Data Sharing Attributes¶

Shared Variable¶

All the threads have access to the shared variable.
By default, in the parallel region, all the variables are considered shared variables except the loop iteration counter variables.

Note

Shared variables should be handled carefully; otherwise, they cause race conditions in the program.

Examples: Shared variable

C/C++

#include <iostream>
#include <omp.h>

using namespace std;

int main()
{
  int N = 10;
  float *a;
  a  = (float*)malloc(sizeof(float) * N);

#pragma omp parallel for shared(a)
  for (int i = 0; i < N; i++)
    {
      a[i] = a[i] + i;
      cout << "value of a in the parallel region" << a[i] << endl;
    }

  for (int i = 0; i < N; i++)
    cout << "value of a after the parallel region " << a[i] << endl;

  return 0;
}

FORTRAN

program main
  use omp_lib
  implicit none

  real(8), dimension(:), allocatable :: a
  integer :: n, i
  n=10

  allocate(a(n))

  !$omp parallel shared(a)
  !$omp do
  do i = 1, n
      a(i) = a(i) + i
      print*, 'value of a in the parallel region', a(i)
  end do
  !$omp end do
  !$omp end parallel

  do i = 1, n
      a(i) = a(i) + i
      print*,'value of a after the parallel region', a(i)
  end do

  deallocate(a)

end program main

Questions

Does the value of vector a change after the parallel loop, if not why, think?
Do we really need to mention shared(a), is it necessary?

Private Variable¶

Each thread will have its own copy of the private variable.
The private variable is only accessible within the parallel region, not outside.
By default, the loop iteration counters are considered as private.
A change made by one thread is not visible to other threads.

Examples: Private variable

C/C++

#include <iostream>
#include <omp.h>

using namespace std;

int main()
{
  int N = 10;
  float *a, b, c;
  b = 1.0;
  c = 2.0;

  a  = (float*)malloc(sizeof(float) * N);

#pragma omp parallel for private(b,c)
  for (int i = 0; i < N; i++)
    {
      b = a[i] + i;
      c = b + 10 * i;
      cout << "value of c in the parallel region " << c << endl;
    }

  cout << "value of c after the parallel region " << c << endl;

  return 0;
}

FORTRAN

program main
  use omp_lib
  implicit none

  real(8), dimension(:), allocatable :: a
  real(8) :: b, c
  integer :: n, i
  n=10
  b=1.0
  c=2.0

  allocate(a(n))

  !$omp parallel private(b,c) shared(a)
  !$omp do
  do i = 1, n
      b = a(i) + i
      c = b + 10 * i
      print*, 'value of c in the parallel region', c
  end do
  !$omp end do
  !$omp end parallel
  print*, 'value of c after the parallel region', c

  deallocate(a)

end program main

Questions

What is the value of the variable c in the parallel region and after the parallel region?
After the parallel region, has variable c been updated or not?

Lastprivate¶

lastprivate is similar to a private clause.
But each thread will have an uninitialized copy of the variables passed as lastprivate.
At the end of the parallel loop or sections, the final variable value will be the last thread accessed value.

Examples: Lastprivate variable

C/C++

#include<iostream>
#include<omp.h>

using namespace std;

int main()
{
  int n = 10;
  int var = 5;
  omp_set_num_threads(10);
#pragma omp parallel for lastprivate(var)
  for(int i = 0; i < n; i++)
    {
      var += omp_get_thread_num();
      cout << " lastprivate in the parallel region " << var << endl;
    }
  cout << "lastprivate after the parallel region " << var <<endl;

  return 0;
}

FORTRAN

program main
  use omp_lib
  implicit none

  real(8) :: var
  integer :: n, i
  n = 10
  var = 5

  call omp_set_num_threads(10)

  !$omp parallel
  !$omp do lastprivate(var)
  do i = 1, n
     var  =  var + omp_get_thread_num()
     print*, 'lastprivate in the parallel region ', var
  end do
  !$omp end do
  !$omp end parallel

  print*, 'lastprivate after the parallel region ', var

end program main

Questions

What is the value of the variable var in the parallel region and after the parallel region?
Do you think the initial value of variable var is considered within the parallel region?

Firstprivate¶

firstprivate is similar to a private clause.
But each thread will have an initialized copy of the variables passed as firstprivate.
Available for parallel constructs, loops, sections and single constructs.

Examples: Firstprivate variable

C/C++

#include<iostream>
#include<omp.h>

using namespace std;

int main()
{
  int n = 10;
  int var = 5;
  omp_set_num_threads(10);
#pragma omp parallel for firstprivate(var)
  for(int i = 0; i < n; i++)
    {
      var += omp_get_thread_num();
      cout << " firstprivate in the parallel region " << var << endl;
    }
  cout << "firstprivate after the parallel region " << var <<endl;

  return 0;
}

FORTRAN

program main
  use omp_lib
  implicit none

  real(8) :: var
  integer :: n, i
  n = 10
  var = 5

  call omp_set_num_threads(10)

  !$omp parallel
  !$omp do firstprivate(var)
  do i = 1, n
     var  =  var + omp_get_thread_num()
     print*, 'firstprivate in the parallel region ', var
  end do
  !$omp end do
  !$omp end parallel

  print*, 'firstprivate after the parallel region ', var

end program main

Questions

What is the value of the variable var in the parallel region and after the parallel region?
Is variable var updated after the parallel region, if not why?