Living Benchmark for Quantum Espresso

Introduction

This set of benchmarks tests the performance of the core program "PWSCF" of the materials modelling suite Quantum Espresso. Quantum Espresso uses density functional theory with a plane-wave (Fourier) basis, and iteratively solves the Kohn-Sham equations with a matrix-free method.

Key computational kernels are 3D FFTs and dense linear algebra (matrix-matrix multiplication, inversion and diagonalisation). When run in parallel, the 3D FFTs may be implemented as 2D and/or 1D FFTs, depending on the decomposition.

Parallel communication between nodes/devices uses MPI, including all-to-all communications. GPU offloading is via OpenACC.

This repo is part of the UKRI Living Benchmarks project. These tests should cover FFTs, MPI and linear algebra.

Maintainers

@pjpbyrne, @pjh503

Overview

Main code/library

Quantum Espresso

Architectures

CPU, GPU.

Languages and programming models

Fortran, OpenACC, MPI.

Seven 'dwarfs'

• Dense linear algebra
• Sparse linear algebra
• Spectral methods
• N-body methods
• Structured grids
• Unstructured grids
• Monte Carlo

Building

The benchmark can be built using Spack or manually using CMake. The full instructions to compile Quantum Espresso are covered in the compilation instructions, but in brief, it can be compiled easily with spack:

spack install quantum-espresso+openmp+mpi+scalapack

will give a basic QE built with somewhat sensible options.

Usage

Each of the folders in benchmark contains one test and these can generally be run with

mpirun -np <number of processors> pw.x -i <input file> > <output file>
../../bin/extract.py <output file>

Further instructions are in the benchmarks folder

Results

At the end of the run, a set of timings will be presented for each calculation:

P: PWSCF_cpu: 6.39 s (r:0, l:None, u:None)
P: PWSCF_wall: 6.45 s (r:0, l:None, u:None)
P: electrons_cpu: 5.75 s (r:0, l:None, u:None)
P: electrons_wall: 5.8 s (r:0, l:None, u:None)
P: c_bands_cpu: 4.61 s (r:0, l:None, u:None)
P: c_bands_wall: 4.65 s (r:0, l:None, u:None)
P: cegterg_cpu: 4.32 s (r:0, l:None, u:None)
P: cegterg_wall: 4.36 s (r:0, l:None, u:None)
P: calbec_cpu: 0.16 s (r:0, l:None, u:None)
P: calbec_wall: 0.16 s (r:0, l:None, u:None)
P: fft_cpu: 0.21 s (r:0, l:None, u:None)
P: fft_wall: 0.21 s (r:0, l:None, u:None)
P: ffts_cpu: 0.0 s (r:0, l:None, u:None)
P: ffts_wall: 0.0 s (r:0, l:None, u:None)
P: fftw_cpu: 3.63 s (r:0, l:None, u:None)
P: fftw_wall: 3.66 s (r:0, l:None, u:None)

The first numbers are the total seconds of CPU and wallclock time.