https://github.com/erichkeane approved this pull request. Seems alright so far. Though if we're going to add the CUDA global, we should have a test for it. https://github.com/llvm/llvm-project/pull/137094