diff --git a/docs/amg4psblas_1.0-guide.pdf b/docs/amg4psblas_1.0-guide.pdf index 2303b96c..60f23ddf 100644 Binary files a/docs/amg4psblas_1.0-guide.pdf and b/docs/amg4psblas_1.0-guide.pdf differ diff --git a/docs/html/index.html b/docs/html/index.html index e720e239..6957ab26 100644 --- a/docs/html/index.html +++ b/docs/html/index.html @@ -123,57 +123,57 @@ class="cmr-12">Method set
 5.3 Method hierarchy_build
 5.4 Method smoothers_build
 5.5 Method build
 5.6 Method apply
 5.7 Method free
 5.8 Method descr
 5.9 Auxiliary Methods
6 Adding new smoother and solver objects to AMG4PSBLAS
7 Error Handling
A License
References
References diff --git a/docs/html/userhtml.css b/docs/html/userhtml.css index 3d2eaa23..f8b025b4 100644 --- a/docs/html/userhtml.css +++ b/docs/html/userhtml.css @@ -194,9 +194,9 @@ div.lstinputlisting{ font-family: monospace; white-space: nowrap; } #TBL-5{border-collapse:collapse;} #TBL-5 colgroup{border-left: 1px solid black;border-right:1px solid black;} #TBL-5{border-collapse:collapse;} -td#TBL-5-10-5{border-left:solid black 0.4pt;border-right:solid black 0.4pt;} td#TBL-5-11-5{border-left:solid black 0.4pt;border-right:solid black 0.4pt;} td#TBL-5-12-5{border-left:solid black 0.4pt;border-right:solid black 0.4pt;} +td#TBL-5-13-5{border-left:solid black 0.4pt;border-right:solid black 0.4pt;} #TBL-6 colgroup{border-left: 1px solid black;border-right:1px solid black;} #TBL-6{border-collapse:collapse;} #TBL-6 colgroup{border-left: 1px solid black;border-right:1px solid black;} @@ -250,17 +250,5 @@ td#TBL-7-7-5{border-left:solid black 0.4pt;border-right:solid black 0.4pt;} #TBL-9{border-collapse:collapse;} #TBL-9 colgroup{border-left: 1px solid black;border-right:1px solid black;} #TBL-9{border-collapse:collapse;} -#TBL-10 colgroup{border-left: 1px solid black;border-right:1px solid black;} -#TBL-10{border-collapse:collapse;} -#TBL-10 colgroup{border-left: 1px solid black;border-right:1px solid black;} -#TBL-10{border-collapse:collapse;} -#TBL-10 colgroup{border-left: 1px solid black;border-right:1px solid black;} -#TBL-10{border-collapse:collapse;} -#TBL-10 colgroup{border-left: 1px solid black;border-right:1px solid black;} -#TBL-10{border-collapse:collapse;} -#TBL-10 colgroup{border-left: 1px solid black;border-right:1px solid black;} -#TBL-10{border-collapse:collapse;} -#TBL-10 colgroup{border-left: 1px solid black;border-right:1px solid black;} -#TBL-10{border-collapse:collapse;} /* end css.sty */ diff --git a/docs/html/userhtml.html b/docs/html/userhtml.html index e720e239..6957ab26 100644 --- a/docs/html/userhtml.html +++ b/docs/html/userhtml.html @@ -123,57 +123,57 @@ class="cmr-12">Method set
 5.3 Method hierarchy_build
 5.4 Method smoothers_build
 5.5 Method build
 5.6 Method apply
 5.7 Method free
 5.8 Method descr
 5.9 Auxiliary Methods
6 Adding new smoother and solver objects to AMG4PSBLAS
7 Error Handling
A License
References
References diff --git a/docs/html/userhtmlli2.html b/docs/html/userhtmlli2.html index 061f8760..8f14145c 100644 --- a/docs/html/userhtmlli2.html +++ b/docs/html/userhtmlli2.html @@ -135,32 +135,32 @@ class="cmr-12">Auxiliary Methods class="cmr-12">  5.9.1 Method: dump
  5.9.2 Method: clone
  5.9.3 Method: sizeof
  5.9.4 Method: allocate_wrk
  5.9.5 Method: free_wrk
A License
References diff --git a/docs/html/userhtmlli4.html b/docs/html/userhtmlli4.html index 31370ba6..38af5368 100644 --- a/docs/html/userhtmlli4.html +++ b/docs/html/userhtmlli4.html @@ -29,7 +29,7 @@ class="cmr-12">]

class="cmr-12">References

+ id="Q1-28-46">

. class="cmr-12">[2]   D. Bertaccini and S. Filippone, Sparse approximate inverse + preconditioners on high performance GPU platforms, Comput. Math. Appl. + 71 (2016), no. 3, 693–711. +

+

+ [3]   M., Computing, 63, 1999, 233–263.

[3][4]   , SIAM, 2000.

[4][5]   State of the Art in Scientific Computing, Lecture Notes in Comput Springer, 2005, 593–602.

+ + +

[5][6]   applications, Applicable Algebra in Engineering, Communications and Computing, 18 (3) 2007, 223–239. - - -

[6][7]   (2), 1999, 792–797.

[7][8]   Workshops, IEEE CS, 2011.

[8][9]   , Applied Numerical Mathematics, Elsevier Science, class="cmr-12">57 (11-12), 2007, 1181-1196.

- [9][10]   , ACM Trans. Math. Softw., 37(3), 2010, art. 30.

[10][11]   , Appl. Algebra Engrg. Comm. Comput., 18(3), 2007, 223–239.

[11][12]   , 2020, arXiv:2006.16147v3. + + +

[12][13]   , ACM Transactions on Mathematical Software, 30, 2004, 196– class="cmr-12">(See also http://www.cise.ufl.edu/~davis/) - - -

[13][14]   SIAM Journal on Matrix Analysis and Applications, 20 (3), 1999, 7

[14][15]   Software, 16 (1) 1990, 1–17.

[15][16]   Transactions on Mathematical Software, 14 (1) 1988, 1–17.

