Re: [FFmpeg-devel] [PATCH] swscale/swscale: don't reject scaling when color parameters are not supported but conversion is not required

[Niklas Haas <ffmpeg@haasn.xyz>]

On Fri, 17 Jan 2025 22:38:03 -0300 James Almer <jamrial@gmail.com> wrote:
> Values in csp, prim, trc, etc, are irrelevant if there's no conversion needed.
>
> Signed-off-by: James Almer <jamrial@gmail.com>
> ---
> ./ffmpeg -i https://0x0.st/8Hr_.mp4 -vf "scale=1920:1080" -f null -
>
> Fails without this patch, because swscale lacks conversion for the signaled
> color primaries.
>
>  libswscale/swscale.c | 13 +++++--------
>  libswscale/utils.h   | 16 +++++++++++-----
>  2 files changed, 16 insertions(+), 13 deletions(-)
>
> diff --git a/libswscale/swscale.c b/libswscale/swscale.c
> index 30fbc590fd..70667ebd1c 100644
> --- a/libswscale/swscale.c
> +++ b/libswscale/swscale.c
> @@ -1442,6 +1442,7 @@ int sws_frame_setup(SwsContext *ctx, const AVFrame *dst, const AVFrame *src)
>      for (int field = 0; field < 2; field++) {
>          SwsFormat src_fmt = ff_fmt_from_frame(src, field);
>          SwsFormat dst_fmt = ff_fmt_from_frame(dst, field);
> +        int src_ret, dst_ret;
>
>          if ((src->flags ^ dst->flags) & AV_FRAME_FLAG_INTERLACED) {
>              err_msg = "Cannot convert interlaced to progressive frames or vice versa.\n";
> @@ -1449,14 +1450,10 @@ int sws_frame_setup(SwsContext *ctx, const AVFrame *dst, const AVFrame *src)
>              goto fail;
>          }
>
> -        if (!ff_test_fmt(&src_fmt, 0)) {
> -            err_msg = "Unsupported input";
> -            ret = AVERROR(ENOTSUP);
> -            goto fail;
> -        }
> -
> -        if (!ff_test_fmt(&dst_fmt, 1)) {
> -            err_msg = "Unsupported output";
> +        src_ret = ff_test_fmt(&src_fmt, 0);
> +        dst_ret = ff_test_fmt(&dst_fmt, 1);
> +        if ((!src_ret || !dst_ret) && !ff_props_equal(&src_fmt, &dst_fmt)) {
> +            err_msg = src_ret ? "Unsupported output" : "Unsupported input";

Nit: Could move the style change to a separate commit.
Nit: Would rename "src_ret" to "src_ok" and "dst_ret" to "dst_ok" to make it
     clearer that 0 indicates failure, not success.

While this is an improvement, it fails to solve the root problem. I think the
correct fix is to defer the check to ff_sws_graph_init(), which can fail in a
more granular way when these properties are actually being needed.

That said, this improves the status quo so LGTM for now.

>              ret = AVERROR(ENOTSUP);
>              goto fail;
>          }
> diff --git a/libswscale/utils.h b/libswscale/utils.h
> index d76fea1133..8e0cdf26ec 100644
> --- a/libswscale/utils.h
> +++ b/libswscale/utils.h
> @@ -98,12 +98,10 @@ static inline int ff_color_equal(const SwsColor *c1, const SwsColor *c2)
>              ff_prim_equal(&c1->gamut, &c2->gamut);
>  }
>
> -/* Tests only the static components of a colorspace, ignoring per-frame data */
> -static inline int ff_fmt_equal(const SwsFormat *fmt1, const SwsFormat *fmt2)
> +/* Tests only the static components of a colorspace, ignoring dimensions and per-frame data */
> +static inline int ff_props_equal(const SwsFormat *fmt1, const SwsFormat *fmt2)
>  {
> -    return fmt1->width      == fmt2->width      &&
> -           fmt1->height     == fmt2->height     &&
> -           fmt1->interlaced == fmt2->interlaced &&
> +    return fmt1->interlaced == fmt2->interlaced &&
>             fmt1->format     == fmt2->format     &&
>             fmt1->range      == fmt2->range      &&
>             fmt1->csp        == fmt2->csp        &&
> @@ -111,6 +109,14 @@ static inline int ff_fmt_equal(const SwsFormat *fmt1, const SwsFormat *fmt2)
>             ff_color_equal(&fmt1->color, &fmt2->color);
>  }
>
> +/* Tests only the static components of a colorspace, ignoring per-frame data */
> +static inline int ff_fmt_equal(const SwsFormat *fmt1, const SwsFormat *fmt2)
> +{
> +    return fmt1->width      == fmt2->width      &&
> +           fmt1->height     == fmt2->height     &&
> +           ff_props_equal(fmt1, fmt2);
> +}
> +
>  static inline int ff_fmt_align(enum AVPixelFormat fmt)
>  {
>      const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(fmt);
> --
> 2.48.1
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".