Systems and methods for improving the execution of computational algorithms

摘要

A system for solving equations of a given scientific code for a large dataset represented as a mesh includes first computing circuitry, second computing circuitry, and communication circuitry. The first computing circuitry includes software to arrange the mesh in a sequential order, and a memory to store the mesh data and the results obtained in the execution of the scientific code. The second computing circuitry includes software to execute the scientific code for one vertex of the mesh, and a cache memory to store all the neighboring vertices to the one vertex. The communication circuitry is between the first and second computing circuitry so that the first computing circuitry fed the second computing circuitry with the vertices of the mesh following the sequential order, and the scientific code is executed for all the vertices of the mesh and its results are returned to the first computing circuitry.

主权项

1. A system for processing a dataset represented as a mesh, comprising:
first computing circuitry including:
software to reorder an original order of vertices of said mesh into a sequential order arranging the vertices according to a sequence of groups of neighboring vertices that are to be processed;a memory to store the mesh dataset and results obtained in execution of a scientific code; second computing circuitry including:
software to execute said scientific code for one vertex of said mesh;a cache memory to store all the neighboring vertices to said one vertex; and communication circuitry between said first and second computing circuitry so that said first computing circuitry feeds the second computing circuitry with the vertices of the mesh following said sequential order, and said scientific code is executed for all the vertices of the mesh and the results obtained in the execution of said scientific code are returned to said first computing circuitry, wherein said mesh is arranged in a sequential order following a breadth-first algorithm starting from a seed vertex selected for reducing the cache memory needed for storing all the neighboring vertices involved in the execution of said scientific code for each vertex, the seed vertex selected is a vertex closest to a center of mass of the mesh assuming that a weight of each vertex is 1, said scientific code is a Computational Fluid Dynamics (CFD) code and said dataset is the mesh of an object that moves through a fluid medium, and said object is an aircraft or a component of an aircraft.