ddemidov / amgcl

90%

master: 88%

LAST BUILD ON BRANCH block-null-space

branch: block-null-space

CHANGE BRANCH

x

Reset

• block-null-space
• 0.5.0
• 0.6.0
• 0.6.1
• 0.6.2
• 0.6.2-r1
• 0.6.3
• 0.6.4
• Mut_et_al
• adaptive_sa
• amg_at_coarse
• appveyor
• bicgstabl_breakdown
• check_Zt_y
• classic_bicgstab
• coarse_solve_v1
• coarse_solve_v2
• codecov
• deflation-def1
• deflation-def2
• delphi
• epetra-adapter
• filtered_matrix
• general_deflation
• hpx-pr-1337
• hpx_backend
• hpx_backend_algorithms
• hpx_backend_better_inner_product
• hpx_backend_dataflow
• hpx_backend_hangs
• machine-precision-zero
• master
• mpi
• no-null-space
• null-space
• parallel_coarse_solve
• python
• ranged_solvers
• runtime
• scalar_with_nullspace
• scale-distributed-matrix
• skip_small_aggregates
• spai1
• standalone_bicgstab
• travis-mpi
• true-err-in-gmres
• unit_deflation_vectors
• v2

Improve work with null-space vectors for non-scalar systems

Adjust block size for coarse levels when null-space is provided.

This is an API breaking change!
Remove `Aggregates` template parameter from aggregation-based coarseners.

Starting with this commit aggregation based coarseners (
    amgcl::coarsening::aggregation,
    amgcl::coarsening::smoothed_aggregation,
    amgcl::coarsening::smoothed_aggr_emin
) always use amgcl::coarsening::pointwise_aggregates for aggregation.

The change is required since scalar problem may become non-scalar if
near null-space vectors are provided. The runtime interface has not
changed since runtime class versions were based on pointwise_aggregates
anyway.