actually,the code to do all the the splat blender is ridiculously easy, and should run pretty fast, since its all done in ram, and has nothing to do with the graphics engine til you lock a texture or quad. In essence it takes the old school approach of "partying on the bitmap", then sending the results off to the "system" for display. You draw what you want (party) in the blending buffer (the bitmap), then send that off to directx or openGL as the case may be. note that since it doesn't use the graphics engine's data structures except for the copy operations, a single blender engine can be used with any graphics system by simply writing the appropriate copy routines, and using the graphics system's data structures for the texture and quad circular buffers.

the purpose of the circular texture and quad buffers is to guarantee there will be an unused mesh or texture available to lock and copy. if you just use a single texture for example, you have to wait for the card to draw a quad before you can lock the texture and copy the contents for the next quad. same idea with locking quads to height map the y values of the vertices.