| 1 |
// ------------------------------------------------------------------------ |
| 2 |
// |
| 3 |
// SPDX-License-Identifier: LGPL-2.1-or-later |
| 4 |
// Copyright (C) 2020 - 2025 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_matrix_free_tools_h |
| 16 |
#define dealii_matrix_free_tools_h |
| 17 |
|
| 18 |
#include <deal.II/base/config.h> |
| 19 |
|
| 20 |
#include <deal.II/grid/tria.h> |
| 21 |
|
| 22 |
#include <deal.II/matrix_free/fe_evaluation.h> |
| 23 |
#include <deal.II/matrix_free/matrix_free.h> |
| 24 |
#include <deal.II/matrix_free/portable_fe_evaluation.h> |
| 25 |
#include <deal.II/matrix_free/portable_matrix_free.h> |
| 26 |
#include <deal.II/matrix_free/vector_access_internal.h> |
| 27 |
|
| 28 |
#include <Kokkos_Core.hpp> |
| 29 |
|
| 30 |
|
| 31 |
DEAL_II_NAMESPACE_OPEN |
| 32 |
|
| 33 |
/** |
| 34 |
* A namespace for utility functions in the context of matrix-free operator |
| 35 |
* evaluation. |
| 36 |
*/ |
| 37 |
namespace MatrixFreeTools |
| 38 |
{ |
| 39 |
namespace internal |
| 40 |
{ |
| 41 |
template <int dim, typename Number, bool is_face_> |
| 42 |
class ComputeMatrixScratchData |
| 43 |
{ |
| 44 |
public: |
| 45 |
using FEEvalType = FEEvaluationData<dim, Number, is_face_>; |
| 46 |
|
| 47 |
std::vector<unsigned int> dof_numbers; |
| 48 |
std::vector<unsigned int> quad_numbers; |
| 49 |
std::vector<unsigned int> n_components; |
| 50 |
std::vector<unsigned int> first_selected_components; |
| 51 |
std::vector<unsigned int> batch_type; |
| 52 |
static const bool is_face = is_face_; |
| 53 |
|
| 54 |
std::function<std::vector<std::unique_ptr<FEEvalType>>( |
| 55 |
const std::pair<unsigned int, unsigned int> &)> |
| 56 |
op_create; |
| 57 |
std::function<void(std::vector<std::unique_ptr<FEEvalType>> &, |
| 58 |
const unsigned int)> |
| 59 |
op_reinit; |
| 60 |
std::function<void(std::vector<std::unique_ptr<FEEvalType>> &)> |
| 61 |
op_compute; |
| 62 |
}; |
| 63 |
} // namespace internal |
| 64 |
|
| 65 |
/** |
| 66 |
* Modify @p additional_data so that cells are categorized |
| 67 |
* according to their boundary IDs, making face integrals in the case of |
| 68 |
* cell-centric loop simpler. |
| 69 |
*/ |
| 70 |
template <int dim, typename AdditionalData> |
| 71 |
void |
| 72 |
categorize_by_boundary_ids(const Triangulation<dim> &tria, |
| 73 |
AdditionalData &additional_data); |
| 74 |
|
| 75 |
|
| 76 |
|
| 77 |
/** |
| 78 |
* Compute the diagonal of a linear operator (@p diagonal_global), given |
| 79 |
* @p matrix_free and the local cell integral operation @p cell_operation. The |
| 80 |
* vector is initialized to the right size in the function. |
| 81 |
* |
| 82 |
* The parameters @p dof_no, @p quad_no, and @p first_selected_component are |
| 83 |
* passed to the constructor of the FEEvaluation that is internally set up. |
| 84 |
* |
| 85 |
* The parameter @p first_vector_component is used to select the right |
| 86 |
* starting block in a block vector. |
| 87 |
*/ |
| 88 |
template <int dim, |
| 89 |
int fe_degree, |
| 90 |
int n_q_points_1d, |
| 91 |
int n_components, |
| 92 |
typename Number, |
| 93 |
typename VectorizedArrayType, |
| 94 |
typename VectorType> |
| 95 |
void |
| 96 |
compute_diagonal( |
| 97 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 98 |
VectorType &diagonal_global, |
| 99 |
const std::function<void(FEEvaluation<dim, |
| 100 |
fe_degree, |
| 101 |
n_q_points_1d, |
| 102 |
n_components, |
| 103 |
Number, |
| 104 |
VectorizedArrayType> &)> |
| 105 |
&cell_operation, |
| 106 |
const unsigned int dof_no = 0, |
| 107 |
const unsigned int quad_no = 0, |
| 108 |
const unsigned int first_selected_component = 0, |
| 109 |
const unsigned int first_vector_component = 0); |
| 110 |
|
| 111 |
/** |
| 112 |
* Same as above but for Portable::MatrixFree. |
| 113 |
*/ |
| 114 |
template <int dim, |
| 115 |
int fe_degree, |
| 116 |
int n_q_points_1d, |
| 117 |
int n_components, |
| 118 |
typename Number, |
| 119 |
typename MemorySpace, |
| 120 |
typename QuadOperation> |
| 121 |
void |
| 122 |
compute_diagonal( |
| 123 |
const Portable::MatrixFree<dim, Number> &matrix_free, |
| 124 |
LinearAlgebra::distributed::Vector<Number, MemorySpace> &diagonal_global, |
| 125 |
const QuadOperation &quad_operation, |
| 126 |
EvaluationFlags::EvaluationFlags evaluation_flags, |
| 127 |
EvaluationFlags::EvaluationFlags integration_flags, |
| 128 |
const unsigned int dof_no = 0, |
| 129 |
const unsigned int quad_no = 0, |
| 130 |
const unsigned int first_selected_component = 0, |
| 131 |
const unsigned int first_vector_component = 0); |
| 132 |
|
| 133 |
/** |
| 134 |
* Same as above but with a class and a function pointer. |
| 135 |
*/ |
| 136 |
template <typename CLASS, |
| 137 |
int dim, |
| 138 |
int fe_degree, |
| 139 |
int n_q_points_1d, |
| 140 |
int n_components, |
| 141 |
typename Number, |
| 142 |
typename VectorizedArrayType, |
| 143 |
typename VectorType> |
| 144 |
void |
| 145 |
compute_diagonal( |
| 146 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 147 |
VectorType &diagonal_global, |
| 148 |
void (CLASS::*cell_operation)(FEEvaluation<dim, |
| 149 |
fe_degree, |
| 150 |
n_q_points_1d, |
| 151 |
n_components, |
| 152 |
Number, |
| 153 |
VectorizedArrayType> &) const, |
| 154 |
const CLASS *owning_class, |
| 155 |
const unsigned int dof_no = 0, |
| 156 |
const unsigned int quad_no = 0, |
| 157 |
const unsigned int first_selected_component = 0, |
| 158 |
const unsigned int first_vector_component = 0); |
| 159 |
|
| 160 |
|
| 161 |
|
| 162 |
/** |
| 163 |
* Compute the diagonal of a linear operator (@p diagonal_global), given |
| 164 |
* @p matrix_free and the local cell integral operation @p cell_operation, |
| 165 |
* interior face integral operation @p face_operation, and boundary face |
| 166 |
* integral operation @p boundary_operation. The |
| 167 |
* vector is initialized to the right size in the function. |
| 168 |
* |
| 169 |
* The parameters @p dof_no, @p quad_no, and @p first_selected_component are |
| 170 |
* passed to the constructor of the FEEvaluation that is internally set up. |
| 171 |
*/ |
| 172 |
template <int dim, |
| 173 |
int fe_degree, |
| 174 |
int n_q_points_1d, |
| 175 |
int n_components, |
| 176 |
typename Number, |
| 177 |
typename VectorizedArrayType, |
| 178 |
typename VectorType> |
| 179 |
void |
| 180 |
compute_diagonal( |
| 181 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 182 |
VectorType &diagonal_global, |
| 183 |
const std::function<void(FEEvaluation<dim, |
| 184 |
fe_degree, |
| 185 |
n_q_points_1d, |
| 186 |
n_components, |
| 187 |
Number, |
| 188 |
VectorizedArrayType> &)> |
| 189 |
&cell_operation, |
| 190 |
const std::function<void(FEFaceEvaluation<dim, |
| 191 |
fe_degree, |
| 192 |
n_q_points_1d, |
| 193 |
n_components, |
| 194 |
Number, |
| 195 |
VectorizedArrayType> &, |
| 196 |
FEFaceEvaluation<dim, |
| 197 |
fe_degree, |
| 198 |
n_q_points_1d, |
| 199 |
n_components, |
| 200 |
Number, |
| 201 |
VectorizedArrayType> &)> |
| 202 |
&face_operation, |
| 203 |
const std::function<void(FEFaceEvaluation<dim, |
| 204 |
fe_degree, |
| 205 |
n_q_points_1d, |
| 206 |
n_components, |
| 207 |
Number, |
| 208 |
VectorizedArrayType> &)> |
| 209 |
&boundary_operation, |
| 210 |
const unsigned int dof_no = 0, |
| 211 |
const unsigned int quad_no = 0, |
| 212 |
const unsigned int first_selected_component = 0, |
| 213 |
const unsigned int first_vector_component = 0); |
| 214 |
|
| 215 |
|
| 216 |
|
| 217 |
/** |
| 218 |
* Same as above but with a class and a function pointer. |
| 219 |
*/ |
| 220 |
template <typename CLASS, |
| 221 |
int dim, |
| 222 |
int fe_degree, |
| 223 |
int n_q_points_1d, |
| 224 |
int n_components, |
| 225 |
typename Number, |
| 226 |
typename VectorizedArrayType, |
| 227 |
typename VectorType> |
| 228 |
void |
| 229 |
compute_diagonal( |
| 230 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 231 |
VectorType &diagonal_global, |
| 232 |
void (CLASS::*cell_operation)(FEEvaluation<dim, |
| 233 |
fe_degree, |
| 234 |
n_q_points_1d, |
| 235 |
n_components, |
| 236 |
Number, |
| 237 |
VectorizedArrayType> &) const, |
| 238 |
void (CLASS::*face_operation)(FEFaceEvaluation<dim, |
| 239 |
fe_degree, |
| 240 |
n_q_points_1d, |
| 241 |
n_components, |
| 242 |
Number, |
| 243 |
VectorizedArrayType> &, |
| 244 |
FEFaceEvaluation<dim, |
| 245 |
fe_degree, |
| 246 |
n_q_points_1d, |
| 247 |
n_components, |
| 248 |
Number, |
| 249 |
VectorizedArrayType> &) |
| 250 |
const, |
| 251 |
void (CLASS::*boundary_operation)(FEFaceEvaluation<dim, |
| 252 |
fe_degree, |
| 253 |
n_q_points_1d, |
| 254 |
n_components, |
| 255 |
Number, |
| 256 |
VectorizedArrayType> &) |
| 257 |
const, |
| 258 |
const CLASS *owning_class, |
| 259 |
const unsigned int dof_no = 0, |
| 260 |
const unsigned int quad_no = 0, |
| 261 |
const unsigned int first_selected_component = 0, |
| 262 |
const unsigned int first_vector_component = 0); |
| 263 |
|
| 264 |
|
| 265 |
|
| 266 |
/** |
| 267 |
* Compute the matrix representation of a linear operator (@p matrix), given |
| 268 |
* @p matrix_free and the local cell integral operation @p cell_operation. |
| 269 |
* Constrained entries on the diagonal are set to one. |
| 270 |
* |
| 271 |
* The parameters @p dof_no, @p quad_no, and @p first_selected_component are |
| 272 |
* passed to the constructor of the FEEvaluation that is internally set up. |
| 273 |
*/ |
| 274 |
template <int dim, |
| 275 |
int fe_degree, |
| 276 |
int n_q_points_1d, |
| 277 |
int n_components, |
| 278 |
typename Number, |
| 279 |
typename VectorizedArrayType, |
| 280 |
typename MatrixType> |
| 281 |
void |
| 282 |
compute_matrix( |
| 283 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 284 |
const AffineConstraints<Number> &constraints, |
| 285 |
MatrixType &matrix, |
| 286 |
const std::function<void(FEEvaluation<dim, |
| 287 |
fe_degree, |
| 288 |
n_q_points_1d, |
| 289 |
n_components, |
| 290 |
Number, |
| 291 |
VectorizedArrayType> &)> |
| 292 |
&cell_operation, |
| 293 |
const unsigned int dof_no = 0, |
| 294 |
const unsigned int quad_no = 0, |
| 295 |
const unsigned int first_selected_component = 0); |
| 296 |
|
| 297 |
|
| 298 |
|
| 299 |
/** |
| 300 |
* Same as above but with a class and a function pointer. |
| 301 |
*/ |
| 302 |
template <typename CLASS, |
| 303 |
int dim, |
| 304 |
int fe_degree, |
| 305 |
int n_q_points_1d, |
| 306 |
int n_components, |
| 307 |
typename Number, |
| 308 |
typename VectorizedArrayType, |
| 309 |
typename MatrixType> |
| 310 |
void |
| 311 |
compute_matrix( |
| 312 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 313 |
const AffineConstraints<Number> &constraints, |
| 314 |
MatrixType &matrix, |
| 315 |
void (CLASS::*cell_operation)(FEEvaluation<dim, |
| 316 |
fe_degree, |
| 317 |
n_q_points_1d, |
| 318 |
n_components, |
| 319 |
Number, |
| 320 |
VectorizedArrayType> &) const, |
| 321 |
const CLASS *owning_class, |
| 322 |
const unsigned int dof_no = 0, |
| 323 |
const unsigned int quad_no = 0, |
| 324 |
const unsigned int first_selected_component = 0); |
| 325 |
|
| 326 |
|
| 327 |
namespace internal |
| 328 |
{ |
| 329 |
/** |
| 330 |
* Compute the diagonal of a linear operator (@p diagonal_global), given |
| 331 |
* @p matrix_free and the local cell, face and boundary integral operation. |
| 332 |
*/ |
| 333 |
template <int dim, |
| 334 |
typename Number, |
| 335 |
typename VectorizedArrayType, |
| 336 |
typename VectorType, |
| 337 |
typename VectorType2> |
| 338 |
void |
| 339 |
compute_diagonal( |
| 340 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 341 |
const internal::ComputeMatrixScratchData<dim, VectorizedArrayType, false> |
| 342 |
&data_cell, |
| 343 |
const internal::ComputeMatrixScratchData<dim, VectorizedArrayType, true> |
| 344 |
&data_face, |
| 345 |
const internal::ComputeMatrixScratchData<dim, VectorizedArrayType, true> |
| 346 |
&data_boundary, |
| 347 |
VectorType &diagonal_global, |
| 348 |
std::vector<VectorType2 *> &diagonal_global_components); |
| 349 |
|
| 350 |
/** |
| 351 |
* Compute the matrix representation of a linear operator (@p matrix), given |
| 352 |
