Performance-driven multi-FPGA Jacobian template calculation optimal deployment method and system

Abstract

The invention discloses a performance-driven multi-FPGA Jacobian template calculation optimal deployment method and system, and the method comprises the steps: carrying out the hexagonal partitioning of Jacobian template calculation, and enabling the data between blocks to be subjected to parallel calculation; carrying out formalized description on N FPGA chips running Jacobian template calculation, and expressing various attributes of different FPGA chips by using mathematics; performing mathematical modeling by using the hexagonal blocks and the formalized description content, determining constraint conditions of the model, and determining a target function according to the total cost, resource occupation and the total time consumption of the FPGA chip model; and writing the attributes of the Jacobian hexagon blocks, the formalized description content of the FPGA chip, various constraint conditions and the target function content, generating a model file, performing optimizing to obtain an optimal result, and realizing an optimal placement strategy by using the optimal result. According to the method and system, the optimal placement of the template calculation on the multiple FPGA chips is realized, the resource waste of the FPGA chips is effectively reduced, the efficient operation of the template calculation is realized, and the engineering expenditure of the template calculation is further saved.

Description

technical field [0001] The invention belongs to the field of electronic technology, and in particular relates to a performance-driven multi-FPGA Jacobian template calculation optimal deployment method and system. Background technique [0002] Stencil Computation is a calculation method that iteratively updates array element values ​​according to a fixed "stencil" form. As a simulation process of scientific calculations such as fluid dynamics and electromagnetics, it is an important method of simulation calculation. According to different application scenarios and calculation methods, there are many types of template calculations. Among the many template calculation categories, Jacobi calculation has a wide range of research and applications in the field of solving linear equations. [0003] figure 1 It is a schematic diagram of 1D Jacobi calculation, in which the horizontal axis is the space dimension i, and the vertical axis is the time dimension t. Each solid black dot i...

AI Technical Summary

Problems solved by technology

[0008] At present, in the field of template computing acceleration based on FPGA, there is a lack of related research on multiple FPGAs running template computing. Compared with a single chip, multi-FPGA chips need to consider the division of template computing tasks, FPGA selection and placement strategies, etc.

Images