I agree with Chris about the need to first understand the current
bottlenecks. Besides, just what performance is going to be like when a
GPU, especially a mobile GPU, is context switching between OpenGL and
OpenCL is still largely unknown.
It really all depends on when mobile hardware gets these capabilities. Geometry shaders is another technology that would do many of the things that OpenCL interoperability can do. Neither of these are available on any mobile hardware today, although the OpenCL group is working on OpenCL ES, which may be the soonest technology available on mobile hardware, even if it would have to be partly implemented in software, using something like LLVM (http://llvm.org/devmtg/2009-10/OpenCLWithLLVM.pdf). So these things will happen. We just have to wait a while. In the meantime, it would be interesting to understand why your content has the performance it does. Can you sample it? It may be the interface to Typed Arrays being slow, which can be significantly improved. Or it may be in some heavily used WebGL API call, which might be optimizable some. We should make sure we have all the fast path logic possible in place before resorting to other technologies to solve performance problems.
begin:vcard fn:Mark Callow n:Callow;Mark org:HI Corporation;Graphics Lab, Research & Development adr:Higashiyama 1-4-4, Meguro-ku;;Meguro Higashiyama Bldg 5F;Tokyo;;153-0043;Japan email;internet:email@example.com title:Chief Architect tel;work:+81 3 3710 9367 x228 tel;fax:+81 3 5773 8660 x-mozilla-html:TRUE url:http://www.hicorp.co.jp, http://www.mascotcapsule.com version:2.1 end:vcard