double lcod() {
// Allocate patches
float *patches = (float*)malloc(N*K*sizeof(float));
// Allocate patches_sr
float *patches_sr = (float*)malloc(N*F*sizeof(float));
// Allocate LR dictionary
float *dict_lr = (float*)malloc(K*C*sizeof(float));
// Allocate HR dictionary
float *dict_hr = (float*)malloc(C*F*sizeof(float));
// Allocate S matrix
float *s = (float*)malloc(C*C*sizeof(float));
// Allocate B
float *b = (float*)malloc(N*C*sizeof(float));
// Allocate Z
float *z = (float*)calloc(N*C, sizeof(float));
// Allocate Z_hat
float *z_hat = (float*)malloc(N*C*sizeof(float));
// Allocate tmp
float *tmp = (float*)malloc(N*C*sizeof(float));
// Perform Coordinate Descent
double start_time = omp_get_wtime();
dot(patches, dict_lr, b, N, K, C);
for (int t=0; t<T; t++) {
st(b, z_hat, N, C, THRESH);
int k;
if (t > 1) {
sub(z_hat, z, tmp, N, C);
k = greedy_k(tmp, N, C);
outer_update(tmp, s, b, N, C, k);
} else {
k = greedy_k(z_hat, N, C);
outer_update(z_hat, s, b, N, C, k);
}
slice_update(z, z_hat, N, C, k);
}
st(b, z, N, C, THRESH);
// Project back to HR space
dot(z, dict_hr, patches_sr, N, C, F);
double duration = omp_get_wtime();
// De-allocate
free(patches);
free(patches_sr);
free(dict_lr);
free(dict_hr);
free(s);
free(b);
free(z);
free(z_hat);
free(tmp);
return duration;
}
int main (int argc, char *argv[])
{
void inidat();
float ***array; /* array for grid */
int taskid, /* this task's unique id */
numtasks, /* number of tasks */
averow,rows,offset,extra, /* for sending rows of data */
dest, source, /* to - from for message send-receive */
left,right, /* neighbor tasks */
msgtype, /* for message types */
rc,start,end, /* misc */
i,x,y,z,it,size,t_sqrt; /* loop variables */
MPI_Status status;
MPI_Datatype dt,dt2;
MPI_Request req, req2,req3,req4,req5;
double t1,t2;
/* First, find out my taskid and how many tasks are running */
MPI_Init(&argc,&argv);
MPI_Comm_size(MPI_COMM_WORLD,&numtasks);
MPI_Comm_rank(MPI_COMM_WORLD,&taskid);
/*Set number of threads */
omp_set_num_threads(atoi(argv[1])); // Use n threads for all consecutive parallel regions
omp_set_nested(1);
if (taskid == 0)
{
//printf("Grid size: X= %d Y= %d Time steps= %d\n",NXPROB,NYPROB,STEPS);
t1 = MPI_Wtime();
}
i = 0;
while(i*i < (NXPROB*NYPROB)/numtasks)
i++;
size = i;
i = 0;
while(i*i<numtasks)
i++;
t_sqrt = i;
MPI_Type_contiguous(size+2,MPI_FLOAT, &dt);
MPI_Type_commit(&dt);
MPI_Type_vector(size+2,1,size+2,MPI_FLOAT,&dt2);
MPI_Type_commit(&dt2);
array = malloc(2*sizeof(float**));
for (i = 0;i<2;i++){
array[i] = malloc((2+size)*sizeof(float*));
array[i][0] = malloc(((2+size)*(2+size))*sizeof(float));
for (x = 1;x<2+size;x++){
array[i][x] = &(array[i][0][x*(2+size)]);
}
}
for (z=0; z<2; z++){
for (x=0; x<2+size; x++){
for (y=0; y<2+size; y++){
array[z][x][y] = 0.0;
}
}
}
z = 0;
inidat(NXPROB,NYPROB,array[z],size*(taskid/t_sqrt),size*(taskid%t_sqrt),size);
for (i = 1; i <= STEPS; i++)
{
if (taskid/t_sqrt != 0) //not first row
{
MPI_Isend(array[z][1],1,dt,taskid-t_sqrt,100, MPI_COMM_WORLD, &req);
MPI_Irecv(array[z][0],1,dt,taskid-t_sqrt,100, MPI_COMM_WORLD, &req2);
}
if (taskid/t_sqrt != t_sqrt-1) //not last row
{
MPI_Isend(array[z][size],1,dt,taskid+t_sqrt,100, MPI_COMM_WORLD, &req);
MPI_Irecv(array[z][size+1],1,dt,taskid+t_sqrt,100, MPI_COMM_WORLD, &req3);
}
if(taskid%t_sqrt != 0) //not last column
{
MPI_Isend(&array[z][0][1],1,dt2,taskid-1,100, MPI_COMM_WORLD, &req);
MPI_Irecv(&array[z][0][0],1,dt2,taskid-1,100, MPI_COMM_WORLD, &req4);
}
if(taskid%t_sqrt != t_sqrt-1) //not last column
{
MPI_Isend(&array[z][0][size],1,dt2,taskid+1,100, MPI_COMM_WORLD, &req);
MPI_Irecv(&array[z][0][size+1],1,dt2,taskid+1,100, MPI_COMM_WORLD, &req5);
}
inner_update(size,array[z],array[1-z]);
if (taskid/t_sqrt != 0)
MPI_Wait(&req2,&status);
if (taskid/t_sqrt != t_sqrt-1)
MPI_Wait(&req3,&status);
if(taskid%t_sqrt != 0)
MPI_Wait(&req4,&status);
if(taskid%t_sqrt != t_sqrt-1)
MPI_Wait(&req5,&status);
outer_update(size,taskid,t_sqrt,array[z],array[1-z]);
z = 1-z;
}
if (taskid == 0){
t2 = MPI_Wtime();
printf("MPI_Wtime measured: %1.2f\n", t2-t1);
}
for (i = 0;i<2;i++){
free(array[i][0]);
free(array[i]);
}
free(array);
MPI_Type_free(&dt);
MPI_Type_free(&dt2);
MPI_Finalize();
}