Improvements of the online TPC GPU tracking in the ALICE HLT in the face of continuous readout in LHC run 3

ALICE is one of the four major experiments at the Large Hadron Collider (LHC) at CERN. Its main goal is the study of matter under extreme pressure and temperature as produced in heavy ion collisions at LHC. The ALICE High Level Trigger (HLT) is an online compute farm that performs a real time event reconstruction of the data delivered by the ALICE detectors. The HLT uses GPUs as hardware accelerators to perform online track reconstruction for the TPC, the main tracking detector. During the Long Shutdown 2, there will be a major upgrade of the ALICE TPC as well as the online / offline computing systems. The new GEM TPC for LHC run 3 will feature continuous readout instead of triggered read out. This poses challenges for the online track reconstruction: it is no longer possible to convert the TPC raw data including the time of the TPC hits to spatial coordinates in advance. The tracking algorithm must be adapted accordingly and needs tighter coupling with coordinate transformation and calibration. We present new features currently developed for the online tracking and first tests performed on the current HLT farm.