Compare commits

..
106 Commits
Author SHA1 Message Date
sfilippone 2e79105695 Fix CUDA defines for compilation 2025-02-21 18:07:58 +01:00
sfilippone 921535c3c9 Update copyright 2025-02-21 16:41:39 +01:00
sfilippone 557809755e Merge branch 'development' of github.com:sfilippone/amg4psblas into development 2025-02-21 16:36:10 +01:00
sfilippone 64105a17c5 Delete changelog 2025-02-21 16:35:45 +01:00
sfilippone 67c32da222 Update docs 2025-02-21 16:33:43 +01:00
Salvatore Filippone f317a571f4 Update README.md 2025-02-21 16:12:17 +01:00
sfilippone b075182ce6 Obsolete AINV_SOLVER_SETirc 2025-02-21 16:06:21 +01:00
sfilippone be469f8844 Added comments on allocat_wrk 2025-02-21 15:39:43 +01:00
Salvatore Filippone 04bcf04a9c Update README.md 2025-02-21 15:29:59 +01:00
sfilippone 725e39586d Fix TLU solver_bld and CUDA test program to print format. 2025-02-21 15:22:51 +01:00
sfilippone e68304e84f Merge branch 'development' of github.com:sfilippone/amg4psblas into development 2025-02-21 15:06:09 +01:00
sfilippone 3972f27eb5 Rework cuda example 2025-02-21 15:05:35 +01:00
fdurastante b21e5aebab Update README.md 2025-02-21 14:33:39 +01:00
Cirdans-Home 6e2d61b51e Initial README for examples 2025-02-21 13:03:43 +00:00
sfilippone 4ee170f0ec Update version 1.2.0 2025-02-21 12:32:36 +01:00
sfilippone 4a9f8e9da0 Move newslv to samples and fix it 2025-02-21 12:32:04 +01:00
sfilippone 94caab8aa9 Removed make veryclean, introduced make distclean 2025-02-20 17:00:32 +01:00
sfilippone 749ab2f6ae Fix extprol. Remove obsolete hybrid aggregator. 2025-02-20 11:53:58 +01:00
sfilippone bf96dd554c Merge branch 'repackage' into development 2025-02-19 17:35:47 +01:00
sfilippone 1aef82023c Merge branch 'repackage' of github.com:sfilippone/amg4psblas into repackage 2025-02-09 16:32:40 +01:00
sfilippone 161b93da64 Fix interface issues of clone_settings 2025-02-09 16:02:00 +01:00
Salvatore Filippone 8d4af1ba9f Fix output formatting bug 2025-01-19 11:15:44 +01:00
sfilippone cd87bdb0c1 Add call to clean_zeros() 2025-01-08 17:19:58 +01:00
Cirdans-Home 53afa87814 Added row-preserving filtering 2024-12-13 17:19:04 +00:00
sfilippone 13c99a0c3f Cleanup use of RICHARDSON 2024-11-18 16:20:35 +01:00
sfilippone 1aa7c8db59 Fix release version of UG 2024-11-18 15:53:52 +01:00
sfilippone 36b57eec24 Define 1.2 as name of UG 2024-11-18 15:51:16 +01:00
fdurastante 417f8beaf9 Added remark on polynomial smoothers 2024-11-18 14:42:31 +01:00
fdurastante d66bf1e2f8 Added remark on polynomial smoothers 2024-11-18 14:41:59 +01:00
sfilippone c590a4088a Merge branch 'repackage' of github.com:sfilippone/amg4psblas into repackage 2024-11-18 12:52:39 +01:00
sfilippone 80dfd1ad3b Modified README 2024-11-18 12:47:39 +01:00
Pasqua D'Ambra 857844474a Update README.md 2024-11-18 10:56:46 +01:00
fdurastante a85997926c Update README.md 2024-11-18 10:08:29 +01:00
fdurastante 70fb39ae55 Added Polynomial Smoother infos 2024-11-18 10:03:21 +01:00
sfilippone 82de529c54 Makefile in amg4psblas/docs 2024-11-18 09:17:11 +01:00
sfilippone db9757a45e Update README Fortran 2003 to 2008 2024-11-15 17:28:14 +01:00
fdurastante 5b654cc221 Update README.md 2024-11-15 17:13:17 +01:00
sfilippone d518f5eac7 Merge branch 'l1-and-0-aggr' into repackage 2024-11-12 10:07:44 +01:00
sfilippone 053d5c4bc0 Changed sample input file 2024-11-11 18:10:34 +01:00
sfilippone 032543d625 Adjusted CBIND test program. 2024-11-11 13:06:48 +01:00
sfilippone 07149a02ad Fixes for krylov->linsolve. Test program to be completed. 2024-11-11 09:11:07 +01:00
sfilippone 08a0c744b1 Switch from KRYLOV to LINSOLVE 2024-11-10 14:00:51 +01:00
sfilippone 60324084d8 New Richardson solver. 2024-11-08 18:25:13 +01:00
sfilippone 83ba79d7ae Fix input file 2024-11-02 10:33:23 +01:00
sfilippone b704d50df1 Fix dump of TPROL and POLY smoother 2024-10-30 10:49:05 +01:00
Cirdans-Home 68a9cceaa0 Fixed l1-Jacobi scaling 2024-10-29 15:51:37 +00:00
Salvatore Filippone aec5a52c7f Missing CBIND clean target 2024-10-16 10:56:53 +02:00
Salvatore Filippone b5f5d356fd Missing CBIND clean target 2024-10-16 10:55:59 +02:00
Cirdans-Home 8966ecb4a6 First implementation of l1-aggregation 2024-10-09 09:43:43 +00:00
Cirdans-Home 3e9a5c0c5b Remove diagonal zeros from aggregation 2024-10-04 12:28:36 +00:00
sfilippone dfc261cf34 Modify error message in smth_bld 2024-09-12 18:45:20 +02:00
sfilippone cab98295e2 Improve handling of pointers in hierarchy_bld 2024-08-06 10:28:02 +02:00
sfilippone 5a83c63810 Unused lines in ilu_solver_bld 2024-08-02 15:00:44 +02:00
sfilippone 2e43f55455 Update sample/advanced/pdegeb 2024-08-02 15:00:28 +02:00
sfilippone 244fcda207 Update test program input 2024-08-02 13:48:42 +02:00
sfilippone ca6fce0765 Fix for test program 2024-08-02 13:46:41 +02:00
sfilippone b6f92354d3 Reworked sample data generation 2024-08-02 12:39:09 +02:00
sfilippone 1b7fe6a9a7 Fix input data in samples pdegen 2024-07-30 09:49:46 +02:00
sfilippone 14ea4d9c15 Fix sample programs for polynomial smoothers 2024-07-29 17:35:01 +02:00
sfilippone 3ee333baac Change name abgdxyz into upd_xyz 2024-07-29 16:59:59 +02:00
sfilippone ecb41dfbbf Update pde3d.inp 2024-07-16 13:41:51 +02:00
sfilippone 33ac3f786b Merge latest changes from polysmooth 2024-07-16 13:12:10 +02:00
sfilippone 474c6a3634 Merge branch 'PolySmooth' into development 2024-07-05 16:50:26 +02:00
sfilippone c1e8bc0c57 Do not use OpenMP code for serial version 2024-07-05 14:23:44 +02:00
sfilippone 2f5072166d Switch off OpenMP in certain sections of MatchBOXP 2024-07-05 13:56:52 +02:00
sfilippone 89e2d53e8b Silence debug print. 2024-07-05 13:56:38 +02:00
sfilippone bfe0a32e09 Merge branch 'mboxomp' into PolySmooth 2024-07-05 13:08:56 +02:00
sfilippone e88d176fed Switch off OpenMP in processExposedVertex 2024-07-05 12:10:26 +02:00
sfilippone c96727a97c Merge branch 'PolySmooth' into mboxomp 2024-07-03 11:01:08 +02:00
sfilippone 6362db0cc5 Try improve OpenMP version of matchbox 2024-07-03 11:00:13 +02:00
sfilippone 9239b16175 Merge branch 'PolySmooth' of github.com:sfilippone/amg4psblas into PolySmooth 2024-07-03 10:58:33 +02:00
sfilippone 96a700cb9d Workaround for save_smoothers bug with gcc-13.3.0 2024-07-03 10:58:00 +02:00
sfilippone 41d91120d4 Merge branch 'PolySmooth' into mboxomp 2024-06-11 09:57:53 +02:00
sfilippone 5d20407b15 Merge branch 'PolySmooth' of github.com:sfilippone/amg4psblas into PolySmooth 2024-06-11 09:56:39 +02:00
sfilippone 322e3f65d1 Merge branch 'PolySmooth' of github.com:sfilippone/amg4psblas into PolySmooth 2024-06-11 09:55:07 +02:00
sfilippone 3ff1ad9372 Fix noise in %memory_use() 2024-06-11 09:54:55 +02:00
sfilippone 818ead5878 Try changes for matching 2024-06-11 09:52:11 +02:00
sfilippone 803d311d1c S versions. Take out parallel in a few places 2024-06-06 17:12:31 +02:00
sfilippone 677e4fe6bc Modify MatchBox names with D in preparation for S version 2024-06-05 15:13:18 +02:00
sfilippone 02a83575a2 Reorganize MatchBox (prepare for S OpenMP) 2024-06-05 13:11:41 +02:00
sfilippone cfbec1f6ea Cosmetic changes to MatchBoxPC 2024-06-05 10:38:30 +02:00
sfilippone e11a134a1f Additional tinkering with OpenMP matchbox 2024-06-04 17:39:31 +02:00
sfilippone 6d05120930 Merge branch 'PolySmooth' of github.com:sfilippone/amg4psblas into PolySmooth 2024-06-04 16:56:07 +02:00
sfilippone bd2d1e3b26 Additional OpenMP tinkering in matchboxp 2024-06-04 16:53:52 +02:00
sfilippone c9605d1b29 Mods for OpenMP 2024-06-04 13:25:12 +02:00
sfilippone 67594f8b07 Fixes for OpenMP 2024-05-30 17:35:15 +02:00
sfilippone 301fb57bb1 Merge branch 'PolySmooth' of github.com:sfilippone/amg4psblas into PolySmooth 2024-05-30 17:26:49 +02:00
sfilippone 13eee99ea3 Use ifdef OPENMP 2024-05-30 17:26:34 +02:00
sfilippone fb802c62cd Merge PSBCXXDEFINES into AMGCXXDEFINES 2024-05-30 17:25:25 +02:00
sfilippone 767b606bb2 Do away with -DOMP 2024-05-30 16:14:32 +02:00
sfilippone 8492c07521 Merge PSBCXXDEFINES into CXXDEFINES 2024-05-30 16:14:12 +02:00
sfilippone 17698c2725 Changes to OpenMP MathcBox version needs -DOMP 2024-05-30 14:43:26 +02:00
Salvatore Filippone 897c5229a6 Improve behaviour of OpenMP matching 2024-05-30 08:13:08 -04:00
Salvatore Filippone ab5eaac5ed Cosmetic changes 2024-05-30 08:12:48 -04:00
sfilippone 234071869d Merge branch 'PolySmooth' of github.com:sfilippone/amg4psblas into PolySmooth 2024-05-10 15:03:38 +02:00
sfilippone 3e3b343131 Fix potential overflow issue in SOC_MAP_BLD 2024-05-10 15:02:36 +02:00
Cirdans-Home 5790aa0cbd Revert "First hardcoded implementation of l1 smooth aggregation"
This reverts commit a17f503486.
2024-04-26 10:55:47 +00:00
Cirdans-Home a17f503486 First hardcoded implementation of l1 smooth aggregation 2024-04-26 10:54:43 +00:00
Cirdans-Home 74dccb6c44 Added timers and removed unuseful spmm 2024-04-22 07:52:14 +00:00
sfilippone e83bde6896 New timings 2024-04-17 14:45:20 +02:00
Salvatore Filippone 83d435b49e Default GLOBAL=.true. for MEMORY_USE 2024-03-16 15:22:24 +01:00
Salvatore Filippone 2ef4459b18 Added "COARSE_INVFILL" 2024-02-21 11:51:18 +01:00
sfilippone c2fd0ac66d Disable MATCHBOX with SERIAL_MPI and add error message 2023-11-26 11:43:10 +01:00
sfilippone 5387e206b1 Fixed sample programs 2023-11-24 16:07:22 +01:00
sfilippone fc34385341 Fix matrix generation for samples 2023-11-16 17:55:42 +01:00
sfilippone 5fbdfb1436 Fix free for jac_solver 2023-11-16 17:55:24 +01:00
377 changed files with 20264 additions and 11068 deletions
+2
View File
@@ -20,3 +20,5 @@ autom4te.cache
# the executable from tests
runs
# Documentation temporary files
docs/src/userguide.pdf
-165
View File
@@ -1,165 +0,0 @@
Changelog. A lot less detailed than usual, at least for past
history.
2022/05/20: Restart ChangeLog. Updated to new name AMG4PSBLAS, now using PSB3.8
2018/10/28: Fix interface to MUMPS and configry machinery. Require PSB 3.6.
2018/10/10: ICTXT argument in prec%init().
2018/07/30: Fixes for Intel compilers. BootCMatch interface in examples.
2018/05/14: Interface for extension of aggregation methods.
2018/02/28: New cartesian distribution for sample programs.
2017/12/15: New WRK component of preconditioner levels, preallocation.
2017/10/25: New example input file formats. Added sample matrices.
2017/10/02: New CBIND in PSBLAS 3.5.0
2017/07/30: Refactored examples. Change default thresholds.
2017/05/31: New internal description of ML.
2017/05/16: Improve build process.
2017/04/20: Force %set interface. Update docs.
2017/04/03: Remove obsolete stuff.
2017/03/17: Fixed level%cnv; add coarse _solver tracker.
2017/02/18: Take out clean_zeros; changed NOFILTER; defined FBGS; take out
n_prec_levs.
2017/02/12: Updated mat_dist usage, dubious SP shell for UMFPACK, fixes
for RPM packaging.
2017/02/02: Fix superlu configury
2016/11/12: Fix hierarchy/smoothers build to handle 1 level.
2016/10/03: Merged changes to hierearchy building.
2016/08/20: Reimplemented decoupled aggregation
2016/07/20: Refactored application of multilevel. Defined V,W and
K-cycles.
2016/05/18: Reworked internals of PRECSET. Defined Forward-Backward
Gauss-Seidel solver. Now available separate PRE and POST smoother
objects.
2016/03/30: MUMPS interface.
2016/02/28: Hybrid Gauss-Seidel method.
2016/02/03: unify integer argument checks.
2015/12/15: defaults single vs. double precision. Use clean_zeros.
2015/12/08: new matdist interface
2015/10/17: configry fixes
2015/10/13: Fixes for SLUDIST versions 3 and 4
2015/05/03: New heap interface
2015/04/21: INTENT fixes
2014/12/21: New error handling
2014/10/27: Added versioncheck to configure.
2014/03/31: New get_diag.
2013/11/07: Merged changes from experimental branch. Fix INCDIR in
makefiles.
2013/07/15: Fixes for UMFPACK 5.4, SuperLU 4.3, SuperLU_Dist 3.3
2013/04/05: CLONE method.
2013/03/08: Reworked SET routines.
2012/12/10: Enable long_integers.
2012/12/05: Split smoother/solver objects.
2012/04/30: New scheme to find dynamically the number of level based on
the size of the coarse matrix
2012/01/10: Done split interface/implementation, plus subdir restructure.
2011/12/13: Start split interface/implementation to improve build time.
2011/11/25: Now works with _vect methods from PSBLAS.
2011/10/24: New test generation methods.
2011/06/15: Dump prolongator/restrictor
2011/04/14: Added MOLD argument(s) to precbld.
2011/03/30: Fixed: descriptive methods, example programs.
2011/03/08: Re-factored modules for ILU methods.
2011/03/04: Make X intent(inout) in APPLY to allow for preconditioners
using SPMM.
2011/03/02: New set methods.
2011/01/07: Fixed UMF interfacing for Z data.
2011/01/04: Added UMF inteface for D data.
2011/01/02: Fix usage of DESC_DATA. Switched all names to F90 ending.
2010/12/16: Fix usage of replicated space descriptor.
2010/11/16: Fix Jacobi smoother in case of empty off-diagonal.
2010/11/04: Defined and tested single real and complex.
2010/11/02: Aligned usage of sparse data type with psblas3.
2009/12/22: Aligned constants with mld2p4 v1.2
2009/12/11: First working version of double multilevel.
2009/12/05: Inttroduction of Smoother/Solver object hierarchy.
2009/09/23: Initial F2003 version.
2009/01/28: Changed names from XbaseprcY to XbaseprecY.
2009/01/27: Changed names from mld_transfer to mld_move_alloc.
2009/01/13: Repackaged the one-level preconditioners. Reorganized the
build routines, taking out mlprec_bld, and switching the
number of levels when needed.
2008/10/27: Changed the definition of prec_type: repackaged with a
onelev-prec-type, containing a baseprec and maps between
index spaces. No performance impact; no changes to
user-level interfaces.
2008/09/18: Changed mld_sizeof to integer(8); updated samples.
2008/08/26: Fixed matrix generation in sample programs.
2008/07/25: missing implicit none in mld_prec_type.
2008/07/23: added HTML documentation
2008/06/13: Fixed aggregation for replicated index spaces.
2008/06/02: Threshold into decoupled aggregation algorithm.
2008/05/27: Single precision version.
2008/03/09: Introduced configure script.
2008/02/08: Merged changes from intermesh branch: we now have an
inter_desc_type object. Cleaned up data allocation and
variable initialization in multilevel prec application.
2008/01/10: Merged various fixes for: prologues, unused variables,
interface details.
2007/12/21: Merge version with prologues and internal docs.
2007/11/15: Created pargen example.
2007/11/14: Fix INTENT(IN) on X vector in preconditioner routines.
2007/10/19: Merged in ILU(P,T). To be tested extensively.
2007/10/17: Merged ILU(K) into trunk.
2007/10/16: Fixed ILU(K), it now performs satisfactorily. Also updated
ILU(0) to be more legible.
2007/10/11: First working version of ILU(K). Still slow, there should
be room for improvement.
2007/10/09: Added benchmark code.
2007/10/09: Added MILU_N_. Beware: values for UMF_ etc. have been
shifted.
2007/10/02: To do: decide whether to name MLD_KRYLOV_MOD or
PSB_KRYLOV_MOD.
2007/10/01: Start of this changelog. MLD2P4 now has a different
structure, to enable a build not embedded in PSBLAS.
+3 -3
View File
@@ -1,10 +1,10 @@
AMG4PSBLAS version 1.1
AMG4PSBLAS version 1.2
Algebraic Multigrid Package
based on PSBLAS (Parallel Sparse BLAS version 3.8)
based on PSBLAS (Parallel Sparse BLAS version 3.9)
(C) Copyright 2022
(C) Copyright 2025
Salvatore Filippone
Pasqua D'Ambra
+1 -1
View File
@@ -75,7 +75,7 @@ CDEFINES=$(AMGCDEFINES)
AMGFDEFINES=@AMGFDEFINES@ $(PSBFDEFINES)
FDEFINES=$(AMGFDEFINES)
CXXDEFINES=@AMGCXXDEFINES@
CXXDEFINES=@AMGCXXDEFINES@ $(PSBCXXDEFINES)
@COMPILERULES@
+1 -1
View File
@@ -49,7 +49,7 @@ PSBLAS_INCLUDES=@PSBLAS_INCLUDES@
PSBLAS_LIBS=@PSBLAS_LIBS@
PSBBASEMODNAME=psb_base_mod
PSBPRECMODNAME=psb_prec_mod
PSBMETHDMODNAME=psb_krylov_mod
PSBMETHDMODNAME=psb_linsolve_mod
PSBUTILMODNAME=psb_util_mod
+8 -5
View File
@@ -3,9 +3,9 @@ include Make.inc
all: objs lib
objs: amgp cbnd
objs: libdir amgp cbnd
lib: libdir objs
lib: objs
cd amgprec && $(MAKE) lib
cd cbind && $(MAKE) lib
@@ -44,8 +44,10 @@ cleanlib:
(cd include; /bin/rm -f *.a *$(.mod) *$(.fh))
(cd modules; /bin/rm -f *.a *$(.mod) *$(.fh))
veryclean: cleanlib
(cd amgprec && $(MAKE) veryclean)
distclean: clean samplesclean
/bin/rm -fr Make.inc
samplesclean: clean
(cd samples/simple/fileread && $(MAKE) clean)
(cd samples/simple/pdegen && $(MAKE) clean)
(cd samples/advanced/fileread && $(MAKE) clean)
@@ -54,5 +56,6 @@ veryclean: cleanlib
check: all
make check -C samples/advanced/pdegen
clean:
clean: cleanlib
(cd amgprec && $(MAKE) clean)
(cd cbind && $(MAKE) clean)
+57 -35
View File
@@ -1,54 +1,76 @@
AMG4PSBLAS
Algebraic Multigrid Package based on PSBLAS (Parallel Sparse BLAS version 3.8)
Salvatore Filippone (University of Rome Tor Vergata and IAC-CNR)
Pasqua D'Ambra (IAC-CNR, Naples, IT)
Fabio Durastante (IAC-CNR, Naples, IT)
# AMG4PSBLAS v1.2
Algebraic Multigrid Package based on [PSBLAS](https://github.com/sfilippone/psblas3) (Parallel Sparse BLAS version 3.9)
---------------------------------------------------------------------
AMG4PSBLAS is a package of parallel algebraic multilevel preconditioners included in the PSCToolkit (Parallel Sparse Computation Toolkit) software framework.
AMG4PSBLAS is a package of Algebraic MultiGrid (AMG)
preconditioners for the iterative solution of large and sparse linear systems.
It is a progress of a software development project started in 2007, named MLD2P4, which originally implemented a multilevel version of some domain decomposition preconditioners of additive-Schwarz type and was based on a parallel decoupled version of the well known smoothed aggregation method to generate the multilevel hierarchy of coarser matrices.
It is an evolution of MLD2P4 (see LICENSE.MLD2P4), but it has been
thoroughly reworked, and it is sufficiently different to warrant a new
project name.
In the last years the package was extended for including new algorithms and functionalities for the setup and application new AMG preconditioners with the final aims of improving efficiency and scalability when tens of thousands cores are used and of boosting reliability in dealing with general symmetric positive definite linear systems.
It is an evolution of MLD2P4 (see [LICENSE.MLD2P4](LICENSE.MLD2P4)), but due to the significant number of changes and the increase in scope, we decided to rename the package as AMG4PSBLAS.
MAIN REFERENCES:
AMG4PSBLAS has been designed to provide scalable and easy-to-use preconditioners in the context of the PSBLAS (Parallel Sparse Basic Linear Algebra Subprograms) computational framework and can be used in conjuction with the Krylov solvers available in this framework. Our package is based on a completely algebraic approach; therefore users level interfaces assume that the system matrix and preconditioners are represented as PSBLAS distributed sparse matrices.
AMG4PSBLAS enables the user to easily specify different features of an algebraic multilevel preconditioner, thus allowing to experiment with different preconditioners for the problem and parallel computers at hand.
P. D'Ambra, D. di Serafino, S. Filippone,
MLD2P4: a Package of Parallel Algebraic Multilevel Domain Decomposition
Preconditioners in Fortran 95,
ACM Transactions on Mathematical Software, 37 (3), 2010, art. 30,
doi: 10.1145/1824801.1824808.
The package employs object-oriented design techniques in Fortran 2008, with interfaces to additional third party libraries such as MUMPS, UMFPACK, SuperLU, and SuperLU_Dist, which can be exploited in building multilevel preconditioners. The parallel implementation is based on a Single Program Multiple Data (SPMD) paradigm; the inter-process communication is based on MPI and is managed mainly through PSBLAS.
## Main Refrerences:
TO COMPILE
The main reference for this project is
> D'Ambra, P., Durastante, F., & Filippone, S. (2021). AMG preconditioners for linear solvers towards extreme scale. SIAM Journal on Scientific Computing, 43(5), S679-S703.
AMG4PSBLAS is the suite of preconditioners for the Parallel Sparse Computation Toolkit ([PSCToolkit](https://psctoolkit.github.io/)) suite of libraries. See the paper:
> D’Ambra, P., Durastante, F., & Filippone, S. (2023). Parallel Sparse Computation Toolkit. Software Impacts, 15, 100463.
The main reference for features inherited from MLD2P4 is
> P. D'Ambra, D. di Serafino, S. Filippone,
> MLD2P4: a Package of Parallel Algebraic Multilevel Domain Decomposition
> Preconditioners in Fortran 95,
> ACM Transactions on Mathematical Software, 37 (3), 2010, art. 30,
> doi: 10.1145/1824801.1824808.
## Installing
Installation requires having a working version of the [PSBLAS](https://github.com/sfilippone/psblas3) library installed.
AMG4PSBLAS has several interfaces to third-party libraries that can be used in the construction and application phases of preconditioners.
In particular, it is possible to link AMG4PSBLAS with the libraries: MUMPS, SuperLU, SuperLU_Dist, UMFPACK. This is _not mandatory_ and the library can run
in isolation and without these features.
0. Unpack the tar file in a directory of your choice (preferrably
outside the main PSBLAS directory).
1. run configure --with-psblas=<ABSOLUTE path of the PSBLAS install directory>
1. run configure `--with-psblas=<ABSOLUTE path of the PSBLAS install directory>`
adding the options for MUMPS, SuperLU, SuperLU_Dist, UMFPACK as desired.
See MLD2P4 User's and Reference Guide (Section 3) for details.
2. Tweak Make.inc if you are not satisfied.
3. make;
See [AMG4PSBLAS User's and Reference Guide](docs/amg4psblas_1.0-guide.pdf) (Section 3) for details.
2. Tweak `Make.inc` if you are not satisfied.
3. run `make`;
4. Go into the test subdirectory and build the examples of your choice.
5. (if desired): make install
5. (if desired): `make install`
>[!CAUTION]
>The single precision version is supported only by MUMPS and SuperLU;
>thus, even if you specify at configure time to use UMFPACK or SuperLU_Dist,
>the corresponding preconditioner options will be available only from
>the double precision version.
NOTES
### CUDA, OpeMP, OpenACC
- The single precision version is supported only by MUMPS and SuperLU;
thus, even if you specify at configure time to use UMFPACK or SuperLU_Dist,
the corresponding preconditioner options will be available only from
the double precision version.
CUDA, OpenMP and OpenACC features are transparently inherited by PSBLAS installation. If PSBLAS has been configured (and installed) with these supports then AMG4PSBLAS will transparently inherit them. It will then be possible to move the computation to GPU accelerator simply by selecting the appropriate variable types. If these have not been activated or installed for PSBLAS then they will not be available for AMG4PSBLAS either and the operation will be purely on CPU/MPI. See also the samples/cuda folder.
### EoCoE - Software as service portal
In the European project “Energy oriented Center of Excellence: toward exascale for energy” we made available a software as service portal: [https://eocoe.psnc.pl/](https://eocoe.psnc.pl/). This permits to test several cutting-edge computational methods for accelerating the transition to the production, storage and management of clean, decarbonized energy. Among them you have the possibility of running PSBLAS+AMG4PSBLAS on some test problems to become familiar with using the software.
## TODO and bugs
- [X] Fix all reamining bugs. Bugs? We dont' have any ! 🤓
> [!NOTE]
> To report bugs 🐛 or issues ❓ please use the [GitHub issue system](https://github.com/sfilippone/amg4psblas/issues).
## The AMG4PSBLAS team.
- Pasqua D'Ambra (IAC-CNR, Naples, IT)
- Fabio Durastante (University of Pisa and IAC-CNR, IT)
- Salvatore Filippone (University of Rome Tor Vergata and IAC-CNR, IT)
The AMG4PSBLAS team.
---------------
Salvatore Filippone
Pasqua D'Ambra
Fabio Durastante
+14
View File
@@ -1,5 +1,19 @@
WHAT'S NEW
AMG4PSBLAS
Version 1.2
1. New polynomial smoothers.
2. Introduced L1-variants
3. Reorganization of sample programs.
Version 1.1
1. Reworked approximate inverse solvers.
Version 1.0
1. Transitioned from MLD2P4
MLD2P4
Version 2.1
1. The multigrid preconditioner now include fully general V- and
W-cycles. We also support K-cycles, both for symmetric and
+31 -21
View File
@@ -81,9 +81,9 @@ module amg_base_prec_type
!
! Version numbers
!
character(len=*), parameter :: amg_version_string_ = "1.1.0"
character(len=*), parameter :: amg_version_string_ = "1.2.0"
integer(psb_ipk_), parameter :: amg_version_major_ = 1
integer(psb_ipk_), parameter :: amg_version_minor_ = 1
integer(psb_ipk_), parameter :: amg_version_minor_ = 2
integer(psb_ipk_), parameter :: amg_patchlevel_ = 0
type amg_ml_parms
@@ -288,17 +288,19 @@ module amg_base_prec_type
!
! Legal values for entry: amg_aggr_prol_
!
integer(psb_ipk_), parameter :: amg_no_smooth_ = 0
integer(psb_ipk_), parameter :: amg_smooth_prol_ = 1
integer(psb_ipk_), parameter :: amg_min_energy_ = 2
integer(psb_ipk_), parameter :: amg_no_smooth_ = 0
integer(psb_ipk_), parameter :: amg_smooth_prol_ = 1
integer(psb_ipk_), parameter :: amg_l1_smooth_prol_ = 2
integer(psb_ipk_), parameter :: amg_min_energy_ = 3
! Disabling min_energy for the time being.
integer(psb_ipk_), parameter :: amg_max_aggr_prol_=amg_smooth_prol_
integer(psb_ipk_), parameter :: amg_max_aggr_prol_= amg_l1_smooth_prol_
!
! Legal values for entry: amg_aggr_filter_
!
integer(psb_ipk_), parameter :: amg_no_filter_mat_ = 0
integer(psb_ipk_), parameter :: amg_filter_mat_ = 1
integer(psb_ipk_), parameter :: amg_max_filter_mat_ = amg_filter_mat_
integer(psb_ipk_), parameter :: amg_no_filter_mat_ = 0
integer(psb_ipk_), parameter :: amg_filter_mat_ = 1
integer(psb_ipk_), parameter :: amg_filter_prow_mat_ = 2
integer(psb_ipk_), parameter :: amg_max_filter_mat_ = amg_filter_prow_mat_
!
! Legal values for entry: amg_aggr_ord_
!
@@ -323,9 +325,10 @@ module amg_base_prec_type
!
! Legal values for entry: amg_poly_variant_
!
integer(psb_ipk_), parameter :: amg_poly_lottes_ = 0
integer(psb_ipk_), parameter :: amg_poly_lottes_beta_ = 1
integer(psb_ipk_), parameter :: amg_poly_new_ = 2
integer(psb_ipk_), parameter :: amg_cheb_4_ = 0
integer(psb_ipk_), parameter :: amg_cheb_4_opt_ = 1
integer(psb_ipk_), parameter :: amg_cheb_1_opt_ = 2
integer(psb_ipk_), parameter :: amg_poly_dbg_ = 8
integer(psb_ipk_), parameter :: amg_poly_rho_est_power_ = 0
@@ -375,10 +378,11 @@ module amg_base_prec_type
character(len=19), parameter, private :: &
& eigen_estimates(0:0)=(/'infinity norm '/)
character(len=15), parameter, private :: &
& aggr_prols(0:3)=(/'unsmoothed ','smoothed ',&
& 'min energy ','bizr. smoothed'/)
& aggr_prols(0:4)=(/'unsmoothed ','smoothed ',&
& 'l1-smoothed ','min energy ','bizr. smoothed'/)
character(len=15), parameter, private :: &
& aggr_filters(0:1)=(/'no filtering ','filtering '/)
& aggr_filters(0:2)=(/'no filtering ','filtering ',&
& 'filtering rsum'/)
character(len=15), parameter, private :: &
& matrix_names(0:1)=(/'distributed ','replicated '/)
character(len=18), parameter, private :: &
@@ -547,6 +551,8 @@ contains
val = amg_no_smooth_
case('SMOOTHED')
val = amg_smooth_prol_
case('L1-SMOOTHED','L1SMOOTHED')
val = amg_l1_smooth_prol_
case('MINENERGY')
val = amg_min_energy_
case('NOPREC')
@@ -569,12 +575,14 @@ contains
val = amg_as_
case('POLY')
val = amg_poly_
case('POLY_LOTTES')
val = amg_poly_lottes_
case('POLY_LOTTES_BETA')
val = amg_poly_lottes_beta_
case('POLY_NEW')
val = amg_poly_new_
case('CHEB_4')
val = amg_cheb_4_
case('CHEB_4_OPT')
val = amg_cheb_4_opt_
case('CHEB_1_OPT')
val = amg_cheb_1_opt_
case('POLY_DBG')
val = amg_poly_dbg_
case('POLY_RHO_EST_POWER')
val = amg_poly_rho_est_power_
case('A_NORMI')
@@ -585,6 +593,8 @@ contains
val = amg_eig_est_
case('FILTER')
val = amg_filter_mat_
case('FILTERROWSUM')
val = amg_filter_prow_mat_
case('NOFILTER','NO_FILTER')
val = amg_no_filter_mat_
case('OUTER_SWEEPS')
+4 -7
View File
@@ -62,9 +62,6 @@ module amg_c_ainv_solver
procedure, pass(sv) :: cseti => amg_c_ainv_solver_cseti
procedure, pass(sv) :: csetc => amg_c_ainv_solver_csetc
procedure, pass(sv) :: csetr => amg_c_ainv_solver_csetr
!!$ procedure, pass(sv) :: seti => amg_c_ainv_solver_seti
!!$ procedure, pass(sv) :: setc => amg_c_ainv_solver_setc
!!$ procedure, pass(sv) :: setr => amg_c_ainv_solver_setr
procedure, pass(sv) :: descr => amg_c_ainv_solver_descr
procedure, pass(sv) :: default => c_ainv_solver_default
procedure, nopass :: stringval => c_ainv_stringval
@@ -91,9 +88,9 @@ module amg_c_ainv_solver
import :: psb_desc_type, psb_cspmat_type, psb_c_base_sparse_mat, &
& amg_c_base_solver_type, psb_dpk_, amg_c_ainv_solver_type, psb_ipk_
Implicit None
class(amg_c_ainv_solver_type), intent(inout) :: sv
class(amg_c_base_solver_type), allocatable, intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
class(amg_c_ainv_solver_type), intent(inout) :: sv
class(amg_c_base_solver_type), intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_c_ainv_solver_clone_settings
end interface
@@ -106,7 +103,7 @@ module amg_c_ainv_solver
Implicit None
! Arguments
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_ainv_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -230,7 +230,7 @@ module amg_c_as_smoother
& psb_desc_type, psb_c_base_sparse_mat, psb_ipk_,&
& psb_i_base_vect_type
implicit none
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_as_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -237,7 +237,7 @@ module amg_c_base_smoother_mod
& psb_c_vect_type, psb_c_base_vect_type, psb_spk_, &
& amg_c_base_smoother_type, psb_ipk_, psb_i_base_vect_type
! Arguments
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_base_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -170,7 +170,7 @@ module amg_c_base_solver_mod
Implicit None
! Arguments
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_base_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+2 -2
View File
@@ -119,7 +119,7 @@ module amg_c_diag_solver
import :: psb_desc_type, psb_cspmat_type, psb_c_base_sparse_mat, &
& psb_c_vect_type, psb_c_base_vect_type, psb_spk_, &
& amg_c_diag_solver_type, psb_ipk_, psb_i_base_vect_type
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_diag_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
@@ -331,7 +331,7 @@ module amg_c_l1_diag_solver
import :: psb_desc_type, psb_cspmat_type, psb_c_base_sparse_mat, &
& psb_c_vect_type, psb_c_base_vect_type, psb_spk_, &
& amg_c_l1_diag_solver_type, psb_ipk_, psb_i_base_vect_type
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_l1_diag_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+2 -2
View File
@@ -181,7 +181,7 @@ module amg_c_gs_solver
& psb_cspmat_type, psb_c_base_sparse_mat, psb_c_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_gs_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
@@ -195,7 +195,7 @@ module amg_c_gs_solver
& psb_cspmat_type, psb_c_base_sparse_mat, psb_c_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_bwgs_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
-125
View File
@@ -1,125 +0,0 @@
!
!
! AMG4PSBLAS version 1.0
! Algebraic Multigrid Package
! based on PSBLAS (Parallel Sparse BLAS version 3.7)
!
! (C) Copyright 2020
!
! Salvatore Filippone
! Pasqua D'Ambra
! Fabio Durastante
!
! Redistribution and use in source and binary forms, with or without
! modification, are permitted provided that the following conditions
! are met:
! 1. Redistributions of source code must retain the above copyright
! notice, this list of conditions and the following disclaimer.
! 2. Redistributions in binary form must reproduce the above copyright
! notice, this list of conditions, and the following disclaimer in the
! documentation and/or other materials provided with the distribution.
! 3. The name of the AMG4PSBLAS group or the names of its contributors may
! not be used to endorse or promote products derived from this
! software without specific written permission.
!
! THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
! ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
! TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
! PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
! BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
! CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
! SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
! INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
! CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
! ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
! POSSIBILITY OF SUCH DAMAGE.
!
!
!
!
! The aggregator object hosts the aggregation method for building
! the multilevel hierarchy. This variant is based on the hybrid method
! presented in
!
! S. Gratton, P. Henon, P. Jiranek and X. Vasseur:
! Reducing complexity of algebraic multigrid by aggregation
! Numerical Lin. Algebra with Applications, 2016, 23:501-518
!
module amg_c_hybrid_aggregator_mod
use amg_c_dec_aggregator_mod
!
! sm - class(amg_T_base_smoother_type), allocatable
! The current level preconditioner (aka smoother).
! parms - type(amg_RTml_parms)
! The parameters defining the multilevel strategy.
! ac - The local part of the current-level matrix, built by
! coarsening the previous-level matrix.
! desc_ac - type(psb_desc_type).
! The communication descriptor associated to the matrix
! stored in ac.
! base_a - type(psb_Tspmat_type), pointer.
! Pointer (really a pointer!) to the local part of the current
! matrix (so we have a unified treatment of residuals).
! We need this to avoid passing explicitly the current matrix
! to the routine which applies the preconditioner.
! base_desc - type(psb_desc_type), pointer.
! Pointer to the communication descriptor associated to the
! matrix pointed by base_a.
! map - Stores the maps (restriction and prolongation) between the
! vector spaces associated to the index spaces of the previous
! and current levels.
!
! Methods:
! Most methods follow the encapsulation hierarchy: they take whatever action
! is appropriate for the current object, then call the corresponding method for
! the contained object.
