/usr/include/trilinos/Tpetra_MultiVectorFiller.hpp is in libtrilinos-tpetra-dev 12.12.1-5.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 | // @HEADER
// ***********************************************************************
//
// Tpetra: Templated Linear Algebra Services Package
// Copyright (2008) Sandia Corporation
//
// Under the terms of Contract DE-AC04-94AL85000 with Sandia Corporation,
// the U.S. Government retains certain rights in this software.
//
// 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. Neither the name of the Corporation nor the names of the
// contributors may be used to endorse or promote products derived from
// this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY SANDIA CORPORATION "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 SANDIA CORPORATION OR THE
// 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.
//
// Questions? Contact Michael A. Heroux (maherou@sandia.gov)
//
// ************************************************************************
// @HEADER
#ifndef __Tpetra_MultiVectorFiller_hpp
#define __Tpetra_MultiVectorFiller_hpp
#include "Tpetra_MultiVector.hpp"
#include "Tpetra_Vector.hpp"
#include "Teuchos_CommHelpers.hpp"
#include <iterator>
#include <set>
namespace Tpetra {
namespace Details {
// \fn sortAndMergeIn
// \brief Sort and merge newEntries into allEntries, and make unique.
//
// \param allEntries [in/out]: Array in which current entries have
// already been stored, and in which the new entries are to be
// stored. Passed by reference in case we need to resize it.
//
// \param currentEntries [in/out]: Current entries, which we assume
// have already been sorted and made unique. Aliases the
// beginning of allEntries.
//
// \param newEntries [in/out] New entries, which have not yet been
// sorted or made unique. This does <i>not</i> alias allEntries.
//
// Sort and make entries of newEntries unique. Resize allEntries if
// necessary to fit the unique entries of newEntries. Merge
// newEntries into allEntries and make the results unique. (This is
// cheaper than sorting the whole array.)
//
// \return A view of all the entries (current and new) in allEntries.
template<class T>
Teuchos::ArrayView<T>
sortAndMergeIn (Teuchos::Array<T>& allEntries,
Teuchos::ArrayView<T> currentEntries,
Teuchos::ArrayView<T> newEntries)
{
using Teuchos::ArrayView;
using Teuchos::as;
typedef typename ArrayView<T>::iterator iter_type;
typedef typename ArrayView<T>::size_type size_type;
std::sort (newEntries.begin(), newEntries.end());
iter_type it = std::unique (newEntries.begin(), newEntries.end());
const size_type numNew = as<size_type> (it - newEntries.begin());
// View of the sorted, made-unique new entries to merge in.
ArrayView<T> newEntriesView = newEntries.view (0, numNew);
const size_type numCur = currentEntries.size();
if (allEntries.size() < numCur + numNew) {
allEntries.resize (numCur + numNew);
}
ArrayView<T> allView = allEntries.view (0, numCur + numNew);
ArrayView<T> newView = allEntries.view (numCur, numNew); // target of copy
std::copy (newEntries.begin(), newEntries.end(), newView.begin());
std::inplace_merge (allView.begin(), newView.begin(), allView.end());
iter_type it2 = std::unique (allView.begin(), allView.end());
const size_type numTotal = as<size_type> (it2 - allView.begin());
return allEntries.view (0, numTotal);
}
/// \class MultiVectorFillerData
/// \brief Implementation of fill and local assembly for MultiVectorFiller.
/// \author Mark Hoemmen
///
/// \tparam MV Specialization of Tpetra::MultiVector.
template<class MV>
class MultiVectorFillerData {
public:
typedef typename MV::scalar_type scalar_type;
typedef typename MV::local_ordinal_type local_ordinal_type;
typedef typename MV::global_ordinal_type global_ordinal_type;
typedef typename MV::node_type node_type;
typedef ::Tpetra::Map<local_ordinal_type, global_ordinal_type, node_type> map_type;
/// \brief Default constructor (sets number of columns to zero).
///
/// \param map [in] Map, which this object may or may not use as a
/// hint to separate local from nonlocal data. This need not be
/// the same Map as that of the multivector output of
/// globalAssemble().
///
/// Before using this object, you should call setNumColumns() to
/// set the number of columns in the output multivector.
/// Otherwise, the two-argument version of sumIntoGlobalValues()
/// won't actually do anything.
MultiVectorFillerData (const Teuchos::RCP<const map_type>& map) :
map_ (map),
numCols_ (0)
{}
/// \brief Constructor.
///
/// \param map [in] Map, which this object may or may not use as a
/// hint to separate local from nonlocal data. This need not be
/// the same Map as that of the multivector output of
/// globalAssemble().
///
/// \param numColumns [in] The (expected) number of columns in the
/// output multivector. You can always change this later by
/// calling setNumColumns().
///
/// \note If the number of columns given here is not the same as
/// the number of columns in the output multivector, you should
/// call setNumColumns() first before inserting any data.
/// Otherwise, the two-argument version of sumIntoGlobalValues()
/// won't do the right thing.
MultiVectorFillerData (const Teuchos::RCP<const map_type>& map,
const size_t numColumns) :
map_ (map),
numCols_ (numColumns),
sourceIndices_ (numColumns),
sourceValues_ (numColumns)
{}
//! Set the number of columns in the output multivector.
void
setNumColumns (const size_t newNumColumns)
{
const size_t oldNumColumns = getNumColumns();
if (newNumColumns >= oldNumColumns) {
for (size_t j = oldNumColumns; j < newNumColumns; ++j) {
sourceIndices_.push_back (Teuchos::Array<global_ordinal_type> ());
sourceValues_.push_back (Teuchos::Array<scalar_type> ());
}
}
else {
// This may not necessarily deallocate any data, but that's OK.
sourceIndices_.resize (newNumColumns);
sourceValues_.resize (newNumColumns);
}
numCols_ = oldNumColumns;
}
void
sumIntoGlobalValues (const Teuchos::ArrayView<const global_ordinal_type>& rows,
const size_t column,
const Teuchos::ArrayView<const scalar_type>& values)
{
if (column >= getNumColumns()) {
for (size_t j = column; j < getNumColumns(); ++j) {
sourceIndices_.push_back (Teuchos::Array<global_ordinal_type> ());
sourceValues_.push_back (Teuchos::Array<scalar_type> ());
}
}
std::copy (rows.begin(), rows.end(), std::back_inserter (sourceIndices_[column]));
std::copy (values.begin(), values.end(), std::back_inserter (sourceValues_[column]));
}
/// Data for each column are stored contiguously in rows and in
/// values. Thus, rows and values are in rowwise order, even
/// though they may be stored in columnwise order in the
/// multivector.
///
/// Be sure that the number of columns is set correctly before
/// calling this.
void
sumIntoGlobalValues (const Teuchos::ArrayView<const global_ordinal_type>& rows,
const Teuchos::ArrayView<const scalar_type>& values)
{
typedef global_ordinal_type GO;
typedef scalar_type ST;
typedef typename Teuchos::ArrayView<const GO>::const_iterator GoIter;
typedef typename Teuchos::ArrayView<const ST>::const_iterator StIter;
const size_t numColumns = getNumColumns();
GoIter rowIter = rows.begin();
StIter valIter = values.begin();
for (size_t j = 0; j < numColumns; ++j) {
GoIter rowIterNext = rowIter + numColumns;
StIter valIterNext = valIter + numColumns;
std::copy (rowIter, rowIterNext, std::back_inserter (sourceIndices_[j]));
std::copy (valIter, valIterNext, std::back_inserter (sourceValues_[j]));
rowIter = rowIterNext;
valIter = valIterNext;
}
}
/// \brief Locally assemble into X.
