https://github.com/huiscliu/sxamg 下面的example ./amg sx: problem size: 236600, 130 x 130 x 14. A: m = 236600, n = 236600, nnz = 1615120 AMG Parameters ----------------------------------------------------------- AMG print level: 3 AMG max num of iter: 1000 AMG tol: 1e-06 AMG ctol: 1e-07 AMG max levels: 30 AMG cycle type: 1 AMG smoother type: 2 AMG smoother order: 1 AMG num of presmoothing: 2 AMG num of postsmoothing: 2 AMG coarsening type: 2 AMG interpolation type: 2 AMG dof on coarsest grid: 10 AMG strong threshold: 0.3000 AMG truncation threshold: 0.2000 AMG max row sum: 0.9000 AMG coarest solver: 1 ----------------------------------------------------------- ### WARNING: Coarse matrix is too dense! ### WARNING: m = n = 191, nnz = 8083! ----------------------------------------------------------- Level Num of rows Num of nonzeros Avg. NNZ / row ----------------------------------------------------------- 0 236600 1615120 6.83 1 118300 2125008 17.96 2 19724 616548 31.26 3 2583 118463 45.86 4 617 31635 51.27 5 191 8083 42.32 ----------------------------------------------------------- Grid complexity = 1.598 | Operator complexity = 2.795 ----------------------------------------------------------- AMG setup time: 2.08698 s AMG setup time: 2.08708 s ----------------------------------------------------------- It Num | ||r||/||b|| | ||r|| | Conv. Factor ----------------------------------------------------------- 0 | 1.000000e+00 | 4.864155e+02 | -.-- 1 | 5.509402e-02 | 2.679858e+01 | 0.0551 2 | 2.312168e-03 | 1.124674e+00 | 0.0420 3 | 9.514382e-05 | 4.627943e-02 | 0.0411 4 | 3.903498e-06 | 1.898722e-03 | 0.0410 5 | 1.604449e-07 | 7.804290e-05 | 0.0411 AMG solve time: 0.969601 s AMG totally time: 3.06794 s AMG residual: 7.80429e-05 AMG relative residual: 1.60445e-07 AMG iterations: 5 2) ./amg-low-level sx: problem size: 88560, 123 x 30 x 24. A: m = 88560, n = 88560, nnz = 605196 AMG Parameters ----------------------------------------------------------- AMG print level: 2 AMG max num of iter: 1000 AMG tol: 1e-06 AMG ctol: 1e-07 AMG max levels: 30 AMG cycle type: 1 AMG smoother type: 2 AMG smoother order: 1 AMG num of presmoothing: 2 AMG num of postsmoothing: 2 AMG coarsening type: 2 AMG interpolation type: 2 AMG dof on coarsest grid: 10 AMG strong threshold: 0.3000 AMG truncation threshold: 0.2000 AMG max row sum: 0.9000 AMG coarest solver: 1 ----------------------------------------------------------- ### WARNING: Coarse matrix is too dense! ### WARNING: m = n = 263, nnz = 17397! ----------------------------------------------------------- Level Num of rows Num of nonzeros Avg. NNZ / row ----------------------------------------------------------- 0 88560 605196 6.83 1 44280 797502 18.01 2 7383 233449 31.62 3 1025 52525 51.24 4 263 17397 66.15 ----------------------------------------------------------- Grid complexity = 1.598 | Operator complexity = 2.819 ----------------------------------------------------------- AMG setup time: 0.707301 s ----------------------------------------------------------- It Num | ||r||/||b|| | ||r|| | Conv. Factor ----------------------------------------------------------- 0 | 1.000000e+00 | 2.975903e+02 | -.-- 1 | 6.597986e-02 | 1.963497e+01 | 0.0660 2 | 3.432394e-03 | 1.021447e+00 | 0.0520 3 | 1.781948e-04 | 5.302905e-02 | 0.0519 4 | 9.220834e-06 | 2.744031e-03 | 0.0517 5 | 4.756874e-07 | 1.415600e-04 | 0.0516 AMG solve time: 0.295148 s AMG residual: 0.00014156 AMG relative residual: 4.75687e-07 AMG iterations: 5 3) ./cg-amg sx: problem size: 178596, 123 x 33 x 44. A: m = 178596, n = 178596, nnz = 1228326 sx: solver: CG, preconditioner: AMG Convergence settings: relative residual: 1.000000e-06, maximal iterations: 2000 Initial residual: 4.226062e+02 itr: 1, residual: 3.076031e+01, relative error: 7.278717e-02 itr: 2, residual: 1.372795e+00, relative error: 3.248403e-03 itr: 3, residual: 3.617905e-02, relative error: 8.560937e-05 itr: 4, residual: 4.572409e-04, relative error: 1.081955e-06 itr: 5, residual: 3.411769e-05, relative error: 8.073163e-08 cg converged: absolute residual: 3.411769e-05, total iterations: 5, time: 0.909855 s verify cg residual: absolute residual: 3.411769e-05, total iterations: 4 转化成矩阵，再采用AMG，加速效果应该比比较明显， YHAMG效果可能会更好。 |
说点什么...