| 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 "BenchKernelD3Q19AaVecCommon.h" |
| 28 | |
| 29 | #include "Memory.h" |
| 30 | #include "Vtk.h" |
| 31 | #include "LikwidIf.h" |
| 32 | #include "Vector.h" |
| 33 | #include "Vector.h" |
| 34 | |
| 35 | #include <inttypes.h> |
| 36 | #include <math.h> |
| 37 | |
| 38 | #ifdef _OPENMP |
| 39 | #include <omp.h> |
| 40 | #endif |
| 41 | |
| 42 | static void KernelEven(LatticeDesc * ld, KernelData * kd, CaseData * cd); |
| 43 | static void KernelOdd( LatticeDesc * ld, KernelData * kd, CaseData * cd); |
| 44 | |
| 45 | #if 0 // {{{ |
| 46 | void DumpPdfs(LatticeDesc * ld, KernelData * kd, int zStart, int zStop, int iter, const char * prefix) |
| 47 | { |
| 48 | return; |
| 49 | |
| 50 | int * lDims = ld->Dims; |
| 51 | int * gDims = kd->GlobalDims; |
| 52 | |
| 53 | int nX = gDims[0]; |
| 54 | int nY = gDims[1]; |
| 55 | int nZ = gDims[2]; |
| 56 | |
| 57 | PdfT pdfs[N_D3Q19]; |
| 58 | |
| 59 | int localZStart = zStart; |
| 60 | int localZStop = zStop; |
| 61 | |
| 62 | if (localZStart == -1) localZStart = 0; |
| 63 | if (localZStop == -1) localZStop = gDims[2] - 1; |
| 64 | |
| 65 | printf("D iter: %d\n", iter); |
| 66 | |
| 67 | for (int dir = 0; dir < 19; ++dir) { |
| 68 | for (int z = localZStop; z >= localZStart; --z) { |
| 69 | printf("D [%2d][%2d][%s] plane % 2d\n", iter, dir, prefix, z); |
| 70 | |
| 71 | for(int y = 0; y < nY; ++y) { |
| 72 | printf("D [%2d][%2d][%s] %2d ", iter, dir, prefix, y); |
| 73 | |
| 74 | for(int x = 0; x < nX; ++x) { |
| 75 | |
| 76 | if (1) { // ld->Lattice[L_INDEX_4(ld->Dims, x, y, z)] != LAT_CELL_OBSTACLE) { |
| 77 | |
| 78 | #define I(x, y, z, dir) P_INDEX_5(gDims, (x), (y), (z), (dir)) |
| 79 | pdfs[dir] = kd->PdfsActive[I(x, y, z, dir)]; |
| 80 | #undef I |
| 81 | // kd->GetNode(kd, x, y, z, pdfs); |
| 82 | } |
| 83 | else { |
| 84 | pdfs[dir] = -F(1.0); |
| 85 | } |
| 86 | |
| 87 | printf("%.16e ", pdfs[dir]); |
| 88 | } |
| 89 | |
| 90 | printf("\n"); |
| 91 | } |
| 92 | } |
| 93 | } |
| 94 | } |
| 95 | #endif // }}} |
| 96 | |
| 97 | void FNAME(D3Q19AaVecKernel)(LatticeDesc * ld, KernelData * kd, CaseData * cd) |
| 98 | { |
| 99 | Assert(ld != NULL); |
| 100 | Assert(kd != NULL); |
| 101 | Assert(cd != NULL); |
| 102 | |
| 103 | Assert(cd->Omega > F(0.0)); |
| 104 | Assert(cd->Omega < F(2.0)); |
| 105 | |
| 106 | KernelDataAa * kda = KDA(kd); |
| 107 | |
| 108 | PdfT * src = kd->PdfsActive; |
| 109 | |
| 110 | int maxIterations = cd->MaxIterations; |
| 111 | |
| 112 | #ifdef VTK_OUTPUT |
| 113 | if (cd->VtkOutput) { |
| 114 | kd->PdfsActive = src; |
| 115 | VtkWrite(ld, kd, cd, -1); |
| 116 | } |
| 117 | #endif |
| 118 | |
| 119 | #ifdef STATISTICS |
| 120 | kd->PdfsActive = src; |
| 121 | KernelStatistics(kd, ld, cd, 0); |
| 122 | #endif |
| 123 | |
| 124 | Assert((maxIterations % 2) == 0); |
| 125 | |
| 126 | X_KERNEL_START(kd); |
| 127 | |
| 128 | for (int iter = 0; iter < maxIterations; iter += 2) { |
| 129 | |
| 130 | // -------------------------------------------------------------------- |
| 131 | // even time step |
| 132 | // -------------------------------------------------------------------- |
| 133 | |
| 134 | X_LIKWID_START("aa-vec-even"); |
| 135 | |
| 136 | #pragma omp parallel |
| 137 | { |
| 138 | KernelEven(ld, kd, cd); |
| 139 | } |
| 140 | |
| 141 | X_LIKWID_STOP("aa-vec-even"); |
| 142 | |
| 143 | // Fixup bounce back PDFs. |
| 144 | #ifdef _OPENMP |
| 145 | #pragma omp parallel for default(none) \ |
| 146 | shared(kd, src) |
| 147 | #endif |
| 148 | for (int i = 0; i < kd->nBounceBackPdfs; ++i) { |
| 149 | src[kd->BounceBackPdfsSrc[i]] = src[kd->BounceBackPdfsDst[i]]; |
| 150 | } |
| 151 | |
| 152 | // save current iteration |
| 153 | kda->Iteration = iter; |
| 154 | |
| 155 | #ifdef VERIFICATION |
| 156 | kd->PdfsActive = src; |
| 157 | KernelAddBodyForce(kd, ld, cd); |
| 158 | #endif |
| 159 | |
| 160 | #ifdef VTK_OUTPUT |
| 161 | if (cd->VtkOutput && (iter % cd->VtkModulus) == 0) { |
| 162 | kd->PdfsActive = src; |
| 163 | VtkWrite(ld, kd, cd, iter); |
| 164 | } |
| 165 | #endif |
| 166 | |
| 167 | #ifdef STATISTICS |
| 168 | kd->PdfsActive = src; |
| 169 | KernelStatistics(kd, ld, cd, iter); |
| 170 | #endif |
| 171 | |
| 172 | // -------------------------------------------------------------------- |
| 173 | // odd time step |
| 174 | // -------------------------------------------------------------------- |
| 175 | |
| 176 | X_LIKWID_START("aa-vec-odd"); |
| 177 | |
| 178 | #pragma omp parallel |
| 179 | { |
| 180 | KernelOdd(ld, kd, cd); |
| 181 | } |
| 182 | |
| 183 | // Stop counters before bounce back. Else computing loop balance will |
| 184 | // be incorrect. |
| 185 | |
| 186 | X_LIKWID_STOP("aa-vec-odd"); |
| 187 | |
| 188 | // Fixup bounce back PDFs. |
| 189 | #ifdef _OPENMP |
| 190 | #pragma omp parallel for default(none) \ |
| 191 | shared(kd, src) |
| 192 | #endif |
| 193 | for (int i = 0; i < kd->nBounceBackPdfs; ++i) { |
| 194 | src[kd->BounceBackPdfsDst[i]] = src[kd->BounceBackPdfsSrc[i]]; |
| 195 | } |
| 196 | |
| 197 | // save current iteration |
| 198 | kda->Iteration = iter + 1; |
| 199 | |
| 200 | #ifdef VERIFICATION |
| 201 | kd->PdfsActive = src; |
| 202 | KernelAddBodyForce(kd, ld, cd); |
| 203 | #endif |
| 204 | |
| 205 | #ifdef VTK_OUTPUT |
| 206 | if (cd->VtkOutput && ((iter + 1) % cd->VtkModulus) == 0) { |
| 207 | kd->PdfsActive = src; |
| 208 | VtkWrite(ld, kd, cd, iter + 1); |
| 209 | } |
| 210 | #endif |
| 211 | |
| 212 | #ifdef STATISTICS |
| 213 | kd->PdfsActive = src; |
| 214 | KernelStatistics(kd, ld, cd, iter + 1); |
| 215 | #endif // }}} |
| 216 | |
| 217 | |
| 218 | } // for (int iter = 0; ... |
| 219 | |
| 220 | X_KERNEL_END(kd); |
| 221 | |
| 222 | #ifdef VTK_OUTPUT |
| 223 | |
| 224 | if (cd->VtkOutput) { |
| 225 | kd->PdfsActive = src; |
| 226 | VtkWrite(ld, kd, cd, maxIterations); |
| 227 | } |
| 228 | |
| 229 | #endif |
| 230 | |
| 231 | return; |
| 232 | } |
| 233 | |
| 234 | static void KernelEven(LatticeDesc * ld, KernelData * kd, CaseData * cd) // {{{ |
