1034784 - The main process uses 4.5 MB more of decoded images in 2.0 than in 1.4

Status: RESOLVED WORKSFORME

(Firefox OS Graveyard :: Gaia::System, defect)

Description

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

According to the memory reports Eric gathered in bug 1028260 we're using 4.5 MB more decoded image data in 2.0.

The diff shows:
7,369,820 B (100.0%) -- explicit
├──4,790,400 B (65.00%) -- images/content

The big ones are:
│  │  ├──1,740,880 B (23.62%) ++ image(app://system.gaiamobile.org/shared/resources/branding/initlogo.png)
│  │  ├──1,581,136 B (21.45%) ++ image(blob:3a9c0b10-0ebb-4a69-a0af-36997897c105)
│  │  ├──1,581,136 B (21.45%) ++ image(blob:48703170-1273-4974-a6e1-58b7dd045836)
│  │  ├──1,544,272 B (20.95%) ++ image(app://system.gaiamobile.org/shared/style/value_selector/images/ui/gradient.png)
│  │  ├────618,576 B (08.39%) ++ image(app://callscreen.gaiamobile.org/shared/style/action_menu/images/ui/gradient.png)
│  │  ├────618,576 B (08.39%) ++ image(app://system.gaiamobile.org/shared/style/confirm/images/ui/gradient.png)

Open questions:

Why is the initlogo hanging around after startup?
Where do these blob URIs come from?
What are these gradients for and why do we need PNGs for them?

Comment 1

[Nicholas Nethercote [inactive]]

Good to see that the per-image reporting is coming in useful :)

Comment 2

[[:fabrice] Fabrice Desré]

The gradients are for backgrounds. We replaced them by plain colors on tarako, and no one complained ;)

But I don't get the same results here (not a diff):

├───2.28 MB (06.19%) -- images/content/raster/used
│   ├──0.61 MB (01.66%) -- image(blob:cb3f1d43-c26c-470e-9df6-aadda1617375)
│   │  ├──0.59 MB (01.59%) ── uncompressed-nonheap
│   │  └──0.02 MB (00.06%) ++ (2 tiny)
│   ├──0.59 MB (01.60%) -- image(app://system.gaiamobile.org/shared/style/confirm/images/ui/gradient.png)
│   │  ├──0.59 MB (01.59%) ── uncompressed-nonheap
│   │  └──0.00 MB (00.01%) ++ (2 tiny)
│   ├──0.59 MB (01.60%) -- image(app://system.gaiamobile.org/shared/style/value_selector/images/ui/gradient.png)
│   │  ├──0.59 MB (01.59%) ── uncompressed-nonheap
│   │  └──0.00 MB (00.01%) ++ (2 tiny)
│   └──0.49 MB (01.33%) -- (8 tiny)
│      ├──0.13 MB (00.35%) ++ image(app://system.gaiamobile.org/style/statusbar/images/icons.png)
│      ├──0.09 MB (00.26%) ++ image(<non-notable images>)
│      ├──0.07 MB (00.18%) ── image(app://system.gaiamobile.org/shared/resources/branding/initlogo.png)/raw
│      ├──0.06 MB (00.15%) ++ image(app://system.gaiamobile.org/style/quick_settings/images/data.png)
│      ├──0.04 MB (00.11%) ++ image(app://system.gaiamobile.org/shared/style/confirm/images/ui/pattern.png)
│      ├──0.04 MB (00.11%) ++ image(app://system.gaiamobile.org/shared/style/value_selector/images/ui/pattern.png)
│      ├──0.04 MB (00.11%) ++ image(app://system.gaiamobile.org/style/themes/default/images/ui/pattern.png)
│      └──0.02 MB (00.06%) ── image(blob:a9f32b16-a04f-4eea-897f-95a65ebd74a8)/raw

And I checked that we remove the logo DOM element (https://mxr.mozilla.org/gaia/source/apps/system/js/init_logo_handler.js#205).

Comment 3

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Ok, I guess we need to see if this reproduces.

Comment 4

[Gregor Wagner [:gwagner]]

Eric, can you share your STR here? Or even better see if this still reproduces?

Comment 5

[Eric Rahm [:erahm]]

STR:
1) Reboot phone
2) Wait until lock screen is shown
3) Measure memory with |MOZ_IGNORE_NUWA_PROCESS=1 tools/get_about_memory.py --minimize|

I'll do some testing to see if I can reproduce.

Comment 6

[Eric Rahm [:erahm]]

After retesting [1] I still see the gradients, initlog, and blobs, so Kyle's questions are still relevant.

