[FFmpeg-devel,v6] lavc: convert frame threading to the receive_frame() pattern

From: Anton Khirnov <anton@khirnov.net>

Reorganize the code such that the frame threading code does not call the
decoders directly, but instead calls back into the generic decoding
code. This avoids duplicating the logic that wraps the decoder
invocation and will be useful in the following commits.
---
 libavcodec/decode.c        |  62 ++++++---
 libavcodec/decode.h        |   7 +
 libavcodec/internal.h      |   7 +
 libavcodec/pthread_frame.c | 279 ++++++++++++++++++++++++-------------
 libavcodec/thread.h        |  18 +--
 5 files changed, 247 insertions(+), 126 deletions(-)

On Wed, Dec 14, 2022 at 02:50:59AM +0100, Timo Rothenpieler wrote:
> From: Anton Khirnov <anton@khirnov.net>
> 
> Reorganize the code such that the frame threading code does not call the
> decoders directly, but instead calls back into the generic decoding
> code. This avoids duplicating the logic that wraps the decoder
> invocation and will be useful in the following commits.
> ---
>  libavcodec/decode.c        |  62 ++++++---
>  libavcodec/decode.h        |   7 +
>  libavcodec/internal.h      |   7 +
>  libavcodec/pthread_frame.c | 279 ++++++++++++++++++++++++-------------
>  libavcodec/thread.h        |  18 +--
>  5 files changed, 247 insertions(+), 126 deletions(-)

i didnt review the code changes but
seems working now

thx

[...]

On 14/12/2022 02:50, Timo Rothenpieler wrote:
> From: Anton Khirnov <anton@khirnov.net>
> 
> Reorganize the code such that the frame threading code does not call the
> decoders directly, but instead calls back into the generic decoding
> code. This avoids duplicating the logic that wraps the decoder
> invocation and will be useful in the following commits.

Will apply series in 24h if nobody objects.

Timo Rothenpieler:
> From: Anton Khirnov <anton@khirnov.net>
> 
> Reorganize the code such that the frame threading code does not call the
> decoders directly, but instead calls back into the generic decoding
> code. This avoids duplicating the logic that wraps the decoder
> invocation and will be useful in the following commits.
> ---
>  libavcodec/decode.c        |  62 ++++++---
>  libavcodec/decode.h        |   7 +
>  libavcodec/internal.h      |   7 +
>  libavcodec/pthread_frame.c | 279 ++++++++++++++++++++++++-------------
>  libavcodec/thread.h        |  18 +--
>  5 files changed, 247 insertions(+), 126 deletions(-)
> 
> diff --git a/libavcodec/decode.c b/libavcodec/decode.c
> index 3e5be501b9..5d2b484b66 100644
> --- a/libavcodec/decode.c
> +++ b/libavcodec/decode.c
> @@ -180,6 +180,11 @@ fail:
>      return ret;
>  }
>  
> +#if !HAVE_THREADS
> +#define ff_thread_get_packet(avctx, pkt) (AVERROR_BUG)
> +#define ff_thread_receive_frame(avctx, frame) (AVERROR_BUG)
> +#endif
> +
>  int ff_decode_get_packet(AVCodecContext *avctx, AVPacket *pkt)
>  {
>      AVCodecInternal *avci = avctx->internal;
> @@ -188,7 +193,14 @@ int ff_decode_get_packet(AVCodecContext *avctx, AVPacket *pkt)
>      if (avci->draining)
>          return AVERROR_EOF;
>  
> -    ret = av_bsf_receive_packet(avci->bsf, pkt);
> +    /* If we are a worker thread, get the next packet from the threading
> +     * context. Otherwise we are the main (user-facing) context, so we get the
> +     * next packet from the input filterchain.
> +     */
> +    if (avctx->internal->is_frame_mt)
> +        ret = ff_thread_get_packet(avctx, pkt);
> +    else
> +        ret = av_bsf_receive_packet(avci->bsf, pkt);
>      if (ret == AVERROR_EOF)
>          avci->draining = 1;
>      if (ret < 0)
> @@ -273,30 +285,25 @@ static inline int decode_simple_internal(AVCodecContext *avctx, AVFrame *frame,
>          return AVERROR_EOF;
>  
>      if (!pkt->data &&
> -        !(avctx->codec->capabilities & AV_CODEC_CAP_DELAY ||
> -          avctx->active_thread_type & FF_THREAD_FRAME))
> +        !(avctx->codec->capabilities & AV_CODEC_CAP_DELAY))
>          return AVERROR_EOF;
>  
>      got_frame = 0;
>  
> -    if (HAVE_THREADS && avctx->active_thread_type & FF_THREAD_FRAME) {
> -        ret = ff_thread_decode_frame(avctx, frame, &got_frame, pkt);
> -    } else {
> -        ret = codec->cb.decode(avctx, frame, &got_frame, pkt);
> -
> -        if (!(codec->caps_internal & FF_CODEC_CAP_SETS_PKT_DTS))
> -            frame->pkt_dts = pkt->dts;
> -        if (avctx->codec->type == AVMEDIA_TYPE_VIDEO) {
> -            if(!avctx->has_b_frames)
> -                frame->pkt_pos = pkt->pos;
> -            //FIXME these should be under if(!avctx->has_b_frames)
> -            /* get_buffer is supposed to set frame parameters */
> -            if (!(avctx->codec->capabilities & AV_CODEC_CAP_DR1)) {
> -                if (!frame->sample_aspect_ratio.num)  frame->sample_aspect_ratio = avctx->sample_aspect_ratio;
> -                if (!frame->width)                    frame->width               = avctx->width;
> -                if (!frame->height)                   frame->height              = avctx->height;
> -                if (frame->format == AV_PIX_FMT_NONE) frame->format              = avctx->pix_fmt;
> -            }
> +    ret = codec->cb.decode(avctx, frame, &got_frame, pkt);
> +
> +    if (!(codec->caps_internal & FF_CODEC_CAP_SETS_PKT_DTS))
> +        frame->pkt_dts = pkt->dts;
> +    if (avctx->codec->type == AVMEDIA_TYPE_VIDEO) {
> +        if(!avctx->has_b_frames)
> +            frame->pkt_pos = pkt->pos;
> +        //FIXME these should be under if(!avctx->has_b_frames)
> +        /* get_buffer is supposed to set frame parameters */
> +        if (!(avctx->codec->capabilities & AV_CODEC_CAP_DR1)) {
> +            if (!frame->sample_aspect_ratio.num)  frame->sample_aspect_ratio = avctx->sample_aspect_ratio;
> +            if (!frame->width)                    frame->width               = avctx->width;
> +            if (!frame->height)                   frame->height              = avctx->height;
> +            if (frame->format == AV_PIX_FMT_NONE) frame->format              = avctx->pix_fmt;
>          }
>      }
>      emms_c();
> @@ -546,7 +553,7 @@ static int decode_simple_receive_frame(AVCodecContext *avctx, AVFrame *frame)
>      return 0;
>  }
>  
> -static int decode_receive_frame_internal(AVCodecContext *avctx, AVFrame *frame)
> +int ff_decode_receive_frame_internal(AVCodecContext *avctx, AVFrame *frame)
>  {
>      AVCodecInternal *avci = avctx->internal;
>      const FFCodec *const codec = ffcodec(avctx->codec);
> @@ -569,6 +576,17 @@ static int decode_receive_frame_internal(AVCodecContext *avctx, AVFrame *frame)
>          return ok;
>      }

You keep detect_colorspace() (the ICC stuff) in the worker threads; this
means that more cmsContexts need to be allocated. Given that the
cmdContexts are not synced between threads, this also might possibly
mean that the output depends upon the number of threads (depending upon
how state based fflcms2 is). Is there a downside to putting this besides
the FrameDecodeData post-processing in the main thread?

