eessppeelllloo/ffmpeg
1
0
Fork 0
mirror of https://git.ffmpeg.org/ffmpeg.git synced 2026-01-06 22:25:36 +01:00
Code Issues Packages Projects Releases Wiki Activity
Files
a6ccaa2eea1900676f975c1dd720638ea2ee828d
ffmpeg/libavcodec/x86/me_cmp_init.c
Andreas Rheinhardt d91b1559e0 avcodec/x86/me_cmp: Replace MMXEXT size 16 funcs by unaligned SSE2 funcs 
Snow calls some of the me_cmp_funcs with insufficient alignment
for the first pointer (see get_block_rd() in snowenc.c);
therefore SSE2 functions which really need this alignment
don't get set for Snow and 542765ce3e
consequently didn't remove MMXEXT functions which are overridden
by these SSE2 functions for normal codecs.

For reference, here is a command line which would segfault
if one simply used the ordinary SSE2 functions for Snow:

./ffmpeg -i mm-short.mpg -an -vcodec snow -t 0.2 -pix_fmt yuv444p \
-vstrict -2 -qscale 2 -flags +qpel -motion_est iter 444iter.avi

This commit adds unaligned SSE2 versions of these functions
and removes the MMXEXT ones. This in particular implies that
sad 16x16 now never uses MMX which allows to remove an emms_c
from ac3enc.c.

Benchmarks (u means unaligned version):
sad_0_c:                                                 8.2 ( 1.00x)
sad_0_mmxext:                                           10.8 ( 0.76x)
sad_0_sse2:                                              6.2 ( 1.33x)
sad_0_sse2u:                                             6.7 ( 1.23x)

vsad_0_c:                                               44.7 ( 1.00x)
vsad_0_mmxext (approx):                                 12.2 ( 3.68x)
vsad_0_sse2 (approx):                                    7.8 ( 5.75x)

vsad_4_c:                                               88.4 ( 1.00x)
vsad_4_mmxext:                                           7.1 (12.46x)
vsad_4_sse2:                                             4.2 (21.15x)
vsad_4_sse2u:                                            5.5 (15.96x)

Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
2025-10-17 13:05:07 +02:00

198 lines
7.8 KiB
C
Raw Blame History

/*
* SIMD-optimized motion estimation
* Copyright (c) 2000, 2001 Fabrice Bellard
* Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
*
* MMX optimization by Nick Kurshev <nickols_k@mail.ru>
*
* 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
*/
#include "libavutil/attributes.h"
#include "libavutil/cpu.h"
#include "libavutil/x86/cpu.h"
#include "libavcodec/me_cmp.h"
#include "libavcodec/mpegvideoenc.h"
int ff_sum_abs_dctelem_sse2(const int16_t *block);
int ff_sum_abs_dctelem_ssse3(const int16_t *block);
int ff_sse8_mmx(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_sse16_mmx(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_sse16_sse2(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_hf_noise8_mmx(const uint8_t *pix1, ptrdiff_t stride, int h);
int ff_hf_noise16_mmx(const uint8_t *pix1, ptrdiff_t stride, int h);
int ff_sad8_mmxext(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_sad16_sse2(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_sad16u_sse2(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_sad8_x2_mmxext(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_sad16_x2_sse2(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_sad8_y2_mmxext(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_sad16_y2_sse2(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_sad8_approx_xy2_mmxext(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_sad8_xy2_sse2(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_sad16_approx_xy2_sse2(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_sad16_xy2_sse2(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_vsad_intra8_mmxext(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_vsad_intra16_sse2(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_vsad_intra16u_sse2(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_vsad8_approx_mmxext(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_vsad16_approx_sse2(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
int ff_vsad16u_approx_sse2(MPVEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h);
#define hadamard_func(cpu) \
int ff_hadamard8_diff_ ## cpu(MPVEncContext *s, const uint8_t *src1, \
const uint8_t *src2, ptrdiff_t stride, int h); \
int ff_hadamard8_diff16_ ## cpu(MPVEncContext *s, const uint8_t *src1, \
const uint8_t *src2, ptrdiff_t stride, int h);
hadamard_func(mmxext)
hadamard_func(sse2)
hadamard_func(ssse3)
#if HAVE_X86ASM
static int nsse16_mmx(MPVEncContext *c, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h)
{
int score1, score2;
if (c)
score1 = c->sse_cmp[0](c, pix1, pix2, stride, h);
else
score1 = ff_sse16_mmx(c, pix1, pix2, stride, h);
score2 = ff_hf_noise16_mmx(pix1, stride, h) + ff_hf_noise8_mmx(pix1+8, stride, h)
- ff_hf_noise16_mmx(pix2, stride, h) - ff_hf_noise8_mmx(pix2+8, stride, h);
if (c)
return score1 + FFABS(score2) * c->c.avctx->nsse_weight;
else
return score1 + FFABS(score2) * 8;
}
static int nsse8_mmx(MPVEncContext *c, const uint8_t *pix1, const uint8_t *pix2,
ptrdiff_t stride, int h)
{
int score1 = ff_sse8_mmx(c, pix1, pix2, stride, h);
int score2 = ff_hf_noise8_mmx(pix1, stride, h) -
ff_hf_noise8_mmx(pix2, stride, h);
if (c)
return score1 + FFABS(score2) * c->c.avctx->nsse_weight;
else
return score1 + FFABS(score2) * 8;
}
#endif /* HAVE_X86ASM */
av_cold void ff_me_cmp_init_x86(MECmpContext *c, AVCodecContext *avctx)
{
int cpu_flags = av_get_cpu_flags();
if (EXTERNAL_MMX(cpu_flags)) {
c->sse[1] = ff_sse8_mmx;
#if HAVE_X86ASM
c->nsse[0] = nsse16_mmx;
c->nsse[1] = nsse8_mmx;
#endif
}
if (EXTERNAL_MMXEXT(cpu_flags)) {
#if !HAVE_ALIGNED_STACK
c->hadamard8_diff[0] = ff_hadamard8_diff16_mmxext;
c->hadamard8_diff[1] = ff_hadamard8_diff_mmxext;
#endif
c->sad[1] = ff_sad8_mmxext;
c->pix_abs[1][0] = ff_sad8_mmxext;
c->pix_abs[1][1] = ff_sad8_x2_mmxext;
c->pix_abs[1][2] = ff_sad8_y2_mmxext;
c->vsad[5] = ff_vsad_intra8_mmxext;
if (!(avctx->flags & AV_CODEC_FLAG_BITEXACT)) {
c->pix_abs[1][3] = ff_sad8_approx_xy2_mmxext;
c->vsad[1] = ff_vsad8_approx_mmxext;
}
}
if (EXTERNAL_SSE2(cpu_flags)) {
c->sse[0] = ff_sse16_sse2;
c->sum_abs_dctelem = ff_sum_abs_dctelem_sse2;
c->pix_abs[0][0] = ff_sad16_sse2;
c->pix_abs[0][1] = ff_sad16_x2_sse2;
c->pix_abs[0][2] = ff_sad16_y2_sse2;
c->pix_abs[0][3] = ff_sad16_xy2_sse2;
#if HAVE_ALIGNED_STACK
c->hadamard8_diff[0] = ff_hadamard8_diff16_sse2;
c->hadamard8_diff[1] = ff_hadamard8_diff_sse2;
#endif
if (avctx->codec_id != AV_CODEC_ID_SNOW) {
c->sad[0] = ff_sad16_sse2;
c->vsad[4] = ff_vsad_intra16_sse2;
if (!(avctx->flags & AV_CODEC_FLAG_BITEXACT)) {
c->vsad[0] = ff_vsad16_approx_sse2;
}
} else {
// Snow does not abide by the alignment requirements
// of blk1, so we use special versions without them for it.
c->sad[0] = ff_sad16u_sse2;
c->vsad[4] = ff_vsad_intra16u_sse2;
if (!(avctx->flags & AV_CODEC_FLAG_BITEXACT)) {
c->vsad[0] = ff_vsad16u_approx_sse2;
}
}
if (avctx->flags & AV_CODEC_FLAG_BITEXACT) {
c->pix_abs[1][3] = ff_sad8_xy2_sse2;
} else {
c->pix_abs[0][3] = ff_sad16_approx_xy2_sse2;
}
}
if (EXTERNAL_SSSE3(cpu_flags)) {
c->sum_abs_dctelem = ff_sum_abs_dctelem_ssse3;
#if HAVE_ALIGNED_STACK
c->hadamard8_diff[0] = ff_hadamard8_diff16_ssse3;
c->hadamard8_diff[1] = ff_hadamard8_diff_ssse3;
#endif
}
}
Reference in New Issue View Git Blame Copy Permalink