Marc Glisse <marc.gli...@inria.fr> writes: >> Realised afterwards that both versions need to check >> GET_MODE_NUNITS (mode) == 2, because we're requiring OP0 and OP1 >> to be scalar. Sorry for not noticing first time. > > I thought that was a consequence of > > XVECLEN (trueop1, 0) == 2 > > (in the lines before your first &&) > ie the result (which has mode "mode") is a vec_select of 2 objects.
Yeah, you're right of course. Richard