Impact of Row Sorting on the Sparse Matrix-Vector Product on GPU

Provided by: IRD India
Topic: Hardware
Format: PDF
Modern Graphics Processing Units (GPUs) allow for a high throughput in processing a massive amount of floating-point operations. The Sparse Matrix-Vector (SpMV) product is of particular interest and subject to intensive research in the GPGPU (General Purpose GPU) computation community, because it is repeatedly used in numerous scientific and engineering applications such as, for example, circuits simulation and ground-water model simulation among others. Bearing this in mind, the authors have studied the impact of the sorting of matrix rows, according to their length (that is, by the number of nonzero values), on the GPU performance in the computing of the SpMV product, using synthetic matrices and benchmark matrices of different dimensions.

Find By Topic