33th ROOT Parallelism, Performance and Programming Model

Thursday 1 Mar 2018, 16:00 → 17:45 Europe/Zurich

4/S-020 (CERN)

Danilo Piparo (CERN)

Present: Guilherme, Stefan, Augusto, Enrico, Lorenzo, Gerri, Philippe, Michael Solokoff, Enric, Jim, Axel, Henry Frederic Schreiner, Andrea Contu

Hydra

- Hit performance bottleneck of analysis: Dalitz plots, Likelihood.

- Dispatch code to device: needs compile time decision

- No convolution, no simultaneous fits available yet, these features are in the roadmap

Questions:

- (Lorenzo) Perhaps an interface for RooFit? There is something similar for Tensorflow. (AA) This can be maybe not immediate. RooFit becomes the high level description language.

- (Lorenzo) Is vectorisation supported? (AA) There is a parallel project to vectorise some operations in Hydra, not yet in production.

- (Lorenzo) What is the future of Hydra? (AA) changing institute and maybe move to Spain.

Potential interplay

- Can the concepts behind Hydra be beneficial for ROOT? Yes! TDF is the convenient interface for the user to do complex and heavy tasks. Right now it is not heterogeneous - we may be inspired by the way Hydra does

- Low level possible item: use tdf for filling hydra datasets from a large set of ROOT files

16:00 → 16:30 Hydra, a framework for data analysis in massivelly parallel and heterogeneous platforms

Hydra is a templatized header-only, C++11-compliant framework for data analysis on massively parallel platforms targeting the field High Energy Physics research.
Hydra supports the description of particle decays via the generation of Monte Carlo samples, multidimensional data fitting, multidimensional adaptive numerical integration, histograming of multidimensional data, s-plots calculation and many other common operations. The project is open source and the code is hosted in GitHub. This contribution aims to summarize the main features and the design of Hydra and presents the huge performance boost obtained on numerical integration, Dalitz's plot analyses and Monte Carlo generation of large samples in a variety of GPUs and multi-core CPUs. It also aims to discuss the possible scenarios to promote and facilitate the access to and the integration of the routines implemented in Hydra in ROOT6 and ROOT7.

Speakers: Antonio Augusto Alves Junior (Unknown), Antonio Augusto Alves Junior (University of Cincinnati (US))

hydra_PPP_meeting.pdf