From patchwork Mon Apr 8 20:13:43 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Andreas Rheinhardt X-Patchwork-Id: 47943 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:3f98:b0:1a7:a0dc:8de5 with SMTP id ay24csp135692pzb; Mon, 8 Apr 2024 13:15:09 -0700 (PDT) X-Forwarded-Encrypted: i=2; AJvYcCVZpTD5tXNF71geMWlaRdlM4r6J+63Vd340stpgiTgKSACtHUWJlmHkwzeJW40YRyPhvjFussGn2JeDk+N8hKfXnOTcQxCLD8QPCg== X-Google-Smtp-Source: AGHT+IFwZKtUCdfqheBQ44+fsZH7WVrS3+vbRETsMFSNAtkfjTEXQTk+KsXCugWkg1bFSaFbhVYF X-Received: by 2002:a17:907:a0c:b0:a51:d66a:6f8 with SMTP id bb12-20020a1709070a0c00b00a51d66a06f8mr2924705ejc.20.1712607309393; Mon, 08 Apr 2024 13:15:09 -0700 (PDT) Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id g10-20020a1709063b0a00b00a4e29be9c8bsi4058536ejf.31.2024.04.08.13.15.08; Mon, 08 Apr 2024 13:15:09 -0700 (PDT) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; dkim=neutral (body hash did not verify) header.i=@outlook.com header.s=selector1 header.b=oGFeS3xX; arc=fail (body hash mismatch); spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org; dmarc=fail (p=NONE sp=QUARANTINE dis=NONE) header.from=outlook.com Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 4DB6F68D275; Mon, 8 Apr 2024 23:15:03 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from EUR01-VE1-obe.outbound.protection.outlook.com (mail-ve1eur01olkn2073.outbound.protection.outlook.com [40.92.66.73]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 13D5A68D352 for ; Mon, 8 Apr 2024 23:15:00 +0300 (EEST) ARC-Seal: i=1; a=rsa-sha256; s=arcselector9901; d=microsoft.com; cv=none; b=Xvt0/celExI4XIKYqy0iyQq7ZIMG1wxoPTNmqsflF7/YpDP8QIzueJw0ICb8fsUSwiC7UV3eEjpm5M/RK/OpeF1QdfrxxhiKGglaa0X8SX778dsruW7JZFDbNjE/3ZAGU6t87jjqWsXMKEibB640++vA+p2CR151H+VlfUlXga7N+3+u69WVhKMWL21PjNCwiP7decDJwB4DyPhRRVElyjy0OcrHfaI/gK6xIO0Zt9OlC+foyKkxk2gn+zeWaLcVUAdClWJtIIytr+P1jQeo1hT94j8Zr8RO+Mku3KZVlAxUwMFZ3ThkXD2b5vLmMSu/xKHhw/3+Epq9KNP5rAwC5Q== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=microsoft.com; s=arcselector9901; h=From:Date:Subject:Message-ID:Content-Type:MIME-Version:X-MS-Exchange-AntiSpam-MessageData-ChunkCount:X-MS-Exchange-AntiSpam-MessageData-0:X-MS-Exchange-AntiSpam-MessageData-1; bh=YzKi1tCOA36bTbZ6Q+NZOeTXgyrDSU1LbsD1Q41CF2Q=; b=moxuonV+xuz2NO01wqXLhGqxV0ki3/BcOsgI752uGMReZMcsEWztEv84U63VftTwwC/6tnkymM6sVAK6I1j603phlVo/RDba6wUheyumIHwdNZUBzSzLenoXWLx33G3xeGuEaCYrm1UGUsfZDQA/ShBTQsKJ+6BPYClNycsjVRZYFbpioHzKqUwdIIOHVLsVhQa715aKxa86/MtDZU6Qtzwqvi13tJ19v4fh2EqcXIbC+fUYZT5GIpWcvLZ3hmaKU12Li59Hcz7NF6vglJXC97HngwUJjH1fZ7qcuHPGlvBHL15cNpTI7fDLK5xwNv5odbS8WzRAbvJR76vhxPWWNg== ARC-Authentication-Results: i=1; mx.microsoft.com 1; spf=none; dmarc=none; dkim=none; arc=none DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=outlook.com; s=selector1; h=From:Date:Subject:Message-ID:Content-Type:MIME-Version:X-MS-Exchange-SenderADCheck; bh=YzKi1tCOA36bTbZ6Q+NZOeTXgyrDSU1LbsD1Q41CF2Q=; b=oGFeS3xXfeyf4XozSp5Z/4Yji+KVHSfPi3flJVBuS4UX+bZKY5s8LEFii6zI29SVe686q3wo9AOak0Fl//quIIffc+bUDM6BshotnhwwVgUhQ+BaiKzkYYR5cVPNqFwALpfwmwq7+Mynhqv//ziUBC2Mig2S58EbNSePX3hrkb4yzTacJHbex7F4L7DqdKzBEYtenKKd1w56+JAbiNTXpgTFKcjZ8RCYkPu4ETa89llG1SGEbcOycsJnIMMyNuUDL9y48/FWRbzZo7LDK7LZ2qIqzi7vdQqqWm027YA32nxZeKN3n9srZakYS7cdHOnNTKen2DDHV33eftkQpxCT3g== Received: from GV1P250MB0737.EURP250.PROD.OUTLOOK.COM (2603:10a6:150:8e::17) by PAXP250MB0494.EURP250.PROD.OUTLOOK.COM (2603:10a6:102:28d::12) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.20.7409.46; Mon, 