From 160556c9ad1662bf175f3c62c581834dadd3b52a Mon Sep 17 00:00:00 2001 From: Paul B Mahol Date: Wed, 30 Sep 2015 23:00:14 +0200 Subject: [PATCH] avfilter/vf_maskedmerge: add SIMD for maskedmerge with 8 bit depth input Signed-off-by: Paul B Mahol --- libavfilter/maskedmerge.h | 40 +++++++++++++++ libavfilter/vf_maskedmerge.c | 59 +++++++++------------- libavfilter/x86/Makefile | 2 + libavfilter/x86/vf_maskedmerge.asm | 73 +++++++++++++++++++++++++++ libavfilter/x86/vf_maskedmerge_init.c | 40 +++++++++++++++ 5 files changed, 179 insertions(+), 35 deletions(-) create mode 100644 libavfilter/maskedmerge.h create mode 100644 libavfilter/x86/vf_maskedmerge.asm create mode 100644 libavfilter/x86/vf_maskedmerge_init.c diff --git a/libavfilter/maskedmerge.h b/libavfilter/maskedmerge.h new file mode 100644 index 0000000000..b47a816c44 --- /dev/null +++ b/libavfilter/maskedmerge.h @@ -0,0 +1,40 @@ +/* + * 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 "avfilter.h" +#include "framesync.h" + +typedef struct MaskedMergeContext { + const AVClass *class; + int width[4], height[4]; + int nb_planes; + int planes; + int half, depth; + FFFrameSync fs; + + void (*maskedmerge)(const uint8_t *bsrc, const uint8_t *osrc, + const uint8_t *msrc, uint8_t *dst, + ptrdiff_t blinesize, ptrdiff_t olinesize, + ptrdiff_t mlinesize, ptrdiff_t dlinesize, + int w, int h, + int half, int shift); +} MaskedMergeContext; + +void ff_maskedmerge_init_x86(MaskedMergeContext *s); diff --git a/libavfilter/vf_maskedmerge.c b/libavfilter/vf_maskedmerge.c index 41bd039e6c..aa11d31720 100644 --- a/libavfilter/vf_maskedmerge.c +++ b/libavfilter/vf_maskedmerge.c @@ -23,24 +23,9 @@ #include "libavutil/opt.h" #include "avfilter.h" #include "formats.h" -#include "framesync.h" #include "internal.h" #include "video.h" - -typedef struct MaskedMergeContext { - const AVClass *class; - int width[4], height[4]; - int nb_planes; - int planes; - int max, half, depth; - FFFrameSync fs; - - void (*maskedmerge)(const uint8_t *bsrc, int blinesize, - const uint8_t *osrc, int olinesize, - const uint8_t *msrc, int mlinesize, - uint8_t *dst, int dlinesize, int w, int h, - int max, int half, int shift); -} MaskedMergeContext; +#include "maskedmerge.h" #define OFFSET(x) offsetof(MaskedMergeContext, x) #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM @@ -109,12 +94,12 @@ static int process_frame(FFFrameSync *fs) continue; } - s->maskedmerge(base->data[p], base->linesize[p], - overlay->data[p], overlay->linesize[p], - mask->data[p], mask->linesize[p], - out->data[p], out->linesize[p], + s->maskedmerge(base->data[p], overlay->data[p], + mask->data[p], out->data[p], + base->linesize[p], overlay->linesize[p], + mask->linesize[p], out->linesize[p], s->width[p], s->height[p], - s->max, s->half, s->depth); + s->half, s->depth); } } out->pts = av_rescale_q(base->pts, s->fs.time_base, outlink->time_base); @@ -122,17 +107,18 @@ static int process_frame(FFFrameSync *fs) return ff_filter_frame(outlink, out); } -static void maskedmerge8(const uint8_t *bsrc, int blinesize, - const uint8_t *osrc, int olinesize, - const uint8_t *msrc, int mlinesize, - uint8_t *dst, int dlinesize, int w, int h, - int max, int half, int shift) +static void maskedmerge8(const uint8_t *bsrc, const uint8_t *osrc, + const uint8_t *msrc, uint8_t *dst, + ptrdiff_t blinesize, ptrdiff_t olinesize, + ptrdiff_t mlinesize, ptrdiff_t dlinesize, + int w, int h, + int half, int shift) { int x, y; for (y = 0; y < h; y++) { for (x = 0; x < w; x++) { - dst[x] = ((256 - msrc[x]) * bsrc[x] + msrc[x] * osrc[x] + 128) >> 8; + dst[x] = bsrc[x] + ((msrc[x] * (osrc[x] - bsrc[x]) + 128) >> 8); } dst += dlinesize; @@ -142,11 +128,12 @@ static void maskedmerge8(const uint8_t *bsrc, int blinesize, } } -static void maskedmerge16(const uint8_t *bbsrc, int blinesize, - const uint8_t *oosrc, int olinesize, - const uint8_t *mmsrc, int mlinesize, - uint8_t *ddst, int dlinesize, int w, int h, - int max, int half, int shift) +static void maskedmerge16(const uint8_t *bbsrc, const uint8_t *oosrc, + const uint8_t *mmsrc, uint8_t *ddst, + ptrdiff_t blinesize, ptrdiff_t olinesize, + ptrdiff_t mlinesize, ptrdiff_t dlinesize, + int w, int h, + int half, int shift) { const uint16_t *bsrc = (const uint16_t *)bbsrc; const uint16_t *osrc = (const uint16_t *)oosrc; @@ -156,7 +143,7 @@ static void maskedmerge16(const uint8_t *bbsrc, int blinesize, for (y = 0; y < h; y++) { for (x = 0; x < w; x++) { - dst[x] = ((max - msrc[x]) * bsrc[x] + msrc[x] * osrc[x] + half) >> shift; + dst[x] = bsrc[x] + ((msrc[x] * (osrc[x] - bsrc[x]) + half) >> shift); } dst += dlinesize / 2; @@ -183,14 +170,16 @@ static int config_input(AVFilterLink *inlink) s->width[0] = s->width[3] = inlink->w; s->depth = desc->comp[0].depth; - s->max = 1 << s->depth; - s->half = s->max / 2; + s->half = (1 << s->depth) / 2; if (desc->comp[0].depth == 8) s->maskedmerge = maskedmerge8; else s->maskedmerge = maskedmerge16; + if (ARCH_X86) + ff_maskedmerge_init_x86(s); + return 0; } diff --git a/libavfilter/x86/Makefile b/libavfilter/x86/Makefile index 5382027f70..db072cd433 100644 --- a/libavfilter/x86/Makefile +++ b/libavfilter/x86/Makefile @@ -4,6 +4,7 @@ OBJS-$(CONFIG_GRADFUN_FILTER) += x86/vf_gradfun_init.o OBJS-$(CONFIG_HQDN3D_FILTER) += x86/vf_hqdn3d_init.o OBJS-$(CONFIG_IDET_FILTER) += x86/vf_idet_init.o OBJS-$(CONFIG_INTERLACE_FILTER) += x86/vf_interlace_init.o +OBJS-$(CONFIG_MASKEDMERGE_FILTER) += x86/vf_maskedmerge_init.o OBJS-$(CONFIG_NOISE_FILTER) += x86/vf_noise.o OBJS-$(CONFIG_PP7_FILTER) += x86/vf_pp7_init.o OBJS-$(CONFIG_PSNR_FILTER) += x86/vf_psnr_init.o @@ -20,6 +21,7 @@ YASM-OBJS-$(CONFIG_GRADFUN_FILTER) += x86/vf_gradfun.o YASM-OBJS-$(CONFIG_HQDN3D_FILTER) += x86/vf_hqdn3d.o YASM-OBJS-$(CONFIG_IDET_FILTER) += x86/vf_idet.o YASM-OBJS-$(CONFIG_INTERLACE_FILTER) += x86/vf_interlace.o +YASM-OBJS-$(CONFIG_MASKEDMERGE_FILTER) += x86/vf_maskedmerge.o YASM-OBJS-$(CONFIG_PP7_FILTER) += x86/vf_pp7.o YASM-OBJS-$(CONFIG_PSNR_FILTER) += x86/vf_psnr.o YASM-OBJS-$(CONFIG_PULLUP_FILTER) += x86/vf_pullup.o diff --git a/libavfilter/x86/vf_maskedmerge.asm b/libavfilter/x86/vf_maskedmerge.asm new file mode 100644 index 0000000000..3b128b4f52 --- /dev/null +++ b/libavfilter/x86/vf_maskedmerge.asm @@ -0,0 +1,73 @@ +;***************************************************************************** +;* x86-optimized functions for maskedmerge filter +;* +;* 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" + +%if ARCH_X86_64 +SECTION_RODATA + +pw_128: times 8 dw 128 +pw_256: times 8 dw 256 + +SECTION .text + +INIT_XMM sse2 +cglobal maskedmerge8, 10, 11, 7, 0, bsrc, osrc, msrc, dst, blinesize, olinesize, mlinesize, dlinesize, w, h + mova m4, [pw_256] + mova m5, [pw_128] + pxor m6, m6 + add bsrcq, wq + add osrcq, wq + add msrcq, wq + add dstq, wq + neg wq +.nextrow: + mov r10q, wq + %define x r10q + + .loop: + movh m0, [bsrcq + x] + movh m1, [osrcq + x] + movh m3, [msrcq + x] + mova m2, m4 + punpcklbw m0, m6 + punpcklbw m1, m6 + punpcklbw m3, m6 + psubw m2, m3 + pmullw m2, m0 + pmullw m1, m3 + paddw m1, m2 + paddw m1, m5 + psrlw m1, 8 + packuswb m1, m1 + movh [dstq + x], m1 + add r10q, mmsize / 2 + jl .loop + + add bsrcq, blinesizeq + add osrcq, olinesizeq + add msrcq, mlinesizeq + add dstq, dlinesizeq + sub hd, 1 + jg .nextrow +REP_RET +%endif diff --git a/libavfilter/x86/vf_maskedmerge_init.c b/libavfilter/x86/vf_maskedmerge_init.c new file mode 100644 index 0000000000..443570047b --- /dev/null +++ b/libavfilter/x86/vf_maskedmerge_init.c @@ -0,0 +1,40 @@ +/* + * 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/attributes.h" +#include "libavutil/cpu.h" +#include "libavutil/x86/cpu.h" +#include "libavfilter/maskedmerge.h" + +void ff_maskedmerge8_sse2(const uint8_t *bsrc, const uint8_t *osrc, + const uint8_t *msrc, uint8_t *dst, + ptrdiff_t blinesize, ptrdiff_t olinesize, + ptrdiff_t mlinesize, ptrdiff_t dlinesize, + int w, int h, + int half, int shift); + +av_cold void ff_maskedmerge_init_x86(MaskedMergeContext *s) +{ + int cpu_flags = av_get_cpu_flags(); + + if (ARCH_X86_64 && EXTERNAL_SSE2(cpu_flags) && s->depth == 8) { + s->maskedmerge = ff_maskedmerge8_sse2; + } +}