The emergence of multi-core and heterogeneous architectures requires the processing of a number of linear algebra algorithms to take advantage of accelerators, such as graphics processors. There is a difficult class of problems involving linear algebra operations for thousands of smalldimensional matrices. Batched computations continue to have a wide range of applications in scientific calculations. Their goal is to more efficiently transfer the application of algorithms to high-performance multi-core architectures. Some operations of linear algebra are presented in this work for complex Hermitian small matrices, which are grouped together under the name of Batched BLAS. In this work the performances of the complex Hermitian Batched matrixmatrix multiplication, matrix-vector multiplication and 2nd rank update operations are presented on NVIDIA Tesla K40c graphics processor with the use of MAGMA library
oai:noad.sci.am:135809
Institute for Informatics and Automation Problems
11th International Conference on Computer Science and Information Technologies CSIT 2017
Mar 3, 2021
Jul 17, 2020
17
https://noad.sci.am/publication/149336
Հրատարակության անուն | Ամսաթիվ |
---|---|
Edita Gichunts, Batched BLAS Problems of Complex Hermitian Small Matrices in theArchitecture of GPU Accelerator | Mar 3, 2021 |
Gichunts Edita
Գիչունց Էդիտա
Ասցատրյան Հրաչյա Գիչունց Էդիտա