Leo Fang

In theory, for a given `RawKernel` (either you use it directly, or get it via `RawModule.get_function`) you can retrieve the `CUFunction` pointer via `RawKernel.kernel.ptr`, but 1. This is not public...

Thanks for sharing your use case @chaoming0625, this is very interesting! Would you be able to point us how you use this capability to make CuPy and Jax interoperable at...

Friendly nudge @chaoming0625 :)

Just FYI, the array API standard does not regulate the sort order and it could be unsorted: https://data-apis.org/array-api/latest/API_specification/set_functions.html Though, in NumPy 2.0 it is still sorted https://numpy.org/devdocs/reference/routines.set.html since all the...

> Thanks for the info @leofang. What I don't get is that, if array-API compliant routines are implemented as shortcuts to `numpy.unique()`, then why not choosing the most efficient `unique()`...

Note to self: I didn't touch NCCL/cuSPARSELt in #4567 because the way that a stream is passed to the library is different: there, the stream is an argument for all...

> * cuFFT: [Fully supported](https://docs.nvidia.com/cuda/cufft/index.html#cuda-graphs-support) Note: This is inaccurate. We got lucky because we have a plan cache, otherwise the plan creation can't be captured.

I am thinking about this too. I think the bigger problem is that the cache keys can contain includes with relative paths (or truncated paths in the bracket form ``...

Right, so when I noted > we might have to either implement it ourselves or use the WIP ([NVIDIA/jitify#131](https://github.com/NVIDIA/jitify/pull/131)). I meant we'll no longer rely on NVRTC to find headers,...

@ryanshrott could you file a new issue with requested information provided (see the issue template)? This warm-up should not exceed ~1 min.