21

So I have this simple code:

function Run () {
  var n = 2*1e7;
  var inside = 0;
  while (n--) {
    if (Math.pow(Math.random(), 2) +
        Math.pow(Math.random(), 2) < 1)
      inside++;
  }

  return inside;
}

var start = Date.now();
Run();
console.log(Date.now() - start);

And it will output some time around 335ms. That's pretty good. But, if I encapsulate the Run function like this:

var d = Date.now();
(function Run () {
  var n = 2*1e7;
  var inside = 0;
  while (n--) {
    if (Math.pow(Math.random(), 2) +
        Math.pow(Math.random(), 2) < 1)
      inside++;
  }

  return inside;
})();
console.log(Date.now() - d);

It will output 18319ms, which is much worse than the case before. Why is this ?

Also, if it matters, I'm running it on Chrome 26.0.1410.63, in the console. On node.js both snippets perform well on the console.

João Pinto Jerónimo
  • 9,586
  • 15
  • 62
  • 86
  • 1
    Probably function declarations are compiled (and optimized) differently from function expressions (like your IEFE) – Bergi Dec 03 '13 at 15:16
  • Can you seed Math.random() in js to make sure your using the same dataset? – buzzsawddog Dec 03 '13 at 15:17
  • @Bergi wouldn't the optimizing compiler figure out that it's really the same thing ? – João Pinto Jerónimo Dec 03 '13 at 15:18
  • @buzzsawddog I think we don't have that in JS – João Pinto Jerónimo Dec 03 '13 at 15:19
  • 1
    You need to look into that, it is conceivable that they will run different due to the `random` factor... – buzzsawddog Dec 03 '13 at 15:19
  • @buzzsawddog both of them run consistently with a low stdev, the only huge difference is from one to the other. – João Pinto Jerónimo Dec 03 '13 at 15:22
  • look at http://jsperf.com/function-declaration-vs-expression-2 – Arun P Johny Dec 03 '13 at 15:23
  • I would still look into it... http://stackoverflow.com/questions/521295/javascript-random-seeds shows some ideas to mock up a seedable random number generator. Run them with a constant to be 100% sure :-) That's just my idea. That will help you rule out the `random` part. – buzzsawddog Dec 03 '13 at 15:27
  • My Chrome version is 32.0.1700.19 and I have the same issue. – Igor Soarez Dec 03 '13 at 15:27
  • The jspref is giving a very negligible difference only – Arun P Johny Dec 03 '13 at 15:28
  • For what its worth, running in Firefox 25.0.1 I ran the first [23661, 22375, 21917] and the second [21707, 21833, 22415] – buzzsawddog Dec 03 '13 at 15:33
  • 5
    @Bergi has identified the basic problem. I think I've seen a duplicate of this problem before: the issue is that function declarations and expression are treated different *in the Chrome REPL console only*. Running on an actual page (like in jsperf) shows no difference, because function epxressions on pages are compiled in advance, while they are not compiled in the console REPL. Instead, it seems that REPL function expressions are interpreted. – apsillers Dec 03 '13 at 15:35

1 Answers1

3

There is no difference with function decleration and function expressions WRT to optimization, that would be ridiculous.

Console code in Google Chrome is wrapped in with statement like this:

 with ((console && console._commandLineAPI) || {}) {
      //Your code is concatenated here
 }

Because function declarations are hoisted, the former code will be effectively this:

function Run () {
  var n = 2*1e7;
  var inside = 0;
  while (n--) {
    if (Math.pow(Math.random(), 2) +
        Math.pow(Math.random(), 2) < 1)
      inside++;
  }

  return inside;
}

with ((console && console._commandLineAPI) || {}) {
  var start = Date.now();
  Run();
  console.log(Date.now() - start);
}

So the declaration is running outside with statement. In fact it is not valid syntax to have function declaration in a block, function declaration can only be a top level statement.

So anyway because of historical reasons V8 is nice and hoists it out instead of throwing syntax error:

var i = 3;

with({i:4}) {
    function test() {
        console.log(i);
    }
}
test();//logs 3 so it is obviously **not** under `with` influence

So because the declaration is not under with statement, it will run much faster. With statement is not optimizable* under V8 and also breaks lexical scoping.

*Not optimizable means the optimizing compiler will not look at the code instead only the generic compiler will generate code for the function. It is comparable to firefox's interpreter vs JIT mode. If you wish to know more about what language features disable optimization in V8, read optimization killers

Esailija
  • 138,174
  • 23
  • 272
  • 326