int main(int argc, char **argv)
{
// OP initialisation
op_init(argc,argv,2);
//MPI for user I/O
int my_rank;
int comm_size;
MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
MPI_Comm_size(MPI_COMM_WORLD, &comm_size);
//timer
double cpu_t1, cpu_t2, wall_t1, wall_t2;
int *becell, *ecell, *bound, *bedge, *edge, *cell;
double *x, *q, *qold, *adt, *res;
int nnode,ncell,nedge,nbedge,niter;
double rms;
/**------------------------BEGIN I/O and PARTITIONING -------------------**/
op_timers(&cpu_t1, &wall_t1);
/* read in grid from disk on root processor */
FILE *fp;
if ( (fp = fopen("new_grid.dat","r")) == NULL) {
op_printf("can't open file new_grid.dat\n"); exit(-1);
}
int g_nnode,g_ncell,g_nedge,g_nbedge;
check_scan(fscanf(fp,"%d %d %d %d \n",&g_nnode, &g_ncell, &g_nedge, &g_nbedge), 4);
int *g_becell = 0, *g_ecell = 0, *g_bound = 0, *g_bedge = 0, *g_edge = 0, *g_cell = 0;
double *g_x = 0,*g_q = 0, *g_qold = 0, *g_adt = 0, *g_res = 0;
// set constants
op_printf("initialising flow field\n");
gam = 1.4f;
gm1 = gam - 1.0f;
cfl = 0.9f;
eps = 0.05f;
double mach = 0.4f;
double alpha = 3.0f*atan(1.0f)/45.0f;
double p = 1.0f;
double r = 1.0f;
double u = sqrt(gam*p/r)*mach;
double e = p/(r*gm1) + 0.5f*u*u;
qinf[0] = r;
qinf[1] = r*u;
qinf[2] = 0.0f;
qinf[3] = r*e;
op_printf("reading in grid \n");
op_printf("Global number of nodes, cells, edges, bedges = %d, %d, %d, %d\n"
,g_nnode,g_ncell,g_nedge,g_nbedge);
if(my_rank == MPI_ROOT) {
g_cell = (int *) malloc(4*g_ncell*sizeof(int));
g_edge = (int *) malloc(2*g_nedge*sizeof(int));
g_ecell = (int *) malloc(2*g_nedge*sizeof(int));
g_bedge = (int *) malloc(2*g_nbedge*sizeof(int));
g_becell = (int *) malloc( g_nbedge*sizeof(int));
g_bound = (int *) malloc( g_nbedge*sizeof(int));
g_x = (double *) malloc(2*g_nnode*sizeof(double));
g_q = (double *) malloc(4*g_ncell*sizeof(double));
g_qold = (double *) malloc(4*g_ncell*sizeof(double));
g_res = (double *) malloc(4*g_ncell*sizeof(double));
g_adt = (double *) malloc( g_ncell*sizeof(double));
for (int n=0; n<g_nnode; n++){
check_scan(fscanf(fp,"%lf %lf \n",&g_x[2*n], &g_x[2*n+1]), 2);
}
for (int n=0; n<g_ncell; n++) {
check_scan(fscanf(fp,"%d %d %d %d \n",&g_cell[4*n ], &g_cell[4*n+1],
&g_cell[4*n+2], &g_cell[4*n+3]), 4);
}
for (int n=0; n<g_nedge; n++) {
check_scan(fscanf(fp,"%d %d %d %d \n",&g_edge[2*n],&g_edge[2*n+1],
&g_ecell[2*n],&g_ecell[2*n+1]), 4);
}
for (int n=0; n<g_nbedge; n++) {
check_scan(fscanf(fp,"%d %d %d %d \n",&g_bedge[2*n],&g_bedge[2*n+1],
&g_becell[n],&g_bound[n]), 4);
}
//initialise flow field and residual
for (int n=0; n<g_ncell; n++) {
for (int m=0; m<4; m++) {
g_q[4*n+m] = qinf[m];
g_res[4*n+m] = 0.0f;
}
}
}
fclose(fp);
nnode = compute_local_size (g_nnode, comm_size, my_rank);
ncell = compute_local_size (g_ncell, comm_size, my_rank);
nedge = compute_local_size (g_nedge, comm_size, my_rank);
nbedge = compute_local_size (g_nbedge, comm_size, my_rank);
op_printf("Number of nodes, cells, edges, bedges on process %d = %d, %d, %d, %d\n"
,my_rank,nnode,ncell,nedge,nbedge);
/*Allocate memory to hold local sets, mapping tables and data*/
cell = (int *) malloc(4*ncell*sizeof(int));
edge = (int *) malloc(2*nedge*sizeof(int));
ecell = (int *) malloc(2*nedge*sizeof(int));
bedge = (int *) malloc(2*nbedge*sizeof(int));
becell = (int *) malloc( nbedge*sizeof(int));
bound = (int *) malloc( nbedge*sizeof(int));
x = (double *) malloc(2*nnode*sizeof(double));
q = (double *) malloc(4*ncell*sizeof(double));
qold = (double *) malloc(4*ncell*sizeof(double));
res = (double *) malloc(4*ncell*sizeof(double));
adt = (double *) malloc( ncell*sizeof(double));
/* scatter sets, mappings and data on sets*/
scatter_int_array(g_cell, cell, comm_size, g_ncell,ncell, 4);
scatter_int_array(g_edge, edge, comm_size, g_nedge,nedge, 2);
scatter_int_array(g_ecell, ecell, comm_size, g_nedge,nedge, 2);
scatter_int_array(g_bedge, bedge, comm_size, g_nbedge,nbedge, 2);
scatter_int_array(g_becell, becell, comm_size, g_nbedge,nbedge, 1);
scatter_int_array(g_bound, bound, comm_size, g_nbedge,nbedge, 1);
scatter_double_array(g_x, x, comm_size, g_nnode,nnode, 2);
scatter_double_array(g_q, q, comm_size, g_ncell,ncell, 4);
scatter_double_array(g_qold, qold, comm_size, g_ncell,ncell, 4);
scatter_double_array(g_res, res, comm_size, g_ncell,ncell, 4);
scatter_double_array(g_adt, adt, comm_size, g_ncell,ncell, 1);
/*Freeing memory allocated to gloabal arrays on rank 0
after scattering to all processes*/
if(my_rank == MPI_ROOT) {
free(g_cell);
free(g_edge);
free(g_ecell);
free(g_bedge);
free(g_becell);
free(g_bound);
free(g_x );
free(g_q);
free(g_qold);
free(g_adt);
free(g_res);
}
op_timers(&cpu_t2, &wall_t2);
op_printf("Max total file read time = %f\n", wall_t2-wall_t1);
/**------------------------END I/O and PARTITIONING -----------------------**/
// declare sets, pointers, datasets and global constants
op_set nodes = op_decl_set(nnode, "nodes");
op_set edges = op_decl_set(nedge, "edges");
op_set bedges = op_decl_set(nbedge, "bedges");
op_set cells = op_decl_set(ncell, "cells");
op_map pedge = op_decl_map(edges, nodes,2,edge, "pedge");
op_map pecell = op_decl_map(edges, cells,2,ecell, "pecell");
op_map pbedge = op_decl_map(bedges,nodes,2,bedge, "pbedge");
op_map pbecell = op_decl_map(bedges,cells,1,becell,"pbecell");
op_map pcell = op_decl_map(cells, nodes,4,cell, "pcell");
op_dat p_bound = op_decl_dat(bedges,1,"int" ,bound,"p_bound");
op_dat p_x = op_decl_dat(nodes ,2,"double",x ,"p_x");
op_dat p_q = op_decl_dat(cells ,4,"double",q ,"p_q");
//op_dat p_qold = op_decl_dat(cells ,4,"double",qold ,"p_qold");
//op_dat p_adt = op_decl_dat(cells ,1,"double",adt ,"p_adt");
//op_dat p_res = op_decl_dat(cells ,4,"double",res ,"p_res");
// p_res, p_adt and p_qold now declared as a temp op_dats during
// the execution of the time-marching loop
op_decl_const2("gam",1,"double",&gam );
op_decl_const2("gm1",1,"double",&gm1 );
op_decl_const2("cfl",1,"double",&cfl );
op_decl_const2("eps",1,"double",&eps );
op_decl_const2("mach",1,"double",&mach );
op_decl_const2("alpha",1,"double",&alpha);
op_decl_const2("qinf",4,"double",qinf );
op_diagnostic_output();
//trigger partitioning and halo creation routines
op_partition("PTSCOTCH", "KWAY", cells, pecell, p_x);
//initialise timers for total execution wall time
op_timers(&cpu_t1, &wall_t1);
niter = 1000;
for(int iter=1; iter<=niter; iter++) {
double* tmp_elem = NULL;
op_dat p_res = op_decl_dat_temp(cells ,4,"double",tmp_elem,"p_res");
op_dat p_adt = op_decl_dat_temp(cells ,1,"double",tmp_elem,"p_adt");
op_dat p_qold = op_decl_dat_temp(cells ,4,"double",qold ,"p_qold");
//save old flow solution
op_par_loop_save_soln("save_soln",cells,
op_arg_dat(p_q,-1,OP_ID,4,"double",OP_READ),
op_arg_dat(p_qold,-1,OP_ID,4,"double",OP_WRITE));
// predictor/corrector update loop
for(int k=0; k<2; k++) {
// calculate area/timstep
op_par_loop_adt_calc("adt_calc",cells,
op_arg_dat(p_x,0,pcell,2,"double",OP_READ),
op_arg_dat(p_x,1,pcell,2,"double",OP_READ),
op_arg_dat(p_x,2,pcell,2,"double",OP_READ),
op_arg_dat(p_x,3,pcell,2,"double",OP_READ),
op_arg_dat(p_q,-1,OP_ID,4,"double",OP_READ),
op_arg_dat(p_adt,-1,OP_ID,1,"double",OP_WRITE));
// calculate flux residual
op_par_loop_res_calc("res_calc",edges,
op_arg_dat(p_x,0,pedge,2,"double",OP_READ),
op_arg_dat(p_x,1,pedge,2,"double",OP_READ),
op_arg_dat(p_q,0,pecell,4,"double",OP_READ),
op_arg_dat(p_q,1,pecell,4,"double",OP_READ),
op_arg_dat(p_adt,0,pecell,1,"double",OP_READ),
op_arg_dat(p_adt,1,pecell,1,"double",OP_READ),
op_arg_dat(p_res,0,pecell,4,"double",OP_INC),
op_arg_dat(p_res,1,pecell,4,"double",OP_INC));
op_par_loop_bres_calc("bres_calc",bedges,
op_arg_dat(p_x,0,pbedge,2,"double",OP_READ),
op_arg_dat(p_x,1,pbedge,2,"double",OP_READ),
op_arg_dat(p_q,0,pbecell,4,"double",OP_READ),
op_arg_dat(p_adt,0,pbecell,1,"double",OP_READ),
op_arg_dat(p_res,0,pbecell,4,"double",OP_INC),
op_arg_dat(p_bound,-1,OP_ID,1,"int",OP_READ));
// update flow field
rms = 0.0;
op_par_loop_update("update",cells,
op_arg_dat(p_qold,-1,OP_ID,4,"double",OP_READ),
op_arg_dat(p_q,-1,OP_ID,4,"double",OP_WRITE),
op_arg_dat(p_res,-1,OP_ID,4,"double",OP_RW),
op_arg_dat(p_adt,-1,OP_ID,1,"double",OP_READ),
op_arg_gbl(&rms,1,"double",OP_INC));
}
//print iteration history
rms = sqrt(rms/(double) g_ncell);
if (iter%100 == 0)
op_printf("%d %10.5e \n",iter,rms);
if (op_free_dat_temp(p_res) < 0)
op_printf("Error: temporary op_dat %s cannot be removed\n",p_res->name);
if (op_free_dat_temp(p_adt) < 0)
op_printf("Error: temporary op_dat %s cannot be removed\n",p_adt->name);
if (op_free_dat_temp(p_qold) < 0)
op_printf("Error: temporary op_dat %s cannot be removed\n",p_qold->name);
}
op_timers(&cpu_t2, &wall_t2);
op_timing_output();
//print total time for niter interations
op_printf("Max total runtime = %f\n",wall_t2-wall_t1);
op_exit();
free(cell);
free(edge);
free(ecell);
free(bedge);
free(becell);
free(bound);
free(x);
free(q);
free(qold);
free(res);
free(adt);
}
int main(int argc, char **argv)
{
// OP initialisation
op_init(argc,argv,2);
//MPI for user I/O
int my_rank;
int comm_size;
MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
MPI_Comm_size(MPI_COMM_WORLD, &comm_size);
//timer
double cpu_t1, cpu_t2, wall_t1, wall_t2;
int *pp;
double *A, *r, *u, *du;
int nnode, nedge;
/**------------------------BEGIN I/O and PARTITIONING ---------------------**/
int g_nnode, g_nedge, g_n, g_e;
g_nnode = (NN-1)*(NN-1);
g_nedge = (NN-1)*(NN-1) + 4*(NN-1)*(NN-2);
int *g_pp = 0;
double *g_A = 0, *g_r = 0, *g_u = 0, *g_du = 0;
op_printf("Global number of nodes, edges = %d, %d\n",g_nnode,g_nedge);
if(my_rank == MPI_ROOT) {
g_pp = (int *)malloc(sizeof(int)*2*g_nedge);
g_A = (double *)malloc(sizeof(double)*g_nedge);
g_r = (double *)malloc(sizeof(double)*g_nnode);
g_u = (double *)malloc(sizeof(double)*g_nnode);
g_du = (double *)malloc(sizeof(double)*g_nnode);
// create matrix and r.h.s., and set coordinates needed for renumbering / partitioning
g_e = 0;
for (int i=1; i<NN; i++) {
for (int j=1; j<NN; j++) {
g_n = i-1 + (j-1)*(NN-1);
g_r[g_n] = 0.0f;
g_u[g_n] = 0.0f;
g_du[g_n] = 0.0f;
g_pp[2*g_e] = g_n;
g_pp[2*g_e+1] = g_n;
g_A[g_e] = -1.0f;
g_e++;
for (int pass=0; pass<4; pass++) {
int i2 = i;
int j2 = j;
if (pass==0) i2 += -1;
if (pass==1) i2 += 1;
if (pass==2) j2 += -1;
if (pass==3) j2 += 1;
if ( (i2==0) || (i2==NN) || (j2==0) || (j2==NN) ) {
g_r[g_n] += 0.25f;
}
else {
g_pp[2*g_e] = g_n;
g_pp[2*g_e+1] = i2-1 + (j2-1)*(NN-1);
g_A[g_e] = 0.25f;
g_e++;
}
}
}
}
}
/* Compute local sizes */
nnode = compute_local_size (g_nnode, comm_size, my_rank);
nedge = compute_local_size (g_nedge, comm_size, my_rank);
op_printf("Number of nodes, edges on process %d = %d, %d\n"
,my_rank,nnode,nedge);
/*Allocate memory to hold local sets, mapping tables and data*/
pp = (int *)malloc(2*sizeof(int)*nedge);
A = (double *) malloc(nedge*sizeof(double));
r = (double *) malloc(nnode*sizeof(double));
u = (double *) malloc(nnode*sizeof(double));
du = (double *) malloc(nnode*sizeof(double));
/* scatter sets, mappings and data on sets*/
scatter_int_array(g_pp, pp, comm_size, g_nedge,nedge, 2);
scatter_double_array(g_A, A, comm_size, g_nedge,nedge, 1);
scatter_double_array(g_r, r, comm_size, g_nnode,nnode, 1);
scatter_double_array(g_u, u, comm_size, g_nnode,nnode, 1);
scatter_double_array(g_du, du, comm_size, g_nnode,nnode, 1);
/*Freeing memory allocated to gloabal arrays on rank 0
after scattering to all processes*/
if(my_rank == MPI_ROOT) {
free(g_pp);
free(g_A);
free(g_r);
free(g_u);
free(g_du);
}
/**------------------------END I/O and PARTITIONING ---------------------**/
// declare sets, pointers, and datasets
op_set nodes = op_decl_set(nnode,"nodes");
op_set edges = op_decl_set(nedge,"edges");
op_map ppedge = op_decl_map(edges,nodes,2,pp, "ppedge");
op_dat p_A = op_decl_dat(edges,1,"double", A, "p_A" );
op_dat p_r = op_decl_dat(nodes,1,"double", r, "p_r" );
op_dat p_u = op_decl_dat(nodes,1,"double", u, "p_u" );
op_dat p_du = op_decl_dat(nodes,1,"double", du,"p_du");
alpha = 1.0f;
op_decl_const(1,"double",&alpha);
op_diagnostic_output();
//trigger partitioning and halo creation routines
op_partition("PTSCOTCH", "KWAY", NULL, NULL, NULL);
//initialise timers for total execution wall time
op_timers(&cpu_t1, &wall_t1);
// main iteration loop
double u_sum, u_max, beta = 1.0f;
for (int iter=0; iter<NITER; iter++) {
op_par_loop(res,"res", edges,
op_arg_dat(p_A, -1,OP_ID, 1,"double", OP_READ),
op_arg_dat(p_u, 1,ppedge, 1,"double", OP_READ),
op_arg_dat(p_du, 0,ppedge, 1,"double", OP_INC),
op_arg_gbl(&beta, 1,"double", OP_READ));
u_sum = 0.0f;
u_max = 0.0f;
op_par_loop(update,"update", nodes,
op_arg_dat(p_r, -1,OP_ID, 1,"double",OP_READ),
op_arg_dat(p_du, -1,OP_ID, 1,"double",OP_RW),
op_arg_dat(p_u, -1,OP_ID, 1,"double",OP_INC),
op_arg_gbl(&u_sum,1,"double",OP_INC),
op_arg_gbl(&u_max,1,"double",OP_MAX));
op_printf("\n u max/rms = %f %f \n\n",u_max, sqrt(u_sum/g_nnode));
