WIP: Pull Request for a complete Solver package #2

Closed
AdityaUjeniya wants to merge 51 commits from (deleted):main into main
331 changed files with 18625 additions and 8575 deletions

View File

@ -1,3 +1,3 @@
CompileFlags: # Tweak the parse settings
Add: [-I/usr/local/include] # treat all files as C++, enable more warnings
CompileFlags:
Add: [-I/usr/local/include, -I/opt/homebrew/include, -D_MPI]
Compiler: clang

View File

@ -1,46 +0,0 @@
#==============================================================================
# Laminar Canal Flow
#==============================================================================
# Problem specific Data:
# ---------------------
name canal # name of flow setup
bcLeft 3 # flags for boundary conditions
bcRight 3 # 1 = no-slip 3 = outflow
bcBottom 1 # 2 = free-slip 4 = periodic
bcTop 1 #
gx 0.0 # Body forces (e.g. gravity)
gy 0.0 #
re 100.0 # Reynolds number
u_init 1.0 # initial value for velocity in x-direction
v_init 0.0 # initial value for velocity in y-direction
p_init 0.0 # initial value for pressure
# Geometry Data:
# -------------
xlength 30.0 # domain size in x-direction
ylength 4.0 # domain size in y-direction
imax 200 # number of interior cells in x-direction
jmax 50 # number of interior cells in y-direction
# Time Data:
# ---------
te 100.0 # final time
dt 0.02 # time stepsize
tau 0.5 # safety factor for time stepsize control (<0 constant delt)
# Pressure Iteration Data:
# -----------------------
itermax 500 # maximal number of pressure iteration in one time step
eps 0.00001 # stopping tolerance for pressure iteration
omg 1.8 # relaxation parameter for SOR iteration
gamma 0.9 # upwind differencing factor gamma
#===============================================================================

View File

@ -1,46 +0,0 @@
#==============================================================================
# Driven Cavity
#==============================================================================
# Problem specific Data:
# ---------------------
name dcavity # name of flow setup
bcLeft 1 # flags for boundary conditions
bcRight 1 # 1 = no-slip 3 = outflow
bcBottom 1 # 2 = free-slip 4 = periodic
bcTop 1 #
gx 0.0 # Body forces (e.g. gravity)
gy 0.0 #
re 500.0 # Reynolds number
u_init 0.0 # initial value for velocity in x-direction
v_init 0.0 # initial value for velocity in y-direction
p_init 0.0 # initial value for pressure
# Geometry Data:
# -------------
xlength 1.0 # domain size in x-direction
ylength 1.0 # domain size in y-direction
imax 100 # number of interior cells in x-direction
jmax 100 # number of interior cells in y-direction
# Time Data:
# ---------
te 25.0 # final time
dt 0.02 # time stepsize
tau 0.5 # safety factor for time stepsize control (<0 constant delt)
# Pressure Iteration Data:
# -----------------------
itermax 1000 # maximal number of pressure iteration in one time step
eps 0.001 # stopping tolerance for pressure iteration
omg 1.7 # relaxation parameter for SOR iteration
gamma 0.9 # upwind differencing factor gamma
#===============================================================================

View File

@ -1,16 +0,0 @@
CC = mpicc
GCC = cc
LINKER = $(CC)
ifeq ($(ENABLE_OPENMP),true)
OPENMP = -fopenmp
#OPENMP = -Xpreprocessor -fopenmp #required on Macos with homebrew libomp
LIBS = # -lomp
endif
VERSION = --version
CFLAGS = -Ofast -std=c99 $(OPENMP)
#CFLAGS = -Ofast -fnt-store=aggressive -std=c99 $(OPENMP) #AMD CLANG
LFLAGS = $(OPENMP)
DEFINES = -D_GNU_SOURCE# -DDEBUG
INCLUDES = -I/usr/local/include

View File

@ -1,61 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#ifdef __linux__
#ifdef _OPENMP
#include <pthread.h>
#include <sched.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/syscall.h>
#include <sys/types.h>
#include <unistd.h>
#define MAX_NUM_THREADS 128
#define gettid() syscall(SYS_gettid)
static int getProcessorID(cpu_set_t* cpu_set)
{
int processorId;
for (processorId = 0; processorId < MAX_NUM_THREADS; processorId++) {
if (CPU_ISSET(processorId, cpu_set)) {
break;
}
}
return processorId;
}
int affinity_getProcessorId()
{
cpu_set_t cpu_set;
CPU_ZERO(&cpu_set);
sched_getaffinity(gettid(), sizeof(cpu_set_t), &cpu_set);
return getProcessorID(&cpu_set);
}
void affinity_pinThread(int processorId)
{
cpu_set_t cpuset;
pthread_t thread;
thread = pthread_self();
CPU_ZERO(&cpuset);
CPU_SET(processorId, &cpuset);
pthread_setaffinity_np(thread, sizeof(cpu_set_t), &cpuset);
}
void affinity_pinProcess(int processorId)
{
cpu_set_t cpuset;
CPU_ZERO(&cpuset);
CPU_SET(processorId, &cpuset);
sched_setaffinity(0, sizeof(cpu_set_t), &cpuset);
}
#endif /*_OPENMP*/
#endif /*__linux__*/

View File

@ -1,14 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#ifndef AFFINITY_H
#define AFFINITY_H
extern int affinity_getProcessorId();
extern void affinity_pinProcess(int);
extern void affinity_pinThread(int);
#endif /*AFFINITY_H*/

View File

@ -1,79 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#include <float.h>
#include <limits.h>
#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include "parameter.h"
#include "progress.h"
#include "solver.h"
#include "timing.h"
#include <mpi.h>
int main(int argc, char** argv)
{
int rank;
double S, E;
Parameter params;
Solver solver;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
initParameter(&params);
if (argc != 2) {
printf("Usage: %s <configFile>\n", argv[0]);
exit(EXIT_SUCCESS);
}
readParameter(&params, argv[1]);
if (rank == 0) {
printParameter(&params);
}
initSolver(&solver, &params);
initProgress(solver.te);
double tau = solver.tau;
double te = solver.te;
double t = 0.0;
S = getTimeStamp();
while (t <= te) {
if (tau > 0.0) {
computeTimestep(&solver);
}
setBoundaryConditions(&solver);
setSpecialBoundaryCondition(&solver);
computeFG(&solver);
computeRHS(&solver);
solve(&solver);
adaptUV(&solver);
/* exit(EXIT_SUCCESS); */
t += solver.dt;
#ifdef VERBOSE
if (rank == 0) {
printf("TIME %f , TIMESTEP %f\n", t, solver.dt);
}
#else
printProgress(t);
#endif
}
E = getTimeStamp();
stopProgress();
if (rank == 0) {
printf("Solution took %.2fs\n", E - S);
}
collectResult(&solver);
MPI_Finalize();
return EXIT_SUCCESS;
}

View File

@ -1,26 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#ifndef __PARAMETER_H_
#define __PARAMETER_H_
typedef struct {
double xlength, ylength;
int imax, jmax;
int itermax;
double eps, omg;
double re, tau, gamma;
double te, dt;
double gx, gy;
char* name;
int bcLeft, bcRight, bcBottom, bcTop;
double u_init, v_init, p_init;
} Parameter;
void initParameter(Parameter*);
void readParameter(Parameter*, const char*);
void printParameter(Parameter*);
#endif

View File

@ -1,60 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <math.h>
#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "progress.h"
static double _end;
static int _current;
static int _rank = -1;
void initProgress(double end)
{
MPI_Comm_rank(MPI_COMM_WORLD, &_rank);
_end = end;
_current = 0;
if (_rank == 0) {
printf("[ ]");
fflush(stdout);
}
}
void printProgress(double current)
{
if (_rank == 0) {
int new = (int)rint((current / _end) * 10.0);
if (new > _current) {
char progress[11];
_current = new;
progress[0] = 0;
for (int i = 0; i < 10; i++) {
if (i < _current) {
sprintf(progress + strlen(progress), "#");
} else {
sprintf(progress + strlen(progress), " ");
}
}
printf("\r[%s]", progress);
}
fflush(stdout);
}
}
void stopProgress()
{
if (_rank == 0) {
printf("\n");
fflush(stdout);
}
}

View File

@ -1,689 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <float.h>
#include <math.h>
#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "allocate.h"
#include "parameter.h"
#include "solver.h"
#include "util.h"
#define P(i, j) p[(j) * (imax + 2) + (i)]
#define F(i, j) f[(j) * (imax + 2) + (i)]
#define G(i, j) g[(j) * (imax + 2) + (i)]
#define U(i, j) u[(j) * (imax + 2) + (i)]
#define V(i, j) v[(j) * (imax + 2) + (i)]
#define RHS(i, j) rhs[(j) * (imax + 2) + (i)]
static int sizeOfRank(int rank, int size, int N)
{
return N / size + ((N % size > rank) ? 1 : 0);
}
static void print(Solver* solver, double* grid)
{
int imax = solver->imax;
for (int i = 0; i < solver->size; i++) {
if (i == solver->rank) {
printf("### RANK %d "
"#######################################################\n",
solver->rank);
for (int j = 0; j < solver->jmaxLocal + 2; j++) {
printf("%02d: ", j);
for (int i = 0; i < solver->imax + 2; i++) {
printf("%12.8f ", grid[j * (imax + 2) + i]);
}
printf("\n");
}
fflush(stdout);
}
MPI_Barrier(MPI_COMM_WORLD);
}
}
static void exchange(Solver* solver, double* grid)
{
MPI_Request requests[4] = { MPI_REQUEST_NULL,
MPI_REQUEST_NULL,
MPI_REQUEST_NULL,
MPI_REQUEST_NULL };
/* exchange ghost cells with top neighbor */
if (solver->rank + 1 < solver->size) {
int top = solver->rank + 1;
double* src = grid + (solver->jmaxLocal) * (solver->imax + 2) + 1;
double* dst = grid + (solver->jmaxLocal + 1) * (solver->imax + 2) + 1;
MPI_Isend(src, solver->imax, MPI_DOUBLE, top, 1, MPI_COMM_WORLD, &requests[0]);
MPI_Irecv(dst, solver->imax, MPI_DOUBLE, top, 2, MPI_COMM_WORLD, &requests[1]);
}
/* exchange ghost cells with bottom neighbor */
if (solver->rank > 0) {
int bottom = solver->rank - 1;
double* src = grid + (solver->imax + 2) + 1;
double* dst = grid + 1;
MPI_Isend(src, solver->imax, MPI_DOUBLE, bottom, 2, MPI_COMM_WORLD, &requests[2]);
MPI_Irecv(dst, solver->imax, MPI_DOUBLE, bottom, 1, MPI_COMM_WORLD, &requests[3]);
}
MPI_Waitall(4, requests, MPI_STATUSES_IGNORE);
}
static void shift(Solver* solver)
{
MPI_Request requests[2] = { MPI_REQUEST_NULL, MPI_REQUEST_NULL };
double* g = solver->g;
/* shift G */
/* receive ghost cells from bottom neighbor */
if (solver->rank > 0) {
int bottom = solver->rank - 1;
MPI_Irecv(g + 1,
solver->imax,
MPI_DOUBLE,
bottom,
0,
MPI_COMM_WORLD,
&requests[0]);
}
if (solver->rank + 1 < solver->size) {
int top = solver->rank + 1;
double* buf = g + (solver->jmaxLocal) * (solver->imax + 2) + 1;
/* send ghost cells to top neighbor */
MPI_Isend(buf, solver->imax, MPI_DOUBLE, top, 0, MPI_COMM_WORLD, &requests[1]);
}
MPI_Waitall(2, requests, MPI_STATUSES_IGNORE);
}
void collectResult(Solver* solver)
{
double* Pall = NULL;
double* Uall = NULL;
double* Vall = NULL;
int *rcvCounts, *displs;
if (solver->rank == 0) {
Pall = allocate(64, (solver->imax + 2) * (solver->jmax + 2) * sizeof(double));
Uall = allocate(64, (solver->imax + 2) * (solver->jmax + 2) * sizeof(double));
Vall = allocate(64, (solver->imax + 2) * (solver->jmax + 2) * sizeof(double));
rcvCounts = (int*)malloc(solver->size * sizeof(int));
displs = (int*)malloc(solver->size * sizeof(int));
rcvCounts[0] = solver->jmaxLocal * (solver->imax + 2);
displs[0] = 0;
int cursor = rcvCounts[0];
for (int i = 1; i < solver->size; i++) {
rcvCounts[i] = sizeOfRank(i, solver->size, solver->jmax) * (solver->imax + 2);
displs[i] = cursor;
cursor += rcvCounts[i];
}
}
int cnt = solver->jmaxLocal * (solver->imax + 2);
double* sendbuffer = solver->p + (solver->imax + 2);
MPI_Gatherv(sendbuffer,
cnt,
MPI_DOUBLE,
Pall,
rcvCounts,
displs,
MPI_DOUBLE,
0,
MPI_COMM_WORLD);
sendbuffer = solver->u + (solver->imax + 2);
MPI_Gatherv(sendbuffer,
cnt,
MPI_DOUBLE,
Uall,
rcvCounts,
displs,
MPI_DOUBLE,
0,
MPI_COMM_WORLD);
sendbuffer = solver->v + (solver->imax + 2);
MPI_Gatherv(sendbuffer,
cnt,
MPI_DOUBLE,
Vall,
rcvCounts,
displs,
MPI_DOUBLE,
0,
MPI_COMM_WORLD);
if (solver->rank == 0) {
writeResult(solver, Pall, Uall, Vall);
}
}
static void printConfig(Solver* solver)
{
if (solver->rank == 0) {
printf("Parameters for #%s#\n", solver->problem);
printf("Boundary conditions Left:%d Right:%d Bottom:%d Top:%d\n",
solver->bcLeft,
solver->bcRight,
solver->bcBottom,
solver->bcTop);
printf("\tReynolds number: %.2f\n", solver->re);
printf("\tGx Gy: %.2f %.2f\n", solver->gx, solver->gy);
printf("Geometry data:\n");
printf("\tDomain box size (x, y): %.2f, %.2f\n",
solver->xlength,
solver->ylength);
printf("\tCells (x, y): %d, %d\n", solver->imax, solver->jmax);
printf("Timestep parameters:\n");
printf("\tDefault stepsize: %.2f, Final time %.2f\n", solver->dt, solver->te);
printf("\tdt bound: %.6f\n", solver->dtBound);
printf("\tTau factor: %.2f\n", solver->tau);
printf("Iterative solver parameters:\n");
printf("\tMax iterations: %d\n", solver->itermax);
printf("\tepsilon (stopping tolerance) : %f\n", solver->eps);
printf("\tgamma factor: %f\n", solver->gamma);
printf("\tomega (SOR relaxation): %f\n", solver->omega);
printf("Communication parameters:\n");
}
for (int i = 0; i < solver->size; i++) {
if (i == solver->rank) {
printf("\tRank %d of %d\n", solver->rank, solver->size);
printf("\tLocal domain size: %dx%d\n", solver->imax, solver->jmaxLocal);
fflush(stdout);
}
}
}
void initSolver(Solver* solver, Parameter* params)
{
MPI_Comm_rank(MPI_COMM_WORLD, &(solver->rank));
MPI_Comm_size(MPI_COMM_WORLD, &(solver->size));
solver->problem = params->name;
solver->bcLeft = params->bcLeft;
solver->bcRight = params->bcRight;
solver->bcBottom = params->bcBottom;
solver->bcTop = params->bcTop;
solver->imax = params->imax;
solver->jmax = params->jmax;
solver->jmaxLocal = sizeOfRank(solver->rank, solver->size, solver->jmax);
solver->xlength = params->xlength;
solver->ylength = params->ylength;
solver->dx = params->xlength / params->imax;
solver->dy = params->ylength / params->jmax;
solver->eps = params->eps;
solver->omega = params->omg;
solver->itermax = params->itermax;
solver->re = params->re;
solver->gx = params->gx;
solver->gy = params->gy;
solver->dt = params->dt;
solver->te = params->te;
solver->tau = params->tau;
solver->gamma = params->gamma;
int imax = solver->imax;
int jmaxLocal = solver->jmaxLocal;
size_t bytesize = (imax + 2) * (jmaxLocal + 2) * sizeof(double);
solver->u = allocate(64, bytesize);
solver->v = allocate(64, bytesize);
solver->p = allocate(64, bytesize);
solver->rhs = allocate(64, bytesize);
solver->f = allocate(64, bytesize);
solver->g = allocate(64, bytesize);
for (int i = 0; i < (imax + 2) * (jmaxLocal + 2); i++) {
solver->u[i] = params->u_init;
solver->v[i] = params->v_init;
solver->p[i] = params->p_init;
solver->rhs[i] = 0.0;
solver->f[i] = 0.0;
solver->g[i] = 0.0;
}
double dx = solver->dx;
double dy = solver->dy;
double inv_sqr_sum = 1.0 / (dx * dx) + 1.0 / (dy * dy);
solver->dtBound = 0.5 * solver->re * 1.0 / inv_sqr_sum;
#ifdef VERBOSE
printConfig(solver);
#endif
}
void computeRHS(Solver* solver)
{
int imax = solver->imax;
int jmaxLocal = solver->jmaxLocal;
double idx = 1.0 / solver->dx;
double idy = 1.0 / solver->dy;
double idt = 1.0 / solver->dt;
double* rhs = solver->rhs;
double* f = solver->f;
double* g = solver->g;
shift(solver);
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = 1; i < imax + 1; i++) {
RHS(i, j) = ((F(i, j) - F(i - 1, j)) * idx + (G(i, j) - G(i, j - 1)) * idy) *
idt;
}
}
}
void solve(Solver* solver)
{
int imax = solver->imax;
int jmax = solver->jmax;
int jmaxLocal = solver->jmaxLocal;
double eps = solver->eps;
int itermax = solver->itermax;
double dx2 = solver->dx * solver->dx;
double dy2 = solver->dy * solver->dy;
double idx2 = 1.0 / dx2;
double idy2 = 1.0 / dy2;
double factor = solver->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
double* p = solver->p;
double* rhs = solver->rhs;
double epssq = eps * eps;
int it = 0;
double res = 1.0;
while ((res >= epssq) && (it < itermax)) {
res = 0.0;
exchange(solver, p);
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = 1; i < imax + 1; i++) {
double r = RHS(i, j) -
((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
(P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2);
P(i, j) -= (factor * r);
res += (r * r);
}
}
if (solver->rank == 0) {
for (int i = 1; i < imax + 1; i++) {
P(i, 0) = P(i, 1);
}
}
if (solver->rank == (solver->size - 1)) {
for (int i = 1; i < imax + 1; i++) {
P(i, jmaxLocal + 1) = P(i, jmaxLocal);
}
}
for (int j = 1; j < jmaxLocal + 1; j++) {
P(0, j) = P(1, j);
P(imax + 1, j) = P(imax, j);
}
MPI_Allreduce(MPI_IN_PLACE, &res, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
res = res / (double)(imax * jmax);
#ifdef DEBUG
if (solver->rank == 0) {
printf("%d Residuum: %e\n", it, res);
}
#endif
it++;
}
#ifdef VERBOSE
if (solver->rank == 0) {
printf("Solver took %d iterations to reach %f\n", it, sqrt(res));
}
#endif
}
static double maxElement(Solver* solver, double* m)
{
int size = (solver->imax + 2) * (solver->jmaxLocal + 2);
double maxval = DBL_MIN;
for (int i = 0; i < size; i++) {
maxval = MAX(maxval, fabs(m[i]));
}
MPI_Allreduce(MPI_IN_PLACE, &maxval, 1, MPI_DOUBLE, MPI_MAX, MPI_COMM_WORLD);
return maxval;
}
void normalizePressure(Solver* solver)
{
int size = (solver->imax + 2) * (solver->jmaxLocal + 2);
double* p = solver->p;
double avgP = 0.0;
for (int i = 0; i < size; i++) {
avgP += p[i];
}
MPI_Allreduce(MPI_IN_PLACE, &avgP, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
avgP /= (solver->imax + 2) * (solver->jmax + 2);
for (int i = 0; i < size; i++) {
p[i] = p[i] - avgP;
}
}
void computeTimestep(Solver* solver)
{
double dt = solver->dtBound;
double dx = solver->dx;
double dy = solver->dy;
double umax = maxElement(solver, solver->u);
double vmax = maxElement(solver, solver->v);
if (umax > 0) {
dt = (dt > dx / umax) ? dx / umax : dt;
}
if (vmax > 0) {
dt = (dt > dy / vmax) ? dy / vmax : dt;
}
solver->dt = dt * solver->tau;
}
void setBoundaryConditions(Solver* solver)
{
int imax = solver->imax;
int jmaxLocal = solver->jmaxLocal;
double* u = solver->u;
double* v = solver->v;
// Left boundary
switch (solver->bcLeft) {
case NOSLIP:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(0, j) = 0.0;
V(0, j) = -V(1, j);
}
break;
case SLIP:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(0, j) = 0.0;
V(0, j) = V(1, j);
}
break;
case OUTFLOW:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(0, j) = U(1, j);
V(0, j) = V(1, j);
}
break;
case PERIODIC:
break;
}
// Right boundary
switch (solver->bcRight) {
case NOSLIP:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(imax, j) = 0.0;
V(imax + 1, j) = -V(imax, j);
}
break;
case SLIP:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(imax, j) = 0.0;
V(imax + 1, j) = V(imax, j);
}
break;
case OUTFLOW:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(imax, j) = U(imax - 1, j);
V(imax + 1, j) = V(imax, j);
}
break;
case PERIODIC:
break;
}
// Bottom boundary
if (solver->rank == 0) {
switch (solver->bcBottom) {
case NOSLIP:
for (int i = 1; i < imax + 1; i++) {
V(i, 0) = 0.0;
U(i, 0) = -U(i, 1);
}
break;
case SLIP:
for (int i = 1; i < imax + 1; i++) {
V(i, 0) = 0.0;
U(i, 0) = U(i, 1);
}
break;
case OUTFLOW:
for (int i = 1; i < imax + 1; i++) {
U(i, 0) = U(i, 1);
V(i, 0) = V(i, 1);
}
break;
case PERIODIC:
break;
}
}
// Top boundary
if (solver->rank == (solver->size - 1)) {
switch (solver->bcTop) {
case NOSLIP:
for (int i = 1; i < imax + 1; i++) {
V(i, jmaxLocal) = 0.0;
U(i, jmaxLocal + 1) = -U(i, jmaxLocal);
}
break;
case SLIP:
for (int i = 1; i < imax + 1; i++) {
V(i, jmaxLocal) = 0.0;
U(i, jmaxLocal + 1) = U(i, jmaxLocal);
}
break;
case OUTFLOW:
for (int i = 1; i < imax + 1; i++) {
U(i, jmaxLocal + 1) = U(i, jmaxLocal);
V(i, jmaxLocal) = V(i, jmaxLocal - 1);
}
break;
case PERIODIC:
break;
}
}
}
void setSpecialBoundaryCondition(Solver* solver)
{
int imax = solver->imax;
int jmaxLocal = solver->jmaxLocal;
double* u = solver->u;
if (strcmp(solver->problem, "dcavity") == 0) {
if (solver->rank == (solver->size - 1)) {
for (int i = 1; i < imax; i++) {
U(i, jmaxLocal + 1) = 2.0 - U(i, jmaxLocal);
}
}
} else if (strcmp(solver->problem, "canal") == 0) {
double ylength = solver->ylength;
double dy = solver->dy;
int rest = solver->jmax % solver->size;
int yc = solver->rank * (solver->jmax / solver->size) + MIN(rest, solver->rank);
double ys = dy * (yc + 0.5);
double y;
/* printf("RANK %d yc: %d ys: %f\n", solver->rank, yc, ys); */
for (int j = 1; j < jmaxLocal + 1; j++) {
y = ys + dy * (j - 0.5);
U(0, j) = y * (ylength - y) * 4.0 / (ylength * ylength);
}
}
/* print(solver, solver->u); */
}
void computeFG(Solver* solver)
{
double* u = solver->u;
double* v = solver->v;
double* f = solver->f;
double* g = solver->g;
int imax = solver->imax;
int jmaxLocal = solver->jmaxLocal;
double gx = solver->gx;
double gy = solver->gy;
double gamma = solver->gamma;
double dt = solver->dt;
double inverseRe = 1.0 / solver->re;
double inverseDx = 1.0 / solver->dx;
double inverseDy = 1.0 / solver->dy;
double du2dx, dv2dy, duvdx, duvdy;
double du2dx2, du2dy2, dv2dx2, dv2dy2;
exchange(solver, u);
exchange(solver, v);
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = 1; i < imax + 1; i++) {
du2dx = inverseDx * 0.25 *
((U(i, j) + U(i + 1, j)) * (U(i, j) + U(i + 1, j)) -
(U(i, j) + U(i - 1, j)) * (U(i, j) + U(i - 1, j))) +
gamma * inverseDx * 0.25 *
(fabs(U(i, j) + U(i + 1, j)) * (U(i, j) - U(i + 1, j)) +
fabs(U(i, j) + U(i - 1, j)) * (U(i, j) - U(i - 1, j)));
duvdy = inverseDy * 0.25 *
((V(i, j) + V(i + 1, j)) * (U(i, j) + U(i, j + 1)) -
(V(i, j - 1) + V(i + 1, j - 1)) * (U(i, j) + U(i, j - 1))) +
gamma * inverseDy * 0.25 *
(fabs(V(i, j) + V(i + 1, j)) * (U(i, j) - U(i, j + 1)) +
fabs(V(i, j - 1) + V(i + 1, j - 1)) *
(U(i, j) - U(i, j - 1)));
du2dx2 = inverseDx * inverseDx * (U(i + 1, j) - 2.0 * U(i, j) + U(i - 1, j));
du2dy2 = inverseDy * inverseDy * (U(i, j + 1) - 2.0 * U(i, j) + U(i, j - 1));
F(i, j) = U(i, j) + dt * (inverseRe * (du2dx2 + du2dy2) - du2dx - duvdy + gx);
duvdx = inverseDx * 0.25 *
((U(i, j) + U(i, j + 1)) * (V(i, j) + V(i + 1, j)) -
(U(i - 1, j) + U(i - 1, j + 1)) * (V(i, j) + V(i - 1, j))) +
gamma * inverseDx * 0.25 *
(fabs(U(i, j) + U(i, j + 1)) * (V(i, j) - V(i + 1, j)) +
fabs(U(i - 1, j) + U(i - 1, j + 1)) *
(V(i, j) - V(i - 1, j)));
dv2dy = inverseDy * 0.25 *
((V(i, j) + V(i, j + 1)) * (V(i, j) + V(i, j + 1)) -
(V(i, j) + V(i, j - 1)) * (V(i, j) + V(i, j - 1))) +
gamma * inverseDy * 0.25 *
(fabs(V(i, j) + V(i, j + 1)) * (V(i, j) - V(i, j + 1)) +
fabs(V(i, j) + V(i, j - 1)) * (V(i, j) - V(i, j - 1)));
dv2dx2 = inverseDx * inverseDx * (V(i + 1, j) - 2.0 * V(i, j) + V(i - 1, j));
dv2dy2 = inverseDy * inverseDy * (V(i, j + 1) - 2.0 * V(i, j) + V(i, j - 1));
G(i, j) = V(i, j) + dt * (inverseRe * (dv2dx2 + dv2dy2) - duvdx - dv2dy + gy);
}
}
/* ----------------------------- boundary of F ---------------------------
*/
for (int j = 1; j < jmaxLocal + 1; j++) {
F(0, j) = U(0, j);
F(imax, j) = U(imax, j);
}
/* ----------------------------- boundary of G ---------------------------
*/
if (solver->rank == 0) {
for (int i = 1; i < imax + 1; i++) {
G(i, 0) = V(i, 0);
}
}
if (solver->rank == (solver->size - 1)) {
for (int i = 1; i < imax + 1; i++) {
G(i, jmaxLocal) = V(i, jmaxLocal);
}
}
}
void adaptUV(Solver* solver)
{
int imax = solver->imax;
int jmaxLocal = solver->jmaxLocal;
double* p = solver->p;
double* u = solver->u;
double* v = solver->v;
double* f = solver->f;
double* g = solver->g;
double factorX = solver->dt / solver->dx;
double factorY = solver->dt / solver->dy;
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = 1; i < imax + 1; i++) {
U(i, j) = F(i, j) - (P(i + 1, j) - P(i, j)) * factorX;
V(i, j) = G(i, j) - (P(i, j + 1) - P(i, j)) * factorY;
}
}
}
void writeResult(Solver* solver, double* p, double* u, double* v)
{
int imax = solver->imax;
int jmax = solver->jmax;
double dx = solver->dx;
double dy = solver->dy;
double x = 0.0, y = 0.0;
FILE* fp;
fp = fopen("pressure.dat", "w");
if (fp == NULL) {
printf("Error!\n");
exit(EXIT_FAILURE);
}
for (int j = 1; j < jmax + 1; j++) {
y = (double)(j - 0.5) * dy;
for (int i = 1; i < imax + 1; i++) {
x = (double)(i - 0.5) * dx;
fprintf(fp, "%.2f %.2f %f\n", x, y, P(i, j));
}
fprintf(fp, "\n");
}
fclose(fp);
fp = fopen("velocity.dat", "w");
if (fp == NULL) {
printf("Error!\n");
exit(EXIT_FAILURE);
}
for (int j = 1; j < jmax + 1; j++) {
y = dy * (j - 0.5);
for (int i = 1; i < imax + 1; i++) {
x = dx * (i - 0.5);
double vel_u = (U(i, j) + U(i - 1, j)) / 2.0;
double vel_v = (V(i, j) + V(i, j - 1)) / 2.0;
double len = sqrt((vel_u * vel_u) + (vel_v * vel_v));
fprintf(fp, "%.2f %.2f %f %f %f\n", x, y, vel_u, vel_v, len);
}
}
fclose(fp);
}

View File

@ -1,49 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#ifndef __SOLVER_H_
#define __SOLVER_H_
#include "parameter.h"
enum BC { NOSLIP = 1, SLIP, OUTFLOW, PERIODIC };
typedef struct {
/* geometry and grid information */
double dx, dy;
int imax, jmax;
int jmaxLocal;
double xlength, ylength;
/* arrays */
double *p, *rhs;
double *f, *g;
double *u, *v;
/* parameters */
double eps, omega;
double re, tau, gamma;
double gx, gy;
/* time stepping */
int itermax;
double dt, te;
double dtBound;
char* problem;
int bcLeft, bcRight, bcBottom, bcTop;
/* mpi */
int rank;
int size;
} Solver;
void initSolver(Solver*, Parameter*);
void computeRHS(Solver*);
void solve(Solver*);
void normalizePressure(Solver*);
void computeTimestep(Solver*);
void setBoundaryConditions(Solver*);
void setSpecialBoundaryCondition(Solver*);
void computeFG(Solver*);
void adaptUV(Solver*);
void collectResult(Solver*);
void writeResult(Solver*, double*, double*, double*);
#endif

View File

@ -1,23 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#ifndef __UTIL_H_
#define __UTIL_H_
#define HLINE \
"------------------------------------------------------------------------" \
"----\n"
#ifndef MIN
#define MIN(x, y) ((x) < (y) ? (x) : (y))
#endif
#ifndef MAX
#define MAX(x, y) ((x) > (y) ? (x) : (y))
#endif
#ifndef ABS
#define ABS(a) ((a) >= 0 ? (a) : -(a))
#endif
#endif // __UTIL_H_

View File

@ -1,5 +0,0 @@
set terminal png size 1800,768 enhanced font ,12
set output 'velocity.png'
set datafile separator whitespace
plot 'velocity.dat' using 1:2:3:4:5 with vectors filled head size 0.01,20,60 lc palette

View File

@ -1,48 +0,0 @@
# C source skeleton
## Build
1. Configure the toolchain and additional options in `config.mk`:
```
# Supported: GCC, CLANG, ICC
TAG ?= GCC
ENABLE_OPENMP ?= false
OPTIONS += -DARRAY_ALIGNMENT=64
#OPTIONS += -DVERBOSE_AFFINITY
#OPTIONS += -DVERBOSE_DATASIZE
#OPTIONS += -DVERBOSE_TIMER
```
The verbosity options enable detailed output about affinity settings, allocation sizes and timer resolution.
2. Build with:
```
make
```
You can build multiple toolchains in the same directory, but notice that the Makefile is only acting on the one currently set.
Intermediate build results are located in the `<TOOLCHAIN>` directory.
To output the executed commands use:
```
make Q=
```
3. Clean up with:
```
make clean
```
to clean intermediate build results.
```
make distclean
```
to clean intermediate build results and binary.
4. (Optional) Generate assembler:
```
make asm
```
The assembler files will also be located in the `<TOOLCHAIN>` directory.

