Fill out python -m numba -s field for CUDA Libraries Test Output (even if GPU is absent)

[jakirkham]

Feature request

When creating a numba environment in a Docker image or other build environment where the CUDA Toolkit libraries are present, but a GPU is absent like so...

conda create -n numba python=3.11 cuda-version=12.2 cuda-nvcc cuda-nvrtc psutil numba

...and running python -m numba -s. The information on CUDA Libraries detection is left out:

CUDA Libraries Test Output:
None

---

However the "CUDA Libraries Test Output:" field...

numba/numba/misc/numba_sysinfo.py

Lines 611 to 612 in 3edb458

	("CUDA Libraries Test Output:",),
	(info.get(_cu_lib_test, "None"),),

...merely calls numba.cuda.cudadrv.libs function test:

numba/numba/misc/numba_sysinfo.py

Line 16 in 3edb458

from numba.cuda import cudadrv

numba/numba/misc/numba_sysinfo.py

Lines 371 to 375 in 3edb458

	output = StringIO()
	with redirect_stdout(output):
	cudadrv.libs.test()
	sys_info[_cu_lib_test] = output.getvalue()
	output.close()

Running this alone is ok:

python -c "from numba.cuda.cudadrv.libs import test; test()"

Finding driver from candidates:
	libcuda.so
	libcuda.so.1
	/usr/lib/libcuda.so
	/usr/lib/libcuda.so.1
	/usr/lib64/libcuda.so
	/usr/lib64/libcuda.so.1
Using loader <class 'ctypes.CDLL'>
	Trying to load driver...	ok
		Loaded from libcuda.so
	Mapped libcuda.so paths:
		/usr/local/cuda-11.8/compat/libcuda.so.520.61.05
Finding nvvm from Conda environment (NVIDIA package)
	Located at /opt/conda/envs/numba/nvvm/lib64/libnvvm.so.4.0.0
	Trying to open library...	ok
Finding nvrtc from Conda environment (NVIDIA package)
	Located at /opt/conda/envs/numba/lib/libnvrtc.so.12.3.107
	Trying to open library...	ok
Finding cudart from Conda environment (NVIDIA package)
	Located at /opt/conda/envs/numba/lib/libcudart.so.12.3.101
	Trying to open library...	ok
Finding cudadevrt from Conda environment (NVIDIA package)
	Located at /opt/conda/envs/numba/lib/libcudadevrt.a
	Checking library...	ok
Finding libdevice from Conda environment (NVIDIA package)
	Located at /opt/conda/envs/numba/nvvm/libdevice/libdevice.10.bc
	Checking library...	ok

---

It appears this happens as Numba skips including this info if a GPU is not found

numba/numba/misc/numba_sysinfo.py

Line 336 in 3edb458

cu.list_devices()[0] # will a device initialise?

However it would be nice to include this information just to provide a sanity check that Numba can find the CUDA Toolkit libraries

[guilhermeleobas]

Hi @jakirkham, thanks for the report. I have labeled this issue as a feature request.