Conversation
There was a problem hiding this comment.
Code Review
This pull request updates the NVIDIA toolchain configuration by adding an nvlink entry, modifies the TritonIR build dependencies, and introduces a compatibility header for AMD. Feedback indicates that adding TritonGPUOpInterfacesIncGen to TritonIR creates a layering violation between the core dialect and GPU-specific interfaces. Additionally, the empty version string for nvlink should be replaced with a valid version to avoid potential build script failures.
| DEPENDS | ||
| TritonTableGen | ||
| TritonCanonicalizeIncGen | ||
| TritonGPUOpInterfacesIncGen |
There was a problem hiding this comment.
Adding TritonGPUOpInterfacesIncGen as a dependency to TritonIR introduces a layering violation. The core Triton dialect should remain independent of GPU-specific interfaces. If this dependency is not strictly required for the compilation of TritonIR, it should be removed to maintain proper architectural separation.
No description provided.