[LbmBenchmarkKernelsPublic.git] / src / BenchKernelD3Q19List.c

// --------------------------------------------------------------------------
//
// Copyright
//   Markus Wittmann, 2016-2017
//   RRZE, University of Erlangen-Nuremberg, Germany
//   markus.wittmann -at- fau.de or hpc -at- rrze.fau.de
//
//   Viktor Haag, 2016
//   LSS, University of Erlangen-Nuremberg, Germany
//
//  This file is part of the Lattice Boltzmann Benchmark Kernels (LbmBenchKernels).
//
//  LbmBenchKernels is free software: you can redistribute it and/or modify
//  it under the terms of the GNU General Public License as published by
//  the Free Software Foundation, either version 3 of the License, or
//  (at your option) any later version.
//
//  LbmBenchKernels is distributed in the hope that it will be useful,
//  but WITHOUT ANY WARRANTY; without even the implied warranty of
//  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
//  GNU General Public License for more details.
//
//  You should have received a copy of the GNU General Public License
//  along with LbmBenchKernels.  If not, see <http://www.gnu.org/licenses/>.
//
// --------------------------------------------------------------------------
#include "BenchKernelD3Q19ListCommon.h"

#include "Memory.h"
#include "Vtk.h"

#include <inttypes.h>
#include <math.h>


void FNAME(D3Q19ListKernel)(LatticeDesc * ld, KernelData * kernelData, CaseData * cd)
{
	Assert(ld != NULL);
	Assert(kernelData != NULL);
	Assert(cd != NULL);

	Assert(cd->Omega > 0.0);
	Assert(cd->Omega < 2.0);

	KernelData * kd = (KernelData *)kernelData;
	KernelDataList * kdl = (KernelDataList *)kernelData;

	PdfT omega = cd->Omega;
	PdfT omegaEven = omega;
// 	PdfT omegaOdd = 8.0*((2.0-omegaEven)/(8.0-omegaEven)); //"standard" trt odd relaxation parameter
	PdfT magicParam = 1.0/12.0;	// 1/4: best stability;	1/12: removes third-order advection error (best advection);	1/6: removes fourth-order diffusion error (best diffusion);	3/16: exact location of bounce back for poiseuille flow
	PdfT omegaOdd = 1.0/( 0.5 + magicParam/(1.0/omega - 0.5) );

	PdfT evenPart = 0.0;
	PdfT oddPart = 0.0;
	PdfT dir_indep_trm = 0.0;

	PdfT w_0 = 1.0 /  3.0;
	PdfT w_1 = 1.0 / 18.0;
	PdfT w_2 = 1.0 / 36.0;

	PdfT w_1_x3 = w_1 * 3.0;	PdfT w_1_nine_half = w_1 * 9.0/2.0;	PdfT w_1_indep = 0.0;
	PdfT w_2_x3 = w_2 * 3.0;	PdfT w_2_nine_half = w_2 * 9.0/2.0;	PdfT w_2_indep = 0.0;

	PdfT ux, uy, uz, ui;
	PdfT dens;

	// Declare pdf_N, pdf_E, pdf_S, pdf_W, ...
	#define X(name, idx, idxinv, x, y, z)	PdfT JOIN(pdf_,name);
	D3Q19_LIST
	#undef X

	PdfT * src = kd->Pdfs[0];
	PdfT * dst = kd->Pdfs[1];
	PdfT * tmp;

	int maxIterations = cd->MaxIterations;
	int nFluid = kdl->nFluid;
	int nCells = kdl->nCells;

	uint32_t adjListIndex;
	uint32_t * adjList = kdl->AdjList;

	#ifdef VTK_OUTPUT
		if (cd->VtkOutput) {
			kd->PdfsActive = src;
			VtkWrite(ld, kd, cd, 0);
		}
	#endif

	#ifdef STATISTICS
		kd->PdfsActive = src;
		KernelStatistics(kd, ld, cd, 0);
	#endif

	// TODO: outer openmp parallel
	for(int iter = 0; iter < maxIterations; ++iter) {


	#ifdef _OPENMP
		#pragma omp parallel for default(none) \
				shared(nFluid, nCells, kd, kdl, adjList, src, dst, w_0, w_1, w_2, omegaEven, omegaOdd, \
				w_1_x3, w_2_x3, w_1_nine_half, w_2_nine_half, cd) \
				private(ux, uy, uz, ui, dens, dir_indep_trm, adjListIndex, \
					pdf_C, \
				  	pdf_N, pdf_E, pdf_S, pdf_W, \
				  	pdf_NE, pdf_SE, pdf_SW, pdf_NW, \
				  	pdf_T, pdf_TN, pdf_TE, pdf_TS, pdf_TW, \
				  	pdf_B, pdf_BN, pdf_BE, pdf_BS, pdf_BW, \
				  	evenPart, oddPart, w_1_indep, w_2_indep)
	#endif
	for (int index = 0; index < nFluid; ++index) {