* @p matrix_free and the local cell, face and boundary integral operation. |
| 353 |
*/ |
| 354 |
template <int dim, |
| 355 |
typename Number, |
| 356 |
typename VectorizedArrayType, |
| 357 |
typename MatrixType> |
| 358 |
void |
| 359 |
compute_matrix( |
| 360 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 361 |
const AffineConstraints<Number> &constraints, |
| 362 |
const internal::ComputeMatrixScratchData<dim, VectorizedArrayType, false> |
| 363 |
&cell_operation, |
| 364 |
const internal::ComputeMatrixScratchData<dim, VectorizedArrayType, true> |
| 365 |
&face_operation, |
| 366 |
const internal::ComputeMatrixScratchData<dim, VectorizedArrayType, true> |
| 367 |
&boundary_operation, |
| 368 |
MatrixType &matrix); |
| 369 |
} // namespace internal |
| 370 |
|
| 371 |
|
| 372 |
|
| 373 |
/** |
| 374 |
* Compute the matrix representation of a linear operator (@p matrix), given |
| 375 |
* @p matrix_free and the local cell integral operation @p cell_operation, |
| 376 |
* interior face integral operation @p face_operation, and boundary face |
| 377 |
* integal operation @p boundary_operation. |
| 378 |
* |
| 379 |
* The parameters @p dof_no, @p quad_no, and @p first_selected_component are |
| 380 |
* passed to the constructor of the FEEvaluation that is internally set up. |
| 381 |
*/ |
| 382 |
template <int dim, |
| 383 |
int fe_degree, |
| 384 |
int n_q_points_1d, |
| 385 |
int n_components, |
| 386 |
typename Number, |
| 387 |
typename VectorizedArrayType, |
| 388 |
typename MatrixType> |
| 389 |
void |
| 390 |
compute_matrix( |
| 391 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 392 |
const AffineConstraints<Number> &constraints, |
| 393 |
MatrixType &matrix, |
| 394 |
const std::function<void(FEEvaluation<dim, |
| 395 |
fe_degree, |
| 396 |
n_q_points_1d, |
| 397 |
n_components, |
| 398 |
Number, |
| 399 |
VectorizedArrayType> &)> |
| 400 |
&cell_operation, |
| 401 |
const std::function<void(FEFaceEvaluation<dim, |
| 402 |
fe_degree, |
| 403 |
n_q_points_1d, |
| 404 |
n_components, |
| 405 |
Number, |
| 406 |
VectorizedArrayType> &, |
| 407 |
FEFaceEvaluation<dim, |
| 408 |
fe_degree, |
| 409 |
n_q_points_1d, |
| 410 |
n_components, |
| 411 |
Number, |
| 412 |
VectorizedArrayType> &)> |
| 413 |
&face_operation, |
| 414 |
const std::function<void(FEFaceEvaluation<dim, |
| 415 |
fe_degree, |
| 416 |
n_q_points_1d, |
| 417 |
n_components, |
| 418 |
Number, |
| 419 |
VectorizedArrayType> &)> |
| 420 |
&boundary_operation, |
| 421 |
const unsigned int dof_no = 0, |
| 422 |
const unsigned int quad_no = 0, |
| 423 |
const unsigned int first_selected_component = 0); |
| 424 |
|
| 425 |
|
| 426 |
|
| 427 |
/** |
| 428 |
* Same as above but with a class and a function pointer. |
| 429 |
*/ |
| 430 |
template <typename CLASS, |
| 431 |
int dim, |
| 432 |
int fe_degree, |
| 433 |
int n_q_points_1d, |
| 434 |
int n_components, |
| 435 |
typename Number, |
| 436 |
typename VectorizedArrayType, |
| 437 |
typename MatrixType> |
| 438 |
void |
| 439 |
compute_matrix( |
| 440 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 441 |
const AffineConstraints<Number> &constraints, |
| 442 |
MatrixType &matrix, |
| 443 |
void (CLASS::*cell_operation)(FEEvaluation<dim, |
| 444 |
fe_degree, |
| 445 |
n_q_points_1d, |
| 446 |
n_components, |
| 447 |
Number, |
| 448 |
VectorizedArrayType> &) const, |
| 449 |
void (CLASS::*face_operation)(FEFaceEvaluation<dim, |
| 450 |
fe_degree, |
| 451 |
n_q_points_1d, |
| 452 |
n_components, |
| 453 |
Number, |
| 454 |
VectorizedArrayType> &, |
| 455 |
FEFaceEvaluation<dim, |
| 456 |
fe_degree, |
| 457 |
n_q_points_1d, |
| 458 |
n_components, |
| 459 |
Number, |
| 460 |
VectorizedArrayType> &) |
| 461 |
const, |
| 462 |
void (CLASS::*boundary_operation)(FEFaceEvaluation<dim, |
| 463 |
fe_degree, |
| 464 |
n_q_points_1d, |
| 465 |
n_components, |
| 466 |
Number, |
| 467 |
VectorizedArrayType> &) |
| 468 |
const, |
| 469 |
const CLASS *owning_class, |
| 470 |
const unsigned int dof_no = 0, |
| 471 |
const unsigned int quad_no = 0, |
| 472 |
const unsigned int first_selected_component = 0); |
| 473 |
|
| 474 |
|
| 475 |
|
| 476 |
/** |
| 477 |
* A wrapper around MatrixFree to help users to deal with DoFHandler |
| 478 |
* objects involving cells without degrees of freedom, i.e., |
| 479 |
* cells using FE_Nothing as element type. In the following we call such |
| 480 |
* cells deactivated. All other cells are activated. In contrast to |
| 481 |
* MatrixFree, this class skips deactivated cells and faces between activated |
| 482 |
* and deactivated cells are treated as boundary faces. |
| 483 |
*/ |
| 484 |
template <int dim, |
| 485 |
typename Number, |
| 486 |
typename VectorizedArrayType = VectorizedArray<Number>> |
| 487 |
class ElementActivationAndDeactivationMatrixFree |
| 488 |
{ |
| 489 |
public: |
| 490 |
/** |
| 491 |
* Struct that helps to configure |
| 492 |
* ElementActivationAndDeactivationMatrixFree. |
| 493 |
*/ |
| 494 |
struct AdditionalData |
| 495 |
{ |
| 496 |
/** |
| 497 |
* Constructor. |
| 498 |
*/ |
| 499 |
AdditionalData(const unsigned int dof_index = 0) |
| 500 |
: dof_index(dof_index) |
| 501 |
{} |
| 502 |
|
| 503 |
/** |
| 504 |
* Index of the DoFHandler within MatrixFree to be used. |
| 505 |
*/ |
| 506 |
unsigned int dof_index; |
| 507 |
}; |
| 508 |
|
| 509 |
/** |
| 510 |
* Reinitialize class based on a given MatrixFree instance. |
| 511 |
* Particularly, the index of the valid FE is determined. |
| 512 |
* |
| 513 |
* @note At the moment, only DoFHandler objects are accepted |
| 514 |
* that are initialized with FECollection objects with at most |
| 515 |
* two finite elements (i.e., `FE_Nothing` and another finite |
| 516 |
* element). |
| 517 |
*/ |
| 518 |
void |
| 519 |
reinit(const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 520 |
const AdditionalData &additional_data = AdditionalData()) |
| 521 |
{ |
| 522 |
this->matrix_free = &matrix_free; |
| 523 |
|
| 524 |
std::vector<unsigned int> valid_fe_indices; |
| 525 |
|
| 526 |
const auto &fe_collection = |
| 527 |
matrix_free.get_dof_handler(additional_data.dof_index) |
| 528 |
.get_fe_collection(); |
| 529 |
|
| 530 |
for (unsigned int i = 0; i < fe_collection.size(); ++i) |
| 531 |
if (fe_collection[i].n_dofs_per_cell() > 0) |
| 532 |
valid_fe_indices.push_back(i); |
| 533 |
|
| 534 |
// TODO: relax this so that arbitrary number of valid |
| 535 |
// FEs can be accepted |
| 536 |
AssertDimension(valid_fe_indices.size(), 1); |
| 537 |
|
| 538 |
fe_index_valid = *valid_fe_indices.begin(); |
| 539 |
} |
| 540 |
|
| 541 |
/** |
| 542 |
* Loop over all activated cells. |
| 543 |
* |
| 544 |
* For the meaning of the parameters see MatrixFree::cell_loop(). |
| 545 |
*/ |
| 546 |
template <typename VectorTypeOut, typename VectorTypeIn> |
| 547 |
void |
| 548 |
cell_loop(const std::function<void( |
| 549 |
const MatrixFree<dim, Number, VectorizedArrayType> &, |
| 550 |
VectorTypeOut &, |
| 551 |
const VectorTypeIn &, |
| 552 |
const std::pair<unsigned int, unsigned int> &)> &cell_operation, |
| 553 |
VectorTypeOut &dst, |
| 554 |
const VectorTypeIn &src, |
| 555 |
const bool zero_dst_vector = false) const |
| 556 |
{ |
| 557 |
const auto ebd_cell_operation = [&](const auto &matrix_free, |
| 558 |
auto &dst, |
| 559 |
const auto &src, |
| 560 |
const auto &range) { |
| 561 |
const auto category = matrix_free.get_cell_range_category(range); |
| 562 |
|
| 563 |
if (category != fe_index_valid) |
| 564 |
return; |
| 565 |
|
| 566 |
cell_operation(matrix_free, dst, src, range); |
| 567 |
}; |
| 568 |
|
| 569 |
matrix_free->template cell_loop<VectorTypeOut, VectorTypeIn>( |
| 570 |
ebd_cell_operation, dst, src, zero_dst_vector); |
| 571 |
} |
| 572 |
|
| 573 |
/** |
| 574 |
* Loop over all activated cells and faces. Faces between activated and |
| 575 |
* deactivated cells are treated as boundary faces with the boundary ID |
| 576 |
* numbers::internal_face_boundary_id. |
| 577 |
* |
| 578 |
* For the meaning of the parameters see MatrixFree::cell_loop(). |
| 579 |
*/ |
| 580 |
template <typename VectorTypeOut, typename VectorTypeIn> |
| 581 |
void |
| 582 |
loop(const std::function< |
| 583 |
void(const MatrixFree<dim, Number, VectorizedArrayType> &, |
| 584 |
VectorTypeOut &, |
| 585 |
const VectorTypeIn &, |
| 586 |
const std::pair<unsigned int, unsigned int> &)> &cell_operation, |
| 587 |
const std::function< |
| 588 |
void(const MatrixFree<dim, Number, VectorizedArrayType> &, |
| 589 |
VectorTypeOut &, |
| 590 |
const VectorTypeIn &, |
| 591 |
const std::pair<unsigned int, unsigned int> &)> &face_operation, |
| 592 |
const std::function< |
| 593 |
void(const MatrixFree<dim, Number, VectorizedArrayType> &, |
| 594 |
VectorTypeOut &, |
| 595 |
const VectorTypeIn &, |
| 596 |
const std::pair<unsigned int, unsigned int> &, |
| 597 |
const bool)> &boundary_operation, |
| 598 |
VectorTypeOut &dst, |
| 599 |
const VectorTypeIn &src, |
| 600 |
const bool zero_dst_vector = false) const |
| 601 |
{ |
| 602 |
const auto ebd_cell_operation = [&](const auto &matrix_free, |
| 603 |
auto &dst, |
| 604 |
const auto &src, |
| 605 |
const auto &range) { |
| 606 |
const auto category = matrix_free.get_cell_range_category(range); |
| 607 |
|
| 608 |
if (category != fe_index_valid) |
| 609 |
return; |
| 610 |
|
| 611 |
cell_operation(matrix_free, dst, src, range); |
| 612 |
}; |
| 613 |
|
| 614 |
const auto ebd_internal_or_boundary_face_operation = |
| 615 |
[&](const auto &matrix_free, |
| 616 |
auto &dst, |
| 617 |
const auto &src, |
| 618 |
const auto &range) { |
| 619 |
const auto category = matrix_free.get_face_range_category(range); |
| 620 |
|
| 621 |
const unsigned int type = |
| 622 |
static_cast<unsigned int>(category.first == fe_index_valid) + |
| 623 |
static_cast<unsigned int>(category.second == fe_index_valid); |
| 624 |
|
| 625 |
if (type == 0) |
| 626 |
return; // deactivated face -> nothing to do |
| 627 |
|
| 628 |
if (type == 1) // boundary face |
| 629 |
boundary_operation( |
| 630 |
matrix_free, dst, src, range, category.first == fe_index_valid); |
| 631 |
else if (type == 2) // internal face |
| 632 |
face_operation(matrix_free, dst, src, range); |
| 633 |
}; |
| 634 |
|
| 635 |
matrix_free->template loop<VectorTypeOut, VectorTypeIn>( |
| 636 |
ebd_cell_operation, |
| 637 |
ebd_internal_or_boundary_face_operation, |
| 638 |
ebd_internal_or_boundary_face_operation, |
| 639 |
dst, |
| 640 |
src, |
| 641 |
zero_dst_vector); |
| 642 |
} |
| 643 |
|
| 644 |
private: |
| 645 |
/** |
| 646 |
* Reference to the underlying MatrixFree object. |
| 647 |
*/ |
| 648 |
ObserverPointer<const MatrixFree<dim, Number, VectorizedArrayType>> |
| 649 |
matrix_free; |
| 650 |
|
| 651 |
/** |
| 652 |
* Index of the valid FE. Currently, only a single one is supported. |
| 653 |
*/ |
| 654 |
unsigned int fe_index_valid; |
| 655 |
}; |
| 656 |
|
| 657 |
// implementations |
| 658 |
|
| 659 |
#ifndef DOXYGEN |
| 660 |
|
| 661 |
template <int dim, typename AdditionalData> |
| 662 |
void |
| 663 |
categorize_by_boundary_ids(const Triangulation<dim> &tria, |
| 664 |
AdditionalData &additional_data) |
| 665 |
{ |
| 666 |
// ... determine if we are on an active or a multigrid level |
| 667 |
const unsigned int level = additional_data.mg_level; |
| 668 |
const bool is_mg = (level != numbers::invalid_unsigned_int); |
| 669 |
|
| 670 |
// ... create empty list for the category of each cell |
| 671 |
if (is_mg) |
| 672 |
additional_data.cell_vectorization_category.assign( |
| 673 |
std::distance(tria.begin(level), tria.end(level)), 0); |
| 674 |
else |
| 675 |
additional_data.cell_vectorization_category.assign(tria.n_active_cells(), |
| 676 |
0); |
| 677 |
|
| 678 |
// ... set up scaling factor |
| 679 |
std::vector<unsigned int> factors(GeometryInfo<dim>::faces_per_cell); |
| 680 |
|
| 681 |
auto bids = tria.get_boundary_ids(); |
| 682 |
std::sort(bids.begin(), bids.end()); |
| 683 |
|
| 684 |
{ |
| 685 |
unsigned int n_bids = bids.size() + 1; |
| 686 |
int offset = 1; |
| 687 |
for (unsigned int i = 0; i < GeometryInfo<dim>::faces_per_cell; |
| 688 |
i++, offset = offset * n_bids) |