[16][17]   .

[17][18]    23.

[18][19]   Library for Parallel Linear Algebra Computation on Sparse Matric class="cmr-12">, ACM Transactions on Mathematical Software, 26 (4), 2000, 527–550. + + +

[19][20]   algebraic multigrid by aggregation, Numerical Lin. Algebra with Applications, 2016, 23:501-518 - - -

[20][21]   , MIT Press, 1998.

[21][22]   Mathematical Software, 5 (3), 1979, 308–323.

[22][23]   ACM Transactions on Mathematical Software, 29 (2), 2003, 110̵

[23][24]   Numerical Linear Algebra with Applications, 15 (5), 2008, 473R

[24][25]   2003.

[25][26]   , Cambridge University Press, 1996.

+ + +

[26][27]   Press, 1998.

[27][28]    class="cmr-12">U. Trottenberg, C. Oosterlee, Multigrid, Academic Press, 2001. - - -

[28][29]   editor, Proceedings of SuperComputing 2000, Dallas, 2000.

[29][30]   (3) 1996, 179–196.

[30][31]   16, (2013) 59–76.

[31][32]   ) provides parallel Algebraic MultiGrid (AMG) preconditioners (se class="cmr-12">e.g., [34, 2728]), to be used in the iterative solution of linear systems, and a version of a Krylov-type cycle (K-cycle) [34, 2324]are available, which can beproposed in [23, 2930], and already included in the previous versions of thepackage [10, 911]; @@ -177,11 +177,11 @@ class="cmr-12">on Compatible Weighted Matching introduced in [3031, 3132]and described indetails in [1112];

computational framework [1819, 1718]. PSBLAS provides basic linear algebra operators multilevel (i.e., AMG) preconditioners with the Krylov solvers in [1617]. The following steps are required:

    @@ -134,8 +134,8 @@ class="cmr-12"> 2-87.
  1. 2-8 for further details of +href="userhtmlsu8.html#x17-16014r7">7 for further details of the preconditioner.
The interfaces for the calls shown above are defined using

-

smoother

smoother

class(amg_x_base_smoother_type)

The user-defined new smoother to be employed in the preconditioner.

solver

solver

class(amg_x_base_solver_type)

The user-defined new solver to be employed in the preconditioner.

PSBLAS error handling routines; for further details see the PSBLA [1617]. @@ -61,6 +61,10 @@ class="cmr-12">. +

+ + + diff --git a/docs/html/userhtmlsu1.html b/docs/html/userhtmlsu1.html index 945fed8d..f6d3e1f0 100644 --- a/docs/html/userhtmlsu1.html +++ b/docs/html/userhtmlsu1.html @@ -36,15 +36,15 @@ class="cmbx-12">BLAS

[1415, 1516, 2122] Many vendors provide optimized versions of BLAS; if no MPI
[2021, 2627] A version of MPI is available on most high-performance computing PSBLAS
[1617, 1819] Parallel Sparse BLAS (PSBLAS) is available from -

[nextMethod smoothers_build

-

-

+

call p,imold])

-

This method builds the smoothers and the coarsest-level solvers for the multilevel preconditioner  5.3). -

Arguments -

a

a

type(psb_in).

The sparse matrix structure containing the local part of the matrix to be preconditioned. Note that , single/double precision version of AMG4PSBLAS class="cmr-12">under use. See the PSBLAS User’s Guide for details [1617].

desc_a

desc_a

type(psb_desc_typein).

The communication descriptor of a. See the PSBLAS User’s Guide @@ -145,14 +145,14 @@ class="cmr-12">. See the PSBLAS User’s Guide class="cmr-12">for details [1617].

info

info

integer, out).

Error code. If no error, 0 is returned. See Section 7 for details.

amold

amold

class(psb_ optional.

The desired dynamic type for internal matrix components; this allows e.g. running on GPUs; it needs not be the same on all @@ -200,14 +200,14 @@ class="cmr-12">allows e.g. running on GPUs; it needs not be the same class="cmr-12">processes. See the PSBLAS User’s Guide for details [1617].

vmold

vmold

class(psb_ optional.

The desired dynamic type for internal vector components; this allows e.g. running on GPUs.

imold

imold

class(psb_i_base_vect_type optional.

The desired dynamic type for internal integer vector components; this allows e.g. running on GPUs.

@@ -262,7 +262,7 @@ class="td11">
-

[next] [up]

-

diff --git a/docs/html/userhtmlsu11.html b/docs/html/userhtmlsu11.html index a9b90003..a2dcad66 100644 --- a/docs/html/userhtmlsu11.html +++ b/docs/html/userhtmlsu11.html @@ -10,7 +10,7 @@ -

[next5.5 Method build

-

-

+

call p,imold])

-

This method builds the preconditioner p according to the requirements made by the @@ -83,17 +83,17 @@ class="cmtt-12">smoothers_build, whose nomenclature would however be somewhat unnatural when dealing with simple one-level preconditioners. -

Arguments -

a

a

type(psb_in).

The sparse matrix structure containing the local part of the matrix to be preconditioned. Note that , single/double precision version of AMG4PSBLAS class="cmr-12">under use. See the PSBLAS User’s Guide for details [1617].

desc_a

desc_a

type(psb_desc_typein).

The communication descriptor of a. See the PSBLAS User’s Guide @@ -153,14 +153,14 @@ class="cmr-12">. See the PSBLAS User’s Guide class="cmr-12">for details [1617].

info

info

integer, out).

Error code. If no error, 0 is returned. See Section 7 for details.

amold

amold

class(psb_ optional.

The desired dynamic type for internal matrix components; this allows e.g. running on GPUs; it needs not be the same on all @@ -208,14 +208,14 @@ class="cmr-12">allows e.g. running on GPUs; it needs not be the same class="cmr-12">processes. See the PSBLAS User’s Guide for details [1617].

vmold

vmold

class(psb_ optional.

The desired dynamic type for internal vector components; this allows e.g. running on GPUs.

imold

imold

class(psb_i_base_vect_type optional.

The desired dynamic type for internal integer vector components; this allows e.g. running on GPUs.

-

The method can be used to build multilevel preconditioners too. @@ -272,7 +272,7 @@ class="cmr-12">The method can be used to build multilevel preconditioners too.

[next] [up]

-

diff --git a/docs/html/userhtmlsu12.html b/docs/html/userhtmlsu12.html index ee0202ce..4cd5142c 100644 --- a/docs/html/userhtmlsu12.html +++ b/docs/html/userhtmlsu12.html @@ -10,7 +10,7 @@ -

[next5.6 Method apply

-

-

+

call p,work])

-

This method computes y = psb_krylov and hence it is completely transparent to the user. -

Arguments -

x

x

type(kindin).

The local part of the vector x. Note that , single/double precision version of AMG4PSBLAS under use.

y

y

type(kindout).

The local part of the vector y. Note that , single/double precision version of AMG4PSBLAS under use.

desc_a

desc_a

type(psb_desc_typein).

The communication descriptor associated to the matrix to be preconditioned.

info

info

integer, out).

Error code. If no error, 0 is returned. See Section 7 for details.

trans

trans

character(len(in).

If trans = -1).

work

work

type(kind target.

Workspace. Its size should be at least 4 , single/double precision version of AMG4PSBLAS under use.

@@ -363,7 +363,7 @@ class="td11">
-

[next] [up]

-

diff --git a/docs/html/userhtmlsu13.html b/docs/html/userhtmlsu13.html index 67da05e0..efc2cee4 100644 --- a/docs/html/userhtmlsu13.html +++ b/docs/html/userhtmlsu13.html @@ -10,7 +10,7 @@ -

[next5.7 Method free

-

-

+

call p,info)

-

This method deallocates the preconditioner data structure p. -

Arguments -

info

info

integer, out).

Error code. If no error, 0 is returned. See Section for details.

@@ -89,7 +89,7 @@ class="td11">
-

[next] [up]

-

diff --git a/docs/html/userhtmlsu14.html b/docs/html/userhtmlsu14.html index 1abdf5b8..d4923cd0 100644 --- a/docs/html/userhtmlsu14.html +++ b/docs/html/userhtmlsu14.html @@ -10,7 +10,7 @@ -

[next5.8 Method descr

-

-

+

call p verbosity])

-

This method prints a description of the preconditioner p to the standard output or to a @@ -68,17 +68,17 @@ class="cmtt-12">build, have been called. -

Arguments -

info

info

integer, out).

Error code. If no error, 0 is returned. See Section 7 for details.

iout

iout

integer,  optional.

The id of the file where the preconditioner description will be printed; the default is the standard output.

root

root

integer,  optional.

The id of the process where the preconditioner description will be printed; the default is psb_root_.

verbosity

verbosity

integer,  optional.

The verbosity level of the description. Default value is 0. For values higher than 0, it prints out further information, e.g., for @@ -169,12 +169,12 @@ class="cmr-12">a distributed multilevel preconditioner the size of the coarsematrices on every process.

-

[next] [up]

-

diff --git a/docs/html/userhtmlsu15.html b/docs/html/userhtmlsu15.html index e9670bb8..d1e34f15 100644 --- a/docs/html/userhtmlsu15.html +++ b/docs/html/userhtmlsu15.html @@ -10,7 +10,7 @@ -

[prev]

class="cmr-12">5.9 Auxiliary Methods -

Various functionalities are implemented as additional methods of the preconditioner object. -

+

5.9.1 Method: dump
-

-

+

call p,global_num])

-

Dump on file. -

Arguments -

info

info

integer, out).

Error code. If no error, 0 is returned. See Section 7 for details.

amold

amold

class(psb_ optional.

The desired dynamic type for internal matrix components; this allows e.g. running on GPUs; it needs not be the same on all @@ -128,21 +128,21 @@ class="cmr-12">allows e.g. running on GPUs; it needs not be the same class="cmr-12">processes. See the PSBLAS User’s Guide for details [1617].

-

+

5.9.2 Method: clone
-

-

+

call p)
-

Create a (deep) copy of the preconditioner object. -

Arguments -

pout

pout

type(amg_out).

The copy of the preconditioner data structure. Note that x must @@ -198,10 +198,10 @@ class="cmr-12">, single/double precision version of AMG4PSBLAS under use.

info

info

integer, out).

Error code. If no error, 0 is returned. See Section 7 for details.

-

+

5.9.3 Method: sizeof
-

-

+

sz =%sizeof()

-

Return memory footprint in bytes. -

+

5.9.4 Method: allocate_wrk
-

-

+

call p vmold])

-

Allocate internal work vectors. Each application of the preconditioner uses a number of work vectors which are allocated internally as necessary; therefore allocation and @@ -296,17 +296,17 @@ class="cmr-12">argument to the apply method. -

Arguments -

info

info

integer, out).

Error code. If no error, 0 is returned. See Section 7 for details.

vmold

vmold

class(psb_ optional.

The desired dynamic type for internal vector components; this allows e.g. running on GPUs.

-

+

5.9.5 Method: free_wrk
-

-

+

call p(info)

-

Deallocate internal work vectors. -

Arguments -

info

info

integer, out).

Error code. If no error, 0 is returned. See Section 7 for details.

diff --git a/docs/html/userhtmlsu2.html b/docs/html/userhtmlsu2.html index 27b9fb8a..2a92a459 100644 --- a/docs/html/userhtmlsu2.html +++ b/docs/html/userhtmlsu2.html @@ -44,7 +44,7 @@ class="cmbx-12">UMFPACK
[1213] A sparse LU factorization package included in the SuiteSparse SuperLU
[1314] A sparse LU factorization package available from _Dist
[2223] A sparse LU factorization package available from the same PSBLAS User’s Guide [1617].

 2-87.  2-87. When the value is of type  2-87).  2-87.

