- update model metadata with correct 'prev_samples_order' value
Bug: 385392415
Test: confirm ThermalHAL loads the model without error
Flag: EXEMPT config update
Change-Id: I02c496af39f0bdef6a005f2622ddfd99c72f67b0
LSTM-VT-PREDICTION-7SEC-BACK:
LSTM predicted current VT 7 seconds back. This is the 2nd sample
of LSTM model output. First sample is for current VT. This can be
compared with current VT to measure LSTM model accuracy for 7 seconds
interval.
LSTM-VT-PREDICTION-NOW:
First sample of LSTM model output
LSTM-VT-PREDICTION-ERROR-7SEC:
Compares LSTM prediction 7 seconds back with current LSTM predicted VT.
Bug: 353345569
Test: Build
Flag: EXEMPT config update
Change-Id: I79abe378fee7be0e6c9311d9e04e61a535e50a09
Signed-off-by: Sayanna Chandula <sayanna@google.com>
Bug: 372110489
Test: loaded to device and boot to home
Flag: EXEMPT config update
Change-Id: Ie05768cf2e8da0180fb22c06cc6322a97544669d
Signed-off-by: Jikai Ma <jikai@google.com>
Increase K_I and PowerWeight to reduce temperature
and charging throttling overshoots.
Bug: 338150302
Test: boot and charging thermal throtlting worked properly.
Change-Id: Ia7b2b5f7b2f8d62d1cde800be149993b284f72d4
Use thb_hda value to find charger type and split vskin-charge into
different categories on basis of charger type.
For each category if charger not active, store time in bucket 0 and if
active store residency in higher buckets.
Thresholds: [-200, 31, 33, 35, 37, 39, 41, 43, 45, 47, 49, 51, 53, 55]
Bug: 341872098
Test: Push locally and verify stats by connecting to different chargers
Change-Id: I15738686a0a6b263ab4b3f2caf1a22d7c31c4518
- Remove stats monitoring for sub-sensors since they don't denote a
particular hotspot.
- Remove stats monitoring for wingboard configs.
- For vskin-legacy, vskin-model, vskin-speaker improve the resolution of
buckets by splitting range [49, 55] into [49, 51, 53, 55]
Bug: 341872098
Test: Push config and verify lshal
Change-Id: Iabff63922c283a595fa7ec2dfa8af539629d4064
- generated legacy formulas with DVT and CUJ data
- removed 'battery' thermistor from sensor combination list
- removed 'LINEAR' formulas
Bug: 342227291
Test: verified thermal hal service starts properly with it
Change-Id: Iff59072d432ff98afe852295745cebd81ec73d15
Signed-off-by: Jikai Ma <Jikai@google.com>
With previous samples being used in vt estimation, the sampling period
between vt estimation calls impact vt accuracy. Currently vt model is
using default cache interval of 2 seconds. This change sets the cache
interval to 7 seconds explicitly to match with passive delay
Bug: 341967304
Test: Check vt estimation invoke calls with frequency thermalservice dump
Change-Id: Ic6a38939b2bec88a66c87517e6efa548566ea877
Signed-off-by: Sayanna Chandula <sayanna@google.com>
Due to defalut passive delay of 2 seconds being used for the VT sensor,
previous samples are ignored at 7 seconds intervals. This change sets
explicit passive_delay for VT sensor to make sure we are not ignoring
previous samples and apply step_ratio
Bug: 342257359
Test: Check VIRTUAL-SKIN-MODEL-UPDATED values at run time
Change-Id: I18a9e4913b8153b8d2c3a313451338a9f1b6ca0c
Signed-off-by: Sayanna Chandula <sayanna@google.com>
Bug: 297083757
Test: Tested on CM4 boot to home and with thermal recorder
Change-Id: Ia6071a8c70e19cebeab909ca6c3e36138c93badc
Signed-off-by: Ramya Subramanian <rsubr@google.com>
Enabling step ration for vt model would dampen the vt which seems
to be very sensitive to some of the thermistor changes
Bug: 338284524
Test: Build and boot on device. Monitor Virtual Temperature
Change-Id: Ia45f683e3bef9dc1becf74c683c408ffdecf6e48
Signed-off-by: Sayanna Chandula <sayanna@google.com>
Reduce temperature and charging throttling overshoots by fine-tuning PID
parameters.
