forked from hypre-space/hypre
-
Notifications
You must be signed in to change notification settings - Fork 0
/
CHANGELOG
1202 lines (768 loc) · 42.1 KB
/
CHANGELOG
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
# Copyright 1998-2019 Lawrence Livermore National Security, LLC and other
# HYPRE Project Developers. See the top-level COPYRIGHT file for details.
#
# SPDX-License-Identifier: (Apache-2.0 OR MIT)
#=============================================================================
# This file chronicles user-level changes, beginning with the most recent.
# =============================================================================
Version 2.23.0 released 2021/10/01
- Added GPU support for multipass interpolation.
- Added new header file variables (HYPRE_DEVELOP_STRING, HYPRE_DEVELOP_NUMBER,
HYPRE_DEVELOP_BRANCH) to help users working with the development branch.
- Added rocsparse triangular solve required for Gauss-Seidel relaxation when
using hip and rocsparse on AMD GPUs.
- Various bug fixes.
#====================================
Version 2.22.1 released 2021/08/20
- Ported Neumann version of AIR to the GPU.
- Re-implemented ParCSR eigenvalue estimation used in Chebyshev relaxation.
- Added IJMatrix Add, Norm, and Transpose functions.
- Simplified the ParCSRMatrix/Vector code by removing ownership info.
- Added GPU support for Chebyshev relaxation.
- Various bug fixes.
#====================================
Version 2.22.0 released 2021/06/22
- Added GPU support for AMS, ADS, and AME solvers.
- Added mixedint support for GPUs.
- Fixed GPU issues with AMG-DD solver.
- Various bug fixes.
#====================================
Version 2.21.0 released 2021/05/25
- Changed GPU defaults and changed default CUDA arch to 70.
- Added new interfaces: HYPRE_SetSpGemmUseCusparse(1) to select cuSPARSE SpGeMM
or hypre's MM (if 0); HYPRE_SetUseGpuRand(1) to use cuRand for PMIS or CPU RNG
(if 0); HYPRE_SetMemoryLocation to select default memory location;
HYPRE_SetExecutionPolicy to select default execution policy.
- Changed configure option: --enable-cub --> --enable-device-memory-pool;
--enable-nvtx --> --enable-gpu-profiling
- Added new configure options: --with-cuda-home=DIR; --with-gpu-arch=ARG
- Added GPU build options to CMake.
- Added a new HYPRE_RELEASE_NUMBER macro.
- Improved support for ParCSR matrices with many zero rows.
- Improved BoomerAMG setup efficiency when there are many off-diagonal coefficients.
- Fixed a bug in the strength matrix generation for nodal AMG.
- Removed unified virtual memory (UVM) requirement from GPU implementation.
- Removed --enable-global-partition option.
- Added UMPIRE support for memory pooling on GPUs.
- Added HIP/AMD support.
- New CUDA based triangular smoothers.
- New parallel ILU solver features and GPU support.
- Added CUDA 11 support.
- Various bug fixes.
#====================================
Version 2.20.0 released 2020/09/24
- New matrix-based interpolation routines for AMG (CPU and GPU)
- Added GPU support for aggressive coarsening in AMG
- New AMG-DD solver
- Improved distributed sparse matrix-matrix and triple-matrix product
performance on GPUs
- IJMatrix/Vector assembly on GPUs (with pointers to GPU memory)
- Updated caliper usage
- Separated C and C++ headers
- Various bug fixes.
#====================================
Version 2.19.0 released 2020/05/26
- Updated to support superlu-dist version to 6.3.1
- Moved reference manual API to online documentation
- New AMG features to keep specified F-points and/or C-points.
- Added GPU support for AMG setup and several interpolation approaches.
- New parallel ILU solvers and smoothers.
- New MGR features.
- Added several interpolation routines based on matrix-matrix interpolations.
- Various bug fixes.
#====================================
Version 2.18.2 released 2019/10/28
- Fixed mixedint bugs.
#====================================
Version 2.18.1 released 2019/10/15
- Various bug fixes.
#====================================
Version 2.18.0 released 2019/09/27
- Ported BoomerAMG setup to run on GPUs. Only a subset of setup phase features
are currently available.
- Updated CMake to support more of the features available in configure/make.
