| 1 |
// ------------------------------------------------------------------------ |
| 2 |
// |
| 3 |
// SPDX-License-Identifier: LGPL-2.1-or-later |
| 4 |
// Copyright (C) 1999 - 2024 by the deal.II authors |
| 5 |
// |
| 6 |
// This file is part of the deal.II library. |
| 7 |
// |
| 8 |
// Part of the source code is dual licensed under Apache-2.0 WITH |
| 9 |
// LLVM-exception OR LGPL-2.1-or-later. Detailed license information |
| 10 |
// governing the source code and code contributions can be found in |
| 11 |
// LICENSE.md and CONTRIBUTING.md at the top level directory of deal.II. |
| 12 |
// |
| 13 |
// ------------------------------------------------------------------------ |
| 14 |
|
| 15 |
#ifndef dealii_vector_templates_h |
| 16 |
#define dealii_vector_templates_h |
| 17 |
|
| 18 |
|
| 19 |
#include <deal.II/base/config.h> |
| 20 |
|
| 21 |
#include <deal.II/base/numbers.h> |
| 22 |
#include <deal.II/base/template_constraints.h> |
| 23 |
|
| 24 |
#include <deal.II/lac/block_vector.h> |
| 25 |
#include <deal.II/lac/exceptions.h> |
| 26 |
#include <deal.II/lac/vector.h> |
| 27 |
#include <deal.II/lac/vector_operations_internal.h> |
| 28 |
|
| 29 |
#ifdef DEAL_II_WITH_PETSC |
| 30 |
# include <deal.II/lac/petsc_vector_base.h> |
| 31 |
#endif |
| 32 |
|
| 33 |
#ifdef DEAL_II_WITH_TRILINOS |
| 34 |
# include <deal.II/lac/trilinos_vector.h> |
| 35 |
#endif |
| 36 |
|
| 37 |
#ifdef DEAL_II_TRILINOS_WITH_TPETRA |
| 38 |
# include <deal.II/lac/trilinos_tpetra_vector.h> |
| 39 |
#endif |
| 40 |
|
| 41 |
|
| 42 |
#include <algorithm> |
| 43 |
#include <cmath> |
| 44 |
#include <cstdio> |
| 45 |
#include <cstring> |
| 46 |
#include <iomanip> |
| 47 |
#include <iostream> |
| 48 |
#include <limits> |
| 49 |
#include <memory> |
| 50 |
|
| 51 |
DEAL_II_NAMESPACE_OPEN |
| 52 |
|
| 53 |
template <typename Number> |
| 54 |
Vector<Number>::Vector(const Vector<Number> &v) |
| 55 |
{ |
| 56 |
*this = v; |
| 57 |
} |
| 58 |
|
| 59 |
|
| 60 |
|
| 61 |
template <typename Number> |
| 62 |
void |
| 63 |
Vector<Number>::apply_givens_rotation(const std::array<Number, 3> &csr, |
| 64 |
const size_type i, |
| 65 |
const size_type k) |
| 66 |
{ |
| 67 |
auto &V = *this; |
| 68 |
const Number t = V(i); |
| 69 |
V(i) = csr[0] * V(i) + csr[1] * V(k); |
| 70 |
V(k) = -csr[1] * t + csr[0] * V(k); |
| 71 |
} |
| 72 |
|
| 73 |
|
| 74 |
|
| 75 |
template <typename Number> |
| 76 |
template <typename OtherNumber> |
| 77 |
Vector<Number>::Vector(const Vector<OtherNumber> &v) |
| 78 |
{ |
| 79 |
*this = v; |
| 80 |
} |
| 81 |
|
| 82 |
|
| 83 |
|
| 84 |
#ifdef DEAL_II_WITH_PETSC |
| 85 |
namespace internal |
| 86 |
{ |
| 87 |
template <typename Number> |
| 88 |
void |
| 89 |
copy_petsc_vector(const PETScWrappers::VectorBase &v, |
| 90 |
::dealii::Vector<Number> &out) |
| 91 |
{ |
| 92 |
if (v.size() == 0) |
| 93 |
{ |
| 94 |
out.reinit(0); |
| 95 |
return; |
| 96 |
} |
| 97 |
// Create a sequential PETSc vector and then copy over the entries into |
| 98 |
// the deal.II vector. |
| 99 |
Vec sequential_vector; |
| 100 |
VecScatter scatter_context; |
| 101 |
|
| 102 |
PetscErrorCode ierr = |
| 103 |
VecScatterCreateToAll(v, &scatter_context, &sequential_vector); |
| 104 |
AssertThrow(ierr == 0, ExcPETScError(ierr)); |
| 105 |
|
| 106 |
ierr = VecScatterBegin( |
| 107 |
scatter_context, v, sequential_vector, INSERT_VALUES, SCATTER_FORWARD); |
| 108 |
AssertThrow(ierr == 0, ExcPETScError(ierr)); |
| 109 |
ierr = VecScatterEnd( |
| 110 |
scatter_context, v, sequential_vector, INSERT_VALUES, SCATTER_FORWARD); |
| 111 |
AssertThrow(ierr == 0, ExcPETScError(ierr)); |
| 112 |
|
| 113 |
const PetscScalar *start_ptr; |
| 114 |
ierr = VecGetArrayRead(sequential_vector, &start_ptr); |
| 115 |
AssertThrow(ierr == 0, ExcPETScError(ierr)); |
| 116 |
|
| 117 |
const PETScWrappers::VectorBase::size_type v_size = v.size(); |
| 118 |
if (out.size() != v_size) |
| 119 |
out.reinit(v_size, true); |
| 120 |
|
| 121 |
internal::VectorOperations::copy(start_ptr, |
| 122 |
start_ptr + out.size(), |
| 123 |
out.begin()); |
| 124 |
ierr = VecRestoreArrayRead(sequential_vector, &start_ptr); |
| 125 |
AssertThrow(ierr == 0, ExcPETScError(ierr)); |
| 126 |
|
| 127 |
ierr = VecScatterDestroy(&scatter_context); |
| 128 |
AssertNothrow(ierr == 0, ExcPETScError(ierr)); |
| 129 |
ierr = VecDestroy(&sequential_vector); |
| 130 |
AssertNothrow(ierr == 0, ExcPETScError(ierr)); |
