Large-Scale Quantum Fourier Transform Simulation For "Songshan" Supercomputer System

Songshan supercomputer system is a new generation of supercomputer independently developed in China.It adopts the international mainstream supercomputer heterogeneous computing mode of universal CPU processor +DCU acceleration device,and the peak computing performance of the whole machine can reach 100 PFLOPS.In order to expand the platform's scientific computing ecology,verify the feasibility of quantum computing simulation research on the platform,and promote the research and progress of quantum algorithm,this paper selects the quantum Fourier transform to carry out simulation on the platform.The main work and innovation points of this paper are as follows:(1)The quantum road map of the quantum Fourier transform is studied,the code is implemented on the CPU cluster and the performance of the algorithm is analyzed.Through the study and research of the quantum Fourier transform algorithm,the quantum Fourier transform simulation of C language version is realized by using the Lib Quantum quantum simulation library.The program performance analysis technology combining static and dynamic is used to locate the program hot spots,and the operating codes of quantum phase shift gate and Hardamard gate which have a lot of intensive computation in the program are analyzed,which lays the foundation for DCU transplantation.(2)The parallel simulation of quantum Fourier transform on CPU+DCU heterogeneous computing platform is realized.Based on HIP heterogeneous programming model,the quantum Fourier transform simulation is implemented in parallel.The CPU side is responsible for the preparation of the initial state of the quantum register and the judgment of the logical branch part.The quantum logic gate operating code in the program is rewritten for HIP devices,and the mapping between computing hotspots and DCU accelerators is realized.(3)Data transmission and access optimization of large-scale quantum Fourier transform simulation on "Songshan" supercomputer system was carried out.In view of the performance and storage bottleneck of the program in large-scale simulation,the parallel optimization of the program was carried out in combination with the characteristics of the target supercomputer architecture.The problem of low transmission efficiency between the host and the device in the computing node was solved by using MPI concurrency.By hiding computation and communication,DCU can avoid being idle for a long time in the process of data transmission and improve the utilization rate of DCU equipment.The shared memory and register space are used to improve the efficiency of data access,realize the global memory oriented merge access,and effectively utilize the multi-level storage structure of DCU acceleration device.The work in this paper has been verified by experiments on the "Songshan" supercomputer system,and the large-scale 44qu-bits quantum Fourier transform simulation has been realized.The quantum Fourier transform simulation accelerated by CPU+DCU is 17.151 times faster than that by CPU simulation,and has good scalability.The work in this paper provides a reference for the simulation and optimization of other quantum algorithms in the Songshan supercomputer system.