nak: LOP2 fixes for SM50
Fixes a few issues with the current implementation:
-
LogicOp2::Pass
argument order is backwards from how SM50 hardware behaves - BNOT src modifiers are ignored
- LOP3 is still emitted by
SSABuilder::sel
andlower_copy_swap
in some cases
I also found that LOP3.LUT
does exist on SM50, we might be able to avoid some of the complexity by just using that. I think we still need the LOP2 stuff for PLOP though.