From: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
To: ffmpeg-devel@ffmpeg.org
Subject: Re: [FFmpeg-devel] [PATCH v3 08/12] lavc/frame_thread_encoder: avoid assigning a whole AVCodecContext
Date: Sun, 24 Mar 2024 12:25:53 +0100
Message-ID: <GV1P250MB07375CED352AE5DFA18655198F372@GV1P250MB0737.EURP250.PROD.OUTLOOK.COM> (raw)
In-Reply-To: <20240324092917.12874-2-anton@khirnov.net>
Anton Khirnov:
> It is highly unsafe, as AVCodecContext contains many allocated fields.
> Everything needed by worked threads should be covered by
> * routing through AVCodecParameters
> * av_opt_copy()
> * copying quantisation matrices manually
>
> avcodec_free_context() can now be used for per-thread contexts.
So now we have to maintain a list of stuff to be copied just to avoid
this "unsafe" copying. Sounds worse to me.
There is another usecase you are breaking: The execute-callbacks. The
user may have used custom ones that use a thread pool which would allow
using both slice- and frame-threading together.
And you also forgot to copy get_encode_buffer.
> ---
> libavcodec/frame_thread_encoder.c | 48 ++++++++++++++++++++++++-------
> 1 file changed, 38 insertions(+), 10 deletions(-)
>
> diff --git a/libavcodec/frame_thread_encoder.c b/libavcodec/frame_thread_encoder.c
> index cda5158117..d34fe022a5 100644
> --- a/libavcodec/frame_thread_encoder.c
> +++ b/libavcodec/frame_thread_encoder.c
> @@ -28,6 +28,7 @@
> #include "libavutil/thread.h"
> #include "avcodec.h"
> #include "avcodec_internal.h"
> +#include "codec_par.h"
> #include "encode.h"
> #include "internal.h"
> #include "pthread_internal.h"
> @@ -111,8 +112,7 @@ static void * attribute_align_arg worker(void *v){
> pthread_mutex_unlock(&c->finished_task_mutex);
> }
> end:
> - ff_codec_close(avctx);
> - av_freep(&avctx);
> + avcodec_free_context(&avctx);
> return NULL;
> }
>
> @@ -121,6 +121,7 @@ av_cold int ff_frame_thread_encoder_init(AVCodecContext *avctx)
> int i=0;
> ThreadContext *c;
> AVCodecContext *thread_avctx = NULL;
> + AVCodecParameters *par = NULL;
> int ret;
>
> if( !(avctx->thread_type & FF_THREAD_FRAME)
> @@ -194,18 +195,27 @@ av_cold int ff_frame_thread_encoder_init(AVCodecContext *avctx)
> }
> }
>
> + par = avcodec_parameters_alloc();
> + if (!par) {
> + ret = AVERROR(ENOMEM);
> + goto fail;
> + }
> +
> + ret = avcodec_parameters_from_context(par, avctx);
> + if (ret < 0)
> + goto fail;
> +
> for(i=0; i<avctx->thread_count ; i++){
> - void *tmpv;
> thread_avctx = avcodec_alloc_context3(avctx->codec);
> if (!thread_avctx) {
> ret = AVERROR(ENOMEM);
> goto fail;
> }
> - tmpv = thread_avctx->priv_data;
> - *thread_avctx = *avctx;
> - thread_avctx->priv_data = tmpv;
> - thread_avctx->internal = NULL;
> - thread_avctx->hw_frames_ctx = NULL;
> +
> + ret = avcodec_parameters_to_context(thread_avctx, par);
> + if (ret < 0)
> + goto fail;
> +
> ret = av_opt_copy(thread_avctx, avctx);
> if (ret < 0)
> goto fail;
> @@ -217,6 +227,22 @@ av_cold int ff_frame_thread_encoder_init(AVCodecContext *avctx)
> thread_avctx->thread_count = 1;
> thread_avctx->active_thread_type &= ~FF_THREAD_FRAME;
>
> +#define DUP_MATRIX(m) \
> + if (avctx->m) { \
> + thread_avctx->m = av_memdup(avctx->m, 64 * sizeof(*avctx->m)); \
> + if (!thread_avctx->m) { \
> + ret = AVERROR(ENOMEM); \
> + goto fail; \
> + } \
> + }
> + DUP_MATRIX(intra_matrix);
