Object-oriented implementation of algebraic multi-grid solver for lattice QCD on SIMD architectures and GPU clusters

Abstract: A portable implementation of elaborated algorithm is important to use variety of architectures in HPC applications. In this work we implement and benchmark an algebraic multi-grid solver for Lattice QCD on three different architectures, Intel Xeon Phi, Fujitsu A64FX, and NVIDIA Tesla V100, in keeping high performance and portability of the code based on the object-oriented paradigm. Some parts of code are specific to an architecture employing appropriate data layout and tuned matrix-vector multiplication kernels, while the implementation of abstract solver algorithm is common to all architectures. Although the performance of the solver depends on tuning of the architecture-dependent part, we observe reasonable scaling behavior and better performance than the mixed precision BiCGSstab solvers.