View File

@ -1,10 +0,0 @@
# Supported: GCC, CLANG, ICC
TAG ?= CLANG
ENABLE_OPENMP ?= false
#Feature options
OPTIONS += -DARRAY_ALIGNMENT=64
# OPTIONS += -DVERBOSE
#OPTIONS += -DVERBOSE_AFFINITY
#OPTIONS += -DVERBOSE_DATASIZE
#OPTIONS += -DVERBOSE_TIMER

View File

@ -1,46 +0,0 @@
#==============================================================================
# Driven Cavity
#==============================================================================
# Problem specific Data:
# ---------------------
name dcavity # name of flow setup
bcN 1 # flags for boundary conditions
bcE 1 # 1 = no-slip 3 = outflow
bcS 1 # 2 = free-slip 4 = periodic
bcW 1 #
gx 0.0 # Body forces (e.g. gravity)
gy 0.0 #
re 1000.0 # Reynolds number
u_init 0.0 # initial value for velocity in x-direction
v_init 0.0 # initial value for velocity in y-direction
p_init 0.0 # initial value for pressure
# Geometry Data:
# -------------
xlength 1.0 # domain size in x-direction
ylength 1.0 # domain size in y-direction
imax 100 # number of interior cells in x-direction
jmax 100 # number of interior cells in y-direction
# Time Data:
# ---------
te 10.0 # final time
dt 0.02 # time stepsize
tau 0.5 # safety factor for time stepsize control (<0 constant delt)
# Pressure Iteration Data:
# -----------------------
itermax 1000 # maximal number of pressure iteration in one time step
eps 0.001 # stopping tolerance for pressure iteration
omg 1.7 # relaxation parameter for SOR iteration
gamma 0.9 # upwind differencing factor gamma
#===============================================================================

View File

@ -1,16 +0,0 @@
CC = mpicc
GCC = cc
LINKER = $(CC)
ifeq ($(ENABLE_OPENMP),true)
OPENMP = -fopenmp
#OPENMP = -Xpreprocessor -fopenmp #required on Macos with homebrew libomp
LIBS = # -lomp
endif
VERSION = --version
CFLAGS = -Ofast -std=c99 $(OPENMP)
#CFLAGS = -Ofast -fnt-store=aggressive -std=c99 $(OPENMP) #AMD CLANG
LFLAGS = $(OPENMP)
DEFINES = -D_GNU_SOURCE# -DDEBUG
INCLUDES = -I/usr/local/include

View File

@ -1,14 +0,0 @@
CC = mpiicc
GCC = gcc
LINKER = $(CC)
ifeq ($(ENABLE_OPENMP),true)
OPENMP = -qopenmp
endif
VERSION = --version
CFLAGS = -O3 -xHost -qopt-zmm-usage=high -std=c99 $(OPENMP)
LFLAGS = $(OPENMP)
DEFINES = -D_GNU_SOURCE
INCLUDES =
LIBS =

View File

@ -1,61 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#ifdef __linux__
#ifdef _OPENMP
#include <pthread.h>
#include <sched.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/syscall.h>
#include <sys/types.h>
#include <unistd.h>
#define MAX_NUM_THREADS 128
#define gettid() syscall(SYS_gettid)
static int getProcessorID(cpu_set_t* cpu_set)
{
int processorId;
for (processorId = 0; processorId < MAX_NUM_THREADS; processorId++) {
if (CPU_ISSET(processorId, cpu_set)) {
break;
}
}
return processorId;
}
int affinity_getProcessorId()
{
cpu_set_t cpu_set;
CPU_ZERO(&cpu_set);
sched_getaffinity(gettid(), sizeof(cpu_set_t), &cpu_set);
return getProcessorID(&cpu_set);
}
void affinity_pinThread(int processorId)
{
cpu_set_t cpuset;
pthread_t thread;
thread = pthread_self();
CPU_ZERO(&cpuset);
CPU_SET(processorId, &cpuset);
pthread_setaffinity_np(thread, sizeof(cpu_set_t), &cpuset);
}
void affinity_pinProcess(int processorId)
{
cpu_set_t cpuset;
CPU_ZERO(&cpuset);
CPU_SET(processorId, &cpuset);
sched_setaffinity(0, sizeof(cpu_set_t), &cpuset);
}
#endif /*_OPENMP*/
#endif /*__linux__*/

View File

@ -1,80 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#include <float.h>
#include <limits.h>
#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include "parameter.h"
#include "progress.h"
#include "solver.h"
#include "timing.h"
int main(int argc, char** argv)
{
int rank;
double S, E;
Parameter params;
Solver solver;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
initParameter(&params);
if (argc != 2) {
printf("Usage: %s <configFile>\n", argv[0]);
exit(EXIT_SUCCESS);
}
readParameter(&params, argv[1]);
if (rank == 0) {
printParameter(&params);
}
initSolver(&solver, &params);
/* debugExchange(&solver); */
/* debugBC(&solver); */
/* exit(EXIT_SUCCESS); */
initProgress(solver.te);
double tau = solver.tau;
double te = solver.te;
double t = 0.0;
S = getTimeStamp();
while (t <= te) {
if (tau > 0.0) {
computeTimestep(&solver);
}
setBoundaryConditions(&solver);
setSpecialBoundaryCondition(&solver);
computeFG(&solver);
computeRHS(&solver);
solve(&solver);
adaptUV(&solver);
t += solver.dt;
#ifdef VERBOSE
if (rank == 0) {
printf("TIME %f , TIMESTEP %f\n", t, solver.dt);
}
#else
printProgress(t);
#endif
}
E = getTimeStamp();
stopProgress();
if (rank == 0) {
printf("Solution took %.2fs\n", E - S);
}
collectResult(&solver);
MPI_Finalize();
return EXIT_SUCCESS;
}

View File

@ -1,108 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "parameter.h"
#include "util.h"
#define MAXLINE 4096
void initParameter(Parameter* param)
{
param->xlength = 1.0;
param->ylength = 1.0;
param->imax = 100;
param->jmax = 100;
param->itermax = 1000;
param->eps = 0.0001;
param->omg = 1.8;
}
void readParameter(Parameter* param, const char* filename)
{
FILE* fp = fopen(filename, "r");
char line[MAXLINE];
int i;
if (!fp) {
fprintf(stderr, "Could not open parameter file: %s\n", filename);
exit(EXIT_FAILURE);
}
while (!feof(fp)) {
line[0] = '\0';
fgets(line, MAXLINE, fp);
for (i = 0; line[i] != '\0' && line[i] != '#'; i++)
;
line[i] = '\0';
char* tok = strtok(line, " ");
char* val = strtok(NULL, " ");
#define PARSE_PARAM(p, f) \
if (strncmp(tok, #p, sizeof(#p) / sizeof(#p[0]) - 1) == 0) { \
param->p = f(val); \
}
#define PARSE_STRING(p) PARSE_PARAM(p, strdup)
#define PARSE_INT(p) PARSE_PARAM(p, atoi)
#define PARSE_REAL(p) PARSE_PARAM(p, atof)
if (tok != NULL && val != NULL) {
PARSE_REAL(xlength);
PARSE_REAL(ylength);
PARSE_INT(imax);
PARSE_INT(jmax);
PARSE_INT(itermax);
PARSE_REAL(eps);
PARSE_REAL(omg);
PARSE_REAL(re);
PARSE_REAL(tau);
PARSE_REAL(gamma);
PARSE_REAL(dt);
PARSE_REAL(te);
PARSE_REAL(gx);
PARSE_REAL(gy);
PARSE_STRING(name);
PARSE_INT(bcN);
PARSE_INT(bcS);
PARSE_INT(bcE);
PARSE_INT(bcW);
PARSE_REAL(u_init);
PARSE_REAL(v_init);
PARSE_REAL(p_init);
}
}
fclose(fp);
}
void printParameter(Parameter* param)
{
printf("Parameters for %s\n", param->name);
printf("Boundary conditions N:%d E:%d S:%d W:%d\n",
param->bcN,
param->bcE,
param->bcS,
param->bcW);
printf("\tReynolds number: %.2f\n", param->re);
printf("\tInit arrays: U:%.2f V:%.2f P:%.2f\n",
param->u_init,
param->v_init,
param->p_init);
printf("Geometry data:\n");
printf("\tDomain box size (x, y): %.2f, %.2f\n", param->xlength, param->ylength);
printf("\tCells (x, y): %d, %d\n", param->imax, param->jmax);
printf("Timestep parameters:\n");
printf("\tDefault stepsize: %.2f, Final time %.2f\n", param->dt, param->te);
printf("\tTau factor: %.2f\n", param->tau);
printf("Iterative solver parameters:\n");
printf("\tMax iterations: %d\n", param->itermax);
printf("\tepsilon (stopping tolerance) : %f\n", param->eps);
printf("\tgamma (stopping tolerance) : %f\n", param->gamma);
printf("\tomega (SOR relaxation): %f\n", param->omg);
}

View File

@ -1,26 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#ifndef __PARAMETER_H_
#define __PARAMETER_H_
typedef struct {
double xlength, ylength;
int imax, jmax;
int itermax;
double eps, omg;
double re, tau, gamma;
double te, dt;
double gx, gy;
char* name;
int bcN, bcS, bcE, bcW;
double u_init, v_init, p_init;
} Parameter;
void initParameter(Parameter*);
void readParameter(Parameter*, const char*);
void printParameter(Parameter*);
#endif

View File

@ -1,60 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <math.h>
#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "progress.h"
static double _end;
static int _current;
static int _rank = -1;
void initProgress(double end)
{
MPI_Comm_rank(MPI_COMM_WORLD, &_rank);
_end = end;
_current = 0;
if (_rank == 0) {
printf("[ ]");
fflush(stdout);
}
}
void printProgress(double current)
{
if (_rank == 0) {
int new = (int)rint((current / _end) * 10.0);
if (new > _current) {
char progress[11];
_current = new;
progress[0] = 0;
for (int i = 0; i < 10; i++) {
if (i < _current) {
sprintf(progress + strlen(progress), "#");
} else {
sprintf(progress + strlen(progress), " ");
}
}
printf("\r[%s]", progress);
}
fflush(stdout);
}
}
void stopProgress()
{
if (_rank == 0) {
printf("\n");
fflush(stdout);
}
}

View File

@ -1,900 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <float.h>
#include <math.h>
#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "allocate.h"
#include "parameter.h"
#include "solver.h"
#include "util.h"
#define P(i, j) p[(j) * (imaxLocal + 2) + (i)]
#define F(i, j) f[(j) * (imaxLocal + 2) + (i)]
#define G(i, j) g[(j) * (imaxLocal + 2) + (i)]
#define U(i, j) u[(j) * (imaxLocal + 2) + (i)]
#define V(i, j) v[(j) * (imaxLocal + 2) + (i)]
#define RHS(i, j) rhs[(j) * (imaxLocal + 2) + (i)]
#define NDIMS 2
#define IDIM 0
#define JDIM 1
static int sizeOfRank(int rank, int size, int N)
{
return N / size + ((N % size > rank) ? 1 : 0);
}
void print(Solver* solver, double* grid)
{
int imaxLocal = solver->imaxLocal;
for (int i = 0; i < solver->size; i++) {
if (i == solver->rank) {
printf(
"### RANK %d #######################################################\n",
solver->rank);
for (int j = 0; j < solver->jmaxLocal + 2; j++) {
printf("%02d: ", j);
for (int i = 0; i < solver->imaxLocal + 2; i++) {
printf("%12.8f ", grid[j * (imaxLocal + 2) + i]);
}
printf("\n");
}
fflush(stdout);
}
MPI_Barrier(MPI_COMM_WORLD);
}
}
static void exchange(Solver* solver, double* grid)
{
double* buf[8];
MPI_Request requests[8];
for (int i = 0; i < 8; i++)
requests[i] = MPI_REQUEST_NULL;
buf[0] = grid + 1; // recv bottom
buf[1] = grid + (solver->imaxLocal + 2) + 1; // send bottom
buf[2] = grid + (solver->jmaxLocal + 1) * (solver->imaxLocal + 2) + 1; // recv top
buf[3] = grid + (solver->jmaxLocal) * (solver->imaxLocal + 2) + 1; // send top
buf[4] = grid + (solver->imaxLocal + 2); // recv left
buf[5] = grid + (solver->imaxLocal + 2) + 1; // send left
buf[6] = grid + (solver->imaxLocal + 2) + (solver->imaxLocal + 1); // recv right
buf[7] = grid + (solver->imaxLocal + 2) + (solver->imaxLocal); // send right
for (int i = 0; i < 2; i++) {
int tag = 0;
if (solver->jNeighbours[i] != MPI_PROC_NULL) {
tag = solver->jNeighbours[i];
}
/* exchange ghost cells with bottom/top neighbor */
MPI_Irecv(buf[i * 2],
1,
solver->jBufferType,
solver->jNeighbours[i],
tag,
solver->comm,
&requests[i * 2]);
MPI_Isend(buf[(i * 2) + 1],
1,
solver->jBufferType,
solver->jNeighbours[i],
solver->rank,
solver->comm,
&requests[i * 2 + 1]);
tag = 0;
if (solver->iNeighbours[i] != MPI_PROC_NULL) {
tag = solver->iNeighbours[i];
}
/* exchange ghost cells with left/right neighbor */
MPI_Irecv(buf[i * 2 + 4],
1,
solver->iBufferType,
solver->iNeighbours[i],
tag,
solver->comm,
&requests[i * 2 + 4]);
MPI_Isend(buf[i * 2 + 5],
1,
solver->iBufferType,
solver->iNeighbours[i],
solver->rank,
solver->comm,
&requests[(i * 2) + 5]);
}
MPI_Waitall(8, requests, MPI_STATUSES_IGNORE);
}
static void shift(Solver* solver)
{
MPI_Request requests[4] = { MPI_REQUEST_NULL,
MPI_REQUEST_NULL,
MPI_REQUEST_NULL,
MPI_REQUEST_NULL };
double* f = solver->f;
double* g = solver->g;
/* shift G */
double* buf = g + 1;
/* receive ghost cells from bottom neighbor */
MPI_Irecv(buf,
1,
solver->jBufferType,
solver->jNeighbours[0],
0,
solver->comm,
&requests[0]);
buf = g + (solver->jmaxLocal) * (solver->imaxLocal + 2) + 1;
/* send ghost cells to top neighbor */
MPI_Isend(buf,
1,
solver->jBufferType,
solver->jNeighbours[1],
0,
solver->comm,
&requests[1]);
/* shift F */
buf = f + (solver->imaxLocal + 2);
/* receive ghost cells from left neighbor */
MPI_Irecv(buf,
1,
solver->iBufferType,
solver->iNeighbours[0],
1,
solver->comm,
&requests[2]);
buf = f + (solver->imaxLocal + 2) + (solver->imaxLocal);
/* send ghost cells to right neighbor */
MPI_Isend(buf,
1,
solver->iBufferType,
solver->iNeighbours[1],
1,
solver->comm,
&requests[3]);
MPI_Waitall(4, requests, MPI_STATUSES_IGNORE);
}
void debugExchange(Solver* solver)
{
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
for (int j = 0; j < jmaxLocal + 2; j++) {
for (int i = 0; i < solver->imaxLocal + 2; i++) {
solver->p[j * (imaxLocal + 2) + i] = solver->rank + 0.01 * i + 0.0001 * j;
}
}
collectResult(solver);
/* print(solver, solver->p); */
}
void debugBC(Solver* solver)
{
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
double* v = solver->v;
// Northern boundary
if (solver->coords[JDIM] == (solver->dims[JDIM] - 1)) { // set top bc
for (int i = 1; i < imaxLocal + 1; i++) {
V(i, jmaxLocal + 1) = 10.0 + solver->rank;
}
}
// Eastern boundary
if (solver->coords[IDIM] == (solver->dims[IDIM] - 1)) { // set right bc
for (int j = 1; j < jmaxLocal + 1; j++) {
V(imaxLocal + 1, j) = 20.0 + solver->rank;
}
}
// Southern boundary
if (solver->coords[JDIM] == 0) { // set bottom bc
for (int i = 1; i < imaxLocal + 1; i++) {
V(i, 0) = 30.0 + solver->rank;
}
}
// Western boundary
if (solver->coords[IDIM] == 0) { // set left bc
for (int j = 1; j < jmaxLocal + 1; j++) {
V(0, j) = 40.0 + solver->rank;
}
}
print(solver, solver->v);
}
static void assembleResult(Solver* solver,
double* src,
double* dst,
int imaxLocal[],
int jmaxLocal[],
int offset[])
{
MPI_Request* requests;
int numRequests = 1;
if (solver->rank == 0) {
numRequests = solver->size + 1;
} else {
numRequests = 1;
}
requests = (MPI_Request*)malloc(numRequests * sizeof(MPI_Request));
/* all ranks send their bulk array */
MPI_Datatype bulkType;
const int ndims = 2;
int oldSizes[ndims] = { solver->jmaxLocal + 2, solver->imaxLocal + 2 };
int newSizes[ndims] = { solver->jmaxLocal, solver->imaxLocal };
int starts[ndims] = { 1, 1 };
MPI_Type_create_subarray(2,
oldSizes,
newSizes,
starts,
MPI_ORDER_C,
MPI_DOUBLE,
&bulkType);
MPI_Type_commit(&bulkType);
MPI_Isend(src, 1, bulkType, 0, 0, solver->comm, &requests[0]);
/* rank 0 assembles the subdomains */
if (solver->rank == 0) {
for (int i = 0; i < solver->size; i++) {
MPI_Datatype domainType;
MPI_Type_vector(jmaxLocal[i],
imaxLocal[i],
solver->imax,
MPI_DOUBLE,
&domainType);
MPI_Type_commit(&domainType);
MPI_Irecv(dst + offset[i],
1,
domainType,
i,
0,
solver->comm,
&requests[i + 1]);
}
}
MPI_Waitall(numRequests, requests, MPI_STATUSES_IGNORE);
}
static int sum(int* sizes, int position)
{
int sum = 0;
for (int i = 0; i < position; i++) {
sum += sizes[i];
}
return sum;
}
void collectResult(Solver* solver)
{
double* Pall = NULL;
double* Uall = NULL;
double* Vall = NULL;
int offset[solver->size];
int imaxLocal[solver->size];
int jmaxLocal[solver->size];
MPI_Gather(&solver->imaxLocal, 1, MPI_INT, imaxLocal, 1, MPI_INT, 0, MPI_COMM_WORLD);
MPI_Gather(&solver->jmaxLocal, 1, MPI_INT, jmaxLocal, 1, MPI_INT, 0, MPI_COMM_WORLD);
if (solver->rank == 0) {
Pall = allocate(64, (solver->imax) * (solver->jmax) * sizeof(double));
Uall = allocate(64, (solver->imax) * (solver->jmax) * sizeof(double));
Vall = allocate(64, (solver->imax) * (solver->jmax) * sizeof(double));
for (int i = 0; i < solver->size; i++) {
int coords[2];
MPI_Cart_coords(solver->comm, i, 2, coords);
int ioffset = sum(imaxLocal, coords[0]);
int joffset = sum(jmaxLocal, coords[1]);
offset[i] = (joffset * solver->imax) + ioffset;
printf("Rank: %d, Coords(i,j): %d %d, Size(i,j): %d %d, Offset(i,j): %d %d\n",
i,
coords[0],
coords[1],
imaxLocal[i],
jmaxLocal[i],
ioffset,
joffset);
}
}
/* collect P */
assembleResult(solver, solver->p, Pall, imaxLocal, jmaxLocal, offset);
/* collect U */
assembleResult(solver, solver->u, Uall, imaxLocal, jmaxLocal, offset);
/* collect V */
assembleResult(solver, solver->v, Vall, imaxLocal, jmaxLocal, offset);
/* write to disk */
if (solver->rank == 0) writeResult(solver, Pall, Uall, Vall);
}
static void printConfig(Solver* solver)
{
if (solver->rank == 0) {
printf("Parameters for #%s#\n", solver->problem);
printf("Boundary conditions N:%d E:%d S:%d W:%d\n",
solver->bcN,
solver->bcE,
solver->bcS,
solver->bcW);
printf("\tReynolds number: %.2f\n", solver->re);
printf("\tGx Gy: %.2f %.2f\n", solver->gx, solver->gy);
printf("Geometry data:\n");
printf("\tDomain box size (x, y): %.2f, %.2f\n",
solver->xlength,
solver->ylength);
printf("\tCells (x, y): %d, %d\n", solver->imax, solver->jmax);
printf("Timestep parameters:\n");
printf("\tDefault stepsize: %.2f, Final time %.2f\n", solver->dt, solver->te);
printf("\tdt bound: %.6f\n", solver->dtBound);
printf("\tTau factor: %.2f\n", solver->tau);
printf("Iterative solver parameters:\n");
printf("\tMax iterations: %d\n", solver->itermax);
printf("\tepsilon (stopping tolerance) : %f\n", solver->eps);
printf("\tgamma factor: %f\n", solver->gamma);
printf("\tomega (SOR relaxation): %f\n", solver->omega);
printf("Communication parameters:\n");
}
for (int i = 0; i < solver->size; i++) {
if (i == solver->rank) {
printf("\tRank %d of %d\n", solver->rank, solver->size);
printf("\tNeighbours (b, t, l, r): %d, %d, %d, %d\n",
solver->jNeighbours[0],
solver->jNeighbours[1],
solver->iNeighbours[0],
solver->iNeighbours[1]);
printf("\tCoordinates %d,%d\n", solver->coords[0], solver->coords[1]);
printf("\tLocal domain size: %dx%d\n", solver->imaxLocal, solver->jmaxLocal);
fflush(stdout);
}
}
}
void initSolver(Solver* solver, Parameter* params)
{
solver->problem = params->name;
solver->bcN = params->bcN;
solver->bcS = params->bcS;
solver->bcW = params->bcW;
solver->bcE = params->bcE;
solver->imax = params->imax;
solver->jmax = params->jmax;
solver->xlength = params->xlength;
solver->ylength = params->ylength;
solver->dx = params->xlength / params->imax;
solver->dy = params->ylength / params->jmax;
solver->eps = params->eps;
solver->omega = params->omg;
solver->itermax = params->itermax;
solver->re = params->re;
solver->gx = params->gx;
solver->gy = params->gy;
solver->dt = params->dt;
solver->te = params->te;
solver->tau = params->tau;
solver->gamma = params->gamma;
/* setup communication */
MPI_Comm_rank(MPI_COMM_WORLD, &(solver->rank));
MPI_Comm_size(MPI_COMM_WORLD, &(solver->size));
int dims[NDIMS] = { 0, 0 };
int periods[NDIMS] = { 0, 0 };
MPI_Dims_create(solver->size, NDIMS, dims);
MPI_Cart_create(MPI_COMM_WORLD, NDIMS, dims, periods, 0, &solver->comm);
MPI_Cart_shift(solver->comm,
IDIM,
1,
&solver->iNeighbours[0],
&solver->iNeighbours[1]);
MPI_Cart_shift(solver->comm,
JDIM,
1,
&solver->jNeighbours[0],
&solver->jNeighbours[1]);
MPI_Cart_get(solver->comm, NDIMS, solver->dims, periods, solver->coords);
solver->imaxLocal = sizeOfRank(solver->rank, dims[IDIM], solver->imax);
solver->jmaxLocal = sizeOfRank(solver->rank, dims[JDIM], solver->jmax);
MPI_Type_contiguous(solver->imaxLocal, MPI_DOUBLE, &solver->jBufferType);
MPI_Type_commit(&solver->jBufferType);
MPI_Type_vector(solver->jmaxLocal,
1,
solver->imaxLocal + 2,
MPI_DOUBLE,
&solver->iBufferType);
MPI_Type_commit(&solver->iBufferType);
/* allocate arrays */
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
size_t bytesize = (imaxLocal + 2) * (jmaxLocal + 2) * sizeof(double);
solver->u = allocate(64, bytesize);
solver->v = allocate(64, bytesize);
solver->p = allocate(64, bytesize);
solver->rhs = allocate(64, bytesize);
solver->f = allocate(64, bytesize);
solver->g = allocate(64, bytesize);
for (int i = 0; i < (imaxLocal + 2) * (jmaxLocal + 2); i++) {
solver->u[i] = params->u_init;
solver->v[i] = params->v_init;
solver->p[i] = params->p_init;
solver->rhs[i] = 0.0;
solver->f[i] = 0.0;
solver->g[i] = 0.0;
}
double dx = solver->dx;
double dy = solver->dy;
double inv_sqr_sum = 1.0 / (dx * dx) + 1.0 / (dy * dy);
solver->dtBound = 0.5 * solver->re * 1.0 / inv_sqr_sum;
#ifdef VERBOSE
printConfig(solver);
#endif
}
void computeRHS(Solver* solver)
{
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
double idx = 1.0 / solver->dx;
double idy = 1.0 / solver->dy;
double idt = 1.0 / solver->dt;
double* rhs = solver->rhs;
double* f = solver->f;
double* g = solver->g;
shift(solver);
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = 1; i < imaxLocal + 1; i++) {
RHS(i, j) = ((F(i, j) - F(i - 1, j)) * idx + (G(i, j) - G(i, j - 1)) * idy) *
idt;
}
}
}
int solve(Solver* solver)
{
int imax = solver->imax;
int jmax = solver->jmax;
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
double eps = solver->eps;
int itermax = solver->itermax;
double dx2 = solver->dx * solver->dx;
double dy2 = solver->dy * solver->dy;
double idx2 = 1.0 / dx2;
double idy2 = 1.0 / dy2;
double factor = solver->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
double* p = solver->p;
double* rhs = solver->rhs;
double epssq = eps * eps;
int it = 0;
double res = 1.0;
while ((res >= epssq) && (it < itermax)) {
res = 0.0;
exchange(solver, p);
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = 1; i < imaxLocal + 1; i++) {
double r = RHS(i, j) -
((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
(P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2);
P(i, j) -= (factor * r);
res += (r * r);
}
}
if (solver->coords[JDIM] == 0) { // set bottom bc
for (int i = 1; i < imaxLocal + 1; i++) {
P(i, 0) = P(i, 1);
}
}
if (solver->coords[JDIM] == (solver->dims[JDIM] - 1)) { // set top bc
for (int i = 1; i < imaxLocal + 1; i++) {
P(i, jmaxLocal + 1) = P(i, jmaxLocal);
}
}
if (solver->coords[IDIM] == 0) { // set left bc
for (int j = 1; j < jmaxLocal + 1; j++) {
P(0, j) = P(1, j);
}
}
if (solver->coords[IDIM] == (solver->dims[IDIM] - 1)) { // set right bc
for (int j = 1; j < jmaxLocal + 1; j++) {
P(imaxLocal + 1, j) = P(imaxLocal, j);
}
}
MPI_Allreduce(MPI_IN_PLACE, &res, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
res = res / (double)(imax * jmax);
#ifdef DEBUG
if (solver->rank == 0) {
printf("%d Residuum: %e\n", it, res);
}
#endif
it++;
}
#ifdef VERBOSE
if (solver->rank == 0) {
printf("Solver took %d iterations to reach %f\n", it, sqrt(res));
}
#endif
if (res < eps) {
return 0;
} else {
return 1;
}
}
static double maxElement(Solver* solver, double* m)
{
int size = (solver->imaxLocal + 2) * (solver->jmaxLocal + 2);
double maxval = DBL_MIN;
for (int i = 0; i < size; i++) {
maxval = MAX(maxval, fabs(m[i]));
}
MPI_Allreduce(MPI_IN_PLACE, &maxval, 1, MPI_DOUBLE, MPI_MAX, MPI_COMM_WORLD);
return maxval;
}
void computeTimestep(Solver* solver)
{
double dt = solver->dtBound;
double dx = solver->dx;
double dy = solver->dy;
double umax = maxElement(solver, solver->u);
double vmax = maxElement(solver, solver->v);
if (umax > 0) {
dt = (dt > dx / umax) ? dx / umax : dt;
}
if (vmax > 0) {
dt = (dt > dy / vmax) ? dy / vmax : dt;
}
solver->dt = dt * solver->tau;
}
void setBoundaryConditions(Solver* solver)
{
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
double* u = solver->u;
double* v = solver->v;
// Northern boundary
if (solver->coords[JDIM] == (solver->dims[JDIM] - 1)) { // set top bc
switch (solver->bcN) {
case NOSLIP:
for (int i = 1; i < imaxLocal + 1; i++) {
V(i, jmaxLocal) = 0.0;
U(i, jmaxLocal + 1) = -U(i, jmaxLocal);
}
break;
case SLIP:
for (int i = 1; i < imaxLocal + 1; i++) {
V(i, jmaxLocal) = 0.0;
U(i, jmaxLocal + 1) = U(i, jmaxLocal);
}
break;
case OUTFLOW:
for (int i = 1; i < imaxLocal + 1; i++) {
U(i, jmaxLocal + 1) = U(i, jmaxLocal);
V(i, jmaxLocal) = V(i, jmaxLocal - 1);
}
break;
case PERIODIC:
break;
}
}
// Southern boundary
if (solver->coords[JDIM] == 0) { // set bottom bc
switch (solver->bcS) {
case NOSLIP:
for (int i = 1; i < imaxLocal + 1; i++) {
V(i, 0) = 0.0;
U(i, 0) = -U(i, 1);
}
break;
case SLIP:
for (int i = 1; i < imaxLocal + 1; i++) {
V(i, 0) = 0.0;
U(i, 0) = U(i, 1);
}
break;
case OUTFLOW:
for (int i = 1; i < imaxLocal + 1; i++) {
U(i, 0) = U(i, 1);
V(i, 0) = V(i, 1);
}
break;
case PERIODIC:
break;
}
}
// Eastern boundary
if (solver->coords[IDIM] == (solver->dims[IDIM] - 1)) { // set right bc
switch (solver->bcE) {
case NOSLIP:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(imaxLocal, j) = 0.0;
V(imaxLocal + 1, j) = -V(imaxLocal, j);
}
break;
case SLIP:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(imaxLocal, j) = 0.0;
V(imaxLocal + 1, j) = V(imaxLocal, j);
}
break;
case OUTFLOW:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(imaxLocal, j) = U(imaxLocal - 1, j);
V(imaxLocal + 1, j) = V(imaxLocal, j);
}
break;
case PERIODIC:
break;
}
}
// Western boundary
if (solver->coords[IDIM] == 0) { // set left bc
switch (solver->bcW) {
case NOSLIP:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(0, j) = 0.0;
V(0, j) = -V(1, j);
}
break;
case SLIP:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(0, j) = 0.0;
V(0, j) = V(1, j);
}
break;
case OUTFLOW:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(0, j) = U(1, j);
V(0, j) = V(1, j);
}
break;
case PERIODIC:
break;
}
}
}
void setSpecialBoundaryCondition(Solver* solver)
{
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
double* u = solver->u;
if (strcmp(solver->problem, "dcavity") == 0) {
if (solver->coords[JDIM] == (solver->dims[JDIM] - 1)) { // set top bc
for (int i = 1; i < imaxLocal + 1; i++) {
U(i, jmaxLocal + 1) = 2.0 - U(i, jmaxLocal);
}
}
} else if (strcmp(solver->problem, "canal") == 0) {
if (solver->coords[IDIM] == 0) { // set left bc
double ylength = solver->ylength;
double dy = solver->dy;
int rest = solver->jmax % solver->size;
int yc = solver->rank * (solver->jmax / solver->size) +
MIN(rest, solver->rank);
double ys = dy * (yc + 0.5);
double y;
/* printf("RANK %d yc: %d ys: %f\n", solver->rank, yc, ys); */
for (int j = 1; j < jmaxLocal + 1; j++) {
y = ys + dy * (j - 0.5);
U(0, j) = y * (ylength - y) * 4.0 / (ylength * ylength);
}
}
}
/* print(solver, solver->u); */
}
void computeFG(Solver* solver)
{
double* u = solver->u;
double* v = solver->v;
double* f = solver->f;
double* g = solver->g;
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
double gx = solver->gx;
double gy = solver->gy;
double gamma = solver->gamma;
double dt = solver->dt;
double inverseRe = 1.0 / solver->re;
double inverseDx = 1.0 / solver->dx;
double inverseDy = 1.0 / solver->dy;
double du2dx, dv2dy, duvdx, duvdy;
double du2dx2, du2dy2, dv2dx2, dv2dy2;
exchange(solver, u);
exchange(solver, v);
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = 1; i < imaxLocal + 1; i++) {
du2dx = inverseDx * 0.25 *
((U(i, j) + U(i + 1, j)) * (U(i, j) + U(i + 1, j)) -
(U(i, j) + U(i - 1, j)) * (U(i, j) + U(i - 1, j))) +
gamma * inverseDx * 0.25 *
(fabs(U(i, j) + U(i + 1, j)) * (U(i, j) - U(i + 1, j)) +
fabs(U(i, j) + U(i - 1, j)) * (U(i, j) - U(i - 1, j)));
duvdy = inverseDy * 0.25 *
((V(i, j) + V(i + 1, j)) * (U(i, j) + U(i, j + 1)) -
(V(i, j - 1) + V(i + 1, j - 1)) * (U(i, j) + U(i, j - 1))) +
gamma * inverseDy * 0.25 *
(fabs(V(i, j) + V(i + 1, j)) * (U(i, j) - U(i, j + 1)) +
fabs(V(i, j - 1) + V(i + 1, j - 1)) *
(U(i, j) - U(i, j - 1)));
du2dx2 = inverseDx * inverseDx * (U(i + 1, j) - 2.0 * U(i, j) + U(i - 1, j));
du2dy2 = inverseDy * inverseDy * (U(i, j + 1) - 2.0 * U(i, j) + U(i, j - 1));
F(i, j) = U(i, j) + dt * (inverseRe * (du2dx2 + du2dy2) - du2dx - duvdy + gx);
duvdx = inverseDx * 0.25 *
((U(i, j) + U(i, j + 1)) * (V(i, j) + V(i + 1, j)) -
(U(i - 1, j) + U(i - 1, j + 1)) * (V(i, j) + V(i - 1, j))) +
gamma * inverseDx * 0.25 *
(fabs(U(i, j) + U(i, j + 1)) * (V(i, j) - V(i + 1, j)) +
fabs(U(i - 1, j) + U(i - 1, j + 1)) *
(V(i, j) - V(i - 1, j)));
dv2dy = inverseDy * 0.25 *
((V(i, j) + V(i, j + 1)) * (V(i, j) + V(i, j + 1)) -
(V(i, j) + V(i, j - 1)) * (V(i, j) + V(i, j - 1))) +
gamma * inverseDy * 0.25 *
(fabs(V(i, j) + V(i, j + 1)) * (V(i, j) - V(i, j + 1)) +
fabs(V(i, j) + V(i, j - 1)) * (V(i, j) - V(i, j - 1)));
dv2dx2 = inverseDx * inverseDx * (V(i + 1, j) - 2.0 * V(i, j) + V(i - 1, j));
dv2dy2 = inverseDy * inverseDy * (V(i, j + 1) - 2.0 * V(i, j) + V(i, j - 1));
G(i, j) = V(i, j) + dt * (inverseRe * (dv2dx2 + dv2dy2) - duvdx - dv2dy + gy);
}
}
/* ----------------------------- boundary of F --------------------------- */
if (solver->coords[IDIM] == 0) { // set left bc
for (int j = 1; j < jmaxLocal + 1; j++) {
F(0, j) = U(0, j);
}
}
if (solver->coords[IDIM] == (solver->dims[IDIM] - 1)) { // set right bc
for (int j = 1; j < jmaxLocal + 1; j++) {
F(imaxLocal, j) = U(imaxLocal, j);
}
}
/* ----------------------------- boundary of G --------------------------- */
if (solver->coords[JDIM] == 0) { // set bottom bc
for (int i = 1; i < imaxLocal + 1; i++) {
G(i, 0) = V(i, 0);
}
}
if (solver->coords[JDIM] == (solver->dims[JDIM] - 1)) { // set top bc
for (int i = 1; i < imaxLocal + 1; i++) {
G(i, jmaxLocal) = V(i, jmaxLocal);
}
}
}
void adaptUV(Solver* solver)
{
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
double* p = solver->p;
double* u = solver->u;
double* v = solver->v;
double* f = solver->f;
double* g = solver->g;
double factorX = solver->dt / solver->dx;
double factorY = solver->dt / solver->dy;
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = 1; i < imaxLocal + 1; i++) {
U(i, j) = F(i, j) - (P(i + 1, j) - P(i, j)) * factorX;
V(i, j) = G(i, j) - (P(i, j + 1) - P(i, j)) * factorY;
}
}
}
void writeResult(Solver* solver, double* p, double* u, double* v)
{
int imax = solver->imax;
int jmax = solver->jmax;
double dx = solver->dx;
double dy = solver->dy;
double x = 0.0, y = 0.0;
FILE* fp;
fp = fopen("pressure.dat", "w");
if (fp == NULL) {
printf("Error!\n");
exit(EXIT_FAILURE);
}
for (int j = 1; j < jmax; j++) {
y = (double)(j - 0.5) * dy;
for (int i = 1; i < imax; i++) {
x = (double)(i - 0.5) * dx;
fprintf(fp, "%.2f %.2f %f\n", x, y, p[j * (imax) + i]);
}
fprintf(fp, "\n");
}
fclose(fp);
fp = fopen("velocity.dat", "w");
if (fp == NULL) {
printf("Error!\n");
exit(EXIT_FAILURE);
}
for (int j = 1; j < jmax; j++) {
y = dy * (j - 0.5);
for (int i = 1; i < imax; i++) {
x = dx * (i - 0.5);
double vel_u = (u[j * (imax) + i] + u[j * (imax) + (i - 1)]) / 2.0;
double vel_v = (v[j * (imax) + i] + v[(j - 1) * (imax) + i]) / 2.0;
double len = sqrt((vel_u * vel_u) + (vel_v * vel_v));
fprintf(fp, "%.2f %.2f %f %f %f\n", x, y, vel_u, vel_v, len);
}
}
fclose(fp);
}

