I've discussed the wear os baseline profile here a while ago, so I'd like to share the results of ap...

# compose-wear

[barat]

I've discussed the wear os baseline profile here a while ago, so I'd like to share the results of applying it to my product.
In my case, since scrolling was slow right after the app was turned on, so I considered applying baseline profile.
Since rooting is required to create a baseline profile on a real wear device, I was created baseline profile on AVD.
After that, the produced baseline-profile.txt was applied to the real device and the performance test was conducted.

StartupBenchmark_None
timeToInitialDisplayMs. min 1259.9, median 1355.6, max 1401.2
StartupBenchmark_Partial
timeToInitialDisplayMs min 928.6, median 954.6, max 982.5

ScrollBenchmark_None
frameDurationCpuMs P50 12.9, P90 34.0, P95 66.9, P99 388.8
ScrollBenchmark_Partial
frameDurationCpuMs P50 12.5, P90 28.7, P95 51.8, P99 156.2

It helped improve the overall speed at start-up, and helped improve the bad case of p90 or higher at scroll.
Since production and application can be done separately, if you do not want rooting, it is best way to make baseline-profile with AVD.

[John Nichol]

Thanks for sharing your result @barat. We also see a huge difference with baseline profiles which is why we recommend them for all apps. @Steve Bower [G] I seem to recall you did some exploration of the benefit of app specific profiles over the generic ones from the library.

[stevebower]

Yes - app specific profiles can make a difference to both the startup time and initial rendering of a page. This is covered really well in Ben's talk at I/O this year, see

https://www.youtube.com/watch?v=EOQB8PTLkpY&t=926s▾

. @barat's 30% improvement on startup is in line with the figures presented there.