Update motion prediction model.
Input events with no movement (r = 0) are now included in the buffer
so that the model can accurately determine when the input device has
become stationary, and a noise floor is added to prevent spurious
predictions when this happens.
Benchmark results:
Old:
timeRecordAndPredict_mean (ns): 17990
timeRecordAndPredict_median (ns): 18024
timeRecordAndPredict_min (ns): 17606
timeRecordAndPredict_standardDeviation: 345
New:
timeRecordAndPredict_mean (ns): 38394
timeRecordAndPredict_median (ns): 38476
timeRecordAndPredict_min (ns): 38083
timeRecordAndPredict_standardDeviation: 187
Bug: 288354672
PiperOrigin-RevId: 549064247
Test: predictions are visible in the motionprediction test app
Test: atest CtsInputTestCases
Test: atest MotionPredictorBenchmark MotionPredictorTest
Test: atest --host libinput_tests
Change-Id: I6c3917591323d7117c4ee2e91abf6c6004178f19
6 files changed