[PATCH] D70172: [CUDA][HIP] Fix assertion due to dtor check on windows

[Reid Kleckner via Phabricator via cfe-commits]

rnk added a comment.

Are we sure using both Itanium and MS C++ ABIs at the same time is really the best way forward here? What are the constraints on CUDA that require the Itanium ABI? I'm sure there are real reasons you can't just use the MS ABI as is, but I'm curious what they are. Was there some RFC or design showing that this is the right way forward?

I wonder if it would be more productive to add new, more expansive attributes, similar to `__attribute__((ms_struct))`, that tag class or function decls as MS or Itanium C++ ABI. CUDA could then leverage this as needed, and it would be much easier to construct test cases for MS/Itanium interop. This is an expansion in scope, but it seems like it could be generally useful, and if we're already going to enter the crazy world of multiple C++ ABIs in a single TU, we might as well bite the bullet and do it in a way that isn't specific to CUDA.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D70172/new/

https://reviews.llvm.org/D70172