Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 8 Jul 2012 13:06:37 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: -atom make targets

On Sun, Jul 08, 2012 at 10:19:38AM +0200, Frank Dittrich wrote:
> [fd@fdn run]$ ./john-linux-x86-any --test=10 --format=md5
> Benchmarking: FreeBSD MD5 [32/32]... DONE
> Raw:	4267 c/s real, 4275 c/s virtual
> 
> [fd@fdn run]$ ./john-generic --test=10 --format=md5
> Benchmarking: FreeBSD MD5 [32/32]... DONE
> Raw:	3191 c/s real, 3198 c/s virtual
...
> So, sse, mmx and any are similar, generic is disappointing.
> That looks like there is something wrong with build generic (or with
> bench.c).

Not really.  -x86-any uses asm code, generic is pure C.  Either may
reasonably be slower on a given CPU.

What this means is that the optimizations I made for the original
Pentium in 1990s are also good for Atom.

One other minor detail is that generic never enables
ARCH_ALLOW_UNALIGNED, though.  Jim's patch will address that, which
should bring the performance you're getting in a generic build a little
bit closer to -x86-any's (but still worse than it).

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.