> I'm probably blind spotting the difference between the then and else > case - what is it?
No difference, just a remnant from an earlier version where there was a difference. Thanks for catching. > So we compute whether we need zeroing - but I failed to spot the > check that we can code generate the required VEC_COND when the > target doesn't support the zero else value? Yeah, I forgot that. Added in v4 which I'll be posting soon. > The patch is unfortunately largely mechanical - easy to miss the > important bits. I'm also a bit concerned about having missed some cases but I guess we have enough time to catch those during stage 3 still. -- Regards Robin