}
op_timers(&cpu_t2, &wall_t2);
//get results data array
op_dat temp = op_mpi_get_data(p_u);
//output the result dat array to files
print_dat_tofile(temp, "out_grid.dat"); //ASCI
//print_dat_tobinfile(temp, "out_grid.bin"); //Binary
//print each mpi process's timing info for each kernel
op_timing_output();
//print total time for niter interations
op_printf("Max total runtime = %f\n",wall_t2-wall_t1);
op_exit();
}
int main(int argc, char **argv)
{
// OP initialisation
op_init(argc,argv,2);
//MPI for user I/O
int my_rank;
int comm_size;
MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
MPI_Comm_size(MPI_COMM_WORLD, &comm_size);
//timer
double cpu_t1, cpu_t2, wall_t1, wall_t2;
int *bnode, *cell, *g_bnode, *g_cell;
double *xm, *g_xm;;
int nnode,ncell,nbnodes,niter, g_nnode, g_ncell, g_nbnodes;
double rms = 1;
// read in grid
op_printf("reading in grid \n");
FILE *fp;
if ( (fp = fopen("FE_grid.dat","r")) == NULL) {
op_printf("can't open file FE_grid.dat\n"); exit(-1);
}
if (fscanf(fp,"%d %d %d \n",&g_nnode, &g_ncell, &g_nbnodes) != 3) {
op_printf("error reading from new_grid.dat\n"); exit(-1);
}
if (my_rank == MPI_ROOT) {
g_cell = (int *) malloc(4*g_ncell*sizeof(int));
g_bnode = (int *) malloc(g_nbnodes*sizeof(int));
g_xm = (double *) malloc(2*g_nnode*sizeof(double));
for (int n=0; n<g_nnode; n++) {
if (fscanf(fp,"%lf %lf \n",&g_xm[2*n], &g_xm[2*n+1]) != 2) {
op_printf("error reading from new_grid.dat\n"); exit(-1);
}
}
for (int n=0; n<g_ncell; n++) {
if (fscanf(fp,"%d %d %d %d \n",&g_cell[4*n ], &g_cell[4*n+1],
&g_cell[4*n+2], &g_cell[4*n+3]) != 4) {
op_printf("error reading from new_grid.dat\n"); exit(-1);
}
}
for (int n=0; n<g_nbnodes; n++) {
if (fscanf(fp,"%d \n",&g_bnode[n]) != 1) {
op_printf("error reading from new_grid.dat\n"); exit(-1);
}
}
}
fclose(fp);
nnode = compute_local_size (g_nnode, comm_size, my_rank);
ncell = compute_local_size (g_ncell, comm_size, my_rank);
nbnodes = compute_local_size (g_nbnodes, comm_size, my_rank);
cell = (int *) malloc(4*ncell*sizeof(int));
bnode = (int *) malloc(nbnodes*sizeof(int));
xm = (double *) malloc(2*nnode*sizeof(double));
scatter_int_array(g_cell, cell, comm_size, g_ncell,ncell, 4);
scatter_int_array(g_bnode, bnode, comm_size, g_nbnodes,nbnodes, 1);
scatter_double_array(g_xm, xm, comm_size, g_nnode,nnode, 2);
if(my_rank == MPI_ROOT) {
free(g_cell);
free(g_xm);
free(g_bnode);
}
// set constants and initialise flow field and residual
op_printf("initialising flow field \n");
double gam = 1.4;
gm1 = gam - 1.0;
gm1i = 1.0/gm1;
wtg1[0] = 0.5;
wtg1[1] = 0.5;
xi1[0] = 0.211324865405187;
xi1[1] = 0.788675134594813;
Ng1[0] = 0.788675134594813;
Ng1[1] = 0.211324865405187;
Ng1[2] = 0.211324865405187;
Ng1[3] = 0.788675134594813;
Ng1_xi[0] = -1;
Ng1_xi[1] = -1;
Ng1_xi[2] = 1;
Ng1_xi[3] = 1;
wtg2[0] = 0.25;
wtg2[1] = 0.25;
wtg2[2] = 0.25;
wtg2[3] = 0.25;
Ng2[0] = 0.622008467928146; Ng2[1] = 0.166666666666667; Ng2[2] = 0.166666666666667; Ng2[3] = 0.044658198738520;
Ng2[4] = 0.166666666666667; Ng2[5] = 0.622008467928146; Ng2[6] = 0.044658198738520; Ng2[7] = 0.166666666666667;
Ng2[8] = 0.166666666666667; Ng2[9] = 0.044658198738520; Ng2[10] = 0.622008467928146; Ng2[11] = 0.166666666666667;
Ng2[12] = 0.044658198738520; Ng2[13] = 0.166666666666667; Ng2[14] = 0.166666666666667; Ng2[15] = 0.622008467928146;
Ng2_xi[0] = -0.788675134594813; Ng2_xi[1] = 0.788675134594813; Ng2_xi[2] = -0.211324865405187;Ng2_xi[3] = 0.211324865405187;
Ng2_xi[4] = -0.788675134594813; Ng2_xi[5] = 0.788675134594813; Ng2_xi[6] = -0.211324865405187; Ng2_xi[7] = 0.211324865405187;
Ng2_xi[8] = -0.211324865405187; Ng2_xi[9] = 0.211324865405187; Ng2_xi[10] = -0.788675134594813; Ng2_xi[11] = 0.788675134594813;
Ng2_xi[12] = -0.211324865405187; Ng2_xi[13] = 0.211324865405187; Ng2_xi[14] = -0.788675134594813; Ng2_xi[15] = 0.788675134594813;
Ng2_xi[16] = -0.788675134594813; Ng2_xi[17] = -0.211324865405187; Ng2_xi[18] = 0.788675134594813; Ng2_xi[19] = 0.211324865405187;
Ng2_xi[20] = -0.211324865405187; Ng2_xi[21] = -0.788675134594813; Ng2_xi[22] = 0.211324865405187; Ng2_xi[23] = 0.788675134594813;
Ng2_xi[24] = -0.788675134594813; Ng2_xi[25] = -0.211324865405187; Ng2_xi[26] = 0.788675134594813; Ng2_xi[27] = 0.211324865405187;