///
/// \param X [in/out] Multivector (overlapping source distribution).
///
/// \param f [in/out] Binary function that defines the combine
/// mode. It must define scalar_type operator (const
/// scalar_type&, const scalar_type&). It need not necessarily
/// be commutative or even associative, but it should be
/// thread-safe in case we decide to parallelize local assembly.
/// We call it via X(i,j) = f(X(i,j), Y(i,j)), so write your
/// possibly nonassociative or noncommutative operation
/// accordingly.
///
/// X is distributed by the source Map (with possible overlap) of
/// the Export operation. The source Map of the Export includes
/// both the elements owned by this object's constructor's input
/// Map, and the indices inserted by sumIntoGlobalValues().
///
/// Precondition: The set of global indices in X's Map equals the
/// the union of the entries of nonlocalIndices_[j] for all valid
/// columns j.
///
/// \note You can get the usual ADD combine mode by supplying f =
/// std::plus<scalar_type>.
template<class BinaryFunction>
void
locallyAssemble (MV& X, BinaryFunction& f)
{
using Teuchos::Array;
using Teuchos::ArrayRCP;
using Teuchos::ArrayView;
using Teuchos::RCP;
typedef local_ordinal_type LO;
typedef global_ordinal_type GO;
typedef scalar_type ST;
typedef node_type NT;
RCP<const ::Tpetra::Map<LO, GO, NT> > map = X.getMap();
Array<LO> localIndices;
const size_t numColumns = getNumColumns();
for (size_t j = 0; j < numColumns; ++j) {
const typename Array<const GO>::size_type numIndices = sourceIndices_[j].size();
// Precompute all the local indices before saving to the
// vector. Hopefully this gives us a little bit of locality
// in the global->local conversion, at the expense of a little
// more storage.
if (localIndices.size() < numIndices) {
localIndices.resize (numIndices);
}
ArrayView<LO> localIndicesView = localIndices.view (0, numIndices);
ArrayView<const GO> globalIndicesView = sourceIndices_[j].view (0, numIndices);
for (typename ArrayView<const GO>::size_type i = 0; i < numIndices; ++i) {
localIndices[i] = map->getLocalElement (globalIndicesView[i]);
}
ArrayRCP<ST> X_j = X.getDataNonConst (j);
ArrayView<const ST> localValues = sourceValues_[j].view (0, numIndices);
for (typename ArrayView<const GO>::size_type i = 0; i < numIndices; ++i) {
const LO localInd = localIndices[i];
X_j[localInd] = f (X_j[localInd], localValues[i]);
}
}
}
//! Overload of locallyAssemble() for the usual ADD combine mode.
void
locallyAssemble (MV& X)
{
std::plus<double> f;
locallyAssemble<std::plus<scalar_type> > (X, f);
}
//! Clear the contents of the vector, making it implicitly a vector of zeros.
void clear() {
Teuchos::Array<Teuchos::Array<global_ordinal_type> > newSourceIndices;
Teuchos::Array<Teuchos::Array<scalar_type> > newSourceValues;
// The standard STL idiom for clearing the contents of a vector completely.
std::swap (sourceIndices_, newSourceIndices);
std::swap (sourceValues_, newSourceValues);
}
//! All source indices (local and nonlocal) of the source Map, sorted and unique.
Teuchos::Array<global_ordinal_type>
getSourceIndices () const
{
using Teuchos::Array;
using Teuchos::ArrayView;
using Teuchos::as;
typedef global_ordinal_type GO;
typedef typename Array<GO>::size_type size_type;
Array<GO> allInds (0); // will resize below
const size_t numCols = getNumColumns();
if (numCols == 1) {
// Special case for 1 column avoids copying indices twice.
// Pick the size of the array exactly the first time so there
// are at most two allocations (the constructor may choose to
// allocate).
const size_type numNew = sourceIndices_[0].size();
allInds.resize (allInds.size() + numNew);
std::copy (sourceIndices_[0].begin(), sourceIndices_[0].end(),
allInds.begin());
std::sort (allInds.begin(), allInds.end());
typename Array<GO>::iterator it =
std::unique (allInds.begin(), allInds.end());
const size_type numFinal = as<size_type> (it - allInds.begin());
allInds.resize (numFinal);
}
else {
// Carefully collect all the row indices one column at a time.
// This ensures that the total allocation size in this routine
// is independent of the number of columns. Also, only sort
// the current column's indices. Use merges to ensure sorted
// order in the collected final result.
ArrayView<GO> curIndsView = allInds.view (0, 0); // will grow
Array<GO> newInds;
for (size_t j = 0; j < numCols; ++j) {
const size_type numNew = sourceIndices_[j].size();
if (numNew > newInds.size()) {
newInds.resize (numNew);
}
ArrayView<GO> newIndsView = newInds.view (0, numNew);
std::copy (sourceIndices_[j].begin(), sourceIndices_[j].end(),
newIndsView.begin());
curIndsView = sortAndMergeIn<GO> (allInds, curIndsView, newIndsView);
}
}
return allInds;
}
private:
Teuchos::RCP<const map_type> map_;
size_t numCols_;
Teuchos::Array<Teuchos::Array<global_ordinal_type> > sourceIndices_;
Teuchos::Array<Teuchos::Array<scalar_type> > sourceValues_;
size_t getNumColumns() const { return numCols_; }
};
/// \class MultiVectorFillerData2
/// \brief Second implementation of fill and local assembly for
/// MultiVectorFiller.
/// \author Mark Hoemmen
///
/// \tparam MV Specialization of Tpetra::MultiVector.
template<class MV>
class MultiVectorFillerData2 : public Teuchos::Describable {
public:
typedef typename MV::scalar_type scalar_type;
typedef typename MV::local_ordinal_type local_ordinal_type;
typedef typename MV::global_ordinal_type global_ordinal_type;
typedef typename MV::node_type node_type;
typedef ::Tpetra::Map<local_ordinal_type, global_ordinal_type, node_type> map_type;
/// \brief Default constructor (sets number of columns to zero).
///
/// \param map [in] Map over which to distribute the initial fill.
/// This need not be the same Map as that of the multivector
/// output of globalAssemble().
///
/// Before using this object, you should call setNumColumns() to
/// set the number of columns in the output multivector.
/// Otherwise, the two-argument version of sumIntoGlobalValues()
/// won't actually do anything.
MultiVectorFillerData2 (const Teuchos::RCP<const map_type>& map,
const Teuchos::EVerbosityLevel verbLevel=Teuchos::VERB_DEFAULT,
const Teuchos::RCP<Teuchos::FancyOStream>& out=Teuchos::null) :
map_ (map),
numCols_ (0),
verbLevel_ (verbLevel),
out_ (out)
{}
/// \brief Constructor.
///
/// \param map [in] Map over which to distribute the initial fill.
/// This need not be the same Map as that of the multivector
/// output of globalAssemble(), but the Map must have the
/// same communicator as the multivector output of
/// globalAssemble().
///
/// \param numColumns [in] The (expected) number of columns in the
/// output multivector. You can always change this later by
/// calling setNumColumns().
///
/// \note If the number of columns given here is not the same as
/// the number of columns in the output multivector, you should
/// call setNumColumns() first before inserting any data.
/// Otherwise, the two-argument version of sumIntoGlobalValues()
/// won't do the right thing.
