Stabilize FPS more

[taisel]

It seems some code re-arranging the last few weeks de-stabilized the fps. I verified it's not the offthread rendering dropping frames, but it's just possibly all the added extra noise causing the drops between cross thread comms. Goal here will be to remove any timers on the main thread when offthread cpu/gpu is used, and to rework the graphics stack slightly.

[taisel]

I'm getting 1 frame dropped every second in Firefox. That pattern of frame drop seems a bit too predictable to be overhead noise. Investigate further and plot eventing timelines to validate.

[taisel]

I'm getting 1 frame dropped every second in Firefox. That pattern of frame drop seems a bit too predictable to be overhead noise. Investigate further and plot eventing timelines to validate.

[taisel]

6ee5efa

We temporarily killed the workaround for iOS, to clean up the code and reduce jank on main thread. The issue for iOS being if the emulator runs in single thread mode AND eats up 100% of a CPU core on an iPhone, then iOS safari refuses to issue requestAnimationFrame calls at all, not even at 1 fps.

[taisel]

6ee5efa

We temporarily killed the workaround for iOS, to clean up the code and reduce jank on main thread. The issue for iOS being if the emulator runs in single thread mode AND eats up 100% of a CPU core on an iPhone, then iOS safari refuses to issue requestAnimationFrame calls at all, not even at 1 fps.

[taisel]

That workaround killing not being the root cause though. The root cause was drawing after every frame copy rather than drawing once per vsync outside of the user callback, and this was a regression.

[taisel]

That workaround killing not being the root cause though. The root cause was drawing after every frame copy rather than drawing once per vsync outside of the user callback, and this was a regression.

[taisel]

Hopefully by the time Apple adds shared memory support to iOS Safari's JavaScript VM, they'll have fixed the iOS 8+ JSC crasher/regression. It's why removing that workaround for the time being is non-consequential for the time being, outside of testing and building ahead of such.

[taisel]

Hopefully by the time Apple adds shared memory support to iOS Safari's JavaScript VM, they'll have fixed the iOS 8+ JSC crasher/regression. It's why removing that workaround for the time being is non-consequential for the time being, outside of testing and building ahead of such.

[taisel]

iOS safari should be using its own "glue code" anyhow for tighter integration, the UI code here anyhow is reference implementation type bare bones.

[taisel]

iOS safari should be using its own "glue code" anyhow for tighter integration, the UI code here anyhow is reference implementation type bare bones.