Ng2_xi[28] = -0.211324865405187; Ng2_xi[29] = -0.788675134594813; Ng2_xi[30] = 0.211324865405187; Ng2_xi[31] = 0.788675134594813;
minf = 0.1;
m2 = minf*minf;
freq = 1;
kappa = 1;
nmode = 0;
mfan = 1.0;
double *phim = (double *)malloc(nnode*sizeof(double));
memset(phim,0,nnode*sizeof(double));
for (int i = 0;i<nnode;i++) {
phim[i] = minf*xm[2*i];
}
double *K = (double *)malloc(4*4*ncell*sizeof(double));
memset(K,0,4*4*ncell*sizeof(double));
double *resm = (double *)malloc(nnode*sizeof(double));
memset(resm,0,nnode*sizeof(double));
double *V = (double *)malloc(nnode*sizeof(double));
memset(V,0,nnode*sizeof(double));
double *P = (double *)malloc(nnode*sizeof(double));
memset(P,0,nnode*sizeof(double));
double *U = (double *)malloc(nnode*sizeof(double));
memset(U,0,nnode*sizeof(double));
// declare sets, pointers, datasets and global constants
op_set nodes = op_decl_set(nnode, "nodes");
op_set bnodes = op_decl_set(nbnodes, "bedges");
op_set cells = op_decl_set(ncell, "cells");
op_map pbnodes = op_decl_map(bnodes,nodes,1,bnode, "pbedge");
op_map pcell = op_decl_map(cells, nodes,4,cell, "pcell");
op_dat p_xm = op_decl_dat(nodes ,2,"double",xm ,"p_x");
op_dat p_phim = op_decl_dat(nodes, 1, "double", phim, "p_phim");
op_dat p_resm = op_decl_dat(nodes, 1, "double", resm, "p_resm");
op_dat p_K = op_decl_dat(cells, 16, "double:soa", K, "p_K");
op_dat p_V = op_decl_dat(nodes, 1, "double", V, "p_V");
op_dat p_P = op_decl_dat(nodes, 1, "double", P, "p_P");
op_dat p_U = op_decl_dat(nodes, 1, "double", U, "p_U");
op_decl_const(1,"double",&gam );
op_decl_const(1,"double",&gm1 );
op_decl_const(1,"double",&gm1i );
op_decl_const(1,"double",&m2 );
op_decl_const(2,"double",wtg1 );
op_decl_const(2,"double",xi1 );
op_decl_const(4,"double",Ng1 );
op_decl_const(4,"double",Ng1_xi );
op_decl_const(4,"double",wtg2 );
op_decl_const(16,"double",Ng2 );
op_decl_const(32,"double",Ng2_xi );
op_decl_const(1,"double",&minf );
op_decl_const(1,"double",&freq );
op_decl_const(1,"double",&kappa );
op_decl_const(1,"double",&nmode );
op_decl_const(1,"double",&mfan );
op_diagnostic_output();
op_partition("PTSCOTCH", "KWAY", cells, pcell, NULL);
// main time-marching loop
niter = 20;
//initialise timers for total execution wall time
op_timers(&cpu_t1, &wall_t1);
for(int iter=1; iter<=niter; iter++) {
op_par_loop(res_calc,"res_calc",cells,
op_arg_dat(p_xm, -4, pcell, 2,"double",OP_READ),
op_arg_dat(p_phim, -4, pcell, 1,"double",OP_READ),
op_arg_dat(p_K, -1, OP_ID, 16,"double:soa",OP_WRITE),
op_arg_dat(p_resm, -4, pcell, 1,"double",OP_INC)
);
op_par_loop(dirichlet,"dirichlet",bnodes,
op_arg_dat(p_resm, 0, pbnodes, 1,"double",OP_WRITE));
double c1 = 0;
double c2 = 0;
double c3 = 0;
double alpha = 0;
double beta = 0;
//c1 = R'*R;
op_par_loop(init_cg, "init_cg", nodes,
op_arg_dat(p_resm, -1, OP_ID, 1, "double", OP_READ),
op_arg_gbl(&c1, 1, "double", OP_INC),
op_arg_dat(p_U, -1, OP_ID, 1, "double", OP_WRITE),
op_arg_dat(p_V, -1, OP_ID, 1, "double", OP_WRITE),
op_arg_dat(p_P, -1, OP_ID, 1, "double", OP_WRITE));
//set up stopping conditions
double res0 = sqrt(c1);
double res = res0;
int inner_iter = 0;
int maxiter = 200;
while (res > 0.1*res0 && inner_iter < maxiter) {
//V = Stiffness*P
op_par_loop(spMV, "spMV", cells,
op_arg_dat(p_V, -4, pcell, 1, "double", OP_INC),
op_arg_dat(p_K, -1, OP_ID, 16, "double:soa", OP_READ),
op_arg_dat(p_P, -4, pcell, 1, "double", OP_READ));
op_par_loop(dirichlet,"dirichlet",bnodes,
op_arg_dat(p_V, 0, pbnodes, 1,"double",OP_WRITE));
c2 = 0;
//c2 = P'*V;
op_par_loop(dotPV, "dotPV", nodes,
op_arg_dat(p_P, -1, OP_ID, 1, "double", OP_READ),
op_arg_dat(p_V, -1, OP_ID, 1, "double", OP_READ),
op_arg_gbl(&c2, 1, "double", OP_INC));
alpha = c1/c2;
//U = U + alpha*P;
//resm = resm-alpha*V;
op_par_loop(updateUR, "updateUR", nodes,
op_arg_dat(p_U, -1, OP_ID, 1, "double", OP_INC),
op_arg_dat(p_resm, -1, OP_ID, 1, "double", OP_INC),
op_arg_dat(p_P, -1, OP_ID, 1, "double", OP_READ),
op_arg_dat(p_V, -1, OP_ID, 1, "double", OP_RW),
op_arg_gbl(&alpha, 1, "double", OP_READ));
c3 = 0;
//c3 = resm'*resm;
op_par_loop(dotR, "dotR", nodes,
op_arg_dat(p_resm, -1, OP_ID, 1, "double", OP_READ),
op_arg_gbl(&c3, 1, "double", OP_INC));
beta = c3/c1;
//P = beta*P+resm;
op_par_loop(updateP, "updateP", nodes,
op_arg_dat(p_resm, -1, OP_ID, 1, "double", OP_READ),
op_arg_dat(p_P, -1, OP_ID, 1, "double", OP_RW),
