Chiebot-Mirror
/
FFmpeg
mirror of https://github.com/FFmpeg/FFmpeg.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
25152 Commits
37 Branches
400 Tags
464 MiB
Tag: Branch: Tree: 1878f685c0
Branches Tags
${ item.name }
Create tag ${ searchTerm }
Create branch ${ searchTerm }
from '1878f685c0'
${ noResults }
FFmpeg/libavcodec/x86/dsputilenc_mmx.c

1439 lines
43 KiB
Raw Normal View History Unescape

split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
/*
* MMX optimized DSP utils
cosmetics: Remove pointless period after copyright statement non-sentences. Originally committed as revision 16684 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
* Copyright (c) 2000, 2001 Fabrice Bellard
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
* Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg 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
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*
* MMX optimization by Nick Kurshev <nickols_k@mail.ru>
*/
Use full path for #includes from another directory. Originally committed as revision 13098 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
#include "libavutil/x86_cpu.h"
#include "libavcodec/dsputil.h"
#include "libavcodec/mpegvideo.h"
moves mid_pred() into mathops.h (with arch specific code split by directory) Originally committed as revision 16681 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
#include "libavcodec/mathops.h"
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
#include "dsputil_mmx.h"
static void get_pixels_mmx(DCTELEM *block, const uint8_t *pixels, int line_size)
{
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile(
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"mov $-128, %%"REG_a" \n\t"
"pxor %%mm7, %%mm7 \n\t"
ASMALIGN(4)
"1: \n\t"
"movq (%0), %%mm0 \n\t"
"movq (%0, %2), %%mm2 \n\t"
"movq %%mm0, %%mm1 \n\t"
"movq %%mm2, %%mm3 \n\t"
"punpcklbw %%mm7, %%mm0 \n\t"
"punpckhbw %%mm7, %%mm1 \n\t"
"punpcklbw %%mm7, %%mm2 \n\t"
"punpckhbw %%mm7, %%mm3 \n\t"
"movq %%mm0, (%1, %%"REG_a") \n\t"
"movq %%mm1, 8(%1, %%"REG_a") \n\t"
"movq %%mm2, 16(%1, %%"REG_a") \n\t"
"movq %%mm3, 24(%1, %%"REG_a") \n\t"
"add %3, %0 \n\t"
"add $32, %%"REG_a" \n\t"
"js 1b \n\t"
: "+r" (pixels)
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r" (block+64), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*2)
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "%"REG_a
);
}
get_pixels_sse2, ~+12% performance compared to mmx Originally committed as revision 15591 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
static void get_pixels_sse2(DCTELEM *block, const uint8_t *pixels, int line_size)
{
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile(
get_pixels_sse2, ~+12% performance compared to mmx Originally committed as revision 15591 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
"pxor %%xmm7, %%xmm7 \n\t"
"movq (%0), %%xmm0 \n\t"
"movq (%0, %2), %%xmm1 \n\t"
"movq (%0, %2,2), %%xmm2 \n\t"
"movq (%0, %3), %%xmm3 \n\t"
"lea (%0,%2,4), %0 \n\t"
"punpcklbw %%xmm7, %%xmm0 \n\t"
"punpcklbw %%xmm7, %%xmm1 \n\t"
"punpcklbw %%xmm7, %%xmm2 \n\t"
"punpcklbw %%xmm7, %%xmm3 \n\t"
"movdqa %%xmm0, (%1) \n\t"
"movdqa %%xmm1, 16(%1) \n\t"
"movdqa %%xmm2, 32(%1) \n\t"
"movdqa %%xmm3, 48(%1) \n\t"
"movq (%0), %%xmm0 \n\t"
"movq (%0, %2), %%xmm1 \n\t"
"movq (%0, %2,2), %%xmm2 \n\t"
"movq (%0, %3), %%xmm3 \n\t"
"punpcklbw %%xmm7, %%xmm0 \n\t"
"punpcklbw %%xmm7, %%xmm1 \n\t"
"punpcklbw %%xmm7, %%xmm2 \n\t"
"punpcklbw %%xmm7, %%xmm3 \n\t"
"movdqa %%xmm0, 64(%1) \n\t"
"movdqa %%xmm1, 80(%1) \n\t"
"movdqa %%xmm2, 96(%1) \n\t"
"movdqa %%xmm3, 112(%1) \n\t"
: "+r" (pixels)
: "r" (block), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3)
);
}
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
static inline void diff_pixels_mmx(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride)
{
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile(
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"pxor %%mm7, %%mm7 \n\t"
"mov $-128, %%"REG_a" \n\t"
ASMALIGN(4)
"1: \n\t"
"movq (%0), %%mm0 \n\t"
"movq (%1), %%mm2 \n\t"
"movq %%mm0, %%mm1 \n\t"
"movq %%mm2, %%mm3 \n\t"
"punpcklbw %%mm7, %%mm0 \n\t"
"punpckhbw %%mm7, %%mm1 \n\t"
"punpcklbw %%mm7, %%mm2 \n\t"
"punpckhbw %%mm7, %%mm3 \n\t"
"psubw %%mm2, %%mm0 \n\t"
"psubw %%mm3, %%mm1 \n\t"
"movq %%mm0, (%2, %%"REG_a") \n\t"
"movq %%mm1, 8(%2, %%"REG_a") \n\t"
"add %3, %0 \n\t"
"add %3, %1 \n\t"
"add $16, %%"REG_a" \n\t"
"jnz 1b \n\t"
: "+r" (s1), "+r" (s2)
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r" (block+64), "r" ((x86_reg)stride)
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "%"REG_a
);
}
static int pix_sum16_mmx(uint8_t * pix, int line_size){
const int h=16;
int sum;
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
x86_reg index= -line_size*h;
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile(
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"pxor %%mm7, %%mm7 \n\t"
"pxor %%mm6, %%mm6 \n\t"
"1: \n\t"
"movq (%2, %1), %%mm0 \n\t"
"movq (%2, %1), %%mm1 \n\t"
"movq 8(%2, %1), %%mm2 \n\t"
"movq 8(%2, %1), %%mm3 \n\t"
"punpcklbw %%mm7, %%mm0 \n\t"
"punpckhbw %%mm7, %%mm1 \n\t"
"punpcklbw %%mm7, %%mm2 \n\t"
"punpckhbw %%mm7, %%mm3 \n\t"
"paddw %%mm0, %%mm1 \n\t"
"paddw %%mm2, %%mm3 \n\t"
"paddw %%mm1, %%mm3 \n\t"
"paddw %%mm3, %%mm6 \n\t"
"add %3, %1 \n\t"
" js 1b \n\t"
"movq %%mm6, %%mm5 \n\t"
"psrlq $32, %%mm6 \n\t"
"paddw %%mm5, %%mm6 \n\t"
"movq %%mm6, %%mm5 \n\t"
"psrlq $16, %%mm6 \n\t"
"paddw %%mm5, %%mm6 \n\t"
"movd %%mm6, %0 \n\t"
"andl $0xFFFF, %0 \n\t"
: "=&r" (sum), "+r" (index)
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r" (pix - index), "r" ((x86_reg)line_size)
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
);
return sum;
}
static int pix_norm1_mmx(uint8_t *pix, int line_size) {
int tmp;
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile (
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"movl $16,%%ecx\n"
"pxor %%mm0,%%mm0\n"
"pxor %%mm7,%%mm7\n"
"1:\n"
"movq (%0),%%mm2\n" /* mm2 = pix[0-7] */
"movq 8(%0),%%mm3\n" /* mm3 = pix[8-15] */
"movq %%mm2,%%mm1\n" /* mm1 = mm2 = pix[0-7] */
