Oral: Optimizing Data Movement in Quantum Circuit Simulations through Graph Partitioning

Quantum circuit simulation is crucial for studying and prototyping quantum algorithms in a controlled environment, facilitating the understanding of quantum behaviors without requiring access to actual quantum hardware. Quantum circuit simulations using a high number of qubits demand significant amounts of data that cannot easily fit within the memory of a single computer. Typically, such large-scale simulations adopt a distributed approach where the state vector of the given simulation spans across thousands of compute nodes. The aggregate memory of all the nodes is sufficient to store the data at the cost of requiring communication across the network. However, the terabytes of data and the lower bandwidths across the network make network communication a critical bottleneck. As such, we present HamFuse, a lightweight framework that focuses on improving data movement by reducing the number of communication steps for a given quantum circuit. Our proposed framework provides a two-step process. We first inspect the quantum circuit, cast it as a graph partitioning problem, optimize on node and network communication and generate the corresponding code. We then execute the optimized implementation for the given circuit on any CPU or GPU based systems. We show that our approach provides significant improvements to time to solutions for a wide range of quantum circuits.