View File

@ -1,56 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#ifndef __SOLVER_H_
#define __SOLVER_H_
#include "parameter.h"
#include <mpi.h>
enum BC { NOSLIP = 1, SLIP, OUTFLOW, PERIODIC };
typedef struct {
/* geometry and grid information */
double dx, dy;
int imax, jmax;
double xlength, ylength;
/* arrays */
double *p, *rhs;
double *f, *g;
double *u, *v;
/* parameters */
double eps, omega;
double re, tau, gamma;
double gx, gy;
/* time stepping */
int itermax;
double dt, te;
double dtBound;
char* problem;
int bcN, bcS, bcW, bcE;
/* mpi */
int rank;
int size;
MPI_Comm comm;
MPI_Datatype iBufferType, jBufferType;
int iNeighbours[2], jNeighbours[2];
int coords[2], dims[2];
int imaxLocal, jmaxLocal;
} Solver;
void initSolver(Solver*, Parameter*);
void computeRHS(Solver*);
int solve(Solver*);
void computeTimestep(Solver*);
void setBoundaryConditions(Solver*);
void setSpecialBoundaryCondition(Solver*);
void computeFG(Solver*);
void adaptUV(Solver*);
void collectResult(Solver*);
void writeResult(Solver*, double*, double*, double*);
void debugExchange(Solver*);
void debugBC(Solver*);
void print(Solver*, double*);
#endif

View File

@ -1,5 +0,0 @@
set terminal png size 1800,768 enhanced font ,12
set output 'velocity.png'
set datafile separator whitespace
plot 'velocity.dat' using 1:2:3:4:5 with vectors filled head size 0.01,20,60 lc palette

View File

@ -1,46 +0,0 @@
#==============================================================================
# Laminar Canal Flow
#==============================================================================
# Problem specific Data:
# ---------------------
name canal # name of flow setup
bcN 1 # flags for boundary conditions
bcE 3 # 1 = no-slip 3 = outflow
bcS 1 # 2 = free-slip 4 = periodic
bcW 3 #
gx 0.0 # Body forces (e.g. gravity)
gy 0.0 #
re 100.0 # Reynolds number
u_init 1.0 # initial value for velocity in x-direction
v_init 0.0 # initial value for velocity in y-direction
p_init 0.0 # initial value for pressure
# Geometry Data:
# -------------
xlength 30.0 # domain size in x-direction
ylength 4.0 # domain size in y-direction
imax 200 # number of interior cells in x-direction
jmax 50 # number of interior cells in y-direction
# Time Data:
# ---------
te 100.0 # final time
dt 0.02 # time stepsize
tau 0.5 # safety factor for time stepsize control (<0 constant delt)
# Pressure Iteration Data:
# -----------------------
itermax 500 # maximal number of pressure iteration in one time step
eps 0.00001 # stopping tolerance for pressure iteration
omg 1.8 # relaxation parameter for SOR iteration
gamma 0.9 # upwind differencing factor gamma
#===============================================================================

View File

@ -1,16 +0,0 @@
CC = mpicc
GCC = cc
LINKER = $(CC)
ifeq ($(ENABLE_OPENMP),true)
OPENMP = -fopenmp
#OPENMP = -Xpreprocessor -fopenmp #required on Macos with homebrew libomp
LIBS = # -lomp
endif
VERSION = --version
CFLAGS = -Ofast -std=c99 $(OPENMP)
#CFLAGS = -Ofast -fnt-store=aggressive -std=c99 $(OPENMP) #AMD CLANG
LFLAGS = $(OPENMP)
DEFINES = -D_GNU_SOURCE# -DDEBUG
INCLUDES = -I/usr/local/include

View File

@ -1,14 +0,0 @@
CC = mpiicc
GCC = gcc
LINKER = $(CC)
ifeq ($(ENABLE_OPENMP),true)
OPENMP = -qopenmp
endif
VERSION = --version
CFLAGS = -O3 -xHost -qopt-zmm-usage=high -std=c99 $(OPENMP)
LFLAGS = $(OPENMP)
DEFINES = -D_GNU_SOURCE
INCLUDES =
LIBS =

View File

@ -1,14 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#ifndef AFFINITY_H
#define AFFINITY_H
extern int affinity_getProcessorId();
extern void affinity_pinProcess(int);
extern void affinity_pinThread(int);
#endif /*AFFINITY_H*/

View File

@ -1,77 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#include <float.h>
#include <limits.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include "parameter.h"
#include "progress.h"
#include "solver.h"
#include "timing.h"
#include <mpi.h>
int main(int argc, char** argv)
{
int rank;
double S, E;
Parameter params;
Solver solver;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
initParameter(&params);
if (argc != 2) {
printf("Usage: %s <configFile>\n", argv[0]);
exit(EXIT_SUCCESS);
}
readParameter(&params, argv[1]);
if (rank == 0) {
printParameter(&params);
}
initSolver(&solver, &params);
initProgress(solver.te);
double tau = solver.tau;
double te = solver.te;
double t = 0.0;
S = getTimeStamp();
while (t <= te) {
if (tau > 0.0) {
computeTimestep(&solver);
}
setBoundaryConditions(&solver);
setSpecialBoundaryCondition(&solver);
computeFG(&solver);
computeRHS(&solver);
solve(&solver);
adaptUV(&solver);
t += solver.dt;
#ifdef VERBOSE
if (rank == 0) {
printf("TIME %f , TIMESTEP %f\n", t, solver.dt);
}
#else
printProgress(t);
#endif
}
E = getTimeStamp();
stopProgress();
if (rank == 0) {
printf("Solution took %.2fs\n", E - S);
}
collectResult(&solver);
MPI_Finalize();
return EXIT_SUCCESS;
}

View File

@ -1,60 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <math.h>
#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "progress.h"
static double _end;
static int _current;
static int _rank = -1;
void initProgress(double end)
{
MPI_Comm_rank(MPI_COMM_WORLD, &_rank);
_end = end;
_current = 0;
if (_rank == 0) {
printf("[ ]");
fflush(stdout);
}
}
void printProgress(double current)
{
if (_rank == 0) {
int new = (int)rint((current / _end) * 10.0);
if (new > _current) {
char progress[11];
_current = new;
progress[0] = 0;
for (int i = 0; i < 10; i++) {
if (i < _current) {
sprintf(progress + strlen(progress), "#");
} else {
sprintf(progress + strlen(progress), " ");
}
}
printf("\r[%s]", progress);
}
fflush(stdout);
}
}
void stopProgress()
{
if (_rank == 0) {
printf("\n");
fflush(stdout);
}
}

View File

@ -1,833 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <float.h>
#include <math.h>
#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "allocate.h"
#include "parameter.h"
#include "solver.h"
#include "util.h"
#define P(i, j) p[(j) * (imaxLocal + 2) + (i)]
#define F(i, j) f[(j) * (imaxLocal + 2) + (i)]
#define G(i, j) g[(j) * (imaxLocal + 2) + (i)]
#define U(i, j) u[(j) * (imaxLocal + 2) + (i)]
#define V(i, j) v[(j) * (imaxLocal + 2) + (i)]
#define RHS(i, j) rhs[(j) * (imaxLocal + 2) + (i)]
#define IDIM 0
#define JDIM 1
static int sizeOfRank(int rank, int size, int N)
{
return N / size + ((N % size > rank) ? 1 : 0);
}
void print(Solver* solver, double* grid)
{
int imaxLocal = solver->imaxLocal;
for (int i = 0; i < solver->size; i++) {
if (i == solver->rank) {
printf(
"### RANK %d #######################################################\n",
solver->rank);
for (int j = 0; j < solver->jmaxLocal + 2; j++) {
printf("%02d: ", j);
for (int i = 0; i < solver->imaxLocal + 2; i++) {
printf("%12.8f ", grid[j * (imaxLocal + 2) + i]);
}
printf("\n");
}
fflush(stdout);
}
MPI_Barrier(MPI_COMM_WORLD);
}
}
static void exchange(Solver* solver, double* grid)
{
int counts[4] = { 1, 1, 1, 1 };
MPI_Neighbor_alltoallw(grid,
counts,
solver->sdispls,
solver->bufferTypes,
grid,
counts,
solver->rdispls,
solver->bufferTypes,
solver->comm);
}
static void shift(Solver* solver)
{
MPI_Request requests[4] = { MPI_REQUEST_NULL,
MPI_REQUEST_NULL,
MPI_REQUEST_NULL,
MPI_REQUEST_NULL };
double* f = solver->f;
double* g = solver->g;
/* shift G */
double* buf = g + 1;
/* receive ghost cells from bottom neighbor */
MPI_Irecv(buf,
1,
solver->bufferTypes[2],
solver->jNeighbours[0],
0,
solver->comm,
&requests[0]);
buf = g + (solver->jmaxLocal) * (solver->imaxLocal + 2) + 1;
/* send ghost cells to top neighbor */
MPI_Isend(buf,
1,
solver->bufferTypes[2],
solver->jNeighbours[1],
0,
solver->comm,
&requests[1]);
/* shift F */
buf = f + (solver->imaxLocal + 2);
/* receive ghost cells from left neighbor */
MPI_Irecv(buf,
1,
solver->bufferTypes[0],
solver->iNeighbours[0],
1,
solver->comm,
&requests[2]);
buf = f + (solver->imaxLocal + 2) + (solver->imaxLocal);
/* send ghost cells to right neighbor */
MPI_Isend(buf,
1,
solver->bufferTypes[0],
solver->iNeighbours[1],
1,
solver->comm,
&requests[3]);
MPI_Waitall(4, requests, MPI_STATUSES_IGNORE);
}
void debugExchange(Solver* solver)
{
for (int i = 0; i < (solver->imaxLocal + 2) * (solver->jmaxLocal + 2); i++) {
solver->p[i] = solver->rank;
}
exchange(solver, solver->p);
print(solver, solver->p);
}
static void assembleResult(Solver* solver,
double* src,
double* dst,
int imaxLocal[],
int jmaxLocal[],
int offset[])
{
MPI_Request* requests;
int numRequests = 1;
if (solver->rank == 0) {
numRequests = solver->size + 1;
} else {
numRequests = 1;
}
requests = (MPI_Request*)malloc(numRequests * sizeof(MPI_Request));
/* all ranks send their bulk array */
MPI_Datatype bulkType;
const int ndims = 2;
int oldSizes[ndims] = { solver->jmaxLocal + 2, solver->imaxLocal + 2 };
int newSizes[ndims] = { solver->jmaxLocal, solver->imaxLocal };
int starts[ndims] = { 1, 1 };
MPI_Type_create_subarray(2,
oldSizes,
newSizes,
starts,
MPI_ORDER_C,
MPI_DOUBLE,
&bulkType);
MPI_Type_commit(&bulkType);
MPI_Isend(src, 1, bulkType, 0, 0, solver->comm, &requests[0]);
/* rank 0 assembles the subdomains */
if (solver->rank == 0) {
for (int i = 0; i < solver->size; i++) {
MPI_Datatype domainType;
MPI_Type_vector(jmaxLocal[i],
imaxLocal[i],
solver->imax,
MPI_DOUBLE,
&domainType);
MPI_Type_commit(&domainType);
MPI_Irecv(dst + offset[i],
1,
domainType,
i,
0,
solver->comm,
&requests[i + 1]);
}
}
MPI_Waitall(numRequests, requests, MPI_STATUSES_IGNORE);
}
static int sum(int* sizes, int position)
{
int sum = 0;
for (int i = 0; i < position; i++) {
sum += sizes[i];
}
return sum;
}
void collectResult(Solver* solver)
{
double* Pall = NULL;
double* Uall = NULL;
double* Vall = NULL;
int offset[solver->size];
int imaxLocal[solver->size];
int jmaxLocal[solver->size];
MPI_Gather(&solver->imaxLocal, 1, MPI_INT, imaxLocal, 1, MPI_INT, 0, MPI_COMM_WORLD);
MPI_Gather(&solver->jmaxLocal, 1, MPI_INT, jmaxLocal, 1, MPI_INT, 0, MPI_COMM_WORLD);
if (solver->rank == 0) {
Pall = allocate(64, (solver->imax) * (solver->jmax) * sizeof(double));
Uall = allocate(64, (solver->imax) * (solver->jmax) * sizeof(double));
Vall = allocate(64, (solver->imax) * (solver->jmax) * sizeof(double));
for (int i = 0; i < solver->size; i++) {
int coords[2];
MPI_Cart_coords(solver->comm, i, 2, coords);
int ioffset = sum(imaxLocal, coords[0]);
int joffset = sum(jmaxLocal, coords[1]);
offset[i] = (joffset * solver->imax) + ioffset;
printf("Rank: %d, Coords(i,j): %d %d, Size(i,j): %d %d, Offset(i,j): %d %d\n",
i,
coords[0],
coords[1],
imaxLocal[i],
jmaxLocal[i],
ioffset,
joffset);
}
}
/* collect P */
assembleResult(solver, solver->p, Pall, imaxLocal, jmaxLocal, offset);
/* collect U */
assembleResult(solver, solver->u, Uall, imaxLocal, jmaxLocal, offset);
/* collect V */
assembleResult(solver, solver->v, Vall, imaxLocal, jmaxLocal, offset);
/* write to disk */
if (solver->rank == 0) writeResult(solver, Pall, Uall, Vall);
}
static void printConfig(Solver* solver)
{
if (solver->rank == 0) {
printf("Parameters for #%s#\n", solver->problem);
printf("Boundary conditions Top:%d Bottom:%d Left:%d Right:%d\n",
solver->bcTop,
solver->bcBottom,
solver->bcLeft,
solver->bcRight);
printf("\tReynolds number: %.2f\n", solver->re);
printf("\tGx Gy: %.2f %.2f\n", solver->gx, solver->gy);
printf("Geometry data:\n");
printf("\tDomain box size (x, y): %.2f, %.2f\n",
solver->xlength,
solver->ylength);
printf("\tCells (x, y): %d, %d\n", solver->imax, solver->jmax);
printf("Timestep parameters:\n");
printf("\tDefault stepsize: %.2f, Final time %.2f\n", solver->dt, solver->te);
printf("\tdt bound: %.6f\n", solver->dtBound);
printf("\tTau factor: %.2f\n", solver->tau);
printf("Iterative solver parameters:\n");
printf("\tMax iterations: %d\n", solver->itermax);
printf("\tepsilon (stopping tolerance) : %f\n", solver->eps);
printf("\tgamma factor: %f\n", solver->gamma);
printf("\tomega (SOR relaxation): %f\n", solver->omega);
printf("Communication parameters:\n");
}
for (int i = 0; i < solver->size; i++) {
if (i == solver->rank) {
printf("\tRank %d of %d\n", solver->rank, solver->size);
printf("\tNeighbours (b, t, l, r): %d, %d, %d, %d\n",
solver->jNeighbours[0],
solver->jNeighbours[1],
solver->iNeighbours[0],
solver->iNeighbours[1]);
printf("\tCoordinates %d,%d\n", solver->coords[0], solver->coords[1]);
printf("\tLocal domain size: %dx%d\n", solver->imaxLocal, solver->jmaxLocal);
fflush(stdout);
}
}
}
void initSolver(Solver* solver, Parameter* params)
{
solver->problem = params->name;
solver->bcTop = params->bcTop;
solver->bcBottom = params->bcBottom;
solver->bcLeft = params->bcLeft;
solver->bcRight = params->bcRight;
solver->imax = params->imax;
solver->jmax = params->jmax;
solver->xlength = params->xlength;
solver->ylength = params->ylength;
solver->dx = params->xlength / params->imax;
solver->dy = params->ylength / params->jmax;
solver->eps = params->eps;
solver->omega = params->omg;
solver->itermax = params->itermax;
solver->re = params->re;
solver->gx = params->gx;
solver->gy = params->gy;
solver->dt = params->dt;
solver->te = params->te;
solver->tau = params->tau;
solver->gamma = params->gamma;
/* setup communication */
MPI_Comm_rank(MPI_COMM_WORLD, &(solver->rank));
MPI_Comm_size(MPI_COMM_WORLD, &(solver->size));
int dims[NDIMS] = { 0, 0 };
int periods[NDIMS] = { 0, 0 };
MPI_Dims_create(solver->size, NDIMS, dims);
MPI_Cart_create(MPI_COMM_WORLD, NDIMS, dims, periods, 0, &solver->comm);
MPI_Cart_shift(solver->comm,
IDIM,
1,
&solver->iNeighbours[0],
&solver->iNeighbours[1]);
MPI_Cart_shift(solver->comm,
JDIM,
1,
&solver->jNeighbours[0],
&solver->jNeighbours[1]);
MPI_Cart_get(solver->comm, NDIMS, solver->dims, periods, solver->coords);
solver->imaxLocal = sizeOfRank(solver->coords[IDIM], dims[IDIM], solver->imax);
solver->jmaxLocal = sizeOfRank(solver->coords[JDIM], dims[JDIM], solver->jmax);
MPI_Datatype jBufferType;
MPI_Type_contiguous(solver->imaxLocal, MPI_DOUBLE, &jBufferType);
MPI_Type_commit(&jBufferType);
MPI_Datatype iBufferType;
MPI_Type_vector(solver->jmaxLocal,
1,
solver->imaxLocal + 2,
MPI_DOUBLE,
&iBufferType);
MPI_Type_commit(&iBufferType);
// in the order of the dimensions i->0, j->1
// first negative direction, then positive direction
size_t dblsize = sizeof(double);
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
solver->bufferTypes[0] = iBufferType; // left
solver->bufferTypes[1] = iBufferType; // right
solver->bufferTypes[2] = jBufferType; // bottom
solver->bufferTypes[3] = jBufferType; // top
solver->sdispls[0] = ((imaxLocal + 2) + 1) * dblsize; // send left
solver->sdispls[1] = ((imaxLocal + 2) + imaxLocal) * dblsize; // send right
solver->sdispls[2] = ((imaxLocal + 2) + 1) * dblsize; // send bottom
solver->sdispls[3] = ((jmaxLocal) * (imaxLocal + 2) + 1) * dblsize; // send top
solver->rdispls[0] = (imaxLocal + 2) * dblsize; // recv left
solver->rdispls[1] = ((imaxLocal + 2) + (imaxLocal + 1)) * dblsize; // recv right
solver->rdispls[2] = 1 * dblsize; // recv bottom
solver->rdispls[3] = ((jmaxLocal + 1) * (imaxLocal + 2) + 1) * dblsize; // recv top
/* allocate arrays */
size_t bytesize = (imaxLocal + 2) * (jmaxLocal + 2) * sizeof(double);
solver->u = allocate(64, bytesize);
solver->v = allocate(64, bytesize);
solver->p = allocate(64, bytesize);
solver->rhs = allocate(64, bytesize);
solver->f = allocate(64, bytesize);
solver->g = allocate(64, bytesize);
for (int i = 0; i < (imaxLocal + 2) * (jmaxLocal + 2); i++) {
solver->u[i] = params->u_init;
solver->v[i] = params->v_init;
solver->p[i] = params->p_init;
solver->rhs[i] = 0.0;
solver->f[i] = 0.0;
solver->g[i] = 0.0;
}
double dx = solver->dx;
double dy = solver->dy;
double inv_sqr_sum = 1.0 / (dx * dx) + 1.0 / (dy * dy);
solver->dtBound = 0.5 * solver->re * 1.0 / inv_sqr_sum;
#ifdef VERBOSE
printConfig(solver);
#endif
}
void computeRHS(Solver* solver)
{
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
double idx = 1.0 / solver->dx;
double idy = 1.0 / solver->dy;
double idt = 1.0 / solver->dt;
double* rhs = solver->rhs;
double* f = solver->f;
double* g = solver->g;
shift(solver);
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = 1; i < imaxLocal + 1; i++) {
RHS(i, j) = ((F(i, j) - F(i - 1, j)) * idx + (G(i, j) - G(i, j - 1)) * idy) *
idt;
}
}
}
int solve(Solver* solver)
{
int imax = solver->imax;
int jmax = solver->jmax;
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
double eps = solver->eps;
int itermax = solver->itermax;
double dx2 = solver->dx * solver->dx;
double dy2 = solver->dy * solver->dy;
double idx2 = 1.0 / dx2;
double idy2 = 1.0 / dy2;
// identical to 1/((2/dx2)+(2/dy2))
double factor = solver->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
double* p = solver->p;
double* rhs = solver->rhs;
double epssq = eps * eps;
int it = 0;
double res = 1.0;
while ((res >= epssq) && (it < itermax)) {
res = 0.0;
exchange(solver, p);
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = 1; i < imaxLocal + 1; i++) {
double r = RHS(i, j) -
((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
(P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2);
P(i, j) -= (factor * r);
res += (r * r);
}
}
if (solver->coords[JDIM] == 0) { // set bottom bc
for (int i = 1; i < imaxLocal + 1; i++) {
P(i, 0) = P(i, 1);
}
}
if (solver->coords[JDIM] == (solver->dims[JDIM] - 1)) { // set top bc
for (int i = 1; i < imaxLocal + 1; i++) {
P(i, jmaxLocal + 1) = P(i, jmaxLocal);
}
}
if (solver->coords[IDIM] == 0) { // set left bc
for (int j = 1; j < jmaxLocal + 1; j++) {
P(0, j) = P(1, j);
}
}
if (solver->coords[IDIM] == (solver->dims[IDIM] - 1)) { // set right bc
for (int j = 1; j < jmaxLocal + 1; j++) {
P(imaxLocal + 1, j) = P(imaxLocal, j);
}
}
MPI_Allreduce(MPI_IN_PLACE, &res, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
res = res / (double)(imax * jmax);
#ifdef DEBUG
if (solver->rank == 0) {
printf("%d Residuum: %e\n", it, res);
}
#endif
it++;
}
#ifdef VERBOSE
if (solver->rank == 0) {
printf("Solver took %d iterations to reach %f\n", it, sqrt(res));
}
#endif
if (res < eps) {
return 0;
} else {
return 1;
}
}
static double maxElement(Solver* solver, double* m)
{
int size = (solver->imaxLocal + 2) * (solver->jmaxLocal + 2);
double maxval = DBL_MIN;
for (int i = 0; i < size; i++) {
maxval = MAX(maxval, fabs(m[i]));
}
MPI_Allreduce(MPI_IN_PLACE, &maxval, 1, MPI_DOUBLE, MPI_MAX, MPI_COMM_WORLD);
return maxval;
}
void computeTimestep(Solver* solver)
{
double dt = solver->dtBound;
double dx = solver->dx;
double dy = solver->dy;
double umax = maxElement(solver, solver->u);
double vmax = maxElement(solver, solver->v);
if (umax > 0) {
dt = (dt > dx / umax) ? dx / umax : dt;
}
if (vmax > 0) {
dt = (dt > dy / vmax) ? dy / vmax : dt;
}
solver->dt = dt * solver->tau;
}
void setBoundaryConditions(Solver* solver)
{
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
double* u = solver->u;
double* v = solver->v;
// Northern boundary
if (solver->coords[JDIM] == (solver->dims[JDIM] - 1)) { // set top bc
switch (solver->bcTop) {
case NOSLIP:
for (int i = 1; i < imaxLocal + 1; i++) {
V(i, jmaxLocal) = 0.0;
U(i, jmaxLocal + 1) = -U(i, jmaxLocal);
}
break;
case SLIP:
for (int i = 1; i < imaxLocal + 1; i++) {
V(i, jmaxLocal) = 0.0;
U(i, jmaxLocal + 1) = U(i, jmaxLocal);
}
break;
case OUTFLOW:
for (int i = 1; i < imaxLocal + 1; i++) {
U(i, jmaxLocal + 1) = U(i, jmaxLocal);
V(i, jmaxLocal) = V(i, jmaxLocal - 1);
}
break;
case PERIODIC:
break;
}
}
// Southern boundary
if (solver->coords[JDIM] == 0) { // set bottom bc
switch (solver->bcBottom) {
case NOSLIP:
for (int i = 1; i < imaxLocal + 1; i++) {
V(i, 0) = 0.0;
U(i, 0) = -U(i, 1);
}
break;
case SLIP:
for (int i = 1; i < imaxLocal + 1; i++) {
V(i, 0) = 0.0;
U(i, 0) = U(i, 1);
}
break;
case OUTFLOW:
for (int i = 1; i < imaxLocal + 1; i++) {
U(i, 0) = U(i, 1);
V(i, 0) = V(i, 1);
}
break;
case PERIODIC:
break;
}
}
// Eastern boundary
if (solver->coords[IDIM] == (solver->dims[IDIM] - 1)) { // set right bc
switch (solver->bcRight) {
case NOSLIP:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(imaxLocal, j) = 0.0;
V(imaxLocal + 1, j) = -V(imaxLocal, j);
}
break;
case SLIP:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(imaxLocal, j) = 0.0;
V(imaxLocal + 1, j) = V(imaxLocal, j);
}
break;
case OUTFLOW:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(imaxLocal, j) = U(imaxLocal - 1, j);
V(imaxLocal + 1, j) = V(imaxLocal, j);
}
break;
case PERIODIC:
break;
}
}
// Western boundary
if (solver->coords[IDIM] == 0) { // set left bc
switch (solver->bcLeft) {
case NOSLIP:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(0, j) = 0.0;
V(0, j) = -V(1, j);
}
break;
case SLIP:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(0, j) = 0.0;
V(0, j) = V(1, j);
}
break;
case OUTFLOW:
for (int j = 1; j < jmaxLocal + 1; j++) {
U(0, j) = U(1, j);
V(0, j) = V(1, j);
}
break;
case PERIODIC:
break;
}
}
}
void setSpecialBoundaryCondition(Solver* solver)
{
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
double* u = solver->u;
if (strcmp(solver->problem, "dcavity") == 0) {
if (solver->coords[JDIM] == (solver->dims[JDIM] - 1)) { // set top bc
for (int i = 1; i < imaxLocal + 1; i++) {
U(i, jmaxLocal + 1) = 2.0 - U(i, jmaxLocal);
}
}
} else if (strcmp(solver->problem, "canal") == 0) {
if (solver->coords[IDIM] == 0) { // set left bc
double ylength = solver->ylength;
double dy = solver->dy;
int rest = solver->jmax % solver->size;
int yc = solver->rank * (solver->jmax / solver->size) +
MIN(rest, solver->rank);
double ys = dy * (yc + 0.5);
double y;
/* printf("RANK %d yc: %d ys: %f\n", solver->rank, yc, ys); */
for (int j = 1; j < jmaxLocal + 1; j++) {
y = ys + dy * (j - 0.5);
U(0, j) = y * (ylength - y) * 4.0 / (ylength * ylength);
}
}
}
/* print(solver, solver->u); */
}
void computeFG(Solver* solver)
{
double* u = solver->u;
double* v = solver->v;
double* f = solver->f;
double* g = solver->g;
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
double gx = solver->gx;
double gy = solver->gy;
double gamma = solver->gamma;
double dt = solver->dt;
double inverseRe = 1.0 / solver->re;
double inverseDx = 1.0 / solver->dx;
double inverseDy = 1.0 / solver->dy;
double du2dx, dv2dy, duvdx, duvdy;
double du2dx2, du2dy2, dv2dx2, dv2dy2;
exchange(solver, u);
exchange(solver, v);
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = 1; i < imaxLocal + 1; i++) {
du2dx = inverseDx * 0.25 *
((U(i, j) + U(i + 1, j)) * (U(i, j) + U(i + 1, j)) -
(U(i, j) + U(i - 1, j)) * (U(i, j) + U(i - 1, j))) +
gamma * inverseDx * 0.25 *
(fabs(U(i, j) + U(i + 1, j)) * (U(i, j) - U(i + 1, j)) +
fabs(U(i, j) + U(i - 1, j)) * (U(i, j) - U(i - 1, j)));
duvdy = inverseDy * 0.25 *
((V(i, j) + V(i + 1, j)) * (U(i, j) + U(i, j + 1)) -
(V(i, j - 1) + V(i + 1, j - 1)) * (U(i, j) + U(i, j - 1))) +
gamma * inverseDy * 0.25 *
(fabs(V(i, j) + V(i + 1, j)) * (U(i, j) - U(i, j + 1)) +
fabs(V(i, j - 1) + V(i + 1, j - 1)) *
(U(i, j) - U(i, j - 1)));
du2dx2 = inverseDx * inverseDx * (U(i + 1, j) - 2.0 * U(i, j) + U(i - 1, j));
du2dy2 = inverseDy * inverseDy * (U(i, j + 1) - 2.0 * U(i, j) + U(i, j - 1));
F(i, j) = U(i, j) + dt * (inverseRe * (du2dx2 + du2dy2) - du2dx - duvdy + gx);
duvdx = inverseDx * 0.25 *
((U(i, j) + U(i, j + 1)) * (V(i, j) + V(i + 1, j)) -
(U(i - 1, j) + U(i - 1, j + 1)) * (V(i, j) + V(i - 1, j))) +
gamma * inverseDx * 0.25 *
(fabs(U(i, j) + U(i, j + 1)) * (V(i, j) - V(i + 1, j)) +
fabs(U(i - 1, j) + U(i - 1, j + 1)) *
(V(i, j) - V(i - 1, j)));
dv2dy = inverseDy * 0.25 *
((V(i, j) + V(i, j + 1)) * (V(i, j) + V(i, j + 1)) -
(V(i, j) + V(i, j - 1)) * (V(i, j) + V(i, j - 1))) +
gamma * inverseDy * 0.25 *
(fabs(V(i, j) + V(i, j + 1)) * (V(i, j) - V(i, j + 1)) +
fabs(V(i, j) + V(i, j - 1)) * (V(i, j) - V(i, j - 1)));
dv2dx2 = inverseDx * inverseDx * (V(i + 1, j) - 2.0 * V(i, j) + V(i - 1, j));
dv2dy2 = inverseDy * inverseDy * (V(i, j + 1) - 2.0 * V(i, j) + V(i, j - 1));
G(i, j) = V(i, j) + dt * (inverseRe * (dv2dx2 + dv2dy2) - duvdx - dv2dy + gy);
}
}
/* ----------------------------- boundary of F --------------------------- */
if (solver->coords[IDIM] == 0) { // set left bc
for (int j = 1; j < jmaxLocal + 1; j++) {
F(0, j) = U(0, j);
}
}
if (solver->coords[IDIM] == (solver->dims[IDIM] - 1)) { // set right bc
for (int j = 1; j < jmaxLocal + 1; j++) {
F(imaxLocal, j) = U(imaxLocal, j);
}
}
/* ----------------------------- boundary of G --------------------------- */
if (solver->coords[JDIM] == 0) { // set bottom bc
for (int i = 1; i < imaxLocal + 1; i++) {
G(i, 0) = V(i, 0);
}
}
if (solver->coords[JDIM] == (solver->dims[JDIM] - 1)) { // set top bc
for (int i = 1; i < imaxLocal + 1; i++) {
G(i, jmaxLocal) = V(i, jmaxLocal);
}
}
}
void adaptUV(Solver* solver)
{
int imaxLocal = solver->imaxLocal;
int jmaxLocal = solver->jmaxLocal;
double* p = solver->p;
double* u = solver->u;
double* v = solver->v;
double* f = solver->f;
double* g = solver->g;
double factorX = solver->dt / solver->dx;
double factorY = solver->dt / solver->dy;
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = 1; i < imaxLocal + 1; i++) {
U(i, j) = F(i, j) - (P(i + 1, j) - P(i, j)) * factorX;
V(i, j) = G(i, j) - (P(i, j + 1) - P(i, j)) * factorY;
}
}
}
void writeResult(Solver* solver, double* p, double* u, double* v)
{
int imax = solver->imax;
int jmax = solver->jmax;
double dx = solver->dx;
double dy = solver->dy;
double x = 0.0, y = 0.0;
FILE* fp;
fp = fopen("pressure.dat", "w");
if (fp == NULL) {
printf("Error!\n");
exit(EXIT_FAILURE);
}
for (int j = 1; j < jmax; j++) {
y = (double)(j - 0.5) * dy;
for (int i = 1; i < imax; i++) {
x = (double)(i - 0.5) * dx;
fprintf(fp, "%.2f %.2f %f\n", x, y, p[j * (imax) + i]);
}
fprintf(fp, "\n");
}
fclose(fp);
fp = fopen("velocity.dat", "w");
if (fp == NULL) {
printf("Error!\n");
exit(EXIT_FAILURE);
}
for (int j = 1; j < jmax; j++) {
y = dy * (j - 0.5);
for (int i = 1; i < imax; i++) {
x = dx * (i - 0.5);
double vel_u = (u[j * (imax) + i] + u[j * (imax) + (i - 1)]) / 2.0;
double vel_v = (v[j * (imax) + i] + v[(j - 1) * (imax) + i]) / 2.0;
double len = sqrt((vel_u * vel_u) + (vel_v * vel_v));
fprintf(fp, "%.2f %.2f %f %f %f\n", x, y, vel_u, vel_v, len);
}
}
fclose(fp);
}

