| Age | Commit message (Collapse) | Author |
|---|
|
SF would set the desired active mode on the specific DisplayDevice the
mode was requested and use it to swicth the active display mode.
Change-Id: Ib4f01ec63b2087ab2a81490051d1e029afc71854
Test: SF unit tests
Test: refresh rate switching is working on device with
more than one display
Bug: 187539899
Change-Id: Ie0e5ffe364f81775c2101ffd0cf2c596051947fa
Merged-In: Ie0e5ffe364f81775c2101ffd0cf2c596051947fa
|
|
- Use GPU finish time as well as actual deadline to determine jank
rate.
- Use dynamic interval to adjust for 60/90hz switching
- Move frame metrics reporting into JankTracker to adjust the
deadline communicated to the app when in stuffing scenario.
- Adjust double-stuffing detection to be a bit more readable.
Test: GraphicsStatsValidationTest.java
Test: adb shell dumpsys gfxinfo
Test: FrameMetricsListenerTest
Test: Log output of FrameMetricsObserver
Bug: 169858044
Change-Id: Ia1cae9f0c5358d1cd3bf043289ea8b4d26154737
|
|
This CL continues the raneming of display "configs" to
display "modes". The goal of this is to have the same
names as in the java code and the public display APIs.
Additionally in this CL we move DisplayConfig (from libui)
to the namespace android::ui. This is to prevent conflict
with the SurfaceFlinger's internal android::DisplayMode.
This is consistent with the neighboring classes which
are also in the ui namespace.
Also the type of the parameter defaultMode of
{s,g}etDesiredDisplayModeSpecs is changed to size_t
for consistency with the rest of the code. Appropriate
error handling is added for this.
Bug: 159590486
Bug: 179158858
Test: presubmit
Change-Id: I31e5be1f2223a9ec9340789ce3dc5738eceaf40f
|
|
Bug: 159590486
Bug: 176148651
Test: atest libsurfaceflinger_unittest
Change-Id: Ie177b213bb7a8ab9f67a51d2b5cf27d8fd97b780
|
|
Compile with -Wextra for additional conversion checks. In each file
which produces compile errors -Wextra is ignored.
Bug: 175126758
Test: m surfaceflinger && m libsurfaceflinger_unittest
Change-Id: Iec6bcc699715a99c8c65d891ab3c0481e884c728
|
|
Add a new event to DisplayEventReceiver that conveys a list
of apps whose frame rates have been overriden.
This event is processed by the DisplayManager to advertise the
new frame rate to the overridden app.
Change-Id: I89fce8b5b4d9db65ec5db7dd2393c384c0fcfd82
Bug: 169271059
Bug: 169271062
Bug: 170503758
Test: manual test using SF backdoor
|
|
Add the ability for SurfaceFlinger to be able to throttle down to
a divider of the refresh rate (i.e. for 30/45 for 90Hz)
Change-Id: I6bfd6f43ee1f30e771a136c558d8ae9a6d7fbe0f
Test: Manually via 1039 SF backdoor
Bug: 170502573
Bug: 169270763
Bug: 169271059
|
|
This change adds the necessary plumbing from SurfaceFlinger to use
FrameTimeline for the BufferQueueLayers.
Bug: 162890590
Test: libsurfaceflinger_unittest
Change-Id: Ibac00ccb6584e9eb8d06eb9682747400b3e92845
|
|
This change is using the duration given to sf/app without
converting them back to the legacy offset. This allows us to get
the expected vsync associated with a wakeup event.
This change also required some refactoring:
- Move the periodic callback interface out of DispSync and in
to DispSyncSource
- Translate legacy offsets to duration in case that duration
is not specified in the build files
- Add support to VSD to expose the deadline timestamp of when
a frame needs to be ready by
Bug: 162888874
Test: SF unit tests
Test: examine systraces
Change-Id: I87a53cc7dea931d3c195eab6842e003ca4516885
|
|
Use a type safe wrapper around uint64_t for display ids.
We also modify the format of the EDID generated physical
display IDs by setting bit 62 to 1, which will indicate
that the display id is stable across reboot.
Bug: 160679868
Test: m && flash device
Test: take screnshot on device
Test: atest surfaceflinger_unittest
Change-Id: Ie2c7c2b737e0882fa4208c059caa85b7ded922b2
|
|
When SurfaceFlinger calculates what is the vsync time, it does
that by asking VSyncReactor what is next anticipated vsync from the
wake up time. To handle negative offsets (when SF wakes up more than
a vsync period before a vsync event) SF adds a vsync period
to that anticipated vsync.
This creates a race condition where the offset used at the
time of invalidate may be different than the offset used by
VSyncReactor calculated at the time of the requestNextVsync().
To fix that, we plumb the first calculated vsync time to
SurfaceFlinger.
Bug: 154303002
Test: Run app transitions and collect systrace
Change-Id: I3f2670c7b0ecb52a85fb07df6d360694b51d5d66
|
|
Print debug information about VsyncReactor internal state when
we expect a vsync event but it is not generated.
Bug: 151892277
Test: adb shell dumpsys SurfaceFlinger --dispsync
Test: simulate a missed vsync and observed logcat
Change-Id: Ia98cb884321d0c25610c4d721f422d9cd05c8c5e
|
|
Bug: 135480141
Test: ./statsd_testdrive 10062
Change-Id: I220f0c96935c96afea194ed8ca69922785cf296e
|
|
This will augment the NDK to respond to display events where the display
refresh rate changes. Consumers of this api will include:
* HWUI, for implementing a policy for determining whether to use
render-ahead,
* Swappy, to potentially replace jumping into Java from native code to
respond to display evnets there.
* Any other native app that would rely on the up-to-date display refresh
rate.
Currently however this is not yet exposed to NDK as CTS is not yet
written. Once CTS is written then this will be formally exposed to NDK.
For now we'll leave these as APEX apis to represent incremental
progress.
Bug: 136262896
Test: builds
Change-Id: I66d393f93eb5d681547411e330ef1b8950a35c5d
|
|
Composer 2.4 adds a new attribute for configs groups. This change
groups configs according to their group and store them in
RefreshRateConfigs.
Test: rev up composer to 2.4 and test refresh rate switching
Test: adb shell /data/nativetest64/libsurfaceflinger_unittest/libsurfaceflinger_unittest
Test: adb shell /data/nativetest64/SurfaceFlinger_test/SurfaceFlinger_test
Bug: 141329414
Fixes: 139751853
Change-Id: Ic0bcd3da4bf6b73efa11a60c2594948ce030362f
|
|
This CL refactors VSYNC injection to interface with the Scheduler API,
and removes otherwise unused EventThread members.
Bug: 128863962
Test: sffakehwc_test
Change-Id: I2ba143bb78baf3e66a47b90163eacf0d0e431124
|
|
When display refresh rate changes, SF fires DISPLAY_EVENT_CONFIG_CHANGED
thru DisplayEventReceiver. If the other end of the pipe doesn't consume
the events it may lead to pipe full and dropping of events. Furthermore,
The only clients interested in this event in DisplayManager and hwui.
To avoid spamming all clients with this event, this change is adding an
explicit register for DISPLAY_EVENT_CONFIG_CHANGED events.
Bug: 131688378
Test: adb shell /data/nativetest64/libsurfaceflinger_unittest/libsurfaceflinger_unittest
Test: trigger config change and observe logcat
Change-Id: I5973a1ecc1f3e3ff8d8a0cba19db0e49ef0d5341
|
|
Add the notion of Touch Events to Scheduler to enhance the algorithm
to move to Performance refresh rate.
Scheduler selects Performance when:
- There is a touch event
- There is a buffer to HWC
This change also removes the behavior of Scheduler to move to Performance
on Choreographer callbacks.
Test: Switch between apps using gesture navigation
Bug: 131906818
Change-Id: I588cfc32449e87744e829dc7c5261a2e4151a8f8
|
|
Bug: 128849305
Change-Id: I97c7ab9a2555f229783cd1c5352d883a627ab193
Test: jank tests
Test: systrace
|
|
A conversion error (uint64_t -> uint32_t) was causing a trunction
of displayId value which lead to invalid displayId in onConfigChanged.
Test: adb shell /data/nativetest64/libsurfaceflinger_unittest/libsurfaceflinger_unittest
Bug: 129159940
Change-Id: I9d24cac0b9e74c7b20421d2b7d9fd65c16cca9e1
|
|
Add a new event to DisplayEventReceiver for display
configuration change. This event is sent by SF when display config
is changed.
Test: adb shell /data/nativetest64/libsurfaceflinger_unittest/libsurfaceflinger_unittest
Bug: 122905403
Change-Id: Ibb7e0ce7b83b91259ccb0e9c982f5e378b0ebfaf
|
|
Do not dipatch vsync callbacks to applications during a config
switch as applications will get choreographer callbacks at uneven
times and will stuff the queue to surface flinger.
Test: test google/perf/jank/UIBench/UIBench-Trace:com.android.uibench.janktests.UiBenchJankTests#testClippedListView -v
Change-Id: I0537933c27af83a9186bcff54c3596d612c748ea
|
|
|
|
|
|
SurfaceFlinger::setRefreshRateTo, called from the Scheduler callbacks,
reads HWC and SF state without locking mStateLock, concurrently with
writes from the main thread. This CL registers ResetIdleTimerCallback
per EventThreadConnection, and locks mStateLock for connections used
off the main thread. Note that ExpiredTimerCallback locks mStateLock
unconditionally, since it is always called from the IdleTimer thread.
