Date: Sun, 05 Aug 2012 00:08:47 +0200 From: magnum <john.magnum@...hmail.com> To: john-dev@...ts.openwall.com Subject: Re: sunmd5 On 2012-08-04 23:58, Solar Designer wrote: > On Sat, Aug 04, 2012 at 11:49:52PM +0200, magnum wrote: >> On 2012-08-04 23:38, Solar Designer wrote: >>> Here's an obvious optimization, removing the modulo division. >>> >>> Benchmarking: SunMD5 [128/128 XOP intrinsics 8x x1024]... DONE >>> Raw: 541 c/s real, 541 c/s virtual >>> >>> or even (best of several invocations): >>> >>> Benchmarking: SunMD5 [128/128 XOP intrinsics 8x x1024]... DONE >>> Raw: 544 c/s real, 544 c/s virtual >> >> Committed, thanks. On my laptop it made a more significant boost, almost 4%. > > Thanks. I think you misread the above. It's +7.5% here (506 to 544). I realised that right after posting :) > I also tried eliminating the sprintf(), but this only hurt performance. > Apparently, it is not called that frequently, whereas my changes affected > register allocation in the inner loop or something. (Not submitting.) I'll run Valgrind again. There must be something more to do, I suppose we are still at 15-20% "waste" in crypt_all(). I think sapG was somewhat comparable to this. I'll revisit that format and see how it looks in Valgrind. magnum
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.