forked from AMReX-Codes/amrex
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathCHANGES
1216 lines (731 loc) · 37.2 KB
/
CHANGES
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
# 21.05
-- Support OpenMP offload with the NVHPC compiler in GNU make (#1980)
-- In GNU make, use -MMD -MP instead of -MM to generate dependency if
supported (#1967)
-- Workaround compile error on Apple M1 (ARM64) (#1969)
-- Changes to EB slopes (#1965)
-- In GNU make, use `LOG_BUILD_TIME=TRUE` to log build time. (#1963)
-- Using rocPRIM in HIP Version of Scan (#1962)
-- Reimplementation of ReduceOps (#1955) The new approach does reduction
in two passes and therefore does not require atomics.
-- CMake: implement logic to support compiler ID IntelLLVM (new in CMake
3.20) (#1959)
-- CPU version of TagBox::buffer (#1957)
-- Fix out of bound error in AsyncWrite (#1753)
-- Use pinned memory ad buffer in Fab data I/O (#1956)
-- Allow user to opt out of renaming MTMPI suffix (#1954)
-- Port FluxRegister::ClearInternalBorders to GPU (#1952)
-- Skip temporary in ParticleToMesh when it is not needed. (#1949)
-- Multilevel version of ParticleToMesh (#1945)
-- Use CUB for block reduction (#1943)
-- Reimplementation of PrefixSum with cub (#1942)
-- Cell-Centered Overset Solver Coarsening (#1915)
-- Hypre Nodal Setup (#1906) Allow CoarseningStrategy::Sigma in Hypre
nodal solver and port setup to GPU.
-- CMake: add support for BOUND_CHECK (#1934)
-- Alias HostVector to PinnedVector and update documentation. (#1931)
-- Make CNS work on CPU too (#1930)
-- std::isfinite/inf for SYCL (#1929)
-- Update RNG for latest oneAPI (#1928)
-- Use pinned memory for data transfer between host and device (#1923)
# 21.04
-- ParmParse parameter, nodal_proj.normalization_threshold, for setting
the normalization threshold of the bottom linear solver. (#1917)
The default value of the threshold has been changed from 1e-10 to
1e-8. (#1914)
-- Fix a bug in BaseFab::resize. (#1916) This caused performance issues
on systems with managed memory and crashes if not.
-- Option to not zero out input multifab in amrex::ParticleToMesh (#1910)
-- CMake: add test_install (#1894)
-- Reimplement increment and incrementwithtotal for particles to take
advantage of GPUs and OpenMP (#1899)
-- Return const& in Particle rdata and idata and disallow returning copies
for rvalues. (#1895)
-- Update ALaplacian for multi-component and reuse (#1891)
-- CMake: IPO/LTO (#1890)
-- Make FabArray::setFab safer (#1884) Instead of taking a raw pointer, it
takes std::unique_ptr or rvalue FAB, so that the ownership is very
clear. It also deletes the old FAB if not null.
-- Flags: RelWithDebInfo (#1879) Add warning control to CMake Build mode
`RelWithDebugInfo` (currently default for WarpX).
-- Build: Windows with Shared Libs (.dll) (#1847)
-- Remove tutorials documentation from main repo; it now lives at
https://github.com/AMReX-Codes/amrex-tutorials (#1877)
-- 2D Poisson and ALaplacian Solvers in 3D Build of AMReX (#1805) New
capability of solving 2D cell-centered Poisson's and ALaplacian
equation with 3D build of amrex.
-- Add FabArray::release and Make BaseFab virtual (#1873)
-- Robin BC in Linear Solver (#1848)
-- LevelBld (#1852) This is a breaking change for Amr/AmrLevel based
codes.
-- Optimization of CArena::freeUnused (#1865)
-- CI for macOS: Build a shared library. This is the default in
Spack. (#1836)
-- Remove tabs and trailing white spaces (#1860)
-- Free unused memory (#1861) If Arena is asked to allocate more than
what's available in the system, try to free unused memory first.
