Python version of Exercise08 much slower than C version about exercises-solutions

[tomdeakin]

Python isn't really built for number crunching, due to a few reasons which are explained pretty well in http://www.python.org/doc/essays/comparisons/. This example does the full serial integration in python.
As it says in the link, Python is good as a 'glue' language which is why it's nice to work with to call OpenCL.

Ultimately, it probably comes down to C being compiled with optimisations on, and Python being interpreted and run-time type checking.

We can either reduce the number of iterations down for the Python version or we have to pick a different way to work out pi, which would mean the serial code wouldn't be the same approach as the kernels.

[tomdeakin]

Just thought I would check, you were you referring to the code in Exercises/Exercise08/Python? This is plain serial code with no OpenCL. I think the Solutions/Exercise08 versions run in similar times which are the OpenCL versions so any difference should be in the pyopencl wrapping of OpenCL calls.

[simonmcs]

Yes I was talking about the plain C and Python on the host. I know it's down to C being compiled and Python interpreted, but with such a big difference in run-times between the two, a typical user might assume the Python version has hung and give up (I nearly did!).

Perhaps we should reduce the iterations on the Python side, or at least print a warning that this will take minutes etc?