8 Apr 2024 20:14:55 +0000 Received: from GV1P250MB0737.EURP250.PROD.OUTLOOK.COM ([fe80::4a3b:cf76:cddc:c68d]) by GV1P250MB0737.EURP250.PROD.OUTLOOK.COM ([fe80::4a3b:cf76:cddc:c68d%4]) with mapi id 15.20.7409.042; Mon, 8 Apr 2024 20:14:55 +0000 From: Andreas Rheinhardt To: ffmpeg-devel@ffmpeg.org Date: Mon, 8 Apr 2024 22:13:43 +0200 Message-ID: X-Mailer: git-send-email 2.40.1 In-Reply-To: References: X-TMN: [OJFGxV/ehQW9to9JEyCCK7F/C0pApbHZV58DUFdYlRY=] X-ClientProxiedBy: ZR0P278CA0124.CHEP278.PROD.OUTLOOK.COM (2603:10a6:910:20::21) To GV1P250MB0737.EURP250.PROD.OUTLOOK.COM (2603:10a6:150:8e::17) X-Microsoft-Original-Message-ID: <20240408201405.2136840-4-andreas.rheinhardt@outlook.com> MIME-Version: 1.0 X-MS-Exchange-MessageSentRepresentingType: 1 X-MS-PublicTrafficType: Email X-MS-TrafficTypeDiagnostic: GV1P250MB0737:EE_|PAXP250MB0494:EE_ X-MS-Office365-Filtering-Correlation-Id: 935ab315-f797-40d5-61b3-08dc58088ea7 X-Microsoft-Antispam: BCL:0; X-Microsoft-Antispam-Message-Info: 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 X-MS-Exchange-AntiSpam-MessageData-ChunkCount: 1 X-MS-Exchange-AntiSpam-MessageData-0: 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 X-OriginatorOrg: outlook.com X-MS-Exchange-CrossTenant-Network-Message-Id: 935ab315-f797-40d5-61b3-08dc58088ea7 X-MS-Exchange-CrossTenant-AuthSource: GV1P250MB0737.EURP250.PROD.OUTLOOK.COM X-MS-Exchange-CrossTenant-AuthAs: Internal X-MS-Exchange-CrossTenant-OriginalArrivalTime: 08 Apr 2024 20:14:55.8190 (UTC) X-MS-Exchange-CrossTenant-FromEntityHeader: Hosted X-MS-Exchange-CrossTenant-Id: 84df9e7f-e9f6-40af-b435-aaaaaaaaaaaa X-MS-Exchange-CrossTenant-RMS-PersistedConsumerOrg: 00000000-0000-0000-0000-000000000000 X-MS-Exchange-Transport-CrossTenantHeadersStamped: PAXP250MB0494 Subject: [FFmpeg-devel] [PATCH v2 05/27] avcodec/vp9: Switch to ProgressFrames X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.29 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: Anton Khirnov , Andreas Rheinhardt Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: d9phwertnuDE This already fixes a race in the vp9-encparams test. In this test, side data is added to the current frame after having been decoded (and therefore after ff_thread_finish_setup() has been called). Yet the update_thread_context callback called ff_thread_ref_frame() and therefore av_frame_ref() with this frame as source frame and the ensuing read was unsynchronised with adding the side data, i.e. there was a data race. By switching to the ProgressFrame API the implicit av_frame_ref() is removed and the race fixed except if this frame is later reused by a show-existing-frame which uses an explicit av_frame_ref(). The vp9-encparams test does not cover this, so this commit already fixes all the races in this test. This decoder kept multiple references to the same ThreadFrames in the same context and therefore had lots of implicit av_frame_ref() even when decoding single-threaded. This incurred lots of small allocations: When decoding an ordinary 10s video in single-threaded mode the number of allocations reported by Valgrind went down from 57,814 to 20,908; for 10 threads it went down from 84,223 to 21,901. Reviewed-by: Anton Khirnov Signed-off-by: Andreas Rheinhardt --- libavcodec/dxva2_vp9.c | 4 +- libavcodec/vaapi_vp9.c | 2 +- libavcodec/vp9.c | 137 +++++++++++------------------------ libavcodec/vp9_mc_template.c | 2 +- libavcodec/vp9block.c | 5 +- libavcodec/vp9dec.h | 6 +- libavcodec/vp9lpf.c | 1 + libavcodec/vp9mvs.c | 4 +- libavcodec/vp9recon.c | 19 ++--- libavcodec/vp9shared.h | 6 +- 10 files changed, 70 insertions(+), 116 deletions(-) diff --git a/libavcodec/dxva2_vp9.c b/libavcodec/dxva2_vp9.c index 1498deb3c8..ca8b3b136d 100644 --- a/libavcodec/dxva2_vp9.c +++ b/libavcodec/dxva2_vp9.c @@ -79,7 +79,7 @@ int ff_dxva2_vp9_fill_picture_parameters(const AVCodecContext *avctx, AVDXVACont pp->Reserved8Bits = 0; for (i = 0; i < 8; i++) { - if (h->refs[i].f->buf[0]) { + if (h->refs[i].f) { fill_picture_entry(&pp->ref_frame_map[i], ff_dxva2_get_surface_index(avctx, ctx, h->refs[i].f, 0), 0); pp->ref_frame_coded_width[i] = h->refs[i].f->width; pp->ref_frame_coded_height[i] = h->refs[i].f->height; @@ -89,7 +89,7 @@ int ff_dxva2_vp9_fill_picture_parameters(const AVCodecContext *avctx, AVDXVACont for (i = 0; i < 3; i++) { uint8_t refidx = h->h.refidx[i]; - if (h->refs[refidx].f->buf[0]) + if (h->refs[refidx].f) fill_picture_entry(&pp->frame_refs[i], ff_dxva2_get_surface_index(avctx, ctx, h->refs[refidx].f, 0), 0); else pp->frame_refs[i].bPicEntry = 0xFF; diff --git a/libavcodec/vaapi_vp9.c b/libavcodec/vaapi_vp9.c index 9dc7d5e72b..b8e760c807 100644 --- a/libavcodec/vaapi_vp9.c +++ b/libavcodec/vaapi_vp9.c @@ -100,7 +100,7 @@ static int vaapi_vp9_start_frame(AVCodecContext *avctx, } for (i = 0; i < 8; i++) { - if (h->refs[i].f->buf[0]) + if (h->refs[i].f) pic_param.reference_frames[i] = ff_vaapi_get_surface_id(h->refs[i].f); else pic_param.reference_frames[i] = VA_INVALID_ID; diff --git a/libavcodec/vp9.c b/libavcodec/vp9.c index 6bcda8bfff..bd52478ce7 100644 --- a/libavcodec/vp9.c +++ b/libavcodec/vp9.c @@ -30,9 +30,9 @@ #include "hwaccel_internal.h" #include "hwconfig.h" #include "profiles.h" +#include "progressframe.h" #include "refstruct.h" #include "thread.h" -#include "threadframe.h" #include "pthread_internal.h" #include "videodsp.h" @@ -100,7 +100,7 @@ static void vp9_tile_data_free(VP9TileData *td) static void vp9_frame_unref(VP9Frame *f) { - ff_thread_release_ext_buffer(&f->tf); + ff_progress_frame_unref(&f->tf); ff_refstruct_unref(&f->extradata); ff_refstruct_unref(&f->hwaccel_picture_private); f->segmentation_map = NULL; @@ -111,7 +111,7 @@ static int vp9_frame_alloc(AVCodecContext *avctx, VP9Frame *f) VP9Context *s = avctx->priv_data; int ret, sz; - ret = ff_thread_get_ext_buffer(avctx, &f->tf, AV_GET_BUFFER_FLAG_REF); + ret = ff_progress_frame_get_buffer(avctx, &f->tf, AV_GET_BUFFER_FLAG_REF); if (ret < 0) return ret; @@ -147,13 +147,9 @@ fail: return ret; } -static int vp9_frame_ref(VP9Frame *dst, VP9Frame *src) +static void vp9_frame_ref(VP9Frame *dst, const VP9Frame *src) { - int ret; - - ret = ff_thread_ref_frame(&dst->tf, &src->tf); - if (ret < 0) - return ret; + ff_progress_frame_ref(&dst->tf, &src->tf); dst->extradata = ff_refstruct_ref(src->extradata); @@ -163,8 +159,13 @@ static int vp9_frame_ref(VP9Frame *dst, VP9Frame *src) ff_refstruct_replace(&dst->hwaccel_picture_private, src->hwaccel_picture_private); +} - return 0; +static void vp9_frame_replace(VP9Frame *dst, const VP9Frame *src) +{ + vp9_frame_unref(dst); + if (src && src->tf.f) + vp9_frame_ref(dst, src); } static int update_size(AVCodecContext *avctx, int w, int h) @@ -589,9 +590,9 @@ static int decode_frame_header(AVCodecContext *avctx, s->s.h.signbias[1] = get_bits1(&s->gb) && !s->s.h.errorres; s->s.h.refidx[2] = get_bits(&s->gb, 