| 235 | { |
| 236 | Assert(ld != NULL); |
| 237 | Assert(kd != NULL); |
| 238 | Assert(cd != NULL); |
| 239 | |
| 240 | Assert(cd->Omega > F(0.0)); |
| 241 | Assert(cd->Omega < F(2.0)); |
| 242 | |
| 243 | KernelDataAa * kda = KDA(kd); |
| 244 | |
| 245 | int nX = ld->Dims[0]; |
| 246 | int nY = ld->Dims[1]; |
| 247 | int nZ = ld->Dims[2]; |
| 248 | |
| 249 | int * gDims = kd->GlobalDims; |
| 250 | |
| 251 | int oX = kd->Offsets[0]; |
| 252 | int oY = kd->Offsets[1]; |
| 253 | int oZ = kd->Offsets[2]; |
| 254 | |
| 255 | int blk[3]; |
| 256 | blk[0] = kda->Blk[0]; |
| 257 | blk[1] = kda->Blk[1]; |
| 258 | blk[2] = kda->Blk[2]; |
| 259 | |
| 260 | PdfT omega = cd->Omega; |
| 261 | PdfT omegaEven = omega; |
| 262 | |
| 263 | PdfT magicParam = F(1.0) / F(12.0); |
| 264 | PdfT omegaOdd = F(1.0) / (F(0.5) + magicParam / (F(1.0) / omega - F(0.5))); |
| 265 | |
| 266 | const PdfT w_0 = F(1.0) / F(3.0); |
| 267 | const PdfT w_1 = F(1.0) / F(18.0); |
| 268 | const PdfT w_2 = F(1.0) / F(36.0); |
| 269 | |
| 270 | const PdfT w_1_x3 = w_1 * F(3.0); const PdfT w_1_nine_half = w_1 * F(9.0) / F(2.0); |
| 271 | const PdfT w_2_x3 = w_2 * F(3.0); const PdfT w_2_nine_half = w_2 * F(9.0) / F(2.0); |
| 272 | |
| 273 | |
| 274 | VPDFT VONE_HALF = VSET(F(0.5)); |
| 275 | VPDFT VTHREE_HALF = VSET(F(3.0) / F(2.0)); |
| 276 | |
| 277 | VPDFT vw_1_indep, vw_2_indep; |
| 278 | VPDFT vw_0 = VSET(w_0); |
| 279 | VPDFT vw_1 = VSET(w_1); |
| 280 | VPDFT vw_2 = VSET(w_2); |
| 281 | |
| 282 | VPDFT vw_1_x3 = VSET(w_1_x3); |
| 283 | VPDFT vw_2_x3 = VSET(w_2_x3); |
| 284 | VPDFT vw_1_nine_half = VSET(w_1_nine_half); |
| 285 | VPDFT vw_2_nine_half = VSET(w_2_nine_half); |
| 286 | |
| 287 | VPDFT vui, vux, vuy, vuz, vdens; |
| 288 | |
| 289 | VPDFT vevenPart, voddPart, vdir_indep_trm; |
| 290 | |
| 291 | VPDFT vomegaEven = VSET(omegaEven); |
| 292 | VPDFT vomegaOdd = VSET(omegaOdd); |
| 293 | |
| 294 | // Declare pdf_N, pdf_E, pdf_S, pdf_W, ... |
| 295 | #define X(name, idx, idxinv, x, y, z) VPDFT JOIN(vpdf_,name); |
| 296 | D3Q19_LIST |
| 297 | #undef X |
| 298 | |
| 299 | PdfT * src = kd->Pdfs[0]; |
| 300 | |
| 301 | int nThreads = 1; |
| 302 | int threadId = 0; |
| 303 | |
| 304 | #ifdef _OPENMP |
| 305 | nThreads = omp_get_max_threads(); |
| 306 | threadId = omp_get_thread_num(); |
| 307 | #endif |
| 308 | |
| 309 | // TODO: Currently only a 1-D decomposition is applied. For achritectures |
| 310 | // with a lot of cores we want at least 2-D. |
| 311 | |
| 312 | int threadStartX = nX / nThreads * threadId; |
| 313 | int threadEndX = nX / nThreads * (threadId + 1); |
| 314 | |
| 315 | if (nX % nThreads > 0) { |
| 316 | if (nX % nThreads > threadId) { |
| 317 | threadStartX += threadId; |
| 318 | threadEndX += threadId + 1; |
| 319 | } |
| 320 | else { |
| 321 | threadStartX += nX % nThreads; |
| 322 | threadEndX += nX % nThreads; |
| 323 | } |
| 324 | } |
| 325 | |
| 326 | AssertMsg((blk[2] % VSIZE == 0) || blk[2] >= nZ, "Blocking in z direction must be a multiple of VSIZE = %d or larger than z dimension.", VSIZE); |
| 327 | |
| 328 | for (int bX = oX + threadStartX; bX < threadEndX + oX; bX += blk[0]) { |
| 329 | for (int bY = oY; bY < nY + oY; bY += blk[1]) { |
| 330 | for (int bZ = oZ; bZ < nZ + oZ; bZ += blk[2]) { |
| 331 | |
| 332 | int eX = MIN(bX + blk[0], threadEndX + oX); |
| 333 | int eY = MIN(bY + blk[1], nY + oY); |
| 334 | int eZ = MIN(bZ + blk[2], nZ + oZ); |
| 335 | |
| 336 | for (int x = bX; x < eX; x += 1) { |
| 337 | for (int y = bY; y < eY; y += 1) { |
| 338 | for (int z = bZ; z < eZ; z += VSIZE) { // LOOP aa-vec-even |
| 339 | |
| 340 | #define I(x, y, z, dir) P_INDEX_5(gDims, (x), (y), (z), (dir)) |
| 341 | |
| 342 | // Load PDFs of local cell: pdf_N = src[I(x, y, z, D3Q19_N)]; ... |
| 343 | #define X(name, idx, idxinv, _x, _y, _z) JOIN(vpdf_,name) = VLDU(&src[I(x, y, z, idx)]); |
| 344 | D3Q19_LIST |
| 345 | #undef X |
| 346 | |
| 347 | |
| 348 | vux = VSUB(VSUB(VSUB(VSUB(VSUB(VADD(VADD(vpdf_E,VADD(vpdf_NE,vpdf_SE)),VADD(vpdf_TE,vpdf_BE)),vpdf_W),vpdf_NW),vpdf_SW),vpdf_TW),vpdf_BW); |
| 349 | vuy = VSUB(VSUB(VSUB(VSUB(VSUB(VADD(VADD(vpdf_N,VADD(vpdf_NE,vpdf_NW)),VADD(vpdf_TN,vpdf_BN)),vpdf_S),vpdf_SE),vpdf_SW),vpdf_TS),vpdf_BS); |
| 350 | vuz = VSUB(VSUB(VSUB(VSUB(VSUB(VADD(VADD(vpdf_T,VADD(vpdf_TE,vpdf_TW)),VADD(vpdf_TN,vpdf_TS)),vpdf_B),vpdf_BE),vpdf_BW),vpdf_BN),vpdf_BS); |
| 351 | |
| 352 | vdens = VADD(VADD(VADD(VADD(VADD(VADD(VADD(VADD(VADD(vpdf_C,VADD(vpdf_N,vpdf_E)),VADD(vpdf_S,vpdf_W)),VADD(vpdf_NE,vpdf_SE)), |
| 353 | VADD(vpdf_SW,vpdf_NW)),VADD(vpdf_T,vpdf_TN)),VADD(vpdf_TE,vpdf_TS)),VADD(vpdf_TW,vpdf_B)), |
| 354 | VADD(vpdf_BN,vpdf_BE)),VADD(vpdf_BS,vpdf_BW)); |
| 355 | |
| 356 | vdir_indep_trm = VSUB(vdens,VMUL(VADD(VADD(VMUL(vux,vux),VMUL(vuy,vuy)),VMUL(vuz,vuz)),VTHREE_HALF)); |
| 357 | |
| 358 | VSTU(&src[I(x, y, z, D3Q19_C)],VSUB(vpdf_C,VMUL(vomegaEven,VSUB(vpdf_C,VMUL(vw_0,vdir_indep_trm))))); |
| 359 | |
| 360 | vw_1_indep = VMUL(vw_1,vdir_indep_trm); |
| 361 | |
| 362 | vui = vuy; |
| 363 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_N,vpdf_S)),VMUL(vui,VMUL(vui,vw_1_nine_half))),vw_1_indep)); |
| 364 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_N,vpdf_S)),VMUL(vui,vw_1_x3))); |
| 365 | VSTU(&src[I(x, y, z, D3Q19_S)],VSUB(VSUB(vpdf_N,vevenPart),voddPart)); |
| 366 | VSTU(&src[I(x, y, z, D3Q19_N)],VADD(VSUB(vpdf_S,vevenPart),voddPart)); |
| 367 | |
| 368 | vui = vux; |
| 369 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_E,vpdf_W)),VMUL(vui,VMUL(vui,vw_1_nine_half))),vw_1_indep)); |
| 370 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_E,vpdf_W)),VMUL(vui,vw_1_x3))); |
| 371 | VSTU(&src[I(x, y, z, D3Q19_W)],VSUB(VSUB(vpdf_E,vevenPart),voddPart)); |
| 372 | VSTU(&src[I(x, y, z, D3Q19_E)],VADD(VSUB(vpdf_W,vevenPart),voddPart)); |
| 373 | |
| 374 | vui = vuz; |
| 375 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_T,vpdf_B)),VMUL(vui,VMUL(vui,vw_1_nine_half))),vw_1_indep)); |
| 376 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_T,vpdf_B)),VMUL(vui,vw_1_x3))); |
| 377 | VSTU(&src[I(x, y, z, D3Q19_B)],VSUB(VSUB(vpdf_T,vevenPart),voddPart)); |
| 378 | VSTU(&src[I(x, y, z, D3Q19_T)],VADD(VSUB(vpdf_B,vevenPart),voddPart)); |
| 379 | |
| 380 | vw_2_indep = VMUL(vw_2,vdir_indep_trm); |
| 381 | |
| 382 | vui = VSUB(vuy,vux); |
| 383 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_NW,vpdf_SE)),VMUL(vui,VMUL(vui,vw_2_nine_half))),vw_2_indep)); |
| 384 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_NW,vpdf_SE)),VMUL(vui,vw_2_x3))); |
| 385 | VSTU(&src[I(x, y, z, D3Q19_SE)],VSUB(VSUB(vpdf_NW,vevenPart),voddPart)); |
| 386 | VSTU(&src[I(x, y, z, D3Q19_NW)],VADD(VSUB(vpdf_SE,vevenPart),voddPart)); |
| 387 | |
| 388 | vui = VADD(vux,vuy); |
| 389 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_NE,vpdf_SW)),VMUL(vui,VMUL(vui,vw_2_nine_half))),vw_2_indep)); |
| 390 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_NE,vpdf_SW)),VMUL(vui,vw_2_x3))); |
| 391 | VSTU(&src[I(x, y, z, D3Q19_SW)],VSUB(VSUB(vpdf_NE,vevenPart),voddPart)); |
| 392 | VSTU(&src[I(x, y, z, D3Q19_NE)],VADD(VSUB(vpdf_SW,vevenPart),voddPart)); |
| 393 | |
| 394 | vui = VSUB(vuz,vux); |
| 395 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_TW,vpdf_BE)),VMUL(vui,VMUL(vui,vw_2_nine_half))),vw_2_indep)); |
| 396 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_TW,vpdf_BE)),VMUL(vui,vw_2_x3))); |
| 397 | VSTU(&src[I(x, y, z, D3Q19_BE)],VSUB(VSUB(vpdf_TW,vevenPart),voddPart)); |
| 398 | VSTU(&src[I(x, y, z, D3Q19_TW)],VADD(VSUB(vpdf_BE,vevenPart),voddPart)); |
| 399 | |
| 400 | vui = VADD(vux,vuz); |
| 401 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_TE,vpdf_BW)),VMUL(vui,VMUL(vui,vw_2_nine_half))),vw_2_indep)); |
| 402 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_TE,vpdf_BW)),VMUL(vui,vw_2_x3))); |
| 403 | VSTU(&src[I(x, y, z, D3Q19_BW)],VSUB(VSUB(vpdf_TE,vevenPart),voddPart)); |
| 404 | VSTU(&src[I(x, y, z, D3Q19_TE)],VADD(VSUB(vpdf_BW,vevenPart),voddPart)); |
| 405 | |
| 406 | vui = VSUB(vuz,vuy); |
| 407 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_TS,vpdf_BN)),VMUL(vui,VMUL(vui,vw_2_nine_half))),vw_2_indep)); |
| 408 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_TS,vpdf_BN)),VMUL(vui,vw_2_x3))); |
| 409 | VSTU(&src[I(x, y, z, D3Q19_BN)],VSUB(VSUB(vpdf_TS,vevenPart),voddPart)); |
| 410 | VSTU(&src[I(x, y, z, D3Q19_TS)],VADD(VSUB(vpdf_BN,vevenPart),voddPart)); |
| 411 | |
| 412 | vui = VADD(vuy,vuz); |
| 413 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_TN,vpdf_BS)),VMUL(vui,VMUL(vui,vw_2_nine_half))),vw_2_indep)); |
| 414 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_TN,vpdf_BS)),VMUL(vui,vw_2_x3))); |
| 415 | VSTU(&src[I(x, y, z, D3Q19_BS)],VSUB(VSUB(vpdf_TN,vevenPart),voddPart)); |
| 416 | VSTU(&src[I(x, y, z, D3Q19_TN)],VADD(VSUB(vpdf_BS,vevenPart),voddPart)); |
| 417 | |
| 418 | #undef I |
| 419 | } } } // x, y, z |
| 420 | } } } // blocked x, y, z |
| 421 | |
| 422 | |
| 423 | |
| 424 | return; |
| 425 | } // }}} |
| 426 | |
| 427 | |
| 428 | static void KernelOdd(LatticeDesc * ld, KernelData * kd, CaseData * cd) // {{{ |
