Either whole-MLP split; both build the same slice program (only the GPU fill differs), so either can share the other's program.
Either whole-MLP split; both build the same slice program (only the GPU fill differs), so either can share the other's program.