| 131 |
} |
| 132 |
} // namespace internal |
| 133 |
|
| 134 |
|
| 135 |
|
| 136 |
template <typename Number> |
| 137 |
Vector<Number>::Vector(const PETScWrappers::VectorBase &v) |
| 138 |
{ |
| 139 |
internal::copy_petsc_vector(v, *this); |
| 140 |
} |
| 141 |
#endif |
| 142 |
|
| 143 |
|
| 144 |
#ifdef DEAL_II_WITH_TRILINOS |
| 145 |
|
| 146 |
template <typename Number> |
| 147 |
Vector<Number>::Vector(const TrilinosWrappers::MPI::Vector &v) |
| 148 |
: values(v.size()) |
| 149 |
{ |
| 150 |
if (size() != 0) |
| 151 |
{ |
| 152 |
// Copy the distributed vector to |
| 153 |
// a local one at all processors |
| 154 |
// that know about the original vector. |
| 155 |
// TODO: There could |
| 156 |
// be a better solution than |
| 157 |
// this, but it has not yet been |
| 158 |
// found. |
| 159 |
TrilinosWrappers::MPI::Vector localized_vector; |
| 160 |
localized_vector.reinit(complete_index_set(size()), |
| 161 |
v.get_mpi_communicator()); |
| 162 |
localized_vector.reinit(v, false, true); |
| 163 |
|
| 164 |
Assert(localized_vector.size() == size(), |
| 165 |
ExcDimensionMismatch(localized_vector.size(), size())); |
| 166 |
|
| 167 |
// get a representation of the vector |
| 168 |
// and copy it |
| 169 |
TrilinosScalar **start_ptr; |
| 170 |
|
| 171 |
int ierr = localized_vector.trilinos_vector().ExtractView(&start_ptr); |
| 172 |
AssertThrow(ierr == 0, ExcTrilinosError(ierr)); |
| 173 |
|
| 174 |
std::copy(start_ptr[0], start_ptr[0] + size(), begin()); |
| 175 |
|
| 176 |
maybe_reset_thread_partitioner(); |
| 177 |
} |
| 178 |
} |
| 179 |
|
| 180 |
#endif |
| 181 |
|
| 182 |
|
| 183 |
#ifdef DEAL_II_TRILINOS_WITH_TPETRA |
| 184 |
|
| 185 |
template <typename Number> |
| 186 |
template <typename OtherNumber, typename MemorySpace> |
| 187 |
Vector<Number>::Vector( |
| 188 |
const LinearAlgebra::TpetraWrappers::Vector<OtherNumber, MemorySpace> &v) |
| 189 |
: values(v.size()) |
| 190 |
{ |
| 191 |
static_assert( |
| 192 |
std::is_same<Number, OtherNumber>::value, |
| 193 |
"TpetraWrappers::Vector and dealii::Vector must use the same number type here."); |
| 194 |
|
| 195 |
if (size() != 0) |
| 196 |
{ |
| 197 |
// Copy the distributed vector to |
| 198 |
// a local one at all processors |
| 199 |
// that know about the original vector. |
| 200 |
LinearAlgebra::TpetraWrappers::TpetraTypes::VectorType<OtherNumber, |
| 201 |
MemorySpace> |
| 202 |
localized_vector( |
| 203 |
complete_index_set(size()) |
| 204 |
.template make_tpetra_map_rcp< |
| 205 |
LinearAlgebra::TpetraWrappers::TpetraTypes::NodeType< |
| 206 |
MemorySpace>>(), |
| 207 |
v.get_mpi_communicator()); |
| 208 |
|
| 209 |
Teuchos::RCP<const LinearAlgebra::TpetraWrappers::TpetraTypes::ImportType< |
| 210 |
MemorySpace>> |
| 211 |
importer = Tpetra::createImport(v.trilinos_vector().getMap(), |
| 212 |
localized_vector.getMap()); |
| 213 |
|
| 214 |
localized_vector.doImport(v.trilinos_vector(), *importer, Tpetra::INSERT); |
| 215 |
|
| 216 |
// get a kokkos view from the localized_vector |
| 217 |
# if DEAL_II_TRILINOS_VERSION_GTE(13, 2, 0) |
| 218 |
auto localized_vector_2d = |
| 219 |
localized_vector.template getLocalView<Kokkos::HostSpace>( |
| 220 |
Tpetra::Access::ReadOnly); |
| 221 |
# else |
| 222 |
localized_vector.template sync<Kokkos::HostSpace>(); |
| 223 |
auto localized_vector_2d = |
| 224 |
localized_vector.template getLocalView<Kokkos::HostSpace>(); |
| 225 |
# endif |
| 226 |
auto localized_vector_1d = |
| 227 |
Kokkos::subview(localized_vector_2d, Kokkos::ALL(), 0); |
| 228 |
const size_t local_length = localized_vector.getLocalLength(); |
| 229 |
|
| 230 |
Kokkos::DefaultHostExecutionSpace exec; |
| 231 |
Kokkos::View<Number *, Kokkos::HostSpace> values_view(values.data(), |
| 232 |
local_length); |
| 233 |
Kokkos::deep_copy(exec, values_view, localized_vector_1d); |
| 234 |
exec.fence(); |
| 235 |
} |
| 236 |
} |
| 237 |
|
| 238 |
#endif |
| 239 |
|
| 240 |
|
| 241 |
template <typename Number> |
| 242 |
inline Vector<Number> & |
| 243 |
Vector<Number>::operator=(const Vector<Number> &v) |
| 244 |
{ |
| 245 |
if (PointerComparison::equal(this, &v)) |
| 246 |
return *this; |
| 247 |
|
| 248 |
if (size() != v.size()) |
| 249 |
reinit(v, true); |
| 250 |
|
| 251 |
if (0 < size()) |
| 252 |
{ |
| 253 |
dealii::internal::VectorOperations::Vector_copy<Number, Number> copier( |
| 254 |
v.begin(), begin()); |
| 255 |