"punpckhbw %%mm0,%%mm1\n" /* mm1 = [pix4-7] */
"punpcklbw %%mm0,%%mm2\n" /* mm2 = [pix0-3] */
"movq %%mm3,%%mm4\n" /* mm4 = mm3 = pix[8-15] */
"punpckhbw %%mm0,%%mm3\n" /* mm3 = [pix12-15] */
"punpcklbw %%mm0,%%mm4\n" /* mm4 = [pix8-11] */
"pmaddwd %%mm1,%%mm1\n" /* mm1 = (pix0^2+pix1^2,pix2^2+pix3^2) */
"pmaddwd %%mm2,%%mm2\n" /* mm2 = (pix4^2+pix5^2,pix6^2+pix7^2) */
"pmaddwd %%mm3,%%mm3\n"
"pmaddwd %%mm4,%%mm4\n"
"paddd %%mm1,%%mm2\n" /* mm2 = (pix0^2+pix1^2+pix4^2+pix5^2,
pix2^2+pix3^2+pix6^2+pix7^2) */
"paddd %%mm3,%%mm4\n"
"paddd %%mm2,%%mm7\n"
"add %2, %0\n"
"paddd %%mm4,%%mm7\n"
"dec %%ecx\n"
"jnz 1b\n"
"movq %%mm7,%%mm1\n"
"psrlq $32, %%mm7\n" /* shift hi dword to lo */
"paddd %%mm7,%%mm1\n"
"movd %%mm1,%1\n"
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "+r" (pix), "=r"(tmp) : "r" ((x86_reg)line_size) : "%ecx" );
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
return tmp;
}
static int sse8_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
int tmp;
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile (
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"movl %4,%%ecx\n"
"shr $1,%%ecx\n"
"pxor %%mm0,%%mm0\n" /* mm0 = 0 */
"pxor %%mm7,%%mm7\n" /* mm7 holds the sum */
"1:\n"
"movq (%0),%%mm1\n" /* mm1 = pix1[0][0-7] */
"movq (%1),%%mm2\n" /* mm2 = pix2[0][0-7] */
"movq (%0,%3),%%mm3\n" /* mm3 = pix1[1][0-7] */
"movq (%1,%3),%%mm4\n" /* mm4 = pix2[1][0-7] */
/* todo: mm1-mm2, mm3-mm4 */
/* algo: subtract mm1 from mm2 with saturation and vice versa */
/* OR the results to get absolute difference */
"movq %%mm1,%%mm5\n"
"movq %%mm3,%%mm6\n"
"psubusb %%mm2,%%mm1\n"
"psubusb %%mm4,%%mm3\n"
"psubusb %%mm5,%%mm2\n"
"psubusb %%mm6,%%mm4\n"
"por %%mm1,%%mm2\n"
"por %%mm3,%%mm4\n"
/* now convert to 16-bit vectors so we can square them */
"movq %%mm2,%%mm1\n"
"movq %%mm4,%%mm3\n"
"punpckhbw %%mm0,%%mm2\n"
"punpckhbw %%mm0,%%mm4\n"
"punpcklbw %%mm0,%%mm1\n" /* mm1 now spread over (mm1,mm2) */
"punpcklbw %%mm0,%%mm3\n" /* mm4 now spread over (mm3,mm4) */
"pmaddwd %%mm2,%%mm2\n"
"pmaddwd %%mm4,%%mm4\n"
"pmaddwd %%mm1,%%mm1\n"
"pmaddwd %%mm3,%%mm3\n"
"lea (%0,%3,2), %0\n" /* pix1 += 2*line_size */
"lea (%1,%3,2), %1\n" /* pix2 += 2*line_size */
"paddd %%mm2,%%mm1\n"
"paddd %%mm4,%%mm3\n"
"paddd %%mm1,%%mm7\n"
"paddd %%mm3,%%mm7\n"
"decl %%ecx\n"
"jnz 1b\n"
"movq %%mm7,%%mm1\n"
"psrlq $32, %%mm7\n" /* shift hi dword to lo */
"paddd %%mm7,%%mm1\n"
"movd %%mm1,%2\n"
: "+r" (pix1), "+r" (pix2), "=r"(tmp)
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r" ((x86_reg)line_size) , "m" (h)
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "%ecx");
return tmp;
}
static int sse16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
int tmp;
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile (
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"movl %4,%%ecx\n"
"pxor %%mm0,%%mm0\n" /* mm0 = 0 */
"pxor %%mm7,%%mm7\n" /* mm7 holds the sum */
"1:\n"
"movq (%0),%%mm1\n" /* mm1 = pix1[0-7] */
"movq (%1),%%mm2\n" /* mm2 = pix2[0-7] */
"movq 8(%0),%%mm3\n" /* mm3 = pix1[8-15] */
"movq 8(%1),%%mm4\n" /* mm4 = pix2[8-15] */
/* todo: mm1-mm2, mm3-mm4 */
/* algo: subtract mm1 from mm2 with saturation and vice versa */
/* OR the results to get absolute difference */
"movq %%mm1,%%mm5\n"
"movq %%mm3,%%mm6\n"
"psubusb %%mm2,%%mm1\n"
"psubusb %%mm4,%%mm3\n"
"psubusb %%mm5,%%mm2\n"
"psubusb %%mm6,%%mm4\n"
"por %%mm1,%%mm2\n"
"por %%mm3,%%mm4\n"
/* now convert to 16-bit vectors so we can square them */
"movq %%mm2,%%mm1\n"
"movq %%mm4,%%mm3\n"
"punpckhbw %%mm0,%%mm2\n"
"punpckhbw %%mm0,%%mm4\n"
"punpcklbw %%mm0,%%mm1\n" /* mm1 now spread over (mm1,mm2) */
"punpcklbw %%mm0,%%mm3\n" /* mm4 now spread over (mm3,mm4) */
"pmaddwd %%mm2,%%mm2\n"
"pmaddwd %%mm4,%%mm4\n"
"pmaddwd %%mm1,%%mm1\n"
"pmaddwd %%mm3,%%mm3\n"
"add %3,%0\n"
"add %3,%1\n"
"paddd %%mm2,%%mm1\n"
"paddd %%mm4,%%mm3\n"
"paddd %%mm1,%%mm7\n"
"paddd %%mm3,%%mm7\n"
"decl %%ecx\n"
"jnz 1b\n"
"movq %%mm7,%%mm1\n"
"psrlq $32, %%mm7\n" /* shift hi dword to lo */
"paddd %%mm7,%%mm1\n"
"movd %%mm1,%2\n"
: "+r" (pix1), "+r" (pix2), "=r"(tmp)
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r" ((x86_reg)line_size) , "m" (h)
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "%ecx");
return tmp;
}
static int sse16_sse2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
int tmp;
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile (
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"shr $1,%2\n"
"pxor %%xmm0,%%xmm0\n" /* mm0 = 0 */
"pxor %%xmm7,%%xmm7\n" /* mm7 holds the sum */
"1:\n"
"movdqu (%0),%%xmm1\n" /* mm1 = pix1[0][0-15] */
"movdqu (%1),%%xmm2\n" /* mm2 = pix2[0][0-15] */
"movdqu (%0,%4),%%xmm3\n" /* mm3 = pix1[1][0-15] */
"movdqu (%1,%4),%%xmm4\n" /* mm4 = pix2[1][0-15] */
/* todo: mm1-mm2, mm3-mm4 */
/* algo: subtract mm1 from mm2 with saturation and vice versa */
/* OR the results to get absolute difference */
"movdqa %%xmm1,%%xmm5\n"
"movdqa %%xmm3,%%xmm6\n"
"psubusb %%xmm2,%%xmm1\n"
"psubusb %%xmm4,%%xmm3\n"
"psubusb %%xmm5,%%xmm2\n"
"psubusb %%xmm6,%%xmm4\n"
"por %%xmm1,%%xmm2\n"
"por %%xmm3,%%xmm4\n"
/* now convert to 16-bit vectors so we can square them */
"movdqa %%xmm2,%%xmm1\n"
"movdqa %%xmm4,%%xmm3\n"
"punpckhbw %%xmm0,%%xmm2\n"
"punpckhbw %%xmm0,%%xmm4\n"
"punpcklbw %%xmm0,%%xmm1\n" /* mm1 now spread over (mm1,mm2) */
"punpcklbw %%xmm0,%%xmm3\n" /* mm4 now spread over (mm3,mm4) */
"pmaddwd %%xmm2,%%xmm2\n"
"pmaddwd %%xmm4,%%xmm4\n"
"pmaddwd %%xmm1,%%xmm1\n"
"pmaddwd %%xmm3,%%xmm3\n"
"lea (%0,%4,2), %0\n" /* pix1 += 2*line_size */
"lea (%1,%4,2), %1\n" /* pix2 += 2*line_size */
"paddd %%xmm2,%%xmm1\n"
"paddd %%xmm4,%%xmm3\n"
"paddd %%xmm1,%%xmm7\n"
"paddd %%xmm3,%%xmm7\n"
"decl %2\n"
"jnz 1b\n"
"movdqa %%xmm7,%%xmm1\n"
"psrldq $8, %%xmm7\n" /* shift hi qword to lo */
"paddd %%xmm1,%%xmm7\n"
"movdqa %%xmm7,%%xmm1\n"
"psrldq $4, %%xmm7\n" /* shift hi dword to lo */
"paddd %%xmm1,%%xmm7\n"
"movd %%xmm7,%3\n"
: "+r" (pix1), "+r" (pix2), "+r"(h), "=r"(tmp)
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r" ((x86_reg)line_size));
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
return tmp;
}
static int hf_noise8_mmx(uint8_t * pix1, int line_size, int h) {
int tmp;
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile (
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"movl %3,%%ecx\n"
"pxor %%mm7,%%mm7\n"
"pxor %%mm6,%%mm6\n"
"movq (%0),%%mm0\n"
"movq %%mm0, %%mm1\n"
"psllq $8, %%mm0\n"
"psrlq $8, %%mm1\n"
"psrlq $8, %%mm0\n"
"movq %%mm0, %%mm2\n"
"movq %%mm1, %%mm3\n"
"punpcklbw %%mm7,%%mm0\n"
"punpcklbw %%mm7,%%mm1\n"
"punpckhbw %%mm7,%%mm2\n"
"punpckhbw %%mm7,%%mm3\n"
"psubw %%mm1, %%mm0\n"
"psubw %%mm3, %%mm2\n"
"add %2,%0\n"
"movq (%0),%%mm4\n"
"movq %%mm4, %%mm1\n"
"psllq $8, %%mm4\n"
"psrlq $8, %%mm1\n"
"psrlq $8, %%mm4\n"
"movq %%mm4, %%mm5\n"
"movq %%mm1, %%mm3\n"
"punpcklbw %%mm7,%%mm4\n"
"punpcklbw %%mm7,%%mm1\n"
"punpckhbw %%mm7,%%mm5\n"
"punpckhbw %%mm7,%%mm3\n"
"psubw %%mm1, %%mm4\n"
"psubw %%mm3, %%mm5\n"
"psubw %%mm4, %%mm0\n"
"psubw %%mm5, %%mm2\n"
"pxor %%mm3, %%mm3\n"
"pxor %%mm1, %%mm1\n"
"pcmpgtw %%mm0, %%mm3\n\t"
"pcmpgtw %%mm2, %%mm1\n\t"
"pxor %%mm3, %%mm0\n"
"pxor %%mm1, %%mm2\n"
"psubw %%mm3, %%mm0\n"
"psubw %%mm1, %%mm2\n"
"paddw %%mm0, %%mm2\n"
"paddw %%mm2, %%mm6\n"
"add %2,%0\n"
"1:\n"
"movq (%0),%%mm0\n"
"movq %%mm0, %%mm1\n"
"psllq $8, %%mm0\n"
"psrlq $8, %%mm1\n"
"psrlq $8, %%mm0\n"
"movq %%mm0, %%mm2\n"
"movq %%mm1, %%mm3\n"
"punpcklbw %%mm7,%%mm0\n"
"punpcklbw %%mm7,%%mm1\n"
"punpckhbw %%mm7,%%mm2\n"
"punpckhbw %%mm7,%%mm3\n"
"psubw %%mm1, %%mm0\n"
"psubw %%mm3, %%mm2\n"
"psubw %%mm0, %%mm4\n"
"psubw %%mm2, %%mm5\n"
"pxor %%mm3, %%mm3\n"
"pxor %%mm1, %%mm1\n"
"pcmpgtw %%mm4, %%mm3\n\t"
"pcmpgtw %%mm5, %%mm1\n\t"
"pxor %%mm3, %%mm4\n"
"pxor %%mm1, %%mm5\n"
"psubw %%mm3, %%mm4\n"
"psubw %%mm1, %%mm5\n"
"paddw %%mm4, %%mm5\n"
"paddw %%mm5, %%mm6\n"
"add %2,%0\n"
"movq (%0),%%mm4\n"
"movq %%mm4, %%mm1\n"
"psllq $8, %%mm4\n"
"psrlq $8, %%mm1\n"
"psrlq $8, %%mm4\n"
"movq %%mm4, %%mm5\n"
"movq %%mm1, %%mm3\n"
"punpcklbw %%mm7,%%mm4\n"
"punpcklbw %%mm7,%%mm1\n"
"punpckhbw %%mm7,%%mm5\n"
"punpckhbw %%mm7,%%mm3\n"
"psubw %%mm1, %%mm4\n"
"psubw %%mm3, %%mm5\n"
"psubw %%mm4, %%mm0\n"
"psubw %%mm5, %%mm2\n"
"pxor %%mm3, %%mm3\n"
"pxor %%mm1, %%mm1\n"
"pcmpgtw %%mm0, %%mm3\n\t"
"pcmpgtw %%mm2, %%mm1\n\t"
"pxor %%mm3, %%mm0\n"
"pxor %%mm1, %%mm2\n"
"psubw %%mm3, %%mm0\n"
"psubw %%mm1, %%mm2\n"
"paddw %%mm0, %%mm2\n"
"paddw %%mm2, %%mm6\n"
"add %2,%0\n"
"subl $2, %%ecx\n"
" jnz 1b\n"
"movq %%mm6, %%mm0\n"
"punpcklwd %%mm7,%%mm0\n"
"punpckhwd %%mm7,%%mm6\n"
"paddd %%mm0, %%mm6\n"
"movq %%mm6,%%mm0\n"
"psrlq $32, %%mm6\n"
"paddd %%mm6,%%mm0\n"
"movd %%mm0,%1\n"
: "+r" (pix1), "=r"(tmp)
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r" ((x86_reg)line_size) , "g" (h-2)
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "%ecx");
return tmp;
}
static int hf_noise16_mmx(uint8_t * pix1, int line_size, int h) {
int tmp;
uint8_t * pix= pix1;
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile (
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"movl %3,%%ecx\n"
"pxor %%mm7,%%mm7\n"
"pxor %%mm6,%%mm6\n"
"movq (%0),%%mm0\n"
"movq 1(%0),%%mm1\n"
"movq %%mm0, %%mm2\n"
"movq %%mm1, %%mm3\n"
"punpcklbw %%mm7,%%mm0\n"
"punpcklbw %%mm7,%%mm1\n"
"punpckhbw %%mm7,%%mm2\n"
"punpckhbw %%mm7,%%mm3\n"
"psubw %%mm1, %%mm0\n"
"psubw %%mm3, %%mm2\n"
"add %2,%0\n"
"movq (%0),%%mm4\n"
"movq 1(%0),%%mm1\n"
"movq %%mm4, %%mm5\n"
"movq %%mm1, %%mm3\n"
"punpcklbw %%mm7,%%mm4\n"
"punpcklbw %%mm7,%%mm1\n"
"punpckhbw %%mm7,%%mm5\n"
"punpckhbw %%mm7,%%mm3\n"
"psubw %%mm1, %%mm4\n"
"psubw %%mm3, %%mm5\n"
"psubw %%mm4, %%mm0\n"
"psubw %%mm5, %%mm2\n"
"pxor %%mm3, %%mm3\n"
"pxor %%mm1, %%mm1\n"
"pcmpgtw %%mm0, %%mm3\n\t"
"pcmpgtw %%mm2, %%mm1\n\t"
"pxor %%mm3, %%mm0\n"
"pxor %%mm1, %%mm2\n"
"psubw %%mm3, %%mm0\n"
"psubw %%mm1, %%mm2\n"
"paddw %%mm0, %%mm2\n"
"paddw %%mm2, %%mm6\n"
"add %2,%0\n"
"1:\n"
"movq (%0),%%mm0\n"
"movq 1(%0),%%mm1\n"
"movq %%mm0, %%mm2\n"
"movq %%mm1, %%mm3\n"
"punpcklbw %%mm7,%%mm0\n"
"punpcklbw %%mm7,%%mm1\n"
"punpckhbw %%mm7,%%mm2\n"
"punpckhbw %%mm7,%%mm3\n"
"psubw %%mm1, %%mm0\n"
"psubw %%mm3, %%mm2\n"
"psubw %%mm0, %%mm4\n"
"psubw %%mm2, %%mm5\n"
"pxor %%mm3, %%mm3\n"
"pxor %%mm1, %%mm1\n"
"pcmpgtw %%mm4, %%mm3\n\t"
"pcmpgtw %%mm5, %%mm1\n\t"
"pxor %%mm3, %%mm4\n"
"pxor %%mm1, %%mm5\n"
"psubw %%mm3, %%mm4\n"
"psubw %%mm1, %%mm5\n"
"paddw %%mm4, %%mm5\n"
"paddw %%mm5, %%mm6\n"
"add %2,%0\n"
"movq (%0),%%mm4\n"
"movq 1(%0),%%mm1\n"
"movq %%mm4, %%mm5\n"
"movq %%mm1, %%mm3\n"
"punpcklbw %%mm7,%%mm4\n"
"punpcklbw %%mm7,%%mm1\n"
"punpckhbw %%mm7,%%mm5\n"
"punpckhbw %%mm7,%%mm3\n"
"psubw %%mm1, %%mm4\n"
"psubw %%mm3, %%mm5\n"
"psubw %%mm4, %%mm0\n"
"psubw %%mm5, %%mm2\n"
"pxor %%mm3, %%mm3\n"
"pxor %%mm1, %%mm1\n"
"pcmpgtw %%mm0, %%mm3\n\t"
"pcmpgtw %%mm2, %%mm1\n\t"
"pxor %%mm3, %%mm0\n"
"pxor %%mm1, %%mm2\n"
"psubw %%mm3, %%mm0\n"
"psubw %%mm1, %%mm2\n"
"paddw %%mm0, %%mm2\n"
"paddw %%mm2, %%mm6\n"
"add %2,%0\n"
"subl $2, %%ecx\n"
" jnz 1b\n"
"movq %%mm6, %%mm0\n"
"punpcklwd %%mm7,%%mm0\n"
"punpckhwd %%mm7,%%mm6\n"
"paddd %%mm0, %%mm6\n"
"movq %%mm6,%%mm0\n"
"psrlq $32, %%mm6\n"
"paddd %%mm6,%%mm0\n"
"movd %%mm0,%1\n"
: "+r" (pix1), "=r"(tmp)
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r" ((x86_reg)line_size) , "g" (h-2)
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "%ecx");
return tmp + hf_noise8_mmx(pix+8, line_size, h);
}
static int nsse16_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
MpegEncContext *c = p;
int score1, score2;
if(c) score1 = c->dsp.sse[0](c, pix1, pix2, line_size, h);
else score1 = sse16_mmx(c, pix1, pix2, line_size, h);
score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h);
if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;
else return score1 + FFABS(score2)*8;
}
static int nsse8_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
MpegEncContext *c = p;
int score1= sse8_mmx(c, pix1, pix2, line_size, h);
int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h);
if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;
else return score1 + FFABS(score2)*8;
}
static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {
int tmp;
assert( (((int)pix) & 7) == 0);
assert((line_size &7) ==0);
