[clang] [llvm] [clang][X86] Return Win64 __int128 indirectly (PR #210358)

[via llvm-commits]

Andarwinux wrote:

That gcc commit should be reverted. The most reasonable approach is consistently using two GPR or one XMM to pass and return.

Since you want to break the ABI, why not just use the highest-performance solution?

I can't think anything more stupid than deliberately sacrificing performance just to comply with some outdated, ridiculous document.

https://github.com/llvm/llvm-project/pull/210358