1020340 - Poor canvas performance on OpenLayers3

Status: NEW

(Core :: Graphics: Canvas2D, defect)

Description

[Edu Logroño]

User Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:29.0) Gecko/20100101 Firefox/29.0 (Beta/Release)
Build ID: 20140506152807

Steps to reproduce:

Hello, 
I'm trying OpenLayers3 examples and I found your browser has poor performance compared to Explorer and Chrome. 

I think problem is Canvas rendering... 

You can compare it -> http://ol3js.org/en/master/examples/synthetic-lines.html 

Do you know what is the problem ? 

Thanks! 


Actual results:

Bad performance

Comment 1

[Maria Grazia Alastra [:mga]]

Confirmed on Windows 7, 64 bits, with Firefox Beta 30 (User Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:30.0) Gecko/20100101 Firefox/30.0, Build ID: 20140605174243)

Comment 2

[yanakey]

Another example - http://openlayers.org/en/v3.0.0/examples/vector-layer.html. Try to drag the map, while no country is selected.
Performance is awful... )-:

Comment 3

[Milan Sreckovic [:milan] (needinfo for best results)]

I get a freeze on the comment 0 site on MacBook Pro (nightly 20141210). With comment 2 site, I get ~12-15fps.

Comment 4

[Lee Salzman [:lsalzman]]

On the synthetic lines test case, and peeking inside what Cairo was doing, it appears to be handing us a fairly gigantic path of 10K points to stroke, which internally is generating a fairly big polygon mesh. The Direct2D backends also seem to choke on this as well. Skia and Skia GL backends seem to do much better and give interactive performance.

After conferring with Bas, it was discussed that the reason for this difference is that Skia is primarily a hairline renderer that does not mess with building meshes. Whereas for Cairo and Direct2D, since we are not caching the mesh that is built up, a lot of time is spent rebuilding this mesh every frame, and not much on the actual rendering itself. Bas also looked into IE11 performance on this test case and found it was similarly not good.

I tried poking at some Cairo "cheats" like turning off AA, messing with cap or join options, but none of that made much of a difference, as expected, since the bottleneck is in processing the mesh itself before any of that becomes an issue.

For the time being, I am not sure there is much we can do about this one for the Direct2D backend, though at least in cases where we rely on Cairo canvas, using Skia as an alternative is one solution.