! As an example: the descr() method prints out a description of the
! level. It starts by invoking the descr() method of the parms object,
! then calls the descr() method of the smoother object.
!
! descr - Prints a description of the object.
! default - Set default values
! dump - Dump to file object contents
! set - Sets various parameters; when a request is unknown
! it is passed to the smoother object for further processing.
! check - Sanity checks.
! sizeof - Total memory occupation in bytes
! get_nzeros - Number of nonzeros
!
!
type, extends(amg_c_dec_aggregator_type) :: amg_c_hybrid_aggregator_type
contains
procedure, pass(ag) :: bld_tprol => amg_c_hybrid_aggregator_build_tprol
procedure, nopass :: fmt => amg_c_hybrid_aggregator_fmt
end type amg_c_hybrid_aggregator_type
interface
subroutine amg_c_hybrid_aggregator_build_tprol(ag,parms,a,desc_a,ilaggr,nlaggr,op_prol,info)
import :: amg_c_hybrid_aggregator_type, psb_desc_type, psb_cspmat_type, psb_spk_, &
& psb_ipk_, psb_long_int_k_, amg_sml_parms
implicit none
class(amg_c_hybrid_aggregator_type), target, intent(inout) :: ag
type(amg_sml_parms), intent(inout) :: parms
type(psb_cspmat_type), intent(in) :: a
type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), allocatable, intent(out) :: ilaggr(:),nlaggr(:)
type(psb_cspmat_type), intent(out) :: op_prol
integer(psb_ipk_), intent(out) :: info
end subroutine amg_c_hybrid_aggregator_build_tprol
end interface
contains
function amg_c_hybrid_aggregator_fmt() result(val)
implicit none
character(len=32) :: val
val = "Hybrid Decoupled aggregation"
end function amg_c_hybrid_aggregator_fmt
end module amg_c_hybrid_aggregator_mod
+1 -1
View File
@@ -123,7 +123,7 @@ contains
Implicit None
! Arguments
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_id_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -144,7 +144,7 @@ module amg_c_ilu_solver
& psb_cspmat_type, psb_c_base_sparse_mat, psb_c_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_ilu_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+2 -1
View File
@@ -109,11 +109,12 @@ module amg_c_inner_mod
end interface amg_map_to_tprol
abstract interface
subroutine amg_caggrmat_var_bld(a,desc_a,ilaggr,nlaggr,parms,&
subroutine amg_caggrmat_var_bld(dol1smoothing,a,desc_a,ilaggr,nlaggr,parms,&
& ac,desc_ac,op_prol,op_restr,t_prol,info)
import :: psb_cspmat_type, psb_desc_type, psb_spk_, psb_ipk_, psb_lpk_, psb_lcspmat_type
import :: amg_c_onelev_type, amg_sml_parms
implicit none
integer(psb_ipk_), intent(in) :: dol1smoothing
type(psb_cspmat_type), intent(in) :: a
type(psb_desc_type), intent(inout) :: desc_a
integer(psb_lpk_), intent(inout) :: ilaggr(:), nlaggr(:)
+4 -4
View File
@@ -79,9 +79,9 @@ module amg_c_invk_solver
import :: psb_desc_type, psb_cspmat_type, psb_c_base_sparse_mat, &
& amg_c_base_solver_type, psb_spk_, amg_c_invk_solver_type, psb_ipk_
Implicit None
class(amg_c_invk_solver_type), intent(inout) :: sv
class(amg_c_base_solver_type), allocatable, intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
class(amg_c_invk_solver_type), intent(inout) :: sv
class(amg_c_base_solver_type), intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_c_invk_solver_clone_settings
end interface
@@ -94,7 +94,7 @@ module amg_c_invk_solver
Implicit None
! Arguments
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_invk_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+4 -4
View File
@@ -79,9 +79,9 @@ module amg_c_invt_solver
import :: psb_desc_type, psb_cspmat_type, psb_c_base_sparse_mat, &
& amg_c_base_solver_type, psb_spk_, amg_c_invt_solver_type, psb_ipk_
Implicit None
class(amg_c_invt_solver_type), intent(inout) :: sv
class(amg_c_base_solver_type), allocatable, intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
class(amg_c_invt_solver_type), intent(inout) :: sv
class(amg_c_base_solver_type), intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_c_invt_solver_clone_settings
end interface
@@ -94,7 +94,7 @@ module amg_c_invt_solver
Implicit None
! Arguments
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_invt_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+4 -4
View File
@@ -151,7 +151,7 @@ module amg_c_jac_smoother
import :: psb_desc_type, amg_c_jac_smoother_type, psb_c_vect_type, psb_spk_, &
& psb_cspmat_type, psb_c_base_sparse_mat, psb_c_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_jac_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
@@ -203,8 +203,8 @@ module amg_c_jac_smoother
subroutine amg_c_jac_smoother_clone_settings(sm,smout,info)
import :: amg_c_jac_smoother_type, psb_spk_, &
& amg_c_base_smoother_type, psb_ipk_
class(amg_c_jac_smoother_type), intent(inout) :: sm
class(amg_c_base_smoother_type), allocatable, intent(inout) :: smout
class(amg_c_jac_smoother_type), intent(inout) :: sm
class(amg_c_base_smoother_type), intent(inout) :: smout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_c_jac_smoother_clone_settings
end interface
@@ -274,7 +274,7 @@ module amg_c_jac_smoother
import :: psb_desc_type, amg_c_l1_jac_smoother_type, psb_c_vect_type, &
& psb_cspmat_type, psb_c_base_sparse_mat, psb_c_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_l1_jac_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
+2 -2
View File
@@ -143,7 +143,7 @@ module amg_c_jac_solver
& psb_cspmat_type, psb_c_base_sparse_mat, psb_c_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_jac_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
@@ -160,7 +160,7 @@ module amg_c_jac_solver
& psb_cspmat_type, psb_c_base_sparse_mat, psb_c_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_l1_jac_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -174,7 +174,7 @@ module amg_c_krm_solver
& psb_cspmat_type, psb_c_base_sparse_mat, psb_c_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_krm_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -163,7 +163,7 @@ module amg_c_mumps_solver
Implicit None
! Arguments
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_mumps_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -256,7 +256,7 @@ contains
Implicit None
! Arguments
type(psb_cspmat_type), intent(inout), target :: a
type(psb_cspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_c_slu_solver_type), intent(inout) :: sv
integer, intent(out) :: info
+4 -7
View File
@@ -62,9 +62,6 @@ module amg_d_ainv_solver
procedure, pass(sv) :: cseti => amg_d_ainv_solver_cseti
procedure, pass(sv) :: csetc => amg_d_ainv_solver_csetc
procedure, pass(sv) :: csetr => amg_d_ainv_solver_csetr
!!$ procedure, pass(sv) :: seti => amg_d_ainv_solver_seti
!!$ procedure, pass(sv) :: setc => amg_d_ainv_solver_setc
!!$ procedure, pass(sv) :: setr => amg_d_ainv_solver_setr
procedure, pass(sv) :: descr => amg_d_ainv_solver_descr
procedure, pass(sv) :: default => d_ainv_solver_default
procedure, nopass :: stringval => d_ainv_stringval
@@ -91,9 +88,9 @@ module amg_d_ainv_solver
import :: psb_desc_type, psb_dspmat_type, psb_d_base_sparse_mat, &
& amg_d_base_solver_type, psb_dpk_, amg_d_ainv_solver_type, psb_ipk_
Implicit None
class(amg_d_ainv_solver_type), intent(inout) :: sv
class(amg_d_base_solver_type), allocatable, intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
class(amg_d_ainv_solver_type), intent(inout) :: sv
class(amg_d_base_solver_type), intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_d_ainv_solver_clone_settings
end interface
@@ -106,7 +103,7 @@ module amg_d_ainv_solver
Implicit None
! Arguments
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_ainv_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -230,7 +230,7 @@ module amg_d_as_smoother
& psb_desc_type, psb_d_base_sparse_mat, psb_ipk_,&
& psb_i_base_vect_type
implicit none
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_as_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -237,7 +237,7 @@ module amg_d_base_smoother_mod
& psb_d_vect_type, psb_d_base_vect_type, psb_dpk_, &
& amg_d_base_smoother_type, psb_ipk_, psb_i_base_vect_type
! Arguments
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_base_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -170,7 +170,7 @@ module amg_d_base_solver_mod
Implicit None
! Arguments
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_base_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+2 -2
View File
@@ -119,7 +119,7 @@ module amg_d_diag_solver
import :: psb_desc_type, psb_dspmat_type, psb_d_base_sparse_mat, &
& psb_d_vect_type, psb_d_base_vect_type, psb_dpk_, &
& amg_d_diag_solver_type, psb_ipk_, psb_i_base_vect_type
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_diag_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
@@ -331,7 +331,7 @@ module amg_d_l1_diag_solver
import :: psb_desc_type, psb_dspmat_type, psb_d_base_sparse_mat, &
& psb_d_vect_type, psb_d_base_vect_type, psb_dpk_, &
& amg_d_l1_diag_solver_type, psb_ipk_, psb_i_base_vect_type
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_l1_diag_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+2 -2
View File
@@ -181,7 +181,7 @@ module amg_d_gs_solver
& psb_dspmat_type, psb_d_base_sparse_mat, psb_d_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_gs_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
@@ -195,7 +195,7 @@ module amg_d_gs_solver
& psb_dspmat_type, psb_d_base_sparse_mat, psb_d_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_bwgs_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
-125
View File
@@ -1,125 +0,0 @@
!
!
! AMG4PSBLAS version 1.0
! Algebraic Multigrid Package
! based on PSBLAS (Parallel Sparse BLAS version 3.7)
!
! (C) Copyright 2020
!
! Salvatore Filippone
! Pasqua D'Ambra
! Fabio Durastante
!
! Redistribution and use in source and binary forms, with or without
! modification, are permitted provided that the following conditions
! are met:
! 1. Redistributions of source code must retain the above copyright
! notice, this list of conditions and the following disclaimer.
! 2. Redistributions in binary form must reproduce the above copyright
! notice, this list of conditions, and the following disclaimer in the
! documentation and/or other materials provided with the distribution.
! 3. The name of the AMG4PSBLAS group or the names of its contributors may
! not be used to endorse or promote products derived from this
! software without specific written permission.
!
! THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
! ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
! TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
! PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
! BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
! CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
! SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
! INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
! CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
! ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
! POSSIBILITY OF SUCH DAMAGE.
!
!
!
!
! The aggregator object hosts the aggregation method for building
! the multilevel hierarchy. This variant is based on the hybrid method
! presented in
!
! S. Gratton, P. Henon, P. Jiranek and X. Vasseur:
! Reducing complexity of algebraic multigrid by aggregation
! Numerical Lin. Algebra with Applications, 2016, 23:501-518
!
module amg_d_hybrid_aggregator_mod
use amg_d_dec_aggregator_mod
!
! sm - class(amg_T_base_smoother_type), allocatable
! The current level preconditioner (aka smoother).
! parms - type(amg_RTml_parms)
! The parameters defining the multilevel strategy.
! ac - The local part of the current-level matrix, built by
! coarsening the previous-level matrix.
! desc_ac - type(psb_desc_type).
! The communication descriptor associated to the matrix
! stored in ac.
! base_a - type(psb_Tspmat_type), pointer.
! Pointer (really a pointer!) to the local part of the current
! matrix (so we have a unified treatment of residuals).
! We need this to avoid passing explicitly the current matrix
! to the routine which applies the preconditioner.
! base_desc - type(psb_desc_type), pointer.
! Pointer to the communication descriptor associated to the
! matrix pointed by base_a.
! map - Stores the maps (restriction and prolongation) between the
! vector spaces associated to the index spaces of the previous
! and current levels.
!
! Methods:
! Most methods follow the encapsulation hierarchy: they take whatever action
! is appropriate for the current object, then call the corresponding method for
! the contained object.
! As an example: the descr() method prints out a description of the
! level. It starts by invoking the descr() method of the parms object,
! then calls the descr() method of the smoother object.
!
! descr - Prints a description of the object.
! default - Set default values
! dump - Dump to file object contents
! set - Sets various parameters; when a request is unknown
! it is passed to the smoother object for further processing.
! check - Sanity checks.
! sizeof - Total memory occupation in bytes
! get_nzeros - Number of nonzeros
!
!
type, extends(amg_d_dec_aggregator_type) :: amg_d_hybrid_aggregator_type
contains
procedure, pass(ag) :: bld_tprol => amg_d_hybrid_aggregator_build_tprol
procedure, nopass :: fmt => amg_d_hybrid_aggregator_fmt
end type amg_d_hybrid_aggregator_type
interface
subroutine amg_d_hybrid_aggregator_build_tprol(ag,parms,a,desc_a,ilaggr,nlaggr,op_prol,info)
import :: amg_d_hybrid_aggregator_type, psb_desc_type, psb_dspmat_type, psb_dpk_, &
& psb_ipk_, psb_long_int_k_, amg_dml_parms
implicit none
class(amg_d_hybrid_aggregator_type), target, intent(inout) :: ag
type(amg_dml_parms), intent(inout) :: parms
type(psb_dspmat_type), intent(in) :: a
type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), allocatable, intent(out) :: ilaggr(:),nlaggr(:)
type(psb_dspmat_type), intent(out) :: op_prol
integer(psb_ipk_), intent(out) :: info
end subroutine amg_d_hybrid_aggregator_build_tprol
end interface
contains
function amg_d_hybrid_aggregator_fmt() result(val)
implicit none
character(len=32) :: val
val = "Hybrid Decoupled aggregation"
end function amg_d_hybrid_aggregator_fmt
end module amg_d_hybrid_aggregator_mod
+1 -1
View File
@@ -123,7 +123,7 @@ contains
Implicit None
! Arguments
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_id_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -144,7 +144,7 @@ module amg_d_ilu_solver
& psb_dspmat_type, psb_d_base_sparse_mat, psb_d_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_ilu_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+2 -1
View File
@@ -109,11 +109,12 @@ module amg_d_inner_mod
end interface amg_map_to_tprol
abstract interface
subroutine amg_daggrmat_var_bld(a,desc_a,ilaggr,nlaggr,parms,&
subroutine amg_daggrmat_var_bld(dol1smoothing,a,desc_a,ilaggr,nlaggr,parms,&
& ac,desc_ac,op_prol,op_restr,t_prol,info)
import :: psb_dspmat_type, psb_desc_type, psb_dpk_, psb_ipk_, psb_lpk_, psb_ldspmat_type
import :: amg_d_onelev_type, amg_dml_parms
implicit none
integer(psb_ipk_), intent(in) :: dol1smoothing
type(psb_dspmat_type), intent(in) :: a
type(psb_desc_type), intent(inout) :: desc_a
integer(psb_lpk_), intent(inout) :: ilaggr(:), nlaggr(:)
+4 -4
View File
@@ -79,9 +79,9 @@ module amg_d_invk_solver
import :: psb_desc_type, psb_dspmat_type, psb_d_base_sparse_mat, &
& amg_d_base_solver_type, psb_dpk_, amg_d_invk_solver_type, psb_ipk_
Implicit None
class(amg_d_invk_solver_type), intent(inout) :: sv
class(amg_d_base_solver_type), allocatable, intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
class(amg_d_invk_solver_type), intent(inout) :: sv
class(amg_d_base_solver_type), intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_d_invk_solver_clone_settings
end interface
@@ -94,7 +94,7 @@ module amg_d_invk_solver
Implicit None
! Arguments
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_invk_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+4 -4
View File
@@ -79,9 +79,9 @@ module amg_d_invt_solver
import :: psb_desc_type, psb_dspmat_type, psb_d_base_sparse_mat, &
& amg_d_base_solver_type, psb_dpk_, amg_d_invt_solver_type, psb_ipk_
Implicit None
class(amg_d_invt_solver_type), intent(inout) :: sv
class(amg_d_base_solver_type), allocatable, intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
class(amg_d_invt_solver_type), intent(inout) :: sv
class(amg_d_base_solver_type), intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_d_invt_solver_clone_settings
end interface
@@ -94,7 +94,7 @@ module amg_d_invt_solver
Implicit None
! Arguments
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_invt_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+4 -4
View File
@@ -151,7 +151,7 @@ module amg_d_jac_smoother
import :: psb_desc_type, amg_d_jac_smoother_type, psb_d_vect_type, psb_dpk_, &
& psb_dspmat_type, psb_d_base_sparse_mat, psb_d_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_jac_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
@@ -203,8 +203,8 @@ module amg_d_jac_smoother
subroutine amg_d_jac_smoother_clone_settings(sm,smout,info)
import :: amg_d_jac_smoother_type, psb_dpk_, &
& amg_d_base_smoother_type, psb_ipk_
class(amg_d_jac_smoother_type), intent(inout) :: sm
class(amg_d_base_smoother_type), allocatable, intent(inout) :: smout
class(amg_d_jac_smoother_type), intent(inout) :: sm
class(amg_d_base_smoother_type), intent(inout) :: smout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_d_jac_smoother_clone_settings
end interface
@@ -274,7 +274,7 @@ module amg_d_jac_smoother
import :: psb_desc_type, amg_d_l1_jac_smoother_type, psb_d_vect_type, &
& psb_dspmat_type, psb_d_base_sparse_mat, psb_d_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_l1_jac_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
+2 -2
View File
@@ -143,7 +143,7 @@ module amg_d_jac_solver
& psb_dspmat_type, psb_d_base_sparse_mat, psb_d_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_jac_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
@@ -160,7 +160,7 @@ module amg_d_jac_solver
& psb_dspmat_type, psb_d_base_sparse_mat, psb_d_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_l1_jac_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -174,7 +174,7 @@ module amg_d_krm_solver
& psb_dspmat_type, psb_d_base_sparse_mat, psb_d_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_krm_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -163,7 +163,7 @@ module amg_d_mumps_solver
Implicit None
! Arguments
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_mumps_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+4 -2
View File
@@ -244,11 +244,12 @@ module amg_d_parmatch_aggregator_mod
end interface
interface
subroutine amg_d_parmatch_unsmth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
subroutine amg_d_parmatch_unsmth_bld(dol1smoothing,ag,a,desc_a,ilaggr,nlaggr,parms,&
& ac,desc_ac,op_prol,op_restr,t_prol,info)
import :: amg_d_parmatch_aggregator_type, psb_desc_type, psb_dspmat_type,&
& psb_ldspmat_type, psb_dpk_, psb_ipk_, psb_lpk_, amg_dml_parms, amg_daggr_data
implicit none
integer(psb_ipk_), intent(in) :: dol1smoothing
class(amg_d_parmatch_aggregator_type), target, intent(inout) :: ag
type(psb_dspmat_type), intent(in) :: a
type(psb_desc_type), intent(inout) :: desc_a
@@ -262,11 +263,12 @@ module amg_d_parmatch_aggregator_mod
end interface
interface
subroutine amg_d_parmatch_smth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
subroutine amg_d_parmatch_smth_bld(dol1smoothing,ag,a,desc_a,ilaggr,nlaggr,parms,&
& ac,desc_ac,op_prol,op_restr,t_prol,info)
import :: amg_d_parmatch_aggregator_type, psb_desc_type, psb_dspmat_type,&
& psb_ldspmat_type, psb_dpk_, psb_ipk_, psb_lpk_, amg_dml_parms, amg_daggr_data
implicit none
integer(psb_ipk_), intent(in) :: dol1smoothing
class(amg_d_parmatch_aggregator_type), target, intent(inout) :: ag
type(psb_dspmat_type), intent(in) :: a
type(psb_desc_type), intent(inout) :: desc_a
+4 -4
View File
@@ -140,7 +140,7 @@ module amg_d_poly_smoother
import :: psb_desc_type, amg_d_poly_smoother_type, psb_d_vect_type, psb_dpk_, &
& psb_dspmat_type, psb_d_base_sparse_mat, psb_d_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_poly_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
@@ -192,8 +192,8 @@ module amg_d_poly_smoother
subroutine amg_d_poly_smoother_clone_settings(sm,smout,info)
import :: amg_d_poly_smoother_type, psb_dpk_, &
& amg_d_base_smoother_type, psb_ipk_
class(amg_d_poly_smoother_type), intent(inout) :: sm
class(amg_d_base_smoother_type), allocatable, intent(inout) :: smout
class(amg_d_poly_smoother_type), intent(inout) :: sm
class(amg_d_base_smoother_type), intent(inout) :: smout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_d_poly_smoother_clone_settings
end interface
@@ -322,7 +322,7 @@ contains
!
