Commit 1207a369 authored by Rafaël Carré's avatar Rafaël Carré Committed by Rémi Denis-Courmont

Remove vlc_memcpy

Signed-off-by: Rémi Denis-Courmont's avatarRémi Denis-Courmont <remi@remlab.net>
parent 31c8cef1
......@@ -4233,8 +4233,6 @@ AC_CONFIG_FILES([
modules/visualization/Makefile
modules/visualization/visual/Makefile
modules/mmx/Makefile
modules/mmxext/Makefile
modules/3dnow/Makefile
modules/sse2/Makefile
modules/altivec/Makefile
modules/arm_neon/Makefile
......
......@@ -878,9 +878,6 @@ static inline void *vlc_memalign(size_t align, size_t size)
VLC_API void vlc_tdestroy( void *, void (*)(void *) );
/* Fast large memory copy */
VLC_API void * vlc_memcpy( void *, const void *, size_t );
/*****************************************************************************
* I18n stuff
*****************************************************************************/
......
......@@ -105,9 +105,5 @@ VLC_API unsigned vlc_CPU( void );
# endif
typedef void *(*vlc_memcpy_t) (void *tgt, const void *src, size_t n);
VLC_API void vlc_fastmem_register(vlc_memcpy_t cpy);
#endif /* !VLC_CPU_H */
libmemcpy3dn_plugin_la_SOURCES = memcpy.c ../mmx/fastmemcpy.h
libmemcpy3dn_plugin_la_CFLAGS = $(AM_CFLAGS)
libmemcpy3dn_plugin_la_LIBADD = $(AM_LIBADD)
libmemcpy3dn_plugin_la_DEPENDENCIES =
if HAVE_WIN32
libvlc_LTLIBRARIES += \
libmemcpy3dn_plugin.la \
$(NULL)
endif
/*****************************************************************************
* memcpy.c : classic memcpy module
*****************************************************************************
* Copyright (C) 2001 the VideoLAN team
* $Id$
*
* Authors: Samuel Hocevar <sam@zoy.org>
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
*****************************************************************************/
#ifdef HAVE_CONFIG_H
# include "config.h"
#endif
#include <vlc_common.h>
#include <vlc_plugin.h>
#include <vlc_cpu.h>
#define HAVE_3DNOW
#include "../mmx/fastmemcpy.h"
static int Activate( vlc_object_t *p_this )
{
if( !(vlc_CPU() & CPU_CAPABILITY_3DNOW) )
return VLC_EGENERIC;
VLC_UNUSED(p_this);
vlc_fastmem_register( fast_memcpy );
return VLC_SUCCESS;
}
vlc_module_begin ()
set_category( CAT_ADVANCED )
set_subcategory( SUBCAT_ADVANCED_MISC )
set_description( N_("3D Now! memcpy") )
add_shortcut( "3dn", "3dnow", "memcpy3dn", "memcpy3dnow" )
set_capability( "memcpy", 100 )
set_callbacks( Activate, NULL )
vlc_module_end ()
......@@ -25,9 +25,7 @@ EXTRA_SUBDIRS = \
mux \
stream_out \
mmx \
mmxext \
sse2 \
3dnow \
altivec \
arm_neon \
lua \
......@@ -41,15 +39,9 @@ endif
if HAVE_MMX
SUBDIRS += mmx
endif
if HAVE_MMXEXT
SUBDIRS += mmxext
endif
if HAVE_SSE2
SUBDIRS += sse2
endif
if HAVE_3DNOW
SUBDIRS += 3dnow
endif
if HAVE_ALTIVEC
SUBDIRS += altivec
endif
......
