On Tue, Apr 21, 2009 at 4:20 PM, Knapp <magick.c...@gmail.com> wrote:
> > Have you seen this? > > http://arstechnica.com/open-source/news/2009/03/google-launches-project-to-boost-python-performance-by-5x.ars > yes, that's a nice goal. I hope it does well! > > Also this makes the python speed point very well. One last thing I > would add is that code speed is often not needed anymore. I have never > in the last 10 years had to worry about code speed while writing a > simple 2d game like pong, tetrus, defender, donky kong or anything > else like that. Yeah, most of the time you don't need it. But when you do, some problems become impossible to solve within a given time, so in those rare cases it's good to have ways available to you. > > > http://www.scipy.org/PerformancePython > > That link is interesting for the comparison down the bottom. One other comparison that could be added is a C version available in a library... which pygame does include a C version in pygame.transform.laplacian. It's not the same code, because it works on surfaces, not general matricies but pretty close. It's also interesting as laplace is also a commonly used benchmark for GPGPU processing. You can use a version written in python using pygpu or pycuda which looks almost identical to the numpy version... and get a 100x speed boost. The performance related thing is that there is a gsoc project this year to make SSE/mmx speedups to numpy at runtime.