), as specified in Tables 76-87. For example, the block-Jacobi smoother using ILU(0) on the blocks is obtained by combining the block-Jacobi smoother object with the @@ -410,8 +410,8 @@ class="cmr-12">ILU(0) solver object. Similarly, the hybrid Gauss-Seidel smoother Table 76) is obtained by combining the block-Jacobi smoother object with a single sweep of the Gauss-Seidel solver object, while the point-Jacobi smoother @@ -429,8 +429,8 @@ class="cmtt-12">SMOOTHER_TYPE’ to appropriate values (see Tables 76), i.e., without setting ’smoother. class="cmr-12">Similar considerations apply to the point-Jacobi, Gauss-Seidel and block-Jacobi coarsest-level solvers, and shortcuts are available in this case too (see Table 5 ??).
@@ -606,16 +605,26 @@ class="cmtt-10x-x-109">character(len=*)

’VCYCLE’ -

’WCYCLE’ -

’KCYCLE’ -

’ADD’

’VCYCLE’

’VCYCLE’ +

’WCYCLE’ +

’KCYCLE’ +

’ADD’

’VCYCLE’

Multilevel cycle: V-cycle, W-cycle, K-cycle, and additive composition.

’OUTER_SWEEPS’ integer

integer

Any integer

number ≥ 1 ments s

+

’MIN_COARSE_SIZE_PER_PROCESS’. +class="cmtt-10x-x-109">’.

Minimum ratio. The aggregation stops if the ratio between the global matrix dimensions at two consecutive levels is lower than or -equal to this threshold (see Note). +equal to this threshold (see Note).

> 1

20

Maximum number of levels. The aggregation stops if the number of levels -reaches this value (see Note). +reaches this value (see Note).

’PAR_AGGR_ALG’

character(len=*)

’DEC’, ’SYMDEC’, ’COUPLED’

’DEC’

Parallel aggregation algorithm. -

the

Parallel aggregation algorithm. +

the SYMDEC option applies decoupled aggregation to the sparsity pattern of A + AT . +class="cmmi-8">T .

what data type

val

default

comments +

’AGGR_TYPE’

’

character(len=*)

’SOC1’

’SOC1’, -’SOC2’, -’MATCHBOXP’

Type of aggregation algorithm: currently, +class="cmtt-10x-x-109">=*)

’SOC1’

’SOC1’, +’SOC2’, +’MATCHBOXP’

Type of aggregation algorithm: currently, for the decoupled aggregation we implement two measures of strength of connection, the one by Vaněk, Mandel and Brezina [29], +href="userhtmlli4.html#XVANEK_MANDEL_BREZINA">30], and the one by Gratton et al [19]. The +href="userhtmlli4.html#XGrHeJi:16">20]. The coupled aggregation is based on a parallel version of the half-approximate matching implemented in the MatchBox-P software -package AGGIUNGERE LINK AL -PACKAGE? +package [8].

’AGGR_SIZE’

integer

Any integer -

number +class="cmtt-10x-x-109">’

integer

Any integer +

number power of 2 and > 2

4

Maximum size of aggregates when the +class="cmmi-10x-x-109">> 2

4

Maximum size of aggregates when the coupled aggregation based on matching is applied. For aggressive coarsening with size of aggregate larger than 8 @@ -865,49 +916,56 @@ we recommend the use of smoothed prolongators. MODIFICARE CODICE +class="cmbx-10x-x-109">CODICE

’AGGR_PROL’

’

character(len=*)

’SMOOTHED’, -’UNSMOOTHED’

’SMOOTHED’

Prolongator used by the aggregation +class="cmtt-10x-x-109">=*)

’SMOOTHED’, +’UNSMOOTHED’

’SMOOTHED’

Prolongator used by the aggregation algorithm: smoothed or unsmoothed (i.e., -tentative prolongator). +tentative prolongator).

Note. The aggregation algorithm stops when at least one of the following criteria is met: the coarse size threshold, the coarse size threshold per process, the
-
minimum coarsening ratio, or the maximum number of levels is reached. Therefore, the actual number of levels may be
+class="td11">
Note. The aggregation algorithm stops when at least one of the following criteria is met: the coarse size threshold,
smaller than the specified maximum number of levels.
+class="td11">
the coarse size threshold per process, the minimum coarsening ratio, or the maximum number of levels is reached.
+
Therefore, the actual number of levels may be smaller than the specified maximum number of levels.
- +
Table 3: Parameters defining the aggregation algorithm.
+ @@ -917,7 +975,7 @@ class="content">Parameters defining the aggregation algorithm.

@@ -925,7 +983,7 @@ class="content">Parameters defining the aggregation algorithm.

+

< style="vertical-align:baseline;" id="TBL-6-5-">
Note. Different thresholds at different levels, such as those used in [29, Section 5.1], can be easily set by invoking the rou-
+href="userhtmlli4.html#XVANEK_MANDEL_BREZINA">30, Section 5.1], can be easily set by invoking the rou-
tine Parameters defining the aggregation algorithm (continued).

@@ -1075,7 +1133,10 @@ class="content">Parameters defining the aggregation algorithm (continued). -

+

+ + +

what

type

val

deault

cots

’AGGR_ORD’

character(len=*)

’NATURAL’ -

’DEGREE’

’NATURAL’

Initial ordering of indices for +class="td11">

Initial ordering of indices for the decoupled aggregation algorithm: either natural ordering or sorted by descending degrees of the nodes in the @@ -994,23 +1052,23 @@ matrix graph.

’AGGR_THRESH’

real( kind_parameter

_parameter

)

Any real -

number 

Any real +

number ∈ [0,1]

0.01

The threshold

0.01

The threshold θ in the decoupled aggregation algorithm, see (??) in @@ -1020,22 +1078,22 @@ bottom of this table.

