[FFmpeg-devel] [PATCH] swresample: try to fix regression

[Martin Storsjö]

On Wed, 8 Feb 2023, Paul B Mahol wrote:

> This one should pass FATE.
>

> diff --git a/libavfilter/af_pan.c b/libavfilter/af_pan.c
> index 80c194b066..067f646805 100644
> --- a/libavfilter/af_pan.c
> +++ b/libavfilter/af_pan.c
> @@ -313,9 +313,7 @@ static int config_props(AVFilterLink *link)
>              pan->channel_map[i] = ch_id;
>          }
> 
> -        av_opt_set_chlayout(pan->swr, "ichl", &link->ch_layout, 0);
> -        av_opt_set_chlayout(pan->swr, "ochl", &pan->out_channel_layout, 0);
> -        av_opt_set_int(pan->swr, "uch", link->ch_layout.nb_channels, 0);
> +        av_opt_set_int(pan->swr, "uch", pan->nb_output_channels, 0);
>          swr_set_channel_mapping(pan->swr, pan->channel_map);
>      } else {
>          // renormalize
> @@ -335,8 +333,6 @@ static int config_props(AVFilterLink *link)
>              for (j = 0; j < link->ch_layout.nb_channels; j++)
>                  pan->gain[i][j] /= t;
>          }
> -        av_opt_set_chlayout(pan->swr, "ichl", &link->ch_layout, 0);
> -        av_opt_set_chlayout(pan->swr, "ochl", &pan->out_channel_layout, 0);
>          swr_set_matrix(pan->swr, pan->gain[0], pan->gain[1] - pan->gain[0]);
>      }
>

This bit looks like a straight revert of the commit that broke fate, so 
that's probably ok.

> diff --git a/libswresample/swresample.c b/libswresample/swresample.c
> index 5884f8d533..7c8d9bfc97 100644
> --- a/libswresample/swresample.c
> +++ b/libswresample/swresample.c
> @@ -312,9 +312,9 @@ av_cold int swr_init(struct SwrContext *s){
>      if (s->out_ch_layout.order == AV_CHANNEL_ORDER_UNSPEC)
>          av_channel_layout_default(&s->out_ch_layout, s->out.ch_count);
> 
> -    s->rematrix = av_channel_layout_compare(&s->out_ch_layout, &s->in_ch_layout) ||
> +    s->rematrix = !s->channel_map && (av_channel_layout_compare(&s->out_ch_layout, &s->in_ch_layout) ||
>                   s->rematrix_volume!=1.0 ||
> -                 s->rematrix_custom;
> +                 s->rematrix_custom);
>
>      if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
>          if(   av_get_bytes_per_sample(s-> in_sample_fmt) <= 2

I really can't say about this part of the patch though...

Can we maybe just do it in two parts - revert the patch that broke fate 
and find someone other than me to comment on the swresample bit?

// Martin