Closed Bug 742760 Opened 13 years ago Closed 10 years ago

drawWindow for CSS transform: 3D renders pixelated poorly

Categories

(Core :: Graphics, defect)

All
macOS
defect
Not set
normal

Tracking

()

RESOLVED FIXED
mozilla36

People

(Reporter: mgalli, Assigned: mstange)

References

Details

Attachments

(3 files, 1 obsolete file)

User Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:11.0) Gecko/20100101 Firefox/11.0 Build ID: 20120312181643 Steps to reproduce: Used a screen grab extension to capture image of iframe with a transform:rotateY and perspective Actual results: The image, for the transformed section ( iframe ) looks poor Expected results: To look normal. BTW in nightly even the live rendering of the iframe transformed looks bad. But I am first making a case about Firefox branch product.
Summary: drawWindow css 3d → drawWindow for CSS transform: 3D renders pixelated poorly
Component: General → Graphics
QA Contact: general → thebes
Are you using the DRAWWINDOW_USE_WIDGET_LAYERS flag? Without this, we will be using BasicLayers for DrawWindow instead of (potentially) an accelerated backend. If you're doing this, or your system doesn't support accelerated layers (see GPU Accelerated Windows section in about:support), then you will be using pixman for the 3d transforms. This is known to be pretty low quality, I'm not sure what we can do about that. Do you have a testcase?
Well IMHO this needs product people here then. I mean, mainly now AMO folks because if Mozilla says the web includes CSS 3D, it is expected we are to take page screen grabs out of these things, so we can keep evangelizing. Attaching Christian Heilmann's personal blog page screenshot taken with a general screenshot taker app from addons.mozilla.org.
See the right part. This is not what users see. This is ony what you get if you use a screen grab app ( like save page as PNG ) from Addons.mozilla.org.
Can you please link me to the addon in question? As for improving the quality of software transforms, we'd need to either use a different library instead of pixman, improve pixman, or write our own. Not sure which would be the best approach here, CC'ing a few other gfx people who might have ideas.
Blocks: 760442
I'd like to note that the site in comment #6 is hopelessly jagged for me even *with* hardware acceleration. Unlike, say: http://thewebrocks.com/demos/html5-3d-css/ which is completely smooth. In addition, hovering over the sidebar in comment #6 causes ugly jumpiness which seems Firefox specific.
FWIW, Firefox Nightly under OSX now renders Heilmann's site nice and smooth. Firefox under Linux was fixed by switching on azure. It then rendered smoothly, whether azure was using cairo or skia, although with cairo enabled I hit bug 793416...
Ok. Correction. Firefox under Linux was fixed by layers.acceleration.force-enabled, not azure. Which is weird since I'd swear I had set force-enabled previously without fixing jaggies. But, at least that makes sense, since that's the reason for jaggies mentioned earlier. Why am I adding more bug spam? Well, I noticed a weird problem on my mac mini 2. Rendering is totally smooth on Firefox Nightly if I hover in and out of the right side of christianheilmann.com 's menu. *but* if I do so a few times, especially, it seems if I move in and out while hover is triggered on the "feed me" image, or the PRESENTATIONS-TWITTER bookmarks, all of a sudden the menu goes jaggy and starts stuttering badly.
FWIW, jfkthame couldn't reproduce on his Mac. Might be Mac Mini 2 specific. He speculated something knocking it unexpectedly into software mode, which is understandable based on the symptoms. It might be he just didn't try it long enough, since sometimes I had to wiggle that menu a good few dozen times before it started happening. Here's a console log piped to uniq -c - jfkthame said he didn't see anything odd. http://m8y.org/tmp/console_uniq.log
Attached patch patch (obsolete) — Splinter Review
Attachment #8523184 - Flags: review?(jmuizelaar)
Comment on attachment 8523184 [details] [diff] [review] patch The Skia patch in bug 1097776 also fixes this.
Attachment #8523184 - Attachment is obsolete: true
Attachment #8523184 - Flags: review?(jmuizelaar)
Assignee: nobody → mstange
Status: UNCONFIRMED → ASSIGNED
Depends on: 1097776
Ever confirmed: true
Hardware: x86 → All
Version: 11 Branch → Trunk
Fixed by bug 1097776.
Status: ASSIGNED → RESOLVED
Closed: 10 years ago
Resolution: --- → FIXED
Target Milestone: --- → mozilla36
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: