Locality-Aware Task Scheduling and Data Distribution for OpenMP Programs on NUMA Systems and Manycore Processors

Listing 2

Program using the proposed interface for selecting data distribution policies.

int main(…) {

  …

  /* Allocate data */

  size_t sz = 8 * UNIT_SIZE;

  /* NUMA system: UNIT_SIZE = PAGE_SIZE */

  /* TILEPro64: UNIT_SIZE = PAGE_SIZE/CACHE_LINE_SIZE */

  void* A = omp_malloc(sz);

  void* B = omp_malloc(sz);

  void* C = omp_malloc_specific(sz, OMP_MALLOC_COARSE);

  void* D = omp_malloc_specific(sz, OMP_MALLOC_COARSE);

  void* E = omp_malloc_specific(sz, OMP_MALLOC_FINE);

  /* Initialize data */

  init(A, B, C, D, E, sz, …);

  /* Work in parallel */

  #pragma omp parallel

  {

  …

…

  export OMP_DATA_DISTRIBUTION=<standard    fine    coarse>

  <invoke program>