| 429 | { |
| 430 | Assert(ld != NULL); |
| 431 | Assert(kd != NULL); |
| 432 | Assert(cd != NULL); |
| 433 | |
| 434 | Assert(cd->Omega > F(0.0)); |
| 435 | Assert(cd->Omega < F(2.0)); |
| 436 | |
| 437 | KernelDataAa * kda = KDA(kd); |
| 438 | |
| 439 | int nX = ld->Dims[0]; |
| 440 | int nY = ld->Dims[1]; |
| 441 | int nZ = ld->Dims[2]; |
| 442 | |
| 443 | int * gDims = kd->GlobalDims; |
| 444 | |
| 445 | int oX = kd->Offsets[0]; |
| 446 | int oY = kd->Offsets[1]; |
| 447 | int oZ = kd->Offsets[2]; |
| 448 | |
| 449 | int blk[3]; |
| 450 | blk[0] = kda->Blk[0]; |
| 451 | blk[1] = kda->Blk[1]; |
| 452 | blk[2] = kda->Blk[2]; |
| 453 | |
| 454 | PdfT omega = cd->Omega; |
| 455 | PdfT omegaEven = omega; |
| 456 | |
| 457 | PdfT magicParam = F(1.0) / F(12.0); |
| 458 | PdfT omegaOdd = F(1.0) / (F(0.5) + magicParam / (F(1.0) / omega - F(0.5))); |
| 459 | |
| 460 | const PdfT w_0 = F(1.0) / F(3.0); |
| 461 | const PdfT w_1 = F(1.0) / F(18.0); |
| 462 | const PdfT w_2 = F(1.0) / F(36.0); |
| 463 | |
| 464 | const PdfT w_1_x3 = w_1 * F(3.0); const PdfT w_1_nine_half = w_1 * F(9.0) / F(2.0); |
| 465 | const PdfT w_2_x3 = w_2 * F(3.0); const PdfT w_2_nine_half = w_2 * F(9.0) / F(2.0); |
| 466 | |
| 467 | VPDFT VONE_HALF = VSET(F(0.5)); |
| 468 | VPDFT VTHREE_HALF = VSET(F(3.0) / F(2.0)); |
| 469 | |
| 470 | VPDFT vw_1_indep, vw_2_indep; |
| 471 | VPDFT vw_0 = VSET(w_0); |
| 472 | VPDFT vw_1 = VSET(w_1); |
| 473 | VPDFT vw_2 = VSET(w_2); |
| 474 | |
| 475 | VPDFT vw_1_x3 = VSET(w_1_x3); |
| 476 | VPDFT vw_2_x3 = VSET(w_2_x3); |
| 477 | VPDFT vw_1_nine_half = VSET(w_1_nine_half); |
| 478 | VPDFT vw_2_nine_half = VSET(w_2_nine_half); |
| 479 | |
| 480 | VPDFT vui, vux, vuy, vuz, vdens; |
| 481 | |
| 482 | VPDFT vevenPart, voddPart, vdir_indep_trm; |
| 483 | |
| 484 | VPDFT vomegaEven = VSET(omegaEven); |
| 485 | VPDFT vomegaOdd = VSET(omegaOdd); |
| 486 | |
| 487 | // Declare pdf_N, pdf_E, pdf_S, pdf_W, ... |
| 488 | #define X(name, idx, idxinv, x, y, z) VPDFT JOIN(vpdf_,name); |
| 489 | D3Q19_LIST |
| 490 | #undef X |
| 491 | |
| 492 | PdfT * src = kd->Pdfs[0]; |
| 493 | |
| 494 | int threadId = 0; |
| 495 | int nThreads = 1; |
| 496 | |
| 497 | #ifdef _OPENMP |
| 498 | threadId = omp_get_thread_num(); |
| 499 | nThreads = omp_get_max_threads(); |
| 500 | #endif |
| 501 | |
| 502 | // TODO: Currently only a 1-D decomposition is applied. For achritectures |
| 503 | // with a lot of cores we want at least 2-D. |
| 504 | int threadStartX = nX / nThreads * threadId; |
| 505 | int threadEndX = nX / nThreads * (threadId + 1); |
| 506 | |
| 507 | if (nX % nThreads > 0) { |
| 508 | if (nX % nThreads > threadId) { |
| 509 | threadStartX += threadId; |
| 510 | threadEndX += threadId + 1; |
| 511 | } |
| 512 | else { |
| 513 | threadStartX += nX % nThreads; |
| 514 | threadEndX += nX % nThreads; |
| 515 | } |
| 516 | } |
| 517 | |
| 518 | AssertMsg((blk[2] % VSIZE == 0) || blk[2] >= nZ, "Blocking in z direction must be a multiple of VSIZE = %d or larger than z dimension.", VSIZE); |