internal::VectorOperations::parallel_for(copier, |
| 256 |
0, |
| 257 |
size(), |
| 258 |
thread_loop_partitioner); |
| 259 |
} |
| 260 |
|
| 261 |
return *this; |
| 262 |
} |
| 263 |
|
| 264 |
|
| 265 |
|
| 266 |
template <typename Number> |
| 267 |
template <typename Number2> |
| 268 |
inline Vector<Number> & |
| 269 |
Vector<Number>::operator=(const Vector<Number2> &v) |
| 270 |
{ |
| 271 |
if (size() != v.size()) |
| 272 |
reinit(v, true); |
| 273 |
|
| 274 |
dealii::internal::VectorOperations::Vector_copy<Number, Number2> copier( |
| 275 |
v.begin(), begin()); |
| 276 |
internal::VectorOperations::parallel_for(copier, |
| 277 |
0, |
| 278 |
size(), |
| 279 |
thread_loop_partitioner); |
| 280 |
|
| 281 |
return *this; |
| 282 |
} |
| 283 |
|
| 284 |
|
| 285 |
|
| 286 |
template <typename Number> |
| 287 |
inline void |
| 288 |
Vector<Number>::reinit(const size_type n, const bool omit_zeroing_entries) |
| 289 |
{ |
| 290 |
do_reinit(n, omit_zeroing_entries, true); |
| 291 |
} |
| 292 |
|
| 293 |
|
| 294 |
|
| 295 |
template <typename Number> |
| 296 |
inline void |
| 297 |
Vector<Number>::grow_or_shrink(const size_type n) |
| 298 |
{ |
| 299 |
values.resize(n); |
| 300 |
|
| 301 |
maybe_reset_thread_partitioner(); |
| 302 |
} |
| 303 |
|
| 304 |
|
| 305 |
|
| 306 |
template <typename Number> |
| 307 |
bool |
| 308 |
Vector<Number>::all_zero() const |
| 309 |
{ |
| 310 |
Assert(size() != 0, ExcEmptyObject()); |
| 311 |
|
| 312 |
for (size_type i = 0; i < size(); ++i) |
| 313 |
if (values[i] != Number()) |
| 314 |
return false; |
| 315 |
return true; |
| 316 |
} |
| 317 |
|
| 318 |
|
| 319 |
|
| 320 |
template <typename Number> |
| 321 |
bool |
| 322 |
Vector<Number>::is_non_negative() const |
| 323 |
{ |
| 324 |
Assert(size() != 0, ExcEmptyObject()); |
| 325 |
|
| 326 |
for (size_type i = 0; i < size(); ++i) |
| 327 |
if (!internal::VectorOperations::is_non_negative(values[i])) |
| 328 |
return false; |
| 329 |
|
| 330 |
return true; |
| 331 |
} |
| 332 |
|
| 333 |
|
| 334 |
|
| 335 |
template <typename Number> |
| 336 |
Vector<Number> & |
| 337 |
Vector<Number>::operator=(const Number s) |
| 338 |
{ |
| 339 |
AssertIsFinite(s); |
| 340 |
if (s != Number()) |
| 341 |
Assert(size() != 0, ExcEmptyObject()); |
| 342 |
|
| 343 |
if (size() > 0) |
| 344 |
{ |
| 345 |
internal::VectorOperations::Vector_set<Number> setter(s, values.begin()); |
| 346 |
internal::VectorOperations::parallel_for(setter, |
| 347 |
0, |
| 348 |
size(), |
| 349 |
thread_loop_partitioner); |
| 350 |
} |
| 351 |
|
| 352 |
return *this; |
| 353 |
} |
| 354 |
|
| 355 |
|
| 356 |
|
| 357 |
template <typename Number> |
| 358 |
Vector<Number> & |
| 359 |
Vector<Number>::operator*=(const Number factor) |
| 360 |
{ |
| 361 |
AssertIsFinite(factor); |
| 362 |
|
| 363 |
Assert(size() != 0, ExcEmptyObject()); |
| 364 |
|
| 365 |
internal::VectorOperations::Vectorization_multiply_factor<Number> |
| 366 |
vector_multiply(values.begin(), factor); |
| 367 |
|
| 368 |
internal::VectorOperations::parallel_for(vector_multiply, |
| 369 |
0, |
| 370 |
size(), |
| 371 |
thread_loop_partitioner); |
| 372 |
|
| 373 |
return *this; |
| 374 |
} |
| 375 |
|
| 376 |
|
| 377 |
|
| 378 |
template <typename Number> |
| 379 |
void |
| 380 |
Vector<Number>::add(const Number a, const Vector<Number> &v) |
| 381 |
{ |
| 382 |
AssertIsFinite(a); |
| 383 |
|
| 384 |
Assert(size() != 0, ExcEmptyObject()); |
| 385 |
Assert(size() == v.size(), ExcDimensionMismatch(size(), v.size())); |
| 386 |
|
| 387 |
internal::VectorOperations::Vectorization_add_av<Number> vector_add_av( |
| 388 |
values.begin(), v.values.begin(), a); |
| 389 |
internal::VectorOperations::parallel_for(vector_add_av, |
| 390 |
0, |
| 391 |
size(), |
| 392 |
thread_loop_partitioner); |
| 393 |
} |
| 394 |
|
| 395 |
|
| 396 |
|
| 397 |
template <typename Number> |
| 398 |
void |
| 399 |
Vector<Number>::sadd(const Number x, const Number a, const Vector<Number> &v) |
| 400 |
{ |
| 401 |
AssertIsFinite(x); |
| 402 |
AssertIsFinite(a); |
| 403 |
|
| 404 |
Assert(size() != 0, ExcEmptyObject()); |
| 405 |
Assert(size() == v.size(), ExcDimensionMismatch(size(), v.size())); |
| 406 |
|
| 407 |
internal::VectorOperations::Vectorization_sadd_xav<Number> vector_sadd_xav( |
| 408 |
values.begin(), v.values.begin(), a, x); |
| 409 |
internal::VectorOperations::parallel_for(vector_sadd_xav, |
| 410 |
0, |
| 411 |
size(), |
| 412 |
