面向异构多核并行机的辐射流体力学并行预条件技术

The numerical simulation of radiation hydrodynamics is one of the key steps in ICF numerical simulation, and thus to research its parallel computation method is of important theoretical and practical significance..In the numerical simulation of radiation hydrodynamics, more than 80% of the computation is in solving large sparse linear algebraic equations, and the high performance computation of the latter is the key problem of the former. With the increase of the problem size, and the development of heterogeneous multi-core parallel computers, many problems occur when studying high performance parallel numerical methods to solve large scale sparse linear systems with heterogeneous multi-core parallel machines. One of the problems is that the performance difference in different levels of heterogeneous multi-core parallel machine architecture is very large, so it is critical challenge to design algorithms with high convergence rate and good parallel scalability. Another problem is that to suit the multi-level architecture of heterogeneous multi-core parallel machines, the designed algorithms are more complicated and thus bring large challenge for convergence analysis. .The project organically combines the architecture characteristic of heterogeneous multi-core parallel machines and the theory of numerical algebraic, and studies the parallel preconditioned iterative algorithms of the numerical simulation of radiation hydrodynamics on heterogeneous multi-core parallel machines, and analyzes their convergence theories.

辐射流体力学数值模拟是ICF数值模拟的关键环节之一，研究其并行计算方法具有非常重要的理论意义和应用价值。.辐射流体力学数值模拟中80%以上的计算量是在求解大型稀疏线性代数方程组，后者的高效并行求解成为前者的关键问题。随着问题规模的逐渐增大，以及异构多核并行计算机的发展，研究出适应异构多核并行机的高效求解大型稀疏线性代数方程组的并行数值方法将面临诸多难题。其一，由于异构多核并行机体系结构各层级的性能差异非常大，在算法设计时要同时保证比较好的收敛速度和并行可扩展性，这就给算法设计带来了巨大的挑战。其二，由于要适应异构多核大型并行机多层次结构的特征，设计的算法复杂度更高，给收敛性分析带来了巨大的挑战。.本项目将异构多核并行计算机体系结构特征与数值代数理论进行有机的结合，研究面向异构多核并行机的辐射流体力学数值模拟的并行预条件迭代算法，并进行收敛性理论分析。