View File

@ -1,58 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#ifndef __SOLVER_H_
#define __SOLVER_H_
#include "parameter.h"
#include <mpi.h>
#define NDIMS 2
enum BC { NOSLIP = 1, SLIP, OUTFLOW, PERIODIC };
typedef struct {
/* geometry and grid information */
double dx, dy;
int imax, jmax;
double xlength, ylength;
/* arrays */
double *p, *rhs;
double *f, *g;
double *u, *v;
/* parameters */
double eps, omega;
double re, tau, gamma;
double gx, gy;
/* time stepping */
int itermax;
double dt, te;
double dtBound;
char* problem;
int bcLeft, bcRight, bcBottom, bcTop;
/* mpi */
int rank;
int size;
MPI_Comm comm;
MPI_Datatype bufferTypes[NDIMS * 2];
MPI_Aint sdispls[NDIMS * 2], rdispls[NDIMS * 2];
int iNeighbours[NDIMS], jNeighbours[NDIMS];
int coords[NDIMS], dims[NDIMS];
int imaxLocal, jmaxLocal;
} Solver;
void initSolver(Solver*, Parameter*);
void computeRHS(Solver*);
int solve(Solver*);
void computeTimestep(Solver*);
void setBoundaryConditions(Solver*);
void setSpecialBoundaryCondition(Solver*);
void computeFG(Solver*);
void adaptUV(Solver*);
void collectResult(Solver*);
void writeResult(Solver*, double*, double*, double*);
void debugExchange(Solver*);
void print(Solver*, double*);
#endif

View File

@ -1,7 +0,0 @@
set terminal png size 1024,768 enhanced font ,12
set output 'p.png'
set datafile separator whitespace
set grid
set hidden3d
splot 'pressure.dat' using 1:2:3 with lines

View File

@ -1,5 +1,5 @@
#=======================================================================================
# Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
# Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
# All rights reserved.
# Use of this source code is governed by a MIT-style
# license that can be found in the LICENSE file.
@ -18,9 +18,11 @@ include $(MAKE_DIR)/include_$(TAG).mk
INCLUDES += -I$(SRC_DIR) -I$(BUILD_DIR)
VPATH = $(SRC_DIR)
SRC = $(wildcard $(SRC_DIR)/*.c)
SRC = $(filter-out $(wildcard $(SRC_DIR)/*-*.c),$(wildcard $(SRC_DIR)/*.c))
ASM = $(patsubst $(SRC_DIR)/%.c, $(BUILD_DIR)/%.s, $(SRC))
OBJ = $(patsubst $(SRC_DIR)/%.c, $(BUILD_DIR)/%.o, $(SRC))
OBJ += $(BUILD_DIR)/comm-$(COMM_TYPE).o
OBJ += $(BUILD_DIR)/solver-$(SOLVER).o
SOURCES = $(SRC) $(wildcard $(SRC_DIR)/*.h)
CPPFLAGS := $(CPPFLAGS) $(DEFINES) $(OPTIONS) $(INCLUDES)
@ -37,9 +39,20 @@ $(BUILD_DIR)/%.s: %.c
$(info ===> GENERATE ASM $@)
$(CC) -S $(CPPFLAGS) $(CFLAGS) $< -o $@
.PHONY: clean distclean tags info asm format
.PHONY: clean distclean vis vis_clean tags info asm format
clean:
vis:
$(info ===> GENERATE VISUALIZATION)
@gnuplot -e "filename='pressure.dat'" ./surface.plot
@gnuplot -e "filename='velocity.dat'" ./vector.plot
@gnuplot -e "filename='residual.dat'" ./residual.plot
vis_clean:
$(info ===> CLEAN VISUALIZATION)
@rm -f *.dat
@rm -f *.png
clean: vis_clean
$(info ===> CLEAN)
@rm -rf $(BUILD_DIR)
@rm -f tags
@ -47,6 +60,8 @@ clean:
distclean: clean
$(info ===> DIST CLEAN)
@rm -f $(TARGET)
@rm -f *.dat
@rm -f *.png
info:
$(info $(CFLAGS))

View File

@ -7,10 +7,10 @@
name canal # name of flow setup
bcN 1 # flags for boundary conditions
bcE 3 # 1 = no-slip 3 = outflow
bcS 1 # 2 = free-slip 4 = periodic
bcW 3 #
bcTop 1 # flags for boundary conditions
bcBottom 1 # 1 = no-slip 3 = outflow
bcLeft 3 # 2 = free-slip 4 = periodic
bcRight 3 #
gx 0.0 # Body forces (e.g. gravity)
gy 0.0 #
@ -27,15 +27,22 @@ p_init 0.0 # initial value for pressure
xlength 30.0 # domain size in x-direction
ylength 4.0 # domain size in y-direction
imax 200 # number of interior cells in x-direction
jmax 50 # number of interior cells in y-direction
jmax 40 # number of interior cells in y-direction
# Time Data:
# ---------
te 100.0 # final time
te 60.0 # final time
dt 0.02 # time stepsize
tau 0.5 # safety factor for time stepsize control (<0 constant delt)
# Multigrid data:
# ---------
levels 2 # Multigrid levels
presmooth 5 # Pre-smoothning iterations
postsmooth 5 # Post-smoothning iterations
# Pressure Iteration Data:
# -----------------------

View File

@ -1,10 +1,17 @@
# Supported: GCC, CLANG, ICC
TAG ?= CLANG
TAG ?= ICC
# Supported: true, false
ENABLE_MPI ?= true
ENABLE_OPENMP ?= false
# Supported: rb, mg
SOLVER ?= mg
# Supported: v1, v2, v3
COMM_TYPE ?= v3
#Feature options
OPTIONS += -DARRAY_ALIGNMENT=64
#OPTIONS += -DVERBOSE
OPTIONS += -DVERBOSE
# OPTIONS += -DTEST
#OPTIONS += -DVERBOSE_AFFINITY
#OPTIONS += -DVERBOSE_DATASIZE
#OPTIONS += -DVERBOSE_TIMER

View File

@ -15,7 +15,7 @@ bcRight 1 #
gx 0.0 # Body forces (e.g. gravity)
gy 0.0 #
re 10.0 # Reynolds number
re 100.0 # Reynolds number
u_init 0.0 # initial value for velocity in x-direction
v_init 0.0 # initial value for velocity in y-direction
@ -26,16 +26,23 @@ p_init 0.0 # initial value for pressure
xlength 1.0 # domain size in x-direction
ylength 1.0 # domain size in y-direction
imax 100 # number of interior cells in x-direction
jmax 100 # number of interior cells in y-direction
imax 128 # number of interior cells in x-direction
jmax 128 # number of interior cells in y-direction
# Time Data:
# ---------
te 5.0 # final time
te 10.0 # final time
dt 0.02 # time stepsize
tau 0.5 # safety factor for time stepsize control (<0 constant delt)
# Multigrid data:
# ---------
levels 2 # Multigrid levels
presmooth 20 # Pre-smoothning iterations
postsmooth 5 # Post-smoothning iterations
# Pressure Iteration Data:
# -----------------------

View File

@ -1,4 +1,10 @@
ifeq ($(ENABLE_MPI),true)
CC = mpicc
DEFINES = -D_MPI
else
CC = cc
endif
GCC = cc
LINKER = $(CC)
@ -9,8 +15,7 @@ LIBS = # -lomp
endif
VERSION = --version
CFLAGS = -Ofast -std=c99 $(OPENMP)
#CFLAGS = -Ofast -fnt-store=aggressive -std=c99 $(OPENMP) #AMD CLANG
LFLAGS = $(OPENMP)
DEFINES = -D_GNU_SOURCE# -DDEBUG
INCLUDES = -I/usr/local/include
CFLAGS = -Ofast -std=c17
LFLAGS = $(OPENMP) -lm
DEFINES += -D_GNU_SOURCE# -DDEBUG
INCLUDES = -I/opt/homebrew/include

View File

@ -1,4 +1,10 @@
ifeq ($(ENABLE_MPI),true)
CC = mpicc
DEFINES = -D_MPI
else
CC = gcc
endif
GCC = gcc
LINKER = $(CC)
@ -9,6 +15,6 @@ endif
VERSION = --version
CFLAGS = -Ofast -ffreestanding -std=c99 $(OPENMP)
LFLAGS = $(OPENMP)
DEFINES = -D_GNU_SOURCE
DEFINES += -D_GNU_SOURCE
INCLUDES =
LIBS =

View File

@ -1,4 +1,10 @@
ifeq ($(ENABLE_MPI),true)
CC = mpiicc
DEFINES = -D_MPI
else
CC = icc
endif
GCC = gcc
LINKER = $(CC)
@ -9,6 +15,6 @@ endif
VERSION = --version
CFLAGS = -O3 -xHost -qopt-zmm-usage=high -std=c99 $(OPENMP)
LFLAGS = $(OPENMP)
DEFINES = -D_GNU_SOURCE
DEFINES += -D_GNU_SOURCE# -DDEBUG
INCLUDES =
LIBS =

View File

@ -0,0 +1,9 @@
set terminal png size 1800,768 enhanced font ,12
set output 'residual.png'
set datafile separator whitespace
set xlabel "Timestep"
set ylabel "Residual"
set logscale y 2
plot 'residual.dat' using 1:2 title "Residual"

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.

View File

@ -1,14 +1,17 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#include <errno.h>
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
void* allocate(int alignment, size_t bytesize)
#include "allocate.h"
void* allocate(size_t alignment, size_t bytesize)
{
int errorCode;
void* ptr;

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
@ -8,6 +8,6 @@
#define __ALLOCATE_H_
#include <stdlib.h>
extern void* allocate(int alignment, size_t bytesize);
extern void* allocate(size_t alignment, size_t bytesize);
#endif

View File

@ -0,0 +1,234 @@
/*
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <stdlib.h>
#include "comm.h"
#ifdef _MPI
// subroutines local to this module
static int sum(int* sizes, int position)
{
int sum = 0;
for (int i = 0; i < position; i += position) {
sum += sizes[i];
}
return sum;
}
static void gatherArray(
Comm* c, int cnt, int* rcvCounts, int* displs, double* src, double* dst)
{
double* sendbuffer = src + (c->imaxLocal + 2);
if (c->rank == 0) {
sendbuffer = src;
}
MPI_Gatherv(sendbuffer,
cnt,
MPI_DOUBLE,
dst,
rcvCounts,
displs,
MPI_DOUBLE,
0,
MPI_COMM_WORLD);
}
#endif // defined _MPI
// exported subroutines
int commIsBoundary(Comm* c, int direction)
{
#ifdef _MPI
switch (direction) {
case LEFT:
return 1;
break;
case RIGHT:
return 1;
break;
case BOTTOM:
return c->rank == 0;
break;
case TOP:
return c->rank == (c->size - 1);
break;
}
#endif
return 1;
}
void commExchange(Comm* c, double* grid)
{
// printf("Rank : %d In exchange \n", c->rank);
#ifdef _MPI
MPI_Request requests[4] = { MPI_REQUEST_NULL,
MPI_REQUEST_NULL,
MPI_REQUEST_NULL,
MPI_REQUEST_NULL };
/* exchange ghost cells with top neighbor */
if (c->rank + 1 < c->size) {
int top = c->rank + 1;
double* src = grid + (c->jmaxLocal) * (c->imaxLocal + 2) + 1;
double* dst = grid + (c->jmaxLocal + 1) * (c->imaxLocal + 2) + 1;
MPI_Isend(src, c->imaxLocal, MPI_DOUBLE, top, 1, MPI_COMM_WORLD, &requests[0]);
MPI_Irecv(dst, c->imaxLocal, MPI_DOUBLE, top, 2, MPI_COMM_WORLD, &requests[1]);
}
/* exchange ghost cells with bottom neighbor */
if (c->rank > 0) {
int bottom = c->rank - 1;
double* src = grid + (c->imaxLocal + 2) + 1;
double* dst = grid + 1;
MPI_Isend(src, c->imaxLocal, MPI_DOUBLE, bottom, 2, MPI_COMM_WORLD, &requests[2]);
MPI_Irecv(dst, c->imaxLocal, MPI_DOUBLE, bottom, 1, MPI_COMM_WORLD, &requests[3]);
}
MPI_Waitall(4, requests, MPI_STATUSES_IGNORE);
#endif
}
void commShift(Comm* c, double* f, double* g)
{
#ifdef _MPI
MPI_Request requests[2] = { MPI_REQUEST_NULL, MPI_REQUEST_NULL };
/* shift G */
/* receive ghost cells from bottom neighbor */
if (c->rank > 0) {
int bottom = c->rank - 1;
MPI_Irecv(g + 1,
c->imaxLocal,
MPI_DOUBLE,
bottom,
0,
MPI_COMM_WORLD,
&requests[0]);
}
if (c->rank + 1 < c->size) {
int top = c->rank + 1;
double* buf = g + (c->jmaxLocal) * (c->imaxLocal + 2) + 1;
/* send ghost cells to top neighbor */
MPI_Isend(buf, c->imaxLocal, MPI_DOUBLE, top, 0, MPI_COMM_WORLD, &requests[1]);
}
MPI_Waitall(2, requests, MPI_STATUSES_IGNORE);
#endif
}
void commCollectResult(Comm* c,
double* ug,
double* vg,
double* pg,
double* u,
double* v,
double* p,
int jmax,
int imax)
{
#ifdef _MPI
int *rcvCounts, *displs;
int cnt = c->jmaxLocal * (imax + 2);
if (c->rank == 0) {
rcvCounts = (int*)malloc(c->size * sizeof(int));
displs = (int*)malloc(c->size * sizeof(int));
}
if (c->rank == 0 && c->size == 1) {
cnt = (c->jmaxLocal + 2) * (imax + 2);
} else if (c->rank == 0 || c->rank == (c->size - 1)) {
cnt = (c->jmaxLocal + 1) * (imax + 2);
}
MPI_Gather(&cnt, 1, MPI_INTEGER, rcvCounts, 1, MPI_INTEGER, 0, MPI_COMM_WORLD);
if (c->rank == 0) {
displs[0] = 0;
int cursor = rcvCounts[0];
for (int i = 1; i < c->size; i++) {
displs[i] = cursor;
cursor += rcvCounts[i];
}
}
gatherArray(c, cnt, rcvCounts, displs, p, pg);
gatherArray(c, cnt, rcvCounts, displs, u, ug);
gatherArray(c, cnt, rcvCounts, displs, v, vg);
#endif
}
void commPartition(Comm* c, int jmax, int imax)
{
#ifdef _MPI
c->imaxLocal = imax;
c->jmaxLocal = sizeOfRank(c->coords[JDIM], c->size, jmax);
c->neighbours[BOTTOM] = c->rank == 0 ? -1 : c->rank - 1;
c->neighbours[TOP] = c->rank == (c->size - 1) ? -1 : c->rank + 1;
c->neighbours[LEFT] = -1;
c->neighbours[RIGHT] = -1;
c->coords[IDIM] = 0;
c->coords[JDIM] = c->rank;
c->dims[IDIM] = 1;
c->dims[JDIM] = c->size;
#else
c->imaxLocal = imax;
c->jmaxLocal = jmax;
#endif
}
void commUpdateDatatypes(Comm* oldcomm, Comm* newcomm, int imaxLocal, int jmaxLocal)
{
#if defined _MPI
newcomm->comm = MPI_COMM_NULL;
int result = MPI_Comm_dup(MPI_COMM_WORLD, &newcomm->comm);
if (result == MPI_ERR_COMM) {
printf("\nNull communicator. Duplication failed !!\n");
}
newcomm->rank = oldcomm->rank;
newcomm->size = oldcomm->size;
newcomm->imaxLocal = imaxLocal / 2;
newcomm->jmaxLocal = jmaxLocal / 2;
newcomm->neighbours[BOTTOM] = newcomm->rank == 0 ? -1 : newcomm->rank - 1;
newcomm->neighbours[TOP] = newcomm->rank == (newcomm->size - 1) ? -1 : newcomm->rank + 1;
newcomm->neighbours[LEFT] = -1;
newcomm->neighbours[RIGHT] = -1;
newcomm->coords[IDIM] = 0;
newcomm->coords[JDIM] = newcomm->rank;
newcomm->dims[IDIM] = 1;
newcomm->dims[JDIM] = newcomm->size;
#endif
newcomm->imaxLocal = imaxLocal;
newcomm->jmaxLocal = jmaxLocal;
}
void commFreeCommunicator(Comm* comm)
{
#ifdef _MPI
MPI_Comm_free(&comm->comm);
#endif
}

View File

@ -0,0 +1,342 @@
/*
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include "comm.h"
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#ifdef _MPI
// subroutines local to this module
static int sum(int* sizes, int init, int offset, int coord)
{
int sum = 0;
for (int i = init - offset; coord > 0; i -= offset, --coord) {
sum += sizes[i];
}
return sum;
}
static void assembleResult(Comm* c, double* src, double* dst, int imax, int jmax)
{
MPI_Request* requests;
int numRequests = 1;
if (c->rank == 0) {
numRequests = c->size + 1;
} else {
numRequests = 1;
}
requests = (MPI_Request*)malloc(numRequests * sizeof(MPI_Request));
/* all ranks send their bulk array, including the external boundary layer */
MPI_Datatype bulkType;
int oldSizes[NDIMS] = { c->jmaxLocal + 2, c->imaxLocal + 2 };
int newSizes[NDIMS] = { c->jmaxLocal, c->imaxLocal };
int starts[NDIMS] = { 1, 1 };
if (commIsBoundary(c, LEFT)) {
newSizes[CIDIM] += 1;
starts[CIDIM] = 0;
}
if (commIsBoundary(c, RIGHT)) {
newSizes[CIDIM] += 1;
}
if (commIsBoundary(c, BOTTOM)) {
newSizes[CJDIM] += 1;
starts[CJDIM] = 0;
}
if (commIsBoundary(c, TOP)) {
newSizes[CJDIM] += 1;
}
MPI_Type_create_subarray(NDIMS,
oldSizes,
newSizes,
starts,
MPI_ORDER_C,
MPI_DOUBLE,
&bulkType);
MPI_Type_commit(&bulkType);
MPI_Isend(src, 1, bulkType, 0, 0, c->comm, &requests[0]);
int newSizesI[c->size];
int newSizesJ[c->size];
MPI_Gather(&newSizes[CIDIM], 1, MPI_INT, newSizesI, 1, MPI_INT, 0, MPI_COMM_WORLD);
MPI_Gather(&newSizes[CJDIM], 1, MPI_INT, newSizesJ, 1, MPI_INT, 0, MPI_COMM_WORLD);
/* rank 0 assembles the subdomains */
if (c->rank == 0) {
for (int i = 0; i < c->size; i++) {
MPI_Datatype domainType;
int oldSizes[NDIMS] = { jmax + 2, imax + 2 };
int newSizes[NDIMS] = { newSizesJ[i], newSizesI[i] };
int coords[NDIMS];
MPI_Cart_coords(c->comm, i, NDIMS, coords);
int starts[NDIMS] = { sum(newSizesJ, i, 1, coords[JDIM]),
sum(newSizesI, i, c->dims[JDIM], coords[IDIM]) };
printf(
"Rank: %d, Coords(i,j): %d %d, Size(i,j): %d %d, Target Size(i,j): %d %d "
"Starts(i,j): %d %d\n",
i,
coords[IDIM],
coords[JDIM],
oldSizes[CIDIM],
oldSizes[CJDIM],
newSizes[CIDIM],
newSizes[CJDIM],
starts[CIDIM],
starts[CJDIM]);
MPI_Type_create_subarray(NDIMS,
oldSizes,
newSizes,
starts,
MPI_ORDER_C,
MPI_DOUBLE,
&domainType);
MPI_Type_commit(&domainType);
MPI_Irecv(dst, 1, domainType, i, 0, c->comm, &requests[i + 1]);
MPI_Type_free(&domainType);
}
}
MPI_Waitall(numRequests, requests, MPI_STATUSES_IGNORE);
}
#endif // defined _MPI
// exported subroutines
int commIsBoundary(Comm* c, int direction)
{
#ifdef _MPI
switch (direction) {
case LEFT:
return c->coords[IDIM] == 0;
break;
case RIGHT:
return c->coords[IDIM] == (c->dims[IDIM] - 1);
break;
case BOTTOM:
return c->coords[JDIM] == 0;
break;
case TOP:
return c->coords[JDIM] == (c->dims[JDIM] - 1);
break;
}
#endif
return 1;
}
void commExchange(Comm* c, double* grid)
{
#ifdef _MPI
MPI_Request requests[8];
for (int i = 0; i < 8; i++)
requests[i] = MPI_REQUEST_NULL;
for (int i = 0; i < NDIRS; i++) {
double* sbuf = grid + c->sdispls[i];
double* rbuf = grid + c->rdispls[i];
int tag = 0;
if (c->neighbours[i] != MPI_PROC_NULL) {
// printf("DEBUG: Rank %d - SendRecv with %d\n", c->rank, c->neighbours[i]);
tag = c->neighbours[i];
}
MPI_Irecv(rbuf,
1,
c->bufferTypes[i],
c->neighbours[i],
tag,
c->comm,
&requests[i * 2]);
MPI_Isend(sbuf,
1,
c->bufferTypes[i],
c->neighbours[i],
c->rank,
c->comm,
&requests[i * 2 + 1]);
}
MPI_Waitall(8, requests, MPI_STATUSES_IGNORE);
#endif
}
void commShift(Comm* c, double* f, double* g)
{
#ifdef _MPI
MPI_Request requests[4] = { MPI_REQUEST_NULL,
MPI_REQUEST_NULL,
MPI_REQUEST_NULL,
MPI_REQUEST_NULL };
/* shift G */
/* receive ghost cells from bottom neighbor */
double* buf = g + 1;
MPI_Irecv(buf,
1,
c->bufferTypes[BOTTOM],
c->neighbours[BOTTOM],
0,
c->comm,
&requests[0]);
/* send ghost cells to top neighbor */
buf = g + (c->jmaxLocal) * (c->imaxLocal + 2) + 1;
MPI_Isend(buf, 1, c->bufferTypes[TOP], c->neighbours[TOP], 0, c->comm, &requests[1]);
/* shift F */
/* receive ghost cells from left neighbor */
buf = f + (c->imaxLocal + 2);
MPI_Irecv(buf,
1,
c->bufferTypes[LEFT],
c->neighbours[LEFT],
1,
c->comm,
&requests[2]);
/* send ghost cells to right neighbor */
buf = f + (c->imaxLocal + 2) + (c->imaxLocal);
MPI_Isend(buf,
1,
c->bufferTypes[RIGHT],
c->neighbours[RIGHT],
1,
c->comm,
&requests[3]);
MPI_Waitall(4, requests, MPI_STATUSES_IGNORE);
#endif
}
void commCollectResult(Comm* c,
double* ug,
double* vg,
double* pg,
double* u,
double* v,
double* p,
int imax,
int jmax)
{
#ifdef _MPI
/* collect P */
assembleResult(c, p, pg, imax, jmax);
/* collect U */
assembleResult(c, u, ug, imax, jmax);
/* collect V */
assembleResult(c, v, vg, imax, jmax);
#endif
}
void commPartition(Comm* c, int jmax, int imax)
{
#ifdef _MPI
int dims[NDIMS] = { 0, 0 };
int periods[NDIMS] = { 0, 0 };
MPI_Dims_create(c->size, NDIMS, dims);
MPI_Cart_create(MPI_COMM_WORLD, NDIMS, dims, periods, 0, &c->comm);
MPI_Cart_shift(c->comm, IDIM, 1, &c->neighbours[LEFT], &c->neighbours[RIGHT]);
MPI_Cart_shift(c->comm, JDIM, 1, &c->neighbours[BOTTOM], &c->neighbours[TOP]);
MPI_Cart_get(c->comm, NDIMS, c->dims, periods, c->coords);
int imaxLocal = sizeOfRank(c->coords[IDIM], dims[IDIM], imax);
int jmaxLocal = sizeOfRank(c->coords[JDIM], dims[JDIM], jmax);
c->imaxLocal = imaxLocal;
c->jmaxLocal = jmaxLocal;
MPI_Datatype jBufferType;
MPI_Type_contiguous(imaxLocal, MPI_DOUBLE, &jBufferType);
MPI_Type_commit(&jBufferType);
MPI_Datatype iBufferType;
MPI_Type_vector(jmaxLocal, 1, imaxLocal + 2, MPI_DOUBLE, &iBufferType);
MPI_Type_commit(&iBufferType);
c->bufferTypes[LEFT] = iBufferType;
c->bufferTypes[RIGHT] = iBufferType;
c->bufferTypes[BOTTOM] = jBufferType;
c->bufferTypes[TOP] = jBufferType;
c->sdispls[LEFT] = (imaxLocal + 2) + 1;
c->sdispls[RIGHT] = (imaxLocal + 2) + imaxLocal;
c->sdispls[BOTTOM] = (imaxLocal + 2) + 1;
c->sdispls[TOP] = jmaxLocal * (imaxLocal + 2) + 1;
c->rdispls[LEFT] = (imaxLocal + 2);
c->rdispls[RIGHT] = (imaxLocal + 2) + (imaxLocal + 1);
c->rdispls[BOTTOM] = 1;
c->rdispls[TOP] = (jmaxLocal + 1) * (imaxLocal + 2) + 1;
#else
c->imaxLocal = imax;
c->jmaxLocal = jmax;
#endif
}
void commUpdateDatatypes(Comm* oldcomm, Comm* newcomm, int imaxLocal, int jmaxLocal)
{
#if defined _MPI
newcomm->comm = MPI_COMM_NULL;
int result = MPI_Comm_dup(oldcomm->comm, &newcomm->comm);
if (result == MPI_ERR_COMM) {
printf("\nNull communicator. Duplication failed !!\n");
}
newcomm->rank = oldcomm->rank;
newcomm->size = oldcomm->size;
memcpy(&newcomm->neighbours, &oldcomm->neighbours, sizeof(oldcomm->neighbours));
memcpy(&newcomm->coords, &oldcomm->coords, sizeof(oldcomm->coords));
memcpy(&newcomm->dims, &oldcomm->dims, sizeof(oldcomm->dims));
newcomm->imaxLocal = imaxLocal/2;
newcomm->jmaxLocal = jmaxLocal/2;
MPI_Datatype jBufferType;
MPI_Type_contiguous(imaxLocal, MPI_DOUBLE, &jBufferType);
MPI_Type_commit(&jBufferType);
MPI_Datatype iBufferType;
MPI_Type_vector(jmaxLocal, 1, imaxLocal + 2, MPI_DOUBLE, &iBufferType);
MPI_Type_commit(&iBufferType);
newcomm->bufferTypes[LEFT] = iBufferType;
newcomm->bufferTypes[RIGHT] = iBufferType;
newcomm->bufferTypes[BOTTOM] = jBufferType;
newcomm->bufferTypes[TOP] = jBufferType;
newcomm->sdispls[LEFT] = (imaxLocal + 2) + 1;
newcomm->sdispls[RIGHT] = (imaxLocal + 2) + imaxLocal;
newcomm->sdispls[BOTTOM] = (imaxLocal + 2) + 1;
newcomm->sdispls[TOP] = jmaxLocal * (imaxLocal + 2) + 1;
newcomm->rdispls[LEFT] = (imaxLocal + 2);
newcomm->rdispls[RIGHT] = (imaxLocal + 2) + (imaxLocal + 1);
newcomm->rdispls[BOTTOM] = 1;
newcomm->rdispls[TOP] = (jmaxLocal + 1) * (imaxLocal + 2) + 1;
#else
newcomm->imaxLocal = imaxLocal;
newcomm->jmaxLocal = jmaxLocal;
#endif
}
void commFreeCommunicator(Comm* comm)
{
#ifdef _MPI
MPI_Comm_free(&comm->comm);
#endif
}

