| 1 | // -------------------------------------------------------------------------- |
| 2 | // |
| 3 | // Copyright |
| 4 | // Markus Wittmann, 2016-2017 |
| 5 | // RRZE, University of Erlangen-Nuremberg, Germany |
| 6 | // markus.wittmann -at- fau.de or hpc -at- rrze.fau.de |
| 7 | // |
| 8 | // Viktor Haag, 2016 |
| 9 | // LSS, University of Erlangen-Nuremberg, Germany |
| 10 | // |
| 11 | // This file is part of the Lattice Boltzmann Benchmark Kernels (LbmBenchKernels). |
| 12 | // |
| 13 | // LbmBenchKernels is free software: you can redistribute it and/or modify |
| 14 | // it under the terms of the GNU General Public License as published by |
| 15 | // the Free Software Foundation, either version 3 of the License, or |
| 16 | // (at your option) any later version. |
| 17 | // |
| 18 | // LbmBenchKernels is distributed in the hope that it will be useful, |
| 19 | // but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 20 | // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 21 | // GNU General Public License for more details. |
| 22 | // |
| 23 | // You should have received a copy of the GNU General Public License |
| 24 | // along with LbmBenchKernels. If not, see <http://www.gnu.org/licenses/>. |
| 25 | // |
| 26 | // -------------------------------------------------------------------------- |
| 27 | #include "BenchKernelD3Q19AaCommon.h" |
| 28 | |
| 29 | #include "Memory.h" |
| 30 | #include "Vtk.h" |
| 31 | #include "LikwidIf.h" |
| 32 | |
| 33 | #include <inttypes.h> |
| 34 | #include <math.h> |
| 35 | |
| 36 | #ifdef _OPENMP |
| 37 | #include <omp.h> |
| 38 | #endif |
| 39 | |
| 40 | void FNAME(D3Q19AaKernel)(LatticeDesc * ld, KernelData * kernelData, CaseData * cd) |
| 41 | { |
| 42 | Assert(ld != NULL); |
| 43 | Assert(kernelData != NULL); |
| 44 | Assert(cd != NULL); |
| 45 | |
| 46 | Assert(cd->Omega > F(0.0)); |
| 47 | Assert(cd->Omega < F(2.0)); |
| 48 | |
| 49 | KernelData * kd = (KernelData *)kernelData; |
| 50 | |
| 51 | |
| 52 | int nX = ld->Dims[0]; |
| 53 | int nY = ld->Dims[1]; |
| 54 | int nZ = ld->Dims[2]; |
| 55 | |
| 56 | int * gDims = kd->GlobalDims; |
| 57 | |
| 58 | int oX = kd->Offsets[0]; |
| 59 | int oY = kd->Offsets[1]; |
| 60 | int oZ = kd->Offsets[2]; |
| 61 | |
| 62 | KernelDataAa * kda = KDA(kd); |
| 63 | |
| 64 | int blk[3]; |
| 65 | blk[0] = kda->Blk[0]; |
| 66 | blk[1] = kda->Blk[1]; |
| 67 | blk[2] = kda->Blk[2]; |
| 68 | |
| 69 | PdfT omega = cd->Omega; |
| 70 | PdfT omegaEven = omega; |
| 71 | PdfT magicParam = F(1.0) / F(12.0); |
| 72 | // 1/4: best stability; |