op_arg_gbl(&beta, 1, "double", OP_READ));
c1 = c3;
res = sqrt(c1);
inner_iter++;
}
rms = 0;
//phim = phim - Stiffness\Load;
op_par_loop(update, "update", nodes,
op_arg_dat(p_phim, -1, OP_ID, 1, "double", OP_RW),
op_arg_dat(p_resm, -1, OP_ID, 1, "double", OP_WRITE),
op_arg_dat(p_U, -1, OP_ID, 1, "double", OP_READ),
op_arg_gbl(&rms, 1, "double", OP_INC));
op_printf("rms = %10.5e iter: %d\n", sqrt(rms)/sqrt(g_nnode), inner_iter);
}
op_timers(&cpu_t2, &wall_t2);
op_timing_output();
op_printf("Max total runtime = %f\n",wall_t2-wall_t1);
op_exit();
/*free(cell);
free(bnode);
free(xm);
free(phim);
free(K);
free(resm);
free(V);
free(P);
free(U);*/
}
int main(int argc, char **argv) {
// OP initialisation
op_init(argc, argv, 2);
// MPI for user I/O
int my_rank;
int comm_size;
MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
MPI_Comm_size(MPI_COMM_WORLD, &comm_size);
// timer
double cpu_t1, cpu_t2, wall_t1, wall_t2;
int *becell, *ecell, *bound, *bedge, *edge, *cell;
double *x, *q, *qold, *adt, *res;
int nnode, ncell, nedge, nbedge, niter;
/**------------------------BEGIN I/O and PARTITIONING -------------------**/
op_timers(&cpu_t1, &wall_t1);
/* read in grid from disk on root processor */
FILE *fp;
if ((fp = fopen("new_grid.dat", "r")) == NULL) {
op_printf("can't open file new_grid.dat\n");
exit(-1);
}
int g_nnode, g_ncell, g_nedge, g_nbedge;
check_scan(
fscanf(fp, "%d %d %d %d \n", &g_nnode, &g_ncell, &g_nedge, &g_nbedge), 4);
int *g_becell = 0, *g_ecell = 0, *g_bound = 0, *g_bedge = 0, *g_edge = 0,
*g_cell = 0;
double *g_x = 0, *g_q = 0, *g_qold = 0, *g_adt = 0, *g_res = 0;
op_printf("reading in grid \n");
op_printf("Global number of nodes, cells, edges, bedges = %d, %d, %d, %d\n",
g_nnode, g_ncell, g_nedge, g_nbedge);
if (my_rank == MPI_ROOT) {
g_cell = (int *)malloc(4 * g_ncell * sizeof(int));
g_edge = (int *)malloc(2 * g_nedge * sizeof(int));
g_ecell = (int *)malloc(2 * g_nedge * sizeof(int));
g_bedge = (int *)malloc(2 * g_nbedge * sizeof(int));
g_becell = (int *)malloc(g_nbedge * sizeof(int));
g_bound = (int *)malloc(g_nbedge * sizeof(int));
g_x = (double *)malloc(2 * g_nnode * sizeof(double));
g_q = (double *)malloc(4 * g_ncell * sizeof(double));
g_qold = (double *)malloc(4 * g_ncell * sizeof(double));
g_res = (double *)malloc(4 * g_ncell * sizeof(double));
g_adt = (double *)malloc(g_ncell * sizeof(double));
for (int n = 0; n < g_nnode; n++) {
check_scan(fscanf(fp, "%lf %lf \n", &g_x[2 * n], &g_x[2 * n + 1]), 2);
}
for (int n = 0; n < g_ncell; n++) {
check_scan(fscanf(fp, "%d %d %d %d \n", &g_cell[4 * n],
&g_cell[4 * n + 1], &g_cell[4 * n + 2],
&g_cell[4 * n + 3]),
4);
}
for (int n = 0; n < g_nedge; n++) {
check_scan(fscanf(fp, "%d %d %d %d \n", &g_edge[2 * n],
&g_edge[2 * n + 1], &g_ecell[2 * n],
&g_ecell[2 * n + 1]),
4);
}
for (int n = 0; n < g_nbedge; n++) {
check_scan(fscanf(fp, "%d %d %d %d \n", &g_bedge[2 * n],
&g_bedge[2 * n + 1], &g_becell[n], &g_bound[n]),
4);
}
// initialise flow field and residual
}
fclose(fp);
nnode = compute_local_size(g_nnode, comm_size, my_rank);
ncell = compute_local_size(g_ncell, comm_size, my_rank);
nedge = compute_local_size(g_nedge, comm_size, my_rank);
nbedge = compute_local_size(g_nbedge, comm_size, my_rank);
op_printf(
"Number of nodes, cells, edges, bedges on process %d = %d, %d, %d, %d\n",
my_rank, nnode, ncell, nedge, nbedge);
/*Allocate memory to hold local sets, mapping tables and data*/
cell = (int *)malloc(4 * ncell * sizeof(int));
edge = (int *)malloc(2 * nedge * sizeof(int));
ecell = (int *)malloc(2 * nedge * sizeof(int));
bedge = (int *)malloc(2 * nbedge * sizeof(int));
becell = (int *)malloc(nbedge * sizeof(int));
bound = (int *)malloc(nbedge * sizeof(int));
x = (double *)malloc(2 * nnode * sizeof(double));
q = (double *)malloc(4 * ncell * sizeof(double));
qold = (double *)malloc(4 * ncell * sizeof(double));
res = (double *)malloc(4 * ncell * sizeof(double));
adt = (double *)malloc(ncell * sizeof(double));
/* scatter sets, mappings and data on sets*/
scatter_int_array(g_cell, cell, comm_size, g_ncell, ncell, 4);
scatter_int_array(g_edge, edge, comm_size, g_nedge, nedge, 2);
scatter_int_array(g_ecell, ecell, comm_size, g_nedge, nedge, 2);
scatter_int_array(g_bedge, bedge, comm_size, g_nbedge, nbedge, 2);
