| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
| |
Bug: 63756964
Test: test-art-host
Change-Id: If8748db7a9e68b6a82656fd9ec8fd8072e92af97
|
| |
|
|
|
|
|
|
|
|
|
| |
For fixed_up_dexfile, the returned dex only needs to contain a single
class. This CL adds filtering during IR creation to only contain that
class. This speeds up wifi tests from 25s to 15s on marlin.
Bug: 63756964
Bug: 72064989
Test: mm test-art-host-gtest-dexlayout_test -j64
Change-Id: I3648ac886b377d95e33e5b40a63a34c2a9dbde58
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Delete all the offset related layout logic, move the logic to
writing. Layout is now controlled by changing the order of the data
vectors.
Calculate all offsets during writing instead of reading. This allows
more flexible layout optimizations without needing to be as careful
about offsets in other sections changing.
Maintained logic for using existing offsets in the case where
changing the layout is not required.
Areas to improve:
How to size the memmap, do we want 2 passes? Currently the algorithm
reserves 10% extra space in case required. This is sufficient for top
1000 apps in the play store. Will consider a two pass approach later.
Bug: 63756964
Bug: 68948395
Bug: 68867570
Bug: 68864106
Bug: 68864168
Test: test/testrunner/testrunner.py --host -j40
Test: test-art-host-gtest
Test: Dexlayout speed profile top 1000 apps
Change-Id: I7dee869da3a010559547f8cfdf93e9aa4c3f47ff
|
|
|
Move all the construction from DexFile out of the constructors of the
basic IR.
Bug: 29921113
Change-Id: I3f79c104ce7183ddde73f143c047061416009a54
Test: test-art-host-gtest
|
