Hello all,
I have written a GPGPU application which uses the SGX chip on my DM8168 with OpenGL ES v2.0 . I send my data in using a texture and some triangles, and render to a Framebuffer Object with a Colorbuffer attachment. I then NEED to get the memory out of OpenGL so I use glReadPixels.
My initial tests however are not good and ended up being very inefficient. On a 800x600 matrix, I was running at 8 fps with my ARM load at 100%! I removed the glReadPixels call and acheived 30 fps easily with very low ARM load. Thus I am sure that my problem lies in extracting data from the SGX.
I fell upon a few different posts while researching and just wanted some suggestions/thoughts on what approach would be best. I see two possible solutions, using the PixMaps that come with my applications or double buffering the FBO (however, this one I am not so sure of!).
I was linked to this source: https://gitorious.org/tigraphics/sgxperf/blobs/1be449b21d588f2817a08e63f37e336b2f1a07fb/sgxperf_gles20_vg.cpp and I think it makes sense. I just wanted to see if anyone had any experience in this before I started digging. As I see it, I need to install the CMem module and use the cmem calls to specify memory for a PixMap... That should let me render directly to user space memory... I then need to pass/copy this memory to GStreamer.
Thank you for your time!
