Philippe Canal's research while affiliated with Fermi National Accelerator Laboratory (Fermilab) and other places

... Because detector simulation is naturally an HTC problem, nontrivial effort is required to adapt the necessary computations to use computer accelerators effectively. Two ongoing projects, AdePT [50] and Celeritas [51,52], the latter led by US national laboratory personnel, have produced viable prototypes of high-performing GPU-native simulation engines. However, there is a significant burden in developing and maintaining coprocessor-friendly code, including portability to different GPUs platforms and, potentially, future non-GPU architectures. ...

... A recently concluded R&D project called GeantV introduced sub-event parallelism using vectorized instructions, primarily targeting many-core CPU architectures, in a new prototype transport engine. The primary findings of this project were that the achievable speedup compared to Geant4 was limited to a factor 2 ± 0.5 and that a small percentage of the speedup actually arose from vectorization [33]. Nevertheless, the project produced several useful developments, such as VecGeom, and lessons, such as the importance of instruction cache locality. ...

... We expect the implementation of the support of RNTuple to be easier than TTree that it is expected to replace, thanks to its design. Data are stored in column of fundamental types (float, int,...) [54], similar to Apache Arrow [55], which should ease support from programming languages other than C ++ like Julia. ...

... Refs. [6][7][8]. Once the storage limit is reached, one is forced to discard parts of the dataset, or only save certain features of the data. Generally, this can be done without impacting the overall scientific program of the experiments, for example by using a data selection system called trigger that only stores data satisfying certain pre-determined characteristics that ensure the dataset will be aligned with the experiment's main scientific goals. ...

... In this respect, generators should be somewhat easier to reengineer efficiently for GPUs than detector simulation software (notably Geant4 [156]), where the abundance of conditional branching of a stochastic nature may lead to "thread divergence" and poor software performance (see, for examples, Refs. [157][158][159][160][161][162]). ...

... The resulting technology has the capability to enable substantial economic and operational gains (including speedup) for High Energy and Nuclear Physics data storage/analysis. In our initial studies, a factor of nearly x4 (3.9) compression was achieved with RHIC/STAR data where ROOT compression managed only x1.4 [6]. ...

... In [55], authors present optimization of SHR3 generator to solve differential equations, which outperformed the Curand XORWOW generator by 79% and 38% for uniform and normal distributions, respectively. Reference [56] compared the performance of PRNGs from the VecRNG package (VecRNG is a part of VecMath, a collection of vectorized algorithms for high-energy physics applications based on the VecCore library [56,57].) with the Curand library PRNGs to generate 10 7 double precision PRNs. They find that the Curand library implementation of the PHILOX4_32_10 generator is five times faster than the VecRNG implementation, while the MRG32k3a generators in both libraries show a similar performance. ...

... Awkward RDataFrame [13] uses the C++ header-only libraries to simplify the process of justin-time (JIT) compilation in ROOT [14]. The ak.from_rdataframe [15] function converts the selected ROOT RDataFrame [16] columns as native Awkward Arrays. The templated headeronly implementation constructs the Form from the primitive data types [17]. ...

... When processing large volumes, sometimes the time spent on reading or writing data comes to the fore. New possibilities for parallelizing these processes are actively used in the optimization of HEP software [5][6][7]. ...

... Besides the scheduler efficiency to dispatch baskets, the global performance is highly impacted by the intrinsic efficiency of the basketizing procedure, involving gather and scatter actions as well as concurrent access. As presented in detail in Ref. [44], the main conclusion is that the basketizing dynamics strongly depends on the complexity of the workflow and on state parameters, such as number of tracks in flight, particle production budget, or percent to completion for a given event. ...