Leo Fang

cc @cryos for vis (since you're also working on wheels)

Tracking more relevant links to code that we want to offer an alternative for: - https://github.com/NVIDIA/nvmath-python/blob/073b168ac0688fa3b84caaa8bb65948bf7db7eae/nvmath/bindings/_internal/cusparse_windows.pyx#L295-L324 - https://pypi.org/project/cupti-python/

> The consequence is that any prior arts that rely on checking if a certain piece exists (ex: nvcc, cuda.h, nvvm, ...) and generalizing it to assume the whole Toolkit...

The bulk of the work is largely done now. Let me close this issue and the remaining tasks can be tracked individually, with the `cuda.pathfinder` label.

Hi, CUDA Python tech lead here 👋 cuFFT has an official Python solution which is exposed as part of [nvmath-python](https://github.com/nvidia/nvmath-python), home for NVIDIA CPU & GPU math libraries. It should...

Hi @Maalvi14 Glad to know! I just checked, Google Colab still has free T4 GPUs for access (it is not the default runner, which is CPU-only, so need to do...

brev.nvidia.com has a wide variety of GPU instances, including T4, but I am not sure if we offer free credits, if Colab does not work for you, I can ask...

We've also moved away from cpdef enum for the same reason. The pollution to the module namespace is too much from the tab completion perspective.

IIRC this is about writing a device function inside a kernel body, something like ```python @jit.rawkernel() def my_kernel(...): def my_dev_func(...): ... ... my_dev_func(...) .... ``` Is this what you need?...

Returning complex values is the right move. Instead of viewing this discussion as introducing a breaking change, I argue this is a NumPy bug (value-dependent casting) that should have been...