696321 - ie.microsoft.com/testdrive/Performance/ParticleAcceleration - Garbage is left in screen canvas drawing.

Status: RESOLVED WONTFIX

(Core :: Graphics, defect)

Description

[Alice0775 White]

Attachment: Screen shot

Build Identifier:
http://hg.mozilla.org/mozilla-central/rev/311fdb9b38b7
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:10.0a1) Gecko/20111020 Firefox/10.0a1 ID:20111020031025

Garbage is left in screen canvas drawing.

Disabled canvas azure helps(i.e I set gfx.canvas.azure.enabled to false).

Reproducible: Always

Steps to Reproduce:
1. Start Firefox with clean profile
2  Open URL ( http://ie.microsoft.com/testdrive/Performance/ParticleAcceleration/ )

Actual Results:
  Garbage is left in screen canvas drawing.

Expected Results:
  No garbage

Regression window(m-c),
Works:
http://hg.mozilla.org/mozilla-central/rev/450e4d9ea2d5
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:7.0a1) Gecko/20110624 Firefox/7.0a1 ID:20110625025936
Fails:
http://hg.mozilla.org/mozilla-central/rev/48ad4ffc4230
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:7.0a1) Gecko/20110625 Firefox/7.0a1 ID:20110625030821
pushlog:
http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=450e4d9ea2d5&tochange=48ad4ffc4230


Regression window(m-i),
Works
http://hg.mozilla.org/integration/mozilla-inbound/rev/c65f1fb0449d
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:7.0a1) Gecko/20110624 Firefox/7.0a1 ID:20110624094307
Crash(cannot run url):
http://hg.mozilla.org/integration/mozilla-inbound/rev/c65f1fb0449d
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:7.0a1) Gecko/20110624 Firefox/7.0a1 ID:20110624094307
http://hg.mozilla.org/integration/mozilla-inbound/rev/11e15aabd2c1
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:7.0a1) Gecko/20110624 Firefox/7.0a1 ID:20110624111756
Fails:
http://hg.mozilla.org/integration/mozilla-inbound/rev/aede8422cdc4
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:7.0a1) Gecko/20110624 Firefox/7.0a1 ID:20110624130320
Pushlog:
http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=c65f1fb0449d&tochange=aede8422cdc4

Suspected:
Bug 651858 - [Azure] Implement Direct2D Azure Backend

Comment 1

[Alice0775 White]

Graphics
Adapter Description: ATI Radeon HD 4300/4500 Series
Vendor ID: 1002
Device ID: 954f
Adapter RAM: 512
Adapter Drivers: aticfx64 aticfx64 aticfx32 aticfx32 atiumd64 atidxx64 atiumdag atidxx32 atiumdva atiumd6a atitmm64
Driver Version: 8.892.0.0
Driver Date: 9-8-2011
Adapter RAM (GPU #2): Unknown
Adapter Drivers (GPU #2): Unknown
Direct2D Enabled: true
DirectWrite Enabled: true (6.1.7601.17563)
ClearType Parameters: Gamma: 2200 Pixel Structure: RGB ClearType Level: 100 Enhanced Contrast: 50 
WebGL Renderer: Google Inc. -- ANGLE (ATI Radeon HD 4300/4500 Series) -- OpenGL ES 2.0 (ANGLE 0.0.0.774)
GPU Accelerated Windows: 1/1 Direct3D 10

Comment 2

[Bas Schouten (:bas.schouten)]

So, something interesting is going on, this is essentially all due to rounding errors, the transform that we have set has an 0.99993473 scale. This causes the rect we end up clearing to start at 0.03 instead of 0. Effectively not fully clearing the top line, I'm investigating this further.

Comment 3

[Bas Schouten (:bas.schouten)]

So, this rounding error works like this: What this demo is going is continuously scaling by x, and then by the reciprocal of x, and assuming the matrix will be identical to what it was. This works most of the time, but for some scales this seems to result in a rounding error. Since the matrix is never reset this rounding error gets cumulatively worse. Due to large number of times this is happening, the error gets bad enough to cause artifacts pretty quickly.

Comment 4

[Bas Schouten (:bas.schouten)]

Roc, what do you think? I'm looking at Demo.js around line 329. It seems to be that's just bad practice, without the save restore, or manually setting the transform back to what it was this would probably just work. Assuming a * s * 1/s to always return a again is simply a false assumption in floating point I think.

Comment 5

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Yes.

Maybe we should detect transform components being very close to integers and snap them to integers?

Comment 6

[Bas Schouten (:bas.schouten)]

IE9 suffers from this issue just as much after running this for a short while.

Comment 7

[Robert O'Callahan (:roc) (email my personal email if necessary)]

OK.

Comment 8

[Ryan]

Well it looks like it was fixed at some point because Firefox 18 doesn't exhibit this issue, but with 19+ it's back again.

Comment 9

[Ryan]

Once again, disabling Azure Gfx Content fixes it.

Comment 10

[Ryan]

Actually strike that last message, changing enabling Azure Canvas in Firefox 18 makes the garbage appear but in 19+ the garbage always shows regardless of the Azure Canvas setting.