libmemcpyaltivec_plugin_la_SOURCES = memcpy.c
libmemcpyaltivec_plugin_la_CFLAGS = $(AM_CFLAGS) $(CFLAGS_memcpyaltivec)
libmemcpyaltivec_plugin_la_LIBADD = $(AM_LIBADD) $(LIBS_memcpyaltivec)
libmemcpyaltivec_plugin_la_DEPENDENCIES =
libi420_yuy2_altivec_plugin_la_SOURCES = \
../video_chroma/i420_yuy2.c \
../video_chroma/i420_yuy2.h
......@@ -11,6 +6,5 @@ libi420_yuy2_altivec_plugin_la_LIBADD = $(AM_LIBADD) $(LIBS_i420_yuy2_altivec)
libi420_yuy2_altivec_plugin_la_DEPENDENCIES =
libvlc_LTLIBRARIES += \
libmemcpyaltivec_plugin.la \
libi420_yuy2_altivec_plugin.la \
$(NULL)
/*****************************************************************************
* memcpy.c : AltiVec memcpy module
*****************************************************************************
* Copyright (C) 2001 the VideoLAN team
* $Id$
*
* Author: Christophe Massiot <massiot@via.ecp.fr>
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
*****************************************************************************/
#ifndef __BUILD_ALTIVEC_ASM__
/*****************************************************************************
* Preamble
*****************************************************************************/
#ifdef HAVE_CONFIG_H
# include "config.h"
#endif
#include <vlc_common.h>
#include <vlc_plugin.h>
#include <vlc_cpu.h>
#ifdef HAVE_ALTIVEC_H
# include <altivec.h>
#endif
/*****************************************************************************
* Local prototypes.
*****************************************************************************/
static void * fast_memcpy ( void * to, const void * from, size_t len );
/*****************************************************************************
* Module initializer.
*****************************************************************************/
static int Activate ( vlc_object_t *p_this )
{
if( !(vlc_CPU() & CPU_CAPABILITY_ALTIVEC) )
return VLC_EGENERIC;
VLC_UNUSED(p_this);
vlc_fastmem_register( fast_memcpy );
return VLC_SUCCESS;
}
/*****************************************************************************
* Module descriptor.
*****************************************************************************/
vlc_module_begin ()
set_description( N_("AltiVec memcpy") )
set_category( CAT_ADVANCED )
set_subcategory( SUBCAT_ADVANCED_MISC )
set_capability( "memcpy", 100 )
set_callbacks( Activate, NULL )
add_shortcut( "altivec" )
vlc_module_end ()
#else
typedef unsigned long size_t;
#endif /* __BUILD_ALTIVEC_ASM__ */
#if defined(CAN_COMPILE_C_ALTIVEC) || defined( __BUILD_ALTIVEC_ASM__ )
#define vector_s16_t vector signed short
#define vector_u16_t vector unsigned short
#define vector_s8_t vector signed char
#define vector_u8_t vector unsigned char
#define vector_s32_t vector signed int
#define vector_u32_t vector unsigned int
#define MMREG_SIZE 16
#define SMALL_MEMCPY(to, from, len) \
{ \
unsigned char * end = to + len; \
while( to < end ) \
{ \
*to++ = *from++; \
} \
}
static void * fast_memcpy( void * _to, const void * _from, size_t len )
{
