Tested using a simple command (apply edge enhance): ./ffmpeg_g -i ~/Downloads/bbb_sunflower_1080p_30fps_normal.mp4 \ -vf convolution="0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:5:1:1:1:0:128:128:128" \ -an -vframes 1000 -f null /dev/null The fps increase from 151 to 270 on my local machine.
Unified · Split
| @@ -0,0 +1,64 @@ | ||
| 1 | +/* | |
| 2 | + * Copyright (c) 2012-2013 Oka Motofumi (chikuzen.mo at gmail dot com) | |
| 3 | + * Copyright (c) 2015 Paul B Mahol | |
| 4 | + * | |
| 5 | + * This file is part of FFmpeg. | |
| 6 | + * | |
| 7 | + * FFmpeg is free software; you can redistribute it and/or | |
| 8 | + * modify it under the terms of the GNU Lesser General Public | |
| 9 | + * License as published by the Free Software Foundation; either | |
| 10 | + * version 2.1 of the License, or (at your option) any later version. | |
| 11 | + * | |
| 12 | + * FFmpeg is distributed in the hope that it will be useful, | |
| 13 | + * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 14 | + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
| 15 | + * Lesser General Public License for more details. | |
| 16 | + * | |
| 17 | + * You should have received a copy of the GNU Lesser General Public | |
| 18 | + * License along with FFmpeg; if not, write to the Free Software | |
| 19 | + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
| 20 | + */ | |
| 21 | +#ifndef AVFILTER_CONVOLUTION_H | |
| 22 | +#define AVFILTER_CONVOLUTION_H | |
| 23 | +#include "avfilter.h" | |
| 24 | + | |
| 25 | +enum MatrixMode { | |
| 26 | + MATRIX_SQUARE, | |
| 27 | + MATRIX_ROW, | |
| 28 | + MATRIX_COLUMN, | |
| 29 | + MATRIX_NBMODES, | |
| 30 | +}; | |
| 31 | + | |
| 32 | +typedef struct ConvolutionContext { | |
| 33 | + const AVClass *class; | |
| 34 | + | |
| 35 | + char *matrix_str[4]; | |
| 36 | + float rdiv[4]; | |
| 37 | + float bias[4]; | |
| 38 | + int mode[4]; | |
| 39 | + float scale; | |
| 40 | + float delta; | |
| 41 | + int planes; | |
| 42 | + | |
| 43 | + int size[4]; | |
| 44 | + int depth; | |
| 45 | + int max; | |
| 46 | + int bpc; | |
| 47 | + int nb_planes; | |
| 48 | + int nb_threads; | |
| 49 | + int planewidth[4]; | |
| 50 | + int planeheight[4]; | |
| 51 | + int matrix[4][49]; | |
| 52 | + int matrix_length[4]; | |
| 53 | + int copy[4]; | |
| 54 | + | |
| 55 | + void (*setup[4])(int radius, const uint8_t *c[], const uint8_t *src, int stride, | |
| 56 | + int x, int width, int y, int height, int bpc); | |
| 57 | + void (*filter[4])(uint8_t *dst, int width, | |
| 58 | + float rdiv, float bias, const int *const matrix, | |
| 59 | + const uint8_t *c[], int peak, int radius, | |
| 60 | + int dstride, int stride); | |
| 61 | +} ConvolutionContext; | |
| 62 | + | |
| 63 | +void ff_convolution_init_x86(ConvolutionContext *s); | |
| 64 | +#endif | |
| @@ -25,48 +25,11 @@ · 24 unmodified lines | ||
| 25 | 25 | #include "libavutil/opt.h" |
| 26 | 26 | #include "libavutil/pixdesc.h" |
| 27 | 27 | #include "avfilter.h" |
| 28 | +#include "convolution.h" | |
| 28 | 29 | #include "formats.h" |
| 29 | 30 | #include "internal.h" |
| 30 | 31 | #include "video.h" |
| 31 | 32 | |
| 32 | −enum MatrixMode { | |
| 33 | − MATRIX_SQUARE, | |
| 34 | − MATRIX_ROW, | |
| 35 | − MATRIX_COLUMN, | |
| 36 | − MATRIX_NBMODES, | |
| 37 | −}; | |
| 38 | − | |