View File

@ -0,0 +1,320 @@
/*
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <stdio.h>
#include <stdlib.h>
#include "comm.h"
#ifdef _MPI
// subroutines local to this module
static int sum(int* sizes, int init, int offset, int coord)
{
int sum = 0;
for (int i = init - offset; coord > 0; i -= offset, --coord) {
sum += sizes[i];
}
return sum;
}
static void assembleResult(Comm* c, double* src, double* dst, int imax, int jmax)
{
MPI_Request* requests;
int numRequests = 1;
if (c->rank == 0) {
numRequests = c->size + 1;
} else {
numRequests = 1;
}
requests = (MPI_Request*)malloc(numRequests * sizeof(MPI_Request));
/* all ranks send their bulk array, including the external boundary layer */
MPI_Datatype bulkType;
int oldSizes[NDIMS] = { c->jmaxLocal + 2, c->imaxLocal + 2 };
int newSizes[NDIMS] = { c->jmaxLocal, c->imaxLocal };
int starts[NDIMS] = { 1, 1 };
if (commIsBoundary(c, LEFT)) {
newSizes[CIDIM] += 1;
starts[CIDIM] = 0;
}
if (commIsBoundary(c, RIGHT)) {
newSizes[CIDIM] += 1;
}
if (commIsBoundary(c, BOTTOM)) {
newSizes[CJDIM] += 1;
starts[CJDIM] = 0;
}
if (commIsBoundary(c, TOP)) {
newSizes[CJDIM] += 1;
}
MPI_Type_create_subarray(NDIMS,
oldSizes,
newSizes,
starts,
MPI_ORDER_C,
MPI_DOUBLE,
&bulkType);
MPI_Type_commit(&bulkType);
MPI_Isend(src, 1, bulkType, 0, 0, c->comm, &requests[0]);
int newSizesI[c->size];
int newSizesJ[c->size];
MPI_Gather(&newSizes[CIDIM], 1, MPI_INT, newSizesI, 1, MPI_INT, 0, MPI_COMM_WORLD);
MPI_Gather(&newSizes[CJDIM], 1, MPI_INT, newSizesJ, 1, MPI_INT, 0, MPI_COMM_WORLD);
/* rank 0 assembles the subdomains */
if (c->rank == 0) {
for (int i = 0; i < c->size; i++) {
MPI_Datatype domainType;
int oldSizes[NDIMS] = { jmax + 2, imax + 2 };
int newSizes[NDIMS] = { newSizesJ[i], newSizesI[i] };
int coords[NDIMS];
MPI_Cart_coords(c->comm, i, NDIMS, coords);
int starts[NDIMS] = { sum(newSizesJ, i, 1, coords[JDIM]),
sum(newSizesI, i, c->dims[JDIM], coords[IDIM]) };
printf(
"Rank: %d, Coords(i,j): %d %d, Size(i,j): %d %d, Target Size(i,j): %d %d "
"Starts(i,j): %d %d\n",
i,
coords[IDIM],
coords[JDIM],
oldSizes[CIDIM],
oldSizes[CJDIM],
newSizes[CIDIM],
newSizes[CJDIM],
starts[CIDIM],
starts[CJDIM]);
MPI_Type_create_subarray(NDIMS,
oldSizes,
newSizes,
starts,
MPI_ORDER_C,
MPI_DOUBLE,
&domainType);
MPI_Type_commit(&domainType);
MPI_Irecv(dst, 1, domainType, i, 0, c->comm, &requests[i + 1]);
MPI_Type_free(&domainType);
}
}
MPI_Waitall(numRequests, requests, MPI_STATUSES_IGNORE);
}
#endif // defined _MPI
// exported subroutines
int commIsBoundary(Comm* c, int direction)
{
#ifdef _MPI
switch (direction) {
case LEFT:
return c->coords[IDIM] == 0;
break;
case RIGHT:
return c->coords[IDIM] == (c->dims[IDIM] - 1);
break;
case BOTTOM:
return c->coords[JDIM] == 0;
break;
case TOP:
return c->coords[JDIM] == (c->dims[JDIM] - 1);
break;
}
#endif
return 1;
}
void commExchange(Comm* c, double* grid)
{
#ifdef _MPI
int counts[NDIRS] = { 1, 1, 1, 1 };
MPI_Neighbor_alltoallw(grid,
counts,
c->sdispls,
c->bufferTypes,
grid,
counts,
c->rdispls,
c->bufferTypes,
c->comm);
#endif
}
void commShift(Comm* c, double* f, double* g)
{
#ifdef _MPI
MPI_Request requests[4] = { MPI_REQUEST_NULL,
MPI_REQUEST_NULL,
MPI_REQUEST_NULL,
MPI_REQUEST_NULL };
/* shift G */
/* receive ghost cells from bottom neighbor */
double* buf = g + 1;
MPI_Irecv(buf,
1,
c->bufferTypes[BOTTOM],
c->neighbours[BOTTOM],
0,
c->comm,
&requests[0]);
/* send ghost cells to top neighbor */
buf = g + (c->jmaxLocal) * (c->imaxLocal + 2) + 1;
MPI_Isend(buf, 1, c->bufferTypes[TOP], c->neighbours[TOP], 0, c->comm, &requests[1]);
/* shift F */
/* receive ghost cells from left neighbor */
buf = f + (c->imaxLocal + 2);
MPI_Irecv(buf,
1,
c->bufferTypes[LEFT],
c->neighbours[LEFT],
1,
c->comm,
&requests[2]);
/* send ghost cells to right neighbor */
buf = f + (c->imaxLocal + 2) + (c->imaxLocal);
MPI_Isend(buf,
1,
c->bufferTypes[RIGHT],
c->neighbours[RIGHT],
1,
c->comm,
&requests[3]);
MPI_Waitall(4, requests, MPI_STATUSES_IGNORE);
#endif
}
void commCollectResult(Comm* c,
double* ug,
double* vg,
double* pg,
double* u,
double* v,
double* p,
int imax,
int jmax)
{
#ifdef _MPI
/* collect P */
assembleResult(c, p, pg, imax, jmax);
/* collect U */
assembleResult(c, u, ug, imax, jmax);
/* collect V */
assembleResult(c, v, vg, imax, jmax);
#endif
}
void commPartition(Comm* c, int jmax, int imax)
{
#ifdef _MPI
int dims[NDIMS] = { 0, 0 };
int periods[NDIMS] = { 0, 0 };
MPI_Dims_create(c->size, NDIMS, dims);
MPI_Cart_create(MPI_COMM_WORLD, NDIMS, dims, periods, 0, &c->comm);
MPI_Cart_shift(c->comm, IDIM, 1, &c->neighbours[LEFT], &c->neighbours[RIGHT]);
MPI_Cart_shift(c->comm, JDIM, 1, &c->neighbours[BOTTOM], &c->neighbours[TOP]);
MPI_Cart_get(c->comm, NDIMS, c->dims, periods, c->coords);
int imaxLocal = sizeOfRank(c->coords[IDIM], dims[IDIM], imax);
int jmaxLocal = sizeOfRank(c->coords[JDIM], dims[JDIM], jmax);
c->imaxLocal = imaxLocal;
c->jmaxLocal = jmaxLocal;
MPI_Datatype jBufferType;
MPI_Type_contiguous(imaxLocal, MPI_DOUBLE, &jBufferType);
MPI_Type_commit(&jBufferType);
MPI_Datatype iBufferType;
MPI_Type_vector(jmaxLocal, 1, imaxLocal + 2, MPI_DOUBLE, &iBufferType);
MPI_Type_commit(&iBufferType);
c->bufferTypes[LEFT] = iBufferType;
c->bufferTypes[RIGHT] = iBufferType;
c->bufferTypes[BOTTOM] = jBufferType;
c->bufferTypes[TOP] = jBufferType;
size_t dblsize = sizeof(double);
c->sdispls[LEFT] = ((imaxLocal + 2) + 1) * dblsize;
c->sdispls[RIGHT] = ((imaxLocal + 2) + imaxLocal) * dblsize;
c->sdispls[BOTTOM] = ((imaxLocal + 2) + 1) * dblsize;
c->sdispls[TOP] = (jmaxLocal * (imaxLocal + 2) + 1) * dblsize;
c->rdispls[LEFT] = (imaxLocal + 2) * dblsize;
c->rdispls[RIGHT] = ((imaxLocal + 2) + (imaxLocal + 1)) * dblsize;
c->rdispls[BOTTOM] = 1 * dblsize;
c->rdispls[TOP] = ((jmaxLocal + 1) * (imaxLocal + 2) + 1) * dblsize;
#else
c->imaxLocal = imax;
c->jmaxLocal = jmax;
#endif
}
void commUpdateDatatypes(Comm* oldcomm, Comm* newcomm, int imaxLocal, int jmaxLocal)
{
#if defined _MPI
int result = MPI_Comm_dup(oldcomm->comm, &newcomm->comm);
if (result == MPI_ERR_COMM) {
printf("\nNull communicator. Duplication failed !!\n");
}
newcomm->rank = oldcomm->rank;
newcomm->size = oldcomm->size;
newcomm->imaxLocal = imaxLocal / 2;
newcomm->jmaxLocal = jmaxLocal / 2;
MPI_Datatype jBufferType;
MPI_Type_contiguous(imaxLocal, MPI_DOUBLE, &jBufferType);
MPI_Type_commit(&jBufferType);
MPI_Datatype iBufferType;
MPI_Type_vector(jmaxLocal, 1, imaxLocal + 2, MPI_DOUBLE, &iBufferType);
MPI_Type_commit(&iBufferType);
newcomm->bufferTypes[LEFT] = iBufferType;
newcomm->bufferTypes[RIGHT] = iBufferType;
newcomm->bufferTypes[BOTTOM] = jBufferType;
newcomm->bufferTypes[TOP] = jBufferType;
newcomm->sdispls[LEFT] = (imaxLocal + 2) + 1;
newcomm->sdispls[RIGHT] = (imaxLocal + 2) + imaxLocal;
newcomm->sdispls[BOTTOM] = (imaxLocal + 2) + 1;
newcomm->sdispls[TOP] = jmaxLocal * (imaxLocal + 2) + 1;
newcomm->rdispls[LEFT] = (imaxLocal + 2);
newcomm->rdispls[RIGHT] = (imaxLocal + 2) + (imaxLocal + 1);
newcomm->rdispls[BOTTOM] = 1;
newcomm->rdispls[TOP] = (jmaxLocal + 1) * (imaxLocal + 2) + 1;
#else
newcomm->imaxLocal = imaxLocal;
newcomm->jmaxLocal = jmaxLocal;
#endif
}
void commFreeCommunicator(Comm* comm)
{
#ifdef _MPI
MPI_Comm_free(&comm->comm);
#endif
}

View File

@ -1,281 +1,34 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include "comm.h"
// subroutines local to this module
static int sizeOfRank(int rank, int size, int N)
int sizeOfRank(int rank, int size, int N)
{
return N / size + ((N % size > rank) ? 1 : 0);
}
static void setupCommunication(Comm* c, int direction, int layer)
{
MPI_Datatype type;
size_t dblsize = sizeof(double);
int imaxLocal = c->imaxLocal;
int jmaxLocal = c->jmaxLocal;
int sizes[NDIMS];
int subSizes[NDIMS];
int starts[NDIMS];
int offset = 0;
sizes[IDIM] = imaxLocal + 2;
sizes[JDIM] = jmaxLocal + 2;
if (layer == HALO) {
offset = 1;
}
switch (direction) {
case LEFT:
subSizes[IDIM] = 1;
subSizes[JDIM] = jmaxLocal;
starts[IDIM] = 1 - offset;
starts[JDIM] = 1;
break;
case RIGHT:
subSizes[IDIM] = 1;
subSizes[JDIM] = jmaxLocal;
starts[IDIM] = imaxLocal + offset;
starts[JDIM] = 1;
break;
case BOTTOM:
subSizes[IDIM] = imaxLocal;
subSizes[JDIM] = 1;
starts[IDIM] = 1;
starts[JDIM] = 1 - offset;
break;
case TOP:
subSizes[IDIM] = imaxLocal;
subSizes[JDIM] = 1;
starts[IDIM] = 1;
starts[JDIM] = jmaxLocal + offset;
break;
}
MPI_Type_create_subarray(NDIMS,
sizes,
subSizes,
starts,
MPI_ORDER_C,
MPI_DOUBLE,
&type);
MPI_Type_commit(&type);
if (layer == HALO) {
c->rbufferTypes[direction] = type;
} else if (layer == BULK) {
c->sbufferTypes[direction] = type;
}
}
static void assembleResult(Comm* c,
double* src,
double* dst,
int imaxLocal[],
int jmaxLocal[],
int offset[],
int jmax,
int imax)
{
MPI_Request* requests;
int numRequests = 1;
if (c->rank == 0) {
numRequests = c->size + 1;
} else {
numRequests = 1;
}
requests = (MPI_Request*)malloc(numRequests * sizeof(MPI_Request));
/* all ranks send their bulk array */
MPI_Datatype bulkType;
int oldSizes[NDIMS] = { c->jmaxLocal + 2, c->imaxLocal + 2 };
int newSizes[NDIMS] = { c->jmaxLocal, c->imaxLocal };
int starts[NDIMS] = { 1, 1 };
MPI_Type_create_subarray(NDIMS,
oldSizes,
newSizes,
starts,
MPI_ORDER_C,
MPI_DOUBLE,
&bulkType);
MPI_Type_commit(&bulkType);
MPI_Isend(src, 1, bulkType, 0, 0, c->comm, &requests[0]);
/* rank 0 assembles the subdomains */
if (c->rank == 0) {
for (int i = 0; i < c->size; i++) {
MPI_Datatype domainType;
int oldSizes[NDIMS] = { jmax, imax };
int newSizes[NDIMS] = { jmaxLocal[i], imaxLocal[i] };
int starts[NDIMS] = { offset[i * NDIMS + JDIM], offset[i * NDIMS + IDIM] };
MPI_Type_create_subarray(NDIMS,
oldSizes,
newSizes,
starts,
MPI_ORDER_C,
MPI_DOUBLE,
&domainType);
MPI_Type_commit(&domainType);
MPI_Irecv(dst, 1, domainType, i, 0, c->comm, &requests[i + 1]);
}
}
MPI_Waitall(numRequests, requests, MPI_STATUSES_IGNORE);
}
static int sum(int* sizes, int position)
{
int sum = 0;
for (int i = 0; i < position; i++) {
sum += sizes[i];
}
return sum;
}
// exported subroutines
void commReduction(double* v, int op)
{
#ifdef _MPI
if (op == MAX) {
MPI_Allreduce(MPI_IN_PLACE, v, 1, MPI_DOUBLE, MPI_MAX, MPI_COMM_WORLD);
} else if (op == SUM) {
MPI_Allreduce(MPI_IN_PLACE, v, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
}
}
int commIsBoundary(Comm* c, int direction)
{
switch (direction) {
case LEFT:
return c->coords[IDIM] == 0;
break;
case RIGHT:
return c->coords[IDIM] == (c->dims[IDIM] - 1);
break;
case BOTTOM:
return c->coords[JDIM] == 0;
break;
case TOP:
return c->coords[JDIM] == (c->dims[JDIM] - 1);
break;
}
return 0;
}
void commExchange(Comm* c, double* grid)
{
int counts[NDIRS] = { 1, 1, 1, 1 };
MPI_Aint displs[NDIRS] = { 0, 0, 0, 0 };
MPI_Neighbor_alltoallw(grid,
counts,
displs,
c->sbufferTypes,
grid,
counts,
displs,
c->rbufferTypes,
c->comm);
}
void commShift(Comm* c, double* f, double* g)
{
MPI_Request requests[4] = { MPI_REQUEST_NULL,
MPI_REQUEST_NULL,
MPI_REQUEST_NULL,
MPI_REQUEST_NULL };
/* shift G */
/* receive ghost cells from bottom neighbor */
MPI_Irecv(g,
1,
c->rbufferTypes[BOTTOM],
c->neighbours[BOTTOM],
0,
c->comm,
&requests[0]);
/* send ghost cells to top neighbor */
MPI_Isend(g, 1, c->sbufferTypes[TOP], c->neighbours[TOP], 0, c->comm, &requests[1]);
/* shift F */
/* receive ghost cells from left neighbor */
MPI_Irecv(f, 1, c->rbufferTypes[LEFT], c->neighbours[LEFT], 1, c->comm, &requests[2]);
/* send ghost cells to right neighbor */
MPI_Isend(f,
1,
c->sbufferTypes[RIGHT],
c->neighbours[RIGHT],
1,
c->comm,
&requests[3]);
MPI_Waitall(4, requests, MPI_STATUSES_IGNORE);
}
void commCollectResult(Comm* c,
double* ug,
double* vg,
double* pg,
double* u,
double* v,
double* p,
int jmax,
int imax)
{
int offset[c->size * NDIMS];
int imaxLocal[c->size];
int jmaxLocal[c->size];
MPI_Gather(&c->imaxLocal, 1, MPI_INT, imaxLocal, 1, MPI_INT, 0, MPI_COMM_WORLD);
MPI_Gather(&c->jmaxLocal, 1, MPI_INT, jmaxLocal, 1, MPI_INT, 0, MPI_COMM_WORLD);
if (c->rank == 0) {
for (int i = 0; i < c->size; i++) {
int coords[NDIMS];
MPI_Cart_coords(c->comm, i, NDIMS, coords);
offset[i * NDIMS + IDIM] = sum(imaxLocal, coords[IDIM]);
offset[i * NDIMS + JDIM] = sum(jmaxLocal, coords[JDIM]);
printf("Rank: %d, Coords(j,i): %d %d, Size(j,i): %d %d "
"Offset(j,i): %d %d\n",
i,
coords[JDIM],
coords[IDIM],
jmaxLocal[i],
imaxLocal[i],
offset[i * NDIMS + JDIM],
offset[i * NDIMS + IDIM]);
}
}
/* collect P */
assembleResult(c, p, pg, imaxLocal, jmaxLocal, offset, jmax, imax);
/* collect U */
assembleResult(c, u, ug, imaxLocal, jmaxLocal, offset, jmax, imax);
/* collect V */
assembleResult(c, v, vg, imaxLocal, jmaxLocal, offset, jmax, imax);
#endif
}
void commPrintConfig(Comm* c)
{
#ifdef _MPI
fflush(stdout);
MPI_Barrier(MPI_COMM_WORLD);
if (commIsMaster(c)) {
@ -290,37 +43,87 @@ void commPrintConfig(Comm* c)
c->neighbours[TOP],
c->neighbours[LEFT],
c->neighbours[RIGHT]);
printf("\tCoordinates (j,i) %d %d\n", c->coords[JDIM], c->coords[IDIM]);
printf("\tLocal domain size (j,i) %dx%d\n", c->jmaxLocal, c->imaxLocal);
printf("\tIs boundary:\n");
printf("\t\tLEFT: %d\n", commIsBoundary(c, LEFT));
printf("\t\tRIGHT: %d\n", commIsBoundary(c, RIGHT));
printf("\t\tBOTTOM: %d\n", commIsBoundary(c, BOTTOM));
printf("\t\tTOP: %d\n", commIsBoundary(c, TOP));
printf("\tCoordinates (i,j) %d %d\n", c->coords[IDIM], c->coords[JDIM]);
printf("\tDims (i,j) %d %d\n", c->dims[IDIM], c->dims[JDIM]);
printf("\tLocal domain size (i,j) %dx%d\n", c->imaxLocal, c->jmaxLocal);
fflush(stdout);
}
}
MPI_Barrier(MPI_COMM_WORLD);
}
#endif
}
void commInit(Comm* c, int jmax, int imax)
void commInit(Comm* c, int argc, char** argv)
{
/* setup communication */
#ifdef _MPI
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &(c->rank));
MPI_Comm_size(MPI_COMM_WORLD, &(c->size));
int dims[NDIMS] = { 0, 0 };
int periods[NDIMS] = { 0, 0 };
MPI_Dims_create(c->size, NDIMS, dims);
MPI_Cart_create(MPI_COMM_WORLD, NDIMS, dims, periods, 0, &c->comm);
MPI_Cart_shift(c->comm, IDIM, 1, &c->neighbours[LEFT], &c->neighbours[RIGHT]);
MPI_Cart_shift(c->comm, JDIM, 1, &c->neighbours[BOTTOM], &c->neighbours[TOP]);
MPI_Cart_get(c->comm, NDIMS, c->dims, periods, c->coords);
c->imaxLocal = sizeOfRank(c->rank, dims[IDIM], imax);
c->jmaxLocal = sizeOfRank(c->rank, dims[JDIM], jmax);
// setup buffer types for communication
setupCommunication(c, LEFT, BULK);
setupCommunication(c, LEFT, HALO);
setupCommunication(c, RIGHT, BULK);
setupCommunication(c, RIGHT, HALO);
setupCommunication(c, BOTTOM, BULK);
setupCommunication(c, BOTTOM, HALO);
setupCommunication(c, TOP, BULK);
setupCommunication(c, TOP, HALO);
#else
c->rank = 0;
c->size = 1;
#endif
}
void commTestInit(Comm* c, double* p, double* f, double* g)
{
int imax = c->imaxLocal;
int jmax = c->jmaxLocal;
int rank = c->rank;
for (int j = 0; j < jmax + 2; j++) {
for (int i = 0; i < imax + 2; i++) {
p[j * (imax + 2) + i] = rank;
f[j * (imax + 2) + i] = rank;
g[j * (imax + 2) + i] = rank;
}
}
}
static void testWriteFile(char* filename, double* grid, int imax, int jmax)
{
FILE* fp = fopen(filename, "w");
if (fp == NULL) {
printf("Error!\n");
exit(EXIT_FAILURE);
}
for (int j = 0; j < jmax + 2; j++) {
for (int i = 0; i < imax + 2; i++) {
fprintf(fp, "%.2f ", grid[j * (imax + 2) + i]);
}
fprintf(fp, "\n");
}
fclose(fp);
}
void commTestWrite(Comm* c, double* p, double* f, double* g)
{
int imax = c->imaxLocal;
int jmax = c->jmaxLocal;
int rank = c->rank;
char filename[30];
snprintf(filename, 30, "ptest-%d.dat", rank);
testWriteFile(filename, p, imax, jmax);
snprintf(filename, 30, "ftest-%d.dat", rank);
testWriteFile(filename, f, imax, jmax);
snprintf(filename, 30, "gtest-%d.dat", rank);
testWriteFile(filename, g, imax, jmax);
}
void commFinalize(Comm* c)
{
#ifdef _MPI
MPI_Finalize();
#endif
}

View File

@ -1,35 +1,48 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#ifndef __COMM_H_
#define __COMM_H_
#if defined(_MPI)
#include <mpi.h>
#endif
enum direction { LEFT = 0, RIGHT, BOTTOM, TOP, NDIRS };
enum dimension { JDIM = 0, IDIM, NDIMS };
enum dimension { IDIM = 0, JDIM, NDIMS };
enum cdimension { CJDIM = 0, CIDIM };
enum layer { HALO = 0, BULK };
enum op { MAX = 0, SUM };
typedef struct {
int rank;
int size;
#if defined(_MPI)
MPI_Comm comm;
MPI_Datatype sbufferTypes[NDIRS];
MPI_Datatype rbufferTypes[NDIRS];
MPI_Datatype bufferTypes[NDIRS];
MPI_Aint sdispls[NDIRS];
MPI_Aint rdispls[NDIRS];
#endif
int neighbours[NDIRS];
int coords[NDIMS], dims[NDIMS];
int imaxLocal, jmaxLocal;
} Comm;
extern void commInit(Comm* c, int jmax, int imax);
extern int sizeOfRank(int rank, int size, int N);
extern void commInit(Comm* c, int argc, char** argv);
extern void commTestInit(Comm* c, double* p, double* f, double* g);
extern void commTestWrite(Comm* c, double* p, double* f, double* g);
extern void commFinalize(Comm* c);
extern void commPartition(Comm* c, int jmax, int imax);
extern void commPrintConfig(Comm*);
extern void commExchange(Comm*, double*);
extern void commShift(Comm* c, double* f, double* g);
extern void commReduction(double* v, int op);
extern int commIsBoundary(Comm* c, int direction);
extern void commUpdateDatatypes(Comm*, Comm*, int, int);
extern void commFreeCommunicator(Comm*);
extern void commCollectResult(Comm* c,
double* ug,
double* vg,

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
@ -12,18 +12,11 @@
#include "allocate.h"
#include "comm.h"
#include "discretization.h"
#include "parameter.h"
#include "solver.h"
#include "util.h"
#define P(i, j) p[(j) * (imaxLocal + 2) + (i)]
#define F(i, j) f[(j) * (imaxLocal + 2) + (i)]
#define G(i, j) g[(j) * (imaxLocal + 2) + (i)]
#define U(i, j) u[(j) * (imaxLocal + 2) + (i)]
#define V(i, j) v[(j) * (imaxLocal + 2) + (i)]
#define RHS(i, j) rhs[(j) * (imaxLocal + 2) + (i)]
static void printConfig(Solver* s)
static void printConfig(Discretization* s)
{
if (commIsMaster(&s->comm)) {
printf("Parameters for #%s#\n", s->problem);
@ -35,38 +28,34 @@ static void printConfig(Solver* s)
printf("\tReynolds number: %.2f\n", s->re);
printf("\tGx Gy: %.2f %.2f\n", s->gx, s->gy);
printf("Geometry data:\n");
printf("\tDomain box size (x, y): %.2f, %.2f\n", s->xlength, s->ylength);
printf("\tCells (x, y): %d, %d\n", s->imax, s->jmax);
printf("\tCell size (dx, dy): %f, %f\n", s->dx, s->dy);
printf("\tDomain box size (x, y): %.2f, %.2f\n",
s->grid.xlength,
s->grid.ylength);
printf("\tCells (x, y): %d, %d\n", s->grid.imax, s->grid.jmax);
printf("\tCell size (dx, dy): %f, %f\n", s->grid.dx, s->grid.dy);
printf("Timestep parameters:\n");
printf("\tDefault stepsize: %.2f, Final time %.2f\n", s->dt, s->te);
printf("\tdt bound: %.6f\n", s->dtBound);
printf("\tTau factor: %.2f\n", s->tau);
printf("Iterative s parameters:\n");
printf("\tMax iterations: %d\n", s->itermax);
printf("\tepsilon (stopping tolerance) : %f\n", s->eps);
printf("\tgamma factor: %f\n", s->gamma);
printf("\tomega (SOR relaxation): %f\n", s->omega);
}
commPrintConfig(&s->comm);
}
void initSolver(Solver* s, Parameter* params)
void initDiscretiztion(Discretization* s, Parameter* params)
{
s->problem = params->name;
s->bcLeft = params->bcLeft;
s->bcRight = params->bcRight;
s->bcBottom = params->bcBottom;
s->bcTop = params->bcTop;
s->imax = params->imax;
s->jmax = params->jmax;
s->xlength = params->xlength;
s->ylength = params->ylength;
s->dx = params->xlength / params->imax;
s->dy = params->ylength / params->jmax;
s->eps = params->eps;
s->omega = params->omg;
s->itermax = params->itermax;
s->grid.imax = params->imax;
s->grid.jmax = params->jmax;
s->grid.xlength = params->xlength;
s->grid.ylength = params->ylength;
s->grid.dx = params->xlength / params->imax;
s->grid.dy = params->ylength / params->jmax;
s->re = params->re;
s->gx = params->gx;
s->gy = params->gy;
@ -75,8 +64,6 @@ void initSolver(Solver* s, Parameter* params)
s->tau = params->tau;
s->gamma = params->gamma;
commInit(&s->comm, s->jmax, s->imax);
/* allocate arrays */
int imaxLocal = s->comm.imaxLocal;
int jmaxLocal = s->comm.jmaxLocal;
@ -98,8 +85,8 @@ void initSolver(Solver* s, Parameter* params)
s->g[i] = 0.0;
}
double dx = s->dx;
double dy = s->dy;
double dx = s->grid.dx;
double dy = s->grid.dy;
double invSqrSum = 1.0 / (dx * dx) + 1.0 / (dy * dy);
s->dtBound = 0.5 * s->re * 1.0 / invSqrSum;
@ -108,12 +95,12 @@ void initSolver(Solver* s, Parameter* params)
#endif
}
void computeRHS(Solver* s)
void computeRHS(Discretization* s)
{
int imaxLocal = s->comm.imaxLocal;
int jmaxLocal = s->comm.jmaxLocal;
double idx = 1.0 / s->dx;
double idy = 1.0 / s->dy;
double idx = 1.0 / s->grid.dx;
double idy = 1.0 / s->grid.dy;
double idt = 1.0 / s->dt;
double* rhs = s->rhs;
double* f = s->f;
@ -129,88 +116,7 @@ void computeRHS(Solver* s)
}
}
int solve(Solver* s)
{
int imax = s->imax;
int jmax = s->jmax;
int imaxLocal = s->comm.imaxLocal;
int jmaxLocal = s->comm.jmaxLocal;
double eps = s->eps;
int itermax = s->itermax;
double dx2 = s->dx * s->dx;
double dy2 = s->dy * s->dy;
double idx2 = 1.0 / dx2;
double idy2 = 1.0 / dy2;
double factor = s->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
double* p = s->p;
double* rhs = s->rhs;
double epssq = eps * eps;
int it = 0;
double res = 1.0;
commExchange(&s->comm, p);
while ((res >= epssq) && (it < itermax)) {
res = 0.0;
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = 1; i < imaxLocal + 1; i++) {
double r = RHS(i, j) -
((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
(P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2);
P(i, j) -= (factor * r);
res += (r * r);
}
}
if (commIsBoundary(&s->comm, BOTTOM)) { // set bottom bc
for (int i = 1; i < imaxLocal + 1; i++) {
P(i, 0) = P(i, 1);
}
}
if (commIsBoundary(&s->comm, TOP)) { // set top bc
for (int i = 1; i < imaxLocal + 1; i++) {
P(i, jmaxLocal + 1) = P(i, jmaxLocal);
}
}
if (commIsBoundary(&s->comm, LEFT)) { // set left bc
for (int j = 1; j < jmaxLocal + 1; j++) {
P(0, j) = P(1, j);
}
}
if (commIsBoundary(&s->comm, RIGHT)) { // set right bc
for (int j = 1; j < jmaxLocal + 1; j++) {
P(imaxLocal + 1, j) = P(imaxLocal, j);
}
}
commReduction(&res, SUM);
res = res / (double)(imax * jmax);
#ifdef DEBUG
if (commIsMaster(&s->comm)) {
printf("%d Residuum: %e\n", it, res);
}
#endif
it++;
}
#ifdef VERBOSE
if (commIsMaster(&s->comm)) {
printf("Solver took %d iterations to reach %f\n", it, sqrt(res));
}
#endif
if (res < eps) {
return 0;
} else {
return 1;
}
}
static double maxElement(Solver* s, double* m)
static double maxElement(Discretization* s, double* m)
{
int imaxLocal = s->comm.imaxLocal;
int jmaxLocal = s->comm.jmaxLocal;
@ -225,11 +131,11 @@ static double maxElement(Solver* s, double* m)
return maxval;
}
void computeTimestep(Solver* s)
void computeTimestep(Discretization* s)
{
double dt = s->dtBound;
double dx = s->dx;
double dy = s->dy;
double dx = s->grid.dx;
double dy = s->grid.dy;
double umax = maxElement(s, s->u);
double vmax = maxElement(s, s->v);
@ -243,7 +149,7 @@ void computeTimestep(Solver* s)
s->dt = dt * s->tau;
}
void setBoundaryConditions(Solver* s)
void setBoundaryConditions(Discretization* s)
{
int imaxLocal = s->comm.imaxLocal;
int jmaxLocal = s->comm.jmaxLocal;
@ -351,7 +257,7 @@ void setBoundaryConditions(Solver* s)
}
}
void setSpecialBoundaryCondition(Solver* s)
void setSpecialBoundaryCondition(Discretization* s)
{
int imaxLocal = s->comm.imaxLocal;
int jmaxLocal = s->comm.jmaxLocal;
@ -365,25 +271,27 @@ void setSpecialBoundaryCondition(Solver* s)
}
} else if (strcmp(s->problem, "canal") == 0) {
if (commIsBoundary(&s->comm, LEFT)) {
double ylength = s->ylength;
double dy = s->dy;
int rest = s->jmax % s->comm.size;
int yc = s->comm.rank * (s->jmax / s->comm.size) + MIN(rest, s->comm.rank);
double ylength = s->grid.ylength;
double dy = s->grid.dy;
int rest = s->grid.jmax % s->comm.dims[JDIM];
int yc = s->comm.rank * (s->grid.jmax / s->comm.dims[JDIM]) +
MIN(rest, s->comm.rank);
double ys = dy * (yc + 0.5);
double y;
/* printf("RANK %d yc: %d ys: %f\n", solver->rank, yc, ys); */
// printf("RANK %d yc: %d ys: %f\n", s->comm.rank, yc, ys);
for (int j = 1; j < jmaxLocal + 1; j++) {
y = ys + dy * (j - 0.5);
U(0, j) = y * (ylength - y) * 4.0 / (ylength * ylength);
}
}
}
/* print(solver, solver->u); */
}
void computeFG(Solver* s)
void computeFG(Discretization* s)
{
double* u = s->u;
double* v = s->v;
@ -398,8 +306,8 @@ void computeFG(Solver* s)
double gamma = s->gamma;
double dt = s->dt;
double inverseRe = 1.0 / s->re;
double inverseDx = 1.0 / s->dx;
double inverseDy = 1.0 / s->dy;
double inverseDx = 1.0 / s->grid.dx;
double inverseDy = 1.0 / s->grid.dy;
double du2dx, dv2dy, duvdx, duvdy;
double du2dx2, du2dy2, dv2dx2, dv2dy2;
@ -475,7 +383,7 @@ void computeFG(Solver* s)
}
}
void adaptUV(Solver* s)
void adaptUV(Discretization* s)
{
int imaxLocal = s->comm.imaxLocal;
int jmaxLocal = s->comm.jmaxLocal;
@ -486,8 +394,8 @@ void adaptUV(Solver* s)
double* f = s->f;
double* g = s->g;
double factorX = s->dt / s->dx;
double factorY = s->dt / s->dy;
double factorX = s->dt / s->grid.dx;
double factorY = s->dt / s->grid.dy;
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = 1; i < imaxLocal + 1; i++) {
@ -497,12 +405,12 @@ void adaptUV(Solver* s)
}
}
void writeResult(Solver* s, double* u, double* v, double* p)
void writeResult(Discretization* s, double* u, double* v, double* p)
{
int imax = s->imax;
int jmax = s->jmax;
double dx = s->dx;
double dy = s->dy;
int imax = s->grid.imax;
int jmax = s->grid.jmax;
double dx = s->grid.dx;
double dy = s->grid.dy;
double x = 0.0, y = 0.0;
FILE* fp;
@ -513,11 +421,11 @@ void writeResult(Solver* s, double* u, double* v, double* p)
exit(EXIT_FAILURE);
}
for (int j = 1; j < jmax; j++) {
for (int j = 1; j <= jmax; j++) {
y = (double)(j - 0.5) * dy;
for (int i = 1; i < imax; i++) {
for (int i = 1; i <= imax; i++) {
x = (double)(i - 0.5) * dx;
fprintf(fp, "%.2f %.2f %f\n", x, y, p[j * (imax) + i]);
fprintf(fp, "%.2f %.2f %f\n", x, y, p[j * (imax + 2) + i]);
}
fprintf(fp, "\n");
}
@ -531,14 +439,14 @@ void writeResult(Solver* s, double* u, double* v, double* p)
exit(EXIT_FAILURE);
}
for (int j = 1; j < jmax; j++) {
for (int j = 1; j <= jmax; j++) {
y = dy * (j - 0.5);
for (int i = 1; i < imax; i++) {
for (int i = 1; i <= imax; i++) {
x = dx * (i - 0.5);
double vel_u = (u[j * (imax) + i] + u[j * (imax) + (i - 1)]) / 2.0;
double vel_v = (v[j * (imax) + i] + v[(j - 1) * (imax) + i]) / 2.0;
double len = sqrt((vel_u * vel_u) + (vel_v * vel_v));
fprintf(fp, "%.2f %.2f %f %f %f\n", x, y, vel_u, vel_v, len);
double velU = (u[j * (imax + 2) + i] + u[j * (imax + 2) + (i - 1)]) / 2.0;
double velV = (v[j * (imax + 2) + i] + v[(j - 1) * (imax + 2) + i]) / 2.0;
double len = sqrt((velU * velU) + (velV * velV));
fprintf(fp, "%.2f %.2f %f %f %f\n", x, y, velU, velV, len);
}
}

