math: avoid runtime conversions of floating-point constants