-- Fortran Interfaces for new FillPatch (#1793)
-- Arena Release Threshold (#1857)
-- Feature/non local parallel copy (#1842) This PR introduces "Multiblock"
capabilities to AMReX.
-- AMREX_NO_PROBINIT (#1846) For Amr/AmrLevel based codes, there is now a
compile time option AMREX_NO_PROBINIT=[TRUE|FALSE] for the support of
the probin file and amrex_probinit function.
-- Add EditorConfig (#1841)
-- Fix ncomp bug in linear solver (#1832)
-- CMake: AMReX_INSTALL (#1831) Add an `AMReX_INSTALL` option that
controls `install()` generation.
# 21.03
-- Support multiple components in the ABecLaplacian solver. (#1825)
-- Add query functions to Arena. (#1823) They can be used to find out the
memory type and if the memory is accessible on device or host.
-- CMake: add -Wno-pass-failed to Clang-based compilers (#1815)
-- Option to cover multiple cuts in EB generation (#1810)
-- The_Async_Arena and Elixir::append (#1804)
-- Add function name output option to TinyProfiler. (#1803)
-- DPCPP: Use codeplay_host_task (#1797)
-- Add ParticleArray classes (#1796)
-- ParallelCopy_nowait & ParallelCopy_finish (#1765)
-- Add FabArray::tags() to return the tags. (#1794)
-- Bump minimum C++ standard from 11 to 14. (#1787)
-- Update the preferred short name of NVIDIA HPC SDK to nvhpc (#1788)
-- Fix precision issue in ParmParse::add (#1783)
-- Overset Solver with Refinement Ratio of 4 (#1778)
-- Refinement Ratio of 4 Support in Nodal Solver (#1774)
-- MSVC: Proper __cplusplus macro (#1773)
# 21.02
-- Abort when MLMG is detected as failing regardless of verbosity
setting. (#1762)
-- Update DPC++ version of Anyof (#1757)
-- Use ParticleReal for particle interpolation (#1763)
-- Remove obsolete functions from AMReX particle classes (#1754, #1755)
-- Make ParticleContainer::Geom always refer to the particle one. (#1470)
-- Gpu Docs Update (#1758)
-- Add IntVect versions of numParticlesOutOfRange. (#1751)
-- Improve the interpolation from cell centroids to face centroids:
(#1750)
-- In ParticleToMesh and MeshToParticle, use nGrowVect() to accommodate
different a different number of grow cells in each dimension. (#1748)
-- HIP: New Warp Size Macro (#1742) Use new the HIP macro for the warp
size on various AMD devices.
-- Add allocator template parameter to ParticleContainer (#1416)
-- Add ability to calculate slopes on face centroids (#1707) And apply the
explicit stencil when computing the laplacian.
-- HIP indirect function and kernel fusing (#1739) Add a GNU Make flag,
HIP_INDIRECT_FUNCTION=[TRUE|FALSE], and modify the Gpu kernel fusing
test.
-- When using CMake 3.15+ allow for MSVC runtime library changes (#1736)
-- RoundRobin: option to not sort processes (#1721)
-- EB: edge centroid (#1730) Add edge centroid to the database and make it
available via EBFArrayBoxFactory.
-- Fix RegridOnly to regrid even with single level grids (#1722)
-- Define: _OPENMP -> AMREX_USE_OMP (#1560) Replace the define `_OPENMP`
with `AMREX_USE_OMP` for all parallel "backend" implementations and
control of MFIter loops.
-- Update Make.nrel to accomodate MPT MPI with CUDA enabled (#1726)
-- Cell-centered Hypre and Petsc setup on GPU (#1709)
-- ParmParse::hasUnusedInputs & getUnusedInputs (#1716) Add functions to
find if there are unused `ParmParse` parameters starting with
`prefix.`. If `prefix` is empty, all unused `ParmParse` parameters are
included.
