1070775 - 25% perf regression Scroll slow and laggy due to landing Bug 1002992

Status: RESOLVED WORKSFORME

(Core :: Graphics: ImageLib, defect)

Description

[Alice0775 White]

+++ This bug was initially created as a clone of Bug #1070772 +++

This bug is for #2 regression below.

Steps To Reproduce:
1. Open testcase index.html
2. Run the following code in ScratchPad with chrome privilege

   (function() {
    function scroll(event) {
      var top = content.document.documentElement.scrollTop ||
               content.document.body.scrollTop;
      
      if (top < 6000) {
        var evt = content.document.createEvent("KeyboardEvent");
        evt.initKeyEvent ("keypress", true, true, window, 
                              null, null, null, null, 
                              40, 0) 
        content.document.dispatchEvent(evt)
      } else {
        content.removeEventListener("scroll", scroll, false);
        content.scrollTo(0,0);
        alert(new Date().getTime() - start);
      }
    }

    content.focus();
    content.scrollTo(0,0);
    var start = new Date().getTime();
    content.addEventListener("scroll", scroll, false);
    scroll();
})();

Actual Results:
Scroll slow and laggy

There are atleast 2 regressions.
#1: 200% slower than previous, Regressed by Bug 994081
#2:  25% slower than the previous, Regressed by Bug 1002992

#1 Regression window(m-i)
Good: 6000 msec
L:\trunk\2014\06\inbound\firefox inbound 06-Jun-2014 1811 1402051313
https://hg.mozilla.org/integration/mozilla-inbound/rev/4511d9ac1000
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:32.0) Gecko/20100101 Firefox/32.0 ID:20140606104153
Bad: 13000 msec
https://hg.mozilla.org/integration/mozilla-inbound/rev/79624417d247
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:32.0) Gecko/20100101 Firefox/32.0 ID:20140606104333
Pushlog:
http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=4511d9ac1000&tochange=79624417d247
Regressed by: Bug 994081 - Convert imgFrame to store and act on a Moz2D SourceSurface instead of a Thebes gfxASurface


#2 Regression window(m-i)
Bad: 13000 msec
https://hg.mozilla.org/integration/mozilla-inbound/rev/e2714481c419
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:32.0) Gecko/20100101 Firefox/32.0 ID:20140606191036
Bad: 16000 msec
https://hg.mozilla.org/integration/mozilla-inbound/rev/47fcf180a0b9
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:32.0) Gecko/20100101 Firefox/32.0 ID:20140606192436
Pushlog:
http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=e2714481c419&tochange=47fcf180a0b9
Regressed by: Bug 1002992 - use a bare frame tree walker for image visibility


Regarding #1 regression, See Bug 1070772.

Comment 1

[Alice0775 White]

testcase
https://dl.dropboxusercontent.com/u/33246373/20140922031026.7z

Comment 2

[Alice0775 White]

Correct:

2. Run the following code in ScratchPad with chrome privilege

   (function() {
    function scroll(event) {
      var top = content.document.documentElement.scrollTop ||
               content.document.body.scrollTop;
      
      if (top < 11000) {
        var evt = content.document.createEvent("KeyboardEvent");
        evt.initKeyEvent ("keypress", true, true, window, 
                              null, null, null, null, 
                              40, 0) 
        content.document.dispatchEvent(evt)
      } else {
        content.removeEventListener("scroll", scroll, false);
        content.scrollTo(0,0);
        alert(new Date().getTime() - start);
      }
    }

    content.focus();
    content.scrollTo(0,0);
    var start = new Date().getTime();
    content.addEventListener("scroll", scroll, false);
    scroll();
})();

Comment 3

[Milan Sreckovic [:milan] (needinfo for best results)]

Timothy, is this regression expected?

Comment 4

[Timothy Nikkel (:tnikkel)]

No, bug 1002992 should have improved perf and produced the same results. It's possible that the new code is slower in some cases (usually it's 3+ times faster) at it's job, or maybe we are doing the image locking/unlocking slightly differently and it leads to worse behaviour (more decoding) in this case. I'll investigate.

Comment 5

[Timothy Nikkel (:tnikkel)]

(In reply to Timothy Nikkel (:tn) from comment #4)
> maybe we are doing the image locking/unlocking
> slightly differently and it leads to worse behaviour (more decoding) in this
> case.