| 73 | // 1/12: removes third-order advection error (best advection); |
| 74 | // 1/6: removes fourth-order diffusion error (best diffusion); |
| 75 | // 3/16: exact location of bounce back for poiseuille flow |
| 76 | |
| 77 | PdfT omegaOdd = F(1.0)/( F(0.5) + magicParam/(F(1.0)/omega - F(0.5)) ); |
| 78 | |
| 79 | PdfT evenPart = F(0.0); |
| 80 | PdfT oddPart = F(0.0); |
| 81 | PdfT dir_indep_trm = F(0.0); |
| 82 | |
| 83 | PdfT w_0 = F(1.0) / F(3.0); |
| 84 | PdfT w_1 = F(1.0) / F(18.0); |
| 85 | PdfT w_2 = F(1.0) / F(36.0); |
| 86 | |
| 87 | PdfT w_1_x3 = w_1 * F(3.0); PdfT w_1_nine_half = w_1 * F(9.0)/F(2.0); PdfT w_1_indep = F(0.0); |
| 88 | PdfT w_2_x3 = w_2 * F(3.0); PdfT w_2_nine_half = w_2 * F(9.0)/F(2.0); PdfT w_2_indep = F(0.0); |
| 89 | |
| 90 | PdfT ux, uy, uz, ui; |
| 91 | PdfT dens; |
| 92 | |
| 93 | // Declare pdf_N, pdf_E, pdf_S, pdf_W, ... |
| 94 | #define X(name, idx, idxinv, x, y, z) PdfT JOIN(pdf_,name); |
| 95 | D3Q19_LIST |
| 96 | #undef X |
| 97 | |
| 98 | PdfT * src = kd->Pdfs[0]; |
| 99 | |
| 100 | int maxIterations = cd->MaxIterations; |
| 101 | |
| 102 | #ifdef VTK_OUTPUT |
| 103 | if (cd->VtkOutput) { |
| 104 | kd->PdfsActive = src; |
| 105 | VtkWrite(ld, kd, cd, -1); |
| 106 | } |
| 107 | #endif |
| 108 | |
| 109 | #ifdef STATISTICS |
| 110 | kd->PdfsActive = src; |
| 111 | KernelStatistics(kd, ld, cd, 0); |
| 112 | #endif |
| 113 | |
| 114 | |
| 115 | int nThreads = 1; |
| 116 | |
| 117 | #ifdef _OPENMP |
| 118 | nThreads = omp_get_max_threads(); |
| 119 | #endif |
| 120 | |
| 121 | X_KERNEL_START(kernelData); |
| 122 | |
| 123 | for (int iter = 0; iter < maxIterations; iter += 2) { |
| 124 | |
| 125 | // -------------------------------------------------------------------- |
| 126 | // even time step |
| 127 | |
| 128 | X_LIKWID_START("aa-even"); |
| 129 | |
| 130 | // {{{ |
| 131 | |
| 132 | #ifdef _OPENMP |
| 133 | #pragma omp parallel for default(none) \ |
| 134 | shared(gDims,src, w_0, w_1, w_2, omegaEven, omegaOdd, \ |
| 135 | w_1_x3, w_2_x3, w_1_nine_half, w_2_nine_half, cd, \ |
| 136 | oX, oY, oZ, nX, nY, nZ, blk, nThreads, ld) \ |
| 137 | private(ux, uy, uz, ui, dens, dir_indep_trm, \ |
| 138 | pdf_C, \ |
| 139 | pdf_N, pdf_E, pdf_S, pdf_W, \ |
| 140 | pdf_NE, pdf_SE, pdf_SW, pdf_NW, \ |
| 141 | pdf_T, pdf_TN, pdf_TE, pdf_TS, pdf_TW, \ |
| 142 | pdf_B, pdf_BN, pdf_BE, pdf_BS, pdf_BW, \ |
| 143 | evenPart, oddPart, w_1_indep, w_2_indep) |
| 144 | #endif |
| 145 | |
| 146 | for (int i = 0; i < nThreads; ++i) { |
| 147 | |
| 148 | int threadStartX = nX / nThreads * i; |
| 149 | int threadEndX = nX / nThreads * (i + 1); |