scatter_int_array(g_becell, becell, comm_size, g_nbedge, nbedge, 1);
scatter_int_array(g_bound, bound, comm_size, g_nbedge, nbedge, 1);
scatter_double_array(g_x, x, comm_size, g_nnode, nnode, 2);
scatter_double_array(g_q, q, comm_size, g_ncell, ncell, 4);
scatter_double_array(g_qold, qold, comm_size, g_ncell, ncell, 4);
scatter_double_array(g_res, res, comm_size, g_ncell, ncell, 4);
scatter_double_array(g_adt, adt, comm_size, g_ncell, ncell, 1);
/*Freeing memory allocated to gloabal arrays on rank 0
after scattering to all processes*/
if (my_rank == MPI_ROOT) {
free(g_cell);
free(g_edge);
free(g_ecell);
free(g_bedge);
free(g_becell);
free(g_bound);
free(g_x);
free(g_q);
free(g_qold);
free(g_adt);
free(g_res);
}
op_timers(&cpu_t2, &wall_t2);
op_printf("Max total file read time = %f\n", wall_t2 - wall_t1);
/**------------------------END I/O and PARTITIONING -----------------------**/
op_set edges = op_decl_set(nedge, "edges");
op_set cells = op_decl_set(ncell, "cells");
op_map pecell = op_decl_map(edges, cells, 2, ecell, "pecell");
op_dat p_res = op_decl_dat(cells, 4, "double", res, "p_res");
int count;
// trigger partitioning and halo creation routines
op_partition("PTSCOTCH", "KWAY", cells, pecell, NULL);
op_diagnostic_output();
// initialise timers for total execution wall time
op_timers(&cpu_t1, &wall_t1);
// indirect reduction
count = 0;
op_par_loop_res_calc("res_calc", edges,
op_arg_dat(p_res, 0, pecell, 4, "double", OP_INC),
op_arg_gbl(&count, 1, "int", OP_INC));
op_printf("number of edges:: %d should be: %d \n", count, g_nedge);
if (count != g_nedge)
op_printf("indirect reduction FAILED\n");
else
op_printf("indirect reduction PASSED\n");
// direct reduction
count = 0;
op_par_loop_update("update", cells,
op_arg_dat(p_res, -1, OP_ID, 4, "double", OP_RW),
op_arg_gbl(&count, 1, "int", OP_INC));
op_printf("number of cells: %d should be: %d \n", count, g_ncell);
if (count != g_ncell)
op_printf("direct reduction FAILED\n");
else
op_printf("direct reduction PASSED\n");
op_timers(&cpu_t2, &wall_t2);
op_timing_output();
op_exit();
free(cell);
free(edge);
free(ecell);
free(bedge);
free(becell);
free(bound);
free(x);
free(q);
free(qold);
free(res);
free(adt);
}
int main(int argc, char **argv)
{
// OP initialisation
op_init(argc,argv,2);
//MPI for user I/O
int my_rank;
int comm_size;
MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
MPI_Comm_size(MPI_COMM_WORLD, &comm_size);
int *becell, *ecell, *bound, *bedge, *edge, *cell;
float *x, *q, *qold, *adt, *res;
int nnode,ncell,nedge,nbedge;
/**------------------------BEGIN I/O -------------------**/
char file[] = "new_grid.dat";
char file_out[] = "new_grid_out.h5";
/* read in grid from disk on root processor */
FILE *fp;
if ( (fp = fopen(file,"r")) == NULL) {
op_printf("can't open file %s\n",file); exit(-1);
}
int g_nnode,g_ncell,g_nedge,g_nbedge;
check_scan(fscanf(fp,"%d %d %d %d \n",&g_nnode, &g_ncell, &g_nedge, &g_nbedge), 4);
int *g_becell = 0, *g_ecell = 0, *g_bound = 0, *g_bedge = 0, *g_edge = 0, *g_cell = 0;
float *g_x = 0,*g_q = 0, *g_qold = 0, *g_adt = 0, *g_res = 0;
// set constants
op_printf("initialising flow field\n");
gam = 1.4f;
gm1 = gam - 1.0f;
cfl = 0.9f;
eps = 0.05f;
float mach = 0.4f;
float alpha = 3.0f*atan(1.0f)/45.0f;
float p = 1.0f;
float r = 1.0f;
float u = sqrt(gam*p/r)*mach;
float e = p/(r*gm1) + 0.5f*u*u;
qinf[0] = r;
qinf[1] = r*u;
qinf[2] = 0.0f;
qinf[3] = r*e;
op_printf("reading in grid \n");
op_printf("Global number of nodes, cells, edges, bedges = %d, %d, %d, %d\n"
,g_nnode,g_ncell,g_nedge,g_nbedge);
if(my_rank == MPI_ROOT) {
g_cell = (int *) malloc(4*g_ncell*sizeof(int));
g_edge = (int *) malloc(2*g_nedge*sizeof(int));
g_ecell = (int *) malloc(2*g_nedge*sizeof(int));
g_bedge = (int *) malloc(2*g_nbedge*sizeof(int));
g_becell = (int *) malloc( g_nbedge*sizeof(int));
g_bound = (int *) malloc( g_nbedge*sizeof(int));
g_x = (float *) malloc(2*g_nnode*sizeof(float));
g_q = (float *) malloc(4*g_ncell*sizeof(float));
g_qold = (float *) malloc(4*g_ncell*sizeof(float));
g_res = (float *) malloc(4*g_ncell*sizeof(float));
g_adt = (float *) malloc( g_ncell*sizeof(float));
for (int n=0; n<g_nnode; n++){
check_scan(fscanf(fp,"%f %f \n",&g_x[2*n], &g_x[2*n+1]), 2);
}
for (int n=0; n<g_ncell; n++) {
check_scan(fscanf(fp,"%d %d %d %d \n",&g_cell[4*n ], &g_cell[4*n+1],