thread_loop_partitioner); |
| 413 |
} |
| 414 |
|
| 415 |
|
| 416 |
|
| 417 |
template <typename Number> |
| 418 |
template <typename Number2> |
| 419 |
Number |
| 420 |
Vector<Number>::operator*(const Vector<Number2> &v) const |
| 421 |
{ |
| 422 |
Assert(size() != 0, ExcEmptyObject()); |
| 423 |
|
| 424 |
if (PointerComparison::equal(this, &v)) |
| 425 |
return norm_sqr(); |
| 426 |
|
| 427 |
Assert(size() == v.size(), ExcDimensionMismatch(size(), v.size())); |
| 428 |
|
| 429 |
Number sum; |
| 430 |
internal::VectorOperations::Dot<Number, Number2> dot(values.begin(), |
| 431 |
v.values.begin()); |
| 432 |
internal::VectorOperations::parallel_reduce( |
| 433 |
dot, 0, size(), sum, thread_loop_partitioner); |
| 434 |
AssertIsFinite(sum); |
| 435 |
|
| 436 |
return sum; |
| 437 |
} |
| 438 |
|
| 439 |
|
| 440 |
|
| 441 |
template <typename Number> |
| 442 |
typename Vector<Number>::real_type |
| 443 |
Vector<Number>::norm_sqr() const |
| 444 |
{ |
| 445 |
Assert(size() != 0, ExcEmptyObject()); |
| 446 |
|
| 447 |
real_type sum; |
| 448 |
internal::VectorOperations::Norm2<Number, real_type> norm2(values.begin()); |
| 449 |
internal::VectorOperations::parallel_reduce( |
| 450 |
norm2, 0, size(), sum, thread_loop_partitioner); |
| 451 |
|
| 452 |
AssertIsFinite(sum); |
| 453 |
|
| 454 |
return sum; |
| 455 |
} |
| 456 |
|
| 457 |
|
| 458 |
|
| 459 |
template <typename Number> |
| 460 |
Number |
| 461 |
Vector<Number>::mean_value() const |
| 462 |
{ |
| 463 |
Assert(size() != 0, ExcEmptyObject()); |
| 464 |
|
| 465 |
Number sum; |
| 466 |
internal::VectorOperations::MeanValue<Number> mean(values.begin()); |
| 467 |
internal::VectorOperations::parallel_reduce( |
| 468 |
mean, 0, size(), sum, thread_loop_partitioner); |
| 469 |
|
| 470 |
return sum / real_type(size()); |
| 471 |
} |
| 472 |
|
| 473 |
|
| 474 |
|
| 475 |
template <typename Number> |
| 476 |
typename Vector<Number>::real_type |
| 477 |
Vector<Number>::l1_norm() const |
| 478 |
{ |
| 479 |
Assert(size() != 0, ExcEmptyObject()); |
| 480 |
|
| 481 |
real_type sum; |
| 482 |
internal::VectorOperations::Norm1<Number, real_type> norm1(values.begin()); |
| 483 |
internal::VectorOperations::parallel_reduce( |
| 484 |
norm1, 0, size(), sum, thread_loop_partitioner); |
| 485 |
|
| 486 |
return sum; |
| 487 |
} |
| 488 |
|
| 489 |
|
| 490 |
|
| 491 |
template <typename Number> |
| 492 |
typename Vector<Number>::real_type |
| 493 |
Vector<Number>::l2_norm() const |
| 494 |
{ |
| 495 |
// if l2_norm()^2 is finite and non-zero, the answer is computed as |
| 496 |
// std::sqrt(norm_sqr()). If norm_sqr() is infinite or zero, the l2 norm |
| 497 |
// might still be finite. In that case, recompute it (this is a rare case, |
| 498 |
// so working on the vector twice is uncritical and paid off by the extended |
| 499 |
// precision) using the BLAS approach with a weight, see e.g. dnrm2.f. |
| 500 |
Assert(size() != 0, ExcEmptyObject()); |
| 501 |
|
| 502 |
real_type norm_square; |
| 503 |
internal::VectorOperations::Norm2<Number, real_type> norm2(values.begin()); |
| 504 |
internal::VectorOperations::parallel_reduce( |
| 505 |
norm2, 0, size(), norm_square, thread_loop_partitioner); |
| 506 |
if (numbers::is_finite(norm_square) && |
| 507 |
norm_square >= std::numeric_limits<real_type>::min()) |
| 508 |
return static_cast<typename Vector<Number>::real_type>( |
| 509 |
std::sqrt(norm_square)); |
| 510 |
else |
| 511 |
{ |
| 512 |
real_type scale = 0.; |
| 513 |
real_type sum = 1.; |
| 514 |
for (size_type i = 0; i < size(); ++i) |
| 515 |
{ |
| 516 |
if (values[i] != Number()) |
| 517 |
{ |
| 518 |
const real_type abs_x = |
| 519 |
numbers::NumberTraits<Number>::abs(values[i]); |
| 520 |
if (scale < abs_x) |
| 521 |
{ |
| 522 |
sum = 1 + sum * (scale / abs_x) * (scale / abs_x); |
| 523 |
scale = abs_x; |
| 524 |
} |
| 525 |
else |
| 526 |
sum += (abs_x / scale) * (abs_x / scale); |
| 527 |
} |
| 528 |
} |
| 529 |
AssertIsFinite(scale * std::sqrt(sum)); |
| 530 |
return static_cast<typename Vector<Number>::real_type>(scale * |
| 531 |
std::sqrt(sum)); |
| 532 |
} |
| 533 |
} |
| 534 |
|
| 535 |
|
| 536 |
|
| 537 |
template <typename Number> |
| 538 |
typename Vector<Number>::real_type |
| 539 |
Vector<Number>::lp_norm(const real_type p) const |
| 540 |
{ |
| 541 |
Assert(size() != 0, ExcEmptyObject()); |
| 542 |
|
| 543 |
if (p == 1.) |
| 544 |
return l1_norm(); |
| 545 |
else if (p == 2.) |
