| 1 |
// Copyright(C) 2021, 2022, 2023, 2024 National Technology & Engineering Solutions |
| 2 |
// of Sandia, LLC (NTESS). Under the terms of Contract DE-NA0003525 with |
| 3 |
// NTESS, the U.S. Government retains certain rights in this software. |
| 4 |
// |
| 5 |
// See packages/seacas/LICENSE for details |
| 6 |
|
| 7 |
#include <numeric> |
| 8 |
|
| 9 |
#include "Cell.h" |
| 10 |
|
| 11 |
#include <Ioss_NodeBlock.h> |
| 12 |
#include <Ioss_SmartAssert.h> |
| 13 |
#include <algorithm> |
| 14 |
#include <fmt/format.h> |
| 15 |
#include <fmt/ranges.h> |
| 16 |
#include <string> |
| 17 |
|
| 18 |
//! \file |
| 19 |
extern unsigned int debug_level; |
| 20 |
|
| 21 |
template <> struct fmt::formatter<Loc> : formatter<std::string> |
| 22 |
{ |
| 23 |
// parse is inherited from formatter<std::string>. |
| 24 |
template <typename FormatContext> auto format(Loc l, FormatContext &ctx) const |
| 25 |
{ |
| 26 |
std::string name = "unknown"; |
| 27 |
switch (l) { |
| 28 |
case Loc::C: name = "Center"; break; |
| 29 |
case Loc::BL: name = "Bottom Left"; break; |
| 30 |
case Loc::B: name = "Bottom"; break; |
| 31 |
case Loc::BR: name = "Bottom Right"; break; |
| 32 |
case Loc::L: name = "Left"; break; |
| 33 |
case Loc::R: name = "Right"; break; |
| 34 |
case Loc::TL: name = "Top Left"; break; |
| 35 |
case Loc::T: name = "Top"; break; |
| 36 |
case Loc::TR: name = "Top Right"; break; |
| 37 |
} |
| 38 |
return formatter<std::string>::format(name, ctx); |
| 39 |
} |
| 40 |
}; |
| 41 |
|
| 42 |
namespace { |
| 43 |
// Iterate over the interior nodes on the specified face. Skips |
| 44 |
// the corner nodes on the I-J intersections. Processes I-K and |
| 45 |
// J-K corners. |
| 46 |
template <typename INT> |
| 47 |
void process_face_nodes(const std::vector<INT> &node_map, std::vector<INT> &nodes, |
| 48 |
std::vector<INT> &procs, const std::vector<int64_t> &face_nodes, |
| 49 |
size_t KK, int rank) |
| 50 |
{ |
| 51 |
for (size_t i = KK; i < face_nodes.size() - KK; i++) { |
| 52 |
auto idx = face_nodes[i]; |
| 53 |
nodes.push_back(node_map[idx + 1]); |
| 54 |
procs.push_back(rank); |
| 55 |
} |
| 56 |
} |
| 57 |
|
| 58 |
// Iterate over the specified corner nodes. |
| 59 |
template <typename INT> |
| 60 |
void process_corner_nodes(const std::vector<INT> &node_map, std::vector<INT> &nodes, |
| 61 |
std::vector<INT> &procs, const std::vector<int64_t> &face_nodes, |
| 62 |
size_t KK, int rank, Loc location) |
| 63 |
{ |
| 64 |
if (location == Loc::BL || location == Loc::TL) { |
| 65 |
for (size_t i = 0; i < KK; i++) { |
| 66 |
auto idx = face_nodes[i]; |
| 67 |
nodes.push_back(node_map[idx + 1]); |
| 68 |
procs.push_back(rank); |
| 69 |
} |
| 70 |
} |
| 71 |
else { |
| 72 |
for (size_t i = face_nodes.size() - KK; i < face_nodes.size(); i++) { |
| 73 |
auto idx = face_nodes[i]; |
| 74 |