’AGGR_FILTER’

character(len=*)

’FILTER’ -

’NOFILTER’

’NOFILTER’

Matrix used in computing the smoothed +class="td11">

Matrix used in computing the smoothed prolongator: filtered or unfiltered (see (??) in Section 
+LU from MUMPS, SuperLU or UMFPACK (plus +triangular solve). Suitable for GPUs (no triangular +solve) approximate inverse solvers INVK(p,q), +INVT(p1,p2,t1,t2) and AINV(t), see [2]. Note that +UMFPACK and SuperLU_Dist are available only in +double precision.

what

a type

e

val

deault

cots

’COARSE_MAT’

character(len=*)

’DIST’ -

’REPL’

’REPL’

Coarsest matrix layout: distributed among the +class="td11">

’DIST’ +

’REPL’

’REPL’

Coarsest matrix layout: distributed among the processes or replicated on each of them.

’COARSE_SOLVE’

character(len=*)

’MUMPS’ -

’UMF’ -

’SLU’ -

’SLUDIST’ -

’JACOBI’ -

’GS’ -

’BJAC’ -

’PCG’

See Note.

Solver used at the coarsest level: sequential LU +class="td11">

’MUMPS’ +

’UMF’ +

’SLU’ +

’SLUDIST’ +

’JACOBI’ +

’GS’ +

’BJAC’ +

’RKR’

See Note.

Solver used at the coarsest level: sequential LU from MUMPS, UMFPACK, or SuperLU (plus triangular solve); distributed LU from MUMPS or SuperLU_Dist (plus triangular solve); point-Jacobi, @@ -1180,62 +1263,95 @@ class="cmbx-10x-x-109">preconditioned Conjugate Gradient coupled with the block-Jacobi preconditioner with ILU(0) on the blocks. -

Note that UMF and SLU require the coarsest matrix -to be replicated, ILU(0) on the blocks. Note that UMF and +SLU require the coarsest matrix to be replicated, +SLUDIST, JACOBI, GS, BJAC and PCG -require it to be distributed, MUMPS can be used with -either a replicated or a distributed matrix. When -any of the previous solvers is specified, the matrix -layout is set to a default value which allows the use -of the solver (see Remark 3, p. 24). Note also that +class="cmtt-10x-x-109">PCG require it to +be distributed, MUMPS can be used with either a +replicated or a distributed matrix. When any of +the previous solvers is specified, the matrix layout +is set to a default value which allows the use of +the solver (see Remark 3, p. 24). Note also that UMFPACK and SuperLU_Dist are available only in double precision.

’COARSE_SUBSOLVE’

character(len=*)

’ILU’ -

’ILUT’ -

’MILU’ -

’MUMPS’ -

’SLU’ -

’UMF’

See Note.

Solver for the diagonal blocks of the coarse matrix, +class="td11">

’ILU’ +

’ILUT’ +

’MILU’ +

’MUMPS’ +

’SLU’ +

’UMF’ +

’INVT’ +

’INVK’ +

’AINV’

See Note.

Solver for the diagonal blocks of the coarse matrix, in case the block Jacobi solver is chosen as coarsest-level solver: ILU(p), ILU(p,t), MILU(p), -LU from MUMPS, SuperLU or UMFPACK -(plus triangular solve). Aggiungere Sparse -Approssimate per GPU? Note that UMFPACK -and SuperLU_Dist are available only in double -precision.

ILU otherwise.
-
Table 5: Parameters defining the coarse-space correction at the coarsest level.
- - - -
- -
- - - -

- - - -
-

-

s + +class="cmbx-10x-x-109">criterio di arresto del PCG? +class="cmtt-10x-x-109">’ +class="cmti-10x-x-109">_parameter

)

what

what

da type

val

e

val

default

t

coments

’COARSE_SWEEPS’

integer

Any integer -

number > 0

10

Number of sweeps when JACOBI, GS or BJAC is -chosen as coarsest-level solver. ’

integer

Any +integer +

number > +0

10

Number of sweeps when JACOBI, GS or BJAC +is chosen as coarsest-level solver. Aggiungere criterio di arresto del PCG?

’COARSE_FILLIN’

integer

Any integer -

number ≥ 0

0

Fill-in level p of the ILU factorizations.

integer

Any +integer +

number ≥ +0

0

Fill-in level p of the ILU factorizations.

’COARSE_ILUTHRS’

’

real( kind_parameter

)

Any real -

number ≥ 0

0

Drop tolerance t in the ILU(p,t) -factorization.

Any real +

number ≥ +0

0

Drop tolerance t in the ILU(p,t) factorization.

+ style="vertical-align:baseline;" id="TBL-7-12-"> +
+ + +
Table 6: Table 5: Parameters defining the coarse-space correction at the coarsest level -(continued).
+(continued). @@ -1399,27 +1492,27 @@ class="content">Parameters defining the coarse-space correction at the co -

-

-

+

what

what

da type

val

e

val

default

t

conts

’SMOOTHER_TYPE’

’

character(len=*)

=*)

’JACOBI’ -

’GS’ -

’BGS’ -

’BJAC’ -

’AS’

’

’FBGS’

’

Type of smoother used in the multilevel preconditioner: point-Jacobi, hybrid @@ -1499,39 +1592,55 @@ class="cmr-7">1-versions? and Additive Schwarz. -

It is ignored by one-level preconditioners.

’SUB_SOLVE’

’

character(len=*)

’JACOBI’ -

’GS’ -

=*)

’JACOBI’ +

’GS’ +

’BGS’ -

’ILU’ -

’ILUT’ -

’MILU’ -

’MUMPS’ -

’SLU’ -

’UMF’

’ILU’ +

’ILUT’ +

’MILU’ +

’MUMPS’ +

’SLU’ +

’UMF’

GS and BGS multilevel class="cmr-10">preconditioners, respectively -

ILU for block-Jacobi preconditioners ℓ1-versions?

-versions?

The local solver to be used with the smoother or one-level preconditioner (see @@ -1583,23 +1692,23 @@ class="cmr-10">or UMFPACK (plus triangular solve). See class="cmr-10">Note for details on hybrid Gauss-Seidel.

’SMOOTHER_SWEEPS’

integer

’

integer

Any integer -

number ≥ 0

1

0

1

Number of sweeps of the smoother or one-level preconditioner. In the multilevel @@ -1624,34 +1733,34 @@ class="cmr-10">, class="cmr-10">respectively.