-- CMake: offer option to exclude Src/Amr from build (#1714)
-- Make bisect more robust (#1708)
-- Support single precision build with configure (#1706) Add new arguments
`--single-precision` and `--single-precision-particles` to `configure`.
-- Support when CMake 3.18 policy 105 is set to new (#1696)
-- Optimized function for filling BaseFab of GpuArray (#1692)
-- BaseFab::minmax (#1691) Add BaseFab::minmax function and use it in
VisMF.
-- New feature in the fextract tool to directly export in the csv format
(#1634)
# 21.01
-- Optimization of MLCellLinOp::applyBC (#1690)
-- Optimization of iMultiFab sum (#1684)
-- Sanitize the input of mkconfig (#1685)
-- New versions of Reduce::Min, Max and MinMax taking callable (#1679)
-- Only issue prefetch if arena uses managed memory (#1678)
-- GNU Make: Fix the library flags (#1641)
-- PolymorphicArray4, a thin layer around Array4 allowing us to access
both AoS and SoA data with operator() (#1643)
-- CMake: dowstream projects must enable Fortran if AMReX_FORTRAN=ON
(#1669)
-- Optimization of FaceLinear interpolater (#1663)
-- Allow alternate form of "check_pair" function (#1667)
-- ParallelFor for Reduction (#1658)
-- Add features to fextract (#1657)
-- Fix parallel add bug when both USE_GPU and USE_OMP are defined (#1654)
-- Write a plotfile to ASCII (#1638)
-- Fix NodalProjector::getGradPhi for multilevel case (#1642)
-- Remove GPU pragma support (#1650) (#1649)
-- Remove custom MFIter reduction code (#1648)
-- Remove support for old versions of DPCPP (#1644)
-- Fused FillBoundary (#1570)
-- Divergence Free FillPatch (#1483)
-- Optimization of constant coefficient MacProjector (#1636)
-- Optimization of nodal solver with constant coefficient (#1629)
-- Expose the index of the neighboring particle to users of the
iterator. (#1628)
-- Implement overset mask for MLPoisson too (#1604)
-- Fuse nodal jacobi smoother for GPU (#1625)
-- Support for SoA data for neighbor particles. (#1590)
-- Fix a bug that causes errors when writing HDF5 plotfile and particle
(#1621)
-- Fortran Inteface: Porting more MultiFab functions (#1609)
-- Remove alternative operators from AMReX (#1599)
-- EB from STL (#1592)
-- AMReX_Config.H generated at built time is included in all
headers. (#1566)
-- MacProjector: Update setDomainBC API (#1589)
-- Reimplement NumParticlesAtLevel (#1581)
-- Add msvc cuda support (#1573)
# 20.12
-- EB2::PolyIF (#1567) The existing PolynomialIF contains a dynamic size
vector. This makes it hard to support it on GPU. This new PolyIF
class uses fixed size array and has GPU support.
-- Tools/Plotfile works for for single precision now. (#1577)
-- Arena: Add accessor to query Arena attributes through info object
(#1575)
-- Allow MacProjection reuse (#1574)
-- Remove SUNDIALS Fortran interface support (#1559)
-- Allow MLEBABecLap to solve for multiple components (#1557)
-- DPCPP: MPI (#1556) Allow MPI for dpcpp build in gnu make assuming
mpiicpx is used.
-- Non-local boundary conditions (#1544)
-- If TheZeroVector() is passed in to SortParticlesByBin, do nothing
instead of crashing. (#1564)
-- Scan size type (#1521) Make the size type a template parameter to
support the scan of more than INT_MAX elements.
-- DPCPP: Fortran (#1554) Switch from gfortran to ifx that comes with
oneapi in GNU make.
-- CMake: generate_buildinfo target (#1545) This commit generates a static
library `buildInfo::<target>` for the passed user target. This target
is then automatically linked. Also, the generated `.cpp` file is now
scoped properly, avoiding collisions.