MultiVectorFillerData2 (const Teuchos::RCP<const map_type>& map,
const size_t numColumns,
const Teuchos::EVerbosityLevel verbLevel=Teuchos::VERB_DEFAULT,
const Teuchos::RCP<Teuchos::FancyOStream>& out=Teuchos::null) :
map_ (map),
numCols_ (numColumns),
localVec_ (new MV (map, numColumns)),
#if 0
nonlocalIndices_ (numColumns),
nonlocalValues_ (numColumns),
#endif // 0
verbLevel_ (verbLevel),
out_ (out)
{}
std::string
description() const
{
std::ostringstream oss;
oss << "Tpetra::MultiVectorFillerData2<"
<< Teuchos::TypeNameTraits<MV>::name () << ">";
return oss.str();
}
void
describe (Teuchos::FancyOStream& out,
const Teuchos::EVerbosityLevel verbLevel =
Teuchos::Describable::verbLevel_default) const
{
using std::endl;
using Teuchos::Array;
using Teuchos::ArrayRCP;
using Teuchos::ArrayView;
using Teuchos::RCP;
using Teuchos::VERB_DEFAULT;
using Teuchos::VERB_NONE;
using Teuchos::VERB_LOW;
using Teuchos::VERB_MEDIUM;
using Teuchos::VERB_HIGH;
using Teuchos::VERB_EXTREME;
// Set default verbosity if applicable.
const Teuchos::EVerbosityLevel vl =
(verbLevel == VERB_DEFAULT) ? VERB_LOW : verbLevel;
RCP<const Teuchos::Comm<int> > comm = map_->getComm();
const int myImageID = comm->getRank();
const int numImages = comm->getSize();
if (vl != VERB_NONE) {
// Don't set the tab level unless we're printing something.
Teuchos::OSTab tab (out);
if (myImageID == 0) { // >= VERB_LOW prints description()
out << this->description() << endl;
}
for (int imageCtr = 0; imageCtr < numImages; ++imageCtr) {
if (myImageID == imageCtr) {
if (vl != VERB_LOW) {
// At verbosity > VERB_LOW, each process prints something.
out << "Process " << myImageID << ":" << endl;
Teuchos::OSTab procTab (out);
// >= VERB_MEDIUM: print the local vector length.
out << "local length=" << localVec_->getLocalLength();
if (vl != VERB_MEDIUM) {
// >= VERB_HIGH: print isConstantStride() and getStride()
if (localVec_->isConstantStride()) {
out << ", constant stride=" << localVec_->getStride() << endl;
}
else {
out << ", not constant stride" << endl;
}
if (vl == VERB_EXTREME) {
// VERB_EXTREME: print all the values in the multivector.
out << "Local values:" << endl;
ArrayRCP<ArrayRCP<const scalar_type> > X = localVec_->get2dView();
for (size_t i = 0; i < localVec_->getLocalLength(); ++i) {
for (size_t j = 0; j < localVec_->getNumVectors(); ++j) {
out << X[j][i];
if (j < localVec_->getNumVectors() - 1) {
out << " ";
}
} // for each column
out << endl;
} // for each row
#if 0
out << "Nonlocal indices and values:" << endl;
for (size_t j = 0; j < (size_t)nonlocalIndices_.size(); ++j) {
ArrayView<const global_ordinal_type> inds = nonlocalIndices_[j]();
ArrayView<const scalar_type> vals = nonlocalValues_[j]();
for (typename ArrayView<const global_ordinal_type>::size_type k = 0;
k < inds.size(); ++k) {
out << "X(" << inds[k] << "," << j << ") = " << vals[k] << endl;
}
}
#endif // 0
} // if vl == VERB_EXTREME
} // if (vl != VERB_MEDIUM)
else { // vl == VERB_LOW
out << endl;
}
} // if vl != VERB_LOW
} // if it is my process' turn to print
} // for each process in the communicator
} // if vl != VERB_NONE
}
//! All source indices (local and nonlocal) of the source Map, sorted and unique.
Teuchos::Array<global_ordinal_type>
getSourceIndices (const Teuchos::Comm<int>& comm,
const Teuchos::RCP<Teuchos::FancyOStream>& out = Teuchos::null,
const bool debug = false) const
{
using Teuchos::Array;
using Teuchos::ArrayView;
using Teuchos::RCP;
using Teuchos::rcp;
using Tpetra::global_size_t;
typedef global_ordinal_type GO;
const char prefix[] = "Tpetra::MultiVectorFiller::getSourceIndices: ";
if (debug && ! out.is_null ()) {
std::ostringstream os;
os << "Proc " << comm.getRank () << ": getSourceIndices" << std::endl;
*out << os.str ();
}
// Get the nonlocal row indices, sorted and made unique.
// It's fair to assume that these are not contiguous.
Array<GO> nonlocalIndices = getSortedUniqueNonlocalIndices (comm, out, debug);
// Get the local row indices, not necessarily sorted or unique.
ArrayView<const GO> localIndices = getLocalIndices ();
// Copy the local indices into the full indices array, and sort
// them there. We'll merge in the nonlocal indices below. This
// can be more efficient than just sorting all the indices, if
// there are a lot of nonlocal indices.
Array<GO> indices (localIndices.size () + nonlocalIndices.size ());
ArrayView<GO> localIndView = indices.view (0, localIndices.size ());
TEUCHOS_TEST_FOR_EXCEPTION
(localIndView.size () > indices.size (), std::logic_error,
prefix << "localIndView.size() = " << localIndView.size ()
<< " > indices.size() = " << indices.size () << ".");
TEUCHOS_TEST_FOR_EXCEPTION
(localIndView.size () != localIndices.size (), std::logic_error,
prefix << "localIndView.size() = " << localIndView.size ()
<< " != localIndices.size() = " << localIndices.size () << ".");
std::copy (localIndices.begin (), localIndices.end (), localIndView.begin ());
std::sort (localIndView.begin (), localIndView.end ());
if (debug && ! out.is_null ()) {
std::ostringstream os;
os << "Proc " << comm.getRank () << ": Right after copying and sorting" << std::endl;
*out << os.str ();
}
// Merge the local and nonlocal indices.
if (nonlocalIndices.size () > 0) {
ArrayView<GO> nonlclIndView =
indices.view (localIndices.size (), nonlocalIndices.size ());
// We need a view of the output array, because
// std::inplace_merge needs all its iterator inputs to have
// the same type. Debug mode builds are pickier than release
// mode builds, because the iterators in a debug mode build
// are of a different type that does run-time checking (they
// aren't just raw pointers).
ArrayView<GO> indView = indices.view (0, indices.size ());
if (debug && ! out.is_null ()) {
std::ostringstream os;
os << "Right before std::copy" << std::endl;
*out << os.str ();
}
std::copy (nonlocalIndices.begin (),
nonlocalIndices.end (),
nonlclIndView.begin ());
if (debug && ! out.is_null ()) {
std::ostringstream os;
os << "Proc " << comm.getRank () << ": Right before std::inplace_merge"
<< std::endl;
*out << os.str ();
}
std::inplace_merge (indView.begin (),
indView.begin () + localIndices.size (),
indView.end ());
}
if (debug && ! out.is_null ()) {
std::ostringstream os;
os << "Proc " << comm.getRank () << ": Done with getSourceIndices"
<< std::endl;
*out << os.str ();
}
return indices;
}
/// \brief Set the number of columns in the output multivector.
///
/// \param newNumColumns [in] The new number of columns in the
/// multivector. Zero is allowed; it means "clear local
/// storage."
///
/// Setting the number of columns to zero effectively clears out
/// all local storage, but may not necessarily deallocate nonlocal
/// storage. Call clear() to clear out all nonlocal storage.
void
setNumColumns (const size_t newNumColumns)
{
using Teuchos::Array;
using Teuchos::Range1D;
using Teuchos::RCP;
const size_t oldNumColumns = numCols_;
if (newNumColumns == oldNumColumns) {
return; // No side effects if no change.
