C++ ops_register_args函数代码示例

// host stub function
void ops_par_loop_reset_field_kernel1_execute(ops_kernel_descriptor *desc) {
  ops_block block = desc->block;
  int dim = desc->dim;
  int *range = desc->range;
  ops_arg arg0 = desc->args[0];
  ops_arg arg1 = desc->args[1];
  ops_arg arg2 = desc->args[2];
  ops_arg arg3 = desc->args[3];

  // Timing
  double t1, t2, c1, c2;

  ops_arg args[4] = {arg0, arg1, arg2, arg3};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 4, range, 139))
    return;
#endif

  if (OPS_diags > 1) {
    OPS_kernels[139].count++;
    ops_timers_core(&c2, &t2);
  }

  // compute locally allocated range for the sub-block
  int start[3];
  int end[3];

  for (int n = 0; n < 3; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
  }

#ifdef OPS_DEBUG
  ops_register_args(args, "reset_field_kernel1");
#endif

  // set up initial pointers and exchange halos if necessary
  int base0 = args[0].dat->base_offset;
  double *__restrict__ density0 = (double *)(args[0].data + base0);

  int base1 = args[1].dat->base_offset;
  const double *__restrict__ density1 = (double *)(args[1].data + base1);

  int base2 = args[2].dat->base_offset;
  double *__restrict__ energy0 = (double *)(args[2].data + base2);

  int base3 = args[3].dat->base_offset;
  const double *__restrict__ energy1 = (double *)(args[3].data + base3);

  // initialize global variable with the dimension of dats
  int xdim0_reset_field_kernel1 = args[0].dat->size[0];
  int ydim0_reset_field_kernel1 = args[0].dat->size[1];
  int xdim1_reset_field_kernel1 = args[1].dat->size[0];
  int ydim1_reset_field_kernel1 = args[1].dat->size[1];
  int xdim2_reset_field_kernel1 = args[2].dat->size[0];
  int ydim2_reset_field_kernel1 = args[2].dat->size[1];
  int xdim3_reset_field_kernel1 = args[3].dat->size[0];
  int ydim3_reset_field_kernel1 = args[3].dat->size[1];

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[139].mpi_time += t1 - t2;
  }

#pragma omp parallel for collapse(2)
  for (int n_z = start[2]; n_z < end[2]; n_z++) {
    for (int n_y = start[1]; n_y < end[1]; n_y++) {
#ifdef intel
#pragma loop_count(10000)
#pragma omp simd aligned(density0, density1, energy0, energy1)
#else
#pragma simd
#endif
      for (int n_x = start[0]; n_x < end[0]; n_x++) {

        density0[OPS_ACC0(0, 0, 0)] = density1[OPS_ACC1(0, 0, 0)];
        energy0[OPS_ACC2(0, 0, 0)] = energy1[OPS_ACC3(0, 0, 0)];
      }
    }
  }
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[139].time += t2 - t1;
  }

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c1, &t1);
    OPS_kernels[139].mpi_time += t1 - t2;
    OPS_kernels[139].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[139].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[139].transfer += ops_compute_transfer(dim, start, end, &arg2);
    OPS_kernels[139].transfer += ops_compute_transfer(dim, start, end, &arg3);
  }
}

// host stub function
void ops_par_loop_update_halo_kernel1_fr2(char const *name, ops_block block, int dim, int* range,
 ops_arg arg0, ops_arg arg1, ops_arg arg2, ops_arg arg3,
 ops_arg arg4, ops_arg arg5, ops_arg arg6, ops_arg arg7) {

  //Timing
  double t1,t2,c1,c2;
  ops_timers_core(&c1,&t1);


  int  offs[8][3];
  ops_arg args[8] = { arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7};



  ops_timing_realloc(51,"update_halo_kernel1_fr2");
  OPS_kernels[51].count++;

  //compute locally allocated range for the sub-block

  int start[3];
  int end[3];

  #ifdef OPS_MPI
  sub_block_list sb = OPS_sub_block_list[block->index];
  if (!sb->owned) return;
  for ( int n=0; n<3; n++ ){
    start[n] = sb->decomp_disp[n];end[n] = sb->decomp_disp[n]+sb->decomp_size[n];
    if (start[n] >= range[2*n]) {
      start[n] = 0;
    }
    else {
      start[n] = range[2*n] - start[n];
    }
    if (sb->id_m[n]==MPI_PROC_NULL && range[2*n] < 0) start[n] = range[2*n];
    if (end[n] >= range[2*n+1]) {
      end[n] = range[2*n+1] - sb->decomp_disp[n];
    }
    else {
      end[n] = sb->decomp_size[n];
    }
    if (sb->id_p[n]==MPI_PROC_NULL && (range[2*n+1] > sb->decomp_disp[n]+sb->decomp_size[n]))
      end[n] += (range[2*n+1]-sb->decomp_disp[n]-sb->decomp_size[n]);
  }
  #else //OPS_MPI
  for ( int n=0; n<3; n++ ){
    start[n] = range[2*n];end[n] = range[2*n+1];
  }
  #endif //OPS_MPI
  #ifdef OPS_DEBUG
  ops_register_args(args, "update_halo_kernel1_fr2");
  #endif

  offs[0][0] = args[0].stencil->stride[0]*1;  //unit step in x dimension
  offs[0][1] = off3D(1, &start[0],
      &end[0],args[0].dat->size, args[0].stencil->stride) - offs[0][0];
  offs[0][2] = off3D(2, &start[0],
      &end[0],args[0].dat->size, args[0].stencil->stride) - offs[0][1] - offs[0][0];

  offs[1][0] = args[1].stencil->stride[0]*1;  //unit step in x dimension
  offs[1][1] = off3D(1, &start[0],
      &end[0],args[1].dat->size, args[1].stencil->stride) - offs[1][0];
  offs[1][2] = off3D(2, &start[0],
      &end[0],args[1].dat->size, args[1].stencil->stride) - offs[1][1] - offs[1][0];

  offs[2][0] = args[2].stencil->stride[0]*1;  //unit step in x dimension
  offs[2][1] = off3D(1, &start[0],
      &end[0],args[2].dat->size, args[2].stencil->stride) - offs[2][0];
  offs[2][2] = off3D(2, &start[0],
      &end[0],args[2].dat->size, args[2].stencil->stride) - offs[2][1] - offs[2][0];

  offs[3][0] = args[3].stencil->stride[0]*1;  //unit step in x dimension
  offs[3][1] = off3D(1, &start[0],
      &end[0],args[3].dat->size, args[3].stencil->stride) - offs[3][0];
  offs[3][2] = off3D(2, &start[0],
      &end[0],args[3].dat->size, args[3].stencil->stride) - offs[3][1] - offs[3][0];

  offs[4][0] = args[4].stencil->stride[0]*1;  //unit step in x dimension
  offs[4][1] = off3D(1, &start[0],
      &end[0],args[4].dat->size, args[4].stencil->stride) - offs[4][0];
  offs[4][2] = off3D(2, &start[0],
      &end[0],args[4].dat->size, args[4].stencil->stride) - offs[4][1] - offs[4][0];

  offs[5][0] = args[5].stencil->stride[0]*1;  //unit step in x dimension
  offs[5][1] = off3D(1, &start[0],
      &end[0],args[5].dat->size, args[5].stencil->stride) - offs[5][0];
  offs[5][2] = off3D(2, &start[0],
      &end[0],args[5].dat->size, args[5].stencil->stride) - offs[5][1] - offs[5][0];

  offs[6][0] = args[6].stencil->stride[0]*1;  //unit step in x dimension
  offs[6][1] = off3D(1, &start[0],
      &end[0],args[6].dat->size, args[6].stencil->stride) - offs[6][0];
  offs[6][2] = off3D(2, &start[0],
      &end[0],args[6].dat->size, args[6].stencil->stride) - offs[6][1] - offs[6][0];



  int off0_0 = offs[0][0];
  int off0_1 = offs[0][1];
  int off0_2 = offs[0][2];
//.........这里部分代码省略.........

// host stub function
void ops_par_loop_update_halo_kernel1_t1(char const *name, ops_block block,
                                         int dim, int *range, ops_arg arg0,
                                         ops_arg arg1, ops_arg arg2,
                                         ops_arg arg3, ops_arg arg4,
                                         ops_arg arg5, ops_arg arg6,
                                         ops_arg arg7) {

  // Timing
  double t1, t2, c1, c2;

  char *p_a[8];
  int offs[8][3];
  ops_arg args[8] = {arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 8, range, 14))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(14, "update_halo_kernel1_t1");
    OPS_kernels[14].count++;
    ops_timers_core(&c2, &t2);
  }

  // compute locally allocated range for the sub-block
  int start[3];
  int end[3];