| 150 | |
| 151 | if (nX % nThreads > 0) { |
| 152 | if (nX % nThreads > i) { |
| 153 | threadStartX += i; |
| 154 | threadEndX += i + 1; |
| 155 | } |
| 156 | else { |
| 157 | threadStartX += nX % nThreads; |
| 158 | threadEndX += nX % nThreads; |
| 159 | } |
| 160 | } |
| 161 | |
| 162 | for (int bX = oX + threadStartX; bX < threadEndX + oX; bX += blk[0]) { |
| 163 | for (int bY = oY; bY < nY + oY; bY += blk[1]) { |
| 164 | for (int bZ = oZ; bZ < nZ + oZ; bZ += blk[2]) { |
| 165 | |
| 166 | int eX = MIN(bX + blk[0], threadEndX + oX); |
| 167 | int eY = MIN(bY + blk[1], nY + oY); |
| 168 | int eZ = MIN(bZ + blk[2], nZ + oZ); |
| 169 | |
| 170 | // printf("%d: %d-%d %d-%d %d-%d %d - %d\n", omp_get_thread_num(), bZ, eZ, bY, eY, bX, eX, threadStartX, threadEndX); |
| 171 | |
| 172 | for (int x = bX; x < eX; ++x) { |
| 173 | for (int y = bY; y < eY; ++y) { |
| 174 | #ifdef INTEL_OPT_DIRECTIVES |
| 175 | #pragma ivdep |
| 176 | #pragma vector always |
| 177 | #pragma simd |
| 178 | #endif |
| 179 | for (int z = bZ; z < eZ; ++z) { // LOOP aa-even |
| 180 | |
| 181 | |
| 182 | if (ld->Lattice[L_INDEX_4(ld->Dims, x - oX, y - oY, z - oZ)] == LAT_CELL_OBSTACLE) { |
| 183 | continue; |
| 184 | } |
| 185 | |
| 186 | #define I(x, y, z, dir) P_INDEX_5(gDims, (x), (y), (z), (dir)) |
| 187 | |
| 188 | |
| 189 | // Load PDFs of local cell: pdf_N = src[I(x, y, z, D3Q19_N)]; ... |
| 190 | #define X(name, idx, idxinv, _x, _y, _z) JOIN(pdf_,name) = src[I(x, y, z, idx)]; |
| 191 | D3Q19_LIST |
| 192 | #undef X |
| 193 | |
| 194 | // #define LID_DRIVEN_CAVITY |
| 195 | |
| 196 | #ifdef LID_DRIVEN_CAVITY |
| 197 | |
| 198 | if (z == nZ - 4 + oZ && x > 3 + oX && x < (nX - 4 + oX) && y > 3 + oY && y < (nY - 4 + oY)) { |
| 199 | ux = F(0.1) * F(0.5)77; |
| 200 | uy = F(0.0); |
| 201 | uz = F(0.0); |
| 202 | |
| 203 | } else { |
| 204 | #endif |
| 205 | ux = pdf_E + pdf_NE + pdf_SE + pdf_TE + pdf_BE - |
| 206 | pdf_W - pdf_NW - pdf_SW - pdf_TW - pdf_BW; |
| 207 | uy = pdf_N + pdf_NE + pdf_NW + pdf_TN + pdf_BN - |
| 208 | pdf_S - pdf_SE - pdf_SW - pdf_TS - pdf_BS; |
| 209 | uz = pdf_T + pdf_TE + pdf_TW + pdf_TN + pdf_TS - |
| 210 | pdf_B - pdf_BE - pdf_BW - pdf_BN - pdf_BS; |
| 211 | #ifdef LID_DRIVEN_CAVITY |
| 212 | } |
| 213 | #endif |
| 214 | |
| 215 | dens = pdf_C + |
| 216 | pdf_N + pdf_E + pdf_S + pdf_W + |
| 217 | pdf_NE + pdf_SE + pdf_SW + pdf_NW + |
| 218 | pdf_T + pdf_TN + pdf_TE + pdf_TS + pdf_TW + |
| 219 | pdf_B + pdf_BN + pdf_BE + pdf_BS + pdf_BW; |
| 220 | |
| 221 | dir_indep_trm = dens - (ux * ux + uy * uy + uz * uz)*F(3.0)/F(2.0); |
| 222 | |
| 223 | // direction: w_0 |
| 224 | src[I(x, y, z, D3Q19_C)] = pdf_C - omegaEven*(pdf_C - w_0*dir_indep_trm); |
| 225 | |
| 226 | // direction: w_1 |
| 227 | w_1_indep = w_1*dir_indep_trm; |
| 228 | |
| 229 | ui = uy; |
| 230 | evenPart = omegaEven*( F(0.5)*(pdf_N + pdf_S) - ui*ui*w_1_nine_half - w_1_indep ); |
| 231 | oddPart = omegaOdd*(F(0.5)*(pdf_N - pdf_S) - ui*w_1_x3 ); |
| 232 | src[I(x, y, z, D3Q19_S)] = pdf_N - evenPart - oddPart; |
| 233 | src[I(x, y, z, D3Q19_N)] = pdf_S - evenPart + oddPart; |
| 234 | |
| 235 | ui = ux; |
| 236 | evenPart = omegaEven*( F(0.5)*(pdf_E + pdf_W) - ui*ui*w_1_nine_half - w_1_indep ); |
| 237 | oddPart = omegaOdd*(F(0.5)*(pdf_E - pdf_W) - ui*w_1_x3 ); |
| 238 | src[I(x, y, z, D3Q19_W)] = pdf_E - evenPart - oddPart; |
| 239 | src[I(x, y, z, D3Q19_E)] = pdf_W - evenPart + oddPart; |
| 240 | |
| 241 | ui = uz; |
| 242 | evenPart = omegaEven*( F(0.5)*(pdf_T + pdf_B) - ui*ui*w_1_nine_half - w_1_indep ); |
| 243 | oddPart = omegaOdd*(F(0.5)*(pdf_T - pdf_B) - ui*w_1_x3 ); |
| 244 | src[I(x, y, z, D3Q19_B)] = pdf_T - evenPart - oddPart; |
| 245 | src[I(x, y, z, D3Q19_T)] = pdf_B - evenPart + oddPart; |
| 246 | |
| 247 | // direction: w_2 |
| 248 | w_2_indep = w_2*dir_indep_trm; |
| 249 | |
| 250 | ui = -ux + uy; |
| 251 | evenPart = omegaEven*( F(0.5)*(pdf_NW + pdf_SE) - ui*ui*w_2_nine_half - w_2_indep ); |
| 252 | oddPart = omegaOdd*(F(0.5)*(pdf_NW - pdf_SE) - ui*w_2_x3 ); |
| 253 | src[I(x, y, z, D3Q19_SE)] = pdf_NW - evenPart - oddPart; |
| 254 | src[I(x, y, z, D3Q19_NW)] = pdf_SE - evenPart + oddPart; |
| 255 | |
| 256 | ui = ux + uy; |
| 257 | evenPart = omegaEven*( F(0.5)*(pdf_NE + pdf_SW) - ui*ui*w_2_nine_half - w_2_indep ); |
| 258 | oddPart = omegaOdd*(F(0.5)*(pdf_NE - pdf_SW) - ui*w_2_x3 ); |
| 259 | src[I(x, y, z, D3Q19_SW)] = pdf_NE - evenPart - oddPart; |
| 260 | src[I(x, y, z, D3Q19_NE)] = pdf_SW - evenPart + oddPart; |
| 261 | |
| 262 | ui = -ux + uz; |
| 263 | evenPart = omegaEven*( F(0.5)*(pdf_TW + pdf_BE) - ui*ui*w_2_nine_half - w_2_indep ); |
| 264 | oddPart = omegaOdd*(F(0.5)*(pdf_TW - pdf_BE) - ui*w_2_x3 ); |
| 265 | src[I(x, y, z, D3Q19_BE)] = pdf_TW - evenPart - oddPart; |
| 266 | src[I(x, y, z, D3Q19_TW)] = pdf_BE - evenPart + oddPart; |
| 267 | |
| 268 | ui = ux + uz; |
| 269 | evenPart = omegaEven*( F(0.5)*(pdf_TE + pdf_BW) - ui*ui*w_2_nine_half - w_2_indep ); |
| 270 | oddPart = omegaOdd*(F(0.5)*(pdf_TE - pdf_BW) - ui*w_2_x3 ); |
| 271 | src[I(x, y, z, D3Q19_BW)] = pdf_TE - evenPart - oddPart; |
| 272 | src[I(x, y, z, D3Q19_TE)] = pdf_BW - evenPart + oddPart; |