- Added RecomputeResidual options to several hybrid solvers.
- Various bug fixes.
#====================================
Version 2.17.0 released 2019/07/10
- Changed open source license to dual MIT/Apache-2.0
- Removed FEI_mv/ml subdirectory and several FEI header files
- Various bug fixes.
#====================================
Version 2.16.0 released 2019/03/20
- Added new '--enable-mixedint' configure option. This introduces a HYPRE_BigInt
type that is set to 'long long int' (when configured with this option) to
enable the solution of linear systems that have more than 2 billion unknowns,
while reducing the memory use and computational costs needed with the previous
option '--enable-bigint' (still available). Note that this option currently
disables Euclid, ParaSails, pilut, and CGC coarsening in BoomerAMG.
- Various GPU optimizations.
- Changed documentation to use Sphinx and Doxygen. The User Manual is available
online through the hypre GitHub repo. The Reference Manual is not yet online.
- Various bug fixes.
#====================================
Version 2.15.1 released 2018/10/19
- Added hypre version functions: HYPRE_Version(), HYPRE_VersionNumber()
- Various bug fixes.
#====================================
Version 2.15.0 released 2018/09/21
- Added new system interface routines to reduce copies and improve threading:
SetBoxValues2() for Struct/SStruct and IJSetValues2().
- GPU-related changes: improved build process; fixed Raja and Kokkos ports of
Struct/SStruct; new boxloop reduction implementation without unified memory.
- Added a new GMRES variant COGMRES, with reduced communication, which is based
on classical Gram-Schmidt with a reorthogonalization option.
- Added functions HYPRE_ParVectorGetValues, HYPRE_ParCSRPCGGetResidual,
HYPRE_ParCSRGMRESGetResidual, HYPRE_ParCSRBiCGSTABGetResidual, and more.
- Various bug fixes.
#====================================
Version 2.14.0 released 2018/03/16
- New GPU release uses CUDA or OpenMP4.5. AMG setup phase is not yet ported.
- New AIR AMG solver.
- Added SuperLU_Dist as a coarse grid solver option.
- Various bug fixes.
#====================================
Version 2.13.0 released 2017/10/20
- Added new multigrid reduction (MGR) solver
- Added BoomerAMG feature SetCpointsToKeep() to force specific points to keep on
all coarse grids, up to a specified coarse level.
- Changed defaults in AMG Hybrid solver to match BoomerAMG defaults
- Updated hypre blas/lapack implementation
- Removed SuperLU/DSuperLU from the distribution and updated the FEI code to
work with external installations of SuperLU/SuperLU_DIST (version 5.2.1)
- Fixed a bug in Euclid
- Fixed Windows Visual Studio compile (version 2017)
#====================================
Version 2.12.1 released 2017/09/29
- Added support for single and quad precision floating point numbers
- Added weighted Jacobi (relax_type 7) to be usable on GPU in BoomerAMG
- Various bug fixes
#====================================
Version 2.12.0 released 2017/05/02
- Added GPU support to hypre. The Struct and SStruct code can use CUDA, RAJA,
or KOKKOS. The ParCSR code uses CUDA. The BoomerAMG setup phase is not yet
ported to the GPU. This release uses unified memory.
#====================================
Version 2.11.2 released 2017/03/13
- Changed the defaults in hypre to HMIS with ext+i(4) interpolation
- Added a routine HYPRE_BoomerAMGSetOldDefault to easily get old defaults
- Added Caliper instrumentation
- Various bug fixes
#====================================
Version 2.11.1 released 2016/06/09
- Fixed one more bug related to SStructSetSharedPart and SetNeighborPart
- Fixed a bug in PFMG red/black relaxation for 2D problems
- Fixed various other bugs in ParCSR and the FEI
- Fixed configure option --without-MLI and removed option --with-examples
- Modified code to enable compilation with a C++ compiler.
- Added several missing Fortran interface routines in Struct and BoomerAMG
- Updated the AME and LOBPCG interfaces to work correctly with both absolute and
relative tolerances.
#====================================
Version 2.11.0 released 2016/03/28
- Added a new function hypre_BoomerAMGSetKeepTranspose, which will store the
transposes of the interpolation operators and then replace the multiplication
of the transpose of P with a more efficient matvec.
