Multigrid preconditioning has proved to deal efficiently with the critical slowing down of standard Krylov solvers. In the literature two slightly different approaches have been developed, referred to as MG-CGR and DD-αAMG. Several libraries implementing multigrid solvers are publicly available for both CPU and GPU codes. In this talk we will consider the QUDA and DDalphaAMG library which implem ... More

A code for the simulation of QCD+QED with C* boundary conditions is presented. This code is based on openQCD-1.6, from which it inherits the core features that ensure its efficiency: the locally-deflated SAP-preconditioned GCR solver, the twisted-mass frequency splitting of the fermion action, the multilevel integrator, the 4th order OMF integrator, the SSE/AVX intrinsics, etc. The photon field is ... More

A code for the simulation of QCD+QED with C* boundary conditions is presented. This code is based on openQCD-1.6, from which it inherits the core features that ensure its efficiency: the locally-deflated SAP-preconditioned GCR solver, the twisted-mass frequency splitting of the fermion action, the multilevel integrator, the 4th order OMF integrator, the SSE/AVX intrinsics, etc. The photon field is ... More

The process of generating ensembles of gauge configurations (and measuring various observables over them) can be tedious and error-prone when done "by hand". In practice, most of this procedure can be automated with the use of a workflow manager. We discuss how this automation can be accomplished using Taxi, a minimal workflow manager built for generating lattice data. We present a case study d ... More

Numerical QCD is often extremely resource demanding and it is not rare to run hundreds of simulations at the same time. Each of these typically requires a job-script file as well as an input file with the physical parameters for the application to be run. Moreover, it is often necessary to resume a simulation either because it crashed or simply because the accumulated statistics is not sufficien ... More

QUDA is an increasingly popular GPU library for both accelerating legacy lattice QCD applications and, having also evolved into a framework in its own right, for directly deploying QCD simulations. Recent work has focused on simplifying the framework to lower the barrier of entry to extending QUDA for new algorithms and methods. In this talk we describe the QUDA framework, with worked examples o ... More

This presentation will provide overviews into different technologies. We will review selected technologies that are currently at the disposal of researchers to handle their data. This will include - Filesystems (such as LUSTRE) and strategies to overcome their bottlenecks - Object storages (such as CEPH) and their differneces to filesystems - Management of data on the larger scale, i.e. bet ... More

I will present the strategies we developed in Grid for the HMC sector, in order to support a variety of behaviours without code replication. I will also discuss the current status of the architecture support in view of the upcoming machines.

Quantities in lattice QCD are estimated from statistical fits to lattice correlation functions. For example, a meson mass may be estimated from a fit to a two-point function, $C(t)$, computed at $p$ timeslices, where each measurement is averaged over a sample of $n$, statistically independent gauge configurations. Correlations among $C(t)$ at nearby time slices means that the statistical pr ... More

High Performance Computing is often performed on scarce, shared computing resources. To ensure computers are used to their full capacity, administrators often incentivize large workloads that are not possible on smaller systems. Measurements in Lattice QCD frequently do not scale to machine-size workloads. By bundling tasks together we can create large jobs suitable for gigantic partitions. ... More

In October, 2016, the US Department of Energy launched an Exascale Computation Project that aims to deploy exascale computing resources for science and engineering in the early 2020's. The project brings together application teams, software developers, and hardware vendors in realizing this goal. Lattice QCD is one of the applications. Members of the US lattice gauge theory community with signi ... More

We present a case study for data analysis in lattice gauge theory. Three features are key to our framework: relational databases for storage of high-level lattice results, fast open-source software for analysis, and a dashboard environment based on the Jupyter notebook. We talk about our recent experiences learning, using, and combining these tools. Finally, we advocate their wider usage in the la ... More

I will discuss the features of the programming language Nim (https://nim-lang.org) that make it especially suitable for HPC applications. In particular I will discuss installation, metaprogramming, optimization, scripting and C/C++ interoperability. I will also give an update on the status of the lattice field theory framework QEX (https://github.com/jcosborn/qex) being developed in Nim.

One of the key requirements for the Lattice QCD Application Development as part of the US Exascale Computing Project is performance portability across multiple architectures. Using the Grid C++ expression template as a starting point, we report on the progress made with regards to the Grid GPU offloading strategies. We present both the successes and issues encountered in using CUDA, OpenACC and Ju ... More

Varying from multi-core CPU processors to many-core GPUs, the present scenario of HPC architectures is extremely heterogeneous. In this context, code portability is increasingly important for easy maintainability of applications; this is relevant in scientific computing where code changes are numerous and frequent. In this talk we present the design and optimization of a state-of-the-art producti ... More

We will present the QCD component of the Unified European Application Benchmark Suite (UEABS). The UEABS is an initiative of PRACE to create and maintain a benchmark suite of application kernels representative of the European scientific computing landscape. Several kernels have been included based on publicly available software packages from various lattice QCD collaborations, including BQCD, open ... More

We present preliminary results from dynamical simulations of QCD in isolation, as well as QCD coupled to QED, with C* boundary conditions. In finite volume, the use of C* boundary conditions allows for a gauge invariant and local formulation of QED without zero modes We start with an overview of the simulation code and its features. In particular we discuss the implementation of the C* boundary ... More

Filtering algorithms for two degenerate quark flavours have advanced to the point that, in 2+1 flavour simulations, the cost of the strange quark is significant compared with the light quarks. This makes efficient filtering algorithms for single flavour actions highly desirable, in particular when considering 1+1+1 flavour simulations for QED+QCD. Here we discuss two methods for filtering the RHMC ... More

As we march towards the exa-scale, the surge of hybrid architectures, long SIMD vector lengths, low bandwidth memory subsystems, caches and networks has made it extremely challenging for Lattice QCD codes to retain both high single node performance and good scaling properties in this heterogeneous environment. In order to have access to efficient kernel implementations and the latest solvers, th ... More

We report the performance of Wilson and Domainwall Kernels on a new Intel Xeon Phi Knights Landing based machine named Oakforest-PACS, which is co-hosted by University of Tokyo and Tsukuba University and is currently fastest in Japan. This machine uses Intel Omni-Path for the internode network. We compare performance with several types of implementation including that makes use of the Grid l ... More