Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sat, 11 Apr 2015 02:34:29 +0300
From: Solar Designer <>
Subject: Re: [GSoC] John the Ripper support for PHC finalists

Hi Agnieszka,

On Sat, Apr 11, 2015 at 01:18:50AM +0200, Agnieszka Bielec wrote:
> 2015-04-09 6:21 GMT+02:00 Solar Designer <>:
> > I suggest that you leave OpenCL alone for a moment (well, maybe right
> > after you finish whatever you're currently doing to it), and instead get
> > POMELO's SIMD code into the tree first.  Both SSE* and AVX2, please.
> I've added SSE2 and isn't faster (bleeding-jumbo)

This is unexpected.  Are you sure the SSE2 (actually AVX, when building
with AVX enabled) code is getting compiled in?  And the non-SSE2 code
isn't getting compiled in?

> pomelo documentation says:
> "In POMELO, function H makes the attack using SIMD expensive
> since the function H uses password-dependent memory accesses."

The original POMELO submission to PHC actually tried to defeat any SIMD
(but was of limited success at that).  POMELO v2 was revised to allow
for up to 256-bit SIMD, and only defeat attempts to use SIMD wider than
that.  256-bit is enough up to AVX2.  Maybe the POMELO specification
document wasn't fully revised to reflect that - I mean, the algorithm
specification was revised, but the surrounding text may contain pieces
like what you quoted above, which now needs further clarification.

> I've added also tunable costs for pomelo_fmt_plug.c


> I don't know what to do with avx2 because
> [ src]$ ./configure | grep -i avx
> checking for AVX... yes
> checking for AVX2... no

I'll create you an account on our machine with AVX2.  I'll e-mail you
off-list about that.



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.