[Chapel Merge] add missing include path 'gpu/cuda' to get_runtime

Branch: refs/heads/main
Revision: 5e1936c
Author: stonea
Link: add missing include path 'gpu/cuda' to get_runtime_includes_and_defines by stonea · Pull Request #19133 · chapel-lang/chapel · GitHub
Log Message:

Merge pull request #19133 from stonea/fix_gpu_runtime_include_path_regression

add missing include path 'gpu/cuda' to get_runtime_includes_and_defines

This aims to solve the regression described here: https://github.com/Cray/chapel-private/issues/2961 where we're producing this error every time we run Chapel when we're using the gpu locale model:

stdchpl.h:78:10: fatal error: 'chpl-gpu-gen-includes.h' file not found

It looks like chapel-lang/chapel#18880 influenced what include paths get included when we link to the runtime library and that's introduced a regression in our GPU tests. It looks like these are now being computed in our Python scripts specifically by the get_runtime_includes_and_defines function in util/chplenv/compile_link_args_utils.

In this PR I've modified that function so it will also add runtime/include/gpu/cuda

[Reviewed by @mppf]

Modified Files:
M util/chplenv/compile_link_args_utils.py

Compare: https://github.com/chapel-lang/chapel/compare/0ad48a037f2d...5e1936c1a82a