View File

@ -0,0 +1,43 @@
/*
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#ifndef __DISCRETIZATION_H_
#define __DISCRETIZATION_H_
#include "comm.h"
#include "grid.h"
#include "parameter.h"
enum BC { NOSLIP = 1, SLIP, OUTFLOW, PERIODIC };
typedef struct {
/* geometry and grid information */
Grid grid;
/* arrays */
double *p, *rhs;
double *f, *g;
double *u, *v;
/* parameters */
double re, tau, gamma;
double gx, gy;
/* time stepping */
double dt, te;
double dtBound;
char* problem;
int bcLeft, bcRight, bcBottom, bcTop;
/* communication */
Comm comm;
} Discretization;
void initDiscretiztion(Discretization*, Parameter*);
void computeRHS(Discretization*);
void normalizePressure(Discretization*);
void computeTimestep(Discretization*);
void setBoundaryConditions(Discretization*);
void setSpecialBoundaryCondition(Discretization*);
void computeFG(Discretization*);
void adaptUV(Discretization*);
void writeResult(Discretization* s, double* u, double* v, double* p);
#endif

View File

@ -0,0 +1,16 @@
/*
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#ifndef __GRID_H_
#define __GRID_H_
typedef struct {
double dx, dy;
int imax, jmax;
double xlength, ylength;
} Grid;
#endif // __GRID_H_

View File

@ -1,95 +1,121 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#include <float.h>
#include <limits.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include "allocate.h"
#include "comm.h"
#include "discretization.h"
#include "parameter.h"
#include "progress.h"
#include "solver.h"
#include "timing.h"
#include <mpi.h>
static void writeResults(Discretization* s)
{
#ifdef _MPI
size_t bytesize = (s->grid.imax + 2) * (s->grid.jmax + 2) * sizeof(double);
double* ug = allocate(64, bytesize);
double* vg = allocate(64, bytesize);
double* pg = allocate(64, bytesize);
commCollectResult(&s->comm, ug, vg, pg, s->u, s->v, s->p, s->grid.imax, s->grid.jmax);
if (commIsMaster(&s->comm)) {
writeResult(s, ug, vg, pg);
}
free(ug);
free(vg);
free(pg);
#else
writeResult(s, s->u, s->v, s->p);
#endif
}
int main(int argc, char** argv)
{
int rank;
double S, E;
Parameter params;
Solver solver;
double timeStart, timeStop;
Parameter p;
Discretization d;
Solver s;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
initParameter(&params);
commInit(&d.comm, argc, argv);
initParameter(&p);
FILE* fp;
if (commIsMaster(&d.comm)) fp = initResidualWriter();
if (argc != 2) {
printf("Usage: %s <configFile>\n", argv[0]);
exit(EXIT_SUCCESS);
}
readParameter(&params, argv[1]);
if (rank == 0) {
printParameter(&params);
readParameter(&p, argv[1]);
commPartition(&d.comm, p.jmax, p.imax);
if (commIsMaster(&d.comm)) {
printParameter(&p);
}
initSolver(&solver, &params);
/* debugExchange(&solver); */
/* exit(EXIT_SUCCESS); */
initProgress(solver.te);
double tau = solver.tau;
double te = solver.te;
initDiscretiztion(&d, &p);
initSolver(&s, &d, &p);
#ifdef TEST
commPrintConfig(&d.comm);
commTestInit(&d.comm, d.p, d.f, d.g);
commExchange(&d.comm, d.p);
commShift(&d.comm, d.f, d.g);
commTestWrite(&d.comm, d.p, d.f, d.g);
writeResults(&d);
commFinalize(&d.comm);
exit(EXIT_SUCCESS);
#endif
#ifndef VERBOSE
initProgress(d.te);
#endif
double tau = d.tau;
double te = d.te;
double t = 0.0;
double res = 0.0;
S = getTimeStamp();
timeStart = getTimeStamp();
while (t <= te) {
if (tau > 0.0) {
computeTimestep(&solver);
}
setBoundaryConditions(&solver);
setSpecialBoundaryCondition(&solver);
computeFG(&solver);
computeRHS(&solver);
solve(&solver);
adaptUV(&solver);
t += solver.dt;
if (tau > 0.0) computeTimestep(&d);
setBoundaryConditions(&d);
setSpecialBoundaryCondition(&d);
computeFG(&d);
computeRHS(&d);
res = solve(&s, d.p, d.rhs);
adaptUV(&d);
if (commIsMaster(&d.comm)) writeResidual(fp, t, res);
t += d.dt;
#ifdef VERBOSE
if (rank == 0) {
printf("TIME %f , TIMESTEP %f\n", t, solver.dt);
if (commIsMaster(s.comm)) {
printf("TIME %f , TIMESTEP %f\n", t, d.dt);
}
#else
printProgress(t);
#endif
}
E = getTimeStamp();
timeStop = getTimeStamp();
#ifndef VERBOSE
stopProgress();
if (rank == 0) {
printf("Solution took %.2fs\n", E - S);
#endif
if (commIsMaster(s.comm)) {
printf("Solution took %.2fs\n", timeStop - timeStart);
}
size_t bytesize = solver.imax * solver.jmax * sizeof(double);
double* ug = allocate(64, bytesize);
double* vg = allocate(64, bytesize);
double* pg = allocate(64, bytesize);
commCollectResult(&solver.comm,
ug,
vg,
pg,
solver.u,
solver.v,
solver.p,
solver.jmax,
solver.imax);
writeResult(&solver, ug, vg, pg);
MPI_Finalize();
if (commIsMaster(&d.comm)) fclose(fp);
writeResults(&d);
commFinalize(s.comm);
return EXIT_SUCCESS;
}

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
@ -21,6 +21,9 @@ void initParameter(Parameter* param)
param->itermax = 1000;
param->eps = 0.0001;
param->omg = 1.8;
param->levels = 5;
param->presmooth = 5;
param->postsmooth = 5;
}
void readParameter(Parameter* param, const char* filename)
@ -72,6 +75,9 @@ void readParameter(Parameter* param, const char* filename)
PARSE_INT(bcRight);
PARSE_INT(bcBottom);
PARSE_INT(bcTop);
PARSE_INT(levels);
PARSE_INT(presmooth);
PARSE_INT(postsmooth);
PARSE_REAL(u_init);
PARSE_REAL(v_init);
PARSE_REAL(p_init);

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
@ -18,6 +18,7 @@ typedef struct {
char* name;
int bcLeft, bcRight, bcBottom, bcTop;
double u_init, v_init, p_init;
int levels, presmooth, postsmooth;
} Parameter;
void initParameter(Parameter*);

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
@ -7,30 +7,24 @@
#include <math.h>
#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <stdlib.h>
#include "progress.h"
static double _end;
static int _current;
static int _rank = -1;
void initProgress(double end)
{
MPI_Comm_rank(MPI_COMM_WORLD, &_rank);
_end = end;
_current = 0;
if (_rank == 0) {
printf("[ ]");
fflush(stdout);
}
}
void printProgress(double current)
{
if (_rank == 0) {
int new = (int)rint((current / _end) * 10.0);
if (new > _current) {
@ -48,13 +42,29 @@ void printProgress(double current)
printf("\r[%s]", progress);
}
fflush(stdout);
}
}
void stopProgress()
{
if (_rank == 0) {
printf("\n");
fflush(stdout);
}
}
FILE* initResidualWriter()
{
FILE* fp;
fp = fopen("residual.dat", "w");
if (fp == NULL) {
printf("Error!\n");
exit(EXIT_FAILURE);
}
return fp;
}
void writeResidual(FILE* fp, double ts, double res)
{
fprintf(fp, "%f, %f\n", ts, res);
}

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
@ -10,5 +10,6 @@
extern void initProgress(double);
extern void printProgress(double);
extern void stopProgress();
extern FILE* initResidualWriter(void);
extern void writeResidual(FILE*, double, double);
#endif

View File

@ -0,0 +1,302 @@
/*
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <stdio.h>
#include <stdlib.h>
#include "allocate.h"
#include "solver.h"
#include "util.h"
#define FINEST_LEVEL 0
#define COARSEST_LEVEL (s->levels - 1)
// #define S(i, j) s[(j) * (imaxLocal + 2) + (i)]
#define E(i, j) e[(j) * (imaxLocal + 2) + (i)]
#define R(i, j) r[(j) * (imaxLocal + 2) + (i)]
#define OLD(i, j) old[(j) * (imaxLocal + 2) + (i)]
static void restrictMG(Solver* s, int level, Comm* comm)
{
int imaxLocal = comm->imaxLocal;
int jmaxLocal = comm->jmaxLocal;
double* r = s->r[level + 1];
double* old = s->r[level];
#ifdef _MPI
commExchange(comm, old);
#endif
for (int j = 1; j < (jmaxLocal / 2) + 1; j++) {
for (int i = 1; i < (imaxLocal / 2) + 1; i++) {
R(i, j) = (OLD(2 * i - 1, 2 * j - 1) + OLD(2 * i, 2 * j - 1) * 2 +
OLD(2 * i + 1, 2 * j - 1) + OLD(2 * i - 1, 2 * j) * 2 +
OLD(2 * i, 2 * j) * 4 + OLD(2 * i + 1, 2 * j) * 2 +
OLD(2 * i - 1, 2 * j + 1) + OLD(2 * i, 2 * j + 1) * 2 +
OLD(2 * i + 1, 2 * j + 1)) /
16.0;
}
}
}
static void prolongate(Solver* s, int level, Comm* comm)
{
int imaxLocal = comm->imaxLocal;
int jmaxLocal = comm->jmaxLocal;
double* old = s->r[level + 1];
double* e = s->r[level];
for (int j = 2; j < jmaxLocal + 1; j += 2) {
for (int i = 2; i < imaxLocal + 1; i += 2) {
E(i, j) = OLD(i / 2, j / 2);
}
}
}
static void correct(Solver* s, double* p, int level, Comm* comm)
{
double* e = s->e[level];
int imaxLocal = comm->imaxLocal;
int jmaxLocal = comm->jmaxLocal;
for (int j = 1; j < jmaxLocal + 1; ++j) {
for (int i = 1; i < imaxLocal + 1; ++i) {
P(i, j) += E(i, j);
}
}
}
static void setBoundaryCondition(Solver* s, double* p, int imaxLocal, int jmaxLocal)
{
#ifdef _MPI
if (commIsBoundary(s->comm, BOTTOM)) { // set bottom bc
for (int i = 1; i < imaxLocal + 1; i++) {
P(i, 0) = P(i, 1);
}
}
if (commIsBoundary(s->comm, TOP)) { // set top bc
for (int i = 1; i < imaxLocal + 1; i++) {
P(i, jmaxLocal + 1) = P(i, jmaxLocal);
}
}
if (commIsBoundary(s->comm, LEFT)) { // set left bc
for (int j = 1; j < jmaxLocal + 1; j++) {
P(0, j) = P(1, j);
}
}
if (commIsBoundary(s->comm, RIGHT)) { // set right bc
for (int j = 1; j < jmaxLocal + 1; j++) {
P(imaxLocal + 1, j) = P(imaxLocal, j);
}
}
#else
for (int i = 1; i < imaxLocal + 1; i++) {
P(i, 0) = P(i, 1);
P(i, jmaxLocal + 1) = P(i, jmaxLocal);
}
for (int j = 1; j < jmaxLocal + 1; j++) {
P(0, j) = P(1, j);
P(imaxLocal + 1, j) = P(imaxLocal, j);
}
#endif
}
static double smooth(Solver* s, double* p, double* rhs, int level, Comm* comm)
{
int imaxLocal = comm->imaxLocal;
int jmaxLocal = comm->jmaxLocal;
int imax = s->grid->imax;
int jmax = s->grid->jmax;
double dx2 = s->grid->dx * s->grid->dx;
double dy2 = s->grid->dy * s->grid->dy;
double idx2 = 1.0 / dx2;
double idy2 = 1.0 / dy2;
double factor = s->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
double* r = s->r[level];
double res = 1.0;
int pass, jsw, isw;
jsw = 1;
for (pass = 0; pass < 2; pass++) {
isw = jsw;
#ifdef _MPI
commExchange(comm, p);
#endif
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = isw; i < imaxLocal + 1; i += 2) {
P(i, j) -= factor *
(RHS(i, j) -
((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
(P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2));
}
isw = 3 - isw;
}
jsw = 3 - jsw;
}
}
static double calculateResidual(Solver* s, double* p, double* rhs, int level, Comm* comm)
{
int imax = s->grid->imax;
int jmax = s->grid->jmax;
int imaxLocal = comm->imaxLocal;
int jmaxLocal = comm->jmaxLocal;
double dx2 = s->grid->dx * s->grid->dx;
double dy2 = s->grid->dy * s->grid->dy;
double idx2 = 1.0 / dx2;
double idy2 = 1.0 / dy2;
double factor = s->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
double* r = s->r[level];
double res = 1.0;
int pass, jsw, isw;
jsw = 1;
for (pass = 0; pass < 2; pass++) {
isw = jsw;
#ifdef _MPI
commExchange(comm, p);
#endif
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = isw; i < imaxLocal + 1; i += 2) {
R(i, j) = RHS(i, j) -
((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
(P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2);
res += (R(i, j) * R(i, j));
}
isw = 3 - isw;
}
jsw = 3 - jsw;
}
#ifdef _MPI
commReduction(&res, SUM);
#endif
res = res / (double)(imax * jmax);
#ifdef DEBUG
if (commIsMaster(s->comm)) {
printf("%d Residuum: %e\n", it, res);
}
#endif
return res;
}
static double multiGrid(Solver* s, double* p, double* rhs, int level, Comm* comm)
{
double res = 0.0;
// coarsest level
if (level == COARSEST_LEVEL) {
for (int i = 0; i < 5; i++) {
smooth(s, p, rhs, level, comm);
}
return res;
}
// pre-smoothing
for (int i = 0; i < s->presmooth; i++) {
smooth(s, p, rhs, level, comm);
if (level == FINEST_LEVEL)
setBoundaryCondition(s, p, comm->imaxLocal, comm->jmaxLocal);
}
// calculate residuals
res = calculateResidual(s, p, rhs, level, comm);
// restrict
restrictMG(s, level, comm);
Comm newcomm;
commUpdateDatatypes(s->comm, &newcomm, comm->imaxLocal, comm->jmaxLocal);
// MGSolver on residual and error.
multiGrid(s, s->e[level + 1], s->r[level + 1], level + 1, &newcomm);
commFreeCommunicator(&newcomm);
// prolongate
prolongate(s, level, comm);
// correct p on finer level using residual
correct(s, p, level, comm);
if (level == FINEST_LEVEL)
setBoundaryCondition(s, p, comm->imaxLocal, comm->jmaxLocal);
// post-smoothing
for (int i = 0; i < s->postsmooth; i++) {
smooth(s, p, rhs, level, comm);
if (level == FINEST_LEVEL)
setBoundaryCondition(s, p, comm->imaxLocal, comm->jmaxLocal);
}
return res;
}
void initSolver(Solver* s, Discretization* d, Parameter* p)
{
s->eps = p->eps;
s->omega = p->omg;
s->itermax = p->itermax;
s->levels = p->levels;
s->grid = &d->grid;
s->comm = &d->comm;
s->presmooth = p->presmooth;
s->postsmooth = p->postsmooth;
int imax = s->grid->imax;
int jmax = s->grid->jmax;
int levels = s->levels;
printf("Using Multigrid solver with %d levels\n", levels);
s->r = malloc(levels * sizeof(double*));
s->e = malloc(levels * sizeof(double*));
size_t size = (imax + 2) * (jmax + 2) * sizeof(double);
for (int j = 0; j < levels; j++) {
s->r[j] = allocate(64, size);
s->e[j] = allocate(64, size);
for (int i = 0; i < (imax + 2) * (jmax + 2); i++) {
s->r[j][i] = 0.0;
s->e[j][i] = 0.0;
}
}
}
double solve(Solver* s, double* p, double* rhs)
{
double res = multiGrid(s, p, rhs, 0, s->comm);
#ifdef VERBOSE
if (commIsMaster(s->comm)) {
printf("Residuum: %.6f\n", res);
}
#endif
return res;
}

View File

@ -0,0 +1,106 @@
/*
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <math.h>
#include <stdio.h>
#include "allocate.h"
#include "comm.h"
#include "discretization.h"
#include "parameter.h"
#include "solver.h"
#include "util.h"
void initSolver(Solver* s, Discretization* d, Parameter* p)
{
s->grid = &d->grid;
s->eps = p->eps;
s->omega = p->omg;
s->itermax = p->itermax;
s->comm = &d->comm;
}
double solve(Solver* s, double* p, double* rhs)
{
int imax = s->grid->imax;
int jmax = s->grid->jmax;
int imaxLocal = s->comm->imaxLocal;
int jmaxLocal = s->comm->jmaxLocal;
double eps = s->eps;
int itermax = s->itermax;
double dx2 = s->grid->dx * s->grid->dx;
double dy2 = s->grid->dy * s->grid->dy;
double idx2 = 1.0 / dx2;
double idy2 = 1.0 / dy2;
double factor = s->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
double epssq = eps * eps;
int pass, jsw, isw;
int it = 0;
double res = 1.0;
while ((res >= epssq) && (it < itermax)) {
jsw = 1;
for (pass = 0; pass < 2; pass++) {
isw = jsw;
commExchange(s->comm, p);
for (int j = 1; j < jmaxLocal + 1; j++) {
for (int i = isw; i < imaxLocal + 1; i += 2) {
double r = RHS(i, j) -
((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
(P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2);
P(i, j) -= (factor * r);
res += (r * r);
}
isw = 3 - isw;
}
jsw = 3 - jsw;
}
if (commIsBoundary(s->comm, BOTTOM)) { // set bottom bc
for (int i = 1; i < imaxLocal + 1; i++) {
P(i, 0) = P(i, 1);
}
}
if (commIsBoundary(s->comm, TOP)) { // set top bc
for (int i = 1; i < imaxLocal + 1; i++) {
P(i, jmaxLocal + 1) = P(i, jmaxLocal);
}
}
if (commIsBoundary(s->comm, LEFT)) { // set left bc
for (int j = 1; j < jmaxLocal + 1; j++) {
P(0, j) = P(1, j);
}
}
if (commIsBoundary(s->comm, RIGHT)) { // set right bc
for (int j = 1; j < jmaxLocal + 1; j++) {
P(imaxLocal + 1, j) = P(imaxLocal, j);
}
}
commReduction(&res, SUM);
res = res / (double)(imax * jmax);
#ifdef DEBUG
if (commIsMaster(s->comm)) {
printf("%d Residuum: %e\n", it, res);
}
#endif
it++;
}
#ifdef VERBOSE
if (commIsMaster(s->comm)) {
printf("Solver took %d iterations to reach %f\n", it, sqrt(res));
}
#endif
return res;
}

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
@ -7,41 +7,23 @@
#ifndef __SOLVER_H_
#define __SOLVER_H_
#include "comm.h"
#include "discretization.h"
#include "grid.h"
#include "mpi.h"
#include "parameter.h"
enum BC { NOSLIP = 1, SLIP, OUTFLOW, PERIODIC };
typedef struct {
/* geometry and grid information */
double dx, dy;
int imax, jmax;
double xlength, ylength;
/* arrays */
double *p, *rhs;
double *f, *g;
double *u, *v;
Grid* grid;
/* parameters */
double eps, omega;
double re, tau, gamma;
double gx, gy;
/* time stepping */
int itermax;
double dt, te;
double dtBound;
char* problem;
int bcLeft, bcRight, bcBottom, bcTop;
int levels, presmooth, postsmooth;
double **r, **e;
/* communication */
Comm comm;
Comm* comm;
} Solver;
void initSolver(Solver*, Parameter*);
void computeRHS(Solver*);
int solve(Solver*);
void normalizePressure(Solver*);
void computeTimestep(Solver*);
void setBoundaryConditions(Solver*);
void setSpecialBoundaryCondition(Solver*);
void computeFG(Solver*);
void adaptUV(Solver*);
void writeResult(Solver* s, double* u, double* v, double* p);
void initSolver(Solver*, Discretization*, Parameter*);
double solve(Solver*, double*, double*);
#endif

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
@ -7,18 +7,16 @@
#include <stdlib.h>
#include <time.h>
double getTimeStamp()
double getTimeStamp(void)
{
struct timespec ts;
clock_gettime(CLOCK_MONOTONIC, &ts);
return (double)ts.tv_sec + (double)ts.tv_nsec * 1.e-9;
}
double getTimeResolution()
double getTimeResolution(void)
{
struct timespec ts;
clock_getres(CLOCK_MONOTONIC, &ts);
return (double)ts.tv_sec + (double)ts.tv_nsec * 1.e-9;
}
double getTimeStamp_() { return getTimeStamp(); }

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
@ -7,8 +7,7 @@
#ifndef __TIMING_H_
#define __TIMING_H_
extern double getTimeStamp();
extern double getTimeResolution();
extern double getTimeStamp_();
extern double getTimeStamp(void);
extern double getTimeResolution(void);
#endif // __TIMING_H_

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
@ -19,4 +19,11 @@
#define ABS(a) ((a) >= 0 ? (a) : -(a))
#endif
#define P(i, j) p[(j) * (imaxLocal + 2) + (i)]
#define F(i, j) f[(j) * (imaxLocal + 2) + (i)]
#define G(i, j) g[(j) * (imaxLocal + 2) + (i)]
#define U(i, j) u[(j) * (imaxLocal + 2) + (i)]
#define V(i, j) v[(j) * (imaxLocal + 2) + (i)]
#define RHS(i, j) rhs[(j) * (imaxLocal + 2) + (i)]
#endif // __UTIL_H_

Binary file not shown.

Before

Width:  |  Height:  |  Size: 9.1 KiB

View File

@ -1,61 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#ifdef __linux__
#ifdef _OPENMP
#include <pthread.h>
#include <sched.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/syscall.h>
#include <sys/types.h>
#include <unistd.h>
#define MAX_NUM_THREADS 128
#define gettid() syscall(SYS_gettid)
static int getProcessorID(cpu_set_t* cpu_set)
{
int processorId;
for (processorId = 0; processorId < MAX_NUM_THREADS; processorId++) {
if (CPU_ISSET(processorId, cpu_set)) {
break;
}
}
return processorId;
}
int affinity_getProcessorId()
{
cpu_set_t cpu_set;
CPU_ZERO(&cpu_set);
sched_getaffinity(gettid(), sizeof(cpu_set_t), &cpu_set);
return getProcessorID(&cpu_set);
}
void affinity_pinThread(int processorId)
{
cpu_set_t cpuset;
pthread_t thread;
thread = pthread_self();
CPU_ZERO(&cpuset);
CPU_SET(processorId, &cpuset);
pthread_setaffinity_np(thread, sizeof(cpu_set_t), &cpuset);
}
void affinity_pinProcess(int processorId)
{
cpu_set_t cpuset;
CPU_ZERO(&cpuset);
CPU_SET(processorId, &cpuset);
sched_setaffinity(0, sizeof(cpu_set_t), &cpuset);
}
#endif /*_OPENMP*/
#endif /*__linux__*/

View File

@ -1,14 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#ifndef AFFINITY_H
#define AFFINITY_H
extern int affinity_getProcessorId();
extern void affinity_pinProcess(int);
extern void affinity_pinThread(int);
#endif /*AFFINITY_H*/

View File

@ -1,54 +0,0 @@
/*
* =======================================================================================
*
* Author: Jan Eitzinger (je), jan.eitzinger@fau.de
* Copyright (c) 2020 RRZE, University Erlangen-Nuremberg
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included
* in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*
* =======================================================================================
*/
#ifndef LIKWID_MARKERS_H
#define LIKWID_MARKERS_H
#ifdef LIKWID_PERFMON
#include <likwid.h>
#define LIKWID_MARKER_INIT likwid_markerInit()
#define LIKWID_MARKER_THREADINIT likwid_markerThreadInit()
#define LIKWID_MARKER_SWITCH likwid_markerNextGroup()
#define LIKWID_MARKER_REGISTER(regionTag) likwid_markerRegisterRegion(regionTag)
#define LIKWID_MARKER_START(regionTag) likwid_markerStartRegion(regionTag)
#define LIKWID_MARKER_STOP(regionTag) likwid_markerStopRegion(regionTag)
#define LIKWID_MARKER_CLOSE likwid_markerClose()
#define LIKWID_MARKER_RESET(regionTag) likwid_markerResetRegion(regionTag)
#define LIKWID_MARKER_GET(regionTag, nevents, events, time, count) \
likwid_markerGetRegion(regionTag, nevents, events, time, count)
#else /* LIKWID_PERFMON */
#define LIKWID_MARKER_INIT
#define LIKWID_MARKER_THREADINIT
#define LIKWID_MARKER_SWITCH
#define LIKWID_MARKER_REGISTER(regionTag)
#define LIKWID_MARKER_START(regionTag)
#define LIKWID_MARKER_STOP(regionTag)
#define LIKWID_MARKER_CLOSE
#define LIKWID_MARKER_GET(regionTag, nevents, events, time, count)
#define LIKWID_MARKER_RESET(regionTag)
#endif /* LIKWID_PERFMON */
#endif /*LIKWID_MARKERS_H*/

View File

@ -1,71 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#include <float.h>
#include <limits.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include "parameter.h"
#include "progress.h"
#include "solver.h"
#include "timing.h"
#include "trace.h"
int main(int argc, char** argv)
{
double timeStart, timeEnd;
Parameter p;
Solver s;
Tracing t;
initParameter(&p);
if (argc != 2) {
printf("Usage: %s <configFile>\n", argv[0]);
exit(EXIT_SUCCESS);
}
readParameter(&p, argv[1]);
printParameter(&p);
initSolver(&s, &p);
initTrace(&t, &p);
#ifndef VERBOSE
initProgress(s.te);
#endif
double tau = s.tau;
double te = s.te;
double time = 0.0;
int nt = 0;
timeStart = getTimeStamp();
while (time <= te) {
if (tau > 0.0) computeTimestep(&s);
setBoundaryConditions(&s);
setSpecialBoundaryCondition(&s);
computeFG(&s);
computeRHS(&s);
if (nt % 100 == 0) normalizePressure(&s);
solve(&s);
adaptUV(&s);
time += s.dt;
nt++;
trace(&t, s.u, s.v, time);
#ifdef VERBOSE
printf("TIME %f , TIMESTEP %f\n", time, s.dt);
#else
printProgress(time);
#endif
}
timeEnd = getTimeStamp();
stopProgress();
printf("Solution took %.2fs\n", timeEnd - timeStart);
writeResult(&s);
return EXIT_SUCCESS;
}

View File

@ -1,47 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#ifndef __SOLVER_H_
#define __SOLVER_H_
#include "parameter.h"
#define U(i, j) u[(j) * (imax + 2) + (i)]
#define V(i, j) v[(j) * (imax + 2) + (i)]
enum BC { NOSLIP = 1, SLIP, OUTFLOW, PERIODIC };
typedef struct {
/* geometry and grid information */
double dx, dy;
int imax, jmax;
double xlength, ylength;
/* arrays */
double *p, *rhs;
double *f, *g;
double *u, *v;
/* parameters */
double eps, omega;
double re, tau, gamma;
double gx, gy;
/* time stepping */
int itermax;
double dt, te;
double dtBound;
char* problem;
int bcLeft, bcRight, bcBottom, bcTop;
} Solver;
void initSolver(Solver*, Parameter*);
void computeRHS(Solver*);
void solve(Solver*);
void normalizePressure(Solver*);
void computeTimestep(Solver*);
void setBoundaryConditions(Solver*);
void setSpecialBoundaryCondition(Solver*);
void computeFG(Solver*);
void adaptUV(Solver*);
void writeResult(Solver*);
#endif