| 273 | |
| 274 | ui = -uy + uz; |
| 275 | evenPart = omegaEven*( F(0.5)*(pdf_TS + pdf_BN) - ui*ui*w_2_nine_half - w_2_indep ); |
| 276 | oddPart = omegaOdd*(F(0.5)*(pdf_TS - pdf_BN) - ui*w_2_x3 ); |
| 277 | src[I(x, y, z, D3Q19_BN)] = pdf_TS - evenPart - oddPart; |
| 278 | src[I(x, y, z, D3Q19_TS)] = pdf_BN - evenPart + oddPart; |
| 279 | |
| 280 | ui = uy + uz; |
| 281 | evenPart = omegaEven*( F(0.5)*(pdf_TN + pdf_BS) - ui*ui*w_2_nine_half - w_2_indep ); |
| 282 | oddPart = omegaOdd*(F(0.5)*(pdf_TN - pdf_BS) - ui*w_2_x3 ); |
| 283 | src[I(x, y, z, D3Q19_BS)] = pdf_TN - evenPart - oddPart; |
| 284 | src[I(x, y, z, D3Q19_TN)] = pdf_BS - evenPart + oddPart; |
| 285 | |
| 286 | #undef I |
| 287 | } } } // z, y, x (from inner to outer) |
| 288 | } } } // z, y, x (from inner to outer) |
| 289 | |
| 290 | } // loop over threads |
| 291 | |
| 292 | // }}} |
| 293 | |
| 294 | X_LIKWID_STOP("aa-even"); |
| 295 | |
| 296 | #ifdef STATISTICS |
| 297 | kd->PdfsActive = src; |
| 298 | KernelStatistics(kd, ld, cd, iter); |
| 299 | #endif |
| 300 | |
| 301 | // Fixup bounce back PDFs. |
| 302 | #ifdef _OPENMP |
| 303 | #pragma omp parallel for default(none) \ |
| 304 | shared(kd, src) |
| 305 | #endif |
| 306 | #ifdef INTEL_OPT_DIRECTIVES |
| 307 | #pragma ivdep |
| 308 | #endif |
| 309 | for (int i = 0; i < kd->nBounceBackPdfs; ++i) { |
| 310 | src[kd->BounceBackPdfsSrc[i]] = src[kd->BounceBackPdfsDst[i]]; |
| 311 | } |
| 312 | |
| 313 | // save current iteration |
| 314 | kda->Iteration = iter; |
| 315 | |
| 316 | #ifdef VERIFICATION |
| 317 | kd->PdfsActive = src; |
| 318 | KernelAddBodyForce(kd, ld, cd); |
| 319 | #endif |
| 320 | |
| 321 | #ifdef VTK_OUTPUT |
| 322 | if (cd->VtkOutput && (iter % cd->VtkModulus) == 0) { |
| 323 | kd->PdfsActive = src; |
| 324 | VtkWrite(ld, kd, cd, iter); |
| 325 | } |
| 326 | #endif |
| 327 | |
| 328 | #ifdef STATISTICS |
| 329 | kd->PdfsActive = src; |
| 330 | KernelStatistics(kd, ld, cd, iter); |
| 331 | #endif |
| 332 | |
| 333 | // -------------------------------------------------------------------- |
| 334 | // odd time step |
| 335 | |
| 336 | |
| 337 | X_LIKWID_START("aa-odd"); |
| 338 | |
| 339 | // {{{ |
| 340 | |
| 341 | #ifdef _OPENMP |
| 342 | #pragma omp parallel for default(none) \ |
| 343 | shared(gDims,src, w_0, w_1, w_2, omegaEven, omegaOdd, \ |
| 344 | w_1_x3, w_2_x3, w_1_nine_half, w_2_nine_half, cd, \ |
| 345 | oX, oY, oZ, nX, nY, nZ, blk, nThreads) \ |
| 346 | private(ux, uy, uz, ui, dens, dir_indep_trm, \ |
| 347 | pdf_C, \ |
| 348 | pdf_N, pdf_E, pdf_S, pdf_W, \ |
| 349 | pdf_NE, pdf_SE, pdf_SW, pdf_NW, \ |
