GPGPU with GLES API

Since BeagleBoards are heterogeneous platforms, why not use them to their full extent? Apparently for the time being the GPU block lies mostly useless and cannot assist with any computations due to non-technical reasons. Normally, OpenCL would be used for accelerating on the GPU, but taking the limitations discussed above it is impossible, yet there is another way! GPGPU acceleration with OpenGLES.

The GPU accelerators from PowerVR (series SGX500) are available in all BB's and a unified acceleration framework for them is the topic of this proposal. The targeted version of OpenGL ES is 2.0, but extending it to the future versions will not be a problem from a technical point of view as the newer versions are backwards compatible. The API of the library will allow for performing various mathematical operations, and will return the results to the user in the same format they input the data in.

Apart from the library, the project will explore the performance gains when using the GPU (compared to the CPU) and provide the user with guidelines on when it is feasible to put the BB's GPU accelerator to use and when to refrain from it.