I don't think it's this. We get the same list of visible images.

So I downloaded the inbound builds (mac) for the push for bug 1002992, and the push immediately before it and then ran the code from comment 2. I got the following results.

For the push immediately before bug 1002992:
4058, 3989, 4050, 3972, 4063, 3997. avg 4021.5, stddev 39.9337.

For the push of bug 1002992:
3912, 3932, 3905, 3942, 3946, 3921. avg 3926.33333, stddev 16.45195.

From these numbers it looks like, if anything, bug 1002992 had a positive impact on this testcase.

Alice, do you think you could double check your results?

Comment 6

[Alice0775 White]

I have to restart the browser for every measurement.

For the push immediately before bug 1002992 (m-i cset e2714481c419):
13288 14664 13667 14787 13821 13915

For the push of bug 1002992 (m-i cset 47fcf180a0b9):
17336 19085 17652 16949 18218 17269 

Without a doubt, bug 1002992 causes regression of the performance.

Comment 7

[Timothy Nikkel (:tnikkel)]

I redid my measurements, this time starting the browser fresh for each measurement. But I got pretty much the same results.

Without bug 1002992:
4020, 4127, 3981, 4004, 4069, 4003. avg 4034. stddev 54.29549.

With bug 1002992:
4037, 3908, 3957, 3987, 3980, 3920. avg 3964.83333. stddev 47.39374.


Other than me testing on mac, and you on Windows, what other differences might there be that might be causing this? I don't easy access to a Windows machine right now to test this unfortunately.

Comment 8

[Milan Sreckovic [:milan] (needinfo for best results)]

Bas, let's get measurement on Windows as well, see if we can reproduce Alice's results locally?

Comment 9

[Alice0775 White]

In local build
last Good: 3254fa7b1c51
First Bad: ce548c7d8fa2

Regressed by: 
  ce548c7d8fa2	Timothy Nikkel — Bug 1002992. Part 5. Use a custom frame tree walker to find visible and almost visible images instead of using the heavier weight display list infrastructure. r=mats

Comment 10

[Timothy Nikkel (:tnikkel)]

I don't have any better ideas for how to investigate this so I created some try builds with some changes that might have an effect. Alice, would you mind testing these?
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-ad4ea4469570/try-win32/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-83667965d4d1/try-win32/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-cf97714b9b48/try-win32/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-a673077f61e0/try-win32/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-bd4e45d9809f/try-win32/

Also, it might be interesting to see what your results look like it you don't restart between every measurement with a regular nightly. If they are the same or different that will be a useful data point. And I assume using a fresh profile doesn't change the results? Thanks!

Comment 11

[Alice0775 White]

ad4ea4469570
13501 12451 13186 12976 12971 13152

83667965d4d1
13077 13134 12906 12721 13174 12963

cf97714b9b48
13195 13200 12691 13252 13059 13153

a673077f61e0
13287 12803 13028 12998 13019 12853

bd4e45d9809f
14363 14841 14675 14634 14313 14096

Comment 12

[Alice0775 White]

FYI, Some improved in Nightly35.0a1, but not Aurora34.0a2.

Progression window(m-i)
Bad about16000:
https://hg.mozilla.org/integration/mozilla-inbound/rev/80164e15bd54
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:35.0) Gecko/20100101 Firefox/35.0 ID:20140912024553
Improved about13000:
https://hg.mozilla.org/integration/mozilla-inbound/rev/c232720a2847
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:35.0) Gecko/20100101 Firefox/35.0 ID:20140912025257
Pushlog:
https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=80164e15bd54&tochange=c232720a2847
Improved by: Bug 1062723

I am not sure,  Bug 1062723 is related to Bug 1002992 or not.
And large regression Bug 1070772 still exist.

Comment 13

[Timothy Nikkel (:tnikkel)]

Thank you.

(In reply to Alice0775 White from comment #11)
> ad4ea4469570
> 13501 12451 13186 12976 12971 13152

This looks about the same as mozilla-central.
This made all image decoding async.

> 83667965d4d1
> 13077 13134 12906 12721 13174 12963

This looks about the same as mozilla-central.
This made all image decoding async, except if there was work completed off main thread we notified about the finished work synchronously.

