diff options
author | Patrick Baumann <patb@google.com> | 2020-05-27 16:39:45 -0700 |
---|---|---|
committer | Alex Buynytskyy <alexbuy@google.com> | 2020-06-15 17:33:49 -0700 |
commit | 2377f33aceae0c0e8b4c06c391653963fbb640dd (patch) | |
tree | d04f671777fd6c9de3c16b5d7c90ae14b4aed613 /tools/aapt2/java/JavaClassGenerator_test.cpp | |
parent | 88d54c764f952af0321112c243f8a44809588ff6 (diff) |
Pre-cache filter results
This change adds a cache layer for the return of
shouldFilterApplicationInternal in AppsFilter. This cuts most operations
that rely on app filtering nearly in half by removing a good deal of
branching that used to exist on the critical path for a crosshatch.
Before this CL:
android.os.PackageManagerPerfTest#testGetApplicationInfoWithFiltering:
getApplicationInfoWithFiltering_mean: 983719
getApplicationInfoWithFiltering_standardDeviation: 120061
getApplicationInfoWithFiltering_median: 1061674
getApplicationInfoWithFiltering_min: 827489
android.multiuser.UserLifecycleTests#createAndStartUser:
Mean (ms): 4243.10
After this CL:
android.os.PackageManagerPerfTest#testGetApplicationInfoWithFiltering:
getApplicationInfoWithFiltering_mean: 426340
getApplicationInfoWithFiltering_standardDeviation: 18861
getApplicationInfoWithFiltering_median: 427002
getApplicationInfoWithFiltering_min: 407887
android.multiuser.UserLifecycleTests#createAndStartUser:
Mean (ms): 3387.25
Note: this is a 2nd attempt at this change that fixes a few from the
previous at ag/11622391. This works with multi-user and updates tests to
ensure this stays true. It also improves peformance when building the
cache by pre-allocating the SparseBooleanArrays to their max sizes.
Test: atest AppEnumerationTests AppsFilterTest PackageManagerPerfTests
Fixes: 150405193
Change-Id: I6e0446068f46af0f22e0259ab7b6cdbbc7e08a22
Diffstat (limited to 'tools/aapt2/java/JavaClassGenerator_test.cpp')
0 files changed, 0 insertions, 0 deletions