METHOD FOR MULTIPLICATION OF SPARSE MATRIX ON THE GPU

文件类型：PDF文档
文件大小：336K
Disclosed is a sparse matrix multiplication method using a GPU. A method performing multiplication between sparse matrices includes a step of dividing each of a first sparse matrix and a second sparse matrix into partial matrices by a central processor unit (CPU), a step of generating a workload table for each partial matrix of the first and second sparse matrices to equally distribute a workload of a thread and a workload of a thread block, and a step of performing multiplication between the first and second sparse matrices by using a graphic processor unit (GPU) based on the partial matrix of the first sparse matrix, the partial matrix of the second sparse matrix, and the workload table.(110) Divide a first sparse matrix and a second sparse matrix into partial matrices(120) Generate a WLT for each partial matrix to equally allocate a workload of a thread block and a workload of a thread(130) Transmit the partial matrix of the first sparse matrix, the partial matrix of the second sparse matrix, and the WLT to a GPU(140) Receive a result multiplied through a cross product from the GPU(AA) Start(BB) EndCOPYRIGHT KIPO 2014