>  
> +    return ret;
> +}
> +
> +static int decode_receive_frame_internal(AVCodecContext *avctx, AVFrame *frame)
> +{
> +    int ret;
> +    if (avctx->active_thread_type & FF_THREAD_FRAME)
> +        ret = ff_thread_receive_frame(avctx, frame);
> +    else
> +        ret = ff_decode_receive_frame_internal(avctx, frame);
> +
>      if (!ret) {
>          frame->best_effort_timestamp = guess_correct_pts(avctx,
>                                                           frame->pts,
> diff --git a/libavcodec/decode.h b/libavcodec/decode.h
> index 906122b4a7..7ba8e3a332 100644
> --- a/libavcodec/decode.h
> +++ b/libavcodec/decode.h
> @@ -58,6 +58,13 @@ typedef struct FrameDecodeData {
>   */
>  int ff_decode_receive_frame(AVCodecContext *avctx, AVFrame *frame);
>  
> +/**
> + * Do the actual decoding and obtain a decoded frame from the decoder, if
> + * available. When frame threading is used, this is invoked by the worker
> + * threads, otherwise by the top layer directly.
> + */
> +int ff_decode_receive_frame_internal(AVCodecContext *avctx, AVFrame *frame);
> +
>  /**
>   * Called by decoders to get the next packet for decoding.
>   *
> diff --git a/libavcodec/internal.h b/libavcodec/internal.h
> index a283c52e01..c87036efc4 100644
> --- a/libavcodec/internal.h
> +++ b/libavcodec/internal.h
> @@ -56,6 +56,13 @@ typedef struct AVCodecInternal {
>       */
>      int is_copy;
>  
> +    /**
> +     * This field is set to 1 when frame threading is being used and the parent
> +     * AVCodecContext of this AVCodecInternal is a worker-thread context (i.e.
> +     * one of those actually doing the decoding), 0 otherwise.
> +     */
> +    int is_frame_mt;
> +
>      /**
>       * An audio frame with less than required samples has been submitted (and
>       * potentially padded with silence). Reject all subsequent frames.
> diff --git a/libavcodec/pthread_frame.c b/libavcodec/pthread_frame.c
> index 62a0b18a8a..d8182cb4b8 100644
> --- a/libavcodec/pthread_frame.c
> +++ b/libavcodec/pthread_frame.c
> @@ -46,6 +46,7 @@
>  #include "libavutil/log.h"
>  #include "libavutil/mem.h"
>  #include "libavutil/opt.h"
> +#include "libavutil/fifo.h"
>  #include "libavutil/thread.h"
>  
>  enum {
> @@ -73,6 +74,12 @@ enum {
>      INITIALIZED,    ///< Thread has been properly set up
>  };
>  
> +typedef struct DecodedFrames {
> +    AVFrame  **f;
> +    size_t  nb_f;
> +    size_t  nb_f_allocated;
> +} DecodedFrames;
> +
>  /**
>   * Context used by codec threads and stored in their AVCodecInternal thread_ctx.
>   */
> @@ -93,8 +100,10 @@ typedef struct PerThreadContext {
>  
>      AVPacket       *avpkt;          ///< Input packet (for decoding) or output (for encoding).
>  
> -    AVFrame *frame;                 ///< Output frame (for decoding) or input (for encoding).
> -    int     got_frame;              ///< The output of got_picture_ptr from the last avcodec_decode_video() call.
> +    /**
> +     * Decoded frames from a single decode iteration.
> +     */
> +    DecodedFrames df;
>      int     result;                 ///< The result of the last codec decode/encode() call.
>  
>      atomic_int state;
> @@ -141,6 +150,14 @@ typedef struct FrameThreadContext {
>      pthread_cond_t async_cond;
>      int async_lock;
>  
> +    DecodedFrames df;
> +    int result;
> +
> +    /**
> +     * Packet to be submitted to the next thread for decoding.
> +     */
> +    AVPacket *next_pkt;
> +
>      int next_decoding;             ///< The next context to submit a packet to.
>      int next_finished;             ///< The next context to return output from.
>  
> @@ -190,6 +207,51 @@ static void thread_set_name(PerThreadContext *p)
>      ff_thread_setname(name);
>  }
>  
> +// get a free frame to decode into
> +static AVFrame *decoded_frames_get_free(DecodedFrames *df)
> +{
> +    if (df->nb_f == df->nb_f_allocated) {
> +        AVFrame **tmp = av_realloc_array(df->f, df->nb_f + 1,
> +                                         sizeof(*df->f));
> +        if (!tmp)
> +            return NULL;
> +        df->f = tmp;
> +
> +        df->f[df->nb_f] = av_frame_alloc();
> +        if (!df->f[df->nb_f])
> +            return NULL;
> +
> +        df->nb_f_allocated++;
> +    }
> +
> +    av_frame_unref(df->f[df->nb_f]);

Instead of unrefing here you should ensure that the frames corresponding
to indices nb_f..nb_f_allocated-1 are always blank.

> +    return df->f[df->nb_f];
> +}
> +
> +static void decoded_frames_pop(DecodedFrames *df, AVFrame *dst)
> +{
> +    AVFrame *tmp_frame = df->f[0];
> +    av_frame_move_ref(dst, tmp_frame);
> +    memmove(df->f, df->f + 1, (df->nb_f - 1) * sizeof(*df->f));
> +    df->f[--df->nb_f] = tmp_frame;
> +}
> +
> +static void decoded_frames_flush(DecodedFrames *df)
> +{
> +    for (int i = 0; i < df->nb_f; i++)

nb_f is a size_t; the iterator or nb_f should be made to match, one way
or another.
Same below.