nodes.push_back(node_map[idx + 1]); |
| 75 |
procs.push_back(rank); |
| 76 |
} |
| 77 |
} |
| 78 |
} |
| 79 |
|
| 80 |
// Return a vector (possibly empty) of the ranks of the cells that surround this |
| 81 |
// cell. Includes the rank that this cell is on (cell_ranks[0]). |
| 82 |
std::vector<int> get_shared_ranks(const std::array<int, 9> &cell_ranks) |
| 83 |
{ |
| 84 |
std::vector<int> ranks(9); |
| 85 |
std::copy(cell_ranks.begin(), cell_ranks.end(), ranks.begin()); |
| 86 |
|
| 87 |
// Set all `-1` (non-neighbor) values to match center rank... |
| 88 |
for (auto &r : ranks) { |
| 89 |
if (r == -1) { |
| 90 |
r = cell_ranks[(int)Loc::C]; |
| 91 |
} |
| 92 |
} |
| 93 |
Ioss::Utils::uniquify(ranks); |
| 94 |
return ranks; |
| 95 |
} |
| 96 |
} // namespace |
| 97 |
|
| 98 |
void Cell::initialize(size_t i, size_t j, std::shared_ptr<UnitCell> unit_cell) |
| 99 |
{ |
| 100 |
m_i = i; |
| 101 |
m_j = j; |
| 102 |
m_unitCell = unit_cell; |
| 103 |
|
| 104 |
// These are not necessarily the correct ranks, but at this point can determine |
| 105 |
// Whether this cell is surrounded by other cells, or is on the boundary. |
| 106 |
set_rank(Loc::C, 0); |
| 107 |
set_rank(Loc::L, m_i > 0 ? 0 : -1); |
| 108 |
set_rank(Loc::B, m_j > 0 ? 0 : -1); |
| 109 |
set_rank(Loc::BL, (m_i > 0 && m_j > 0) ? 0 : -1); |
| 110 |
} |
| 111 |
|
| 112 |
std::pair<double, double> Cell::get_coordinate_range(enum Axis axis) const |
| 113 |
{ |
| 114 |
if (axis == Axis::X) { |
| 115 |
return m_unitCell->minmax_x; |
| 116 |
} |
| 117 |
if (axis == Axis::Y) { |
| 118 |
return m_unitCell->minmax_y; |
| 119 |
} |
| 120 |
return std::make_pair(0.0, 0.0); |
| 121 |
} |
| 122 |
|
| 123 |
// Number of nodes that will be added to global node count when this cell is added to |
| 124 |
// grid -- accounts for coincident nodes if cell has neighbor(s) |
| 125 |
size_t Cell::added_node_count(enum Mode mode, bool equivalence_nodes) const |
| 126 |
{ |
| 127 |
// If no neighbors (to -I, -J), then all nodes would be added... |
| 128 |
auto count = m_unitCell->m_region->get_property("node_count").get_int(); |
| 129 |
|
| 130 |
if (equivalence_nodes) { |
| 131 |
if (mode == Mode::GLOBAL) { |
| 132 |
if (has_neighbor_i()) { |
| 133 |
count -= (m_unitCell->cell_JJ * m_unitCell->cell_KK); |
| 134 |
} |
| 135 |
|
| 136 |
if (has_neighbor_j()) { |
| 137 |
count -= (m_unitCell->cell_II * m_unitCell->cell_KK); |
| 138 |
} |
| 139 |
|
| 140 |
if (has_neighbor_i() && has_neighbor_j()) { |
| 141 |
count += m_unitCell->cell_KK; |
| 142 |
} |
| 143 |
} |
| 144 |
else if (mode == Mode::PROCESSOR) { |
| 145 |
if (has_neighbor_i() && !processor_boundary(Loc::L)) { |
| 146 |
count -= (m_unitCell->cell_JJ * m_unitCell->cell_KK); |
| 147 |
} |
| 148 |
|
| 149 |
if (has_neighbor_j() && !processor_boundary(Loc::B)) { |
| 150 |