> cf97714b9b48
> 13195 13200 12691 13252 13059 13153

This looks about the same as mozilla-central.
This made all image decoding so some decoding synchronously.

> a673077f61e0
> 13287 12803 13028 12998 13019 12853

This was plain mozilla-central, so we had a baseline.

> bd4e45d9809f
> 14363 14841 14675 14634 14313 14096

This is slower than mozilla-central.
This made us use the old display list based image visibility pass.

So this shows that the frame walker based image visibility pass is faster. So this contradicts your earlier results.

I remember that mozregression used to (still does?) get confused by the distinction between pgo and not pgo, maybe that's what happened here?

Comment 14

[Timothy Nikkel (:tnikkel)]

(In reply to Alice0775 White from comment #12)
> https://hg.mozilla.org/integration/mozilla-inbound/
> pushloghtml?fromchange=80164e15bd54&tochange=c232720a2847
> Improved by: Bug 1062723
> 
> I am not sure,  Bug 1062723 is related to Bug 1002992 or not.
> And large regression Bug 1070772 still exist.

I think this speed up is unrelated to bug 1002992. But good to know!

Comment 15

[Alice0775 White]

(In reply to Timothy Nikkel (:tn) from comment #10)

> Also, it might be interesting to see what your results look like it you
> don't restart between every measurement with a regular nightly. If they are
> the same or different that will be a useful data point. 

https://hg.mozilla.org/mozilla-central/rev/1735ff2bb23e
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:35.0) Gecko/20100101 Firefox/35.0 ID:20140925030203
12842 5201 5101 5150 5131 5034

https://hg.mozilla.org/releases/mozilla-aurora/rev/e970bc96f8b5
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:34.0) Gecko/20100101 Firefox/34.0 ID:20140925004001
20210 21434 20884 21584

Nightly35.0a1: It become fast if I do not restart browser.
Aurora34.0a2: painfully slow :( regardless of restarting browser

> And I assume using a fresh profile doesn't change the results? Thanks!
Almost same.

Comment 16

[Alice0775 White]

(In reply to Timothy Nikkel (:tn) from comment #13)

> I remember that mozregression used to (still does?) get confused by the
> distinction between pgo and not pgo, maybe that's what happened here?

I did not use mozregression tool.
I used zip file of tinderbox build[1] and its archive[2].

[1] http://ftp.mozilla.org/pub/mozilla.org/firefox/tinderbox-builds/mozilla-inbound-win32/
[2] http://inbound-archive.pub.build.mozilla.org/pub/mozilla.org/firefox/tinderbox-builds/mozilla-inbound-win32/

Comment 17

[Alice0775 White]

(In reply to Timothy Nikkel (:tn) from comment #13)
> Thank you.
> 
> (In reply to Alice0775 White from comment #11)

> > a673077f61e0
> > 13287 12803 13028 12998 13019 12853
> 
> This was plain mozilla-central, so we had a baseline.
> 
> > bd4e45d9809f
> > 14363 14841 14675 14634 14313 14096
> 
> This is slower than mozilla-central.
> This made us use the old display list based image visibility pass.
> 


> So this shows that the frame walker based image visibility pass is faster.
> So this contradicts your earlier results.
 
I do not think so. 
Because, Bug 1062723 is incuded in mozilla-central now, so I think the pref regression of Bug 1002992 is hidden by Bug 1062723.

Comment 18

[Timothy Nikkel (:tnikkel)]

Okay, to get rid of the confounding effect of bug 1062723 I made two more pushes to try to test. Both pushes are based on http://hg.mozilla.org/mozilla-central/rev/59d4326311e0 which is just a little before bug 1062723. One has the frame walker, one uses the display list.

They should be ready in about an hour from when I post this comment.

https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-0a3bfbbfd895/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-b0591923b6a3/

Comment 19

[Alice0775 White]

0a3bfbbfd895
14429 14663 15413 13896 13904 14446

b0591923b6a3
15841 13608 14996 15498 15100 15119

Comment 20

[Timothy Nikkel (:tnikkel)]

(In reply to Alice0775 White from comment #19)
> 0a3bfbbfd895
> 14429 14663 15413 13896 13904 14446

This is display list builder. avg 14458.5, stddev 561.74612.

> b0591923b6a3
> 15841 13608 14996 15498 15100 15119

This is frame walker. avg 15027, stddev 763.24989.

The standard deviation to is too big to conclude that either one is faster than the either, so it looks like a tie. Unless we get more data.

Comment 21

[Timothy Nikkel (:tnikkel)]

I don't have any better ideas for how to track down what is going on here, so I produced some try builds to try to bisect when bug 1002992 had a regressing effect on mozilla-central by disabling it on different revisions.

These should be ready in an hour.
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-a120b09a9856/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-4d914299ad38/

Comment 22

[Timothy Nikkel (:tnikkel)]

Sorry, build error. Use these links instead.

https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-a120b09a9856/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-171ce5d00304/

Comment 23

[Alice0775 White]

(In reply to Timothy Nikkel (:tn) from comment #22)
> Sorry, build error. Use these links instead.
> 
> https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-
> a120b09a9856/
> https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-
> 171ce5d00304/

a120b09a9856
15926 17181 15769 16355 15709 16150

171ce5d00304
14014 15147 14180 14891 13863 14177

Comment 24

[Timothy Nikkel (:tnikkel)]

More try builds. In about an hour.

https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-8526961d91b4/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-3e0c3f227c02/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-8e3e717be941/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-b808965b53ab/

Comment 25

[Alice0775 White]

8526961d91b4
20642 19962 20028 20205 20283 20722

3e0c3f227c02
17516 16198 17338 16686 16203 17798

8e3e717be941
13746 13541 13864 13347 13032 13697

b808965b53ab
15237 16192 14942 15955 16658 15258

Comment 26

[Timothy Nikkel (:tnikkel)]

Those results muddy the picture even more. They suggest that the frame walker went from being 25% regression, then to a 10% regression, then back to 25% regression, and then to today where it is either the same or slightly better. More try builds to try to make some sense of this.

https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-4945d624571e/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-b36d78faa22a/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-67f71fcc749c/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-c7ebd51d36fe/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-0f43f94793e6/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-1e3de144bed7/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-2307e701a7df/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-4760961f1ac7/

Comment 27

[Alice0775 White]

4945d624571e
5047 5166 6030 4783 5130 5546

b36d78faa22a
5629 5429 6392 5979 5663 5513

67f71fcc749c
17675 16775 17028 17132 17257 16397

c7ebd51d36fe
13846 13663 14129 13696 13647 13005

0f43f94793e6
14330 14320 14627 15063 13813 15030

1e3de144bed7
16549 16868 17157 17196 17563 16277

2307e701a7df
19267 19002 19629 17937 18357 18810

4760961f1ac7
16677 16391 16331 16114 17556 16724

Comment 28

[Timothy Nikkel (:tnikkel)]

More try builds.

https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-13d4527a56cd/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-1acb4e806af6/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-387c0103854c/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-bcbacf1949c2/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-caa32647a226/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-ec46f3a855f8/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-e0222b14df4c/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-29ae16b82a27/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-e6e1735ec21e/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-99777486f0f9/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-430e8b1364e3/

Comment 29

[Alice0775 White]

13d4527a56cd
13063 13762 13791 14080 14430 13686

1acb4e806af6
15520 15229 14525 14792 15379 15626 

387c0103854c
16738 16706 16409 16957 17313 16456 

bcbacf1949c2
16846 15251 15357 15011 16158 16666 

caa32647a226
9935 10806 10435 10157 10034 9824 

ec46f3a855f8
15165 14304 13240 13946 13630 14743 

e0222b14df4c
14080 14029 13537 13478 14213 15457 

29ae16b82a27
10768 11449 11384 10300 12363 11473 

e6e1735ec21e
13547 13014 14030 13813 14594 14313 

99777486f0f9
23031 22968 22431 25169 24515 23288 

430e8b1364e3
15796 15819 15413 15588 14653 15057

Comment 30

[Timothy Nikkel (:tnikkel)]

Thanks again. A bunch more builds. Some are there already, the rest will be there in an hour.


http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-ca39e671836d/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-1c7170a407cf/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-a89aa46e0a93/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-201e72654bec/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-8587be73c923/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-df104ebaa26b/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-fd3c566228d6/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-3136b8867e64/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-de2c479c75e1/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-7d2735b73de9/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-5de96d3dffd5/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-0b858bc58791/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-edfab9de74cd/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-9ab2c0d41928/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-f6e6b9191d0e/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-c5463b044ccb/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-b33c443b079c/
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-3cfe1d898bdc/