sm%pdegree = 1
sm%rho_ba = -done
sm%variant = amg_poly_lottes_
sm%variant = amg_cheb_4_
sm%rho_estimate = amg_poly_rho_est_power_
sm%rho_estimate_iterations = 20
if (allocated(sm%sv)) then
+1 -1
View File
@@ -256,7 +256,7 @@ contains
Implicit None
! Arguments
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_slu_solver_type), intent(inout) :: sv
integer, intent(out) :: info
+1 -1
View File
@@ -259,7 +259,7 @@ contains
Implicit None
! Arguments
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_sludist_solver_type), intent(inout) :: sv
integer, intent(out) :: info
+1 -1
View File
@@ -259,7 +259,7 @@ contains
Implicit None
! Arguments
type(psb_dspmat_type), intent(inout), target :: a
type(psb_dspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_d_umf_solver_type), intent(inout) :: sv
integer, intent(out) :: info
+4 -7
View File
@@ -62,9 +62,6 @@ module amg_s_ainv_solver
procedure, pass(sv) :: cseti => amg_s_ainv_solver_cseti
procedure, pass(sv) :: csetc => amg_s_ainv_solver_csetc
procedure, pass(sv) :: csetr => amg_s_ainv_solver_csetr
!!$ procedure, pass(sv) :: seti => amg_s_ainv_solver_seti
!!$ procedure, pass(sv) :: setc => amg_s_ainv_solver_setc
!!$ procedure, pass(sv) :: setr => amg_s_ainv_solver_setr
procedure, pass(sv) :: descr => amg_s_ainv_solver_descr
procedure, pass(sv) :: default => s_ainv_solver_default
procedure, nopass :: stringval => s_ainv_stringval
@@ -91,9 +88,9 @@ module amg_s_ainv_solver
import :: psb_desc_type, psb_sspmat_type, psb_s_base_sparse_mat, &
& amg_s_base_solver_type, psb_dpk_, amg_s_ainv_solver_type, psb_ipk_
Implicit None
class(amg_s_ainv_solver_type), intent(inout) :: sv
class(amg_s_base_solver_type), allocatable, intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
class(amg_s_ainv_solver_type), intent(inout) :: sv
class(amg_s_base_solver_type), intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_s_ainv_solver_clone_settings
end interface
@@ -106,7 +103,7 @@ module amg_s_ainv_solver
Implicit None
! Arguments
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_ainv_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -230,7 +230,7 @@ module amg_s_as_smoother
& psb_desc_type, psb_s_base_sparse_mat, psb_ipk_,&
& psb_i_base_vect_type
implicit none
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_as_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -237,7 +237,7 @@ module amg_s_base_smoother_mod
& psb_s_vect_type, psb_s_base_vect_type, psb_spk_, &
& amg_s_base_smoother_type, psb_ipk_, psb_i_base_vect_type
! Arguments
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_base_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -170,7 +170,7 @@ module amg_s_base_solver_mod
Implicit None
! Arguments
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_base_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+2 -2
View File
@@ -119,7 +119,7 @@ module amg_s_diag_solver
import :: psb_desc_type, psb_sspmat_type, psb_s_base_sparse_mat, &
& psb_s_vect_type, psb_s_base_vect_type, psb_spk_, &
& amg_s_diag_solver_type, psb_ipk_, psb_i_base_vect_type
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_diag_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
@@ -331,7 +331,7 @@ module amg_s_l1_diag_solver
import :: psb_desc_type, psb_sspmat_type, psb_s_base_sparse_mat, &
& psb_s_vect_type, psb_s_base_vect_type, psb_spk_, &
& amg_s_l1_diag_solver_type, psb_ipk_, psb_i_base_vect_type
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_l1_diag_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+2 -2
View File
@@ -181,7 +181,7 @@ module amg_s_gs_solver
& psb_sspmat_type, psb_s_base_sparse_mat, psb_s_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_gs_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
@@ -195,7 +195,7 @@ module amg_s_gs_solver
& psb_sspmat_type, psb_s_base_sparse_mat, psb_s_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_bwgs_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
-125
View File
@@ -1,125 +0,0 @@
!
!
! AMG4PSBLAS version 1.0
! Algebraic Multigrid Package
! based on PSBLAS (Parallel Sparse BLAS version 3.7)
!
! (C) Copyright 2020
!
! Salvatore Filippone
! Pasqua D'Ambra
! Fabio Durastante
!
! Redistribution and use in source and binary forms, with or without
! modification, are permitted provided that the following conditions
! are met:
! 1. Redistributions of source code must retain the above copyright
! notice, this list of conditions and the following disclaimer.
! 2. Redistributions in binary form must reproduce the above copyright
! notice, this list of conditions, and the following disclaimer in the
! documentation and/or other materials provided with the distribution.
! 3. The name of the AMG4PSBLAS group or the names of its contributors may
! not be used to endorse or promote products derived from this
! software without specific written permission.
!
! THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
! ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
! TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
! PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
! BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
! CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
! SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
! INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
! CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
! ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
! POSSIBILITY OF SUCH DAMAGE.
!
!
!
!
! The aggregator object hosts the aggregation method for building
! the multilevel hierarchy. This variant is based on the hybrid method
! presented in
!
! S. Gratton, P. Henon, P. Jiranek and X. Vasseur:
! Reducing complexity of algebraic multigrid by aggregation
! Numerical Lin. Algebra with Applications, 2016, 23:501-518
!
module amg_s_hybrid_aggregator_mod
use amg_s_dec_aggregator_mod
!
! sm - class(amg_T_base_smoother_type), allocatable
! The current level preconditioner (aka smoother).
! parms - type(amg_RTml_parms)
! The parameters defining the multilevel strategy.
! ac - The local part of the current-level matrix, built by
! coarsening the previous-level matrix.
! desc_ac - type(psb_desc_type).
! The communication descriptor associated to the matrix
! stored in ac.
! base_a - type(psb_Tspmat_type), pointer.
! Pointer (really a pointer!) to the local part of the current
! matrix (so we have a unified treatment of residuals).
! We need this to avoid passing explicitly the current matrix
! to the routine which applies the preconditioner.
! base_desc - type(psb_desc_type), pointer.
! Pointer to the communication descriptor associated to the
! matrix pointed by base_a.
! map - Stores the maps (restriction and prolongation) between the
! vector spaces associated to the index spaces of the previous
! and current levels.
!
! Methods:
! Most methods follow the encapsulation hierarchy: they take whatever action
! is appropriate for the current object, then call the corresponding method for
! the contained object.
! As an example: the descr() method prints out a description of the
! level. It starts by invoking the descr() method of the parms object,
! then calls the descr() method of the smoother object.
!
! descr - Prints a description of the object.
! default - Set default values
! dump - Dump to file object contents
! set - Sets various parameters; when a request is unknown
! it is passed to the smoother object for further processing.
! check - Sanity checks.
! sizeof - Total memory occupation in bytes
! get_nzeros - Number of nonzeros
!
!
type, extends(amg_s_dec_aggregator_type) :: amg_s_hybrid_aggregator_type
contains
procedure, pass(ag) :: bld_tprol => amg_s_hybrid_aggregator_build_tprol
procedure, nopass :: fmt => amg_s_hybrid_aggregator_fmt
end type amg_s_hybrid_aggregator_type
interface
subroutine amg_s_hybrid_aggregator_build_tprol(ag,parms,a,desc_a,ilaggr,nlaggr,op_prol,info)
import :: amg_s_hybrid_aggregator_type, psb_desc_type, psb_sspmat_type, psb_spk_, &
& psb_ipk_, psb_long_int_k_, amg_sml_parms
implicit none
class(amg_s_hybrid_aggregator_type), target, intent(inout) :: ag
type(amg_sml_parms), intent(inout) :: parms
type(psb_sspmat_type), intent(in) :: a
type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), allocatable, intent(out) :: ilaggr(:),nlaggr(:)
type(psb_sspmat_type), intent(out) :: op_prol
integer(psb_ipk_), intent(out) :: info
end subroutine amg_s_hybrid_aggregator_build_tprol
end interface
contains
function amg_s_hybrid_aggregator_fmt() result(val)
implicit none
character(len=32) :: val
val = "Hybrid Decoupled aggregation"
end function amg_s_hybrid_aggregator_fmt
end module amg_s_hybrid_aggregator_mod
+1 -1
View File
@@ -123,7 +123,7 @@ contains
Implicit None
! Arguments
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_id_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -144,7 +144,7 @@ module amg_s_ilu_solver
& psb_sspmat_type, psb_s_base_sparse_mat, psb_s_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_ilu_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+2 -1
View File
@@ -109,11 +109,12 @@ module amg_s_inner_mod
end interface amg_map_to_tprol
abstract interface
subroutine amg_saggrmat_var_bld(a,desc_a,ilaggr,nlaggr,parms,&
subroutine amg_saggrmat_var_bld(dol1smoothing,a,desc_a,ilaggr,nlaggr,parms,&
& ac,desc_ac,op_prol,op_restr,t_prol,info)
import :: psb_sspmat_type, psb_desc_type, psb_spk_, psb_ipk_, psb_lpk_, psb_lsspmat_type
import :: amg_s_onelev_type, amg_sml_parms
implicit none
integer(psb_ipk_), intent(in) :: dol1smoothing
type(psb_sspmat_type), intent(in) :: a
type(psb_desc_type), intent(inout) :: desc_a
integer(psb_lpk_), intent(inout) :: ilaggr(:), nlaggr(:)
+4 -4
View File
@@ -79,9 +79,9 @@ module amg_s_invk_solver
import :: psb_desc_type, psb_sspmat_type, psb_s_base_sparse_mat, &
& amg_s_base_solver_type, psb_spk_, amg_s_invk_solver_type, psb_ipk_
Implicit None
class(amg_s_invk_solver_type), intent(inout) :: sv
class(amg_s_base_solver_type), allocatable, intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
class(amg_s_invk_solver_type), intent(inout) :: sv
class(amg_s_base_solver_type), intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_s_invk_solver_clone_settings
end interface
@@ -94,7 +94,7 @@ module amg_s_invk_solver
Implicit None
! Arguments
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_invk_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+4 -4
View File
@@ -79,9 +79,9 @@ module amg_s_invt_solver
import :: psb_desc_type, psb_sspmat_type, psb_s_base_sparse_mat, &
& amg_s_base_solver_type, psb_spk_, amg_s_invt_solver_type, psb_ipk_
Implicit None
class(amg_s_invt_solver_type), intent(inout) :: sv
class(amg_s_base_solver_type), allocatable, intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
class(amg_s_invt_solver_type), intent(inout) :: sv
class(amg_s_base_solver_type), intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_s_invt_solver_clone_settings
end interface
@@ -94,7 +94,7 @@ module amg_s_invt_solver
Implicit None
! Arguments
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_invt_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+4 -4
View File
@@ -151,7 +151,7 @@ module amg_s_jac_smoother
import :: psb_desc_type, amg_s_jac_smoother_type, psb_s_vect_type, psb_spk_, &
& psb_sspmat_type, psb_s_base_sparse_mat, psb_s_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_jac_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
@@ -203,8 +203,8 @@ module amg_s_jac_smoother
subroutine amg_s_jac_smoother_clone_settings(sm,smout,info)
import :: amg_s_jac_smoother_type, psb_spk_, &
& amg_s_base_smoother_type, psb_ipk_
class(amg_s_jac_smoother_type), intent(inout) :: sm
class(amg_s_base_smoother_type), allocatable, intent(inout) :: smout
class(amg_s_jac_smoother_type), intent(inout) :: sm
class(amg_s_base_smoother_type), intent(inout) :: smout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_s_jac_smoother_clone_settings
end interface
@@ -274,7 +274,7 @@ module amg_s_jac_smoother
import :: psb_desc_type, amg_s_l1_jac_smoother_type, psb_s_vect_type, &
& psb_sspmat_type, psb_s_base_sparse_mat, psb_s_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_l1_jac_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
+2 -2
View File
@@ -143,7 +143,7 @@ module amg_s_jac_solver
& psb_sspmat_type, psb_s_base_sparse_mat, psb_s_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_jac_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
@@ -160,7 +160,7 @@ module amg_s_jac_solver
& psb_sspmat_type, psb_s_base_sparse_mat, psb_s_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_l1_jac_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -174,7 +174,7 @@ module amg_s_krm_solver
& psb_sspmat_type, psb_s_base_sparse_mat, psb_s_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_krm_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -163,7 +163,7 @@ module amg_s_mumps_solver
Implicit None
! Arguments
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_mumps_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+4 -2
View File
@@ -244,11 +244,12 @@ module amg_s_parmatch_aggregator_mod
end interface
interface
subroutine amg_s_parmatch_unsmth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
subroutine amg_s_parmatch_unsmth_bld(dol1smoothing,ag,a,desc_a,ilaggr,nlaggr,parms,&
& ac,desc_ac,op_prol,op_restr,t_prol,info)
import :: amg_s_parmatch_aggregator_type, psb_desc_type, psb_sspmat_type,&
& psb_lsspmat_type, psb_dpk_, psb_ipk_, psb_lpk_, amg_sml_parms, amg_saggr_data
implicit none
integer(psb_ipk_), intent(in) :: dol1smoothing
class(amg_s_parmatch_aggregator_type), target, intent(inout) :: ag
type(psb_sspmat_type), intent(in) :: a
type(psb_desc_type), intent(inout) :: desc_a
@@ -262,11 +263,12 @@ module amg_s_parmatch_aggregator_mod
end interface
interface
subroutine amg_s_parmatch_smth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
subroutine amg_s_parmatch_smth_bld(dol1smoothing,ag,a,desc_a,ilaggr,nlaggr,parms,&
& ac,desc_ac,op_prol,op_restr,t_prol,info)
import :: amg_s_parmatch_aggregator_type, psb_desc_type, psb_sspmat_type,&
& psb_lsspmat_type, psb_dpk_, psb_ipk_, psb_lpk_, amg_sml_parms, amg_saggr_data
implicit none
integer(psb_ipk_), intent(in) :: dol1smoothing
class(amg_s_parmatch_aggregator_type), target, intent(inout) :: ag
type(psb_sspmat_type), intent(in) :: a
type(psb_desc_type), intent(inout) :: desc_a
+4 -4
View File
@@ -140,7 +140,7 @@ module amg_s_poly_smoother
import :: psb_desc_type, amg_s_poly_smoother_type, psb_s_vect_type, psb_spk_, &
& psb_sspmat_type, psb_s_base_sparse_mat, psb_s_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_poly_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
@@ -192,8 +192,8 @@ module amg_s_poly_smoother
subroutine amg_s_poly_smoother_clone_settings(sm,smout,info)
import :: amg_s_poly_smoother_type, psb_spk_, &
& amg_s_base_smoother_type, psb_ipk_
class(amg_s_poly_smoother_type), intent(inout) :: sm
class(amg_s_base_smoother_type), allocatable, intent(inout) :: smout
class(amg_s_poly_smoother_type), intent(inout) :: sm
class(amg_s_base_smoother_type), intent(inout) :: smout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_s_poly_smoother_clone_settings
end interface
@@ -322,7 +322,7 @@ contains
!
sm%pdegree = 1
sm%rho_ba = -sone
sm%variant = amg_poly_lottes_
sm%variant = amg_cheb_4_
sm%rho_estimate = amg_poly_rho_est_power_
sm%rho_estimate_iterations = 20
if (allocated(sm%sv)) then
+1 -1
View File
@@ -256,7 +256,7 @@ contains
Implicit None
! Arguments
type(psb_sspmat_type), intent(inout), target :: a
type(psb_sspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_s_slu_solver_type), intent(inout) :: sv
integer, intent(out) :: info
+4 -7
View File
@@ -62,9 +62,6 @@ module amg_z_ainv_solver
procedure, pass(sv) :: cseti => amg_z_ainv_solver_cseti
procedure, pass(sv) :: csetc => amg_z_ainv_solver_csetc
procedure, pass(sv) :: csetr => amg_z_ainv_solver_csetr
!!$ procedure, pass(sv) :: seti => amg_z_ainv_solver_seti
!!$ procedure, pass(sv) :: setc => amg_z_ainv_solver_setc
!!$ procedure, pass(sv) :: setr => amg_z_ainv_solver_setr
procedure, pass(sv) :: descr => amg_z_ainv_solver_descr
procedure, pass(sv) :: default => z_ainv_solver_default
procedure, nopass :: stringval => z_ainv_stringval
@@ -91,9 +88,9 @@ module amg_z_ainv_solver
import :: psb_desc_type, psb_zspmat_type, psb_z_base_sparse_mat, &
& amg_z_base_solver_type, psb_dpk_, amg_z_ainv_solver_type, psb_ipk_
Implicit None
class(amg_z_ainv_solver_type), intent(inout) :: sv
class(amg_z_base_solver_type), allocatable, intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
class(amg_z_ainv_solver_type), intent(inout) :: sv
class(amg_z_base_solver_type), intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_z_ainv_solver_clone_settings
end interface
@@ -106,7 +103,7 @@ module amg_z_ainv_solver
Implicit None
! Arguments
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_ainv_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -230,7 +230,7 @@ module amg_z_as_smoother
& psb_desc_type, psb_z_base_sparse_mat, psb_ipk_,&
& psb_i_base_vect_type
implicit none
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_as_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -237,7 +237,7 @@ module amg_z_base_smoother_mod
& psb_z_vect_type, psb_z_base_vect_type, psb_dpk_, &
& amg_z_base_smoother_type, psb_ipk_, psb_i_base_vect_type
! Arguments
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_base_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -170,7 +170,7 @@ module amg_z_base_solver_mod
Implicit None
! Arguments
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_base_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+2 -2
View File
@@ -119,7 +119,7 @@ module amg_z_diag_solver
import :: psb_desc_type, psb_zspmat_type, psb_z_base_sparse_mat, &
& psb_z_vect_type, psb_z_base_vect_type, psb_dpk_, &
& amg_z_diag_solver_type, psb_ipk_, psb_i_base_vect_type
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_diag_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
@@ -331,7 +331,7 @@ module amg_z_l1_diag_solver
import :: psb_desc_type, psb_zspmat_type, psb_z_base_sparse_mat, &
& psb_z_vect_type, psb_z_base_vect_type, psb_dpk_, &
& amg_z_l1_diag_solver_type, psb_ipk_, psb_i_base_vect_type
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_l1_diag_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+2 -2
View File
@@ -181,7 +181,7 @@ module amg_z_gs_solver
& psb_zspmat_type, psb_z_base_sparse_mat, psb_z_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_gs_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
@@ -195,7 +195,7 @@ module amg_z_gs_solver
& psb_zspmat_type, psb_z_base_sparse_mat, psb_z_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_bwgs_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
-125
View File
@@ -1,125 +0,0 @@
!
