Comment by dominikh

Note that the source you're seeing there is the fallback implementation, which is only used if there is no instruction for FMA in the architecture you're compiling for. On AMD64, for example, the call to math.FMA will be replaced by the VFMADD231SD instruction.