’SUB_OVR’

integer

’

integer

Any integer -

number ≥ 0

1

0

1

Number of overlap layers, for Additive Schwarz only.


Table 7: Table 6: Parameters defining the smoother or the details of the one-level -preconditioner.
+preconditioner.
@@ -1661,27 +1770,27 @@ preconditioner.
-

-

-

+

what

what

da type

val

e

val

default

t

conts

’SUB_RESTR’

’

character(len=*)

=*)

’HALO’ -

’NONE’

’

’HALO’

’

Type of restriction operator, for Additive Schwarz only: ’NONE’ for neglecting it. -

Note that HALO must be chosen for the @@ -1758,29 +1867,29 @@ class="cmr-10">classical Addditive Schwarz smoother and class="cmr-10">its RAS variant.

’SUB_PROL’

’

character(len=*)

=*)

’SUM’ -

’NONE’

’

’NONE’

’

Type of prolongation operator, for Additive Schwarz only: ’ for neglecting them. -

Note that ’SUM’ for its RAS variant.

’SUB_FILLIN’

integer

’

integer

Any integer -

number ≥ 0

0

0

0

Fill-in level p of the incomplete LU @@ -1835,27 +1944,27 @@ class="cmr-10">of the incomplete LU class="cmr-10">factorizations.

’SUB_ILUTHRS’

’

real( kind_parameter

)

_parameter

)

Any real number ≥ 0

0

0

0

Drop tolerance t in the ILU() factorization.

’MUMPS_LOC_GLOB’

’

character(len=*)

=*)

’LOCAL_SOLVER’ -

’GLOBAL_SOLVER’

’

’GLOBAL_SOLVER’

’

Whether MUMPS should be used as a distributed solver, or as a serial solver acting @@ -1895,20 +2004,20 @@ class="cmr-10">only on the part of the matrix local to each process.

’MUMPS_IPAR_ENTRY’

integer

’

integer

Any integer number

0

number

0

Set an entry in the MUMPS integer control array, as chosen via the optional argument.

’MUMPS_RPAR_ENTRY’

real

Any real number

0

’

real

Any real number

0

Set an entry in the MUMPS real control array, as chosen via the optional class="cmr-10">argument.


Table 8: Table 7: Parameters defining the smoother or the details of the one-level preconditioner -(continued).
+(continued).
@@ -1955,7 +2064,7 @@ class="content">Parameters defining the smoother or the details of the one-level -

[next] [up]

-

diff --git a/docs/html/userhtmlsu9.html b/docs/html/userhtmlsu9.html index cf7ad525..5e246357 100644 --- a/docs/html/userhtmlsu9.html +++ b/docs/html/userhtmlsu9.html @@ -10,7 +10,7 @@ -

[nextMethod hierarchy_build

-

-

+

call p,info)

-

This method builds the hierarchy of matrices and restriction/prolongation operators for the multilevel preconditioner init and set. -

Arguments -

a

a

type(psb_in).

The sparse matrix structure containing the local part of the matrix to be preconditioned. Note that , single/double precision version of AMG4PSBLAS class="cmr-12">under use. See the PSBLAS User’s Guide for details [1617].

desc_a

desc_a

type(psb_desc_typein).

The communication descriptor of a. See the PSBLAS User’s Guide @@ -131,14 +131,14 @@ class="cmr-12">. See the PSBLAS User’s Guide class="cmr-12">for details [1617].

info

info

integer, out).

Error code. If no error, 0 is returned. See Section 7 for details.

@@ -165,7 +165,7 @@ class="td11">
-

[next] [up]