!
! AMG4PSBLAS version 1.0
! Algebraic Multigrid Package
! based on PSBLAS (Parallel Sparse BLAS version 3.7)
!
! (C) Copyright 2020
!
! Salvatore Filippone
! Pasqua D'Ambra
! Fabio Durastante
!
! Redistribution and use in source and binary forms, with or without
! modification, are permitted provided that the following conditions
! are met:
! 1. Redistributions of source code must retain the above copyright
! notice, this list of conditions and the following disclaimer.
! 2. Redistributions in binary form must reproduce the above copyright
! notice, this list of conditions, and the following disclaimer in the
! documentation and/or other materials provided with the distribution.
! 3. The name of the AMG4PSBLAS group or the names of its contributors may
! not be used to endorse or promote products derived from this
! software without specific written permission.
!
! THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
! ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
! TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
! PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AMG4PSBLAS GROUP OR ITS CONTRIBUTORS
! BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
! CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
! SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
! INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
! CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
! ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
! POSSIBILITY OF SUCH DAMAGE.
!
!
!
!
! The aggregator object hosts the aggregation method for building
! the multilevel hierarchy. This variant is based on the hybrid method
! presented in
!
! S. Gratton, P. Henon, P. Jiranek and X. Vasseur:
! Reducing complexity of algebraic multigrid by aggregation
! Numerical Lin. Algebra with Applications, 2016, 23:501-518
!
module amg_z_hybrid_aggregator_mod
use amg_z_dec_aggregator_mod
!
! sm - class(amg_T_base_smoother_type), allocatable
! The current level preconditioner (aka smoother).
! parms - type(amg_RTml_parms)
! The parameters defining the multilevel strategy.
! ac - The local part of the current-level matrix, built by
! coarsening the previous-level matrix.
! desc_ac - type(psb_desc_type).
! The communication descriptor associated to the matrix
! stored in ac.
! base_a - type(psb_Tspmat_type), pointer.
! Pointer (really a pointer!) to the local part of the current
! matrix (so we have a unified treatment of residuals).
! We need this to avoid passing explicitly the current matrix
! to the routine which applies the preconditioner.
! base_desc - type(psb_desc_type), pointer.
! Pointer to the communication descriptor associated to the
! matrix pointed by base_a.
! map - Stores the maps (restriction and prolongation) between the
! vector spaces associated to the index spaces of the previous
! and current levels.
!
! Methods:
! Most methods follow the encapsulation hierarchy: they take whatever action
! is appropriate for the current object, then call the corresponding method for
! the contained object.
! As an example: the descr() method prints out a description of the
! level. It starts by invoking the descr() method of the parms object,
! then calls the descr() method of the smoother object.
!
! descr - Prints a description of the object.
! default - Set default values
! dump - Dump to file object contents
! set - Sets various parameters; when a request is unknown
! it is passed to the smoother object for further processing.
! check - Sanity checks.
! sizeof - Total memory occupation in bytes
! get_nzeros - Number of nonzeros
!
!
type, extends(amg_z_dec_aggregator_type) :: amg_z_hybrid_aggregator_type
contains
procedure, pass(ag) :: bld_tprol => amg_z_hybrid_aggregator_build_tprol
procedure, nopass :: fmt => amg_z_hybrid_aggregator_fmt
end type amg_z_hybrid_aggregator_type
interface
subroutine amg_z_hybrid_aggregator_build_tprol(ag,parms,a,desc_a,ilaggr,nlaggr,op_prol,info)
import :: amg_z_hybrid_aggregator_type, psb_desc_type, psb_zspmat_type, psb_dpk_, &
& psb_ipk_, psb_long_int_k_, amg_dml_parms
implicit none
class(amg_z_hybrid_aggregator_type), target, intent(inout) :: ag
type(amg_dml_parms), intent(inout) :: parms
type(psb_zspmat_type), intent(in) :: a
type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), allocatable, intent(out) :: ilaggr(:),nlaggr(:)
type(psb_zspmat_type), intent(out) :: op_prol
integer(psb_ipk_), intent(out) :: info
end subroutine amg_z_hybrid_aggregator_build_tprol
end interface
contains
function amg_z_hybrid_aggregator_fmt() result(val)
implicit none
character(len=32) :: val
val = "Hybrid Decoupled aggregation"
end function amg_z_hybrid_aggregator_fmt
end module amg_z_hybrid_aggregator_mod
+1 -1
View File
@@ -123,7 +123,7 @@ contains
Implicit None
! Arguments
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_id_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -144,7 +144,7 @@ module amg_z_ilu_solver
& psb_zspmat_type, psb_z_base_sparse_mat, psb_z_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_ilu_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+2 -1
View File
@@ -109,11 +109,12 @@ module amg_z_inner_mod
end interface amg_map_to_tprol
abstract interface
subroutine amg_zaggrmat_var_bld(a,desc_a,ilaggr,nlaggr,parms,&
subroutine amg_zaggrmat_var_bld(dol1smoothing,a,desc_a,ilaggr,nlaggr,parms,&
& ac,desc_ac,op_prol,op_restr,t_prol,info)
import :: psb_zspmat_type, psb_desc_type, psb_dpk_, psb_ipk_, psb_lpk_, psb_lzspmat_type
import :: amg_z_onelev_type, amg_dml_parms
implicit none
integer(psb_ipk_), intent(in) :: dol1smoothing
type(psb_zspmat_type), intent(in) :: a
type(psb_desc_type), intent(inout) :: desc_a
integer(psb_lpk_), intent(inout) :: ilaggr(:), nlaggr(:)
+4 -4
View File
@@ -79,9 +79,9 @@ module amg_z_invk_solver
import :: psb_desc_type, psb_zspmat_type, psb_z_base_sparse_mat, &
& amg_z_base_solver_type, psb_dpk_, amg_z_invk_solver_type, psb_ipk_
Implicit None
class(amg_z_invk_solver_type), intent(inout) :: sv
class(amg_z_base_solver_type), allocatable, intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
class(amg_z_invk_solver_type), intent(inout) :: sv
class(amg_z_base_solver_type), intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_z_invk_solver_clone_settings
end interface
@@ -94,7 +94,7 @@ module amg_z_invk_solver
Implicit None
! Arguments
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_invk_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+4 -4
View File
@@ -79,9 +79,9 @@ module amg_z_invt_solver
import :: psb_desc_type, psb_zspmat_type, psb_z_base_sparse_mat, &
& amg_z_base_solver_type, psb_dpk_, amg_z_invt_solver_type, psb_ipk_
Implicit None
class(amg_z_invt_solver_type), intent(inout) :: sv
class(amg_z_base_solver_type), allocatable, intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
class(amg_z_invt_solver_type), intent(inout) :: sv
class(amg_z_base_solver_type), intent(inout) :: svout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_z_invt_solver_clone_settings
end interface
@@ -94,7 +94,7 @@ module amg_z_invt_solver
Implicit None
! Arguments
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_invt_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+4 -4
View File
@@ -151,7 +151,7 @@ module amg_z_jac_smoother
import :: psb_desc_type, amg_z_jac_smoother_type, psb_z_vect_type, psb_dpk_, &
& psb_zspmat_type, psb_z_base_sparse_mat, psb_z_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_jac_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
@@ -203,8 +203,8 @@ module amg_z_jac_smoother
subroutine amg_z_jac_smoother_clone_settings(sm,smout,info)
import :: amg_z_jac_smoother_type, psb_dpk_, &
& amg_z_base_smoother_type, psb_ipk_
class(amg_z_jac_smoother_type), intent(inout) :: sm
class(amg_z_base_smoother_type), allocatable, intent(inout) :: smout
class(amg_z_jac_smoother_type), intent(inout) :: sm
class(amg_z_base_smoother_type), intent(inout) :: smout
integer(psb_ipk_), intent(out) :: info
end subroutine amg_z_jac_smoother_clone_settings
end interface
@@ -274,7 +274,7 @@ module amg_z_jac_smoother
import :: psb_desc_type, amg_z_l1_jac_smoother_type, psb_z_vect_type, &
& psb_zspmat_type, psb_z_base_sparse_mat, psb_z_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_l1_jac_smoother_type), intent(inout) :: sm
integer(psb_ipk_), intent(out) :: info
+2 -2
View File
@@ -143,7 +143,7 @@ module amg_z_jac_solver
& psb_zspmat_type, psb_z_base_sparse_mat, psb_z_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_jac_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
@@ -160,7 +160,7 @@ module amg_z_jac_solver
& psb_zspmat_type, psb_z_base_sparse_mat, psb_z_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_l1_jac_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -174,7 +174,7 @@ module amg_z_krm_solver
& psb_zspmat_type, psb_z_base_sparse_mat, psb_z_base_vect_type,&
& psb_ipk_, psb_i_base_vect_type
implicit none
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_krm_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -163,7 +163,7 @@ module amg_z_mumps_solver
Implicit None
! Arguments
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_mumps_solver_type), intent(inout) :: sv
integer(psb_ipk_), intent(out) :: info
+1 -1
View File
@@ -256,7 +256,7 @@ contains
Implicit None
! Arguments
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_slu_solver_type), intent(inout) :: sv
integer, intent(out) :: info
+1 -1
View File
@@ -259,7 +259,7 @@ contains
Implicit None
! Arguments
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_sludist_solver_type), intent(inout) :: sv
integer, intent(out) :: info
+1 -1
View File
@@ -259,7 +259,7 @@ contains
Implicit None
! Arguments
type(psb_zspmat_type), intent(inout), target :: a
type(psb_zspmat_type), intent(in), target :: a
Type(psb_desc_type), Intent(inout) :: desc_a
class(amg_z_umf_solver_type), intent(inout) :: sv
integer, intent(out) :: info
+28 -18
View File
@@ -67,14 +67,13 @@ void dMatchBoxPC(MilanLongInt NLVer, MilanLongInt NLEdge,
#endif
#define TIME_TRACKER
#ifdef TIME_TRACKER
double tmr = MPI_Wtime();
#endif
#undef TIME_TRACKER
#ifdef TIME_TRACKER
double tmr = MPI_Wtime();
#endif
// Rimosso per tornare al vecchio matching #define OMP
#ifdef OMP
fprintf(stderr,"Warning: using buggy OpenMP matching!\n");
#if defined(OPENMP)
//fprintf(stderr,"Warning: using buggy OpenMP matching!\n");
dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(NLVer, NLEdge,
verLocPtr, verLocInd, edgeLocWeight,
verDistance, Mate,
@@ -93,11 +92,11 @@ void dMatchBoxPC(MilanLongInt NLVer, MilanLongInt NLEdge,
#endif
#ifdef TIME_TRACKER
tmr = MPI_Wtime() - tmr;
fprintf(stderr, "Elaboration time: %f for %ld nodes\n", tmr, NLVer);
#endif
#ifdef TIME_TRACKER
tmr = MPI_Wtime() - tmr;
fprintf(stderr, "Elaboration time: %f for %ld nodes\n", tmr, NLVer);
#endif
#endif
}
@@ -115,13 +114,24 @@ void sMatchBoxPC(MilanLongInt NLVer, MilanLongInt NLEdge,
fprintf(stderr,"MatchBoxPC: rank %d nlver %ld nledge %ld [ %ld %ld ]\n",
myRank,NLVer, NLEdge,verDistance[0],verDistance[1]);
#endif
#if defined(OPENMP)
//fprintf(stderr,"Warning: using buggy OpenMP matching!\n");
salgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(NLVer, NLEdge,
verLocPtr, verLocInd, edgeLocWeight,
verDistance, Mate,
myRank, numProcs, C_comm,
msgIndSent, msgActualSent, msgPercent,
ph0_time, ph1_time, ph2_time,
ph1_card, ph2_card );
#else
salgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateC(NLVer, NLEdge,
verLocPtr, verLocInd, edgeLocWeight,
verDistance, Mate,
myRank, numProcs, C_comm,
msgIndSent, msgActualSent, msgPercent,
ph0_time, ph1_time, ph2_time,
ph1_card, ph2_card );
verLocPtr, verLocInd, edgeLocWeight,
verDistance, Mate,
myRank, numProcs, C_comm,
msgIndSent, msgActualSent, msgPercent,
ph0_time, ph1_time, ph2_time,
ph1_card, ph2_card );
#endif
#endif
}
+405 -250
View File
@@ -59,7 +59,7 @@
#include <assert.h>
#include <map>
#include <vector>
#ifdef OMP
#ifdef OPENMP
// OpenMP is included and used if and only if the OpenMP version of the matching
// is required
#include "omp.h"
@@ -82,6 +82,8 @@ const int BundleTag = 9; // Predefined tag
static vector<MilanLongInt> DEFAULT_VECTOR;
#if !defined(SERIAL_MPI)
// MPI type map
template <typename T>
MPI_Datatype TypeMap();
@@ -93,6 +95,7 @@ template <>
inline MPI_Datatype TypeMap<double>() { return MPI_DOUBLE; }
template <>
inline MPI_Datatype TypeMap<float>() { return MPI_FLOAT; }
#endif
#ifdef __cplusplus
extern "C"
@@ -178,264 +181,416 @@ extern "C"
#define MilanRealMin MINUS_INFINITY
#endif
#ifdef OMP
/* These functions are only used in the experimental OMP implementation, if that
is disabled there is no reason to actually compile or reference them. */
// Function of find the owner of a ghost vertex using binary search:
MilanInt findOwnerOfGhost(MilanLongInt vtxIndex, MilanLongInt *mVerDistance,
MilanInt myRank, MilanInt numProcs);
// Function of find the owner of a ghost vertex using binary search:
MilanInt findOwnerOfGhost(MilanLongInt vtxIndex, MilanLongInt *mVerDistance,
MilanInt myRank, MilanInt numProcs);
MilanLongInt firstComputeCandidateMateD(MilanLongInt adj1,
MilanLongInt adj2,
MilanLongInt *verLocInd,
MilanReal *edgeLocWeight);
MilanLongInt firstComputeCandidateMate(MilanLongInt adj1,
MilanLongInt adj2,
MilanLongInt *verLocInd,
MilanReal *edgeLocWeight);
void queuesTransfer(vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
vector<MilanLongInt> &privateQLocalVtx,
vector<MilanLongInt> &privateQGhostVtx,
vector<MilanLongInt> &privateQMsgType,
vector<MilanInt> &privateQOwner);
bool isAlreadyMatched(MilanLongInt node,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
vector<MilanLongInt> &GMate,
MilanLongInt *Mate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap);
MilanLongInt computeCandidateMateD(MilanLongInt adj1,
MilanLongInt adj2,
MilanReal *edgeLocWeight,
MilanLongInt k,
MilanLongInt *verLocInd,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
vector<MilanLongInt> &GMate,
MilanLongInt *Mate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap);
void initialize(MilanLongInt NLVer, MilanLongInt NLEdge,
MilanLongInt StartIndex, MilanLongInt EndIndex,
MilanLongInt *numGhostEdgesPtr,
MilanLongInt *numGhostVerticesPtr,
MilanLongInt *S,
MilanLongInt *verLocInd,
MilanLongInt *verLocPtr,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
vector<MilanLongInt> &Counter,
vector<MilanLongInt> &verGhostPtr,
vector<MilanLongInt> &verGhostInd,
vector<MilanLongInt> &tempCounter,
vector<MilanLongInt> &GMate,
vector<MilanLongInt> &Message,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
MilanLongInt *&candidateMate,
vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU,
vector<MilanLongInt> &privateQLocalVtx,
vector<MilanLongInt> &privateQGhostVtx,
vector<MilanLongInt> &privateQMsgType,
vector<MilanInt> &privateQOwner);
void clean(MilanLongInt NLVer,
MilanInt myRank,
MilanLongInt MessageIndex,
vector<MPI_Request> &SRequest,
vector<MPI_Status> &SStatus,
MilanInt BufferSize,
MilanLongInt *Buffer,
MilanLongInt msgActual,
MilanLongInt *msgActualSent,
MilanLongInt msgInd,
MilanLongInt *msgIndSent,
MilanLongInt NumMessagesBundled,
MilanReal *msgPercent);
void PARALLEL_COMPUTE_CANDIDATE_MATE_BD(MilanLongInt NLVer,
MilanLongInt *verLocPtr,
MilanLongInt *verLocInd,
MilanInt myRank,
MilanReal *edgeLocWeight,
MilanLongInt *candidateMate);
void PARALLEL_PROCESS_EXPOSED_VERTEX_BD(MilanLongInt NLVer,
MilanLongInt *candidateMate,
MilanLongInt *verLocInd,
MilanLongInt *verLocPtr,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
MilanLongInt *Mate,
vector<MilanLongInt> &GMate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
MilanReal *edgeLocWeight,
MilanLongInt *myCardPtr,
MilanLongInt *msgIndPtr,
MilanLongInt *NumMessagesBundledPtr,
MilanLongInt *SPtr,
MilanLongInt *verDistance,
MilanLongInt *PCounter,
vector<MilanLongInt> &Counter,
MilanInt myRank,
MilanInt numProcs,
vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
vector<MilanLongInt> &privateQLocalVtx,
vector<MilanLongInt> &privateQGhostVtx,
vector<MilanLongInt> &privateQMsgType,
vector<MilanInt> &privateQOwner);
void PROCESS_CROSS_EDGE(MilanLongInt *edge,
MilanLongInt *SPtr);
void processMatchedVerticesD(
MilanLongInt NLVer,
vector<MilanLongInt> &UChunkBeingProcessed,
vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
MilanLongInt *myCardPtr,
MilanLongInt *msgIndPtr,
MilanLongInt *NumMessagesBundledPtr,
MilanLongInt *SPtr,
MilanLongInt *verLocPtr,
MilanLongInt *verLocInd,
MilanLongInt *verDistance,
MilanLongInt *PCounter,
vector<MilanLongInt> &Counter,
MilanInt myRank,
MilanInt numProcs,
MilanLongInt *candidateMate,
vector<MilanLongInt> &GMate,
MilanLongInt *Mate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
MilanReal *edgeLocWeight,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
vector<MilanLongInt> &privateQLocalVtx,
vector<MilanLongInt> &privateQGhostVtx,
vector<MilanLongInt> &privateQMsgType,
vector<MilanInt> &privateQOwner);
void processMatchedVerticesAndSendMessagesD(
MilanLongInt NLVer,
vector<MilanLongInt> &UChunkBeingProcessed,
vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
MilanLongInt *myCardPtr,
MilanLongInt *msgIndPtr,