This CL also adds a thread annotation to setRefreshRateTo, and refactors
it accordingly.
Bug: 123715322
Test: libsurfaceflinger_unittest
Test: Boot with scheduler enabled
Change-Id: Id62c48ae22da38f292ffc18e8731a1c49a0a083c
|
|
This CL replaces ISurfaceComposer::{eDisplayIdMain,eDisplayIdHdmi} with
the stable 64-bit display IDs generated by SF. Note that the 64-bit IDs
fall back to the old values if the HWC API for display identification is
not supported.
Bug: 74619554
Test: LocalDisplayAdapter and Choreographer receive 64-bit IDs
Test: 64-bit IDs fall back to 0 and 1 on HWC 2.2 and below
Change-Id: I3c08eff6eb8bb179ecce596ab2820a2aa44c8649
|
|
This has been a minor issue for a while, and was causing a few failures
a week in the continuous testing infrastructure.
I managed to reproduce it locally with 10k iterations of one test (to
even have a good chance of catching one failure!) -- on a phone that was
stuck in a startup boot crash loop! (Other significant activity), and
from there diagnose it to an extremely rare chance that threads would
take 20-30ms to resume execution after being unblocked, significantly
more than the 10ms maximum time allowed.
This patch increases the maximum wait time to 100ms which should
eliminate the flakes (or at least reduce them to an substantially
smaller percentage of the time).
This led to the discovery that the EventThreadTest destructor was using
the call for expecting an event when actually it should have been making
a call using a lower default threshhold to detect unexpected calls,
which I fixed.
With the tests otherwise using the calls as intended, I verified that
the typical test time was not substantially slower than it was before (in
fact it was faster after the EventThreadTest fix!)
Bug: 112104412
Test: atest libsurfaceflinger_unittest # Not slower for a typical run
Test: EventControlThreadTest.signalsVSyncEnabledThenDisabled * 100k
Change-Id: Ie778567955d746922ef514baff1a552c34794b8d
|
|
The test was invoking a VSYNC callback right after checking that VSYNC
callbacks are not enabled.
Bug: 123530318
Test: vsyncRequestIsIgnoredIfDisplayIsDisconnected
Change-Id: I5cc997266450e1ecb6ed2cd3901472984053477a
|
|
This CL creates/destroys EventThread::VSyncState when the display is
connected/disconnected.
Bug: 74619554
Test: libsurfaceflinger_unittest
Test: dumpsys SurfaceFlinger --vsync
Change-Id: I46dad4857222b6d2932ab568c4d66325edc3371b
|
|
This CL simplifies the logic to enable and disable VSYNC in the
EventThread loop.
Bug: 74619554
Test: libsurfaceflinger_unittest
Test: dumpsys SurfaceFlinger --vsync
Change-Id: I7637fcf1982d60cfced5dae68c74556f0ee67a0f
|
|
This CL ties the resync callback to an EventThreadConnection instead
of an EventThread. This is a step towards having IDisplayEventConnection
subscribe to a given display rather than the primary display implicitly.
Each display will then have SurfaceFlinger::VsyncState that resyncs
independently at potentially different rates. Callbacks have weak
references to the per-display VsyncState owned by SurfaceFlinger.
Bug: 74619554
Test: Boot and turn display on/off repeatedly
Change-Id: Ic7cc64e2004fa07a5d54431fc330995048a4ed20
|
|
When device is idle, refresh rate is set to 60. When not it's set to 90.
See go/surface-flinger-scheduler for more info.
Test: All SF tests pass.
Bug: 113612090
Bug: 122347908
Change-Id: Ica6e483118db276f72d3cb4e79535303c76f99d7
|
|
This is part of Scheduler refactoring. EventThread::Connection should be
a wrapper that translates IDisplayEventConnection into EventThread calls.
Test: SF tests pass.
Bug: 113612090
Change-Id: I2bcf0c45a33638c59f7828085c563dbc52b2d66e
|
|
Scheduler (see go/surface-flinger-scheduler) is going to live in its own
directory. This CL just moves the relevant files into that directory. No
changes to business logic.
Test: all SF tests pass.
Change-Id: Iff0717f9867316b28e68fd8311bd9fdc4e029951
|
|
EventThread uses DisplayDevice::DisplayType constants, which will be
removed in a follow-up CL. This CL replaces them with local constants as
a stopgap until stable display IDs are propagated through the SF/WM
interface.
Bug: 74619554
Test: libsurfaceflinger_unittest
Change-Id: I68be363dc58c5e3aa17d05fba156d520a01fa775
|
|
Add a unit test to cover EventThread.cpp
Test: atest libsurfaceflinger_unittest
Bug: 74827900
Change-Id: If9479cd9deedff836068cb53e7da2cb64041aea1
|