#ifdef OPS_MPI
  sub_block_list sb = OPS_sub_block_list[block->index];
#endif
#ifdef OPS_DEBUG
  ops_register_args(args, "update_halo_kernel1_t1");
#endif

  int arg_idx[3];
  int arg_idx_base[3];
#ifdef OPS_MPI
  if (compute_ranges(args, 8, block, range, start, end, arg_idx) < 0)
    return;
#else  // OPS_MPI
  for (int n = 0; n < 3; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
    arg_idx[n] = start[n];
  }
#endif // OPS_MPI
  for (int n = 0; n < 3; n++) {
    arg_idx_base[n] = arg_idx[n];
  }
  offs[0][0] = args[0].stencil->stride[0] * 1; // unit step in x dimension
  offs[0][1] =
      off3D(1, &start[0], &end[0], args[0].dat->size, args[0].stencil->stride) -
      offs[0][0];
  offs[0][2] =
      off3D(2, &start[0], &end[0], args[0].dat->size, args[0].stencil->stride) -
      offs[0][1] - offs[0][0];

  offs[1][0] = args[1].stencil->stride[0] * 1; // unit step in x dimension
  offs[1][1] =
      off3D(1, &start[0], &end[0], args[1].dat->size, args[1].stencil->stride) -
      offs[1][0];
  offs[1][2] =
      off3D(2, &start[0], &end[0], args[1].dat->size, args[1].stencil->stride) -
      offs[1][1] - offs[1][0];

  offs[2][0] = args[2].stencil->stride[0] * 1; // unit step in x dimension
  offs[2][1] =
      off3D(1, &start[0], &end[0], args[2].dat->size, args[2].stencil->stride) -
      offs[2][0];
  offs[2][2] =
      off3D(2, &start[0], &end[0], args[2].dat->size, args[2].stencil->stride) -
      offs[2][1] - offs[2][0];

  offs[3][0] = args[3].stencil->stride[0] * 1; // unit step in x dimension
  offs[3][1] =
      off3D(1, &start[0], &end[0], args[3].dat->size, args[3].stencil->stride) -
      offs[3][0];
  offs[3][2] =
      off3D(2, &start[0], &end[0], args[3].dat->size, args[3].stencil->stride) -
      offs[3][1] - offs[3][0];

  offs[4][0] = args[4].stencil->stride[0] * 1; // unit step in x dimension
  offs[4][1] =
      off3D(1, &start[0], &end[0], args[4].dat->size, args[4].stencil->stride) -
      offs[4][0];
  offs[4][2] =
      off3D(2, &start[0], &end[0], args[4].dat->size, args[4].stencil->stride) -
      offs[4][1] - offs[4][0];

  offs[5][0] = args[5].stencil->stride[0] * 1; // unit step in x dimension
  offs[5][1] =
      off3D(1, &start[0], &end[0], args[5].dat->size, args[5].stencil->stride) -
      offs[5][0];
  offs[5][2] =
      off3D(2, &start[0], &end[0], args[5].dat->size, args[5].stencil->stride) -
      offs[5][1] - offs[5][0];

//.........这里部分代码省略.........

// host stub function
void ops_par_loop_update_halo_kernel3_plus_4_a(char const *name,
                                               ops_block block, int dim,
                                               int *range, ops_arg arg0,
                                               ops_arg arg1, ops_arg arg2) {

  // Timing
  double t1, t2, c1, c2;

  char *p_a[3];
  int offs[3][2];
  ops_arg args[3] = {arg0, arg1, arg2};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 3, range, 33))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(33, "update_halo_kernel3_plus_4_a");
    OPS_kernels[33].count++;
    ops_timers_core(&c2, &t2);
  }

  // compute locally allocated range for the sub-block
  int start[2];
  int end[2];

#ifdef OPS_MPI
  sub_block_list sb = OPS_sub_block_list[block->index];
#endif
#ifdef OPS_DEBUG
  ops_register_args(args, "update_halo_kernel3_plus_4_a");
#endif

  int arg_idx[2];
  int arg_idx_base[2];
#ifdef OPS_MPI
  if (compute_ranges(args, 3, block, range, start, end, arg_idx) < 0)
    return;
#else  // OPS_MPI
  for (int n = 0; n < 2; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
    arg_idx[n] = start[n];
  }
#endif // OPS_MPI
  for (int n = 0; n < 2; n++) {
    arg_idx_base[n] = arg_idx[n];
  }
  offs[0][0] = args[0].stencil->stride[0] * 1; // unit step in x dimension
  offs[0][1] =
      off2D(1, &start[0], &end[0], args[0].dat->size, args[0].stencil->stride) -
      offs[0][0];

  offs[1][0] = args[1].stencil->stride[0] * 1; // unit step in x dimension
  offs[1][1] =
      off2D(1, &start[0], &end[0], args[1].dat->size, args[1].stencil->stride) -
      offs[1][0];

  int off0_0 = offs[0][0];
  int off0_1 = offs[0][1];
  int dat0 = (OPS_soa ? args[0].dat->type_size : args[0].dat->elem_size);
  int off1_0 = offs[1][0];
  int off1_1 = offs[1][1];
  int dat1 = (OPS_soa ? args[1].dat->type_size : args[1].dat->elem_size);

