On Thu, Feb 11, 2021 at 08:04:55PM +1000, Nicholas Piggin wrote: > Excerpts from Christophe Leroy's message of February 11, 2021 5:41 pm: > > As modern powerpc implement branch folding, that's even more efficient.
Ah, it seems you mean what Arm calls branch folding. Sure, power4 already did that, and this has not changed. > I think POWER will speculate conditional traps as non faulting always > so it should be just as good if not better than the branch. Right, these are not branch instructions, so are not branch predicted; all trap instructions are assumed to fall through, like all other non-branch instructions. Segher