void * retval = _to;
unsigned char * to = (unsigned char *)_to;
unsigned char * from = (unsigned char *)_from;
if( len > 16 )
{
/* Align destination to MMREG_SIZE -boundary */
register unsigned long int delta;
delta = ((unsigned long)to)&(MMREG_SIZE-1);
if( delta )
{
delta = MMREG_SIZE - delta;
len -= delta;
SMALL_MEMCPY(to, from, delta);
}
if( len & ~(MMREG_SIZE-1) )
{
vector_u8_t perm, ref0, ref1, tmp;
perm = vec_lvsl( 0, from );
ref0 = vec_ld( 0, from );
ref1 = vec_ld( 15, from );
from += 16;
len -= 16;
tmp = vec_perm( ref0, ref1, perm );
while( len & ~(MMREG_SIZE-1) )
{
ref0 = vec_ld( 0, from );
ref1 = vec_ld( 15, from );
from += 16;
len -= 16;
vec_st( tmp, 0, to );
tmp = vec_perm( ref0, ref1, perm );
to += 16;
}
vec_st( tmp, 0, to );
to += 16;
}
}
if( len )
{
SMALL_MEMCPY( to, from, len );
}
return retval;
}
#endif
#if !defined(CAN_COMPILE_C_ALTIVEC) && !defined(__BUILD_ALTIVEC_ASM__)
/*
* The asm code is generated with:
*
* gcc-2.95 -fvec -D__BUILD_ALTIVEC_ASM__ -O9 -fomit-frame-pointer -mregnames -S * memcpyaltivec.c
*
* sed 's/.L/._L/g' memcpyaltivec.s |
* awk '{args=""; len=split ($2, arg, ",");
* for (i=1; i<=len; i++) { a=arg[i]; if (i<len) a=a",";
* args = args sprintf ("%-6s", a) }
* printf ("\t\"\t%-16s%-24s\\n\"\n", $1, args) }' |
* unexpand -a
*/
static void * fast_memcpy( void * _to, const void * _from, size_t len )
{
asm (" \n"
" cmplwi %cr0, %r5, 16 \n"
" mr %r9, %r3 \n"
" bc 4, 1, ._L3 \n"
" andi. %r0, %r3, 15 \n"
" bc 12, 2, ._L4 \n"
" subfic %r0, %r0, 16 \n"
" add %r11, %r3, %r0 \n"
" cmplw %cr0, %r3, %r11 \n"
" subf %r5, %r0, %r5 \n"
" bc 4, 0, ._L4 \n"
" ._L7: \n"
" lbz %r0, 0(%r4) \n"
" stb %r0, 0(%r9) \n"
" addi %r9, %r9, 1 \n"
" cmplw %cr0, %r9, %r11 \n"
" addi %r4, %r4, 1 \n"
" bc 12, 0, ._L7 \n"
" ._L4: \n"
" rlwinm. %r0, %r5, 0, 0, 27 \n"
" bc 12, 2, ._L3 \n"
" addi %r5, %r5, -16 \n"
" li %r11, 15 \n"
" lvsl %v12, 0, %r4 \n"
" lvx %v1, 0, %r4 \n"
" lvx %v0, %r11, %r4 \n"
" rlwinm. %r0, %r5, 0, 0, 27 \n"
" vperm %v13, %v1, %v0, %v12 \n"
" addi %r4, %r4, 16 \n"
" bc 12, 2, ._L11 \n"
" ._L12: \n"
" addi %r5, %r5, -16 \n"
" li %r11, 15 \n"
" lvx %v1, 0, %r4 \n"
" lvx %v0, %r11, %r4 \n"
" rlwinm. %r0, %r5, 0, 0, 27 \n"
" stvx %v13, 0, %r9 \n"
" vperm %v13, %v1, %v0, %v12 \n"
" addi %r4, %r4, 16 \n"
" addi %r9, %r9, 16 \n"
" bc 4, 2, ._L12 \n"
" ._L11: \n"
" stvx %v13, 0, %r9 \n"
" addi %r9, %r9, 16 \n"
" ._L3: \n"
" cmpwi %cr0, %r5, 0 \n"
" bclr 12, 2 \n"
" add %r5, %r9, %r5 \n"
" cmplw %cr0, %r9, %r5 \n"
" bclr 4, 0 \n"
" ._L17: \n"
" lbz %r0, 0(%r4) \n"
" stb %r0, 0(%r9) \n"
" addi %r9, %r9, 1 \n"
" cmplw %cr0, %r9, %r5 \n"
" addi %r4, %r4, 1 \n"
" bc 12, 0, ._L17 \n"
);
}
#endif
......@@ -9,9 +9,6 @@ SOURCES_inhibit = inhibit.c
SOURCES_sqlite = sqlite.c
SOURCES_xml = xml/libxml.c
SOURCES_memcpy = memcpy.c
libvlc_LTLIBRARIES += libmemcpy_plugin.la
libgnutls_plugin_la_SOURCES = gnutls.c dhparams.h
libgnutls_plugin_la_CFLAGS = $(AM_CFLAGS) $(GNUTLS_CFLAGS) $(GCRYPT_CFLAGS)
libgnutls_plugin_la_LIBADD = $(AM_LIBADD) $(GNUTLS_LIBS) $(GCRYPT_LIBS)
......
