Chiebot-Mirror
/
FFmpeg
mirror of https://github.com/FFmpeg/FFmpeg.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

avfilter/vf_framerate: add SIMD functions for frame blending

Browse Source 
Blend function speedups on x86_64 Core i5 4460:

ffmpeg -f lavfi -i allyuv -vf framerate=60:threads=1 -f null none

C:     447548411 decicycles in Blend,    2048 runs,      0 skips
SSSE3: 130020087 decicycles in Blend,    2048 runs,      0 skips
AVX2:  128508221 decicycles in Blend,    2048 runs,      0 skips

ffmpeg -f lavfi -i allyuv -vf format=yuv420p12,framerate=60:threads=1 -f null none

C:     228932745 decicycles in Blend,    2048 runs,      0 skips
SSE4:  123357781 decicycles in Blend,    2048 runs,      0 skips
AVX2:  121215353 decicycles in Blend,    2048 runs,      0 skips

Signed-off-by: Marton Balint <cus@passwd.hu>
pull/277/head
Marton Balint 7 years ago
parent 2cbe6bac03
commit 4d95c6d5d7
5 changed files with 268 additions and 54 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 74
      libavfilter/framerate.h
  2. 70
      libavfilter/vf_framerate.c
  3. 2
      libavfilter/x86/Makefile
  4. 134
      libavfilter/x86/vf_framerate.asm
  5. 42
      libavfilter/x86/vf_framerate_init.c

74
libavfilter/framerate.h
Unescape View File

@ -0,0 +1,74 @@
/*
* 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
*/
#ifndef AVFILTER_FRAMERATE_H
#define AVFILTER_FRAMERATE_H
#include "libavutil/pixelutils.h"
#include "avfilter.h"
#define BLEND_FUNC_PARAMS const uint8_t *src1, ptrdiff_t src1_linesize, \
const uint8_t *src2, ptrdiff_t src2_linesize, \
uint8_t *dst, ptrdiff_t dst_linesize, \
ptrdiff_t width, ptrdiff_t height, \
int factor1, int factor2, int half
#define BLEND_FACTOR_DEPTH8 7
#define BLEND_FACTOR_DEPTH16 15
typedef void (*blend_func)(BLEND_FUNC_PARAMS);
typedef struct FrameRateContext {
const AVClass *class;
// parameters
AVRational dest_frame_rate; ///< output frames per second
int flags; ///< flags affecting frame rate conversion algorithm
double scene_score; ///< score that denotes a scene change has happened
int interp_start; ///< start of range to apply linear interpolation
int interp_end; ///< end of range to apply linear interpolation
int line_size[4]; ///< bytes of pixel data per line for each plane
int vsub;
AVRational srce_time_base; ///< timebase of source
AVRational dest_time_base; ///< timebase of destination
av_pixelutils_sad_fn sad; ///< Sum of the absolute difference function (scene detect only)
double prev_mafd; ///< previous MAFD (scene detect only)
int blend_factor_max;
int bitdepth;
AVFrame *work;
AVFrame *f0; ///< last frame
AVFrame *f1; ///< current frame
int64_t pts0; ///< last frame pts in dest_time_base
int64_t pts1; ///< current frame pts in dest_time_base
int64_t delta; ///< pts1 to pts0 delta
double score; ///< scene change score (f0 to f1)
int flush; ///< 1 if the filter is being flushed
int64_t start_pts; ///< pts of the first output frame
int64_t n; ///< output frame counter
blend_func blend;
} FrameRateContext;
void ff_framerate_init(FrameRateContext *s);
void ff_framerate_init_x86(FrameRateContext *s);
#endif /* AVFILTER_FRAMERATE_H */

