On 2026-08-04 05:10, Melissa Wen wrote:
>
>
> On 01/08/2026 03:41, Alex Hung wrote:
>> I tested above on
>> - DCN35 - issue fiexed
>> - DCN401 - no issue and no regression.
>>
>> Does it pass DCN321 that you mentioned abouve?
>
> Yes, with this change the kms_colorop passes on DCN321 and so far I don't see
> any visual regression on games.
>
> I also checked that DCN2+ uses different CM helpers for degamma and gamma as
> well - but with 12 segments, which is the original distribution of
> _translate_curve_to_degamma_hw_format.
> Just to say that the proposed approach isn't new.
>
>>
>> DCN301 doesn't use cm3_helper_translate_curve_to_degamma_hw_format() so it
>> wasn't affected.
>
> Oh my, I wasn't expecting that! Banding was affecting both DCN301 and DCN321.
> It was my fault, I overlooked DCN301, I'll extend the fix when we agree in
> the proper fix after Harry's review.
>
The fix looks right to me. Values in non-linear space are essentially optimally
distributed
already, so for a "degamma" operation we would want something close to a
uniform distribution
of LUT entries. It makes no sense to use a distribution that's weighted toward
dark values.
That is something we would want to do for linear to non-linear conversion, like
the shaper,
which already uses a different function and presumably a non-uniform segment
distribution.
So, in short, I agree this change is correct.
Once you put that in a patch feel free to add my
Reviewed-by: Harry Wentland <[email protected]>
Harry
> Thanks for the feedback,
>
> Melissa
>
>>
>> The changes look good to me, but let's ask @Wentland, Harry for review too.
>
