Test `numba_debug` in the libNVVM path too?

[leofang]

Follow-up of #1216. We currently test the NVRTC path, but cuda.core.Program also covers libNVVM (and nvJitLink!) and we should get them tested too.

[leofang]

@ayermolo is this something you can help with? 🙂

[ayermolo]

@leofang My understanding from: https://github.com/NVIDIA/numba-cuda/blob/358f3c4a38577d5bc61e5dbe228dddd3e8505d01/numba_cuda/numba/cuda/cudadrv/nvvm.py#L229
Is that for nvvm path Numba Cuda is not using Program python class. It sets options internally and passes them directly to:
err = self.driver.nvvmCompileProgram(
self._handle, self.n_options, self.option_ptrs
)

Since the flag is numba specific, should it still be tested on generic libNVVM path in cyda-python Program?

[leofang]

Yes. Our eventual goal (NVIDIA/numba-cuda#128) is to replace everything that numba-cuda has internally to wrap CUDA functionalities by cuda-core, so once Program is ready for consuming and compiling NVVM IR (there is a bit of gaps that @abhilash1910 is filling in PR #981) we'll make this happen, so it is good to get this capability plumbed through (as suggested in the team sync today by @kkraus14).

[ayermolo]

NVIDIA/numba-cuda#128

Ah IC. Thanks for elaborating. Yeah I can put it on my queue.
What is the timeline for this?

[leofang]

Whenever you get some free cycles I guess? 🙂