We introduce qclab++, a light-weight, fully-templated C++ package for GPU-accelerated quantum circuit simulations. The code offers a high degree of portability as it has no external dependencies and the GPU kernels are generated through OpenMP offloading. qclab++ is designed for performance and numerical stability through highly optimized gate simulation algorithms for 1-qubit, controlled 1-qubit, and 2-qubit gates. Furthermore, we also introduce qclab, a quantum circuit toolbox for Matlab with a syntax that mimics qclab++. This provides users the flexibility and ease of use of a scripting language like Matlab for studying their quantum algorithms, while offering high-performance GPU acceleration when required. As such, the qclab++ library offers a unique combination of features. We compare the CPU simulator in qclab++ with the GPU kernels generated by OpenMP and observe a speedup of over $40\times$ . Furthermore, we also compare qclab++ to other circuit simulation packages, such as cirq-qsim and qibo, in a series of benchmarks conducted on NERSC's Perlmutter system and illustrate its competitiveness.We introduce qclab++, a light-weight, fully-templated C++ package for GPU-accelerated quantum circuit simulations. The code offers a high degree of portability as it has no external dependencies and the GPU kernels are generated through OpenMP offloading. qclab++ is designed for performance and numerical stability through highly optimized gate simulation algorithms for 1-qubit, controlled 1-qubit, and 2-qubit gates. Furthermore, we also introduce qclab, a quantum circuit toolbox for Matlab with a syntax that mimics qclab++. This provides users the flexibility and ease of use of a scripting language like Matlab for studying their quantum algorithms, while offering high-performance GPU acceleration when required. As such, the qclab++ library offers a unique combination of features. We compare the CPU simulator in qclab++ with the GPU kernels generated by OpenMP and observe a speedup of over $40\times$. Furthermore, we also compare qclab++ to other circuit simulation packages, such as cirq-qsim and qibo, in a series of benchmarks conducted on NERSC's Perlmutter system and illustrate its competitiveness.

arXiv:2303.00123Feb 28, 20232 citations

Study of GPU Acceleration in Genetic Algorithms for Quantum Circuit Synthesis

Martin Lukáč, Georgiy Krylov

In this work we present a comparative study of several GPU accelerated elements of a Genetic Algorithm (GA) for the synthesis of quantum circuits on the level of Electro-Magnetic (EM) pulses. The novelty in our approach is in the implementation: a) a completely GPU accelerated quantum simulator, b) GPU accelerated genetic operators and fitness evaluation and finally c) a set of GPU implemented optimizations for GPU accelerated evolutionary search optimization for the synthesis of quantum circuits. The reason for using EM pulses model for synthesis is the observation that this model requires the largest amount of elementary rotations to implement quantumlogic gates and thus provides a good measure to evaluate the efficiency of the acceleration by the GPU processor. The reason to use a GA is the advantage of pseudo evolutionary search in very large problem space such as the one defined by the Ising model where the EM realized quantum circuits are evolved. As a result of the several GPU optimizations several new circuits implementations are presented and their cost is compared to thecurrently known Ising model implementations.In this work we present a comparative study of several GPU accelerated elements of a Genetic Algorithm (GA) for the synthesis of quantum circuits on the level of Electro-Magnetic (EM) pulses. The novelty in our approach is in the implementation: a) a completely GPU accelerated quantum simulator, b) GPU accelerated genetic operators and fitness evaluation and finally c) a set of GPU implemented optimizations for GPU accelerated evolutionary search optimization for the synthesis of quantum circuits. The reason for using EM pulses model for synthesis is the observation that this model requires the largest amount of elementary rotations to implement quantumlogic gates and thus provides a good measure to evaluate the efficiency of the acceleration by the GPU processor. The reason to use a GA is the advantage of pseudo evolutionary search in very large problem space such as the one defined by the Ising model where the EM realized quantum circuits are evolved. As a result of the several GPU optimizations several new circuits implementations are presented and their cost is compared to thecurrently known Ising model implementations.

IEEE International Symposium on Multiple-Valued Logic (ISMVL)May 1, 20174 citations

An efficient quantum circuit implementation of Shor’s algorithm for GPU accelerated simulation

XinJian Tan, Peng Gao

In this study, we introduce a novel implementation of Shor’s algorithm specifically designed for the Graphics Processing Unit (GPU) acceleration framework. Our focus lies on achieving efficient execution of the modular multiplication circuit through GPU simulation. To seamlessly integrate our design into the PyQPanda library framework, we made necessary modifications, making a deliberate trade-off by sacrificing a small number of quantum resources to leverage the advantages of GPU acceleration. Subsequently, we conducted simulations and rigorously validated the functionality of our circuit using the PyQPanda library, resulting in a significant speedup compared to a central processing unit-only mode.In this study, we introduce a novel implementation of Shor’s algorithm specifically designed for the Graphics Processing Unit (GPU) acceleration framework. Our focus lies on achieving efficient execution of the modular multiplication circuit through GPU simulation. To seamlessly integrate our design into the PyQPanda library framework, we made necessary modifications, making a deliberate trade-off by sacrificing a small number of quantum resources to leverage the advantages of GPU acceleration. Subsequently, we conducted simulations and rigorously validated the functionality of our circuit using the PyQPanda library, resulting in a significant speedup compared to a central processing unit-only mode.

AIP AdvancesFeb 1, 20246 citations

40\times

. Furthermore, we also compare qclab++ to other circuit simulation packages, such as cirq-qsim and qibo, in a series of benchmarks conducted on NERSC's Perlmutter system and illustrate its competitiveness.