Original Post
I've written a 2D graphics class library using textured quads with position/colored vertices and an orthoganal view, and I'm having some trouble with small errors in rendering. Because all textures have to be powers of two, I have a texture wrapper class that accepts input about the size of the sprite within the texture and the number of frames (for example, in a walk cycle for a 32x32 sprite, you could just have the 4 frames in one 64x64 image), and can provide the texture coordinates for the four corners of a given frame. Unfortunately, there's a bit of error, since this isn't exactly what DX was made to do. The only way I can think of to fix this would be to use a transparent buffer of a couple pixels around each frame, which could probably be done only by modifying the texture class, but it would also end up wasting a good deal of memory, and it would only half-solve the problem. Anyone have any other ideas how I can get this thing pixel perfect without giving up untransformed vertices?