3); s->s.h.signbias[2] = get_bits1(&s->gb) && !s->s.h.errorres; - if (!s->s.refs[s->s.h.refidx[0]].f->buf[0] || - !s->s.refs[s->s.h.refidx[1]].f->buf[0] || - !s->s.refs[s->s.h.refidx[2]].f->buf[0]) { + if (!s->s.refs[s->s.h.refidx[0]].f || + !s->s.refs[s->s.h.refidx[1]].f || + !s->s.refs[s->s.h.refidx[2]].f) { av_log(avctx, AV_LOG_ERROR, "Not all references are available\n"); return AVERROR_INVALIDDATA; } @@ -611,7 +612,8 @@ static int decode_frame_header(AVCodecContext *avctx, // Note that in this code, "CUR_FRAME" is actually before we // have formally allocated a frame, and thus actually represents // the _last_ frame - s->s.h.use_last_frame_mvs &= s->s.frames[CUR_FRAME].tf.f->width == w && + s->s.h.use_last_frame_mvs &= s->s.frames[CUR_FRAME].tf.f && + s->s.frames[CUR_FRAME].tf.f->width == w && s->s.frames[CUR_FRAME].tf.f->height == h; if (get_bits1(&s->gb)) // display size skip_bits(&s->gb, 32); @@ -1240,16 +1242,12 @@ static av_cold int vp9_decode_free(AVCodecContext *avctx) VP9Context *s = avctx->priv_data; int i; - for (i = 0; i < 3; i++) { + for (int i = 0; i < 3; i++) vp9_frame_unref(&s->s.frames[i]); - av_frame_free(&s->s.frames[i].tf.f); - } ff_refstruct_pool_uninit(&s->frame_extradata_pool); for (i = 0; i < 8; i++) { - ff_thread_release_ext_buffer(&s->s.refs[i]); - av_frame_free(&s->s.refs[i].f); - ff_thread_release_ext_buffer(&s->next_refs[i]); - av_frame_free(&s->next_refs[i].f); + ff_progress_frame_unref(&s->s.refs[i]); + ff_progress_frame_unref(&s->next_refs[i]); } free_buffers(s); @@ -1384,7 +1382,7 @@ static int decode_tiles(AVCodecContext *avctx, // FIXME maybe we can make this more finegrained by running the // loopfilter per-block instead of after each sbrow // In fact that would also make intra pred left preparation easier? - ff_thread_report_progress(&s->s.frames[CUR_FRAME].tf, row >> 3, 0); + ff_progress_frame_report(&s->s.frames[CUR_FRAME].tf, row >> 3); } } return 0; @@ -1561,12 +1559,13 @@ static int vp9_decode_frame(AVCodecContext *avctx, AVFrame *frame, int ret, i, j, ref; int retain_segmap_ref = s->s.frames[REF_FRAME_SEGMAP].segmentation_map && (!s->s.h.segmentation.enabled || !s->s.h.segmentation.update_map); + const VP9Frame *src; AVFrame *f; if ((ret = decode_frame_header(avctx, data, size, &ref)) < 0) { return ret; } else if (ret == 0) { - if (!s->s.refs[ref].f->buf[0]) { + if (!s->s.refs[ref].f) { av_log(avctx, AV_LOG_ERROR, "Requested reference %d not available\n", ref); return AVERROR_INVALIDDATA; } @@ -1574,33 +1573,19 @@ static int vp9_decode_frame(AVCodecContext *avctx, AVFrame *frame, return ret; frame->pts = pkt->pts; frame->pkt_dts = pkt->dts; - for (i = 0; i < 8; i++) { - if (s->next_refs[i].f->buf[0]) - ff_thread_release_ext_buffer(&s->next_refs[i]); - if (s->s.refs[i].f->buf[0] && - (ret = ff_thread_ref_frame(&s->next_refs[i], &s->s.refs[i])) < 0) - return ret; - } + for (int i = 0; i < 8; i++) + ff_progress_frame_replace(&s->next_refs[i], &s->s.refs[i]); *got_frame = 1; return pkt->size; } data += ret; size -= ret; - if (!retain_segmap_ref || s->s.h.keyframe || s->s.h.intraonly) { - if (s->s.frames[REF_FRAME_SEGMAP].tf.f->buf[0]) - vp9_frame_unref(&s->s.frames[REF_FRAME_SEGMAP]); - if (!s->s.h.keyframe && !s->s.h.intraonly && !s->s.h.errorres && s->s.frames[CUR_FRAME].tf.f->buf[0] && - (ret = vp9_frame_ref(&s->s.frames[REF_FRAME_SEGMAP], &s->s.frames[CUR_FRAME])) < 0) - return ret; - } - if (s->s.frames[REF_FRAME_MVPAIR].tf.f->buf[0]) - vp9_frame_unref(&s->s.frames[REF_FRAME_MVPAIR]); - if (!s->s.h.intraonly && !s->s.h.keyframe && !s->s.h.errorres && s->s.frames[CUR_FRAME].tf.f->buf[0] && - (ret = vp9_frame_ref(&s->s.frames[REF_FRAME_MVPAIR], &s->s.frames[CUR_FRAME])) < 0) - return ret; - if (s->s.frames[CUR_FRAME].tf.f->buf[0]) - vp9_frame_unref(&s->s.frames[CUR_FRAME]); + src = !s->s.h.keyframe && !s->s.h.intraonly && !s->s.h.errorres ? &s->s.frames[CUR_FRAME] : NULL; + if (!retain_segmap_ref || s->s.h.keyframe || s->s.h.intraonly) + vp9_frame_replace(&s->s.frames[REF_FRAME_SEGMAP], src); + vp9_frame_replace(&s->s.frames[REF_FRAME_MVPAIR], src); + vp9_frame_unref(&s->s.frames[CUR_FRAME]); if ((ret = vp9_frame_alloc(avctx, &s->s.frames[CUR_FRAME])) < 0) return ret; f = s->s.frames[CUR_FRAME].tf.f; @@ -1610,7 +1595,8 @@ static int vp9_decode_frame(AVCodecContext *avctx, AVFrame *frame, f->flags &= ~AV_FRAME_FLAG_KEY; f->pict_type = (s->s.h.keyframe || s->s.h.intraonly) ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P; - if (s->s.frames[REF_FRAME_SEGMAP].tf.f->buf[0] && + // Non-existent frames have the implicit dimension 0x0 != CUR_FRAME + if (!s->s.frames[REF_FRAME_MVPAIR].tf.f || (s->s.frames[REF_FRAME_MVPAIR].tf.f->width != s->s.frames[CUR_FRAME].tf.f->width || s->s.frames[REF_FRAME_MVPAIR].tf.f->height != s->s.frames[CUR_FRAME].tf.f->height)) { vp9_frame_unref(&s->s.frames[REF_FRAME_SEGMAP]); @@ -1618,15 +1604,9 @@ static int vp9_decode_frame(AVCodecContext *avctx, AVFrame *frame, // ref frame setup for (i = 0; i < 8; i++) { - if (s->next_refs[i].f->buf[0]) - ff_thread_release_ext_buffer(&s->next_refs[i]); - if (s->s.h.refreshrefmask & (1 << i)) { - ret = ff_thread_ref_frame(&s->next_refs[i], &s->s.frames[CUR_FRAME].tf); - } else if (s->s.refs[i].f->buf[0]) { - ret = ff_thread_ref_frame(&s->next_refs[i], &s->s.refs[i]); - } - if (ret < 0) - return ret; + ff_progress_frame_replace(&s->next_refs[i], + s->s.h.refreshrefmask & (1 << i) ? + &s->s.frames[CUR_FRAME].tf : &s->s.refs[i]); } if (avctx->hwaccel) { @@ -1736,7 +1716,7 @@ static int vp9_decode_frame(AVCodecContext *avctx, AVFrame *frame, { ret = decode_tiles(avctx, data, size); if (ret < 0) { - ff_thread_report_progress(&s->s.frames[CUR_FRAME].tf, INT_MAX, 0); + ff_progress_frame_report(&s->s.frames[CUR_FRAME].tf, INT_MAX); return ret; } } @@ -1752,7 +1732,7 @@ static int vp9_decode_frame(AVCodecContext *avctx, AVFrame *frame, ff_thread_finish_setup(avctx); } } while (s->pass++ == 1); - ff_thread_report_progress(&s->s.frames[CUR_FRAME].tf, INT_MAX, 0); + ff_progress_frame_report(&s->s.frames[CUR_FRAME].tf, INT_MAX); if (s->td->error_info < 0) { av_log(avctx, AV_LOG_ERROR, "Failed to decode tile data\n"); @@ -1767,13 +1747,8 @@ static int vp9_decode_frame(AVCodecContext *avctx, AVFrame *frame, finish: // ref frame setup - for (i = 0; i < 8; i++) { - if (s->s.refs[i].f->buf[0]) - ff_thread_release_ext_buffer(&s->s.refs[i]); - if (s->next_refs[i].f->buf[0] && - (ret = ff_thread_ref_frame(&s->s.refs[i], &s->next_refs[i])) < 0) - return ret; - } + for (int i = 0; i < 8; i++) + ff_progress_frame_replace(&s->s.refs[i], &s->next_refs[i]); if (!s->s.h.invisible) { if ((ret = av_frame_ref(frame, s->s.frames[CUR_FRAME].tf.f)) < 0) @@ -1792,7 +1767,7 @@ static void vp9_decode_flush(AVCodecContext *avctx) for (i = 0; i < 3; i++) vp9_frame_unref(&s->s.frames[i]); for (i = 0; i < 8; i++) - ff_thread_release_ext_buffer(&s->s.refs[i]); + ff_progress_frame_unref(&s->s.refs[i]); if (FF_HW_HAS_CB(avctx, flush)) FF_HW_SIMPLE_CALL(avctx, flush); @@ -1814,42 +1789,18 @@ static av_cold int vp9_decode_init(AVCodecContext *avctx) } #endif - for (int i = 0; i < 3; i++) { - s->s.frames[i].tf.f = av_frame_alloc(); - if (!s->s.frames[i].tf.f) - return AVERROR(ENOMEM); - } - for (int i = 0; i < 8; i++) { - s->s.refs[i].f = av_frame_alloc(); - s->next_refs[i].f = av_frame_alloc(); - if (!s->s.refs[i].f || !s->next_refs[i].f) - return AVERROR(ENOMEM); - } return 0; } #if HAVE_THREADS static int vp9_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src) { - int i, ret; VP9Context *s = dst->priv_data, *ssrc = src->priv_data; - for (i = 0; i < 3; i++) { - if (s->s.frames[i].tf.f->buf[0]) - vp9_frame_unref(&s->s.frames[i]); - if (ssrc->s.frames[i].tf.f->buf[0]) { - if ((ret = vp9_frame_ref(&s->s.frames[i], &ssrc->s.frames[i])) < 0) - return ret; - } - } - for (i = 0; i < 8; i++) { - if (s->s.refs[i].f->buf[0]) - ff_thread_release_ext_buffer(&s->s.refs[i]); - if (ssrc->next_refs[i].f->buf[0]) { - if ((ret = ff_thread_ref_frame(&s->s.refs[i], &ssrc->next_refs[i])) < 0) - return ret; - } - } + for (int i = 0; i < 3; i++) + vp9_frame_replace(&s->s.frames[i], &ssrc->s.frames[i]); + for (int i = 0; i < 8; i++) + ff_progress_frame_replace(&s->s.refs[i], &ssrc->next_refs[i]); ff_refstruct_replace(&s->frame_extradata_pool, ssrc->frame_extradata_pool); s->frame_extradata_pool_size = ssrc->frame_extradata_pool_size; @@ -1889,7 +1840,7 @@ const FFCodec ff_vp9_decoder = { .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS | AV_CODEC_CAP_SLICE_THREADS, .caps_internal = FF_CODEC_CAP_INIT_CLEANUP | FF_CODEC_CAP_SLICE_THREAD_HAS_MF | - FF_CODEC_CAP_ALLOCATE_PROGRESS, + FF_CODEC_CAP_USES_PROGRESSFRAMES, .flush = vp9_decode_flush, UPDATE_THREAD_CONTEXT(vp9_decode_update_thread_context), .p.profiles = NULL_IF_CONFIG_SMALL(ff_vp9_profiles), diff --git a/libavcodec/vp9_mc_template.c b/libavcodec/vp9_mc_template.c index e654c0e5ed..81e4ed59c7 100644 --- a/libavcodec/vp9_mc_template.c +++ b/libavcodec/vp9_mc_template.c @@ -36,7 +36,7 @@ static void FN(inter_pred)(VP9TileData *td) const VP9Context *s = td->s; VP9Block *b = td->b; int row = td->row, col = td->col; - const ThreadFrame *tref1 = &s->s.refs[s->s.h.refidx[b->ref[0]]], *tref2; + const ProgressFrame *tref1 = &s->s.refs[s->s.h.refidx[b->ref[0]]], *tref2; const AVFrame *ref1 = tref1->f, *ref2; int w1 = ref1->width, h1 = ref1->height, w2, h2; ptrdiff_t ls_y = td->y_stride, ls_uv = td->uv_stride; diff --git a/libavcodec/vp9block.c b/libavcodec/vp9block.c index 5743f048cc..3a694763ce 100644 --- a/libavcodec/vp9block.c +++ b/libavcodec/vp9block.c @@ -22,8 +22,9 @@ */ #include "libavutil/avassert.h" +#include "libavutil/frame.h" -#include "threadframe.h" +#include "progressframe.h" #include "vp89_rac.h" #include "vp9.h" #include "vp9data.h" @@ -113,7 +114,7 @@ static void decode_mode(VP9TileData *td) uint8_t *refsegmap = s->s.frames[REF_FRAME_SEGMAP].segmentation_map; if (!s->s.frames[REF_FRAME_SEGMAP].uses_2pass) - ff_thread_await_progress(&s->s.frames[REF_FRAME_SEGMAP].tf, row >> 3, 0); + ff_progress_frame_await(&s->s.frames[REF_FRAME_SEGMAP].tf, row >> 3); for (y = 0; y < h4; y++) { int idx_base = (y + row) * 8 * s->sb_cols + col; for (x = 0; x < w4; x++) diff --git a/libavcodec/vp9dec.h b/libavcodec/vp9dec.h index 013aac49eb..81dc801052 100644 --- a/libavcodec/vp9dec.h +++ b/libavcodec/vp9dec.h @@ -29,8 +29,8 @@ #include #include "libavutil/mem_internal.h" +#include "libavutil/pixfmt.h" #include "libavutil/thread.h" -#include "libavutil/internal.h" #include "get_bits.h" #include "videodsp.h" @@ -120,7 +120,7 @@ typedef struct VP9Context { int w, h; enum AVPixelFormat pix_fmt, last_fmt, gf_fmt; unsigned sb_cols, sb_rows, rows, cols; - ThreadFrame next_refs[8]; + ProgressFrame next_refs[8]; struct { uint8_t lim_lut[64]; @@ -245,7 +245,7 @@ void ff_vp9_decode_block(VP9TileData *td, int row, int col, VP9Filter *lflvl, ptrdiff_t yoff, ptrdiff_t uvoff, enum BlockLevel bl, enum BlockPartition bp); -void ff_vp9_loopfilter_sb(AVCodecContext *avctx, VP9Filter *lflvl, +void ff_vp9_loopfilter_sb(struct AVCodecContext *avctx, VP9Filter *lflvl, int row, int col, ptrdiff_t yoff, ptrdiff_t uvoff); void ff_vp9_intra_recon_8bpp(VP9TileData *td, diff --git a/libavcodec/vp9lpf.c b/libavcodec/vp9lpf.c index 414cede852..afeebebf59 100644 --- a/libavcodec/vp9lpf.c +++ b/libavcodec/vp9lpf.c @@ -21,6 +21,7 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "avcodec.h" #include "vp9dec.h" static av_always_inline void filter_plane_cols(VP9Context *s, int col, int ss_h, int ss_v, diff --git a/libavcodec/vp9mvs.c b/libavcodec/vp9mvs.c index b93d878d6f..b706d1660f 100644 --- a/libavcodec/vp9mvs.c +++ b/libavcodec/vp9mvs.c @@ -21,7 +21,7 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ -#include "threadframe.h" +#include "progressframe.h" #include "vp89_rac.h" #include "vp9data.h" #include "vp9dec.h" @@ -175,7 +175,7 @@ static void find_ref_mvs(VP9TileData *td, VP9mvrefPair *mv = &s->s.frames[REF_FRAME_MVPAIR].mv[row * s->sb_cols * 8 + col]; if (!s->s.frames[REF_FRAME_MVPAIR].uses_2pass) - ff_thread_await_progress(&s->s.frames[REF_FRAME_MVPAIR].tf, row >> 3, 0); + ff_progress_frame_await(&s->s.frames[REF_FRAME_MVPAIR].tf, row >> 3); if (mv->ref[0] == ref) RETURN_MV(mv->mv[0]); else if (mv->ref[1] == ref) diff --git a/libavcodec/vp9recon.c b/libavcodec/vp9recon.c index 073c04b47d..ef08ed17c8 100644 --- a/libavcodec/vp9recon.c +++ b/libavcodec/vp9recon.c @@ -22,9 +22,10 @@ */ #include "libavutil/avassert.h" +#include "libavutil/frame.h" #include "libavutil/mem_internal.h" -#include "threadframe.h" +#include "progressframe.h" #include "videodsp.h" #include "vp9data.h" #include "vp9dec.h" @@ -298,7 +299,7 @@ void ff_vp9_intra_recon_16bpp(VP9TileData *td, ptrdiff_t y_off, ptrdiff_t uv_off static av_always_inline void mc_luma_unscaled(VP9TileData *td, const vp9_mc_func (*mc)[2], uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *ref, ptrdiff_t ref_stride, - const ThreadFrame *ref_frame, + const ProgressFrame *ref_frame, ptrdiff_t y, ptrdiff_t x, const VP9mv *mv, int bw, int bh, int w, int h, int bytesperpixel) { @@ -314,7 +315,7 @@ static av_always_inline void mc_luma_unscaled(VP9TileData *td, const vp9_mc_func // we use +7 because the last 7 pixels of each sbrow can be changed in // the longest loopfilter of the next sbrow th = (y + bh + 4 * !!my + 7) >> 6; - ff_thread_await_progress(ref_frame, FFMAX(th, 0), 0); + ff_progress_frame_await(ref_frame, FFMAX(th, 0)); // The arm/aarch64 _hv filters read one more row than what actually is // needed, so switch to emulated edge one pixel sooner vertically // (!!my * 5) than horizontally (!!mx * 4). @@ -336,7 +337,7 @@ static av_always_inline void mc_chroma_unscaled(VP9TileData *td, const vp9_mc_fu ptrdiff_t dst_stride, const uint8_t *ref_u, ptrdiff_t src_stride_u, const uint8_t *ref_v, ptrdiff_t