Some floating-point instructions clear all bits in FLG,
they will have to be fixed to only clear some LH bits.