From: Andreas Rheinhardt <andreas.rheinhardt@outlook.com> To: ffmpeg-devel@ffmpeg.org Subject: Re: [FFmpeg-devel] [PATCH] avcodec/libaomenc: remove one memcpy when queueing packets Date: Thu, 25 Aug 2022 18:29:11 +0200 Message-ID: <DB6PR0101MB2214FAB7B46119A143EF04A58F729@DB6PR0101MB2214.eurprd01.prod.exchangelabs.com> (raw) In-Reply-To: <20220824225209.4076-1-jamrial@gmail.com> James Almer: > Don't use an intermediary buffer. Achieve this by replacing FrameListData with > a PacketList, and by allocating and populating every packet's payload before > inserting them into the list. > > Signed-off-by: James Almer <jamrial@gmail.com> > --- > libavcodec/libaomenc.c | 195 +++++++++++++++-------------------------- > 1 file changed, 70 insertions(+), 125 deletions(-) > > diff --git a/libavcodec/libaomenc.c b/libavcodec/libaomenc.c > index 485f554165..f9476b3ddf 100644 > --- a/libavcodec/libaomenc.c > +++ b/libavcodec/libaomenc.c > @@ -38,6 +38,7 @@ > > #include "av1.h" > #include "avcodec.h" > +#include "bytestream.h" > #include "bsf.h" > #include "codec_internal.h" > #include "encode.h" > @@ -46,24 +47,6 @@ > #include "packet_internal.h" > #include "profiles.h" > > -/* > - * Portion of struct aom_codec_cx_pkt from aom_encoder.h. > - * One encoded frame returned from the library. > - */ > -struct FrameListData { > - void *buf; /**< compressed data buffer */ > - size_t sz; /**< length of compressed data */ > - int64_t pts; /**< time stamp to show frame > - (in timebase units) */ > - unsigned long duration; /**< duration to show frame > - (in timebase units) */ > - uint32_t flags; /**< flags for this frame */ > - uint64_t sse[4]; > - int have_sse; /**< true if we have pending sse[] */ > - uint64_t frame_number; > - struct FrameListData *next; > -}; > - > typedef struct AOMEncoderContext { > AVClass *class; > AVBSFContext *bsf; > @@ -71,7 +54,8 @@ typedef struct AOMEncoderContext { > struct aom_image rawimg; > struct aom_fixed_buf twopass_stats; > unsigned twopass_stats_size; > - struct FrameListData *coded_frame_list; > + PacketList coded_frame_list; > + AVPacket *pkt; Renaming this variable to avpkt would improve clarity by simplifying distinguishing it from the aom_codec_cx_pkt packets. > int cpu_used; > int auto_alt_ref; > int arnr_max_frames; > @@ -283,33 +267,6 @@ static av_cold void dump_enc_cfg(AVCodecContext *avctx, > av_log(avctx, level, "\n"); > } > > -static void coded_frame_add(void *list, struct FrameListData *cx_frame) > -{ > - struct FrameListData **p = list; > - > - while (*p) > - p = &(*p)->next; > - *p = cx_frame; > - cx_frame->next = NULL; > -} > - > -static av_cold void free_coded_frame(struct FrameListData *cx_frame) > -{ > - av_freep(&cx_frame->buf); > - av_freep(&cx_frame); > -} > - > -static av_cold void free_frame_list(struct FrameListData *list) > -{ > - struct FrameListData *p = list; > - > - while (p) { > - list = list->next; > - free_coded_frame(p); > - p = list; > - } > -} > - > static av_cold int codecctl_int(AVCodecContext *avctx, > #ifdef UENUM1BYTE > aome_enc_control_id id, > @@ -432,7 +389,8 @@ static av_cold int aom_free(AVCodecContext *avctx) > aom_codec_destroy(&ctx->encoder); > av_freep(&ctx->twopass_stats.buf); > av_freep(&avctx->stats_out); > - free_frame_list(ctx->coded_frame_list); > + avpriv_packet_list_free(&ctx->coded_frame_list); > + av_packet_free(&ctx->pkt); > av_bsf_free(&ctx->bsf); > return 0; > } > @@ -1042,6 +1000,10 @@ static av_cold int aom_init(AVCodecContext *avctx, > return ret; > } > > + ctx->pkt = av_packet_alloc(); > + if (!ctx->pkt) > + return AVERROR(ENOMEM); > + This encoder does not have the INIT_CLEANUP flag set, so everything leaks in case the above allocation fails. In fact, it seems like there are already leaks in several errors paths in this function. > if (enccfg.rc_end_usage == AOM_CBR || > enccfg.g_pass != AOM_RC_ONE_PASS) { > cpb_props->max_bitrate = avctx->rc_max_rate; > @@ -1053,25 +1015,40 @@ static av_cold int aom_init(AVCodecContext *avctx, > return 0; > } > > -static inline void cx_pktcpy(AOMContext *ctx, > - struct FrameListData *dst, > +static inline int cx_pktcpy(AVCodecContext *avctx, We should not override the compiler's inlining behaviour unless we have a good reason to do so, so you could remove it while at it. > + AVPacket *dst, Wrong indentation. > const struct aom_codec_cx_pkt *src) > { > - dst->pts = src->data.frame.pts; > - dst->duration = src->data.frame.duration; > - dst->flags = src->data.frame.flags; > - dst->sz = src->data.frame.sz; > - dst->buf = src->data.frame.buf; > + AOMContext *ctx = avctx->priv_data; > + int av_unused pict_type; > + int ret; > + > + av_packet_unref(dst); Can dst ever be non-blank here (i.e. before the unref)? > + ret = ff_get_encode_buffer(avctx, dst, src->data.frame.sz, 0); > + if (ret < 0) { > + av_log(avctx, AV_LOG_ERROR, > + "Error getting output packet of size %"SIZE_SPECIFIER".\n", src->data.frame.sz); > + return ret; > + } > + memcpy(dst->data, src->data.frame.buf, src->data.frame.sz); > + dst->pts = dst->dts = src->data.frame.pts; > + > + if (src->data.frame.flags & AOM_FRAME_IS_KEY) { > + dst->flags |= AV_PKT_FLAG_KEY; > #ifdef AOM_FRAME_IS_INTRAONLY > - dst->frame_number = ++ctx->frame_number; > - dst->have_sse = ctx->have_sse; > + pict_type = AV_PICTURE_TYPE_I; > + } else if (src->data.frame.flags & AOM_FRAME_IS_INTRAONLY) { > + pict_type = AV_PICTURE_TYPE_I; > + } else { > + pict_type = AV_PICTURE_TYPE_P; > + } > + > if (ctx->have_sse) { > - /* associate last-seen SSE to the frame. */ > - /* Transfers ownership from ctx to dst. */ > - memcpy(dst->sse, ctx->sse, sizeof(dst->sse)); > + ff_side_data_set_encoder_stats(dst, 0, ctx->sse + 1, 3, pict_type); This function can fail. > ctx->have_sse = 0; > - } > #endif > + } > + return 0; > } > > /** > @@ -1081,50 +1058,32 @@ static inline void cx_pktcpy(AOMContext *ctx, > * @return packet data size on success > * @return a negative AVERROR on error > */ > -static int storeframe(AVCodecContext *avctx, struct FrameListData *cx_frame, > - AVPacket *pkt) > +static int storeframe(AVCodecContext *avctx, AVPacket *dst, AVPacket *src) > { > AOMContext *ctx = avctx->priv_data; > - int av_unused pict_type; > - int ret = ff_get_encode_buffer(avctx, pkt, cx_frame->sz, 0); > - if (ret < 0) { > - av_log(avctx, AV_LOG_ERROR, > - "Error getting output packet of size %"SIZE_SPECIFIER".\n", cx_frame->sz); > - return ret; > - } > - memcpy(pkt->data, cx_frame->buf, pkt->size); > - pkt->pts = pkt->dts = cx_frame->pts; > + const uint8_t *sd; > + size_t size; > + int ret; > > - if (!!(cx_frame->flags & AOM_FRAME_IS_KEY)) { > - pkt->flags |= AV_PKT_FLAG_KEY; > -#ifdef AOM_FRAME_IS_INTRAONLY > - pict_type = AV_PICTURE_TYPE_I; > - } else if (cx_frame->flags & AOM_FRAME_IS_INTRAONLY) { > - pict_type = AV_PICTURE_TYPE_I; > - } else { > - pict_type = AV_PICTURE_TYPE_P; > - } > - > - ff_side_data_set_encoder_stats(pkt, 0, cx_frame->sse + 1, > - cx_frame->have_sse ? 3 : 0, pict_type); > + av_packet_move_ref(dst, src); > > - if (cx_frame->have_sse) { > + sd = av_packet_get_side_data(dst, AV_PKT_DATA_QUALITY_STATS, &size); > + if (sd && size >= 4 + 4 + 8 * 3) { > int i; > + sd += 4 + 4; > for (i = 0; i < 3; ++i) { > - avctx->error[i] += cx_frame->sse[i + 1]; > + avctx->error[i] += bytestream_get_le64(&sd); > } > - cx_frame->have_sse = 0; > -#endif > } > > if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) { > - ret = av_bsf_send_packet(ctx->bsf, pkt); > + ret = av_bsf_send_packet(ctx->bsf, dst); > if (ret < 0) { > av_log(avctx, AV_LOG_ERROR, "extract_extradata filter " > "failed to send input packet\n"); > return ret; > } > - ret = av_bsf_receive_packet(ctx->bsf, pkt); > + ret = av_bsf_receive_packet(ctx->bsf, dst); > > if (ret < 0) { > av_log(avctx, AV_LOG_ERROR, "extract_extradata filter " > @@ -1132,7 +1091,7 @@ static int storeframe(AVCodecContext *avctx, struct FrameListData *cx_frame, > return ret; > } > } > - return pkt->size; > + return