int main(int argc, char const *argv[])
{
/* code */
double start, end;
start = gethrtime_x86();
empty();
end = gethrtime_x86();
printf("Time: %0.8f\n", end-start );
return 0;
}
void *work_thread(void *id){
int i,j,k;
double pivotVal;
double hrtime1, hrtime2;
int task_id = *((int *) id);
barrier(task_num); //wait for all threads to come and then start
if(task_id == 0){
hrtime1 = gethrtime_x86();
}
for(i=0; i<nsize; i++){
if(task_id == i % task_num){
getPivot(nsize,i); // select corresponding thread to find pivot in row
}
barrier(task_num); //wait for all threads finish
pivotVal = matrix[i][i];
for (j = i + 1 ; j < nsize; j++){
if(task_id == j % task_num){
pivotVal = matrix[j][i];
matrix[j][i] = 0.0;
for (k = i + 1 ; k < nsize; k++){
matrix[j][k] -= pivotVal * matrix[i][k];
}
B[j] -= pivotVal * B[i];
}
}
barrier(task_num);
}
hrtime2 = gethrtime_x86();
if(task_id==0){
printf("Hrtime = %f seconds\n", hrtime2 - hrtime1);
}
return NULL;
}
int main(int argc, char **argv){
setbuf(stdout,0);
int num = sysconf(_SC_NPROCESSORS_CONF);
printf("system has %d processor(s)\n", num);
cpu_set_t mask,get;
CPU_ZERO(&mask);
CPU_SET(0, &mask);
int i,j ;
int ite = 0;
ite_converge = 0;
hrtimer_converge = 0.0;
int retcode,allone;
void *retval;
double hrtimer_tmp,hrtime_start,hrtime_end,l2normtotal;
if ( argc != 4) {printf("[usage] SOR <matrix> <thread> <w>\n"); return 0;}
if (pthread_setaffinity_np(pthread_self(), sizeof(mask), &mask) < 0) {
fprintf(stderr, "set thread affinity failed\n");
}
thread_num = atoi(argv[2]);
w = atof(argv[3]);
nsize = initMatrix(argv[1]);
initRHS(nsize);
initResult(nsize);
int workload = (int)ceil((double)nsize/(double)thread_num) ;
pthread_t thread[thread_num];
int *seq = (int *)malloc(thread_num*sizeof(int));
for(i=0;i<thread_num;i++)
seq[i] = i;
shuffle(seq,thread_num);
for(j=1;j<thread_num;j++)
{
retcode = pthread_create(&thread[j], NULL, task, new int(j));
if (retcode != 0)
fprintf (stderr, "create thread failed %d\n", retcode);
}
sense_reverse_barrier(0); // start at the same time
hrtime_start = gethrtime_x86();
printf("Finish reading file. Start computing...\n");
while(converge == 0)
{
ticket_acquire(&flags[0].tlock);
if(flags[0].test == 1)
{
solve(0,workload-1,0,1);
flags[0].flag = 0;
flags[0].test = 0;
sense_reverse_barrier(0);
}
else
{
hrtimer_tmp = gethrtime_x86();
ite ++;
solve(0,workload-1,0,0);
hrtimer_compute[0] += gethrtime_x86() - hrtimer_tmp;
}
if(token == 0 && flags[0].flag == 1&&converge==0)
{
//observe all flags
allone = 1;
for(j=0;j<thread_num;j++)
{
allone = allone & flags[j].flag;
}
if(allone == 1)
{
ite_converge ++;
shuffle(seq,thread_num);
for(i=0;i<thread_num;i++)
if(seq[i] != 0)
{
ticket_acquire(&flags[seq[i]].tlock);
flags[seq[i]].test = 1;
ticket_release(&flags[seq[i]].tlock);
}
hrtimer_tmp = gethrtime_x86();
solve(0,workload-1,0,1);
hrtimer_converge += gethrtime_x86() - hrtimer_tmp;
sense_reverse_barrier(0);
l2normtotal = 0.0;
for(j=0;j<thread_num;j++)
l2normtotal += flags[j].l2norm;
if(l2normtotal < EPS)
{
converge = 1;
}
else
{
token = (token + 1) % thread_num;
}
}
}
ticket_release(&flags[0].tlock);
}
printf ("#%d finished\n",0);
