On 18 August 2014 17:50, Alan Lawrence <alan.lawre...@arm.com> wrote: > Well, you're right that it could be. So I presented the wrong justification. > > Clearly we would benefit from some better cost infrastructure here, ideally > that is expressive, taken into account at all appropriate stages of the > compiler, and tunable per core. I imagine that steps (patches) towards such > infrastructure would be welcomed by both AArch64 maintainers and more > widely. > > In the meantime, however, we must work with what we have. I'll still argue > that we should remove the '!' (as per patch), however. As James has said, > even if your add is more expensive in SIMD registers, the '!' still doesn't > express that; and leaving it in affects code-generation on all cores. And it > is inconsistent with other instructions.
Agreed and OK. /Marcus