Optimization help...

Odds are the normal.normalise is the costly bit, with a square root to calculate. Assuming your face normals are all pre-normalised then you could just divide your normal by the number of face normals you''ve added together.

Thanks for the quick reply.

Oddly enough, when I comment out the normalization I gain no noticeable speed whatsoever. I'm profiling it, and it takes the same amount of time.

It is probably the if statement inside the inner loop... But I'm not sure what to do to make it more efficient. Any tips?

[Edit: Spelling]

[edited by - James Trotter on June 6, 2004 3:36:51 PM]

quote:

---

Original post by PouyaCatnip
Odds are the normal.normalise is the costly bit, with a square root to calculate. Assuming your face normals are all pre-normalised then you could just divide your normal by the number of face normals you''ve added together.

---

No. Simple example:
n1 = (1, 0, 0)
n2 = (0, 1, 0)
n = (n1 + n2) / 2 = (1, 1, 0) / 2 = (0.5, 0.5, 0)
length(n) = sqrt(0.25 + 0.25) != 1

quote:

---

Original post by BitMaster
No. Simple example:

---

Dang. Silly me, I should have thought that one through.

OK, so the function calculates face and vertex normals. Do need face normals or are you just using them to get to vertex normals.

Anyways.. this can be optimized quite a bit. It''s basicly just 3 single loops. First go trough all vertices and set normal to zero. Then go trough all faces and calculate face normal and add it to coresponding 3 vertices. The last loop is simply to normalize all normals.

You should never let your fears become the boundaries of your dreams.

Hi, DarkWing.

Thanks for the insightful comment. I really hadn''t thought of doing it that way. I''ll try it out, and inform you of the results.
Much appreciated!

quote:

---

Original post by James Trotter
Hi, DarkWing.

Thanks for the insightful comment. I really hadn''t thought of doing it that way. I''ll try it out, and inform you of the results.
Much appreciated!

---

Show us the code after you implemented this and we''ll see if anything else can be done.


You should never let your fears become the boundaries of your dreams.

It works elegantly. Performance has increased ten-fold in that function. It''s no longer the bottleneck. But there are other places in the code that needs optimization. I''ll have a look at it, and if there''s anything I need help with I might be back.

Thanks a million, DarkWing!

quote:

---

Original post by _DarkWIng_
Show us the code after you implemented this and we''ll see if anything else can be done.


You should never let your fears become the boundaries of your dreams.

---

Sure, no problem. Here''s the code:

void CTerrain::CalculateNormals() {
	if (!Vertex || !Index) {
		Log->Output("Error in CTerrain::CalculateNormals(): Missing data!");
		return;
	}

	if (Normal) {
		Log->Output("Error in CTerrain::CalculateNormals(): Normals already exist!");
		return;
	}

	unsigned int numPolygons = (unsigned int)ceil(double(numIndex / 3.0));
	Vector3d *FaceNormal = new Vector3d[numPolygons + 1];
	Normal = new Vector3d[numVertex];

	int Counter = 0;
	for (unsigned int i = 0; i < numIndex; i += 3) { // Calculate the face normals

		FaceNormal[Counter] = Cross(Vertex[Index[i]] - Vertex[Index[i + 1]],
			Vertex[Index[i]] - Vertex[Index[i + 2]]);

		Normal[Index[i]] += FaceNormal[Counter];
		Normal[Index[i + 1]] += FaceNormal[Counter];
		Normal[Index[i + 2]] += FaceNormal[Counter];
	}

	for (i = 0; i < numVertex; i++)
		Normal[i].Normalize();

	delete[] FaceNormal;
}