- Added a new function hypre_BoomerAMGSetRAP2, which will replace the triple
matrix product R*A*P by two matrix-matrix products. This has shown to be
faster on BG/Q. Note that setting this function will prevent the keeping of
transposes.
- Added various changes that can improve performance on certain architectures,
particularly in the BoomerAMG Setup when using threading. While some of them
will be in effect automatically, using the configure options --with-openmp
--enable-hopscotch can lead to additional improvements if atomic operations
are available. The configure option --enable-persistent allows the use of
persistent communication if available. (Many of these changes are described in
"High-Performance Algebraic Multigrid Solver Optimized for Multi-Core Based
Distributed Parallel Systems" by J. Park, M. Smelyanskiy, U.M. Yang,
D. Mudigere, P. Dubey, published in Proceedings of SC15.)
- Modified configure to conform to xSDK standards in IDEAS project
(https://ideas-productivity.org/focus-areas/xsdk/)
- Fixed a bug related to SStructSetSharedPart and SetNeighborPart
- Fixed a bug in elasticity interpolation for BoomerAMG
- Fixed an MPI bug that was causing Parasails to hang occasionally
- Removed Babel from the release
#====================================
Version 2.10.1 released 2015/09/11
- Fixed various bugs in AMG and the IJ interface.
- Changed function names to set interpolation truncation parameters for
mult-additive versions as follows:
HYPRE_BoomerAMGSetAddTruncFactor to HYPRE_BoomerAMGSetMultAddTruncFactor
HYPRE_BoomerAMGSetAddPMaxElmts to HYPRE_BoomerAMGSetMultAddPMAxElmts
- Replaced the function HYPRE_BoomerAMGSetNonGalerkTol, which sets parameters
for the nonGalerkin version, with HYPRE_BoomerAMGSetNonGalerkinTol and
HYPRE_BommerAMGSetLevelNonGalerkinTol
#====================================
Version 2.10.0b released 2015/01/22
- Added new interpolation that allows the incorporation of near null-space
vectors for systems problems.
- Added additive V-cycle and new mult-additive and simplified mult-additive
V-cycle.
- Fixed several bugs in BoomerAMG and removed some inefficiencies.
- Fixed potential bug and memory leaks in the AMS and ADS solvers.
- Added threading to the IJ interface and some AMG setup routines.
- Added new non-Galerkin coarse grid construction scheme to reduce communication
and complexity.
- Extended GLVis visualization to all 2D and 3D example codes.
- Changed the configure settings to use --disable-global-partition by default.
The equivalent old option --with-no-global-partition is still available, but
users should migrate to the new one.
#====================================
Version 2.9.4a released 2013/11/21
- Minor bug fix in AMS and other minor changes.
#====================================
Version 2.9.3a released 2013/10/15
- Extended AMS to work for AMR problems on quad/hex meshes by removing an
assumption that the discrete gradient G only has +-1 entries. Also changed
default smoother in AMS to l1-GS.
- Added a hi-D capability to Struct and SStruct. Use the --enable-maxdim option
to configure the library for higher-than-3D usage.
- Added complex system support to a subset of the library, including most of the
matrix-vector class functions and the Krylov solvers. Currently does not
support the FEI or babel components. Use the --enable-complex option to set.
#====================================
Version 2.9.2a released 2013/02/15
- Added HYPRE_AMSSet*AMGCoarseRelaxType for specifying the coarsest level
relaxation for the AMG subspace solvers in AMS. Since the subspace matrices
are often singular, the default value uses l1-GS on the coarsest level, which
is different from BoomerAMG's default (exact solve).
- Fixed a bug in the IJ interface related to setting off-processor values when
hypre is configured to use the assumed partition.
- Fixed a bug in BoomerAMG when using C-F L1-Jacobi on coarsest grid, and fixed
a memory bug related to interpolation.
#====================================
Version 2.9.1a released 2013/01/16
- Added a HYPRE_BoomerAMGSetRedundant() switch to allow for an agglomeration
approach for solving the coarsest grid system.
- Changed how BoomerAMG handles zero diagonal entries.
- Made a few minor bug fixes in BoomerAMG.
- Fixed a configuration problem with external blas and lapack.