#define SUM(in0, in1, out0, out1) \
"movq (%0), %%mm2\n"\
"movq 8(%0), %%mm3\n"\
"add %2,%0\n"\
"movq %%mm2, " #out0 "\n"\
"movq %%mm3, " #out1 "\n"\
"psubusb " #in0 ", %%mm2\n"\
"psubusb " #in1 ", %%mm3\n"\
"psubusb " #out0 ", " #in0 "\n"\
"psubusb " #out1 ", " #in1 "\n"\
"por %%mm2, " #in0 "\n"\
"por %%mm3, " #in1 "\n"\
"movq " #in0 ", %%mm2\n"\
"movq " #in1 ", %%mm3\n"\
"punpcklbw %%mm7, " #in0 "\n"\
"punpcklbw %%mm7, " #in1 "\n"\
"punpckhbw %%mm7, %%mm2\n"\
"punpckhbw %%mm7, %%mm3\n"\
"paddw " #in1 ", " #in0 "\n"\
"paddw %%mm3, %%mm2\n"\
"paddw %%mm2, " #in0 "\n"\
"paddw " #in0 ", %%mm6\n"
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile (
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"movl %3,%%ecx\n"
"pxor %%mm6,%%mm6\n"
"pxor %%mm7,%%mm7\n"
"movq (%0),%%mm0\n"
"movq 8(%0),%%mm1\n"
"add %2,%0\n"
Simplify vsad_intra16_mmx() Originally committed as revision 13188 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"jmp 2f\n"
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"1:\n"
SUM(%%mm4, %%mm5, %%mm0, %%mm1)
Simplify vsad_intra16_mmx() Originally committed as revision 13188 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"2:\n"
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
SUM(%%mm0, %%mm1, %%mm4, %%mm5)
"subl $2, %%ecx\n"
"jnz 1b\n"
"movq %%mm6,%%mm0\n"
"psrlq $32, %%mm6\n"
"paddw %%mm6,%%mm0\n"
"movq %%mm0,%%mm6\n"
"psrlq $16, %%mm0\n"
"paddw %%mm6,%%mm0\n"
"movd %%mm0,%1\n"
: "+r" (pix), "=r"(tmp)
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r" ((x86_reg)line_size) , "m" (h)
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "%ecx");
return tmp & 0xFFFF;
}
#undef SUM
static int vsad_intra16_mmx2(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {
int tmp;
assert( (((int)pix) & 7) == 0);
assert((line_size &7) ==0);
#define SUM(in0, in1, out0, out1) \
"movq (%0), " #out0 "\n"\
"movq 8(%0), " #out1 "\n"\
"add %2,%0\n"\
"psadbw " #out0 ", " #in0 "\n"\
"psadbw " #out1 ", " #in1 "\n"\
"paddw " #in1 ", " #in0 "\n"\
"paddw " #in0 ", %%mm6\n"
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile (
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"movl %3,%%ecx\n"
"pxor %%mm6,%%mm6\n"
"pxor %%mm7,%%mm7\n"
"movq (%0),%%mm0\n"
"movq 8(%0),%%mm1\n"
"add %2,%0\n"
Simplify vsad_intra16_mmx2() Originally committed as revision 13189 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"jmp 2f\n"
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"1:\n"
SUM(%%mm4, %%mm5, %%mm0, %%mm1)
Simplify vsad_intra16_mmx2() Originally committed as revision 13189 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"2:\n"
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
SUM(%%mm0, %%mm1, %%mm4, %%mm5)
"subl $2, %%ecx\n"
"jnz 1b\n"
"movd %%mm6,%1\n"
: "+r" (pix), "=r"(tmp)
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r" ((x86_reg)line_size) , "m" (h)
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "%ecx");
return tmp;
}
#undef SUM
static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
int tmp;
assert( (((int)pix1) & 7) == 0);
assert( (((int)pix2) & 7) == 0);
assert((line_size &7) ==0);
#define SUM(in0, in1, out0, out1) \
"movq (%0),%%mm2\n"\
"movq (%1)," #out0 "\n"\
"movq 8(%0),%%mm3\n"\
"movq 8(%1)," #out1 "\n"\
"add %3,%0\n"\
"add %3,%1\n"\
"psubb " #out0 ", %%mm2\n"\
"psubb " #out1 ", %%mm3\n"\
"pxor %%mm7, %%mm2\n"\
"pxor %%mm7, %%mm3\n"\
"movq %%mm2, " #out0 "\n"\
"movq %%mm3, " #out1 "\n"\
"psubusb " #in0 ", %%mm2\n"\
"psubusb " #in1 ", %%mm3\n"\
"psubusb " #out0 ", " #in0 "\n"\
"psubusb " #out1 ", " #in1 "\n"\
"por %%mm2, " #in0 "\n"\
"por %%mm3, " #in1 "\n"\
"movq " #in0 ", %%mm2\n"\
"movq " #in1 ", %%mm3\n"\
"punpcklbw %%mm7, " #in0 "\n"\
"punpcklbw %%mm7, " #in1 "\n"\
"punpckhbw %%mm7, %%mm2\n"\
"punpckhbw %%mm7, %%mm3\n"\
"paddw " #in1 ", " #in0 "\n"\
"paddw %%mm3, %%mm2\n"\
"paddw %%mm2, " #in0 "\n"\
"paddw " #in0 ", %%mm6\n"
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile (
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"movl %4,%%ecx\n"
"pxor %%mm6,%%mm6\n"
"pcmpeqw %%mm7,%%mm7\n"
"psllw $15, %%mm7\n"
"packsswb %%mm7, %%mm7\n"
"movq (%0),%%mm0\n"
"movq (%1),%%mm2\n"
"movq 8(%0),%%mm1\n"
"movq 8(%1),%%mm3\n"
"add %3,%0\n"
"add %3,%1\n"
"psubb %%mm2, %%mm0\n"
"psubb %%mm3, %%mm1\n"
"pxor %%mm7, %%mm0\n"
"pxor %%mm7, %%mm1\n"
Simplify vsad16_mmx(). Originally committed as revision 13191 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"jmp 2f\n"
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"1:\n"
SUM(%%mm4, %%mm5, %%mm0, %%mm1)
Simplify vsad16_mmx(). Originally committed as revision 13191 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"2:\n"
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
SUM(%%mm0, %%mm1, %%mm4, %%mm5)
"subl $2, %%ecx\n"
"jnz 1b\n"
"movq %%mm6,%%mm0\n"
"psrlq $32, %%mm6\n"
"paddw %%mm6,%%mm0\n"
"movq %%mm0,%%mm6\n"
"psrlq $16, %%mm0\n"
"paddw %%mm6,%%mm0\n"
"movd %%mm0,%2\n"
: "+r" (pix1), "+r" (pix2), "=r"(tmp)
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r" ((x86_reg)line_size) , "m" (h)
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "%ecx");
return tmp & 0x7FFF;
}
#undef SUM
static int vsad16_mmx2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
int tmp;
assert( (((int)pix1) & 7) == 0);
assert( (((int)pix2) & 7) == 0);
assert((line_size &7) ==0);
#define SUM(in0, in1, out0, out1) \
"movq (%0)," #out0 "\n"\
"movq (%1),%%mm2\n"\
"movq 8(%0)," #out1 "\n"\
"movq 8(%1),%%mm3\n"\
"add %3,%0\n"\
"add %3,%1\n"\
"psubb %%mm2, " #out0 "\n"\
"psubb %%mm3, " #out1 "\n"\
"pxor %%mm7, " #out0 "\n"\
"pxor %%mm7, " #out1 "\n"\
"psadbw " #out0 ", " #in0 "\n"\
"psadbw " #out1 ", " #in1 "\n"\
"paddw " #in1 ", " #in0 "\n"\
"paddw " #in0 ", %%mm6\n"
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile (
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"movl %4,%%ecx\n"
"pxor %%mm6,%%mm6\n"
"pcmpeqw %%mm7,%%mm7\n"
"psllw $15, %%mm7\n"
"packsswb %%mm7, %%mm7\n"
"movq (%0),%%mm0\n"
"movq (%1),%%mm2\n"
"movq 8(%0),%%mm1\n"
"movq 8(%1),%%mm3\n"
"add %3,%0\n"
"add %3,%1\n"
"psubb %%mm2, %%mm0\n"
"psubb %%mm3, %%mm1\n"
"pxor %%mm7, %%mm0\n"
"pxor %%mm7, %%mm1\n"
