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

Abstract

We are developing a new code set "Bridge++" for lattice simulations. It is aimed to be an extensible, 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.

Practical Implementation of Lattice QCD Simulation on SIMD Machines with Intel AVX-512

“…Our base code is the Bridge++ code set [12,13] which is described in C++ based on an object oriented design and allows us to replace fermion matrices and solver algorithms independently. In the original Bridge++, hereafter called the Bridge++ core library, the data are in double precision and in a fixed data layout with array of structure (AoS).…”

Lattice QCD code set Bridge++ on arithmetic accelerators

“…In QCD simulations, we often encounter bottlenecks in the steps (4) and (6), because of a limited bandwidth between host and device. This is a severe problem in the fermion matrix inversion.…”

“…At the construction of an AField_dev instance, a memory space on the device is allocated. The linear algebraic operations as well as the fermion operators corresponding to the above step (5) are also prepared with the same interfaces as on the host. The solver algorithms are implemented with the C++ template for both host and device field classes.…”

“…To keep applicability to a wide range of physics problems and numerical algorithms, the components of the code specific to each hardware should be separated and encapsulated into reusable ingredients with the least interference. A guideline to develop such a code set is the object-oriented programming (OOP) [3,4,5]. In 2009, we have started a project to develop a general-purpose code set for lattice QCD simulations that is widely applicable while based on uniform design policy.…”

“…In Ref [5]. we quoted incorrect performance data for Tesla K40, caused by that the position of time counter was changed by compiler optimization.…”