760888 - IE10 Test Drive: ParticleAcceleration Canvas 2D demo is slow

Status: NEW

(Core :: Graphics: Canvas2D, defect)

Description

[Chris Peterson [:cpeterson]]

STR:
1. Load https://ie.microsoft.com/testdrive/Performance/ParticleAcceleration/

On my MacBook Pro, the test frame rate for Chrome 19 is about 3x faster than Nightly 15.0a1 (2012-06-02):

* Firefox 15: FPS: 9, Callback Efficiency: 10%, Score:   ~9000
* Chrome 19: FPS: 32, Callback Efficiency: 24%, Score: ~100000

The test's HTML comments say:

"This demo uses mathematical models to draw the particles in 3D, using the 2D Context HTML5 Canvas. The faster your underlying browser and computer, the higher your score and faster the particles will spin."

Comment 1

[Chris Peterson [:cpeterson]]

* Safari 5.1.7 (6534.57.2) on Mac OS X 10.6.8:
FPS: 9, Callback Efficiency: 10%, Score: ~9000

Comment 2

[Guilherme Lima]

Is this bug still valid? I got 60fps (Win7 - d3d - direct2d)

Comment 3

[Kevin Brosnan [Ex-Mozilla]]

This bug is filed against OS X.

Comment 4

[Chris Peterson [:cpeterson]]

I retested OS X and this bug is still a problem:

* Firefox 22: DrawTime: ~90ms, FPS: ~10, Callback Efficiency: ~1%, Score: ~400
* Chrome 25: DrawTime: ~3ms, FPS: 60, Callback Efficiency: ~11%, Score: ~64,000
* Safari 6.0.3: DrawTime: ~5ms, FPS: 60, Callback Efficiency: ~1%, Score ~6,000

The difference in frame rate is clear, but I'm not sure what "Callback Efficiency" is measuring. Chrome and Safari both render 60 FPS, but Chrome's "Callback Efficiency" (and "Score") is 10x Safari's.

Comment 5

[Mayank Bansal]

See also  bug 806044 .

Comment 6

[Artem S. Tashkinov]

This demo is still painfully slow under Linux with NVIDIA binary drivers even with:

    user_pref("gfx.canvas.azure.accelerated", true);
    user_pref("gfx.xrender.enabled", true);
    user_pref("layers.acceleration.force-enabled", true);

Firefox 58: 16 fps
Google Chrome 64: 60 fps

Comment 7

[Kannan Vijayan [:djvj]]

Bas, do you have an understanding of where this perf hit comes from?  We're doing significantly worse, so there must be some architectural issue there, or is this a driver support issue?

Comment 8

[Artem S. Tashkinov]

I also filed bug 1435986.

Comment 9

[Bas Schouten (:bas.schouten)]

(In reply to Kannan Vijayan [:djvj] from comment #7)
> Bas, do you have an understanding of where this perf hit comes from?  We're
> doing significantly worse, so there must be some architectural issue there,
> or is this a driver support issue?

Sure, on OS X we don't use Direct2D, on Linux we don't use acceleration at all. Why Chrome is faster? Maybe they use accelerated Skia where we don't? Jeff would know.

Comment 10

[Artem S. Tashkinov]

Could be solved.

The working URL: https://web.archive.org/web/20231106001735/https://testdrive-archive.azurewebsites.net/Performance/ParticleAcceleration/