  // set up initial pointers and exchange halos if necessary
  int base0 = args[0].dat->base_offset +
              (OPS_soa ? args[0].dat->type_size : args[0].dat->elem_size) *
                  start[0] * args[0].stencil->stride[0];
  base0 = base0 +
          (OPS_soa ? args[0].dat->type_size : args[0].dat->elem_size) *
              args[0].dat->size[0] * start[1] * args[0].stencil->stride[1];
  p_a[0] = (char *)args[0].data + base0;

  int base1 = args[1].dat->base_offset +
              (OPS_soa ? args[1].dat->type_size : args[1].dat->elem_size) *
                  start[0] * args[1].stencil->stride[0];
  base1 = base1 +
          (OPS_soa ? args[1].dat->type_size : args[1].dat->elem_size) *
              args[1].dat->size[0] * start[1] * args[1].stencil->stride[1];
  p_a[1] = (char *)args[1].data + base1;

  p_a[2] = args[2].data;

  // initialize global variable with the dimension of dats
  xdim0 = args[0].dat->size[0];
  xdim1 = args[1].dat->size[0];

  // Halo Exchanges
  ops_H_D_exchanges_host(args, 3);
  ops_halo_exchanges(args, 3, range);
  ops_H_D_exchanges_host(args, 3);

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[33].mpi_time += t1 - t2;
  }

//.........这里部分代码省略.........

// host stub function
void ops_par_loop_advec_cell_kernel2_zdir(char const *name, ops_block block,
                                          int dim, int *range, ops_arg arg0,
                                          ops_arg arg1, ops_arg arg2,
                                          ops_arg arg3) {

  // Timing
  double t1, t2, c1, c2;

  int offs[4][3];
  ops_arg args[4] = {arg0, arg1, arg2, arg3};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 4, range, 118))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(118, "advec_cell_kernel2_zdir");
    OPS_kernels[118].count++;
    ops_timers_core(&c1, &t1);
  }

#ifdef OPS_MPI
  sub_block_list sb = OPS_sub_block_list[block->index];
#endif

  // compute locally allocated range for the sub-block

  int start[3];
  int end[3];
  int arg_idx[3];

#ifdef OPS_MPI
  if (!sb->owned)
    return;
  for (int n = 0; n < 3; n++) {
    start[n] = sb->decomp_disp[n];
    end[n] = sb->decomp_disp[n] + sb->decomp_size[n];
    if (start[n] >= range[2 * n]) {
      start[n] = 0;
    } else {
      start[n] = range[2 * n] - start[n];
    }
    if (sb->id_m[n] == MPI_PROC_NULL && range[2 * n] < 0)
      start[n] = range[2 * n];
    if (end[n] >= range[2 * n + 1]) {
      end[n] = range[2 * n + 1] - sb->decomp_disp[n];
    } else {
      end[n] = sb->decomp_size[n];
    }
    if (sb->id_p[n] == MPI_PROC_NULL &&
        (range[2 * n + 1] > sb->decomp_disp[n] + sb->decomp_size[n]))
      end[n] += (range[2 * n + 1] - sb->decomp_disp[n] - sb->decomp_size[n]);
    if (end[n] < start[n])
      end[n] = start[n];
  }
#else
  for (int n = 0; n < 3; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
  }
#endif
#ifdef OPS_DEBUG
  ops_register_args(args, "advec_cell_kernel2_zdir");
#endif

  offs[0][0] = args[0].stencil->stride[0] * 1; // unit step in x dimension
  offs[0][1] =
      off3D(1, &start[0], &end[0], args[0].dat->size, args[0].stencil->stride) -
      offs[0][0];
  offs[0][2] =
      off3D(2, &start[0], &end[0], args[0].dat->size, args[0].stencil->stride) -
      offs[0][1] - offs[0][0];

