Comparison of the computational and analytical evaluations for multiple acceleration of multiple linear system solution by block LU-decomposition

Authors: Kuksenko S. P., Surovtsev R. S.

Annotation: The results of the performance evaluation of noncommercial linear algebra libraries BOOST and Eigen on example of LU- decomposition are presented. It is revealed that the Eigen algorithm up to 3–53 times faster depending on variable type and Microsoft Visual C++ project configuration. Implementation of the block LU-decomposition algorithm for the multiple linear systems solution using Eigen library functions is carried out. The computational evaluations with the earlier analytical evaluations are compared. It is shown that the evaluations of acceleration are similar in type of the dependence on the number of solved linear algebraic systems, but differ on values and in type of the dependence on the number of right-hand sides.

Keywords: multiple solution, linear algebraic system, boost, eigen, block lu-decomposition, analytical and computational evaluation