Handling GeoM projection on CPU may seem like a weird choice, given
how fast GPU is, but it pays off:
* You only have to do a very small subset of the actual matrix
multiply.
* You don't have to construct a matrix in the vertex shader.
* Six fewer float32 values per vertex.
* You do still have to do the matrix computation for each vertex,
though.
Signed-off-by: Seebs <seebs@seebs.net>