MilanLongInt *NumMessagesBundledPtr,
MilanLongInt *SPtr,
MilanLongInt *verLocPtr,
MilanLongInt *verLocInd,
MilanLongInt *verDistance,
MilanLongInt *PCounter,
vector<MilanLongInt> &Counter,
MilanInt myRank,
MilanInt numProcs,
MilanLongInt *candidateMate,
vector<MilanLongInt> &GMate,
MilanLongInt *Mate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
MilanReal *edgeLocWeight,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
vector<MilanLongInt> &privateQLocalVtx,
vector<MilanLongInt> &privateQGhostVtx,
vector<MilanLongInt> &privateQMsgType,
vector<MilanInt> &privateQOwner,
MPI_Comm comm,
MilanLongInt *msgActual,
vector<MilanLongInt> &Message);
void sendBundledMessages(MilanLongInt *numGhostEdgesPtr,
MilanInt *BufferSizePtr,
MilanLongInt *Buffer,
vector<MilanLongInt> &PCumulative,
vector<MilanLongInt> &PMessageBundle,
vector<MilanLongInt> &PSizeInfoMessages,
MilanLongInt *PCounter,
MilanLongInt NumMessagesBundled,
MilanLongInt *msgActualPtr,
MilanLongInt *MessageIndexPtr,
MilanInt numProcs,
MilanInt myRank,
MPI_Comm comm,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
vector<MPI_Request> &SRequest,
vector<MPI_Status> &SStatus);
void processMessagesD(
MilanLongInt NLVer,
MilanLongInt *Mate,
MilanLongInt *candidateMate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
vector<MilanLongInt> &GMate,
vector<MilanLongInt> &Counter,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
MilanLongInt *myCardPtr,
MilanLongInt *msgIndPtr,
MilanLongInt *msgActualPtr,
MilanReal *edgeLocWeight,
MilanLongInt *verDistance,
MilanLongInt *verLocPtr,
MilanLongInt k,
MilanLongInt *verLocInd,
MilanInt numProcs,
MilanInt myRank,
MPI_Comm comm,
vector<MilanLongInt> &Message,
MilanLongInt numGhostEdges,
MilanLongInt u,
MilanLongInt v,
MilanLongInt *SPtr,
vector<MilanLongInt> &U);
void extractUChunk(
vector<MilanLongInt> &UChunkBeingProcessed,
vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU);
void queuesTransfer(vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
vector<MilanLongInt> &privateQLocalVtx,
vector<MilanLongInt> &privateQGhostVtx,
vector<MilanLongInt> &privateQMsgType,
vector<MilanInt> &privateQOwner);
MilanLongInt firstComputeCandidateMateS(MilanLongInt adj1,
MilanLongInt adj2,
MilanLongInt *verLocInd,
MilanFloat *edgeLocWeight);
bool isAlreadyMatched(MilanLongInt node,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
vector<MilanLongInt> &GMate,
MilanLongInt *Mate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap);
MilanLongInt computeCandidateMateS(MilanLongInt adj1,
MilanLongInt adj2,
MilanFloat *edgeLocWeight,
MilanLongInt k,
MilanLongInt *verLocInd,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
vector<MilanLongInt> &GMate,
MilanLongInt *Mate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap);
void PARALLEL_COMPUTE_CANDIDATE_MATE_BS(MilanLongInt NLVer,
MilanLongInt *verLocPtr,
MilanLongInt *verLocInd,
MilanInt myRank,
MilanFloat *edgeLocWeight,
MilanLongInt *candidateMate);
void PARALLEL_PROCESS_EXPOSED_VERTEX_BS(MilanLongInt NLVer,
MilanLongInt *candidateMate,
MilanLongInt *verLocInd,
MilanLongInt *verLocPtr,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
MilanLongInt *Mate,
vector<MilanLongInt> &GMate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
MilanFloat *edgeLocWeight,
MilanLongInt *myCardPtr,
MilanLongInt *msgIndPtr,
MilanLongInt *NumMessagesBundledPtr,
MilanLongInt *SPtr,
MilanLongInt *verDistance,
MilanLongInt *PCounter,
vector<MilanLongInt> &Counter,
MilanInt myRank,
MilanInt numProcs,
vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
vector<MilanLongInt> &privateQLocalVtx,
vector<MilanLongInt> &privateQGhostVtx,
vector<MilanLongInt> &privateQMsgType,
vector<MilanInt> &privateQOwner);
void processMatchedVerticesS(
MilanLongInt NLVer,
vector<MilanLongInt> &UChunkBeingProcessed,
vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
MilanLongInt *myCardPtr,
MilanLongInt *msgIndPtr,
MilanLongInt *NumMessagesBundledPtr,
MilanLongInt *SPtr,
MilanLongInt *verLocPtr,
MilanLongInt *verLocInd,
MilanLongInt *verDistance,
MilanLongInt *PCounter,
vector<MilanLongInt> &Counter,
MilanInt myRank,
MilanInt numProcs,
MilanLongInt *candidateMate,
vector<MilanLongInt> &GMate,
MilanLongInt *Mate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
MilanFloat *edgeLocWeight,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
vector<MilanLongInt> &privateQLocalVtx,
vector<MilanLongInt> &privateQGhostVtx,
vector<MilanLongInt> &privateQMsgType,
vector<MilanInt> &privateQOwner);
void processMatchedVerticesAndSendMessagesS(
MilanLongInt NLVer,
vector<MilanLongInt> &UChunkBeingProcessed,
vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
MilanLongInt *myCardPtr,
MilanLongInt *msgIndPtr,
MilanLongInt *NumMessagesBundledPtr,
MilanLongInt *SPtr,
MilanLongInt *verLocPtr,
MilanLongInt *verLocInd,
MilanLongInt *verDistance,
MilanLongInt *PCounter,
vector<MilanLongInt> &Counter,
MilanInt myRank,
MilanInt numProcs,
MilanLongInt *candidateMate,
vector<MilanLongInt> &GMate,
MilanLongInt *Mate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
MilanFloat *edgeLocWeight,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
vector<MilanLongInt> &privateQLocalVtx,
vector<MilanLongInt> &privateQGhostVtx,
vector<MilanLongInt> &privateQMsgType,
vector<MilanInt> &privateQOwner,
MPI_Comm comm,
MilanLongInt *msgActual,
vector<MilanLongInt> &Message);
void processMessagesS(
MilanLongInt NLVer,
MilanLongInt *Mate,
MilanLongInt *candidateMate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
vector<MilanLongInt> &GMate,
vector<MilanLongInt> &Counter,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
MilanLongInt *myCardPtr,
MilanLongInt *msgIndPtr,
MilanLongInt *msgActualPtr,
MilanFloat *edgeLocWeight,
MilanLongInt *verDistance,
MilanLongInt *verLocPtr,
MilanLongInt k,
MilanLongInt *verLocInd,
MilanInt numProcs,
MilanInt myRank,
MPI_Comm comm,
vector<MilanLongInt> &Message,
MilanLongInt numGhostEdges,
MilanLongInt u,
MilanLongInt v,
MilanLongInt *SPtr,
vector<MilanLongInt> &U);
void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
MilanLongInt NLVer, MilanLongInt NLEdge,
MilanLongInt *verLocPtr, MilanLongInt *verLocInd, MilanReal *edgeLocWeight,
MilanLongInt *verDistance,
MilanLongInt *Mate,
MilanInt myRank, MilanInt numProcs, MPI_Comm comm,
MilanLongInt *msgIndSent, MilanLongInt *msgActualSent, MilanReal *msgPercent,
MilanReal *ph0_time, MilanReal *ph1_time, MilanReal *ph2_time,
MilanLongInt *ph1_card, MilanLongInt *ph2_card);
MilanLongInt computeCandidateMate(MilanLongInt adj1,
MilanLongInt adj2,
MilanReal *edgeLocWeight,
MilanLongInt k,
MilanLongInt *verLocInd,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
vector<MilanLongInt> &GMate,
MilanLongInt *Mate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap);
void initialize(MilanLongInt NLVer, MilanLongInt NLEdge,
MilanLongInt StartIndex, MilanLongInt EndIndex,
MilanLongInt *numGhostEdgesPtr,
MilanLongInt *numGhostVerticesPtr,
MilanLongInt *S,
MilanLongInt *verLocInd,
MilanLongInt *verLocPtr,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
vector<MilanLongInt> &Counter,
vector<MilanLongInt> &verGhostPtr,
vector<MilanLongInt> &verGhostInd,
vector<MilanLongInt> &tempCounter,
vector<MilanLongInt> &GMate,
vector<MilanLongInt> &Message,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
MilanLongInt *&candidateMate,
vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU,
vector<MilanLongInt> &privateQLocalVtx,
vector<MilanLongInt> &privateQGhostVtx,
vector<MilanLongInt> &privateQMsgType,
vector<MilanInt> &privateQOwner);
void clean(MilanLongInt NLVer,
MilanInt myRank,
MilanLongInt MessageIndex,
vector<MPI_Request> &SRequest,
vector<MPI_Status> &SStatus,
MilanInt BufferSize,
MilanLongInt *Buffer,
MilanLongInt msgActual,
MilanLongInt *msgActualSent,
MilanLongInt msgInd,
MilanLongInt *msgIndSent,
MilanLongInt NumMessagesBundled,
MilanReal *msgPercent);
void PARALLEL_COMPUTE_CANDIDATE_MATE_B(MilanLongInt NLVer,
MilanLongInt *verLocPtr,
MilanLongInt *verLocInd,
MilanInt myRank,
MilanReal *edgeLocWeight,
MilanLongInt *candidateMate);
void PARALLEL_PROCESS_EXPOSED_VERTEX_B(MilanLongInt NLVer,
MilanLongInt *candidateMate,
MilanLongInt *verLocInd,
MilanLongInt *verLocPtr,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
MilanLongInt *Mate,
vector<MilanLongInt> &GMate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
MilanReal *edgeLocWeight,
MilanLongInt *myCardPtr,
MilanLongInt *msgIndPtr,
MilanLongInt *NumMessagesBundledPtr,
MilanLongInt *SPtr,
MilanLongInt *verDistance,
MilanLongInt *PCounter,
vector<MilanLongInt> &Counter,
MilanInt myRank,
MilanInt numProcs,
vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
vector<MilanLongInt> &privateQLocalVtx,
vector<MilanLongInt> &privateQGhostVtx,
vector<MilanLongInt> &privateQMsgType,
vector<MilanInt> &privateQOwner);
void PROCESS_CROSS_EDGE(MilanLongInt *edge,
MilanLongInt *SPtr);
void processMatchedVertices(
MilanLongInt NLVer,
vector<MilanLongInt> &UChunkBeingProcessed,
vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
MilanLongInt *myCardPtr,
MilanLongInt *msgIndPtr,
MilanLongInt *NumMessagesBundledPtr,
MilanLongInt *SPtr,
MilanLongInt *verLocPtr,
MilanLongInt *verLocInd,
MilanLongInt *verDistance,
MilanLongInt *PCounter,
vector<MilanLongInt> &Counter,
MilanInt myRank,
MilanInt numProcs,
MilanLongInt *candidateMate,
vector<MilanLongInt> &GMate,
MilanLongInt *Mate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
MilanReal *edgeLocWeight,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
vector<MilanLongInt> &privateQLocalVtx,
vector<MilanLongInt> &privateQGhostVtx,
vector<MilanLongInt> &privateQMsgType,
vector<MilanInt> &privateQOwner);
void processMatchedVerticesAndSendMessages(
MilanLongInt NLVer,
vector<MilanLongInt> &UChunkBeingProcessed,
vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
MilanLongInt *myCardPtr,
MilanLongInt *msgIndPtr,
MilanLongInt *NumMessagesBundledPtr,
MilanLongInt *SPtr,
MilanLongInt *verLocPtr,
MilanLongInt *verLocInd,
MilanLongInt *verDistance,
MilanLongInt *PCounter,
vector<MilanLongInt> &Counter,
MilanInt myRank,
MilanInt numProcs,
MilanLongInt *candidateMate,
vector<MilanLongInt> &GMate,
MilanLongInt *Mate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
MilanReal *edgeLocWeight,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
vector<MilanLongInt> &privateQLocalVtx,
vector<MilanLongInt> &privateQGhostVtx,
vector<MilanLongInt> &privateQMsgType,
vector<MilanInt> &privateQOwner,
MPI_Comm comm,
MilanLongInt *msgActual,
vector<MilanLongInt> &Message);
void sendBundledMessages(MilanLongInt *numGhostEdgesPtr,
MilanInt *BufferSizePtr,
MilanLongInt *Buffer,
vector<MilanLongInt> &PCumulative,
vector<MilanLongInt> &PMessageBundle,
vector<MilanLongInt> &PSizeInfoMessages,
MilanLongInt *PCounter,
MilanLongInt NumMessagesBundled,
MilanLongInt *msgActualPtr,
MilanLongInt *MessageIndexPtr,
MilanInt numProcs,
MilanInt myRank,
MPI_Comm comm,
vector<MilanLongInt> &QLocalVtx,
vector<MilanLongInt> &QGhostVtx,
vector<MilanLongInt> &QMsgType,
vector<MilanInt> &QOwner,
vector<MPI_Request> &SRequest,
vector<MPI_Status> &SStatus);
void processMessages(
MilanLongInt NLVer,
MilanLongInt *Mate,
MilanLongInt *candidateMate,
map<MilanLongInt, MilanLongInt> &Ghost2LocalMap,
vector<MilanLongInt> &GMate,
vector<MilanLongInt> &Counter,
MilanLongInt StartIndex,
MilanLongInt EndIndex,
MilanLongInt *myCardPtr,
MilanLongInt *msgIndPtr,
MilanLongInt *msgActualPtr,
MilanReal *edgeLocWeight,
MilanLongInt *verDistance,
MilanLongInt *verLocPtr,
MilanLongInt k,
MilanLongInt *verLocInd,
MilanInt numProcs,
MilanInt myRank,
MPI_Comm comm,
vector<MilanLongInt> &Message,
MilanLongInt numGhostEdges,
MilanLongInt u,
MilanLongInt v,
MilanLongInt *SPtr,
vector<MilanLongInt> &U);
void extractUChunk(
vector<MilanLongInt> &UChunkBeingProcessed,
vector<MilanLongInt> &U,
vector<MilanLongInt> &privateU);
void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
MilanLongInt NLVer, MilanLongInt NLEdge,
MilanLongInt *verLocPtr, MilanLongInt *verLocInd, MilanReal *edgeLocWeight,
MilanLongInt *verDistance,
MilanLongInt *Mate,
MilanInt myRank, MilanInt numProcs, MPI_Comm comm,
MilanLongInt *msgIndSent, MilanLongInt *msgActualSent, MilanReal *msgPercent,
MilanReal *ph0_time, MilanReal *ph1_time, MilanReal *ph2_time,
MilanLongInt *ph1_card, MilanLongInt *ph2_card);
#endif
void salgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
MilanLongInt NLVer, MilanLongInt NLEdge,
MilanLongInt *verLocPtr, MilanLongInt *verLocInd, MilanFloat *edgeLocWeight,
MilanLongInt *verDistance,
MilanLongInt *Mate,
MilanInt myRank, MilanInt numProcs, MPI_Comm comm,
MilanLongInt *msgIndSent, MilanLongInt *msgActualSent, MilanReal *msgPercent,
MilanReal *ph0_time, MilanReal *ph1_time, MilanReal *ph2_time,
MilanLongInt *ph1_card, MilanLongInt *ph2_card);
#ifndef OMP
//Function of find the owner of a ghost vertex using binary search:
inline MilanInt findOwnerOfGhost(MilanLongInt vtxIndex, MilanLongInt *mVerDistance,
MilanInt myRank, MilanInt numProcs);
#endif
void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateC(
MilanLongInt NLVer, MilanLongInt NLEdge,
@@ -1303,16 +1303,16 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateC(
// SINGLE PRECISION VERSION
void salgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateC(
MilanLongInt NLVer, MilanLongInt NLEdge,
MilanLongInt* verLocPtr, MilanLongInt* verLocInd,
MilanFloat* edgeLocWeight,
MilanLongInt* verDistance,
MilanLongInt* Mate,
MilanInt myRank, MilanInt numProcs, MPI_Comm comm,
MilanLongInt* msgIndSent, MilanLongInt* msgActualSent,
MilanReal* msgPercent,
MilanReal* ph0_time, MilanReal* ph1_time, MilanReal* ph2_time,
MilanLongInt* ph1_card, MilanLongInt* ph2_card ) {
MilanLongInt NLVer, MilanLongInt NLEdge,
MilanLongInt* verLocPtr, MilanLongInt* verLocInd,
MilanFloat* edgeLocWeight,
MilanLongInt* verDistance,
MilanLongInt* Mate,
MilanInt myRank, MilanInt numProcs, MPI_Comm comm,
MilanLongInt* msgIndSent, MilanLongInt* msgActualSent,
MilanReal* msgPercent,
MilanReal* ph0_time, MilanReal* ph1_time, MilanReal* ph2_time,
MilanLongInt* ph1_card, MilanLongInt* ph2_card ) {
#if !defined(SERIAL_MPI)
#ifdef PRINT_DEBUG_INFO_
cout<<"\n("<<myRank<<")Within algoEdgeApproxDominatingEdgesLinearSearchMessageBundling()"; fflush(stdout);
@@ -1,5 +1,4 @@
#include "MatchBoxPC.h"
#ifdef OMP
// ***********************************************************************
//
// MatchboxP: A C++ library for approximate weighted matching
@@ -126,8 +125,10 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
fflush(stdout);
#endif
MilanLongInt StartIndex = verDistance[myRank]; // The starting vertex owned by the current rank
MilanLongInt EndIndex = verDistance[myRank + 1] - 1; // The ending vertex owned by the current rank
// The starting vertex owned by the current rank
MilanLongInt StartIndex = verDistance[myRank];
// The ending vertex owned by the current rank
MilanLongInt EndIndex = verDistance[myRank + 1] - 1;
MPI_Status computeStatus;
@@ -145,7 +146,8 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
// only one message will be sent in the initialization phase -
// one of: REQUEST/FAILURE/SUCCESS
vector<MilanLongInt> QLocalVtx, QGhostVtx, QMsgType;
vector<MilanInt> QOwner; // Changed by Fabio to be an integer, addresses needs to be integers!
// Changed by Fabio to be an integer, addresses needs to be integers!
vector<MilanInt> QOwner;
MilanLongInt *PCounter = new MilanLongInt[numProcs];
for (int i = 0; i < numProcs; i++)
@@ -153,7 +155,8 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
MilanLongInt NumMessagesBundled = 0;
// TODO when the last computational section will be refactored this could be eliminated
MilanInt ghostOwner = 0; // Changed by Fabio to be an integer, addresses needs to be integers!
// Changed by Fabio to be an integer, addresses needs to be integers!
MilanInt ghostOwner = 0;
MilanLongInt *candidateMate = nullptr;
#ifdef PRINT_DEBUG_INFO_
cout << "\n(" << myRank << ")NV: " << NLVer << " Edges: " << NLEdge;
@@ -168,9 +171,12 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
MilanLongInt myCard = 0;
// Build the Ghost Vertex Set: Vg
map<MilanLongInt, MilanLongInt> Ghost2LocalMap; // Map each ghost vertex to a local vertex
vector<MilanLongInt> Counter; // Store the edge count for each ghost vertex
MilanLongInt numGhostVertices = 0, numGhostEdges = 0; // Number of Ghost vertices
// Map each ghost vertex to a local vertex
map<MilanLongInt, MilanLongInt> Ghost2LocalMap;
// Store the edge count for each ghost vertex
vector<MilanLongInt> Counter;
// Number of Ghost vertices
MilanLongInt numGhostVertices = 0, numGhostEdges = 0;
#ifdef PRINT_DEBUG_INFO_
cout << "\n(" << myRank << ")About to compute Ghost Vertices...";
@@ -237,7 +243,7 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
* PARALLEL_COMPUTE_CANDIDATE_MATE_B is now totally parallel.
*/
PARALLEL_COMPUTE_CANDIDATE_MATE_B(NLVer,
PARALLEL_COMPUTE_CANDIDATE_MATE_BD(NLVer,
verLocPtr,
verLocInd,
myRank,
@@ -262,7 +268,7 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
* TODO: Test when it's actually more efficient to execute this code
* in parallel.
