Hi, I''m working on a terrain egine and right now I got a brutforce version up and runing (e.g. drawing every vertex, always). It''s a bit slow thought to I need to speed it up. I can see two ways of doing this, CVA and Octree. I got to beasic idea on how both work, but how make make them work together? In the octree example I have they simply store the vertexes in each node (well, end node) and then draws it using normal glVertex3f calls. I guess I could many CVAs, but wouldn''t that hurt performance? Is there a way to tell what parts of the CVA to draw, whitout needing to send it again? Or would it be better to just use normal vertex arrays? Thanks in advance Leon Ljunggren

You just need to make four big arrays. One for vertices, colors, normals and texcoords (if you''re using all four). For each area you want to draw(each leaf node in a quad/octree) you will need an array of indices. Then for each node set up your indices to draw a ''square'' inside the large array.

You''ll need an array of indices:

if you use triangle strips, you would supply pairs of vertices

So the first index would be 0 (that''s the easy bit )

Next you would need an index for the vertex ''below'', on the next row. Normally in a single CVA this would be : 0 + n, where n is the width of each row, eg:

0, n, 1, 1+n, 2, 2+n, 3, 3+n

BUT, remember this will be INSIDE a bigger array so for the next row add the width of the larger array (be careful not to go beyond the edge of the large array)

When you have all the indices set up to render a square area, you will need to supply an address to the pointer functions.

Set all the pointers (glTexCoordPointer, glColorPointer, glNormalPointer, glVertexPointer) to the ''top-left corner'' address (they expect a memory location which is at the start of a vertex array) of the VA you want to draw.

glDrawElements also has a start address, use the same address as above

And there you go, you should now be able to draw any square(or rectangular!) area of an array.

This saves you setting up a separate array for each node when you implement your octree.

Hope at least some of that made sense, I know it works coz my engine does it this way.

Good luck

______________________________

Don''t ask me, I''m just the code monkey

first before u do anything make sure u are geometry bound.
how to?
make the window small eg 200x100 does the framerate increase dramatically? if so youre fillrate bound and doing CVA,VAR whatever aint gonna help matters much.

http://uk.geocities.com/sloppyturds/kea/kea.html
http://uk.geocities.com/sloppyturds/gotterdammerung.html

zedzeek: Yes it's geometry, if I render every point of a 32x32 heightmap I get a fps of around 200, rendering every point of a 1024x1024 heightmap and I'm lucky if I get 0.5 fps.
But thanks for the tip, it could have been it.

BlueCat: Thanks, I think I understand the general pricipe (basicaly you make a CVA, then tell it which vertises to draw this time, correct?), but I don't really understand how to do it.
Lets say I wanted to render the trinagles in the picture. My vertex array would look something like this (3D, just happen to have y = 0): { {0,0,0}, {1,0,0}, {0,0,1}, {1,0,1}, {0,0,2}, {1,0,2}.... {1,0,6}, {2,0,6}, {1,0,5), {2,0,5}, {1,0,4}, {2,0,4}... etc

Now lets say I want to render the trinagles from the (3,3) and forward. How would I do that? I din't quite understand how that indices array was done. Did you mean that I would make a array like. {35, 49, 37, 47, 39, 66... etc?
How would I then pass it so that I can use it to read the corrent verteises from the CVA?
Wouldn't passing this array defete the point with a CVA? E.g wouldn't it be better just to use multiple CVs, one fore each node?


Thanks

[edited by - Spearhawk on October 23, 2002 6:57:55 PM]

OK, firstly I have tried both ways and one big array gave me much better performance. I too have a 1024x1024 heightmap with octree nodes approx 20x20x20 so that would be a lot of separate arrays! No reason not to try it yourself though

I wasn''t very clear about index arrays, the idea is that you have an array of vertices (you can also have arrays for normals, colors and texture coordinates but you should have at least a vertex array) which you fill with x,y,z values.