count -= (m_unitCell->cell_II * m_unitCell->cell_KK); |
| 151 |
} |
| 152 |
|
| 153 |
if (has_neighbor_i() && has_neighbor_j() && !processor_boundary(Loc::L) && |
| 154 |
!processor_boundary(Loc::B)) { |
| 155 |
count += m_unitCell->cell_KK; |
| 156 |
} |
| 157 |
|
| 158 |
// Now the "corner case" ;-) If there is a processor boundary below, but the cell to the BL is |
| 159 |
// on the same rank as this cell, then we have already counted the IJ-line nodes, so need to |
| 160 |
// subtract that count... |
| 161 |
if (processor_boundary(Loc::B) && processor_boundary(Loc::L) && |
| 162 |
rank(Loc::BL) == rank(Loc::C)) { |
| 163 |
count -= m_unitCell->cell_KK; |
| 164 |
} |
| 165 |
|
| 166 |
// Now the other "corner case" |
| 167 |
if (processor_boundary(Loc::B) && rank(Loc::BR) == rank(Loc::C)) { |
| 168 |
count -= m_unitCell->cell_KK; |
| 169 |
} |
| 170 |
} |
| 171 |
} |
| 172 |
return count; |
| 173 |
} |
| 174 |
|
| 175 |
std::array<int, 9> Cell::categorize_processor_boundary_nodes(int the_rank) const |
| 176 |
{ |
| 177 |
// Create a "unit cell" to categorize processor boundary nodes... |
| 178 |
std::array<int, 9> bnd_nodes{0}; |
| 179 |
|
| 180 |
// Bottom... |
| 181 |
if (rank(Loc::B) == the_rank) { |
| 182 |
bnd_nodes[(int)Loc::B] = 1; |
| 183 |
if ((rank(Loc::BL) != rank(Loc::C)) && (rank(Loc::L) != rank(Loc::C))) { |
| 184 |
bnd_nodes[(int)Loc::BL] = 1; |
| 185 |
} |
| 186 |
if (rank(Loc::BR) != rank(Loc::C)) { |
| 187 |
bnd_nodes[(int)Loc::BR] = 1; |
| 188 |
} |
| 189 |
} |
| 190 |
|
| 191 |
// Left |
| 192 |
if (rank(Loc::L) == the_rank) { |
| 193 |
bnd_nodes[(int)Loc::L] = 1; |
| 194 |
bnd_nodes[(int)Loc::TL] = 1; |
| 195 |
if ((rank(Loc::BL) != rank(Loc::C)) && (rank(Loc::B) != rank(Loc::C))) { |
| 196 |
bnd_nodes[(int)Loc::BL] = 1; |
| 197 |
} |
| 198 |
} |
| 199 |
|
| 200 |
// Top |
| 201 |
if (rank(Loc::T) == the_rank) { |
| 202 |
bnd_nodes[(int)Loc::T] = 1; |
| 203 |
bnd_nodes[(int)Loc::TR] = 1; |
| 204 |
if (rank(Loc::L) != rank(Loc::C)) { |
| 205 |
bnd_nodes[(int)Loc::TL] = 1; |
| 206 |
} |
| 207 |
} |
| 208 |
|
| 209 |
// Right |
| 210 |
if (rank(Loc::R) == the_rank) { |
| 211 |
bnd_nodes[(int)Loc::R] = 1; |
| 212 |
bnd_nodes[(int)Loc::TR] = 1; |
| 213 |
if ((rank(Loc::BR) != rank(Loc::C)) && (rank(Loc::B) != rank(Loc::C))) { |
| 214 |
bnd_nodes[(int)Loc::BR] = 1; |
| 215 |
} |
| 216 |
} |
| 217 |
|
| 218 |
// Bottom Left |
| 219 |
if (rank(Loc::BL) == the_rank) { |
| 220 |
// If left and bottom *don't* match the_rank, then need to add this node |
| 221 |
if ((rank(Loc::L) != the_rank) && (rank(Loc::B) != the_rank) && |
| 222 |
(rank(Loc::L) != rank(Loc::C)) && (rank(Loc::B) != rank(Loc::C))) { |
| 223 |
bnd_nodes[(int)Loc::BL] = 1; |
| 224 |
} |
| 225 |
} |
| 226 |
|
| 227 |
// Bottom Right |
| 228 |
if (rank(Loc::BR) == the_rank) { |
| 229 |
// If bottom *doesn't* match the_rank, then need to add this node |