*/
PARALLEL_PROCESS_EXPOSED_VERTEX_B(NLVer,
PARALLEL_PROCESS_EXPOSED_VERTEX_BD(NLVer,
candidateMate,
verLocInd,
verLocPtr,
@@ -314,7 +320,7 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
vector<MilanLongInt> UChunkBeingProcessed;
UChunkBeingProcessed.reserve(UCHUNK);
processMatchedVertices(NLVer,
processMatchedVerticesD(NLVer,
UChunkBeingProcessed,
U,
privateU,
@@ -423,7 +429,7 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
/////////////////////////// PROCESS MATCHED VERTICES //////////////////////////////
///////////////////////////////////////////////////////////////////////////////////
processMatchedVerticesAndSendMessages(NLVer,
processMatchedVerticesAndSendMessagesD(NLVer,
UChunkBeingProcessed,
U,
privateU,
@@ -483,8 +489,8 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
///////////////////////////////////////////////////////////////////////////////////
/////////////////////////// PROCESS MESSAGES //////////////////////////////////////
///////////////////////////////////////////////////////////////////////////////////
processMessages(NLVer,
//startTime = MPI_Wtime();
processMessagesD(NLVer,
Mate,
candidateMate,
Ghost2LocalMap,
@@ -549,6 +555,488 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
*ph2_card = myCard; // Cardinality at the end of Phase-2
}
// End of algoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMate
void salgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
MilanLongInt NLVer, MilanLongInt NLEdge,
MilanLongInt *verLocPtr, MilanLongInt *verLocInd,
MilanFloat *edgeLocWeight,
MilanLongInt *verDistance,
MilanLongInt *Mate,
MilanInt myRank, MilanInt numProcs, MPI_Comm comm,
MilanLongInt *msgIndSent, MilanLongInt *msgActualSent,
MilanReal *msgPercent,
MilanReal *ph0_time, MilanReal *ph1_time, MilanReal *ph2_time,
MilanLongInt *ph1_card, MilanLongInt *ph2_card)
{
/*
* verDistance: it's a vector long as the number of processors.
* verDistance[i] contains the first node index of the i-th processor
* verDistance[i + 1] contains the last node index of the i-th processor
* NLVer: number of elements in the LocPtr
* NLEdge: number of edges assigned to the current processor
*
* Contains the portion of matrix assigned to the processor in
* Yale notation
* verLocInd: contains the positions on row of the matrix
* verLocPtr: i-th value is the position of the first element on the i-th row and
* i+1-th value is the position of the first element on the i+1-th row
*/
#if !defined(SERIAL_MPI)
#ifdef PRINT_DEBUG_INFO_
cout << "\n(" << myRank << ")Within algoEdgeApproxDominatingEdgesLinearSearchMessageBundling()";
fflush(stdout);
#endif
#ifdef PRINT_DEBUG_INFO_
cout << "\n(" << myRank << ") verDistance [" ;
for (int i = 0; i < numProcs; i++)
cout << verDistance[i] << "," << verDistance[i+1];
cout << "]\n";
fflush(stdout);
#endif
#ifdef DEBUG_HANG_
if (myRank == 0) {
cout << "\n(" << myRank << ") verDistance [" ;
for (int i = 0; i < numProcs; i++)
cout << verDistance[i] << "," ;
cout << verDistance[numProcs]<< "]\n";
}
fflush(stdout);
#endif
// The starting vertex owned by the current rank
MilanLongInt StartIndex = verDistance[myRank];
// The ending vertex owned by the current rank
MilanLongInt EndIndex = verDistance[myRank + 1] - 1;
MPI_Status computeStatus;
MilanLongInt msgActual = 0, msgInd = 0;
MilanFloat heaviestEdgeWt = 0.0f; // Assumes positive weight
MilanReal startTime, finishTime;
startTime = MPI_Wtime();
// Data structures for sending and receiving messages:
vector<MilanLongInt> Message; // [ u, v, message_type ]
Message.resize(3, -1);
// Data structures for Message Bundling:
// Although up to two messages can be sent along any cross edge,
// only one message will be sent in the initialization phase -
// one of: REQUEST/FAILURE/SUCCESS
vector<MilanLongInt> QLocalVtx, QGhostVtx, QMsgType;
// Changed by Fabio to be an integer, addresses needs to be integers!
vector<MilanInt> QOwner;
MilanLongInt *PCounter = new MilanLongInt[numProcs];
for (int i = 0; i < numProcs; i++)
PCounter[i] = 0;
MilanLongInt NumMessagesBundled = 0;
// TODO when the last computational section will be refactored this could be eliminated
// Changed by Fabio to be an integer, addresses needs to be integers!
MilanInt ghostOwner = 0;
MilanLongInt *candidateMate = nullptr;
#ifdef PRINT_DEBUG_INFO_
cout << "\n(" << myRank << ")NV: " << NLVer << " Edges: " << NLEdge;
fflush(stdout);
cout << "\n(" << myRank << ")StartIndex: " << StartIndex << " EndIndex: " << EndIndex;
fflush(stdout);
#endif
// Other Variables:
MilanLongInt u = -1, v = -1, w = -1, i = 0;
MilanLongInt k = -1, adj1 = -1, adj2 = -1;
MilanLongInt k1 = -1, adj11 = -1, adj12 = -1;
MilanLongInt myCard = 0;
// Build the Ghost Vertex Set: Vg
// Map each ghost vertex to a local vertex
map<MilanLongInt, MilanLongInt> Ghost2LocalMap;
// Store the edge count for each ghost vertex
vector<MilanLongInt> Counter;
// Number of Ghost vertices
MilanLongInt numGhostVertices = 0, numGhostEdges = 0;
#ifdef PRINT_DEBUG_INFO_
cout << "\n(" << myRank << ")About to compute Ghost Vertices...";
fflush(stdout);
#endif
#ifdef DEBUG_HANG_
if (myRank == 0)
cout << "\n(" << myRank << ")About to compute Ghost Vertices...";
fflush(stdout);
#endif
// Define Adjacency Lists for Ghost Vertices:
// cout<<"Building Ghost data structures ... \n\n";
vector<MilanLongInt> verGhostPtr, verGhostInd, tempCounter;
// Mate array for ghost vertices:
vector<MilanLongInt> GMate; // Proportional to the number of ghost vertices
MilanLongInt S;
MilanLongInt privateMyCard = 0;
vector<MilanLongInt> PCumulative, PMessageBundle, PSizeInfoMessages;
vector<MPI_Request> SRequest; // Requests that are used for each send message
vector<MPI_Status> SStatus; // Status of sent messages, used in MPI_Wait
MilanLongInt MessageIndex = 0; // Pointer for current message
MilanInt BufferSize;
MilanLongInt *Buffer;
vector<MilanLongInt> privateQLocalVtx, privateQGhostVtx, privateQMsgType;
vector<MilanInt> privateQOwner;
vector<MilanLongInt> U, privateU;
initialize(NLVer, NLEdge, StartIndex,
EndIndex, &numGhostEdges,
&numGhostVertices, &S,
verLocInd, verLocPtr,
Ghost2LocalMap, Counter,
verGhostPtr, verGhostInd,
tempCounter, GMate,
Message, QLocalVtx,
QGhostVtx, QMsgType, QOwner,
candidateMate, U,
privateU,
privateQLocalVtx,
privateQGhostVtx,
privateQMsgType,
privateQOwner);
finishTime = MPI_Wtime();
*ph0_time = finishTime - startTime; // Time taken for Phase-0: Initialization
#ifdef DEBUG_HANG_
cout << myRank << " Finished initialization" << endl;
fflush(stdout);
#endif
startTime = MPI_Wtime();
/////////////////////////////////////////////////////////////////////////////////////////
//////////////////////////////////// INITIALIZATION /////////////////////////////////////
/////////////////////////////////////////////////////////////////////////////////////////
// Compute the Initial Matching Set:
/*
* OMP PARALLEL_COMPUTE_CANDIDATE_MATE_B has been splitted from
* PARALLEL_PROCESS_EXPOSED_VERTEX_B in order to better parallelize
* the two.
* PARALLEL_COMPUTE_CANDIDATE_MATE_B is now totally parallel.
*/
PARALLEL_COMPUTE_CANDIDATE_MATE_BS(NLVer,
verLocPtr,
verLocInd,
myRank,
edgeLocWeight,
candidateMate);
#ifdef DEBUG_HANG_
cout << myRank << " Finished Exposed Vertex" << endl;
fflush(stdout);
#if 0
cout << myRank << " candidateMate after parallelCompute " <<endl;
for (int i=0; i<NLVer; i++) {
cout << candidateMate[i] << " " ;
}
cout << endl;
#endif
#endif
/*
* PARALLEL_PROCESS_EXPOSED_VERTEX_B
* TODO: write comment
*
* TODO: Test when it's actually more efficient to execute this code
* in parallel.
*/
PARALLEL_PROCESS_EXPOSED_VERTEX_BS(NLVer,
candidateMate,
verLocInd,
verLocPtr,
StartIndex,
EndIndex,
Mate,
GMate,
Ghost2LocalMap,
edgeLocWeight,
&myCard,
&msgInd,
&NumMessagesBundled,
&S,
verDistance,
PCounter,
Counter,
myRank,
numProcs,
U,
privateU,
QLocalVtx,
QGhostVtx,
QMsgType,
QOwner,
privateQLocalVtx,
privateQGhostVtx,
privateQMsgType,
privateQOwner);
tempCounter.clear(); // Do not need this any more
#ifdef DEBUG_HANG_
cout << myRank << " Finished Exposed Vertex" << endl;
fflush(stdout);
#if 0
cout << myRank << " Mate after Exposed Vertices " <<endl;
for (int i=0; i<NLVer; i++) {
cout << Mate[i] << " " ;
}
cout << endl;
#endif
#endif
///////////////////////////////////////////////////////////////////////////////////
/////////////////////////// PROCESS MATCHED VERTICES //////////////////////////////
///////////////////////////////////////////////////////////////////////////////////
// TODO what would be the optimal UCHUNK
vector<MilanLongInt> UChunkBeingProcessed;
UChunkBeingProcessed.reserve(UCHUNK);
processMatchedVerticesS(NLVer,
UChunkBeingProcessed,
U,
privateU,
StartIndex,
EndIndex,
&myCard,
&msgInd,
&NumMessagesBundled,
&S,
verLocPtr,
verLocInd,
verDistance,
PCounter,
Counter,
myRank,
numProcs,
candidateMate,
GMate,
Mate,
Ghost2LocalMap,
edgeLocWeight,
QLocalVtx,
QGhostVtx,
QMsgType,
QOwner,
privateQLocalVtx,
privateQGhostVtx,
privateQMsgType,
privateQOwner);
#ifdef DEBUG_HANG_
cout << myRank << " Finished Process Vertices" << endl;
fflush(stdout);
#if 0
cout << myRank << " Mate after Matched Vertices " <<endl;
for (int i=0; i<NLVer; i++) {
cout << Mate[i] << " " ;
}
cout << endl;
#endif
#endif
/////////////////////////////////////////////////////////////////////////////////////////
///////////////////////////// SEND BUNDLED MESSAGES /////////////////////////////////////
/////////////////////////////////////////////////////////////////////////////////////////
sendBundledMessages(&numGhostEdges,
&BufferSize,
Buffer,
PCumulative,
PMessageBundle,
PSizeInfoMessages,
PCounter,
NumMessagesBundled,
&msgActual,
&MessageIndex,
numProcs,
myRank,
comm,
QLocalVtx,
QGhostVtx,
QMsgType,
QOwner,
SRequest,
SStatus);
///////////////////////// END OF SEND BUNDLED MESSAGES //////////////////////////////////
finishTime = MPI_Wtime();
*ph1_time = finishTime - startTime; // Time taken for Phase-1
#ifdef DEBUG_HANG_
cout << myRank << " Finished sendBundles" << endl;
fflush(stdout);
#endif
*ph1_card = myCard; // Cardinality at the end of Phase-1
startTime = MPI_Wtime();
/////////////////////////////////////////////////////////////////////////////////////////
//////////////////////////////////////// MAIN LOOP //////////////////////////////////////
/////////////////////////////////////////////////////////////////////////////////////////
// Main While Loop:
#ifdef PRINT_DEBUG_INFO_
cout << "\n(" << myRank << "=========================************===============================" << endl;
fflush(stdout);
fflush(stdout);
#endif
#ifdef PRINT_DEBUG_INFO_
cout << "\n(" << myRank << ")Entering While(true) loop..";
fflush(stdout);
#endif
#ifdef PRINT_DEBUG_INFO_
cout << "\n(" << myRank << "=========================************===============================" << endl;
fflush(stdout);
fflush(stdout);
#endif
while (true) {
#ifdef DEBUG_HANG_
//if (myRank == 0)
cout << "\n(" << myRank << ") Main loop" << endl;
fflush(stdout);
#endif
///////////////////////////////////////////////////////////////////////////////////
/////////////////////////// PROCESS MATCHED VERTICES //////////////////////////////
///////////////////////////////////////////////////////////////////////////////////
processMatchedVerticesAndSendMessagesS(NLVer,
UChunkBeingProcessed,
U,
privateU,
StartIndex,
EndIndex,
&myCard,
&msgInd,
&NumMessagesBundled,
&S,
verLocPtr,
verLocInd,
verDistance,
PCounter,
Counter,
myRank,
numProcs,
candidateMate,
GMate,
Mate,
Ghost2LocalMap,
edgeLocWeight,
QLocalVtx,
QGhostVtx,
QMsgType,
QOwner,
privateQLocalVtx,
privateQGhostVtx,
privateQMsgType,
privateQOwner,
comm,
&msgActual,
Message);
///////////////////////// END OF PROCESS MATCHED VERTICES /////////////////////////
//// BREAK IF NO MESSAGES EXPECTED /////////
#ifdef DEBUG_HANG_
#if 0
cout << myRank << " Mate after ProcessMatchedAndSend phase "<<S <<endl;
for (int i=0; i<NLVer; i++) {
cout << Mate[i] << " " ;
}
cout << endl;
#endif
#endif
#ifdef PRINT_DEBUG_INFO_
cout << "\n(" << myRank << ")Deciding whether to break: S= " << S << endl;
#endif
if (S == 0) {
#ifdef DEBUG_HANG_
cout << "\n(" << myRank << ") Breaking out" << endl;
fflush(stdout);
#endif
break;
}
///////////////////////////////////////////////////////////////////////////////////
/////////////////////////// PROCESS MESSAGES //////////////////////////////////////
///////////////////////////////////////////////////////////////////////////////////
processMessagesS(NLVer,
Mate,
candidateMate,
Ghost2LocalMap,
GMate,
Counter,
StartIndex,
EndIndex,
&myCard,
&msgInd,
&msgActual,
edgeLocWeight,
verDistance,
verLocPtr,
k,
verLocInd,
numProcs,
myRank,
comm,
Message,
numGhostEdges,
u,
v,
&S,
U);
///////////////////////// END OF PROCESS MESSAGES /////////////////////////////////
#ifdef DEBUG_HANG_
#if 0
cout << myRank << " Mate after ProcessMessages phase "<<S <<endl;
for (int i=0; i<NLVer; i++) {
cout << Mate[i] << " " ;
}
cout << endl;
#endif
#endif
#ifdef PRINT_DEBUG_INFO_
cout << "\n(" << myRank << ")Finished Message processing phase: S= " << S;
fflush(stdout);
cout << "\n(" << myRank << ")** SENT : ACTUAL= " << msgActual;
fflush(stdout);
cout << "\n(" << myRank << ")** SENT : INDIVIDUAL= " << msgInd << endl;
fflush(stdout);
#endif
} // End of while (true)
clean(NLVer,
myRank,
MessageIndex,
SRequest,
SStatus,
BufferSize,
Buffer,
msgActual,
msgActualSent,
msgInd,
msgIndSent,
NumMessagesBundled,
msgPercent);
finishTime = MPI_Wtime();
*ph2_time = finishTime - startTime; // Time taken for Phase-2
*ph2_card = myCard; // Cardinality at the end of Phase-2
}
#endif
#endif
@@ -177,23 +177,24 @@ subroutine amg_c_dec_aggregator_mat_bld(ag,parms,a,desc_a,ilaggr,nlaggr,&
select case (parms%aggr_prol)
case (amg_no_smooth_)
call amg_caggrmat_nosmth_bld(a,desc_a,ilaggr,nlaggr,&
& parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
call amg_caggrmat_nosmth_bld(parms%aggr_prol,a,desc_a,ilaggr,&
nlaggr,parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
case(amg_smooth_prol_)
case(amg_smooth_prol_,amg_l1_smooth_prol_)
call amg_caggrmat_smth_bld(a,desc_a,ilaggr,nlaggr, &
& parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
call amg_caggrmat_smth_bld(parms%aggr_prol,a,desc_a,&
ilaggr,nlaggr,parms,ac,desc_ac,op_prol,&
op_restr,t_prol,info)
!!$ case(amg_biz_prol_)
!!$
!!$ call amg_caggrmat_biz_bld(a,desc_a,ilaggr,nlaggr, &
!!$ & parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
case(amg_min_energy_)
call amg_caggrmat_minnrg_bld(a,desc_a,ilaggr,nlaggr, &
& parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
call amg_caggrmat_minnrg_bld(parms%aggr_prol,a,desc_a,ilaggr,&
nlaggr,parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
case default
info = psb_err_internal_error_
@@ -216,6 +216,7 @@ subroutine amg_c_ptap_bld(a_csr,desc_a,nlaggr,parms,ac,&
call ac_csr%set_nrows(desc_ac%get_local_rows())
call ac_csr%set_ncols(desc_ac%get_local_cols())
call ac_csr%clean_zeros(info)
call ac%mv_from(ac_csr)
call ac%set_asb()
@@ -420,6 +421,7 @@ subroutine amg_c_lc_ptap_bld(a_csr,desc_a,nlaggr,parms,ac,&
call ac_csr%set_nrows(desc_ac%get_local_rows())
call ac_csr%set_ncols(desc_ac%get_local_cols())
call ac_csr%clean_zeros(info)
call ac%mv_from(ac_csr)
call ac%set_asb()
@@ -629,6 +631,7 @@ subroutine amg_lc_ptap_bld(a_csr,desc_a,nlaggr,parms,ac,&
call ac_csr%set_nrows(desc_ac%get_local_rows())
call ac_csr%set_ncols(desc_ac%get_local_cols())
call ac_csr%clean_zeros(info)
call ac%mv_from(ac_csr)
call ac%set_asb()
+1
View File
@@ -142,6 +142,7 @@ subroutine amg_c_rap(a_csr,desc_a,nlaggr,parms,ac,&
call ac_csr%set_nrows(desc_ac%get_local_rows())
call ac_csr%set_ncols(desc_ac%get_local_cols())
call ac_csr%clean_zeros(info)
call ac%mv_from(ac_csr)
call ac%set_asb()
@@ -250,7 +250,7 @@ subroutine amg_c_soc1_map_bld(iorder,theta,clean_zeros,a,desc_a,nlaggr,ilaggr,in
! we will not reset.
if (j>nr) cycle step1
if (ilaggr(j) > 0) cycle step1
if (abs(val(k)) > theta*sqrt(abs(diag(i)*diag(j)))) then
if ((abs(val(k)) > theta*sqrt(abs(diag(i)*diag(j)))).and.(diag(i).ne.czero)) then
ip = ip + 1
icol(ip) = icol(k)
end if
@@ -275,7 +275,8 @@ subroutine amg_c_soc1_map_bld(iorder,theta,clean_zeros,a,desc_a,nlaggr,ilaggr,in
disjoint = all(ilaggr(icol(1:ip)) == -(nr+1)).or.(ip==0)
if (disjoint) then
locnaggr(kk) = locnaggr(kk) + 1
itmp = (bnds(kk)-1+locnaggr(kk))*nths+kk
itmp = (bnds(kk)-1+locnaggr(kk)) !be careful about overflow
itmp = itmp*nths+kk
if (itmp < (bnds(kk)-1+locnaggr(kk))) then
!$omp atomic update
info = max(12345678,info)
@@ -356,7 +357,7 @@ subroutine amg_c_soc1_map_bld(iorder,theta,clean_zeros,a,desc_a,nlaggr,ilaggr,in
do k=1, nz
j = icol(k)
if ((1<=j).and.(j<=nr)) then
if (abs(val(k)) > theta*sqrt(abs(diag(i)*diag(j)))) then
if ((abs(val(k)) > theta*sqrt(abs(diag(i)*diag(j)))).and.(diag(i).ne.czero)) then
ip = ip + 1
icol(ip) = icol(k)
end if
@@ -544,4 +545,3 @@ subroutine amg_c_soc1_map_bld(iorder,theta,clean_zeros,a,desc_a,nlaggr,ilaggr,in
return
end subroutine amg_c_soc1_map_bld
@@ -309,7 +309,8 @@ subroutine amg_c_soc2_map_bld(iorder,theta,clean_zeros,a,desc_a,nlaggr,ilaggr,in
!
if (disjoint) then
locnaggr(kk) = locnaggr(kk) + 1
itmp = (bnds(kk)-1+locnaggr(kk))*nths+kk
itmp = (bnds(kk)-1+locnaggr(kk)) !be careful about overflow
itmp = itmp*nths+kk
if (itmp < (bnds(kk)-1+locnaggr(kk))) then
!$omp atomic update
info = max(12345678,info)
@@ -69,6 +69,7 @@
!