> +        av_frame_unref(df->f[i]);
> +    df->nb_f = 0;
> +}
> +
> +static void decoded_frames_free(DecodedFrames *df)
> +{
> +    for (int i = 0; i < df->nb_f_allocated; i++)
> +        av_frame_free(&df->f[i]);
> +    av_freep(&df->f);
> +    df->nb_f           = 0;
> +    df->nb_f_allocated = 0;
> +}
> +
>  /**
>   * Codec worker thread.
>   *
> @@ -202,6 +264,7 @@ static attribute_align_arg void *frame_worker_thread(void *arg)
>      PerThreadContext *p = arg;
>      AVCodecContext *avctx = p->avctx;
>      const FFCodec *codec = ffcodec(avctx->codec);
> +    int ret;
>  
>      thread_set_name(p);
>  
> @@ -236,16 +299,31 @@ FF_ENABLE_DEPRECATION_WARNINGS
>              p->hwaccel_serializing = 1;
>          }
>  
> -        av_frame_unref(p->frame);
> -        p->got_frame = 0;
> -        p->result = codec->cb.decode(avctx, p->frame, &p->got_frame, p->avpkt);
> +        ret = 0;
> +        while (ret >= 0) {
> +            AVFrame *frame;
>  
> -        if ((p->result < 0 || !p->got_frame) && p->frame->buf[0])
> -            ff_thread_release_buffer(avctx, p->frame);
> +            /* get the frame which will store the output */
> +            frame = decoded_frames_get_free(&p->df);
> +            if (!frame) {
> +                p->result = AVERROR(ENOMEM);
> +                goto alloc_fail;
> +            }
> +
> +            /* do the actual decoding */
> +            ret = ff_decode_receive_frame_internal(avctx, frame);
> +            if (ret == 0)
> +                p->df.nb_f++;
> +            else if (ret < 0 && frame->buf[0])
> +                ff_thread_release_buffer(avctx, frame);
> +
> +            p->result = (ret == AVERROR(EAGAIN)) ? 0 : ret;
> +        }
>  
>          if (atomic_load(&p->state) == STATE_SETTING_UP)
>              ff_thread_finish_setup(avctx);
>  
> +alloc_fail:
>          if (p->hwaccel_serializing) {
>              /* wipe hwaccel state to avoid stale pointers lying around;
>               * the state was transferred to FrameThreadContext in
> @@ -441,23 +519,25 @@ static void release_delayed_buffers(PerThreadContext *p)
>  #endif
>  
>  static int submit_packet(PerThreadContext *p, AVCodecContext *user_avctx,
> -                         AVPacket *avpkt)
> +                         AVPacket *in_pkt)
>  {
>      FrameThreadContext *fctx = p->parent;
>      PerThreadContext *prev_thread = fctx->prev_thread;
> -    const AVCodec *codec = p->avctx->codec;
> -    int ret;
> -
> -    if (!avpkt->size && !(codec->capabilities & AV_CODEC_CAP_DELAY))
> -        return 0;
> +    int err;
>  
>      pthread_mutex_lock(&p->mutex);
>  
> -    ret = update_context_from_user(p->avctx, user_avctx);
> -    if (ret) {
> +    av_packet_unref(p->avpkt);
> +    av_packet_move_ref(p->avpkt, in_pkt);
> +    if (!p->avpkt->size)
> +        p->avctx->internal->draining = 1;
> +
> +    err = update_context_from_user(p->avctx, user_avctx);
> +    if (err < 0) {
>          pthread_mutex_unlock(&p->mutex);
> -        return ret;
> +        return err;
>      }
> +
>      atomic_store_explicit(&p->debug_threads,
>                            (p->avctx->debug & FF_DEBUG_THREADS) != 0,
>                            memory_order_relaxed);
> @@ -467,7 +547,6 @@ static int submit_packet(PerThreadContext *p, AVCodecContext *user_avctx,
>  #endif
>  
>      if (prev_thread) {
> -        int err;
>          if (atomic_load(&prev_thread->state) == STATE_SETTING_UP) {
>              pthread_mutex_lock(&prev_thread->progress_mutex);
>              while (atomic_load(&prev_thread->state) == STATE_SETTING_UP)
> @@ -475,10 +554,16 @@ static int submit_packet(PerThreadContext *p, AVCodecContext *user_avctx,
>              pthread_mutex_unlock(&prev_thread->progress_mutex);
>          }
>  
> -        err = update_context_from_thread(p->avctx, prev_thread->avctx, 0);
> -        if (err) {
> -            pthread_mutex_unlock(&p->mutex);
> -            return err;
> +        /* codecs without delay might not be prepared to be called repeatedly here during
> +         * flushing (vp3/theora), and also don't need to be, since from this point on, they
> +         * will always return EOF anyway */
> +        if (!p->avctx->internal->draining || (p->avctx->codec->capabilities & AV_CODEC_CAP_DELAY))
> +        {
> +            err = update_context_from_thread(p->avctx, prev_thread->avctx, 0);
> +            if (err) {
> +                pthread_mutex_unlock(&p->mutex);
> +                return err;
> +            }
>          }
>      }
>  
> @@ -488,14 +573,6 @@ static int submit_packet(PerThreadContext *p, AVCodecContext *user_avctx,
>      FFSWAP(void*,            p->avctx->hwaccel_context,             fctx->stash_hwaccel_context);
>      FFSWAP(void*,            p->avctx->internal->hwaccel_priv_data, fctx->stash_hwaccel_priv);
>  
> -    av_packet_unref(p->avpkt);
> -    ret = av_packet_ref(p->avpkt, avpkt);
> -    if (ret < 0) {
> -        pthread_mutex_unlock(&p->mutex);
> -        av_log(p->avctx, AV_LOG_ERROR, "av_packet_ref() failed in submit_packet()\n");
> -        return ret;
> -    }
> -
>      atomic_store(&p->state, STATE_SETTING_UP);
>      pthread_cond_signal(&p->input_cond);
>      pthread_mutex_unlock(&p->mutex);
> @@ -539,57 +616,42 @@ FF_ENABLE_DEPRECATION_WARNINGS
>  #endif
>  
>      fctx->prev_thread = p;
> -    fctx->next_decoding++;
> +    fctx->next_decoding = (fctx->next_decoding + 1) % p->avctx->thread_count;
>  
>      return 0;
>  }
>  
> -int ff_thread_decode_frame(AVCodecContext *avctx,
> -                           AVFrame *picture, int *got_picture_ptr,
> -                           AVPacket *avpkt)
> +int ff_thread_receive_frame(AVCodecContext *avctx, AVFrame *frame)
>  {
>      FrameThreadContext *fctx = avctx->internal->thread_ctx;
> -    int finished = fctx->next_finished;
> -    PerThreadContext *p;
> -    int err;
> +    int ret = 0;
>  
>      /* release the async lock, permitting blocked hwaccel threads to
>       * go forward while we are in this function */
>      async_unlock(fctx);
>  
> -    /*
> -     * Submit a packet to the next decoding thread.
> -     */
> +    /* submit packets to threads while there are no buffered results to return */
> +    while (!fctx->df.nb_f && !fctx->result) {
> +        PerThreadContext *p;
>  
> -    p = &fctx->threads[fctx->next_decoding];
> -    err = submit_packet(p, avctx, avpkt);
> -    if (err)
> -        goto finish;
> -
> -    /*
> -     * If we're still receiving the initial packets, don't return a frame.
> -     */
> -
> -    if (fctx->next_decoding > (avctx->thread_count-1-(avctx->codec_id == AV_CODEC_ID_FFV1)))
> -        fctx->delaying = 0;
> +        /* get a packet to be submitted to the next thread */
> +        av_packet_unref(fctx->next_pkt);
> +        ret = ff_decode_get_packet(avctx, fctx->next_pkt);
> +        if (ret < 0 && ret != AVERROR_EOF)
> +            goto finish;
>  
> -    if (fctx->delaying) {
> -        *got_picture_ptr=0;
> -        if (avpkt->size) {
> -            err = avpkt->size;
> +        ret = submit_packet(&fctx->threads[fctx->next_decoding], avctx,
> +                            fctx->next_pkt);
> +        if (ret < 0)
>              goto finish;
> -        }
> -    }
>  
> -    /*
> -     * Return the next available frame from the oldest thread.
> -     * If we're at the end of the stream, then we have to skip threads that
> -     * didn't output a frame/error, because we don't want to accidentally signal
> -     * EOF (avpkt->size == 0 && *got_picture_ptr == 0 && err >= 0).
> -     */
> +        /* do not return any frames until all threads have something to do */
> +        if (fctx->next_decoding != fctx->next_finished &&
> +            !avctx->internal->draining)
> +            continue;
>  
> -    do {
> -        p = &fctx->threads[finished++];
> +        p                   = &fctx->threads[fctx->next_finished];
> +        fctx->next_finished = (fctx->next_finished + 1) % avctx->thread_count;
>  
>          if (atomic_load(&p->state) != STATE_INPUT_READY) {
>              pthread_mutex_lock(&p->progress_mutex);
> @@ -598,35 +660,28 @@ int ff_thread_decode_frame(AVCodecContext *avctx,
>              pthread_mutex_unlock(&p->progress_mutex);
>          }
>  
> -        av_frame_move_ref(picture, p->frame);
> -        *got_picture_ptr = p->got_frame;
> -        picture->pkt_dts = p->avpkt->dts;
> -        err = p->result;
> -
> -        /*
> -         * A later call with avkpt->size == 0 may loop over all threads,
> -         * including this one, searching for a frame/error to return before being
> -         * stopped by the "finished != fctx->next_finished" condition.
> -         * Make sure we don't mistakenly return the same frame/error again.
> -         */
> -        p->got_frame = 0;
> -        p->result = 0;
> -
> -        if (finished >= avctx->thread_count) finished = 0;
> -    } while (!avpkt->size && !*got_picture_ptr && err >= 0 && finished != fctx->next_finished);
> +        update_context_from_thread(avctx, p->avctx, 1);
>  
> -    update_context_from_thread(avctx, p->avctx, 1);
> +        fctx->result = p->result;
> +        p->result    = 0;
>  
> -    if (fctx->next_decoding >= avctx->thread_count) fctx->next_decoding = 0;
> +        if (p->df.nb_f)
> +            FFSWAP(DecodedFrames, fctx->df, p->df);
> +    }
>  
> -    fctx->next_finished = finished;
> +    /* a thread may return multiple frames AND an error
> +     * we first return all the frames, then the error */
> +    if (fctx->df.nb_f) {
> +        decoded_frames_pop(&fctx->df, frame);
> +        ret = 0;
> +    } else {
> +        ret = fctx->result;
> +        fctx->result = 0;
> +    }
>  
> -    /* return the size of the consumed packet if no error occurred */
> -    if (err >= 0)
> -        err = avpkt->size;
>  finish:
>      async_lock(fctx);
> -    return err;
> +    return ret;
>  }
>  
>  void ff_thread_report_progress(ThreadFrame *f, int n, int field)
> @@ -726,7 +781,6 @@ static void park_frame_worker_threads(FrameThreadContext *fctx, int thread_count
>                  pthread_cond_wait(&p->output_cond, &p->progress_mutex);
>              pthread_mutex_unlock(&p->progress_mutex);
>          }
> -        p->got_frame = 0;
>      }
>  
>      async_lock(fctx);
> @@ -784,11 +838,12 @@ void ff_frame_thread_free(AVCodecContext *avctx, int thread_count)
>  
>              av_buffer_unref(&ctx->internal->pool);
>              av_packet_free(&ctx->internal->last_pkt_props);
> +            av_packet_free(&ctx->internal->in_pkt);
>              av_freep(&ctx->internal);
>              av_buffer_unref(&ctx->hw_frames_ctx);
>          }
>  
> -        av_frame_free(&p->frame);
> +        decoded_frames_free(&p->df);
>  
>          ff_pthread_free(p, per_thread_offsets);
>          av_packet_free(&p->avpkt);
> @@ -796,6 +851,9 @@ void ff_frame_thread_free(AVCodecContext *avctx, int thread_count)
>          av_freep(&p->avctx);
>      }
>  
> +    decoded_frames_free(&fctx->df);
> +    av_packet_free(&fctx->next_pkt);
> +
>      av_freep(&fctx->threads);
>      ff_pthread_free(fctx, thread_ctx_offsets);
>  
> @@ -854,13 +912,17 @@ static av_cold int init_thread(PerThreadContext *p, int *threads_to_free,
>      if (err < 0)
>          return err;
>  
> -    if (!(p->frame = av_frame_alloc()) ||
> -        !(p->avpkt = av_packet_alloc()))
> +    if (!(p->avpkt = av_packet_alloc()))
>          return AVERROR(ENOMEM);
>  
> +    copy->internal->is_frame_mt = 1;
>      if (!first)
>          copy->internal->is_copy = 1;
>  
> +    copy->internal->in_pkt = av_packet_alloc();
> +    if (!copy->internal->in_pkt)
> +        return AVERROR(ENOMEM);
> +
>      copy->internal->last_pkt_props = av_packet_alloc();
>      if (!copy->internal->last_pkt_props)
>          return AVERROR(ENOMEM);
> @@ -920,6 +982,10 @@ int ff_frame_thread_init(AVCodecContext *avctx)
>          return err;
>      }
>  
> +    fctx->next_pkt = av_packet_alloc();
> +    if (!fctx->next_pkt)
> +        return AVERROR(ENOMEM);

