Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 11 Mar 2015 22:11:49 +0100
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: bitslice MD*/SHA*, AVX2

On 2015-03-11 21:55, Solar Designer wrote:
> For comparison, a non-OpenMP build of john-1.8.0-jumbo-1 on this machine
> achieves:
> 
> solar@...l:~/j/john-1.8.0-jumbo-1/run$ ./john -te -form=raw-md5
> Benchmarking: Raw-MD5 [MD5 128/128 AVX 12x]... DONE
> Raw:    31923K c/s real, 31923K c/s virtual
> 
> So the speed is finally comparable.  That's AVX vs. AVX2, so a
> straightforward implementation of MD5 with AVX2 will likely run faster
> yet, but the results so far are not conclusively anti-bitslice.

Just for the record, Jumbo-1 had an unfortunate regression from a patch
meant to mitigate "address-sanitizer" crashes, that was applied without
due care about consequences. An extra strncpy() had significant impact
on the fastest hashes.

I have since fixed this. Current bleeding-jumbo on same machine:

magnum@...l:src [bleeding-jumbo]$ ../run/john -test -form:raw-md5
Benchmarking: Raw-MD5 [MD5 128/128 AVX 12x]... DONE
Raw:	37842K c/s real, 37842K c/s virtual

Regardless, I think the bitslice task is relevant and interesting.

magnum

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.