Benchmarks on bullhead-userdebug show that there is a negligent
performance impact when using sparse entries on a 30% loaded
sparse type of 1000 resources.
Benchmark Time CPU Iterations
-----------------------------------------------------------------------------------
BM_SparseEntryGetResourceSparseLarge 255 ns 254 ns 2751408
BM_SparseEntryGetResourceNotSparseLarge 254 ns 254 ns 2756534
Bug: 27381711
Test: make libandroidfw_tests aapt2_tests
Change-Id: I051ea22f2f6b2bc3696e446adc9e2a34be18009f
Use Google3 naming style to match new
projects' and open source google projects' style.
Preferred to do this in a massive CL so as to avoid
style inconsistencies that plague legacy code bases.
This is a relatively NEW code base, may as well keep
it up to date.
Test: name/style refactor - existing tests pass
Change-Id: Ie80ecb78d46ec53efdfca2336bb57d96cbb7fb87
PngCrunching now has a slightly better heuristic of choosing to encode
an image as a palette or RGB. For small images, RGB compresses much better
than a palette.
The original PNG is used as-is (minus some optional chunks being stripped)
if the resulting crunched PNG is larger than the original.
9-patch handling is abstracted away from PNGs, paving the way
for other 9-patches, like WebP.
TODO: handle PNGs with 9-patch chunks already present, which
should just be passed through. This will allow for 3rd party
tools to generate 9-patches.
TODO: implement cheap transparency: when one color is used to represent
transparent, and all other colors are opaque.
Bug:30053276
Change-Id: I5167f53b91d1efa462d9f03d6b9108d9b541c0c1
An early refactor. Some ideas became clearer as
development continued. Now the various phases are much
clearer and more easily reusable.
Also added a ton of tests!
Change-Id: Ic8f0a70c8222370352e63533b329c40457c0903e