#====================================
Version 2.9.0b released 2012/10/30
- Changed the behavior of the BoomerAMG interpolation routines when a zero
diagonal is encountered to avoid divide-by-zero issues.
#====================================
Version 2.8.3a released 2012/09/28
- Fixed some memory allocation problems in SStruct, including code related to
the SetPeriodic() routine.
- Updated the CMake build system to build the FEI component of hypre also.
- Flattened the directory structure for the FEI code.
#====================================
Version 2.8.2a released 2012/05/10
- Added threading to multipass interpolation in BoomerAMG.
- Fixed a coarsening problem in PFMG when the diagonal is negative.
- Added a new option to BoomerAMG to specify a minimum coarse grid size.
#====================================
Version 2.8.1a released 2012/03/19
- For non-cell data, fixed a memory problem with SStructGraphAddEntries() and an
accumulation problem with SStructMatrixAddValues().
- Improved the penalty scaling factor for AMS's discrete divergence term (which is
added to the matrix to robustly handle problems with pure void regions).
- Added threading to various routines in BoomerAMG.
- Fixed the SStructGridSetPeriodic() routine to work for all object types,
including HYPRE_PARCSR which previously did not work.
- Made two changes to GMRES: 1) Added GMRESSetSkipRealResidualCheck to skip the
evaluation and check of the final residual in GMRES. This can be useful in
situations where restart is not expected to be beneficial. 2) Independent of the
above, added a check to see if consecutive real GMRES residuals decrease (in the
"false convergence" case). If not, we conclude that restart leads to pollution
from round-off errors and exit GMRES.
- Implemented a CMake build system for the library and the test drivers.
#====================================
Version 2.8.0b released 2011/11/14
- Added the new Auxiliary-space Divergence Solver (ADS) which targets div-div
problems discretized with lowest order Raviart-Thomas "face" finite elements.
ADS is very similar to AMS, and in general its relationship to AMS is
analogous to the relationship between AMS and AMG.
- Added a redundant coarse-grid solve option to BoomerAMG, which can improve
solution times on machines with high communication latencies.
- Extended the AMS and ADS solvers to support (arbitrary) high-order H(curl) and
H(div) discretization methods. The high-order interface allows the users to
set directly the Nedelec and Raviart-Thomas interpolation matrices, which can
be used to precondition other types of discretizations (e.g. ones based on the
second family of Nedelec elements).
- Added a Euclid preconditioner option to the Fortran interfaces for the ParCSR
Krylov solvers.
- Fixed a bug with periodic problems in the SStruct interface.
- Fixed a bug in CGNR.
- Added a new higher order FEM example (ex16).
- Added visualization capabilities to the examples based on the GLVis package.
#====================================
Version 2.7.1a released 2011/03/18
- Fixed a long-standing O(P) memory allocation problem in the communication
setup routines for Struct.
- Fixed some OpenMP threading problems in Struct.
- Made some optimizations to the constant coefficient version of PFMG.
- Added a 64-bit version of Example 15 (file 'ex15big.c' in examples/)
#====================================
Version 2.7.0b released 2011/01/27
- Added a new --enable-bigint feature that, when turned on, requires the user to
pass integers of size 'long long int' to hypre and enables ParCSR matrices with
more than 2 billion unknowns (previously a limitation due to 32-bit integers).
This feature is currently not available in the FEI or babel sections of hypre.
Added example 'ex5big.c' as an illustration of how to use this new feature.
- Changed the Fortran interface to use a Fortran MPI communicator. Previously,
users were required to call HYPRE_MPI_Comm_f2c() to convert their Fortran
communicator to a C communicator. Fortran users will need to remove this call
from their code to use the new release.
- Added a new version of the l1-SGS smoother in BoomerAMG as a replacement for
relax type 8. This smoother behaves similarly to relax type 6 on matrices
with relatively small off-processor entries, but is guaranteed to always be
convergent. The new smoother is also available in AMS as relax type 4.
- Fixed some bugs in the coarsening routines for BoomerAMG.
- Relaxed the power-of-two restrictions for the PFMG and SysPFMG solvers. See
the reference manual for more details. SMG still requires a power-of-two.
