@brecht So after some further investigation, the issue could be eliminated by hard coding the random offset into the ints https://projects.blender.org/blender/blender/src/commit/8fa578dcc2a0587cd…
The test was performed on an RTX 3070.
There are a few things we can try: do not call fmod when all(fabs(p) < 1000000.0f))
I mean we don't necessarily need to use fmod. If were talking…
Huh, interesting. Though I don't think we can do much about the M2 CPU not being able to do fmod quickly. And it's also not like it's some kind of obscure function that we could just avoid to use.
Actually regarding the possible seed input, if we were to add it I plan to make it a [0.0, 1.0] value. That way it'd be compatible with a possible implementation in some other textures like…
Is there really no other way to solve this? Some other way of offsetting the coordinates where we don't just add to them, but change them in a way that preserves precision in some…
@brecht Ok, then there's just the problem of OSL left. We currently use the build in Perlin noise function provided by OSL which doesn't support different seeds. Should I just port the GLSL…
@brecht Then we could also add a seed input without proper versioning. Just to be on the same page, I'd also do the versioning code if wanted.
Actually there is way to add a seed input while keeping the noise output the same. Since a seed input of 0 would result in the same output as the current Fac (or red channel) output, having…
I'm currently too busy to immediately start work on this but I could add a seed input in the near future. But before I start work on that I'd want to make sure that everyone's on board with this…
@OmarEmaraDev I agree, but even then the difference would be immediately obvious. I think what's important for most users is whether or not the output visibly changes at all, rather than how much…
If we choose to accept some breaking changes, perhaps we could add the offsets after scaling octaves, which means the first octave will remain the same, and the breaking change will be "less…
There are actually issues much worse than just the pixelation:
The red output on the left shows the correct…
Brecht is right, but I'd like to add some explanations as to why it happens and why it happens much earlier for Perlin noise compared to the other noise textures.
One of the properties of…
@studio_punkt You mentioned that you used the exact same node setup in 3.6.
When you opened the file in 4.x did you change the Detail value to 15.0 yourself or was it set to 15.0 automatically?
This is neither a performance regression nor unexpected behavior.
Let me explain: In 3.6 the Voronoi node didn't have a Detail input, which is equivalent to setting the Detail value to…