speedups necessary about three.js HOT 17 CLOSED

Here is already something I was working on:
http://jsperf.com/threejs-canvasrenderer-improvements

from three.js.

Looks good!

Does it win a lot to have the empty in the same method? I mean, maybe just add3Points()/add4Points()?

I don't think core classes are going to change any time soon. I think the API is starting to get quite stable. We change from time to time materials related stuff, but the rest I think it's quite stable.

from three.js.

`changed it

from three.js.

Uh, I see... :)

from three.js.

I updated the add3points algoritm to use ternary operators and now It only uses half the time in both ie9 and chrome 10. I didn't test in firefox opera or safari.

http://jsperf.com/threejs-canvasrenderer-improvements/2

from three.js.

Same results here. 60% faster on Chromium 8 and Firefox 4b7. Great! :)

from three.js.

I've implemented the optimisations gero3. It's hard to notice the improvements becase the bottleneck it's the context API, but everything seems correct :)

from three.js.

I know but those where the bottlenecks without the context API

from three.js.

Yeah for sure. Thanks!
Now we just need the browsers to accelerate context...

from three.js.

Actually the profiler in google chrome 10 seems to say they already have in the new chrome build because the bottlenecks seems to be around the render function and the projectscene function

from three.js.

Also the use of instanceof seems to be an overhead

http://jsperf.com/test-instanceof/2

from three.js.

typetest is 65% faster in my case (Chromium 8 / Ubuntu)

from three.js.

Interesting... but, although I'm not sure we're using this still, we lose all the prototype chain - instanceof checks for the object and the objects it extends.

gero3: Do you think all these optimisations will still be valid when stuff like crankshaft lands?

from three.js.

I think they will as I tested these optimizations already in Google chrome 10 which has already the crankshaft engine. Also the crankshaft does the same optimizations as the TraceMonkey, only on a different way which already is in FF4 for a longer time.

from three.js.

I don't think you use any of these anymore but it is indeed a problem if persons want to add their own material or rendererable by extending the working solution

from three.js.

The two functions that together use the most cpu are now:
29.95%
setupMatrices
22.40%
renderBuffer

(of 41% spend in total).

from three.js.

Okay I'll look into to improve them

from three.js.