- Fixed the constant-coefficient PFMG solvers. Also changed the non-Galerkin
scheme slightly for the constant-coefficient variable-diagonal PFMG solver to
simplify the code and enable easier threading.
- Fixed a bug in the Struct multigrid solvers for periodic problems when the
period is larger than the grid.
- Fixed some OpenMP problems in the Struct and SStruct code.
#====================================
Version 2.6.1a released 2010/03/30
- Made several bug fixes related to the new SetSharedPart() and FEM() routines:
Fixed a memory leak. Fixed to allow SetSharedPart() or SetNeighborPart()
calls from any processor, even if the processor doesn't own data on 'part'.
- Fixed a bug in the void-zone version of AMS, which sometimes produced a
non-symmetric preconditioner.
- Added new smoothers to BoomerAMG (Chebyshev, FCF-Jacobi, and CG) and
AMS (Chebyshev).
- Made some performance improvements to the implementation of the assumed
partition algorithm used in SStruct.
#====================================
Version 2.6.0b released 2010/02/04
- Added a new interface routine to SStruct called GridSetSharedPart() which is
used to describe how regions inside a part are shared with regions in other
parts in a block-structured grid setting. The routine is similar to
SetNeighborPart(), but allows more flexibility in finite-element settings.
- Added new finite element routines to the SStruct interface as an alternative
to using stencils: GridSetFEMOrdering(), GraphSetFEM(), GraphSetFEMSparsity(),
MatrixAddFEMValues(), and VectorAddFEMValues().
- Added new examples, ex13.c, ex14.c, and ex15.c, to illustrate the use of the
new SetSharedPart and FEM routines, the new rectangular matrix feature in
SStruct, and the use of the AMS solver through the SStruct interface.
- Removed support for SetNeighborBox() in the SStruct interface. This was
replaced by SetNeighborPart() in release 2.4.0b.
- Added the assumed partition algorithm to the implementation of the SStruct
interface to improve scaling at huge processor counts.
- Changed the behavior of IJMatrixSetValues so that it is deterministic and
consistent with the behavior of SStructMatrixSetValues when setting values
owned by other processors (a similar change was made to IJVectorSetValues).
- Added user documentation for building hypre under Windows.
- Added user and reference documentation for the LOBPCG eigensolver.
- Did some cleaning up in the 'configure' script.
- Changed to install only one hypre library now: libHYPRE.a (or libHYPRE.so).
- Added the release number to the hypre shared library soname and changed to
resolve all symbols in the shared library.
- Updated the non-babel fortran interface.
- Added configure options to explicitly set the mangling scheme for Fortran.
- Added a configure option to --disable-fortran for users who do not have a
working fortran compiler. This is most useful for Mac users.
- Improved threading for BoomerAMG.
#====================================
Version 2.5.0a released, 2009/10/12.
- Added new 2-stage interpolation operators to be used with aggressive
coarsening.
- Added a new routine HYPRE_BoomerAMGSetMaxCoarseSize that allows to set the
maximal system size on the coarsest level.
- Added a "recompute residual" option to PCG which makes it behave like a
"restarted CG". This, can be useful in high-accuracy computation where the
CG residuals drift from r = b - Ax.
- Added a residual-based stopping criteria in PCG for high-accuracy
computations.
- Added a Fortran interface for AMS.
- Initial release of a more robust version of AMS for singular problems with
zero-conductivity regions. This requires a list of the interior nodes in the
zero conductivity regions and uses a periodic projection onto a compatible
subspace.
- Added Distributed SuperLU in the FEI.
- Added support for multiple right-hand-sides in the FEI.
- Added C and Fortran nodal versions of Example 1: ex12.c and ex12f.f.
- Added a LOBPCG example: /examples/ex11.c.
- Removed the make uninstall option.
- The HTML version of the reference manual is no longer part of the hypre
distribution. However, it can still be generated by typing make in the docs/
directory (requires doc++).
#====================================
Version 2.4.0b released 2008/08/08
- Fixed some Fortran interface bugs.
- Added two additional Krylov solvers: Flexible GMRES and LGMRES
- Added a fortran example: /examples/ex5f.f
- Updated SuperLU to version 3.0
- Improved relaxation options in AMS.
- Changed 'make clean' to erase the hypre/ directory.