-- nvcc --Werror cross-execution-space-call (#1540) This is turned on by
default in GNU make, and can be turned on in CMake.
-- fcompare: Add option for an absolute tolerance check (#1537)
-- Remove call to cudaDeviceSetCacheConfig (#1536) No longer set CUDA
device cache configuration to prefer L1 cache.
-- Assert no nested MFIters (#1530) Add assertion to catch nested MFIters
(e.g., MultiFab functions are called inside MFIter).
-- Free MPI_Datatype defined by AMReX (#1527) This fixes a one-time memory
leak in MPI_Datatypes defined by AMReX as reported in #1525.
-- make clean & make cleanconfig (#1520) `make clean` now does what `make
realclean` does. `make realclean` is kept. `make cleanconfig` is
introduced to do what `make clean` used to do. For most users, `make
clean` is the one that should be used.
-- hypre: Adjust singular matrices before hypre solves (#1519)
-- CMake: overhaul options (#1490)
-- New Reduce::Sum function taking lambda function (#1508)
-- GNU Make: AMREX_CUDA_ARCH (#1507) Option to use AMREX_CUDA_ARCH in
addition to CUDA_ARCH to set CUDA arch. If AMREX_CUDA is set, it will
be used over CUDA_ARCH.
-- Add option for device sync at beginning and end of TinyProfiler region
(#1505) Now when setting
`tiny_profiler.device_synchronize_on_region_end = 1` in the inputs
file, we will synchronize before calling nvtxRangePop() and
nvtxRangePush(), which means that TINY_PROFILE regions will include the
full kernel time, rather than just the kernel launch time.
-- CMake; bump up hypre minimi required version (#1503)
# 20.11
-- The default hypre interface in MLMG is now IJ matrix for both EB and
non-EB. Previously, it was semi-struct for non-EB. (#1492)
-- USE_SAVE_TEMPS option in GNU Make system for HIP. (#1492)
-- The device version of `Error`, `Abort` and `Warning` is no-op, unless
`NDEBUG` is defined. (#1492)
-- Ascent: Use Default Name for Mesh Ghost (#1488)
-- Add fill method to GpuArray (#1472)
-- Add Gpu::Atomic::AddNoRet (#1469)
-- Option to make the code NOT enforce solvability in MLMG for singular
problems. (#1471)
-- CMake: improve HDF5 support (#1468)
-- CMake: FindPETSc must not overwrite CMAKE_Fortran_FLAGS (#1464)
-- CMake: prevent in-source builds (#1453)
-- change char * to char const* so SWFFT code compiles with USE_CUDA=TRUE
(#1454)
-- Fix when small_volfrac is used -- we need to adjust the data in cut
cell neighbors as well as regular neighbors. (#1451)
-- Allow users to set a Geometry object for ParticleContainer
independently of the AmrCore / AmrLevel object. (#1446)
-- HDF5 plotfile write bug fix and performance improvement (#1448)
-- CMake: add HIP support (#1316)
-- NodalProjector: use volume-weighted average down (#1444)
-- Fix race conditions in EB interpolation from cell centers to faces
(#1443)
-- Allow tagging value to vary by level (#1441)
-- Hypre with overset (#1439)
-- Hypre IJ interface: Enable access to additional solvers and
preconditioners available in Hypre (#1437)
-- Fix CPU version of uninitailiedFillNImpl in PODVector (#1435)
-- CMake: Fix missing compile time definitions for HYPRE/PETSc (#1436)
-- Fix bug in EB extdir slopes (#1434)
-- Make the id and cpu members of amrex_particle private, as they should
no longer be accessed directly. (#1433)
-- Disable OpenMPI C++ binding in GNU make system (#1398)
-- Fix a long standing bug in eb levelset (#1432)
# 20.10
-- CMake: fix ENABLE_PROFPARSER and ENABLE_SENSEI options (#1428)
-- Remove EB LSCore and levelset, and reimplement signed distance function
capability. (#1429 & #1425)
-- Remove flags setCGVerbose and setCGMaxIter since those are more
accurately called setBottomVerbose and setBottomMaxIter. The
"setCG..." flags are misleading since they actually apply to non-CG
bottom solvers as well. (#1413)
-- Remove GetPosition and SetPosition from ParIter. These were added for
WarpX, but they are not used any more. (#1408)
-- Only check the requested MPI Threading level if AMReX also initialized
MPI. (#1406)
-- DPCPP support for mulitple GPUs (#1392)
-- Random Number Generation on Device (#1363) In order to support RNG on
device with DPC++, we have to change the API because DPC++ does not
support global device variables.
-- Remove redundant particle tests and old/obsolete/non-test tests (#1389
& #1381)
-- Gpu kernel fusing is integrated in MFIter. (#1332)
-- New runtime parameter, amrex.max_gpu_streams. (#1386)
-- Switch to `use mpi` from `include 'mpif.h'` (#1385)
-- Remove reliance on managed memory from
AMReX_ParticleCommunication.H/.cpp (#1380)
-- CMake: re-organize tutorials (#1333)
-- Implement Particle in a way that does not involve UB. (#1337)
-- amrex::EB_average_down(): Fix indexing error. (#1360)
-- Only add one copy of the ghost particle regardless of how many isects
we have (#1359)
-- Add non-EB slopes and extend the generality of EB slope routines in 2D
and 3D (#1350)
-- Async IO: Check Runtime Threading (#1351)
-- Changes to make AMR codes compiled with EB but not building any
Geometry work as if it is all regular. (#1349)
-- Fix mkconfig.py (#1344)
-- Add -Xcudafe --diag_suppress=esa_on_defaulted_function_ignored to
suppress spurious warnings from nvcc. (#1342)
-- EB extend_domain_face (#1321)
-- Add parameter that stores OpenMP support version (#1325)
# 20.09
-- Fix interpolation coefficients for EB stencil in cell-centered linear
solvers when the EB surface at the domain boundary is tilted.
-- Add a parameter to `AMReX_omp_mod` that can be used by Fortran code to
check whether AMReX was built with OpenMP support, and, if yes, what
the version number (_OPENMP) was at the time AMReX was compiled.
-- CMake: re-write genex evaluation functions.
-- Extend the number of unique particles per cpu we can have at once.
-- Add a new method to the `amrex_distromap` type in
F_Interfaces. `get_pmap` fills a caller-owned array of PEs.
-- Fix a long standing bug in GNU make system on the use of HOST and
HOSTNAME.
-- Add Scan::InclusiveSum and ExclusiveSum for CPU to avoid ifdef.
-- Add check for empty probin_file in Amr::restart.
-- Update Sundials interface, documentation, and build to be version
agnostic.
-- Updates for ROCm 3.6 and 3.7.
-- Add section to Make.unknown for intel mpi.
-- CMake: re-factor third party libraries setup.
-- Port TagBoxArray to GPU.
-- Fix a bug in matching of COMP_VERSION to correctly treat CCE > 9.
-- By default, EB outside the domain will be generated by extending from
the domain faces.
-- Fix an issues due to roundoff error in determining whether particles
are inside the domain.
-- Remove Perilla because it's incompatible with our GPU strategy and it's
no longer being maintained.
-- Switch the overset mask convention to that 1 means unknown and 0 means
known. The new convention is more convenient for AMR-Wind.
-- Remove old solvers, Src/LinearSolvers/C_CellMG and C_TensorMG. They
have been superseded by Src/LinearSolvers/MLMG.
-- MacProjector: allow for re-use of the object and enhance multi-level
algorithm.
-- Add asserts for bounds checking to Array1D and Array2D.
-- Optimization of the construction of SFC.
-- Refactoring of particle buffer map.
-- Optimization of FPinfo and complementIn.
-- Move the Amr/Extrapolater used in IAMR/PeleLM to C++/GPU compliant.
