摘要 |
PROBLEM TO BE SOLVED: To provide a parallel processing method of an operation for performing a parallel matrix operation at a high speed in a shared memory scalar computer. SOLUTION: First of all, the computer is broken down into blocks E and HLU. Next, a block B is updated by using an upper triangle part of the block E, and a block D is updated by using a lower triangle part of the block E. By this time, blocks F and I are already updated by LU breakdown. Blocks A, C, G and I are updated by using the blocks B, D, F and H, and an upper triangle part of the block E is updated, and the blocks D and F are finally updated. The block E is updated for the second time, and the blocks B and H are updated by using the result. The block E is finally updated, and processing is finished by finishing replacement processing of a pivot. These block processing is divided into a plurality of threads, and is executed in parallel. COPYRIGHT: (C)2004,JPO
|