I’m trying to run bunnymark using openfl html5 with the pixi.js backend to compare it against a standard js/ts pixi.js bunnymark test to gauge whether or not it’s a viable option vs plain js/ts.
I’ve installed haxe/openfl as per the insctructions on the official site and downloaded the bunnymark from https://github.com/openfl/openfl-samples.
However performance wise the best I seem to be able to get is 6k bunnies dropping to 30fps for desktop and 500 bunnies at 15fps for android using the CopyPixelsTest. This pales in comparison to pixi.js using js/ts which i’m getting 50k+ at 60fps for desktop and 5k+ at 60fps for android.
Is that expected performance for openFL or have I missed out something and it’s not really using pixi.js to render or have I missed something to get similar performance to standard js/ts pixi.js?
I’m using FlashDevelop 4.6 to compile with 3.1.3.
What do you get if you use -Dwebgl?
There are some differences in the Pixi-based test (for example, using a small canvas and CSS scaling instead of matching the fullscreen) that do make a big (deceiving) difference as well
Thanks, I wasn’t aware of the webgl flag I will test it and post back the results in comparison to previous.
On the scaling side are you saying that openfl uses a large resolution and scales down or that pixi js/ts uses a small resolution and scales up? Based purely on the output it looked to me that openFL used a smaller canvas for mobile and a huge one for desktop.
If anyone can post an optimized bunnymark in regards to html5 and in particular mobile, making use of the most recent openFl changes it would be really appreciated.
There’s been a big focus on accuracy lately, I haven’t swept through and compared performance in a while. I see regressions, but I think it’s performing pretty well, here’s a sample: