This change-set puts 93d08acaacec951dbb302f77eeae51974985b6b2 functionality
under -add-omp-offload-notes switch that is OFF by default.
CUDA toolchain is not able to handle ELF images with LLVMOMPOFFLOAD
notes for unknown reason (see https://reviews.llvm.org/D99551#2950272).
I disable the ELF notes embedding until the CUDA issue is triaged and resolved.
Details
Details
Diff Detail
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Comment Actions
I would prefer to revert 93d08acaacec951dbb302f77eeae51974985b6b2 now and then fix the CUDA toolchain rather than making it someone else's problem.
Comment Actions
@jdoerfert, can we please keep the code upstream? I've spent some time already to resolve the conflicts downstream, and I do not really want to do it once again. I am taking the AR to triage the CUDA issue.
Comment Actions
As discussed, LG, we will look into the NVIDIA GPU problem now to get rid of this again.
clang/tools/clang-offload-wrapper/ClangOffloadWrapper.cpp | ||
---|---|---|
79 | I'd have probably gone with an explicit false here but it doesn't make much difference. |
I'd have probably gone with an explicit false here but it doesn't make much difference.