I see. I made a mistake when saying some nonzero seeds are not reproducible; my checking was wrong. For the seed 0, it does happen to be reproducible for some reason in some cases; I understand. N…
Sorry, to clarify, do you mean the issue of Volcano at seed 0 is not reproducible or issues other than that are not reproducible? Thanks for your investigation btw!
And to make things more weird. Only in volcano preset it happens. Nothing weird happens in all other presets I tried.
Good observation. In the initial test, it seems nonzero values work. However, in a bunch of jobs I run, I found some nonzero seeds are not reproducible, but some zero seeds are reproducible.…
Thanks for your investigation. I am surprised that any hash function should use the runtime address, though.
In my opinion, anything that is not debuggable (i.e. reproducible on the same platform) is a bug, unless there is an alternative deterministic implementation (e.g., cpu impl of cuda code). Can…
after second thought, I don't think rounding the results is a good idea because as the number of vertices increase, the probability of any of them differ increases. As what we want is not the…
Thanks for your reply. The two explanations you found are pretty old, 13 years ago, and 6 years ago. I tried the second example, Mac and Linux no longer differs. But then I google for more…
Thanks for your investigation. Regarding your suggestion to sort the vertex, I am outputting the summation of vertices coords after modulus, which is already permutation invariant, so it means the…