loop_impl.c

enum class BindKind {
  teams,
  parallel,
  thread,
};

typedef void (*body_t)(int);

void loop_init_teams(int lb, int ub, body_t body, bool HasInnerLoop) {
  if (HasInnerLoop) {
    #pragma omp distribute
    for (int i = lb; i < ub; ++i)
      body(i);
  } else {
    #pragma omp distribute parallel for simd
    for (int i = lb; i < ub; ++i)
      body(i);
  }
}

void loop_init_parallel(int lb, int ub, body_t body, bool HasInnerLoop) {
  #pragma omp for simd
  for (int i = lb; i < ub; ++i)
    body(i);
}

void loop_init_thread(int lb, int ub, body_t body, bool HasInnerLoop) {
  if (HasInnerLoop) {
    #pragma omp parallel for simd num_threads(some_heuristic())
    for (int i = lb; i < ub; ++i)
      body(i);
  } else {
    #pragma omp parallel for simd
    for (int i = lb; i < ub; ++i)
      body(i);
  }
}

void loop_init(BindKind kind, int lb, int ub, body_t body, bool HasInnerLoop) {
  switch (kind) {
    case BindKind::teams:
      return loop_init_teams(lb, ub, body, HasInnerLoop);
    case BindKind::parallel:
      return loop_init_parallel(lb, ub, body, HasInnerLoop);
    default:
      return loop_init_thread(lb, ub, body, HasInnerLoop);
  }
}