| 546 |
return l2_norm(); |
| 547 |
|
| 548 |
real_type sum; |
| 549 |
internal::VectorOperations::NormP<Number, real_type> normp(values.begin(), p); |
| 550 |
internal::VectorOperations::parallel_reduce( |
| 551 |
normp, 0, size(), sum, thread_loop_partitioner); |
| 552 |
|
| 553 |
if (numbers::is_finite(sum) && sum >= std::numeric_limits<real_type>::min()) |
| 554 |
return std::pow(sum, static_cast<real_type>(1. / p)); |
| 555 |
else |
| 556 |
{ |
| 557 |
real_type scale = 0.; |
| 558 |
real_type sum = 1.; |
| 559 |
for (size_type i = 0; i < size(); ++i) |
| 560 |
{ |
| 561 |
if (values[i] != Number()) |
| 562 |
{ |
| 563 |
const real_type abs_x = |
| 564 |
numbers::NumberTraits<Number>::abs(values[i]); |
| 565 |
if (scale < abs_x) |
| 566 |
{ |
| 567 |
sum = 1. + sum * std::pow(scale / abs_x, p); |
| 568 |
scale = abs_x; |
| 569 |
} |
| 570 |
else |
| 571 |
sum += std::pow(abs_x / scale, p); |
| 572 |
} |
| 573 |
} |
| 574 |
return scale * std::pow(sum, static_cast<real_type>(1. / p)); |
| 575 |
} |
| 576 |
} |
| 577 |
|
| 578 |
|
| 579 |
|
| 580 |
template <typename Number> |
| 581 |
typename Vector<Number>::real_type |
| 582 |
Vector<Number>::linfty_norm() const |
| 583 |
{ |
| 584 |
Assert(size() != 0, ExcEmptyObject()); |
| 585 |
|
| 586 |
real_type max = 0.; |
| 587 |
|
| 588 |
for (size_type i = 0; i < size(); ++i) |
| 589 |
max = std::max(numbers::NumberTraits<Number>::abs(values[i]), max); |
| 590 |
|
| 591 |
return max; |
| 592 |
} |
| 593 |
|
| 594 |
|
| 595 |
|
| 596 |
template <typename Number> |
| 597 |
Number |
| 598 |
Vector<Number>::add_and_dot(const Number a, |
| 599 |
const Vector<Number> &V, |
| 600 |
const Vector<Number> &W) |
| 601 |
{ |
| 602 |
Assert(size() != 0, ExcEmptyObject()); |
| 603 |
AssertDimension(size(), V.size()); |
| 604 |
AssertDimension(size(), W.size()); |
| 605 |
|
| 606 |
Number sum; |
| 607 |
internal::VectorOperations::AddAndDot<Number> adder(values.begin(), |
| 608 |
V.values.begin(), |
| 609 |
W.values.begin(), |
| 610 |
a); |
| 611 |
internal::VectorOperations::parallel_reduce( |
| 612 |
adder, 0, size(), sum, thread_loop_partitioner); |
| 613 |
AssertIsFinite(sum); |
| 614 |
|
| 615 |
return sum; |
| 616 |
} |
| 617 |
|
| 618 |
|
| 619 |
|
| 620 |
template <typename Number> |
| 621 |
void |
| 622 |
Vector<Number>::extract_subvector_to( |
| 623 |
const ArrayView<const types::global_dof_index> &indices, |
| 624 |
const ArrayView<Number> &elements) const |
| 625 |
{ |
| 626 |
AssertDimension(indices.size(), elements.size()); |
| 627 |
for (unsigned int i = 0; i < indices.size(); ++i) |
| 628 |
{ |
| 629 |
AssertIndexRange(indices[i], size()); |
| 630 |
elements[i] = (*this)[indices[i]]; |
| 631 |
} |
| 632 |
} |
| 633 |
|
| 634 |
|
| 635 |
|
| 636 |
template <typename Number> |
| 637 |
Vector<Number> & |
| 638 |
Vector<Number>::operator+=(const Vector<Number> &v) |
| 639 |
{ |
| 640 |
Assert(size() != 0, ExcEmptyObject()); |
| 641 |
Assert(size() == v.size(), ExcDimensionMismatch(size(), v.size())); |
| 642 |
|
| 643 |
internal::VectorOperations::Vectorization_add_v<Number> vector_add( |
| 644 |
values.begin(), v.values.begin()); |
| 645 |
internal::VectorOperations::parallel_for(vector_add, |
| 646 |
0, |
| 647 |
size(), |
| 648 |
thread_loop_partitioner); |
| 649 |
return *this; |
| 650 |
} |
| 651 |
|
| 652 |
|
| 653 |
|
| 654 |
template <typename Number> |
| 655 |
Vector<Number> & |
| 656 |
Vector<Number>::operator-=(const Vector<Number> &v) |
| 657 |
{ |
| 658 |
Assert(size() != 0, ExcEmptyObject()); |
| 659 |
Assert(size() == v.size(), ExcDimensionMismatch(size(), v.size())); |
| 660 |
|
| 661 |
internal::VectorOperations::Vectorization_subtract_v<Number> vector_subtract( |
| 662 |
values.begin(), v.values.begin()); |
| 663 |
internal::VectorOperations::parallel_for(vector_subtract, |
| 664 |
0, |
| 665 |
size(), |
| 666 |
thread_loop_partitioner); |
| 667 |
|
| 668 |
return *this; |
| 669 |
} |
| 670 |
|
| 671 |
|
| 672 |
|
| 673 |
template <typename Number> |
| 674 |
void |
| 675 |
Vector<Number>::add(const Number v) |
| 676 |
{ |
| 677 |
Assert(size() != 0, ExcEmptyObject()); |
| 678 |
|
| 679 |
internal::VectorOperations::Vectorization_add_factor<Number> vector_add( |
| 680 |
values.begin(), v); |
| 681 |
internal::VectorOperations::parallel_for(vector_add, |
| 682 |
0, |
| 683 |
size(), |
| 684 |
thread_loop_partitioner); |
| 685 |
} |
| 686 |