Simplify vsad16_mmx2(). Originally committed as revision 13193 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"jmp 2f\n"
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"1:\n"
SUM(%%mm4, %%mm5, %%mm0, %%mm1)
Simplify vsad16_mmx2(). Originally committed as revision 13193 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"2:\n"
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
SUM(%%mm0, %%mm1, %%mm4, %%mm5)
"subl $2, %%ecx\n"
"jnz 1b\n"
"movd %%mm6,%2\n"
: "+r" (pix1), "+r" (pix2), "=r"(tmp)
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r" ((x86_reg)line_size) , "m" (h)
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "%ecx");
return tmp;
}
#undef SUM
static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
x86_reg i=0;
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile(
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"1: \n\t"
"movq (%2, %0), %%mm0 \n\t"
"movq (%1, %0), %%mm1 \n\t"
"psubb %%mm0, %%mm1 \n\t"
"movq %%mm1, (%3, %0) \n\t"
"movq 8(%2, %0), %%mm0 \n\t"
"movq 8(%1, %0), %%mm1 \n\t"
"psubb %%mm0, %%mm1 \n\t"
"movq %%mm1, 8(%3, %0) \n\t"
"add $16, %0 \n\t"
"cmp %4, %0 \n\t"
" jb 1b \n\t"
: "+r" (i)
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r"(src1), "r"(src2), "r"(dst), "r"((x86_reg)w-15)
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
);
for(; i<w; i++)
dst[i+0] = src1[i+0]-src2[i+0];
}
huffyuv: add some const qualifiers Originally committed as revision 20290 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
static void sub_hfyu_median_prediction_mmx2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top){
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
x86_reg i=0;
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
uint8_t l, lt;
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile(
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"1: \n\t"
"movq -1(%1, %0), %%mm0 \n\t" // LT
"movq (%1, %0), %%mm1 \n\t" // T
"movq -1(%2, %0), %%mm2 \n\t" // L
"movq (%2, %0), %%mm3 \n\t" // X
"movq %%mm2, %%mm4 \n\t" // L
"psubb %%mm0, %%mm2 \n\t"
"paddb %%mm1, %%mm2 \n\t" // L + T - LT
"movq %%mm4, %%mm5 \n\t" // L
"pmaxub %%mm1, %%mm4 \n\t" // max(T, L)
"pminub %%mm5, %%mm1 \n\t" // min(T, L)
"pminub %%mm2, %%mm4 \n\t"
"pmaxub %%mm1, %%mm4 \n\t"
"psubb %%mm4, %%mm3 \n\t" // dst - pred
"movq %%mm3, (%3, %0) \n\t"
"add $8, %0 \n\t"
"cmp %4, %0 \n\t"
" jb 1b \n\t"
: "+r" (i)
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r"(src1), "r"(src2), "r"(dst), "r"((x86_reg)w)
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
);
l= *left;
lt= *left_top;
dst[0]= src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt)&0xFF);
*left_top= src1[w-1];
*left = src2[w-1];
}
#define DIFF_PIXELS_1(m,a,t,p1,p2)\
"mov"#m" "#p1", "#a" \n\t"\
"mov"#m" "#p2", "#t" \n\t"\
"punpcklbw "#a", "#t" \n\t"\
"punpcklbw "#a", "#a" \n\t"\
"psubw "#t", "#a" \n\t"\
#define DIFF_PIXELS_8(m0,m1,mm,p1,p2,stride,temp) {\
uint8_t *p1b=p1, *p2b=p2;\
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile(\
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
DIFF_PIXELS_1(m0, mm##0, mm##7, (%1), (%2))\
DIFF_PIXELS_1(m0, mm##1, mm##7, (%1,%3), (%2,%3))\
DIFF_PIXELS_1(m0, mm##2, mm##7, (%1,%3,2), (%2,%3,2))\
"add %4, %1 \n\t"\
"add %4, %2 \n\t"\
DIFF_PIXELS_1(m0, mm##3, mm##7, (%1), (%2))\
DIFF_PIXELS_1(m0, mm##4, mm##7, (%1,%3), (%2,%3))\
DIFF_PIXELS_1(m0, mm##5, mm##7, (%1,%3,2), (%2,%3,2))\
DIFF_PIXELS_1(m0, mm##6, mm##7, (%1,%4), (%2,%4))\
"mov"#m1" "#mm"0, %0 \n\t"\
DIFF_PIXELS_1(m0, mm##7, mm##0, (%1,%3,4), (%2,%3,4))\
"mov"#m1" %0, "#mm"0 \n\t"\
: "+m"(temp), "+r"(p1b), "+r"(p2b)\
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
: "r"((x86_reg)stride), "r"((x86_reg)stride*3)\
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
);\
}
//the "+m"(temp) is needed as gcc 2.95 sometimes fails to compile "=m"(temp)
#define DIFF_PIXELS_4x8(p1,p2,stride,temp) DIFF_PIXELS_8(d, q, %%mm, p1, p2, stride, temp)
#define DIFF_PIXELS_8x8(p1,p2,stride,temp) DIFF_PIXELS_8(q, dqa, %%xmm, p1, p2, stride, temp)
#define LBUTTERFLY2(a1,b1,a2,b2)\
"paddw " #b1 ", " #a1 " \n\t"\
"paddw " #b2 ", " #a2 " \n\t"\
"paddw " #b1 ", " #b1 " \n\t"\
"paddw " #b2 ", " #b2 " \n\t"\
"psubw " #a1 ", " #b1 " \n\t"\
"psubw " #a2 ", " #b2 " \n\t"
#define HADAMARD8(m0, m1, m2, m3, m4, m5, m6, m7)\
LBUTTERFLY2(m0, m1, m2, m3)\
LBUTTERFLY2(m4, m5, m6, m7)\
LBUTTERFLY2(m0, m2, m1, m3)\
LBUTTERFLY2(m4, m6, m5, m7)\
LBUTTERFLY2(m0, m4, m1, m5)\
LBUTTERFLY2(m2, m6, m3, m7)\
#define HADAMARD48 HADAMARD8(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm6, %%mm7)
#define MMABS_MMX(a,z)\
"pxor " #z ", " #z " \n\t"\
"pcmpgtw " #a ", " #z " \n\t"\
"pxor " #z ", " #a " \n\t"\
"psubw " #z ", " #a " \n\t"
#define MMABS_MMX2(a,z)\
"pxor " #z ", " #z " \n\t"\
"psubw " #a ", " #z " \n\t"\
"pmaxsw " #z ", " #a " \n\t"
#define MMABS_SSSE3(a,z)\
"pabsw " #a ", " #a " \n\t"
#define MMABS_SUM(a,z, sum)\
MMABS(a,z)\
"paddusw " #a ", " #sum " \n\t"
#define MMABS_SUM_8x8_NOSPILL\
MMABS(%%xmm0, %%xmm8)\
MMABS(%%xmm1, %%xmm9)\
MMABS_SUM(%%xmm2, %%xmm8, %%xmm0)\
MMABS_SUM(%%xmm3, %%xmm9, %%xmm1)\
MMABS_SUM(%%xmm4, %%xmm8, %%xmm0)\
MMABS_SUM(%%xmm5, %%xmm9, %%xmm1)\
MMABS_SUM(%%xmm6, %%xmm8, %%xmm0)\
MMABS_SUM(%%xmm7, %%xmm9, %%xmm1)\
"paddusw %%xmm1, %%xmm0 \n\t"
Change semantic of CONFIG_*, HAVE_* and ARCH_*. They are now always defined to either 0 or 1. Originally committed as revision 16590 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
#if ARCH_X86_64
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
#define MMABS_SUM_8x8_SSE2 MMABS_SUM_8x8_NOSPILL
#else
#define MMABS_SUM_8x8_SSE2\
"movdqa %%xmm7, (%1) \n\t"\
MMABS(%%xmm0, %%xmm7)\
MMABS(%%xmm1, %%xmm7)\
MMABS_SUM(%%xmm2, %%xmm7, %%xmm0)\
MMABS_SUM(%%xmm3, %%xmm7, %%xmm1)\
MMABS_SUM(%%xmm4, %%xmm7, %%xmm0)\
MMABS_SUM(%%xmm5, %%xmm7, %%xmm1)\
MMABS_SUM(%%xmm6, %%xmm7, %%xmm0)\
"movdqa (%1), %%xmm2 \n\t"\
MMABS_SUM(%%xmm2, %%xmm7, %%xmm1)\
"paddusw %%xmm1, %%xmm0 \n\t"
#endif
/* FIXME: HSUM_* saturates at 64k, while an 8x8 hadamard or dct block can get up to
* about 100k on extreme inputs. But that's very unlikely to occur in natural video,
