From patchwork Tue Oct 8 13:45:40 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Jun Zhao X-Patchwork-Id: 15575 Return-Path: X-Original-To: patchwork@ffaux-bg.ffmpeg.org Delivered-To: patchwork@ffaux-bg.ffmpeg.org Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org [79.124.17.100]) by ffaux.localdomain (Postfix) with ESMTP id D9655447282 for ; Tue, 8 Oct 2019 17:12:45 +0300 (EEST) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id B21E8687F36; Tue, 8 Oct 2019 17:12:45 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mail-pg1-f194.google.com (mail-pg1-f194.google.com [209.85.215.194]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 5A1FF68018B for ; Tue, 8 Oct 2019 17:12:38 +0300 (EEST) Received: by mail-pg1-f194.google.com with SMTP id e1so10310157pgj.6 for ; Tue, 08 Oct 2019 07:12:38 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=from:to:cc:subject:date:message-id; bh=MR8Ni8ImsgxUY8wc69VU70RjOOORDPoCECH0fcze3LA=; b=G26dRqUZ10uO76Snkg6SmRmHoXD0hVnraDQxOGejSCDTFDIzFnKCYQV7L8cHb6q28T f75fVxrmY/MtfKzDP4iqsg+GYZCHg/F0yWpzKpRFFIf/a2sVqBt8I5xoDIpqrU+BCPUJ fJ1ep4cnVPYzwONk4TOO7t7jjop6hZRImuBLXVS0z7BjDDV9n1gryhkwhMJNhz8hFBfT TE1gA+uN7NDeR08phd+Jrn9g3SaNqBAcs9s4vdBQErpEEYfi556JOEB6wllU/gxz2cM7 q5l+Rn0ikZ2l3iWNHylK1TFRkB4vJBkS+io3ZXIZM3LJzu2OAojN2ifsi0eyB2yXyyIZ QugQ== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id; bh=MR8Ni8ImsgxUY8wc69VU70RjOOORDPoCECH0fcze3LA=; b=XpM7RuEs5mo/lI81BjHujRvMZgXiqrJ3dmD+CXZarrs6FkMFVIimQPQCd/rIyNMdZ6 IBGDRQjGryQHOycFuvGZBWrZFswiQ6IDGyIFMpfVZWFW9zYbGRqvlaqduvtCQI//N9HM YFCIWaCRqDFgs3JqBG5A5tad8mPhm6nZOlgnqTpbitn5jfk9dgmG7yVV+7Cb5x0p7eKY jrIO4ZusZhRrAfA8Y+gAmV4eeCd1mtHP83fN1ZSKi5c3rhCANRV9ya1mZUOwYmW8+2Lw TsG2NMu75o2XRUPhHU7ZB5eEc0iSmvPPUo+l76zEf6CFVR1jdbFsfVKNB3KYlc2K4iTM LHgA== X-Gm-Message-State: APjAAAU+eezVv8wuSl3q+WhnxVjAwH7EeDNFOcX9OwPs5McLBOau9wZa 0huzBlMOZFJo5Ajjx/WNo7rA6NVZ X-Google-Smtp-Source: APXvYqycg+x9Dul30LJqLH72v+jyrEM6n/9tf5lhSsQIKUniz4b2jMgqmpYbYCwT6yxm+A62s+JH1Q== X-Received: by 2002:a62:75d2:: with SMTP id q201mr38944873pfc.43.1570542346067; Tue, 08 Oct 2019 06:45:46 -0700 (PDT) Received: from localhost.localdomain ([47.90.47.25]) by smtp.gmail.com with ESMTPSA id l72sm2450998pjb.7.2019.10.08.06.45.44 (version=TLS1_2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128); Tue, 08 Oct 2019 06:45:45 -0700 (PDT) From: Jun Zhao To: ffmpeg-devel@ffmpeg.org Date: Tue, 8 Oct 2019 21:45:40 +0800 Message-Id: <1570542340-9853-1-git-send-email-mypopydev@gmail.com> X-Mailer: git-send-email 1.7.1 Subject: [FFmpeg-devel] [PATCH V2] lavfi/hqdn3d: add slice thread optimization X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Jun Zhao MIME-Version: 1.0 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" From: Jun Zhao Enabled one thread per planar, used the test command for 1080P video (YUV420P format) as follow: ffmpeg -i 1080p.mp4 -an -vf hqdn3d -f null /dev/nul This optimization improved the performance about 30% in 1080P YUV420P case (from 110fps to 143fps), also pass the framemd5 check and FATE. Signed-off-by: Jun Zhao --- libavfilter/vf_hqdn3d.c | 56 +++++++++++++++++++++++++++++++++------------- libavfilter/vf_hqdn3d.h | 2 +- 2 files changed, 41 insertions(+), 17 deletions(-) diff --git