Implement tiled copies for AMD -> Intel PRIME
Submitted by Jason Ekstrand
Assigned to mes..@..op.org
When running the display server (X11 or Wayland) on an Intel GPU and using an AMD GPU as the primary, it wouldn't be terribly difficult to exchange X or Y-tiled images instead of linear ones. This would certainly improve performance on the Intel GPU and probably somewhat on the AMD GPU since anything is better than linear. The idea would be to replace the vkCmdCopyImageToBuffer with a compute pipeline that binds the PRIME image as a buffer and does the swizzling in the shader to get X or Y-tiling.