I'm interested in (1) using my graphics card for computations and (2) low-level shader code, and I'm looking for guidance on how I would go about doing the following operations with libdrm (sans Mesa):

* client-side initialization * loading textures into the GPU * loading a bytecode shader into the GPU * creating a texture on the GPU without uploading data * running the shader using the uploaded textures as inputs and the non-uploaded texture as the output * transferring the output texture into main memory * tear-down, finalization

The GPU in question is a Radeon RV730, if any hardware-dependent discussion is needed. I've tried to understand the workflow by looking at the Mesa and Plymouth sources and the documentation I've found for the DRI/DRM, but I could still use some guidance.