Source code

Revision control

Other Tools

<!doctype html>
<script src="util.js"></script>
<div id="x"></div>
<div id="y"></div>
* Test the performance of getElementById.
onload = function() {
document.getElementById("x").counterProp = 0;
document.getElementById("y").counterProp = 0;
var count = 1000000; // About 50-60ms
var ids = ["x", "y"];
for (var i = count; i > 0; --i) {
* We jump through some hoops here, because the JIT knows that
* getElementById is side-effect-free. That means we need to prevent it
* being dead-code eliminated or loop-hoisted.
* We avoid dead-code elimination by using the return value in a way that
* the JIT can't eliminate (i.e. by changing state on it that can be
* observed later). The use of += instead of just assigning is to avoid
* issues with the JIT detecting that later writes dominate earlier ones
* and eliminating the earlier ones by having all writes contribute to the
* state.
* We avoid loop-hoisting by making the arg to the getElementById call
* depend on the loop variable, which is obviously not loop-invariant.
* A sufficiently smart compiler could probably still figure out that
* there are only two possible return values, loop-hoist them both, and
* just grab the right thing inside the loop based on the value of the
* loop variable, and if that starts happening we'll need to see what we
* can do to fool the compiler into actually doing the many getElementById
* calls involved.
document.getElementById(ids[i % 2]).counterProp += i;