shiltian wrote: I don't think OpenMP is more special than HIP here. Anything exposed to the host should not be internalized. In addition, OpenMP actually also heavily uses internalization as well in OpenMPOpt. It is likely that this change exposes something bad in the downstream.
The motivation of this change was to apply ThinLTO for OpenMP. Given our current approach to link device runtime, it will be treated as a regular input file and then goes through the backend as well, which is completely unnecessary. Its whole purpose should be for function import. After that, it should be discarded. https://github.com/llvm/llvm-project/pull/138365 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
