Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 19 Oct 2008 05:53:41 +0400
From: Solar Designer <>
Subject: Re: fast freebsd MD5 implementation

On Fri, Oct 17, 2008 at 01:52:02PM +0200, Simon Marechal wrote:
> And for more spam, this adds an ICC target

I compiled this on linux-x86-64 with gcc 4.3.1 after some trivial
changes (patch attached).

> gcc:
> Benchmarking: FreeBSD MD5 [12x]... DONE
> Raw:    2640 c/s real, 2613 c/s virtual

What version of gcc is that?  I am not getting results this poor on
x86-64 with gcc 4.3.1.  Indeed, it is possible that gcc's shortcomings
manifest themselves in a worse way when fewer registers are available.

> icc:
> Benchmarking: FreeBSD MD5 [12x]... DONE
> Raw:    10380 c/s real, 10380 c/s virtual

Here's what I am getting:

Athlon 64 3000+ 2.0 GHz
old code: 8970 c/s
4x: 7060
8x: 4500
12x: 5100
16x: 5100
20x: 5020

Q6600 2.4 GHz (one core):
old code: 10200 c/s
4x: 14844
8x: 9592
12x: 9374
16x: 3200
20x: 3100

So even with gcc there's an improvement on Core 2, but only when no
extra parallelism is introduced (beyond "one instance" of SSE2).

This time all 5 builds actually crack all of my 120 test hashes.



View attachment "john-" of type "text/plain" (2061 bytes)

To unsubscribe, e-mail and reply
to the automated confirmation request that will be sent to you.

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.