Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 28 Jun 2012 11:45:39 +0200
From: Tavis Ormandy <>
Subject: Re: Re: Failed self test for raw-sha1-ng (linux-x86-sse2i OMP)

magnum <> wrote:

> On 2012-06-28 09:35, Tavis Ormandy wrote:
> > I see the bug, that's embarassing. Give me a minute to send a push
> > request.
> > 
> It got a little slower benchmark now, hopefully just because of new test
> vectors.
> magnum

Let me see if I can win that back :-)

Re-ordering the W loads closer to first use seems to help gcc allocate the
registers a bit. I'm not 100% happy with the code it generates, but it looks
a little better. I think I can also nudge up the number of crypts with the
cmp_one changes.

On my linux-x86-native (xeon) machine, before:

$ ../run/john --test=30 --format=raw-sha1-ng
Benchmarking: Raw SHA-1 (pwlen <= 15) [128/128 SSE4.1 intrinsics 4x]... DONE
Raw:	12730K c/s real, 12803K c/s virtual


$ ../run/john --test=30 --format=raw-sha1-ng
Benchmarking: Raw SHA-1 (pwlen <= 15) [128/128 SSE4.1 intrinsics 4x]... DONE
Raw:	13103K c/s real, 13160K c/s virtual

Please verify you get similar improvement, because I havn't tried it on x64.

Push request sent.


------------------------------------- | pgp encrypted mail preferred

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.