| 689 |
factors[i] = offset; |
| 690 |
} |
| 691 |
|
| 692 |
const auto to_category = [&](const auto &cell) { |
| 693 |
unsigned int c_num = 0; |
| 694 |
for (unsigned int i = 0; i < GeometryInfo<dim>::faces_per_cell; ++i) |
| 695 |
{ |
| 696 |
const auto face = cell->face(i); |
| 697 |
if (face->at_boundary() && !cell->has_periodic_neighbor(i)) |
| 698 |
c_num += |
| 699 |
factors[i] * (1 + std::distance(bids.begin(), |
| 700 |
std::find(bids.begin(), |
| 701 |
bids.end(), |
| 702 |
face->boundary_id()))); |
| 703 |
} |
| 704 |
return c_num; |
| 705 |
}; |
| 706 |
|
| 707 |
if (!is_mg) |
| 708 |
{ |
| 709 |
for (auto cell = tria.begin_active(); cell != tria.end(); ++cell) |
| 710 |
{ |
| 711 |
if (cell->is_locally_owned()) |
| 712 |
additional_data |
| 713 |
.cell_vectorization_category[cell->active_cell_index()] = |
| 714 |
to_category(cell); |
| 715 |
} |
| 716 |
} |
| 717 |
else |
| 718 |
{ |
| 719 |
for (auto cell = tria.begin(level); cell != tria.end(level); ++cell) |
| 720 |
{ |
| 721 |
if (cell->is_locally_owned_on_level()) |
| 722 |
additional_data.cell_vectorization_category[cell->index()] = |
| 723 |
to_category(cell); |
| 724 |
} |
| 725 |
} |
| 726 |
|
| 727 |
// ... finalize set up of matrix_free |
| 728 |
additional_data.hold_all_faces_to_owned_cells = true; |
| 729 |
additional_data.cell_vectorization_categories_strict = true; |
| 730 |
additional_data.mapping_update_flags_faces_by_cells = |
| 731 |
additional_data.mapping_update_flags_inner_faces | |
| 732 |
additional_data.mapping_update_flags_boundary_faces; |
| 733 |
} |
| 734 |
|
| 735 |
namespace internal |
| 736 |
{ |
| 737 |
template <typename Number> |
| 738 |
struct LocalCSR |
| 739 |
{ |
| 740 |
std::vector<unsigned int> row_lid_to_gid; |
| 741 |
std::vector<unsigned int> row; |
| 742 |
std::vector<unsigned int> col; |
| 743 |
std::vector<Number> val; |
| 744 |
|
| 745 |
std::vector<unsigned int> inverse_lookup_rows; |
| 746 |
std::vector<std::pair<unsigned int, unsigned int>> inverse_lookup_origins; |
| 747 |
}; |
| 748 |
|
| 749 |
template <int dim, typename VectorizedArrayType, bool is_face> |
| 750 |
class ComputeDiagonalHelper |
| 751 |
{ |
| 752 |
public: |
| 753 |
using FEEvaluationType = |
| 754 |
FEEvaluationData<dim, VectorizedArrayType, is_face>; |
| 755 |
|
| 756 |
using Number = typename VectorizedArrayType::value_type; |
| 757 |
static const unsigned int n_lanes = VectorizedArrayType::size(); |
| 758 |
|
| 759 |
ComputeDiagonalHelper() |
| 760 |
: phi(nullptr) |
| 761 |
, matrix_free(nullptr) |
| 762 |
, dofs_per_component(0) |
| 763 |
, n_components(0) |
| 764 |
{} |
| 765 |
|
| 766 |
ComputeDiagonalHelper(const ComputeDiagonalHelper &) |
| 767 |
: phi(nullptr) |
| 768 |
, matrix_free(nullptr) |
| 769 |
, dofs_per_component(0) |
| 770 |
, n_components(0) |
| 771 |
{} |
| 772 |
|
| 773 |
void |
| 774 |
initialize( |
| 775 |
FEEvaluationType &phi, |
| 776 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 777 |
const unsigned int n_components) |
| 778 |
{ |
| 779 |
// if we are in hp mode and the number of unknowns changed, we must |
| 780 |
// clear the map of entries |
| 781 |
if (dofs_per_component != |
| 782 |
phi.get_shape_info().dofs_per_component_on_cell) |
| 783 |
{ |
| 784 |
locally_relevant_constraints_hn_map.clear(); |
| 785 |
dofs_per_component = |
| 786 |
phi.get_shape_info().dofs_per_component_on_cell; |
| 787 |
} |
| 788 |
this->n_components = n_components; |
| 789 |
this->dofs_per_cell = n_components * dofs_per_component; |
| 790 |
this->phi = φ |
| 791 |
this->matrix_free = &matrix_free; |
| 792 |
} |
| 793 |
|
| 794 |
void |
| 795 |
reinit(const unsigned int cell) |
| 796 |
{ |
| 797 |
// STEP 1: get relevant information from FEEvaluation |
| 798 |
const auto &matrix_free = *this->matrix_free; |
| 799 |
const auto &dof_info = phi->get_dof_info(); |
| 800 |
const unsigned int n_fe_components = dof_info.start_components.back(); |
| 801 |
|
| 802 |
// if we have a block vector with components with the same DoFHandler, |
| 803 |
// each component is described with same set of constraints and |
| 804 |
// we consider the shift in components only during access of the global |
| 805 |
// vector |
| 806 |
const unsigned int first_selected_component = |
| 807 |
n_fe_components == 1 ? 0 : phi->get_first_selected_component(); |
| 808 |
|
| 809 |
this->n_lanes_filled = |
| 810 |
is_face ? matrix_free.n_active_entries_per_face_batch(cell) : |
| 811 |
matrix_free.n_active_entries_per_cell_batch(cell); |
| 812 |
|
| 813 |
// STEP 2: setup CSR storage of transposed locally-relevant |
| 814 |
// constraint matrix |
| 815 |
|
| 816 |
// (constrained local index, global index of dof |
| 817 |
// constraints, weight) |
| 818 |
std::vector<std::tuple<unsigned int, unsigned int, Number>> |
| 819 |
locally_relevant_constraints, locally_relevant_constraints_tmp; |
| 820 |
locally_relevant_constraints.reserve(dofs_per_cell); |
| 821 |
std::vector<unsigned int> constraint_position; |
| 822 |
std::vector<unsigned char> is_constrained_hn; |
| 823 |
|
| 824 |
const std::array<unsigned int, n_lanes> &cells = |
| 825 |
this->phi->get_cell_ids(); |
| 826 |
|
| 827 |
for (unsigned int v = 0; v < n_lanes_filled; ++v) |
| 828 |
{ |
| 829 |
Assert(cells[v] != numbers::invalid_unsigned_int, |
| 830 |
ExcInternalError()); |
| 831 |
|
| 832 |
const unsigned int *dof_indices; |
| 833 |
unsigned int index_indicators, next_index_indicators; |
| 834 |
|
| 835 |
const unsigned int start = |
| 836 |
cells[v] * n_fe_components + first_selected_component; |
| 837 |
dof_indices = |
| 838 |
dof_info.dof_indices.data() + dof_info.row_starts[start].first; |
| 839 |
index_indicators = dof_info.row_starts[start].second; |
| 840 |
next_index_indicators = dof_info.row_starts[start + 1].second; |
| 841 |
|
| 842 |
// STEP 2a: setup locally-relevant constraint matrix in a |
| 843 |
// coordinate list (COO) |
| 844 |
locally_relevant_constraints.clear(); |
| 845 |
|
| 846 |
if (n_components == 1 || n_fe_components == 1) |
| 847 |
{ |
| 848 |
unsigned int ind_local = 0; |
| 849 |
for (; index_indicators != next_index_indicators; |
| 850 |
++index_indicators, ++ind_local) |
| 851 |
{ |
| 852 |
const std::pair<unsigned short, unsigned short> indicator = |
| 853 |
dof_info.constraint_indicator[index_indicators]; |
| 854 |
|
| 855 |
for (unsigned int j = 0; j < indicator.first; |
| 856 |
++j, ++ind_local) |
| 857 |
locally_relevant_constraints.emplace_back(ind_local, |
| 858 |
dof_indices[j], |
| 859 |
1.0); |
| 860 |
|
| 861 |
dof_indices += indicator.first; |
| 862 |
|
| 863 |
const Number *data_val = |
| 864 |
matrix_free.constraint_pool_begin(indicator.second); |
| 865 |
const Number *end_pool = |
| 866 |
matrix_free.constraint_pool_end(indicator.second); |
| 867 |
|
| 868 |
for (; data_val != end_pool; ++data_val, ++dof_indices) |
| 869 |
locally_relevant_constraints.emplace_back(ind_local, |
| 870 |
*dof_indices, |
| 871 |
*data_val); |
| 872 |
} |
| 873 |
|
| 874 |
AssertIndexRange(ind_local, dofs_per_component + 1); |
| 875 |
|
| 876 |
for (; ind_local < dofs_per_component; |
| 877 |
++dof_indices, ++ind_local) |
| 878 |
locally_relevant_constraints.emplace_back(ind_local, |
| 879 |
*dof_indices, |
| 880 |
1.0); |
| 881 |
} |
| 882 |
else |
| 883 |
{ |
| 884 |
// case with vector-valued finite elements where all |
| 885 |
// components are included in one single vector. Assumption: |
| 886 |
// first come all entries to the first component, then all |
| 887 |
// entries to the second one, and so on. This is ensured by |
| 888 |
// the way MatrixFree reads out the indices. |
| 889 |
for (unsigned int comp = 0; comp < n_components; ++comp) |
| 890 |
{ |
| 891 |
unsigned int ind_local = 0; |
| 892 |
|
| 893 |
// check whether there is any constraint on the current |
| 894 |
// cell |
| 895 |
for (; index_indicators != next_index_indicators; |
| 896 |
++index_indicators, ++ind_local) |
| 897 |
{ |
| 898 |
const std::pair<unsigned short, unsigned short> |
| 899 |
indicator = |
| 900 |
dof_info.constraint_indicator[index_indicators]; |
| 901 |
|
| 902 |
// run through values up to next constraint |
| 903 |
for (unsigned int j = 0; j < indicator.first; |
| 904 |
++j, ++ind_local) |
| 905 |
locally_relevant_constraints.emplace_back( |
| 906 |
comp * dofs_per_component + ind_local, |
| 907 |
dof_indices[j], |
| 908 |
1.0); |
| 909 |
dof_indices += indicator.first; |
| 910 |
|
| 911 |
const Number *data_val = |
| 912 |
matrix_free.constraint_pool_begin(indicator.second); |
| 913 |
const Number *end_pool = |
| 914 |
matrix_free.constraint_pool_end(indicator.second); |
| 915 |
|
| 916 |
for (; data_val != end_pool; ++data_val, ++dof_indices) |
| 917 |
locally_relevant_constraints.emplace_back( |
| 918 |
comp * dofs_per_component + ind_local, |
| 919 |
*dof_indices, |
| 920 |
*data_val); |
| 921 |
} |
| 922 |
|
| 923 |
AssertIndexRange(ind_local, dofs_per_component + 1); |
| 924 |
|
| 925 |
// get the dof values past the last constraint |
| 926 |
for (; ind_local < dofs_per_component; |
| 927 |
++dof_indices, ++ind_local) |
| 928 |
locally_relevant_constraints.emplace_back( |
| 929 |
comp * dofs_per_component + ind_local, |
| 930 |
*dof_indices, |
| 931 |
1.0); |
| 932 |
|
| 933 |
if (comp + 1 < n_components) |
| 934 |
next_index_indicators = |
| 935 |
dof_info.row_starts[start + comp + 2].second; |
| 936 |
} |
| 937 |
} |
| 938 |
|
| 939 |
// we only need partial sortedness for the algorithm below in that |
| 940 |
// all entries for a particular row must be adjacent. this is |
| 941 |
// ensured by the way we fill the field, but check it again |
| 942 |
for (unsigned int i = 1; i < locally_relevant_constraints.size(); |
| 943 |
++i) |
| 944 |
Assert(std::get<0>(locally_relevant_constraints[i]) >= |
| 945 |
std::get<0>(locally_relevant_constraints[i - 1]), |
| 946 |
ExcInternalError()); |
| 947 |
|
| 948 |
// STEP 2c: apply hanging-node constraints |
| 949 |
if (dof_info.hanging_node_constraint_masks.size() > 0 && |
| 950 |
dof_info.hanging_node_constraint_masks_comp.size() > 0 && |
| 951 |
dof_info.hanging_node_constraint_masks_comp |
| 952 |
[phi->get_active_fe_index()][first_selected_component]) |
| 953 |
{ |
| 954 |
const auto mask = |
| 955 |
dof_info.hanging_node_constraint_masks[cells[v]]; |
| 956 |
|
| 957 |
// cell has hanging nodes |
| 958 |
if (mask != dealii::internal::MatrixFreeFunctions:: |
| 959 |
unconstrained_compressed_constraint_kind) |
| 960 |
{ |
| 961 |
// check if hanging node internpolation matrix has been set |
| 962 |
// up |
| 963 |
if (locally_relevant_constraints_hn_map.find(mask) == |
| 964 |
locally_relevant_constraints_hn_map.end()) |
| 965 |
fill_constraint_type_into_map(mask); |
| 966 |
|
| 967 |
const auto &locally_relevant_constraints_hn = |
| 968 |
locally_relevant_constraints_hn_map[mask]; |
| 969 |
|
| 970 |
locally_relevant_constraints_tmp.clear(); |
| 971 |
if (locally_relevant_constraints_tmp.capacity() < |
| 972 |
locally_relevant_constraints.size()) |
| 973 |
locally_relevant_constraints_tmp.reserve( |
| 974 |
locally_relevant_constraints.size() + |
| 975 |
locally_relevant_constraints_hn.size()); |
| 976 |
|
| 977 |
// combine with other constraints: to avoid binary |
| 978 |
// searches, we first build a list of where constraints |
| 979 |
// are pointing to, and then merge the two lists |
| 980 |
constraint_position.assign(dofs_per_cell, |
| 981 |
numbers::invalid_unsigned_int); |
| 982 |
for (auto &a : locally_relevant_constraints) |
| 983 |
if (constraint_position[std::get<0>(a)] == |
| 984 |
numbers::invalid_unsigned_int) |
| 985 |
constraint_position[std::get<0>(a)] = |
| 986 |
std::distance(locally_relevant_constraints.data(), |
| 987 |
&a); |
| 988 |
is_constrained_hn.assign(dofs_per_cell, false); |
| 989 |
for (auto &hn : locally_relevant_constraints_hn) |
| 990 |
is_constrained_hn[std::get<0>(hn)] = 1; |
| 991 |
|
| 992 |
// not constrained from hanging nodes |
| 993 |
for (const auto &a : locally_relevant_constraints) |
| 994 |
if (is_constrained_hn[std::get<0>(a)] == 0) |
| 995 |
locally_relevant_constraints_tmp.push_back(a); |
| 996 |
|
