On Wo, 2023-02-01 at 02:59 +0000, Xiang, Haihao wrote:
> On Di, 2023-01-31 at 14:14 +0100, Anton Khirnov wrote:
> > input_image is freed by the time the output packet is constructed, so we
> > need to store copies in VAAPIEncodePicture.
> > ---
> > Is this better?
> > ---
> > libavcodec/vaapi_encode.c | 22 +++++++++++++++++-----
> > libavcodec/vaapi_encode.h | 4 ++++
> > 2 files changed, 21 insertions(+), 5 deletions(-)
> >
> > diff --git a/libavcodec/vaapi_encode.c b/libavcodec/vaapi_encode.c
> > index 6787b90e8d6..bfca315a7ad 100644
> > --- a/libavcodec/vaapi_encode.c
> > +++ b/libavcodec/vaapi_encode.c
> > @@ -695,7 +695,7 @@ static int vaapi_encode_output(AVCodecContext *avctx,
> > pkt->flags |= AV_PKT_FLAG_KEY;
> >
> > pkt->pts = pic->pts;
> > - pkt->duration = pic->input_image->duration;
> > + pkt->duration = pic->duration;
> >
> > vas = vaUnmapBuffer(ctx->hwctx->display, pic->output_buffer);
> > if (vas != VA_STATUS_SUCCESS) {
> > @@ -706,10 +706,11 @@ static int vaapi_encode_output(AVCodecContext *avctx,
> > }
> >
> > // for no-delay encoders this is handled in generic codec
> > - if (avctx->codec->capabilities & AV_CODEC_CAP_DELAY) {
> > - err = ff_encode_reordered_opaque(avctx, pkt, pic->input_image);
> > - if (err < 0)
> > - goto fail;
> > + if (avctx->codec->capabilities & AV_CODEC_CAP_DELAY &&
> > + avctx->flags & AV_CODEC_FLAG_COPY_OPAQUE) {
> > + pkt->opaque = pic->opaque;
> > + pkt->opaque_ref = pic->opaque_ref;
> > + pic->opaque_ref = NULL;
> > }
> >
> > av_buffer_unref(&pic->output_buffer_ref);
> > @@ -785,6 +786,8 @@ static int vaapi_encode_free(AVCodecContext *avctx,
> > av_frame_free(&pic->input_image);
> > av_frame_free(&pic->recon_image);
> >
> > + av_buffer_unref(&pic->opaque_ref);
> > +
> > av_freep(&pic->param_buffers);
> > av_freep(&pic->slices);
> > // Output buffer should already be destroyed.
> > @@ -1152,6 +1155,15 @@ static int vaapi_encode_send_frame(AVCodecContext
> > *avctx, AVFrame *frame)
> >
> > pic->input_surface = (VASurfaceID)(uintptr_t)frame->data[3];
> > pic->pts = frame->pts;
> > + pic->duration = frame->duration;
> > +
> > + if (avctx->flags & AV_CODEC_FLAG_COPY_OPAQUE) {
> > + err = av_buffer_replace(&pic->opaque_ref, frame->opaque_ref);
> > + if (err < 0)
> > + goto fail;
> > +
> > + pic->opaque = frame->opaque;
> > + }
> >
> > av_frame_move_ref(pic->input_image, frame);
> >
> > diff --git a/libavcodec/vaapi_encode.h b/libavcodec/vaapi_encode.h
> > index 359f954ffff..a1e639f56b0 100644
> > --- a/libavcodec/vaapi_encode.h
> > +++ b/libavcodec/vaapi_encode.h
> > @@ -75,8 +75,12 @@ typedef struct VAAPIEncodePicture {
> > int64_t display_order;
> > int64_t encode_order;
> > int64_t pts;
> > + int64_t duration;
> > int force_idr;
> >
> > + void *opaque;
> > + AVBufferRef *opaque_ref;
> > +
> > #if VA_CHECK_VERSION(1, 0, 0)
> > // ROI regions.
> > VAEncROI *roi;
>
> LGTM and the command for transcoding with vaapi works well now.
>
Pushed because this patch fixed critical regression in vaapi path.
Thanks
Haihao
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
