Polar downscale antiring bug

https://github.com/mpv-player/mpv/issues/12375

changed milestone to %v6.323+

added bug shaders labels

assigned to @haasn

It seems that changing shifted range from "[0.07,1.07]" to "[0.08,1.08]" fixed the issue for me. See https://github.com/mpv-player/mpv/issues/12375#issuecomment-1715997303

But I don't understand why it can only be trigger by --dscale antiringing

not sure if 0.08 is enough in this case though, since kernel weights are multiplied to 0.08^33 as well. I guess 0.1 is a much safer choice (0.1**33 * 1/1024 is still in range)

mentioned in commit bjin/libplacebo@9fc5ee57

mentioned in merge request !579 (merged)

But I don't understand why it can only be trigger by --dscale antiringing

When downscaling kernel size increases, as we sample more pixels to produce single output one. This is what correct_downscaling does. So we end up with bigger filter size that apparently overflows.

mentioned in commit bjin/libplacebo@e706d20c

I understand how correct-downscaling works. Increasing filter kernel size will summing more samples, but it's an underflow issue instead of overflow. So I guess it's not relevant in this case.

Anyhow, the fix seems to work fine so far.

mentioned in commit bjin/libplacebo@2e248c1a

closed with merge request !579 (merged)