JDK-8088687 : [Macos] [canvas] Performance problem with Canvas shaped clipping on Mavericks
  • Type: Bug
  • Component: javafx
  • Sub-Component: graphics
  • Affected Version: 8u40
  • Priority: P4
  • Status: Open
  • Resolution: Unresolved
  • Submitted: 2014-07-16
  • Updated: 2018-09-05
The Version table provides details related to the release that this issue/RFE will be addressed.

Unresolved : Release in which this issue/RFE will be addressed.
Resolved: Release in which this issue/RFE has been resolved.
Fixed : Release in which this issue/RFE has been fixed. The release containing this fix may be available for download as an Early Access Release or a General Availability Release.

To download the current JDK release, click here.
Other
tbdUnresolved
Related Reports
Relates :  
JDK-8098079 - NGCanvas should take advantage of new 1-step clip support
Relates :  
JDK-8090206 - Mac: GuiMark2 Bitmap runs at 3 fps with non-rectangular clips even after RT-30107
Description
We are seeing performance issues running the Bitmap and Vector Canvas toys on MacOS under Mavericks with the oval clipping setting.

On the same machine I get 20fps for large oval and 36-40 fps for small oval under 10.8, but I only get 6fps for either under Mavericks.
Comments
SQE is OK to defer this from 8u40.
19-11-2014
Just noticed that this is marked major so we have to get approval for deferment from 8u40. Since this is more of a "didn't get a performance *increase* that we expected" than a regression or a performance loss, I'm not sure why it is major, or why we need to get it solved in 8u40.
18-11-2014
For the record I updated my Mavericks partition to 10.9.5 yesterday and noticed no improvements here...
19-09-2014
I wrote a test case TestAlternatingImages.java to see if there was a difference in the performance of having to bind new textures frequently on Mavericks (click on the scene to switch between whether the two banks of images are batched with each other, or alternating) since the shaped clipping support relies heavily on rendering to a temporary texture and then blitting to the back buffer as opposed to the unclipped version that would do all blits directly to the destination. Unfortunately this test case, while it does show that the technique is a major performance issue, shows the same 10x performance differential on 10.8 and 10.9 so it doesn't point out any performance "differences" that are new in Mavericks.
22-07-2014
The essential mystery of this issue is why we saw no improvement from the fixes for RT-30107 so I am linking to that bug as "related". This is possibly related to a lack of improvement for the same fix on an older Mac Mini so I am also linking RT-37920 as related.
22-07-2014
Note, from RT-30107, running on Mavericks: 600 monsters on Retina, 15-inch, Early 2013/2.4 GHz Intel Core i7/NVIDIA GeForce GT 650M 1024 MB Rectangle clip: 60 FPS Oval clip: 6.0 FPS Small oval clip: 6.5 FPS Same test, same hardware, running on Mac 10.10 Yosemite Dev Preview 2: Rectangle clip: 60 FPS Oval clip: 12.51 FPS Small oval clip: 16.14 FPS
16-07-2014