What this benchmark does

• A gate-by-gate simulation of a quantum computer with a full state vector formulation.
• It simulates a simplified version of Shor's algorithm with increasing number of qubits until resources are exhausted.
• The score (states/s) is how fast the Approximate Quantum Fourier Transform can be computed (the modular exponentiation is not timed).
• The goal is to quantify the ability of a computer to simulate ideal quantum circuits.

Why quantum factorization simulation as a benchmark for HPC

• Quantum factorization is a widely known & relevant problem, and it is known to be hard to simulate.
• Each additional qubit doubles RAM usage, CPU power and internode communication: good to test large machines.
• The output is reproducible, easy to validate and understand.
• Runs in a reasonable time: 1/2‒3 hours.
• Minimal portable code with less than ~300 lines of C and MPI, very few dependencies.
• It just runs: no input or specialized knowledge from user.

How to compile and run

This is system dependent. For example, in systems with mpicc and Slurm one usually compiles with

mpicc -Ofast quansimbench.c -o quansimbench -lm

and runs the program with

sbatch quansimbench.batch

using a Slurm batch script such as:

#SBATCH –o output-%J.txt
#SBATCH --nodes=8     # 8 nodes (must be a power of 2)
#SBATCH –n 64         # 8 ranks per node (must be a power of 2)
#SBATCH –p normal     # what queue
#SBATCH –t 02:00:00   # usually less than 3 hours
mpirun ./quansimbench

In all systems, the number of nodes and number of cores must be a power of two.

The code contains comments with hints to optional addition of OpenMP as well.

How to test the ability of your computer to simulate quantum AQFTs

1. First run the benchmark until completion with increasing number of nodes/ranks until you get the largest number of States/s in the last row. That usually corresponds to the largest number of nodes/ranks you can run. If the program hangs, read the last visible row.

2. Verify that you get the "Pass=yes" result on all rows.