!
! Arguments:
! dol1smoothing - fictitious integer argument, it is not used inside
! a - type(psb_cspmat_type), input.
! The sparse matrix structure containing the local part of
! the fine-level matrix.
@@ -104,8 +105,8 @@
! Error code.
!
!
subroutine amg_caggrmat_minnrg_bld(a,desc_a,ilaggr,nlaggr,parms,&
& ac,desc_ac,op_prol,op_restr,t_prol,info)
subroutine amg_caggrmat_minnrg_bld(dol1smoothing,a,desc_a,ilaggr,nlaggr,&
parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
use psb_base_mod
use amg_base_prec_type
use amg_c_inner_mod, amg_protect_name => amg_caggrmat_minnrg_bld
@@ -113,6 +114,7 @@ subroutine amg_caggrmat_minnrg_bld(a,desc_a,ilaggr,nlaggr,parms,&
implicit none
! Arguments
integer(psb_ipk_), intent(in) :: dol1smoothing
type(psb_cspmat_type), intent(in) :: a
type(psb_desc_type), intent(inout) :: desc_a
integer(psb_lpk_), intent(inout) :: ilaggr(:), nlaggr(:)
@@ -171,6 +173,13 @@ subroutine amg_caggrmat_minnrg_bld(a,desc_a,ilaggr,nlaggr,parms,&
filter_mat = (parms%aggr_filter == amg_filter_mat_)
if (dol1smoothing.ne.amg_no_smooth_) then
info=psb_err_fatal_;
call psb_errpush(info,name,a_err='Are you trying to smooth an unsmoothed aggregation?')
goto 9999
end if
!NEEDS TO BE REWORKED !!
! naggr: number of local aggregates
@@ -300,7 +309,7 @@ subroutine amg_caggrmat_minnrg_bld(a,desc_a,ilaggr,nlaggr,parms,&
!!$ endif
!!$ enddo
!!$ if (jd == -1) then
!!$ write(0,*) 'Wrong input: we need the diagonal!!!!', i
!!$ write(0,*) name,': Warning: there is no diagonal element', i
!!$ else
!!$ acsrf%val(jd)=acsrf%val(jd)-tmp
!!$ end if
@@ -94,10 +94,11 @@
!
! info - integer, output.
! Error code.
! dol1smoothing - optional, this is here just for interfacing reasons. It is not used by the
! code
!
!
subroutine amg_caggrmat_nosmth_bld(a,desc_a,ilaggr,nlaggr,parms,&
& ac,desc_ac,op_prol,op_restr,t_prol,info)
subroutine amg_caggrmat_nosmth_bld(dol1smoothing,a,desc_a,ilaggr,nlaggr,&
parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
use psb_base_mod
use amg_base_prec_type
use amg_c_inner_mod, amg_protect_name => amg_caggrmat_nosmth_bld
@@ -105,6 +106,7 @@ subroutine amg_caggrmat_nosmth_bld(a,desc_a,ilaggr,nlaggr,parms,&
implicit none
! Arguments
integer(psb_ipk_), intent(in) :: dol1smoothing
type(psb_cspmat_type), intent(in) :: a
type(psb_desc_type), intent(inout) :: desc_a
integer(psb_lpk_), intent(inout) :: ilaggr(:), nlaggr(:)
@@ -137,6 +139,12 @@ subroutine amg_caggrmat_nosmth_bld(a,desc_a,ilaggr,nlaggr,parms,&
ctxt = desc_a%get_context()
call psb_info(ctxt, me, np)
if (dol1smoothing.ne.amg_no_smooth_) then
info=psb_err_fatal_;
call psb_errpush(info,name,a_err='Are you trying to smooth an unsmoothed aggregation?')
goto 9999
end if
nglob = desc_a%get_global_rows()
nrow = desc_a%get_local_rows()
ncol = desc_a%get_local_cols()
@@ -69,6 +69,8 @@
!
!
! Arguments:
! dol1smooth - Integer taking the type of smoother that has to be used
! on the tentative prolongator
! a - type(psb_cspmat_type), input.
! The sparse matrix structure containing the local part of
! the fine-level matrix.
@@ -102,16 +104,18 @@
! info - integer, output.
! Error code.
!
subroutine amg_caggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
& ac,desc_ac,op_prol,op_restr,t_prol,info)
subroutine amg_caggrmat_smth_bld(dol1smoothing,a,desc_a,ilaggr,nlaggr,&
parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
use psb_base_mod
use amg_base_prec_type
use amg_c_inner_mod, amg_protect_name => amg_caggrmat_smth_bld
use amg_c_base_aggregator_mod
! use, intrinsic :: ieee_arithmetic
implicit none
! Arguments
integer(psb_ipk_), intent(in) :: dol1smoothing
type(psb_cspmat_type), intent(in) :: a
type(psb_desc_type), intent(inout) :: desc_a
integer(psb_lpk_), intent(inout) :: ilaggr(:), nlaggr(:)
@@ -132,7 +136,7 @@ subroutine amg_caggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
type(psb_c_coo_sparse_mat) :: coo_prol, coo_restr
type(psb_c_csr_sparse_mat) :: acsr1, acsrf, csr_prol, acsr
complex(psb_spk_), allocatable :: adiag(:)
real(psb_spk_), allocatable :: arwsum(:)
real(psb_spk_), allocatable :: arwsum(:),l1rwsum(:)
integer(psb_ipk_) :: ierr(5)
logical :: filter_mat
integer(psb_ipk_) :: debug_level, debug_unit, err_act
@@ -141,6 +145,7 @@ subroutine amg_caggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
logical, parameter :: debug_new=.false.
character(len=80) :: filename
logical, parameter :: do_timings=.false.
logical :: do_l1correction=.false.
integer(psb_ipk_), save :: idx_spspmm=-1, idx_phase1=-1, idx_gtrans=-1, idx_phase2=-1, idx_refine=-1
integer(psb_ipk_), save :: idx_phase3=-1, idx_cdasb=-1, idx_ptap=-1
@@ -173,6 +178,9 @@ subroutine amg_caggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
if ((do_timings).and.(idx_ptap==-1)) &
& idx_ptap = psb_get_timer_idx("DEC_SMTH_BLD: ptap_bld ")
! check if we have to use Jacobi or l1-Jacobi to smooth the tentative prolongator
if (dol1smoothing.eq.amg_l1_smooth_prol_) do_l1correction=.true.
nglob = desc_a%get_global_rows()
nrow = desc_a%get_local_rows()
@@ -185,7 +193,7 @@ subroutine amg_caggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
naggrm1 = sum(nlaggr(1:me))
naggrp1 = sum(nlaggr(1:me+1))
filter_mat = (parms%aggr_filter == amg_filter_mat_)
filter_mat = (parms%aggr_filter == amg_filter_mat_).or.(parms%aggr_filter == amg_filter_prow_mat_)
!
! naggr: number of local aggregates
@@ -200,6 +208,24 @@ subroutine amg_caggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
if (info == psb_success_) &
& call psb_halo(adiag,desc_a,info)
if (info == psb_success_) call a%cp_to(acsr)
!
! Do the l1-correction on the diagonal if it is requested
!
if (do_l1correction) then
allocate(l1rwsum(nrow))
call acsr%arwsum(l1rwsum)
if (info == psb_success_) &
& call psb_realloc(ncol,l1rwsum,info)
if (info == psb_success_) &
& call psb_halo(l1rwsum,desc_a,info)
! \tilde{D}_{i,i} = \sum_{j \ne i} |a_{i,j}|
!$OMP parallel do private(i) schedule(static)
do i=1,size(adiag)
adiag(i) = adiag(i) + l1rwsum(i) - abs(adiag(i))
end do
!$OMP end parallel do
end if
if(info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='sp_getdiag')
@@ -230,9 +256,15 @@ subroutine amg_caggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
enddo
if (jd == -1) then
! if (.not.do_l1correction)
write(0,*) 'Wrong input: we need the diagonal!!!!', i
else
else if (parms%aggr_filter == amg_filter_mat_) then
! We perform filtering in the standard way assuming that A is an M-matrix
acsrf%val(jd)=acsrf%val(jd)-tmp
else if (parms%aggr_filter == amg_filter_prow_mat_) then
! We are probably doing l1-correction, hence we want to preserve the
! row sum of the matrix: note the change in sign
acsrf%val(jd)=acsrf%val(jd)+tmp
end if
enddo
!$OMP end parallel do
@@ -240,7 +272,6 @@ subroutine amg_caggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
call acsrf%clean_zeros(info)
end if
!$OMP parallel do private(i) schedule(static)
do i=1,size(adiag)
if (adiag(i) /= czero) then
@@ -252,14 +283,17 @@ subroutine amg_caggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
!$OMP end parallel do
if (parms%aggr_omega_alg == amg_eig_est_) then
if (parms%aggr_eig == amg_max_norm_) then
if ( (parms%aggr_filter == amg_filter_prow_mat_).and.(do_l1correction) ) then
! For l1-Jacobi this can be estimated with 1:
! this makes sense only if we are preserving the row-sum!
parms%aggr_omega_val = done
else if (parms%aggr_eig == amg_max_norm_) then
allocate(arwsum(nrow))
call acsr%arwsum(arwsum)
anorm = maxval(abs(adiag(1:nrow)*arwsum(1:nrow)))
call psb_amx(ctxt,anorm)
omega = 4.d0/(3.d0*anorm)
parms%aggr_omega_val = omega
else
info = psb_err_internal_error_
call psb_errpush(info,name,a_err='invalid amg_aggr_eig_')
@@ -322,6 +356,7 @@ subroutine amg_caggrmat_smth_bld(a,desc_a,ilaggr,nlaggr,parms,&
if (debug_level >= psb_debug_outer_) &
& write(debug_unit,*) me,' ',trim(name),&
& 'Done smooth_aggregate '
if (allocated(l1rwsum)) deallocate(l1rwsum)
call psb_erractionrestore(err_act)
return
@@ -177,23 +177,24 @@ subroutine amg_d_dec_aggregator_mat_bld(ag,parms,a,desc_a,ilaggr,nlaggr,&
select case (parms%aggr_prol)
case (amg_no_smooth_)
call amg_daggrmat_nosmth_bld(a,desc_a,ilaggr,nlaggr,&
& parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
call amg_daggrmat_nosmth_bld(parms%aggr_prol,a,desc_a,ilaggr,&
nlaggr,parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
case(amg_smooth_prol_)
case(amg_smooth_prol_,amg_l1_smooth_prol_)
call amg_daggrmat_smth_bld(a,desc_a,ilaggr,nlaggr, &
& parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
call amg_daggrmat_smth_bld(parms%aggr_prol,a,desc_a,&
ilaggr,nlaggr,parms,ac,desc_ac,op_prol,&
op_restr,t_prol,info)
!!$ case(amg_biz_prol_)
!!$
!!$ call amg_daggrmat_biz_bld(a,desc_a,ilaggr,nlaggr, &
!!$ & parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
case(amg_min_energy_)
call amg_daggrmat_minnrg_bld(a,desc_a,ilaggr,nlaggr, &
& parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
call amg_daggrmat_minnrg_bld(parms%aggr_prol,a,desc_a,ilaggr,&
nlaggr,parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
case default
info = psb_err_internal_error_
@@ -184,20 +184,23 @@ subroutine amg_d_parmatch_aggregator_mat_bld(ag,parms,a,desc_a,ilaggr,nlaggr,&
!
select case (parms%aggr_prol)
case (amg_no_smooth_)
call amg_d_parmatch_unsmth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
& ac,desc_ac,op_prol,op_restr,t_prol,info)
call amg_d_parmatch_unsmth_bld(parms%aggr_prol,ag,a,desc_a,&
ilaggr,nlaggr,parms,ac,desc_ac,op_prol,op_restr,&
t_prol,info)
case(amg_smooth_prol_)
call amg_d_parmatch_smth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
& ac,desc_ac,op_prol,op_restr,t_prol,info)
case(amg_smooth_prol_,amg_l1_smooth_prol_)
call amg_d_parmatch_smth_bld(parms%aggr_prol,ag,a,desc_a,&
ilaggr,nlaggr,parms,ac,desc_ac,op_prol,op_restr,&
t_prol,info)
!!$ case(amg_biz_prol_)
!!$ call amg_daggrmat_biz_bld(a,desc_a,ilaggr,nlaggr, &
!!$ & parms,ac,desc_ac,op_prol,op_restr,info)
case(amg_min_energy_)
call amg_daggrmat_minnrg_bld(a,desc_a,ilaggr,nlaggr, &
& parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
call amg_daggrmat_minnrg_bld(parms%aggr_prol,a,desc_a,&
ilaggr,nlaggr,parms,ac,desc_ac,op_prol,op_restr,&
t_prol,info)
case default
info = psb_err_internal_error_
@@ -69,6 +69,8 @@
!
!
! Arguments:
! dol1smoothing - Select between l1-Jacobi and Jacobi as smoother for the
! tentative prolongator
! a - type(psb_dspmat_type), input.
! The sparse matrix structure containing the local part of
! the fine-level matrix.
@@ -102,8 +104,8 @@
! info - integer, output.
! Error code.
!
subroutine amg_d_parmatch_smth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
& ac,desc_ac,op_prol,op_restr,t_prol,info)
subroutine amg_d_parmatch_smth_bld(dol1smoothing,ag,a,desc_a,ilaggr,nlaggr,&
parms,ac,desc_ac,op_prol,op_restr,t_prol,info)
use psb_base_mod
use amg_base_prec_type
use amg_d_inner_mod
@@ -116,6 +118,7 @@ subroutine amg_d_parmatch_smth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
implicit none
! Arguments
integer(psb_ipk_), intent(in) :: dol1smoothing
class(amg_d_parmatch_aggregator_type), target, intent(inout) :: ag
type(psb_dspmat_type), intent(in) :: a
type(psb_desc_type), intent(inout) :: desc_a
@@ -137,7 +140,7 @@ subroutine amg_d_parmatch_smth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
type(psb_d_coo_sparse_mat) :: coo_prol, coo_restr
type(psb_d_csr_sparse_mat) :: acsrf, csr_prol, acsr, tcsr
real(psb_dpk_), allocatable :: adiag(:)
real(psb_dpk_), allocatable :: arwsum(:)
real(psb_dpk_), allocatable :: arwsum(:),l1rwsum(:)
logical :: filter_mat
integer(psb_ipk_) :: debug_level, debug_unit, err_act
integer(psb_ipk_), parameter :: ncmax=16
@@ -145,6 +148,7 @@ subroutine amg_d_parmatch_smth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
logical, parameter :: debug_new=.false., dump_r=.false., dump_p=.false., debug=.false.
character(len=80) :: filename
logical, parameter :: do_timings=.false.
logical :: do_l1correction=.false.
integer(psb_ipk_), save :: idx_spspmm=-1, idx_phase1=-1, idx_gtrans=-1, idx_phase2=-1, idx_refine=-1, idx_phase3=-1
integer(psb_ipk_), save :: idx_cdasb=-1, idx_ptap=-1
@@ -166,6 +170,10 @@ subroutine amg_d_parmatch_smth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
ncol = desc_a%get_local_cols()
theta = parms%aggr_thresh
! Check if we have to perform l1-Jacobi or Jacobi as smoother
if(dol1smoothing.eq.amg_l1_smooth_prol_) do_l1correction=.true.
!write(0,*) me,' ',trim(name),' Start ',idx_spspmm
if ((do_timings).and.(idx_spspmm==-1)) &
& idx_spspmm = psb_get_timer_idx("PMC_SMTH_BLD: par_spspmm")
@@ -217,6 +225,19 @@ subroutine amg_d_parmatch_smth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
if (info == psb_success_) &
& call psb_halo(adiag,desc_a,info)
if (info == psb_success_) call a%cp_to(acsr)
! Get the l1-diagonal of D
if (do_l1correction) then
allocate(l1rwsum(nrow))
call acsr%arwsum(l1rwsum)
if (info == psb_success_) &
& call psb_realloc(ncol,l1rwsum,info)
if (info == psb_success_) &
& call psb_halo(l1rwsum,desc_a,info)
! \tilde{D}_{i,i} = \sum_{j \ne i} |a_{i,j}|
do i=1,size(adiag)
adiag(i) = adiag(i) + l1rwsum(i) - abs(adiag(i))
end do
end if
if(info /= psb_success_) then
call psb_errpush(psb_err_from_subroutine_,name,a_err='sp_getdiag')
@@ -246,7 +267,7 @@ subroutine amg_d_parmatch_smth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
enddo
if (jd == -1) then
write(0,*) 'Wrong input: we need the diagonal!!!!', i
write(0,*) name,': Warning: there is no diagonal element', i
else
acsrf%val(jd)=acsrf%val(jd)-tmp
end if
@@ -267,7 +288,10 @@ subroutine amg_d_parmatch_smth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
if (parms%aggr_omega_alg == amg_eig_est_) then
if (parms%aggr_eig == amg_max_norm_) then
if (do_l1correction) then
! For l1-Jacobi this can be estimated with 1
parms%aggr_omega_val = done
else if (parms%aggr_eig == amg_max_norm_) then
allocate(arwsum(nrow))
call acsr%arwsum(arwsum)
anorm = maxval(abs(adiag(1:nrow)*arwsum(1:nrow)))
@@ -373,6 +397,7 @@ subroutine amg_d_parmatch_smth_bld(ag,a,desc_a,ilaggr,nlaggr,parms,&
end block
end if
if (allocated(l1rwsum)) deallocate(l1rwsum)
if (do_timings) call psb_toc(idx_phase2)
if (debug_level >= psb_debug_outer_) &

Some files were not shown because too many files have changed in this diff Show More