This was previously implemented with a macro and we were using __builtin_copysign(), which takes double inputs for the float version of copysign(). Reviewed-and-Tested-by: Aaron Watry <awatry@gmail.com> llvm-svn: 217045
Tom Stellard