Quantum Computer Simulation on GPU Cluster Incorporating Data Locality

摘要

Quantum computer simulation provides researchers with tools for verification of quantum algorithms. GPU (Graphics Processing Units) cluster is an advisable platform for this task. However, the high cost of communication between GPUs makes the simulation inefficiency. To overcome this drawback, we propose the following two methods. (1) A method for GPU cluster quantum simulation to improve the data locality is introduced, and two schemes for data exchanging are proposed. (2) A novel data distribution method for quantum computer simulation on GPU cluster is proposed. Experimental results show that the simulation of 33-qubit Quantum Fourier Transform algorithm using 4 nodes outperforms the serial program of the CPU cluster with a speedup of 129 times.