| 997 |
// dof is constrained by hanging nodes: build transitive |
| 998 |
// closure |
| 999 |
for (const auto &hn : locally_relevant_constraints_hn) |
| 1000 |
if (constraint_position[std::get<1>(hn)] != |
| 1001 |
numbers::invalid_unsigned_int) |
| 1002 |
{ |
| 1003 |
AssertIndexRange(constraint_position[std::get<1>(hn)], |
| 1004 |
locally_relevant_constraints.size()); |
| 1005 |
auto other = locally_relevant_constraints.begin() + |
| 1006 |
constraint_position[std::get<1>(hn)]; |
| 1007 |
AssertDimension(std::get<0>(*other), std::get<1>(hn)); |
| 1008 |
|
| 1009 |
for (; other != locally_relevant_constraints.end() && |
| 1010 |
std::get<0>(*other) == std::get<1>(hn); |
| 1011 |
++other) |
| 1012 |
locally_relevant_constraints_tmp.emplace_back( |
| 1013 |
std::get<0>(hn), |
| 1014 |
std::get<1>(*other), |
| 1015 |
std::get<2>(hn) * std::get<2>(*other)); |
| 1016 |
} |
| 1017 |
|
| 1018 |
std::swap(locally_relevant_constraints, |
| 1019 |
locally_relevant_constraints_tmp); |
| 1020 |
} |
| 1021 |
} |
| 1022 |
|
| 1023 |
// STEP 2d: transpose COO |
| 1024 |
std::sort(locally_relevant_constraints.begin(), |
| 1025 |
locally_relevant_constraints.end(), |
| 1026 |
[](const auto &a, const auto &b) { |
| 1027 |
if (std::get<1>(a) < std::get<1>(b)) |
| 1028 |
return true; |
| 1029 |
return (std::get<1>(a) == std::get<1>(b)) && |
| 1030 |
(std::get<0>(a) < std::get<0>(b)); |
| 1031 |
}); |
| 1032 |
|
| 1033 |
// STEP 2e: translate COO to CRS |
| 1034 |
auto &c_pool = c_pools[v]; |
| 1035 |
{ |
| 1036 |
c_pool.row_lid_to_gid.clear(); |
| 1037 |
c_pool.row.clear(); |
| 1038 |
c_pool.row.push_back(0); |
| 1039 |
c_pool.col.clear(); |
| 1040 |
c_pool.val.clear(); |
| 1041 |
|
| 1042 |
if (locally_relevant_constraints.size() > 0) |
| 1043 |
c_pool.row_lid_to_gid.emplace_back( |
| 1044 |
std::get<1>(locally_relevant_constraints.front())); |
| 1045 |
for (const auto &j : locally_relevant_constraints) |
| 1046 |
{ |
| 1047 |
if (c_pool.row_lid_to_gid.back() != std::get<1>(j)) |
| 1048 |
{ |
| 1049 |
c_pool.row_lid_to_gid.push_back(std::get<1>(j)); |
| 1050 |
c_pool.row.push_back(c_pool.val.size()); |
| 1051 |
} |
| 1052 |
|
| 1053 |
c_pool.col.emplace_back(std::get<0>(j)); |
| 1054 |
c_pool.val.emplace_back(std::get<2>(j)); |
| 1055 |
} |
| 1056 |
|
| 1057 |
if (c_pool.val.size() > 0) |
| 1058 |
c_pool.row.push_back(c_pool.val.size()); |
| 1059 |
|
| 1060 |
c_pool.inverse_lookup_rows.clear(); |
| 1061 |
c_pool.inverse_lookup_rows.resize(1 + dofs_per_cell); |
| 1062 |
for (const unsigned int i : c_pool.col) |
| 1063 |
c_pool.inverse_lookup_rows[1 + i]++; |
| 1064 |
// transform to offsets |
| 1065 |
std::partial_sum(c_pool.inverse_lookup_rows.begin(), |
| 1066 |
c_pool.inverse_lookup_rows.end(), |
| 1067 |
c_pool.inverse_lookup_rows.begin()); |
| 1068 |
AssertDimension(c_pool.inverse_lookup_rows.back(), |
| 1069 |
c_pool.col.size()); |
| 1070 |
|
| 1071 |
c_pool.inverse_lookup_origins.resize(c_pool.col.size()); |
| 1072 |
std::vector<unsigned int> inverse_lookup_count(dofs_per_cell); |
| 1073 |
for (unsigned int row = 0; row < c_pool.row.size() - 1; ++row) |
| 1074 |
for (unsigned int col = c_pool.row[row]; |
| 1075 |
col < c_pool.row[row + 1]; |
| 1076 |
++col) |
| 1077 |
{ |
| 1078 |
const unsigned int index = c_pool.col[col]; |
| 1079 |
c_pool.inverse_lookup_origins |
| 1080 |
[c_pool.inverse_lookup_rows[index] + |
| 1081 |
inverse_lookup_count[index]] = std::make_pair(row, col); |
| 1082 |
++inverse_lookup_count[index]; |
| 1083 |
} |
| 1084 |
} |
| 1085 |
} |
| 1086 |
|
| 1087 |
// STEP 3: compute element matrix A_e, apply |
| 1088 |
// locally-relevant constraints C_e^T * A_e * C_e, and get the |
| 1089 |
// the diagonal entry |
| 1090 |
// (C_e^T * A_e * C_e)(i,i) |
| 1091 |
// or |
| 1092 |
// C_e^T(i,:) * A_e * C_e(:,i). |
| 1093 |
// |
| 1094 |
// Since, we compute the element matrix column-by-column and as a |
| 1095 |
// result never actually have the full element matrix, we actually |
| 1096 |
// perform following steps: |
| 1097 |
// 1) loop over all columns of the element matrix |
| 1098 |
// a) compute column i |
| 1099 |
// b) compute for each j (rows of C_e^T): |
| 1100 |
// (C_e^T(j,:) * A_e(:,i)) * C_e(i,j) |
| 1101 |
// or |
| 1102 |
// (C_e^T(j,:) * A_e(:,i)) * C_e^T(j,i) |
| 1103 |
// This gives a contribution the j-th entry of the |
| 1104 |
// locally-relevant diagonal and comprises the multiplication |
| 1105 |
// by the locally-relevant constraint matrix from the left and |
| 1106 |
// the right. There is no contribution to the j-th vector |
| 1107 |
// entry if the j-th row of C_e^T is empty or C_e^T(j,i) is |
| 1108 |
// zero. |
| 1109 |
|
| 1110 |
// set size locally-relevant diagonal |
| 1111 |
for (unsigned int v = 0; v < n_lanes_filled; ++v) |
| 1112 |
diagonals_local_constrained[v].assign( |
| 1113 |
c_pools[v].row_lid_to_gid.size() * |
| 1114 |
(n_fe_components == 1 ? n_components : 1), |
| 1115 |
Number(0.0)); |
| 1116 |
|
| 1117 |
// check if fast path can be taken via FEEvaluation |
| 1118 |
bool use_fast_path = true; |
| 1119 |
|
| 1120 |
for (unsigned int v = 0; v < n_lanes_filled; ++v) |
| 1121 |
{ |
| 1122 |
auto &c_pool = c_pools[v]; |
| 1123 |
|
| 1124 |
for (unsigned int i = 0; i < c_pool.row.size() - 1; ++i) |
| 1125 |
{ |
| 1126 |
if ((c_pool.row[i + 1] - c_pool.row[i]) > 1) |
| 1127 |
{ |
| 1128 |
use_fast_path = false; |
| 1129 |
break; |
| 1130 |
} |
| 1131 |
else if (((c_pool.row[i + 1] - c_pool.row[i]) == 1) && |
| 1132 |
(c_pool.val[c_pool.row[i]] != 1.0)) |
| 1133 |
{ |
| 1134 |
use_fast_path = false; |
| 1135 |
break; |
| 1136 |
} |
| 1137 |
} |
| 1138 |
|
| 1139 |
if (use_fast_path == false) |
| 1140 |
break; |
| 1141 |
} |
| 1142 |
|
| 1143 |
this->has_simple_constraints_ = use_fast_path; |
| 1144 |
} |
| 1145 |
|
| 1146 |
void |
| 1147 |
fill_constraint_type_into_map( |
| 1148 |
const dealii::internal::MatrixFreeFunctions::compressed_constraint_kind |
| 1149 |
mask) |
| 1150 |
{ |
| 1151 |
auto &constraints_hn = locally_relevant_constraints_hn_map[mask]; |
| 1152 |
|
| 1153 |
// assume that we constrain one face, i.e., (fe_degree + 1)^(dim-1) |
| 1154 |
// unknowns - we might have more or less entries, but this is a good |
| 1155 |
// first guess |
| 1156 |
const unsigned int degree = |
| 1157 |
phi->get_shape_info().data.front().fe_degree; |
| 1158 |
constraints_hn.reserve(Utilities::pow(degree + 1, dim - 1)); |
| 1159 |
|
| 1160 |
// 1) collect hanging-node constraints for cell assuming |
| 1161 |
// scalar finite element |
| 1162 |
values_dofs.resize(dofs_per_component); |
| 1163 |
std::array< |
| 1164 |
dealii::internal::MatrixFreeFunctions::compressed_constraint_kind, |
| 1165 |
VectorizedArrayType::size()> |
| 1166 |
constraint_mask; |
| 1167 |
constraint_mask.fill(dealii::internal::MatrixFreeFunctions:: |
| 1168 |
unconstrained_compressed_constraint_kind); |
| 1169 |
constraint_mask[0] = mask; |
| 1170 |
|
| 1171 |
for (unsigned int i = 0; i < dofs_per_component; ++i) |
| 1172 |
{ |
| 1173 |
for (unsigned int j = 0; j < dofs_per_component; ++j) |
| 1174 |
values_dofs[j] = VectorizedArrayType(); |
| 1175 |
values_dofs[i] = Number(1); |
| 1176 |
|
| 1177 |
dealii::internal::FEEvaluationHangingNodesFactory< |
| 1178 |
dim, |
| 1179 |
Number, |
| 1180 |
VectorizedArrayType>::apply(1, |
| 1181 |
degree, |
| 1182 |
phi->get_shape_info(), |
| 1183 |
false, |
| 1184 |
constraint_mask, |
| 1185 |
values_dofs.data()); |
| 1186 |
|
| 1187 |
const Number tolerance = |
| 1188 |
std::max(Number(1e-12), |
| 1189 |
std::numeric_limits<Number>::epsilon() * 16); |
| 1190 |
for (unsigned int j = 0; j < dofs_per_component; ++j) |
| 1191 |
if (std::abs(values_dofs[j][0]) > tolerance && |
| 1192 |
(j != i || |
| 1193 |
std::abs(values_dofs[j][0] - Number(1)) > tolerance)) |
| 1194 |
constraints_hn.emplace_back(j, i, values_dofs[j][0]); |
| 1195 |
} |
| 1196 |
|
| 1197 |
// 2) extend for multiple components |
| 1198 |
const unsigned int n_hn_constraints = constraints_hn.size(); |
| 1199 |
constraints_hn.resize(n_hn_constraints * n_components); |
| 1200 |
|
| 1201 |
for (unsigned int c = 1; c < n_components; ++c) |
| 1202 |
for (unsigned int i = 0; i < n_hn_constraints; ++i) |
| 1203 |
constraints_hn[c * n_hn_constraints + i] = std::make_tuple( |
| 1204 |
std::get<0>(constraints_hn[i]) + c * dofs_per_component, |
| 1205 |
std::get<1>(constraints_hn[i]) + c * dofs_per_component, |
| 1206 |
std::get<2>(constraints_hn[i])); |
| 1207 |
} |
| 1208 |
|
| 1209 |
void |
| 1210 |
prepare_basis_vector(const unsigned int i) |
| 1211 |
{ |
| 1212 |
this->i = i; |
| 1213 |
|
| 1214 |
// compute i-th column of element stiffness matrix: |
| 1215 |
// this could be simply performed as done at the moment with |
| 1216 |
// matrix-free operator evaluation applied to a ith-basis vector |
| 1217 |
VectorizedArrayType *dof_values = phi->begin_dof_values(); |
| 1218 |
for (unsigned int j = 0; j < dofs_per_cell; ++j) |
| 1219 |
dof_values[j] = VectorizedArrayType(); |
| 1220 |
dof_values[i] = Number(1); |
| 1221 |
} |
| 1222 |
|
| 1223 |
void |
| 1224 |
zero_basis_vector() |
| 1225 |
{ |
| 1226 |
VectorizedArrayType *dof_values = phi->begin_dof_values(); |
| 1227 |
for (unsigned int j = 0; j < dofs_per_cell; ++j) |
| 1228 |
dof_values[j] = VectorizedArrayType(); |
| 1229 |
} |
| 1230 |
|
| 1231 |
void |
| 1232 |
submit() |
| 1233 |
{ |
| 1234 |
// if we have a block vector with components with the same DoFHandler, |
| 1235 |
// we need to figure out which component and which DoF within the |
| 1236 |
// component are we currently considering |
| 1237 |
const unsigned int n_fe_components = |
| 1238 |
phi->get_dof_info().start_components.back(); |
| 1239 |
const unsigned int comp = |
| 1240 |
n_fe_components == 1 ? i / dofs_per_component : 0; |
| 1241 |
const unsigned int i_comp = |
| 1242 |
n_fe_components == 1 ? (i % dofs_per_component) : i; |
| 1243 |
|
| 1244 |
// apply local constraint matrix from left and from right: |
| 1245 |
// loop over all rows of transposed constrained matrix |
| 1246 |
for (unsigned int v = 0; v < n_lanes_filled; ++v) |
| 1247 |
{ |
| 1248 |
const auto &c_pool = c_pools[v]; |
| 1249 |
|
| 1250 |
for (unsigned int jj = c_pool.inverse_lookup_rows[i_comp]; |
| 1251 |
jj < c_pool.inverse_lookup_rows[i_comp + 1]; |
| 1252 |
++jj) |
| 1253 |
{ |
| 1254 |
const unsigned int j = c_pool.inverse_lookup_origins[jj].first; |
| 1255 |
// apply constraint matrix from the left |
| 1256 |
Number temp = 0.0; |
| 1257 |
for (unsigned int k = c_pool.row[j]; k < c_pool.row[j + 1]; ++k) |
| 1258 |
temp += c_pool.val[k] * |
| 1259 |
phi->begin_dof_values()[comp * dofs_per_component + |
| 1260 |
c_pool.col[k]][v]; |
| 1261 |
|
| 1262 |
// apply constraint matrix from the right |
| 1263 |
diagonals_local_constrained |
| 1264 |
[v][j + comp * c_pools[v].row_lid_to_gid.size()] += |
| 1265 |
temp * c_pool.val[c_pool.inverse_lookup_origins[jj].second]; |
| 1266 |
} |
| 1267 |
} |
| 1268 |
} |
| 1269 |
|
| 1270 |
template <typename VectorType> |
| 1271 |
inline void |
| 1272 |
distribute_local_to_global(std::vector<VectorType *> &diagonal_global) |
| 1273 |
{ |
| 1274 |
// STEP 4: assembly results: add into global vector |
| 1275 |
const unsigned int n_fe_components = |
| 1276 |
phi->get_dof_info().start_components.back(); |
| 1277 |
|
| 1278 |
if (n_fe_components == 1) |
| 1279 |
AssertDimension(diagonal_global.size(), n_components); |
| 1280 |
|
| 1281 |
for (unsigned int v = 0; v < n_lanes_filled; ++v) |
| 1282 |
// if we have a block vector with components with the same |
| 1283 |
// DoFHandler, we need to loop over all components manually and |
| 1284 |
// need to apply the correct shift |
| 1285 |
for (unsigned int j = 0; j < c_pools[v].row.size() - 1; ++j) |
| 1286 |
for (unsigned int comp = 0; |
| 1287 |
comp < (n_fe_components == 1 ? |
| 1288 |
static_cast<unsigned int>(n_components) : |
| 1289 |
1); |
| 1290 |
++comp) |
| 1291 |
::dealii::internal::vector_access_add( |
| 1292 |
*diagonal_global[n_fe_components == 1 ? comp : 0], |
| 1293 |
c_pools[v].row_lid_to_gid[j], |
| 1294 |
diagonals_local_constrained |
| 1295 |
