[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <5e72c2cf8512ad6a7463dbdcf3fb304a@smtp.hushmail.com>
Date: Thu, 4 Jul 2013 21:49:01 +0200
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: bug: GPU use in CPU-only formats
On 4 Jul, 2013, at 20:41 , magnum <john.magnum@...hmail.com> wrote:
>> It works like expected with non-opencl formats (my laptop does not switch to discrete GPU). But opencl formats fail:
>>
>> OpenCL error (CL_INVALID_CONTEXT) in file (common-opencl.c) at line (560) - (Error while creating program)
>>
>>> Have you committed somewhere?
>>
>> Committed now to temporary branch "opencl-init".
>
> I'm starting to think we should drop that branch and try to straighten things out properly. Here's what I think we should do:
>
> 1. (This is a side track) Re-factor all non-static functions in common-opencl.c so they are prefixed with opencl_, eg:
>
> init_opencl_devices() -> opencl_passive_init()
> clean_opencl_environment() -> opencl_done()
>
> 2. Re-work opencl_passive_init() so it does the platform/device enumeration and option parsing but does *not* set up any queue or context (I am confident that's what wakes up the GTX 570).
>
> 3. Add such setting up of queue and context in opencl_init_opt().
>
> And then we keep calling opencl_passive_init() from john.c but it won't wake up the nvidia. I'll think this over for a while.
I tried this in yet another temporary branch "opencl-init2". It does not work. Again, probably something really trivial and I'm just unfocused.
$ ../run/john -t=0 -form:md5crypt-opencl
Warning: doing quick benchmarking - the performance numbers will be inaccurate
OpenCL error (CL_INVALID_DEVICE) in file (common-opencl.c) at line (1367) - (Error querying CL_DEVICE_TYPE)
It doesn't wake up the GPU when running a non-opencl format though.
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.
