High CPU Usage when waiting for V-Sync

[jwagner]

I have an application that produces new data in a thread and then triggers a rerender via ctx.request_repaint();. Fresh data is produced at about 100 Hz. I'm using eframe with the WGPU backend.

The main thread is using a lot of CPU for rendering even when the ui is blank. On Mac about 30% CPU, on Linux it's 100% when running with the display at 60 Hz and vsync on. A perf profile shows it's mostly spinning in some nvidia shared objects, probably waiting for time to pass until the next vsync.

The exact behavior changes based on many factors. If I ctx.request_repaint_after(Duration::from_micros(1)); the CPU usage drops by ~95% but the frame rate is only 45 fps. If I set the display to 120 Hz (display refresh is happening more often than request_repaint), the I observe the same drop.

When I turn off vsync cpu usage also goes down, but the frame rate is unbounded and there is the potential for tearing which isn't ideal either.

If I had to guess the main thread starts spinning when the presentation queue of the Vulkan backend (in the Linux case) is full. Is there a portable way to avoid this that doesn't involve a huge latency penalty?