70
libavfilter/vf_framerate.c
Unescape View File

@ -38,52 +38,7 @@
#include "avfilter.h"
#include "internal.h"
#include "video.h"
#define BLEND_FUNC_PARAMS const uint8_t *src1, ptrdiff_t src1_linesize, \
const uint8_t *src2, ptrdiff_t src2_linesize, \
uint8_t *dst, ptrdiff_t dst_linesize, \
ptrdiff_t width, ptrdiff_t height, \
int factor1, int factor2, int half
#define BLEND_FACTOR_DEPTH8 7
#define BLEND_FACTOR_DEPTH16 15
typedef void (*blend_func)(BLEND_FUNC_PARAMS);
typedef struct FrameRateContext {
const AVClass *class;
// parameters
AVRational dest_frame_rate; ///< output frames per second
int flags; ///< flags affecting frame rate conversion algorithm
double scene_score; ///< score that denotes a scene change has happened
int interp_start; ///< start of range to apply linear interpolation
int interp_end; ///< end of range to apply linear interpolation
int line_size[4]; ///< bytes of pixel data per line for each plane
int vsub;
AVRational srce_time_base; ///< timebase of source
AVRational dest_time_base; ///< timebase of destination
av_pixelutils_sad_fn sad; ///< Sum of the absolute difference function (scene detect only)
double prev_mafd; ///< previous MAFD (scene detect only)
int blend_factor_max;
int bitdepth;
AVFrame *work;
AVFrame *f0; ///< last frame
AVFrame *f1; ///< current frame
int64_t pts0; ///< last frame pts in dest_time_base
int64_t pts1; ///< current frame pts in dest_time_base
int64_t delta; ///< pts1 to pts0 delta
double score; ///< scene change score (f0 to f1)
int flush; ///< 1 if the filter is being flushed
int64_t start_pts; ///< pts of the first output frame
int64_t n; ///< output frame counter
blend_func blend;
} FrameRateContext;
#include "framerate.h"
#define OFFSET(x) offsetof(FrameRateContext, x)
#define V AV_OPT_FLAG_VIDEO_PARAM
@ -246,7 +201,7 @@ static int blend_frames(AVFilterContext *ctx, int interpolate)
av_frame_copy_props(s->work, s->f0);
ff_dlog(ctx, "blend_frames() INTERPOLATE to create work frame\n");
ctx->internal->execute(ctx, filter_slice, &td, NULL, FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
ctx->internal->execute(ctx, filter_slice, &td, NULL, FFMIN(FFMAX(1, outlink->h >> 2), ff_filter_get_nb_threads(ctx)));
return 1;
}
return 0;
@ -366,6 +321,19 @@ static void blend_frames16_c(BLEND_FUNC_PARAMS)
}
}
void ff_framerate_init(FrameRateContext *s)
{
if (s->bitdepth == 8) {
s->blend_factor_max = 1 << BLEND_FACTOR_DEPTH8;
s->blend = blend_frames_c;
} else {
s->blend_factor_max = 1 << BLEND_FACTOR_DEPTH16;
s->blend = blend_frames16_c;
}
if (ARCH_X86)
ff_framerate_init_x86(s);
}
static int config_input(AVFilterLink *inlink)
{
AVFilterContext *ctx = inlink->dst;
@ -387,13 +355,7 @@ static int config_input(AVFilterLink *inlink)
s->srce_time_base = inlink->time_base;
if (s->bitdepth == 8) {
s->blend_factor_max = 1 << BLEND_FACTOR_DEPTH8;
s->blend = blend_frames_c;
} else {
s->blend_factor_max = 1 << BLEND_FACTOR_DEPTH16;
s->blend = blend_frames16_c;
}
ff_framerate_init(s);
return 0;
}

2
libavfilter/x86/Makefile
Unescape View File

@ -5,6 +5,7 @@ OBJS-$(CONFIG_COLORSPACE_FILTER) += x86/colorspacedsp_init.o
OBJS-$(CONFIG_EQ_FILTER) += x86/vf_eq.o
OBJS-$(CONFIG_FSPP_FILTER) += x86/vf_fspp_init.o
OBJS-$(CONFIG_GRADFUN_FILTER) += x86/vf_gradfun_init.o
OBJS-$(CONFIG_FRAMERATE_FILTER) += x86/vf_framerate_init.o
OBJS-$(CONFIG_HFLIP_FILTER) += x86/vf_hflip_init.o
OBJS-$(CONFIG_HQDN3D_FILTER) += x86/vf_hqdn3d_init.o
OBJS-$(CONFIG_IDET_FILTER) += x86/vf_idet_init.o
@ -31,6 +32,7 @@ X86ASM-OBJS-$(CONFIG_AFIR_FILTER) += x86/af_afir.o
X86ASM-OBJS-$(CONFIG_BLEND_FILTER) += x86/vf_blend.o
X86ASM-OBJS-$(CONFIG_BWDIF_FILTER) += x86/vf_bwdif.o
X86ASM-OBJS-$(CONFIG_COLORSPACE_FILTER) += x86/colorspacedsp.o
X86ASM-OBJS-$(CONFIG_FRAMERATE_FILTER) += x86/vf_framerate.o
X86ASM-OBJS-$(CONFIG_FSPP_FILTER) += x86/vf_fspp.o
X86ASM-OBJS-$(CONFIG_GRADFUN_FILTER) += x86/vf_gradfun.o
X86ASM-OBJS-$(CONFIG_HFLIP_FILTER) += x86/vf_hflip.o