Comment 31

[Alice0775 White]

ca39e671836d
14880 15561 14967 15578 14695 15734 

1c7170a407cf
13367 13433 12993 14233 13558 14493 

a89aa46e0a93
17213 19150 17479 18748 18641 17716 

201e72654bec
16000 18952 17799 18675 17332 16622 

8587be73c923
17049 19203 17576 19185 18345 17623 

df104ebaa26b
18115 18114 17999 17668 18675 17211 

fd3c566228d6
17509 15463 17783 17344 16861 17270 

3136b8867e64
16774 17633 16926 16951 18074 16278 

de2c479c75e1
17748 18624 18617 17061 18776 18977 

7d2735b73de9
20061 20013 19943 19700 18990 20960 

5de96d3dffd5
19453 19867 19395 18983 20498 19233 

0b858bc58791
16055 14614 15356 14952 15549 18758 

edfab9de74cd
14238 14631 14080 14993 13370 16974 

9ab2c0d41928
17734 20233 18227 17731 17502 17757 

f6e6b9191d0e
19422 18959 17586 19501 19077 17632 

c5463b044ccb
16733 16067 17041 16743 15462 16323 

b33c443b079c
16603 15457 16792 15277 16004 17360 

3cfe1d898bdc
29193 30627 28840 30301 31468 30010

Comment 32

[Timothy Nikkel (:tnikkel)]

Thanks. More builds. I hope not many more.

https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-0b4cbfa5dde3/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-d89aa5d3d790/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-3b3a4b0b69b7/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-c5abab9cdddc/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-61dc5da60bf0/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-78e0bdb5c558/

Comment 33

[Alice0775 White]

0b4cbfa5dde3
14733 14700 15304 14674 13901 14150 

d89aa5d3d790
15398 16512 16845 15411 16894 16261 

3b3a4b0b69b7
30489 31455 29959 29417 30385 31278 

c5abab9cdddc
19991 19254 19005 19129 19125 19060 

61dc5da60bf0
20006 20341 20027 19508 20060 19843 

78e0bdb5c558
17530 18135 17150 19259 18396 18368

Comment 34

[Timothy Nikkel (:tnikkel)]

Thanks. More builds. Not sure if I'm making any progress, just trying random guesses.

https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-615458fdbe45/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-8fc2ce5dc00c/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-77d775ca5ac2/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-bfa27e268aa4/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-da4e7565750f/

Comment 35

[Alice0775 White]

615458fdbe45
18645 18107 18829 17641 19932 18353 

8fc2ce5dc00c
18339 19411 19044 18483 18559 18354 

77d775ca5ac2
17621 18549 18581 18275 17885 18738 

bfa27e268aa4
18877 18691 17914 19030 18790 17187 

da4e7565750f
17965 18437 18135 18329 18223 19092

Comment 36

[Timothy Nikkel (:tnikkel)]

I've learned a few things, not as many as I would have liked though. I've learned that the display list builder and the frame walker are a tie on current mozilla-central. I've learned that before bug 994081 landed, the display list builder and the frame walker are also tie. I learned that on no platform can I reproduce what Alice is seeing (that the frame walker is a regression when it landed); I have tested on mac, linux, and Windows. I learned that doing both the frame walker and the display list builder fixes the regression (and it doesn't matter which order you do them in). I learned that you don't even need to do the whole display list builder pass, it is enough to build a display list, and then iterate over it (and not do anything while you iterate over it). However if you don't iterate over the list it seems to be a huge regression, worse then the original regression. I learned a lot of things have no effect.