Did you test whether cleanup works when this allocation fails? I doubt
it does. You are entering ff_frame_thread_free() here with thread_count
> 0, yet you have not initialized a single worker thread.

> +
>      fctx->async_lock = 1;
>      fctx->delaying = 1;
>  
> @@ -964,17 +1030,28 @@ void ff_thread_flush(AVCodecContext *avctx)
>      fctx->next_decoding = fctx->next_finished = 0;
>      fctx->delaying = 1;
>      fctx->prev_thread = NULL;
> +
> +    decoded_frames_flush(&fctx->df);
> +
>      for (i = 0; i < avctx->thread_count; i++) {
>          PerThreadContext *p = &fctx->threads[i];
> -        // Make sure decode flush calls with size=0 won't return old frames
> -        p->got_frame = 0;
> -        av_frame_unref(p->frame);
> -        p->result = 0;
> +
> +        decoded_frames_flush(&p->df);
>  
>  #if FF_API_THREAD_SAFE_CALLBACKS
>          release_delayed_buffers(p);
>  #endif
>  
> +        av_packet_unref(p->avctx->internal->last_pkt_props);
> +        av_packet_unref(p->avctx->internal->in_pkt);
> +
> +        p->avctx->pts_correction_last_pts =
> +        p->avctx->pts_correction_last_dts = INT64_MIN;
> +
> +        p->avctx->internal->draining = 0;
> +        p->avctx->internal->draining_done = 0;
> +        p->avctx->internal->nb_draining_errors = 0;
> +
>          if (ffcodec(avctx->codec)->flush)
>              ffcodec(avctx->codec)->flush(p->avctx);
>      }
> @@ -1193,3 +1270,15 @@ void ff_thread_release_ext_buffer(AVCodecContext *avctx, ThreadFrame *f)
>      f->owner[0] = f->owner[1] = NULL;
>      ff_thread_release_buffer(avctx, f->f);
>  }
> +
> +int ff_thread_get_packet(AVCodecContext *avctx, AVPacket *pkt)
> +{
> +    PerThreadContext *p = avctx->internal->thread_ctx;
> +
> +    if (p->avpkt->buf) {
> +        av_packet_move_ref(pkt, p->avpkt);
> +        return 0;
> +    }
> +
> +    return avctx->internal->draining ? AVERROR_EOF : AVERROR(EAGAIN);
> +}
> diff --git a/libavcodec/thread.h b/libavcodec/thread.h
> index d5673f25ea..7ae69990fb 100644
> --- a/libavcodec/thread.h
> +++ b/libavcodec/thread.h
> @@ -40,17 +40,12 @@
>  void ff_thread_flush(AVCodecContext *avctx);
>  
>  /**
> - * Submit a new frame to a decoding thread.
> - * Returns the next available frame in picture. *got_picture_ptr
> - * will be 0 if none is available.
> - * The return value on success is the size of the consumed packet for
> - * compatibility with FFCodec.decode. This means the decoder
> - * has to consume the full packet.
> + * Submit available packets for decoding to worker threads, return a
> + * decoded frame if available. Returns AVERROR(EAGAIN) if none is available.
>   *
> - * Parameters are the same as FFCodec.decode.
> + * Parameters are the same as FFCodec.receive_frame.
>   */
> -int ff_thread_decode_frame(AVCodecContext *avctx, AVFrame *picture,
> -                           int *got_picture_ptr, AVPacket *avpkt);
> +int ff_thread_receive_frame(AVCodecContext *avctx, AVFrame *frame);
>  
>  /**
>   * If the codec defines update_thread_context(), call this
> @@ -99,6 +94,11 @@ int ff_thread_get_buffer(AVCodecContext *avctx, AVFrame *f, int flags);
>   */
>  void ff_thread_release_buffer(AVCodecContext *avctx, AVFrame *f);
>  
> +/**
> + * Get a packet for decoding. This gets invoked by the worker threads.
> + */
> +int ff_thread_get_packet(AVCodecContext *avctx, AVPacket *pkt);
> +
>  int ff_thread_init(AVCodecContext *s);
>  int ff_slice_thread_execute_with_mainfunc(AVCodecContext *avctx,
>          int (*action_func2)(AVCodecContext *c, void *arg, int jobnr, int threadnr),

Generally, I don't see the advantage of this. The only decoder that
would benefit from it is SMVJPEG, yet for all other codecs it will just
involve more allocs+frees.

