Speaker
Description
I will present our recent work on developing a scalable quantum simulator prototype using the open-source QuEST toolkit. We tackle two major scalability challenges: increasing the number of qubits in simulations and enhancing simulation speed. Initially focusing on optimization, we refactored QuEST’s core data structure to improve cache locality and multithreaded performance on single-node, homogeneous systems. Recognizing that traditional OpenMP scheduling does not address memory locality, we implemented NUMA-aware allocation and a manual task scheduler, ensuring threads are bound to their respective NUMA nodes for optimal memory access. Additional performance improvements were achieved through the integration of AVX-512 SIMD instructions, FMA operations, and prefetching techniques. These optimizations resulted in up to 97% speedup for single-qubit gates and more than a twofold acceleration for the QFT circuit. In this talk I will discuss how these advances lay the foundation for extending our approach to heterogeneous clusters and accelerator-based systems such as GPUs and FPGAs. This work not only contributes to the field of quantum computing but also provides valuable insights for HPC computing, making it relevant to researchers across computer science disciplines.