			#define I(index, dir)	P_INDEX_3((nCells), (index), (dir))

#ifdef PROP_MODEL_PUSH

			// Load PDFs of local cell: pdf_N = src[I(x, y, z, D3Q19_N)]; ...
			#define X(name, idx, idxinv, _x, _y, _z)	JOIN(pdf_,name) = src[I(index, idx)];
			D3Q19_LIST
			#undef X

#elif PROP_MODEL_PULL

			adjListIndex = index * N_D3Q19_IDX;

			pdf_C = src[P_INDEX_3(nCells, index, D3Q19_C)];

			// Load PDFs of local cell: pdf_N = src[adjList[adjListIndex + D3Q19_N]]; ...
			#define X(name, idx, idxinv, _x, _y, _z)	JOIN(pdf_,name) = src[adjList[adjListIndex + idx]];
			D3Q19_LIST_WO_C
			#undef X

#else
	#error No implementation for PROP_MODEL_NAME.
#endif

// #define LID_DRIVEN_CAVITY

#ifdef LID_DRIVEN_CAVITY
			int nX = kd->Dims[0];
			int nY = kd->Dims[1];
			int nZ = kd->Dims[2];

			int x = kdl->Coords[C_INDEX_X(index)];
			int y = kdl->Coords[C_INDEX_Y(index)];
			int z = kdl->Coords[C_INDEX_Z(index)];

			if (z == nZ - 4 && x > 3 && x < (nX - 4) && y > 3 && y < (nY - 4)) {
				ux = 0.1 * 0.577;
				uy = 0.0;
				uz = 0.0;
			} else {
#endif
				ux = pdf_E + pdf_NE + pdf_SE + pdf_TE + pdf_BE -
					 pdf_W - pdf_NW - pdf_SW - pdf_TW - pdf_BW;
				uy = pdf_N + pdf_NE + pdf_NW + pdf_TN + pdf_BN -
					 pdf_S - pdf_SE - pdf_SW - pdf_TS - pdf_BS;
				uz = pdf_T + pdf_TE + pdf_TW + pdf_TN + pdf_TS -
					 pdf_B - pdf_BE - pdf_BW - pdf_BN - pdf_BS;
#ifdef LID_DRIVEN_CAVITY
			}
#endif

			dens = pdf_C +
				   pdf_N  + pdf_E  + pdf_S  + pdf_W  +
				   pdf_NE + pdf_SE + pdf_SW + pdf_NW +
				   pdf_T  + pdf_TN + pdf_TE + pdf_TS + pdf_TW +
				   pdf_B  + pdf_BN + pdf_BE + pdf_BS + pdf_BW;

			dir_indep_trm = dens - (ux * ux + uy * uy + uz * uz)*3.0/2.0;

#ifdef PROP_MODEL_PUSH

			adjListIndex = index * N_D3Q19_IDX;

			// direction: w_0
			dst[I(index, D3Q19_C)             ]  = pdf_C - omegaEven*(pdf_C - w_0*dir_indep_trm);

			// direction: w_1
			w_1_indep = w_1*dir_indep_trm;

			ui = uy;
			evenPart = omegaEven*( 0.5*(pdf_N + pdf_S) - ui*ui*w_1_nine_half - w_1_indep );
			oddPart = omegaOdd*(0.5*(pdf_N - pdf_S) - ui*w_1_x3 );
			dst[adjList[adjListIndex + D3Q19_N]]  = pdf_N - evenPart - oddPart;
			dst[adjList[adjListIndex + D3Q19_S]]  = pdf_S - evenPart + oddPart;

			ui = ux;
			evenPart = omegaEven*( 0.5*(pdf_E + pdf_W) - ui*ui*w_1_nine_half - w_1_indep );
			oddPart = omegaOdd*(0.5*(pdf_E - pdf_W) - ui*w_1_x3 );
			dst[adjList[adjListIndex + D3Q19_E]]  = pdf_E - evenPart - oddPart;
			dst[adjList[adjListIndex + D3Q19_W]]  = pdf_W - evenPart + oddPart;

			ui = uz;
			evenPart = omegaEven*( 0.5*(pdf_T + pdf_B) - ui*ui*w_1_nine_half - w_1_indep );
			oddPart = omegaOdd*(0.5*(pdf_T - pdf_B) - ui*w_1_x3 );
			dst[adjList[adjListIndex + D3Q19_T]]  = pdf_T - evenPart - oddPart;
			dst[adjList[adjListIndex + D3Q19_B]]  = pdf_B - evenPart + oddPart;

