Regression in timeToFirstFrameRasterizedMicros on Mac_android_complex_layout__start_up

[zanderso]

From the mid-500s ms up to high 700s ms.

https://flutter-flutter-perf.skia.org/e/?begin=1624501862&end=1627353183&keys=Xc8b23a2e76f54575dead81d68735842f&requestType=0&xbaroffset=25119

Suspecting the engine roll here: #86825

Maybe flutter/engine#27629 ? /cc @dnfield

[dnfield]

I guess it's possible that creating the additional surface is causing this. I'll try to look.

[dnfield]

I'm pretty sure it was the pbuffer commit.

I have an idea to fix forward that I'd like to test - if I can't prove it quickly I'll revert.

[zanderso]

These memory benchmarks may also be related:

https://flutter-flutter-perf.skia.org/e/?begin=1624501262&end=1627399444&keys=Xa8fa28eb4feeb4575c2e5a574479b7c7&requestType=0&xbaroffset=25120

[dnfield]

So before the pbuffer patch, the ShaderWarmUp routine did not actually produce any shaders, because it ended up working with a raster based surface. Now, however, it does produce shaders.

[dnfield]

Specifically, this line here:

flutter/packages/flutter/lib/src/painting/shader_warm_up.dart

Line 95 in ea1182e

final ui.Image image = await picture.toImage(size.width.ceil(), size.height.ceil());

[dnfield]

I'm inclined to drop the default shader warmup routine from the framework - @zanderso @chinmaygarde WDYT?

Internal customers that are using this have their own shader warmup specializations. However, this never does anything on Android or iOS.

[dnfield]

Both the memory and the time to first frame regressions improved after removing the default shader warm-up routine.

[github-actions]

This thread has been automatically locked since there has not been any recent activity after it was closed. If you are still experiencing a similar issue, please open a new bug, including the output of flutter doctor -v and a minimal reproduction of the issue.