These scripts were written as a learning exercise for me to learn OpenCL. They were inspired by the excellent _OpenCL Programming Guide_ by authors including Apple's Aaftab Munshi, AMD's Benedict Gaster, Intel's Timothy Mattson, NVIDIA's James Fung published July 2011
I've been a fan of Andreas Kloeckner's work since 2009 and PyCUDA. PyOpenCL provides a complete set of features for OpenCL 1.1(as of August 2011). PyOpenCL provides a fantastic degree of brevity, starkly contrasted to the burdensome boilerplate typically required in OpenCL applications written with the C API. I don't exploit all of the possible brevity, but during my learning
phase, I lean on the pythonic wisdom that explicit is better than implicit. This doesn't mean that I always go for the lengthiest solution possible, but I'm attempting to strike what feels like a natural middle ground between the possible verbosity(sometimes good for learners) and ease of coding(good for everyone).
An example of this is that I launch kernels by calling them as methods to the OpenCL program object rather than setting their arguments manually and
enqueuing the execution. Contents:
Hello World
Most of these scripts were written for execution on vertex, one of the most powerful on the UAB hospital network with a dual quad-core Xeon CPU and two NVIDIA GPUs, a Tesla 1060 and one minimal one for the display. There are thus two OpenCL platforms to choose from. Enumerating and querying
the devices is explored followed by a simple kernel for vector vector addition.
laplace
Since undergrad, i've repeatedly used a simple Jacobian iteration solution to the laplace equation in order to learn about programming. This method provides a nice balance
between a toy learning problem and one with actual use in scientific computing. I implement a steady state solver in two dimensional space using constants
provided in Chapra's textbook: _Numerical Methods for Engineers_ p. 856
August 2011
Robert L. Cloud
rcloud@gmail.com
http://www.robertlouiscloud.com