- Andreas

On 20.12.2022 12:45, Andreas Rheinhardt wrote:
> Timo Rothenpieler:
>> From: Anton Khirnov <anton@khirnov.net>
>>
>> Reorganize the code such that the frame threading code does not call the
>> decoders directly, but instead calls back into the generic decoding
>> code. This avoids duplicating the logic that wraps the decoder
>> invocation and will be useful in the following commits.
>> ---
>>   libavcodec/decode.c        |  62 ++++++---
>>   libavcodec/decode.h        |   7 +
>>   libavcodec/internal.h      |   7 +
>>   libavcodec/pthread_frame.c | 279 ++++++++++++++++++++++++-------------
>>   libavcodec/thread.h        |  18 +--
>>   5 files changed, 247 insertions(+), 126 deletions(-)
>>
>> diff --git a/libavcodec/decode.c b/libavcodec/decode.c
>> index 3e5be501b9..5d2b484b66 100644
>> --- a/libavcodec/decode.c
>> +++ b/libavcodec/decode.c
>> @@ -180,6 +180,11 @@ fail:
>>       return ret;
>>   }
>>   
>> +#if !HAVE_THREADS
>> +#define ff_thread_get_packet(avctx, pkt) (AVERROR_BUG)
>> +#define ff_thread_receive_frame(avctx, frame) (AVERROR_BUG)
>> +#endif
>> +
>>   int ff_decode_get_packet(AVCodecContext *avctx, AVPacket *pkt)
>>   {
>>       AVCodecInternal *avci = avctx->internal;
>> @@ -188,7 +193,14 @@ int ff_decode_get_packet(AVCodecContext *avctx, AVPacket *pkt)
>>       if (avci->draining)
>>           return AVERROR_EOF;
>>   
>> -    ret = av_bsf_receive_packet(avci->bsf, pkt);
>> +    /* If we are a worker thread, get the next packet from the threading
>> +     * context. Otherwise we are the main (user-facing) context, so we get the
>> +     * next packet from the input filterchain.
>> +     */
>> +    if (avctx->internal->is_frame_mt)
>> +        ret = ff_thread_get_packet(avctx, pkt);
>> +    else
>> +        ret = av_bsf_receive_packet(avci->bsf, pkt);
>>       if (ret == AVERROR_EOF)
>>           avci->draining = 1;
>>       if (ret < 0)
>> @@ -273,30 +285,25 @@ static inline int decode_simple_internal(AVCodecContext *avctx, AVFrame *frame,
>>           return AVERROR_EOF;
>>   
>>       if (!pkt->data &&
>> -        !(avctx->codec->capabilities & AV_CODEC_CAP_DELAY ||
>> -          avctx->active_thread_type & FF_THREAD_FRAME))
>> +        !(avctx->codec->capabilities & AV_CODEC_CAP_DELAY))
>>           return AVERROR_EOF;
>>   
>>       got_frame = 0;
>>   
>> -    if (HAVE_THREADS && avctx->active_thread_type & FF_THREAD_FRAME) {
>> -        ret = ff_thread_decode_frame(avctx, frame, &got_frame, pkt);
>> -    } else {
>> -        ret = codec->cb.decode(avctx, frame, &got_frame, pkt);
>> -
>> -        if (!(codec->caps_internal & FF_CODEC_CAP_SETS_PKT_DTS))
>> -            frame->pkt_dts = pkt->dts;
>> -        if (avctx->codec->type == AVMEDIA_TYPE_VIDEO) {
>> -            if(!avctx->has_b_frames)
>> -                frame->pkt_pos = pkt->pos;
>> -            //FIXME these should be under if(!avctx->has_b_frames)
>> -            /* get_buffer is supposed to set frame parameters */
>> -            if (!(avctx->codec->capabilities & AV_CODEC_CAP_DR1)) {
>> -                if (!frame->sample_aspect_ratio.num)  frame->sample_aspect_ratio = avctx->sample_aspect_ratio;
>> -                if (!frame->width)                    frame->width               = avctx->width;
>> -                if (!frame->height)                   frame->height              = avctx->height;
>> -                if (frame->format == AV_PIX_FMT_NONE) frame->format              = avctx->pix_fmt;
>> -            }
>> +    ret = codec->cb.decode(avctx, frame, &got_frame, pkt);
>> +
>> +    if (!(codec->caps_internal & FF_CODEC_CAP_SETS_PKT_DTS))
>> +        frame->pkt_dts = pkt->dts;
>> +    if (avctx->codec->type == AVMEDIA_TYPE_VIDEO) {
>> +        if(!avctx->has_b_frames)
>> +            frame->pkt_pos = pkt->pos;
>> +        //FIXME these should be under if(!avctx->has_b_frames)
>> +        /* get_buffer is supposed to set frame parameters */
>> +        if (!(avctx->codec->capabilities & AV_CODEC_CAP_DR1)) {
>> +            if (!frame->sample_aspect_ratio.num)  frame->sample_aspect_ratio = avctx->sample_aspect_ratio;
>> +            if (!frame->width)                    frame->width               = avctx->width;
>> +            if (!frame->height)                   frame->height              = avctx->height;
>> +            if (frame->format == AV_PIX_FMT_NONE) frame->format              = avctx->pix_fmt;
>>           }
>>       }
>>       emms_c();
>> @@ -546,7 +553,7 @@ static int decode_simple_receive_frame(AVCodecContext *avctx, AVFrame *frame)
>>       return 0;
>>   }
>>   
>> -static int decode_receive_frame_internal(AVCodecContext *avctx, AVFrame *frame)
>> +int ff_decode_receive_frame_internal(AVCodecContext *avctx, AVFrame *frame)
>>   {
>>       AVCodecInternal *avci = avctx->internal;
>>       const FFCodec *const codec = ffcodec(avctx->codec);
>> @@ -569,6 +576,17 @@ static int decode_receive_frame_internal(AVCodecContext *avctx, AVFrame *frame)
>>           return ok;
>>       }
> 
> You keep detect_colorspace() (the ICC stuff) in the worker threads; this
> means that more cmsContexts need to be allocated. Given that the
> cmdContexts are not synced between threads, this also might possibly
> mean that the output depends upon the number of threads (depending upon
> how state based fflcms2 is). Is there a downside to putting this besides
> the FrameDecodeData post-processing in the main thread?

Just to keep as much processing in the threads as possible.
The initial patch had even more stuff inside of the threads (everything, 
including the post processing), which actually broke some stuff.

If that's causing issues for even the colorspace detection, the worker 
thread can just call the callback directly as it used to, and all logic 
and functions here can be kept as it was.