| 230 |
if ((rank(Loc::B) != the_rank) && (rank(Loc::B) != rank(Loc::C))) { |
| 231 |
bnd_nodes[(int)Loc::BR] = 1; |
| 232 |
} |
| 233 |
} |
| 234 |
|
| 235 |
// Top Left |
| 236 |
if (rank(Loc::TL) == the_rank) { |
| 237 |
// If left *doesn't* match the_rank, then need to add this node |
| 238 |
if ((rank(Loc::L) != the_rank) && (rank(Loc::L) != rank(Loc::C))) { |
| 239 |
bnd_nodes[(int)Loc::TL] = 1; |
| 240 |
} |
| 241 |
} |
| 242 |
|
| 243 |
// Top Right |
| 244 |
if (rank(Loc::TR) == the_rank) { |
| 245 |
bnd_nodes[(int)Loc::TR] = 1; |
| 246 |
} |
| 247 |
|
| 248 |
return bnd_nodes; |
| 249 |
} |
| 250 |
|
| 251 |
size_t Cell::processor_boundary_node_count() const |
| 252 |
{ |
| 253 |
// Get list of ranks that this cell shares nodes with... |
| 254 |
auto ranks = get_shared_ranks(m_ranks); |
| 255 |
if (ranks.size() == 1) { |
| 256 |
// `ranks` contains center node, so if size == 1, does not touch |
| 257 |
// any other processor. |
| 258 |
return 0; |
| 259 |
} |
| 260 |
|
| 261 |
// Iterate `ranks` and for each rank, "color" the `bnd_nodes` that that rank touches... |
| 262 |
// Skip center. |
| 263 |
size_t b_count = 0; |
| 264 |
for (auto the_rank : ranks) { |
| 265 |
if (the_rank == rank(Loc::C)) { |
| 266 |
continue; |
| 267 |
} |
| 268 |
|
| 269 |
// a "unit cell" categorizing processor boundary nodes... |
| 270 |
// Size is 9. Value is '1' if nodes at this location are shared with rank `rank` |
| 271 |
auto bnd_nodes = categorize_processor_boundary_nodes(the_rank); |
| 272 |
|
| 273 |
// Now count how many nodes we have added... |
| 274 |
// Edges (B, T, L, R) without corners |
| 275 |
b_count += (m_unitCell->cell_II - 2) * (bnd_nodes[(int)Loc::B] + bnd_nodes[(int)Loc::T]); |
| 276 |
b_count += (m_unitCell->cell_JJ - 2) * (bnd_nodes[(int)Loc::L] + bnd_nodes[(int)Loc::R]); |
| 277 |
|
| 278 |
// Now the corners (BL, BR, TL, TR) |
| 279 |
b_count += bnd_nodes[(int)Loc::BL] + bnd_nodes[(int)Loc::BR] + bnd_nodes[(int)Loc::TL] + |
| 280 |
bnd_nodes[(int)Loc::TR]; |
| 281 |
} |
| 282 |
|
| 283 |
// The counts above only account for a single KK plane. Now multiply by `m_unitCell->KK` to get |
| 284 |
// total count. |
| 285 |
b_count *= m_unitCell->cell_KK; |
| 286 |
m_communicationNodeCount = b_count; |
| 287 |
return b_count; |
| 288 |
} |
| 289 |
|
| 290 |
template void Cell::populate_node_communication_map(const std::vector<int> &node_map, |
| 291 |
std::vector<int> &nodes, |
| 292 |
std::vector<int> &procs) const; |
| 293 |
template void Cell::populate_node_communication_map(const std::vector<int64_t> &node_map, |
| 294 |
std::vector<int64_t> &nodes, |
| 295 |
std::vector<int64_t> &procs) const; |
| 296 |
|
| 297 |
template <typename INT> |
| 298 |
void Cell::populate_node_communication_map(const std::vector<INT> &node_map, |
| 299 |
