The Intel math library does not presently support these functions. I looked into what LLVM libc did, and it returned -0.0 for all of the values, and from reading the source code, this should be true across all architectures.