ci(build): pin MSVC toolset to 14.44 for Windows CUDA portable build#457
Merged
Conversation
GitHub's windows-large runner image drifted to an MSVC toolset newer than the pinned CUDA 12.9.1 toolkit supports. Without a toolset pin, TheMrMilchmann/ setup-msvc-dev activates the newest toolset, so nvcc's host_config.h version gate rejects it during CMake's CUDA compiler probe (C1189 'unsupported Microsoft Visual Studio version'), and forcing past that gate instead crashes nvcc's front-end (cudafe++ ACCESS_VIOLATION) — the newer MSVC is genuinely incompatible with CUDA 12.9. Pin the VC++ toolset to 14.44 (VS 2022 17.14, MSVC 19.44), which ships on the runner image and is within CUDA 12.9's supported host-compiler range. Code was unchanged since the last green run; this is purely runner toolchain drift. Verified on branch: the Windows CUDA leg builds and packages pie-x86_64-windows-cuda.zip end-to-end.
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (1)
WalkthroughThis change updates the GitHub Actions workflow for the Windows CUDA portable build leg, pinning the MSVC toolset version to "14.44" in the setup-msvc-dev step, along with added explanatory comments. ChangesCI Windows CUDA Toolchain Pin
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Root cause
The
portable / x86_64-windows-cudajob broke on runner image drift, not code — the same job was green on the last run. Thewindows-large-x64runner drifted to Visual Studio 18, which now ships two MSVC toolsets: 14.44.35207 (MSVC 19.44) and 14.51.36231 (MSVC 19.51).TheMrMilchmann/setup-msvc-devwith no toolset pin activates the newest (14.51), which the pinned CUDA 12.9.1 toolkit does not support:host_config.hversion gate rejects it during CMake's CUDA compiler probe —C1189: unsupported Microsoft Visual Studio version.-allow-unsupported-compilerjust moves the failure downstream: nvcc's front-end crashes —nvcc error : 'cudafe++' died with status 0xC0000005 (ACCESS_VIOLATION). So the newer MSVC is genuinely incompatible with CUDA 12.9, which is exactly why the gate exists.Fix
Pin the VC++ toolset to 14.44 (MSVC 19.44) in the Windows CUDA step. 14.44 ships on the current image (confirmed via the run's vcvars dump) and is within CUDA 12.9.1's supported host-compiler range, so nvcc's probe and the ggml-cuda build both succeed. The
-allow-unsupported-compilerapproach was evaluated and dropped (it segfaults cudafe++).Verification
Green test run: https://github.com/pie-project/pie/actions/runs/28537465006 —
portable / x86_64-windows-cudabuilt and packagedpie-x86_64-windows-cuda.zipend-to-end in 13m20s (cargo build → package → upload). Tested with a windows-cuda-only matrix + throwaway tag; this PR restores the full matrix and carries only the toolset pin.Notes
Summary by CodeRabbit