[FFmpeg-devel] [PATCH] avcodec/cuviddec: fix AV1 decoding error

["zhilizhao(赵志立)"]

> On Jun 23, 2022, at 10:14 PM, Timo Rothenpieler <timo at rothenpieler.org> wrote:
> 
> On 23/06/2022 11:10, Zhao Zhili wrote:
>> From: Zhao Zhili <zhilizhao at tencent.com>
>> Regression from 45e3b6a68. cuvidParseVideoData report unknown
>> error with AV1CodecConfigurationRecord.
>> ---
>>  libavcodec/cuviddec.c | 11 +++++++++++
>>  1 file changed, 11 insertions(+)
>> diff --git a/libavcodec/cuviddec.c b/libavcodec/cuviddec.c
>> index cb3cda7e24..584fff40ed 100644
>> --- a/libavcodec/cuviddec.c
>> +++ b/libavcodec/cuviddec.c
>> @@ -953,6 +953,17 @@ static av_cold int cuvid_decode_init(AVCodecContext *avctx)
>>          extradata_size = avctx->extradata_size;
>>      }
>>  +#if CONFIG_AV1_CUVID_DECODER && defined(CUVID_HAS_AV1_SUPPORT)
>> +    if (avctx->codec->id == AV_CODEC_ID_AV1 &&
>> +            extradata_size > 4 &&
>> +            extradata && extradata[0] & 0x80) {
>> +        // skip AV1CodecConfigurationRecord to keep configOBUs only,
>> +        // otherwise cuvidParseVideoData report unknown error.
>> +        extradata += 4;
>> +        extradata_size -= 4;
>> +    }
>> +#endif
> 
> Checking the defines shouldn't be necessary. The codec ID should never be AV1 if it's not enabled, and the code does not contain anything AV1 specific.

I have thought about this. On one hand, the conditional compile is
unnecessary. On the other hand, it can remove dead code explicitly
when the condition is false, instead of depending on compiler.

I’m OK with remove the condition to keep code simple.

> 
>>      ctx->cuparse_ext = av_mallocz(sizeof(*ctx->cuparse_ext)
>>              + FFMAX(extradata_size - (int)sizeof(ctx->cuparse_ext->raw_seqhdr_data), 0));
>>      if (!ctx->cuparse_ext) {
> 
> I'm not quite sure I understand what causes this.
> Something about the mp4 demuxer changed, yes.
> But what about other containers? Why wasn't this an issue before?

It’s a regression from the point of fftools. If we take libavcodec
as standalone lib, the issue exist before 45e3b6a68.

> 
> The code that was removed in 45e3b6a68 seems to just have always skipped the first 4 byte unconditionally.

In the context of mp4 demuxer, it’s guaranteed the extradata is
AV1CodecConfigurationRecord. There is no guarantee that the
extradata is AV1CodecConfigurationRecord in the context of
decoder. So that’s why the check `extradata[0] & 0x80` exist.

AV1CodecConfigurationRecord is used by mp4 and mkv both. And
there is a similar patch in mkv: ffd1316

    avformat/matroskadec: remove special handling of av1c extradata

    Libavcodec can now handle the standard AV1CodecConfigurationRecord
    extradata as-is.

> 
> A little more info in the commit message would be nice here. Specially a sentence or two about the "& 80" part.

Will add comments about the “& 80” part.

Thanks for the review!

> Code LGTM otherwise.
> _______________________________________________
> 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".