Update imports in dispatch module to handle missing GPU

[oliverholworthy]

Use HAS_GPU in dispatch module to avoid some imports that raise exceptions. Followup to #98

Motivation

Importing the dispatch module without GPU available (e.g. in a docker container running without gpu configuration) raises an exception.

This module is imported by merlin.systems and makes it more difficult to experiment with triton without GPU features enabled.

import merlin.core.dispatch

Results in the following error (partial error output)

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
File /usr/local/lib/python3.8/dist-packages/cuda/_cuda/ccuda.pyx:3553, in cuda._cuda.ccuda._cuInit()

File /usr/local/lib/python3.8/dist-packages/cuda/_cuda/ccuda.pyx:424, in cuda._cuda.ccuda.cuPythonInit()

RuntimeError: Failed to dlopen libcuda.so
Exception ignored in: 'cuda._lib.ccudart.utils.cudaPythonGlobal.lazyInit'
Traceback (most recent call last):
  File "cuda/_cuda/ccuda.pyx", line 3553, in cuda._cuda.ccuda._cuInit
  File "cuda/_cuda/ccuda.pyx", line 424, in cuda._cuda.ccuda.cuPythonInit
RuntimeError: Failed to dlopen libcuda.so
---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
[...]
RuntimeError: Function "cuDeviceGetCount" not found

Implementation Details

The dask_cudf import is the one responsible for the exception. And the RuntimeError raised cannot be caught with a simple try/except.

This PR wraps some of the imports with a check for the HAS_GPU variable from the compat module added in #98

[github-actions]

Documentation preview

https://nvidia-merlin.github.io/core/review/pr-99