Dissertation
Dissertation > Industrial Technology > Radio electronics, telecommunications technology > Microelectronics, integrated circuit (IC) > LSI,ultra LSI

The Acceleration of Matrix Inversion Based on Two-dimensional Mesh NoC

Author HeYingYing
Tutor GaoMingLun;SongYuZuo
School Hefei University of Technology
Course Detection Technology and Automation
Keywords Two - dimensional mesh NoC Isomorphic multicore Matrix inversion Parallel algorithms FPGA prototyping
CLC TN47
Type Master's thesis
Year 2010
Downloads 42
Quotes 0
Download Dissertation

Supercomputer development gave people a lot of inspiration , guidelines multicore research supercomputer reference system based on PC reference system standards that refer to the organizational structure and history of the development of supercomputers multicore systems research . The supercomputer philosophy , the parallelization of the program is the core issue , the core of the parallel program is a parallel algorithm . The nature of the problem of high-density computing matrix operations , on the basis of the relationship between the software and hardware of multicore preliminary exploration of large matrix operations in multi-core systems optimized implementation techniques , specific to the large dimensions of the complex matrix calculation ( sub-blocks, the inverse transpose triangulation ) objects , the research matrix operations parallel algorithm , as well as high-density computing isomorphic multicore structure parallel algorithm decomposition , and then mix and task allocation process . Through load balanced multi-core systems Nuclear , looking for the minimum routing path , optimizing the allocation of resources , to achieve high-density computing tasks driven , fully tap the parallel capabilities of multi-core systems , improve multicore system to calculate the speed of the target . The main content of the paper is as follows: 1 reference design is based on a two-dimensional grid structure NoC isomorphism multi-core systems . According to the theory of parallel algorithms compare multiple existing matrix inversion algorithm to select a high degree of parallelism based on the block LU decomposition matrix inversion methods , mining parallel capabilities. 3 . Follow the balanced multi-core systems Nuclear load, reduce synchronization overhead , optimize resource allocation principle , the matrix inverse parallel algorithm is mapped to the two-dimensional mesh NoC multicore architecture , give full play to the parallel capabilities of multi-core systems , improve the matrix inversion computing speed. Use FPGA prototyping eventually be mapped from the matrix inversion parallel algorithm isomorphic multicore acceleration effect . The results show that this task allocation method is feasible acceleration effect .

Related Dissertations
More Dissertations