I am a software developer based near Melbourne, Australia. Most of my days are spent coding in Python and JavaScript, commercially as an engineer at Mozilla as well as for a variety of open-source projects. I also maintain a strong interest in logic programming, mainly as a result of my doctoral thesis. Read more about me and check out my curriculum vitae if you want to know more.


Tue, 06 May 2014

PyPy.js: Now faster than CPython

OK OK, I couldn't resist that title but it probably goes a bit far. Let me try for a little more nuance:

PyPy.js: Now faster than CPython, on a single carefully-tuned benchmark, after JIT warmup.

It has been the better part of a year since I first started hacking on PyPy.js, an experiment in bringing a fast and compliant python interpreter to the web. I've been pretty quiet during that time but have certainly been keeping busy. Some of the big changes since my previous update include:

The result, while still rough in a lot of places, is nonetheless an exciting milestone: the full PyPy interpreter, compiled down to javascript and re-targeted to emit asmjs from its JIT, running the pystone benchmark faster in a browser than the native CPython interpreter runs it on bare metal.

Continue reading...


Thu, 08 Aug 2013

PyPy.js Update: A Proof-of-Concept JIT

Two weeks ago I hatched a plan to port Python to the Web Platform by using Emscripten to translate PyPy into JavaScript. My hope is to produce something than can run in the browser with performance comparable to a standard Python interpreter, a hope which hinges on two key ingredients:

  • PyPy's powerful just-in-time compiler, which can optimize the hot loops of your program into efficient native code.
  • The recent work on asm.js, a low-level subset of JavaScript that can act like an efficient virtual machine in the browser.

By translating the PyPy interpreter into asm.js code, and by having its JIT backend emit specialized asm.js code at runtime, it should theoretically be possible to have an in-browser Python implementation whose hot loops perform within a factor of two of native code.

I'm excited to report a small but important milestone on the road to making this a reality.

It's certainly not a full Python interpreter, and it comes with many caveats and question-marks and todos, but I have been able to produce a simple demo interpreter, with JIT, that approaches the theoretical factor-of-two comparison to native code under some circumstances. There's a long way to go, but this seems like a very promising start.

TL;DR? Feel free to jump straight to the important graph.

Continue reading...