iteration[0] = ite;
for(j=1;j<thread_num;j++)
{
retcode = pthread_join(thread[j], &retval);
iteration[j] = *(int *)retval;
if (retcode != 0)
fprintf (stderr, "join failed %d\n", retcode);
}
hrtime_end = gethrtime_x86();
CPU_ZERO(&get);
if (pthread_getaffinity_np(pthread_self(), sizeof(get), &get) < 0)
fprintf(stderr, "get thread affinity failed\n");
for (j = 0; j < 24; j++)
if (CPU_ISSET(j, &get))
printf("thread %d is running in processor %d sched_getcpu %d\n", -1, j,sched_getcpu());
#ifdef VERIFY
FILE *res;
res = fopen ("ASOR result", "w");
for (i = 0; i < nsize; i++) {
fprintf (res, "[%d] = %lf\n", i+1, X[i]);
}
#endif
double total = hrtime_end-hrtime_start;
printf("Total time:%.16lfs\n",total);
printf("Computation time:%.16lfs\n",max(thread_num,hrtimer_compute));
printf("Convergence time:%.16lfs\n",hrtimer_converge);
printf("Syn time:%.16lfs\n",total-max(thread_num,hrtimer_compute)-hrtimer_converge);
for(i=0;i<thread_num;i++)
printf("Iteration[%d]: %d\t\tComputation time:%.16lfs\n",i,iteration[i],hrtimer_compute[i]);
printf("#Converge iteration:%d\n",ite_converge);
printf("sizeof(cflag) = %lu\n",sizeof(cflag));
return 0;
}
static void *
task(void *arg)
{
int index = *((int *) arg);
cpu_set_t mask,get;
CPU_ZERO(&mask);
CPU_SET(index, &mask);
int i,j,k,loop,allone;
double l2normtotal;
int counter = 0;
int workload = (int)ceil((double)nsize/(double)thread_num) ;
int start = workload*(index);
int end = (index + 1)*workload-1;
double hrtimer_tmp;
int *seq = (int *)malloc(thread_num*sizeof(int));
for(i=0;i<thread_num;i++)
seq[i] = i;
if (pthread_setaffinity_np(pthread_self(), sizeof(mask), &mask) < 0) {
fprintf(stderr, "set thread affinity failed\n");
}
sense_reverse_barrier(index);
int ite = 0;
while(converge==0)
{
ticket_acquire(&flags[index].tlock);
if(flags[index].test == 1)
{
solve(start,end,index,1);
flags[index].flag = 0;
flags[index].test = 0;
sense_reverse_barrier(index);
}
else
{
hrtimer_tmp = gethrtime_x86();
ite ++;
solve(start,end,index,0);
hrtimer_compute[index] += gethrtime_x86() - hrtimer_tmp;
}
if(token == index && flags[index].flag == 1&&converge==0)
{
//observe all flags
allone = 1;
for(j=0;j<thread_num;j++)
{
allone = allone & flags[j].flag;
}
if(allone == 1)
{
ite_converge ++;
shuffle(seq,thread_num);
for(i=0;i<thread_num;i++)
if(seq[i] != index)
{
ticket_acquire(&flags[seq[i]].tlock);
flags[seq[i]].test = 1;
ticket_release(&flags[seq[i]].tlock);
}
hrtimer_tmp = gethrtime_x86();
solve(start,end,index,1);
hrtimer_converge += gethrtime_x86() - hrtimer_tmp;
sense_reverse_barrier(index);
l2normtotal = 0.0;
for(j=0;j<thread_num;j++)
{
l2normtotal += flags[j].l2norm;
}
if(l2normtotal < EPS)
{
converge = 1;
}
else
{
token = (token + 1) % thread_num;
}
}
}
ticket_release(&flags[index].tlock);
}
printf ("#%d finished\n",index);
CPU_ZERO(&get);
if (pthread_getaffinity_np(pthread_self(), sizeof(get), &get) < 0) {
fprintf(stderr, "get thread affinity failed\n");
}
for (j = 0; j < 24; j++) {
if (CPU_ISSET(j, &get)) {
printf("thread %d is running in processor %d sched_getcpu %d\n", index, j,sched_getcpu());
}
}
return new int(ite);
}