			// direction: w_2
			w_2_indep = w_2*dir_indep_trm;

			ui = -ux + uy;
			evenPart = omegaEven*( 0.5*(pdf_NW + pdf_SE) - ui*ui*w_2_nine_half - w_2_indep );
			oddPart = omegaOdd*(0.5*(pdf_NW - pdf_SE) - ui*w_2_x3 );
			dst[adjList[adjListIndex + D3Q19_NW]] = pdf_NW - evenPart - oddPart;
			dst[adjList[adjListIndex + D3Q19_SE]] = pdf_SE - evenPart + oddPart;

			ui = ux + uy;
			evenPart = omegaEven*( 0.5*(pdf_NE + pdf_SW) - ui*ui*w_2_nine_half - w_2_indep );
			oddPart = omegaOdd*(0.5*(pdf_NE - pdf_SW) - ui*w_2_x3 );
			dst[adjList[adjListIndex + D3Q19_NE]] = pdf_NE - evenPart - oddPart;
			dst[adjList[adjListIndex + D3Q19_SW]] = pdf_SW - evenPart + oddPart;

			ui = -ux + uz;
			evenPart = omegaEven*( 0.5*(pdf_TW + pdf_BE) - ui*ui*w_2_nine_half - w_2_indep );
			oddPart = omegaOdd*(0.5*(pdf_TW - pdf_BE) - ui*w_2_x3 );
			dst[adjList[adjListIndex + D3Q19_TW]] = pdf_TW - evenPart - oddPart;
			dst[adjList[adjListIndex + D3Q19_BE]] = pdf_BE - evenPart + oddPart;

			ui = ux + uz;
			evenPart = omegaEven*( 0.5*(pdf_TE + pdf_BW) - ui*ui*w_2_nine_half - w_2_indep );
			oddPart = omegaOdd*(0.5*(pdf_TE - pdf_BW) - ui*w_2_x3 );
			dst[adjList[adjListIndex + D3Q19_TE]] = pdf_TE - evenPart - oddPart;
			dst[adjList[adjListIndex + D3Q19_BW]] = pdf_BW - evenPart + oddPart;

			ui = -uy + uz;
			evenPart = omegaEven*( 0.5*(pdf_TS + pdf_BN) - ui*ui*w_2_nine_half - w_2_indep );
			oddPart = omegaOdd*(0.5*(pdf_TS - pdf_BN) - ui*w_2_x3 );
			dst[adjList[adjListIndex + D3Q19_TS]] = pdf_TS - evenPart - oddPart;
			dst[adjList[adjListIndex + D3Q19_BN]] = pdf_BN - evenPart + oddPart;

			ui = uy + uz;
			evenPart = omegaEven*( 0.5*(pdf_TN + pdf_BS) - ui*ui*w_2_nine_half - w_2_indep );
			oddPart = omegaOdd*(0.5*(pdf_TN - pdf_BS) - ui*w_2_x3 );
			dst[adjList[adjListIndex + D3Q19_TN]] = pdf_TN - evenPart - oddPart;
			dst[adjList[adjListIndex + D3Q19_BS]] = pdf_BS - evenPart + oddPart;

#elif PROP_MODEL_PULL

			// direction: w_0
			dst[I(index, D3Q19_C )]  = pdf_C - omegaEven*(pdf_C - w_0*dir_indep_trm);

			// direction: w_1
			w_1_indep = w_1*dir_indep_trm;

			ui = uy;
			evenPart = omegaEven*( 0.5*(pdf_N + pdf_S) - ui*ui*w_1_nine_half - w_1_indep );
			oddPart = omegaOdd*(0.5*(pdf_N - pdf_S) - ui*w_1_x3 );
			dst[I(index, D3Q19_N )]  = pdf_N - evenPart - oddPart;
			dst[I(index, D3Q19_S )]  = pdf_S - evenPart + oddPart;

