[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 9 Sep 2013 00:23:45 +0200
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: Daniel's weekly report #12
On 8 sep 2013, at 23:28, Dániel Bali <balijanosdaniel@...il.com> wrote:
> What I have trouble with is how to handle different length input in OpenCL formats. What if the input keys are 8 character long and what if they are 12?
Yes? What is the trouble?
Not sure what you mean but we have at least three variants:
1. Key length is enumerated on CPU side, stored in a separate array and sent to GPU (ie. key[index] and length[index]).
2. Key is null-terminated and length is enumerated on GPU side. Just key[index].
3. Some formats use a packed format. CPU side store keys right after each other (not even separated by a single null) in one array, and have another array show where key[index] is located and how long it is. See for example raw-md5-opencl. This significantly reduces the size of key transfer: While the two variants above always transfer a size of (max_length * keys per crypt) where max_length typically is 55 or even more, this packed variant just need to transfer actual data with no padding, plus one byte per key for index/length.
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.
