Code Speedup Experiments
This is a log of the speed up experiments that we are conducting.
Here are some experiments we could conduct. For each, we would compare against a pure Python implementation: Fib hand optimized to Java, no frames or other call setup support as seen in PyBaseCode.
- Fib pure Python, but with indy.
- Nested loops. Array multiplication?
Guard failure of a loop, call into interpreter (PyBytecode) with well-known offset and synthesized frame. (Limited CPS.)
- Etc
Related Speedups
Before venturing into these experiments, I tried to speedup the PyByteCode. You can view the page on that at CodeSpeedupExperiments/PyByteCode
Repository
All the work is being done at a mercurial repo at https://bitbucket.org/shashank/jython-experiments/