[FFmpeg-devel] [PATCH] avfilter/vf_convolution: add x86 SIMD for filter_3x3()

Song, Ruiling ruiling.song at intel.com
Mon Jul 15 04:48:33 EEST 2019


> -----Original Message-----
> From: Song, Ruiling
> Sent: Tuesday, July 9, 2019 9:15 AM
> To: ffmpeg-devel at ffmpeg.org
> Cc: Song, Ruiling <ruiling.song at intel.com>
> Subject: [PATCH] avfilter/vf_convolution: add x86 SIMD for filter_3x3()
> 
> 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.
> 
> Signed-off-by: Ruiling Song <ruiling.song at intel.com>
Ping?

> ---
>  libavfilter/convolution.h             |  64 +++++++++++
>  libavfilter/vf_convolution.c          |  41 +------
>  libavfilter/x86/Makefile              |   2 +
>  libavfilter/x86/vf_convolution.asm    | 158 ++++++++++++++++++++++++++
>  libavfilter/x86/vf_convolution_init.c |  46 ++++++++
>  5 files changed, 273 insertions(+), 38 deletions(-)
>  create mode 100644 libavfilter/convolution.h
>  create mode 100644 libavfilter/x86/vf_convolution.asm
>  create mode 100644 libavfilter/x86/vf_convolution_init.c
> 
> diff --git a/libavfilter/convolution.h b/libavfilter/convolution.h
> new file mode 100644
> index 0000000000..fc6aad58fd
> --- /dev/null
> +++ b/libavfilter/convolution.h
> @@ -0,0 +1,64 @@
> +/*
> + * Copyright (c) 2012-2013 Oka Motofumi (chikuzen.mo at gmail dot com)
> + * 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
> + */
> +#ifndef AVFILTER_CONVOLUTION_H
> +#define AVFILTER_CONVOLUTION_H
> +#include "avfilter.h"
> +
> +enum MatrixMode {
> +    MATRIX_SQUARE,
> +    MATRIX_ROW,
> +    MATRIX_COLUMN,
> +    MATRIX_NBMODES,
> +};
> +
> +typedef struct ConvolutionContext {
> +    const AVClass *class;
> +
> +    char *matrix_str[4];
> +    float rdiv[4];
> +    float bias[4];
> +    int mode[4];
> +    float scale;
> +    float delta;
> +    int planes;
> +
> +    int size[4];
> +    int depth;
> +    int max;
> +    int bpc;
> +    int nb_planes;
> +    int nb_threads;
> +    int planewidth[4];
> +    int planeheight[4];
> +    int matrix[4][49];
> +    int matrix_length[4];
> +    int copy[4];
> +
> +    void (*setup[4])(int radius, const uint8_t *c[], const uint8_t *src, int stride,
> +                     int x, int width, int y, int height, int bpc);
> +    void (*filter[4])(uint8_t *dst, int width,
> +                      float rdiv, float bias, const int *const matrix,
> +                      const uint8_t *c[], int peak, int radius,
> +                      int dstride, int stride);
> +} ConvolutionContext;
> +
> +void ff_convolution_init_x86(ConvolutionContext *s);
> +#endif
> diff --git a/libavfilter/vf_convolution.c b/libavfilter/vf_convolution.c
> index 1305569c88..e3bf1df79f 100644
> --- a/libavfilter/vf_convolution.c
> +++ b/libavfilter/vf_convolution.c
> @@ -25,48 +25,11 @@
>  #include "libavutil/opt.h"
>  #include "libavutil/pixdesc.h"
>  #include "avfilter.h"
> +#include "convolution.h"
>  #include "formats.h"
>  #include "internal.h"
>  #include "video.h"
> 
> -enum MatrixMode {
> -    MATRIX_SQUARE,
> -    MATRIX_ROW,
> -    MATRIX_COLUMN,
> -    MATRIX_NBMODES,
> -};
> -
> -typedef struct ConvolutionContext {
> -    const AVClass *class;
> -
> -    char *matrix_str[4];
> -    float rdiv[4];
> -    float bias[4];
> -    int mode[4];
> -    float scale;
> -    float delta;
> -    int planes;
> -
> -    int size[4];
> -    int depth;
> -    int max;
> -    int bpc;
> -    int nb_planes;
> -    int nb_threads;
> -    int planewidth[4];
> -    int planeheight[4];
> -    int matrix[4][49];
> -    int matrix_length[4];
> -    int copy[4];
> -