| 519 | |
| 520 | for (int bX = oX + threadStartX; bX < threadEndX + oX; bX += blk[0]) { |
| 521 | for (int bY = oY; bY < nY + oY; bY += blk[1]) { |
| 522 | for (int bZ = oZ; bZ < nZ + oZ; bZ += blk[2]) { |
| 523 | |
| 524 | int eX = MIN(bX + blk[0], threadEndX + oX); |
| 525 | int eY = MIN(bY + blk[1], nY + oY); |
| 526 | int eZ = MIN(bZ + blk[2], nZ + oZ); |
| 527 | |
| 528 | for (int x = bX; x < eX; ++x) { |
| 529 | for (int y = bY; y < eY; ++y) { |
| 530 | for (int z = bZ; z < eZ; z += VSIZE) { // LOOP aa-vec-odd |
| 531 | |
| 532 | #define I(x, y, z, dir) P_INDEX_5(gDims, (x), (y), (z), (dir)) |
| 533 | |
| 534 | |
| 535 | #define X(name, idx, idxinv, _x, _y, _z) JOIN(vpdf_,name) = VLDU(&src[I(x - _x, y - _y, z - _z, idxinv)]); |
| 536 | D3Q19_LIST |
| 537 | #undef X |
| 538 | |
| 539 | vux = VSUB(VSUB(VSUB(VSUB(VSUB(VADD(VADD(vpdf_E,VADD(vpdf_NE,vpdf_SE)),VADD(vpdf_TE,vpdf_BE)),vpdf_W),vpdf_NW),vpdf_SW),vpdf_TW),vpdf_BW); |
| 540 | vuy = VSUB(VSUB(VSUB(VSUB(VSUB(VADD(VADD(vpdf_N,VADD(vpdf_NE,vpdf_NW)),VADD(vpdf_TN,vpdf_BN)),vpdf_S),vpdf_SE),vpdf_SW),vpdf_TS),vpdf_BS); |
| 541 | vuz = VSUB(VSUB(VSUB(VSUB(VSUB(VADD(VADD(vpdf_T,VADD(vpdf_TE,vpdf_TW)),VADD(vpdf_TN,vpdf_TS)),vpdf_B),vpdf_BE),vpdf_BW),vpdf_BN),vpdf_BS); |
| 542 | |
| 543 | vdens = VADD(VADD(VADD(VADD(VADD(VADD(VADD(VADD(VADD(vpdf_C,VADD(vpdf_N,vpdf_E)),VADD(vpdf_S,vpdf_W)),VADD(vpdf_NE,vpdf_SE)), |
| 544 | VADD(vpdf_SW,vpdf_NW)),VADD(vpdf_T,vpdf_TN)),VADD(vpdf_TE,vpdf_TS)),VADD(vpdf_TW,vpdf_B)),VADD(vpdf_BN,vpdf_BE)),VADD(vpdf_BS,vpdf_BW)); |
| 545 | |
| 546 | vdir_indep_trm = VSUB(vdens,VMUL(VADD(VADD(VMUL(vux,vux),VMUL(vuy,vuy)),VMUL(vuz,vuz)),VTHREE_HALF)); |
| 547 | |
| 548 | VSTU(&src[I(x, y, z, D3Q19_C)],VSUB(vpdf_C,VMUL(vomegaEven,VSUB(vpdf_C,VMUL(vw_0,vdir_indep_trm))))); |
| 549 | |
| 550 | vw_1_indep = VMUL(vw_1,vdir_indep_trm); |
| 551 | |
| 552 | vui = vuy; |
| 553 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_N,vpdf_S)),VMUL(vui,VMUL(vui,vw_1_nine_half))),vw_1_indep)); |
| 554 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_N,vpdf_S)),VMUL(vui,vw_1_x3))); |
| 555 | VSTU(&src[I(x, y + 1, z, D3Q19_N)], VSUB(VSUB(vpdf_N,vevenPart),voddPart)); |
| 556 | VSTU(&src[I(x, y - 1, z, D3Q19_S)], VADD(VSUB(vpdf_S,vevenPart),voddPart)); |
| 557 | |
| 558 | vui = vux; |
| 559 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_E,vpdf_W)),VMUL(vui,VMUL(vui,vw_1_nine_half))),vw_1_indep)); |
| 560 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_E,vpdf_W)),VMUL(vui,vw_1_x3))); |
| 561 | VSTU(&src[I(x + 1, y, z, D3Q19_E)], VSUB(VSUB(vpdf_E,vevenPart),voddPart)); |
| 562 | VSTU(&src[I(x - 1, y, z, D3Q19_W)], VADD(VSUB(vpdf_W,vevenPart),voddPart)); |
| 563 | |
| 564 | vui = vuz; |
| 565 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_T,vpdf_B)),VMUL(vui,VMUL(vui,vw_1_nine_half))),vw_1_indep)); |
| 566 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_T,vpdf_B)),VMUL(vui,vw_1_x3))); |
| 567 | VSTU(&src[I(x, y, z + 1, D3Q19_T)], VSUB(VSUB(vpdf_T,vevenPart),voddPart)); |