			ui = ux;
			evenPart = omegaEven*( 0.5*(pdf_E + pdf_W) - ui*ui*w_1_nine_half - w_1_indep );
			oddPart = omegaOdd*(0.5*(pdf_E - pdf_W) - ui*w_1_x3 );
			dst[I(index, D3Q19_E )]  = pdf_E - evenPart - oddPart;
			dst[I(index, D3Q19_W )]  = pdf_W - evenPart + oddPart;

			ui = uz;
			evenPart = omegaEven*( 0.5*(pdf_T + pdf_B) - ui*ui*w_1_nine_half - w_1_indep );
			oddPart = omegaOdd*(0.5*(pdf_T - pdf_B) - ui*w_1_x3 );
			dst[I(index, D3Q19_T )]  = pdf_T - evenPart - oddPart;
			dst[I(index, D3Q19_B )]  = pdf_B - evenPart + oddPart;

			// direction: w_2
			w_2_indep = w_2*dir_indep_trm;

			ui = -ux + uy;
			evenPart = omegaEven*( 0.5*(pdf_NW + pdf_SE) - ui*ui*w_2_nine_half - w_2_indep );
			oddPart = omegaOdd*(0.5*(pdf_NW - pdf_SE) - ui*w_2_x3 );
			dst[I(index, D3Q19_NW)] = pdf_NW - evenPart - oddPart;
			dst[I(index, D3Q19_SE)] = pdf_SE - evenPart + oddPart;

			ui = ux + uy;
			evenPart = omegaEven*( 0.5*(pdf_NE + pdf_SW) - ui*ui*w_2_nine_half - w_2_indep );
			oddPart = omegaOdd*(0.5*(pdf_NE - pdf_SW) - ui*w_2_x3 );
			dst[I(index, D3Q19_NE)] = pdf_NE - evenPart - oddPart;
			dst[I(index, D3Q19_SW)] = pdf_SW - evenPart + oddPart;

			ui = -ux + uz;
			evenPart = omegaEven*( 0.5*(pdf_TW + pdf_BE) - ui*ui*w_2_nine_half - w_2_indep );
			oddPart = omegaOdd*(0.5*(pdf_TW - pdf_BE) - ui*w_2_x3 );
			dst[I(index, D3Q19_TW)] = pdf_TW - evenPart - oddPart;
			dst[I(index, D3Q19_BE)] = pdf_BE - evenPart + oddPart;

			ui = ux + uz;
			evenPart = omegaEven*( 0.5*(pdf_TE + pdf_BW) - ui*ui*w_2_nine_half - w_2_indep );
			oddPart = omegaOdd*(0.5*(pdf_TE - pdf_BW) - ui*w_2_x3 );
			dst[I(index, D3Q19_TE)] = pdf_TE - evenPart - oddPart;
			dst[I(index, D3Q19_BW)] = pdf_BW - evenPart + oddPart;

			ui = -uy + uz;
			evenPart = omegaEven*( 0.5*(pdf_TS + pdf_BN) - ui*ui*w_2_nine_half - w_2_indep );
			oddPart = omegaOdd*(0.5*(pdf_TS - pdf_BN) - ui*w_2_x3 );
			dst[I(index, D3Q19_TS)] = pdf_TS - evenPart - oddPart;
			dst[I(index, D3Q19_BN)] = pdf_BN - evenPart + oddPart;

			ui = uy + uz;
			evenPart = omegaEven*( 0.5*(pdf_TN + pdf_BS) - ui*ui*w_2_nine_half - w_2_indep );
			oddPart = omegaOdd*(0.5*(pdf_TN - pdf_BS) - ui*w_2_x3 );
			dst[I(index, D3Q19_TN)] = pdf_TN - evenPart - oddPart;
			dst[I(index, D3Q19_BS)] = pdf_BS - evenPart + oddPart;

#endif
			#undef I
		} // loop over fluid nodes

		#ifdef VERIFICATION
			kd->PdfsActive = dst;
			KernelAddBodyForce(kd, ld, cd);
		#endif

		#ifdef VTK_OUTPUT
			if (cd->VtkOutput && (iter % cd->VtkModulus) == 0) {
				kd->PdfsActive = dst;
				VtkWrite(ld, kd, cd, iter);
			}
		#endif

		#ifdef STATISTICS
			kd->PdfsActive = dst;
			KernelStatistics(kd, ld, cd, iter);
		#endif

		// swap grids
		tmp = src;
		src = dst;
		dst = tmp;

	} // for (int iter = 0; ...