> -    void (*setup[4])(int radius, const uint8_t *c[], const uint8_t *src, int stride,
> -                     int x, int width, int y, int height, int bpc);
> -    void (*filter[4])(uint8_t *dst, int width,
> -                      float rdiv, float bias, const int *const matrix,
> -                      const uint8_t *c[], int peak, int radius,
> -                      int dstride, int stride);
> -} ConvolutionContext;
> -
>  #define OFFSET(x) offsetof(ConvolutionContext, x)
>  #define FLAGS
> AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
> 
> @@ -625,6 +588,8 @@ static int config_input(AVFilterLink *inlink)
>                      s->filter[p] = filter16_7x7;
>              }
>          }
> +        if (ARCH_X86_64)
> +            ff_convolution_init_x86(s);
>      } else if (!strcmp(ctx->filter->name, "prewitt")) {
>          if (s->depth > 8)
>              for (p = 0; p < s->nb_planes; p++)
> diff --git a/libavfilter/x86/Makefile b/libavfilter/x86/Makefile
> index 6b0361bed2..8dc0b0e6d4 100644
> --- a/libavfilter/x86/Makefile
> +++ b/libavfilter/x86/Makefile
> @@ -5,6 +5,7 @@ OBJS-$(CONFIG_ANLMDN_FILTER)                 +=
> x86/af_anlmdn_init.o
>  OBJS-$(CONFIG_BLEND_FILTER)                  += x86/vf_blend_init.o
>  OBJS-$(CONFIG_BWDIF_FILTER)                  += x86/vf_bwdif_init.o
>  OBJS-$(CONFIG_COLORSPACE_FILTER)             += x86/colorspacedsp_init.o
> +OBJS-$(CONFIG_CONVOLUTION_FILTER)            +=
> x86/vf_convolution_init.o
>  OBJS-$(CONFIG_EQ_FILTER)                     += x86/vf_eq.o
>  OBJS-$(CONFIG_FSPP_FILTER)                   += x86/vf_fspp_init.o
>  OBJS-$(CONFIG_GBLUR_FILTER)                  += x86/vf_gblur_init.o
> @@ -40,6 +41,7 @@ X86ASM-OBJS-$(CONFIG_ANLMDN_FILTER)          +=
> x86/af_anlmdn.o
>  X86ASM-OBJS-$(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-$(CONFIG_CONVOLUTION_FILTER)     +=
> x86/vf_convolution.o
>  X86ASM-OBJS-$(CONFIG_FRAMERATE_FILTER)       += x86/vf_framerate.o
>  X86ASM-OBJS-$(CONFIG_FSPP_FILTER)            += x86/vf_fspp.o
>  X86ASM-OBJS-$(CONFIG_GBLUR_FILTER)           += x86/vf_gblur.o
> diff --git a/libavfilter/x86/vf_convolution.asm
> b/libavfilter/x86/vf_convolution.asm
> new file mode 100644
> index 0000000000..41a17f6d2d
> --- /dev/null
> +++ b/libavfilter/x86/vf_convolution.asm
> @@ -0,0 +1,158 @@
> +;*********************************************************
> ********************
> +;* x86-optimized functions for convolution filter
> +;*
> +;* 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_RODATA
> +half:   dd 0.5
> +
> +SECTION .text
> +
> +; void filter_3x3_sse4(uint8_t *dst, int width,
> +;                      float rdiv, float bias, const int *const matrix,
> +;                      const uint8_t *c[], int peak, int radius,
> +;                      int dstride, int stride)
> +
> +
> +%macro PROCESS_V 1
> +    movss m2, [matrixq + 4 * %1]
> +    VBROADCASTSS m2, m2
> +    movss m3, [c%1q + xq]
> +    punpcklbw m3, m6
> +    punpcklwd m3, m6
> +    pmulld m2, m3
> +    paddd m4, m2
> +%endmacro
> +
> +%macro PROCESS_S 1
> +    mov   ptrd, [c%1q + xq]
> +    imul  ptrd, [matrixq + 4 * %1]
> +    add   rd, ptrd
> +%endmacro
> +
> +%macro FILTER_3X3 0
> +%if UNIX64
> +cglobal filter_3x3, 4, 15, 7, dst, width, matrix, ptr, c0, c1, c2, c3, c4, c5, c6, c7,
> c8, r, x
> +%else
> +cglobal filter_3x3, 4, 15, 7, dst, width, rdiv, bias, matrix, ptr, c0, c1, c2, c3, c4,
> c5, c6, c7, c8, r, x
> +%endif
> +
> +%if WIN64
> +    SWAP m0, m2
> +    SWAP m1, m3
> +    mov  r2q, matrixmp
> +    mov  r3q, ptrmp