- Fixed a bug in preconditioned BiCGStab.
#====================================
Version 2.3.2a released, 2008/05/08.
- Fixed a couple of bugs associated with SStructGridSetNeighborPart.
- Added absolute convergence tolerance stopping criteria to GMRES, CG, and
BiCGStab (See HYPRE_GMRESSetAbsoluteTolerance, HYPRE_PCGSetAbsoluteTolerance,
or HYPRE_BiCGSTABSetAbsoluteTolerance in the reference manual for details).
- Updated error checking for CG and BiCGStab.
- Fixed a bug in AMG when the maximum number of levels is set to 1.
- Added two new solver types to AMS: a Hiptmair-like smoother (type 0), and
one based on a block 4x4 coarse-grid matrix (type 20).
#====================================
Version 2.3.1a released, 2008/01/31.
- A bug fix (uninitialized variables introduced when AMS was added to
the FEI preconditioner list).
- Fixed a memory allocation bug in new SStruct SetNeighborPart feature.
- Added a new option for the Schwarz smoothers
(HYPRE_BoomerAMGSetSchwarzUseNonSymm and HYPRE_SchwarzSetNonSymm) to use LU
instead of Cholesky LAPACK functions to invert the matrix blocks.
- Fixed a bug with the Schwarz smoothers and the ESSL compile.
#====================================
Version 2.3.0a released, 2008/01/28.
- Completed the implementation of SStructGridSetNeighborBox to work for all
variable types (not just cell-centered). In the process, we discovered that
the interface design for this routine can result in ambiguous coordinate
transformations between parts, so the new SetNeighborPart routine (below)
should be used instead.
- Introduced the SStructGridSetNeighborPart routine. Users migrating from
SetNeighborBox to this new routine only need to change the name ("Box" to
"Part") and append an additional argument 'index_dir'. See the reference
manual for details.
- The SStructSplit solver can now be used as a standalone solver (before, it
could only be used as a preconditioner). Added HYPRE_Jacobi as an option to
SetStructSolver.
- Added Get(Box)Values routines for Struct/SStruct Matrix.
- Added nodal interpolation schemes to BoomerAMG for systems problems.
- Added capability to change Euclid runtime parameters using individual
functions.
- Improved use of Euclid as smoother in BoomerAMG.
- Fixed two potential bugs in BoomerAMG.
- Fixed a major bug with the OpenMP code in hypre (beginning with v1.7.5b,
OpenMP support was effectively disabled in the code). Also made several
additional OpenMP changes/fixes.
#====================================
Version 2.2.0b released, 2007/09/20.
- Renovate the FEI block preconditioner which will be useful for ALE3D
simulation of mixed variable type problem
#====================================
Version 2.1.0a released, 2007/06/25.
- Added the CGC and CGC-E coarsening algorithms, which have been developed
and implemented by Michael Griebel, Bram Metsch and Alex Schweitzer of
the University of Bonn.
- Added several new solver types to AMS and improved the AMS/FEI integration.
- Better handling of the relative change test in GMRES.
- Improved weights for Struct and SStruct weighted Jacobi.
- Added a 'make check' target that does a simple code verification test.
- Added truncation for interpolation according to the number of elements to the
ParCSR Hybrid AMG Solver.
- Renovated the hypre fei implementation for use in ARES.
#====================================
Version 2.0.0 released, 2006/12/15.
- Minor documentation updates.
#====================================
Version 1.14.0b released, 2006/11/30.
- Added python support to Babel interface.
- Changed the name of several internal header files to use the '_hypre_' prefix.
This should not affect most users, since these are internal headers.
- Added AME, a new Maxwell eigensolver based on AMS and LOBPCG.
- Changes to Struct code to hopefully improve efficiency of MG solver setup.
- Efficiency improvements for new long-range interpolation algorithms in
BoomerAMG, including:
* extended interpolation is now like modified, i.e no absolute values
* added an extended interpolation which only extends when there are no
common coarse neighbors
* changed standard interpolation to only extend at strong F neighbors
(instead of all neighbors) for better efficiency
#====================================
Version 1.13.0b released, 2006/10/03.
