stencil_block_loops.hpp

/*
 * 4-D loop code.
 * Generated automatically from the following pseudo-code:
 *
 * omp loop(bnv,bxv) { loop(byv) { prefetch(L1) loop(bzv) { calc(cluster(bt)); } } }
 *
 */

// Number of iterations to get from begin_bnv to (but not including) end_bnv, stepping by step_bnv.
const idx_t num_bnv = ((end_bnv - begin_bnv) + (step_bnv - 1)) / step_bnv;

// Number of iterations to get from begin_bxv to (but not including) end_bxv, stepping by step_bxv.
const idx_t num_bxv = ((end_bxv - begin_bxv) + (step_bxv - 1)) / step_bxv;

// Number of iterations in loop collapsed across bnv, bxv dimensions.
const idx_t num_bnv_bxv = (idx_t)num_bnv * (idx_t)num_bxv;

// Distribute iterations among OpenMP threads.
//_Pragma("omp parallel for schedule(static,1) proc_bind(close)")
for (idx_t loop_index_bnv_bxv = 0; loop_index_bnv_bxv < num_bnv_bxv; loop_index_bnv_bxv++) {

  // Zero-based, unit-stride index var for bnv.
  idx_t index_bnv = loop_index_bnv_bxv / (num_bxv);

  // Zero-based, unit-stride index var for bxv.
  idx_t index_bxv = (loop_index_bnv_bxv) % num_bxv;

  // This value of index_bnv covers bnv from start_bnv to stop_bnv-1.
  const idx_t start_bnv = begin_bnv + (index_bnv * step_bnv);
  const idx_t stop_bnv = std::min(start_bnv + step_bnv, end_bnv);

  // This value of index_bxv covers bxv from start_bxv to stop_bxv-1.
  const idx_t start_bxv = begin_bxv + (index_bxv * step_bxv);
  const idx_t stop_bxv = std::min(start_bxv + step_bxv, end_bxv);

  // Number of iterations to get from begin_byv to (but not including) end_byv, stepping by step_byv.
  const idx_t num_byv = ((end_byv - begin_byv) + (step_byv - 1)) / step_byv;
  for (idx_t loop_index_byv = 0; loop_index_byv < num_byv; loop_index_byv++) {

    // Zero-based, unit-stride index var for byv.
    idx_t index_byv = loop_index_byv;

    // This value of index_byv covers byv from start_byv to stop_byv-1.
    const idx_t start_byv = begin_byv + (index_byv * step_byv);
    const idx_t stop_byv = std::min(start_byv + step_byv, end_byv);

    // Number of iterations to get from begin_bzv to (but not including) end_bzv, stepping by step_bzv.
    const idx_t num_bzv = ((end_bzv - begin_bzv) + (step_bzv - 1)) / step_bzv;

    // Prime prefetch to L1.
    _Pragma("noprefetch")
      for (idx_t loop_index_bzv = 0; loop_index_bzv < PFDL1; loop_index_bzv++) {

	// Zero-based, unit-stride index var for bzv.
	idx_t index_bzv = loop_index_bzv;

	// This value of index_bzv covers bzv from start_bzv to stop_bzv-1.
	const idx_t start_bzv = begin_bzv + (index_bzv * step_bzv);
	const idx_t stop_bzv = std::min(start_bzv + step_bzv, end_bzv);

	// Prefetch to L1.
	prefetch_cluster<L1>(context, bt, start_bnv, start_bxv, start_byv, start_bzv, stop_bnv, stop_bxv, stop_byv, stop_bzv);
      }

    // Computation loop.
    _Pragma("noprefetch")
    for (idx_t loop_index_bzv = 0; loop_index_bzv < num_bzv; loop_index_bzv++) {

      // Zero-based, unit-stride index var for bzv.
      idx_t index_bzv = loop_index_bzv;

      // This value of index_bzv covers bzv from start_bzv to stop_bzv-1.
      const idx_t start_bzv = begin_bzv + (index_bzv * step_bzv);
      const idx_t stop_bzv = std::min(start_bzv + step_bzv, end_bzv);

      // Computation.
      calc_cluster(context, bt, start_bnv, start_bxv, start_byv, start_bzv, stop_bnv, stop_bxv, stop_byv, stop_bzv);

      // Prefetch loop index var.
      idx_t loop_index_bzv_pfL1 = loop_index_bzv + PFDL1;

      // Zero-based, unit-stride prefetch index var for bzv.
      idx_t index_bzv_pfL1 = loop_index_bzv_pfL1;

      // This value of index_bzv_pfL1 covers bzv from start_bzv_pfL1 to stop_bzv_pfL1-1.
      const idx_t start_bzv_pfL1 = begin_bzv + (index_bzv_pfL1 * step_bzv);
      const idx_t stop_bzv_pfL1 = std::min(start_bzv_pfL1 + step_bzv, end_bzv);

      // Prefetch to L1.
      prefetch_cluster_bzv<L1>(context, bt, start_bnv, start_bxv, start_byv, start_bzv_pfL1, stop_bnv, stop_bxv, stop_byv, stop_bzv_pfL1);
    }
  }
 }
// End of generated code.