}
RCP<MV> newLclVec;
if (newNumColumns > oldNumColumns) {
newLclVec = rcp (new MV (map_, newNumColumns));
// Assign the contents of the old local multivector to the
// first oldNumColumns columns of the new local multivector,
// then get rid of the old local multivector.
RCP<MV> newLclVecView =
newLclVec->subViewNonConst (Range1D (0, oldNumColumns-1));
*newLclVecView = *localVec_;
}
else {
if (newNumColumns == 0) {
// Tpetra::MultiVector doesn't let you construct a
// multivector with zero columns.
newLclVec = Teuchos::null;
}
else {
newLclVec = localVec_->subViewNonConst (Range1D (0, newNumColumns-1));
}
}
// Leave most side effects until the end, for exception safety.
localVec_ = newLclVec;
numCols_ = newNumColumns;
}
/// \brief Set entry (rows[i],columnIndex) to values[i], for all i.
///
/// \param rows [in] Which rows of the MultiVector to modify.
/// \param columnIndex [in] Which column of the MultiVector to modify.
/// \param values [in] The new values.
void
sumIntoGlobalValues (const Teuchos::ArrayView<const global_ordinal_type>& rows,
const size_t columnIndex,
const Teuchos::ArrayView<const scalar_type>& values)
{
using Teuchos::ArrayView;
typedef local_ordinal_type LO;
typedef global_ordinal_type GO;
typedef scalar_type ST;
typedef decltype (rows.size ()) size_type;
const char prefix[] = "Tpetra::MultiVectorFiller::sumIntoGlobalValues: ";
if (map_.is_null ()) {
return; // the calling process is not participating
}
const size_type numEnt = rows.size ();
TEUCHOS_TEST_FOR_EXCEPTION
(numEnt != values.size (), std::invalid_argument, prefix
<< "rows.size() = " << numEnt << " != values.size() = "
<< values.size () << ".");
// FIXME (31 Aug 2015) Don't do this.
if (columnIndex >= getNumColumns()) {
// Automatically expand the number of columns. This
// implicitly ensures that localVec_ is not null.
setNumColumns (columnIndex + 1);
}
// It would be a lot more efficient for users to get the Kokkos
// view outside of their fill loop, and fill into it directly.
auto X_j = localVec_->getVector (columnIndex);
auto X_j_2d = X_j->template getLocalView<typename MV::dual_view_type::t_host::memory_space> ();
auto X_j_1d = Kokkos::subview (X_j_2d, Kokkos::ALL (), 0);
const map_type& theMap = *map_;
for (size_type k = 0; k < numEnt; ++k) {
const ST val = values[k];
const GO gblRowInd = rows[k];
const LO lclRowInd = theMap.getLocalElement (gblRowInd);
if (lclRowInd == Teuchos::OrdinalTraits<LO>::invalid ()) {
// The row doesn't belong to the calling process, so stuff
// its data in nonlocals_.
auto& innerMap = nonlocals_[columnIndex];
auto innerIter = innerMap.find (gblRowInd);
if (innerIter == innerMap.end ()) {
innerMap.insert (std::make_pair (gblRowInd, values[k]));
} else {
innerIter->second += val;
}
}
else {
// FIXME (mfh 27 Feb 2012) Allow different combine modes.
// The current combine mode just adds to the current value
// at that location.
X_j_1d[lclRowInd] += val;
}
}
}
/// \brief Set entry (rows[i],j) to values[i], for all i and j.
///
/// Data for each column are stored contiguously in rows and in
/// values. Thus, rows and values are in rowwise order, even
/// though they may be stored in columnwise order in the
/// multivector.
///
/// Be sure that the number of columns is set correctly before
/// calling this.
void
sumIntoGlobalValues (const Teuchos::ArrayView<const global_ordinal_type>& rows,
const Teuchos::ArrayView<const scalar_type>& values)
{
using Teuchos::ArrayView;
typedef typename ArrayView<const global_ordinal_type>::size_type size_type;
const size_t numCols = getNumColumns();
for (size_t j = 0; j < numCols; ++j) {
const size_type offset = numCols*j;
const size_type len = numCols;
this->sumIntoGlobalValues (rows.view (offset, len), j, values.view (offset, len));
}
}
/// \brief Locally assemble into X, with user-specified combine mode.
///
/// \param X [in/out] Multivector (overlapping source distribution).
///
/// \param f [in/out] Binary function that defines the combine
/// mode. It must define scalar_type operator (const
/// scalar_type&, const scalar_type&). It need not necessarily
/// be commutative or even associative, but it should be
/// thread-safe in case we decide to parallelize local assembly.
/// We call it via X(i,j) = f(X(i,j), Y(i,j)), so write your
/// possibly nonassociative or noncommutative operation
/// accordingly.
///
/// X is distributed by the source Map (with possible overlap) of
/// the Export operation. The source Map of the Export includes
/// both the elements owned by this object's constructor's input
/// Map, and the indices inserted by sumIntoGlobalValues().
///
/// Precondition: The set of global indices in X's Map equals the
/// union of the global indices in map_ and (the union of the
/// entries of nonlocalIndices_[j] for all valid columns j).
///
/// \note You can get the usual Tpetra::ADD combine mode by
/// supplying f = std::plus<scalar_type>.
template<class BinaryFunction>
void
locallyAssemble (MV& X, BinaryFunction& f)
{
typedef scalar_type ST;
typedef local_ordinal_type LO;
typedef global_ordinal_type GO;
typedef typename MV::dual_view_type::t_host::memory_space host_memory_space;
if (X.getMap ().is_null ()) {
return; // this process is not participating
}
const map_type& srcMap = * (X.getMap ());
for (size_t j = 0; j < X.getNumVectors (); ++j) {
auto X_j = X.getVector (j);
auto X_j_2d = X_j->template getLocalView<host_memory_space> ();
auto X_j_1d = Kokkos::subview (X_j_2d, Kokkos::ALL (), 0);
// First, assemble in the local components from localVec_.
if (! localVec_.is_null () && ! localVec_->getMap ().is_null ()
&& j < localVec_->getNumVectors ()) {
auto lcl_j = localVec_->getVector (j);
auto lcl_j_2d = lcl_j->template getLocalView<host_memory_space> ();
auto lcl_j_1d = Kokkos::subview (lcl_j_2d, Kokkos::ALL (), 0);
// Iterate over all rows of localVec_. i_lcl is a local
// index with respect to localVec_'s Map, which may differ
// from X's Map.
const map_type& lclMap = * (localVec_->getMap ());
const LO lclNumRows = static_cast<LO> (lcl_j->getLocalLength ());
for (LO i_lcl = 0; i_lcl < lclNumRows; ++i_lcl) {
const GO i_gbl = lclMap.getGlobalElement (i_lcl);
const LO i_X = srcMap.getLocalElement (i_gbl);
X_j_1d(i_X) = f (X_j_1d(i_X), lcl_j_1d(i_lcl));
}
}
// Second, assemble in the nonlocal components.
auto outerIter = nonlocals_.find (j);
if (outerIter != nonlocals_.end ()) {
auto beg = outerIter->second.begin ();
auto end = outerIter->second.end ();
for (auto innerIter = beg; innerIter != end; ++innerIter) {
const GO gblRowInd = innerIter->first;
const LO lclRowInd = srcMap.getLocalElement (gblRowInd);
if (lclRowInd != Teuchos::OrdinalTraits<LO>::invalid ()) {
const ST val = innerIter->second;
X_j_1d(lclRowInd) = f (X_j_1d(lclRowInd), val);
}
}
}
}
}
/// \brief Do the "local" (not MPI) part of assembly
///
/// Call this method for the usual Tpetra::ADD combine mode. If
/// you want a different combine mode, call the overload that
/// takes two arguments, one of which is a binary function.
void
locallyAssemble (MV& X)
{
std::plus<double> f;
locallyAssemble<std::plus<scalar_type> > (X, f);
}
/// \brief Clear the contents of the multivector.
