13:18 <zmike> alright, some facts: radeonsi has a swapchain size of 4 vs zink at 3, so that's a factor

13:18 <zmike> increasing the swapchain size on zink doesn't entirely fix the issue, but it does massively reduce acquire time so the diff is only about 2x

13:19 <zmike> what's maybe odd is that radeonsi gets XCB_PRESENT_COMPLETE_MODE_SKIP for complete notify about 50% of the time whereas in zink it's more like 2-3% of the time?

13:31 <daniels> koooooppppeeeerrrrrrr

13:33 <zmike> not sure how kopper explains the skip event diff

13:35 <daniels> depends on what it sets as the msc and when it submits … also FIFO thread if it’s using that as the present mode

13:35 <zmike> no thread since it's immediate

13:35 <zmike> msc is always 0 from vk

13:35 <zmike> dri actually calcs one

14:09 <daniels> hmm

14:09 <daniels> Xwl or Xorg?

14:20 <zmike> xorg

14:40 <zmike> more progress: it seems like vk presentation is actually a significant bottleneck due to 1) draining the event queue before present 2) forcing a full roundtrip for present instead of just emit+flush like dri does

14:41 <zmike> tweaking both of those on top of swapchain resize gets me to 17k fps