[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CABob6iqpF5TqvvE48-92x3qgb7+f1WbVacVpi0NFE5bcHi7bZQ@mail.gmail.com>
Date: Thu, 18 Aug 2011 16:45:04 +0200
From: Lukas Odzioba <lukas.odzioba@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: john-1.7.8-mscashcuda-0.diff
2011/8/18 Solar Designer <solar@...nwall.com>:
> Lukas,
>
> I've just spent a few minutes with john-1.7.8-mscashcuda-0.diff.
> Unfortunately, it does not work for me:
>
> [user@...alhost src]$ time ../run/john -te -fo=mscashcuda
> Benchmarking: MSCASHCUDA []... FAILED (get_hash[0](0))
>
> real 0m0.681s
> user 0m0.009s
> sys 0m0.509s
>
> I guess it might be something similar to the issue with
> john-1.7.8-allcuda-0.2.diff, where 3 of 4 hash types fail for me, but
> work for you. You mentioned it's related to GPU family differences, and
> that you know how to fix it, so I think you don't need additional detail.
I'll try to fix that, and I must admit that OpenCL much architecture
neutral, or I am doing something wrong.
>
> Another thing I noticed is that the valid() function is unreliable: it
> doesn't handle the case when strrchr() returns NULL, and it uses a char
> expression as array subscript (should use the ARCH_INDEX() macro there).
Thanks! This will be solved.
> I guess that's normal for a patch that you haven't even announced in
> here yet, and I hope you'll provide and announce a revised version soon.
> The 20k c/s you mentioned (on the wiki) you're getting on your card
> sounds good to me.
>
> Oh, and I think you need to rename s/mscash/mscash2/g everything in this
> patch (and its name).
It was just quick draft of fast MSCash not slow MSCash2, so I didn't
spend much time on that and renaming is not needed. It is 20M not 20K
c/s, and as rest fast hashes for 80-85% time gpu idles.
Thanks for code review. I apreciate.
Lukas
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.