///
/// This fills the vector with zeros, and also removes nonlocal
/// data. It does <i>not</i> deallocate all storage. For that,
/// you need to set the number of columns to zero.
void clear ()
{
std::map<size_t, std::map<global_ordinal_type, scalar_type> > newNonlcls;
// The standard STL idiom for clearing the contents of a
// container completely. Setting the size to zero may not
// necessarily deallocate data.
std::swap (nonlocals_, newNonlcls);
// Don't actually deallocate the multivector of local entries.
// Just fill it with zero. This is because the caller hasn't
// reset the number of columns.
if (! localVec_.is_null ()) {
localVec_->putScalar (Teuchos::ScalarTraits<scalar_type>::zero());
}
}
private:
//! Map that tells us which entries are locally owned.
Teuchos::RCP<const map_type> map_;
//! The number of columns in the (output) multivector.
size_t numCols_;
//! Multivector of locally stored (i.e., owned by \c map_) entries.
Teuchos::RCP<MV> localVec_;
//! Map from column index to (map from row index to value).
std::map<size_t, std::map<global_ordinal_type, scalar_type> > nonlocals_;
//! Verbosity level of this object (mainly used for debugging output).
Teuchos::EVerbosityLevel verbLevel_;
//! Output stream (mainly used for debugging output).
Teuchos::RCP<Teuchos::FancyOStream> out_;
//! The number of columns in the (output) multivector.
size_t getNumColumns() const { return numCols_; }
//! The locally owned row indices, not necessarily sorted or unique.
Teuchos::ArrayView<const global_ordinal_type>
getLocalIndices() const
{
return map_->getNodeElementList ();
}
//! The inserted nonlocal row indices, sorted and made unique.
Teuchos::Array<global_ordinal_type>
getSortedUniqueNonlocalIndices (const Teuchos::Comm<int>& comm,
const Teuchos::RCP<Teuchos::FancyOStream>& out = Teuchos::null,
const bool debug = false) const
{
using Teuchos::Array;
using Teuchos::ArrayView;
using Teuchos::as;
typedef global_ordinal_type GO;
if (debug && ! out.is_null ()) {
std::ostringstream os;
os << "Proc " << comm.getRank () << ": getSortedUniqueNonlocalIndices"
<< std::endl;
*out << os.str ();
}
// FIXME (mfh 30 Aug 2015) The std::set algorithm is harder to
// thread-parallelize, but it should be easier to make the new
// test pass with this.
std::set<GO> indsOut;
const size_t numCols = getNumColumns ();
for (size_t j = 0; j < numCols; ++j) {
auto outerIter = nonlocals_.find (j);
if (outerIter != nonlocals_.end ()) {
auto beg = outerIter->second.begin ();
auto end = outerIter->second.end ();
for (auto innerIter = beg; innerIter != end; ++innerIter) {
// *innerIter: (global row index, value) pair.
indsOut.insert (innerIter->first);
}
}
}
if (debug && ! out.is_null ()) {
std::ostringstream os;
os << "Proc " << comm.getRank () << ": Made nonlocals set" << std::endl;
*out << os.str ();
}
Array<GO> allNonlocals (indsOut.begin (), indsOut.end ());
if (debug && ! out.is_null ()) {
std::ostringstream os;
os << "Proc " << comm.getRank () << ": Done with "
"getSortedUniqueNonlocalIndices" << std::endl;
*out << os.str ();
}
return allNonlocals;
}
};
} // namespace Details
} // namespace Tpetra
namespace Tpetra {
/// \class MultiVectorFiller
/// \brief Adds nonlocal sum-into functionality to Tpetra::MultiVector.
/// \author Mark Hoemmen
///
/// MultiVector does not allow modifications to rows not owned by
/// the calling process. This makes the implementation of
/// MultiVector simple, fast, and local, but it complicates
/// operations like assembly of the right-hand side in the finite
/// element method. CrsMatrix already supports insertion into
/// nonowned rows, so that makes assembly not consistent for sparse
/// matrices and vectors. MultiVectorFiller remedies this by
/// specializing for the case of filling a (multi)vector. It
/// provides a constructor and sumIntoGlobalValues() methods that
/// look just like those of MultiVector, so you can swap it in at
/// the assembly stage. Then, you can call globalAssemble() to turn
/// the MultiVectorFiller into a standard MultiVector.
///
/// The globalAssemble() method will redistribute data if necessary.
/// You can force it to use the Map provided to the constructor (or
/// the Map provided to the last call to globalAssemble(), which is
/// a common use case for doing nonlinear solves where the sparse
/// matrix structure doesn't change).
///
/// \tparam MV Specialization of Tpetra::MultiVector.
template<class MV>
class MultiVectorFiller {
public:
typedef typename MV::scalar_type scalar_type;
typedef typename MV::local_ordinal_type local_ordinal_type;
typedef typename MV::global_ordinal_type global_ordinal_type;
typedef typename MV::node_type node_type;
typedef ::Tpetra::Map<local_ordinal_type, global_ordinal_type, node_type> map_type;
/// \brief Constructor.
///
/// The constructor takes the same arguments as MV's constructor.
/// We did this on purpose, so that you can construct
/// MultiVectorFiller as you would a standard MV.
///
/// \param map [in] A Map with the same communicator and Kokkos
/// Node as the output multivector of globalAssemble(). This
/// need not be the same Map as that output multivector's Map.
///
/// \param numCols [in] Expected number of columns in the output
/// multivector of globalAssemble().
///
/// \note If the number of columns given here is not the same as
/// the number of columns in the output multivector, the
/// two-argument version of sumIntoGlobalValues() won't do
/// the right thing. Use the three-argument version of that
/// method if you don't know how many columns there will be in
/// advance.
///
/// \note Not providing the output multivector in the constructor
/// gives globalAssemble() more flexibility. For example, its
/// output multivector may have any distribution with the same
/// global number of rows. Furthermore, decoupling insertion
/// from the output multivector lets this object store
/// preassembled data in whatever format it likes. It doesn't
/// have to insert directly into the output multivector until
/// assembly time. (This may improve performance by amortizing
/// the cost of global->local index conversions, for example.)
MultiVectorFiller (const Teuchos::RCP<const map_type>& map,
const size_t numCols);
/// \brief Assemble all the data (local and nonlocal) into X_out.
///
/// You can call this method multiple times with different
/// multivector output arguments. If those arguments have the
/// same Map, this method will attempt to reuse the Export object
/// each time. It will only reuse if the new target Map is the
/// same as (in the sense of Map::isSameAs()) the previous target
/// Map, unless you force reuse with the second argument (that
/// saves a few global reductions for the check).
///
/// \param X_out [in/out] MultiVector with a nonoverlapping Map.
/// That Map need not be the same as the one with which this
/// object was created.
