x86 Floating point exceptions - incorrect support?