a/libavfilter/vf_hqdn3d.c b/libavfilter/vf_hqdn3d.c index d6c14bb..6dd8fd8 100644 --- a/libavfilter/vf_hqdn3d.c +++ b/libavfilter/vf_hqdn3d.c @@ -223,7 +223,9 @@ static av_cold void uninit(AVFilterContext *ctx) av_freep(&s->coefs[1]); av_freep(&s->coefs[2]); av_freep(&s->coefs[3]); - av_freep(&s->line); + av_freep(&s->line[0]); + av_freep(&s->line[1]); + av_freep(&s->line[2]); av_freep(&s->frame_prev[0]); av_freep(&s->frame_prev[1]); av_freep(&s->frame_prev[2]); @@ -271,9 +273,11 @@ static int config_input(AVFilterLink *inlink) s->vsub = desc->log2_chroma_h; s->depth = desc->comp[0].depth; - s->line = av_malloc_array(inlink->w, sizeof(*s->line)); - if (!s->line) - return AVERROR(ENOMEM); + for (i = 0; i < 3; i++) { + s->line[i] = av_malloc_array(inlink->w, sizeof(*s->line[i])); + if (!s->line[i]) + return AVERROR(ENOMEM); + } for (i = 0; i < 4; i++) { s->coefs[i] = precalc_coefs(s->strength[i], s->depth); @@ -287,14 +291,38 @@ static int config_input(AVFilterLink *inlink) return 0; } +struct ThreadData { + AVFrame *in, *out; + int direct; +}; + +static int do_denoise(AVFilterContext *ctx, void *data, int job_nr, int n_jobs) +{ + HQDN3DContext *s = ctx->priv; + struct ThreadData *td = data; + AVFrame *out = td->out; + AVFrame *in = td->in; + int direct = td->direct; + + denoise(s, in->data[job_nr], out->data[job_nr], + s->line[job_nr], &s->frame_prev[job_nr], + AV_CEIL_RSHIFT(in->width, (!!job_nr * s->hsub)), + AV_CEIL_RSHIFT(in->height, (!!job_nr * s->vsub)), + in->linesize[job_nr], out->linesize[job_nr], + s->coefs[job_nr ? CHROMA_SPATIAL : LUMA_SPATIAL], + s->coefs[job_nr ? CHROMA_TMP : LUMA_TMP]); + + return 0; +} + static int filter_frame(AVFilterLink *inlink, AVFrame *in) { AVFilterContext *ctx = inlink->dst; - HQDN3DContext *s = ctx->priv; AVFilterLink *outlink = ctx->outputs[0]; AVFrame *out; - int c, direct = av_frame_is_writable(in) && !ctx->is_disabled; + int direct = av_frame_is_writable(in) && !ctx->is_disabled; + struct ThreadData td; if (direct) { out = in; @@ -308,15 +336,11 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in) av_frame_copy_props(out, in); } - for (c = 0; c < 3; c++) { - denoise(s, in->data[c], out->data[c], - s->line, &s->frame_prev[c], - AV_CEIL_RSHIFT(in->width, (!!c * s->hsub)), - AV_CEIL_RSHIFT(in->height, (!!c * s->vsub)), - in->linesize[c], out->linesize[c], - s->coefs[c ? CHROMA_SPATIAL : LUMA_SPATIAL], - s->coefs[c ? CHROMA_TMP : LUMA_TMP]); - } + td.in = in; + td.out = out; + td.direct = direct; + /* one thread per planar */ + ctx->internal->execute(ctx, do_denoise, &td, NULL, 3); if (ctx->is_disabled) { av_frame_free(&out); @@ -370,5 +394,5 @@ AVFilter ff_vf_hqdn3d = { .query_formats = query_formats, .inputs = avfilter_vf_hqdn3d_inputs, .outputs = avfilter_vf_hqdn3d_outputs, - .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL, + .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS, }; diff --git a/libavfilter/vf_hqdn3d.h b/libavfilter/vf_hqdn3d.h index 03a79a1..3279bbc 100644 --- a/libavfilter/vf_hqdn3d.h +++ b/libavfilter/vf_hqdn3d.h @@ -31,7 +31,7 @@ typedef struct HQDN3DContext { const AVClass *class; int16_t *coefs[4]; - uint16_t *line; + uint16_t *line[3]; uint16_t *frame_prev[3]; double strength[4]; int hsub, vsub;