DiaQ: Efficient State-Vector Quantum Simulation

TL;DR

DiaQ introduces a quantum-specific sparse matrix format and a high-performance C++ library that significantly accelerates digital quantum simulations by exploiting inherent sparsity patterns, especially on multi-core and SIMD architectures.

Contribution

The paper presents DiaQ, a novel sparse matrix format tailored for quantum circuit sparsity, and a C++ library with multi-core and SIMD optimizations, improving simulation efficiency.

Findings

Performance improvements of ~26.67% for GHZ-28 circuits.

Performance improvements of ~32.72% for QFT-29 circuits.

Demonstrated efficiency gains on benchmarks from SupermarQ and QASMBench.

Abstract

In the current era of Noisy Intermediate Scale Quantum (NISQ) computing, efficient digital simulation of quantum systems holds significant importance for quantum algorithm development, verification and validation. However, analysis of sparsity within these simulations remains largely unexplored. In this paper, we present a novel observation regarding the prevalent sparsity patterns inherent in quantum circuits. We introduce DiaQ, a new sparse matrix format tailored to exploit this quantum-specific sparsity, thereby enhancing simulation performance. Our contribution extends to the development of libdiaq, a numerical library implemented in C++ with OpenMP for multi-core acceleration and SIMD vectorization, featuring essential mathematical kernels for digital quantum simulations. Furthermore, we integrate DiaQ with SV-Sim, a state vector simulator, yielding substantial performance improvements across various quantum circuits (e.g., ~26.67% for GHZ-28 and ~32.72% for QFT-29 with multi-core parallelization and SIMD vectorization on Frontier). Evaluations conducted on benchmarks from SupermarQ and QASMBench demonstrate that DiaQ represents a significant step towards achieving highly efficient quantum simulations.