yxsamliu wrote: > In general, having different C++ ABIs between the host and device seems like > an ongoing source of tension and bugs.
I agree. However completely switching to Microsoft ABI on device side does not work with existing device libraries since they assume Itanium mangling. Therefore I only changes record layout to be compatible with host, in the hope that the generated LLVM IR is correct for such a combination. I added more tests about member accessing and virtual function calls. It seems the IR is correct. I think clang codegen is generic enough to handle Itanium ABI with Microsoft record layout. https://github.com/llvm/llvm-project/pull/87651 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits