Hello,

after finding out that my old frustum culling implementation didn't work at all except for the top-down view I was using for that one game I've been using it, I found this article. Now I'm trying to implement the "clip space approach", but the matrix being created there is from OpenGL. Now I know almost nothing about OpenGL, but enough to know that matrix differs in some aspects. The question is: What exactly is the difference, and what do I need to do in order to get a correct layout?

m_frustums[0].Set(m.m31 + m.m41, m.m32 + m.m42, m.m33 + m.m43, m.m34 + m.m44); // near
m_frustums[1].Set(-m.m31 + m.m41, -m.m32 + m.m42, -m.m33 + m.m43, -m.m34 + m.m44); // far
m_frustums[2].Set(m.m21 + m.m41, m.m22 + m.m42, m.m23 + m.m43, m.m24 + m.m44); // bottom
m_frustums[3].Set(-m.m21 + m.m41, -m.m22 + m.m42, -m.m23 + m.m43, -m.m24 + m.m44); // top
m_frustums[4].Set(m.m11 + m.m41, m.m12 + m.m42, m.m13 + m.m43, m.m14 + m.m44); // left
m_frustums[5].Set(-m.m11 + m.m41, -m.m12 + m.m42, -m.m13 + m.m43, -m.m14 + m.m44); // right

This is my code for setting the 6 planes for the frustum, using the OpenGL values from the article. It seems fine so far, but some of the planes appear to be mirrored on certain axis in certain positions... How do I need to change the values here to have it work for DirectX?

Have a look at Matrix Layouts, DirectX and OpenGL.

The big difference is that OpenGL is column major, while DirectX is row major.

The big difference is that OpenGL is column major, while DirectX is row major.

I kind of get that, from what I understand that means I'd have to transpose the matrix before using it, right? I tried just that, yet it didn't seem to do anything. I do understand the basics of matrix calculations, but not well enough that I can really meanfully debug this "mess". Any code-example, even pseudo, on how you'd convert your DX matrix to work here?

Frustum hmmm ? Not that I have a direct answer (I don't use OpenGL), but the normalized device coordinates are slightly different for OpenGL: in DirectX z goes from 0..1, in OpenGL from -1..1. I think this is relevant here.

Yep: There you go(GetPlanesFromMatrix in particular). At least the near plane looks different.

Yep: There you go(GetPlanesFromMatrix in particular). At least the near plane looks different.

Thats really odd, that was the matrix I've been using to being with.

			m_frustums[0].Set(m.m13, m.m23, m.m33, m.m43); // near
			m_frustums[1].Set(m.m14 - m.m13, m.m24 - m.m23, m.m34 - m.m33, m.m44 - m.m43); // far
			m_frustums[2].Set(m.m14 + m.m11, m.m24 + m.m21, m.m34 + m.m31, m.m44 + m.m41); // left
			m_frustums[3].Set(m.m14 - m.m11, m.m24 - m.m21, m.m34 - m.m31, m.m44 - m.m41); // right
			m_frustums[4].Set(m.m14 + m.m12, m.m24 + m.m22, m.m34 + m.m32, m.m44 + m.m42); // bottom
			m_frustums[5].Set(m.m14 - m.m12, m.m24 - m.m22, m.m34 - m.m32, m.m44 - m.m42); // top
			for(unsigned int i = 0; i < 6; i++)
			{
				m_frustums[i].Normalize();
			}

I thought, since this one didn't work, that there must be some other method available that is working. But now I've searched high and low, and almost all DX-articles feature exactly this set of planes... yet, it still isn't working for me. As is, its kind of like things are reversed in a strange way: If I'm at the origin, nothing shows up at all. If I move backwarsd, things start to pop up in front of my face. If I turn the camera upwards the y-axis, objects slowly disappear from beyond, and vice versa. I'm really getting desperate about this, I spent almost the whole day trying to figure it out. Any more ideas, or some sample applications that have a working application and show the process from begin to end?