> + DUP_MATRIX(chroma_intra_matrix);
> + DUP_MATRIX(inter_matrix);
> +
> +#undef DUP_MATRIX
> +
> + thread_avctx->stats_in = avctx->stats_in;
> +
> if ((ret = avcodec_open2(thread_avctx, avctx->codec, NULL)) < 0)
> goto fail;
> av_assert0(!thread_avctx->internal->frame_thread_encoder);
> @@ -227,12 +253,14 @@ av_cold int ff_frame_thread_encoder_init(AVCodecContext *avctx)
> }
> }
>
> + avcodec_parameters_free(&par);
> +
> avctx->active_thread_type = FF_THREAD_FRAME;
>
> return 0;
> fail:
> - ff_codec_close(thread_avctx);
> - av_freep(&thread_avctx);
> + avcodec_parameters_free(&par);
> + avcodec_free_context(&thread_avctx);
> avctx->thread_count = i;
> av_log(avctx, AV_LOG_ERROR, "ff_frame_thread_encoder_init failed\n");
> ff_frame_thread_encoder_free(avctx);
_______________________________________________
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:[~2024-03-24 11:26 UTC|newest]
Thread overview: 29+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-03-22 20:28 [FFmpeg-devel] [PATCH 01/12] tests/fate/ffmpeg: evaluate thread count in fate-run.sh rather than make Anton Khirnov
2024-03-22 20:28 ` [FFmpeg-devel] [PATCH 02/12] fftools/ffmpeg_demux: only call filter_codec_opts() when we have a decoder Anton Khirnov
2024-03-22 20:28 ` [FFmpeg-devel] [PATCH 03/12] fftools/cmdutils: do not use a random codec's private options Anton Khirnov
2024-03-22 20:28 ` [FFmpeg-devel] [PATCH 04/12] fftools/ffmpeg_dec: apply decoder options manually Anton Khirnov
2024-03-22 20:28 ` [FFmpeg-devel] [PATCH 05/12] fftools/ffmpeg_{demux, dec}: pass -bitexact through DecoderFlags Anton Khirnov
2024-03-22 20:28 ` [FFmpeg-devel] [PATCH 06/12] fftools/ffmpeg_dec: apply cropping manually Anton Khirnov
2024-03-23 2:53 ` James Almer
2024-03-23 7:20 ` Anton Khirnov
2024-03-22 20:28 ` [FFmpeg-devel] [PATCH 07/12] fftools/ffmpeg_filter: remove display matrix if we have applied it Anton Khirnov
2024-03-22 20:28 ` [FFmpeg-devel] [PATCH 08/12] lavc/frame_thread_encoder: avoid assigning a whole AVCodecContext Anton Khirnov
2024-03-22 20:42 ` Andreas Rheinhardt
2024-03-23 14:00 ` [FFmpeg-devel] [PATCH v2 " Anton Khirnov
2024-03-23 14:11 ` Andreas Rheinhardt
2024-03-24 9:27 ` Anton Khirnov
2024-03-24 10:19 ` Andreas Rheinhardt
2024-03-24 10:33 ` Anton Khirnov
2024-03-24 11:10 ` Andreas Rheinhardt
2024-03-24 11:23 ` Anton Khirnov
2024-03-24 0:11 ` [FFmpeg-devel] [PATCH " Michael Niedermayer
2024-03-24 9:29 ` [FFmpeg-devel] [PATCH v3 " Anton Khirnov
2024-03-24 11:25 ` Andreas Rheinhardt [this message]
2024-03-24 11:36 ` Anton Khirnov
2024-03-27 10:27 ` [FFmpeg-devel] [PATCH v4 " Anton Khirnov
2024-03-22 20:28 ` [FFmpeg-devel] [PATCH 09/12] lavc/decode: move sd_global_map to avcodec Anton Khirnov
2024-03-22 20:28 ` [FFmpeg-devel] [PATCH 10/12] lavc/encode: map AVCodecContext.decoded_side_data to coded_side_data Anton Khirnov
2024-03-23 2:33 ` James Almer
2024-03-23 7:20 ` Anton Khirnov
2024-03-22 20:28 ` [FFmpeg-devel] [PATCH 11/12] fftools/ffmpeg_enc: stop copying demuxer side data to the muxer Anton Khirnov
2024-03-22 20:28 ` [FFmpeg-devel] [PATCH 12/12] fftools/ffmpeg_demux: make InputStream.autorotate private Anton Khirnov
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=GV1P250MB07375CED352AE5DFA18655198F372@GV1P250MB0737.EURP250.PROD.OUTLOOK.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