Open Bug 974607 Opened 11 years ago Updated 2 years ago

Don't use subpixel aa w/D2D on transparent surfaces if possible

Categories

(Core :: Graphics, defect)

x86
Windows 7
defect

Tracking

()

REOPENED

People

(Reporter: jrmuizel, Assigned: jrmuizel)

References

Details

(Whiteboard: [Australis:P-])

Attachments

(1 file, 1 obsolete file)

The D2D path for this is slow and we're getting hurt a lot on the australis customize transition. We have a couple of different options here: 1. Disable subpixel-aa on transparent surfaces in chrome. 2. Disable subpixel-aa on transparent surfaces everywhere. I think it's ok to not have subpixel-aa on tab titles, but it would be nice if we can still have subpixel-aa on the url bar text. I'm not exactly sure what would need to happen for that to be possible.
Blocks: 974407
Summary: Don't use subpixel aa on transparent surfaces if possible → Don't use subpixel aa w/D2D on transparent surfaces if possible
We may, in fact, just be able to do better.
Attached patch Something like this (obsolete) — Splinter Review
The advantage of doing something like this is pretty substantial, according to the CART test results in bug 974407. I'm marking this as a P2 Australis bug - I do believe we want this.
Blocks: 873060
Whiteboard: [Australis:P2]
Is this something that should go through review now, or is it still a WIP patch?
Flags: needinfo?(jmuizelaar)
Still a WIP. There are bugs that might not be easy to solve.
Flags: needinfo?(jmuizelaar)
Do we still want to call this P2? Sounded like the there were bigger wins from the about:customizing preloading, and that this was small peanuts?
(Calling this Jeff's since he's got the WIP patch)
Assignee: nobody → jmuizelaar
(In reply to Justin Dolske [:Dolske] from comment #6) > Do we still want to call this P2? Sounded like the there were bigger wins > from the about:customizing preloading, and that this was small peanuts? I think so? According to [1], approximately 50% of our Windows 7 users have d2d layer acceleration enabled, meaning that about half of our Windows users will experience the slowdown from the subpixel AA. It's a pretty substantial change, according to the compare-talos I did in bug 974407[2]. So, based on user exposure and potential payoff, I think we should try to chase this one down. [1]: http://people.mozilla.org/~bjacob/gfx_features_stats/#win7 [2]: http://compare-talos.mattn.ca/?oldRevs=fa48d6ebc486&newRev=306fd5b6afb6&server=graphs.mozilla.org&submit=true
So it turns out I can toggle subpixel AA by setting opacity of 0.99 on text. I'm going to whip up a patch to do that during the transition to see what that buys us.
I've pushed a POC patch to try that uses CSS to disable subpixel AA during the customize mode transition. It's a pretty big hammer - I basically just set opacity to 0.99 on all labels during the transition. Baseline: https://tbpl.mozilla.org/?tree=Try&rev=21e77b801df3 Patch: https://tbpl.mozilla.org/?tree=Try&rev=7660be8d3aa9 Compare talos: http://compare-talos.mattn.ca/?oldRevs=21e77b801df3&newRev=7660be8d3aa9&server=graphs.mozilla.org&submit=true
jrmuizel: mattwoodrow wants to know what the problems are remaining with this patch, since he says it looks sane.
Flags: needinfo?(jmuizelaar)
Apparently, the opacity trick was a bit silly since it seems to cost more to pull that trick than to do the subpixel AA itself. mattwoodrow suggested applying transform:perspective(1px) on the text instead, so I'm trying that next. Baseline: https://tbpl.mozilla.org/?tree=Try&rev=21e77b801df3 Patch: https://tbpl.mozilla.org/?tree=Try&rev=b5d485143452 Compare talos: http://compare-talos.mattn.ca/?oldRevs=21e77b801df3&newRev=b5d485143452&server=graphs.mozilla.org&submit=true
(In reply to Jeff Muizelaar [:jrmuizel] from comment #0) > I think it's ok to not have subpixel-aa on tab titles Why do you think that's ok? Sounds like a significant regression to me... except for Aero Glass where we already get grayscale AA, unfortunately. This is primary UI that users see all the time, with text that users read all the time. We shouldn't dismiss this as less important than the customize transition.
(In reply to Dão Gottwald [:dao] from comment #13) > (In reply to Jeff Muizelaar [:jrmuizel] from comment #0) > > I think it's ok to not have subpixel-aa on tab titles > > Why do you think that's ok? Sounds like a significant regression to me... > except for Aero Glass where we already get grayscale AA, unfortunately. This > is primary UI that users see all the time, with text that users read all the > time. We shouldn't dismiss this as less important than the customize > transition. I was only talking about Aero Glass.
Flags: needinfo?(jmuizelaar)
But we already get grayscale AA on Glass. What would this bug change?
(In reply to Dão Gottwald [:dao] from comment #15) > But we already get grayscale AA on Glass. What would this bug change? That is a little unclear at this point.
Wow - a pretty substantial win if I use mattwoodrow's transform:perspective(1px) hack. Especially on Windows 8. Given our time constraints, I think that's the route I'm going to pursue.
CART results on OS X and Linux don't show the same improvement (in fact, CART results come in worse in some cases): http://compare-talos.mattn.ca/?oldRevs=71efe1f6d4a7&newRev=fdc2ed44e85d&server=graphs.mozilla.org&submit=true I think this will be a Windows-only improvement.
Also, I'm going to extract the work I'm doing to disable subpixel AA during the transition into a separate bug. It's possible that jrmuizel may want to follow-up on some of this for non-Australis performance things, but I don't think it's a blocker for Australis anymore. I've filed bug 977796 for the split out work. jrmuizel, if you don't feel this is still worth pursuing for other reasons, feel free to close this bug.
Whiteboard: [Australis:P2]
No longer blocks: 873060
If transparent surfaces cause this, I'd suggest removing them as much as possible from the Browser UI. For example, the subviews has a rgba(255,255,255,0.97) background, which could be changed to white.
P- per comment 19 (still showing up in the tracker through the other bug).
Whiteboard: [Australis:P-]
Attached patch subaaSplinter Review
I'd like to try landing this to see if it causes any bugs for anyone.
Attachment #8379050 - Attachment is obsolete: true
Attachment #8404008 - Flags: review?(bas)
Attachment #8404008 - Flags: review?(bas)
Per policy at https://wiki.mozilla.org/Bug_Triage/Projects/Bug_Handling/Bug_Husbandry#Inactive_Bugs. If this bug is not an enhancement request or a bug not present in a supported release of Firefox, then it may be reopened.
Status: NEW → RESOLVED
Closed: 6 years ago
Resolution: --- → INACTIVE
Status: RESOLVED → REOPENED
Resolution: INACTIVE → ---
Severity: normal → S3
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: