I have the following code
int main()
{
compute();
return 0;
}
void compute()
{
double Ptr;
vector <double> Vector;
for (int u=0; u<N; u++)
{
computeFor(u, Vector, Ptr);
Vector.push_back(EnergyPtr);
}
}
void computeFor(int i, vector<double> &List, int u, vector<double> &Vector, double &Ptr)
{
int Num = List.size();
int pos[2][Num];
for ( int i=0; i<Num; i++)
{
pos[0][i] = List[i]->position[0];
pos[1][i] = List[i]->position[1];
}
double tmp[Num];
for (int i=0; i<Num; i++)
{
tmp[i]=0;
}
double trg[Num];
for (int i=0; i<Num; i++)
{
trg[i]=0;
}
for (int i=0; i<Num; i++)
{
trg[i] +=3*tmp[i];
tmp[i]=0.0;
}
double total = 0.0;
for (int i=0; i<Num; i++)
{
trg[i] +=3*List[i]->getDec();
total +=trg[i];
}
std::stringstream ss;
ss << "out" << u << ".txt";
std::ofstream fout(ss.str().c_str());
fout<<"Total is "<<total<<endl;
Ptr = total;
}
I want to parallelize the above, which computes total and then prints the value in a for loop to txt files out0.txt, out1.txt, out2.txt, etc. I want to just create one txt file that prints out all the values by storing the values in a vector
How would I convert this to OpenMPI? I think I understand the pragma omp parallel and pragma omp critical part, but I'm confused as to how to use pragma for since there are multiple for loops:
void compute()
{
double Ptr;
vector <double> Vector;
int my_rank = omp_get_thread_num();
int thread_count = omp_get_num_threads();
#pragma omp parallel num_threads(thread_count)
{
compute(List, v, Vector, Ptr);
printf("Hello from thread %d of %d\n", my_rank, thread_count);
#pragma omp critical
Vector.push_back(Ptr);
}
/*for (int u=0; u<N; u++)
{
compute(u, Vector, Ptr);
Vector.push_back(EnergyPtr);
}*/
ofstream foutp("VectorParallel.txt");
for (vector<double>::iterator i = Vector.begin(); i < Vector.end(); i++)
{
foutp<<*i<<endl;
}
}
For the computeFor function, do I just enter # pragma omp parallel for numthreads(threadcount) once? I am also confused as to how it works. If I have 8 threads, I believe only one thread will execute computeFor as specified in #pragma omp parallel num_threads(thread_count), right? If so, are new threads created in # pragma omp for? Or do the 8 threads carry out the work in # pragma omp for for each call to the computeFor function?
void computeFor(int i, vector<double> &List, int u, vector<double> &Vector, double &Ptr)
{
int Num = List.size();
int pos[2][Num];
# pragma omp for
for ( int i=0; i<Num; i++)
{
pos[0][i] = List[i]->position[0];
pos[1][i] = List[i]->position[1];
}
double tmp[Num];
for (int i=0; i<Num; i++)
{
tmp[i]=0;
}
double trg[Num];
for (int i=0; i<Num; i++)
{
trg[i]=0;
}
for (int i=0; i<Num; i++)
{
trg[i] +=3*tmp[i];
tmp[i]=0.0;
}
double total = 0.0;
for (int i=0; i<Num; i++)
{
trg[i] +=3*List[i]->getDec();
total +=trg[i];
}
std::stringstream ss;
ss << "out" << u << ".txt";
std::ofstream fout(ss.str().c_str());
fout<<"Total is "<<total<<endl;
Ptr = total;
}