/*****************************************************************************
* memcpy.c : default memcpy plugin for vlc
*****************************************************************************
* Copyright (C) 2000, 2001 the VideoLAN team
* $Id$
*
* Authors: Samuel Hocevar <sam@zoy.org>
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
*****************************************************************************/
/*****************************************************************************
* Preamble
*****************************************************************************/
#ifdef HAVE_CONFIG_H
# include "config.h"
#endif
#include <vlc_common.h>
#include <vlc_plugin.h>
static int OpenDummy(vlc_object_t *);
/*****************************************************************************
* Module descriptor
*****************************************************************************/
vlc_module_begin ()
set_shortname( N_("Dummy") )
set_description( N_("libc memcpy") )
set_capability( "memcpy", 50 )
set_callbacks( OpenDummy, NULL )
add_shortcut( "c", "libc" )
vlc_module_end ()
static int OpenDummy( vlc_object_t *obj )
{
(void) obj;
return VLC_SUCCESS;
}
libmemcpymmx_plugin_la_SOURCES = memcpy.c fastmemcpy.h
libmemcpymmx_plugin_la_CFLAGS = $(AM_CFLAGS)
libmemcpymmx_plugin_la_LIBADD = $(AM_LIBADD)
libmemcpymmx_plugin_la_DEPENDENCIES =
libi420_rgb_mmx_plugin_la_SOURCES = \
../video_chroma/i420_rgb.c \
../video_chroma/i420_rgb.h \
......@@ -31,7 +26,3 @@ libvlc_LTLIBRARIES += \
libi420_yuy2_mmx_plugin.la \
libi422_yuy2_mmx_plugin.la \
$(NULL)
if HAVE_WIN32
libvlc_LTLIBRARIES += libmemcpymmx_plugin.la
endif
/*****************************************************************************
* fastmemcpy.h : fast memcpy routines
*****************************************************************************
* $Id$
*
* Authors: various Linux kernel hackers
* various MPlayer hackers
* Nick Kurshev <nickols_k@mail.ru>
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
*****************************************************************************/
/*
aclib - advanced C library ;)
This file contains functions which improve and expand standard C-library
*/
#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
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.
*/
#undef HAVE_SSE
#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 up to 12% versus
standard (non MMX-optimized) version.
Note: on K6-2+ it speedups memory copying up to 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)\
{\
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");\
}
#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
#ifdef HAVE_SSE
VLC_SSE
#else
VLC_MMX
#endif
static 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", "xmm0", "xmm1", "xmm2", "xmm3");
((const unsigned char *)from)+=64;
((unsigned char *)to)+=64;
}
else
/*
Only if SRC is aligned on 16-byte boundary.
It allows using 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", "xmm0", "xmm1", "xmm2", "xmm3");
((const unsigned char *)from)+=64;
((unsigned char *)to)+=64;
}
#else
/* Align destination at BLOCK_SIZE boundary */
for(; ((uintptr_t)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", "mm0", "mm1", "mm2", "mm3",
"mm4", "mm5", "mm6", "mm7");
from = (const void *) (((const unsigned char *)from)+64);
to = (void *) (((unsigned char *)to)+64);
}
/* printf(" %p %p\n", (uintptr_t)from&1023, (uintptr_t)to&1023); */
/* Pure Assembly cuz gcc is a bit unpredictable ;) */
# if 0
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"