[Common] Default CUDA_HOME to /usr/local/cuda when dynamically loading cuDNN and NVRTC

[denera]

Description

This PR ensures that we always look for cuDNN and NVRTC under /usr/local/cuda when importing Transformer Engine without requiring CUDA_HOME or CUDA_PATH to be set to the default CUDA installation path. The user should only need to set these variables if CUDA is installed in a non-standard/custom path.

Type of change

• Documentation change (change only to the documentation, either a fix or a new content)
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Infra/Build change
• Code refractor

Changes

• Dynamically loading cuDNN and NVRTC now always falls back on checking /usr/local/cuda when CUDA_HOME or CUDA_PATH variables are not set.

Checklist:

• I have read and followed the contributing guidelines
• The functionality is complete
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[denera]

/te-ci pytorch

[ksivaman]

LGTM! Let's run the full CI though not just pytorch @denera

[denera]

/te-ci

[timmoon10]

Looks reasonable. It would be better to check in LD_LIBRARY_PATH before trying /usr/local/cuda, although this is a nit.