[Bf-committers] Non-Standard sRGB OETF transform T51216

[Sergey Sharybin]

Ok, just to answer the question and avoid dangling question flying around
the mailing list.

Guess you're talking about issue reported in [1]. The root of the issue
there is coming from OCIO itself, which has numerical stability issue
in lookupLinear_1D (and it's SSE variant) when index is far above the
possible LUT index.

Doing something like this [2] (and surely it's SSE variant) will solve the
issue. But this needs to be double checked from the CPU ticks point of view
-- currently it does some unwanted casts which ideally need to be avoid).

[1] https://developer.blender.org/T51217
[2] https://hastebin.com/posiduhesa.diff

On Sat, Apr 15, 2017 at 1:16 PM, Sergey Sharybin <sergey.vfx at gmail.com>
wrote:

> This is because OCIO is rather a joke and takes much much much more time
> to perform inverse transform. And it doesn't even cache anything, so it
> performs expensive inversion for every pixel, every frame slowing
> everyone's workflow like hell.
>
> The new inverted transform is calculated from the original forward
> transfom by simply transposing it. I am not sure what is wrong there and
> your feedback is rather vague. Please attach .blend configured in a way
> which shows that something became broken after that change.
>
> On Sat, Apr 15, 2017 at 12:55 AM, Troy Sobotka <troy.sobotka at gmail.com>
> wrote:
>
>> Whoops... wasn't T51216, but rather
>> https://git.blender.org/gitweb/gitweb.cgi/blender.git/commit
>> /6fb874369c31649de7232235b0114344bfdb8e92
>>
>> Anyways, the values and ranges are off and it breaks HDRIs.
>>
>> With respect,
>> TJS
>>
>> On Fri, Apr 14, 2017 at 3:53 PM Troy Sobotka <troy.sobotka at gmail.com>
>> wrote:
>>
>> > Any reason that the nonstandard sRGB OETF transform was expanded to
>> > include an "inverted" LUT as well?
>> >
>> > Why wasn't OCIO's native inversion sufficient? Why is the inverse
>> building
>> > upon the nonstandard SPI based sRGB transform versus a proper 0.0 to 1.0
>> > sRGB OETF?
>> >
>> > With respect,
>> > TJS
>> >
>> _______________________________________________
>> Bf-committers mailing list
>> Bf-committers at blender.org
>> https://lists.blender.org/mailman/listinfo/bf-committers
>>
>
>
>
> --
> With best regards, Sergey Sharybin
>



-- 
With best regards, Sergey Sharybin