>>   
>> +    return ret;
>> +}
>> +
>> +static int decode_receive_frame_internal(AVCodecContext *avctx, AVFrame *frame)
>> +{
>> +    int ret;
>> +    if (avctx->active_thread_type & FF_THREAD_FRAME)
>> +        ret = ff_thread_receive_frame(avctx, frame);
>> +    else
>> +        ret = ff_decode_receive_frame_internal(avctx, frame);
>> +
>>       if (!ret) {
>>           frame->best_effort_timestamp = guess_correct_pts(avctx,
>>                                                            frame->pts,
>> diff --git a/libavcodec/decode.h b/libavcodec/decode.h
>> index 906122b4a7..7ba8e3a332 100644
>> --- a/libavcodec/decode.h
>> +++ b/libavcodec/decode.h
>> @@ -58,6 +58,13 @@ typedef struct FrameDecodeData {
>>    */
>>   int ff_decode_receive_frame(AVCodecContext *avctx, AVFrame *frame);
>>   
>> +/**
>> + * Do the actual decoding and obtain a decoded frame from the decoder, if
>> + * available. When frame threading is used, this is invoked by the worker
>> + * threads, otherwise by the top layer directly.
>> + */
>> +int ff_decode_receive_frame_internal(AVCodecContext *avctx, AVFrame *frame);
>> +
>>   /**
>>    * Called by decoders to get the next packet for decoding.
>>    *
>> diff --git a/libavcodec/internal.h b/libavcodec/internal.h
>> index a283c52e01..c87036efc4 100644
>> --- a/libavcodec/internal.h
>> +++ b/libavcodec/internal.h
>> @@ -56,6 +56,13 @@ typedef struct AVCodecInternal {
>>        */
>>       int is_copy;
>>   
>> +    /**
>> +     * This field is set to 1 when frame threading is being used and the parent
>> +     * AVCodecContext of this AVCodecInternal is a worker-thread context (i.e.
>> +     * one of those actually doing the decoding), 0 otherwise.
>> +     */
>> +    int is_frame_mt;
>> +
>>       /**
>>        * An audio frame with less than required samples has been submitted (and
>>        * potentially padded with silence). Reject all subsequent frames.
>> diff --git a/libavcodec/pthread_frame.c b/libavcodec/pthread_frame.c
>> index 62a0b18a8a..d8182cb4b8 100644
>> --- a/libavcodec/pthread_frame.c
>> +++ b/libavcodec/pthread_frame.c
>> @@ -46,6 +46,7 @@
>>   #include "libavutil/log.h"
>>   #include "libavutil/mem.h"
>>   #include "libavutil/opt.h"
>> +#include "libavutil/fifo.h"
>>   #include "libavutil/thread.h"
>>   
>>   enum {
>> @@ -73,6 +74,12 @@ enum {
>>       INITIALIZED,    ///< Thread has been properly set up
>>   };
>>   
>> +typedef struct DecodedFrames {
>> +    AVFrame  **f;
>> +    size_t  nb_f;
>> +    size_t  nb_f_allocated;
>> +} DecodedFrames;
>> +
>>   /**
>>    * Context used by codec threads and stored in their AVCodecInternal thread_ctx.
>>    */
>> @@ -93,8 +100,10 @@ typedef struct PerThreadContext {
>>   
>>       AVPacket       *avpkt;          ///< Input packet (for decoding) or output (for encoding).
>>   
>> -    AVFrame *frame;                 ///< Output frame (for decoding) or input (for encoding).
>> -    int     got_frame;              ///< The output of got_picture_ptr from the last avcodec_decode_video() call.
>> +    /**
>> +     * Decoded frames from a single decode iteration.
>> +     */
>> +    DecodedFrames df;
>>       int     result;                 ///< The result of the last codec decode/encode() call.
>>   
>>       atomic_int state;
>> @@ -141,6 +150,14 @@ typedef struct FrameThreadContext {
>>       pthread_cond_t async_cond;
>>       int async_lock;
>>   
>> +    DecodedFrames df;
>> +    int result;
>> +
>> +    /**
>> +     * Packet to be submitted to the next thread for decoding.
>> +     */
>> +    AVPacket *next_pkt;
>> +
>>       int next_decoding;             ///< The next context to submit a packet to.
>>       int next_finished;             ///< The next context to return output from.
>>   
>> @@ -190,6 +207,51 @@ static void thread_set_name(PerThreadContext *p)
>>       ff_thread_setname(name);
>>   }
>>   
>> +// get a free frame to decode into
>> +static AVFrame *decoded_frames_get_free(DecodedFrames *df)
>> +{
>> +    if (df->nb_f == df->nb_f_allocated) {
>> +        AVFrame **tmp = av_realloc_array(df->f, df->nb_f + 1,
>> +                                         sizeof(*df->f));
>> +        if (!tmp)
>> +            return NULL;
>> +        df->f = tmp;
>> +
>> +        df->f[df->nb_f] = av_frame_alloc();
>> +        if (!df->f[df->nb_f])
>> +            return NULL;
>> +
>> +        df->nb_f_allocated++;
>> +    }
>> +
>> +    av_frame_unref(df->f[df->nb_f]);
> 
> Instead of unrefing here you should ensure that the frames corresponding
> to indices nb_f..nb_f_allocated-1 are always blank.
> 
>> +    return df->f[df->nb_f];
>> +}
>> +
>> +static void decoded_frames_pop(DecodedFrames *df, AVFrame *dst)
>> +{
>> +    AVFrame *tmp_frame = df->f[0];
>> +    av_frame_move_ref(dst, tmp_frame);
>> +    memmove(df->f, df->f + 1, (df->nb_f - 1) * sizeof(*df->f));
>> +    df->f[--df->nb_f] = tmp_frame;
>> +}
>> +
>> +static void decoded_frames_flush(DecodedFrames *df)
>> +{
>> +    for (int i = 0; i < df->nb_f; i++)
> 
> nb_f is a size_t; the iterator or nb_f should be made to match, one way
> or another.
> Same below.
> 
>> +        av_frame_unref(df->f[i]);
>> +    df->nb_f = 0;
>> +}
>> +
>> +static void decoded_frames_free(DecodedFrames *df)
>> +{
>> +    for (int i = 0; i < df->nb_f_allocated; i++)
>> +        av_frame_free(&df->f[i]);
>> +    av_freep(&df->f);
>> +    df->nb_f           = 0;
>> +    df->nb_f_allocated = 0;
>> +}
>> +
>>   /**
>>    * Codec worker thread.
>>    *
>> @@ -202,6 +264,7 @@ static attribute_align_arg void *frame_worker_thread(void *arg)
>>       PerThreadContext *p = arg;
>>       AVCodecContext *avctx = p->avctx;
>>       const FFCodec *codec = ffcodec(avctx->codec);
>> +    int ret;
>>   
>>       thread_set_name(p);
>>   
>> @@ -236,16 +299,31 @@ FF_ENABLE_DEPRECATION_WARNINGS
>>               p->hwaccel_serializing = 1;
>>           }
>>   
>> -        av_frame_unref(p->frame);
>> -        p->got_frame = 0;
>> -        p->result = codec->cb.decode(avctx, p->frame, &p->got_frame, p->avpkt);
>> +        ret = 0;
>> +        while (ret >= 0) {
>> +            AVFrame *frame;
>>   
>> -        if ((p->result < 0 || !p->got_frame) && p->frame->buf[0])
>> -            ff_thread_release_buffer(avctx, p->frame);
>> +            /* get the frame which will store the output */
>> +            frame = decoded_frames_get_free(&p->df);
>> +            if (!frame) {
>> +                p->result = AVERROR(ENOMEM);
>> +                goto alloc_fail;
>> +            }
>> +
>> +            /* do the actual decoding */
>> +            ret = ff_decode_receive_frame_internal(avctx, frame);
>> +            if (ret == 0)
>> +                p->df.nb_f++;
>> +            else if (ret < 0 && frame->buf[0])
>> +                ff_thread_release_buffer(avctx, frame);
>> +
>> +            p->result = (ret == AVERROR(EAGAIN)) ? 0 : ret;
>> +        }
>>   
>>           if (atomic_load(&p->state) == STATE_SETTING_UP)
>>               ff_thread_finish_setup(avctx);
>>   
>> +alloc_fail:
>>           if (p->hwaccel_serializing) {
>>               /* wipe hwaccel state to avoid stale pointers lying around;
>>                * the state was transferred to FrameThreadContext in
>> @@ -441,23 +519,25 @@ static void release_delayed_buffers(PerThreadContext *p)
>>   #endif
>>   
>>   static int submit_packet(PerThreadContext *p, AVCodecContext *user_avctx,
>> -                         AVPacket *avpkt)
>> +                         AVPacket *in_pkt)
>>   {
>>       FrameThreadContext *fctx = p->parent;
>>       PerThreadContext *prev_thread = fctx->prev_thread;
>> -    const AVCodec *codec = p->avctx->codec;
>> -    int ret;
>> -
>> -    if (!avpkt->size && !(codec->capabilities & AV_CODEC_CAP_DELAY))
>> -        return 0;
>> +    int err;
>>   
>>       pthread_mutex_lock(&p->mutex);
>>   
>> -    ret = update_context_from_user(p->avctx, user_avctx);
>> -    if (ret) {
>> +    av_packet_unref(p->avpkt);
>> +    av_packet_move_ref(p->avpkt, in_pkt);
>> +    if (!p->avpkt->size)
>> +        p->avctx->internal->draining = 1;
>> +
>> +    err = update_context_from_user(p->avctx, user_avctx);
>> +    if (err < 0) {
>>           pthread_mutex_unlock(&p->mutex);
>> -        return ret;
>> +        return err;
>>       }
>> +
>>       atomic_store_explicit(&p->debug_threads,
>>                             (p->avctx->debug & FF_DEBUG_THREADS) != 0,
>>                             memory_order_relaxed);
>> @@ -467,7 +547,6 @@ static int submit_packet(PerThreadContext *p, AVCodecContext *user_avctx,
>>   #endif
>>   
>>       if (prev_thread) {
>> -        int err;
>>           if (atomic_load(&prev_thread->state) == STATE_SETTING_UP) {
>>               pthread_mutex_lock(&prev_thread->progress_mutex);
>>               while (atomic_load(&prev_thread->state) == STATE_SETTING_UP)
>> @@ -475,10 +554,16 @@ static int submit_packet(PerThreadContext *p, AVCodecContext *user_avctx,
>>               pthread_mutex_unlock(&prev_thread->progress_mutex);
>>           }
>>   
>> -        err = update_context_from_thread(p->avctx, prev_thread->avctx, 0);
>> -        if (err) {
>> -            pthread_mutex_unlock(&p->mutex);
>> -            return err;
>> +        /* codecs without delay might not be prepared to be called repeatedly here during
>> +         * flushing (vp3/theora), and also don't need to be, since from this point on, they
>> +         * will always return EOF anyway */
>> +        if (!p->avctx->internal->draining || (p->avctx->codec->capabilities & AV_CODEC_CAP_DELAY))
>> +        {
>> +            err = update_context_from_thread(p->avctx, prev_thread->avctx, 0);
>> +            if (err) {
>> +                pthread_mutex_unlock(&p->mutex);
>> +                return err;
>> +            }
>>           }
>>       }
>>   
>> @@ -488,14 +573,6 @@ static int submit_packet(PerThreadContext *p, AVCodecContext *user_avctx,
>>       FFSWAP(void*,            p->avctx->hwaccel_context,             fctx->stash_hwaccel_context);
>>       FFSWAP(void*,            p->avctx->internal->hwaccel_priv_data, fctx->stash_hwaccel_priv);
>>   
>> -    av_packet_unref(p->avpkt);
>> -    ret = av_packet_ref(p->avpkt, avpkt);
>> -    if (ret < 0) {
>> -        pthread_mutex_unlock(&p->mutex);
>> -        av_log(p->avctx, AV_LOG_ERROR, "av_packet_ref() failed in submit_packet()\n");
>> -        return ret;
>> -    }
>> -
>>       atomic_store(&p->state, STATE_SETTING_UP);
>>       pthread_cond_signal(&p->input_cond);
>>       pthread_mutex_unlock(&p->mutex);
>> @@ -539,57 +616,42 @@ FF_ENABLE_DEPRECATION_WARNINGS
>>   #endif
>>   
>>       fctx->prev_thread = p;
>> -    fctx->next_decoding++;
>> +    fctx->next_decoding = (fctx->next_decoding + 1) % p->avctx->thread_count;