| 350 | pdf_T, pdf_TN, pdf_TE, pdf_TS, pdf_TW, \ |
| 351 | pdf_B, pdf_BN, pdf_BE, pdf_BS, pdf_BW, \ |
| 352 | evenPart, oddPart, w_1_indep, w_2_indep) |
| 353 | #endif |
| 354 | |
| 355 | for (int i = 0; i < nThreads; ++i) { |
| 356 | |
| 357 | int threadStartX = nX / nThreads * i; |
| 358 | int threadEndX = nX / nThreads * (i + 1); |
| 359 | |
| 360 | if (nX % nThreads > 0) { |
| 361 | if (nX % nThreads > i) { |
| 362 | threadStartX += i; |
| 363 | threadEndX += i + 1; |
| 364 | } |
| 365 | else { |
| 366 | threadStartX += nX % nThreads; |
| 367 | threadEndX += nX % nThreads; |
| 368 | } |
| 369 | } |
| 370 | |
| 371 | for (int bX = oX + threadStartX; bX < threadEndX + oX; bX += blk[0]) { |
| 372 | for (int bY = oY; bY < nY + oY; bY += blk[1]) { |
| 373 | for (int bZ = oZ; bZ < nZ + oZ; bZ += blk[2]) { |
| 374 | |
| 375 | // Must do everything here, else it would break collapse. |
| 376 | int eZ = MIN(bZ + blk[2], nZ + oZ); |
| 377 | int eY = MIN(bY + blk[1], nY + oY); |
| 378 | int eX = MIN(bX + blk[0], threadEndX + oX); |
| 379 | |
| 380 | for (int x = bX; x < eX; ++x) { |
| 381 | for (int y = bY; y < eY; ++y) { |
| 382 | #ifdef INTEL_OPT_DIRECTIVES |
| 383 | #pragma ivdep |
| 384 | #pragma vector always |
| 385 | #pragma simd |
| 386 | #endif |
| 387 | for (int z = bZ; z < eZ; ++z) { // LOOP aa-odd |
| 388 | |
| 389 | #define I(x, y, z, dir) P_INDEX_5(gDims, (x), (y), (z), (dir)) |
| 390 | |
| 391 | // Load PDFs of local cell: pdf_N = src[I(x, y, z, D3Q19_N)]; ... |
| 392 | #define X(name, idx, idxinv, _x, _y, _z) JOIN(pdf_,name) = src[I(x - _x, y - _y, z - _z, idxinv)]; |
| 393 | D3Q19_LIST |
| 394 | #undef X |
| 395 | |
| 396 | |
| 397 | // #define LID_DRIVEN_CAVITY |
| 398 | |
| 399 | #ifdef LID_DRIVEN_CAVITY |
| 400 | |
| 401 | if (z == nZ - 4 + oZ && x > 3 + oX && x < (nX - 4 + oX) && y > 3 + oY && y < (nY - 4 + oY)) { |
| 402 | ux = F(0.1) * F(0.5)77; |
| 403 | uy = F(0.0); |
| 404 | uz = F(0.0); |
| 405 | |
| 406 | } else { |
| 407 | #endif |
| 408 | ux = pdf_E + pdf_NE + pdf_SE + pdf_TE + pdf_BE - |
| 409 | pdf_W - pdf_NW - pdf_SW - pdf_TW - pdf_BW; |
| 410 | uy = pdf_N + pdf_NE + pdf_NW + pdf_TN + pdf_BN - |
| 411 | pdf_S - pdf_SE - pdf_SW - pdf_TS - pdf_BS; |
| 412 | uz = pdf_T + pdf_TE + pdf_TW + pdf_TN + pdf_TS - |
| 413 | pdf_B - pdf_BE - pdf_BW - pdf_BN - pdf_BS; |
| 414 | #ifdef LID_DRIVEN_CAVITY |
| 415 | } |
| 416 | #endif |
| 417 | |
| 418 | dens = pdf_C + |
| 419 | pdf_N + pdf_E + pdf_S + pdf_W + |
| 420 | pdf_NE + pdf_SE + pdf_SW + pdf_NW + |
| 421 | pdf_T + pdf_TN + pdf_TE + pdf_TS + pdf_TW + |