-

diff --git a/docs/src/bibliography.tex b/docs/src/bibliography.tex index a6da642e..bad19c07 100644 --- a/docs/src/bibliography.tex +++ b/docs/src/bibliography.tex @@ -12,6 +12,11 @@ P.~R.~Amestoy, C.~Ashcraft, O.~Boiteau, A.~Buttari, J.~L'Excellent, C.~Weisbecke SIAM Journal on Scientific Computing, volume 37 (3), 2015, A1452--A1474. See also {\tt http://mumps.enseeiht.fr}. % +\bibitem{BERTACCINIFILIPPONE} +D. Bertaccini\ and\ S. Filippone, +{\em Sparse approximate inverse preconditioners on high performance GPU platforms}, +Comput. Math. Appl. {\bf 71} (2016), no.~3, 693--711. +% \bibitem{BREZINA_VANEK} M.~Brezina, P.~Van\v{e}k, {\em A Black-Box Iterative Solver Based on a Two-Level Schwarz Method}, diff --git a/docs/src/overview.tex b/docs/src/overview.tex index b0aa1a3d..961ef7c3 100644 --- a/docs/src/overview.tex +++ b/docs/src/overview.tex @@ -25,7 +25,7 @@ coarse-level matrices and operators, without explicitly using any information on geometry of the original problem, e.g., the discretization of a PDE. To this end, two different coarsening strategies, based on aggregation, are available: \begin{itemize} -\item a decoupled version of the well known smoothed aggregation procedure proposed in~\cite{BREZINA_VANEK,VANEK_MANDEL_BREZINA}, and already included in the previous versions of the package~\cite{BDDF2007,MLD2P4_TOMS}; +\item a decoupled version of the well known smoothed aggregation procedure proposed in~\cite{BREZINA_VANEK,VANEK_MANDEL_BREZINA}, and already included in the previous versions of the package~\cite{MLD2P4_TOMS,BDDF2007}; \item the first parallel implementation of a coupled version of Coarsening based on Compatible Weighted Matching introduced in~\cite{DV2013,DFV2018} and described in details in~\cite{DDF2020}; \end{itemize} diff --git a/docs/src/userguide.tex b/docs/src/userguide.tex index 0325730b..1b1b0d65 100644 --- a/docs/src/userguide.tex +++ b/docs/src/userguide.tex @@ -16,6 +16,7 @@ \usepackage{html} \usepackage{ifthen} \usepackage{graphicx} +\usepackage{subfig} \newtheorem{theorem}{Theorem} \newtheorem{corollary}{Corollary} \usepackage{rotating} diff --git a/docs/src/userinterface.tex b/docs/src/userinterface.tex index 492446e2..52649de2 100644 --- a/docs/src/userinterface.tex +++ b/docs/src/userinterface.tex @@ -223,10 +223,10 @@ See also Sec.~\ref{sec:adding}. \fortinline|what| & \textsc{data type} & \fortinline|val| & \textsc{default} & \textsc{comments} \\ \hline \fortinline|'ML_CYCLE'| & \fortinline|character(len=*)| - & \texttt{'VCYCLE'} \par \texttt{'WCYCLE'} \par \texttt{'KCYCLE'} \par \texttt{'ADD'} - & \texttt{'VCYCLE'} + & \fortinline|'VCYCLE'| \par \fortinline|'WCYCLE'| \par \fortinline|'KCYCLE'| \par \fortinline|'ADD'| + & \fortinline|'VCYCLE'| &Multilevel cycle: V-cycle, W-cycle, K-cycle, and additive composition. \\ \hline - \fortinline|'OUTER_SWEEPS'| & \texttt{integer} & + \fortinline|'OUTER_SWEEPS'| & \fortinline|integer| & Any integer \par number $\ge 1$ & 1 & Number of multilevel cycles. \\ \hline @@ -279,54 +279,63 @@ be applied. & 20 & Maximum number of levels. The aggregation stops if the number of levels reaches this value (see Note). \\ \hline - \fortinline|'PAR_AGGR_ALG'| & \fortinline|character(len=*)| \hspace*{-3mm} - & \texttt{'DEC'}, \texttt{'SYMDEC'}, \texttt{'COUPLED'} - & \texttt{'DEC'} - & Parallel aggregation algorithm. \par the - \fortinline|SYMDEC| option applies decoupled - aggregation to the sparsity pattern - of $A+A^T$.\\ \hline - +& \texttt{'DEC'}, \texttt{'SYMDEC'}, \texttt{'COUPLED'} +& \texttt{'DEC'} +& Parallel aggregation algorithm. \par the +\fortinline|SYMDEC| option applies decoupled +aggregation to the sparsity pattern +of $A+A^T$.\\ \hline +\ifpdf +\end{tabular} +\end{center} +\esideways +\bsideways +\ContinuedFloat +\begin{center} +\begin{tabular}{|p{5.7cm}|l|p{2.3cm}|p{2.5cm}|p{5.9cm}|} +\hline +\fi +\fortinline|what| & \textsc{data type} & \fortinline|val| & \textsc{default} & +\textsc{comments} \\ \hline \fortinline|'AGGR_TYPE'| & \fortinline|character(len=*)| \hspace*{-3mm} - & \textbf{\texttt{'SOC1'}} & - \textbf{\texttt{'SOC1'}}, - \textbf{\texttt{'SOC2'}}, - \textbf{\texttt{'MATCHBOXP'}} - & Type of aggregation algorithm: currently, - for the decoupled aggregation we implement two measures of strength of - connection, the one by Van\v{e}k, Mandel - and Brezina~\cite{VANEK_MANDEL_BREZINA}, - and the one by Gratton et al~\cite{GrHeJi:16}. The coupled - aggregation is based on a parallel version of the half-approximate - matching implemented in the MatchBox-P software package - {\bf AGGIUNGERE LINK AL PACKAGE?}\\ \hline +& \fortinline|'SOC1'| & +\fortinline|'SOC1'|, +\fortinline|'SOC2'|, +\fortinline|'MATCHBOXP'| +& Type of aggregation algorithm: currently, +for the decoupled aggregation we implement two measures of strength of +connection, the one by Van\v{e}k, Mandel +and Brezina~\cite{VANEK_MANDEL_BREZINA}, +and the one by Gratton et al~\cite{GrHeJi:16}. The coupled +aggregation is based on a parallel version of the half-approximate +matching implemented in the MatchBox-P software package~\cite{MatchBoxP}.\\ \hline \fortinline|'AGGR_SIZE'| & \fortinline|integer| \hspace*{-3mm} - & Any integer \par number power of $2$ and $> 2$ - & 4 - & Maximum size of aggregates when the coupled aggregation based on - matching is applied. For aggressive coarsening with size of - aggregate larger than $8$ we recommend the use of smoothed prolongators. - {\bf MODIFICARE CODICE}\\ \hline +& Any integer \par number power of $2$ and $> 2$ +& 4 +& Maximum size of aggregates when the coupled aggregation based on +matching is applied. For aggressive coarsening with size of +aggregate larger than $8$ we recommend the use of smoothed prolongators. +{\bf MODIFICARE CODICE}\\ \hline \fortinline|'AGGR_PROL'| & \fortinline|character(len=*)| \hspace*{-3mm} - & \texttt{'SMOOTHED'}, \texttt{'UNSMOOTHED'} & \texttt{'SMOOTHED'} - & Prolongator used by the aggregation algorithm: smoothed or unsmoothed - (i.e., tentative prolongator). \\ +& \fortinline|'SMOOTHED'|, \fortinline|'UNSMOOTHED'| & \fortinline|'SMOOTHED'| +& Prolongator used by the aggregation algorithm: smoothed or unsmoothed +(i.e., tentative prolongator). \\ \hline \multicolumn{5}{|l|}{{\bfseries Note.