The core of this project is to test CPU cache locality by creating a large list of ints, but only multiplying every n'th member. For a much more in-depth article about what I'm trying to test, see:

http://igoro.com/archive/gallery-of-processor-cache-effects/

./mew.py <amount of data to generate> <number of items to skip> <number to * by>

For example, if we want to create a list of 128 numbers, and multiply every 16th number in that list by 3:

./mew.py 128 16 3

However, Python is slower than C. There are some pre-compiled binaries included in the bin directory. (compiled for OSX Lion x64, Ubuntu 12.04 x64) Running those binaries is the same as the Python version, but returns results about 100x faster.

cd bin
./mew_nosum_osx.bin 128 16 3

What does this thing do? It runs the mew-c program with several different conditions, and then writes the average runtime to a .csv file for each condition. Running mew_runner.py will make your CPU hot, and create a file called out.csv each time you run it.

The plot.py program will read the .csv files that mew_runner.py creates and will create some nice looking graphs. Run it with any number of csv files as args: python plot.py run1.csv run2.csv.

Short for Multiplexer http://en.wikipedia.org/wiki/Multiplexer

I estimate that a list of 65536 numbers requires about 2097224 bytes of RAM (~2MB). To get interesting results, try creating data sets that are smaller than your CPU L2 cache, and larger, and then compare different skip sizes for each.

The problem with python, is that it can be much slower than C. I decided to compile my python code down to C-bytecode using the amazing new pypy translate.py jit compiler.

See this gist for a performance comparison between pure python, and pypy:

https://gist.github.com/2253894

Using the Pypy jit translator allows you to compile your Python code down to raw, compiled C code as long as you write your python code "C compatible". See: http://doc.pypy.org/en/latest/coding-guide.html

Watch David Beazley's Pycon 2012 keynote for a really great overview of Pypy http://pyvideo.org/video/659/keynote-david-beazley

• Download the Pypy source (1.8, as of today) https://bitbucket.org/pypy/pypy/get/release-1.8.tar.bz2
• Expand this somewhere, I put it under ~/bin/pypy
• Run the translate.py app against the mew.py file with:

python ~/bin/pypy/pypy-pypy-2346207d9946/pypy/translator/goal/translate.py mew_sum.py -Ojit