Commit | Line | Data |
---|---|---|
10988083 MW |
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 "BenchKernelD3Q19Common.h" | |
28 | ||
29 | #include "Memory.h" | |
30 | #include "Vtk.h" | |
31 | ||
32 | #include <inttypes.h> | |
33 | #include <math.h> | |
34 | ||
35 | ||
36 | // Forward definition. | |
37 | void FNAME(D3Q19Kernel)(LatticeDesc * ld, struct KernelData_ * kd, CaseData * cd); | |
38 | ||
39 | void FNAME(D3Q19BlkKernel)(LatticeDesc * ld, struct KernelData_ * kd, CaseData * cd); | |
40 | ||
41 | ||
42 | ||
43 | static void FNAME(BcGetPdf)(KernelData * kd, int x, int y, int z, int dir, PdfT * pdf) | |
44 | { | |
45 | Assert(kd != NULL); | |
46 | Assert(kd->PdfsActive != NULL); | |
47 | Assert(kd->PdfsActive == kd->Pdfs[0] || kd->PdfsActive == kd->Pdfs[1]); | |
48 | Assert(pdf != NULL); | |
49 | ||
50 | Assert(x >= 0); | |
51 | Assert(y >= 0); | |
52 | Assert(z >= 0); | |
53 | Assert(x < kd->Dims[0]); | |
54 | Assert(y < kd->Dims[1]); | |
55 | Assert(z < kd->Dims[2]); | |
56 | Assert(dir >= 0); | |
57 | Assert(dir < N_D3Q19); | |
58 | ||
59 | int oX = kd->Offsets[0]; | |
60 | int oY = kd->Offsets[1]; | |
61 | int oZ = kd->Offsets[2]; | |
62 | ||
63 | #ifdef PROP_MODEL_PUSH | |
64 | int nx = x; | |
65 | int ny = y; | |
66 | int nz = z; | |
67 | #elif PROP_MODEL_PULL | |
68 | int nx = x - D3Q19_X[dir]; | |
69 | int ny = y - D3Q19_Y[dir]; | |
70 | int nz = z - D3Q19_Z[dir]; | |
71 | #endif | |
72 | ||
73 | #define I(x, y, z, dir) P_INDEX_5(kd->GlobalDims, (x), (y), (z), (dir)) | |
74 | *pdf = kd->PdfsActive[I(nx + oX, ny + oY, nz + oZ, dir)]; | |
75 | #undef I | |
76 | ||
77 | return; | |
78 | } | |
79 | ||
80 | static void FNAME(BcSetPdf)(KernelData * kd, int x, int y, int z, int dir, PdfT pdf) | |
81 | { | |
82 | Assert(kd != NULL); | |
83 | Assert(kd->PdfsActive != NULL); | |
84 | Assert(kd->PdfsActive == kd->Pdfs[0] || kd->PdfsActive == kd->Pdfs[1]); | |
85 | Assert(x >= 0); | |
86 | Assert(y >= 0); | |
87 | Assert(z >= 0); | |
88 | Assert(x < kd->Dims[0]); | |
89 | Assert(y < kd->Dims[1]); | |
90 | Assert(z < kd->Dims[2]); | |
91 | Assert(dir >= 0); | |
92 | Assert(dir < N_D3Q19); | |
93 | ||
94 | int oX = kd->Offsets[0]; | |
95 | int oY = kd->Offsets[1]; | |
96 | int oZ = kd->Offsets[2]; | |
97 | ||
98 | #ifdef PROP_MODEL_PUSH | |
99 | int nx = x; | |
100 | int ny = y; | |
101 | int nz = z; | |
102 | #elif PROP_MODEL_PULL | |
103 | int nx = x - D3Q19_X[dir]; | |
104 | int ny = y - D3Q19_Y[dir]; | |
105 | int nz = z - D3Q19_Z[dir]; | |
106 | #endif | |
107 | ||
108 | #define I(x, y, z, dir) P_INDEX_5(kd->GlobalDims, (x), (y), (z), (dir)) | |
109 | kd->PdfsActive[I(nx + oX, ny + oY, nz + oZ, dir)] = pdf; | |
110 | #undef I | |
111 | ||
112 | ||
113 | return; | |
114 | } | |
115 | ||
116 | ||
117 | static void FNAME(GetNode)(KernelData * kd, int x, int y, int z, PdfT * pdfs) | |
118 | { | |
119 | Assert(kd != NULL); | |
120 | Assert(kd->PdfsActive != NULL); | |
