In one set of the fixed cases we actually were never closing the
fd. In the other set we would crash with a fatal error. Now both
have been updated to handle the failures and correctly close the
fd to make sure it all works.
However, I don't think this is the route case of the leaking fds
from the bug since I don't believe we are seeing any of the
corresponding error messages in the logs.
Test: Manual running on phone and watching logs.
Bug: 187240173
Change-Id: I18babe91b0ec83ca63199f06d3d02ec10e8aea85
The initialize call can be made from both an GrallocUploadThread
and the RenderThread. So we need a mutex to make sure both
threads don't try to initialize in parallel.
Bug: 187218890
Test: manual running of device, guessing this could fix bug.
Change-Id: I203afd91fad5eacc131c839dff1a1065864b76f8
Previously we would submit a command buffer for the
frames draws and then immediately submit a second
command buffer that had no commands and just an attached
semaphore. This merges those two submit calls into Skia
into one call.
Test: manual running of system of benches
Bug: 175913056
Change-Id: If0d054960de8b459814cbfa9289756f8ee9d4a93
(cherry picked from commit f16ba60194)
* Degrade error when changing the rendering backend from Properties to
no longer be fatal.
* Support using a single graphics queue for the Vulkan backend. Cloud
devices only support a single queue and some Vulkan implementations such
as Swiftshader expose only one queue for simplying their implementation,
so we'll need to support one.
Bug: 175618060
Bug: 162628999
Test: hwui_unit_tests
Test: experiment on Pixel 4 enforcing a single queue (settings app,
maps, sysui)
Change-Id: I495fcabc3c89bd62bbb833998eab4944c6660f6f
The difference between recording & direct contexts does not matter
to Android right now because they don't use DDLs, so there's no
functional change here.
Change-Id: Iedcdf5c3ce0ccd3c7e1724d8fb45b7882d8c2129
This change also fixed that the graphics queue should be the one to wait
on upon surface destruction.
Bug: 162628999
Bug: 167693610
Test: no new regressions in CtsUiRenderingTestCases with Vulkan backend
Change-Id: I919748b85b2194bb4f13e1af6950d6493a17921d
The VkInstance and VkDevice are constructs that can be shared between
threads and avoid the extra driver allocations associated with them.
This CL also fixes a bug where Vulkan uploads were occuring on the
thread that invoked the upload and not occuring on the upload thread.
Bug: b/162628999
Test: atest CtsUiRenderingTestCases
Change-Id: I0289938333e2c3dafcc2ce18e72dca3acf4d5b79
This is part of an effort to distinguish between different types of
contexts in GPU Skia. When using a DeferredDisplayList (DDL)
recorder, the context you get is not a direct context and cannot
be used for operations like uploading or reading textures.
Since Android does not use DDLs, it is not directly affected by this
change but other APIs, such as SkImage::MakeFromTexture are
being migrated to require a GrDirectContext to increase sanity.
Change-Id: I9afbdf3c026a9f9cb6ad2aad904915e189e584d6
Revert submission 10449863-revert-fatvector
Reason for revert: b/149254345
Original change (of introducing FatVector) was reverted as a stop-gap solution to fix the aforementioned bug.
The bug was caused by an ABI lock between Surface's definition (that changed with Region) and lib-imsvt prebuilt.
Enabling this change now as we have re-compiled the prebuilt with the change enabled. Doing that via a revert of the revert.
Reverted Changes:
I8ac66acb8:Revert "hwui: remove FatVector"
Ib60dbf3ef:Revert "libui: rewrite Region with FatVector"
Original changes:
I09dc2fddd:hwui: remove FatVector
I265c6c831:libui: rewrite Region with FatVector
bug: 149254345
Change-Id: I9fb5e8908434abb9a763992e922445a2dc37e865
Revert "libui: rewrite Region with FatVector"
Revert submission 10248126-fatvector-region
Reason for revert: b/149254345
Reverted Changes:
I09dc2fddd:hwui: remove FatVector
I265c6c831:libui: rewrite Region with FatVector
also fix wrong license in FatVector.h
Test: boots
Bug: 149254345
Change-Id: I8ac66acb8b635324051edd41c5d4092c223157ff
Remove all Skia and HWUI types from SurfaceTexture
implementation.
Move SurfaceTexture to libgui (ag/9578265).
Define private C++ API for SurfaceTexture, which is consumed
by DeferredLayerUpdater.
Move AutoBackendTextureRelease/Skia code from SurfaceTexture
to HWUI.
Test: pass CtsUiRenderingTestCases and CtsViewTestCases
Bug: 136263580
Change-Id: I3f971bb490f64a3ac0b2a66a89ba935bf7f08213
1. Android support all transforms as exposed by client composition.
2. Properly exposing min/max image size is tracked in http://b/134182502.
3. Current image size and max buffer count are cached in Surface already.
4. Remove surface support check since hwui does not have YUV + CPU consumer.
5. Since HWUI must require those image usage flags, just use them directly,
and those are always supported by Vulkan driver and enforced by cts.