///
/// \param forceReuseMap [in] If true, assume that X_out has the
/// same Map (in the sense of Map::isSameAs()) as the target Map
/// in the previous call to this method. If this method was not
/// called before, then assume that X_out has the same Map as
/// the argument to the constructor of this object.
void
globalAssemble (MV& X_out, const bool forceReuseMap = false);
//! Clear contents, in preparation for another fill cycle.
void clear () {
data_.clear ();
}
void
describe (Teuchos::FancyOStream& out,
const Teuchos::EVerbosityLevel verbLevel=Teuchos::Describable::verbLevel_default) const
{
data_.describe (out, verbLevel);
}
/// \brief Sum data into the multivector.
///
/// \param rows [in] Array of global rows for which to insert
/// values. Must have the same length as values.
///
/// \param column [in] Index of the column in which to insert.
///
/// \param values [in] Array of values to insert. Must have the
/// same length as rows. rows[i] is the row in which values[i]
/// is to be inserted.
void
sumIntoGlobalValues (Teuchos::ArrayView<const global_ordinal_type> rows,
size_t column,
Teuchos::ArrayView<const scalar_type> values)
{
data_.sumIntoGlobalValues (rows, column, values);
}
/// \brief Sum data into the multivector.
///
/// In rows and values, the data for each column are stored
/// contiguously. Thus, each array is really a matrix in rowwise
/// order. (The multivector may use columnwise order internally.)
///
/// Be sure that the number of columns is set correctly before
/// calling this.
///
/// \param rows [in] Array of global rows for which to insert
/// values. Must have the same length as values.
///
/// \param values [in] Array of values to insert. Must have the
/// same length as rows. rows[i] is the row in which values[i]
/// is to be inserted.
void
sumIntoGlobalValues (Teuchos::ArrayView<const global_ordinal_type> rows,
Teuchos::ArrayView<const scalar_type> values)
{
data_.sumIntoGlobalValues (rows, values);
}
private:
//! Map with which this object was created ("ctor" == "constructor").
Teuchos::RCP<const map_type> ctorMap_;
/// \brief Source Map of the last call to globalAssemble().
///
/// A possibly overlapping Map that describes the distribution of
/// input data, and is the source of the Export.
Teuchos::RCP<const map_type> sourceMap_;
/// \brief The target Map of the last call to globalAssemble().
///
/// A nonoverlapping Map which is the target of the Export, and
/// describes the distribution of the output multivector of
/// globalAssemble().
Teuchos::RCP<const map_type> targetMap_;
/// \brief The source MV of the Export operation in globalAssemble().
///
/// The globalAssemble() method uses this multivector as the
/// source of the Export operation. The Export redistributes the
/// data from a possibly overlapping distribution (reflecting how
/// elements were inserted) to a nonoverlapping distribution (that
/// of the output multivector \c X_out of globalAssemble()). This
/// is the simplest way to implement redistribution.
///
/// We avoid resizing and reallocating \c sourceVec_ by using a
/// contiguous subview as the source of the Export, if \c X_out
/// has fewer columns than \c sourceVec_.
Teuchos::RCP<MV> sourceVec_;
/// \brief Storage for inserted indices and values.
///
/// We separate this to facilitate experimentation with different
/// storage formats.
Tpetra::Details::MultiVectorFillerData2<MV> data_;
typedef Tpetra::Export<local_ordinal_type, global_ordinal_type, node_type> export_type;
Teuchos::RCP<export_type> exporter_;
/// \brief Assemble the local data into \c X_in.
///
/// This method is called by globalAssemble(), in which \c X_in
/// is the multivector with the (possibly overlapping) source
/// distribution.
void locallyAssemble (MV& X_in) {
data_.locallyAssemble (X_in);
}
//! All source indices (local and nonlocal) of the source Map.
Teuchos::Array<global_ordinal_type>
getSourceIndices (const Teuchos::Comm<int>& comm,
const Teuchos::RCP<Teuchos::FancyOStream>& out = Teuchos::null,
const bool debug = false) const
{
return data_.getSourceIndices (comm, out, debug);
}
/// \brief Compute the source Map (for the Export) from the source indices.
///
/// indexBase, comm, and node are the same as the input arguments
/// of the Map constructors.
///
/// \note This is a collective operation.
///
/// \return The (possibly overlapping) source Map.
Teuchos::RCP<const map_type>
computeSourceMap (const global_ordinal_type indexBase,
const Teuchos::RCP<const Teuchos::Comm<int> >& comm,
const Teuchos::RCP<Teuchos::FancyOStream>& out = Teuchos::null,
const bool debug = false);
};
template<class MV>
MultiVectorFiller<MV>::MultiVectorFiller (const Teuchos::RCP<const typename MultiVectorFiller<MV>::map_type>& map, const size_t numCols)
: ctorMap_ (map),
sourceMap_ (Teuchos::null),
targetMap_ (Teuchos::null),
data_ (map, numCols),
exporter_ (Teuchos::null)
{}
template<class MV>
Teuchos::RCP<const typename MultiVectorFiller<MV>::map_type>
MultiVectorFiller<MV>::
computeSourceMap (const global_ordinal_type indexBase,
const Teuchos::RCP<const Teuchos::Comm<int> >& comm,
const Teuchos::RCP<Teuchos::FancyOStream>& out,
const bool debug)
{
using Teuchos::Array;
using Teuchos::ArrayView;
using Teuchos::rcp;
typedef global_ordinal_type GO;
if (debug && ! out.is_null ()) {
out->pushTab ();
std::ostringstream os;
const int myRank = comm->getRank ();
os << "Proc " << myRank << ": computeSourceMap" << std::endl;
*out << os.str ();
}
Array<GO> indices = getSourceIndices (*comm, out, debug);
if (debug && ! out.is_null ()) {
std::ostringstream os;
const int myRank = comm->getRank ();
os << "Proc " << myRank << ": computeSourceMap: About to create Map"
<< std::endl;
*out << os.str ();
out->popTab ();
}
// Passing "invalid" for the numGlobalElements argument of the Map
// constructor tells the Map to compute the global number of
// elements itself.
typedef Tpetra::global_size_t GST;
const GST INV = Teuchos::OrdinalTraits<GST>::invalid ();
return rcp (new map_type (INV, indices, indexBase, comm));
}
template<class MV>
void
MultiVectorFiller<MV>::
globalAssemble (MV& X_out, const bool forceReuseMap)
{
using Teuchos::ArrayView;
using Teuchos::Array;
using Teuchos::Range1D;
using Teuchos::RCP;
using Teuchos::rcp;
RCP<Teuchos::FancyOStream> outPtr;
const bool debug = false;
if (debug) {
outPtr = Teuchos::getFancyOStream (Teuchos::rcpFromRef (std::cerr));
outPtr->pushTab ();
}
const size_t numVecs = X_out.getNumVectors ();
if (numVecs == 0) {
// Nothing to do! Of course, this does not check for whether
// X_out has the right number of rows. That's OK, though.
// Compare to the definition of the BLAS' _AXPY for an input
// vector containing NaNs, but multiplied by alpha=0.
return;
}
//
// Get the target Map of the Export. If X_out's Map is the same
// as the target Map from last time, then we may be able to
// recycle the Export from last time, if the source Map is also
// the same.