[v][j + comp * c_pools[v].row_lid_to_gid.size()]); |
| 1296 |
} |
| 1297 |
|
| 1298 |
bool |
| 1299 |
has_simple_constraints() const |
| 1300 |
{ |
| 1301 |
return has_simple_constraints_; |
| 1302 |
} |
| 1303 |
|
| 1304 |
private: |
| 1305 |
FEEvaluationType *phi; |
| 1306 |
const MatrixFree<dim, Number, VectorizedArrayType> *matrix_free; |
| 1307 |
|
| 1308 |
unsigned int dofs_per_component; |
| 1309 |
unsigned int dofs_per_cell; |
| 1310 |
unsigned int n_components; |
| 1311 |
|
| 1312 |
unsigned int i; |
| 1313 |
|
| 1314 |
unsigned int n_lanes_filled; |
| 1315 |
|
| 1316 |
std::array<internal::LocalCSR<Number>, n_lanes> c_pools; |
| 1317 |
|
| 1318 |
// local storage: buffer so that we access the global vector once |
| 1319 |
// note: may be larger then dofs_per_cell in the presence of |
| 1320 |
// constraints! |
| 1321 |
std::array<std::vector<Number>, n_lanes> diagonals_local_constrained; |
| 1322 |
|
| 1323 |
std::map< |
| 1324 |
dealii::internal::MatrixFreeFunctions::compressed_constraint_kind, |
| 1325 |
std::vector<std::tuple<unsigned int, unsigned int, Number>>> |
| 1326 |
locally_relevant_constraints_hn_map; |
| 1327 |
|
| 1328 |
// scratch array |
| 1329 |
AlignedVector<VectorizedArrayType> values_dofs; |
| 1330 |
|
| 1331 |
bool has_simple_constraints_; |
| 1332 |
}; |
| 1333 |
|
| 1334 |
template <bool is_face, |
| 1335 |
int dim, |
| 1336 |
typename Number, |
| 1337 |
typename VectorizedArrayType> |
| 1338 |
bool |
| 1339 |
is_fe_nothing( |
| 1340 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 1341 |
const std::pair<unsigned int, unsigned int> &range, |
| 1342 |
const unsigned int dof_no, |
| 1343 |
const unsigned int quad_no, |
| 1344 |
const unsigned int first_selected_component, |
| 1345 |
const unsigned int fe_degree, |
| 1346 |
const unsigned int n_q_points_1d, |
| 1347 |
const bool is_interior_face = true) |
| 1348 |
{ |
| 1349 |
const unsigned int static_n_q_points = |
| 1350 |
is_face ? Utilities::pow(n_q_points_1d, dim - 1) : |
| 1351 |
Utilities::pow(n_q_points_1d, dim); |
| 1352 |
|
| 1353 |
unsigned int active_fe_index = 0; |
| 1354 |
if (!is_face) |
| 1355 |
active_fe_index = matrix_free.get_cell_active_fe_index(range, dof_no); |
| 1356 |
else if (is_interior_face) |
| 1357 |
active_fe_index = |
| 1358 |
matrix_free.get_face_range_category(range, dof_no).first; |
| 1359 |
else |
| 1360 |
active_fe_index = |
| 1361 |
matrix_free.get_face_range_category(range, dof_no).second; |
| 1362 |
|
| 1363 |
const auto init_data = dealii::internal:: |
| 1364 |
extract_initialization_data<is_face, dim, Number, VectorizedArrayType>( |
| 1365 |
matrix_free, |
| 1366 |
dof_no, |
| 1367 |
first_selected_component, |
| 1368 |
quad_no, |
| 1369 |
fe_degree, |
| 1370 |
static_n_q_points, |
| 1371 |
active_fe_index, |
| 1372 |
numbers::invalid_unsigned_int /*active_quad_index*/, |
| 1373 |
numbers::invalid_unsigned_int /*face_type*/); |
| 1374 |
|
| 1375 |
return init_data.shape_info->dofs_per_component_on_cell == 0; |
| 1376 |
} |
| 1377 |
|
| 1378 |
|
| 1379 |
|
| 1380 |
/** |
| 1381 |
* Portable compute kernel for MatrixFreeTools::compute_diagonal(). |
| 1382 |
*/ |
| 1383 |
template <int dim, |
| 1384 |
int fe_degree, |
| 1385 |
int n_q_points_1d, |
| 1386 |
int n_components, |
| 1387 |
typename Number, |
| 1388 |
typename QuadOperation> |
| 1389 |
class ComputeDiagonalCellAction |
| 1390 |
{ |
| 1391 |
public: |
| 1392 |
ComputeDiagonalCellAction( |
| 1393 |
const QuadOperation &quad_operation, |
| 1394 |
const EvaluationFlags::EvaluationFlags evaluation_flags, |
| 1395 |
const EvaluationFlags::EvaluationFlags integration_flags) |
| 1396 |
: m_quad_operation(quad_operation) |
| 1397 |
, m_evaluation_flags(evaluation_flags) |
| 1398 |
, m_integration_flags(integration_flags) |
| 1399 |
{} |
| 1400 |
|
| 1401 |
KOKKOS_FUNCTION void |
| 1402 |
operator()(const typename Portable::MatrixFree<dim, Number>::Data *data, |
| 1403 |
const Portable::DeviceVector<Number> &, |
| 1404 |
Portable::DeviceVector<Number> &dst) const |
| 1405 |
{ |
| 1406 |
Portable:: |
| 1407 |
FEEvaluation<dim, fe_degree, n_q_points_1d, n_components, Number> |
| 1408 |
fe_eval(data); |
| 1409 |
const typename Portable::MatrixFree<dim, Number>::PrecomputedData |
| 1410 |
*gpu_data = data->precomputed_data; |
| 1411 |
const int cell = data->cell_index; |
| 1412 |
|
| 1413 |
constexpr int dofs_per_cell = decltype(fe_eval)::tensor_dofs_per_cell; |
| 1414 |
typename decltype(fe_eval)::value_type |
| 1415 |
diagonal[dofs_per_cell / n_components] = {}; |
| 1416 |
for (unsigned int i = 0; i < dofs_per_cell; ++i) |
| 1417 |
{ |
| 1418 |
const auto c = i % n_components; |
| 1419 |
|
| 1420 |
Kokkos::parallel_for( |
| 1421 |
Kokkos::TeamThreadRange(data->team_member, |
| 1422 |
dofs_per_cell / n_components), |
| 1423 |
[&](unsigned int j) { |
| 1424 |
typename decltype(fe_eval)::value_type val = {}; |
| 1425 |
|
| 1426 |
if constexpr (n_components == 1) |
| 1427 |
{ |
| 1428 |
val = (i == j) ? 1 : 0; |
| 1429 |
} |
| 1430 |
else |
| 1431 |
{ |
| 1432 |
val[c] = (i / n_components == j) ? 1 : 0; |
| 1433 |
} |
| 1434 |
|
| 1435 |
fe_eval.submit_dof_value(val, j); |
| 1436 |
}); |
| 1437 |
|
| 1438 |
data->team_member.team_barrier(); |
| 1439 |
|
| 1440 |
Portable::internal:: |
| 1441 |
resolve_hanging_nodes<dim, fe_degree, false, Number>( |
| 1442 |
data->team_member, |
| 1443 |
gpu_data->constraint_weights, |
| 1444 |
gpu_data->constraint_mask(cell * n_components + c), |
| 1445 |
Kokkos::subview(data->shared_data->values, Kokkos::ALL, c)); |
| 1446 |
|
| 1447 |
fe_eval.evaluate(m_evaluation_flags); |
| 1448 |
fe_eval.apply_for_each_quad_point(m_quad_operation); |
| 1449 |
fe_eval.integrate(m_integration_flags); |
| 1450 |
|
| 1451 |
Portable::internal:: |
| 1452 |
resolve_hanging_nodes<dim, fe_degree, true, Number>( |
| 1453 |
data->team_member, |
| 1454 |
gpu_data->constraint_weights, |
| 1455 |
gpu_data->constraint_mask(cell * n_components + c), |
| 1456 |
Kokkos::subview(data->shared_data->values, Kokkos::ALL, c)); |
| 1457 |
|
| 1458 |
Kokkos::single(Kokkos::PerTeam(data->team_member), [&] { |
| 1459 |
if constexpr (n_components == 1) |
| 1460 |
diagonal[i] = fe_eval.get_dof_value(i); |
| 1461 |
else |
| 1462 |
diagonal[i / n_components][i % n_components] = |
| 1463 |
fe_eval.get_dof_value(i / n_components)[i % n_components]; |
| 1464 |
}); |
| 1465 |
|
| 1466 |
data->team_member.team_barrier(); |
| 1467 |
} |
| 1468 |
|
| 1469 |
Kokkos::single(Kokkos::PerTeam(data->team_member), [&] { |
| 1470 |
for (unsigned int i = 0; i < dofs_per_cell / n_components; ++i) |
| 1471 |
fe_eval.submit_dof_value(diagonal[i], i); |
| 1472 |
}); |
| 1473 |
|
| 1474 |
data->team_member.team_barrier(); |
| 1475 |
|
| 1476 |
// We need to do the same as distribute_local_to_global but without |
| 1477 |
// constraints since we have already taken care of them earlier |
| 1478 |
if (gpu_data->use_coloring) |
| 1479 |
{ |
| 1480 |
Kokkos::parallel_for( |
| 1481 |
Kokkos::TeamThreadRange(data->team_member, dofs_per_cell), |
| 1482 |
[&](const int &i) { |
| 1483 |
dst[gpu_data->local_to_global(i, cell)] += |
| 1484 |
data->shared_data->values(i % (dofs_per_cell / n_components), |
| 1485 |
i / (dofs_per_cell / n_components)); |
| 1486 |
}); |
| 1487 |
} |
| 1488 |
else |
| 1489 |
{ |
| 1490 |
Kokkos::parallel_for( |
| 1491 |
Kokkos::TeamThreadRange(data->team_member, dofs_per_cell), |
| 1492 |
[&](const int &i) { |
| 1493 |
Kokkos::atomic_add(&dst[gpu_data->local_to_global(i, cell)], |
| 1494 |
data->shared_data->values( |
| 1495 |
i % (dofs_per_cell / n_components), |
| 1496 |
i / (dofs_per_cell / n_components))); |
| 1497 |
}); |
| 1498 |
} |
| 1499 |
}; |
| 1500 |
|
| 1501 |
static constexpr unsigned int n_local_dofs = |
| 1502 |
dealii::Utilities::pow(fe_degree + 1, dim) * n_components; |
| 1503 |
static constexpr unsigned int n_q_points = |
| 1504 |
dealii::Utilities::pow(n_q_points_1d, dim); |
| 1505 |
|
| 1506 |
private: |
| 1507 |
const QuadOperation m_quad_operation; |
| 1508 |
const EvaluationFlags::EvaluationFlags m_evaluation_flags; |
| 1509 |
const EvaluationFlags::EvaluationFlags m_integration_flags; |
| 1510 |
}; |
| 1511 |
|
| 1512 |
} // namespace internal |
| 1513 |
|
| 1514 |
|
| 1515 |
|
| 1516 |
template <int dim, |
| 1517 |
int fe_degree, |
| 1518 |
int n_q_points_1d, |
| 1519 |
int n_components, |
| 1520 |
typename Number, |
| 1521 |
typename MemorySpace, |
| 1522 |
typename QuadOperation> |
| 1523 |
void |
| 1524 |
compute_diagonal( |
| 1525 |
const Portable::MatrixFree<dim, Number> &matrix_free, |
| 1526 |
LinearAlgebra::distributed::Vector<Number, MemorySpace> &diagonal_global, |
| 1527 |
const QuadOperation &quad_operation, |
| 1528 |
EvaluationFlags::EvaluationFlags evaluation_flags, |
| 1529 |
EvaluationFlags::EvaluationFlags integration_flags, |
| 1530 |
const unsigned int dof_no, |
| 1531 |
const unsigned int quad_no, |
| 1532 |
const unsigned int first_selected_component, |
| 1533 |
const unsigned int first_vector_component) |
| 1534 |
{ |
| 1535 |
Assert(dof_no == 0, ExcNotImplemented()); |
| 1536 |
Assert(quad_no == 0, ExcNotImplemented()); |
| 1537 |
Assert(first_selected_component == 0, ExcNotImplemented()); |
| 1538 |
Assert(first_vector_component == 0, ExcNotImplemented()); |
| 1539 |
|
| 1540 |
matrix_free.initialize_dof_vector(diagonal_global); |
| 1541 |
|
| 1542 |
|
| 1543 |
internal::ComputeDiagonalCellAction<dim, |
| 1544 |
fe_degree, |
| 1545 |
n_q_points_1d, |
| 1546 |
n_components, |
| 1547 |
Number, |
| 1548 |
QuadOperation> |
| 1549 |
cell_action(quad_operation, evaluation_flags, integration_flags); |
| 1550 |
LinearAlgebra::distributed::Vector<Number, MemorySpace> dummy; |
| 1551 |
matrix_free.cell_loop(cell_action, dummy, diagonal_global); |
| 1552 |
|
| 1553 |
matrix_free.set_constrained_values(Number(1.), diagonal_global); |
| 1554 |
} |
| 1555 |
|
| 1556 |
template <int dim, |
| 1557 |
int fe_degree, |
| 1558 |
int n_q_points_1d, |
| 1559 |
int n_components, |
| 1560 |
typename Number, |
| 1561 |
typename VectorizedArrayType, |
| 1562 |
typename VectorType> |
| 1563 |
void |
| 1564 |
compute_diagonal( |
| 1565 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 1566 |
VectorType &diagonal_global, |
| 1567 |
const std::function<void(FEEvaluation<dim, |
| 1568 |
fe_degree, |
| 1569 |
n_q_points_1d, |
| 1570 |
n_components, |
| 1571 |
Number, |
| 1572 |
VectorizedArrayType> &)> |
| 1573 |
&cell_operation, |
| 1574 |
const unsigned int dof_no, |
| 1575 |
const unsigned int quad_no, |
| 1576 |
const unsigned int first_selected_component, |
| 1577 |
const unsigned int first_vector_component) |
| 1578 |
{ |
| 1579 |
compute_diagonal<dim, |
| 1580 |
fe_degree, |
| 1581 |
n_q_points_1d, |
| 1582 |
n_components, |
| 1583 |
Number, |
| 1584 |
VectorizedArrayType, |
| 1585 |
VectorType>(matrix_free, |
| 1586 |
diagonal_global, |
| 1587 |
cell_operation, |
| 1588 |
{}, |
| 1589 |
{}, |
| 1590 |
dof_no, |
| 1591 |
quad_no, |
| 1592 |
first_selected_component, |
| 1593 |
first_vector_component); |
| 1594 |
} |
| 1595 |
|
| 1596 |
template <typename CLASS, |
| 1597 |
int dim, |
| 1598 |
int fe_degree, |
| 1599 |
int n_q_points_1d, |
| 1600 |
int n_components, |
| 1601 |
typename Number, |
| 1602 |
typename VectorizedArrayType, |
| 1603 |
typename VectorType> |
| 1604 |
void |
| 1605 |
compute_diagonal( |
| 1606 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 1607 |
VectorType &diagonal_global, |
| 1608 |
void (CLASS::*cell_operation)(FEEvaluation<dim, |
| 1609 |
fe_degree, |
| 1610 |
n_q_points_1d, |
| 1611 |
n_components, |
| 1612 |
Number, |
| 1613 |
VectorizedArrayType> &) const, |
| 1614 |
const CLASS *owning_class, |
| 1615 |
const unsigned int dof_no, |
| 1616 |
const unsigned int quad_no, |
| 1617 |
const unsigned int first_selected_component, |
| 1618 |
const unsigned int first_vector_component) |
| 1619 |
{ |
| 1620 |
compute_diagonal<dim, |
| 1621 |
fe_degree, |
| 1622 |
n_q_points_1d, |
| 1623 |
n_components, |
| 1624 |
Number, |
| 1625 |
VectorizedArrayType, |
| 1626 |
VectorType>( |
| 1627 |
matrix_free, |
| 1628 |
diagonal_global, |
| 1629 |