* and it's even more unlikely to not have any alternative mvs/modes with lower cost. */
#define HSUM_MMX(a, t, dst)\
"movq "#a", "#t" \n\t"\
"psrlq $32, "#a" \n\t"\
"paddusw "#t", "#a" \n\t"\
"movq "#a", "#t" \n\t"\
"psrlq $16, "#a" \n\t"\
"paddusw "#t", "#a" \n\t"\
"movd "#a", "#dst" \n\t"\
#define HSUM_MMX2(a, t, dst)\
"pshufw $0x0E, "#a", "#t" \n\t"\
"paddusw "#t", "#a" \n\t"\
"pshufw $0x01, "#a", "#t" \n\t"\
"paddusw "#t", "#a" \n\t"\
"movd "#a", "#dst" \n\t"\
#define HSUM_SSE2(a, t, dst)\
"movhlps "#a", "#t" \n\t"\
"paddusw "#t", "#a" \n\t"\
"pshuflw $0x0E, "#a", "#t" \n\t"\
"paddusw "#t", "#a" \n\t"\
"pshuflw $0x01, "#a", "#t" \n\t"\
"paddusw "#t", "#a" \n\t"\
"movd "#a", "#dst" \n\t"\
#define HADAMARD8_DIFF_MMX(cpu) \
static int hadamard8_diff_##cpu(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){\
Remove DECLARE_ALIGNED_{8,16} macros These macros are redundant. All uses are replaced with the generic DECLARE_ALIGNED macro instead. Originally committed as revision 22233 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
DECLARE_ALIGNED(8, uint64_t, temp)[13];\
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
int sum;\
\
assert(h==8);\
\
DIFF_PIXELS_4x8(src1, src2, stride, temp[0]);\
\
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile(\
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
HADAMARD48\
\
"movq %%mm7, 96(%1) \n\t"\
\
TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)\
Make LOAD4/STORE4 macros more generic. Patch by Victor Pollex victor pollex web de Original thread: [PATCH] mmx implementation of vc-1 inverse transformations Date: 06/21/2008 03:37 PM Originally committed as revision 14108 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
STORE4(8, 0(%1), %%mm0, %%mm3, %%mm7, %%mm2)\
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
\
"movq 96(%1), %%mm7 \n\t"\
TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)\
Make LOAD4/STORE4 macros more generic. Patch by Victor Pollex victor pollex web de Original thread: [PATCH] mmx implementation of vc-1 inverse transformations Date: 06/21/2008 03:37 PM Originally committed as revision 14108 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
STORE4(8, 64(%1), %%mm4, %%mm7, %%mm0, %%mm6)\
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
\
: "=r" (sum)\
: "r"(temp)\
);\
\
DIFF_PIXELS_4x8(src1+4, src2+4, stride, temp[4]);\
\
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile(\
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
HADAMARD48\
\
"movq %%mm7, 96(%1) \n\t"\
\
TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)\
Make LOAD4/STORE4 macros more generic. Patch by Victor Pollex victor pollex web de Original thread: [PATCH] mmx implementation of vc-1 inverse transformations Date: 06/21/2008 03:37 PM Originally committed as revision 14108 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
STORE4(8, 32(%1), %%mm0, %%mm3, %%mm7, %%mm2)\
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
\
"movq 96(%1), %%mm7 \n\t"\
TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)\
"movq %%mm7, %%mm5 \n\t"/*FIXME remove*/\
"movq %%mm6, %%mm7 \n\t"\
"movq %%mm0, %%mm6 \n\t"\
\
Make LOAD4/STORE4 macros more generic. Patch by Victor Pollex victor pollex web de Original thread: [PATCH] mmx implementation of vc-1 inverse transformations Date: 06/21/2008 03:37 PM Originally committed as revision 14108 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
LOAD4(8, 64(%1), %%mm0, %%mm1, %%mm2, %%mm3)\
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
\
HADAMARD48\
"movq %%mm7, 64(%1) \n\t"\
MMABS(%%mm0, %%mm7)\
MMABS(%%mm1, %%mm7)\
MMABS_SUM(%%mm2, %%mm7, %%mm0)\
MMABS_SUM(%%mm3, %%mm7, %%mm1)\
MMABS_SUM(%%mm4, %%mm7, %%mm0)\
MMABS_SUM(%%mm5, %%mm7, %%mm1)\
MMABS_SUM(%%mm6, %%mm7, %%mm0)\
"movq 64(%1), %%mm2 \n\t"\
MMABS_SUM(%%mm2, %%mm7, %%mm1)\
"paddusw %%mm1, %%mm0 \n\t"\
"movq %%mm0, 64(%1) \n\t"\
\
Make LOAD4/STORE4 macros more generic. Patch by Victor Pollex victor pollex web de Original thread: [PATCH] mmx implementation of vc-1 inverse transformations Date: 06/21/2008 03:37 PM Originally committed as revision 14108 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
LOAD4(8, 0(%1), %%mm0, %%mm1, %%mm2, %%mm3)\
LOAD4(8, 32(%1), %%mm4, %%mm5, %%mm6, %%mm7)\
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
\
HADAMARD48\
"movq %%mm7, (%1) \n\t"\
MMABS(%%mm0, %%mm7)\
MMABS(%%mm1, %%mm7)\
MMABS_SUM(%%mm2, %%mm7, %%mm0)\
MMABS_SUM(%%mm3, %%mm7, %%mm1)\
MMABS_SUM(%%mm4, %%mm7, %%mm0)\
MMABS_SUM(%%mm5, %%mm7, %%mm1)\
MMABS_SUM(%%mm6, %%mm7, %%mm0)\
"movq (%1), %%mm2 \n\t"\
MMABS_SUM(%%mm2, %%mm7, %%mm1)\
"paddusw 64(%1), %%mm0 \n\t"\
"paddusw %%mm1, %%mm0 \n\t"\
\
HSUM(%%mm0, %%mm1, %0)\
\
: "=r" (sum)\
: "r"(temp)\
);\
return sum&0xFFFF;\
}\
WRAPPER8_16_SQ(hadamard8_diff_##cpu, hadamard8_diff16_##cpu)
#define HADAMARD8_DIFF_SSE2(cpu) \
static int hadamard8_diff_##cpu(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){\
Remove DECLARE_ALIGNED_{8,16} macros These macros are redundant. All uses are replaced with the generic DECLARE_ALIGNED macro instead. Originally committed as revision 22233 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
DECLARE_ALIGNED(16, uint64_t, temp)[4];\
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
int sum;\
\
assert(h==8);\
\
DIFF_PIXELS_8x8(src1, src2, stride, temp[0]);\
\
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile(\
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
HADAMARD8(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7)\
TRANSPOSE8(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7, (%1))\
HADAMARD8(%%xmm0, %%xmm5, %%xmm7, %%xmm3, %%xmm6, %%xmm4, %%xmm2, %%xmm1)\
MMABS_SUM_8x8\
HSUM_SSE2(%%xmm0, %%xmm1, %0)\
: "=r" (sum)\
: "r"(temp)\
);\
return sum&0xFFFF;\
}\
WRAPPER8_16_SQ(hadamard8_diff_##cpu, hadamard8_diff16_##cpu)
#define MMABS(a,z) MMABS_MMX(a,z)
#define HSUM(a,t,dst) HSUM_MMX(a,t,dst)
HADAMARD8_DIFF_MMX(mmx)
#undef MMABS
#undef HSUM
#define MMABS(a,z) MMABS_MMX2(a,z)
#define MMABS_SUM_8x8 MMABS_SUM_8x8_SSE2
#define HSUM(a,t,dst) HSUM_MMX2(a,t,dst)
HADAMARD8_DIFF_MMX(mmx2)
