Many-core pipeline parallelism method adopting implicit solution

Abstract

The invention provides a many-core pipeline parallelism method adopting an implicit solution. The method comprises the following steps: performing data partition on an variable a<i,j> on p computing cores of a many-core processor in the i direction to enable the variable a<i,j> to be partitioned p segments, wherein p is an integer greater than 1 and {a<i,j>,i=n<k>,..., n<k+1>-1} is stored on a computing core pk, wherein K is equal to 0,...,p-1, and 1 is equal to n0<n1<...<n<p-1>=n+1; using communication of registers among computing cores to realize pipeline parallelism.

Description

technical field [0001] The invention relates to the technical field of computers, in particular to a many-core pipeline parallel method using implicit solution. Background technique [0002] Implicit algorithm (implicit algorithm) refers to that in the process of converting differential equations into algebraic equations, if the variables at that time are needed to solve the function value at a certain time, then this discrete method is an implicit algorithm. [0003] In numerical calculation, the implicit algorithm is widely used because of its good stability and fast convergence speed. The discretization of differential equations by implicit algorithm finally boils down to solving large sparse matrix linear algebraic equations in parallel in blocks. If the two-step scanning format is used, that is, the LU-SGS method is used to solve the linear algebraic equation system, the recursive equation will be obtained, and after the boundary conditions are given, the recursive met...

AI Technical Summary

Problems solved by technology

However, many-core processors have a high degree of parallelism, each computing core has limited storage resources, and the data interaction between computing cores is complex. Therefore, it is more difficult to implement the parallelism of implicit algorithms on many-core processors.

[0005] At present, most applications on many-core processors use explicit numerical algorithms. Many-core applications of implicit algorithms are very limited. They can only achieve partial parallelism, and cannot solve the entire calculation area as a whole, so the strict implicitness of the algorithm cannot be guaranteed. , which has a great impact on the stability and convergence of numerical calculations

Images