} The aggregation algorithm stops when -at least one of the following criteria is met: -the coarse size threshold, the coarse size threshold per process, the} \\ -\multicolumn{5}{|l|}{minimum coarsening ratio, or the maximum number -of levels is reached. Therefore, the actual number of levels may be} \\ -\multicolumn{5}{|l|}{smaller than the specified maximum number -of levels. } \\ + at least one of the following criteria is met: + the coarse size threshold, } \\ +\multicolumn{5}{|l|}{the coarse size threshold per process, the minimum coarsening ratio, or the maximum number + of levels is reached.} \\ +\multicolumn{5}{|l|}{Therefore, the actual number of levels may be smaller than the specified maximum number + of levels. } \\ \hline \end{tabular} -\end{center} \caption{Parameters defining the aggregation algorithm. \label{tab:p_aggregation}} +\end{center} \esideways \bsideways @@ -378,14 +387,14 @@ the parameter \texttt{ilev}.} \\ \fortinline|what| & \textsc{data type} & \fortinline|val| & \textsc{default} & \textsc{comments} \\ \hline \fortinline|'COARSE_MAT'| & \fortinline|character(len=*)| - & \texttt{'DIST'} \par \texttt{'REPL'} - & \texttt{'REPL'} + & \fortinline|'DIST'| \par \fortinline|'REPL'| + & \fortinline|'REPL'| & Coarsest matrix layout: distributed among the processes or replicated on each of them. \\ \hline \fortinline|'COARSE_SOLVE'| & \fortinline|character(len=*)| - & \texttt{'MUMPS'} \par \texttt{'UMF'} \par - \texttt{'SLU'} \par \texttt{'SLUDIST'} \par - \texttt{'JACOBI'} \par \texttt{'GS'} \par \texttt{'BJAC'} \par \texttt{'PCG'} + & \fortinline|'MUMPS'| \par \fortinline|'UMF'| \par + \fortinline|'SLU'| \par \fortinline|'SLUDIST'| \par + \fortinline|'JACOBI'| \par \fortinline|'GS'| \par \fortinline|'BJAC'| \par \fortinline|'RKR'| & See~Note. & Solver used at the coarsest level: sequential LU from MUMPS, UMFPACK, or SuperLU @@ -394,7 +403,7 @@ the parameter \texttt{ilev}.} \\ (plus triangular solve); point-Jacobi, hybrid Gauss-Seidel or block-Jacobi and related $\ell_1$-versions; {\bf preconditioned Conjugate Gradient coupled with the block-Jacobi preconditioner - with ILU(0) on the blocks}. \par + with ILU(0) on the blocks}. Note that \texttt{UMF} and \texttt{SLU} require the coarsest matrix to be replicated, \texttt{SLUDIST}, \texttt{JACOBI}, \texttt{GS}, \texttt{BJAC} and \texttt{PCG} require it to be @@ -404,15 +413,28 @@ the parameter \texttt{ilev}.} \\ value which allows the use of the solver (see Remark 3, p.~24). Note also that UMFPACK and SuperLU\_Dist are available only in double precision. \\ \hline +\ifpdf +\end{tabular} +\end{center} +\esideways +\bsideways +\ContinuedFloat +\begin{center} +\begin{tabular}{|p{3.9cm}|l|p{1.7cm}|p{1.7cm}|p{8.6cm}|} +\hline +\fi \fortinline|'COARSE_SUBSOLVE'| & \fortinline|character(len=*)| - & \texttt{'ILU'} \par \texttt{'ILUT'} \par \texttt{'MILU'} \par - \texttt{'MUMPS'} \par \texttt{'SLU'} \par \texttt{'UMF'} + & \fortinline|'ILU'| \par \fortinline|'ILUT'| \par \fortinline|'MILU'| \par + \fortinline|'MUMPS'| \par \fortinline|'SLU'| \par \fortinline|'UMF'| \par + \fortinline|'INVT'| \par \fortinline|'INVK'| \par \fortinline|'AINV'| & See~Note. & Solver for the diagonal blocks of the coarse matrix, in case the block Jacobi solver is chosen as coarsest-level solver: ILU($p$), ILU($p,t$), MILU($p$), LU from MUMPS, SuperLU or UMFPACK - (plus triangular solve). {\bf Aggiungere Sparse Approssimate per GPU?} + (plus triangular solve). Suitable for GPUs (no triangular solve) + approximate inverse solvers INVK($p,q$), INVT($p_1,p2,t_1,t_2$) and + AINV($t$), see~\cite{BERTACCINIFILIPPONE}. Note that UMFPACK and SuperLU\_Dist are available only in double precision. \\ \hline @@ -425,16 +447,16 @@ the parameter \texttt{ilev}.} \\ then \texttt{MUMPS} if installed, then \texttt{SLU} if installed, \texttt{ILU} otherwise.}\\ \hline -\end{tabular} -\end{center} -\caption{Parameters defining the coarse-space correction at the coarsest -level.\label{tab:p_coarse}} -\esideways - -\bsideways -\begin{center} -\begin{tabular}{|p{3.9cm}|l|p{2cm}|p{1.5cm}|p{7.5cm}|} -\hline +%\end{tabular} +%\end{center} +%\caption{Parameters defining the coarse-space correction at the coarsest +%level.\label{tab:p_coarse}} +%\esideways +% +%\bsideways +%\begin{center} +%\begin{tabular}{|p{3.9cm}|l|p{2cm}|p{1.5cm}|p{7.5cm}|} +%\hline \fortinline|what| & \textsc{data type} & \fortinline|val| & \textsc{default} & \textsc{comments} \\ \hline \fortinline|'COARSE_SWEEPS'| & \fortinline|integer| @@ -476,10 +498,10 @@ level (continued).\label{tab:p_coarse_1}} Additive Schwarz. \par It is ignored by one-level preconditioners. \\ \hline \fortinline|'SUB_SOLVE'| & \fortinline|character(len=*)| - & \texttt{'JACOBI'} \par - \texttt{'GS'} \par \texttt{'BGS'} \par \texttt{'ILU'} \par - \texttt{'ILUT'} \par \texttt{'MILU'} \par - \par \texttt{'MUMPS'} \par \texttt{'SLU'} \par \texttt{'UMF'} + & \fortinline|'JACOBI'| \par + \fortinline|'GS'| \par \texttt{'BGS'} \par \fortinline|'ILU'| \par + \fortinline|'ILUT'| \par \fortinline|'MILU'| \par + \par \fortinline|'MUMPS'| \par \fortinline|'SLU'| \par \fortinline|'UMF'| & \texttt{GS} and \texttt{BGS} for pre- and post-smoothers of multilevel preconditioners, respectively \par \texttt{ILU} for block-Jacobi and Additive Schwarz