[1] |adb reboot && sleep 60 && MOZ_IGNORE_NUWA_PROCESS=1 tools/get_about_memory.py --minimize|

But it looks like the diff of 1.4 -> 2.0 actually shows an overall *decrease* in image memory (-1.29 MB). The stats are a little misleading due to us having better reporting in 2.0 and negative signs (-) looking like the tree dashes (─):

5.11 MB (100.0%) -- explicit
├──-1.29 MB (-25.29%) -- images/content       <--- this is an overall decrease
│  ├──-7.43 MB (-145.37%) -- used             <--- this decrease is because we changed how we report
│  │  ├──-6.68 MB (-130.78%) ── uncompressed-nonheap [-]
│  │  ├──-0.57 MB (-11.21%) ── raw [-]
│  │  └──-0.17 MB (-3.38%) ── uncompressed-heap [-]
│  └───6.14 MB (120.08%) -- raster/used       <--- this increase is because we changed how we report
│      ├──1.66 MB (32.49%) -- image(app://system.gaiamobile.org/shared/resources/branding/initlogo.png)
│      │  ├──1.46 MB (28.66%) ── uncompressed-nonheap [+]
│      │  ├──0.20 MB (03.82%) ── raw [+]
│      │  └──0.00 MB (00.00%) ── uncompressed-heap [+]
│      ├──1.51 MB (29.51%) -- image(blob:05d249c4-d760-40d6-8420-52a314b2d2ce)
│      │  ├──1.46 MB (28.66%) ── uncompressed-nonheap [+]
│      │  └──0.04 MB (00.84%) ++ (2 tiny)
│      ├──1.47 MB (28.82%) -- image(app://system.gaiamobile.org/shared/style/value_selector/images/ui/gradient.png)
│      │  ├──1.46 MB (28.66%) ── uncompressed-nonheap [+]
│      │  └──0.01 MB (00.15%) ++ (2 tiny)
│      ├──0.59 MB (11.54%) -- image(app://system.gaiamobile.org/shared/style/confirm/images/ui/gradient.png)
│      │  ├──0.59 MB (11.47%) ── uncompressed-nonheap [+]
│      │  └──0.00 MB (00.08%) ++ (2 tiny)
│      ├──0.30 MB (05.81%) -- image(app://system.gaiamobile.org/style/statusbar/images/icons.png)
│      │  ├──0.28 MB (05.50%) ── uncompressed-nonheap [+]
│      │  └──0.02 MB (00.31%) ++ (2 tiny)
│      ├──0.19 MB (03.76%) ++ (6 tiny)
│      ├──0.16 MB (03.03%) -- image(app://system.gaiamobile.org/style/statusbar/images/signal-searching.png)
│      │  ├──0.14 MB (02.81%) ── uncompressed-heap [+]
│      │  └──0.01 MB (00.23%) ── raw [+]
│      ├──0.14 MB (02.75%) -- image(<non-notable images>)
│      │  ├──0.09 MB (01.78%) ── uncompressed-heap [+]
│      │  └──0.05 MB (00.98%) ── raw [+]
│      └──0.12 MB (02.37%) -- image(app://system.gaiamobile.org/style/quick_settings/images/data.png)
│         ├──0.12 MB (02.29%) ── uncompressed-nonheap [+]
│         └──0.00 MB (00.08%) ++ (2 tiny)

Comment 7

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

(In reply to Eric Rahm [:erahm] from comment #6)
> But it looks like the diff of 1.4 -> 2.0 actually shows an overall
> *decrease* in image memory (-1.29 MB). The stats are a little misleading due
> to us having better reporting in 2.0 and negative signs (-) looking like the
> tree dashes (─):

The original set of diffs showed an increase ...

Comment 8

[Eric Rahm [:erahm]]

(In reply to Kyle Huey [:khuey] ([e-mail]) from comment #7)
> The original set of diffs showed an increase ...

Those reports weren't generated with |--minimize|, which certainly could account for the discrepancy.

Comment 9

[Eric Rahm [:erahm]]

Attachment: Memory report for 2.0

Memory report for 2.0 generated using:
  |adb reboot && sleep 60 && MOZ_IGNORE_NUWA_PROCESS=1 tools/get_about_memory.py --minimize|

Comment 10

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

initlogo.png is bug 1035432

The large blob URI is the wallpaper.  I assume that lives forever?  If so this isn't that big of a deal.

The two gradients appear to be the same image.  They would share compressed memory if they were the same URL, and probably uncompressed too, although they might be decoded to two different sizes.

Comment 11

[(inactive after 6/18) Alive Kuo [:alive]]

(In reply to Kyle Huey [:khuey] ([e-mail]) from comment #10)
> initlogo.png is bug 1035432
> 
> The large blob URI is the wallpaper.  I assume that lives forever?  If so
> this isn't that big of a deal.

Wallpaper lives forever.

> 
> The two gradients appear to be the same image.  They would share compressed
> memory if they were the same URL, and probably uncompressed too, although
> they might be decoded to two different sizes.

For gradient we could try to access the same URL, does this help?
But for the gradient lives in callscreen, I guess we shouldn't. It appears because callscreen is in process right now. Don't think we should share the same image between callscreen and system.

Comment 12

[Gregor Wagner [:gwagner]]

What are the next steps here? Is the bug description and the blocking flag still accurate?

Comment 13

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

No, it's not.  I'll file new bugs for the remaining issues.