Make an array called myvertexarray, eg 9x9 = 81 vertices

The 1st X,Y and Z values would be at:

myvertexarray[0],myvertexarray[1],myvertexarray[2]

The X,Y and Z values at the start of the 2nd row would be at:

myvertexarray[27],myvertexarray[28],myvertexarray[29]

As an example, imagine you just wanted to draw one triangle, you could do it this way:

glBegin(GL_TRIANGLES);
glVertex3f(myvertexarray[0],myvertexarray[1],myvertexarray[2]);
glVertex3f(myvertexarray[3],myvertexarray[4],myvertexarray[5]);
glVertex3f(myvertexarray[27],myvertexarray[28],myvertexarray[29]);
glEnd;

or this way:

create an index array (let''s call it ''myindexarray''), fill it with these 3 values:

0, 1, 9

To render the triangle:

glVertexPointer(3, GL_FLOAT, 0, @myvertexarray); (Search the OpenGL docs if you want to know what all the parameters are)

glDrawElements(GL_TRIANGLES, 3, GL_UNSIGNED_INT, @myindexarray);

This will use the index array to look up the vertex data in the vertex array and send each one as though it were a glVertex call. Notice the ''GL_TRIANGLES'' which tells OpenGL to expect vertices 3 at a time.

To draw the entire array you would fill ''myindexarray'' sequentially with indices (3 at a time) to draw each triangle, you can also use GL_TRIANGLE_STRIP but wait until you have the triangles working



If you understood that bit, try this!



You should be able to see a red grid (4x4 vertices) at vertex 30. To render these you would need to put the following values in your index array:

30,31,39, 39,31,40, 31,32,40, 40,32,41, 32,33,41, 41,33,42,
39,40,48, 48,40,49, 40,41,49, 49,41,50, 41,42,50, 50,42,51,
48,49,57, 57,49,58, 49,50,58, 58,50,59, 50,51,59, 59,51,60

BUT what you actually need to do is use the address: myvertexarray + (30*3) in glVertexPointer and adjust your indices relative to that memory location

Like so:

0,1,9, 9,1,10, 1,2,10, 10,2,11, 2,3,11, 11,3,12,
9,10,18, 18,10,19, 10,11,19, 19,11,20, 11,12,20, 20,12,21,
18,19,27, 27,19,28, 19,20,28, 28,20,29, 20,21,29, 29,21,30

''WHY?'' I hear you ask, well there is a good reason. Those indices you put in your index array are OK when you only have 81 vertices, but your heightmap has 1048576 and indices are 16 bit values (You don''t really want to be using 32 bit indices now do you?).

So you can see, as I said in my previous post, you have to add the width of the larger vertex array (9) to reference the next row, you DON''T add the width of the smaller square (in this case 4)

Hope that helps

If I''ve made any stoopid mistakes, I''m sorry, it''s late



______________________________

Don''t ask me, I''m just the code monkey

quote:Original post by BlueCat
Hope that helps

Ya, thanks! I wrote a small sample program to test it and it works like a charm. Now on to the reall challange, making it work with my egine.
Thanks once again, it was really usefull.

>>zedzeek: Yes it''s geometry, if I render every point of a 32x32 heightmap I get a fps of around 200, rendering every point of a 1024x1024 heightmap and I''m lucky if I get 0.5 fps.
But thanks for the tip, it could have been it.<<

i said make the window smaller, not draw less points

http://uk.geocities.com/sloppyturds/kea/kea.html
http://uk.geocities.com/sloppyturds/gotterdammerung.html

Sorry, zedzeek. I misunderstood you post, when you siad fillrate I took it to mean the screen refresh rate. Which could be tested by drawing less points (e.g if it goes far faster than the screen fillrate, it''s not the refresh rate hindering it).
Anyway, rendering 1048576 vertises will requier an equal amount of function calls (twice or trice if color and texture). So eliminating as many of them as possible has to be good and even if it''s not, it''s much neater this way

id render such a terrain with triangle strip vertexarrays