#ifdef VTK_OUTPUT
	if (cd->VtkOutput) {
		kd->PdfsActive = src;
		VtkWrite(ld, kd, cd, maxIterations);
	}
#endif

#ifdef STATISTICS
	kd->PdfsActive = src;
	KernelStatistics(kd, ld, cd, maxIterations);
#endif

	return;
}
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 "BenchKernelD3Q19ListCommon.h"
	28
	29	#include "Memory.h"
	30	#include "Vtk.h"
	31
	32	#include <inttypes.h>
	33	#include <math.h>
	34
	35
	36	void FNAME(D3Q19ListKernel)(LatticeDesc * ld, KernelData * kernelData, CaseData * cd)
	37	{
	38	Assert(ld != NULL);
	39	Assert(kernelData != NULL);
	40	Assert(cd != NULL);
	41
	42	Assert(cd->Omega > 0.0);
	43	Assert(cd->Omega < 2.0);
	44
	45	KernelData * kd = (KernelData *)kernelData;
	46	KernelDataList * kdl = (KernelDataList *)kernelData;
	47
	48	PdfT omega = cd->Omega;
	49	PdfT omegaEven = omega;
	50	// PdfT omegaOdd = 8.0*((2.0-omegaEven)/(8.0-omegaEven)); //"standard" trt odd relaxation parameter
	51	PdfT magicParam = 1.0/12.0; // 1/4: best stability; 1/12: removes third-order advection error (best advection); 1/6: removes fourth-order diffusion error (best diffusion); 3/16: exact location of bounce back for poiseuille flow
	52	PdfT omegaOdd = 1.0/( 0.5 + magicParam/(1.0/omega - 0.5) );
	53
	54	PdfT evenPart = 0.0;
	55	PdfT oddPart = 0.0;
	56	PdfT dir_indep_trm = 0.0;
	57
	58	PdfT w_0 = 1.0 / 3.0;
	59	PdfT w_1 = 1.0 / 18.0;
	60	PdfT w_2 = 1.0 / 36.0;
	61
	62	PdfT w_1_x3 = w_1 * 3.0; PdfT w_1_nine_half = w_1 * 9.0/2.0; PdfT w_1_indep = 0.0;
	63	PdfT w_2_x3 = w_2 * 3.0; PdfT w_2_nine_half = w_2 * 9.0/2.0; PdfT w_2_indep = 0.0;
	64
65	PdfT ux, uy, uz, ui;
66	PdfT dens;
67
68	// Declare pdf_N, pdf_E, pdf_S, pdf_W, ...
69	#define X(name, idx, idxinv, x, y, z) PdfT JOIN(pdf_,name);
70	D3Q19_LIST
71	#undef X
72
73	PdfT * src = kd->Pdfs[0];
74	PdfT * dst = kd->Pdfs[1];
75	PdfT * tmp;
76
77	int maxIterations = cd->MaxIterations;
78	int nFluid = kdl->nFluid;
79	int nCells = kdl->nCells;
80
81	uint32_t adjListIndex;
82	uint32_t * adjList = kdl->AdjList;
83
84	#ifdef VTK_OUTPUT
85	if (cd->VtkOutput) {
86	kd->PdfsActive = src;
87	VtkWrite(ld, kd, cd, 0);
88	}
89	#endif
90
91	#ifdef STATISTICS
92	kd->PdfsActive = src;
93	KernelStatistics(kd, ld, cd, 0);
94	#endif
95
96	// TODO: outer openmp parallel
97	for(int iter = 0; iter < maxIterations; ++iter) {
98
99
100
101	#ifdef _OPENMP
102	#pragma omp parallel for default(none) \
103	shared(nFluid, nCells, kd, kdl, adjList, src, dst, w_0, w_1, w_2, omegaEven, omegaOdd, \
104	w_1_x3, w_2_x3, w_1_nine_half, w_2_nine_half, cd) \
105	private(ux, uy, uz, ui, dens, dir_indep_trm, adjListIndex, \
106	pdf_C, \
107	pdf_N, pdf_E, pdf_S, pdf_W, \
108	pdf_NE, pdf_SE, pdf_SW, pdf_NW, \
109	pdf_T, pdf_TN, pdf_TE, pdf_TS, pdf_TW, \
110	pdf_B, pdf_BN, pdf_BE, pdf_BS, pdf_BW, \
111	evenPart, oddPart, w_1_indep, w_2_indep)
