Could you add some info of the incorrect texture name in the exception? Which like "Tried to set an undefined texture: textureName". This may help to debug, thanks!

And another question:

PowerVR SGX uses TBDR, so sorting objs front-to-back makes no sense on such device.

But for Tegra(IMR)、Adreno(TBR)、Mali(TBR), will "Config.stateOrientedSorting = false" gain better performace?

I have heard about this before, but i have no device to test it now. Thanks for your test, and there are some details i will be interested in:

1, Will ART improve the cpu bound only(mainly)? For a game, which means AI、physics、draw calls...

2, What about the runtime-debuging?

3, How larger of the installed-apk will increase when enable ART?

4, From the result of QSort test of the article, seems there is also an obvious performance gap between JNI and ART.

class MyListener implements CollisionListener{

  @Override
  public void collision(CollisionEvent event) {
    int[] ids = event.getPolygonIDs();
    Object3D[] objs = event.getTargets();
  }

  @Override
  public boolean requiresPolygonIDs() {
    // TODO Auto-generated method stub
    return true;
  }
}

//Assume a car collide with 2 walls;
public void test(){

     MyListener listener = new MyListener();
   
    Object3D wall_1;
    wall_1.setCollisionMode(Object3D.COLLISION_CHECK_OTHERS);
   
    Object3D wall_2;
    wall_2.setCollisionMode(Object3D.COLLISION_CHECK_OTHERS);
   
    Object3D car;
    car.setCollisionMode(Object3D.COLLISION_CHECK_SELF);
   
    wall_1.addCollisionListener(listener);
    wall_2.addCollisionListener(listener);
    car.checkForCollisionEllipsoid(translation, ellipsoid, recursionDepth);
  }
 
When the car collides with both of wall_1 and wall_2 at the same time, the objs = event.getTargets() will contain wall_1 and wall_2, and the ids = event.getPolygonIDs() will contain some polygonIDs of wall_1 and some polygonIDs of wall_2, right?

If that's correct, how could i judge which polygonIDs belongs to wall_1?(and which belongs to wall_2)

Or should i assign 2 different listeners to 2 walls?

I found there is only a Object3D.scale(); Are there some methods like scaleX()\scaleY()\scaleZ()... etc?

NEON is ARM's Single Input Multiple Data (SIMD) architecture available on most Android ARM powered devices. It is a powerful engine for accelerating processing of media, augmented reality, gaming graphics and physics engines, DSP, etc.

And there is an  open source project of NEON targeting Android and Linux apps. Project Ne10 has been set up to provide a set of common, useful functions which have been heavily optimized for the ARM Architecture and provide consistent well tested behavior that can be easily incorporated into applications.

Maybe jPCT-AE can benefit form this.

ARM's introduce:
http://www.arm.com/community/software-enablement/google/solution-center-android/advanced-app-dev.php

Project Ne10:
http://projectne10.github.io/Ne10/

Cool!

First, I clear ZBuffer only per frame. Second, I test with 4 devices with differents GPU, all of them are incorrect.

1, The correct scene, a road with left wall and right wall.


2, The incorrect scene(far):


3, The incorrect scene(close), the road、left wall、right wall seems mixed together, and this position is 5\6 of the road(ring):


4, The incorrect scene(through the mixed-wall-road, the white cubes are waypoints which should upon the road,  but there is none of road\wall.


5, Going to fly a while, the start position of the road appear.