While fixing RT-29954, we again ran into a problem rendering into an RTT whose content size is bigger than the viewport into which we want to render. A partial fix (more of a workaround, really) was implemented in the ES2 pipeline to allow ParallelCamera to work for for systems without NPOT support -- specifically BeagleBoard.

This solution is incomplete in that it doesn't work correctly for PerspectiveCamera. Also, by setting the forcepowerof2 flag, I verified that if we ever find a D3D system without NPOT suport, it would break as well.

A longer term solution should involve having the RTT itself be aware of the viewport into which it is rendered. Note that this is an issue that affects more than just the camera transforms. See RT-26216, RT-26217, RT-26218 for other examples of why not having that information is a problem. Note that there could be other cases besides lack of NPOT support where an RTT could be bigger than the viewport into which you are rendering (e.g., reusing an RTT that is bigger than you need to avoid recreating it). It is possible we just haven't hit those cases yet in testing on BeagleBoard.