Proper Scope and Domain of GPUGC with PyCuda

Hi, I am new to the GPU and CUDA paradigm.

Where can I read (other then wikipedia) about the proper scope/domains
that can be computed via CUDA .  Obviously parallel computation
algorithms are the niche domain... Can I datamine a server or log file
to run mapreduce tasks.... send parallel requests a webserver and
expect to get understandable meta information back?

JR

feel free to respond at my wiki...
www.cluster-fork.info


On 5/3/09, [email protected] <[email protected]> wrote:
> Send PyCuda mailing list submissions to
>       [email protected]
>
> To subscribe or unsubscribe via the World Wide Web, visit
>       http://tiker.net/mailman/listinfo/pycuda_tiker.net
> or, via email, send a message with subject or body 'help' to
>       [email protected]
>
> You can reach the person managing the list at
>       [email protected]
>
> When replying, please edit your Subject line so it is more specific
> than "Re: Contents of PyCuda digest..."
>
>
> Today's Topics:
>
>    1. 3D Texture References (Daniel Lepage)
>    2. Re: 3D Texture References (Andreas Kl?ckner)
>    3. Re: 3D Texture References (Daniel Lepage)
>    4. Re: PyCuda (Andreas Kl?ckner)
>
>
> ----------------------------------------------------------------------
>
> Message: 1
> Date: Sat, 2 May 2009 23:40:25 -0400
> From: Daniel Lepage <[email protected]>
> Subject: [PyCuda] 3D Texture References
> To: [email protected]
> Message-ID:
>       <[email protected]>
> Content-Type: text/plain; charset=ISO-8859-1
>
> How can I take a three-dimensional numpy.ndarray and bind it to a
> texture<float,3>?
>
> My thought was to do the following (given a 3D ndarray data and a
> texture reference texref):
>
> descr = cuda.ArrayDescriptor3D()
> h, w, d = data.shape
> descr.height = h
> descr.width = w
> descr.depth = d
> descr.format = cuda.array_format.FLOAT
> descr.num_channels = 1
>
> ary = cuda.Array(descr)
>
> copy = cuda.Memcpy3D()
> copy.set_src_host(data)
> copy.set_dst_array(ary)
> copy.width_in_bytes = copy.src_pitch = copy.dst_pitch = data.strides[0]
> copy.height = copy.src_height = h
> copy.depth = d
> copy()
>
> texref.set_array(ary)
>
>
> but this results in:
> Traceback (most recent call last):
>   File "tex3d_test.py", line 42, in <module>
>     copy()
> pycuda._driver.LaunchError: cuMemcpy3D failed: launch failed
> terminate called after throwing an instance of 'cuda::error'
>   what():  cuModuleUnload failed: launch failed
> Aborted
>
> Thanks,
> Dan Lepage
>
>
>
> ------------------------------
>
> Message: 2
> Date: Sat, 2 May 2009 23:50:33 -0400
> From: Andreas Kl?ckner <[email protected]>
> Subject: Re: [PyCuda] 3D Texture References
> To: [email protected]
> Message-ID: <[email protected]>
> Content-Type: text/plain; charset="iso-8859-1"
>
> On Samstag 02 Mai 2009, Daniel Lepage wrote:
>> How can I take a three-dimensional numpy.ndarray and bind it to a
>> texture<float,3>?
>
> Check test_3d_texture in test/test_driver.py.
>
> Andreas
> -------------- next part --------------
> A non-text attachment was scrubbed...
> Name: not available
> Type: application/pgp-signature
> Size: 189 bytes
> Desc: This is a digitally signed message part.
> URL:
> <http://tiker.net/pipermail/pycuda_tiker.net/attachments/20090502/a9210dc6/attachment-0001.bin>
>
> ------------------------------
>
> Message: 3
> Date: Sun, 3 May 2009 00:15:46 -0400
> From: Daniel Lepage <[email protected]>
> Subject: Re: [PyCuda] 3D Texture References
> To: [email protected]
> Message-ID:
>       <[email protected]>
> Content-Type: text/plain; charset=ISO-8859-1
>
> Ah, I stupidly grabbed the wrong stride. Thanks!
>
> --
> Dan Lepage
>
> On Sat, May 2, 2009 at 11:50 PM, Andreas Kl?ckner
> <[email protected]> wrote:
>> On Samstag 02 Mai 2009, Daniel Lepage wrote:
>>> How can I take a three-dimensional numpy.ndarray and bind it to a
>>> texture<float,3>?
>>
>> Check test_3d_texture in test/test_driver.py.
>>
>> Andreas
>>
>> _______________________________________________
>> PyCuda mailing list
>> [email protected]
>> http://tiker.net/mailman/listinfo/pycuda_tiker.net
>>
>>
>
>
>
> ------------------------------
>
> Message: 4
> Date: Sun, 3 May 2009 10:16:06 -0400
> From: Andreas Kl?ckner <[email protected]>
> Subject: Re: [PyCuda] PyCuda
> To: James Rubino <[email protected]>, [email protected]
> Message-ID: <[email protected]>
> Content-Type: text/plain; charset="iso-8859-15"
>
> On Sonntag 03 Mai 2009, James Rubino wrote:
>> Thanks for your reply.
>> My understanding of the multiprocessing is still a bit primitive.
>>
>> My understanding is that python is CPU bound because of the GIL.
>> Breaking away from the CPU via the GPU is what I meant by "work around".
>
> The GPU threads you launch from PyCuda aren't constrained by the GIL.
>
>> If there is an official PyCuda forum online I will gladly send my
>> questions
>> there.
>
> There's a mailing list.
> http://tiker.net/mailman/listinfo/pycuda_tiker.net
>
> Andreas
> -------------- next part --------------
> A non-text attachment was scrubbed...
> Name: not available
> Type: application/pgp-signature
> Size: 189 bytes
> Desc: This is a digitally signed message part.
> URL:
> <http://tiker.net/pipermail/pycuda_tiker.net/attachments/20090503/c4535ae4/attachment-0001.bin>
>
> ------------------------------
>
> _______________________________________________
> PyCuda mailing list
> [email protected]
> http://tiker.net/mailman/listinfo/pycuda_tiker.net
>
>
> End of PyCuda Digest, Vol 11, Issue 2
> *************************************
>


-- 
www.jcrubino.net
www.linkedin.com/in/jimmyru

_______________________________________________
PyCuda mailing list
[email protected]
http://tiker.net/mailman/listinfo/pycuda_tiker.net

Reply via email to