-- Less surprising behavior for the ok() method of FabArray. This makes
the `ok()` method of FabArray return `false` instead of crashing if the
`define()` method has yet to be called.
-- Optimization of regrid.
-- Special FillBoundary for Almo.
-- Fix the average_down_faces calls in linear solvers to see periodicity.
-- Make sure the calculation of divu at inflow face in nodal projection
does not use tangential velocities on an inflow face.
# 20.08
-- New average_down_faces functions that take periodic boundary into account.
-- Update the documentation for Nsight.
-- Fix a bug with HDF5 boxcenter data.
-- New runtime parameter, `eb2.extend_domain_face`, that can be used to extraploate EB information
from domain faces.
-- New GNU Make options for GCC and Clang, `WARN_ALL` and `WARN_ERROR`. `WARN_ALL` turns on more
compiler warnings. `WARN_ERROR` turns warnings into errors.
-- Fix various compiler warnings.
-- Fix memory fragmentation issue with communication buffer.
-- Semi-coarsening support in linear solvers.
-- Fix virtual particle issues.
-- Capability to fuse GPU kernel launches.
-- Make ghost particles support soa data.
-- CMake: ability to export build tree.
-- Reset amrex_geometry_module initialization state when finalizing.
-- Remove Long version of Gatherv.
-- Optimization of TagBoxArray::collate.
-- CMake: fix missing PETSc setup in Config file.
-- BndryFunc for nodal data.
-- MLMG Hypre fix for singular problems.
-- Add particle copy filters.
-- Async output support for particles.
-- BackgroundThread class.
# 20.07
-- Nodal hypre solver now supports coarsening before hypre is called.
-- A new ParmParse parameter, `amrex.the_arena_is_managed` is introduced to set the default memory
type of `The_Arena`.
-- MFIter is now OMP thread safe when it is compiled with GPU support.
-- Overset support is added to cell-center and tensor linear solvers.
-- Remove compile time flag, BACKTRACE, for simplicity.
-- Updates on Blueprint and Ascent.
-- HDF5 support in GNU Make and CMake.
-- CMake support for DPC++.
-- Add -DNDEBUG for non-debug build.
-- Support for Windows.
-- Support for ParallelContext in Particles.
-- Non const iterator for neighbor list.
-- Tool for comparing single-level plotfiles of all nodalities.
-- GNU Make: Use -M instead the perl script to generate dependencies.
-- Particle: Remove directly accessing the m_idata and m_rdata structs internally.
-- Extend the makeSFC method to take an additional argument where the user can override the number
of processes a BoxArray is distributed to.
-- Turn tiling off by default for gpu.
# 20.06
-- Set a non-default CMAKE_INSTALL_PREFIX only if AMReX is the
top-level CMake project. Otherwise, the default of a downstream
project using AMReX as subproject will be placed in a quite
confusing, temporary directory.
-- C order support for Array2D and Array3D.
-- Disable unqualified abs function within amrex namespace.
-- Updates for SENSEI 3.2.
-- CMake: link to OpenMP::Fortran only when needed.
-- DPC++ support.
-- Add non-const data() to GpuArray.
-- Fix HIP version of scan.
-- Return immediately from GPU memcpy if size is zero.
-- Documentation on EB linear solvers.
# 20.05
-- AMReX's git history has been rewritten. If you have problems due
to this, follow the instruction here. https://xkcd.com/1597/ We
also have a new CONTRIBUTING.md.
-- EB linear solvers: You now need to specify the location of
diffusion coefficient (face center or face centroid) and the
location of the unknowns (cell center or cell centroid).
-- Optimization of makeKnapSack and makeSFC.
-- Unification of GPU and CPU versions of neighbor particle /
neighbor list implementations.
-- New HostDeviceParallelFor functions that can run on either host
or device depending on the launch region flag.
-- Add CUPTI trace for CUDA kernel timing.
