Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <d90044813a39c3a7aa3906c590a6e4f1@smtp.hushmail.com>
Date: Thu, 01 Nov 2012 08:12:33 +0100
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: mscash2-opencl problems with GT650M

On 11/01/12 03:46, Sayantan Datta wrote:
> Hello magnum,
>
> On Wed, Oct 31, 2012 at 2:48 AM, magnum <john.magnum@...hmail.com
> <mailto:john.magnum@...hmail.com>> wrote:
>
>        HANDLE_CLERROR(__clWaitForEvents(1,&evnt),"SYNC FAILED");
>
>
> Yes it is correct. However does it solves the problem? If yes then it is
> fine.

It solves the problem of hiding the actual error code. The original code 
would send true|false to HANDLE_CL_ERROR() instead of the error code. I 
can't imagine why you did that?

The original code gave this error (passing "false" to HANDLE_CLERROR()):
OpenCL error (UNKNOWN ERROR :() in file (common_opencl_pbkdf2.c) at line 
(313) - (SYNC FAILED)

With my change I get this:
OpenCL error (CL_OUT_OF_RESOURCES) in file (common_opencl_pbkdf2.c) at 
line (313) - (SYNC FAILED)

> Alternately you may comment this statement altogether without any
> foreseeable problem(tested on 4890,7970 and CPU). I think clFinish()
> would implicitly wait for all events to complete.
>
>   Also I get the same error while profiling on AMD GPUs. So in general I
> comment it out anyways during profiling.

That's strange. Why do you get an error when waiting for the specific 
event, but not on clFinish? Some problem is probably still there but hidden.

If I comment out the line altogether I would expect the 
CL_OUT_OF_RESOURCES to occur at clFinish. But while that's probably 
still the case I do not get the hint, just a no-longer existing command 
queue:
OpenCL error (CL_INVALID_COMMAND_QUEUE) in file (common_opencl_pbkdf2.c) 
at line (315) - (clFinish error)

Anyway, trying to address the root problem (provided it is indeed 
resources) I tried lowering MAX_KEYS_PER_CRYPT but it does not help. I 
also tried using NULL for local work size to no avail. What could be the 
problem? What resources could be low? I have 48K of local memory and 1G 
of global.

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.