Just as a relevant resource for other people trying to get OpenCL savvy in a hurry, I've been watching a series of talks that Andreas Klockner gave in Jan. 2011. There are 4 one hour sessions: Lecture 1 - http://www.youtube.com/watch?v=X9mflbX1NL8 (Intro stuff) Lecture 2 - http://www.youtube.com/watch?v=MqvfCE_bKOg (PyOpenCL) Lecture 3 - http://www.youtube.com/watch?v=TAvKmV7CuUw (Code generation) Lecture 4 - http://www.youtube.com/watch?v=SsuJ0LvZW1Q (Multi-GPU, PyCUDA) Hope it helps someone :) -Dan 2011/5/17 James Bergstra :

Exciting work.  Can I draw your attention to related ongoing work with Andreas Klochner and Frederic Bastien on the 'gpundarray' also slated for compyte? The goal is essentially to provide a c-based datatype to supersede PyOpenCL's and PyCUDA's GpuArray. This datatype would be used as the gpu-side container by PyCUDA, PyOpenCL, Theano, and hopefully other projects too. It will extend the existing GpuArray by supporting strided memory layouts like numpy's ndarray. For ongoing design discussions - feel free to join the list: http://lists.tiker.net/listinfo/gpundarray James Bergstra On Tue, May 17, 2011 at 11:52 AM, Ralf Gommers wrote:

...

2011/5/17 Stéfan van der Walt

...

Hi all,

As you may know, we now have a GSoC student working on a new backend engine for scikits.image:

Sounds great! Perhaps Pieter's blog can be added to Planet Scipy for some more exposure?

Cheers, Ralf

...

http://pythonimaging.blogspot.com/2011/04/gsoc2011.html

One of these will be a PyOpenCL-backend, which I hope will be able to leverage this PyOpenCL-subproject:

https://github.com/inducer/compyte

Cheers Stéfan

-- http://www-etud.iro.umontreal.ca/~bergstrj