Posted by
Michael Bien on
Feb 05, 2011; 6:19pm
URL: https://forum.jogamp.org/JOGL-Applet-invokeAndWait-Exception-tp2425308p2432653.html
Hi William,
optimizing bottlenecks would be a start. The most difficult part however
is to find a way to parallelize them. For a proof of concept i would
recommend to pick the easiest tasks first :)
I didn't run many benchmarks yet but one of the first things i did with
jocl was to try to find out when the additional copy to GPU mem and back
pays of. E.g for simple sorting its already after about 30k elements
(
http://jogamp.org/jocl/www/gpusort.png). The chances are high that you
may indeed be able to kill some bottlenecks with jocl.
In best case there would be no copy at all and the physics engine would
pass all the data directly to GL... but thats hard.
The nice thing about physics is that you always find a way to put more
particles into the scene :)
btw we just deployed a few webstart demos
http://jogamp.org/jocl-demos/www/-michael
On 02/05/2011 05:25 PM, William Bittle [via jogamp] wrote:
> Thanks for the reply. Yeah the new JNLP applet configuration works great, I
> just wanted to test the fallback option. I'll just go ahead and use the new
> method in the meantime.
>
> Thank you very much for the comments. The project isn't very visible right
> now but I hope that changes. You are more the than welcome to reference the
> project. I'm not really doing anything stellar with JOGL but I did use an
> FBO to render to texture and then both a Vertex/Fragment shader in GLSL for
> the blur effect on the "HUD".
>
> As far as JOCL, I am interested to see if it can improve performance.
> Bottlenecks are collision detection, continuous collision detection, and
> collision resolution. I have played with some simple Java multithreading
> with inconclusive benefits. Did you have anything specific in mind (Vector
> operations, Matrix solving, etc.)?
>
--
http://michael-bien.com/