Status Update
Comments
ph...@gmail.com <ph...@gmail.com> #2
ph...@gmail.com <ph...@gmail.com> #3
Branch: androidx-main
commit e7b4b06e67951ab47157d06ff314a5c8b2e8fdc6
Author: Andrei Shikov <ashikov@google.com>
Date: Mon Jan 08 15:54:51 2024
Count recursive local declarations as captures
Local declarations were already counted as captures for composable lambdas before, but the traversal ordering missed the recursive captures, converting some of them into singletons.
Test: Compiler test
Fixes: 318745941
Change-Id: I9097d1be71fb67b73e5027f723fd187c4272f6b4
M compose/compiler/compiler-hosted/integration-tests/src/androidUnitTest/kotlin/androidx/compose/compiler/plugins/kotlin/ComposeBytecodeCodegenTest.kt
M compose/compiler/compiler-hosted/integration-tests/src/jvmTest/kotlin/androidx/compose/compiler/plugins/kotlin/AbstractCodegenTest.kt
M compose/compiler/compiler-hosted/src/main/java/androidx/compose/compiler/plugins/kotlin/lower/ComposerLambdaMemoization.kt
se...@google.com <se...@google.com> #4
Rahul, don't know if this is a real issue or a benchmark being coded poorly thing. Passing along due to profileinstaller.
ca...@gmail.com <ca...@gmail.com> #5
ra...@google.com <ra...@google.com> #6
Hi Marcelo, can you please take a look at this when you get a chance ? If not, I can come and take a look when I get back.
ma...@google.com <ma...@google.com> #7
I'm pretty sure the offending PR here is aosp/2333568 - that is the only PR touching the profile transcoding.
It fixes an important bug with compressed profiles but it seems that the new code is slower than before.
cc...@google.com <cc...@google.com> #8
If you look in the dashboard, the commit link is wrong, since it doesn't encompass all builds since the previous (9385400). Correct build range link is here, which includes Ben's CL:
(If a build goes missing for any reason, it corrupts the commit link. In general, you can always go to dashboard, and create range link like the above)
be...@google.com <be...@google.com> #9
Yes, it looks like my change introduced this regression.
If I read
That would mean the change makes opening the file slower by about 1000 times.
Chris: Given this runs on a background thread and is not blocking app startup, is it urgent to address this or can it wait for the next release cycle?
be...@google.com <be...@google.com> #11
Uploaded aosp/2355662 to address the problem.
be...@google.com <be...@google.com> #12
The CL mentioned in
mo...@google.com <mo...@google.com>
ap...@google.com <ap...@google.com> #13
Branch: androidx-main
commit b7fc1b0247a4f4883b37c1998e3f12b4f0c0b7b0
Author: Ben Weiss <benweiss@google.com>
Date: Wed Dec 14 15:49:17 2022
More detailed compressed profile handling
Just replacing openFd with open dramatically decreased the performance
of opening a profile. This new handling goes back to openFd and
introduces a diagnostic code to hint at updating bundletool.
Bug: 261998144
Test: :profileinstaller:profileinstaller:test
RelNote: Add diagnostics for compressed profiles
Change-Id: I8641387ad6073dde588d0d764da4a3b24b1c8ee4
M profileinstaller/profileinstaller/api/current.txt
M profileinstaller/profileinstaller/api/public_plus_experimental_current.txt
M profileinstaller/profileinstaller/api/restricted_current.txt
M profileinstaller/profileinstaller/src/main/java/androidx/profileinstaller/DeviceProfileWriter.java
M profileinstaller/profileinstaller/src/main/java/androidx/profileinstaller/ProfileInstaller.java
be...@google.com <be...@google.com> #14
The landed CL has moved the benchmark down to normal levels again.
pr...@google.com <pr...@google.com> #16
The following release(s) address this bug.It is possible this bug has only been partially addressed:
androidx.profileinstaller:profileinstaller:1.3.0-beta01
Description
Alert on dashboard:https://androidx-perf.skia.org/t/?begin=1670529681&end=1670529682&subset=all
CLs in build:https://android-build.googleplex.com/builds/branch-dashboard/aosp-androidx-main?build_id=9385668
The commit points to a CL that has no real change (just reformatting), so something else is going on.
Sean, can you look and see why this benchmark is going crazy?