OpenCL for Python

Question

I'm looking for a good OpenCL wrapper\library for Python, with good documentation. I tried to search some... but couldn't find one good enough.

Answer 1

The most popular and best documented option seems to be PyOpenCL. It claims to be a complete wrapper for OpenCL and the documentation looks good.

Answer 2

Both CLyther and PyOpenCL look nicely documented to me.

Answer 3

pycl is a ctypes binding to OpenCL (hosted on bitbucket)

Its primary goal is simple: wrap OpenCL in such a way that as many Python implementations can use it as feasible. It is currently tested on CPython 2.{5,6,7}, 3.2, and PyPy 1.5. It is known to largely not work on Jython, whose ctypes library is still immature.

To achieve this, we eschew extension modules and dependencies outside of the standard library. Ideally things like NumPy arrays and PIL images should Just Work, but they shouldn't be required

It's currently pretty new. I've been playing with it recently and it's working nicely. Not done timings compared to PyOpenCL, but the performance appears to be perfectly reasonable

I didn't work for me with OS X 10.6's OpenCL 1.0 (and an AMD Radeon 5870, see here), but works perfectly under 10.7 and OpenCL 1.1

As for documentation, it's a fairly direct bindings to the C libraries, so any documentation those will apply. Also "Adventures in PyOpenCL" is good, and applies to almost all bindings

Answer 4

We are developing opencl4py, higher level bindings. This project uses CFFI, so it works on Pypy.

The major issue we encountered with pyopencl is that 'import pyopencl' does OpenCL initialization and takes the whole virtual memory in case of NVIDIA driver, preventing from correct forking and effectively disabling multiprocessing (yes, we claim that using pyopencl disables multiprocessing at least with NVIDIA). opencl4py uses lazy OpenCL initialization, resolving this "import hell".

Later, it gained some nice features like super easy binary program caching, etc. Unfortunately, the documentation is somewhat brief. The best way to learn how it works is go through the tests.