Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 4 Jul 2012 20:29:15 +0400
From: Solar Designer <>
Subject: Re: OSX problems with Jumbo-6 OpenCL

On Wed, Jul 04, 2012 at 12:55:13PM -0300, Claudio Andr?? wrote:
> I removed the byte_addressable and self-tested here (.cl attached).


> /* MD5 OpenCL kernel based on Solar Designer's MD5 algorithm implementation at:

We should also have a copyright and license statement for the OpenCL
code here.

> #define F(x, y, z)			((z) ^ ((x) & ((y) ^ (z))))
> #define G(x, y, z)			((y) ^ ((z) & ((x) ^ (y))))

We should use bitselect() here.

>     (a) = (((a) << (s)) | (((a) & 0xffffffff) >> (32 - (s)))); \

...and rotate() here.

I understand that speeding this up does not matter yet, though - and we
did take care of these little things in "slow" hashes.  When we approach
optimizing this for real (and when that matters), we'll need to reverse
several MD5 steps as well (or more than just several).


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.