  offs[1][0] = args[1].stencil->stride[0] * 1; // unit step in x dimension
  offs[1][1] =
      off3D(1, &start[0], &end[0], args[1].dat->size, args[1].stencil->stride) -
      offs[1][0];
  offs[1][2] =
      off3D(2, &start[0], &end[0], args[1].dat->size, args[1].stencil->stride) -
      offs[1][1] - offs[1][0];

  offs[2][0] = args[2].stencil->stride[0] * 1; // unit step in x dimension
  offs[2][1] =
      off3D(1, &start[0], &end[0], args[2].dat->size, args[2].stencil->stride) -
      offs[2][0];
  offs[2][2] =
      off3D(2, &start[0], &end[0], args[2].dat->size, args[2].stencil->stride) -
      offs[2][1] - offs[2][0];

  offs[3][0] = args[3].stencil->stride[0] * 1; // unit step in x dimension
  offs[3][1] =
      off3D(1, &start[0], &end[0], args[3].dat->size, args[3].stencil->stride) -
      offs[3][0];
  offs[3][2] =
      off3D(2, &start[0], &end[0], args[3].dat->size, args[3].stencil->stride) -
      offs[3][1] - offs[3][0];

  int off0_0 = offs[0][0];
//.........这里部分代码省略.........

// host stub function
void ops_par_loop_initialise_chunk_kernel_z_execute(
    ops_kernel_descriptor *desc) {
  ops_block block = desc->block;
  int dim = desc->dim;
  int *range = desc->range;
  ops_arg arg0 = desc->args[0];
  ops_arg arg1 = desc->args[1];
  ops_arg arg2 = desc->args[2];

  // Timing
  double t1, t2, c1, c2;

  ops_arg args[3] = {arg0, arg1, arg2};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 3, range, 5))
    return;
#endif

  if (OPS_diags > 1) {
    OPS_kernels[5].count++;
    ops_timers_core(&c2, &t2);
  }

  // compute locally allocated range for the sub-block
  int start[3];
  int end[3];

  for (int n = 0; n < 3; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
  }

#ifdef OPS_DEBUG
  ops_register_args(args, "initialise_chunk_kernel_z");
#endif

  // set up initial pointers and exchange halos if necessary
  int base0 = args[0].dat->base_offset;
  double *__restrict__ vertexz = (double *)(args[0].data + base0);

  int base1 = args[1].dat->base_offset;
  const int *__restrict__ zz = (int *)(args[1].data + base1);

  int base2 = args[2].dat->base_offset;
  double *__restrict__ vertexdz = (double *)(args[2].data + base2);

  // initialize global variable with the dimension of dats
  int xdim0_initialise_chunk_kernel_z = args[0].dat->size[0];
  int ydim0_initialise_chunk_kernel_z = args[0].dat->size[1];
  int xdim1_initialise_chunk_kernel_z = args[1].dat->size[0];
  int ydim1_initialise_chunk_kernel_z = args[1].dat->size[1];
  int xdim2_initialise_chunk_kernel_z = args[2].dat->size[0];
  int ydim2_initialise_chunk_kernel_z = args[2].dat->size[1];

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[5].mpi_time += t1 - t2;
  }

#pragma omp parallel for collapse(2)
  for (int n_z = start[2]; n_z < end[2]; n_z++) {
    for (int n_y = start[1]; n_y < end[1]; n_y++) {
#ifdef intel
#pragma loop_count(10000)
#pragma omp simd aligned(vertexz, zz, vertexdz)
#else
#pragma simd
#endif
      for (int n_x = start[0]; n_x < end[0]; n_x++) {

        int z_min = field.z_min - 2;

        double min_z, d_z;
        d_z = (grid.zmax - grid.zmin) / (double)grid.z_cells;
        min_z = grid.zmin + d_z * field.back;

        vertexz[OPS_ACC0(0, 0, 0)] =
            min_z + d_z * (zz[OPS_ACC1(0, 0, 0)] - z_min);
        vertexdz[OPS_ACC2(0, 0, 0)] = (double)d_z;
      }
    }
  }
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[5].time += t2 - t1;
  }

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c1, &t1);
    OPS_kernels[5].mpi_time += t1 - t2;
    OPS_kernels[5].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[5].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[5].transfer += ops_compute_transfer(dim, start, end, &arg2);
  }
}

// host stub function
void ops_par_loop_tea_leaf_init_zero2_kernel_execute(
    ops_kernel_descriptor *desc) {
  ops_block block = desc->block;
  int dim = desc->dim;
  int *range = desc->range;
  ops_arg arg0 = desc->args[0];
  ops_arg arg1 = desc->args[1];