HADAMARD8_DIFF_SSE2(sse2)
#undef MMABS
#undef MMABS_SUM_8x8
#undef HSUM
Change semantic of CONFIG_*, HAVE_* and ARCH_*. They are now always defined to either 0 or 1. Originally committed as revision 16590 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
#if HAVE_SSSE3
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
#define MMABS(a,z) MMABS_SSSE3(a,z)
#define MMABS_SUM_8x8 MMABS_SUM_8x8_NOSPILL
HADAMARD8_DIFF_SSE2(ssse3)
#undef MMABS
#undef MMABS_SUM_8x8
#endif
#define DCT_SAD4(m,mm,o)\
"mov"#m" "#o"+ 0(%1), "#mm"2 \n\t"\
"mov"#m" "#o"+16(%1), "#mm"3 \n\t"\
"mov"#m" "#o"+32(%1), "#mm"4 \n\t"\
"mov"#m" "#o"+48(%1), "#mm"5 \n\t"\
MMABS_SUM(mm##2, mm##6, mm##0)\
MMABS_SUM(mm##3, mm##7, mm##1)\
MMABS_SUM(mm##4, mm##6, mm##0)\
MMABS_SUM(mm##5, mm##7, mm##1)\
#define DCT_SAD_MMX\
"pxor %%mm0, %%mm0 \n\t"\
"pxor %%mm1, %%mm1 \n\t"\
DCT_SAD4(q, %%mm, 0)\
DCT_SAD4(q, %%mm, 8)\
DCT_SAD4(q, %%mm, 64)\
DCT_SAD4(q, %%mm, 72)\
"paddusw %%mm1, %%mm0 \n\t"\
HSUM(%%mm0, %%mm1, %0)
#define DCT_SAD_SSE2\
"pxor %%xmm0, %%xmm0 \n\t"\
"pxor %%xmm1, %%xmm1 \n\t"\
DCT_SAD4(dqa, %%xmm, 0)\
DCT_SAD4(dqa, %%xmm, 64)\
"paddusw %%xmm1, %%xmm0 \n\t"\
HSUM(%%xmm0, %%xmm1, %0)
#define DCT_SAD_FUNC(cpu) \
static int sum_abs_dctelem_##cpu(DCTELEM *block){\
int sum;\
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile(\
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
DCT_SAD\
:"=r"(sum)\
:"r"(block)\
);\
return sum&0xFFFF;\
}
#define DCT_SAD DCT_SAD_MMX
#define HSUM(a,t,dst) HSUM_MMX(a,t,dst)
#define MMABS(a,z) MMABS_MMX(a,z)
DCT_SAD_FUNC(mmx)
#undef MMABS
#undef HSUM
#define HSUM(a,t,dst) HSUM_MMX2(a,t,dst)
#define MMABS(a,z) MMABS_MMX2(a,z)
DCT_SAD_FUNC(mmx2)
#undef HSUM
#undef DCT_SAD
#define DCT_SAD DCT_SAD_SSE2
#define HSUM(a,t,dst) HSUM_SSE2(a,t,dst)
DCT_SAD_FUNC(sse2)
#undef MMABS
Change semantic of CONFIG_*, HAVE_* and ARCH_*. They are now always defined to either 0 or 1. Originally committed as revision 16590 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
#if HAVE_SSSE3
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
#define MMABS(a,z) MMABS_SSSE3(a,z)
DCT_SAD_FUNC(ssse3)
#undef MMABS
#endif
#undef HSUM
#undef DCT_SAD
static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2, int size){
int sum;
Do not misuse long as the size of a register in x86. typedef x86_reg as the appropriate size and use it instead. Originally committed as revision 13081 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
x86_reg i=size;
Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax. Originally committed as revision 15627 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
__asm__ volatile(
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
"pxor %%mm4, %%mm4 \n"
"1: \n"
"sub $8, %0 \n"
"movq (%2,%0), %%mm2 \n"
"movq (%3,%0,2), %%mm0 \n"
"movq 8(%3,%0,2), %%mm1 \n"
"punpckhbw %%mm2, %%mm3 \n"
"punpcklbw %%mm2, %%mm2 \n"
"psraw $8, %%mm3 \n"
"psraw $8, %%mm2 \n"
"psubw %%mm3, %%mm1 \n"
"psubw %%mm2, %%mm0 \n"
"pmaddwd %%mm1, %%mm1 \n"
"pmaddwd %%mm0, %%mm0 \n"
"paddd %%mm1, %%mm4 \n"
"paddd %%mm0, %%mm4 \n"
"jg 1b \n"
"movq %%mm4, %%mm3 \n"
"psrlq $32, %%mm3 \n"
"paddd %%mm3, %%mm4 \n"
"movd %%mm4, %1 \n"
:"+r"(i), "=r"(sum)
:"r"(pix1), "r"(pix2)
);
return sum;
}
#define PHADDD(a, t)\
"movq "#a", "#t" \n\t"\
"psrlq $32, "#a" \n\t"\
"paddd "#t", "#a" \n\t"
/*
pmulhw: dst[0-15]=(src[0-15]*dst[0-15])[16-31]
pmulhrw: dst[0-15]=(src[0-15]*dst[0-15] + 0x8000)[16-31]
pmulhrsw: dst[0-15]=(src[0-15]*dst[0-15] + 0x4000)[15-30]
*/
#define PMULHRW(x, y, s, o)\
"pmulhw " #s ", "#x " \n\t"\
"pmulhw " #s ", "#y " \n\t"\
"paddw " #o ", "#x " \n\t"\
"paddw " #o ", "#y " \n\t"\
"psraw $1, "#x " \n\t"\
"psraw $1, "#y " \n\t"
#define DEF(x) x ## _mmx
#define SET_RND MOVQ_WONE
#define SCALE_OFFSET 1
Rename template included sources from .h to _template.c. There are multiple source files that are #include'd rather than compiled, as they are used as template for generation of similar code, like asm-optimised code. Some of these files are right now named with a .h extension, although they are not header in any reasonable sense. Rename them so that instead of being named with .h extension they are named with _template.c as final part. Originally committed as revision 15730 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
#include "dsputil_mmx_qns_template.c"
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
#undef DEF
#undef SET_RND
#undef SCALE_OFFSET
#undef PMULHRW
#define DEF(x) x ## _3dnow
#define SET_RND(x)
#define SCALE_OFFSET 0
#define PMULHRW(x, y, s, o)\
"pmulhrw " #s ", "#x " \n\t"\
"pmulhrw " #s ", "#y " \n\t"
Rename template included sources from .h to _template.c. There are multiple source files that are #include'd rather than compiled, as they are used as template for generation of similar code, like asm-optimised code. Some of these files are right now named with a .h extension, although they are not header in any reasonable sense. Rename them so that instead of being named with .h extension they are named with _template.c as final part. Originally committed as revision 15730 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
#include "dsputil_mmx_qns_template.c"
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
#undef DEF
#undef SET_RND
#undef SCALE_OFFSET
#undef PMULHRW
Change semantic of CONFIG_*, HAVE_* and ARCH_*. They are now always defined to either 0 or 1. Originally committed as revision 16590 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
#if HAVE_SSSE3
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
#undef PHADDD
#define DEF(x) x ## _ssse3
#define SET_RND(x)
#define SCALE_OFFSET -1
#define PHADDD(a, t)\
"pshufw $0x0E, "#a", "#t" \n\t"\
"paddd "#t", "#a" \n\t" /* faster than phaddd on core2 */
#define PMULHRW(x, y, s, o)\
"pmulhrsw " #s ", "#x " \n\t"\
"pmulhrsw " #s ", "#y " \n\t"
Rename template included sources from .h to _template.c. There are multiple source files that are #include'd rather than compiled, as they are used as template for generation of similar code, like asm-optimised code. Some of these files are right now named with a .h extension, although they are not header in any reasonable sense. Rename them so that instead of being named with .h extension they are named with _template.c as final part. Originally committed as revision 15730 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
