PoS - Proceedings of Science
Volume 414 - 41st International Conference on High Energy physics (ICHEP2022) - Computing and Data Handling
New RooFit Developments to Speed up your Analysis
Z. Wolffs*, P. Bos, C. Burgard, E. Michalainas, L. Moneta, J. Rembser and W. Verkerke
Full text: pdf
Pre-published on: November 28, 2022
Published on:
Abstract
As the field of high energy physics moves to an era of precision measurements its models become ever more complex and so do the challenges for computational frameworks that intend to fit these models to data. This report describes two computational optimizations with which RooFit intends to address this challenge: parallelization and batched computations. For the former, a problem-agnostic parallelization framework was devised with generality in mind such that it could be seamlessly applied at various stages of the existing Minuit2 minimization routine. In the results shown in this report parallelization was applied at the gradient calculation stage. The batched computations approach on the other hand required an overhaul of the current manner in which RooFit prepares its computatational graph for the evaluation of likelihoods. This report includes initial benchmarks of the batched computations strategy run on a CPU with vector instructions. Both strategies show significant performance improvements and the parallelization approach at its current state also proves to be robust enough to consistently fit state of the art physics models to real LHC data. Future developments are targeted towards combining both technologies in RooFit in a production-ready state in a ROOT release in the near future.
DOI: https://doi.org/10.22323/1.414.0249
How to cite

Metadata are provided both in "article" format (very similar to INSPIRE) as this helps creating very compact bibliographies which can be beneficial to authors and readers, and in "proceeding" format which is more detailed and complete.

Open Access
Creative Commons LicenseCopyright owned by the author(s) under the term of the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.