[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20120811222809.GA1831@openwall.com>
Date: Sun, 12 Aug 2012 02:28:09 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: Current -fixes GPU formats vs TS
Lukas -
On Sun, Aug 12, 2012 at 12:05:26AM +0200, Lukas Odzioba wrote:
> 2012/8/11 Lukas Odzioba <lukas.odzioba@...il.com>:
> > I'll merge those changes and send final patch to magnum.
> > I added in my last patch proper get_binary and get_hash functions,
> > and removed crack checking from gpu code. So your changes to this code
> > won't be included in the final version.
>
> Here's patch. Moving crack checking to cpu seems to not affect speed.
Why do you #include "sha2.h" and how is this going to work in the -fixes
branch?
The CUDA and OpenCL implementations are now inconsistent in where the
comparisons are done.
Did you do your speed testing on your cards only or also on bull's?
I think there may be a bit more of a difference (maybe 1% or so?) with
the faster GPUs.
I think you could transfer e.g. 64-bit partial hashes to CPU only, and
when these match, do a full recomputation and comparison in cmp_exact()
(on CPU). But this is beyond what's acceptable for the -fixes branch
now. Also, you're right that polishing this format (beyond making it
work, which we did) is not a priority right now.
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.