[&](auto &phi) { (owning_class->*cell_operation)(phi); }, |
| 1630 |
dof_no, |
| 1631 |
quad_no, |
| 1632 |
first_selected_component, |
| 1633 |
first_vector_component); |
| 1634 |
} |
| 1635 |
|
| 1636 |
template <int dim, |
| 1637 |
int fe_degree, |
| 1638 |
int n_q_points_1d, |
| 1639 |
int n_components, |
| 1640 |
typename Number, |
| 1641 |
typename VectorizedArrayType, |
| 1642 |
typename VectorType> |
| 1643 |
void |
| 1644 |
compute_diagonal( |
| 1645 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 1646 |
VectorType &diagonal_global, |
| 1647 |
const std::function<void(FEEvaluation<dim, |
| 1648 |
fe_degree, |
| 1649 |
n_q_points_1d, |
| 1650 |
n_components, |
| 1651 |
Number, |
| 1652 |
VectorizedArrayType> &)> |
| 1653 |
&cell_operation, |
| 1654 |
const std::function<void(FEFaceEvaluation<dim, |
| 1655 |
fe_degree, |
| 1656 |
n_q_points_1d, |
| 1657 |
n_components, |
| 1658 |
Number, |
| 1659 |
VectorizedArrayType> &, |
| 1660 |
FEFaceEvaluation<dim, |
| 1661 |
fe_degree, |
| 1662 |
n_q_points_1d, |
| 1663 |
n_components, |
| 1664 |
Number, |
| 1665 |
VectorizedArrayType> &)> |
| 1666 |
&face_operation, |
| 1667 |
const std::function<void(FEFaceEvaluation<dim, |
| 1668 |
fe_degree, |
| 1669 |
n_q_points_1d, |
| 1670 |
n_components, |
| 1671 |
Number, |
| 1672 |
VectorizedArrayType> &)> |
| 1673 |
&boundary_operation, |
| 1674 |
const unsigned int dof_no, |
| 1675 |
const unsigned int quad_no, |
| 1676 |
const unsigned int first_selected_component, |
| 1677 |
const unsigned int first_vector_component) |
| 1678 |
{ |
| 1679 |
std::vector<typename dealii::internal::BlockVectorSelector< |
| 1680 |
VectorType, |
| 1681 |
IsBlockVector<VectorType>::value>::BaseVectorType *> |
| 1682 |
diagonal_global_components(n_components); |
| 1683 |
|
| 1684 |
for (unsigned int d = 0; d < n_components; ++d) |
| 1685 |
diagonal_global_components[d] = dealii::internal:: |
| 1686 |
BlockVectorSelector<VectorType, IsBlockVector<VectorType>::value>:: |
| 1687 |
get_vector_component(diagonal_global, d + first_vector_component); |
| 1688 |
|
| 1689 |
const auto &dof_info = matrix_free.get_dof_info(dof_no); |
| 1690 |
|
| 1691 |
if (dof_info.start_components.back() == 1) |
| 1692 |
for (unsigned int comp = 0; comp < n_components; ++comp) |
| 1693 |
{ |
| 1694 |
Assert(diagonal_global_components[comp] != nullptr, |
| 1695 |
ExcMessage("The finite element underlying this FEEvaluation " |
| 1696 |
"object is scalar, but you requested " + |
| 1697 |
std::to_string(n_components) + |
| 1698 |
" components via the template argument in " |
| 1699 |
"FEEvaluation. In that case, you must pass an " |
| 1700 |
"std::vector<VectorType> or a BlockVector to " + |
| 1701 |
"read_dof_values and distribute_local_to_global.")); |
| 1702 |
dealii::internal::check_vector_compatibility( |
| 1703 |
*diagonal_global_components[comp], matrix_free, dof_info); |
| 1704 |
} |
| 1705 |
else |
| 1706 |
{ |
| 1707 |
dealii::internal::check_vector_compatibility( |
| 1708 |
*diagonal_global_components[0], matrix_free, dof_info); |
| 1709 |
} |
| 1710 |
|
| 1711 |
using FEEvalType = FEEvaluation<dim, |
| 1712 |
fe_degree, |
| 1713 |
n_q_points_1d, |
| 1714 |
n_components, |
| 1715 |
Number, |
| 1716 |
VectorizedArrayType>; |
| 1717 |
|
| 1718 |
using FEFaceEvalType = FEFaceEvaluation<dim, |
| 1719 |
fe_degree, |
| 1720 |
n_q_points_1d, |
| 1721 |
n_components, |
| 1722 |
Number, |
| 1723 |
VectorizedArrayType>; |
| 1724 |
|
| 1725 |
internal::ComputeMatrixScratchData<dim, VectorizedArrayType, false> |
| 1726 |
data_cell; |
| 1727 |
|
| 1728 |
data_cell.dof_numbers = {dof_no}; |
| 1729 |
data_cell.quad_numbers = {quad_no}; |
| 1730 |
data_cell.n_components = {n_components}; |
| 1731 |
data_cell.first_selected_components = {first_selected_component}; |
| 1732 |
data_cell.batch_type = {0}; |
| 1733 |
|
| 1734 |
data_cell.op_create = |
| 1735 |
[&](const std::pair<unsigned int, unsigned int> &range) { |
| 1736 |
std::vector< |
| 1737 |
std::unique_ptr<FEEvaluationData<dim, VectorizedArrayType, false>>> |
| 1738 |
phi; |
| 1739 |
|
| 1740 |
if (!internal::is_fe_nothing<false>(matrix_free, |
| 1741 |
range, |
| 1742 |
dof_no, |
| 1743 |
quad_no, |
| 1744 |
first_selected_component, |
| 1745 |
fe_degree, |
| 1746 |
n_q_points_1d)) |
| 1747 |
phi.emplace_back(std::make_unique<FEEvalType>( |
| 1748 |
matrix_free, range, dof_no, quad_no, first_selected_component)); |
| 1749 |
|
| 1750 |
return phi; |
| 1751 |
}; |
| 1752 |
|
| 1753 |
data_cell.op_reinit = [](auto &phi, const unsigned batch) { |
| 1754 |
if (phi.size() == 1) |
| 1755 |
static_cast<FEEvalType &>(*phi[0]).reinit(batch); |
| 1756 |
}; |
| 1757 |
|
| 1758 |
if (cell_operation) |
| 1759 |
data_cell.op_compute = [&](auto &phi) { |
| 1760 |
cell_operation(static_cast<FEEvalType &>(*phi[0])); |
| 1761 |
}; |
| 1762 |
|
| 1763 |
internal::ComputeMatrixScratchData<dim, VectorizedArrayType, true> |
| 1764 |
data_face; |
| 1765 |
|
| 1766 |
data_face.dof_numbers = {dof_no, dof_no}; |
| 1767 |
data_face.quad_numbers = {quad_no, quad_no}; |
| 1768 |
data_face.n_components = {n_components, n_components}; |
| 1769 |
data_face.first_selected_components = {first_selected_component, |
| 1770 |
first_selected_component}; |
| 1771 |
data_face.batch_type = {1, 2}; |
| 1772 |
|
| 1773 |
data_face.op_create = |
| 1774 |
[&](const std::pair<unsigned int, unsigned int> &range) { |
| 1775 |
std::vector< |
| 1776 |
std::unique_ptr<FEEvaluationData<dim, VectorizedArrayType, true>>> |
| 1777 |
phi; |
| 1778 |
|
| 1779 |
if (!internal::is_fe_nothing<true>(matrix_free, |
| 1780 |
range, |
| 1781 |
dof_no, |
| 1782 |
quad_no, |
| 1783 |
first_selected_component, |
| 1784 |
fe_degree, |
| 1785 |
n_q_points_1d, |
| 1786 |
true) && |
| 1787 |
!internal::is_fe_nothing<true>(matrix_free, |
| 1788 |
range, |
| 1789 |
dof_no, |
| 1790 |
quad_no, |
| 1791 |
first_selected_component, |
| 1792 |
fe_degree, |
| 1793 |
n_q_points_1d, |
| 1794 |
false)) |
| 1795 |
{ |
| 1796 |
phi.emplace_back( |
| 1797 |
std::make_unique<FEFaceEvalType>(matrix_free, |
| 1798 |
range, |
| 1799 |
true, |
| 1800 |
dof_no, |
| 1801 |
quad_no, |
| 1802 |
first_selected_component)); |
| 1803 |
phi.emplace_back( |
| 1804 |
std::make_unique<FEFaceEvalType>(matrix_free, |
| 1805 |
range, |
| 1806 |
false, |
| 1807 |
dof_no, |
| 1808 |
quad_no, |
| 1809 |
first_selected_component)); |
| 1810 |
} |
| 1811 |
|
| 1812 |
return phi; |
| 1813 |
}; |
| 1814 |
|
| 1815 |
data_face.op_reinit = [](auto &phi, const unsigned batch) { |
| 1816 |
if (phi.size() == 2) |
| 1817 |
{ |
| 1818 |
static_cast<FEFaceEvalType &>(*phi[0]).reinit(batch); |
| 1819 |
static_cast<FEFaceEvalType &>(*phi[1]).reinit(batch); |
| 1820 |
} |
| 1821 |
}; |
| 1822 |
|
| 1823 |
if (face_operation) |
| 1824 |
data_face.op_compute = [&](auto &phi) { |
| 1825 |
face_operation(static_cast<FEFaceEvalType &>(*phi[0]), |
| 1826 |
static_cast<FEFaceEvalType &>(*phi[1])); |
| 1827 |
}; |
| 1828 |
|
| 1829 |
internal::ComputeMatrixScratchData<dim, VectorizedArrayType, true> |
| 1830 |
data_boundary; |
| 1831 |
|
| 1832 |
data_boundary.dof_numbers = {dof_no}; |
| 1833 |
data_boundary.quad_numbers = {quad_no}; |
| 1834 |
data_boundary.n_components = {n_components}; |
| 1835 |
data_boundary.first_selected_components = {first_selected_component}; |
| 1836 |
data_boundary.batch_type = {1}; |
| 1837 |
|
| 1838 |
data_boundary |
| 1839 |
.op_create = [&](const std::pair<unsigned int, unsigned int> &range) { |
| 1840 |
std::vector< |
| 1841 |
std::unique_ptr<FEEvaluationData<dim, VectorizedArrayType, true>>> |
| 1842 |
phi; |
| 1843 |
|
| 1844 |
if (!internal::is_fe_nothing<true>(matrix_free, |
| 1845 |
range, |
| 1846 |
dof_no, |
| 1847 |
quad_no, |
| 1848 |
first_selected_component, |
| 1849 |
fe_degree, |
| 1850 |
n_q_points_1d, |
| 1851 |
true)) |
| 1852 |
phi.emplace_back(std::make_unique<FEFaceEvalType>( |
| 1853 |
matrix_free, range, true, dof_no, quad_no, first_selected_component)); |
| 1854 |
|
| 1855 |
return phi; |
| 1856 |
}; |
| 1857 |
|
| 1858 |
data_boundary.op_reinit = [](auto &phi, const unsigned batch) { |
| 1859 |
if (phi.size() == 1) |
| 1860 |
static_cast<FEFaceEvalType &>(*phi[0]).reinit(batch); |
| 1861 |
}; |
| 1862 |
|
| 1863 |
if (boundary_operation) |
| 1864 |
data_boundary.op_compute = [&](auto &phi) { |
| 1865 |
boundary_operation(static_cast<FEFaceEvalType &>(*phi[0])); |
| 1866 |
}; |
| 1867 |
|
| 1868 |
internal::compute_diagonal(matrix_free, |
| 1869 |
data_cell, |
| 1870 |
data_face, |
| 1871 |
data_boundary, |
| 1872 |
diagonal_global, |
| 1873 |
diagonal_global_components); |
| 1874 |
} |
| 1875 |
|
| 1876 |
namespace internal |
| 1877 |
{ |
| 1878 |
template <int dim, |
| 1879 |
typename Number, |
| 1880 |
typename VectorizedArrayType, |
| 1881 |
typename VectorType, |
| 1882 |
typename VectorType2> |
| 1883 |
void |
| 1884 |
compute_diagonal( |
| 1885 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 1886 |
const internal::ComputeMatrixScratchData<dim, VectorizedArrayType, false> |
| 1887 |
&data_cell, |
| 1888 |
const internal::ComputeMatrixScratchData<dim, VectorizedArrayType, true> |
| 1889 |
&data_face, |
| 1890 |
const internal::ComputeMatrixScratchData<dim, VectorizedArrayType, true> |
| 1891 |
&data_boundary, |
| 1892 |
VectorType &diagonal_global, |
| 1893 |
std::vector<VectorType2 *> &diagonal_global_components) |
| 1894 |
{ |
| 1895 |
// TODO: can we remove diagonal_global_components as argument? |
| 1896 |
|
| 1897 |
int dummy = 0; |
| 1898 |
|
| 1899 |
using Helper = |
| 1900 |
internal::ComputeDiagonalHelper<dim, VectorizedArrayType, false>; |
| 1901 |
|
| 1902 |
using HelperFace = |
| 1903 |
internal::ComputeDiagonalHelper<dim, VectorizedArrayType, true>; |
| 1904 |
|
| 1905 |
Threads::ThreadLocalStorage<std::vector<Helper>> scratch_data; |
| 1906 |
Threads::ThreadLocalStorage<std::vector<HelperFace>> |
| 1907 |
scratch_data_internal; |
| 1908 |
Threads::ThreadLocalStorage<std::vector<HelperFace>> scratch_data_bc; |
| 1909 |
|
| 1910 |
const auto batch_operation = |
| 1911 |
[&](auto &data, |
| 1912 |
auto &scratch_data, |
| 1913 |
const std::pair<unsigned int, unsigned int> &range) { |
| 1914 |
if (!data.op_compute) |
| 1915 |
return; // nothing to do |
| 1916 |
|
| 1917 |
auto phi = data.op_create(range); |
| 1918 |
|
| 1919 |
const unsigned int n_blocks = phi.size(); |
| 1920 |
|
| 1921 |
auto &helpers = scratch_data.get(); |
| 1922 |
helpers.resize(n_blocks); |
| 1923 |
|
| 1924 |
for (unsigned int b = 0; b < n_blocks; ++b) |
| 1925 |
helpers[b].initialize(*phi[b], matrix_free, data.n_components[b]); |
| 1926 |
|
| 1927 |
for (unsigned int batch = range.first; batch < range.second; ++batch) |
| 1928 |
{ |
| 1929 |
data.op_reinit(phi, batch); |
| 1930 |
|
| 1931 |
for (unsigned int b = 0; b < n_blocks; ++b) |
| 1932 |
helpers[b].reinit(batch); |
| 1933 |
|
| 1934 |
if (n_blocks > 1) |
| 1935 |
{ |
| 1936 |
Assert(std::all_of(helpers.begin(), |
| 1937 |
helpers.end(), |
| 1938 |
[](const auto &helper) { |
| 1939 |
return helper.has_simple_constraints(); |
| 1940 |
}), |
| 1941 |
ExcNotImplemented()); |
| 1942 |
} |
| 1943 |
|
| 1944 |
for (unsigned int b = 0; b < n_blocks; ++b) |
| 1945 |
{ |
| 1946 |
for (unsigned int i = 0; |
| 1947 |
i < phi[b]->get_shape_info().dofs_per_component_on_cell * |
| 1948 |
data.n_components[b]; |
| 1949 |
++i) |
| 1950 |
{ |
| 1951 |
for (unsigned int bb = 0; bb < n_blocks; ++bb) |
| 1952 |
if (b == bb) |
| 1953 |
helpers[bb].prepare_basis_vector(i); |
| 1954 |
else |
| 1955 |
helpers[bb].zero_basis_vector(); |
| 1956 |
|
| 1957 |
data.op_compute(phi); |
| 1958 |
helpers[b].submit(); |
| 1959 |
} |
| 1960 |
|
| 1961 |
helpers[b].distribute_local_to_global( |
| 1962 |
diagonal_global_components); |
| 1963 |
} |
| 1964 |
} |
| 1965 |
}; |
| 1966 |
|
| 1967 |
const auto cell_operation_wrapped = |
| 1968 |
[&](const auto &, auto &, const auto &, const auto range) { |
| 1969 |
batch_operation(data_cell, scratch_data, range); |
| 1970 |
}; |
| 1971 |
|
| 1972 |
const auto face_operation_wrapped = |
| 1973 |
[&](const auto &, auto &, const auto &, const auto range) { |
| 1974 |
