Hi,
Probably I am overlooking something very simple… at the moment I have some weird theories that the subdivide POP is acting in a somewhat non-deterministic way, thats where the title of this thread comes from, but probably I am making some mistake.
See the attached toe:
mixQuestion.toe (9.6 KB)
I am using a box, subdivide it to create a sphere-ish thing. I also subdivide the box with a higher crease weight parameter. I do this because I thought I get 2 shapes and wanted to use the GLSL Copy POP to blend between these two shapes in a shader. It seems that the connectivity/order of the 2 shapes is completely different though.
While writing this Post, I realized that the same problem should appear when I just use the Blend POP. I tried it and, well it does!
Any comments appreciated.
Hi Patrik,
I don’t think we have a list of known issues but this is one of those, going to bump it up in my current list!
The only other POPs which is non deterministic is the line thick POP (basically the line MAT in POPs)
Also Sprinkle POP in volume mode has both deterministic and non deterministic flavor.
Hi,
Looking at this again because I was really doubtful if ‘non-deterministic’ is the right word here.
Seems like it is.
With a Box POP as input, the problem starts appearing with Depth >=3. When you adjust the Crease Weight, The point indizes of the Vertices shuffle around.
The weird thing is that (starting at Crease Weight = 0) if you increase the crease weight and then decrease it back to 0, you end up with a different data then you started with.
See attached.
subDivEx.toe (3.7 KB)
Yes, every time it cooks you can end with a different point order and it’s more likely with more points - it’s currently still using an atomic operation in the algorithm, and there’s no guarantee the same threads are executed in the same order
Will be fixed in next released build (Fixed in builds 2023.31732 and above)
