Floating point error mitigation