This change introduces "is APF enabled" system properties, and
uses them as trigger conditions for APF nodes / pre-APF nodes.
By default APF nodes are active, and pre-APF nodes are inactive,
which maintains the situation before this change.
A future experiment will flip "is APF enabled" system properties,
deactivating APF nodes and activating pre-APF nodes. For more
info see go/pixel-perf-apf-experiment.
Bug: 365855872
Test: set vendor.powerhal.apf_enabled=false, vendor.powerhal.apf_disabled=true, observe auto_prefer_fit
procfs value staying "false" and instead preferred_idle_mask_mid
changing to 0x1ff during app launches
Flag: NONE powerhint change does not support flags
Change-Id: I85b5dc812132f229c2cc8583aaadd31d7eedc178
Signed-off-by: Dmitry Skiba <dskiba@google.com>
Set the CPU frequencies of FIXED_PERFORMANCE to align with Thermal CPU Light level.
Bug: 327070837
Test: Boot to home
Flag: NONE powerhal doesn't have a flag yet
Change-Id: Ibb698b608c21c2a4f81354f852cb077345907947
These streams do not disable TA or FG prefer idle, so restore it.
Bug: 236868627
Flag: NONE powerhint change does not support flags
Test: build pass
Change-Id: Ia0f12911110856b82b858bed11fb5d120388285a
Enable auto prefer fit for high prio tasks when app launch, and tune
uclmap min settings for different task prio.
Bug: 353885521
Flag: NONE powerhint change does not support flags
Test: build pass
Change-Id: I1b65a1543085dc3244a36d90c01ccb3ad9211604
The big core mask on zuma should be 0x100.
Bug: 248621289
Test: build pass
Flag: NONE powerhint change does not support flags
Change-Id: Id9dbd933505d3eaa2553ed1851883a3042028f16
Some frequencies uses unavailable frequency other soc ex: ag/22293847, the patch update them to explicitly show the actual actions of driver.
Bug: 361294204
Change-Id: I1adc41aca636650b4eaad179d73bde030146347c
FLAG: NONE powerhint do not need flag
Also disable prefer idle for TA and FG by default.
Bug: 236868627
Flag: NONE powerhint change does not support flags
Test: build pass
Change-Id: I0d06e707900ec9a85decbdecbd184532743c7de0
- Update freq for legacy mismatched frequency with the rule of scaling_max_freq will automatically round down, scaling_min_freq will round up. All powerhint behavior should not change.
- Remove some min/max frequency not used in any nodes.
Bug: 361294204
FLAG: NONE powerhint do not need flag
Change-Id: I0242977f05f859ad5c3c38673579e296e74c78ca
Current CAM uclamp min is beyond MID clusters capacity when
CAM2 energy model is used. This causes most of camera HAL threads
scheduled on big cluster only with CAM2 EM profile.
Lowering CAM uclamp min so it is within MID clusters capacity for
both energe model
Flag: EXEMPT tuning changes
Test: check camera trace with proper uclamp value
Bug: 350629738
Change-Id: I40ce59d0a0d99ea6ba016a24245f44713485bc52
Port from ag/28607900
This CL defines CAMERA_MULTICAM_BOOST (with the same settings as the
existing CAMERA_ZOOMING_BOOST). Having both definitions is necessary to
be able to transition from zooming boost to multicam boost. Once the
multicam boost flag is fully rolled out, we can remove the
CAMERA_ZOOMING_BOOST.
Bug: 347829135
Test: adb logcat
Flag: EXEMPT resource file only change
Change-Id: I66f82dec5733a147ecbf073beb5b6c36f4860bb4
Bug: 344685639
Test: Checked the trace file running on the device.
Flag: NONE powerhal doesn't have a flag yet
Change-Id: I8b3918ee11a34315155561bfd0b0a7e9a8f9f32d
Also reset the mask for launch and other camera streaming modes that
do not use it.
Bug: 248621289
Test: build pass
Change-Id: Id2f8f203359d4551d17f9b6ba02ee1c8e805fa3c
Set big core to group_cfs_skip_mask for sys group when screen is off,
so that sys group tasks will run less in big core.
Bug: 328210236
Test: build pass
Change-Id: I9ccb3035a480e88960951222c45d0ca8352082d9
This reverts previous JankCheckTimeFactor from 1.43 to 1.2.
Since the current reported duration doesn't include the GPU
part, the JankCheckTimeFactor needs to be smaller for matching
jank frame detection.
Bug: 204204746
Test: adb shell dumpsys android.hardware.power.Ipower/default;
check adpf sessions' traces.
Change-Id: I9c7587b6b1507fae685561fcf008c5ab9ae73532
Update it to align with the HWUI target duration scale factor,
so the adjusted jank check time duration is same as expected frame
timeline's duration.
HWUI target duration scale factor is 70%, and its inverse value is
approximately 1.43.
Bug: 204204746
Test: adb shell dumpsys android.hardware.power.IPower/default;
Check the device adpf sessions' traces.
Change-Id: I8565fa1352769e6f879ea5fc12412d00f254da3b
In auto prefer idle mode, a task in top-app will be granted prefer
idle if it satisfied the conditions. So turn auto prefer idle off
if top-app prefer_idle is explictly disabled. Also turn auto prefer
off when other group enables prefer_idle explictly.
Bug: 236868627
Test: build pass
Change-Id: Id0dc5bda64686826f72aad4549b8edcb1ceb4502
Enables a signal indicating that the app has janked as
detected by the ADPF API.
Bug: 290625326
Test: perf / power tests
Change-Id: Iaf0f131cab90e4225642af243303473f56a34320
Set default value to 0xff to exclude big core for all priorities, but
restore high and mid to 0x1ff when app launch.
Bug: 248621289
Test: value set correctly
Change-Id: If9696e02f2a54f07d5acb974ff0301ea936c53e0
This is part of an optimization that limits the length and prevelance of
this power hint. Strict, ultra-agression minimizes power and latency
versus always-on mid aggression.
Bug: 297964952
Change-Id: I384f2e04f69d570e10c8e52f2818c1ecde6be1a0
This is part of an optimization that limits the length and prevelance of
this power hint. Strict, ultra-agression minimizes power and latency
versus always-on mid aggression.
Bug: 297964952
Change-Id: I384f2e04f69d570e10c8e52f2818c1ecde6be1a0