| 422 | pdf_B + pdf_BN + pdf_BE + pdf_BS + pdf_BW; |
| 423 | |
| 424 | dir_indep_trm = dens - (ux * ux + uy * uy + uz * uz)*F(3.0)/F(2.0); |
| 425 | |
| 426 | // direction: w_0 |
| 427 | src[I(x, y, z, D3Q19_C)] = pdf_C - omegaEven*(pdf_C - w_0*dir_indep_trm); |
| 428 | |
| 429 | // direction: w_1 |
| 430 | w_1_indep = w_1*dir_indep_trm; |
| 431 | |
| 432 | ui = uy; |
| 433 | evenPart = omegaEven*( F(0.5)*(pdf_N + pdf_S) - ui*ui*w_1_nine_half - w_1_indep ); |
| 434 | oddPart = omegaOdd*(F(0.5)*(pdf_N - pdf_S) - ui*w_1_x3 ); |
| 435 | src[I(x, y + 1, z, D3Q19_N)] = pdf_N - evenPart - oddPart; |
| 436 | src[I(x, y - 1, z, D3Q19_S)] = pdf_S - evenPart + oddPart; |
| 437 | |
| 438 | ui = ux; |
| 439 | evenPart = omegaEven*( F(0.5)*(pdf_E + pdf_W) - ui*ui*w_1_nine_half - w_1_indep ); |
| 440 | oddPart = omegaOdd*(F(0.5)*(pdf_E - pdf_W) - ui*w_1_x3 ); |
| 441 | src[I(x + 1, y, z, D3Q19_E)] = pdf_E - evenPart - oddPart; |
| 442 | src[I(x - 1, y, z, D3Q19_W)] = pdf_W - evenPart + oddPart; |
| 443 | |
| 444 | ui = uz; |
| 445 | evenPart = omegaEven*( F(0.5)*(pdf_T + pdf_B) - ui*ui*w_1_nine_half - w_1_indep ); |
| 446 | oddPart = omegaOdd*(F(0.5)*(pdf_T - pdf_B) - ui*w_1_x3 ); |
| 447 | src[I(x, y, z + 1, D3Q19_T)] = pdf_T - evenPart - oddPart; |
| 448 | src[I(x, y, z - 1, D3Q19_B)] = pdf_B - evenPart + oddPart; |
| 449 | |
| 450 | // direction: w_2 |
| 451 | w_2_indep = w_2*dir_indep_trm; |
| 452 | |
| 453 | ui = -ux + uy; |
| 454 | evenPart = omegaEven*( F(0.5)*(pdf_NW + pdf_SE) - ui*ui*w_2_nine_half - w_2_indep ); |
| 455 | oddPart = omegaOdd*(F(0.5)*(pdf_NW - pdf_SE) - ui*w_2_x3 ); |
| 456 | src[I(x - 1, y + 1, z, D3Q19_NW)] = pdf_NW - evenPart - oddPart; |
| 457 | src[I(x + 1, y - 1, z, D3Q19_SE)] = pdf_SE - evenPart + oddPart; |
| 458 | |
| 459 | ui = ux + uy; |
| 460 | evenPart = omegaEven*( F(0.5)*(pdf_NE + pdf_SW) - ui*ui*w_2_nine_half - w_2_indep ); |
| 461 | oddPart = omegaOdd*(F(0.5)*(pdf_NE - pdf_SW) - ui*w_2_x3 ); |
| 462 | src[I(x + 1, y + 1, z, D3Q19_NE)] = pdf_NE - evenPart - oddPart; |
| 463 | src[I(x - 1, y - 1, z, D3Q19_SW)] = pdf_SW - evenPart + oddPart; |
| 464 | |
| 465 | ui = -ux + uz; |
| 466 | evenPart = omegaEven*( F(0.5)*(pdf_TW + pdf_BE) - ui*ui*w_2_nine_half - w_2_indep ); |
| 467 | oddPart = omegaOdd*(F(0.5)*(pdf_TW - pdf_BE) - ui*w_2_x3 ); |
| 468 | src[I(x - 1, y, z + 1, D3Q19_TW)] = pdf_TW - evenPart - oddPart; |
| 469 | src[I(x + 1, y, z - 1, D3Q19_BE)] = pdf_BE - evenPart + oddPart; |
| 470 | |
| 471 | ui = ux + uz; |
| 472 | evenPart = omegaEven*( F(0.5)*(pdf_TE + pdf_BW) - ui*ui*w_2_nine_half - w_2_indep ); |
