Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 24 Jul 2011 05:24:31 +0400
From: Solar Designer <solar@...nwall.com>
To: john-users@...ts.openwall.com
Subject: Re: 1.7.8 to jumbo2 speed regression for FreeBSD md5

On Sat, Jul 23, 2011 at 02:04:39PM -0500, James Nobis wrote:
> On my Phenom II X6 1090T @ 3.2Ghz (Black edition) with all the auto OC
> boost disabled I ran run/john -test against a build with make
> linux-x86-64 on each.
> 
> john-1.7.8
> Benchmarking: FreeBSD MD5 [32/64 X2]... DONE
> Raw:	14537 c/s real, 14566 c/s virtual
> 
> john-1.7.8-jumbo2
> Benchmarking: FreeBSD MD5 [8x]... DONE
> Raw:	6656 c/s real, 6656 c/s virtual

I've revised the gcc version check in 1.7.8-jumbo-4 to only use the new
code with gcc 4.4+, assuming that your version of gcc is somewhere
between 4.0 and 4.3.x.  Did this help?

Ideally, we'd test all of: 4.0, 4.1, 4.2, 4.3, 4.4 to determine which is
the lowest gcc version number where the new code is beneficial...  We
could have to do this on different CPUs, though.  Recent Intel CPUs tend
to perform better at SSE2 than AMD's do, so the same version of gcc may
be good enough for Intel, but not good enough for AMD.

With gcc 4.5.0, the SSE2 intrinsics code definitely performs faster for
me, but that's on Intel CPUs only (no AMDs here).

Alexander

Powered by blists - more mailing lists

Your e-mail address:

Powered by Openwall GNU/*/Linux - Powered by OpenVZ