George Hotz
George Hotz
Amazing! Let me look this over
Why did mish brake, that seems odd?
Ahh, interesting. I reverted that change
Overall, this is a pretty big change. I think it needs tests exercising each of the tiling code paths.
Still lacking boundary tests of the different codepaths of the tiler.
Huh, that segfault doesn't even seem related! I wouldn't worry about that. Think this is getting close. Though it's wasting a lot of RAM with not tiling closer to even,...
I merged your tests into master (except tiler), you can rebase for shorter diff.
And now the segfault is present on master!
Can you rebase? Test failure should be gone.