Bug: 303547723
Test: boot and charging thermal throtlting worked properly.
Change-Id: Icaeef76998347167976a1ef3606f417db3c5a380
For mmWAVE test cases, legacy equation is under predicting heavily
resulting in virtua-skin switching to vt legacy even though the
model output was more accurate. This change increases the threshold
to address the same
Bug: 338284524
Test: Build and boot on device
Change-Id: I1c7a14a1d9042a2370c30490969d29294fc95f79
Signed-off-by: Sayanna Chandula <sayanna@google.com>
- Update VT models trained with dvt and cuj data
- Use backup sensor during undersampling
- Remove battery thermistor from VT model prediction
Bug: 337512118
Bug: 338596757
Test: Build and boot on device. Check VT through dumpsys
adb shell dumpsys thermalservice
Change-Id: Id26238f7baf58e23ef5e1c771d2ee6cf8968ec8e
Signed-off-by: Sayanna Chandula <sayanna@google.com>
Replacec cdevceiling and hardlimit information with freuqencies in
thermal config
Bug: 336698561
Test: Flash Build and boot to home, run adb dumpsys android.hardware.thermal.IThermal/default, check the cdevceiling and hardlimit number for gpu
Change-Id: If5d105208ea055230dbbf4b6ecbac799b04a18db
Signed-off-by: Peter (YM) <peterym@google.com>
Expose virtual charge sensors to which throttling is binded to and
hide those which don't have.
Unhide:
- VIRTUAL-SKIN-CHARGE-WIRED
- VIRTUAL-SKIN-CHARGE-PERSIST
Hide:
- VIRTUAL-SKIN-CHARGE
Bug: 336193005
Test: Verify dumpsys
Change-Id: I4d83f6af17fccb7de8d053c9d981163c9dde0ea3
Bug: 320714468
Test: side load to device and successfully start thermal hal service.
Change-Id: I8189d7d7c94d590cc06ac82893d6b3dfd730585a
Signed-off-by: Jikai Ma <jikai@google.com>
- Remove backup sensor and enable under sampling
- Clamp VT_MODEL prediction based on training set max_surface
- Use VT_MODEL_LEGACY_DIFF to decide on switch
- Enable smooth transition with weights
Bug: 336265406
Test: Build and boot on device. Test smooth transition
Change-Id: If30a87a24b12b0af5527d9c05be7a14bb06dcfac
Signed-off-by: Sayanna Chandula <sayanna@google.com>
Test: Boot to home and flash on device
Bug: 330780045
Change-Id: I428e14f354a5897601cbcaf8f7bbbeb6a0cc5d8f
Signed-off-by: Peter (YM) <peterym@google.com>
- add tflite model files
- add lstm model sensors to thermal config
- add lstm model sensor to virtual skin prediction
Bug: 308179047
Test: build and validate thermal service
Change-Id: Iaff613d75b3766bfbdf40b391e453a94172872a3
Signed-off-by: Jikai Ma <jikai@google.com>
Reduce threshold to 7C to gather more traces to analyze
Bug: 333110670
Test: Build and boot on device
Change-Id: I2a9ff381055fedaea18791fc030832a8b0877105
Signed-off-by: Sayanna Chandula <sayanna@google.com>
Given that we have very few traces/events uploaded for 12C threshold,
this change reduces the threshold to 10C to help collect more data
to analyze infield abnormalities.
Bug: 333110670
Test: Build and boot on device
Change-Id: Icd3357e543bfd612b856d843a7cf56486361ed81
Signed-off-by: Sayanna Chandula <sayanna@google.com>
- Models prepared with both PD and Cert data from EVT1.1 tests
- Used leaky_relu activation
Bug: 332971287
Test: Build and boot on device
Change-Id: I07fea17ab21d6efd66d1d74629886051ea92aa94
Signed-off-by: Sayanna Chandula <sayanna@google.com>
Remove the below power rails as they're going to be replaced with
different ones for better ODPM coverage.
- S8S_VDD_G3D_L2
- L9S_GNSS_CORE
Bug: 331556280
Test: Push config and verify thermal logs
Change-Id: Ic5c6cb7c4ddae258c89cbe3cb50a104f7044e016