112	#endif
113	for (int index = 0; index < nFluid; ++index) {
114
115	#define I(index, dir) P_INDEX_3((nCells), (index), (dir))
116
117	#ifdef PROP_MODEL_PUSH
118
119	// Load PDFs of local cell: pdf_N = src[I(x, y, z, D3Q19_N)]; ...
120	#define X(name, idx, idxinv, _x, _y, _z) JOIN(pdf_,name) = src[I(index, idx)];
121	D3Q19_LIST
122	#undef X
123
124	#elif PROP_MODEL_PULL
125
126	adjListIndex = index * N_D3Q19_IDX;
127
128	pdf_C = src[P_INDEX_3(nCells, index, D3Q19_C)];
129
130	// Load PDFs of local cell: pdf_N = src[adjList[adjListIndex + D3Q19_N]]; ...
131	#define X(name, idx, idxinv, _x, _y, _z) JOIN(pdf_,name) = src[adjList[adjListIndex + idx]];
132	D3Q19_LIST_WO_C
133	#undef X
134
135	#else
136	#error No implementation for PROP_MODEL_NAME.
137	#endif
138
139	// #define LID_DRIVEN_CAVITY
140
141	#ifdef LID_DRIVEN_CAVITY
142	int nX = kd->Dims[0];
143	int nY = kd->Dims[1];
144	int nZ = kd->Dims[2];
145
146	int x = kdl->Coords[C_INDEX_X(index)];
147	int y = kdl->Coords[C_INDEX_Y(index)];
148	int z = kdl->Coords[C_INDEX_Z(index)];
149
150	if (z == nZ - 4 && x > 3 && x < (nX - 4) && y > 3 && y < (nY - 4)) {
151	ux = 0.1 * 0.577;
152	uy = 0.0;
153	uz = 0.0;
154	} else {
155	#endif
156	ux = pdf_E + pdf_NE + pdf_SE + pdf_TE + pdf_BE -
157	pdf_W - pdf_NW - pdf_SW - pdf_TW - pdf_BW;
158	uy = pdf_N + pdf_NE + pdf_NW + pdf_TN + pdf_BN -
159	pdf_S - pdf_SE - pdf_SW - pdf_TS - pdf_BS;
160	uz = pdf_T + pdf_TE + pdf_TW + pdf_TN + pdf_TS -
161	pdf_B - pdf_BE - pdf_BW - pdf_BN - pdf_BS;
162	#ifdef LID_DRIVEN_CAVITY
163	}
164	#endif
165
166	dens = pdf_C +
167	pdf_N + pdf_E + pdf_S + pdf_W +
168	pdf_NE + pdf_SE + pdf_SW + pdf_NW +
169	pdf_T + pdf_TN + pdf_TE + pdf_TS + pdf_TW +
170	pdf_B + pdf_BN + pdf_BE + pdf_BS + pdf_BW;
171
172	dir_indep_trm = dens - (ux * ux + uy * uy + uz * uz)*3.0/2.0;
173
174	#ifdef PROP_MODEL_PUSH
175
176	adjListIndex = index * N_D3Q19_IDX;
177
178	// direction: w_0
179	dst[I(index, D3Q19_C) ] = pdf_C - omegaEven(pdf_C - w_0dir_indep_trm);
180
181	// direction: w_1
182	w_1_indep = w_1*dir_indep_trm;
183
184	ui = uy;
185	evenPart = omegaEven( 0.5(pdf_N + pdf_S) - uiuiw_1_nine_half - w_1_indep );
186	oddPart = omegaOdd(0.5(pdf_N - pdf_S) - ui*w_1_x3 );
187	dst[adjList[adjListIndex + D3Q19_N]] = pdf_N - evenPart - oddPart;
188	dst[adjList[adjListIndex + D3Q19_S]] = pdf_S - evenPart + oddPart;
189
190	ui = ux;
191	evenPart = omegaEven( 0.5(pdf_E + pdf_W) - uiuiw_1_nine_half - w_1_indep );
192	oddPart = omegaOdd(0.5(pdf_E - pdf_W) - ui*w_1_x3 );
193	dst[adjList[adjListIndex + D3Q19_E]] = pdf_E - evenPart - oddPart;
194	dst[adjList[adjListIndex + D3Q19_W]] = pdf_W - evenPart + oddPart;
195
196	ui = uz;
197	evenPart = omegaEven( 0.5(pdf_T + pdf_B) - uiuiw_1_nine_half - w_1_indep );
198	oddPart = omegaOdd(0.5(pdf_T - pdf_B) - ui*w_1_x3 );
199	dst[adjList[adjListIndex + D3Q19_T]] = pdf_T - evenPart - oddPart;
200	dst[adjList[adjListIndex + D3Q19_B]] = pdf_B - evenPart + oddPart;
201
202	// direction: w_2
203	w_2_indep = w_2*dir_indep_trm;
204
205	ui = -ux + uy;
206	evenPart = omegaEven( 0.5(pdf_NW + pdf_SE) - uiuiw_2_nine_half - w_2_indep );
207	oddPart = omegaOdd(0.5(pdf_NW - pdf_SE) - ui*w_2_x3 );
208	dst[adjList[adjListIndex + D3Q19_NW]] = pdf_NW - evenPart - oddPart;
209	dst[adjList[adjListIndex + D3Q19_SE]] = pdf_SE - evenPart + oddPart;
210
211	ui = ux + uy;
212	evenPart = omegaEven( 0.5(pdf_NE + pdf_SW) - uiuiw_2_nine_half - w_2_indep );
213	oddPart = omegaOdd(0.5(pdf_NE - pdf_SW) - ui*w_2_x3 );
214	dst[adjList[adjListIndex + D3Q19_NE]] = pdf_NE - evenPart - oddPart;
215	dst[adjList[adjListIndex + D3Q19_SW]] = pdf_SW - evenPart + oddPart;
216
217	ui = -ux + uz;
218	evenPart = omegaEven( 0.5(pdf_TW + pdf_BE) - uiuiw_2_nine_half - w_2_indep );
219	oddPart = omegaOdd(0.5(pdf_TW - pdf_BE) - ui*w_2_x3 );
220	dst[adjList[adjListIndex + D3Q19_TW]] = pdf_TW - evenPart - oddPart;
221	dst[adjList[adjListIndex + D3Q19_BE]] = pdf_BE - evenPart + oddPart;
222
223	ui = ux + uz;
224	evenPart = omegaEven( 0.5(pdf_TE + pdf_BW) - uiuiw_2_nine_half - w_2_indep );
225	oddPart = omegaOdd(0.5(pdf_TE - pdf_BW) - ui*w_2_x3 );
226	dst[adjList[adjListIndex + D3Q19_TE]] = pdf_TE - evenPart - oddPart;
227	dst[adjList[adjListIndex + D3Q19_BW]] = pdf_BW - evenPart + oddPart;
228
229	ui = -uy + uz;
230	evenPart = omegaEven( 0.5(pdf_TS + pdf_BN) - uiuiw_2_nine_half - w_2_indep );
231	oddPart = omegaOdd(0.5(pdf_TS - pdf_BN) - ui*w_2_x3 );
232	dst[adjList[adjListIndex + D3Q19_TS]] = pdf_TS - evenPart - oddPart;
233	dst[adjList[adjListIndex + D3Q19_BN]] = pdf_BN - evenPart + oddPart;
234
235	ui = uy + uz;
236	evenPart = omegaEven( 0.5(pdf_TN + pdf_BS) - uiuiw_2_nine_half - w_2_indep );
237	oddPart = omegaOdd(0.5(pdf_TN - pdf_BS) - ui*w_2_x3 );
238	dst[adjList[adjListIndex + D3Q19_TN]] = pdf_TN - evenPart - oddPart;
239	dst[adjList[adjListIndex + D3Q19_BS]] = pdf_BS - evenPart + oddPart;
240
241	#elif PROP_MODEL_PULL
242
243	// direction: w_0
244	dst[I(index, D3Q19_C )] = pdf_C - omegaEven(pdf_C - w_0dir_indep_trm);
245
246	// direction: w_1
247	w_1_indep = w_1*dir_indep_trm;
248
249	ui = uy;
250	evenPart = omegaEven( 0.5(pdf_N + pdf_S) - uiuiw_1_nine_half - w_1_indep );
251	oddPart = omegaOdd(0.5(pdf_N - pdf_S) - ui*w_1_x3 );
252	dst[I(index, D3Q19_N )] = pdf_N - evenPart - oddPart;
253	dst[I(index, D3Q19_S )] = pdf_S - evenPart + oddPart;
254
255	ui = ux;
256	evenPart = omegaEven( 0.5(pdf_E + pdf_W) - uiuiw_1_nine_half - w_1_indep );
257	oddPart = omegaOdd(0.5(pdf_E - pdf_W) - ui*w_1_x3 );
258	dst[I(index, D3Q19_E )] = pdf_E - evenPart - oddPart;
259	dst[I(index, D3Q19_W )] = pdf_W - evenPart + oddPart;
260
261	ui = uz;
262	evenPart = omegaEven( 0.5(pdf_T + pdf_B) - uiuiw_1_nine_half - w_1_indep );
263	oddPart = omegaOdd(0.5(pdf_T - pdf_B) - ui*w_1_x3 );
264	dst[I(index, D3Q19_T )] = pdf_T - evenPart - oddPart;
265	dst[I(index, D3Q19_B )] = pdf_B - evenPart + oddPart;
266
267	// direction: w_2
268	w_2_indep = w_2*dir_indep_trm;
269
270	ui = -ux + uy;
271	evenPart = omegaEven( 0.5(pdf_NW + pdf_SE) - uiuiw_2_nine_half - w_2_indep );
272	oddPart = omegaOdd(0.5(pdf_NW - pdf_SE) - ui*w_2_x3 );
273	dst[I(index, D3Q19_NW)] = pdf_NW - evenPart - oddPart;
274	dst[I(index, D3Q19_SE)] = pdf_SE - evenPart + oddPart;
275
276	ui = ux + uy;
277	evenPart = omegaEven( 0.5(pdf_NE + pdf_SW) - uiuiw_2_nine_half - w_2_indep );
278	oddPart = omegaOdd(0.5(pdf_NE - pdf_SW) - ui*w_2_x3 );
279	dst[I(index, D3Q19_NE)] = pdf_NE - evenPart - oddPart;
280	dst[I(index, D3Q19_SW)] = pdf_SW - evenPart + oddPart;
281
282	ui = -ux + uz;
283	evenPart = omegaEven( 0.5(pdf_TW + pdf_BE) - uiuiw_2_nine_half - w_2_indep );
284	oddPart = omegaOdd(0.5(pdf_TW - pdf_BE) - ui*w_2_x3 );
285	dst[I(index, D3Q19_TW)] = pdf_TW - evenPart - oddPart;
286	dst[I(index, D3Q19_BE)] = pdf_BE - evenPart + oddPart;
287
288	ui = ux + uz;
289	evenPart = omegaEven( 0.5(pdf_TE + pdf_BW) - uiuiw_2_nine_half - w_2_indep );
290	oddPart = omegaOdd(0.5(pdf_TE - pdf_BW) - ui*w_2_x3 );
291	dst[I(index, D3Q19_TE)] = pdf_TE - evenPart - oddPart;
292	dst[I(index, D3Q19_BW)] = pdf_BW - evenPart + oddPart;
293
294	ui = -uy + uz;
295	evenPart = omegaEven( 0.5(pdf_TS + pdf_BN) - uiuiw_2_nine_half - w_2_indep );
296	oddPart = omegaOdd(0.5(pdf_TS - pdf_BN) - ui*w_2_x3 );
297	dst[I(index, D3Q19_TS)] = pdf_TS - evenPart - oddPart;
298	dst[I(index, D3Q19_BN)] = pdf_BN - evenPart + oddPart;
299
300	ui = uy + uz;
301	evenPart = omegaEven( 0.5(pdf_TN + pdf_BS) - uiuiw_2_nine_half - w_2_indep );
302	oddPart = omegaOdd(0.5(pdf_TN - pdf_BS) - ui*w_2_x3 );
303	dst[I(index, D3Q19_TN)] = pdf_TN - evenPart - oddPart;
304	dst[I(index, D3Q19_BS)] = pdf_BS - evenPart + oddPart;
305
306	#endif
307	#undef I
308	} // loop over fluid nodes
309
310	#ifdef VERIFICATION
311	kd->PdfsActive = dst;
312	KernelAddBodyForce(kd, ld, cd);
313	#endif
314
315	#ifdef VTK_OUTPUT
316	if (cd->VtkOutput && (iter % cd->VtkModulus) == 0) {
317	kd->PdfsActive = dst;
318	VtkWrite(ld, kd, cd, iter);
319	}
320	#endif
321
322	#ifdef STATISTICS
323	kd->PdfsActive = dst;
324	KernelStatistics(kd, ld, cd, iter);
325	#endif
326
327	// swap grids
328	tmp = src;
329	src = dst;
330	dst = tmp;
331
332	} // for (int iter = 0; ...
333
334	#ifdef VTK_OUTPUT
335	if (cd->VtkOutput) {
336	kd->PdfsActive = src;
337	VtkWrite(ld, kd, cd, maxIterations);
338	}
339	#endif
340
341	#ifdef STATISTICS
342	kd->PdfsActive = src;
343	KernelStatistics(kd, ld, cd, maxIterations);
344	#endif
345
346	return;
347	}