[FFmpeg-devel] [PATCH] h264: fix reference reordering

[Zhao Zhili]

> On Mar 24, 2025, at 04:15, averne <averne381 at gmail.com> wrote:
> 
> Fixes hardware decoders depending on long_term_pic_idx.
> Relevant ITU conformance test vectors: MR6_BT_B, MR8_BT_B
> 
> ---
> According to the specification (8.2.4.3 Modification process for
> reference picture lists), the reference reordering process does not
> affect frame indices (PicNum for short refs and LongTermPicNum for long
> ones).
> 
> Hardware decoders commonly use this values in order to identify refs,
> so modifying the frame idx passed to the the hardware would throw off
> the decode engine.
> 
> Fixes: two ITU conformance test samples (MR6_BT_B and MR8_BT_B, found at
> https://www.itu.int/net/itu-t/sigdb/spevideo/VideoForm-s.aspx?val=102002641),
> on my WIP hwaccel backend (https://github.com/averne/FFmpeg/tree/envideo).
> Potentially fixes more hwaccels using this value (dxva2, vaapi, vdpau,
> vulkan).

The patch indeed fixed vdpau decoding of the two samples. Pushed, thanks!

Cc Lynne. Vulkan decoding still failed for the two samples. My machine is Nvidia
GeForce RTX 3060 with driver version 535.183.01.

> 
> This patch was verified to not cause any regression of the software
> decoder.
> 
> Signed-off-by: averne <averne381 at gmail.com>
> ---
> libavcodec/h264_refs.c | 5 ++---
> 1 file changed, 2 insertions(+), 3 deletions(-)
> 
> diff --git a/libavcodec/h264_refs.c b/libavcodec/h264_refs.c
> index 050ca9229218d68c1f1c5acb8fed8e24c61fd5a1..74840e5909212d6a750518acc416221ffe90f722 100644
> --- a/libavcodec/h264_refs.c
> +++ b/libavcodec/h264_refs.c
> @@ -337,7 +337,7 @@ int ff_h264_build_ref_list(H264Context *h, H264SliceContext *sl)
>                         break;
>                 }
>                 if (i >= 0)
> -                    ref->pic_id = pred;
> +                    pic_id = pred;
>                 break;
>             }
>             case 2: {
> @@ -354,7 +354,6 @@ int ff_h264_build_ref_list(H264Context *h, H264SliceContext *sl)
>                 ref = h->long_ref[long_idx];
>                 assert(!(ref && !ref->reference));
>                 if (ref && (ref->reference & pic_structure)) {
> -                    ref->pic_id = pic_id;
>                     assert(ref->long_ref);
>                     i = 0;
>                 } else {
> @@ -376,7 +375,7 @@ int ff_h264_build_ref_list(H264Context *h, H264SliceContext *sl)
>                 for (i = index; i + 1 < sl->ref_count[list]; i++) {
>                     if (sl->ref_list[list][i].parent &&
>                         ref->long_ref == sl->ref_list[list][i].parent->long_ref &&
> -                        ref->pic_id   == sl->ref_list[list][i].pic_id)
> +                        pic_id        == sl->ref_list[list][i].pic_id)
>                         break;
>                 }
>                 for (; i > index; i--) {
> 
> ---
> base-commit: 26f2f03e0de278f296fbd1e8a09c43245b65f5e3
> change-id: 20250323-h264-reorder-38465068ea4e
> 
> Best regards,
> -- 
> averne <averne381 at gmail.com>
> 
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".