🔒 Locked

Vertex Array speed

Started by Schinizer Oct 4, 2010 at 10:38 PM 6 replies 1.8k views

Original Post

Oct 4, 2010 at 10:38 PM

Hi all,

Just to ask, is there performance difference for these 2 code?

Say that I am drawing 1000 quads. vert is the pointer to the vertices and texCoord is pointer to the texture coordinates. count is the number of vertices.

Method 1:

glEnableClientState(GL_VERTEX_ARRAY);glEnableClientState(GL_TEXTURE_COORD_ARRAY);glVertexPointer(2, GL_FLOAT, 0, vert);glTexCoordPointer(2, GL_FLOAT, 0, texCoord);glDrawArrays(GL_QUADS, 0, count);glDisableClientState(GL_VERTEX_ARRAY);glDisableClientState(GL_TEXTURE_COORD_ARRAY);

Method 2:

glEnableClientState(GL_VERTEX_ARRAY);glEnableClientState(GL_TEXTURE_COORD_ARRAY);glVertexPointer(2, GL_FLOAT, 0, vert);glTexCoordPointer(2, GL_FLOAT, 0, texCoord);for(int a=0; a<count; a+=4)     glDrawArray(GL_QUADS, a, a+3);glDisableClientState(GL_VERTEX_ARRAY);glDisableClientState(GL_TEXTURE_COORD_ARRAY);

karwosts

Oct 4, 2010 at 11:19 PM

You should definitely prefer method 1. #2 is exactly the same thing, but just creates a bunch more commands for the driver and gpu to process, there's no benefit and a definite potential loss of performance. That's one of the big reason to use these commands is so you don't have to spam commands for each primitive.

Hodgman

Oct 4, 2010 at 11:39 PM

You should prefer #1, but you should also be using VBOs instead of VAs ;)

Schinizer

Oct 5, 2010 at 1:22 AM

Ok so how do you handle texture bindings? Say there is 5 different textures among the 1000 quads and you have to do alpha sorting. Any references will be appreciated :D

GamerSg

Oct 5, 2010 at 1:50 AM

You could use couple of methods

i) Multi-texturing allowing you to bind more than 1 texture at once (usually for multi texturing the same polygon)
ii) Texture Atlas index (More suited but it's newer and might not be as well supported on older HW)

Ive not used any of the methods before so i cant get into detail

Rasmadrak

Oct 5, 2010 at 2:06 AM

Usually you sort stuff after materials (textures + shaders) and render all type A first, then type B, type C ... etc.

21st Century Moose

Oct 5, 2010 at 12:02 PM

Use something like this:

int firstvertex = 0;int numvertexes = 0;int lasttexture = -1;qsort (obj, count, sizeof (*obj), UserProvidedObjectSortFunction);for (int i = 0; i < count; i++, obj++){	if (obj->texture != lasttexture)	{	    glBindTexture (GL_TEXTURE_2D, obj->texture);            if (numvertexes)            {	        glDrawArrays (GL_QUADS, firstvertex, numvertexes);	        numvertexex = 0;	    }	    lasttexture = obj->texture;	    firstvertex += numvertexes;	}	numvertexes += 4;}if (numvertexes)    glDrawArrays (GL_QUADS, firstvertex, numvertexes);

I've just typed this in off the top of my head, by the way, so don't copy and paste it directly; it's intention is to illustrate the principle rather than to be strictly correct.

onfu

Oct 5, 2010 at 7:22 PM

When you need to render multiple textures, the way you organize your objects prior to OpenGl instructions is even more important. It's up to you to manage the way you supply the data, but always (as said above) try to lump polygons that sample from the same texture together.

If you take your 2nd example in the original post, the performance would be insanely bad if you were not only firing one draw cal per quad, but also binding backwards and forwards through your textures arbitrarily at the same time as well.

rule of thumb:
* minimize the number of instructions
* minimize state changes

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.