| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
| |
bug:5461107
Tiles were being destroyed, and subsequently dereferenced in TransferQueue
Change-Id: I4fea289e5fda03a69f07554f57120c4c5bf7b016
|
|
|
|
|
|
|
|
|
|
|
| |
bug:5453156
simplified m_owner management - ONLY modified on UI thread
Previously deferring clear of m_owner, now set immediately since texture thread
doesn't use need stale owner info
Change-Id: I9b6f7bfd27af44a818d378ff750c0170d8122cfd
|
|
|
|
|
|
|
|
|
|
|
|
| |
bug:5409902
Tiles were being discarded from the queue (and simply unsuccessfully
added). This caused them to get stuck in the 'ValidatedUntransferred' state.
Now if a tile isn't added successfully, or if it's discarded, it removes its
painting texture and will have to repaint from scratch.
Change-Id: I551e00fb8a6be3b0f3cabeabaa91e8b8b30019d5
|
|\ |
|
| |
| |
| |
| |
| | |
bug:5395019
Change-Id: Ifc16eb59fec62e5f4b95ad4e35f140969226506d
|
|\ \
| |/ |
|
| |
| |
| |
| |
| |
| |
| |
| | |
This relied on the hacky Surface Texture change:
https://android-git.corp.google.com/g/#/c/138516/
bug:5347539
Change-Id: I78e0f677a9c201ba49e11ddb24fa095471cbf3c9
|
|\ \
| |/ |
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Basically, move unnecessary function calls out from the loop.
And move the clear related code into DEBUG only.
This is saving us some CPU overhead.
bug:5347539
Change-Id: I237162fd6a502e93ea32c9856f907c5a9ddb2e78
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
bug:5369978
Manage invalidations, painting, transferring, and swapping with a
state machine.
notes:
* readyFor shouldn't be needed if all of the events that would cause a readyfor
fail mark the tile dirty (such as stealing, scale change)
* changing the scale of a page should discard all textures
* m_dirty should be more gracefully worked into the state machine
* a tile may drop dirtiness notifications if it's already painting, since upon
completion m_dirty will be cleared - (issues for layers only)
Change-Id: I5909fb5d208da2fb276e223c56bf143741a9a24c
|
|
|
|
|
| |
bug:5347539
Change-Id: I12b8b45792d5aa684295b897be308181a7d44631
|
|
|
|
|
|
|
|
|
|
|
|
| |
The default for WebView should be still be GPU upload.
That means Gmail will be default to using GPU upload.
bug:5347539
The browser change is in
https://android-git.corp.google.com/g/#/c/137606/
Change-Id: Ic7d42331511b24cf8a58f2f5fb64aaffc18cae5f
|
|
|
|
|
|
|
|
|
| |
This can save the draw call perf up to about 20-40% for each quad.
We should make more draw calls using NEAREST mode when it is 1 to 1 mapping.
bug:5347539
Change-Id: I1ae206716f2b1352775e0079e25e54f46bc18578
|
|
|
|
|
|
|
|
| |
Clean up the unnecessary Surface Texture operation in CPU upload code path.
bug:5347539
Change-Id: Id8cf1d3e472dce2470bc62a844b95e8d5d2a6371
|
|
|
|
|
|
| |
bug:5044597
Change-Id: Ib70e48f4503a85952afb28da3a6c500b0c011bb8
|
|
|
|
|
|
|
|
|
|
| |
Currently disable the wait b/c b/5332112.
For now it should be the same functionality as the original workaround by
using glDraw.
bug:5270526
Change-Id: Ia6d6739909994d6ed756c6b195e6cf7ee9518c54
|
|
|
|
|
| |
bug:5270526
Change-Id: I790a3fc995edc73192b2bcb24eaa29e99434f2fb
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| | |
bug:5269460
Deallocate the graphics memory backing a BaseTileTexture on onTrimMemory
signals, and accordingly allocate it lazily, as needed.
Change-Id: I52039723f47e6470e4fe8dd987d384017005390f
|
|\ \
| |/
|/| |
|
| |
| |
| |
| |
| |
| | |
This is just for debugging anything related to the driver/HW.
Change-Id: I26757f8259ec6b6c2406f78bf7aa9c5f527e31e7
|
|/
|
|
|
|
|
|
|
|
|
|
| |
bug:2522049
allocate textures and tiles using the gldraw count when they were most recently
prepared
remaining issues:
-layers still flicker (presumably from texture stealing)
-layers aren't double buffered yet
Change-Id: Iccdf68326d7d476269d4e3a13903aaab249ee92d
|
|
|
|
|
|
|
|
|
| |
The lock over the surface texture operation is unnecessary.
They just need to be around the transfer queue operation.
bug:5185840
Change-Id: Ia09ec5f6049421047aac1944efbf601cdc179a52
|
|
|
|
|
|
|
|
| |
When the stride is the same as the bitmap width, we can just use single memcpy
to copy the whole content.
bug:5196685
Change-Id: I01c9a673c8e95e14eb23f3d7df4002c886b7554d
|
|
|
|
|
|
| |
bug:5161750
Change-Id: I33735f258abf850fa3afd7769385b38691365da3
|
|
|
|
|
| |
bug: 5013645
Change-Id: I255a19c68d8a98231b1500d93d21649d15d70820
|
|
|
|
|
|
| |
bug:5044597
Change-Id: Id8cd326774a174a537f2b104d934539269395151
|
|
|
|
|
|
|
|
|
|
| |
It seems like the FBO operation is out of sync in some GL driver.
Instead of doing a readPixel, we can also force a read after write by calling
an extra draw.
bug:5118763
Change-Id: Idaf259f84b0dcc0f3bc94bcf1bd1a4adf5250350
|
|
|
|
|
|
|
|
| |
Improve the sync part for the Shared Surface Texture.
Clean up the interface as much as possible.
bug:5044597
Change-Id: I123cc520fb447aeb6e72e90149d1658a657740f7
|
|
Instead of assigning each tile a Surface Texture, now just a normal GL texture.
The content from skia bitmap will first drawn into a queue of Surface Textures,
at draw time, blit them into each tile's GL texture.
Added extra lock to protect the queue operation and Surface Texture operation.
bug:5044597
Change-Id: I25f46228b93a3e99526daf52d8bd0d8d8fcc2879
|