std::vector<INT> &nodes, std::vector<INT> &procs) const |
| 300 |
{ |
| 301 |
if (m_communicationNodeCount == 0) { |
| 302 |
return; |
| 303 |
} |
| 304 |
|
| 305 |
nodes.reserve(m_communicationNodeCount); |
| 306 |
procs.reserve(m_communicationNodeCount); |
| 307 |
|
| 308 |
// Get list of ranks that this cell shares nodes with... |
| 309 |
auto ranks = get_shared_ranks(m_ranks); |
| 310 |
SMART_ASSERT(ranks.size() > 1); |
| 311 |
|
| 312 |
auto KK = m_unitCell->cell_KK; |
| 313 |
|
| 314 |
for (auto shared_rank : ranks) { |
| 315 |
if (shared_rank == rank(Loc::C)) { |
| 316 |
continue; |
| 317 |
} |
| 318 |
|
| 319 |
// a "unit cell" categorizing processor boundary nodes... |
| 320 |
// Size is 9. Value is '1' if nodes at this location are shared with rank `rank` |
| 321 |
auto bnd_nodes = categorize_processor_boundary_nodes(shared_rank); |
| 322 |
|
| 323 |
// Handle Edges, but skip nodes on corners. They are handled later. |
| 324 |
if (bnd_nodes[(int)Loc::B] == 1) { |
| 325 |
process_face_nodes(node_map, nodes, procs, m_unitCell->min_J_face, KK, shared_rank); |
| 326 |
} |
| 327 |
|
| 328 |
if (bnd_nodes[(int)Loc::T] == 1) { |
| 329 |
process_face_nodes(node_map, nodes, procs, m_unitCell->max_J_face, KK, shared_rank); |
| 330 |
} |
| 331 |
|
| 332 |
if (bnd_nodes[(int)Loc::L] == 1) { |
| 333 |
process_face_nodes(node_map, nodes, procs, m_unitCell->min_I_face, KK, shared_rank); |
| 334 |
} |
| 335 |
|
| 336 |
if (bnd_nodes[(int)Loc::R] == 1) { |
| 337 |
process_face_nodes(node_map, nodes, procs, m_unitCell->max_I_face, KK, shared_rank); |
| 338 |
} |
| 339 |
|
| 340 |
// Now the corners... |
| 341 |
if (bnd_nodes[(int)Loc::BL] == 1) { |
| 342 |
process_corner_nodes(node_map, nodes, procs, m_unitCell->min_J_face, KK, shared_rank, |
| 343 |
Loc::BL); |
| 344 |
} |
| 345 |
|
| 346 |
if (bnd_nodes[(int)Loc::BR] == 1) { |
| 347 |
process_corner_nodes(node_map, nodes, procs, m_unitCell->min_J_face, KK, shared_rank, |
| 348 |
Loc::BR); |
| 349 |
} |
| 350 |
|
| 351 |
if (bnd_nodes[(int)Loc::TL] == 1) { |
| 352 |
process_corner_nodes(node_map, nodes, procs, m_unitCell->max_J_face, KK, shared_rank, |
| 353 |
Loc::TL); |
| 354 |
} |
| 355 |
|
| 356 |
if (bnd_nodes[(int)Loc::TR] == 1) { |
| 357 |
process_corner_nodes(node_map, nodes, procs, m_unitCell->max_J_face, KK, shared_rank, |
| 358 |
Loc::TR); |
| 359 |
} |
| 360 |
} |
| 361 |
SMART_ASSERT(nodes.size() == procs.size())(nodes.size())(procs.size()); |
| 362 |
SMART_ASSERT(nodes.size() == m_communicationNodeCount)(nodes.size())(m_communicationNodeCount); |
| 363 |
} |
| 364 |
|
| 365 |
std::vector<int> Cell::categorize_nodes(enum Mode mode) const |
| 366 |
{ |
| 367 |
auto nodes = m_unitCell->categorize_nodes(has_neighbor_i(), has_neighbor_j()); |
| 368 |
if (mode == Mode::PROCESSOR) { |
| 369 |
// If there is a processor boundary to the left, then need to change categorization of |