  // Timing
  double t1, t2, c1, c2;

  ops_arg args[2] = {arg0, arg1};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 2, range, 16))
    return;
#endif

  if (OPS_diags > 1) {
    OPS_kernels[16].count++;
    ops_timers_core(&c2, &t2);
  }

  // compute locally allocated range for the sub-block
  int start[2];
  int end[2];

  for (int n = 0; n < 2; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
  }

#ifdef OPS_DEBUG
  ops_register_args(args, "tea_leaf_init_zero2_kernel");
#endif

  // set up initial pointers and exchange halos if necessary
  int base0 = args[0].dat->base_offset;
  double *__restrict__ p = (double *)(args[0].data + base0);

  int base1 = args[1].dat->base_offset;
  double *__restrict__ z = (double *)(args[1].data + base1);

  // initialize global variable with the dimension of dats
  int xdim0_tea_leaf_init_zero2_kernel = args[0].dat->size[0];
  int xdim1_tea_leaf_init_zero2_kernel = args[1].dat->size[0];

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[16].mpi_time += t1 - t2;
  }

#pragma omp parallel for
  for (int n_y = start[1]; n_y < end[1]; n_y++) {
#ifdef intel
#pragma loop_count(10000)
#pragma omp simd aligned(p, z)
#else
#pragma simd
#endif
    for (int n_x = start[0]; n_x < end[0]; n_x++) {

      p[OPS_ACC0(0, 0)] = 0.0;
      z[OPS_ACC1(0, 0)] = 0.0;
    }
  }
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[16].time += t2 - t1;
  }

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c1, &t1);
    OPS_kernels[16].mpi_time += t1 - t2;
    OPS_kernels[16].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[16].transfer += ops_compute_transfer(dim, start, end, &arg1);
  }
}

// host stub function
void ops_par_loop_advec_mom_kernel_y2(char const *name, ops_block block, int dim, int* range,
 ops_arg arg0, ops_arg arg1, ops_arg arg2, ops_arg arg3) {

  char *p_a[4];
  int  offs[4][2];
  ops_arg args[4] = { arg0, arg1, arg2, arg3};



  #ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args,4,range,18)) return;
  #endif

  ops_timing_realloc(18,"advec_mom_kernel_y2");
  OPS_kernels[18].count++;

  //compute locally allocated range for the sub-block
  int start[2];
  int end[2];

  #ifdef OPS_MPI
  sub_block_list sb = OPS_sub_block_list[block->index];
  if (!sb->owned) return;
  for ( int n=0; n<2; n++ ){
    start[n] = sb->decomp_disp[n];end[n] = sb->decomp_disp[n]+sb->decomp_size[n];
    if (start[n] >= range[2*n]) {
      start[n] = 0;
    }
    else {
      start[n] = range[2*n] - start[n];
    }
    if (sb->id_m[n]==MPI_PROC_NULL && range[2*n] < 0) start[n] = range[2*n];
    if (end[n] >= range[2*n+1]) {
      end[n] = range[2*n+1] - sb->decomp_disp[n];
    }
    else {
      end[n] = sb->decomp_size[n];
    }
    if (sb->id_p[n]==MPI_PROC_NULL && (range[2*n+1] > sb->decomp_disp[n]+sb->decomp_size[n]))
      end[n] += (range[2*n+1]-sb->decomp_disp[n]-sb->decomp_size[n]);
  }
  #else //OPS_MPI
  for ( int n=0; n<2; n++ ){
    start[n] = range[2*n];end[n] = range[2*n+1];
  }
  #endif //OPS_MPI
  #ifdef OPS_DEBUG
  ops_register_args(args, "advec_mom_kernel_y2");
  #endif

  offs[0][0] = args[0].stencil->stride[0]*1;  //unit step in x dimension
  offs[0][1] = off2D(1, &start[0],
      &end[0],args[0].dat->size, args[0].stencil->stride) - offs[0][0];

  offs[1][0] = args[1].stencil->stride[0]*1;  //unit step in x dimension
  offs[1][1] = off2D(1, &start[0],
      &end[0],args[1].dat->size, args[1].stencil->stride) - offs[1][0];

  offs[2][0] = args[2].stencil->stride[0]*1;  //unit step in x dimension
  offs[2][1] = off2D(1, &start[0],
      &end[0],args[2].dat->size, args[2].stencil->stride) - offs[2][0];

  offs[3][0] = args[3].stencil->stride[0]*1;  //unit step in x dimension
  offs[3][1] = off2D(1, &start[0],
      &end[0],args[3].dat->size, args[3].stencil->stride) - offs[3][0];