| 39 | −typedef struct ConvolutionContext { | |
| 40 | − const AVClass *class; | |
| 41 | − | |
| 42 | − char *matrix_str[4]; | |
| 43 | − float rdiv[4]; | |
| 44 | − float bias[4]; | |
| 45 | − int mode[4]; | |
| 46 | − float scale; | |
| 47 | − float delta; | |
| 48 | − int planes; | |
| 49 | − | |
| 50 | − int size[4]; | |
| 51 | − int depth; | |
| 52 | − int max; | |
| 53 | − int bpc; | |
| 54 | − int nb_planes; | |
| 55 | − int nb_threads; | |
| 56 | − int planewidth[4]; | |
| 57 | − int planeheight[4]; | |
| 58 | − int matrix[4][49]; | |
| 59 | − int matrix_length[4]; | |
| 60 | − int copy[4]; | |
| 61 | − | |
| 62 | − void (*setup[4])(int radius, const uint8_t *c[], const uint8_t *src, int stride, | |
| 63 | − int x, int width, int y, int height, int bpc); | |
| 64 | − void (*filter[4])(uint8_t *dst, int width, | |
| 65 | − float rdiv, float bias, const int *const matrix, | |
| 66 | − const uint8_t *c[], int peak, int radius, | |
| 67 | − int dstride, int stride); | |
| 68 | −} ConvolutionContext; | |
| 69 | − | |
| 70 | 33 | #define OFFSET(x) offsetof(ConvolutionContext, x) |
| 71 | 34 | #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM |
| 72 | 35 | |
| @@ -625,6 +588,8 @@ static int config_input(AVFilterLink *inlink) · 552 unmodified lines | ||
| 625 | 588 | s->filter[p] = filter16_7x7; |
| 626 | 589 | } |
| 627 | 590 | } |
| 591 | + if (ARCH_X86_64) | |
| 592 | + ff_convolution_init_x86(s); | |
| 628 | 593 | } else if (!strcmp(ctx->filter->name, "prewitt")) { |
| 629 | 594 | if (s->depth > 8) |
| 630 | 595 | for (p = 0; p < s->nb_planes; p++) |
| @@ -5,6 +5,7 @@ OBJS-$(CONFIG_ANLMDN_FILTER) += x86/af_anlmdn_init.o · 4 unmodified lines | ||
| 5 | 5 | OBJS-$(CONFIG_BLEND_FILTER) += x86/vf_blend_init.o |
| 6 | 6 | OBJS-$(CONFIG_BWDIF_FILTER) += x86/vf_bwdif_init.o |
| 7 | 7 | OBJS-$(CONFIG_COLORSPACE_FILTER) += x86/colorspacedsp_init.o |
| 8 | +OBJS-$(CONFIG_CONVOLUTION_FILTER) += x86/vf_convolution_init.o | |
| 8 | 9 | OBJS-$(CONFIG_EQ_FILTER) += x86/vf_eq.o |
| 9 | 10 | OBJS-$(CONFIG_FSPP_FILTER) += x86/vf_fspp_init.o |
| 10 | 11 | OBJS-$(CONFIG_GBLUR_FILTER) += x86/vf_gblur_init.o |
| @@ -40,6 +41,7 @@ X86ASM-OBJS-$(CONFIG_ANLMDN_FILTER) += x86/af_anlmdn.o · 29 unmodified lines | ||
| 40 | 41 | X86ASM-OBJS-$(CONFIG_BLEND_FILTER) += x86/vf_blend.o |
| 41 | 42 | X86ASM-OBJS-$(CONFIG_BWDIF_FILTER) += x86/vf_bwdif.o |
| 42 | 43 | X86ASM-OBJS-$(CONFIG_COLORSPACE_FILTER) += x86/colorspacedsp.o |
| 44 | +X86ASM-OBJS-$(CONFIG_CONVOLUTION_FILTER) += x86/vf_convolution.o | |
| 43 | 45 | X86ASM-OBJS-$(CONFIG_FRAMERATE_FILTER) += x86/vf_framerate.o |
| 44 | 46 | X86ASM-OBJS-$(CONFIG_FSPP_FILTER) += x86/vf_fspp.o |
| 45 | 47 | X86ASM-OBJS-$(CONFIG_GBLUR_FILTER) += x86/vf_gblur.o |
| @@ -0,0 +1,156 @@ | ||
| 1 | +;***************************************************************************** | |
| 2 | +;* x86-optimized functions for convolution filter | |
| 3 | +;* | |
| 4 | +;* This file is part of FFmpeg. | |
| 5 | +;* | |
| 6 | +;* FFmpeg is free software; you can redistribute it and/or | |
| 7 | +;* modify it under the terms of the GNU Lesser General Public | |
| 8 | +;* License as published by the Free Software Foundation; either | |
| 9 | +;* version 2.1 of the License, or (at your option) any later version. | |
| 10 | +;* | |
| 11 | +;* FFmpeg is distributed in the hope that it will be useful, | |