| 370 |
// all nodes on the left to '0' |
| 371 |
if (processor_boundary(Loc::L)) { |
| 372 |
const auto &min_I_face = m_unitCell->min_I_face; |
| 373 |
for (const auto &node : min_I_face) { |
| 374 |
nodes[node] -= 1; |
| 375 |
} |
| 376 |
} |
| 377 |
if (processor_boundary(Loc::B)) { |
| 378 |
const auto &min_J_face = m_unitCell->min_J_face; |
| 379 |
for (const auto &node : min_J_face) { |
| 380 |
nodes[node] -= 2; |
| 381 |
} |
| 382 |
} |
| 383 |
|
| 384 |
// Now the "corner case" ;-) If there is a processor boundary below, but the cell to the BL is |
| 385 |
// on the same rank as this cell, then we have already counted the IJ-line nodes, so need to |
| 386 |
// categorize those nodes as already accounted for in a previous map... |
| 387 |
if (processor_boundary(Loc::B) && processor_boundary(Loc::L) && rank(Loc::BL) == rank(Loc::C)) { |
| 388 |
// Want KK() nodes -- First KK of min_i and of min_j. But since they match, can "unzero" |
| 389 |
// min_i[0..KK) |
| 390 |
for (size_t i = 0; i < m_unitCell->cell_KK; i++) { |
| 391 |
nodes[m_unitCell->min_I_face[i]] = -1; |
| 392 |
} |
| 393 |
} |
| 394 |
// Now the other "corner case" |
| 395 |
if (processor_boundary(Loc::B) && rank(Loc::BR) == rank(Loc::C)) { |
| 396 |
// Want KK() nodes -- First KK of max_i and Last KK of min_j. But since they match, can |
| 397 |
// "unzero" max_i[0..KK) |
| 398 |
for (size_t i = 0; i < m_unitCell->cell_KK; i++) { |
| 399 |
nodes[m_unitCell->max_I_face[i]] = -1; |
| 400 |
} |
| 401 |
} |
| 402 |
} |
| 403 |
return nodes; |
| 404 |
} |
| 405 |
|
| 406 |
template std::vector<int64_t> Cell::generate_node_map(Mode, bool, int64_t) const; |
| 407 |
template std::vector<int> Cell::generate_node_map(Mode, bool, int) const; |
| 408 |
|
| 409 |
template <typename INT> |
| 410 |
std::vector<INT> Cell::generate_node_map(Mode mode, bool equivalence_nodes, INT /*dummy*/) const |
| 411 |
{ |
| 412 |
// Size is node_count + 1 to handle the 1-based connectivity values. |
| 413 |
size_t cell_node_count = m_unitCell->m_region->get_property("node_count").get_int(); |
| 414 |
std::vector<INT> map(cell_node_count + 1); |
| 415 |
|
| 416 |
INT offset = mode == Mode::PROCESSOR ? m_localNodeIdOffset : m_globalNodeIdOffset; |
| 417 |
|
| 418 |
if (!equivalence_nodes || !(has_neighbor_i() || has_neighbor_j())) { |
| 419 |
std::iota(map.begin(), map.end(), offset); |
| 420 |
} |
| 421 |
else if (has_neighbor_i() || has_neighbor_j()) { |
| 422 |
// At least one neighboring cell and the nodes are being equivalenced |
| 423 |
// Generate map for the "non-neighbored" nodes (not contiguous with a neighbor cell) |
| 424 |
auto categorized_nodes = categorize_nodes(mode); |
| 425 |
SMART_ASSERT(categorized_nodes.size() == cell_node_count) |
| 426 |
(categorized_nodes.size())(cell_node_count); |