#include "dsputil_mmx_qns_template.c"
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
#undef DEF
#undef SET_RND
#undef SCALE_OFFSET
#undef PMULHRW
#undef PHADDD
#endif //HAVE_SSSE3
void dsputilenc_init_mmx(DSPContext* c, AVCodecContext *avctx)
{
Remove duplicated MM_* macros for CPU capabilities from dsputil.h. Add missing one for FF_MM_ALTIVEC to avcodec.h. Rename all the occurences of MM_* to the corresponding FF_MM_*. Originally committed as revision 15770 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
if (mm_flags & FF_MM_MMX) {
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
const int dct_algo = avctx->dct_algo;
if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){
Remove duplicated MM_* macros for CPU capabilities from dsputil.h. Add missing one for FF_MM_ALTIVEC to avcodec.h. Rename all the occurences of MM_* to the corresponding FF_MM_*. Originally committed as revision 15770 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
if(mm_flags & FF_MM_SSE2){
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
c->fdct = ff_fdct_sse2;
Rename FF_MM_MMXEXT to FF_MM_MMX2, for both clarity and consistency with libswscale. Originally committed as revision 18330 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
}else if(mm_flags & FF_MM_MMX2){
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
c->fdct = ff_fdct_mmx2;
}else{
c->fdct = ff_fdct_mmx;
}
}
c->get_pixels = get_pixels_mmx;
c->diff_pixels = diff_pixels_mmx;
c->pix_sum = pix_sum16_mmx;
c->diff_bytes= diff_bytes_mmx;
c->sum_abs_dctelem= sum_abs_dctelem_mmx;
c->hadamard8_diff[0]= hadamard8_diff16_mmx;
c->hadamard8_diff[1]= hadamard8_diff_mmx;
c->pix_norm1 = pix_norm1_mmx;
Remove duplicated MM_* macros for CPU capabilities from dsputil.h. Add missing one for FF_MM_ALTIVEC to avcodec.h. Rename all the occurences of MM_* to the corresponding FF_MM_*. Originally committed as revision 15770 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
c->sse[0] = (mm_flags & FF_MM_SSE2) ? sse16_sse2 : sse16_mmx;
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
c->sse[1] = sse8_mmx;
c->vsad[4]= vsad_intra16_mmx;
c->nsse[0] = nsse16_mmx;
c->nsse[1] = nsse8_mmx;
if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
c->vsad[0] = vsad16_mmx;
}
if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
c->try_8x8basis= try_8x8basis_mmx;
}
c->add_8x8basis= add_8x8basis_mmx;
c->ssd_int8_vs_int16 = ssd_int8_vs_int16_mmx;
Rename FF_MM_MMXEXT to FF_MM_MMX2, for both clarity and consistency with libswscale. Originally committed as revision 18330 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
if (mm_flags & FF_MM_MMX2) {
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
c->sum_abs_dctelem= sum_abs_dctelem_mmx2;
c->hadamard8_diff[0]= hadamard8_diff16_mmx2;
c->hadamard8_diff[1]= hadamard8_diff_mmx2;
c->vsad[4]= vsad_intra16_mmx2;
if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
c->vsad[0] = vsad16_mmx2;
}
c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_mmx2;
}
Remove duplicated MM_* macros for CPU capabilities from dsputil.h. Add missing one for FF_MM_ALTIVEC to avcodec.h. Rename all the occurences of MM_* to the corresponding FF_MM_*. Originally committed as revision 15770 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
if(mm_flags & FF_MM_SSE2){
get_pixels_sse2, ~+12% performance compared to mmx Originally committed as revision 15591 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
c->get_pixels = get_pixels_sse2;
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
c->sum_abs_dctelem= sum_abs_dctelem_sse2;
c->hadamard8_diff[0]= hadamard8_diff16_sse2;
c->hadamard8_diff[1]= hadamard8_diff_sse2;
add CONFIG_LPC to the build system for lpc dsputil functions. fixes build problems when lpc.c is not compiled. Originally committed as revision 20285 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
#if CONFIG_LPC
Move autocorrelation function from flacenc.c to lpc.c. Also rename the corresponding dsputil functions and remove their dependency on the FLAC encoder. Fixes Issue1486. Originally committed as revision 20266 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
c->lpc_compute_autocorr = ff_lpc_compute_autocorr_sse2;
add CONFIG_LPC to the build system for lpc dsputil functions. fixes build problems when lpc.c is not compiled. Originally committed as revision 20285 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
#endif
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
}
Change semantic of CONFIG_*, HAVE_* and ARCH_*. They are now always defined to either 0 or 1. Originally committed as revision 16590 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
#if HAVE_SSSE3
Remove duplicated MM_* macros for CPU capabilities from dsputil.h. Add missing one for FF_MM_ALTIVEC to avcodec.h. Rename all the occurences of MM_* to the corresponding FF_MM_*. Originally committed as revision 15770 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
if(mm_flags & FF_MM_SSSE3){
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
c->try_8x8basis= try_8x8basis_ssse3;
}
c->add_8x8basis= add_8x8basis_ssse3;
c->sum_abs_dctelem= sum_abs_dctelem_ssse3;
c->hadamard8_diff[0]= hadamard8_diff16_ssse3;
c->hadamard8_diff[1]= hadamard8_diff_ssse3;
}
#endif
Remove duplicated MM_* macros for CPU capabilities from dsputil.h. Add missing one for FF_MM_ALTIVEC to avcodec.h. Rename all the occurences of MM_* to the corresponding FF_MM_*. Originally committed as revision 15770 to svn://svn.ffmpeg.org/ffmpeg/trunk
16 years ago
if(mm_flags & FF_MM_3DNOW){
split encoding part of dsputil_mmx into its own file Originally committed as revision 12223 to svn://svn.ffmpeg.org/ffmpeg/trunk
17 years ago
if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
c->try_8x8basis= try_8x8basis_3dnow;
}
c->add_8x8basis= add_8x8basis_3dnow;
}
}
dsputil_init_pix_mmx(c, avctx);
}