Skip to content

Reduce latency and remove latency policy and render time estimator settings

Xaver Hugl requested to merge work/zamundaaa/latency-policy into master

As KWin measures render times properly now, these settings and estimations should no longer be necessary, so this commit replaces them with one hardcoded algorithm that should prevent most if not all dropped frames and reduce latency.

Draft because I did some latency measurements and the results are weird:

  • with FIFO, master is at 50ms and this MR is at 41.6ms (so a reduction of exactly one frame with my 120Hz display), which is amazing
  • with Mailbox (~800fps with a test client that doesn't tax the GPU a lot), master is at 32ms and this MR is at 28ms. That's only the median however, there's a lot of spikes to 49ms (which is exactly 2.5 frames more) that master does not have and that I can't explain that yet. There's no dropped frames in the commit thread, and no huge scheduling hickups that master would be immune against

It would also be great to test this on weaker hardware to make sure this doesn't regress things there.

Merge request reports