john-dev - Re: Bleeding device selection

Follow @Openwall on Twitter for new release announcements and other news

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <5b5eb6a805d185e03331461340d1e5b7@smtp.hushmail.com>
Date: Mon, 18 Feb 2013 23:40:12 +0100
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: Bleeding device selection

On 18 Feb, 2013, at 20:59 , Claudio André <claudioandre.br@...il.com> wrote:
> Em 18-02-2013 16:50, magnum escreveu:
>> On 18 Feb, 2013, at 20:42 , Claudio André <claudioandre.br@...il.com> wrote:
>> Em 18-02-2013 16:40, magnum escreveu:
>>>> On 18 Feb, 2013, at 19:55 , magnum <john.magnum@...hmail.com> wrote:
>>>>> Claudio,
>>>>> 
>>>>> Maybe I do something wrong but I currently get these weird results from bleeding on Bull:
>>>>> ...
>>>>> magnum@...l:src [bleeding-jumbo]$ ../run/john -t -fo:opencl -dev=1
>>>>> Device 2: AMD FX(tm)-8120 Eight-Core Processor
>>>>> OpenCL error (CL_INVALID_CONTEXT) in file (common-opencl.c) at line (552) - (Error while creating program)
>>>>> 
>>>>> 
>>>>> Not sure what happens, it used to work fine a while ago.
>>>> BTW that example used the new --format=opencl magic, but that is not the cause of the problem. Without that patch, you still get the error for the first opencl format when running "john -t -dev=1".
>>>> 
>>>> magnum
>>> The test i mentioned earlier was:
>>> ../run/john -t=0 -fo:opencl -de:1 (on bull)
>> Ah, of course, thanks. After the sorting of OpenCL format regitration, DES is the first one. I think I'll put those two last for now, or disable them altogether until Sayanyan fixes them.
>> 
>> magnum
> 
> The "first" fix attached. mscash 2 works. But, for sure, more tests needed.
> 
> Claudio
> <ver.txt>

I committed this along with a fix for pwsafe. The latter now uses a split kernel, but only asked for max workgroup size for one of the kernels. The good thing about my crappy CPU driver is it triggers such bugs.

Unfortunately it also triggers it in DES and I can't find how/where to fix it:

Benchmarking: Traditional DES [OpenCL]... OpenCL error (CL_INVALID_WORK_GROUP_SIZE) in file (opencl_DES_bs_b.c) at line (410) - (Enque Kernel Failed)

The BF format works on CPU here though, so it apparently obeys that limit.

magnum

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.