batch_operation(data_face, scratch_data_internal, range); |
| 1975 |
}; |
| 1976 |
|
| 1977 |
const auto boundary_operation_wrapped = |
| 1978 |
[&](const auto &, auto &, const auto &, const auto range) { |
| 1979 |
batch_operation(data_boundary, scratch_data_bc, range); |
| 1980 |
}; |
| 1981 |
|
| 1982 |
if (data_face.op_compute || data_boundary.op_compute) |
| 1983 |
matrix_free.template loop<VectorType, int>(cell_operation_wrapped, |
| 1984 |
face_operation_wrapped, |
| 1985 |
boundary_operation_wrapped, |
| 1986 |
diagonal_global, |
| 1987 |
dummy, |
| 1988 |
false); |
| 1989 |
else |
| 1990 |
matrix_free.template cell_loop<VectorType, int>(cell_operation_wrapped, |
| 1991 |
diagonal_global, |
| 1992 |
dummy, |
| 1993 |
false); |
| 1994 |
} |
| 1995 |
} // namespace internal |
| 1996 |
|
| 1997 |
template <typename CLASS, |
| 1998 |
int dim, |
| 1999 |
int fe_degree, |
| 2000 |
int n_q_points_1d, |
| 2001 |
int n_components, |
| 2002 |
typename Number, |
| 2003 |
typename VectorizedArrayType, |
| 2004 |
typename VectorType> |
| 2005 |
void |
| 2006 |
compute_diagonal( |
| 2007 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 2008 |
VectorType &diagonal_global, |
| 2009 |
void (CLASS::*cell_operation)(FEEvaluation<dim, |
| 2010 |
fe_degree, |
| 2011 |
n_q_points_1d, |
| 2012 |
n_components, |
| 2013 |
Number, |
| 2014 |
VectorizedArrayType> &) const, |
| 2015 |
void (CLASS::*face_operation)(FEFaceEvaluation<dim, |
| 2016 |
fe_degree, |
| 2017 |
n_q_points_1d, |
| 2018 |
n_components, |
| 2019 |
Number, |
| 2020 |
VectorizedArrayType> &, |
| 2021 |
FEFaceEvaluation<dim, |
| 2022 |
fe_degree, |
| 2023 |
n_q_points_1d, |
| 2024 |
n_components, |
| 2025 |
Number, |
| 2026 |
VectorizedArrayType> &) |
| 2027 |
const, |
| 2028 |
void (CLASS::*boundary_operation)(FEFaceEvaluation<dim, |
| 2029 |
fe_degree, |
| 2030 |
n_q_points_1d, |
| 2031 |
n_components, |
| 2032 |
Number, |
| 2033 |
VectorizedArrayType> &) |
| 2034 |
const, |
| 2035 |
const CLASS *owning_class, |
| 2036 |
const unsigned int dof_no, |
| 2037 |
const unsigned int quad_no, |
| 2038 |
const unsigned int first_selected_component, |
| 2039 |
const unsigned int first_vector_component) |
| 2040 |
{ |
| 2041 |
compute_diagonal<dim, |
| 2042 |
fe_degree, |
| 2043 |
n_q_points_1d, |
| 2044 |
n_components, |
| 2045 |
Number, |
| 2046 |
VectorizedArrayType, |
| 2047 |
VectorType>( |
| 2048 |
matrix_free, |
| 2049 |
diagonal_global, |
| 2050 |
[&](auto &phi) { (owning_class->*cell_operation)(phi); }, |
| 2051 |
[&](auto &phi_m, auto &phi_p) { |
| 2052 |
(owning_class->*face_operation)(phi_m, phi_p); |
| 2053 |
}, |
| 2054 |
[&](auto &phi) { (owning_class->*boundary_operation)(phi); }, |
| 2055 |
dof_no, |
| 2056 |
quad_no, |
| 2057 |
first_selected_component, |
| 2058 |
first_vector_component); |
| 2059 |
} |
| 2060 |
|
| 2061 |
namespace internal |
| 2062 |
{ |
| 2063 |
/** |
| 2064 |
* If value type of matrix and constrains equals, return a reference |
| 2065 |
* to the given AffineConstraint instance. |
| 2066 |
*/ |
| 2067 |
template < |
| 2068 |
typename MatrixType, |
| 2069 |
typename Number, |
| 2070 |
std::enable_if_t<std::is_same_v< |
| 2071 |
std::remove_const_t< |
| 2072 |
std::remove_reference_t<typename MatrixType::value_type>>, |
| 2073 |
std::remove_const_t<std::remove_reference_t<Number>>>> * = nullptr> |
| 2074 |
const AffineConstraints<typename MatrixType::value_type> & |
| 2075 |
create_new_affine_constraints_if_needed( |
| 2076 |
const MatrixType &, |
| 2077 |
const AffineConstraints<Number> &constraints, |
| 2078 |
std::unique_ptr<AffineConstraints<typename MatrixType::value_type>> &) |
| 2079 |
{ |
| 2080 |
return constraints; |
| 2081 |
} |
| 2082 |
|
| 2083 |
/** |
| 2084 |
* If value type of matrix and constrains do not equal, a new |
| 2085 |
* AffineConstraint instance with the value type of the matrix is |
| 2086 |
* created and a reference to it is returned. |
| 2087 |
*/ |
| 2088 |
template < |
| 2089 |
typename MatrixType, |
| 2090 |
typename Number, |
| 2091 |
std::enable_if_t<!std::is_same_v< |
| 2092 |
std::remove_const_t< |
| 2093 |
std::remove_reference_t<typename MatrixType::value_type>>, |
| 2094 |
std::remove_const_t<std::remove_reference_t<Number>>>> * = nullptr> |
| 2095 |
const AffineConstraints<typename MatrixType::value_type> & |
| 2096 |
create_new_affine_constraints_if_needed( |
| 2097 |
const MatrixType &, |
| 2098 |
const AffineConstraints<Number> &constraints, |
| 2099 |
std::unique_ptr<AffineConstraints<typename MatrixType::value_type>> |
| 2100 |
&new_constraints) |
| 2101 |
{ |
| 2102 |
new_constraints = |
| 2103 |
std::make_unique<AffineConstraints<typename MatrixType::value_type>>(); |
| 2104 |
new_constraints->copy_from(constraints); |
| 2105 |
|
| 2106 |
return *new_constraints; |
| 2107 |
} |
| 2108 |
} // namespace internal |
| 2109 |
|
| 2110 |
template <int dim, |
| 2111 |
int fe_degree, |
| 2112 |
int n_q_points_1d, |
| 2113 |
int n_components, |
| 2114 |
typename Number, |
| 2115 |
typename VectorizedArrayType, |
| 2116 |
typename MatrixType> |
| 2117 |
void |
| 2118 |
compute_matrix( |
| 2119 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 2120 |
const AffineConstraints<Number> &constraints_in, |
| 2121 |
MatrixType &matrix, |
| 2122 |
const std::function<void(FEEvaluation<dim, |
| 2123 |
fe_degree, |
| 2124 |
n_q_points_1d, |
| 2125 |
n_components, |
| 2126 |
Number, |
| 2127 |
VectorizedArrayType> &)> |
| 2128 |
&cell_operation, |
| 2129 |
const unsigned int dof_no, |
| 2130 |
const unsigned int quad_no, |
| 2131 |
const unsigned int first_selected_component) |
| 2132 |
{ |
| 2133 |
compute_matrix<dim, |
| 2134 |
fe_degree, |
| 2135 |
n_q_points_1d, |
| 2136 |
n_components, |
| 2137 |
Number, |
| 2138 |
VectorizedArrayType, |
| 2139 |
MatrixType>(matrix_free, |
| 2140 |
constraints_in, |
| 2141 |
matrix, |
| 2142 |
cell_operation, |
| 2143 |
{}, |
| 2144 |
{}, |
| 2145 |
dof_no, |
| 2146 |
quad_no, |
| 2147 |
first_selected_component); |
| 2148 |
} |
| 2149 |
|
| 2150 |
template <typename CLASS, |
| 2151 |
int dim, |
| 2152 |
int fe_degree, |
| 2153 |
int n_q_points_1d, |
| 2154 |
int n_components, |
| 2155 |
typename Number, |
| 2156 |
typename VectorizedArrayType, |
| 2157 |
typename MatrixType> |
| 2158 |
void |
| 2159 |
compute_matrix( |
| 2160 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 2161 |
const AffineConstraints<Number> &constraints, |
| 2162 |
MatrixType &matrix, |
| 2163 |
void (CLASS::*cell_operation)(FEEvaluation<dim, |
| 2164 |
fe_degree, |
| 2165 |
n_q_points_1d, |
| 2166 |
n_components, |
| 2167 |
Number, |
| 2168 |
VectorizedArrayType> &) const, |
| 2169 |
const CLASS *owning_class, |
| 2170 |
const unsigned int dof_no, |
| 2171 |
const unsigned int quad_no, |
| 2172 |
const unsigned int first_selected_component) |
| 2173 |
{ |
| 2174 |
compute_matrix<dim, |
| 2175 |
fe_degree, |
| 2176 |
n_q_points_1d, |
| 2177 |
n_components, |
| 2178 |
Number, |
| 2179 |
VectorizedArrayType, |
| 2180 |
MatrixType>( |
| 2181 |
matrix_free, |
| 2182 |
constraints, |
| 2183 |
matrix, |
| 2184 |
[&](auto &phi) { (owning_class->*cell_operation)(phi); }, |
| 2185 |
dof_no, |
| 2186 |
quad_no, |
| 2187 |
first_selected_component); |
| 2188 |
} |
| 2189 |
|
| 2190 |
template <int dim, |
| 2191 |
int fe_degree, |
| 2192 |
int n_q_points_1d, |
| 2193 |
int n_components, |
| 2194 |
typename Number, |
| 2195 |
typename VectorizedArrayType, |
| 2196 |
typename MatrixType> |
| 2197 |
void |
| 2198 |
compute_matrix( |
| 2199 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 2200 |
const AffineConstraints<Number> &constraints_in, |
| 2201 |
MatrixType &matrix, |
| 2202 |
const std::function<void(FEEvaluation<dim, |
| 2203 |
fe_degree, |
| 2204 |
n_q_points_1d, |
| 2205 |
n_components, |
| 2206 |
Number, |
| 2207 |
VectorizedArrayType> &)> |
| 2208 |
&cell_operation, |
| 2209 |
const std::function<void(FEFaceEvaluation<dim, |
| 2210 |
fe_degree, |
| 2211 |
n_q_points_1d, |
| 2212 |
n_components, |
| 2213 |
Number, |
| 2214 |
VectorizedArrayType> &, |
| 2215 |
FEFaceEvaluation<dim, |
| 2216 |
fe_degree, |
| 2217 |
n_q_points_1d, |
| 2218 |
n_components, |
| 2219 |
Number, |
| 2220 |
VectorizedArrayType> &)> |
| 2221 |
&face_operation, |
| 2222 |
const std::function<void(FEFaceEvaluation<dim, |
| 2223 |
fe_degree, |
| 2224 |
n_q_points_1d, |
| 2225 |
n_components, |
| 2226 |
Number, |
| 2227 |
VectorizedArrayType> &)> |
| 2228 |
&boundary_operation, |
| 2229 |
const unsigned int dof_no, |
| 2230 |
const unsigned int quad_no, |
| 2231 |
const unsigned int first_selected_component) |
| 2232 |
{ |
| 2233 |
using FEEvalType = FEEvaluation<dim, |
| 2234 |
fe_degree, |
| 2235 |
n_q_points_1d, |
| 2236 |
n_components, |
| 2237 |
Number, |
| 2238 |
VectorizedArrayType>; |
| 2239 |
|
| 2240 |
using FEFaceEvalType = FEFaceEvaluation<dim, |
| 2241 |
fe_degree, |
| 2242 |
n_q_points_1d, |
| 2243 |
n_components, |
| 2244 |
Number, |
| 2245 |
VectorizedArrayType>; |
| 2246 |
|
| 2247 |
internal::ComputeMatrixScratchData<dim, VectorizedArrayType, false> |
| 2248 |
data_cell; |
| 2249 |
|
| 2250 |
data_cell.dof_numbers = {dof_no}; |
| 2251 |
data_cell.quad_numbers = {quad_no}; |
| 2252 |
data_cell.n_components = {n_components}; |
| 2253 |
data_cell.first_selected_components = {first_selected_component}; |
| 2254 |
data_cell.batch_type = {0}; |
| 2255 |
|
| 2256 |
data_cell.op_create = |
| 2257 |
[&](const std::pair<unsigned int, unsigned int> &range) { |
| 2258 |
std::vector< |
| 2259 |
std::unique_ptr<FEEvaluationData<dim, VectorizedArrayType, false>>> |
| 2260 |
phi; |
| 2261 |
|
| 2262 |
if (!internal::is_fe_nothing<false>(matrix_free, |
| 2263 |
range, |
| 2264 |
dof_no, |
| 2265 |
quad_no, |
| 2266 |
first_selected_component, |
| 2267 |
fe_degree, |
| 2268 |
n_q_points_1d)) |
| 2269 |
phi.emplace_back(std::make_unique<FEEvalType>( |
| 2270 |
matrix_free, range, dof_no, quad_no, first_selected_component)); |
| 2271 |
|
| 2272 |
return phi; |
| 2273 |
}; |
| 2274 |
|
| 2275 |
data_cell.op_reinit = [](auto &phi, const unsigned batch) { |
| 2276 |
if (phi.size() == 1) |
| 2277 |
static_cast<FEEvalType &>(*phi[0]).reinit(batch); |
| 2278 |
}; |
| 2279 |
|
| 2280 |
if (cell_operation) |
| 2281 |
data_cell.op_compute = [&](auto &phi) { |
| 2282 |
cell_operation(static_cast<FEEvalType &>(*phi[0])); |
| 2283 |
}; |
| 2284 |
|
| 2285 |
internal::ComputeMatrixScratchData<dim, VectorizedArrayType, true> |
| 2286 |
data_face; |
| 2287 |
|
| 2288 |
data_face.dof_numbers = {dof_no, dof_no}; |
| 2289 |
data_face.quad_numbers = {quad_no, quad_no}; |
| 2290 |
data_face.n_components = {n_components, n_components}; |
| 2291 |
data_face.first_selected_components = {first_selected_component, |
| 2292 |
first_selected_component}; |
| 2293 |
data_face.batch_type = {1, 2}; |
| 2294 |
|
| 2295 |
data_face.op_create = |
| 2296 |
[&](const std::pair<unsigned int, unsigned int> &range) { |
| 2297 |
std::vector< |
| 2298 |
std::unique_ptr<FEEvaluationData<dim, VectorizedArrayType, true>>> |
| 2299 |
phi; |
| 2300 |
|
| 2301 |
if (!internal::is_fe_nothing<true>(matrix_free, |
| 2302 |
range, |
| 2303 |
dof_no, |
| 2304 |
quad_no, |
| 2305 |
first_selected_component, |
| 2306 |
fe_degree, |
| 2307 |
n_q_points_1d, |
| 2308 |
true) && |
| 2309 |
!internal::is_fe_nothing<true>(matrix_free, |
| 2310 |
range, |
| 2311 |
dof_no, |
| 2312 |
quad_no, |
| 2313 |
first_selected_component, |
| 2314 |
fe_degree, |
| 2315 |
n_q_points_1d, |
| 2316 |
false)) |
| 2317 |
{ |
| 2318 |
phi.emplace_back( |
| 2319 |
std::make_unique<FEFaceEvalType>(matrix_free, |
| 2320 |
range, |
| 2321 |
true, |
| 2322 |
dof_no, |
| 2323 |
quad_no, |
| 2324 |
first_selected_component)); |
| 2325 |
phi.emplace_back( |
| 2326 |
std::make_unique<FEFaceEvalType>(matrix_free, |
| 2327 |
range, |
| 2328 |
false, |
| 2329 |
dof_no, |
| 2330 |
quad_no, |
| 2331 |
first_selected_component)); |
| 2332 |
} |
| 2333 |
|
| 2334 |
return phi; |
| 2335 |
}; |
| 2336 |
|
| 2337 |
data_face.op_reinit = [](auto &phi, const unsigned batch) { |
| 2338 |
if (phi.size() == 2) |
| 2339 |
{ |
| 2340 |
static_cast<FEFaceEvalType &>(*phi[0]).reinit(batch); |