| 12 | +;* but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 13 | +;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
| 14 | +;* Lesser General Public License for more details. | |
| 15 | +;* | |
| 16 | +;* You should have received a copy of the GNU Lesser General Public | |
| 17 | +;* License along with FFmpeg; if not, write to the Free Software | |
| 18 | +;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
| 19 | +;****************************************************************************** | |
| 20 | + | |
| 21 | +%include "libavutil/x86/x86util.asm" | |
| 22 | + | |
| 23 | +SECTION_RODATA | |
| 24 | +half: dd 0.5 | |
| 25 | + | |
| 26 | +SECTION .text | |
| 27 | + | |
| 28 | +; void filter_3x3_sse4(uint8_t *dst, int width, | |
| 29 | +; float rdiv, float bias, const int *const matrix, | |
| 30 | +; const uint8_t *c[], int peak, int radius, | |
| 31 | +; int dstride, int stride) | |
| 32 | + | |
| 33 | + | |
| 34 | +%macro PROCESS_V 1 | |
| 35 | + movss m2, [matrixq + 4 * %1] | |
| 36 | + VBROADCASTSS m2, m2 | |
| 37 | + movss m3, [c%1q + xq] | |
| 38 | + punpcklbw m3, m6 | |
| 39 | + punpcklwd m3, m6 | |
| 40 | + pmulld m2, m3 | |
| 41 | + paddd m4, m2 | |
| 42 | +%endmacro | |
| 43 | + | |
| 44 | +%macro PROCESS_S 1 | |
| 45 | + movzx ptrd, byte [c%1q + xq] | |
| 46 | + imul ptrd, [matrixq + 4 * %1] | |
| 47 | + add rd, ptrd | |
| 48 | +%endmacro | |
| 49 | + | |
| 50 | +%macro FILTER_3X3 0 | |
| 51 | +%if UNIX64 | |
| 52 | +cglobal filter_3x3, 4, 15, 7, dst, width, matrix, ptr, c0, c1, c2, c3, c4, c5, c6, c7, c8, r, x | |
| 53 | +%else | |
| 54 | +cglobal filter_3x3, 4, 15, 7, dst, width, rdiv, bias, matrix, ptr, c0, c1, c2, c3, c4, c5, c6, c7, c8, r, x | |
| 55 | +%endif | |
| 56 | + | |
| 57 | +%if WIN64 | |
| 58 | + SWAP m0, m2 | |
| 59 | + SWAP m1, m3 | |
| 60 | + mov r2q, matrixmp | |
| 61 | + mov r3q, ptrmp | |
| 62 | + DEFINE_ARGS dst, width, matrix, ptr, c0, c1, c2, c3, c4, c5, c6, c7, c8, r, x | |
| 63 | +%endif | |
| 64 | + movsxdifnidn widthq, widthd | |
| 65 | + VBROADCASTSS m0, m0 | |
| 66 | + VBROADCASTSS m1, m1 | |
| 67 | + pxor m6, m6 | |
| 68 | + movss m5, [half] | |
| 69 | + VBROADCASTSS m5, m5 | |
| 70 | + mov c0q, [ptrq + 0*gprsize] | |
| 71 | + mov c1q, [ptrq + 1*gprsize] | |
| 72 | + mov c2q, [ptrq + 2*gprsize] | |
| 73 | + mov c3q, [ptrq + 3*gprsize] | |
| 74 | + mov c4q, [ptrq + 4*gprsize] | |
| 75 | + mov c5q, [ptrq + 5*gprsize] | |
| 76 | + mov c6q, [ptrq + 6*gprsize] | |
| 77 | + mov c7q, [ptrq + 7*gprsize] | |
| 78 | + mov c8q, [ptrq + 8*gprsize] | |
| 79 | + | |
| 80 | + xor xq, xq | |
| 81 | + cmp widthq, mmsize/4 | |
| 82 | + jl .loop2 | |
| 83 | + | |
| 84 | + mov rq, widthq | |
| 85 | + and rq, mmsize/4-1 | |
| 86 | + sub widthq, rq | |
| 87 | + | |
| 88 | +.loop1: | |
| 89 | + pxor m4, m4 ; sum = 0; | |
| 90 | + | |
| 91 | + PROCESS_V 0 | |
| 92 | + PROCESS_V 1 | |
| 93 | + PROCESS_V 2 | |
| 94 | + PROCESS_V 3 | |
| 95 | + PROCESS_V 4 | |
| 96 | + PROCESS_V 5 | |
| 97 | + PROCESS_V 6 | |
| 98 | + PROCESS_V 7 | |
| 99 | + PROCESS_V 8 | |
| 100 | + | |
| 101 | + cvtdq2ps m4, m4 | |
| 102 | + mulps m4, m0 ; sum *= rdiv | |
| 103 | + addps m4, m1 ; sum += bias | |
| 104 | + addps m4, m5 ; sum += 0.5 | |
| 105 | + cvttps2dq m4, m4 | |
| 106 | + packssdw m4, m4 | |
| 107 | + packuswb m4, m4 | |
