Andrew Ho
afadd435f5
review comments
2026-01-29 12:50:04 -08:00
Andrew Ho
36d7629b26
residual vector requires hypre >= 2.15.0
2026-01-29 12:41:53 -08:00
Andrew Ho
8c2ffb9d26
added Set/GetUseTwoNorm
2026-01-20 13:25:08 -08:00
Andrew Ho
a024fb10bc
more initial solver configurations
2026-01-07 17:17:19 -08:00
Andrew Ho
531e8d7ad9
Added a way to get the residual vector r and ||r||_p
2026-01-07 17:02:25 -08:00
Andrew Ho
9008d5a050
Improved hypre's documentation and added getter methods (when possible) for solver parameters
2026-01-07 15:36:54 -08:00
Andrew Ho
d1ca02310d
Merge branch 'master' into nbeams/gko-updates-distrib
2025-12-16 12:05:14 -08:00
camierjs
387005e37a
with style
2025-12-09 15:25:57 -08:00
camierjs
1d81f19c37
minor argument type safety casts
2025-12-09 15:16:11 -08:00
Andrew Ho
67311104a5
Merge branch 'master' into nbeams/gko-updates-distrib
2025-10-06 16:40:50 -07:00
Tzanio Kolev
bce6e2ca76
Merge branch 'master' into najlkin/parform-impro
2025-10-05 13:25:11 -07:00
Victor A. P. Magri
59579f2cdb
Fix build with upcoming hypre-3.0
2025-09-26 12:34:18 -04:00
Tzanio Kolev
40d1550fd6
Merge branch 'master' into najlkin/parform-impro
2025-09-17 03:31:41 -07:00
Jan Nikl
193f8a6801
Partially reverted const modifiers in HypreParMatrix::Copy(Bool)CSR().
2025-09-16 16:19:03 -07:00
nbeams
d630bf5250
Add distributed support for Ginkgo solvers and preconditioners
2025-09-10 22:21:27 +00:00
Veselin Dobrev
def35c8a15
In HypreParMatrix::EliminateBC, use stream synchronization instead of
...
device synchronization and do it only when hypre is using GPU-aware MPI.
Remove device synchronization before CUDA/HIP free which implicitly
perform the same synchronization.
2025-06-28 08:05:30 -07:00
Andrew Ho
8dcd0d6349
add device syncs in EliminateBC
...
This should fix potential race conditions in some situations
2025-06-26 15:57:18 -07:00
Jan Nikl
9dab032bd0
Merge branch 'master' into najlkin/parform-impro
2025-04-24 15:40:44 -07:00
Andrew Ho
f3360ac118
switch to MFEM_MPI_CXX_BOOL
...
This is an internal macro which switches between MPI_CXX_BOOL and
MPI_C_BOOL if MPI_CXX_BOOL happens to not be defined (ex.: in Windows).
2025-03-14 14:19:17 -07:00
Andrew Ho
940943268b
replace MPI_C_BOOL with MPI_CXX_BOOL
2025-03-14 10:06:57 -07:00
Tzanio Kolev
41b65d6333
Merge branch 'master' into najlkin/parform-impro
2025-02-04 14:55:19 -08:00
Tzanio Kolev
e4c2152419
Update Copyright to 2010-2025
2025-02-03 12:17:26 -08:00
Jan Nikl
613d5dd826
Fixed constness in some HyperParMatrix constructors.
2025-01-08 17:56:42 -08:00
Andrew Ho
51a75893f7
Changes from Veselin
2024-12-09 08:21:57 -08:00
Andrew Ho
f1304374e0
Changed back to not use hypre's internal state tracker for calling init/finalize.
...
This will mean users can initialize hypre themself, and mfem won't
automatically finalize hypre for them in this situation.
2024-12-05 22:43:05 -08:00
Andrew Ho
79834fa003
use hypre's internal state manager when possible, allow re-initialization
2024-12-03 09:50:07 -08:00
Andrew Ho
25a1f0b3c6
Merge branch 'master' into master
2024-12-03 09:18:14 -08:00
adam-sim-dev
1f6abc2e0b
Merge branch 'master' into warning
2024-11-13 05:21:58 +08:00
Will Pazner
e6fb9d0ff9
Merge pull request #4548 from mfem/ParMatrixFrobeniusNorm
...
