[Gpundarray] Fwd: OpenCL merged

[Arnaud Bergeron]

Le 10 novembre 2011 16:51, Frédéric Bastien <nouiz at nouiz.org> a écrit :
> I commited change to your script to cover more case. I rename it to
> setup_opencl.py and merge it to the trunk.
>
> I have as you the segmentation fault.

I fixed a bunch of bugs in the OpenCL and most tests pass now (with no
segfaults).

There are still two that don't and it seems that the copy code does
not work correctly at some point (but I have no idea where).

> Fred
>
>
> ---------- Forwarded message ----------
> From: Arnaud Bergeron <abergeron at gmail.com>
> Date: 2011/11/10
> Subject: Re: [Gpundarray] OpenCL merged
> To: Frédéric Bastien <nouiz at nouiz.org>
>
>
> Le 10 novembre 2011 12:02, Frédéric Bastien <nouiz at nouiz.org> a écrit :
>> Can you tell me how to compile it for opencl?
>
> I use the setup2.py in branch build_ocl of my fork.  You might have to
> adjust some include paths for it to work, but otherwise it's fine.
>
>> Fred
>>
>> 2011/11/9 Arnaud Bergeron <abergeron at gmail.com>:
>>> Le 8 novembre 2011 15:59, Frédéric Bastien <nouiz at nouiz.org> a écrit :
>>>> Hi,
>>>>
>>>> I just pushed change to make all generic cuda/opencl code support the
>>>> offset attribute. So the opencl back-end should have all what it need
>>>> to have all code working as in cuda except for element wise operation.
>>>> There is also the copy code that can't be used by both at the same
>>>> time. I don't know if it was written for opencl.
>>>
>>> The changes are not enough apparently.  The code still segfaults when
>>> running the tests.  I'll try to go over this tomorrow.
>>>
>>>> We have our paper accepted to the NIPS workshop big learn. It will be
>>>> a poster and a talk. The "chief reviewer" made it a talk to raise the
>>>> awareness of the project even if we miss important feature like
>>>> reduction.
>>>>
>>>> The opencl code have been merged in the trunk from the branch.
>>>>
>>>> Arnaud, can you tell me how to compile the code with opencl instead of
>>>> cuda? We should try to test with the opencl backend. This allow to
>>>> "prove" that we can have shared code between CUDA and OpenCL.
>>>>
>>>> thanks
>>>>
>>>> Fred
>>>>
>>>> 2011/10/18 Andreas Kloeckner <lists at informa.tiker.net>:
>>>>> On Thu, 13 Oct 2011 16:03:29 -0400, Frédéric Bastien <nouiz at nouiz.org> wrote:
>>>>>> Hi,
>>>>>>
>>>>>> the branch with OpenCL have been refactored a little and merged in the
>>>>>> trunk. Not all code implemented functionnality is supported with the
>>>>>> OpenCL back-end. There is a change needed related to contex. Arnaud
>>>>>> will make a ticket about this.
>>>>>>
>>>>>> Andreas, do you see other change needed to be able to use it in
>>>>>> pycuda/pyopencl? The refactoring should have more localized the use of
>>>>>> nvcc.
>>>>>
>>>>> I'll try to push out releases of Py{CUDA,OpenCL} this weekend, and then
>>>>> we should try and do the merge to see what breaks.
>>>>>
>>>>> Andreas
>>>>>
>>>>
>>>> _______________________________________________
>>>> Gpundarray mailing list
>>>> Gpundarray at tiker.net
>>>> http://lists.tiker.net/listinfo/gpundarray
>>>>
>>>
>>>
>>>
>>> --
>>> La brigade SnW veut vous recruter - http://www.brigadesnw.ca
>>>
>>
>
>
>
> --
> La brigade SnW veut vous recruter - http://www.brigadesnw.ca
>
> _______________________________________________
> Gpundarray mailing list
> Gpundarray at tiker.net
> http://lists.tiker.net/listinfo/gpundarray
>



-- 
La brigade SnW veut vous recruter - http://www.brigadesnw.ca