Comment 37

[Timothy Nikkel (:tnikkel)]

More builds. Thanks in advance.

https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-7c5fb9366e61/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-5ebd432a91df/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-e5b541c88d39/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-d45f61fbfeee/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-34d8effab9a8/

Comment 38

[Alice0775 White]

7c5fb9366e61
19422 20523 19046 20348 20234 19573 

5ebd432a91df
18851 19319 19062 18451 19306 19484 

e5b541c88d39
18115 17668 18114 18407 16575 17065 

d45f61fbfeee
4222 4503 4539 4404 4486 4258 

34d8effab9a8
4500 4983 4717 4947 4567 4649

Comment 39

[Timothy Nikkel (:tnikkel)]

Alice, do you think you could provide a profile of this running on your machine? Instructions are here https://developer.mozilla.org/en-US/docs/Mozilla/Performance/Profiling_with_the_Built-in_Profiler

Comment 40

[Alice0775 White]

(In reply to Timothy Nikkel (:tn) from comment #39)
> Alice, do you think you could provide a profile of this running on your
> machine? Instructions are here
> https://developer.mozilla.org/en-US/docs/Mozilla/Performance/
> Profiling_with_the_Built-in_Profiler


> Ctrl+Shift+2 - Take a profile and launch Cleopatra to view it

Nothing happens...

Comment 41

[Timothy Nikkel (:tnikkel)]

(In reply to Alice0775 White from comment #40)
> (In reply to Timothy Nikkel (:tn) from comment #39)
> > Alice, do you think you could provide a profile of this running on your
> > machine? Instructions are here
> > https://developer.mozilla.org/en-US/docs/Mozilla/Performance/
> > Profiling_with_the_Built-in_Profiler
> 
> 
> > Ctrl+Shift+2 - Take a profile and launch Cleopatra to view it
> 
> Nothing happens...

It seems like nothing happens, but if you just wait a little while a new tab will eventually open with the profile. (It happens to me too, and it was super confusing until I figured that out.)

Comment 42

[Timothy Nikkel (:tnikkel)]

And make sure the Firefox window has focus (not the scratch pad window), and make sure to close the modal alert dialog. If that doesn't work just click on the profiler button in the tool bar to bring up the profiler panel with buttons for start/stop/analyze.

Comment 43

[Alice0775 White]

Okay
http://people.mozilla.org/~bgirard/cleopatra/#report=50a418e13e2bcc99a78bcd460b8f2482eb461a1c

Comment 44

[Timothy Nikkel (:tnikkel)]

Thank you for that, very informative.

More builds to try.

https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-cfe0518ab437/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-ecf236d04bbe/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-24124b11d610/
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-37b216c563db/

Comment 45

[Alice0775 White]

cfe0518ab437
19138 19400 19297 18664 18715 18397 

ecf236d04bbe
17633 18004 18832 17802 17434 17301 

24124b11d610
17250 17551 16784 17900 17351 18600 

37b216c563db
18498 19547 18420 19197 19180 20051

Comment 46

[Timothy Nikkel (:tnikkel)]

Alice, I'm curious to know if you see any regression when the pref image.high_quality_downscaling.enabled is set to false. ie did bug 1002992 have any effect? Same for bug 994081. And if current nightly is the same or different from the start of June.

Comment 47

[Alice0775 White]

(In reply to Timothy Nikkel (:tn) from comment #46)
> Alice, I'm curious to know if you see any regression when the pref
> image.high_quality_downscaling.enabled is set to false. ie did bug 1002992
> have any effect? Same for bug 994081. And if current nightly is the same or
> different from the start of June.

Nightly(2014-06-06)
5725 5554 5648 6193 5757 5991 

Nightly(2014-06-06) with image.high_quality_downscaling.enabled = false
5183 5413 5522 4540 5398 5211 


Nightly(2014-06-07)
15622 15827 16165 16318 14925 16527 

Nightly(2014-06-07) with image.high_quality_downscaling.enabled = false
4169 4335 4321 4322 4322 4219 


Nightly(2014-10-01)
11751 13463 12412 11471 11673 12406 

Nightly(2014-10-01) with image.high_quality_downscaling.enabled = false
14157 14134 14085 14104 14200 15350

Comment 48

[Alice0775 White]

FYI,
I field a bug.
Bug 1076725 - Scrolling speed has become 200% slower after landing Bug 1044702 under condition of image.high_quality_downscaling.enabled = false

Comment 49

[Bas Schouten (:bas.schouten)]

This looks under control.

Comment 50

[Timothy Nikkel (:tnikkel)]

Alice, do you think you could re-test this testcase to see if the situation has improved for you? There have been a lot of image related changes over the last year.

Comment 51

[Alice0775 White]

I can no longer reproduced on Latest Nightly45 without e10s