For the GPU, we had to rewrite everything from the ground up, and we still have work to do. (This is also in part why it’s so much faster than a CPU - it is a way simpler renderer.) Every feature that we want to support on the GPU, we need to code separate and try to make sure, as much as possible, that it produces a similar result to the production renderer. Whatever we have to do for the Xeon Phi, the efforts involved will be much much much less.