HypreParMatrix Frobenius Norm
2024-11-12 11:39:54 -08:00
adam-sim-dev
470698bb28
Merge branch 'master' into warning
2024-11-04 08:27:54 +08:00
Tucker Hartland
73efadce18
adding comment indicating that matrix data on host for the indicated hypre version
2024-11-01 14:47:40 -07:00
Tucker Hartland and Will Pazner
9dc7331d9c
Style update
...
Updating style as suggested by Will.
Co-authored-by: Will Pazner <11493037+pazner@users.noreply.github.com >
2024-11-01 10:25:04 -07:00
Tucker Hartland
d5d2cbba19
computing FNorm without making use of the expensive hypre_MergeDiagAndOffd call. Summing local Frobenius norms of diag/offd components before a single all-to-all communication.
2024-11-01 08:57:07 -07:00
Tucker Hartland
2464a9d1ca
adding a means to compute Frobenius norms when mfem is built with hypre version < 2.19.0
2024-10-31 15:21:55 -07:00
adam-sim-dev
b4dfb8660d
static_cast size_t to int
2024-10-31 18:07:45 +08:00
Will Pazner
6cedf73dd5
Small style adjustments to HypreParMatrix::FNorm
2024-10-30 09:21:01 -07:00
Tucker Hartland
e46a3dc294
NormFro --> FNorm for consistency with DenseMatrix::FNorm
2024-10-25 11:20:17 -07:00
Tucker Hartland
7f2ee0b7c9
minor update
2024-10-25 10:10:09 -07:00
Tucker Hartland
324a86d4e7
adding a NormFro method to the HypreParMatrix class which is a wrapper of the hypre function hypre_PARCSRMatrixNormFro
2024-10-25 09:56:02 -07:00
Veselin Dobrev
36f15e5784
Fix one more test failure when using HYPRE built with GPU + UVM
...
The fix uses a newly added constructor for HypreParVector that creates
an alias to existing Vector + offset.
Also, fix some doxygen comments.
This commit shouls also address issue #4351 (not tested yet).
2024-10-20 12:36:53 -07:00
Veselin Dobrev
5c1d4f24a3
Fix test failures when using HYPRE built with GPU + UVM
2024-10-17 18:48:39 -07:00
Andrew Ho
48c1be8fb5
Changes suggested by Will
...
- fixed accidentally copying the singleton
- deleted copy/move constructors to prevent future accidental copying
2024-09-27 14:39:30 -07:00
Andrew Ho
345957aae2
Have Hypre track if it was initialized in addition to finalized.
...
Moved where HYPRE_Init gets called.
2024-09-25 18:27:10 -07:00
Andrew Ho
f61629c5a5
Whitespace formatting from astyle
2024-09-13 18:02:16 -07:00
Andrew Ho
b3fff9267a
Moved Hypre::Finalize earlier in Device::~Device, only finalize HYPRE if it's been initialized.
...
Moving Hypre::Finalize earlier ensures that early return paths in
Device::~Device don't prevent Hypre::Finalize from executing.
2024-09-13 17:55:09 -07:00
Tom Stitt
a74663d634
Use HYPRE's SpMV since it is currently faster than rocSPARSE's
2024-06-04 10:51:23 -07:00
Tzanio Kolev
0406101e29
Merge pull request #4271 from mfem/symmatcoeff-project-fix
...
Fix `SymmetricMatrixCoefficient::ProjectSymmetric` bug
2024-05-22 14:03:43 -07:00
Veselin Dobrev
1bc5a0c5e4
Fix use-after-delete bug in one of the HypreParMatrix constructors
2024-05-21 21:53:12 -07:00
Will Pazner
8da512d5cf
Factor out hypre row and col start comparison logic
...
Create new static functions GetPartitioningArraySize and
RowAndColStartsAreEqual.
2024-05-16 14:46:20 -07:00
Will Pazner
8be0dee008
Use MPI_Allreduce in HypreParMatrix ctor to determine whether to reorder
...
This ensures the same decision is made across all ranks
2024-05-16 14:30:11 -07:00
Andrew Ho