yade-dev team mailing list archive

["Sergei D." <sj2001@xxxxxxxxx>]

22.03.2011 14:58, Bruno Chareyre wrote:
> but parallelizing a serial algo the non-intrusive way is
> exactly what OpenMP is supposed to do
No. OpenMP is supposed to run a loops in a few threads. Not anymore.
Parallelizing of a _algo_ is a problem of a developers.
Proof in the attached file. Compile it and run (may be several times) by

g++ -fopenmp -lgomp insertion-sort-test.cpp -o insertion-sort-test
./insertion-sort-test

You should see few "insertion sort failed" and negative values on wrong 
places in resulting sorted vector.


--

Best regards,
Sergei D.

// Compilation:
// g++ -fopenmp -lgomp insertion-sort-test.cpp -o insertion-sort-test
#include <iostream>
#include <vector>
#include <omp.h>
void print_vector(const std::vector<int>& v)
{
	for (int i=0; i<v.size(); i++)
		std::cout << v[i] <<',';
	std::cout<<std::endl;
}
void check_sorting(std::vector<int>& v)
{
	for (int i=1; i<v.size(); i++) {
		if (v[i]<v[i-1]) {
			std::cout << "sorting failed" << std::endl;
			v[i]*=-1;
		}
	}
}
int main()
{
	std::vector<int> arr;
	for (int i=50; i>0; --i)
		arr.push_back(i);

	print_vector(arr);

	std::cout << "max_threads="<<omp_get_max_threads()<< std::endl;
	#pragma omp parallel for schedule(static)
	for (int i = 1; i < arr.size(); i++) 
	{
		std::cout << "hello from thread "<<omp_get_thread_num()<< std::endl;
		 int key = arr[i];
		 int j = i - 1;
		 while (j >= 0 && arr[j] > key)
		 {
			   arr[j + 1] = arr[j];
			   j--;
		 }
		 arr[j + 1] = key;
	}
	check_sorting(arr);
	print_vector(arr);
}