>>   
>>       return 0;
>>   }
>>   
>> -int ff_thread_decode_frame(AVCodecContext *avctx,
>> -                           AVFrame *picture, int *got_picture_ptr,
>> -                           AVPacket *avpkt)
>> +int ff_thread_receive_frame(AVCodecContext *avctx, AVFrame *frame)
>>   {
>>       FrameThreadContext *fctx = avctx->internal->thread_ctx;
>> -    int finished = fctx->next_finished;
>> -    PerThreadContext *p;
>> -    int err;
>> +    int ret = 0;
>>   
>>       /* release the async lock, permitting blocked hwaccel threads to
>>        * go forward while we are in this function */
>>       async_unlock(fctx);
>>   
>> -    /*
>> -     * Submit a packet to the next decoding thread.
>> -     */
>> +    /* submit packets to threads while there are no buffered results to return */
>> +    while (!fctx->df.nb_f && !fctx->result) {
>> +        PerThreadContext *p;
>>   
>> -    p = &fctx->threads[fctx->next_decoding];
>> -    err = submit_packet(p, avctx, avpkt);
>> -    if (err)
>> -        goto finish;
>> -
>> -    /*
>> -     * If we're still receiving the initial packets, don't return a frame.
>> -     */
>> -
>> -    if (fctx->next_decoding > (avctx->thread_count-1-(avctx->codec_id == AV_CODEC_ID_FFV1)))
>> -        fctx->delaying = 0;
>> +        /* get a packet to be submitted to the next thread */
>> +        av_packet_unref(fctx->next_pkt);
>> +        ret = ff_decode_get_packet(avctx, fctx->next_pkt);
>> +        if (ret < 0 && ret != AVERROR_EOF)
>> +            goto finish;
>>   
>> -    if (fctx->delaying) {
>> -        *got_picture_ptr=0;
>> -        if (avpkt->size) {
>> -            err = avpkt->size;
>> +        ret = submit_packet(&fctx->threads[fctx->next_decoding], avctx,
>> +                            fctx->next_pkt);
>> +        if (ret < 0)
>>               goto finish;
>> -        }
>> -    }
>>   
>> -    /*
>> -     * Return the next available frame from the oldest thread.
>> -     * If we're at the end of the stream, then we have to skip threads that
>> -     * didn't output a frame/error, because we don't want to accidentally signal
>> -     * EOF (avpkt->size == 0 && *got_picture_ptr == 0 && err >= 0).
>> -     */
>> +        /* do not return any frames until all threads have something to do */
>> +        if (fctx->next_decoding != fctx->next_finished &&
>> +            !avctx->internal->draining)
>> +            continue;
>>   
>> -    do {
>> -        p = &fctx->threads[finished++];
>> +        p                   = &fctx->threads[fctx->next_finished];
>> +        fctx->next_finished = (fctx->next_finished + 1) % avctx->thread_count;
>>   
>>           if (atomic_load(&p->state) != STATE_INPUT_READY) {
>>               pthread_mutex_lock(&p->progress_mutex);
>> @@ -598,35 +660,28 @@ int ff_thread_decode_frame(AVCodecContext *avctx,
>>               pthread_mutex_unlock(&p->progress_mutex);
>>           }
>>   
>> -        av_frame_move_ref(picture, p->frame);
>> -        *got_picture_ptr = p->got_frame;
>> -        picture->pkt_dts = p->avpkt->dts;
>> -        err = p->result;
>> -
>> -        /*
>> -         * A later call with avkpt->size == 0 may loop over all threads,
>> -         * including this one, searching for a frame/error to return before being
>> -         * stopped by the "finished != fctx->next_finished" condition.
>> -         * Make sure we don't mistakenly return the same frame/error again.
>> -         */
>> -        p->got_frame = 0;
>> -        p->result = 0;
>> -
>> -        if (finished >= avctx->thread_count) finished = 0;
>> -    } while (!avpkt->size && !*got_picture_ptr && err >= 0 && finished != fctx->next_finished);
>> +        update_context_from_thread(avctx, p->avctx, 1);
>>   
>> -    update_context_from_thread(avctx, p->avctx, 1);
>> +        fctx->result = p->result;
>> +        p->result    = 0;
>>   
>> -    if (fctx->next_decoding >= avctx->thread_count) fctx->next_decoding = 0;
>> +        if (p->df.nb_f)
>> +            FFSWAP(DecodedFrames, fctx->df, p->df);
>> +    }
>>   
>> -    fctx->next_finished = finished;
>> +    /* a thread may return multiple frames AND an error
>> +     * we first return all the frames, then the error */
>> +    if (fctx->df.nb_f) {
>> +        decoded_frames_pop(&fctx->df, frame);
>> +        ret = 0;
>> +    } else {
>> +        ret = fctx->result;
>> +        fctx->result = 0;
>> +    }
>>   
>> -    /* return the size of the consumed packet if no error occurred */
>> -    if (err >= 0)
>> -        err = avpkt->size;
>>   finish:
>>       async_lock(fctx);
>> -    return err;
>> +    return ret;
>>   }
>>   
>>   void ff_thread_report_progress(ThreadFrame *f, int n, int field)
>> @@ -726,7 +781,6 @@ static void park_frame_worker_threads(FrameThreadContext *fctx, int thread_count
>>                   pthread_cond_wait(&p->output_cond, &p->progress_mutex);
>>               pthread_mutex_unlock(&p->progress_mutex);
>>           }
>> -        p->got_frame = 0;
>>       }
>>   
>>       async_lock(fctx);
>> @@ -784,11 +838,12 @@ void ff_frame_thread_free(AVCodecContext *avctx, int thread_count)
>>   
>>               av_buffer_unref(&ctx->internal->pool);
>>               av_packet_free(&ctx->internal->last_pkt_props);
>> +            av_packet_free(&ctx->internal->in_pkt);
>>               av_freep(&ctx->internal);
>>               av_buffer_unref(&ctx->hw_frames_ctx);
>>           }
>>   
>> -        av_frame_free(&p->frame);
>> +        decoded_frames_free(&p->df);
>>   
>>           ff_pthread_free(p, per_thread_offsets);
>>           av_packet_free(&p->avpkt);
>> @@ -796,6 +851,9 @@ void ff_frame_thread_free(AVCodecContext *avctx, int thread_count)
>>           av_freep(&p->avctx);
>>       }
>>   
>> +    decoded_frames_free(&fctx->df);
>> +    av_packet_free(&fctx->next_pkt);
>> +
>>       av_freep(&fctx->threads);
>>       ff_pthread_free(fctx, thread_ctx_offsets);
>>   
>> @@ -854,13 +912,17 @@ static av_cold int init_thread(PerThreadContext *p, int *threads_to_free,
>>       if (err < 0)
>>           return err;
>>   
>> -    if (!(p->frame = av_frame_alloc()) ||
>> -        !(p->avpkt = av_packet_alloc()))
>> +    if (!(p->avpkt = av_packet_alloc()))
>>           return AVERROR(ENOMEM);
>>   
>> +    copy->internal->is_frame_mt = 1;
>>       if (!first)
>>           copy->internal->is_copy = 1;
>>   
>> +    copy->internal->in_pkt = av_packet_alloc();
>> +    if (!copy->internal->in_pkt)
>> +        return AVERROR(ENOMEM);
>> +
>>       copy->internal->last_pkt_props = av_packet_alloc();
>>       if (!copy->internal->last_pkt_props)
>>           return AVERROR(ENOMEM);
>> @@ -920,6 +982,10 @@ int ff_frame_thread_init(AVCodecContext *avctx)
>>           return err;
>>       }
>>   
>> +    fctx->next_pkt = av_packet_alloc();
>> +    if (!fctx->next_pkt)
>> +        return AVERROR(ENOMEM);
> 
> Did you test whether cleanup works when this allocation fails? I doubt
> it does. You are entering ff_frame_thread_free() here with thread_count
>> 0, yet you have not initialized a single worker thread.
> 
>> +
>>       fctx->async_lock = 1;
>>       fctx->delaying = 1;
>>   
>> @@ -964,17 +1030,28 @@ void ff_thread_flush(AVCodecContext *avctx)
>>       fctx->next_decoding = fctx->next_finished = 0;
>>       fctx->delaying = 1;
>>       fctx->prev_thread = NULL;
>> +
>> +    decoded_frames_flush(&fctx->df);
>> +
>>       for (i = 0; i < avctx->thread_count; i++) {
>>           PerThreadContext *p = &fctx->threads[i];
>> -        // Make sure decode flush calls with size=0 won't return old frames
>> -        p->got_frame = 0;
>> -        av_frame_unref(p->frame);
>> -        p->result = 0;
>> +
>> +        decoded_frames_flush(&p->df);
>>   
>>   #if FF_API_THREAD_SAFE_CALLBACKS
>>           release_delayed_buffers(p);
>>   #endif
>>   
>> +        av_packet_unref(p->avctx->internal->last_pkt_props);
>> +        av_packet_unref(p->avctx->internal->in_pkt);
>> +
>> +        p->avctx->pts_correction_last_pts =
>> +        p->avctx->pts_correction_last_dts = INT64_MIN;
>> +
>> +        p->avctx->internal->draining = 0;
>> +        p->avctx->internal->draining_done = 0;
>> +        p->avctx->internal->nb_draining_errors = 0;
>> +
>>           if (ffcodec(avctx->codec)->flush)
>>               ffcodec(avctx->codec)->flush(p->avctx);
>>       }
>> @@ -1193,3 +1270,15 @@ void ff_thread_release_ext_buffer(AVCodecContext *avctx, ThreadFrame *f)
>>       f->owner[0] = f->owner[1] = NULL;
>>       ff_thread_release_buffer(avctx, f->f);
>>   }
>> +
>> +int ff_thread_get_packet(AVCodecContext *avctx, AVPacket *pkt)
>> +{
>> +    PerThreadContext *p = avctx->internal->thread_ctx;
>> +
>> +    if (p->avpkt->buf) {
>> +        av_packet_move_ref(pkt, p->avpkt);
>> +        return 0;
>> +    }
>> +
>> +    return avctx->internal->draining ? AVERROR_EOF : AVERROR(EAGAIN);
>> +}
>> diff --git a/libavcodec/thread.h b/libavcodec/thread.h
>> index d5673f25ea..7ae69990fb 100644
>> --- a/libavcodec/thread.h
>> +++ b/libavcodec/thread.h
>> @@ -40,17 +40,12 @@
>>   void ff_thread_flush(AVCodecContext *avctx);
>>   
>>   /**
>> - * Submit a new frame to a decoding thread.
>> - * Returns the next available frame in picture. *got_picture_ptr
>> - * will be 0 if none is available.
>> - * The return value on success is the size of the consumed packet for
>> - * compatibility with FFCodec.decode. This means the decoder
>> - * has to consume the full packet.
>> + * Submit available packets for decoding to worker threads, return a
>> + * decoded frame if available. Returns AVERROR(EAGAIN) if none is available.
>>    *
>> - * Parameters are the same as FFCodec.decode.
>> + * Parameters are the same as FFCodec.receive_frame.
>>    */
>> -int ff_thread_decode_frame(AVCodecContext *avctx, AVFrame *picture,
>> -                           int *got_picture_ptr, AVPacket *avpkt);
>> +int ff_thread_receive_frame(AVCodecContext *avctx, AVFrame *frame);
>>   
>>   /**
>>    * If the codec defines update_thread_context(), call this
>> @@ -99,6 +94,11 @@ int ff_thread_get_buffer(AVCodecContext *avctx, AVFrame *f, int flags);
>>    */
>>   void ff_thread_release_buffer(AVCodecContext *avctx, AVFrame *f);
>>   
>> +/**
>> + * Get a packet for decoding. This gets invoked by the worker threads.
>> + */
>> +int ff_thread_get_packet(AVCodecContext *avctx, AVPacket *pkt);
>> +
>>   int ff_thread_init(AVCodecContext *s);
>>   int ff_slice_thread_execute_with_mainfunc(AVCodecContext *avctx,
>>           int (*action_func2)(AVCodecContext *c, void *arg, int jobnr, int threadnr),
> 
> Generally, I don't see the advantage of this. The only decoder that
> would benefit from it is SMVJPEG, yet for all other codecs it will just
> involve more allocs+frees.

