void *work_thread (void *lp) {
int task_id = *((int *) lp);
int begin, end;
struct timeval start, finish;
int i;
/*get the divided task*/
begin = (nsize * task_id) / task_num + 1;
end = (nsize * (task_id + 1)) / task_num;
if(task_id==0) gettimeofday (&start, NULL);
fprintf (stderr, "thread %d: begin %d, end %d\n", task_id, begin, end);
barrier (task_num);
/* initialization */
if (task_id == 0)
initRHS(nsize, begin, end);
barrier (task_num);
initResult(nsize, begin, end);
barrier (task_num);
/* Gauss Compute */
computeGauss(nsize, task_id);
barrier (task_num);
/* Gauss computation done */
if(task_id==0) {
/* Since there are dependencies in computing equation stage
the upper part need the results of upper part), it should be done by thread 0 solely. */
solveGauss(nsize);
gettimeofday (&finish, NULL);
printf ("Elapsed time: %.2f seconds\n",
(((finish.tv_sec * 1000000.0) + finish.tv_usec) -
((start.tv_sec * 1000000.0) + start.tv_usec)) / 1000000.0);
}
}
int main(int argc, char *argv[])
{
int i;
struct timeval start, finish;
double error;
pthread_attr_t attr;
pthread_t *tid;
int *id;
if (argc < 2) {
fprintf(stderr, "usage: %s <matrixfile>\n", argv[0]);
exit(-1);
}
// for getting the threads
if(argc == 3) {
task_num = strtol(argv[2], NULL, 10);
}
nsize = initMatrix(argv[1]);
initRHS(nsize);
initResult(nsize);
// create threads
id = (int *) malloc (sizeof (int) * task_num);
tid = (pthread_t *) malloc (sizeof (pthread_t) * task_num);
if (!id || !tid)
errexit ("out of shared memory");
pthread_attr_init (&attr);
pthread_attr_setscope (&attr, PTHREAD_SCOPE_SYSTEM);
for (i = 1; i < task_num; i++) {
id[i] = i;
pthread_create (&tid[i], &attr, work_thread, &id[i]);
}
id[0]=0;
work_thread(&id[0]);
// wait for all threads to finish
for (i = 1; i < task_num; i++)
pthread_join (tid[i], NULL);
solveGauss(nsize);
error = 0.0;
for (i = 0; i < nsize; i++) {
double error__ = (X__[i]==0.0) ? 1.0 : fabs((X[i]-X__[i])/X__[i]);
if (error < error__) {
error = error__;
}
}
fprintf(stdout, "Error: %e\n", error);
}
int main(int argc,char *argv[])
{
int i;
pthread_t *tid;
int *id;
double mhz;
hrtime_t hrcycle1, hrcycle2;
while((i = getopt(argc,argv,"s:p:")) != -1){
switch(i){
case 's':
{
int s;
s = atoi(optarg);
if (s > 0){
nsize = s;
} else {
fprintf(stderr,"Entered size is negative, hence using the default (%d)\n",(int)NSIZE);
}
}
break;
case 'p':
{
int p;
p = atoi(optarg);
if (p > 0){
task_num = p;
} else {
fprintf(stderr,"Entered task number is negative, hence using the default value: 1\n");
}
}
break;
default:
assert(0);
break;
}
}
#ifdef CHECK_CORRECTNESS
nsize = 3;
allocate_memory(nsize);
initCheck(nsize);
#else
allocate_memory(nsize);
initMatrix(nsize);
#endif
hrcycle1 = gethrcycle_x86();
/* create thread*/
id = (int *) malloc (sizeof (int) * task_num);
tid = (pthread_t *) malloc (sizeof (pthread_t *) * task_num);
if (!id || !tid)
errexit ("out of shared memory");
for(i=1; i<task_num; i++){
id[i] = i;
pthread_create(&tid[i], NULL, work_thread, &id[i]);
}
id[0] = 0;
work_thread(&id[0]);
for(i=1; i<task_num; i++){
pthread_join(tid[i], NULL);
}
#if VERIFY
solveGauss(nsize);
#endif
hrcycle2 = gethrcycle_x86();
mhz = getMHZ_x86();
printf("Here hrcycle = %lld, mhz = %f HZ\n", hrcycle2 - hrcycle1, mhz);
#if VERIFY
for(i = 0; i < nsize; i++)
printf("%6.5f %5.5f\n",B[i],C[i]);
#endif
return 0;
}
int main(int argc, char *argv[])
{
int i;
struct timeval start, finish;
double error;
pthread_attr_t attr;
pthread_t *tid;
int *id;
//Ensure that the program takes two parameters
// first param is the input matrix and the second is the number of processors for the parallel run
if (argc != 3) {
fprintf(stderr, "usage: %s <matrixfile> <#ofProcesses>\n", argv[0]);
exit(-1);
}
task_num = atoi(argv[2]);
nsize = initMatrix(argv[1]);
initRHS(nsize);
initResult(nsize);
gettimeofday(&start, 0);
// create threads
id = (int *) malloc (sizeof (int) * task_num);
tid = (pthread_t *) malloc (sizeof (pthread_t) * task_num);
if (!id || !tid)
errexit ("out of shared memory");
pthread_attr_init (&attr);
pthread_attr_setscope (&attr, PTHREAD_SCOPE_SYSTEM);
for (i = 1; i < task_num; i++) {
id[i] = i;
pthread_create (&tid[i], &attr, work_thread, &id[i]);
}
id[0]=0;
work_thread(&id[0]);
// wait for all threads to finish
for (i = 1; i < task_num; i++)
pthread_join (tid[i], NULL);
gettimeofday(&finish, 0);
solveGauss(nsize);
fprintf(stdout, "Time: %f seconds\n", (finish.tv_sec - start.tv_sec) + (finish.tv_usec - start.tv_usec)*0.000001);
error = 0.0;
for (i = 0; i < nsize; i++) {
double error__ = (X__[i]==0.0) ? 1.0 : fabs((X[i]-X__[i])/X__[i]);
if (error < error__) {
error = error__;
}
}
fprintf(stdout, "Error: %e\n", error);
//File output
// float timeOutput = ((finish.tv_sec - start.tv_sec) + (finish.tv_usec - start.tv_usec)*0.000001);
// FILE *pthreadTimingFile = fopen("pthread1Timing.txt", "a");
// fprintf(pthreadTimingFile, "%s %d %f %G\n", argv[1], task_num, timeOutput, error);
// fclose(pthreadTimingFile);
return 0;
}
