[compiler-rt] [compiler-rt] Stop using x86 builtin on AArch64 with GCC (PR #93890)

[Simon Tatham via llvm-commits]

https://github.com/statham-arm commented:

This certainly seems like a thing that should be fixed. But can you explain why the right decision is _not_ to make both conditions just `#if __has_builtin(__builtin_copysignXXX)`? Why is it _ever_ sensible to add the extra clause that overrides `__has_builtin` returning false, based on the compiler version?

If the answer is (for example) that a particular version of gcc or clang lies about not having a builtin, then should there be a comment explaining that?

https://github.com/llvm/llvm-project/pull/93890