Skip to content

Latest commit

 

History

History
6 lines (3 loc) · 506 Bytes

File metadata and controls

6 lines (3 loc) · 506 Bytes

Preconditioned-Conjugate-Gradient-Method-in-CUDA

Implementation of Incomplete-Cholesky preconditioned conjugate-gradient for symmetric PSD systems using CUDA 12.0.

The Matrix-Vector multiplication has been implemented in CSR format with one warp taking care of one row of matrix. The sparse lower and upper triangular solver have been implemented using cuSPARSE library. Considering the IC decomposition is deprecated in the new version of cuSPARSE, it has been calculated manually on the host side.