GPU access on Windows Mobile

I am building an app for Windows Mobile 6.5 and I was wondering if there is any way to hardware accelerate various calculations. I would like to have the GPU do some of the work for the app, instead of relying on the CPU to do everything.

I would like to use C#, but if that is not possible, then C++ is just fine.

Thanks for any guidance!

EDIT-

An example of the types of calculations I want to offload to the GPU would be things like calculating the locations of 25-100 different rectangles so they can be placed on the screen. This is just a simple example, but I've currently been doing these 开发者_JAVA百科kinds of calculations on a seperate thread, so I figured (since it's geometry calculations) it would be a prime candidate for GPU.

To fully answer your question I would need more details about what calculations you are trying to perform, but the short answer is no, the GPUs in Windows Mobile devices and the SDK Microsoft exposes are not suitable for GPGPU(General-Purpose Computation on Graphics Hardware).

GPGPU really only became practical when GPUs started providing programmable vertex and pixel shaders with DirectX9(and limited support in 8). The GPUs used with Windows Mobile 6.5 devices are much more similar to those around DirectX8, and do not have programmable vertex and pixel shaders: http://msdn.microsoft.com/en-us/library/aa920048.aspx

Even on modern desktop graphics cards with GPGPU libraries such as CUDA, getting performance increases when offloading calculations to the GPU is not a trivial task. The calculations must be inherently suited to GPUS( ie able to run massively in parallel, and enough calculations performed on any memory to offset the cost of transferring it to the GPU and back ).

That does not mean it is impossible to speed up calculations with the GPU on Windows Mobile 6.5, however. There is a small set problems that can be mapped to a fixed functions pipeline without shaders. If you can figure out how to solve your problem by rending polygons and reading back the resulting image, then you can use the GPU to do it, but it is unlikely that the calculations you need to do would be suitable, or that it would be worth the effort of attempting.