Why is my scanner code so slow with Pthread or OpenMP? - multithreading

Question

I want to scan one array and get the scanning result .
In My code , I make a disorder array with the shuffle function. then scan it get some number ( which is bigger than 60000 ). I split the array into threadnum-part ,every thread get one part to deal with. It seems that there are no shared memory between different thread.
So, why the two parallel code is so slow, since it is too slow ,I think the padding things maybe not the main reason. Could anyone give me some tips ? I am a beginner in parallel programming ,thank you.
here is my testing code with three part: serialization/Pthread/OpenMP. you can copy it and test on your own machine.
serialization code :
#include <stdlib.h>
#include <stdio.h>
#include <time.h>
#define N (65532)
#define threadnum 2
typedef struct{
int *mydata;
int *myres;
int val;
int datalen;
int reslen;
int tid;
}arg_t;
int randint(int i, int k){
int ret;
if(i> k){
int t= i;i=k;k=t;
}
ret = i + rand()%(k-i+1);
return ret;
}
void shuffle(int *org,int n){
int j=0,i;
for(i = n-1;i !=0;i--){
j = randint(0,i);
int t = org[j];org[j] = org[i];org[i] =t;
}
}
void scan(arg_t *arg){
int i;
arg->reslen=0;
for(i=0;i<arg->datalen;i++){
//if(arg->mydata[i] > arg->val){
arg->myres[arg->reslen] = arg->mydata[i];
arg->reslen +=(arg->mydata[i] > arg->val);
//}
}
}
int main(){
struct timeval begin,end;
int i,A[N],*res,reslen,perthread;
double diff_usec;
arg_t args[threadnum];
for(i=0;i<N;i++){
A[i] = i+1;
}
shuffle(A,N);
gettimeofday(&begin,NULL);
res = malloc(sizeof(int)*688);
reslen = 0;
for(i=0;i<N;i++){
//if(arg->mydata[i] > arg->val){
res[reslen] =A[i];
reslen +=(A[i] > 60000);
//}
}
gettimeofday(&end,NULL);
diff_usec = (((end).tv_sec*1000000L + (end).tv_usec)- ((begin).tv_sec*1000000L+(begin).tv_usec));
printf("\n%.4lf %d\n",diff_usec,reslen);
return 0;
}
Pthread code :
#include <stdlib.h>
#include <stdio.h>
#include <time.h>
#include <pthread.h>
#define N 65532
#define threadnum 2
typedef struct{
int *mydata;
int *myres;
int val;
int datalen;
int reslen;
int tid;
}arg_t;
int randint(int i, int k){
int ret;
if(i> k){
int t= i;i=k;k=t;
}
ret = i + rand()%(k-i+1);
return ret;
}
void shuffle(int *org,int n){
int j=0,i;
for(i = n-1;i !=0;i--){
j = randint(0,i);
int t = org[j];org[j] = org[i];org[i] =t;
}
}
void scan(arg_t *arg){
int i;
arg->reslen=0;
for(i=0;i<arg->datalen;i++){
//if(arg->mydata[i] > arg->val){
arg->myres[arg->reslen] = arg->mydata[i];
arg->reslen +=(arg->mydata[i] > arg->val);
//}
}
}
int get_time(struct timespec *begin , struct timespec *end){
return 1000 * (end -> tv_sec - begin -> tv_sec) + (end -> tv_sec - begin->tv_sec)/1000000;
}
int main(){
struct timeval begin,end;
//struct timespec begin,end;
int i,A[N],*res,reslen,perthread;
arg_t args[threadnum];
double diff_usec;
pthread_t tid[threadnum];
for(i=0;i<N;i++){
A[i] = i+1;
}
shuffle(A,N);
gettimeofday(&begin,NULL);
//clock_gettime(CLOCK_REALTIME,&begin);
perthread = N/threadnum;
for(i=0;i<threadnum;i++){
args[i].mydata = A+i*perthread;
args[i].myres = malloc(sizeof(int)*perthread);
args[i].datalen = (i == threadnum-1)?(N-(threadnum-1)*perthread):perthread;
args[i].val = 60000;
args[i].tid = i;
pthread_create(&tid[i],NULL,scan,(void*)&args[i]);
}
reslen =0;
for(i=0;i<threadnum;i++){
pthread_join(tid[i],NULL);
reslen += args[i].reslen;
}
res = malloc(sizeof(int)*reslen);
reslen=0;
for(i=0;i<threadnum;i++){
memcpy(res+reslen,args[i].myres,args[i].reslen);
reslen += args[i].reslen;
}
//clock_gettime(CLOCK_REALTIME,&end);
gettimeofday(&end,NULL);
diff_usec = (((end).tv_sec*1000000L + (end).tv_usec)- ((begin).tv_sec*1000000L+(begin).tv_usec));
//printf("\n%dms %d\n",get_time(&begin,&end),reslen);
printf("\n%.4lfms %d\n",diff_usec,reslen);
return 0;
}
OpenMP code :
#include <stdlib.h>
#include <stdio.h>
#include <omp.h>
#include <time.h>
#define N 65532
#define threadnum 16
typedef struct{
int *mydata;
int *myres;
int val;
int datalen;
int reslen;
}arg_t;
int get_time(struct timespec *begin , struct timespec *end){
return 1000 * (end -> tv_sec - begin -> tv_sec) + (end -> tv_sec - begin->tv_sec)/1000000;
}
int randint(int i, int k){
int ret;
if(i> k){
int t= i;i=k;k=t;
}
ret = i + rand()%(k-i+1);
return ret;
}
void shuffle(int *org,int n){
int j=0,i;
for(i = n-1;i !=0;i--){
j = randint(0,i);
int t = org[j];org[j] = org[i];org[i] =t;
}
}
void scan(arg_t *arg){
int i;
arg->reslen=0;
for(i=0;i<arg->datalen;i++){
//if(arg->mydata[i] > arg->val){
arg->myres[arg->reslen] = arg->mydata[i];
arg->reslen +=(arg->mydata[i] > arg->val);
//}
}
}
int main(){
struct timeval begin,end;
//struct timespec begin,end;
int i,A[N],*res,reslen,perthread;
double diff_usec;
arg_t args[threadnum];
for(i=0;i<N;i++){
A[i] = i+1;
}
shuffle(A,N);
gettimeofday(&begin,NULL);
//clock_gettime(CLOCK_REALTIME,&begin);
perthread = N/threadnum;
for(i=0;i<threadnum;i++){
args[i].mydata = A+i*perthread;
args[i].myres = malloc(sizeof(int)*perthread);
args[i].datalen = (i == threadnum-1)?(N-(threadnum-1)*perthread):perthread;
args[i].val = 60000;
}
omp_set_num_threads(threadnum);
#pragma omp parallel
{
scan(&args[omp_get_thread_num()]);
}
reslen =0;
for(i=0;i<threadnum;i++){
reslen += args[i].reslen;
}
res = malloc(sizeof(int)*reslen);
reslen=0;
for(i=0;i<threadnum;i++){
memcpy(res+reslen,args[i].myres,args[i].reslen);
reslen += args[i].reslen;
}
gettimeofday(&end,NULL);
//clock_gettime(CLOCK_REALTIME,&end);
diff_usec = (((end).tv_sec*1000000L + (end).tv_usec)- ((begin).tv_sec*1000000L+(begin).tv_usec));
printf("\n%.4lfms %d\n",diff_usec,reslen);
return 0;
}