|
| 687 |
|
| 688 |
|
| 689 |
template <typename Number> |
| 690 |
void |
| 691 |
Vector<Number>::add(const Number a, |
| 692 |
const Vector<Number> &v, |
| 693 |
const Number b, |
| 694 |
const Vector<Number> &w) |
| 695 |
{ |
| 696 |
AssertIsFinite(a); |
| 697 |
AssertIsFinite(b); |
| 698 |
|
| 699 |
Assert(size() != 0, ExcEmptyObject()); |
| 700 |
Assert(size() == v.size(), ExcDimensionMismatch(size(), v.size())); |
| 701 |
Assert(size() == w.size(), ExcDimensionMismatch(size(), w.size())); |
| 702 |
|
| 703 |
internal::VectorOperations::Vectorization_add_avpbw<Number> vector_add( |
| 704 |
values.begin(), v.values.begin(), w.values.begin(), a, b); |
| 705 |
internal::VectorOperations::parallel_for(vector_add, |
| 706 |
0, |
| 707 |
size(), |
| 708 |
thread_loop_partitioner); |
| 709 |
} |
| 710 |
|
| 711 |
|
| 712 |
|
| 713 |
template <typename Number> |
| 714 |
void |
| 715 |
Vector<Number>::sadd(const Number x, const Vector<Number> &v) |
| 716 |
{ |
| 717 |
AssertIsFinite(x); |
| 718 |
|
| 719 |
Assert(size() != 0, ExcEmptyObject()); |
| 720 |
Assert(size() == v.size(), ExcDimensionMismatch(size(), v.size())); |
| 721 |
|
| 722 |
internal::VectorOperations::Vectorization_sadd_xv<Number> vector_sadd( |
| 723 |
values.begin(), v.values.begin(), x); |
| 724 |
internal::VectorOperations::parallel_for(vector_sadd, |
| 725 |
0, |
| 726 |
size(), |
| 727 |
thread_loop_partitioner); |
| 728 |
} |
| 729 |
|
| 730 |
|
| 731 |
|
| 732 |
template <typename Number> |
| 733 |
void |
| 734 |
Vector<Number>::scale(const Vector<Number> &s) |
| 735 |
{ |
| 736 |
Assert(size() != 0, ExcEmptyObject()); |
| 737 |
Assert(size() == s.size(), ExcDimensionMismatch(size(), s.size())); |
| 738 |
|
| 739 |
internal::VectorOperations::Vectorization_scale<Number> vector_scale( |
| 740 |
values.begin(), s.values.begin()); |
| 741 |
internal::VectorOperations::parallel_for(vector_scale, |
| 742 |
0, |
| 743 |
size(), |
| 744 |
thread_loop_partitioner); |
| 745 |
} |
| 746 |
|
| 747 |
|
| 748 |
|
| 749 |
template <typename Number> |
| 750 |
template <typename Number2> |
| 751 |
void |
| 752 |
Vector<Number>::scale(const Vector<Number2> &s) |
| 753 |
{ |
| 754 |
Assert(size() != 0, ExcEmptyObject()); |
| 755 |
Assert(size() == s.size(), ExcDimensionMismatch(size(), s.size())); |
| 756 |
|
| 757 |
for (size_type i = 0; i < size(); ++i) |
| 758 |
values[i] *= Number(s.values[i]); |
| 759 |
} |
| 760 |
|
| 761 |
|
| 762 |
|
| 763 |
template <typename Number> |
| 764 |
void |
| 765 |
Vector<Number>::equ(const Number a, const Vector<Number> &u) |
| 766 |
{ |
| 767 |
AssertIsFinite(a); |
| 768 |
|
| 769 |
Assert(size() != 0, ExcEmptyObject()); |
| 770 |
Assert(size() == u.size(), ExcDimensionMismatch(size(), u.size())); |
| 771 |
|
| 772 |
internal::VectorOperations::Vectorization_equ_au<Number> vector_equ( |
| 773 |
values.begin(), u.values.begin(), a); |
| 774 |
internal::VectorOperations::parallel_for(vector_equ, |
| 775 |
0, |
| 776 |
size(), |
| 777 |
thread_loop_partitioner); |
| 778 |
} |
| 779 |
|
| 780 |
|
| 781 |
|
| 782 |
template <typename Number> |
| 783 |
template <typename Number2> |
| 784 |
void |
| 785 |
Vector<Number>::equ(const Number a, const Vector<Number2> &u) |
| 786 |
{ |
| 787 |
AssertIsFinite(a); |
| 788 |
|
| 789 |
Assert(size() != 0, ExcEmptyObject()); |
| 790 |
Assert(size() == u.size(), ExcDimensionMismatch(size(), u.size())); |
| 791 |
|
| 792 |
// set the result vector to a*u. we have to |
| 793 |
// convert the elements of u to the type of |
| 794 |
// the result vector. this is necessary |
| 795 |
// because |
| 796 |
// operator*(complex<float>,complex<double>) |
| 797 |
// is not defined by default |
| 798 |
for (size_type i = 0; i < size(); ++i) |
| 799 |
values[i] = a * Number(u.values[i]); |
| 800 |
} |
| 801 |
|
| 802 |
|
| 803 |
|
| 804 |
template <typename Number> |
| 805 |
Vector<Number> & |
| 806 |
Vector<Number>::operator=(const BlockVector<Number> &v) |
| 807 |
{ |
| 808 |
if (v.size() != size()) |
| 809 |
reinit(v.size(), true); |
| 810 |
|
| 811 |
size_type this_index = 0; |
| 812 |
for (size_type b = 0; b < v.n_blocks(); ++b) |
| 813 |
for (size_type i = 0; i < v.block(b).size(); ++i, ++this_index) |
| 814 |
values[this_index] = v.block(b)(i); |
| 815 |
|
| 816 |
return *this; |
| 817 |
} |
| 818 |
|
| 819 |
|
| 820 |
|
| 821 |
#ifdef DEAL_II_WITH_PETSC |
| 822 |
template <typename Number> |
