rjmccall wrote: The AMDGPU `TargetInfo` is responsible for ensuring that the choices it makes match the ABI. It is inconceivable that just dropping in the converted struct type that Clang IRGen uses as a direct parameter or result type would be a correct way to implement this; even if by chance it happens to match today (quite unlikely given all the complexities of unions, bit-fields, and padding), it is not resilient to innocuous changes to type conversion like the one in this patch. Clang would be completely within its rights to lower a struct type to `[N x i8]`, and frankly I think we're gradually heading in that direction in LLVM.
https://github.com/llvm/llvm-project/pull/96422 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