121 | Assert(kd->PdfsActive == kd->Pdfs[0] || kd->PdfsActive == kd->Pdfs[1]); | |
122 | Assert(pdfs != NULL); | |
123 | Assert(x >= 0); | |
124 | Assert(y >= 0); | |
125 | Assert(z >= 0); | |
126 | Assert(x < kd->Dims[0]); | |
127 | Assert(y < kd->Dims[1]); | |
128 | Assert(z < kd->Dims[2]); | |
129 | ||
130 | int oX = kd->Offsets[0]; | |
131 | int oY = kd->Offsets[1]; | |
132 | int oZ = kd->Offsets[2]; | |
133 | ||
134 | ||
135 | #define I(x, y, z, dir) P_INDEX_5(kd->GlobalDims, (x), (y), (z), (dir)) | |
136 | #ifdef PROP_MODEL_PUSH | |
137 | #define X(name, idx, idxinv, _x, _y, _z) pdfs[idx] = kd->PdfsActive[I(x + oX, y + oY, z + oZ, idx)]; | |
138 | #elif PROP_MODEL_PULL | |
139 | #define X(name, idx, idxinv, _x, _y, _z) pdfs[idx] = kd->PdfsActive[I(x + oX - (_x), y + oY - (_y), z + oZ - (_z), idx)]; | |
140 | #endif | |
141 | D3Q19_LIST | |
142 | #undef X | |
143 | #undef I | |
144 | ||
145 | #if 0 // DETECT NANs | |
146 | ||
147 | for (int d = 0; d < 19; ++d) { | |
148 | if (isnan(pdfs[d])) { | |
149 | printf("%d %d %d %d nan! get node\n", x, y, z, d); | |
150 | ||
151 | for (int d2 = 0; d2 < 19; ++d2) { | |
152 | printf("%d: %e\n", d2, pdfs[d2]); | |
153 | } | |
154 | ||
155 | exit(1); | |
156 | } | |
157 | } | |
158 | ||
159 | #endif | |
160 | ||
161 | return; | |
162 | } | |
163 | ||
164 | ||
165 | static void FNAME(SetNode)(KernelData * kd, int x, int y, int z, PdfT * pdfs) | |
166 | { | |
167 | Assert(kd != NULL); | |
168 | Assert(kd->PdfsActive != NULL); | |
169 | Assert(kd->PdfsActive == kd->Pdfs[0] || kd->PdfsActive == kd->Pdfs[1]); | |
170 | Assert(pdfs != NULL); | |
171 | ||
172 | Assert(x >= 0); | |
173 | Assert(y >= 0); | |
174 | Assert(z >= 0); | |
175 | Assert(x < kd->Dims[0]); | |
176 | Assert(y < kd->Dims[1]); | |
177 | Assert(z < kd->Dims[2]); | |
178 | ||
179 | int oX = kd->Offsets[0]; | |
180 | int oY = kd->Offsets[1]; | |
181 | int oZ = kd->Offsets[2]; | |
182 | ||
183 | #define I(x, y, z, dir) P_INDEX_5(kd->GlobalDims, (x), (y), (z), (dir)) | |
184 | #ifdef PROP_MODEL_PUSH | |
185 | #define X(name, idx, idxinv, _x, _y, _z) kd->PdfsActive[I(x + oX, y + oY, z + oZ, idx)] = pdfs[idx]; | |
186 | #elif PROP_MODEL_PULL | |
187 | #define X(name, idx, idxinv, _x, _y, _z) kd->PdfsActive[I(x + oX - (_x), y + oY - (_y), z + oZ - (_z), idx)] = pdfs[idx]; | |
188 | #endif | |
189 | D3Q19_LIST | |
190 | #undef X | |
191 | #undef I | |
192 | ||
193 | return; | |
194 | } | |
195 | ||
196 | ||
197 | static void ParameterUsage() | |
198 | { | |
199 | printf("Kernel parameters:\n"); | |
200 | printf(" [-blk <n>] [-blk-[xyz] <n>]\n"); | |
201 | ||
202 | return; | |
203 | } | |
204 | ||
205 | static void ParseParameters(Parameters * params, int * blk) | |
206 | { | |
207 | Assert(blk != NULL); | |
208 | ||
209 | blk[0] = 0; blk[1] = 0; blk[2] = 0; | |
210 | ||
211 | #define ARG_IS(param) (!strcmp(params->KernelArgs[i], param)) | |
212 | #define NEXT_ARG_PRESENT() \ | |
213 | do { \ | |
214 | if (i + 1 >= params->nKernelArgs) { \ | |
215 | printf("ERROR: argument %s requires a parameter.\n", params->KernelArgs[i]); \ | |
216 | exit(1); \ | |
217 | } \ | |
218 | } while (0) | |
219 | ||
220 | ||
221 | for (int i = 0; i < params->nKernelArgs; ++i) { | |
222 | if (ARG_IS("-blk") || ARG_IS("--blk")) { | |
223 | NEXT_ARG_PRESENT(); | |
224 | ||
225 | int tmp = strtol(params->KernelArgs[++i], NULL, 0); | |
226 | ||
227 | if (tmp <= 0) { | |
228 | printf("ERROR: blocking parameter must be > 0.\n"); | |
229 | exit(1); | |
230 | } | |
231 | ||
232 | blk[0] = blk[1] = blk[2] = tmp; | |
233 | } | |
234 | else if (ARG_IS("-blk-x") || ARG_IS("--blk-x")) { | |
235 | NEXT_ARG_PRESENT(); | |
236 | ||
237 | int tmp = strtol(params->KernelArgs[++i], NULL, 0); | |
238 | ||
239 | if (tmp <= 0) { | |
240 | printf("ERROR: blocking parameter must be > 0.\n"); | |
241 | exit(1); | |
242 | } | |
243 | ||
244 | blk[0] = tmp; | |
245 | } | |
246 | else if (ARG_IS("-blk-y") || ARG_IS("--blk-y")) { | |
247 | NEXT_ARG_PRESENT(); | |
248 | ||
249 | int tmp = strtol(params->KernelArgs[++i], NULL, 0); | |
250 | ||
251 | if (tmp <= 0) { | |
252 | printf("ERROR: blocking parameter must be > 0.\n"); | |
253 | exit(1); | |
254 | } | |
255 | ||
256 | blk[1] = tmp; | |
257 | } | |
258 | else if (ARG_IS("-blk-z") || ARG_IS("--blk-z")) { | |
259 | NEXT_ARG_PRESENT(); | |
260 | ||
261 | int tmp = strtol(params->KernelArgs[++i], NULL, 0); | |
262 | ||
263 | if (tmp <= 0) { | |
264 | printf("ERROR: blocking parameter must be > 0.\n"); | |
265 | exit(1); | |
266 | } | |
267 | ||
268 | blk[2] = tmp; | |
269 | } | |
270 | else if (ARG_IS("-h") || ARG_IS("-help") || ARG_IS("--help")) { | |
271 | ParameterUsage(); | |
272 | exit(1); | |
273 | } | |
274 | else { | |
275 | printf("ERROR: unknown kernel parameter.\n"); | |
276 | ParameterUsage(); | |
277 | exit(1); | |
278 | } | |
279 | } | |
280 | ||
281 | #undef ARG_IS | |
282 | #undef NEXT_ARG_PRESENT | |
283 | ||
284 | return; | |
285 | } | |
286 | ||
287 | ||
288 | void FNAME(D3Q19BlkInit)(LatticeDesc * ld, KernelData ** kernelData, Parameters * params) | |
289 | { | |
290 | KernelDataEx * kdex = NULL; | |
291 | MemAlloc((void **)&kdex, sizeof(KernelDataEx)); | |
292 | ||
293 | kdex->Blk[0] = 0; kdex->Blk[1] = 0; kdex->Blk[2] = 0; | |
294 | ||
295 | KernelData * kd = &kdex->kd; | |
296 | *kernelData = kd; | |
297 | ||
298 | kd->nObstIndices = ld->nObst; | |
299 | ||
300 | // Ajust the dimensions according to padding, if used. | |
301 | kd->Dims[0] = ld->Dims[0]; | |
302 | kd->Dims[1] = ld->Dims[1]; | |
303 | kd->Dims[2] = ld->Dims[2]; | |
304 | ||
305 | ||
306 | int * lDims = ld->Dims; | |
307 | int * gDims = kd->GlobalDims; | |
308 | ||
309 | gDims[0] = lDims[0] + 2; | |
310 | gDims[1] = lDims[1] + 2; | |
311 | gDims[2] = lDims[2] + 2; | |
312 | ||
313 | kd->Offsets[0] = 1; | |
314 | kd->Offsets[1] = 1; | |
315 | kd->Offsets[2] = 1; | |
316 | ||
317 | int lX = lDims[0]; | |
318 | int lY = lDims[1]; | |
319 | int lZ = lDims[2]; | |
320 | ||
321 | int gX = gDims[0]; | |
322 | int gY = gDims[1]; | |
323 | int gZ = gDims[2]; | |
324 | ||
325 | int oX = kd->Offsets[0]; | |
326 | int oY = kd->Offsets[1]; | |
327 | int oZ = kd->Offsets[2]; | |
328 | ||
329 | int blk[3] = { 0 }; | |
330 | ||
331 | int nCells = gX * gY * gZ; | |
332 | ||
333 | PdfT * pdfs[2]; | |
334 | ||
335 | ParseParameters(params, blk); | |
336 | ||
337 | if (blk[0] == 0) blk[0] = gX; | |
338 | if (blk[1] == 0) blk[1] = gY; | |
339 | if (blk[2] == 0) blk[2] = gZ; | |
340 | ||
341 | printf("# blocking x: %3d y: %3d z: %3d\n", blk[0], blk[1], blk[2]); | |
342 | ||
343 | ||
344 | kdex->Blk[0] = blk[0]; kdex->Blk[1] = blk[1]; kdex->Blk[2] = blk[2]; | |
345 | ||
346 | ||
347 | printf("# allocating data for %d LB nodes with padding (%lu bytes = %f MiB for both lattices)\n", | |
348 | nCells, 2 * sizeof(PdfT) * nCells * N_D3Q19, | |
349 | 2 * sizeof(PdfT) * nCells * N_D3Q19 / 1024.0 / 1024.0); | |
350 | ||
351 | MemAlloc((void **)&pdfs[0], sizeof(PdfT) * nCells * N_D3Q19); | |
352 | MemAlloc((void **)&pdfs[1], sizeof(PdfT) * nCells * N_D3Q19); | |
353 | ||
354 | kd->Pdfs[0] = pdfs[0]; | |
355 | kd->Pdfs[1] = pdfs[1]; | |
356 | ||
357 | // Initialize PDFs with some (arbitrary) data for correct NUMA placement. | |
358 | // This depends on the chosen data layout. | |
359 | // The structure of the loop should resemble the same "execution layout" | |
360 | // as in the kernel! | |
361 | #ifdef _OPENMP | |
362 | #pragma omp parallel for collapse(3) | |
363 | #endif | |
364 | ||
365 | for (int bZ = 0; bZ < gZ; bZ += blk[2]) { | |
366 | for (int bY = 0; bY < gY; bY += blk[1]) { | |
367 | for (int bX = 0; bX < gX; bX += blk[0]) { | |
368 | ||
369 | // Must do everything here, else it would break collapse. | |
370 | int eZ = MIN(bZ + blk[2], gZ); | |
371 | int eY = MIN(bY + blk[1], gY); | |
372 | int eX = MIN(bX + blk[0], gX); | |
373 | ||
374 | for (int z = bZ; z < eZ; ++z) { | |
375 | for (int y = bY; y < eY; ++y) { | |
376 | for (int x = bX; x < eX; ++x) { | |
377 | ||
378 | for (int d = 0; d < N_D3Q19; ++d) { | |
379 | pdfs[0][P_INDEX_5(gDims, x, y, z, d)] = 1.0; | |
380 | pdfs[1][P_INDEX_5(gDims, x, y, z, d)] = 1.0; | |
381 | } | |
382 | ||
383 | } | |
384 | } | |
385 | } | |
386 | } | |
387 | } | |
388 | } | |
389 | ||
390 | // Initialize all PDFs to some standard value. | |
391 | for (int z = 0; z < gZ; ++z) { | |
392 | for (int y = 0; y < gY; ++y) { | |
393 | for (int x = 0; x < gX; ++x) { | |
394 | for (int d = 0; d < N_D3Q19; ++d) { | |
395 | pdfs[0][P_INDEX_5(gDims, x, y, z, d)] = 0.0; | |
396 | pdfs[1][P_INDEX_5(gDims, x, y, z, d)] = 0.0; | |
397 | } | |
398 | } | |
399 | } | |
400 | } | |
401 | ||
402 | ||
403 | // Count how many *PDFs* need bounce back treatment. | |
404 | ||
405 | uint64_t nPdfs = ((uint64_t)19) * gX * gY * gZ; | |
406 | ||
407 | if (nPdfs > ((2LU << 31) - 1)) { | |
408 | printf("ERROR: number of PDFs exceed 2^31.\n"); | |
409 | exit(1); | |
410 | } | |
411 | ||
412 | // Compiler bug? Incorrect computation of nBounceBackPdfs when using icc 15.0.2. | |
413 | // Works when declaring nBounceBackPdfs as int64_t or using volatile. | |
414 | volatile int nBounceBackPdfs = 0; | |
415 | // int64_t nBounceBackPdfs = 0; | |
416 | int nx, ny, nz, px, py, pz; | |
417 | ||
418 | // TODO: apply blocking? | |
419 | ||
420 | for (int z = 0; z < lZ; ++z) { | |
421 | for (int y = 0; y < lY; ++y) { | |
422 | for (int x = 0; x < lX; ++x) { | |
423 | ||
424 | if (ld->Lattice[L_INDEX_4(ld->Dims, x, y, z)] != LAT_CELL_OBSTACLE) { | |
425 | for (int d = 0; d < N_D3Q19; ++d) { | |
426 | #ifdef PROP_MODEL_PUSH | |
427 | nx = x + D3Q19_X[d]; | |
428 | ny = y + D3Q19_Y[d]; | |
429 | nz = z + D3Q19_Z[d]; | |
430 | #elif PROP_MODEL_PULL | |
431 | nx = x - D3Q19_X[d]; | |
432 | ny = y - D3Q19_Y[d]; | |
433 | nz = z - D3Q19_Z[d]; | |
434 | #else | |
435 | #error PROP_MODEL_NAME unknown. | |
436 | #endif | |
437 | // Check if neighbor is inside the lattice. | |
438 | // if(nx < 0 || ny < 0 || nz < 0 || nx >= lX || ny >= lY || nz >= lZ) { | |
439 | // continue; | |
440 | // } | |
441 | if ((nx < 0 || nx >= lX) && ld->PeriodicX) { | |
442 | ++nBounceBackPdfs; // Compiler bug --> see above | |
443 | } | |
444 | else if ((ny < 0 || ny >= lY) && ld->PeriodicY) { | |
445 | ++nBounceBackPdfs; // Compiler bug --> see above | |
446 | } | |
447 | else if ((nz < 0 || nz >= lZ) && ld->PeriodicZ) { | |
448 | ++nBounceBackPdfs; // Compiler bug --> see above | |
449 | } | |
450 | else if (nx < 0 || ny < 0 || nz < 0 || nx >= lX || ny >= lY || nz >= lZ) { | |
451 | continue; | |
452 | } | |
453 | else if (ld->Lattice[L_INDEX_4(lDims, nx, ny, nz)] == LAT_CELL_OBSTACLE) { | |
454 | ++nBounceBackPdfs; // Compiler bug --> see above | |
455 | } | |
456 | } | |
457 | } | |
458 | } | |
459 | } | |
460 | } | |
461 | ||
462 | ||
463 | printf("# allocating %d indices for bounce back pdfs (%s for source and destination array)\n", nBounceBackPdfs, ByteToHuman(sizeof(int) * nBounceBackPdfs * 2)); | |
464 | ||
465 | MemAlloc((void **) & (kd->BounceBackPdfsSrc), sizeof(int) * nBounceBackPdfs + 100); | |
466 | MemAlloc((void **) & (kd->BounceBackPdfsDst), sizeof(int) * nBounceBackPdfs + 100); | |
467 | ||
468 | kd->nBounceBackPdfs = nBounceBackPdfs; | |
469 | nBounceBackPdfs = 0; | |
470 | ||
471 | int srcIndex; | |
472 | int dstIndex; | |
473 | ||
474 | for (int z = 0; z < lZ; ++z) { | |
475 | for (int y = 0; y < lY; ++y) { | |
476 | for (int x = 0; x < lX; ++x) { | |
477 | ||
478 | if (ld->Lattice[L_INDEX_4(ld->Dims, x, y, z)] != LAT_CELL_OBSTACLE) { | |
479 | for (int d = 0; d < N_D3Q19; ++d) { | |
480 | #ifdef PROP_MODEL_PUSH | |
481 | nx = x + D3Q19_X[d]; | |
482 | ny = y + D3Q19_Y[d]; | |
483 | nz = z + D3Q19_Z[d]; | |
484 | #elif PROP_MODEL_PULL | |
485 | nx = x - D3Q19_X[d]; | |
486 | ny = y - D3Q19_Y[d]; | |
487 | nz = z - D3Q19_Z[d]; | |
488 | #else | |
489 | #error PROP_MODEL_NAME unknown. | |
490 | #endif | |
491 | ||
492 | if ( ((nx < 0 || nx >= lX) && ld->PeriodicX) || | |
493 | ((ny < 0 || ny >= lY) && ld->PeriodicY) || | |
494 | ((nz < 0 || nz >= lZ) && ld->PeriodicZ) | |
495 | ){ | |
496 | // Implement periodic boundary in X direction. | |
497 | ||
498 | // If the target node reached through propagation is outside the lattice | |
499 | // the kernel stores it in some buffer around the domain. | |
500 | // From this position the PDF must be transported to the other side of the | |
501 | // geometry. | |
502 | ||
503 | // Take PDF from outside the domain. | |
504 | ||
505 | // x periodic | |
506 | if (nx < 0) { | |
507 | px = lX - 1; | |
508 | } | |
509 | else if (nx >= lX) { | |
510 | px = 0; | |
511 | } else { | |
512 | px = nx; | |
513 | } | |
514 | ||
515 | // y periodic | |
516 | if (ny < 0) { | |
517 | py = lY - 1; | |
518 | } | |
519 | else if (ny >= lY) { | |
520 | py = 0; | |
521 | } else { | |
522 | py = ny; | |
523 | } | |
524 | ||
525 | // z periodic | |
526 | if (nz < 0) { | |
527 | pz = lZ - 1; | |
528 | } | |
529 | else if (nz >= lZ) { | |
530 | pz = 0; | |
531 | } else { | |
532 | pz = nz; | |
533 | } | |
534 | ||
535 | if (ld->Lattice[L_INDEX_4(lDims, px, py, pz)] == LAT_CELL_OBSTACLE) { | |
536 | #ifdef PROP_MODEL_PUSH | |
537 | srcIndex = P_INDEX_5(gDims, nx + oX, ny + oY, nz + oZ, d); | |
538 | dstIndex = P_INDEX_5(gDims, x + oX, y + oY, z + oZ, D3Q19_INV[d]); | |
539 | #elif PROP_MODEL_PULL | |
540 | srcIndex = P_INDEX_5(gDims, x + oX, y + oY, z + oZ, D3Q19_INV[d]); | |
541 | dstIndex = P_INDEX_5(gDims, nx + oX, ny + oY, nz + oZ, d); | |
542 | #endif | |
543 | } | |
544 | else { | |
545 | ||
546 | #ifdef PROP_MODEL_PUSH | |
547 | srcIndex = P_INDEX_5(gDims, nx + oX, ny + oY, nz + oZ, d); | |
548 | // Put it on the other side back into the domain. | |
549 | dstIndex = P_INDEX_5(gDims, px + oX, py + oY, pz + oZ, d); | |
550 | #elif PROP_MODEL_PULL | |
551 | srcIndex = P_INDEX_5(gDims, px + oX, py + oY, pz + oZ, d); | |
552 | // Put it on the other side back into the ghost layer. | |
553 | dstIndex = P_INDEX_5(gDims, nx + oX, ny + oY, nz + oZ, d); | |
554 | #endif | |
555 | ||
556 | VerifyMsg(nBounceBackPdfs < kd->nBounceBackPdfs, "nBBPdfs %d < kd->nBBPdfs %d xyz: %d %d %d d: %d\n", nBounceBackPdfs, kd->nBounceBackPdfs, x, y, z, d); | |
557 | ||
558 | } | |
559 | ||
560 | kd->BounceBackPdfsSrc[nBounceBackPdfs] = srcIndex; | |
561 | kd->BounceBackPdfsDst[nBounceBackPdfs] = dstIndex; | |
562 | ||
563 | ++nBounceBackPdfs; | |
564 | ||
565 | } | |
566 | else if (nx < 0 || ny < 0 || nz < 0 || nx >= lX || ny >= lY || nz >= lZ) { | |
567 | continue; | |
568 | } | |
569 | else if (ld->Lattice[L_INDEX_4(lDims, nx, ny, nz)] == LAT_CELL_OBSTACLE) { | |
570 | #ifdef PROP_MODEL_PUSH | |
571 | srcIndex = P_INDEX_5(gDims, nx + oX, ny + oY, nz + oZ, d); | |
572 | dstIndex = P_INDEX_5(gDims, x + oX, y + oY, z + oZ, D3Q19_INV[d]); | |
573 | #elif PROP_MODEL_PULL | |
574 | srcIndex = P_INDEX_5(gDims, x + oX, y + oY, z + oZ, D3Q19_INV[d]); | |
575 | dstIndex = P_INDEX_5(gDims, nx + oX, ny + oY, nz + oZ, d); | |
576 | // srcIndex = P_INDEX_5(gDims, x + oX, y + oY, z + oZ, d); | |
577 | // dstIndex = P_INDEX_5(gDims, nx + oX, ny + oY, nz + oZ, D3Q19_INV[d]); | |
578 | #endif | |
579 | ||
580 | VerifyMsg(nBounceBackPdfs < kd->nBounceBackPdfs, "nBBPdfs %d < kd->nBBPdfs %d xyz: %d %d %d d: %d\n", nBounceBackPdfs, kd->nBounceBackPdfs, x, y, z, d); | |
581 | ||
582 | kd->BounceBackPdfsSrc[nBounceBackPdfs] = srcIndex; | |
583 | kd->BounceBackPdfsDst[nBounceBackPdfs] = dstIndex; | |
584 | ||
585 | ++nBounceBackPdfs; | |
586 | } | |
587 | } | |
588 | } | |
589 | } | |
590 | } | |
591 | } | |
592 | ||
593 | ||
594 | // Fill remaining KernelData structures | |
595 | kd->GetNode = FNAME(GetNode); | |
596 | kd->SetNode = FNAME(SetNode); | |
597 | ||
598 | kd->BoundaryConditionsGetPdf = FNAME(BcGetPdf); | |
599 | kd->BoundaryConditionsSetPdf = FNAME(BcSetPdf); | |
600 | ||
601 | kd->Kernel = FNAME(D3Q19BlkKernel); | |
602 | ||
603 | kd->DstPdfs = NULL; | |
604 | kd->PdfsActive = kd->Pdfs[0]; | |
605 | ||
606 | return; | |
607 | } | |
608 | ||
609 | void FNAME(D3Q19BlkDeinit)(LatticeDesc * ld, KernelData ** kernelData) | |
610 | { | |
611 | MemFree((void **) & ((*kernelData)->Pdfs[0])); | |
612 | MemFree((void **) & ((*kernelData)->Pdfs[1])); | |
613 | ||
614 | MemFree((void **) & ((*kernelData)->BounceBackPdfsSrc)); | |
615 | MemFree((void **) & ((*kernelData)->BounceBackPdfsDst)); | |
616 | ||
617 | MemFree((void **)kernelData); | |
618 | ||
619 | return; | |
620 | } | |
621 | ||
622 | // Kernels without blocking perform the same initialization/deinitialization as with | |
623 | // blocking, except that a different kernel is called. Hence, no arguments are allowed. | |
624 | ||
625 | void FNAME(D3Q19Init)(LatticeDesc * ld, KernelData ** kernelData, Parameters * params) | |
626 | { | |
627 | Parameters p; | |
628 | ||
629 | if (params->nKernelArgs != 0) { | |
630 | printf("ERROR: unknown kernel parameter.\n"); | |
631 | printf("This kernels accepts no parameters.\n"); | |
632 | exit(1); | |
633 | } | |
634 | ||
635 | // Setup an empty parameters structure. | |
636 | p.nArgs = params->nArgs; | |
637 | p.Args = params->Args; | |
638 | p.nKernelArgs = 0; | |
639 | p.KernelArgs = NULL; | |
640 | ||
641 | // Call init routine for blocking kernel and override the | |
642 | // kernel function to be called later on. | |
643 | FNAME(D3Q19BlkInit)(ld, kernelData, &p); | |
644 | ||
645 | (*kernelData)->Kernel = FNAME(D3Q19Kernel); | |
646 | ||
647 | return; | |
648 | ||
649 | } | |
650 | ||
651 | void FNAME(D3Q19Deinit)(LatticeDesc * ld, KernelData ** kernelData) | |
652 | { | |
653 | FNAME(D3Q19BlkDeinit)(ld, kernelData); | |
654 | return; | |
655 | } |