Well, just to make sure there is nothing that can be messed up here, there is the rest of the relevant code: I'll assume my matrices are OK since everything draws just fine... start with the plane:

        void Plane::Set(float a, float b, float c, float d)
        {
			m_vNormal.x = a;
			m_vNormal.y = b;
			m_vNormal.z = c;
			m_d = d;
			m_vNormal.Normalize();
        }
		
		float Plane::Dot(const Vector3& v) const
		{
			return m_vNormal.x * v.x + m_vNormal.y * v.y + m_vNormal.z * v.z + m_d;
		}

		void Plane::Normalize(void)
		{   
			float norm = m_vNormal.length();
			if(norm)
			{
				m_vNormal /= norm;
				m_d /= norm;
			}
			else
			{
				m_vNormal = math::Vector3(0.0f, 0.0f, 0.0f);
				m_d = 0.0f;
			}
}

Those are the functions being used on the plane. Next, the box check:

		bool AABB::InsidePlane(const Plane& plane) const
		{
	        Vector3 vCenter = m_vCenter;
			//const Vector3& vNormal = plane.GetNormal();

	  //      if(vNormal.x >= 0.0f)  
		 //       vCenter.x += m_vSize.x; 
   //         else 
		 //       vCenter.x -= m_vSize.x;  

   //         if(vNormal.y >= 0.0f)  
		 //       vCenter.y += m_vSize.y;
   //         else 
		 //       vCenter.y -= m_vSize.y; 

   //         if(vNormal.z >= 0.0f)  
		 //       vCenter.z += m_vSize.z;
   //         else 
		 //       vCenter.z -= m_vSize.z;

	        float posDot = plane.Dot(vCenter);

	        return posDot >= 0.0f;
		}

for testing purposes, I'm treating all AABBs as dots, just to make sure. That code tests the frustum:

        bool Frustum::TestVolume(const IVolume& volume) const
        {
	        for ( int i = 1; i < 6; i++ )
            {
                if ( !volume.InsidePlane(m_frustums[i]) )
			        return false;
            }
            return true;
        }

And here is how I set up the frustum in the camera:

        void Camera::CalculateMatrices(void)
        {
	        m_mView = math::MatLookAt(m_vPosition, m_vLookAt, m_vUp);
			if(m_bOrtho)
				m_mProjection = math::MatOrthoLH((float)m_vScreenSize.x, (float)m_vScreenSize.y, m_near, m_far);
			else
				m_mProjection = math::MatPerspFovLH(m_fovy, m_vScreenSize.x / (float)m_vScreenSize.y, m_near, m_far);
	        m_mViewProjection = m_mView * m_mProjection;
	        m_frustum.CalculateFrustum(m_mViewProjection);
        }

So uhm... see anything obviously wrong here?

EDIT:

YEEESH!

        void Plane::Set(float a, float b, float c, float d)
        {
			m_vNormal.x = a;
			m_vNormal.y = b;
			m_vNormal.z = c;
			m_d = d;
                        m_vNormal.Normalize(); // wtf was I thinking?
        }

So that was the problem all along. When setting the plane, the normal was normalized, and thus the "d" component wasn't adjusted at all... I didn't normalize afterwards anyway, but thats why it didn't do you-know-what. Damn, I think I'll start with unit-test, at least for my math lib, that was a wasted day -.-

Devil's in the details. Well, glad you got it working.

For testing math: It sometimes helps to use a (working!) library to check against. Not really a unit test, but less to write.

For testing math: It sometimes helps to use a (working!) library to check against. Not really a unit test, but less to write.

Good advice, though in that case I was way to focused on where I though the problem arised. I was using the D3DX9-math libary at some points, but only to verify if my matrices where correct. I didn't even spent a tiny thought on that my planes could have a problem. I was deluded by the fact that I was using the plane class at other places sucessfully (like my water surface calculation), but fixing that plane issue also fixed a small visual error there, which I couldn't debug yet. Oh, well, debugging sucks sometimes :/