[defen]

Back when I wanted to investigate the numeric performance of v8 I wrote a Runge-Kutta integrator + Lorenz attractor in C and in JavaScript as a simple-but-not-entirely-trivial benchmark. I was actually pretty impressed with how fast the v8 version was. On the downside, it's fairly non-idiomatic js and not that much nicer to look at than the C. Doing a million steps on my machine takes 0.65 seconds in node.js v0.8.4, 0.41 seconds in C compiled with gcc -O0, and 0.13 seconds with gcc -O3. Here is the code if anyone is interested. Note that it's not commented, not thread-safe, and doesn't free memory, so use at your own risk :)

https://gist.github.com/anonymous/5066486

    gcc strange.c rk4.c; ./a.out

    node strange.js

[masklinn]

> Back when I wanted to investigate the numeric performance of v8

Straightforward numerical computations really isn't a good jit benchmark, because numerical computations are by far the easiest thing to JIT, and JITted perfs are going to be much closer to AOT than in the general case (unless the problem can be vectorized an the AOT compiler is vectorizing, I don't think JITs can usually vectorize)

[defen]

Yup...I wanted to try to quickly measure just how good the JIT was for that kind of stuff, to see if we were getting to the point where it's feasible to do physics in the browser. Turns out it's "fast enough", and yet still 5x slower than equivalent stock C.