Hideo Matsufuru

OpenCL vs OpenACC: Lessons from Development of Lattice QCD Simulation Code☆

Abstract OpenCL and OpenACC are generic frameworks for heterogeneous programming using CPU and accelerator devices such as GPUs. They have contrasting features: the former explicitly controls devices through API functions, while the latter generates such procedures along a guide of the directives inserted by a programmer. In this paper, we apply these two frameworks to a general-purpose code set for numerical simulations of lattice QCD, which is a computational physics of elementary particles based on the Monte Carlo method. The fermion matrix inversion, which is usually the most time-consuming part of the lattice QCD simulations, is offloaded to the accelerator devices. From a viewpoint of constructing reusable components based on the object-oriented programming and also tuning the code to achieve high performance, we discuss feasibility of these frameworks through the practical implementations.

First Application of Lattice QCD to Pezy-SC Processor

Pezy-SC processor is a novel new architecture developed by Pezy Computing K. K. that has achieved large computational power with low electric power consumption. It works as an accelerator device similarly to GPGPUs. A programming environment that resembles OpenCL is provided. Using a hybrid parallel system Suiren installed at KEK, we port and tune a simulation code of lattice QCD, which is computational elementary particle physics based on Monte Carlo method. We offload an iterative solver of a linear equation for a fermion matrix, which is in general the most time consuming part of the lattice QCD simulations. On single and multiple Pezy-SC devices, the sustained performance is measured for the matrix multiplications and a BiCGStab solver. We examine how the data layout affects the performance. The results demonstrate that the Pezy-SC processors provide a feasible environment to perform numerical lattice QCD simulations.

Development of Lattice QCD Simulation Code Set “Bridge++” on Accelerators

Abstract We are developing a new code set “Bridge++” for lattice QCD (Quantum Chromodynamics) simulations. It aims at an extensible, readable, and portable workbench, while achieving high performance. Bridge++ covers popular lattice actions and numerical algorithms. The code set is constructed in C++ with an object oriented programming. In this paper, we describe our code design focusing on the use of accelerators such as GPGPUs. For portability our implemen- tation employs OpenCL to control the devices while encapsulates the details of manipulation by providing generalized interfaces. The code is successfully applied to several recent accelerators.

Bridge++: an object-oriented C++ code for lattice simulations

We are developing a new code set Bridge++ for lattice simulations. It is aimed to be an ex-tensible, readable, and portable workbench, while it keeps a high performance. Bridge++ covers conventional lattice actions and numerical algorithms. The code set is constructed in C++ with an object oriented programming. We explain our strategy and the basic design of Bridge++. We also present our current status of this project, including the sustained performance on several systems.

Lattice simulation of SU(2) gauge theory with chirally symmetric fermions

We numerically study the SU(2) gauge theory with two dynamical flavors of the domain-wall fermions in fundamental representation. The meson spectra and the residual mass are measured on three lattice volumes and at two values of gauge coupling so as to investigate the finite volume effect. On generated configurations, eigenvalues of the overlap fermion operator are determined and compared to the random matrix theory. To quantify the effect of violation of the exact chiral symmetry, we measure the correlation between the eigenvectors of the domain-wall and the overlap operators.

Development of an object oriented lattice QCD code 'Bridge++'

We are developing a new lattice QCD code set Bridge++ aiming at extensible, readable, and portable workbench for QCD simulations, while keeping a high performance at the same time. Bridge++ covers conventional lattice actions and numerical algorithms. The code set is constructed in C++ with an object oriented programming. In this paper we describe fundamental ingredients of the code and the current status of development.

Simulation of SU(2) gauge theory with improved domain-wall fermions

We investigate the SU(2) gauge theory with the domain-wall fermions. Our previous simulations with many flavors of the standard domain-wall fermions suffered from rather large residual mass that obstruct the study in the small mass region. To improve the domain-wall fermion action, we employ the link smearing technique that consists of the HYP smearing and the stout projection. On the previously generated configurations with the two flavors of the standard domain-wall fermions, we confirm that the improved domain-wall fermion operator indeed reduce the residual mass to a factor of five. Dynamical simulations with two flavors of improved domain-wall fermions are in progress.

SU(2) gauge theory with domain-wall fermions in fundamental and adjoint representations

We numerically study the SU(2) gauge theories with fermions in fundamental and adjoint representations. For the fundamental fermions, we perform dynamical simulations with N f = 2, 4, 6, and 8 using the domain-wall fermions to examine the N f dependence of the meson masses and decay constants. For the adjoint fermions, we investigate, for advance preparation for simulations at a fixed topological charge, the properties of the vacuum in the presence of two-flavors of Wilson fermions with and without the twisted mass ghosts. The meson spectrum is calculated to clarify the Aoki phase structure.

Lattice QCD code Bridge++ on multi-thread and many core accelerators

We are developing an object oriented code set Bridge++ for simulation of lattice gauge theories. It aims to be an extensible, readable, and portable workbench while keeping sufficiently high performance in actual productive runs. This paper describes the status of two extensions. One is multi-threading with OpenMP in the recently released version 1.2. The other is a design to use arithmetic accelerator devices such as GPGPUs, which is still under developing version. Feasibility test is performed with OpenCL.