Results and Try it Yourself
Here are the results from four benchmarks to give you a taste:
(Graph courtesy of Brendan Eich.)
The tests are:
- SunSpider, 1.83x faster (released by the WebKit team)
- SunSpider ubench, 22.5x faster (released by the WebKit team for testing Squirrelfish)
- Image Manipulation Demo, 6.46x faster (Change the contrast and brightness of images dynamically)
If you want to try these out for yourself, just snag a nightly of Firefox 3.1, open about:config, and set the following preference to true:
You should be, happily, in just-in-time tracing land. It’s still buggy (hence the reason for hiding behind the preference wall) but it should be good enough to handle most web sites.
What’s especially exciting is that this code is working on x86, x86-64, and ARM – which means that these improvements won’t be limited to just the desktop – you’ll be able to receive them on your mobile devices as well.
How Tracing Works
In simple terms tracing works by watching for commonly-repeated actions (such as loops, function calls, or type checking) and tries to optimize their resulting native code into the lowest number of actions. The premise is rather simple – and it’s an advance that we’ll probably see proliferate to many interpreters and engines in the upcoming years.
Some of the improvements made by tracing include:
- Function Inlining: Removing the overhead of function calls by simply replacing them with their resulting native code.
- Type Inference: Removing checks surrounding common operators (like “+”) when the types contained within a variable are already known. This means that the engine will have already pre-determined, for example, that two strings need to be concated when it sees the “+” operator.
The code for this project has come from a number of places – one of which is coming from some code contributed to Mozilla, from Adobe: Tamarin Tracing, specifically the nanojit code that’s able to work a lot of this just-in-time magic.
The work began just about 60 days ago, working with Andreas Gal of UC Irvine, to integrate the nanojit technology into Spidermonkey. You can hear more about the development from those that were involved: Andreas Gal, Mike Shaver, and Brendan Eich.
If you want to try running your own copy of TraceMonkey on the command-line, just follow the steps outlined here.
There is still a ton of work to be done. The incredible speed-ups that we're seeing are only just the beginning. A lot can be done to improve how registers are currently being allocated which will provide even more speed-ups.
One area that I'm especially excited about is in relation to Canvas. The primary thing holding back most extensive Canvas development hasn't been rendering - but the processor limitations of the language (performing the challenging mathematical operations related to vectors, matrices, or collision detection). I expect this area to absolutely explode after the release of Firefox 3.1 as we start to see this work take hold.