//
RCP<const map_type> targetMap;
bool assumeSameTargetMap = false;
if (targetMap_.is_null()) {
targetMap_ = X_out.getMap();
targetMap = targetMap_;
assumeSameTargetMap = false;
}
else {
if (forceReuseMap) {
targetMap = targetMap_;
assumeSameTargetMap = true;
}
else {
// If X_out's Map is the same as targetMap_, we may be able to
// reuse the Export. Constructing the Export may be more
// expensive than calling isSameAs() (which requires just a
// few reductions and reading through the lists of owned
// global indices), so it's worth checking.
if (targetMap_->isSameAs (*(X_out.getMap()))) {
assumeSameTargetMap = true;
targetMap = targetMap_;
}
}
}
if (debug && ! outPtr.is_null ()) {
std::ostringstream os;
const Teuchos::Comm<int>& comm = * (X_out.getMap ()->getComm ());
const int myRank = comm.getRank ();
os << "Proc " << myRank << ": Right before getting source Map" << std::endl;
*outPtr << os.str ();
}
//
// Get the source Map of the Export. If the source Map of the
// Export is the same as last time, then we may be able to recycle
// the Export from last time, if the target Map is also the same.
//
RCP<const map_type> sourceMap;
bool computedSourceMap = false;
{
if (forceReuseMap && ! sourceMap_.is_null()) {
sourceMap = sourceMap_;
}
else {
sourceMap = computeSourceMap (ctorMap_->getIndexBase (),
ctorMap_->getComm (),
outPtr, debug);
computedSourceMap = true;
}
}
if (computedSourceMap) {
sourceMap_ = sourceMap;
}
if (debug && ! outPtr.is_null ()) {
std::ostringstream os;
const Teuchos::Comm<int>& comm = * (X_out.getMap ()->getComm ());
const int myRank = comm.getRank ();
os << "Proc " << myRank << ": Right before computing Export" << std::endl;
*outPtr << os.str ();
}
//
// Now that we have the source and target Maps of the Export, we
// can check whether we can recycle the Export from last time.
//
const bool mustComputeExport =
(exporter_.is_null() || (assumeSameTargetMap && ! computedSourceMap));
if (mustComputeExport) {
exporter_ = rcp (new export_type (sourceMap_, targetMap_));
}
if (debug && ! outPtr.is_null ()) {
std::ostringstream os;
const Teuchos::Comm<int>& comm = * (X_out.getMap ()->getComm ());
const int myRank = comm.getRank ();
os << "Proc " << myRank << ": Right after computing Export" << std::endl;
*outPtr << os.str ();
}
// Source multivector for the Export.
RCP<MV> X_in;
const bool mustReallocateVec = sourceVec_.is_null() ||
sourceVec_->getNumVectors() < numVecs || ! assumeSameTargetMap;
if (mustReallocateVec) {
// Reallocating nonlocalVec_ ensures that it has the right Map.
sourceVec_ = rcp (new MV (sourceMap_, numVecs));
X_in = sourceVec_;
} else {
if (sourceVec_->getNumVectors() == numVecs) {
X_in = sourceVec_;
} else { // sourceVec_ has more vectors than needed.
X_in = sourceVec_->subViewNonConst (Range1D (0, numVecs-1));
}
}
if (debug && ! outPtr.is_null ()) {
std::ostringstream os;
const Teuchos::Comm<int>& comm = * (X_out.getMap ()->getComm ());
const int myRank = comm.getRank ();
os << "Proc " << myRank << ": Right before locallyAssemble" << std::endl;
*outPtr << os.str ();
}
// "Locally assemble" the data into X_in by summing together
// entries with the same indices.
locallyAssemble (*X_in);
if (debug && ! outPtr.is_null ()) {
std::ostringstream os;
const Teuchos::Comm<int>& comm = * (X_out.getMap ()->getComm ());
const int myRank = comm.getRank ();
os << "Proc " << myRank << ": Right after locallyAssemble" << std::endl;
*outPtr << os.str ();
}
// Do the Export.
const Tpetra::CombineMode combineMode = Tpetra::ADD;
X_out.doExport (*X_in, *exporter_, combineMode);
if (debug && ! outPtr.is_null ()) {
std::ostringstream os;
const Teuchos::Comm<int>& comm = * (X_out.getMap ()->getComm ());
const int myRank = comm.getRank ();
os << "Proc " << myRank << ": Right after Export" << std::endl;
*outPtr << os.str ();
}
// Clean out the locally owned data for next time.
X_in->putScalar (Teuchos::ScalarTraits<scalar_type>::zero ());
// FIXME (mfh 27 Aug 2015) Clean out the remote data for next
// time. See Bug 6398.
if (debug && ! outPtr.is_null ()) {
std::ostringstream os;
const Teuchos::Comm<int>& comm = * (X_out.getMap ()->getComm ());
const int myRank = comm.getRank ();
os << "Proc " << myRank << ": Done with globalAssemble" << std::endl;
*outPtr << os.str ();
outPtr->pushTab ();
}
}
namespace Test {
/// \class MultiVectorFillerTester
/// \brief Tests for MultiVectorFiller
/// \author Mark Hoemmen
///
/// \tparam MV A specialization of Tpetra::MultiVector.
template<class MV>
class MultiVectorFillerTester {
public:
typedef typename MV::scalar_type scalar_type;
typedef typename MV::local_ordinal_type local_ordinal_type;
typedef typename MV::global_ordinal_type global_ordinal_type;
typedef typename MV::node_type node_type;
typedef ::Tpetra::Map<local_ordinal_type, global_ordinal_type, node_type> map_type;
/// \brief Test global assembly when constructor Map = target Map.
///
/// Constructor Map = target Map is a common case for finite
/// element assembly. This method current only tests the
/// version of MultiVectorFiller::sumIntoGlobalValues() that
/// works on one column at a time.
///
/// If any test fails, this method throws an exception.
static void
testSameMap (const Teuchos::RCP<const map_type>& targetMap,
const global_ordinal_type eltSize, // Must be odd
const size_t numCols,
const Teuchos::RCP<Teuchos::FancyOStream>& outStream=Teuchos::null,
const Teuchos::EVerbosityLevel verbosityLevel=Teuchos::VERB_DEFAULT)
{
using Teuchos::Array;
using Teuchos::ArrayRCP;
using Teuchos::ArrayView;
using Teuchos::as;
using Teuchos::Comm;
using Teuchos::FancyOStream;
using Teuchos::getFancyOStream;
using Teuchos::oblackholestream;
using Teuchos::ptr;
using Teuchos::RCP;
using Teuchos::rcp;
using Teuchos::rcpFromRef;
using Teuchos::REDUCE_SUM;
using Teuchos::reduceAll;
using std::cerr;
using std::endl;
typedef local_ordinal_type LO;
typedef global_ordinal_type GO;
typedef scalar_type ST;
typedef Teuchos::ScalarTraits<ST> STS;
TEUCHOS_TEST_FOR_EXCEPTION(eltSize % 2 == 0, std::invalid_argument,
"Element size (eltSize) argument must be odd.");
TEUCHOS_TEST_FOR_EXCEPTION(numCols == 0, std::invalid_argument,
"Number of columns (numCols) argument must be nonzero.");
// Default behavior is to print nothing out.
RCP<FancyOStream> out = outStream.is_null() ?
getFancyOStream (rcp (new oblackholestream)) : outStream;
const Teuchos::EVerbosityLevel verbLevel =
(verbosityLevel == Teuchos::VERB_DEFAULT) ?
Teuchos::VERB_NONE : verbosityLevel;
//RCP<MV> X = rcp (new MV (targetMap, numCols));
const GO indexBase = targetMap->getIndexBase();
Array<GO> rows (eltSize);
Array<ST> values (eltSize);
std::fill (values.begin(), values.end(), STS::one());
// Make this a pointer so we can free its contents, in case
// those contents depend on the input to globalAssemble().
