Date: Sun, 24 Jun 2012 02:47:06 +0400 From: Solar Designer <solar@...nwall.com> To: john-dev@...ts.openwall.com Subject: Re: Some relbench results for linux-x86-sse2 (jumbo5 vs. upcoming jumbo6) On Sat, Jun 23, 2012 at 11:51:57PM +0200, Frank Dittrich wrote: > I'll attach 3 files with benchmark output, the j6 one (current git),and > two j5 (original and modified by benchmark-unify). j5-sse2.txt: Benchmarking: FreeBSD MD5 [32/32]... DONE Raw: 4283 c/s real, 4283 c/s virtual j6-sse2.txt: Benchmarking: FreeBSD MD5 [128/128 SSE2 intrinsics 12x]... DONE Raw: 6540 c/s real, 6540 c/s virtual Where's the regression? I see algorithm change and a speedup here. (Low speed either way, but that's expected for that machine.) j5-sse2.txt: Benchmarking: MediaWiki -- md5($s.'-'.md5($p)) [SSE2 32x4]... DONE Many salts: 3708K c/s real, 3708K c/s virtual Only one salt: 1361K c/s real, 1361K c/s virtual j6-sse2.txt: Benchmarking: MediaWiki md5($s.'-'.md5($p)) [128/128 SSE2 intrinsics 10x4x3]... DONE Many salts: 2266K c/s real, 2266K c/s virtual Only one salt: 1377K c/s real, 1377K c/s virtual Also an algorithm change, and one that resulted in a slowdown this time. Need to test these on more/newer machines, I guess (including also 32-bit). Thanks, Alexander
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.