| Commit message (Collapse) | Author | Age | Files | Lines |
|\
| |
| |
| |
| |
| |
| | |
recursive" into jb-dev
* commit '67169ebc1651937450492ebea558fedda74ca50a':
Make viewport dirtyness calculations recursive
|
| |
| |
| |
| |
| |
| |
| |
| | |
If a layer marks the viewport dirty, it should mark all of its children as well,
since they likely have also moved.
bug:6547356
Change-Id: I0db2704f35e7c8a78108d6ed8372201a7f025994
|
|/
|
|
| |
Change-Id: I2a53b157d40c630e7ca1361f8e0c8aca1d6e8a4e
|
|
|
|
|
| |
bug:6605853
Change-Id: I65583e61197452ac70c5adf5f967182f0fd7578a
|
|
|
|
|
|
|
| |
Also isolates layer translation fudging so as to not affect child transforms.
bug:6585341
Change-Id: If416c6fad90ee37fbe3438e01706b34e683165b7
|
|\ |
|
| |
| |
| |
| |
| |
| | |
bug:5683630
Change-Id: I43f738f2649a79b4ad7865ed27375c07195fa9b9
|
|/
|
|
|
|
|
|
|
| |
If the current clip covers the entire view, we don't have to calculate framework
invals from swapping layers - we already draw them immediately.
bug:6530473
Change-Id: I4b639e06781371e37c246f3220b79f28e7d95649
|
|
|
|
|
|
|
|
|
| |
Previously, hundreds of layers could take 4+ ms to update transforms. Now, only
those that are fixed, have animations, or are scrollable calculate transforms.
bug:6530473
Change-Id: Ic9b14b76190c5bd0e5a405914a624d92b7ca7980
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| | |
Layer can now use the offset to paint, resulting in correct translates for
overflow:scroll layers in the Layer::draw() path.
bug:6548000
Change-Id: I1dfb88b15491e72e6c3841ce2953e114124e40fd
|
|/
|
|
|
| |
bug:6537803
Change-Id: Ie0bb6e64644fd82c0ed6cf92c592c434f67fdc0d
|
|
|
|
|
|
|
|
|
| |
Draw transforms are used at painting time for merged surfaces, so they shouldn't
be manipulated in the draw path. The fudging isn't necessary for these surfaces
in the first place, as merged surfaces are always aligned to pixel coordinates.
bug:6156523
Change-Id: I3d954869188008f1a9e35915df893d75b49be6a2
|
|
|
|
|
|
| |
bug:6522485 bug:6486951
Change-Id: Ia707efed9311e4352eed3ab005ec7ca93b8444e4
|
|
|
|
|
| |
bug:6479200 bug:6323847 bug:4124445
Change-Id: I1a4058ba6c69d3e285b6274d99a6eafcbf1cdc6f
|
|
|
|
|
|
|
|
|
|
| |
This will reduce the need to go to single surface rendering mode,
as well as improving jank by reducing the number of surfaces we need.
(tested on a set of popular sites, we now need up to 16 times less
surfaces, commonly 4-5 times less)
bug:5683630 bug:6499283 bug:6366440
Change-Id: I761c6a60279e5b21dca8bc1faccd956be58e3702
|
|\ |
|
| |
| |
| |
| |
| |
| |
| | |
This was causing content to go missing.
bug:6504906
This reverts commit 0b072353e2ad89087c944dfb4acb1eff87bbac49
|
|/
|
|
|
| |
bug:6331876 bug:6366440 bug:6182286
Change-Id: I8390f734ef5894219ea50a7b3e36a029e0e7931d
|
|
|
|
|
|
|
|
| |
This will reduce the need to go to single surface rendering mode.
bug:5683630 bug:6499283
Change-Id: I482060b38d856c81ae8eff296e76252892e6a610
|
|
|
|
|
|
|
|
|
| |
Remove old bounds calculation method.
Depends on frameworks/base change: https://android-git.corp.google.com/g/#/c/190474/
bug:6486310
Change-Id: Ieeb09778f810598c38127c66ae12b2b29abcf8e4
|
|
|
|
|
| |
bug:6331760
Change-Id: Iddc044c11bf7fcb96e061e346481d0228803cc9e
|
|
|
|
|
| |
bug:6221600
Change-Id: I974eb846200d43bd3635ed1787a5753f87a547dd
|
|
|
|
|
|
|
|
|
|
| |
- extract image for body background (a lot more memory-friendly)
- implement tiling / repeat
- handles background color
bug:1352305
Change-Id: I0efa27e09416e3c3848a4a53ced650cbb3d9f7ce
|
|
|
|
|
|
|
|
|
|
|
| |
Use the inverse of draw tranform to convert from content to layers
coordinate space.
Also, if the draw transform isn't limited to 2D space, return the entire
content area. Above conversion doesn't preserve 3D.
Bug: 6077575
Change-Id: I8393de0597ae5cd739b3df10c087cedd234ac9f0
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| | |
Bug: 6445600
Clip fast partial invals to the screen instead of an arbitrary size
limit
Change-Id: Iafcbcb427566db87420b8bfc54eb9148530913ce
|
|/
|
|
|
|
|
|
|
| |
The rects are referring to the same rect under different coordinates,
so we just prefix the coordinates' name to differentiate the rects.
This is pure refactor, no functional change.
bug:6338456
Change-Id: Ic072a4f5aa56a25751e0151a697c0e31bca94ef5
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This changes how partial invals are done by adding a hybrid mode.
What we used to do is generate a SkPicture for the new area. This
SkPicture would possibly be larger than the actual inval, depending
on various merge rules (more SkPictures == slower to draw a tile)
The new code rewrites PictureSet entirely, preserving many of the old rules
but cleans up the code and adds the concept of a "PrerenderedInval". This is
a partial inval that WebKit has rasterized. By having WebKit produce both
a SkPicture and a SkBitmap, we avoid needing to play back the picture and
avoid overdrawing. We take this SkBitmap, and simply update the front
textures with it. This gives us full partial invals through the entire
system without hitting any driver bugs, and with minimal copies. And while
the SkPicture may be larger than the inval, the SkBitmap that is rasterized
is not - it matches the area webkit has said is dirty.
Change-Id: Ieb7ecc9db0d4f679102fda004a43399f9b319ebc
|
|
|
|
|
|
| |
bug:1352305
Change-Id: Id9caaae9b9442729110b52c75004f634d8284db4
|
|\ |
|
| |
| |
| |
| |
| | |
bug:6317028
Change-Id: Ibec896f64371c29a25aaae2490a4e5cf97b5b54a
|
|/
|
|
|
|
|
|
|
|
| |
Multiple paths were redundant, as they both synchronized the layers, and drew
the base if needed.
depends on frameworks/base change: https://android-git.corp.google.com/g/#/c/179964/
largely reverts 3adbe453816c9282bfe1d212e813661ce2590955
Change-Id: Ic60f79521dc62e7c2281302bc51a3dfde16d396f
|
|
Change-Id: Idc67155cfa99784dcd931e705336bfa063ecae46
|