- Babel-based interface updated to Babel version 1.0.0, plus various
improvements, such as an interface to the hypre error handler. Pre-built
interfaces are included for C, Fortran, and C++.
- Various modifications to BoomerAMG:
* Eliminated the possibility to set relaxation on the finest grid
* Added various new long-range interpolation algorithms: standard (with and
without weight separation), extended, and F-F interpolation. These routines
will improve convergence for the low complexity coarsening algorithms PMIS
and HMIS.
* Added the possibility to set a maximal number of weights per row <n> when
truncating the interpolation operator. When this option is used, only the
absolutely largest <n> weights per row are kept.
- Modifications to the fac user interface to permit easier problem set up.
- Shared libraries can be built by using the --enable-shared option when
configuring. NOTE: the directory where the shared libraries reside MUST be
defined in the LD_LIBRARY_PATH variable.
- Incorporated the GNU Lesser General Public License.
- Now using Roundup for issue tracking; send email to [email protected]
#====================================
Version 1.12.0a released, 2006/07/26.
- First release of AMS, a new solver for (semi)definite Maxwell problems.
- First release of Maxwell, a semi-structured Maxwell solver.
- Modifications to the fac user interface to permit easier set up of the problem.
#====================================
Version 1.11.1b released, 2006/05/30.
- Corrected FEI build to allow user to specify a system FEI and build internal
HYPRE routines.
#====================================
Version 1.11.0b released, 2006/04/24.
- Added configure option --without-FEI to disable build of internal FEI
routines.
- Added a new system for handling error codes in hypre.
- Added configure option --print-errors to enable printing of HYPRE error
messages.
- Modified configure to only depend on having a C compiler available.
- Added compatible relaxation on one processor.
- Updated all F90_HYPRE_* files to be consistent with HYPRE_* files.
- Updated ex5.c to include:
* an option to print the system matrix and rhs
* a description of how to read in a matrix and rhs in IJ format.
- Corrected error that occurred when more than one Euclid solver object is used
at the same time
- Added Maxwell capabilities via Sandia's ML solvers; see details in
FEI_mv/fei-hypre/CHANGELOG.
- Replaced PILUT with Euclid in test code sstruct.c.
#====================================
Version 1.10.0b released, 2005/12/19.
- configure updates
* always runs --without-blas --without-lapack unless otherwise specified
* cross-compiling enabled
* added --with-no-global-partition option
- makefile updates
* changed eigensolvers subdirectory to multivector
- Added a set of simple example codes covering the Struct, SStruct and IJ
interfaces. See README.html in the examples subdirectory for more details.
- The --with-no-global-partition option was added to reduce storage and improve
run times on very large numbers of processors (such as BlueGene/L). Currently
only the IJ interface solvers have been optimized. For small or modest
numbers of processors (typically < 1000-5000 depending on the machine), this
option is not needed (and will likely be slower).
- Added aggressive coarsening, which can be applied to all available coarsening
schemes (via HYPRE_BoomerAMGSetNumAggLevels and HYPRE_BoomerAMGSetNumPaths)
- Added multipass interpolation and direct interpolation.
- Added the capability of nodal coarsening for systems AMG.
- expanded coverage of Babel-based interface
#====================================
Version 1.9.0b released, 2005/02/10.
- Added updated LOBPCG source code and test files.
- Added a vector Laplace problem for testing systems AMG.
- Added FAC to semi-struct
#====================================
Version 1.9.0a released, 2005/01/18.
- NOTE: Link lines may need to be changed depending on configure options used.
- The build system introduced in v1.8.3a was too difficult to maintain, so we
have reverted to a system similar to the previous one. In particular, all of
the new automake and libtool stuff was removed. In the new system, each
directory (except 'config') has ONLY a 'Makefile' (no more 'Makefile.in' or
'Makefile.am' files). When the 'configure' script is run, the single file
'config/Makefile.config' is created. This file is included in all Makefiles
in all sub-directories. Shared libraries CANNOT be created with this system.
- The makefile targets are now more inline with GNU standards. To get details,
type 'make help'.
- The makefile target 'nofei' has been removed and is now run by default.
- The library files 'libHYPRE_blas.a' and 'libHYPRE_lapack.a' are no longer
created when 'configure' is run '--without-blas' and/or '--without-lapack'.