| 108 | + movss [dstq + xq], m4 | |
| 109 | + | |
| 110 | + add xq, mmsize/4 | |
| 111 | + cmp xq, widthq | |
| 112 | + jl .loop1 | |
| 113 | + | |
| 114 | + add widthq, rq | |
| 115 | + cmp xq, widthq | |
| 116 | + jge .end | |
| 117 | + | |
| 118 | +.loop2: | |
| 119 | + ; reuse r to hold sum, init with zero | |
| 120 | + xor rd, rd | |
| 121 | + | |
| 122 | + PROCESS_S 0 | |
| 123 | + PROCESS_S 1 | |
| 124 | + PROCESS_S 2 | |
| 125 | + PROCESS_S 3 | |
| 126 | + PROCESS_S 4 | |
| 127 | + PROCESS_S 5 | |
| 128 | + PROCESS_S 6 | |
| 129 | + PROCESS_S 7 | |
| 130 | + PROCESS_S 8 | |
| 131 | + | |
| 132 | + pxor m4, m4 | |
| 133 | + cvtsi2ss m4, rd | |
| 134 | + mulss m4, m0 ; sum *= rdiv | |
| 135 | + addss m4, m1 ; sum += bias | |
| 136 | + addss m4, m5 ; sum += 0.5 | |
| 137 | + ; we don't have simple scalar instructions to convert | |
| 138 | + ; from 32bit to 8bit with saturation, so here | |
| 139 | + ; just use packed version SSE instructions for simplicity. | |
| 140 | + cvttps2dq m4, m4 ; trunc to integer | |
| 141 | + packssdw m4, m4 | |
| 142 | + packuswb m4, m4 | |
| 143 | + movd rd, m4 | |
| 144 | + mov [dstq + xq], rb | |
| 145 | + | |
| 146 | + add xq, 1 | |
| 147 | + cmp xq, widthq | |
| 148 | + jl .loop2 | |
| 149 | +.end: | |
| 150 | + RET | |
| 151 | +%endmacro | |
| 152 | + | |
| 153 | +%if ARCH_X86_64 | |
| 154 | +INIT_XMM sse4 | |
| 155 | +FILTER_3X3 | |
| 156 | +%endif | |
| @@ -0,0 +1,46 @@ | ||
| 1 | +/* | |
| 2 | + * | |
| 3 | + * This file is part of FFmpeg. | |
| 4 | + * | |
| 5 | + * FFmpeg is free software; you can redistribute it and/or | |
| 6 | + * modify it under the terms of the GNU Lesser General Public | |
| 7 | + * License as published by the Free Software Foundation; either | |
| 8 | + * version 2.1 of the License, or (at your option) any later version. | |
| 9 | + * | |
| 10 | + * FFmpeg is distributed in the hope that it will be useful, | |
| 11 | + * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 12 | + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
| 13 | + * Lesser General Public License for more details. | |
| 14 | + * | |
| 15 | + * You should have received a copy of the GNU Lesser General Public | |
| 16 | + * License along with FFmpeg; if not, write to the Free Software | |
| 17 | + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
| 18 | + */ | |
| 19 | + | |
| 20 | +#include "config.h" | |
| 21 | + | |
| 22 | +#include "libavutil/attributes.h" | |
| 23 | +#include "libavutil/cpu.h" | |
| 24 | +#include "libavutil/x86/cpu.h" | |
| 25 | +#include "libavfilter/convolution.h" | |
| 26 | + | |
| 27 | +void ff_filter_3x3_sse4(uint8_t *dst, int width, | |
| 28 | + float rdiv, float bias, const int *const matrix, | |
| 29 | + const uint8_t *c[], int peak, int radius, | |
| 30 | + int dstride, int stride); | |
| 31 | + | |
| 32 | +av_cold void ff_convolution_init_x86(ConvolutionContext *s) | |
| 33 | +{ | |
| 34 | +#if ARCH_X86_64 | |
| 35 | + int i; | |
| 36 | + int cpu_flags = av_get_cpu_flags(); | |
| 37 | + for (i = 0; i < 4; i++) { | |
| 38 | + if (s->mode[i] == MATRIX_SQUARE) { | |
| 39 | + if (s->matrix_length[i] == 9) { | |
| 40 | + if (EXTERNAL_SSE4(cpu_flags)) | |
| 41 | + s->filter[i] = ff_filter_3x3_sse4; | |
| 42 | + } | |
| 43 | + } | |
| 44 | + } | |
| 45 | +#endif | |
| 46 | +} | |