Radeon Gallium3D Picks Up A Nice Performance Optimization For iGPU/dGPU PRIME Setups
With this latest code for Mesa 21.3-devel to the RadeonSI driver, copies from the discrete GPU to the integrated GPU for presentation on the screen are now done asynchronously. These async copies are done using the SDMA engine and in turn allows the discrete GPU to handle more work. An Async compute context is also used as a fallback to async SDMA copies. The patch summed it up quite simply, "Doing this copy using SDMA frees up the dGPU to do more interesting things while the copy is happening; for instance the rendering of the next frame."
In the merge request for this functionality, AMD's Marek Olsak noted that this is "significant" for PRIME setups and should end up being a clear performance win for those with AMD Radeon PRIME setups where needing to copy from the dGPU to iGPU (as opposed to the systems that can mux the output straight from the dGPU to the screen).
Mesa 21.3 is shaping up to be a great quarterly feature update that soon is going into feature freeze and should see its stable debut in roughly one month's time.
38 Comments
