Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20150917154610.GA16763@openwall.com>
Date: Thu, 17 Sep 2015 18:46:10 +0300
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: wordlist mgetl() (was: Judy array)

On Thu, Sep 17, 2015 at 06:36:01PM +0300, Solar Designer wrote:
> On Thu, Sep 17, 2015 at 04:15:40PM +0200, magnum wrote:
> > Here's a theory: Does the E5420 have a significant penalty for unaligned 
> > SIMD loads?
> 
> Yes.
> 
> > If so, you should try this (wordlist.c:140)
> > 
> > -#if defined(SIMD_COEF_32) && !defined(_MSC_VER) && \
> > +#if 0 && defined(SIMD_COEF_32) && !defined(_MSC_VER) && \
> > 
> > The alternative code might be faster.
> 
> I've just tried, with and without mmap.  I am getting similar speeds as
> before (a sub-second difference one way or the other).  mmap is still
> slower than pre-loading (with -mem=0).  That's for the 29M testcase;
> maybe I should test on something more focused on wordlist performance.

When cracking just 1 hash, mmap is faster than pre-loading either way,
and more so when SIMD is kept enabled.  The time differences are small,
but are measurable across multiple runs.

So it appears that mmap is somehow negatively affected by us having a
lot of other memory activity, to a greater extent than pre-loading.

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.