Bug: 136110987
Test: CtsUiRenderingTestCases and CtsGraphicsTestCases
Change-Id: I89bd9afb69a46e3deca09890ce6348c6b61edddf
We were deleting the VkSemaphore objects too quickly when
importing/exporting the semaphores. Even though the semaphore payload
gets reset on these operations the VkSemaphore still needs to be
finished its use in Vulkan before being deleted.
Test: manual build and testing of vulkan apps and vulakn ImageConsumer
Bug: b/130643604
Change-Id: I7f03087e477d812c0174ede3a10f12dc1df72ee1
Anytime dequeueNativeBuffer fails, it sets mDequeuedIndex to -1.
This is causing a crash latter, when getCurrentSkSurface tries
to index mNativeBuffers[mDequeuedIndex].
This CL removes mDequeuedIndex and improves Vulkan swapchain
error handling.
Test: Ran Camera app
Bug: 129024275
Change-Id: Ieeb685c3a1b33f23ce2334d286199a44ace53165
When Vulkan pipeline dequeues next frame at the beginning of
DrawFrame, the dequeue fence has been signalled in most cases.
This CL avoids additional work and saves about 0.3ms per frame.
There is no need to create VkSemaphore and commit an empty command
buffer to the queue if the fence has already signalled.
Bug: 128998567
Test: Ran systrace on SelfieCity and observed better performance
Change-Id: I3532b785fae90308d922a29f1698f5dbcbd79079
If dequeueBuffer fence has not fired yet, then submit to the queue
and block GPU on VkSemaphore immediatelly.
It is not clear why this CL makes a difference, because semaphore
was already waited on.
Fixes: 128805387
Fixes: 128806388
Fixes: 128807485
Fixes: 128879779
Fixes: 128417453
Bug: 128998567
Test: Ran SelfieCity, Camera360, Settings app
Change-Id: I880ba6fa65509ba4c6faca0dc933551efcee423b
Replace it with a newer, fancier, WorkQueue-inspired
one that's just a global common thread pool.
Test: hwuiunit passes
Change-Id: Ib5d03104a08bbac9a4ec67a1bfc0db2b35d6700f
Instead of relying on Vulkan swapchains this CL enables HWUI to directly
manage the native window. This allows us to preallocate buffers using
our own strategy as well as having no longer having to jump through an
unecessary translation layer that resulted in code that was hard to
reason about and also introduced inefficiencies.
Bug: 123541940
Bug: 119687951
Test: CtsUiRenderingTestCases and CtsGraphicsTestCases
Change-Id: I7e5930748795e7ca4a998ab2c608c3c9b6363037
This CL should fix application startup regression for Vulkan
detected by "Hermetic Startup: EmptyActivity" test.
EGL drivers are loaded in a temp thread to leave more time
in RenderThread for other work. Loading EGL drivers
on the RenderThread may cause a perf regression.
Test: Ran cold-dropcache-test test.
Bug: 122659224
Bug: 123361175
Change-Id: I8ca818e98fac196a41d079be15594caca5cb1bab
This CL allows for more than one VulkanManager to exist.
VulkanManager ctor are public allowing for classes other
than RenderThread to instantiate it.
Secondary VulkanManager can be used to render on a thread
other than RT.
Test: Ran HWUI unit tests and several apps
Change-Id: Ibfd76c86ff67e01617a500902bba7431b928f5c0
SkiaVulkanPipeline::mVkSurface can become obsolete if
RenderThread destroys Vulkan context. This CL enables
RenderThread to notify active Vulkan pipelines that their
surface is invalid.
Improve error handling, when trying to draw a frame with null
VulkanSurface.
Bug: 123640274
Bug: 123541940
Test: Ran several apps
Change-Id: If7fba00713d097192c96179df36e90b54f4f8090
This updates to use Skia's new api which takes the vulkan apiVersion
instead of the instance version. This is technically more correct since
the application apiVersion is really the only client modifiable version
value in vulkan.
This change also updates the webview structs to use the apiVersion as
well.
Test: manual build and testing.
Change-Id: I6ce7c20949eb7242f7bbe69955b54c0785696891
Bug: 115613038
Test: Turning on vulkan with appropriate webview apk does not crash and
sort of works.
Change-Id: If1504da7a35e4bd74a994ab2c2a351e6bc415a18
Previously we hardcode wide color gamut in HWUI as scRGB color space with FP16
pixel format. However, the hardware composer doesn't support this combination.
This patch plumbs wide color gamut composition preference from composer API to
HWUI such that HWUI can now pick the combination of color space and pixel
format for the surface.
BUG: 111436479
Test: Build, flash and boot, verify with a demo app.
Change-Id: I7a8b4d8deca72ef40069dba9d23a3f5e90dbfe5a
Set NATIVE_WINDOW_SCALING_MODE_FREEZE scaling mode on the surface
after Vulkan swapchain is created. This is the scaling mode used
by GL as well. Window resize flow depends on preserving this
behavior, because SurfaceComposerClient::Transaction::setSize
calls are not postponed until a new buffer is enqueued.
Test: StatusBar is resized correctly without any glitch.
Bug: 118501234
Change-Id: Icfb6805bc362a742e14b993a0b825e5447d6eeb4
