[OpenMP] Issues with Work-sharing region errors in ludcmp, fdtd-2d

[4fterlook]

In the ludcmp benchmark, the barrier is inside a parallel for loop which results in a error during compiling. There're other benchmarks that have similar problems. `#pragma scop b[0] = 1.0;

pragma omp parallel

{

pragma omp for private (j, k, w)

for (i = 0; i < _PB_N; i++)
{
  for (j = i+1; j <= _PB_N; j++)
    {
    w = A[j][i];
    for (k = 0; k < i; k++)
      w = w- A[j][k] * A[k][i];
      A[j][i] = w / A[i][i];
  }
  #pragma omp barrier
  for (j = i+1; j <= _PB_N; j++)
    {
      w = A[i+1][j];
      for (k = 0; k <= i; k++)
        w = w  - A[i+1][k] * A[k][j];
      A[i+1][j] = w;
    }
}
y[0] = b[0];
#pragma omp for private (j, w)
for (i = 1; i <= _PB_N; i++)
{
  w = b[i];
  for (j = 0; j < i; j++)
    w = w - A[i][j] * y[j];
  y[i] = w;
}
x[_PB_N] = y[_PB_N] / A[_PB_N][_PB_N];
#pragma omp for private (j, w)
for (i = 0; i <= _PB_N - 1; i++)
{
  w = y[_PB_N - 1 - (i)];
  for (j = _PB_N - i; j <= _PB_N; j++)
    w = w - A[_PB_N - 1 - i][j] * x[j];
  x[_PB_N - 1 - i] = w / A[_PB_N - 1 - (i)][_PB_N - 1-(i)];
}

}

pragma endscop`

This part of code just doesn't compile, it gives errors about work-sharing region may not be closely nested inside of work-sharing, critical, ordered, master or explicit task region

Is there any way to fix this issue? Is it because of a mistake in implementation or it's just tricks in openmp that I don't know?

[avacoder42]

yes, I have the same error. Did you find a solution ?