I've hit on an interesting new (to me at least) method for rendering batches of 2D quads. I load a texture map with several related tiles on it, for example, all the different pictures for a particular character in the game. I build a vertex buffer, with a different quad for each tile in the texture map. Each quad has the same coordinates, same everything, except the uv texture coordinates, which map to the quad's tile. So tile 0 in the texture maps to quad 0, tile 1 to quad 1, and so on. You move the quad using matrices. The result is, that on each render cycle, you never need to create a vertex buffer, it's already created. Where you really get an advantage is if you want to draw several of these sprites, all in different stages of their animations. If you draw them together, you don't need to change textures, because they're all using the same texture. You don't need to change vertex buffers, because they're all using the same buffer. And, as I said, you don't need to create or alter any vertex buffers; they're already made, and movement is accomplished through transformations. The result is that you can print the sprites out pretty efficiently, without having to change any render settings. Does anyone think talking about this might make a good gamedev article?

You need to batch them as well. Drawing one at a time will be slow. Its much better to have a few hundred in a vertex buffer and use one draw for them all, and using matrix transforms sort of gets in the way of batching.

Quote:Original post by Stru
You need to batch them as well. Drawing one at a time will be slow. Its much better to have a few hundred in a vertex buffer and use one draw for them all, and using matrix transforms sort of gets in the way of batching.

But if you don't use transforms, don't you have to keep remaking your vertex buffers again and again? Isn't that slow in and of itself?