diff options
| author | Misha Wagner <mishaw@google.com> | 2021-09-20 13:42:27 +0100 |
|---|---|---|
| committer | Misha Wagner <mishaw@google.com> | 2021-09-20 18:16:31 +0000 |
| commit | 7d611c750c3d518794b2e580539bf9a22222a41f (patch) | |
| tree | dfaf68c2ff225e86cee7e12787ac47a39e447366 /rs/java/android/renderscript/ProgramFragmentFixedFunction.java | |
| parent | f81d50e255b53cb2e3e54d7159978a8035b162f7 (diff) | |
Change CacheOomRanker's "uses" feature.
The previous implementation was incorrect, as an app goes in and out of
the cache several times during oom_adj calculation. If all apps started
at the same time, then this would be OK - as they all get incremented
the same amount, plus get incremented when they actually go in and out
of the cache. This would result in the relative counts being correct.
However, as all apps aren't started at the same time, the old
implementation is instead a measure of how long the app has been
started.
So instead we use mSetProcState, which isn't changed on each oom_adj
calculation. This measure should accurately reflect the number of times
the process is used - whether that's by a user app open, or a content
receiver call, or an intent being broadcast.
See go/sim-v-impl for more details.
Test: atest CacheOomRankerTest
Bug: 196031723
Change-Id: I49d02362b8277f8472355e78412f79256c2289a1
Merged-In: I49d02362b8277f8472355e78412f79256c2289a1
(cherry picked from commit 480cc3123c5ee6ae4fff56599124794a0ff25aa4)
Diffstat (limited to 'rs/java/android/renderscript/ProgramFragmentFixedFunction.java')
0 files changed, 0 insertions, 0 deletions