NewtonTreeCollision do not optimize

[FSA]

No it does still not optimize.
There you can see the collision mesh:

[Julio Jerez]

I am sure it is doing it.
you can set a break point on this function.
..\newton-dynamics\coreLibrary_300\source\core\dgPolyhedra.cpp

void dgPolyhedra::Optimize (const dgFloat64* const array, dgInt32 strideInBytes, dgFloat64 tol)

or check the history of that file.

I just test the version that is checked with one of the complex mesh in the sand box, and after I fixed that bug now the optimization is so much cleaner.
it manage to resolve some of the difficult area where it did no do a clean job, but now is does almost a perfect commove optimization.
this is how it looks like now.

optimizeBug.png (558.88 KiB) Viewed 3259 times

are you sure you are displaying the debug collision and no you visual mesh wire frame?
set a break point on that functions to see if the mesh is optimized.

[FSA]

I display the collision mesh as following:

Code: Select all

void DebugShowGeometryCollisionImp (void* userData, int vertexCount, const dFloat* faceVertec, int id)
{
   int ii = vertexCount - 1;
   dVector p0(faceVertec[ii * 3], faceVertec[ii * 3 + 1], faceVertec[ii * 3 + 2]);
   for (int i = 0; i < vertexCount; i ++) {
      dVector p1 (faceVertec[i * 3], faceVertec[i * 3 + 1], faceVertec[i * 3 + 2]);

      Vector3 s,e;
      s.x = p0.m_x;
      s.y = p0.m_y;
      s.z = p0.m_z;

      e.x = p1.m_x;
      e.y = p1.m_y;
      e.z = p1.m_z;

      g_pLine->AddLine(s, e, Color(1.0f,0.0f,0.55f));
      p0 = p1;
   }
}


[FSA]

And: Yes the function is called.

[Julio Jerez]

in this function ..\newton-dynamics\coreLibrary_300\source\physics\dgCollisionBVH.cpp

Code: Select all

void dgCollisionBVH::EndBuild(dgInt32 optimize)
{
   dgVector p0;
   dgVector p1;

   bool state = optimize ? true : false;

// place a break point here and check the m_faceCount before and after the call to  end
   m_builder->End(state);
   Create (*m_builder, state);
   
   GetAABB (p0, p1);
   SetCollisionBBox (p0, p1);

   delete m_builder;
   m_builder = NULL;
}

set the break point as see if the variable m_faceCount is reduced after call to builder->end().

[FSA]

Facecount before: 1200
Facecount after: 1200

No change.

[Julio Jerez]

It did not optimized anything, that means that state parameter is 0. you must be calling it with optimize = 0

[FSA]

I call it with 1.

[FSA]

Second picture. dgCollisionBVH:

[Julio Jerez]

That makes no sense, I tested with the same mesh that you sent me and it worked. I just tested again checkout a fresh version from SVN and it works.
do you have a binary executable test that I can run using a DLL? so that I can see what is wrong

[FSA]

I've made a small demo. I will send you a pm for more infortmaion in about 5 minutes

[Julio Jerez]

Ha I see why it is not optimizing, you are passing a unique material ID for each triangle

Code: Select all

PhysicBody* RigidBodyHandler::CreateLevelRigidBody(Model* CollisionModel, Matrix mTranslation, bool bOptimize)
{
..
      for(int i = 0; i < indexcount; i += 3)
   {
..
// here you are passing a unique material ID to each triangle
      NewtonTreeCollisionAddFace(collision, 3, (float*)vArray[0], sizeof(Vector3), i+1);
   }

}

the collision tree optimizes optimize the mesh by passes, It makes batched of face with equal Material ID, and optimizes the separate.
the in combine then all together, only face that share common Matirial ID can be optimized.

if you want that shape to be optimize dyou can eiteh pass the same id
NewtonTreeCollisionAddFace(collision, 3, (float*)vArray[0], sizeof(Vector3), ID);

or you can pass per Face IDs for example each face of the Box can have different IDs that you can se by applying different material; in your editor or modeler.

It is important for the engine to respect the boundaries of material ID, otherwise the collision respond will not be what the user expect.

[FSA]

Perfect! It works now Thank you.