| 427 |
offset++; // To deal with 1-based node numbers. |
| 428 |
for (size_t n = 0; n < cell_node_count; n++) { |
| 429 |
if (categorized_nodes[n] == 0) { |
| 430 |
map[n + 1] = offset++; |
| 431 |
} |
| 432 |
} |
| 433 |
} |
| 434 |
|
| 435 |
if (equivalence_nodes && has_neighbor_i() && |
| 436 |
(mode == Mode::GLOBAL || (mode == Mode::PROCESSOR && rank(Loc::C) == rank(Loc::L)))) { |
| 437 |
// Get the neighbor cell... |
| 438 |
// iterate my unit cell's min_I_face() nodes to get index into map |
| 439 |
// At this index, set value to this cells min_I_nodes() node |
| 440 |
// which was created by the neighbor when he was processed... |
| 441 |
SMART_ASSERT(min_I_nodes.size() == m_unitCell->min_I_face.size()) |
| 442 |
(m_i)(m_j)(min_I_nodes.size())(m_unitCell->min_I_face.size()); |
| 443 |
|
| 444 |
for (size_t i = 0; i < m_unitCell->min_I_face.size(); i++) { |
| 445 |
auto idx = m_unitCell->min_I_face[i] + 1; |
| 446 |
auto val = min_I_nodes[i]; |
| 447 |
map[idx] = (INT)val; |
| 448 |
} |
| 449 |
} |
| 450 |
|
| 451 |
if (equivalence_nodes && has_neighbor_j() && |
| 452 |
(mode == Mode::GLOBAL || (mode == Mode::PROCESSOR && rank(Loc::C) == rank(Loc::B)))) { |
| 453 |
SMART_ASSERT(min_J_nodes.size() == m_unitCell->min_J_face.size()) |
| 454 |
(m_i)(m_j)(min_J_nodes.size())(m_unitCell->min_J_face.size()); |
| 455 |
|
| 456 |
for (size_t i = 0; i < m_unitCell->min_J_face.size(); i++) { |
| 457 |
auto idx = m_unitCell->min_J_face[i] + 1; |
| 458 |
auto val = min_J_nodes[i]; |
| 459 |
map[idx] = (INT)val; |
| 460 |
} |
| 461 |
} |
| 462 |
|
| 463 |
if (mode == Mode::PROCESSOR) { |
| 464 |
// Now the "corner case" ;-) If there is a processor boundary below, but the cell to the BL is |
| 465 |
// on the same rank as this cell, then we have already counted the IJ-line nodes, so need to |
| 466 |
// categorize those nodes as already accounted for in a previous map... |
| 467 |
if (processor_boundary(Loc::B) && processor_boundary(Loc::L) && rank(Loc::BL) == rank(Loc::C)) { |
| 468 |
// Want KK() nodes -- First KK of min_i and of min_j. But since they match, can "unzero" |
| 469 |
// min_i[0..KK) |
| 470 |
auto KK = m_unitCell->cell_KK; |
| 471 |
for (size_t i = 0; i < KK; i++) { |
| 472 |
auto idx = m_unitCell->min_J_face[i] + 1; |
| 473 |
auto val = min_J_nodes[i]; |
| 474 |
map[idx] = (INT)val; |
| 475 |
} |
| 476 |
} |
| 477 |
// Now the other "corner case" |
| 478 |
if (processor_boundary(Loc::B) && rank(Loc::BR) == rank(Loc::C)) { |
| 479 |
// Want KK() nodes -- First KK of max_i and Last KK of min_j. But since they match, can |
| 480 |
// "unzero" max_i[0..KK) |
| 481 |
auto KK = m_unitCell->cell_KK; |
| 482 |
auto j_offset = min_J_nodes.size() - KK; |
| 483 |
for (size_t i = 0; i < KK; i++) { |
| 484 |
auto idx = m_unitCell->min_J_face[j_offset + i] + 1; |
| 485 |
