[FFmpeg-devel] [PATCH] libavcodec/libfdk-aacenc: Scale VBR mode with FF_QP2LAMBDA

[Gyan Doshi]

On 2023-02-22 09:39 pm, Timo Rothenpieler wrote:
> On 22.02.2023 03:12, JonHGee wrote:
>> libavcodec/libfdk-aacenc: VBR mode currently does not account for 
>> scaling when using -aq options with libfdk, resulting in clamping to 
>> vbr mode 5 whenever a value >0 is provided.  Adjusting for the 
>> scaling factor for proper VBR support.
>>
>> ---
>>   libavcodec/libfdk-aacenc.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/libavcodec/libfdk-aacenc.c b/libavcodec/libfdk-aacenc.c
>> index 54549de473..da211baf51 100644
>> --- a/libavcodec/libfdk-aacenc.c
>> +++ b/libavcodec/libfdk-aacenc.c
>> @@ -230,7 +230,7 @@ static av_cold int aac_encode_init(AVCodecContext 
>> *avctx)
>>       }
>>         if (avctx->flags & AV_CODEC_FLAG_QSCALE || s->vbr) {
>> -        int mode = s->vbr ? s->vbr : avctx->global_quality;
>> +        int mode = s->vbr ? s->vbr : avctx->global_quality / 
>> FF_QP2LAMBDA;
>>           if (mode <  1 || mode > 5) {
>>               av_log(avctx, AV_LOG_WARNING,
>>                      "VBR quality %d out of range, should be 1-5\n", 
>> mode);
>
> Won't this break every existing command line and API client that has 
> passed a value according to the current scale?
>
> Also, what binds stronger here?
> It this "(s->vbr ? s->vbr : avctx->global_quality) / FF_QP2LAMBDA" or 
> "s->vbr ? s->vbr : (avctx->global_quality / FF_QP2LAMBDA)"?
>
> In any case, this does not look correct to me.
> Where would the sudden multiplication with FF_QP2LAMBDA come from in 
> the first place?

From

fftools\ffmpeg_mux_init.c
619:        ost->enc_ctx->global_quality = FF_QP2LAMBDA * qscale

Regards,
Gyan