  //Timing
  double t1,t2,c1,c2;
  ops_timers_core(&c2,&t2);

  int off0_0 = offs[0][0];
  int off0_1 = offs[0][1];
  int dat0 = args[0].dat->elem_size;
  int off1_0 = offs[1][0];
  int off1_1 = offs[1][1];
  int dat1 = args[1].dat->elem_size;
  int off2_0 = offs[2][0];
  int off2_1 = offs[2][1];
  int dat2 = args[2].dat->elem_size;
  int off3_0 = offs[3][0];
  int off3_1 = offs[3][1];
  int dat3 = args[3].dat->elem_size;

  //set up initial pointers and exchange halos if necessary
  int d_m[OPS_MAX_DIM];
  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[0].dat->d_m[d] + OPS_sub_dat_list[args[0].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[0].dat->d_m[d];
  #endif //OPS_MPI
  int base0 = dat0 * 1 * 
    (start[0] * args[0].stencil->stride[0] - args[0].dat->base[0] - d_m[0]);
  base0 = base0+ dat0 *
    args[0].dat->size[0] *
    (start[1] * args[0].stencil->stride[1] - args[0].dat->base[1] - d_m[1]);
  p_a[0] = (char *)args[0].data + base0;

//.........这里部分代码省略.........

// host stub function
void ops_par_loop_ideal_gas_kernel_execute(ops_kernel_descriptor *desc) {
  ops_block block = desc->block;
  int dim = desc->dim;
  int *range = desc->range;
  ops_arg arg0 = desc->args[0];
  ops_arg arg1 = desc->args[1];
  ops_arg arg2 = desc->args[2];
  ops_arg arg3 = desc->args[3];

  // Timing
  double t1, t2, c1, c2;

  ops_arg args[4] = {arg0, arg1, arg2, arg3};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 4, range, 8))
    return;
#endif

  if (OPS_diags > 1) {
    OPS_kernels[8].count++;
    ops_timers_core(&c2, &t2);
  }

  // compute locally allocated range for the sub-block
  int start[2];
  int end[2];

  for (int n = 0; n < 2; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
  }

#ifdef OPS_DEBUG
  ops_register_args(args, "ideal_gas_kernel");
#endif

  // set up initial pointers and exchange halos if necessary
  int base0 = args[0].dat->base_offset;
  const double *__restrict__ density = (double *)(args[0].data + base0);

  int base1 = args[1].dat->base_offset;
  const double *__restrict__ energy = (double *)(args[1].data + base1);

  int base2 = args[2].dat->base_offset;
  double *__restrict__ pressure = (double *)(args[2].data + base2);

  int base3 = args[3].dat->base_offset;
  double *__restrict__ soundspeed = (double *)(args[3].data + base3);

  // initialize global variable with the dimension of dats
  int xdim0_ideal_gas_kernel = args[0].dat->size[0];
  int xdim1_ideal_gas_kernel = args[1].dat->size[0];
  int xdim2_ideal_gas_kernel = args[2].dat->size[0];
  int xdim3_ideal_gas_kernel = args[3].dat->size[0];

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[8].mpi_time += t1 - t2;
  }

#pragma omp parallel for
  for (int n_y = start[1]; n_y < end[1]; n_y++) {
#ifdef intel
#pragma loop_count(10000)
#pragma omp simd aligned(density, energy, pressure, soundspeed)
#else
#pragma simd
#endif
    for (int n_x = start[0]; n_x < end[0]; n_x++) {

      double sound_speed_squared, v, pressurebyenergy, pressurebyvolume;

      v = 1.0 / density[OPS_ACC0(0, 0)];
      pressure[OPS_ACC2(0, 0)] =
          (1.4 - 1.0) * density[OPS_ACC0(0, 0)] * energy[OPS_ACC1(0, 0)];
      pressurebyenergy = (1.4 - 1.0) * density[OPS_ACC0(0, 0)];
      pressurebyvolume =
          -1 * density[OPS_ACC0(0, 0)] * pressure[OPS_ACC2(0, 0)];
      sound_speed_squared =
          v * v *
          (pressure[OPS_ACC2(0, 0)] * pressurebyenergy - pressurebyvolume);
      soundspeed[OPS_ACC3(0, 0)] = sqrt(sound_speed_squared);
    }
  }
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[8].time += t2 - t1;
  }

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c1, &t1);
    OPS_kernels[8].mpi_time += t1 - t2;
    OPS_kernels[8].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[8].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[8].transfer += ops_compute_transfer(dim, start, end, &arg2);
    OPS_kernels[8].transfer += ops_compute_transfer(dim, start, end, &arg3);
  }
//.........这里部分代码省略.........

