE_MMX2
|
||||
#define MOVNTQ "movntq"
|
||||
#else
|
||||
#define MOVNTQ "movq"
|
||||
#endif
|
||||
|
||||
#undef MIN_LEN
|
||||
#ifdef HAVE_MMX1
|
||||
#define MIN_LEN 0x800 /* 2K blocks */
|
||||
#else
|
||||
#define MIN_LEN 0x40 /* 64-byte blocks */
|
||||
#endif
|
||||
|
||||
void * fast_memcpy(void * to, const void * from, size_t len)
|
||||
static inline void * RENAME(fast_memcpy)(void * to, const void * from, size_t len)
|
||||
{
|
||||
void *retval;
|
||||
size_t i;
|
||||
@ -360,7 +353,3 @@ void * fast_memcpy(void * to, const void * from, size_t len)
|
||||
if(len) small_memcpy(to, from, len);
|
||||
return retval;
|
||||
}
|
||||
|
||||
|
||||
#endif /* #if defined( HAVE_MMX2 ) || defined( HAVE_3DNOW ) || defined( HAVE_MMX ) */
|
||||
#endif /* USE_FASTMEMCPY */
|
||||
|
||||
Loading…
Reference in New Issue
Block a user