2026-10-14T21:08:03Z https://www.scidb.cn/oai

10.57760/sciencedb.46390 2026-08-21T17:17:24Z

2026-08-21 Dataset for CPU/CUDA performance testing of AMGCL, AMGX, HYPRE, and UNAP in incompressible CFD solvers. doi:10.57760/sciencedb.46390 en This dataset is designed for comprehensive testing and performance analysis of sparse linear system solvers, preconditioners, and CPU/CUDA heterogeneous computing platforms. The warehouse mainly uses sparse matrix in Matrix Market format as input, and organizes multiple experiments from single solver functional validation to multi library unified control variable control around four solver libraries: AMGCL, AMGX, HYPRE, and UNAP. The test covers algebraic multigrid (AMG), incomplete decomposition (ILU, including ILU0, DILU, etc.), Jacobi, LDU, and various Krylov subspace iterative solvers. Based on the capabilities of different solver libraries, the experiment includes a combination of PBiCGSTAB/BicGSTAB, PCG, GMRES, FGMRES, and AMG direct solution or AMG preconditioning. Among them, AMGCL testing provides ILU0, Jacobi, and AMG+PBiCGSTAB configurations; AMGX testing covers ILU, Jacobi, PMIS/aggregated AMG, DILU, PBiCGSTAB, PCG, and FGMRES configurations in CUDA environment; HYPRE testing covers BoomerAMG, PCG, BiCGSTAB, GMRES, and their AMG combinations; UNAP testing covers the LDU solution call chain of CPU/CUDA and includes Jacobi+PBiCGSTAB testing for unified comparison with other libraries.  AMGCL; AMGX; HYPRE; UNAP; Matrix Market zhao cheng peng PUBLIC https://creativecommons.org/licenses/by-sa/4.0/ dataset Science Data Bank