[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CA+TsHUBf3zSqXSu+LM8QvnqfVf3-x65n1MUSdDK8ONsThXfTgA@mail.gmail.com>
Date: Tue, 4 Dec 2012 10:45:09 +0530
From: Sayantan Datta <std2048@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: bitslice DES on GPU
On Fri, Oct 19, 2012 at 2:01 AM, Solar Designer <solar@...nwall.com> wrote:
>
> > Since the 96 index array is almost constant the indexing could be
> > done prior to execution. However it might require some radical approach
> > like pre compiling the kernels manually.
>
> Yes, or patching binary kernels at runtime. This is something I had in
> mind before we started. I previously tested this on CPU, for a 7%
> speedup over what we currently have in JtR. On GPU, the speedup will
> likely be much greater. You may measure it by temporarily (in a copy of
> the code) eliminating E[] and hard-coding the indices for a certain salt
> value (it is easiest to do this for salt=0 - just take the indices out
> of the LM hash code and use the test hash with ".." for the salt).
>
> Alexander
>
Did that. Performance increased from 67M to 110M using the zeroed salt with
2501 iterations on 7970.
Content of type "text/html" skipped
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.
