Julia interface to hypre linear solvers (https://github.com/hypre-space/hypre)
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 

4.6 KiB

HYPRE.jl changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Changed

  • PartitionedArrays.jl dependency upgraded from release series 0.3.x to release series 0.5.x. (#17, #18)
  • PartitionedArrays.jl support is now moved to a package extension. (#23)
  • CEnum.jl dependency upgraded to release series 0.5.x (release series 0.4.x still allowed). (#17, #18)

v1.5.0 - 2023-05-26

Changed

  • PartitionedArrays.jl dependency upgraded from version 0.2.x to version 0.3.x. (#16)

v1.4.0 - 2023-01-20

Added

  • New function HYPRE.GetFinalRelativeResidualNorm(s::HYPRESolver) for getting the final residual norm from a solver. This function dispatches on the solver to the corresponding C API wrapper LibHYPRE.HYPRE_${Solver}GetFinalRelativeResidualNorm. (#14)
  • New function HYPRE.GetNumIterations(s::HYPRESolver) for getting the number of iterations from a solver. This function dispatches on the solver to the corresponding C API wrapper LibHYPRE.HYPRE_${Solver}GetNumIterations. (#14)

v1.3.1 - 2023-01-14

Fixed

  • Solvers now keep an reference to the added preconditioner to make sure the preconditioner is not finalized before the solver. This fixes crashes (segfaults) that could happen in case no other reference to the preconditioner existed in the program. (#12)
  • The proper conversion methods for ccall are now defined for HYPREMatrix, HYPREVector, and HYPRESolver such that they can be passed direcly to HYPRE_* functions and let ccall guarantee the GC preservation of these objects. Although not observed in practice, this fixes a possible race condition where the matrix/vector/solver could be finalized too early. (#13)

v1.3.0 - 2022-12-30

Added

  • Rectangular matrices can now be assembled by the new method HYPRE.assemble!(::HYPREMatrixAssembler, i::Vector, j::Vector, a::Matrix) where i are the rows and j the columns. (#7)

Fixed

  • All created HYPRE objects (HYPREMatrix, HYPREVector, and HYPRESolvers) are now kept track of internally and explicitly finalized (if they haven't been GC'd) before finalizing HYPRE. This fixes a "race condition" where MPI and/or HYPRE would finalize before these Julia objects are garbage collected and finalized. (#8)

Deprecated

  • The method HYPRE.assemble!(A::HYPREMatrixAssembler, ij::Vector, a::Matrix) have been deprecated in favor of HYPRE.assemble!(A::HYPREMatrixAssembler, i::Vector, j::Vector, a::Matrix), i.e. it is now required to explicitly pass rows and column indices individually. The motivation behind this is to support assembling of rectangular matrices. Note that HYPRE.assemble!(A::HYPREAssembler, ij::Vector, a::Matrix, b::Vector) is still supported, where ij are used as row and column indices for a, as well as row indices for b. (#6)

v1.2.0 - 2022-10-12

Added

  • Added assembler interface to assemble HYPREMatrix and/or HYPREVector directly without an intermediate sparse structure in Julia. (#5)

v1.1.0 - 2022-10-05

Added

  • Added support for MPI.jl version 0.20.x (in addition to the existing version 0.19.x support). (#2)

v1.0.0 - 2022-07-28

Initial release of HYPRE.jl.