The given development is devoted to the optimization of Strassen algorithm for MIC architecture. In the course of the project, several versions of Strassen algorithm parallelization have been implemented and tested. The analysis of the experimental results has shown that the most productive methods of parallelization are different for different matrices size and different numbers of threads. Thus, the system for automatic parallelization configuration selection depending on the input parameters, primarily the matrices size and the number of threads, has been realized. This system also controls and limits the effective maximum number of threads and the threshold parameters.
Purpose: optimization of matrix multiplication by Strassen algorithm in systems built on MIC architecture
Specification
Client: | Intel, Nizhny Novgorod, Russia |
Area of use: | fast matrix multiplication |
Type (platform): | Intel® MIC Architecture |
Technologies and algorithms in use: | MIC, C/C++ |
More information: | The article on our website |