Leo Fang
@asi1024 would it be possible for you to take another look? All tests have passed. Thanks! 🙂
cc: @ev-br for vis
Good question. I took a cursory look. Setting it in Python would end up flipping [`WarningState::sync_debug_mode`](https://github.com/pytorch/pytorch/blob/5614023f5e00f16e0173c503fdc70957fb151cf7/c10/cuda/CUDAFunctions.h#L69), but it's impractical as the call sites have to ensure to call [`warn_or_error_on_sync`](https://github.com/pytorch/pytorch/blob/5614023f5e00f16e0173c503fdc70957fb151cf7/c10/cuda/CUDAFunctions.h#L35) to...
@kmaehashi should we do this in v14?
Thanks for your help! > I also observed one deprecation warning regarding Cython, which is currently not a blocker but may become one in the future: > > > The...
> My suggestion is to use the compiler flag added in [cython/cython#6243](https://github.com/cython/cython/pull/6243) It seems the flag has not made it to the 3.0.x releases...
FYI @kmaehashi we noticed (https://github.com/conda-forge/cupy-feedstock/pull/295#issuecomment-2564498963) that this PR blocks the Python 3.13 support (https://github.com/cupy/cupy/issues/8651) even for the regular (with-GIL) builds, because Cython 0.29.x simply has no wheel or conda package...
> * Use Numba Now that CuPy has its own JIT compiler, I don't think this is that important anymore 🙂
@keithroe @jakirkham I think from CUDA C/C++ point of view manipulating AoS/SoA in CuPy is totally possible. It should not be confused with the need of supporting arbitrary dtypes as...