RCP<MultiVectorFiller<MV> > filler =
rcp (new MultiVectorFiller<MV> (targetMap, numCols));
TEUCHOS_TEST_FOR_EXCEPTION(! targetMap->isContiguous(),
std::invalid_argument, "MultiVectorFiller test currently only works "
"for contiguous Maps.");
const GO minGlobalIndex = targetMap->getMinGlobalIndex();
const GO maxGlobalIndex = targetMap->getMaxGlobalIndex();
const GO minAllGlobalIndex = targetMap->getMinAllGlobalIndex();
const GO maxAllGlobalIndex = targetMap->getMaxAllGlobalIndex();
for (size_t j = 0; j < numCols; ++j) {
for (GO i = minGlobalIndex; i <= maxGlobalIndex; ++i) {
// Overlap over processes, without running out of bounds.
const GO start = std::max (i - eltSize/2, minAllGlobalIndex);
const GO end = std::min (i + eltSize/2, maxAllGlobalIndex);
const GO len = end - start + 1;
TEUCHOS_TEST_FOR_EXCEPTION(len > eltSize, std::logic_error,
"At start,end = " << start << "," << end << ", len = " << len
<< " > eltSize = " << eltSize << ".");
for (GO k = 0; k < len; ++k) {
rows[k] = start + k;
}
if (verbLevel == Teuchos::VERB_EXTREME) {
*out << "Inserting: "
<< Teuchos::toString (rows.view(0,len)) << ", "
<< Teuchos::toString (values.view(0, len)) << std::endl;
}
filler->sumIntoGlobalValues (rows.view(0, len), j, values.view(0, len));
}
}
if (verbLevel == Teuchos::VERB_EXTREME) {
*out << "Filler:" << std::endl;
filler->describe (*out, verbLevel);
*out << std::endl;
}
MV X_out (targetMap, numCols);
filler->globalAssemble (X_out);
filler = Teuchos::null;
if (verbLevel == Teuchos::VERB_EXTREME) {
*out << "X_out:" << std::endl;
X_out.describe (*out, verbLevel);
*out << std::endl;
}
// Create multivector for comparison.
MV X_expected (targetMap, numCols);
const scalar_type two = STS::one() + STS::one();
for (size_t j = 0; j < numCols; ++j) {
ArrayRCP<ST> X_j = X_expected.getDataNonConst (j);
// Each entry of the column should have the value eltSize,
// except for the first and last few entries in the whole
// column (globally, not locally).
for (GO i = minGlobalIndex; i <= maxGlobalIndex; ++i) {
const LO localIndex = targetMap->getLocalElement (i);
TEUCHOS_TEST_FOR_EXCEPTION(i == Teuchos::OrdinalTraits<LO>::invalid(),
std::logic_error, "Global index " << i << " is not in the "
"multivector's Map.");
if (i <= minAllGlobalIndex + eltSize/2) {
X_j[localIndex] = two + as<ST>(i) - as<ST>(indexBase);
}
else if (i >= maxAllGlobalIndex - eltSize/2) {
X_j[localIndex] = two + as<ST>(maxAllGlobalIndex) - as<ST>(i);
}
else {
X_j[localIndex] = as<ST>(eltSize);
}
} // for each global index which my process owns
} // for each column of the multivector
if (verbLevel == Teuchos::VERB_EXTREME) {
*out << "X_expected:" << std::endl;
X_expected.describe (*out, verbLevel);
*out << std::endl;
}
Array<GO> errorLocations;
for (size_t j = 0; j < numCols; ++j) {
ArrayRCP<const ST> X_out_j = X_out.getData (j);
ArrayRCP<const ST> X_expected_j = X_expected.getData (j);
// Each entry of the column should have the value eltSize,
// except for the first and last few entries in the whole
// column (globally, not locally).
for (GO i = minGlobalIndex; i <= maxGlobalIndex; ++i) {
const LO localIndex = targetMap->getLocalElement (i);
TEUCHOS_TEST_FOR_EXCEPTION(i == Teuchos::OrdinalTraits<LO>::invalid(),
std::logic_error, "Global index " << i << " is not in the "
"multivector's Map.");
// The floating-point additions should be exact in this
// case, except for very large values of eltSize.
if (X_out_j[localIndex] != X_expected_j[localIndex]) {
errorLocations.push_back (i);
}
} // for each global index which my process owns
const typename Array<GO>::size_type localNumErrors = errorLocations.size();
typename Array<GO>::size_type globalNumErrors = 0;
RCP<const Comm<int> > comm = targetMap->getComm();
reduceAll (*comm, REDUCE_SUM, localNumErrors, ptr (&globalNumErrors));
if (globalNumErrors != 0) {
std::ostringstream os;
os << "Proc " << comm->getRank() << ": " << localNumErrors
<< " incorrect value" << (localNumErrors != 1 ? "s" : "")
<< ". Error locations: [ ";
std::copy (errorLocations.begin(), errorLocations.end(),
std::ostream_iterator<GO> (os, " "));
os << "].";
// Iterate through all processes in the communicator,
// printing out each process' local errors.
for (int p = 0; p < comm->getSize(); ++p) {
if (p == comm->getRank()) {
cerr << os.str() << endl;
}
// Barriers to let output finish.
comm->barrier();
comm->barrier();
comm->barrier();
}
TEUCHOS_TEST_FOR_EXCEPTION(globalNumErrors != 0, std::logic_error,
"Over all procs: " << globalNumErrors << " total error"
<< (globalNumErrors != 1 ? "s" : "") << ".");
} // if there were any errors in column j
} // for each column j
}
};
//! Instantiate a MultiVectorFillerTester and run the test.
template<class ScalarType,
class LocalOrdinalType,
class GlobalOrdinalType,
class NodeType>
void
testMultiVectorFiller (const Teuchos::RCP<const Teuchos::Comm<int> >& comm,
const size_t unknownsPerNode,
const GlobalOrdinalType unknownsPerElt,
const size_t numCols,
const Teuchos::RCP<Teuchos::FancyOStream>& outStream,
const Teuchos::EVerbosityLevel verbLevel)
{
using Tpetra::createContigMapWithNode;
using Teuchos::FancyOStream;
using Teuchos::getFancyOStream;
using Teuchos::oblackholestream;
using Teuchos::ParameterList;
using Teuchos::RCP;
using Teuchos::rcp;
using std::cerr;
using std::endl;
typedef ScalarType ST;
typedef LocalOrdinalType LO;
typedef GlobalOrdinalType GO;
typedef NodeType NT;
typedef ::Tpetra::Map<LO, GO, NT> map_type;
typedef Tpetra::MultiVector<ST, LO, GO, NT> MV;
typedef Tpetra::global_size_t GST;
RCP<FancyOStream> out = outStream.is_null() ?
getFancyOStream (rcp (new oblackholestream)) : outStream;
const GST INV = Teuchos::OrdinalTraits<GST>::invalid ();
const GO indexBase = 0;
RCP<const map_type> targetMap =
rcp (new map_type (INV, unknownsPerNode, indexBase, comm));
std::ostringstream os;
try {
MultiVectorFillerTester<MV>::testSameMap (targetMap, unknownsPerElt, numCols, out, verbLevel);
} catch (std::exception& e) {
os << e.what();
}
for (int p = 0; p < comm->getSize(); ++p) {
if (p == comm->getRank()) {
*out << "On process " << comm->getRank() << ": " << os.str() << endl;
}
comm->barrier();
comm->barrier();
comm->barrier();
}
}
/// \brief Test the sortAndMergeIn() utility function.
///
/// If any test fails, this function throws std::logic_error with
/// an informative message. If all tests pass, this function
/// returns with no externally visible side effects.
void
testSortAndMergeIn ();
} // namespace Test
} // namespace Tpetra
#endif // __Tpetra_MultiVectorFiller_hpp
|