Extreme JavaScript Performance(slideshare.net)
slideshare.net
Extreme JavaScript Performance
http://www.slideshare.net/madrobby/extreme-javascript-performance
8 comments
Maybe this was aimed as compiler writers (Google Closure), not the poor programmers?
Stupid microoptimisations.
Not only one should never optimize prematurely, and that is noted in the slides, but one should also never optimize like that (at least in the JS context). The interpreters/compilers/whatever will evolve for the readable cases, not for obscure idiocies, and programs using these techniques today will just run slower tomorrow.
On a large Javascript codebase these optimizations actually make a difference. It could take SECONDS off your page load time, which is huge.
These would be useful in something like the Closure Compiler http://code.google.com/closure/compiler/. Most seem like too much of a micro-optimization to use in real code though (especially given that they may change over time).
I'd be more interested in seeing these techniques applied to Javascript that is coded up with a framework (say jQuery for arguments sake).
For example, what is the performance penalty of:
For example, what is the performance penalty of:
var arr = [1, 2, 3];
for(var i=0; i < arr.length; i++){ //stuff }
// versus
$.each( arr, function(i, val){ // stuff } );
$("a").each( function(){
$(this).click(
$(this).find("img").hide();
);
});
// versus
$("a").each( function(){
var targetImage = $(this).find("img");
$(this).click(
$(targetImage).hide();
);
});Didn't even know about the function.toString() method, but it seems to that actually the non-firefox implementations deserve the wtf. (javascript:(function(){return 2*3;}).toString() => function(){return 6;}). Can't an interpreter optimize anymore?
The one I wish they had measured was createElement compared to innerHTML replacement.
In my job I was always shocked to see massive HTML strings appended or replaced in an objects innerHTML, and instead wrote lots of createElements and createTextNodes. Apparently, for performance's sake (and especially in loops) it is MUCH more efficient to use a string and replace methods to alter the content.
In my job I was always shocked to see massive HTML strings appended or replaced in an objects innerHTML, and instead wrote lots of createElements and createTextNodes. Apparently, for performance's sake (and especially in loops) it is MUCH more efficient to use a string and replace methods to alter the content.
See also: createDocumentFragment()
http://ejohn.org/blog/dom-documentfragments/
http://ejohn.org/blog/dom-documentfragments/
Author's page with PDF of slides
http://mir.aculo.us/ http://script.aculo.us/downloads/extremejs.pdf
Slideshare is terribly unresponsive at times, and the registration's capcha doesn't work properly.
http://mir.aculo.us/ http://script.aculo.us/downloads/extremejs.pdf
Slideshare is terribly unresponsive at times, and the registration's capcha doesn't work properly.
How come JIT compilers seem to have a problem with compiling try/catch blocks? Also, why does caching the window object increase performance by significant margins in some of the browsers?
It has to do with JavaScript's "scope chain" method of looking up symbol bindings. If something isn't in the current scope, it has to incrementally check each "scope" (such a bastardization..) for the binding it's looking for. If it works its way back up to the global (window) object's "scope" and doesn't find it, it gives up and returns undefined.
"with", try/catch, etc. all introduce another level of scope, which means anything you reference that isn't in that scope directly takes more time to look up. Yeah, it's pretty retarded.
"with", try/catch, etc. all introduce another level of scope, which means anything you reference that isn't in that scope directly takes more time to look up. Yeah, it's pretty retarded.
Why can't scope be resolved at compile-time?
I realize that things like window.x can be added at runtime, but local scopes ought to be simple to rule out.... as long as there are no calls to eval() .....
eval(), which can introduce local bindings at runtime from arbitrary, runtime-constructed strings....
Ugh.
And since aliases to eval can also be created, in any scope, from other dynamically eval()'d code strings, you can't even be sure that any nonlocal symbol won't resolve to eval.
So the only way you can be sure that a level of scope can be skipped in the lookup chain is if there are no function calls made on any redefinable nonlocal symbols.
If there were a subset of javascript where eval was a keyword instead of an identifier, then this would be easier.
I realize that things like window.x can be added at runtime, but local scopes ought to be simple to rule out.... as long as there are no calls to eval() .....
eval(), which can introduce local bindings at runtime from arbitrary, runtime-constructed strings....
Ugh.
And since aliases to eval can also be created, in any scope, from other dynamically eval()'d code strings, you can't even be sure that any nonlocal symbol won't resolve to eval.
So the only way you can be sure that a level of scope can be skipped in the lookup chain is if there are no function calls made on any redefinable nonlocal symbols.
If there were a subset of javascript where eval was a keyword instead of an identifier, then this would be easier.
If you don't use eval, with, and a couple of other things, V8 and most of the other good engines will actually optimize that scope level somewhat. Introducing eval, with, etc will make that scope level and any more therein to be much more dynamic/slow.
"local scopes ought to be simple to rule out"
If it's easy to do, it is probably only a matter of time until the major implementations include it. They seem to be competing on speed, after all.
If it's easy to do, it is probably only a matter of time until the major implementations include it. They seem to be competing on speed, after all.
I don't think I'd inflict ~~(1*"12.5") on someone for an insignificant gain when a parseInt() is obvious. (I also suspect that at least one of those implementations constant folded it.)