steakhal wrote: > But we support Intel as target triple, that's just a normal cross-compile. The problem is if something in the compiler (accidentally) depends on the host endianness. How about #213281? https://github.com/llvm/llvm-project/pull/212124