On 29 July 2011 10:58, Dmitry Melnik <d...@ispras.ru> wrote: > This patch adds two define_insn patterns for NEON vabd instruction to make > combine pass recognize expressions matching (vabs (vsub ...)) patterns as > vabd.
Interesting but I would be a bit defensive and make sure that this matches only if -ffast-math in the FP case. You are sort of relying on the fact that vsub wouldn't be generated without ffast-math but I'd rather be defensive about it . (This is in case it's not clear in the non-intrinsics case). I've had a couple of conversations about what the intrinsics behaviour should in such cases with folks. Should we try to match vabs (vsub) even for intrinsics and generate a vabd or desist from doing this and generate only what was asked for. My personal preference is the former but it would be interesting to see what others think . BTW was SPEC2k built with -Ofast ? Maybe then you'll see a bit of vectorization. cheers Ramana
