Closed Bug 1533545 Opened 8 months ago Closed 6 months ago

Malloc contention on WrWorker threads is too high.

Categories

(Core :: Graphics: WebRender, defect, P3)

defect

Tracking

()

RESOLVED FIXED
mozilla68
Tracking Status
firefox68 --- fixed

People

(Reporter: jrmuizel, Assigned: nical)

References

(Blocks 1 open bug)

Details

(Whiteboard: [wr-q2][wr-april])

Attachments

(3 files)

Blocks: wr-67
Priority: -- → P3
Blocks: 1495170

Things to try:

  • Opt into per-worker-thread arena with jemalloc.
  • Lower the thread count from num-cpus to num-cpus - 1 with a maximum of maybe 6 or 8. We get diminishing returns from having many threads due to this contention issue and the per-thread overhead of a few things like managing fonts.
  • Recycle the tile pixels.
See Also: → 1533546
Assignee: nobody → jmuizelaar
Assignee: jmuizelaar → nical.bugzilla
Whiteboard: [wr-q2][wr-april]
Attachment #9058516 - Attachment description: Bug 1533545 - Limit the number of WebRender Workers to 8. r=jrmuizel → Bug 1533545 - Limit the number of WebRender Workers to 8. r=kats
Pushed by nsilva@mozilla.com:
https://hg.mozilla.org/integration/mozilla-inbound/rev/76927efc43f5
Register thread local arenas in WebRender worker threads. r=jrmuizel
https://hg.mozilla.org/integration/mozilla-inbound/rev/d0260a2dac6c
Limit the number of WebRender Workers to 8. r=kats
Pushed by nsilva@mozilla.com:
https://hg.mozilla.org/integration/mozilla-inbound/rev/ed474f3cbc38
Register thread local arenas in WebRender worker threads. r=jrmuizel
https://hg.mozilla.org/integration/mozilla-inbound/rev/1edf22da902a
Limit the number of WebRender Workers to 8. r=kats
Pushed by nsilva@mozilla.com:
https://hg.mozilla.org/integration/mozilla-inbound/rev/4615f45cd4fe
Backout because of build bustage on a CLOSED TREE. r=me

wait, wasnt all these changes backed out in comment 8, and not relanded?

(In reply to Raul Gurzau (:RaulGurzau) from comment #9)

https://hg.mozilla.org/mozilla-central/rev/ed474f3cbc38
https://hg.mozilla.org/mozilla-central/rev/1edf22da902a
https://hg.mozilla.org/mozilla-central/rev/4615f45cd4fe

Isnt the third change the backout of the other two?

Flags: needinfo?(rgurzau)
Status: RESOLVED → REOPENED
Flags: needinfo?(nical.bugzilla)
Resolution: FIXED → ---
Pushed by nsilva@mozilla.com:
https://hg.mozilla.org/integration/mozilla-inbound/rev/a2f7510851a4
Register thread local arenas in WebRender worker threads. r=jrmuizel

Isnt the third change the backout of the other two?

I had (confusingly) only backed out one of the two commits.

Flags: needinfo?(rgurzau)
Status: REOPENED → RESOLVED
Closed: 6 months ago6 months ago
Resolution: --- → FIXED

Big perf improvements noticed! \0/

== Change summary for alert #20681 (as of Sun, 28 Apr 2019 05:19:44 GMT) ==

Improvements:

25% tsvg_static linux64-shippable-qr opt e10s stylo 59.99 -> 45.28
12% tsvgx linux64-shippable-qr opt e10s stylo 281.98 -> 248.04
10% tsvg_static windows10-64-shippable-qr opt e10s stylo 40.93 -> 36.67
8% tsvgr_opacity linux64-shippable-qr opt e10s stylo 103.70 -> 95.52
6% tsvgx windows10-64-shippable-qr opt e10s stylo 194.91 -> 183.11

For up to date results, see: https://treeherder.mozilla.org/perf.html#/alerts?id=20681

Blocks: 1567291
You need to log in before you can comment on or make changes to this bug.