| 823 |
Vector<Number> & |
| 824 |
Vector<Number>::operator=(const PETScWrappers::VectorBase &v) |
| 825 |
{ |
| 826 |
internal::copy_petsc_vector(v, *this); |
| 827 |
return *this; |
| 828 |
} |
| 829 |
#endif |
| 830 |
|
| 831 |
|
| 832 |
#ifdef DEAL_II_WITH_TRILINOS |
| 833 |
|
| 834 |
template <typename Number> |
| 835 |
Vector<Number> & |
| 836 |
Vector<Number>::operator=(const TrilinosWrappers::MPI::Vector &v) |
| 837 |
{ |
| 838 |
if (v.size() != size()) |
| 839 |
reinit(v.size(), true); |
| 840 |
if (size() != 0) |
| 841 |
{ |
| 842 |
// Copy the distributed vector to |
| 843 |
// a local one at all processors |
| 844 |
// that know about the original vector. |
| 845 |
// TODO: There could |
| 846 |
// be a better solution than |
| 847 |
// this, but it has not yet been |
| 848 |
// found. |
| 849 |
TrilinosWrappers::MPI::Vector localized_vector; |
| 850 |
localized_vector.reinit(complete_index_set(size()), |
| 851 |
v.get_mpi_communicator()); |
| 852 |
localized_vector.reinit(v, false, true); |
| 853 |
|
| 854 |
Assert(localized_vector.size() == size(), |
| 855 |
ExcDimensionMismatch(localized_vector.size(), size())); |
| 856 |
|
| 857 |
// get a representation of the vector |
| 858 |
// and copy it |
| 859 |
TrilinosScalar **start_ptr; |
| 860 |
|
| 861 |
int ierr = localized_vector.trilinos_vector().ExtractView(&start_ptr); |
| 862 |
AssertThrow(ierr == 0, ExcTrilinosError(ierr)); |
| 863 |
|
| 864 |
std::copy(start_ptr[0], start_ptr[0] + size(), begin()); |
| 865 |
} |
| 866 |
|
| 867 |
return *this; |
| 868 |
} |
| 869 |
|
| 870 |
#endif |
| 871 |
|
| 872 |
|
| 873 |
|
| 874 |
#ifdef DEAL_II_TRILINOS_WITH_TPETRA |
| 875 |
|
| 876 |
template <typename Number> |
| 877 |
template <typename OtherNumber, typename MemorySpace> |
| 878 |
Vector<Number> & |
| 879 |
Vector<Number>::operator=( |
| 880 |
const LinearAlgebra::TpetraWrappers::Vector<OtherNumber, MemorySpace> &v) |
| 881 |
{ |
| 882 |
static_assert( |
| 883 |
std::is_same<Number, OtherNumber>::value, |
| 884 |
"TpetraWrappers::Vector and dealii::Vector must use the same number type here."); |
| 885 |
|
| 886 |
if (v.size() != size()) |
| 887 |
reinit(v.size(), true); |
| 888 |
|
| 889 |
if (size() != 0) |
| 890 |
{ |
| 891 |
// Copy the distributed vector to |
| 892 |
// a local one at all processors |
| 893 |
// that know about the original vector. |
| 894 |
LinearAlgebra::TpetraWrappers::TpetraTypes::VectorType<OtherNumber, |
| 895 |
MemorySpace> |
| 896 |
localized_vector( |
| 897 |
complete_index_set(size()) |
| 898 |
.template make_tpetra_map_rcp< |
| 899 |
LinearAlgebra::TpetraWrappers::TpetraTypes::NodeType< |
| 900 |
MemorySpace>>(), |
| 901 |
v.get_mpi_communicator()); |
| 902 |
|
| 903 |
Teuchos::RCP<const LinearAlgebra::TpetraWrappers::TpetraTypes::ImportType< |
| 904 |
MemorySpace>> |
| 905 |
importer = Tpetra::createImport(v.trilinos_vector().getMap(), |
| 906 |
localized_vector.getMap()); |
| 907 |
|
| 908 |
localized_vector.doImport(v.trilinos_vector(), *importer, Tpetra::INSERT); |
| 909 |
|
| 910 |
// get a kokkos view from the localized_vector |
| 911 |
# if DEAL_II_TRILINOS_VERSION_GTE(13, 2, 0) |
| 912 |
auto localized_vector_2d = |
| 913 |
localized_vector.template getLocalView<Kokkos::HostSpace>( |
| 914 |
Tpetra::Access::ReadOnly); |
| 915 |
# else |
| 916 |
localized_vector.template sync<Kokkos::HostSpace>(); |
| 917 |
auto localized_vector_2d = |
| 918 |
localized_vector.template getLocalView<Kokkos::HostSpace>(); |
| 919 |
# endif |
| 920 |
auto localized_vector_1d = |
| 921 |
Kokkos::subview(localized_vector_2d, Kokkos::ALL(), 0); |
| 922 |
const size_t local_length = localized_vector.getLocalLength(); |
| 923 |
|
| 924 |
Kokkos::DefaultHostExecutionSpace exec; |
| 925 |
Kokkos::View<Number *, Kokkos::HostSpace> values_view(values.data(), |
| 926 |
local_length); |
| 927 |
Kokkos::deep_copy(exec, values_view, localized_vector_1d); |
| 928 |
exec.fence(); |
| 929 |
} |
| 930 |
|
| 931 |
return *this; |
| 932 |
} |
| 933 |
|
| 934 |
#endif |
| 935 |
|
| 936 |
|
| 937 |
|
| 938 |
template <typename Number> |
| 939 |
template <typename Number2> |
| 940 |
bool |
| 941 |
Vector<Number>::operator==(const Vector<Number2> &v) const |
| 942 |
{ |
| 943 |
Assert(size() == v.size(), ExcDimensionMismatch(size(), v.size())); |
| 944 |
|
| 945 |
// compare the two vector. we have to |
| 946 |
// convert the elements of v to the type of |