This created confusion before and really didn't make sense anyway. These
options are now consistent with the '--without-MPI' option. Also, there is
no longer any danger of name conflicts with external blas or lapack libraries
as was the case before.
- Added new parallel coarsening schemes PMIS and HMIS with reduced complexities
as well as the option of using one pass of Ruge-Stueben only.
- Added the capability to set values from any processor through the IJ matrix
and vector interfaces.
- Added support to the PFMG solver for matrices with constant coefficients.
- Reverted back to the first approach for setting periodic conditions. The
periodic flag now contains the periodicity value, and is not just a boolean.
#====================================
Version 1.8.3a released, 2004/01/21.
- The build system has been revamped to enable the building of shared
libraries (in addition to static libraries, as before). The new
system more closely follows GNU standards, and will hopefully show
improved flexibility and platform support. Users can build hypre
as before by typing 'configure' followed by 'make'. A few makefile
target changes to note are:
* 'make clean' replaces 'make veryclean'
* 'make all' no longer builds the drivers in the 'test' directory.
If needed, build them from within the 'test' directory or
use the 'make test' target.
Technical notes: We are now using GNU automake-1.8.2 to support
Makefile.in generation; autoconf-2.59 to generate configure; and
libtool-1.5 to provide shared library support. The GNU tools,
autoheader, libtoolize, and aclocal are also used. These tools are
only used by hypre developers and are *not* needed to build hypre.
- Separated the lapack and blas routines into two libraries. Blas and
lapack support (whether to use the routines distributed with hypre or
use already installed libraries) is determined by configure, and can
be overridden with '--with-blas=<lib>' and '--with-lapack=<lib>'.
NOTE: This change affects your link line!
- A new directory 'lib' contains a new Hypre library, 'libHYPRE.a' (and
'libHYPRE.so' if configured for shared libraries) which contain
all of the libHYPRE_*.a objects with in a single library.
- Fixed the PFMG and SMG solvers to work in the case of pure Neumann
boundary conditions.
#====================================
Version 1.8.2b released, 2003/10/22.
- Performance improvements for Struct solvers when there are many
boxes per processor in the grid. For 125 boxes or more in 3D,
the speedup in the setup phases of PFMG and SMG can be significant
(we have seen as much as a factor of 30).
- New non-Galerkin option in PFMG can improve convergence and memory
usage. This option maintains either a 5-pt stencil in 2D or a 7-pt
stencil in 3D on all grid levels.
- New non-symmetric red/black smoother in PFMG can improve
convergence on some problems.
- New optimizations in the smoothers in PFMG lead to as much as 30%
faster times.
- Added a new interpolation option to AMG that is useful for solving
hyperbolic equations. It treats fine weak connections like fine
strong connections.
#====================================
Version 1.8.1b released, 2003/05/21.
- Minor bug fixes and performance improvements.
#====================================
Version 1.8.0b released, 2003/04/08.
- New Babelized HYPRE interface introduced. Babel is a tool that
provides language interoperability. HYPRE will eventually migrate
to using Babel exclusively. See 'docs/babel_transition_info.txt' for
information on how to use the new interface.
- PFMG: bug fix for red/black relaxation.
- Eigenvalue solver LOBPCG has been added. See 'docs/lobpcg_info.txt'
for documentation.
#====================================
Version 1.7.11a released, 2003/03/14.
- BiCGSTAB has been added to struct and sstruct interface
- added automatic determination of relaxation parameters for use with symmetric
smoothers in BoomerAMG
#====================================
Version 1.7.10b released, 2002/12/02.
- New Hybrid solvers available for ParCSR and Struct.
- IJ: Modified the 'GetValues' part of the matrix interface to make
it possible to get matrix values without knowing the nonzero
structure. In particular, added a new option to IJMatrixGetValues,
and added new routines IJMatrixGetRowCounts, IJMatrixGetLocalRange,
and IJVectorGetLocalRange.
- SStruct: Fixed a bug when connecting a part to itself with SetNeighborBox.
- FEI: Added a few more smoothers and amg methods in mli, fixed a few memory
leaks uncovered by running ale3d, added new aggregation coarsening for
handling multiple materials. For more details, see
FEI_mv/fei-hypre/CHANGELOG.