Is it possible to add asynchronous GPU operation for rendering, like flush_ready for flush_cb? It will increase the throughput if CPU and GPU can work in parallel, e.g. GPU is rendering one area while CPU can prepare for next drawing, once GPU finishes it can be started immediately for next rendering. As CPU rendering and GPU rendering might be mixed, it must wait for GPU completion if next drawing action is performed by CPU.
I like the idea. We need to carefully place the “synchronization points” but it can really work and provide some speed up.
Have you already made some experiments with it? If so can you open an issue on GitHub or send a Pull request?