View File

@ -1,208 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "trace.h"
#define U(i, j) u[(j) * (imax + 2) + (i)]
#define V(i, j) v[(j) * (imax + 2) + (i)]
static int ts = 0;
static void printState(Tracing* t)
{
printf("Cursor: %d Total particles: %d\n", t->cursor, t->totalParticles);
}
static void advanceParticles(
Tracing* t, double delt, double* restrict u, double* restrict v)
{
double delx = t->grid.dx;
double dely = t->grid.dy;
double* m = t->memorypool;
int* p = t->particles;
int imax = t->grid.imax;
int jmax = t->grid.jmax;
for (int i = 0; i < t->totalParticles; i++) {
int particleId = p[i];
double x = m[particleId * NCOORD + X];
double y = m[particleId * NCOORD + Y];
// printf("P%d - X %f Y %f\n", i, x, y);
// Interpolate U
int iCoord = (int)(x / delx) + 1;
int jCoord = (int)((y + 0.5 * dely) / dely) + 1;
double x1 = (double)(iCoord - 1) * delx;
double y1 = ((double)(jCoord - 1) - 0.5) * dely;
double x2 = (double)iCoord * delx;
double y2 = ((double)jCoord - 0.5) * dely;
// printf("U - iCoord %d jCoord %d\n", iCoord, jCoord);
double un = (1.0 / (delx * dely)) *
((x2 - x) * (y2 - y) * U(iCoord - 1, jCoord - 1) +
(x - x1) * (y2 - y) * U(iCoord, jCoord - 1) +
(x2 - x) * (y - y1) * U(iCoord - 1, jCoord) +
(x - x1) * (y - y1) * U(iCoord, jCoord));
double xn = x + delt * un;
m[particleId * NCOORD + X] = xn;
// Interpolate V
iCoord = (int)((x + 0.5 * delx) / delx) + 1;
jCoord = (int)(y / dely) + 1;
x1 = ((double)(iCoord - 1) - 0.5) * delx;
y1 = (double)(jCoord - 1) * dely;
x2 = ((double)iCoord - 0.5) * delx;
y2 = (double)jCoord * dely;
// printf("V - iCoord %d jCoord %d\n", iCoord, jCoord);
double vn = (1.0 / (delx * dely)) *
((x2 - x) * (y2 - y) * V(iCoord - 1, jCoord - 1) +
(x - x1) * (y2 - y) * V(iCoord, jCoord - 1) +
(x2 - x) * (y - y1) * V(iCoord - 1, jCoord) +
(x - x1) * (y - y1) * V(iCoord, jCoord));
double yn = y + delt * vn;
m[particleId * NCOORD + Y] = yn;
printf("P%i VEL %f %f dt %f OP %f %f NP %f %f\n", i, un, vn, delt, x, y, xn, yn);
}
double xlength = t->grid.xlength;
double ylength = t->grid.ylength;
int cntNew = 0;
int tmp[t->totalParticles];
// Check for particles to remove
for (int i = 0; i < t->totalParticles; i++) {
int particleId = p[i];
double x = m[particleId * NCOORD + X];
double y = m[particleId * NCOORD + Y];
if (!((x < 0.0) || (x > xlength) || (y < 0.0) || (y > ylength))) {
tmp[cntNew++] = i;
}
}
t->totalParticles = cntNew;
memcpy(t->particles, tmp, cntNew * sizeof(int));
}
static void injectParticles(Tracing* t)
{
double* line = t->line;
double* m = t->memorypool;
for (int i = 0; i < t->numParticles; i++) {
printf("Inject %d as %d mem %d\n", i, t->totalParticles, t->cursor);
t->particles[t->totalParticles] = t->cursor;
m[(t->cursor) * NCOORD + X] = line[i * NCOORD + X];
m[(t->cursor) * NCOORD + Y] = line[i * NCOORD + Y];
t->cursor++;
t->totalParticles++;
}
}
static void writeParticles(Tracing* t)
{
FILE* fp;
double* m = t->memorypool;
int* p = t->particles;
char filename[50];
snprintf(filename, 50, "particles_%d.dat", ts++);
fp = fopen(filename, "w");
if (fp == NULL) {
printf("Error!\n");
exit(EXIT_FAILURE);
}
for (int i = 0; i < t->totalParticles; i++) {
int particleId = p[i];
double x = m[particleId * NCOORD + X];
double y = m[particleId * NCOORD + Y];
fprintf(fp, "%f %f\n", x, y);
}
fclose(fp);
}
void trace(Tracing* t, double* restrict u, double* restrict v, double time)
{
if (time >= t->traceStart) {
if ((time - t->lastUpdate[INJECT]) > t->traceInject) {
printf("Inject at %f\n", time);
printState(t);
injectParticles(t);
t->lastUpdate[INJECT] = time;
}
if ((time - t->lastUpdate[WRITE]) > t->traceWrite) {
printf("Write at %f\n", time);
writeParticles(t);
t->lastUpdate[WRITE] = time;
}
advanceParticles(t, time - t->lastUpdate[ADVANCE], u, v);
t->lastUpdate[ADVANCE] = time;
}
}
void initTrace(Tracing* t, Parameter* p)
{
size_t numParticles = p->nparticles;
size_t totalParticles = (size_t)(p->te - p->traceStart) / (size_t)p->traceInject;
totalParticles += 2;
totalParticles *= numParticles;
double x1 = p->lineX1;
double y1 = p->lineY1;
double x2 = p->lineX2;
double y2 = p->lineY2;
for (int i = 0; i < NUMTIMERS; i++) {
t->lastUpdate[i] = p->traceStart;
}
t->grid.imax = p->imax;
t->grid.jmax = p->jmax;
t->grid.xlength = p->xlength;
t->grid.ylength = p->ylength;
t->grid.dx = p->xlength / p->imax;
t->grid.dy = p->ylength / p->jmax;
t->numParticles = numParticles;
t->totalParticles = 0;
t->cursor = 0;
t->traceStart = p->traceStart;
t->traceWrite = p->traceWrite;
t->traceInject = p->traceInject;
t->particles = (int*)malloc(totalParticles * sizeof(int));
t->memorypool = (double*)malloc(totalParticles * NCOORD * sizeof(double));
t->line = (double*)malloc(numParticles * NCOORD * sizeof(double));
double* line = t->line;
for (int i = 0; i < numParticles; i++) {
double spacing = (double)i / (double)(numParticles - 1);
double x = spacing * x1 + (1.0 - spacing) * x2;
double y = spacing * y1 + (1.0 - spacing) * y2;
printf("S: %f x: %f y: %f\n", spacing, x, y);
line[i * NCOORD + X] = x;
line[i * NCOORD + Y] = y;
}
}
void freeTrace(Tracing* t) { free(t->line); }

View File

@ -1,32 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#ifndef __TRACE_H_
#define __TRACE_H_
#include "grid.h"
#include "parameter.h"
typedef enum COORD { X = 0, Y, NCOORD } COORD;
typedef enum { ADVANCE = 0, INJECT, WRITE, NUMTIMERS } TIMER;
typedef struct Tracing {
double traceStart;
double traceWrite;
double traceInject;
double dt;
double lastUpdate[NUMTIMERS];
double* memorypool;
double* line;
int cursor;
int* particles;
int numParticles;
int totalParticles;
Grid grid;
} Tracing;
extern void initTrace(Tracing* t, Parameter* p);
extern void trace(Tracing* t, double* u, double* v, double time);
#endif

View File

@ -1,23 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#ifndef __UTIL_H_
#define __UTIL_H_
#define HLINE \
"------------------------------------------------------------------------" \
"----\n"
#ifndef MIN
#define MIN(x, y) ((x) < (y) ? (x) : (y))
#endif
#ifndef MAX
#define MAX(x, y) ((x) > (y) ? (x) : (y))
#endif
#ifndef ABS
#define ABS(a) ((a) >= 0 ? (a) : -(a))
#endif
#endif // __UTIL_H_

View File

@ -1,7 +0,0 @@
set terminal png size 1024,768 enhanced font ,12
set output 'p.png'
set datafile separator whitespace
set grid
set hidden3d
splot 'pressure.dat' using 1:2:3 with lines

View File

@ -1,5 +1,5 @@
#=======================================================================================
# Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
# Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
# All rights reserved.
# Use of this source code is governed by a MIT-style
# license that can be found in the LICENSE file.
@ -18,9 +18,10 @@ include $(MAKE_DIR)/include_$(TAG).mk
INCLUDES += -I$(SRC_DIR) -I$(BUILD_DIR)
VPATH = $(SRC_DIR)
SRC = $(wildcard $(SRC_DIR)/*.c)
SRC = $(filter-out $(wildcard $(SRC_DIR)/*-*.c),$(wildcard $(SRC_DIR)/*.c))
ASM = $(patsubst $(SRC_DIR)/%.c, $(BUILD_DIR)/%.s, $(SRC))
OBJ = $(patsubst $(SRC_DIR)/%.c, $(BUILD_DIR)/%.o, $(SRC))
OBJ += $(BUILD_DIR)/solver-$(SOLVER).o
SOURCES = $(SRC) $(wildcard $(SRC_DIR)/*.h)
CPPFLAGS := $(CPPFLAGS) $(DEFINES) $(OPTIONS) $(INCLUDES)
@ -37,9 +38,22 @@ $(BUILD_DIR)/%.s: %.c
$(info ===> GENERATE ASM $@)
$(CC) -S $(CPPFLAGS) $(CFLAGS) $< -o $@
.PHONY: clean distclean tags info asm format
.PHONY: clean distclean vis vis_clean tags info asm format
clean:
vis:
$(info ===> GENERATE VISUALIZATION)
@gnuplot -e "filename='pressure.dat'" ./surface.plot
@gnuplot -e "filename='velocity.dat'" ./vector.plot
@gnuplot -e "filename='residual.dat'" ./residual.plot
vis_clean:
$(info ===> CLEAN VISUALIZATION)
@rm -f *.dat
@rm -f *.png
@rm -f ./vis_files/*.dat
@rm -f ./vis_files/*.gif
clean: vis_clean
$(info ===> CLEAN)
@rm -rf $(BUILD_DIR)
@rm -f tags
@ -47,6 +61,8 @@ clean:
distclean: clean
$(info ===> DIST CLEAN)
@rm -f $(TARGET)
@rm -f *.dat
@rm -f *.png
info:
$(info $(CFLAGS))

View File

@ -36,6 +36,13 @@ te 100.0 # final time
dt 0.02 # time stepsize
tau 0.5 # safety factor for time stepsize control (<0 constant delt)
# Multigrid data:
# ---------
levels 3 # Multigrid levels
presmooth 5 # Pre-smoothning iterations
postsmooth 5 # Post-smoothning iterations
# Pressure Iteration Data:
# -----------------------

View File

@ -1,12 +1,12 @@
# Supported: GCC, CLANG, ICC
TAG ?= CLANG
TAG ?= ICC
ENABLE_OPENMP ?= false
# Supported: sor, rb, mg
SOLVER ?= mg
# Run in debug settings
DEBUG ?= false
#Feature options
OPTIONS += -DARRAY_ALIGNMENT=64
OPTIONS += -DVERBOSE
#OPTIONS += -DDEBUG
#OPTIONS += -DBOUNDCHECK
#OPTIONS += -DVERBOSE_AFFINITY
#OPTIONS += -DVERBOSE_DATASIZE
#OPTIONS += -DVERBOSE_TIMER

View File

@ -15,7 +15,7 @@ bcRight 1 #
gx 0.0 # Body forces (e.g. gravity)
gy 0.0 #
re 10.0 # Reynolds number
re 100.0 # Reynolds number
u_init 0.0 # initial value for velocity in x-direction
v_init 0.0 # initial value for velocity in y-direction
@ -26,8 +26,8 @@ p_init 0.0 # initial value for pressure
xlength 1.0 # domain size in x-direction
ylength 1.0 # domain size in y-direction
imax 40 # number of interior cells in x-direction
jmax 40 # number of interior cells in y-direction
imax 128 # number of interior cells in x-direction
jmax 128 # number of interior cells in y-direction
# Time Data:
# ---------
@ -36,11 +36,19 @@ te 10.0 # final time
dt 0.02 # time stepsize
tau 0.5 # safety factor for time stepsize control (<0 constant delt)
# Pressure Iteration Data:
# Multigrid data:
# ---------
levels 2 # Multigrid levels
presmooth 20 # Pre-smoothning iterations
postsmooth 5 # Post-smoothning iterations
# Solver Data:
# -----------------------
itermax 1000 # maximal number of pressure iteration in one time step
eps 0.001 # stopping tolerance for pressure iteration
rho 0.5
omg 1.7 # relaxation parameter for SOR iteration
gamma 0.9 # upwind differencing factor gamma
#===============================================================================

View File

@ -2,16 +2,18 @@ CC = clang
GCC = cc
LINKER = $(CC)
ifeq ($(ENABLE_OPENMP),true)
ifeq ($(strip $(ENABLE_OPENMP)),true)
OPENMP = -fopenmp
#OPENMP = -Xpreprocessor -fopenmp #required on Macos with homebrew libomp
LIBS = # -lomp
endif
ifeq ($(strip $(DEBUG)),true)
CFLAGS = -O0 -g -std=c17
else
CFLAGS = -O3 -std=c17 $(OPENMP)
endif
VERSION = --version
# CFLAGS = -O3 -std=c17 $(OPENMP)
CFLAGS = -Ofast -std=c17
#CFLAGS = -Ofast -fnt-store=aggressive -std=c99 $(OPENMP) #AMD CLANG
LFLAGS = $(OPENMP) -lm
DEFINES = -D_GNU_SOURCE# -DDEBUG
DEFINES = -D_GNU_SOURCE
INCLUDES =

View File

@ -0,0 +1,9 @@
set terminal png size 1800,768 enhanced font ,12
set output 'residual.png'
set datafile separator whitespace
set xlabel "Timestep"
set ylabel "Residual"
set logscale y 2
plot 'residual.dat' using 1:2 title "Residual"

View File

@ -1,61 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#ifdef __linux__
#ifdef _OPENMP
#include <pthread.h>
#include <sched.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/syscall.h>
#include <sys/types.h>
#include <unistd.h>
#define MAX_NUM_THREADS 128
#define gettid() syscall(SYS_gettid)
static int getProcessorID(cpu_set_t* cpu_set)
{
int processorId;
for (processorId = 0; processorId < MAX_NUM_THREADS; processorId++) {
if (CPU_ISSET(processorId, cpu_set)) {
break;
}
}
return processorId;
}
int affinity_getProcessorId()
{
cpu_set_t cpu_set;
CPU_ZERO(&cpu_set);
sched_getaffinity(gettid(), sizeof(cpu_set_t), &cpu_set);
return getProcessorID(&cpu_set);
}
void affinity_pinThread(int processorId)
{
cpu_set_t cpuset;
pthread_t thread;
thread = pthread_self();
CPU_ZERO(&cpuset);
CPU_SET(processorId, &cpuset);
pthread_setaffinity_np(thread, sizeof(cpu_set_t), &cpuset);
}
void affinity_pinProcess(int processorId)
{
cpu_set_t cpuset;
CPU_ZERO(&cpuset);
CPU_SET(processorId, &cpuset);
sched_setaffinity(0, sizeof(cpu_set_t), &cpuset);
}
#endif /*_OPENMP*/
#endif /*__linux__*/

View File

@ -1,14 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#ifndef AFFINITY_H
#define AFFINITY_H
extern int affinity_getProcessorId();
extern void affinity_pinProcess(int);
extern void affinity_pinThread(int);
#endif /*AFFINITY_H*/

View File

@ -1,14 +1,17 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#include <errno.h>
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
void* allocate(int alignment, size_t bytesize)
#include "allocate.h"
void* allocate(size_t alignment, size_t bytesize)
{
int errorCode;
void* ptr;

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
@ -8,6 +8,6 @@
#define __ALLOCATE_H_
#include <stdlib.h>
extern void* allocate(int alignment, size_t bytesize);
extern void* allocate(size_t alignment, size_t bytesize);
#endif

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
@ -11,22 +11,17 @@
#include <string.h>
#include "allocate.h"
#include "discretization.h"
#include "parameter.h"
#include "solver.h"
#include "util.h"
#define P(i, j) p[(j) * (imax + 2) + (i)]
#define F(i, j) f[(j) * (imax + 2) + (i)]
#define G(i, j) g[(j) * (imax + 2) + (i)]
#define RHS(i, j) rhs[(j) * (imax + 2) + (i)]
static void print(Solver* solver, double* grid)
static void print(Discretization* d, double* grid)
{
int imax = solver->imax;
int imax = d->grid.imax;
for (int j = 0; j < solver->jmax + 2; j++) {
for (int j = 0; j < d->grid.jmax + 2; j++) {
printf("%02d: ", j);
for (int i = 0; i < solver->imax + 2; i++) {
for (int i = 0; i < d->grid.imax + 2; i++) {
printf("%12.8f ", grid[j * (imax + 2) + i]);
}
printf("\n");
@ -34,92 +29,86 @@ static void print(Solver* solver, double* grid)
fflush(stdout);
}
static void printConfig(Solver* solver)
static void printConfig(Discretization* d)
{
printf("Parameters for #%s#\n", solver->problem);
printf("Parameters for #%s#\n", d->problem);
printf("Boundary conditions Left:%d Right:%d Bottom:%d Top:%d\n",
solver->bcLeft,
solver->bcRight,
solver->bcBottom,
solver->bcTop);
printf("\tReynolds number: %.2f\n", solver->re);
printf("\tGx Gy: %.2f %.2f\n", solver->gx, solver->gy);
d->bcLeft,
d->bcRight,
d->bcBottom,
d->bcTop);
printf("\tReynolds number: %.2f\n", d->re);
printf("\tGx Gy: %.2f %.2f\n", d->gx, d->gy);
printf("Geometry data:\n");
printf("\tDomain box size (x, y): %.2f, %.2f\n", solver->xlength, solver->ylength);
printf("\tCells (x, y): %d, %d\n", solver->imax, solver->jmax);
printf("\tDomain box size (x, y): %.2f, %.2f\n", d->grid.xlength, d->grid.ylength);
printf("\tCells (x, y): %d, %d\n", d->grid.imax, d->grid.jmax);
printf("Timestep parameters:\n");
printf("\tDefault stepsize: %.2f, Final time %.2f\n", solver->dt, solver->te);
printf("\tdt bound: %.6f\n", solver->dtBound);
printf("\tTau factor: %.2f\n", solver->tau);
printf("Iterative solver parameters:\n");
printf("\tMax iterations: %d\n", solver->itermax);
printf("\tepsilon (stopping tolerance) : %f\n", solver->eps);
printf("\tgamma factor: %f\n", solver->gamma);
printf("\tomega (SOR relaxation): %f\n", solver->omega);
printf("\tDefault stepsize: %.2f, Final time %.2f\n", d->dt, d->te);
printf("\tdt bound: %.6f\n", d->dtBound);
printf("\tTau factor: %.2f\n", d->tau);
printf("Iterative d parameters:\n");
printf("\tgamma factor: %f\n", d->gamma);
}
void initSolver(Solver* solver, Parameter* params)
void initDiscretization(Discretization* d, Parameter* p)
{
solver->problem = params->name;
solver->bcLeft = params->bcLeft;
solver->bcRight = params->bcRight;
solver->bcBottom = params->bcBottom;
solver->bcTop = params->bcTop;
solver->imax = params->imax;
solver->jmax = params->jmax;
solver->xlength = params->xlength;
solver->ylength = params->ylength;
solver->dx = params->xlength / params->imax;
solver->dy = params->ylength / params->jmax;
solver->eps = params->eps;
solver->omega = params->omg;
solver->itermax = params->itermax;
solver->re = params->re;
solver->gx = params->gx;
solver->gy = params->gy;
solver->dt = params->dt;
solver->te = params->te;
solver->tau = params->tau;
solver->gamma = params->gamma;
d->problem = p->name;
d->bcLeft = p->bcLeft;
d->bcRight = p->bcRight;
d->bcBottom = p->bcBottom;
d->bcTop = p->bcTop;
d->grid.imax = p->imax;
d->grid.jmax = p->jmax;
d->grid.xlength = p->xlength;
d->grid.ylength = p->ylength;
d->grid.dx = p->xlength / p->imax;
d->grid.dy = p->ylength / p->jmax;
d->re = p->re;
d->gx = p->gx;
d->gy = p->gy;
d->dt = p->dt;
d->te = p->te;
d->tau = p->tau;
d->gamma = p->gamma;
int imax = solver->imax;
int jmax = solver->jmax;
int imax = d->grid.imax;
int jmax = d->grid.jmax;
size_t size = (imax + 2) * (jmax + 2) * sizeof(double);
solver->u = allocate(64, size);
solver->v = allocate(64, size);
solver->p = allocate(64, size);
solver->rhs = allocate(64, size);
solver->f = allocate(64, size);
solver->g = allocate(64, size);
d->u = allocate(64, size);
d->v = allocate(64, size);
d->p = allocate(64, size);
d->rhs = allocate(64, size);
d->f = allocate(64, size);
d->g = allocate(64, size);
for (int i = 0; i < (imax + 2) * (jmax + 2); i++) {
solver->u[i] = params->u_init;
solver->v[i] = params->v_init;
solver->p[i] = params->p_init;
solver->rhs[i] = 0.0;
solver->f[i] = 0.0;
solver->g[i] = 0.0;
d->u[i] = p->u_init;
d->v[i] = p->v_init;
d->p[i] = p->p_init;
d->rhs[i] = 0.0;
d->f[i] = 0.0;
d->g[i] = 0.0;
}
double dx = solver->dx;
double dy = solver->dy;
double dx = d->grid.dx;
double dy = d->grid.dy;
double invSqrSum = 1.0 / (dx * dx) + 1.0 / (dy * dy);
solver->dtBound = 0.5 * solver->re * 1.0 / invSqrSum;
d->dtBound = 0.5 * d->re * 1.0 / invSqrSum;
#ifdef VERBOSE
printConfig(solver);
printConfig(d);
#endif
}
void computeRHS(Solver* solver)
void computeRHS(Discretization* d)
{
int imax = solver->imax;
int jmax = solver->jmax;
double idx = 1.0 / solver->dx;
double idy = 1.0 / solver->dy;
double idt = 1.0 / solver->dt;
double* rhs = solver->rhs;
double* f = solver->f;
double* g = solver->g;
int imax = d->grid.imax;
int jmax = d->grid.jmax;
double idx = 1.0 / d->grid.dx;
double idy = 1.0 / d->grid.dy;
double idt = 1.0 / d->dt;
double* rhs = d->rhs;
double* f = d->f;
double* g = d->g;
for (int j = 1; j < jmax + 1; j++) {
for (int i = 1; i < imax + 1; i++) {
@ -129,63 +118,9 @@ void computeRHS(Solver* solver)
}
}
void solve(Solver* solver)
static double maxElement(Discretization* d, double* m)
{
int imax = solver->imax;
int jmax = solver->jmax;
double eps = solver->eps;
int itermax = solver->itermax;
double dx2 = solver->dx * solver->dx;
double dy2 = solver->dy * solver->dy;
double idx2 = 1.0 / dx2;
double idy2 = 1.0 / dy2;
double factor = solver->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
double* p = solver->p;
double* rhs = solver->rhs;
double epssq = eps * eps;
int it = 0;
double res = 1.0;
while ((res >= epssq) && (it < itermax)) {
res = 0.0;
for (int j = 1; j < jmax + 1; j++) {
for (int i = 1; i < imax + 1; i++) {
double r = RHS(i, j) -
((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
(P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2);
P(i, j) -= (factor * r);
res += (r * r);
}
}
for (int i = 1; i < imax + 1; i++) {
P(i, 0) = P(i, 1);
P(i, jmax + 1) = P(i, jmax);
}
for (int j = 1; j < jmax + 1; j++) {
P(0, j) = P(1, j);
P(imax + 1, j) = P(imax, j);
}
res = res / (double)(imax * jmax);
#ifdef DEBUG
printf("%d Residuum: %e\n", it, res);
#endif
it++;
}
#ifdef VERBOSE
printf("Solver took %d iterations to reach %f\n", it, sqrt(res));
#endif
}
static double maxElement(Solver* solver, double* m)
{
int size = (solver->imax + 2) * (solver->jmax + 2);
int size = (d->grid.imax + 2) * (d->grid.jmax + 2);
double maxval = DBL_MIN;
for (int i = 0; i < size; i++) {
@ -195,10 +130,10 @@ static double maxElement(Solver* solver, double* m)
return maxval;
}
void normalizePressure(Solver* solver)
void normalizePressure(Discretization* d)
{
int size = (solver->imax + 2) * (solver->jmax + 2);
double* p = solver->p;
int size = (d->grid.imax + 2) * (d->grid.jmax + 2);
double* p = d->p;
double avgP = 0.0;
for (int i = 0; i < size; i++) {
@ -211,13 +146,13 @@ void normalizePressure(Solver* solver)
}
}
void computeTimestep(Solver* solver)
void computeTimestep(Discretization* d)
{
double dt = solver->dtBound;
double dx = solver->dx;
double dy = solver->dy;
double umax = maxElement(solver, solver->u);
double vmax = maxElement(solver, solver->v);
double dt = d->dtBound;
double dx = d->grid.dx;
double dy = d->grid.dy;
double umax = maxElement(d, d->u);
double vmax = maxElement(d, d->v);
if (umax > 0) {
dt = (dt > dx / umax) ? dx / umax : dt;
@ -226,18 +161,18 @@ void computeTimestep(Solver* solver)
dt = (dt > dy / vmax) ? dy / vmax : dt;
}
solver->dt = dt * solver->tau;
d->dt = dt * d->tau;
}
void setBoundaryConditions(Solver* solver)
void setBoundaryConditions(Discretization* d)
{
int imax = solver->imax;
int jmax = solver->jmax;
double* u = solver->u;
double* v = solver->v;
int imax = d->grid.imax;
int jmax = d->grid.jmax;
double* u = d->u;
double* v = d->v;
// Left boundary
switch (solver->bcLeft) {
switch (d->bcLeft) {
case NOSLIP:
for (int j = 1; j < jmax + 1; j++) {
U(0, j) = 0.0;
@ -261,7 +196,7 @@ void setBoundaryConditions(Solver* solver)
}
// Right boundary
switch (solver->bcRight) {
switch (d->bcRight) {
case NOSLIP:
for (int j = 1; j < jmax + 1; j++) {
U(imax, j) = 0.0;
@ -285,7 +220,7 @@ void setBoundaryConditions(Solver* solver)
}
// Bottom boundary
switch (solver->bcBottom) {
switch (d->bcBottom) {
case NOSLIP:
for (int i = 1; i < imax + 1; i++) {
V(i, 0) = 0.0;
@ -309,7 +244,7 @@ void setBoundaryConditions(Solver* solver)
}
// Top boundary
switch (solver->bcTop) {
switch (d->bcTop) {
case NOSLIP:
for (int i = 1; i < imax + 1; i++) {
V(i, jmax) = 0.0;
@ -333,19 +268,19 @@ void setBoundaryConditions(Solver* solver)
}
}
void setSpecialBoundaryCondition(Solver* solver)
void setSpecialBoundaryCondition(Discretization* d)
{
int imax = solver->imax;
int jmax = solver->jmax;
double mDy = solver->dy;
double* u = solver->u;
int imax = d->grid.imax;
int jmax = d->grid.jmax;
double mDy = d->grid.dy;
double* u = d->u;
if (strcmp(solver->problem, "dcavity") == 0) {
if (strcmp(d->problem, "dcavity") == 0) {
for (int i = 1; i < imax; i++) {
U(i, jmax + 1) = 2.0 - U(i, jmax);
}
} else if (strcmp(solver->problem, "canal") == 0) {
double ylength = solver->ylength;
} else if (strcmp(d->problem, "canal") == 0) {
double ylength = d->grid.ylength;
double y;
for (int j = 1; j < jmax + 1; j++) {
@ -355,21 +290,21 @@ void setSpecialBoundaryCondition(Solver* solver)
}
}
void computeFG(Solver* solver)
void computeFG(Discretization* d)
{
double* u = solver->u;
double* v = solver->v;
double* f = solver->f;
double* g = solver->g;
int imax = solver->imax;
int jmax = solver->jmax;
double gx = solver->gx;
double gy = solver->gy;
double gamma = solver->gamma;
double dt = solver->dt;
double inverseRe = 1.0 / solver->re;
double inverseDx = 1.0 / solver->dx;
double inverseDy = 1.0 / solver->dy;
double* u = d->u;
double* v = d->v;
double* f = d->f;
double* g = d->g;
int imax = d->grid.imax;
int jmax = d->grid.jmax;
double gx = d->gx;
double gy = d->gy;
double gamma = d->gamma;
double dt = d->dt;
double inverseRe = 1.0 / d->re;
double inverseDx = 1.0 / d->grid.dx;
double inverseDy = 1.0 / d->grid.dy;
double du2dx, dv2dy, duvdx, duvdy;
double du2dx2, du2dy2, dv2dx2, dv2dy2;
@ -428,17 +363,17 @@ void computeFG(Solver* solver)
}
}
void adaptUV(Solver* solver)
void adaptUV(Discretization* d)
{
int imax = solver->imax;
int jmax = solver->jmax;
double* p = solver->p;
double* u = solver->u;
double* v = solver->v;
double* f = solver->f;
double* g = solver->g;
double factorX = solver->dt / solver->dx;
double factorY = solver->dt / solver->dy;
int imax = d->grid.imax;
int jmax = d->grid.jmax;
double* p = d->p;
double* u = d->u;
double* v = d->v;
double* f = d->f;
double* g = d->g;
double factorX = d->dt / d->grid.dx;
double factorY = d->dt / d->grid.dy;
for (int j = 1; j < jmax + 1; j++) {
for (int i = 1; i < imax + 1; i++) {
@ -448,15 +383,15 @@ void adaptUV(Solver* solver)
}
}
void writeResult(Solver* solver)
void writeResult(Discretization* d)
{
int imax = solver->imax;
int jmax = solver->jmax;
double dx = solver->dx;
double dy = solver->dy;
double* p = solver->p;
double* u = solver->u;
double* v = solver->v;
int imax = d->grid.imax;
int jmax = d->grid.jmax;
double dx = d->grid.dx;
double dy = d->grid.dy;
double* p = d->p;
double* u = d->u;
double* v = d->v;
double x = 0.0, y = 0.0;
FILE* fp;
@ -489,10 +424,10 @@ void writeResult(Solver* solver)
y = dy * (j - 0.5);
for (int i = 1; i < imax + 1; i++) {
x = dx * (i - 0.5);
double vel_u = (U(i, j) + U(i - 1, j)) / 2.0;
double vel_v = (V(i, j) + V(i, j - 1)) / 2.0;
double len = sqrt((vel_u * vel_u) + (vel_v * vel_v));
fprintf(fp, "%.2f %.2f %f %f %f\n", x, y, vel_u, vel_v, len);
double velU = (U(i, j) + U(i - 1, j)) / 2.0;
double velV = (V(i, j) + V(i, j - 1)) / 2.0;
double len = sqrt((velU * velU) + (velV * velV));
fprintf(fp, "%.2f %.2f %f %f %f\n", x, y, velU, velV, len);
}
}

View File

@ -0,0 +1,40 @@
/*
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#ifndef __DISCRETIZATION_H_
#define __DISCRETIZATION_H_
#include "grid.h"
#include "parameter.h"
enum BC { NOSLIP = 1, SLIP, OUTFLOW, PERIODIC };
typedef struct {
/* geometry and grid information */
Grid grid;
/* arrays */
double *p, *rhs;
double *f, *g;
double *u, *v;
/* parameters */
double re, tau, gamma;
double gx, gy;
/* time stepping */
double dt, te;
double dtBound;
char* problem;
int bcLeft, bcRight, bcBottom, bcTop;
} Discretization;
extern void initDiscretization(Discretization*, Parameter*);
extern void computeRHS(Discretization*);
extern void normalizePressure(Discretization*);
extern void computeTimestep(Discretization*);
extern void setBoundaryConditions(Discretization*);
extern void setSpecialBoundaryCondition(Discretization*);
extern void computeFG(Discretization*);
extern void adaptUV(Discretization*);
extern void writeResult(Discretization*);
#endif

View File

@ -0,0 +1,16 @@
/*
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#ifndef __GRID_H_
#define __GRID_H_
typedef struct {
double dx, dy;
int imax, jmax;
double xlength, ylength;
} Grid;
#endif // __GRID_H_

View File

@ -4,23 +4,23 @@
* Author: Jan Eitzinger (je), jan.eitzinger@fau.de
* Copyright (c) 2020 RRZE, University Erlangen-Nuremberg
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included
* in all copies or substantial portions of the Software.
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*
* =======================================================================================
*/

