[FFmpeg-devel] [PATCH 2/3] avcodec/libopenh264enc: make the profile configuablable correctly

[Martin Storsjö]

On Thu, 13 Jan 2022, lance.lmwang at gmail.com wrote:

> From: Limin Wang <lance.lmwang at gmail.com>
>
> If the version of libopenh264 >= 1.8, we can't configured main profile as
> expected, below is the testing cli:
>
> ffmpeg -y -f lavfi -i testsrc -c:v libopenh264 -profile:v main -frames:v 1 test.ts
> It'll report:
> [libopenh264 @ 0x5638300] Unsupported profile, select EProfileIdc PRO_BASELINE in libopenh264.
>
> Signed-off-by: Limin Wang <lance.lmwang at gmail.com>
> ---
> libavcodec/libopenh264enc.c | 8 +++++---
> 1 file changed, 5 insertions(+), 3 deletions(-)
>
> diff --git a/libavcodec/libopenh264enc.c b/libavcodec/libopenh264enc.c
> index a55bef8..995ee37 100644
> --- a/libavcodec/libopenh264enc.c
> +++ b/libavcodec/libopenh264enc.c
> @@ -220,26 +220,27 @@ static av_cold int svc_encode_init(AVCodecContext *avctx)
> #endif
>
>     switch (s->profile) {
> -#if OPENH264_VER_AT_LEAST(1, 8)
>     case FF_PROFILE_H264_HIGH:
> +        s->profile = PRO_HIGH;

I don't think we should reuse the s->profile field for this value here.

In practice, both FF_PROFILE_H264_HIGH and PRO_HIGH have the same values, 
but they're enums from different namespaces, so would it be clearer to use 
one variable for profiles set with FF_PROFILE_* and one with the PRO_* 
values?

>         param.iEntropyCodingModeFlag = 1;
>         av_log(avctx, AV_LOG_VERBOSE, "Using CABAC, "
>                 "select EProfileIdc PRO_HIGH in libopenh264.\n");
>         break;
> -#else
>     case FF_PROFILE_H264_MAIN:
> +        s->profile = PRO_MAIN;
>         param.iEntropyCodingModeFlag = 1;
>         av_log(avctx, AV_LOG_VERBOSE, "Using CABAC, "
>                 "select EProfileIdc PRO_MAIN in libopenh264.\n");
>         break;
> -#endif
>     case FF_PROFILE_H264_CONSTRAINED_BASELINE:
>     case FF_PROFILE_UNKNOWN:
> +        s->profile = PRO_BASELINE;
>         param.iEntropyCodingModeFlag = 0;
>         av_log(avctx, AV_LOG_VERBOSE, "Using CAVLC, "
>                "select EProfileIdc PRO_BASELINE in libopenh264.\n");
>         break;
>     default:
> +        s->profile = PRO_BASELINE;
>         param.iEntropyCodingModeFlag = 0;
>         av_log(avctx, AV_LOG_WARNING, "Unsupported profile, "
>                "select EProfileIdc PRO_BASELINE in libopenh264.\n");
> @@ -251,6 +252,7 @@ static av_cold int svc_encode_init(AVCodecContext *avctx)
>     param.sSpatialLayers[0].fFrameRate          = param.fMaxFrameRate;
>     param.sSpatialLayers[0].iSpatialBitrate     = param.iTargetBitrate;
>     param.sSpatialLayers[0].iMaxSpatialBitrate  = param.iMaxBitrate;
> +    param.sSpatialLayers[0].uiProfileIdc        = s->profile;

So this assignment is what was missing, and was what caused the incorrect 
conclusion in d3a7bdd4ac54349aea9150a234478635d50ebd87? I think it'd be 
good to explicitly spell this out in the commit message, saying that 
d3a7bdd4ac54349aea9150a234478635d50ebd87 was based on incorrect 
conclusions because we had missed to set uiProfileIdc.

// Martin