aramyaso i can't directly map paired singles to altivec ops but altivec gives me 32 free registers for my emitted instructions to manipulate. fair, i guess. let's see if i can do 1:1 regalloc lol
as there is no way of directly moving fprs <-> vectors i will have to regularly load and store values to do the swap. if i manage to only talk to cache during those operations i think it can be not that slow
the main culprit is the paired single instructions with Rc=1 that will get me to update condition register aaaaaaaa idk how much load/store this will cost (fortunately it seems rare enough that Rc is set for paired singles?)