| 473 | oddPart = omegaOdd*(F(0.5)*(pdf_TE - pdf_BW) - ui*w_2_x3 ); |
| 474 | src[I(x + 1, y, z + 1, D3Q19_TE)] = pdf_TE - evenPart - oddPart; |
| 475 | src[I(x - 1, y, z - 1, D3Q19_BW)] = pdf_BW - evenPart + oddPart; |
| 476 | |
| 477 | ui = -uy + uz; |
| 478 | evenPart = omegaEven*( F(0.5)*(pdf_TS + pdf_BN) - ui*ui*w_2_nine_half - w_2_indep ); |
| 479 | oddPart = omegaOdd*(F(0.5)*(pdf_TS - pdf_BN) - ui*w_2_x3 ); |
| 480 | src[I(x, y - 1, z + 1, D3Q19_TS)] = pdf_TS - evenPart - oddPart; |
| 481 | src[I(x, y + 1, z - 1, D3Q19_BN)] = pdf_BN - evenPart + oddPart; |
| 482 | |
| 483 | ui = uy + uz; |
| 484 | evenPart = omegaEven*( F(0.5)*(pdf_TN + pdf_BS) - ui*ui*w_2_nine_half - w_2_indep ); |
| 485 | oddPart = omegaOdd*(F(0.5)*(pdf_TN - pdf_BS) - ui*w_2_x3 ); |
| 486 | src[I(x, y + 1, z + 1, D3Q19_TN)] = pdf_TN - evenPart - oddPart; |
| 487 | src[I(x, y - 1, z - 1, D3Q19_BS)] = pdf_BS - evenPart + oddPart; |
| 488 | |
| 489 | |
| 490 | #undef I |
| 491 | } } } // z, y, x (from inner to outer) |
| 492 | } } } // z, y, x (from inner to outer) |
| 493 | } // loop over threads |
| 494 | |
| 495 | // }}} |
| 496 | |
| 497 | // Stop counters before bounce back. Else computing loop balance will be incorrect. |
| 498 | |
| 499 | X_LIKWID_STOP("aa-odd"); |
| 500 | |
| 501 | // Fixup bounce back PDFs. |
| 502 | #ifdef _OPENMP |
| 503 | #pragma omp parallel for default(none) \ |
| 504 | shared(kd, src) |
| 505 | #endif |
| 506 | #ifdef INTEL_OPT_DIRECTIVES |
| 507 | #pragma ivdep |
| 508 | #endif |
| 509 | for (int i = 0; i < kd->nBounceBackPdfs; ++i) { |
| 510 | src[kd->BounceBackPdfsDst[i]] = src[kd->BounceBackPdfsSrc[i]]; |
| 511 | } |
| 512 | |
| 513 | // save current iteration |
| 514 | kda->Iteration = iter + 1; |
| 515 | |
| 516 | #ifdef VERIFICATION |
| 517 | kd->PdfsActive = src; |
| 518 | KernelAddBodyForce(kd, ld, cd); |
| 519 | #endif |
| 520 | |
| 521 | #ifdef VTK_OUTPUT |
| 522 | if (cd->VtkOutput && (iter + 1 % cd->VtkModulus) == 0) { |
| 523 | kd->PdfsActive = src; |
| 524 | VtkWrite(ld, kd, cd, iter + 1); |
| 525 | } |
| 526 | #endif |
| 527 | |
| 528 | #ifdef STATISTICS |
| 529 | kd->PdfsActive = src; |
| 530 | KernelStatistics(kd, ld, cd, iter + 1); |
| 531 | #endif // }}} |
| 532 | |
| 533 | |
| 534 | } // for (int iter = 0; ... |
| 535 | |
| 536 | X_KERNEL_END(kernelData); |
| 537 | |
| 538 | #ifdef VTK_OUTPUT |
| 539 | |
| 540 | if (cd->VtkOutput) { |
| 541 | kd->PdfsActive = src; |
| 542 | VtkWrite(ld, kd, cd, maxIterations); |
| 543 | } |
| 544 | |
| 545 | #endif |
| 546 | |
| 547 | return; |
| 548 | } |
| 549 | |