| 568 | VSTU(&src[I(x, y, z - 1, D3Q19_B)], VADD(VSUB(vpdf_B,vevenPart),voddPart)); |
| 569 | |
| 570 | vw_2_indep = VMUL(vw_2,vdir_indep_trm); |
| 571 | |
| 572 | vui = VSUB(vuy,vux); |
| 573 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_NW,vpdf_SE)),VMUL(vui,VMUL(vui,vw_2_nine_half))),vw_2_indep)); |
| 574 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_NW,vpdf_SE)),VMUL(vui,vw_2_x3))); |
| 575 | VSTU(&src[I(x - 1, y + 1, z, D3Q19_NW)], VSUB(VSUB(vpdf_NW,vevenPart),voddPart)); |
| 576 | VSTU(&src[I(x + 1, y - 1, z, D3Q19_SE)], VADD(VSUB(vpdf_SE,vevenPart),voddPart)); |
| 577 | |
| 578 | vui = VADD(vux,vuy); |
| 579 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_NE,vpdf_SW)),VMUL(vui,VMUL(vui,vw_2_nine_half))),vw_2_indep)); |
| 580 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_NE,vpdf_SW)),VMUL(vui,vw_2_x3))); |
| 581 | VSTU(&src[I(x + 1, y + 1, z, D3Q19_NE)], VSUB(VSUB(vpdf_NE,vevenPart),voddPart)); |
| 582 | VSTU(&src[I(x - 1, y - 1, z, D3Q19_SW)], VADD(VSUB(vpdf_SW,vevenPart),voddPart)); |
| 583 | |
| 584 | vui = VSUB(vuz,vux); |
| 585 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_TW,vpdf_BE)),VMUL(vui,VMUL(vui,vw_2_nine_half))),vw_2_indep)); |
| 586 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_TW,vpdf_BE)),VMUL(vui,vw_2_x3))); |
| 587 | VSTU(&src[I(x - 1, y, z + 1, D3Q19_TW)], VSUB(VSUB(vpdf_TW,vevenPart),voddPart)); |
| 588 | VSTU(&src[I(x + 1, y, z - 1, D3Q19_BE)], VADD(VSUB(vpdf_BE,vevenPart),voddPart)); |
| 589 | |
| 590 | vui = VADD(vux,vuz); |
| 591 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_TE,vpdf_BW)),VMUL(vui,VMUL(vui,vw_2_nine_half))),vw_2_indep)); |
| 592 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_TE,vpdf_BW)),VMUL(vui,vw_2_x3))); |
| 593 | VSTU(&src[I(x + 1, y, z + 1, D3Q19_TE)], VSUB(VSUB(vpdf_TE,vevenPart),voddPart)); |
| 594 | VSTU(&src[I(x - 1, y, z - 1, D3Q19_BW)], VADD(VSUB(vpdf_BW,vevenPart),voddPart)); |
| 595 | |
| 596 | vui = VSUB(vuz,vuy); |
| 597 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_TS,vpdf_BN)),VMUL(vui,VMUL(vui,vw_2_nine_half))),vw_2_indep)); |
| 598 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_TS,vpdf_BN)),VMUL(vui,vw_2_x3))); |
| 599 | VSTU(&src[I(x, y - 1, z + 1, D3Q19_TS)], VSUB(VSUB(vpdf_TS,vevenPart),voddPart)); |
| 600 | VSTU(&src[I(x, y + 1, z - 1, D3Q19_BN)], VADD(VSUB(vpdf_BN,vevenPart),voddPart)); |
| 601 | |
| 602 | vui = VADD(vuy,vuz); |
| 603 | vevenPart = VMUL(vomegaEven,VSUB(VSUB(VMUL(VONE_HALF,VADD(vpdf_TN,vpdf_BS)),VMUL(vui,VMUL(vui,vw_2_nine_half))),vw_2_indep)); |
| 604 | voddPart = VMUL(vomegaOdd,VSUB(VMUL(VONE_HALF,VSUB(vpdf_TN,vpdf_BS)),VMUL(vui,vw_2_x3))); |
| 605 | VSTU(&src[I(x, y + 1, z + 1, D3Q19_TN)], VSUB(VSUB(vpdf_TN,vevenPart),voddPart)); |
| 606 | VSTU(&src[I(x, y - 1, z - 1, D3Q19_BS)], VADD(VSUB(vpdf_BS,vevenPart),voddPart)); |
| 607 | |
| 608 | #undef I |
| 609 | } } } // x, y, z |
| 610 | } } } // blocked x, y, z |
| 611 | |
| 612 | return; |
| 613 | |
| 614 | } // }}} |