Adaptive parallel LU decomposition method aiming at circuit simulation

Abstract

The invention discloses an adaptive parallel LU decomposition method aiming at circuit simulation and belongs to the field of EDA (electronic design automation). The method is characterized in that LU decomposition is carried out on a circuit matrix by automatically selecting a serial or parallel method based on the characteristics of the circuit matrix, so that each circuit matrix can carry out LU decomposition under the optical performance. In the parallel method, data dependency is extracted from the structure of each circuit matrix, and parallel dispatching is carried out on the tasks in the LU decomposition process based on the data dependency, so as to accelerating the circuit simulation. Test results on the circuit matrixes show that the method provided by the invention can accurately determine the serial or parallel method suitable for each circuit matrix, and for the matrixes suitable for the parallel method, the decomposition speed of the method is 2.11-8.38 times that of LUdecomposition software KLU in geometric average when the number of parallel threads is 1-8.

Description

technical field [0001] The invention relates to an adaptive parallel LU decomposition method for circuit simulation, and belongs to the technical field of electronic design automation (EDA). Background technique [0002] In circuit simulation, solving the linear equations Ax=b is a very time-consuming step, because this step will be repeated many times in the Newton-Raphson iteration and transient iteration of circuit simulation; The scale follows the development of Moore's Law, making the order of the matrix A established by the circuit simulation reach the order of tens of thousands, and it can reach the order of millions or even tens of millions in the post-simulation. This makes solving Ax=b extremely time-consuming. At present, the post-simulation of AD, PLL and other circuits of normal scale usually takes several days to several months. Therefore, how to speed up the solution of Ax=b has become an urgent problem in circuit simulation. [0003] The solution methods o...

AI Technical Summary

Problems solved by technology

The iterative method usually has high requirements on the matrix A, which may be difficult to meet in general calculation problems

Images