PWR006: Avoid privatization of read-only variables

Issue

Read-only variables should be shared instead of private to reduce memory consumption and unnecessary data copying.

Relevance

Since a read-only variable is never written to, it can be safely shared without any risk of race conditions. Sharing variables is more efficient than privatizing them from a memory perspective so it should be favored whenever possible.

Actions

Set the scope of the read-only variable to shared.

Code example

In the following code, arrays A and B are never written to. However, they are privatized and thus each thread will hold a copy of each array, effectively using more memory and taking more time to create private copies.

#define SIZE 5

void foo() {
  int A[SIZE] = {1, 2, 3, 4, 5};
  int B[SIZE] = {5, 4, 3, 2, 1};
  int sum[SIZE];

  #pragma omp parallel for shared(sum) firstprivate(A, B)
  for (int i = 0; i < SIZE; i++) {
    sum[i] = A[i] + B[i];
  }
}

To save memory, change their scope to shared. This may also prevent memory issues when using arrays, as codes may easily run out of memory for a high number of threads.

#define SIZE 5

void foo() {
  int A[SIZE] = {1, 2, 3, 4, 5};
  int B[SIZE] = {5, 4, 3, 2, 1};
  int sum[SIZE];

  #pragma omp parallel for shared(sum, A, B)
  for (int i = 0; i < SIZE; i++) {
    sum[i] = A[i] + B[i];
  }
}

Related resources

• PWR006 examples at GitHub 
• OpenMP 4.5 Complete Specifications, November 2015 [last checked May 2019]

References

• Privatization
• Overhead (memory overhead)