Abstract
Physics event generators are essential components of the data analysis software chain of high energy physics experiments, and important consumers of their CPU resources. Improving the software performance of these packages on modern hardware architectures, such as those deployed at HPC centers, is essential in view of the upcoming HL-LHC physics programme. In this paper, we describe an ongoing activity to reengineer the Madgraph5_aMC@NLO physics event generator, primarily to port it and allow its efficient execution on GPUs, but also to modernize it and optimize its performance on vector CPUs. We describe the motivation, engineering process and software architecture design of our developments, as well as the current challenges and future directions for this project. This paper is based on our submission to vCHEP2021 in March 2021, complemented with a few preliminary results that we presented during the conference. Further details and updated results will be given in later publications.
Highlights
MadGraph5_aMC@next-to-leading order (NLO) [1] is a physics event generator software used in the data processing workflows of High Energy Physics (HEP) experiments, such as ATLAS and CMS at CERN’s Large Hadron Collider (LHC)
We have presented an ongoing project aiming at reengineering the MG5aMC physics event generator, primarily to develop a new CUDA back-end for graphics processing units (GPUs), and to optimize the performance of its C++ back-end on CPUs
We have described its motivation, its iterative engineering process involving code-generating metacode, its software architecture based on event-level data parallelism to efficiently exploit SIMT on GPUs and SIMD on CPUs, as well as some current challenges and future directions
Summary
MadGraph5_aMC@NLO [1] (in the following, MG5aMC) is a physics event generator software used in the data processing workflows of High Energy Physics (HEP) experiments, such as ATLAS and CMS at CERN’s Large Hadron Collider (LHC). While the majority of CPU resources in the distributed computing environments of the LHC experiments is spent on detector simulation and event reconstruction workflows, physics event generators are large consumers of CPU time, accounting for an estimated 12% of overall CPU budgets for ATLAS and 5% for CMS [2]. Previous work for porting some of its components to CUDA [4] on Nvidia GPUs was done around 10 years ago [5,6,7,8,9], but never reached production quality This effort has been restarted in early 2020 by the authors of this paper (including one of main authors of the MG5aMC software), eventually as part of the activities of a larger team. Further details and updated results for this ongoing activity will be presented in later publications
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