dst->size; > } > > /** > @@ -1148,16 +1107,14 @@ static int queue_frames(AVCodecContext *avctx, AVPacket *pkt_out) > AOMContext *ctx = avctx->priv_data; > const struct aom_codec_cx_pkt *pkt; > const void *iter = NULL; > - int size = 0; > + int ret, size = 0; > > - if (ctx->coded_frame_list) { > - struct FrameListData *cx_frame = ctx->coded_frame_list; > + if (!avpriv_packet_list_get(&ctx->coded_frame_list, ctx->pkt)) { > /* return the leading frame if we've already begun queueing */ > - size = storeframe(avctx, cx_frame, pkt_out); > - if (size < 0) > - return size; > - ctx->coded_frame_list = cx_frame->next; > - free_coded_frame(cx_frame); > + ret = storeframe(avctx, pkt_out, ctx->pkt); > + if (ret < 0) > + goto fail; > + size = ret; > } > > /* consume all available output from the encoder before returning. buffers > @@ -1165,37 +1122,21 @@ static int queue_frames(AVCodecContext *avctx, AVPacket *pkt_out) > while ((pkt = aom_codec_get_cx_data(&ctx->encoder, &iter))) { > switch (pkt->kind) { > case AOM_CODEC_CX_FRAME_PKT: > + ret = cx_pktcpy(avctx, ctx->pkt, pkt); > + if (ret < 0) > + goto fail; > if (!size) { > - struct FrameListData cx_frame; > - > /* avoid storing the frame when the list is empty and we haven't yet > * provided a frame for output */ > - av_assert0(!ctx->coded_frame_list); > - cx_pktcpy(ctx, &cx_frame, pkt); > - size = storeframe(avctx, &cx_frame, pkt_out); > - if (size < 0) > - return size; > + av_assert0(!ctx->coded_frame_list.head); > + ret = storeframe(avctx, pkt_out, ctx->pkt); > + if (ret < 0) > + goto fail; > + size = ret; > } else { > - struct FrameListData *cx_frame = > - av_malloc(sizeof(struct FrameListData)); > - > - if (!cx_frame) { > - av_log(avctx, AV_LOG_ERROR, > - "Frame queue element alloc failed\n"); > - return AVERROR(ENOMEM); > - } > - cx_pktcpy(ctx, cx_frame, pkt); > - cx_frame->buf = av_malloc(cx_frame->sz); > - > - if (!cx_frame->buf) { > - av_log(avctx, AV_LOG_ERROR, > - "Data buffer alloc (%"SIZE_SPECIFIER" bytes) failed\n", > - cx_frame->sz); > - av_freep(&cx_frame); > - return AVERROR(ENOMEM); > - } > - memcpy(cx_frame->buf, pkt->data.frame.buf, pkt->data.frame.sz); I am shocked to see that there were two memcpies. > - coded_frame_add(&ctx->coded_frame_list, cx_frame); > + ret = avpriv_packet_list_put(&ctx->coded_frame_list, ctx->pkt, NULL, 0); > + if (ret < 0) > + goto fail; wtf: Any error that queue_frames() returns will be translated to "got_packet = 1" by the caller (with return code 0). Error handling in this encoder seems to be a joke. > } > break; > case AOM_CODEC_STATS_PKT: > @@ -1236,6 +1177,10 @@ static int queue_frames(AVCodecContext *avctx, AVPacket *pkt_out) > } > > return size; > +fail: > + av_packet_unref(ctx->pkt); > + av_packet_unref(pkt_out); > + return ret; > } > > static enum AVPixelFormat aomfmt_to_pixfmt(struct aom_image *img) _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
next prev parent reply other threads:[~2022-08-25 16:29 UTC|newest] Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top 2022-08-24 22:52 James Almer 2022-08-25 16:29 ` Andreas Rheinhardt [this message] 2022-08-25 16:56 ` James Almer 2022-08-25 17:46 ` [FFmpeg-devel] [PATCH v2] " James Almer
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=DB6PR0101MB2214FAB7B46119A143EF04A58F729@DB6PR0101MB2214.eurprd01.prod.exchangelabs.com \ --to=andreas.rheinhardt@outlook.com \ --cc=ffmpeg-devel@ffmpeg.org \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: link
Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel This inbox may be cloned and mirrored by anyone: git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git # If you have public-inbox 1.1+ installed, you may # initialize and index your mirror using the following commands: public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \ ffmpegdev@gitmailbox.com public-inbox-index ffmpegdev Example config snippet for mirrors. AGPL code for this site: git clone https://public-inbox.org/public-inbox.git