Pedro Côrte-Real

In fact the non-dither non-sse code seems to match the new darktable (sse and non-sse) code just fine even though that one is implemented differently. So I suspect there's indeed...

I've disabled all that when generating these differences. There is no base curve and no WB. The only modules applied are the input profile (I used a linear one) and...

I did some more testing and our code also produces similar differences to your code. So maybe this is just a normal level of difference. It sounds quite high to...

Using a linear output as well the differences between our code and the rawspeed non-sse code are smaller and look mostly like a random splatter of red and blue pixels....

Note that I'm seeing these differences without dithering.

Thanks for looking into this. Are you saying that what's being fixed would never be noticeable in the final image? I'm struggling to define what is a reasonable level of...

There is no denoise in these tests, but I have also wondered if the demosaic algo is doing something funny. The completely linear and without WB tests give differences of...

Did you get any chance to look at this again? I've cooked up a darktable regression testing setup and it's showing up differences between most files and I think that...

Doing things like this is much easier after the rework done in #5629. The matching logic becomes centralized in a single place so it's easy to do this, glob matching...

@LebedevRI has mentioned interest in getting these files working in darktable in case you want to try something with him.