1371646 - Amazon access book result navigation profile

Status: NEW

(Core :: Layout, defect, P3)

Description

[Filip Andrei]

1.Access Amazon
2.Search "Lord of the rings"
3.Scroll down to a Book result
4.Start Gecko profile
5.Access book result
6.Share profile



Gecko profile:

https://perfht.ml/2r21Liz

Comment 1

[Bobby Chien]

Jonathan, could you help to have first profiling in this bug? Thanks.

Comment 2

[Jonathan Kew [:jfkthame]]

If I'm understanding this profile at all, it looks like we spend quite a bit of time in nsCSSRuleProcessor (particularly in RulesMatching), so perhaps Stylo will help here?

Besides that, there's a lot of JS running, and it appears to trigger layout flushes (e.g. by querying things like element.offsetHeight) quite a number of times.

Comment 3

[Naveed Ihsanullah [:naveed]]

Can we retest with Stylo enabled?

Comment 4

[Filip Andrei]

Hi Naveed,

Sure that we can but can you please point me how to enable Stylo because I didn't find it as a preference and we are not building Firefox locally.

Comment 5

[Jonathan Kew [:jfkthame]]

Using Nightly, you should be able to enable it by setting layout.css.servo.enabled to true in about:config (and restart the browser, I assume it's not a "live" switch).

Comment 6

[Filip Andrei]

Here is the profile with stylo enabled and for reference i used an interval of 1ms.If I should change this please let me know

https://perfht.ml/2gN8AUE

Comment 7

[Daniel Holbert [:dholbert]]

Thanks! I'm not super-familiar with the servo restyle functions, but if I'm correct in assuming that "ProcessPendingRestyles" is a reasonable thing to filter for (to find total restyle time), then that profile shows a 50% drop (203ms fewer) in restyle time as compared to the one in comment 0.

Original profile, filtered for ProcessPendingRestyles: https://perfht.ml/2tKSAU7  (412ms)
Stylo profile, filtered for ProcessPendingRestyles:    https://perfht.ml/2tKMinK  (209ms)

If this bug was originally prompted by us being slower than Chrome on the STR in comment 0, I'm curious how much slower & whether this apparent stylo speedup has narrowed the gap.

Comment 8

[Daniel Holbert [:dholbert]]

Filip, given that we've improved here (if you have stylo enabled), is it worth keeping this bug open?

It's unclear from comment 0 how bad the problem was here, and whether the 200ms speedup has narrowed the gap enough for this bug to be closed.

Comment 9

[Daniel Holbert [:dholbert]]

Calling this [qf:p3] for the moment, given that we've gotten some improvements (comment 7) and it's not yet clear in this bug how much of a performance gap remains here.

Comment 10

[Filip Andrei]

Hi Daniel,

I will create a benchmark with stylo enabled to compare the results because at this point the performance is the same without stylo enabled.
I will get back to you as soon as I have some results

Comment 11

[Daniel Holbert [:dholbert]]

OK, thanks!