134
libavfilter/x86/vf_framerate.asm
Unescape View File

@ -0,0 +1,134 @@
;*****************************************************************************
;* x86-optimized functions for framerate filter
;*
;* Copyright (C) 2018 Marton Balint
;*
;* Based on vf_blend.asm, Copyright (C) 2015 Paul B Mahol
;*
;* 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/x86/x86util.asm"
SECTION .text
%macro XSPLAT 3
%if cpuflag(avx2)
vpbroadcast%3 %1, %2
%else
movd %1, %2
%ifidn %3, d
SPLATD %1
%else
SPLATW %1, %1
%endif
%endif
%endmacro
%macro BLEND_INIT 0-1
%if ARCH_X86_64
cglobal blend_frames%1, 6, 9, 5, src1, src1_linesize, src2, src2_linesize, dst, dst_linesize, width, end, x
mov widthd, dword widthm
%else
cglobal blend_frames%1, 5, 7, 5, src1, src1_linesize, src2, src2_linesize, dst, end, x
%define dst_linesizeq r5mp
%define widthq r6mp
%endif
mov endd, dword r7m
add src1q, widthq
add src2q, widthq
add dstq, widthq
neg widthq
%endmacro
%macro BLEND_LOOP 4
.nextrow:
mov xq, widthq
.loop:
movu m0, [src1q + xq]
movu m1, [src2q + xq]
SBUTTERFLY %1%2, 0, 1, 4 ; aAbBcCdD
; eEfFgGhH
pmadd%3 m0, m2
pmadd%3 m1, m2
padd%2 m0, m3
padd%2 m1, m3
psrl%2 m0, %4 ; 0A0B0C0D
psrl%2 m1, %4 ; 0E0F0G0H
packus%2%1 m0, m1 ; ABCDEFGH
movu [dstq + xq], m0
add xq, mmsize
jl .loop
add src1q, src1_linesizeq
add src2q, src2_linesizeq
add dstq, dst_linesizeq
sub endd, 1
jg .nextrow
REP_RET
%endmacro
%macro BLEND_FRAMES 0
BLEND_INIT
XSPLAT m2, r8m, w ; factor1
XSPLAT m3, r9m, w ; factor2
psllw m3, 8
por m2, m3 ; interleaved factors
XSPLAT m3, r10m, w ; half
BLEND_LOOP b, w, ubsw, 7
%endmacro
%macro BLEND_FRAMES16 0
BLEND_INIT 16
XSPLAT m2, r8m, d ; factor1
XSPLAT m3, r9m, d ; factor2
pslld m3, 16
por m2, m3 ; interleaved factors
XSPLAT m3, r10m, d ; half
BLEND_LOOP w, d, wd, 15
%endmacro
INIT_XMM ssse3
BLEND_FRAMES
INIT_XMM sse4
BLEND_FRAMES16
%if HAVE_AVX2_EXTERNAL
INIT_YMM avx2
BLEND_FRAMES
BLEND_FRAMES16
%endif

42
libavfilter/x86/vf_framerate_init.c
Unescape View File

@ -0,0 +1,42 @@
/*
* 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/cpu.h"
#include "libavutil/x86/cpu.h"
#include "libavfilter/framerate.h"
void ff_blend_frames_ssse3(BLEND_FUNC_PARAMS);
void ff_blend_frames_avx2(BLEND_FUNC_PARAMS);
void ff_blend_frames16_sse4(BLEND_FUNC_PARAMS);
void ff_blend_frames16_avx2(BLEND_FUNC_PARAMS);
void ff_framerate_init_x86(FrameRateContext *s)
{
int cpu_flags = av_get_cpu_flags();
if (s->bitdepth == 8) {
if (EXTERNAL_AVX2_FAST(cpu_flags))
s->blend = ff_blend_frames_avx2;
else if (EXTERNAL_SSSE3(cpu_flags))
s->blend = ff_blend_frames_ssse3;
} else {
if (EXTERNAL_AVX2_FAST(cpu_flags))
s->blend = ff_blend_frames16_avx2;
else if (EXTERNAL_SSE4(cpu_flags))
s->blend = ff_blend_frames16_sse4;
}
}
Write Preview
Loading…
Cancel
Save