[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20110818171544.GB2461@openwall.com>
Date: Thu, 18 Aug 2011 21:15:44 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: john-1.7.8-mscashcuda-0.diff
On Thu, Aug 18, 2011 at 05:17:48PM +0200, Lukas Odzioba wrote:
> I've just tought that with pbkdf2 and mscash already working coding
> mscash2 will be easier.
I don't see how coding mscash on GPU helps you implement mscash2.
For mscash2 you need PBKDF2 with SHA-1 on GPU, but the rest of it you do
on CPU. That is, you should not have mscash on GPU to implement
mscash2. Only the PBKDF2 step should be on GPU. In fact, you may fully
reuse this same GPU kernel for more than just mscash2 - also for
something else based on PBKDF2 with SHA-1. For steps other than PBKDF2,
you can even use code "optimized" for simplicity rather than for
performance, such as:
http://openwall.info/wiki/john/MSCash2_simple_code
It simply won't matter unless/until you make your GPU code run
thousands of times faster than the CPU code, which is unlikely.
> Making patches faster is not finished yet, and
> I am going to work on that for a few weeks.
Now that I erroneously announced you having an experimental
implementation of mscash2 on GPU, I'd appreciate it if you actually
implement mscash2 on GPU first (or rather just the PBKDF2 portion of it,
with the rest implemented on CPU). Would you be able to do that in a
few days (before GSoC final evals)?
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.