| 2341 |
static_cast<FEFaceEvalType &>(*phi[1]).reinit(batch); |
| 2342 |
} |
| 2343 |
}; |
| 2344 |
|
| 2345 |
if (face_operation) |
| 2346 |
data_face.op_compute = [&](auto &phi) { |
| 2347 |
face_operation(static_cast<FEFaceEvalType &>(*phi[0]), |
| 2348 |
static_cast<FEFaceEvalType &>(*phi[1])); |
| 2349 |
}; |
| 2350 |
|
| 2351 |
internal::ComputeMatrixScratchData<dim, VectorizedArrayType, true> |
| 2352 |
data_boundary; |
| 2353 |
|
| 2354 |
data_boundary.dof_numbers = {dof_no}; |
| 2355 |
data_boundary.quad_numbers = {quad_no}; |
| 2356 |
data_boundary.n_components = {n_components}; |
| 2357 |
data_boundary.first_selected_components = {first_selected_component}; |
| 2358 |
data_boundary.batch_type = {1}; |
| 2359 |
|
| 2360 |
data_boundary |
| 2361 |
.op_create = [&](const std::pair<unsigned int, unsigned int> &range) { |
| 2362 |
std::vector< |
| 2363 |
std::unique_ptr<FEEvaluationData<dim, VectorizedArrayType, true>>> |
| 2364 |
phi; |
| 2365 |
|
| 2366 |
if (!internal::is_fe_nothing<true>(matrix_free, |
| 2367 |
range, |
| 2368 |
dof_no, |
| 2369 |
quad_no, |
| 2370 |
first_selected_component, |
| 2371 |
fe_degree, |
| 2372 |
n_q_points_1d, |
| 2373 |
true)) |
| 2374 |
phi.emplace_back(std::make_unique<FEFaceEvalType>( |
| 2375 |
matrix_free, range, true, dof_no, quad_no, first_selected_component)); |
| 2376 |
|
| 2377 |
return phi; |
| 2378 |
}; |
| 2379 |
|
| 2380 |
data_boundary.op_reinit = [](auto &phi, const unsigned batch) { |
| 2381 |
if (phi.size() == 1) |
| 2382 |
static_cast<FEFaceEvalType &>(*phi[0]).reinit(batch); |
| 2383 |
}; |
| 2384 |
|
| 2385 |
if (boundary_operation) |
| 2386 |
data_boundary.op_compute = [&](auto &phi) { |
| 2387 |
boundary_operation(static_cast<FEFaceEvalType &>(*phi[0])); |
| 2388 |
}; |
| 2389 |
|
| 2390 |
internal::compute_matrix( |
| 2391 |
matrix_free, constraints_in, data_cell, data_face, data_boundary, matrix); |
| 2392 |
} |
| 2393 |
|
| 2394 |
namespace internal |
| 2395 |
{ |
| 2396 |
template <int dim, |
| 2397 |
typename Number, |
| 2398 |
typename VectorizedArrayType, |
| 2399 |
typename MatrixType> |
| 2400 |
void |
| 2401 |
compute_matrix( |
| 2402 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 2403 |
const AffineConstraints<Number> &constraints_in, |
| 2404 |
const internal::ComputeMatrixScratchData<dim, VectorizedArrayType, false> |
| 2405 |
&data_cell, |
| 2406 |
const internal::ComputeMatrixScratchData<dim, VectorizedArrayType, true> |
| 2407 |
&data_face, |
| 2408 |
const internal::ComputeMatrixScratchData<dim, VectorizedArrayType, true> |
| 2409 |
&data_boundary, |
| 2410 |
MatrixType &matrix) |
| 2411 |
{ |
| 2412 |
std::unique_ptr<AffineConstraints<typename MatrixType::value_type>> |
| 2413 |
constraints_for_matrix; |
| 2414 |
const AffineConstraints<typename MatrixType::value_type> &constraints = |
| 2415 |
internal::create_new_affine_constraints_if_needed( |
| 2416 |
matrix, constraints_in, constraints_for_matrix); |
| 2417 |
|
| 2418 |
const auto batch_operation = |
| 2419 |
[&matrix_free, &constraints, &matrix]( |
| 2420 |
auto &data, const std::pair<unsigned int, unsigned int> &range) { |
| 2421 |
if (!data.op_compute) |
| 2422 |
return; // nothing to do |
| 2423 |
|
| 2424 |
auto phi = data.op_create(range); |
| 2425 |
|
| 2426 |
const unsigned int n_blocks = phi.size(); |
| 2427 |
|
| 2428 |
if (n_blocks == 0) |
| 2429 |
return; |
| 2430 |
|
| 2431 |
Table<1, unsigned int> dofs_per_cell(n_blocks); |
| 2432 |
|
| 2433 |
Table<1, std::vector<types::global_dof_index>> dof_indices(n_blocks); |
| 2434 |
Table<2, std::vector<types::global_dof_index>> dof_indices_mf( |
| 2435 |
n_blocks, VectorizedArrayType::size()); |
| 2436 |
Table<1, std::vector<unsigned int>> lexicographic_numbering(n_blocks); |
| 2437 |
Table<2, |
| 2438 |
std::array<FullMatrix<typename MatrixType::value_type>, |
| 2439 |
VectorizedArrayType::size()>> |
| 2440 |
matrices(n_blocks, n_blocks); |
| 2441 |
|
| 2442 |
for (unsigned int b = 0; b < n_blocks; ++b) |
| 2443 |
{ |
| 2444 |
const auto &fe = matrix_free.get_dof_handler(data.dof_numbers[b]) |
| 2445 |
.get_fe(phi[b]->get_active_fe_index()); |
| 2446 |
|
| 2447 |
const auto component_base = |
| 2448 |
matrix_free.get_dof_info(data.dof_numbers[b]) |
| 2449 |
.component_to_base_index[data.first_selected_components[b]]; |
| 2450 |
const auto component_in_base = |
| 2451 |
data.first_selected_components[b] - |
| 2452 |
matrix_free.get_dof_info(data.dof_numbers[b]) |
| 2453 |
.start_components[component_base]; |
| 2454 |
|
| 2455 |
const auto &shape_info = matrix_free.get_shape_info( |
| 2456 |
data.dof_numbers[b], |
| 2457 |
data.quad_numbers[b], |
| 2458 |
component_base, |
| 2459 |
phi[b]->get_active_fe_index(), |
| 2460 |
phi[b]->get_active_quadrature_index()); |
| 2461 |
|
| 2462 |
dofs_per_cell[b] = |
| 2463 |
shape_info.dofs_per_component_on_cell * data.n_components[b]; |
| 2464 |
|
| 2465 |
dof_indices[b].resize(fe.n_dofs_per_cell()); |
| 2466 |
|
| 2467 |
for (unsigned int v = 0; v < VectorizedArrayType::size(); ++v) |
| 2468 |
dof_indices_mf[b][v].resize(dofs_per_cell[b]); |
| 2469 |
|
| 2470 |
lexicographic_numbering[b].insert( |
| 2471 |
lexicographic_numbering[b].begin(), |
| 2472 |
shape_info.lexicographic_numbering.begin() + |
| 2473 |
component_in_base * shape_info.dofs_per_component_on_cell, |
| 2474 |
shape_info.lexicographic_numbering.begin() + |
| 2475 |
(component_in_base + data.n_components[b]) * |
| 2476 |
shape_info.dofs_per_component_on_cell); |
| 2477 |
} |
| 2478 |
|
| 2479 |
for (unsigned int bj = 0; bj < n_blocks; ++bj) |
| 2480 |
for (unsigned int bi = 0; bi < n_blocks; ++bi) |
| 2481 |
std::fill_n(matrices[bi][bj].begin(), |
| 2482 |
VectorizedArrayType::size(), |
| 2483 |
FullMatrix<typename MatrixType::value_type>( |
| 2484 |
dofs_per_cell[bi], dofs_per_cell[bj])); |
| 2485 |
|
| 2486 |
for (auto batch = range.first; batch < range.second; ++batch) |
| 2487 |
{ |
| 2488 |
data.op_reinit(phi, batch); |
| 2489 |
|
| 2490 |
const unsigned int n_filled_lanes = |
| 2491 |
data.is_face ? |
| 2492 |
matrix_free.n_active_entries_per_face_batch(batch) : |
| 2493 |
matrix_free.n_active_entries_per_cell_batch(batch); |
| 2494 |
|
| 2495 |
for (unsigned int v = 0; v < n_filled_lanes; ++v) |
| 2496 |
for (unsigned int b = 0; b < n_blocks; ++b) |
| 2497 |
{ |
| 2498 |
unsigned int const cell_index = |
| 2499 |
(data.batch_type[b] == 0) ? |
| 2500 |
(batch * VectorizedArrayType::size() + v) : |
| 2501 |
((data.batch_type[b] == 1) ? |
| 2502 |
matrix_free.get_face_info(batch).cells_interior[v] : |
| 2503 |
matrix_free.get_face_info(batch).cells_exterior[v]); |
| 2504 |
|
| 2505 |
const auto cell_iterator = matrix_free.get_cell_iterator( |
| 2506 |
cell_index / VectorizedArrayType::size(), |
| 2507 |
cell_index % VectorizedArrayType::size(), |
| 2508 |
data.dof_numbers[b]); |
| 2509 |
|
| 2510 |
if (matrix_free.get_mg_level() != |
| 2511 |
numbers::invalid_unsigned_int) |
| 2512 |
cell_iterator->get_mg_dof_indices(dof_indices[b]); |
| 2513 |
else |
| 2514 |
cell_iterator->get_dof_indices(dof_indices[b]); |
| 2515 |
|
| 2516 |
for (unsigned int j = 0; j < dofs_per_cell[b]; ++j) |
| 2517 |
dof_indices_mf[b][v][j] = |
| 2518 |
dof_indices[b][lexicographic_numbering[b][j]]; |
| 2519 |
} |
| 2520 |
|
| 2521 |
for (unsigned int bj = 0; bj < n_blocks; ++bj) |
| 2522 |
{ |
| 2523 |
for (unsigned int j = 0; j < dofs_per_cell[bj]; ++j) |
| 2524 |
{ |
| 2525 |
for (unsigned int bi = 0; bi < n_blocks; ++bi) |
| 2526 |
for (unsigned int i = 0; i < dofs_per_cell[bi]; ++i) |
| 2527 |
phi[bi]->begin_dof_values()[i] = |
| 2528 |
(bj == bi) ? static_cast<Number>(i == j) : 0.0; |
| 2529 |
|
| 2530 |
data.op_compute(phi); |
| 2531 |
|
| 2532 |
for (unsigned int bi = 0; bi < n_blocks; ++bi) |
| 2533 |
for (unsigned int i = 0; i < dofs_per_cell[bi]; ++i) |
| 2534 |
for (unsigned int v = 0; v < n_filled_lanes; ++v) |
| 2535 |
matrices[bi][bj][v](i, j) = |
| 2536 |
phi[bi]->begin_dof_values()[i][v]; |
| 2537 |
} |
| 2538 |
|
| 2539 |
for (unsigned int v = 0; v < n_filled_lanes; ++v) |
| 2540 |
for (unsigned int bi = 0; bi < n_blocks; ++bi) |
| 2541 |
if (bi == bj) |
| 2542 |
// specialization for blocks on the diagonal |
| 2543 |
// to writing into diagonal elements of the |
| 2544 |
// matrix if the corresponding degree of freedom |
| 2545 |
// is constrained, see also the documentation |
| 2546 |
// of AffineConstraints::distribute_local_to_global() |
| 2547 |
constraints.distribute_local_to_global( |
| 2548 |
matrices[bi][bi][v], dof_indices_mf[bi][v], matrix); |
| 2549 |
else |
| 2550 |
constraints.distribute_local_to_global( |
| 2551 |
matrices[bi][bj][v], |
| 2552 |
dof_indices_mf[bi][v], |
| 2553 |
dof_indices_mf[bj][v], |
| 2554 |
matrix); |
| 2555 |
} |
| 2556 |
} |
| 2557 |
}; |
| 2558 |
|
| 2559 |
const auto cell_operation_wrapped = |
| 2560 |
[&](const auto &, auto &, const auto &, const auto range) { |
| 2561 |
batch_operation(data_cell, range); |
| 2562 |
}; |
| 2563 |
|
| 2564 |
const auto face_operation_wrapped = |
| 2565 |
[&](const auto &, auto &, const auto &, const auto range) { |
| 2566 |
batch_operation(data_face, range); |
| 2567 |
}; |
| 2568 |
|
| 2569 |
const auto boundary_operation_wrapped = |
| 2570 |
[&](const auto &, auto &, const auto &, const auto range) { |
| 2571 |
batch_operation(data_boundary, range); |
| 2572 |
}; |
| 2573 |
|
| 2574 |
if (data_face.op_compute || data_boundary.op_compute) |
| 2575 |
{ |
| 2576 |
matrix_free.template loop<MatrixType, MatrixType>( |
| 2577 |
cell_operation_wrapped, |
| 2578 |
face_operation_wrapped, |
| 2579 |
boundary_operation_wrapped, |
| 2580 |
matrix, |
| 2581 |
matrix); |
| 2582 |
} |
| 2583 |
else |
| 2584 |
matrix_free.template cell_loop<MatrixType, MatrixType>( |
| 2585 |
cell_operation_wrapped, matrix, matrix); |
| 2586 |
|
| 2587 |
matrix.compress(VectorOperation::add); |
| 2588 |
} |
| 2589 |
} // namespace internal |
| 2590 |
|
| 2591 |
template <typename CLASS, |
| 2592 |
int dim, |
| 2593 |
int fe_degree, |
| 2594 |
int n_q_points_1d, |
| 2595 |
int n_components, |
| 2596 |
typename Number, |
| 2597 |
typename VectorizedArrayType, |
| 2598 |
typename MatrixType> |
| 2599 |
void |
| 2600 |
compute_matrix( |
| 2601 |
const MatrixFree<dim, Number, VectorizedArrayType> &matrix_free, |
| 2602 |
const AffineConstraints<Number> &constraints, |
| 2603 |
MatrixType &matrix, |
| 2604 |
void (CLASS::*cell_operation)(FEEvaluation<dim, |
| 2605 |
fe_degree, |
| 2606 |
n_q_points_1d, |
| 2607 |
n_components, |
| 2608 |
Number, |
| 2609 |
VectorizedArrayType> &) const, |
| 2610 |
void (CLASS::*face_operation)(FEFaceEvaluation<dim, |
| 2611 |
fe_degree, |
| 2612 |
n_q_points_1d, |
| 2613 |
n_components, |
| 2614 |
Number, |
| 2615 |
VectorizedArrayType> &, |
| 2616 |
FEFaceEvaluation<dim, |
| 2617 |
fe_degree, |
| 2618 |
n_q_points_1d, |
| 2619 |
n_components, |
| 2620 |
Number, |
| 2621 |
VectorizedArrayType> &) |
| 2622 |
const, |
| 2623 |
void (CLASS::*boundary_operation)(FEFaceEvaluation<dim, |
| 2624 |
fe_degree, |
| 2625 |
n_q_points_1d, |
| 2626 |
n_components, |
| 2627 |
Number, |
| 2628 |
VectorizedArrayType> &) |
| 2629 |
const, |
| 2630 |
const CLASS *owning_class, |
| 2631 |
const unsigned int dof_no, |
| 2632 |
const unsigned int quad_no, |
| 2633 |
const unsigned int first_selected_component) |
| 2634 |
{ |
| 2635 |
compute_matrix<dim, |
| 2636 |
fe_degree, |
| 2637 |
n_q_points_1d, |
| 2638 |
n_components, |
| 2639 |
Number, |
| 2640 |
VectorizedArrayType, |
| 2641 |
MatrixType>( |
| 2642 |
matrix_free, |
| 2643 |
constraints, |
| 2644 |
matrix, |
| 2645 |
[&](auto &phi) { (owning_class->*cell_operation)(phi); }, |
| 2646 |
[&](auto &phi_m, auto &phi_p) { |
| 2647 |
(owning_class->*face_operation)(phi_m, phi_p); |
| 2648 |
}, |
| 2649 |
[&](auto &phi) { (owning_class->*boundary_operation)(phi); }, |
| 2650 |
dof_no, |
| 2651 |
quad_no, |
| 2652 |
first_selected_component); |
| 2653 |
} |
| 2654 |
|
| 2655 |
#endif // DOXYGEN |
| 2656 |
|
| 2657 |
} // namespace MatrixFreeTools |
| 2658 |
|
| 2659 |
DEAL_II_NAMESPACE_CLOSE |
| 2660 |
|
| 2661 |
|
| 2662 |
#endif |