This reverts commit 6506692fe619ef8a1f7c6ea829d9a9eceb31622d.
Details
Details
Diff Detail
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Comment Actions
Help figuring out the proper lit incantations to filter these tests out on non-sm80 would be most welcome.
I could not find what's the error, but setting ptx version might solve.
mlir/test/Integration/GPU/CUDA/TensorCore/transform-mma-sync-matmul-f16-f16-accum.mlir | ||
---|---|---|
15 ↗ | (On Diff #534874) | Have you tried setting the PTX version? gpu-to-cubin sets StringRef properties="+ptx60" by default. This version does not have mma.sync instructions. |
Comment Actions
Harbormaster says green but I have no idea whether these are the buildbots that @mehdi_amini mentioned.
Could you please share a link that I should remember to look at when doing this kind of work?
Comment Actions
link for monitoring post-submit NVIDIA buildbots: https://lab.llvm.org/buildbot/#/builders/61