Heterogeneous Computing with GPUs for Trigger Decision in CMS Experiment at the LHC

Using current technology, the Compact Muon Solenoid (CMS) experiment at the Large Hadron Collider (LHC) can record only 10$^3$ events per second out of 10$^9$ events occurring when protons collide at the interaction point. This huge rejection is achieved in real time by diligent selection implemented in a 2-level trigger conditioning, Level 1 (L1) and High-Level Trigger (HLT). For the high luminosity LHC operation, the computing for trigger will be based on heterogeneous resources including substantial usage of GPUs. For the integration of GPUs into the existing computing eco-system, a general framework for heterogeneous computing is being developed. This article describes the scheme of utilising fully reconstructed charged particle tracks for online event reconstruction using the faster and scalable new architecture.