> +    DEFINE_ARGS dst, width, matrix, ptr, c0, c1, c2, c3, c4, c5, c6, c7, c8, r, x
> +%endif
> +    movsxdifnidn widthq, widthd
> +    VBROADCASTSS m0, m0
> +    VBROADCASTSS m1, m1
> +    pxor  m6, m6
> +    movss m5, [half]
> +    VBROADCASTSS m5, m5
> +    mov   c0q, [ptrq + 0*gprsize]
> +    mov   c1q, [ptrq + 1*gprsize]
> +    mov   c2q, [ptrq + 2*gprsize]
> +    mov   c3q, [ptrq + 3*gprsize]
> +    mov   c4q, [ptrq + 4*gprsize]
> +    mov   c5q, [ptrq + 5*gprsize]
> +    mov   c6q, [ptrq + 6*gprsize]
> +    mov   c7q, [ptrq + 7*gprsize]
> +    mov   c8q, [ptrq + 8*gprsize]
> +
> +    xor   xq, xq
> +    mov   rq, widthq
> +    and   rq, mmsize/4-1
> +    sub   widthq, rq
> +
> +.loop1:
> +    movss m2, [matrixq + 4*0]
> +    VBROADCASTSS m2, m2
> +    movss m3, [c0q + xq]
> +    punpcklbw m3, m6
> +    punpcklwd m3, m6
> +    pmulld m4, m2, m3
> +
> +    PROCESS_V 1
> +    PROCESS_V 2
> +    PROCESS_V 3
> +    PROCESS_V 4
> +    PROCESS_V 5
> +    PROCESS_V 6
> +    PROCESS_V 7
> +    PROCESS_V 8
> +
> +    cvtdq2ps  m4, m4
> +    mulps     m4, m0     ; sum *= rdiv
> +    addps     m4, m1     ; sum += bias
> +    addps     m4, m5     ; sum += 0.5
> +    cvttps2dq m4, m4
> +    packssdw  m4, m4
> +    packuswb  m4, m4
> +    movss     [dstq + xq], m4
> +
> +    add xq, mmsize/4
> +    cmp xq, widthq
> +    jl .loop1
> +
> +    add widthq, rq
> +    cmp xq, widthq
> +    jge .end
> +
> +.loop2:
> +    mov   ptrd, [c0q + xq]
> +    imul  ptrd, [matrixq + 4*0]
> +    ; reuse r to hold sum
> +    mov   rd, ptrd
> +
> +    PROCESS_S 1
> +    PROCESS_S 2
> +    PROCESS_S 3
> +    PROCESS_S 4
> +    PROCESS_S 5
> +    PROCESS_S 6
> +    PROCESS_S 7
> +    PROCESS_S 8
> +
> +    pxor      m4, m4
> +    cvtsi2ss  m4, rd
> +    mulss     m4, m0     ; sum *= rdiv
> +    addss     m4, m1     ; sum += bias
> +    addss     m4, m5     ; sum += 0.5
> +    ; we don't have simple scalar instructions to convert
> +    ; from 32bit to 8bit with saturation, so here
> +    ; just use packed version SSE instructions for simplity.
> +    cvttps2dq m4, m4     ; trunc to integer
> +    packssdw  m4, m4
> +    packuswb  m4, m4
> +    movd      rd, m4
> +    mov       [dstq + xq], rb
> +
> +    add xq, 1
> +    cmp xq, widthq
> +    jl .loop2
> +.end:
> +    RET
> +%endmacro
> +
> +%if ARCH_X86_64
> +INIT_XMM sse4
> +FILTER_3X3
> +%endif
> diff --git a/libavfilter/x86/vf_convolution_init.c
> b/libavfilter/x86/vf_convolution_init.c
> new file mode 100644
> index 0000000000..51432406ed
> --- /dev/null
> +++ b/libavfilter/x86/vf_convolution_init.c
> @@ -0,0 +1,46 @@
> +/*
> + *
> + * 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 "config.h"
> +
> +#include "libavutil/attributes.h"
> +#include "libavutil/cpu.h"
> +#include "libavutil/x86/cpu.h"
> +#include "libavfilter/convolution.h"
> +
> +void ff_filter_3x3_sse4(uint8_t *dst, int width,
> +                        float rdiv, float bias, const int *const matrix,
> +                        const uint8_t *c[], int peak, int radius,
> +                        int dstride, int stride);
> +
> +av_cold void ff_convolution_init_x86(ConvolutionContext *s)
> +{
> +#if ARCH_X86_64
> +    int i;
> +    int cpu_flags = av_get_cpu_flags();
> +    for (i = 0; i < 4; i++) {
> +        if (s->mode[i] == MATRIX_SQUARE) {
> +            if (s->matrix_length[i] == 9) {
> +                if (EXTERNAL_SSE4(cpu_flags))
> +                    s->filter[i] = ff_filter_3x3_sse4;
> +            }
> +        }
> +    }
> +#endif
> +}
> --
> 2.17.1



More information about the ffmpeg-devel mailing list