1920761 - Make whole cell store buffer tracing more efficient

Status: RESOLVED FIXED

(Core :: JavaScript: GC, task, P3)

Description

[Jon Coppeard (:jonco)]

Currently we iterate through the whole cell store buffer using a linked list, from the most recently allocated to the least recently allocated. However the elements of this list are all allocated from a LifoAlloc so mostly follow each other in memory. We can use the LifoAlloc to iterate them and avoid pointer chasing.

The linked list is still necessary for sweeping.

Comment 1

[Jon Coppeard (:jonco)]

Attachment: Bug 1920761 - Part 1: Remove constant members from ArenaCellSte r?sfink

These should be actual constants not members which are const.

Comment 2

[Jon Coppeard (:jonco)]

Attachment: Bug 1920761 - Part 2: Step directly through the elements of the whole cell store buffer rather than using a linked list r?sfink

Comment 3

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/748d63c0f0a2
Part 1: Remove constant members from ArenaCellSte r=sfink
https://hg.mozilla.org/integration/autoland/rev/01a47e72987a
Part 2: Step directly through the elements of the whole cell store buffer rather than using a linked list r=sfink

Comment 4

[Atila Butkovits]

https://hg.mozilla.org/mozilla-central/rev/748d63c0f0a2
https://hg.mozilla.org/mozilla-central/rev/01a47e72987a

Comment 5

[Alex Finder]

(In reply to Pulsebot from comment #3)

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/748d63c0f0a2
Part 1: Remove constant members from ArenaCellSte r=sfink
https://hg.mozilla.org/integration/autoland/rev/01a47e72987a
Part 2: Step directly through the elements of the whole cell store buffer
rather than using a linked list r=sfink

Perfherder has detected a browsertime performance change from push 01a47e72987a3b02eb1d54a9a2011504f3b8038b.

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)	Performance Profiles
21%	expedia LastVisualChange	linux1804-64-shippable-qr	fission warm webrender	2,772.56 -> 2,184.71	Before/After
17%	expedia PerceptualSpeedIndex	linux1804-64-shippable-qr	fission warm webrender	1,265.24 -> 1,052.40	Before/After
15%	expedia LastVisualChange	macosx1015-64-shippable-qr	fission warm webrender	2,335.18 -> 1,978.94	Before/After
14%	expedia SpeedIndex	linux1804-64-shippable-qr	fission warm webrender	725.18 -> 620.41	Before/After
14%	expedia PerceptualSpeedIndex	macosx1015-64-shippable-qr	fission warm webrender	1,164.41 -> 1,005.68	Before/After
...	...	...	...	...	...
4%	expedia ContentfulSpeedIndex	linux1804-64-shippable-qr	fission warm webrender	433.72 -> 415.76	Before/After

Details of the alert can be found in the alert summary, including links to graphs and comparisons for each of the affected tests.

If you need the profiling jobs you can trigger them yourself from treeherder job view or ask a sheriff to do that for you.

You can run these tests on try with ./mach try perf --alert 2372

For more information on performance sheriffing please see our FAQ.