Improve Box2D performances with TeaVM

[fgnm]

Hi! JBox2D used by libGDX-GWT is commonly known to be a big bottleneck for any 2d physics game made in libGDX for Web, java implementation is just slow compared to the native one, but nowadays there's emscripten and the work you made with Bullet extension for TeaVM shows very great performances! So I was wondering if the same could be made with Box2D to improve performances too.

Thanks :)

[xpenatan]

Hey, sure. Can you create an example where its slow with gwt backend? It will be easier to see if emscripten will give performance boost.

[fgnm]

Currently on my not-so-powerful laptop you're examples already shows interesting scenarios. The Bullet example with 500 3D boxes run at same fps of the Box2DTestCollection here: https://xpenatan.github.io/gdx-html5-tools/teavm/gdx-tests/ (the test with the pyramid of boxes). When boxes are moving fps drops between 20/30fps, while the bullet test runs always between 40/50fps.. On a not-so-powerful phone (OnePlus 6T) the Box2D Pyramid boxes test cannot even run, like 2/3fps before freezing the entire browser, while the Bullet test is at least able to run at 20/30fps and no browser freezing. The only explanation I can give of such behavior is emscripten.. Tell me what do you think!

[xpenatan]

The test collection use ShapeRenderer and not sprites and that also drop FPS.

I got emscripten to compile box2d c++ classes. The next step is to figure out how to bind it to libgdx custom box2d classes and that may take some time.

[xpenatan]

@fgnm Got some c++ box2d to work =D. Need to bind more methods to get pyramid fully working

[fgnm]

Cool! Can't wait to try once finished to check performances :)

[xpenatan]

Hi @fgnm https://xpenatan.github.io/gdx-html5-tools/teavm/test-box2d/

Test it and see if there is any fps increase. There is one issue that the mouse joint is not deleting and I'll fix it later.