-ffast-math
A compiler flag that lets the compiler treat floats as if they were real numbers (regroup, assume no NaN or infinity) for speed, at the cost of exact IEEE results.
GCC/Clang -ffast-math enables reassociation, assumes no NaNs or infinities, allows reciprocal approximations and contraction, and may set flush-to-zero at startup; MSVC's nearest equivalent is /fp:fast.
See also