&g_cell[4*n+2], &g_cell[4*n+3]), 4);
}
for (int n=0; n<g_nedge; n++) {
check_scan(fscanf(fp,"%d %d %d %d \n",&g_edge[2*n],&g_edge[2*n+1],
&g_ecell[2*n],&g_ecell[2*n+1]), 4);
}
for (int n=0; n<g_nbedge; n++) {
check_scan(fscanf(fp,"%d %d %d %d \n",&g_bedge[2*n],&g_bedge[2*n+1],
&g_becell[n],&g_bound[n]), 4);
}
//initialise flow field and residual
for (int n=0; n<g_ncell; n++) {
for (int m=0; m<4; m++) {
g_q[4*n+m] = qinf[m];
g_res[4*n+m] = 0.0f;
}
}
}
fclose(fp);
nnode = compute_local_size (g_nnode, comm_size, my_rank);
ncell = compute_local_size (g_ncell, comm_size, my_rank);
nedge = compute_local_size (g_nedge, comm_size, my_rank);
nbedge = compute_local_size (g_nbedge, comm_size, my_rank);
op_printf("Number of nodes, cells, edges, bedges on process %d = %d, %d, %d, %d\n"
,my_rank,nnode,ncell,nedge,nbedge);
/*Allocate memory to hold local sets, mapping tables and data*/
cell = (int *) malloc(4*ncell*sizeof(int));
edge = (int *) malloc(2*nedge*sizeof(int));
ecell = (int *) malloc(2*nedge*sizeof(int));
bedge = (int *) malloc(2*nbedge*sizeof(int));
becell = (int *) malloc( nbedge*sizeof(int));
bound = (int *) malloc( nbedge*sizeof(int));
x = (float *) malloc(2*nnode*sizeof(float));
q = (float *) malloc(4*ncell*sizeof(float));
qold = (float *) malloc(4*ncell*sizeof(float));
res = (float *) malloc(4*ncell*sizeof(float));
adt = (float *) malloc( ncell*sizeof(float));
/* scatter sets, mappings and data on sets*/
scatter_int_array(g_cell, cell, comm_size, g_ncell,ncell, 4);
scatter_int_array(g_edge, edge, comm_size, g_nedge,nedge, 2);
scatter_int_array(g_ecell, ecell, comm_size, g_nedge,nedge, 2);
scatter_int_array(g_bedge, bedge, comm_size, g_nbedge,nbedge, 2);
scatter_int_array(g_becell, becell, comm_size, g_nbedge,nbedge, 1);
scatter_int_array(g_bound, bound, comm_size, g_nbedge,nbedge, 1);
scatter_float_array(g_x, x, comm_size, g_nnode,nnode, 2);
scatter_float_array(g_q, q, comm_size, g_ncell,ncell, 4);
scatter_float_array(g_qold, qold, comm_size, g_ncell,ncell, 4);
scatter_float_array(g_res, res, comm_size, g_ncell,ncell, 4);
scatter_float_array(g_adt, adt, comm_size, g_ncell,ncell, 1);
/*Freeing memory allocated to gloabal arrays on rank 0
after scattering to all processes*/
if(my_rank == MPI_ROOT) {
free(g_cell);
free(g_edge);
free(g_ecell);
free(g_bedge);
free(g_becell);
free(g_bound);
free(g_x );
free(g_q);
free(g_qold);
free(g_adt);
free(g_res);
}
/**------------------------END I/O -----------------------**/
/* FIXME: It's not clear to the compiler that sth. is going on behind the
scenes here. Hence theses variables are reported as unused */
op_set nodes = op_decl_set(nnode, "nodes");
op_set edges = op_decl_set(nedge, "edges");
op_set bedges = op_decl_set(nbedge, "bedges");
op_set cells = op_decl_set(ncell, "cells");
op_map pedge = op_decl_map(edges, nodes,2,edge, "pedge");
op_map pecell = op_decl_map(edges, cells,2,ecell, "pecell");
op_map pbedge = op_decl_map(bedges,nodes,2,bedge, "pbedge");
op_map pbecell = op_decl_map(bedges,cells,1,becell,"pbecell");
op_map pcell = op_decl_map(cells, nodes,4,cell, "pcell");
op_dat p_bound = op_decl_dat(bedges,1,"int" ,bound,"p_bound");
op_dat p_x = op_decl_dat(nodes ,2,"float",x ,"p_x");
op_dat p_q = op_decl_dat(cells ,4,"float",q ,"p_q");
op_dat p_qold = op_decl_dat(cells ,4,"float",qold ,"p_qold");
op_dat p_adt = op_decl_dat(cells ,1,"float",adt ,"p_adt");
op_dat p_res = op_decl_dat(cells ,4,"float",res ,"p_res");
op_decl_const(1,"float",&gam );
op_decl_const(1,"float",&gm1 );
op_decl_const(1,"float",&cfl );
op_decl_const(1,"float",&eps );
op_decl_const(1,"float",&mach );
op_decl_const(1,"float",&alpha);
op_decl_const(4,"float",qinf );
op_dump_to_hdf5(file_out);
op_write_const_hdf5("gam", 1,"float",(char *)&gam, "new_grid_out.h5");
op_write_const_hdf5("gm1", 1,"float",(char *)&gm1, "new_grid_out.h5");
op_write_const_hdf5("cfl", 1,"float",(char *)&cfl, "new_grid_out.h5");
op_write_const_hdf5("eps", 1,"float",(char *)&eps, "new_grid_out.h5");
op_write_const_hdf5("mach", 1,"float",(char *)&mach, "new_grid_out.h5");
op_write_const_hdf5("alpha",1,"float",(char *)&alpha,"new_grid_out.h5");
op_write_const_hdf5("qinf", 4,"float",(char *)qinf, "new_grid_out.h5");
//create halos - for sanity check
op_halo_create();
op_exit();
}