View File

@ -1,15 +1,14 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
*/
#include <float.h>
#include <limits.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include "discretization.h"
#include "parameter.h"
#include "progress.h"
#include "solver.h"
@ -17,50 +16,61 @@
int main(int argc, char** argv)
{
double S, E;
Parameter params;
Solver solver;
initParameter(&params);
double timeStart, timeStop;
Parameter p;
Discretization d;
Solver s;
initParameter(&p);
FILE* fp;
fp = initResidualWriter();
if (argc != 2) {
printf("Usage: %s <configFile>\n", argv[0]);
exit(EXIT_SUCCESS);
}
readParameter(&params, argv[1]);
printParameter(&params);
initSolver(&solver, &params);
readParameter(&p, argv[1]);
printParameter(&p);
initDiscretization(&d, &p);
initSolver(&s, &d, &p);
#ifndef VERBOSE
initProgress(solver.te);
initProgress(d.te);
#endif
double tau = solver.tau;
double te = solver.te;
double tau = d.tau;
double te = d.te;
double t = 0.0;
int nt = 0;
double res = 0.0;
S = getTimeStamp();
timeStart = getTimeStamp();
while (t <= te) {
if (tau > 0.0) computeTimestep(&solver);
setBoundaryConditions(&solver);
setSpecialBoundaryCondition(&solver);
computeFG(&solver);
computeRHS(&solver);
if (nt % 100 == 0) normalizePressure(&solver);
solveRB(&solver);
adaptUV(&solver);
t += solver.dt;
if (tau > 0.0) computeTimestep(&d);
setBoundaryConditions(&d);
setSpecialBoundaryCondition(&d);
computeFG(&d);
computeRHS(&d);
if (nt % 100 == 0) normalizePressure(&d);
res = solve(&s, d.p, d.rhs);
adaptUV(&d);
writeResidual(fp, t, res);
t += d.dt;
nt++;
#ifdef VERBOSE
printf("TIME %f , TIMESTEP %f\n", t, solver.dt);
printf("TIME %f , TIMESTEP %f\n", t, d.dt);
#else
printProgress(t);
#endif
}
E = getTimeStamp();
fclose(fp);
timeStop = getTimeStamp();
stopProgress();
printf("Solution took %.2fs\n", E - S);
writeResult(&solver);
printf("Solution took %.2fs\n", timeStop - timeStart);
writeResult(&d);
return EXIT_SUCCESS;
}

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
@ -9,21 +9,23 @@
#include <string.h>
#include "parameter.h"
#include "util.h"
#define MAXLINE 4096
void initParameter(Parameter* param)
{
param->xlength = 1.0;
param->ylength = 1.0;
param->imax = 100;
param->jmax = 100;
param->imax = 128;
param->jmax = 128;
param->itermax = 1000;
param->eps = 0.0001;
param->omg = 1.7;
param->re = 100.0;
param->gamma = 0.9;
param->tau = 0.5;
param->levels = 5;
param->presmooth = 5;
param->postsmooth = 5;
}
void readParameter(Parameter* param, const char* filename)
@ -61,6 +63,7 @@ void readParameter(Parameter* param, const char* filename)
PARSE_INT(imax);
PARSE_INT(jmax);
PARSE_INT(itermax);
PARSE_INT(levels);
PARSE_REAL(eps);
PARSE_REAL(omg);
PARSE_REAL(re);
@ -78,6 +81,8 @@ void readParameter(Parameter* param, const char* filename)
PARSE_REAL(u_init);
PARSE_REAL(v_init);
PARSE_REAL(p_init);
PARSE_INT(presmooth);
PARSE_INT(postsmooth);
}
}
@ -108,4 +113,5 @@ void printParameter(Parameter* param)
printf("\tepsilon (stopping tolerance) : %f\n", param->eps);
printf("\tgamma (stopping tolerance) : %f\n", param->gamma);
printf("\tomega (SOR relaxation): %f\n", param->omg);
printf("\tMultiGrid levels : %d\n", param->levels);
}

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
@ -10,14 +10,15 @@
typedef struct {
double xlength, ylength;
int imax, jmax;
int itermax;
double eps, omg;
int itermax, levels;
double eps, omg, rho;
double re, tau, gamma;
double te, dt;
double gx, gy;
char* name;
int bcLeft, bcRight, bcBottom, bcTop;
double u_init, v_init, p_init;
int presmooth, postsmooth;
} Parameter;
void initParameter(Parameter*);

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
@ -49,3 +49,22 @@ void stopProgress()
printf("\n");
fflush(stdout);
}
FILE* initResidualWriter()
{
FILE* fp;
fp = fopen("residual.dat", "w");
if (fp == NULL) {
printf("Error!\n");
exit(EXIT_FAILURE);
}
return fp;
}
void writeResidual(FILE* fp, double ts, double res)
{
fprintf(fp, "%f, %f\n", ts, res);
}

View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved.
* Use of this source code is governed by a MIT-style
* license that can be found in the LICENSE file.
@ -9,6 +9,8 @@
extern void initProgress(double);
extern void printProgress(double);
extern void stopProgress();
extern void stopProgress(void);
extern FILE* initResidualWriter(void);
extern void writeResidual(FILE*, double, double);
#endif

View File

@ -0,0 +1,221 @@
/*
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <stdio.h>
#include <stdlib.h>
#include "allocate.h"
#include "solver.h"
#include "util.h"
#define FINEST_LEVEL 0
#define COARSEST_LEVEL (s->levels - 1)
#define S(i, j) s[(j) * (imax + 2) + (i)]
#define E(i, j) e[(j) * (imax + 2) + (i)]
#define R(i, j) r[(j) * (imax + 2) + (i)]
#define OLD(i, j) old[(j) * (imax + 2) + (i)]
static void restrictMG(Solver* s, int level, int imax, int jmax)
{
double* r = s->r[level + 1];
double* old = s->r[level];
for (int j = 1; j < jmax + 1; j++) {
for (int i = 1; i < imax + 1; i++) {
R(i, j) = (OLD(2 * i - 1, 2 * j - 1) + OLD(2 * i, 2 * j - 1) * 2 +
OLD(2 * i + 1, 2 * j - 1) + OLD(2 * i - 1, 2 * j) * 2 +
OLD(2 * i, 2 * j) * 4 + OLD(2 * i + 1, 2 * j) * 2 +
OLD(2 * i - 1, 2 * j + 1) + OLD(2 * i, 2 * j + 1) * 2 +
OLD(2 * i + 1, 2 * j + 1)) /
16.0;
}
}
}
static void prolongate(Solver* s, int level, int imax, int jmax)
{
double* old = s->r[level + 1];
double* e = s->r[level];
for (int j = 2; j < jmax + 1; j += 2) {
for (int i = 2; i < imax + 1; i += 2) {
E(i, j) = OLD(i / 2, j / 2);
}
}
}
static void correct(Solver* s, double* p, int level, int imax, int jmax)
{
double* e = s->e[level];
for (int j = 1; j < jmax + 1; ++j) {
for (int i = 1; i < imax + 1; ++i) {
P(i, j) += E(i, j);
}
}
}
static void setBoundaryCondition(double* p, int imax, int jmax)
{
for (int i = 1; i < imax + 1; i++) {
P(i, 0) = P(i, 1);
P(i, jmax + 1) = P(i, jmax);
}
for (int j = 1; j < jmax + 1; j++) {
P(0, j) = P(1, j);
P(imax + 1, j) = P(imax, j);
}
}
static double smooth(Solver* s, double* p, double* rhs, int level, int imax, int jmax)
{
double dx2 = s->grid->dx * s->grid->dx;
double dy2 = s->grid->dy * s->grid->dy;
double idx2 = 1.0 / dx2;
double idy2 = 1.0 / dy2;
double factor = s->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
double* r = s->r[level];
double res = 1.0;
int pass, jsw, isw;
jsw = 1;
for (pass = 0; pass < 2; pass++) {
isw = jsw;
for (int j = 1; j < jmax + 1; j++) {
for (int i = isw; i < imax + 1; i += 2) {
P(i, j) -= factor * (RHS(i, j) -
((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
(P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2));
}
isw = 3 - isw;
}
jsw = 3 - jsw;
}
}
static double calculateResidual(Solver* s, double* p, double* rhs, int level, int imax, int jmax)
{
double dx2 = s->grid->dx * s->grid->dx;
double dy2 = s->grid->dy * s->grid->dy;
double idx2 = 1.0 / dx2;
double idy2 = 1.0 / dy2;
double factor = s->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
double* r = s->r[level];
double res = 1.0;
int pass, jsw, isw;
jsw = 1;
for (pass = 0; pass < 2; pass++) {
isw = jsw;
for (int j = 1; j < jmax + 1; j++) {
for (int i = isw; i < imax + 1; i += 2) {
R(i, j) = RHS(i, j) -
((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
(P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2);
res += (R(i, j) * R(i, j));
}
isw = 3 - isw;
}
jsw = 3 - jsw;
}
res = res / (double)(imax * jmax);
return res;
}
static double multiGrid(Solver* s, double* p, double* rhs, int level, int imax, int jmax)
{
double res = 0.0;
// coarsest level
if (level == COARSEST_LEVEL) {
for (int i = 0; i < 5; i++) {
smooth(s, p, rhs, level, imax, jmax);
}
return res;
}
// pre-smoothing
for (int i = 0; i < s->presmooth; i++) {
smooth(s, p, rhs, level, imax, jmax);
if (level == FINEST_LEVEL) setBoundaryCondition(p, imax, jmax);
}
res = calculateResidual(s, p, rhs, level, imax, jmax);
// restrict
restrictMG(s, level, imax, jmax);
// MGSolver on residual and error.
multiGrid(s, s->e[level + 1], s->r[level + 1], level + 1, imax / 2, jmax / 2);
// prolongate
prolongate(s, level, imax, jmax);
// correct p on finer level using residual
correct(s, p, level, imax, jmax);
if (level == FINEST_LEVEL) setBoundaryCondition(p, imax, jmax);
// post-smoothing
for (int i = 0; i < s->postsmooth; i++) {
smooth(s, p, rhs, level, imax, jmax);
if (level == FINEST_LEVEL) setBoundaryCondition(p, imax, jmax);
}
return res;
}
void initSolver(Solver* s, Discretization* d, Parameter* p)
{
s->eps = p->eps;
s->omega = p->omg;
s->itermax = p->itermax;
s->levels = p->levels;
s->grid = &d->grid;
s->presmooth = p->presmooth;
s->postsmooth = p->postsmooth;
int imax = s->grid->imax;
int jmax = s->grid->jmax;
int levels = s->levels;
printf("Using Multigrid solver with %d levels\n", levels);
s->r = malloc(levels * sizeof(double*));
s->e = malloc(levels * sizeof(double*));
size_t size = (imax + 2) * (jmax + 2) * sizeof(double);
for (int j = 0; j < levels; j++) {
s->r[j] = allocate(64, size);
s->e[j] = allocate(64, size);
for (int i = 0; i < (imax + 2) * (jmax + 2); i++) {
s->r[j][i] = 0.0;
s->e[j][i] = 0.0;
}
}
}
double solve(Solver* s, double* p, double* rhs)
{
double res = multiGrid(s, p, rhs, 0, s->grid->imax, s->grid->jmax);
#ifdef VERBOSE
printf("Residuum: %.6f\n", res);
#endif
return res;
}

View File

@ -0,0 +1,78 @@
/*
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include "solver.h"
#include "util.h"
void initSolver(Solver* s, Discretization* d, Parameter* p)
{
s->grid = &d->grid;
s->itermax = p->itermax;
s->eps = p->eps;
s->omega = p->omg;
}
double solve(Solver* solver, double* p, double* rhs)
{
int imax = solver->grid->imax;
int jmax = solver->grid->jmax;
double eps = solver->eps;
int itermax = solver->itermax;
double dx2 = solver->grid->dx * solver->grid->dx;
double dy2 = solver->grid->dy * solver->grid->dy;
double idx2 = 1.0 / dx2;
double idy2 = 1.0 / dy2;
double factor = solver->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
double epssq = eps * eps;
int it = 0;
double res = 1.0;
int pass, jsw, isw;
while ((res >= epssq) && (it < itermax)) {
res = 0.0;
jsw = 1;
for (pass = 0; pass < 2; pass++) {
isw = jsw;
for (int j = 1; j < jmax + 1; j++) {
for (int i = isw; i < imax + 1; i += 2) {
double r = RHS(i, j) -
((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
(P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2);
P(i, j) -= (factor * r);
res += (r * r);
}
isw = 3 - isw;
}
jsw = 3 - jsw;
}
for (int i = 1; i < imax + 1; i++) {
P(i, 0) = P(i, 1);
P(i, jmax + 1) = P(i, jmax);
}
for (int j = 1; j < jmax + 1; j++) {
P(0, j) = P(1, j);
P(imax + 1, j) = P(imax, j);
}
res = res / (double)(imax * jmax);
#ifdef DEBUG
printf("%d Residuum: %e\n", it, res);
#endif
it++;
}
#ifdef VERBOSE
printf("Solver took %d iterations to reach %f\n", it, sqrt(res));
#endif
return res;
}

View File

@ -0,0 +1,70 @@
/*
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include "solver.h"
#include "util.h"
void initSolver(Solver* s, Discretization* d, Parameter* p)
{
s->grid = &d->grid;
s->itermax = p->itermax;
s->eps = p->eps;
s->omega = p->omg;
}
double solve(Solver* solver, double* p, double* rhs)
{
int imax = solver->grid->imax;
int jmax = solver->grid->jmax;
double eps = solver->eps;
int itermax = solver->itermax;
double dx2 = solver->grid->dx * solver->grid->dx;
double dy2 = solver->grid->dy * solver->grid->dy;
double idx2 = 1.0 / dx2;
double idy2 = 1.0 / dy2;
double factor = solver->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
double epssq = eps * eps;
int it = 0;
double res = 1.0;
while ((res >= epssq) && (it < itermax)) {
res = 0.0;
for (int j = 1; j < jmax + 1; j++) {
for (int i = 1; i < imax + 1; i++) {
double r = RHS(i, j) -
((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
(P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2);
P(i, j) -= (factor * r);
res += (r * r);
}
}
for (int i = 1; i < imax + 1; i++) {
P(i, 0) = P(i, 1);
P(i, jmax + 1) = P(i, jmax);
}
for (int j = 1; j < jmax + 1; j++) {
P(0, j) = P(1, j);
P(imax + 1, j) = P(imax, j);
}
res = res / (double)(imax * jmax);
#ifdef DEBUG
printf("%d Residuum: %e\n", it, res);
#endif
it++;
}
#ifdef VERBOSE
printf("Solver took %d iterations to reach %f\n", it, sqrt(res));
#endif
return res;
}

View File

@ -1,564 +0,0 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#include <float.h>
#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "allocate.h"
#include "parameter.h"
#include "solver.h"
#include "util.h"
#define P(i, j) p[(j) * (imax + 2) + (i)]
#define F(i, j) f[(j) * (imax + 2) + (i)]
#define G(i, j) g[(j) * (imax + 2) + (i)]
#define U(i, j) u[(j) * (imax + 2) + (i)]
#define V(i, j) v[(j) * (imax + 2) + (i)]
#define RHS(i, j) rhs[(j) * (imax + 2) + (i)]
static void print(Solver* solver, double* grid)
{
int imax = solver->imax;
for (int j = 0; j < solver->jmax + 2; j++) {
printf("%02d: ", j);
for (int i = 0; i < solver->imax + 2; i++) {
printf("%12.8f ", grid[j * (imax + 2) + i]);
}
printf("\n");
}
fflush(stdout);
}
static void printConfig(Solver* solver)
{
printf("Parameters for #%s#\n", solver->problem);
printf("Boundary conditions Left:%d Right:%d Bottom:%d Top:%d\n",
solver->bcLeft,
solver->bcRight,
solver->bcBottom,
solver->bcTop);
printf("\tReynolds number: %.2f\n", solver->re);
printf("\tGx Gy: %.2f %.2f\n", solver->gx, solver->gy);
printf("Geometry data:\n");
printf("\tDomain box size (x, y): %.2f, %.2f\n", solver->xlength, solver->ylength);
printf("\tCells (x, y): %d, %d\n", solver->imax, solver->jmax);
printf("Timestep parameters:\n");
printf("\tDefault stepsize: %.2f, Final time %.2f\n", solver->dt, solver->te);
printf("\tdt bound: %.6f\n", solver->dtBound);
printf("\tTau factor: %.2f\n", solver->tau);
printf("Iterative solver parameters:\n");
printf("\tMax iterations: %d\n", solver->itermax);
printf("\tepsilon (stopping tolerance) : %f\n", solver->eps);
printf("\tgamma factor: %f\n", solver->gamma);
printf("\tomega (SOR relaxation): %f\n", solver->omega);
}
void initSolver(Solver* solver, Parameter* params)
{
solver->problem = params->name;
solver->bcLeft = params->bcLeft;
solver->bcRight = params->bcRight;
solver->bcBottom = params->bcBottom;
solver->bcTop = params->bcTop;
solver->imax = params->imax;
solver->jmax = params->jmax;
solver->xlength = params->xlength;
solver->ylength = params->ylength;
solver->dx = params->xlength / params->imax;
solver->dy = params->ylength / params->jmax;
solver->eps = params->eps;
solver->omega = params->omg;
solver->itermax = params->itermax;
solver->re = params->re;
solver->gx = params->gx;
solver->gy = params->gy;
solver->dt = params->dt;
solver->te = params->te;
solver->tau = params->tau;
solver->gamma = params->gamma;
int imax = solver->imax;
int jmax = solver->jmax;
size_t size = (imax + 2) * (jmax + 2) * sizeof(double);
solver->u = allocate(64, size);
solver->v = allocate(64, size);
solver->p = allocate(64, size);
solver->rhs = allocate(64, size);
solver->f = allocate(64, size);
solver->g = allocate(64, size);
for (int i = 0; i < (imax + 2) * (jmax + 2); i++) {
solver->u[i] = params->u_init;
solver->v[i] = params->v_init;
solver->p[i] = params->p_init;
solver->rhs[i] = 0.0;
solver->f[i] = 0.0;
solver->g[i] = 0.0;
}
double dx = solver->dx;
double dy = solver->dy;
double invSqrSum = 1.0 / (dx * dx) + 1.0 / (dy * dy);
solver->dtBound = 0.5 * solver->re * 1.0 / invSqrSum;
#ifdef VERBOSE
printConfig(solver);
#endif
}
void computeRHS(Solver* solver)
{
int imax = solver->imax;
int jmax = solver->jmax;
double idx = 1.0 / solver->dx;
double idy = 1.0 / solver->dy;
double idt = 1.0 / solver->dt;
double* rhs = solver->rhs;
double* f = solver->f;
double* g = solver->g;
for (int j = 1; j < jmax + 1; j++) {
for (int i = 1; i < imax + 1; i++) {
RHS(i, j) = idt *
((F(i, j) - F(i - 1, j)) * idx + (G(i, j) - G(i, j - 1)) * idy);
}
}
}
void solve(Solver* solver)
{
int imax = solver->imax;
int jmax = solver->jmax;
double eps = solver->eps;
int itermax = solver->itermax;
double dx2 = solver->dx * solver->dx;
double dy2 = solver->dy * solver->dy;
double idx2 = 1.0 / dx2;
double idy2 = 1.0 / dy2;
double factor = solver->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
double* p = solver->p;
double* rhs = solver->rhs;
double epssq = eps * eps;
int it = 0;
double res = 1.0;
while ((res >= epssq) && (it < itermax)) {
res = 0.0;
for (int j = 1; j < jmax + 1; j++) {
for (int i = 1; i < imax + 1; i++) {
double r = RHS(i, j) -
((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
(P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2);
P(i, j) -= (factor * r);
res += (r * r);
}
}
for (int i = 1; i < imax + 1; i++) {
P(i, 0) = P(i, 1);
P(i, jmax + 1) = P(i, jmax);
}
for (int j = 1; j < jmax + 1; j++) {
P(0, j) = P(1, j);
P(imax + 1, j) = P(imax, j);
}
res = res / (double)(imax * jmax);
#ifdef DEBUG
printf("%d Residuum: %e\n", it, res);
#endif
it++;
}
#ifdef VERBOSE
printf("Solver took %d iterations to reach %f\n", it, sqrt(res));
#endif
}
void solveRB(Solver* solver)
{
int imax = solver->imax;
int jmax = solver->jmax;
double eps = solver->eps;
int itermax = solver->itermax;
double dx2 = solver->dx * solver->dx;
double dy2 = solver->dy * solver->dy;
double idx2 = 1.0 / dx2;
double idy2 = 1.0 / dy2;
double factor = solver->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
double* p = solver->p;
double* rhs = solver->rhs;
double epssq = eps * eps;
int it = 0;
double res = 1.0;
int pass, jsw, isw;
while ((res >= epssq) && (it < itermax)) {
res = 0.0;
jsw = 1;
for (pass = 0; pass < 2; pass++) {
isw = jsw;
for (int j = 1; j < jmax + 1; j++) {
for (int i = isw; i < imax + 1; i += 2) {
double r = RHS(i, j) -
((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
(P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2);
P(i, j) -= (factor * r);
res += (r * r);
}
isw = 3 - isw;
}
jsw = 3 - jsw;
}
for (int i = 1; i < imax + 1; i++) {
P(i, 0) = P(i, 1);
P(i, jmax + 1) = P(i, jmax);
}
for (int j = 1; j < jmax + 1; j++) {
P(0, j) = P(1, j);
P(imax + 1, j) = P(imax, j);
}
res = res / (double)(imax * jmax);
#ifdef DEBUG
printf("%d Residuum: %e\n", it, res);
#endif
it++;
}
#ifdef VERBOSE
printf("Solver took %d iterations to reach %f\n", it, sqrt(res));
#endif
}
static double maxElement(Solver* solver, double* m)
{
int size = (solver->imax + 2) * (solver->jmax + 2);
double maxval = DBL_MIN;
for (int i = 0; i < size; i++) {
maxval = MAX(maxval, fabs(m[i]));
}
return maxval;
}
void normalizePressure(Solver* solver)
{
int size = (solver->imax + 2) * (solver->jmax + 2);
double* p = solver->p;
double avgP = 0.0;
for (int i = 0; i < size; i++) {
avgP += p[i];
}
avgP /= size;
for (int i = 0; i < size; i++) {
p[i] = p[i] - avgP;
}
}
void computeTimestep(Solver* solver)
{
double dt = solver->dtBound;
double dx = solver->dx;
double dy = solver->dy;
double umax = maxElement(solver, solver->u);
double vmax = maxElement(solver, solver->v);
if (umax > 0) {
dt = (dt > dx / umax) ? dx / umax : dt;
}
if (vmax > 0) {
dt = (dt > dy / vmax) ? dy / vmax : dt;
}
solver->dt = dt * solver->tau;
}
void setBoundaryConditions(Solver* solver)
{
int imax = solver->imax;
int jmax = solver->jmax;
double* u = solver->u;
double* v = solver->v;
// Left boundary
switch (solver->bcLeft) {
case NOSLIP:
for (int j = 1; j < jmax + 1; j++) {
U(0, j) = 0.0;
V(0, j) = -V(1, j);
}
break;
case SLIP:
for (int j = 1; j < jmax + 1; j++) {
U(0, j) = 0.0;
V(0, j) = V(1, j);
}
break;
case OUTFLOW:
for (int j = 1; j < jmax + 1; j++) {
U(0, j) = U(1, j);
V(0, j) = V(1, j);
}
break;
case PERIODIC:
break;
}
// Right boundary
switch (solver->bcRight) {
case NOSLIP:
for (int j = 1; j < jmax + 1; j++) {
U(imax, j) = 0.0;
V(imax + 1, j) = -V(imax, j);
}
break;
case SLIP:
for (int j = 1; j < jmax + 1; j++) {
U(imax, j) = 0.0;
V(imax + 1, j) = V(imax, j);
}
break;
case OUTFLOW:
for (int j = 1; j < jmax + 1; j++) {
U(imax, j) = U(imax - 1, j);
V(imax + 1, j) = V(imax, j);
}
break;
case PERIODIC:
break;
}
// Bottom boundary
switch (solver->bcBottom) {
case NOSLIP:
for (int i = 1; i < imax + 1; i++) {
V(i, 0) = 0.0;
U(i, 0) = -U(i, 1);
}
break;
case SLIP:
for (int i = 1; i < imax + 1; i++) {
V(i, 0) = 0.0;
U(i, 0) = U(i, 1);
}
break;
case OUTFLOW:
for (int i = 1; i < imax + 1; i++) {
U(i, 0) = U(i, 1);
V(i, 0) = V(i, 1);
}
break;
case PERIODIC:
break;
}
// Top boundary
switch (solver->bcTop) {
case NOSLIP:
for (int i = 1; i < imax + 1; i++) {
V(i, jmax) = 0.0;
U(i, jmax + 1) = -U(i, jmax);
}
break;
case SLIP:
for (int i = 1; i < imax + 1; i++) {
V(i, jmax) = 0.0;
U(i, jmax + 1) = U(i, jmax);
}
break;
case OUTFLOW:
for (int i = 1; i < imax + 1; i++) {
U(i, jmax + 1) = U(i, jmax);
V(i, jmax) = V(i, jmax - 1);
}
break;
case PERIODIC:
break;
}
}
void setSpecialBoundaryCondition(Solver* solver)
{
int imax = solver->imax;
int jmax = solver->jmax;
double mDy = solver->dy;
double* u = solver->u;
if (strcmp(solver->problem, "dcavity") == 0) {
for (int i = 1; i < imax; i++) {
U(i, jmax + 1) = 2.0 - U(i, jmax);
}
} else if (strcmp(solver->problem, "canal") == 0) {
double ylength = solver->ylength;
double y;
for (int j = 1; j < jmax + 1; j++) {
y = mDy * (j - 0.5);
U(0, j) = y * (ylength - y) * 4.0 / (ylength * ylength);
}
}
}
void computeFG(Solver* solver)
{
double* u = solver->u;
double* v = solver->v;
double* f = solver->f;
double* g = solver->g;
int imax = solver->imax;
int jmax = solver->jmax;
double gx = solver->gx;
double gy = solver->gy;
double gamma = solver->gamma;
double dt = solver->dt;
double inverseRe = 1.0 / solver->re;
double inverseDx = 1.0 / solver->dx;
double inverseDy = 1.0 / solver->dy;
double du2dx, dv2dy, duvdx, duvdy;
double du2dx2, du2dy2, dv2dx2, dv2dy2;
for (int j = 1; j < jmax + 1; j++) {
for (int i = 1; i < imax + 1; i++) {
du2dx = inverseDx * 0.25 *
((U(i, j) + U(i + 1, j)) * (U(i, j) + U(i + 1, j)) -
(U(i, j) + U(i - 1, j)) * (U(i, j) + U(i - 1, j))) +
gamma * inverseDx * 0.25 *
(fabs(U(i, j) + U(i + 1, j)) * (U(i, j) - U(i + 1, j)) +
fabs(U(i, j) + U(i - 1, j)) * (U(i, j) - U(i - 1, j)));
duvdy = inverseDy * 0.25 *
((V(i, j) + V(i + 1, j)) * (U(i, j) + U(i, j + 1)) -
(V(i, j - 1) + V(i + 1, j - 1)) * (U(i, j) + U(i, j - 1))) +
gamma * inverseDy * 0.25 *
(fabs(V(i, j) + V(i + 1, j)) * (U(i, j) - U(i, j + 1)) +
fabs(V(i, j - 1) + V(i + 1, j - 1)) *
(U(i, j) - U(i, j - 1)));
du2dx2 = inverseDx * inverseDx * (U(i + 1, j) - 2.0 * U(i, j) + U(i - 1, j));
du2dy2 = inverseDy * inverseDy * (U(i, j + 1) - 2.0 * U(i, j) + U(i, j - 1));
F(i, j) = U(i, j) + dt * (inverseRe * (du2dx2 + du2dy2) - du2dx - duvdy + gx);
duvdx = inverseDx * 0.25 *
((U(i, j) + U(i, j + 1)) * (V(i, j) + V(i + 1, j)) -
(U(i - 1, j) + U(i - 1, j + 1)) * (V(i, j) + V(i - 1, j))) +
gamma * inverseDx * 0.25 *
(fabs(U(i, j) + U(i, j + 1)) * (V(i, j) - V(i + 1, j)) +
fabs(U(i - 1, j) + U(i - 1, j + 1)) *
(V(i, j) - V(i - 1, j)));
dv2dy = inverseDy * 0.25 *
((V(i, j) + V(i, j + 1)) * (V(i, j) + V(i, j + 1)) -
(V(i, j) + V(i, j - 1)) * (V(i, j) + V(i, j - 1))) +
gamma * inverseDy * 0.25 *
(fabs(V(i, j) + V(i, j + 1)) * (V(i, j) - V(i, j + 1)) +
fabs(V(i, j) + V(i, j - 1)) * (V(i, j) - V(i, j - 1)));
dv2dx2 = inverseDx * inverseDx * (V(i + 1, j) - 2.0 * V(i, j) + V(i - 1, j));
dv2dy2 = inverseDy * inverseDy * (V(i, j + 1) - 2.0 * V(i, j) + V(i, j - 1));
G(i, j) = V(i, j) + dt * (inverseRe * (dv2dx2 + dv2dy2) - duvdx - dv2dy + gy);
}
}
/* ---------------------- boundary of F --------------------------- */
for (int j = 1; j < jmax + 1; j++) {
F(0, j) = U(0, j);
F(imax, j) = U(imax, j);
}
/* ---------------------- boundary of G --------------------------- */
for (int i = 1; i < imax + 1; i++) {
G(i, 0) = V(i, 0);
G(i, jmax) = V(i, jmax);
}
}
void adaptUV(Solver* solver)
{
int imax = solver->imax;
int jmax = solver->jmax;
double* p = solver->p;
double* u = solver->u;
double* v = solver->v;
double* f = solver->f;
double* g = solver->g;
double factorX = solver->dt / solver->dx;
double factorY = solver->dt / solver->dy;
for (int j = 1; j < jmax + 1; j++) {
for (int i = 1; i < imax + 1; i++) {
U(i, j) = F(i, j) - (P(i + 1, j) - P(i, j)) * factorX;
V(i, j) = G(i, j) - (P(i, j + 1) - P(i, j)) * factorY;
}
}
}
void writeResult(Solver* solver)
{
int imax = solver->imax;
int jmax = solver->jmax;
double dx = solver->dx;
double dy = solver->dy;
double* p = solver->p;
double* u = solver->u;
double* v = solver->v;
double x = 0.0, y = 0.0;
FILE* fp;
fp = fopen("pressure.dat", "w");
if (fp == NULL) {
printf("Error!\n");
exit(EXIT_FAILURE);
}
for (int j = 1; j < jmax + 1; j++) {
y = (double)(j - 0.5) * dy;
for (int i = 1; i < imax + 1; i++) {
x = (double)(i - 0.5) * dx;
fprintf(fp, "%.2f %.2f %f\n", x, y, P(i, j));
}
fprintf(fp, "\n");
}
fclose(fp);
fp = fopen("velocity.dat", "w");
if (fp == NULL) {
printf("Error!\n");
exit(EXIT_FAILURE);
}
for (int j = 1; j < jmax + 1; j++) {
y = dy * (j - 0.5);
for (int i = 1; i < imax + 1; i++) {
x = dx * (i - 0.5);
double vel_u = (U(i, j) + U(i - 1, j)) / 2.0;
double vel_v = (V(i, j) + V(i, j - 1)) / 2.0;
double len = sqrt((vel_u * vel_u) + (vel_v * vel_v));
fprintf(fp, "%.2f %.2f %f %f %f\n", x, y, vel_u, vel_v, len);
}
}
fclose(fp);
}

View File

@ -1,46 +1,27 @@
/*
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
* All rights reserved. This file is part of nusif-solver.
* Use of this source code is governed by a MIT style
* license that can be found in the LICENSE file.
*/
#ifndef __SOLVER_H_
#define __SOLVER_H_
#include "discretization.h"
#include "grid.h"
#include "parameter.h"
enum BC { NOSLIP = 1, SLIP, OUTFLOW, PERIODIC };
typedef struct {
/* geometry and grid information */
double dx, dy;
int imax, jmax;
double xlength, ylength;
/* arrays */
double *p, *rhs;
double *f, *g;
double *u, *v;
Grid* grid;
/* parameters */
double eps, omega;
double re, tau, gamma;
double gx, gy;
/* time stepping */
double eps, omega, rho;
int itermax;
double dt, te;
double dtBound;
char* problem;
int bcLeft, bcRight, bcBottom, bcTop;
int levels;
double **r, **e;
int presmooth, postsmooth;
} Solver;
extern void initSolver(Solver*, Parameter*);
extern void computeRHS(Solver*);
extern void solve(Solver*);
extern void solveRB(Solver*);
extern void solveRBA(Solver*);
extern void normalizePressure(Solver*);
extern void computeTimestep(Solver*);
extern void setBoundaryConditions(Solver*);
extern void setSpecialBoundaryCondition(Solver*);
extern void computeFG(Solver*);
extern void adaptUV(Solver*);
extern void writeResult(Solver*);
extern void initSolver(Solver*, Discretization*, Parameter*);
extern double solve(Solver*, double*, double*);
#endif

Some files were not shown because too many files have changed in this diff Show More