Hi everyone,

If I'm calculating the bounding sphere of my mesh (using D3DXComputeBoundingSphere), should I do it before or after I have transformed the mesh using the world matrix? I'm guessing after, but I just want to check!

Thanks!

You can do it any time you wish, including once-only at load time. The great thing about a bounding sphere is that you only need to transform it's midpoint to get it to a different space.

Hi George,

This is a bit tricky, when you do it is not important, it's important that you transform/ rotate etc. at the right time.

I found out by playing around quite I while. Here's what I do now, in pseudo code:

- lock the vertexbuffer of the mesh

- applying scaling and rotations on the vertices, for this mesh instance

- compute boundingsphere for complete meshinstance (0,0,0 being in local space, mesh center)

- do the same for each submesh (attr of the d3dxmesh)

- translate the worldpostion for both the full mesh instance as all submeshes

Keep in mind future plans, if you're gonna use mesh instances, then you should not limit yourself by calculation it on mesh base, but do it on mesh instance base. For example I do this within my mesh instance class (using a specific mesh). Which also has the rotation, translation, etc. matrices for the mesh instance.

It's important to calculate it in local space and keep track of scaling, to prevent weird results.

I also calculate it for the submeshes (attributes of the d3dxmesh).

	/**
	 * Creates a new sphere that represents the given sphere modified by the given matrix.
	 *
	 * \param _sSphere The sphere to modify.
	 * \param _mMat The matrix by which to modify the given sphere.
	 * \return Returns a sphere that has been scaled and moved by the given matrix.
	 */
	CSphere LSE_FCALL CSphere::CreateSphereFromSphereAndMatrix( const CSphere &_sSphere, const CMatrix4x4 &_mMat ) {
		CSphere sRet;
		CVector3 vRow;
		_mMat.GetRow( 0UL, vRow );
		sRet.r = vRow.LenSq();
		_mMat.GetRow( 1UL, vRow );
		sRet.r = CStd::Max( sRet.r, vRow.LenSq() );
		_mMat.GetRow( 2UL, vRow );
		sRet.r = CStd::Max( sRet.r, vRow.LenSq() );
		sRet.r = CMathLib::Sqrt( sRet.r ) * _sSphere.r;
		sRet.c = _mMat * _sSphere.c;
		return sRet;
	}

The great thing about a bounding sphere is that you only need to transform it's midpoint to get it to a different space.

Be careful there, things get really hairy when dealing with non-uniform scaling.

I'd generally suggest that you avoid non-uniform scaling everywhere, for simplicity, but if you can't, bounding boxes start to look a whole lot better.

The great thing about a bounding sphere is that you only need to transform it's midpoint to get it to a different space.

Be careful there, things get really hairy when dealing with non-uniform scaling.

I'd generally suggest that you avoid non-uniform scaling everywhere, for simplicity, but if you can't, bounding boxes start to look a whole lot better.

This is true.

Since the OP talked about transforming the mesh using the world matrix, and since you generally don't see scaling in a world matrix, it can probably be discounted as a specific problem for this question, but it is still correct to highlight it.

since you generally don't see scaling in a world matrix