| 947 |
// the result vector. this is necessary |
| 948 |
// because |
| 949 |
// operator==(complex<float>,complex<double>) |
| 950 |
// is not defined by default |
| 951 |
for (size_type i = 0; i < size(); ++i) |
| 952 |
if (values[i] != Number(v.values[i])) |
| 953 |
return false; |
| 954 |
|
| 955 |
return true; |
| 956 |
} |
| 957 |
|
| 958 |
|
| 959 |
|
| 960 |
template <typename Number> |
| 961 |
void |
| 962 |
Vector<Number>::print(std::ostream &out, |
| 963 |
const unsigned int precision, |
| 964 |
const bool scientific, |
| 965 |
const bool across) const |
| 966 |
{ |
| 967 |
Assert(size() != 0, ExcEmptyObject()); |
| 968 |
AssertThrow(out.fail() == false, ExcIO()); |
| 969 |
|
| 970 |
std::ios::fmtflags old_flags = out.flags(); |
| 971 |
unsigned int old_precision = out.precision(precision); |
| 972 |
|
| 973 |
out.precision(precision); |
| 974 |
if (scientific) |
| 975 |
out.setf(std::ios::scientific, std::ios::floatfield); |
| 976 |
else |
| 977 |
out.setf(std::ios::fixed, std::ios::floatfield); |
| 978 |
|
| 979 |
if (across) |
| 980 |
for (size_type i = 0; i < size(); ++i) |
| 981 |
out << values[i] << ' '; |
| 982 |
else |
| 983 |
for (size_type i = 0; i < size(); ++i) |
| 984 |
out << values[i] << std::endl; |
| 985 |
out << std::endl; |
| 986 |
|
| 987 |
AssertThrow(out.fail() == false, ExcIO()); |
| 988 |
// reset output format |
| 989 |
out.flags(old_flags); |
| 990 |
out.precision(old_precision); |
| 991 |
} |
| 992 |
|
| 993 |
|
| 994 |
|
| 995 |
template <typename Number> |
| 996 |
void |
| 997 |
Vector<Number>::block_write(std::ostream &out) const |
| 998 |
{ |
| 999 |
AssertThrow(out.fail() == false, ExcIO()); |
| 1000 |
|
| 1001 |
out << std::to_string(size()) << "\n["; |
| 1002 |
out.write(reinterpret_cast<const char *>(begin()), |
| 1003 |
reinterpret_cast<const char *>(end()) - |
| 1004 |
reinterpret_cast<const char *>(begin())); |
| 1005 |
out << ']'; |
| 1006 |
|
| 1007 |
AssertThrow(out.fail() == false, ExcIO()); |
| 1008 |
} |
| 1009 |
|
| 1010 |
|
| 1011 |
|
| 1012 |
template <typename Number> |
| 1013 |
void |
| 1014 |
Vector<Number>::block_read(std::istream &in) |
| 1015 |
{ |
| 1016 |
AssertThrow(in.fail() == false, ExcIO()); |
| 1017 |
|
| 1018 |
size_type sz; |
| 1019 |
|
| 1020 |
char buf[16]; |
| 1021 |
|
| 1022 |
|
| 1023 |
in.getline(buf, 16, '\n'); |
| 1024 |
sz = std::atoi(buf); |
| 1025 |
|
| 1026 |
// fast initialization, since the |
| 1027 |
// data elements are overwritten anyway |
| 1028 |
reinit(sz, true); |
| 1029 |
|
| 1030 |
char c; |
| 1031 |
in.read(&c, 1); |
| 1032 |
AssertThrow(c == '[', ExcIO()); |
| 1033 |
|
| 1034 |
in.read(reinterpret_cast<char *>(begin()), |
| 1035 |
reinterpret_cast<const char *>(end()) - |
| 1036 |
reinterpret_cast<const char *>(begin())); |
| 1037 |
|
| 1038 |
in.read(&c, 1); |
| 1039 |
AssertThrow(c == ']', ExcIO()); |
| 1040 |
} |
| 1041 |
|
| 1042 |
|
| 1043 |
|
| 1044 |
template <typename Number> |
| 1045 |
IndexSet |
| 1046 |
Vector<Number>::locally_owned_elements() const |
| 1047 |
{ |
| 1048 |
return complete_index_set(size()); |
| 1049 |
} |
| 1050 |
|
| 1051 |
|
| 1052 |
|
| 1053 |
template <typename Number> |
| 1054 |
std::size_t |
| 1055 |
Vector<Number>::memory_consumption() const |
| 1056 |
{ |
| 1057 |
return sizeof(*this) + values.memory_consumption() - sizeof(values); |
| 1058 |
} |
| 1059 |
|
| 1060 |
|
| 1061 |
|
| 1062 |
template <typename Number> |
| 1063 |
void |
| 1064 |
Vector<Number>::maybe_reset_thread_partitioner() |
| 1065 |
{ |
| 1066 |
if (size() >= 4 * internal::VectorImplementation::minimum_parallel_grain_size) |
| 1067 |
{ |
| 1068 |
if (thread_loop_partitioner == nullptr) |
| 1069 |
thread_loop_partitioner = |
| 1070 |
std::make_shared<parallel::internal::TBBPartitioner>(); |
| 1071 |
} |
| 1072 |
else |
| 1073 |
thread_loop_partitioner.reset(); |
| 1074 |
} |
| 1075 |
|
| 1076 |
|
| 1077 |
|
| 1078 |
template <typename Number> |
| 1079 |
void |
| 1080 |
Vector<Number>::do_reinit(const size_type new_size, |
| 1081 |
const bool omit_zeroing_entries, |
| 1082 |
const bool reset_partitioner) |
| 1083 |
{ |
| 1084 |
values.resize_fast(new_size); |
| 1085 |
if (!omit_zeroing_entries) |
| 1086 |
values.fill(); |
| 1087 |
|
| 1088 |
if (reset_partitioner) |
| 1089 |
maybe_reset_thread_partitioner(); |
| 1090 |
} |
| 1091 |
|
| 1092 |
|
| 1093 |
DEAL_II_NAMESPACE_CLOSE |
| 1094 |
|
| 1095 |
#endif |