Acts GPU R&D - Optimization of GPU tracking pipeline

Description

A Common Tracking Software (Acts) is a general track reconstruction software toolkit for High Energy Physics experiments. In the upcoming High luminosity LHC experiments, the large number of particle interactions will significantly increase the track reconstruction time on CPUs. Therefore, the Acts collaboration has launched several GPU R&D lines (vecmem, detray, and traccc) to accelerate the track reconstruction: vecmem is a memory management toolkit which provides users with convenient GPU interface. detray is a geometry builder which translates the CPU geometry into GPU one. traccc incorporates the other R&D lines to demonstrate GPU tracking pipeline which includes hit clusterization, seed finding, and track fitting.

The goal of project will be the optimization of traccc GPU pipeline with CUDA API. The track reconstruction algorithms can be improved by multi-threading with CUDA-MPS and partitioning the GPU device with Multi-Instance GPU (MIG). The caching of event data model (EDM), which contains information on event, is also critical to the throughput. Their memory access or allocation in the GPU device can be optimized by using a proper caching allocator provided by vecmem.

Expected Results

Evaluation Task

Interested students can contact mentor for evaluation task. It is also possible to submit past projects on GPU (e.g. github link)

Requirements

Mentors

Additional Information

Corresponding Project

Participating Organizations