OpenMP code C++ is slower thatn c++ - c++

Question

i have the following part of code, i run it on sample of N=3000, the c++ sequential code is faster by 3 seconds which is not good at all.
this code is filling the array jsd[N] with calculated values and i want to locate the maximum value and its location.
so
1- is this openmp conversion correct, and is there any better suggstion to make it more profissional
2- why it is slower that the equavilant c++ code, also the more threads i create the more it get slow.
thanks in advance
double maxval = 0;
int pos = -1;
double jsd[N];
#pragma omp parallel for num_threads(4)
for (int i = 0; i < N; i++) {
double Hl = obj.function1(sequenceVctr, i, LEFT);
double Hr = obj.function1(sequenceVctr, i, RIGHT);
jsd[i] = obj.function2(H, i + 1, N, Hl, Hr);
if (jsd[i] >= maxval) {
#pragma omp critical
{
maxval = jsd[i];
pos = i;
}
}
} // for
update:
here is the new code but still slow and get slower in more threads.
i update the code as following. but still get slower for more threads
double maxval = 0;
int pos = -1;
double jsd[N];
#pragma omp parallel num_threads(50)
for (int i = 0; i < N; i++) {
double Hl = obj.function1(sequenceVctr, i, LEFT);
double Hr = obj.function1(sequenceVctr, i, RIGHT);
jsd[i]= obj.function2(H, i + 1, N, Hl, Hr);
} // for
#pragma omp master
{
vector<double> jsd2 (jsd,jsd+N);
vector<double>::iterator jsditer;
jsditer = std::max_element(jsd2.begin(), jsd2.end());
maxval=*jsditer;
pos=std::distance(jsd2.begin(),jsditer) ;
// cout<<"pos"<<pos<<endl;
}
#pragma omp barrier

Answer 1

The first optimization I would suggest is to first compute all jsd values in the loop, then find the maximum element via std::max_element().
This way you are not forcing the threads to synchronise.
The second thing I would do is move over to Intel TBB instead of OpenMP and use parallel_reduce().
But the biggest question is, how complex are the objective functions you are evaluating.