Date: Mon, 17 Aug 2015 15:29:58 -0500 From: JimF <jfoug@....net> To: john-dev@...ts.openwall.com Subject: Re: Sybase-PROP (Re: FMT_OMP_BAD) On Mon, 17 Aug 2015 14:50:42 -0500, Solar Designer <solar@...nwall.com> wrote: > On Mon, Aug 17, 2015 at 02:32:08PM -0500, JimF wrote: >> I find it to be overall faster but I was not having much OMP scale issue >> before: > > Oh. Due to what the problem was, I am not surprised that it affected > different systems differently. In fact, I am a bit surprised that it > did result in that poor OpenMP scaling on super. I would have expected > uses of Rot2() to become fast on all threads eventually (after some > cache line bouncing between the CPUs), but somehow this wasn't happening > on super. The code was unsafe (it needed memory barriers), so it could > have misbehaved, but I am surprised the cache coherence mechanisms > apparently never entered a stable state even when running for a second. > I also wonder if adding the proper memory barriers would have avoided > the slowdown. This would be curious to test, and it has implications on > lots of other (correct) code. I am surprised that it worked at all, with the static's the way they were. As for me not seeing a problem with OMP scaling, I wonder if my build was actually producing invalid results, possibly not detecting thread interference and not accounting for it. To me, it sounds like this was happening on on super, thus your scaling was looking very poor. > In other words, you're getting slightly worse OpenMP scaling now than > before? But overall better speeds, due to improved single-thread speed? The slowdown is really within expectations, since the faster you go, the further away from 'ideal' you expect to get. Dropping .25x/8 is not that far from expectations when going up 25% or more in overall speed. Some of that may have been able to be removed if the OMP_SCALE was not set back so far, I have not investigated. Trying to squeeze out the absolute ultimate speed from this, vs trying to find actual speedups (like you did changing the rot2) usually is a losing battle ;)
Powered by blists - more mailing lists
Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.