-- Implement AMReX_FORTRAN (BL_NO_FORT) option in CMake.
-- Fix CUDA build of tools.
-- Fix invalid git version in CMake.
# 20.04
-- BaseFab can no longer be used on device. Many BaseFab functions
now requires RunOn::Device or RunOn::Host template parameter when
GPU support is enabled.
-- FillPatch* funcitons are now templated to support FabArray in
addition to MultiFab.
-- Fix -pthread for nvcc by default in Make.unknown
-- EB now has the option to build coarse level from GeometryShop
when coarsening from fine level fails. This could be used for
hyperbolic system with EB resolved to the finest level
-- CMake: add C++ 17 among valid options for AMREX_CXX_STANDARD
-- Add option to set fixed number of iterations in the Fortran
interface of linear solver
-- Fabs in alias FabArray now carries EB information.
-- Add AMREX_LOOP_3D and 4D macros.
-- Make amrex::Random() return numbers in the same interval for both
CPU and GPU: [0:1)
-- More flux register functionality for flash
-- Chunked partition in particle redistribution to use less memory.
# 20.03
-- AmrCore can now be built without using ParmParse.
-- Improve BL_NO_FORT support.
-- Fix OMP bug in MFIter constructors taking BoxArray and
DistributionMapping.
-- Cray compiler has been re-enabled in CMake.
-- For device only code, the users can now use #if AMREX_DIVICE_COMPILE.
-- Fix an integer overflow issue in TileArray cache.
# 20.02
-- RealVect updated for GPU.
-- Support for restarting from HDF5 particle checkpoint file.
-- Improvement of MacProjector and NodalProjector
# 20.01
-- GPU host device version of swap.
-- Fix an edge case of gpu block reduce where the block size is less
than a warp.
-- Always check MPI buffer size for overflow.
-- CMake support for PETSc.
-- Include C sources in the CUDA proprecessing.
-- It's now safe to construct a zero components Fab.
-- CXXSTD: option to set c++ standard in GNU make system
-- Improve performance of the nodal linear solver on GPU.
# 19.12
-- Fix potential deadlocks in amrex::Random.
-- MultiFab::Swap that does shallow swap if it can, does deep swap
otherwise.
-- FabArray::arena function that returns the Arena used to build.
-- Nodal projection sync residuals are implemented for RAP.
-- GPU support for initializing Fabs to NaNs.
# 19.11
-- AmrCore::regrid will now simply return instead of abort if the
base level is already the maximal level.
-- Add NodalProjector that wraps MLMG for convenience.
-- Enforce in AmrCore that blocking factor must be a power of 2.
-- Add support to write plotfile in HDF5 format.
-- FabArray's Reduce functions can now take lambda working on Array4
in addition of FAB.
-- Add AMREX_REAL_MIN, AMREX_REAL_MAX and AMREX_REAL_LOWEST for
convenience.
-- GPU launch functions and macros are now safe with empty boxes.
-- Remove FabView from BaseFab.
-- Avoid race conditions in GPU reduction code called multiple
times.
-- MLMG tensor solver is fixed for AMR.
# 19.10
-- Add inhomogeneous Neumann BC to MLMG for convenience.
-- Fix a corner case of tensor solver.
-- Various fixes for single precision and particle real.
-- Add amrex::Long that has a size of 8.
-- Add int8_t support in typecheck.
-- C++ user literal _rt for amrex::Real and _prt for
amrex::ParticleReal.
-- EB MAC projector now requires the user to specify whether the
face velocities are defined on face centers or centroids.
# 19.09
-- First pass of Sr/EB on GPU.
-- Optimization of particle redistribution.
-- EB integral: add integral of x*y*z
-- HostDevice::Atomic::Add that does atomic add for both device and
host omp code.
-- makeFineMask: optional arguments specifying coarse and fine flag
values.
-- Optimization of FabArray communication for GPU by merging small
kernels.
# 19.08
-- MacProjector has the option to set maximal coarsening level.