Being unable to support newer encoders because only the old (and I 
thought deprecated, but apprently not, given encoders are getting ported 
back to it??) is a pretty bad situation, and could hold back some progress.

So adding the ability for frame threading to work for everything seems 
natural and neccesary to me.

On 12/20/2022 1:06 PM, Timo Rothenpieler wrote:
>>
>> Generally, I don't see the advantage of this. The only decoder that
>> would benefit from it is SMVJPEG, yet for all other codecs it will just
>> involve more allocs+frees.
> 
> Being unable to support newer encoders because only the old (and I 
> thought deprecated, but apprently not, given encoders are getting ported 
> back to it??) is a pretty bad situation, and could hold back some progress.

Decoders. And what was deprecated and removed was the public API, not 
the internal callback.
For 1:1 decoders, or decoders with no delay or buffering, the decoupled 
input/output callback is overkill, so you'll keep seeing new decoders 
added using the old one in the future.
But similarly, new decoders can and will be added using the decoupled IO 
callback, so this change is necessary so any future decoder that intends 
to have frame threading isn't forced to use the old callback when it 
could benefit from the new. An obvious case would be the (hopefully) 
eventual native AV1 decoder.

> 
> So adding the ability for frame threading to work for everything seems 
> natural and neccesary to me.

Yes.