src_stride_v, - const ThreadFrame *ref_frame, + const ProgressFrame *ref_frame, ptrdiff_t y, ptrdiff_t x, const VP9mv *mv, int bw, int bh, int w, int h, int bytesperpixel) { @@ -353,7 +354,7 @@ static av_always_inline void mc_chroma_unscaled(VP9TileData *td, const vp9_mc_fu // we use +7 because the last 7 pixels of each sbrow can be changed in // the longest loopfilter of the next sbrow th = (y + bh + 4 * !!my + 7) >> (6 - s->ss_v); - ff_thread_await_progress(ref_frame, FFMAX(th, 0), 0); + ff_progress_frame_await(ref_frame, FFMAX(th, 0)); // The arm/aarch64 _hv filters read one more row than what actually is // needed, so switch to emulated edge one pixel sooner vertically // (!!my * 5) than horizontally (!!mx * 4). @@ -407,7 +408,7 @@ static av_always_inline void mc_luma_scaled(VP9TileData *td, vp9_scaled_mc_func const vp9_mc_func (*mc)[2], uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *ref, ptrdiff_t ref_stride, - const ThreadFrame *ref_frame, + const ProgressFrame *ref_frame, ptrdiff_t y, ptrdiff_t x, const VP9mv *in_mv, int px, int py, int pw, int ph, int bw, int bh, int w, int h, int bytesperpixel, @@ -444,7 +445,7 @@ static av_always_inline void mc_luma_scaled(VP9TileData *td, vp9_scaled_mc_func // we use +7 because the last 7 pixels of each sbrow can be changed in // the longest loopfilter of the next sbrow th = (y + refbh_m1 + 4 + 7) >> 6; - ff_thread_await_progress(ref_frame, FFMAX(th, 0), 0); + ff_progress_frame_await(ref_frame, FFMAX(th, 0)); // The arm/aarch64 _hv filters read one more row than what actually is // needed, so switch to emulated edge one pixel sooner vertically // (y + 5 >= h - refbh_m1) than horizontally (x + 4 >= w - refbw_m1). @@ -467,7 +468,7 @@ static av_always_inline void mc_chroma_scaled(VP9TileData *td, vp9_scaled_mc_fun ptrdiff_t dst_stride, const uint8_t *ref_u, ptrdiff_t src_stride_u, const uint8_t *ref_v, ptrdiff_t src_stride_v, - const ThreadFrame *ref_frame, + const ProgressFrame *ref_frame, ptrdiff_t y, ptrdiff_t x, const VP9mv *in_mv, int px, int py, int pw, int ph, int bw, int bh, int w, int h, int bytesperpixel, @@ -514,7 +515,7 @@ static av_always_inline void mc_chroma_scaled(VP9TileData *td, vp9_scaled_mc_fun // we use +7 because the last 7 pixels of each sbrow can be changed in // the longest loopfilter of the next sbrow th = (y + refbh_m1 + 4 + 7) >> (6 - s->ss_v); - ff_thread_await_progress(ref_frame, FFMAX(th, 0), 0); + ff_progress_frame_await(ref_frame, FFMAX(th, 0)); // The arm/aarch64 _hv filters read one more row than what actually is // needed, so switch to emulated edge one pixel sooner vertically // (y + 5 >= h - refbh_m1) than horizontally (x + 4 >= w - refbw_m1). diff --git a/libavcodec/vp9shared.h b/libavcodec/vp9shared.h index b445a2a746..805668416f 100644 --- a/libavcodec/vp9shared.h +++ b/libavcodec/vp9shared.h @@ -29,8 +29,8 @@ #include "libavutil/mem_internal.h" +#include "progressframe.h" #include "vp9.h" -#include "threadframe.h" enum BlockPartition { PARTITION_NONE, // [ ] <-. @@ -63,7 +63,7 @@ typedef struct VP9mvrefPair { } VP9mvrefPair; typedef struct VP9Frame { - ThreadFrame tf; + ProgressFrame tf; void *extradata; ///< RefStruct reference uint8_t *segmentation_map; VP9mvrefPair *mv; @@ -164,7 +164,7 @@ typedef struct VP9BitstreamHeader { typedef struct VP9SharedContext { VP9BitstreamHeader h; - ThreadFrame refs[8]; + ProgressFrame refs[8]; #define CUR_FRAME 0 #define REF_FRAME_MVPAIR 1 #define REF_FRAME_SEGMAP 2