Fast simulation of quantum algorithms using circuit optimization

TL;DR

This paper introduces a specialized compiler pass integrated with a high-performance quantum simulator to optimize quantum circuit simulation, significantly reducing simulation time by about 50%.

Contribution

It presents a novel compiler optimization technique that leverages simulator-specific instructions to enhance simulation efficiency, bridging the gap between simulation and compilation frameworks.

Findings

Simulation time is typically halved with the new optimization.

The approach reduces communication overhead in distributed simulation.

The method is applicable to arbitrary quantum circuits.

Abstract

Classical simulators play a major role in the development and benchmark of quantum algorithms and practically any software framework for quantum computation provides the option of running the algorithms on simulators. However, the development of quantum simulators was substantially separated from the rest of the software frameworks which, instead, focus on usability and compilation. Here, we demonstrate the advantage of co-developing and integrating simulators and compilers by proposing a specialized compiler pass to reduce the simulation time for arbitrary circuits. While the concept is broadly applicable, we present a concrete implementation based on the Intel Quantum Simulator, a high-performance distributed simulator. As part of this work, we extend its implementation with additional functionalities related to the representation of quantum states. The communication overhead is reduced by changing the order in which state amplitudes are stored in the distributed memory, a concept analogous to the distinction between local and global qubits for distributed Schroedinger-type simulators. We then implement a compiler pass to exploit the novel functionalities by introducing special instructions governing data movement as part of the quantum circuit. Those instructions target unique capabilities of simulators and have no analogue in actual quantum devices. To quantify the advantage, we compare the time required to simulate random circuits with and without our optimization. The simulation time is typically halved.