-- OwnerMask and OverlapMask are now built on GPU.
-- Random number generation on GPU no longer requires an call to
initialize seeds. They are done in amrex::Initialize and resize
is not needed when launching with more threads than the size of
seed array.
-- Optimization of NeighborParticles.
# 19.07
-- Geometry/CoordSys: Static class members have been removed.
Functions such as `isPeriodic`, `isAnyPeriodic`, `isAllPeriodic`,
`ProbLo`, `ProbHi`, `IsCartesian`, `IsRZ`, and `IsSPHERICAL` are
now non-static. The default Geometry can been obtained via
`DefaultGeometry()` function.
-- Funciton amrex::Loop and amrex:LoopConcurrent are added to help
looping over cells in a Box.
-- Nodal solver: A bug in nodal owner mask that could affect
covergence was fixed.
-- Fortran amrex_amrcore_module: add amrex_set_boxarray,
amrex_set_distromap, and amrex_set_finest_level.
-- Option to allocate MultiFab data in arenas different from
The_Arena().
-- Option to tag MultiFabs and query their memory usages.
-- AsyncFab has been deleted.
# 19.06
-- Geometry/CoordSys: New constructor and define function are added.
We plan to remove static data from Geometry and its base CoordSys
in the next release. This means most of the static functions in
Geometry class will become non-static. These include
`isPeriodic`, `isAnyPeriodic`, `isAllPeriodic`, `ProbLo`,
`ProbHi`, `IsCartesian`, `IsRZ`, and `IsSPHERICAL`. You can
prepare codes now for the upcoming changes. In almost all cases,
all you have to do is replace `Geometry::function` with
`DefaultGeometry().function`.
-- Add tensor linear solvers for viscous flows with and without EB.
-- Algoim: A modified version of Algoim is added. This removes the
dependence of 3d EB nodal MLMG solver on the official Algoim code
and Blitz.
-- EB: Fix the potential ownership issue is IndexSpace.
-- Add a new FillPatchTwoLevels for EB.
# 19.05
-- Interface of virtual function Interpolater::interp has changed.
-- Add Fortran funcntions for shared memory reductions.
-- EB nodal linear solver: The default bottom solver is now bicgcg.
It starts with bicgstab and switches cg if bicgstab fails.
-- EB nodal linear solver: Finished the implementation of matrix
coarsening with RAP.
-- Particle: Option for adding runtime struct of array components.
-- Add a trilinear interpolation option for 3D.
-- Add more support for non-uniform grow vectors to MultiFab and
FabArray.
-- AmrMesh: n_error_buf is now IntVect. This allows for different
number of error buffer cells in different directions.
-- Using PETSc as bottom solver no longer requires Hypre.
# 19.04
-- SUNDIALS 4 support.
-- Option to disable device sync in MFIter.
-- Functions for prefetch unified memory.
-- GNU Make system: It is now an error if a function with return
type does not return.
-- GNU Make system: Pass C++ flags to nvcc when compiling C codes.
-- Allocate 3/4 of gpu memory upfront in The_Arena.
-- BaseFab now has Allocator template parameter.
-- Random number generation on gpu.
# 19.03
-- Deprecated Fortran framework F_BaseLib and linear solver F_MG
have been removed. A number of tools written in F_BaseLib have
been migrated to Tools/Plotfile/. Note that the Fortran wrappers
to C++ (F_Interfaces) are still there.
-- In addition to launch macros, there are now functions for
lauching, amrex::launch, amrex::For and amrex::ParallelFor.
These improves debuggability.
-- If compiled with BASEFAB_IS_PINNED, pinned memory will be used
for BaseFab meta-data. On systems with nvlink, this improves the
performance a little bit.
-- Performance improvement in particle redistribution.
-- Particle NeighborList has been moved from tutorial to AMReX
proper.
-- Cell-centered MLMG linear solver have been ported to GPU.
# 19.02
-- Array4 class that contains the important information in BaseFab