Distributed linear solvers

So this is probably ready - please review and criticise 😉

CC @oberhuber @hanousek

added 38 commits

• df7d6df4...40a6fbec - 10 commits from branch develop
• da1d02b1 - Implemented DistributedArray
• 00971191 - Implemented DistributedVector
• 20dc15de - Basic implementation of DistributedMatrix
• e55b52d3 - Implemented vectorProduct for distributed matrices
• 9e6b9ade - Refactoring Subrange and Partitioner
• 4af3efc6 - Updated DistributedVector
• 55339b85 - Updated DistributedMatrix, simplified usage of Partitioner
• 76111496 - Added tnl-benchmark-distributed-spmv
• 51d16ef0 - Refactoring benchmarks - common getHardwareMetadata function
• 616f0cbe - Rewritten tnl-benchmark-linear-solvers
• 59516c25 - Refactoring the Benchmarks class to make it possible to add extra columns with results
• deafccb3 - Updating tnl-benchmark-linear-solvers: log iterations and residues
• df8fcd53 - Implemented DistributedArrayView and DistributedVectorView
• 9bbe9ac8 - Refactoring distributed SpMV
• f0a853ae - Refactoring diagonal preconditioner
• f26a9b8a - Refactoring preconditioners - the solve method can return just void
• 061710da - Refactoring ILU preconditioners: added TriangularSolve.h
• 163ac7f8 - Updating linear solvers and preconditioners for distributed matrices/vectors
• 356abdcb - Optimized ILUT
• 67a99cb6 - Refactoring GMRES variants
• 44d1be85 - Distributed GMRES
• 905c3fcb - Optimization in GMRES: avoiding a priori known scalar product
• 48e6fe8f - Distributed BiCGstab(ell)
• 36107967 - Updated tnl-benchmark-linear-solvers
• 601e8203 - Added command-line flags to select the solvers and preconditioners in tnl-benchmark-linear-solver
• 54018f38 - Benchmarks: return the mean time instead of total time
• f531f98f - Fixed setup of the preconditioner in tnl-benchmark-linear-solvers
• 66fcb734 - Improved ILU(0) in CUDA - separate storage for the L and U factors

Compare with previous version

added 24 commits

• 449d7efe - Rewritten tnl-benchmark-linear-solvers
• 3f7ce6f1 - Refactoring the Benchmarks class to make it possible to add extra columns with results
• 726cf4b1 - Updating tnl-benchmark-linear-solvers: log iterations and residues
• fb97bcd0 - Benchmarks: return the mean time instead of total time
• 7a8ba0db - Implemented DistributedArrayView and DistributedVectorView
• 663f3b2d - Refactoring distributed SpMV
• bc9a2b30 - Refactoring diagonal preconditioner
• dc4b2498 - Refactoring preconditioners - the solve method can return just void
• 63dda575 - Refactoring ILU preconditioners: added TriangularSolve.h
• 8ae818f1 - Updating linear solvers and preconditioners for distributed matrices/vectors
• e7c80efe - Optimized ILUT
• 0a561076 - Refactoring GMRES variants
• ee6823f0 - Distributed GMRES
• e252921b - Optimization in GMRES: avoiding a priori known scalar product
• 725d6bb2 - Distributed BiCGstab(ell)
• 36870661 - Updated tnl-benchmark-linear-solvers
• 2367e150 - Added command-line flags to select the solvers and preconditioners in tnl-benchmark-linear-solver
• 4b5548b1 - Improved ILU(0) in CUDA - separate storage for the L and U factors
• a813195e - Refactoring: added SparseOperations.h containing the copySparseMatrix function
• 3e1c1eec - Cleaned up Matrices/CMakeLists.txt
• e9984112 - Refactoring copySparseMatrix: implementation of cross-device copy
• 15f4dfee - Added copyAdjacencyStructure function
• ea079456 - Fixed cross-device overload of copySparseMatrix_impl
• 12b37fb0 - Refactoring: moved reorderMatrix from benchmarks into SparseOperatiosn.h as copySparseMatrix

Compare with previous version

merged

mentioned in commit 774735aa