auto val = min_J_nodes[j_offset + i]; |
| 486 |
map[idx] = (INT)val; |
| 487 |
} |
| 488 |
} |
| 489 |
} |
| 490 |
// Can now clean out the `min_I_nodes` and `min_J_nodes` lists since the data will no longer be |
| 491 |
// needed. |
| 492 |
Ioss::Utils::clear(min_I_nodes); |
| 493 |
Ioss::Utils::clear(min_J_nodes); |
| 494 |
|
| 495 |
return map; |
| 496 |
} |
| 497 |
|
| 498 |
template void Cell::populate_neighbor(Loc location, const std::vector<int64_t> &map, |
| 499 |
const Cell &neighbor) const; |
| 500 |
template void Cell::populate_neighbor(Loc location, const std::vector<int> &map, |
| 501 |
const Cell &neighbor) const; |
| 502 |
|
| 503 |
template <typename INT> |
| 504 |
void Cell::populate_neighbor(Loc location, const std::vector<INT> &map, const Cell &neighbor) const |
| 505 |
{ |
| 506 |
switch (location) { |
| 507 |
case Loc::L: |
| 508 |
neighbor.min_I_nodes.resize(m_unitCell->max_I_face.size()); |
| 509 |
for (size_t i = 0; i < m_unitCell->max_I_face.size(); i++) { |
| 510 |
auto idx = m_unitCell->max_I_face[i] + 1; |
| 511 |
auto val = map[idx]; |
| 512 |
neighbor.min_I_nodes[i] = val; |
| 513 |
} |
| 514 |
if (debug_level & 8) { |
| 515 |
fmt::print("\nCell {} {}\n", neighbor.m_i, neighbor.m_j); |
| 516 |
fmt::print("min_I_nodes: {}\n", fmt::join(neighbor.min_I_nodes, " ")); |
| 517 |
} |
| 518 |
break; |
| 519 |
|
| 520 |
case Loc::B: |
| 521 |
neighbor.min_J_nodes.resize(m_unitCell->max_J_face.size()); |
| 522 |
for (size_t i = 0; i < m_unitCell->max_J_face.size(); i++) { |
| 523 |
auto idx = m_unitCell->max_J_face[i] + 1; |
| 524 |
auto val = map[idx]; |
| 525 |
neighbor.min_J_nodes[i] = val; |
| 526 |
} |
| 527 |
if (debug_level & 8) { |
| 528 |
fmt::print("min_J_nodes: {}\n", fmt::join(neighbor.min_J_nodes, " ")); |
| 529 |
} |
| 530 |
break; |
| 531 |
|
| 532 |
case Loc::BR: { |
| 533 |
neighbor.min_J_nodes.resize(m_unitCell->max_J_face.size()); |
| 534 |
auto KK = m_unitCell->cell_KK; |
| 535 |
auto j_offset = neighbor.min_J_nodes.size() - KK; |
| 536 |
for (size_t i = 0; i < KK; i++) { |
| 537 |
auto idx = m_unitCell->max_J_face[i] + 1; |
| 538 |
auto val = map[idx]; |
| 539 |
neighbor.min_J_nodes[j_offset + i] = val; |
| 540 |
} |
| 541 |
} break; |
| 542 |
|
| 543 |
case Loc::BL: { |
| 544 |
neighbor.min_J_nodes.resize(m_unitCell->max_J_face.size()); |
| 545 |
auto KK = m_unitCell->cell_KK; |
| 546 |
auto j_offset = neighbor.min_J_nodes.size() - KK; |
| 547 |
for (size_t i = 0; i < KK; i++) { |
| 548 |
auto idx = m_unitCell->max_J_face[j_offset + i] + 1; |
| 549 |
auto val = map[idx]; |
| 550 |
neighbor.min_J_nodes[i] = val; |
| 551 |
} |
| 552 |
} break; |
| 553 |
|
| 554 |
default: |
| 555 |
fmt::print(stderr, "\nINTERNAL ERROR: Unhandled direction in populate_neighbor(): {}\n", |
| 556 |
location); |
| 557 |
exit(EXIT_FAILURE); |
| 558 |
} |
| 559 |
} |