A lot of time spent in memory allocation/deallocation on the leaves benchmark
Categories
(Core :: Graphics: WebRender, task, P3)
Tracking
()
People
(Reporter: nical, Unassigned)
References
(Blocks 1 open bug)
Details
Profile: https://share.firefox.dev/46snaEW
The profile was coolected by Bas: using https://browserbench.org/MotionMark1.2/developer.html
- select motionmark/leaves
- "Keep at fixed complexity"
- set the boz that appears next to motionmark/leaves at 2500
- run
The largest bottleneck appears to be the content thread however we spend a fair amount of time on scene building and frame building with timings in the realm of 10-12ms and a large portion of both are in memory allocation/deallocation (mostly growing vectors).
Chrome is doing better than us at this test, but looks like something webrender should be able to handle pretty well.
Reporter | ||
Comment 1•1 year ago
|
||
Jeff noted that the memory measurements in the profile can inflate the cost of allocation and deallocations, so while it's probably still bad, it may be bad to a lesser extent than what the profile suggests. We'd need to profile using etw, perf or xcode to know for sure.
Updated•11 months ago
|
Description
•