GCC mangles long double like __float128 in order to support compatibility with ABI variants which had a different interpretation of long double. This fixes PR23791. llvm-svn: 239421