// host stub function
void ops_par_loop_advec_mom_kernel_post_pre_advec_z_execute(
    ops_kernel_descriptor *desc) {
  ops_block block = desc->block;
  int dim = desc->dim;
  int *range = desc->range;
  ops_arg arg0 = desc->args[0];
  ops_arg arg1 = desc->args[1];
  ops_arg arg2 = desc->args[2];
  ops_arg arg3 = desc->args[3];
  ops_arg arg4 = desc->args[4];

  // Timing
  double t1, t2, c1, c2;

  ops_arg args[5] = {arg0, arg1, arg2, arg3, arg4};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 5, range, 136))
    return;
#endif

  if (OPS_diags > 1) {
    OPS_kernels[136].count++;
    ops_timers_core(&c2, &t2);
  }

  // compute locally allocated range for the sub-block
  int start[3];
  int end[3];

  for (int n = 0; n < 3; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
  }

#ifdef OPS_DEBUG
  ops_register_args(args, "advec_mom_kernel_post_pre_advec_z");
#endif

  // set up initial pointers and exchange halos if necessary
  int base0 = args[0].dat->base_offset;
  double *__restrict__ node_mass_post = (double *)(args[0].data + base0);

  int base1 = args[1].dat->base_offset;
  const double *__restrict__ post_vol = (double *)(args[1].data + base1);

  int base2 = args[2].dat->base_offset;
  const double *__restrict__ density1 = (double *)(args[2].data + base2);

  int base3 = args[3].dat->base_offset;
  double *__restrict__ node_mass_pre = (double *)(args[3].data + base3);

  int base4 = args[4].dat->base_offset;
  const double *__restrict__ node_flux = (double *)(args[4].data + base4);

  // initialize global variable with the dimension of dats
  int xdim0_advec_mom_kernel_post_pre_advec_z = args[0].dat->size[0];
  int ydim0_advec_mom_kernel_post_pre_advec_z = args[0].dat->size[1];
  int xdim1_advec_mom_kernel_post_pre_advec_z = args[1].dat->size[0];
  int ydim1_advec_mom_kernel_post_pre_advec_z = args[1].dat->size[1];
  int xdim2_advec_mom_kernel_post_pre_advec_z = args[2].dat->size[0];
  int ydim2_advec_mom_kernel_post_pre_advec_z = args[2].dat->size[1];
  int xdim3_advec_mom_kernel_post_pre_advec_z = args[3].dat->size[0];
  int ydim3_advec_mom_kernel_post_pre_advec_z = args[3].dat->size[1];
  int xdim4_advec_mom_kernel_post_pre_advec_z = args[4].dat->size[0];
  int ydim4_advec_mom_kernel_post_pre_advec_z = args[4].dat->size[1];

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[136].mpi_time += t1 - t2;
  }

#pragma omp parallel for collapse(2)
  for (int n_z = start[2]; n_z < end[2]; n_z++) {
    for (int n_y = start[1]; n_y < end[1]; n_y++) {
#ifdef intel
#pragma loop_count(10000)
#pragma omp simd aligned(node_mass_post, post_vol, density1, node_mass_pre,    \
                         node_flux)
#else
#pragma simd
#endif
      for (int n_x = start[0]; n_x < end[0]; n_x++) {

        node_mass_post[OPS_ACC0(0, 0, 0)] =
            0.125 *
            (density1[OPS_ACC2(0, -1, 0)] * post_vol[OPS_ACC1(0, -1, 0)] +
             density1[OPS_ACC2(0, 0, 0)] * post_vol[OPS_ACC1(0, 0, 0)] +
             density1[OPS_ACC2(-1, -1, 0)] * post_vol[OPS_ACC1(-1, -1, 0)] +
             density1[OPS_ACC2(-1, 0, 0)] * post_vol[OPS_ACC1(-1, 0, 0)] +
             density1[OPS_ACC2(0, -1, -1)] * post_vol[OPS_ACC1(0, -1, -1)] +
             density1[OPS_ACC2(0, 0, -1)] * post_vol[OPS_ACC1(0, 0, -1)] +
             density1[OPS_ACC2(-1, -1, -1)] * post_vol[OPS_ACC1(-1, -1, -1)] +
             density1[OPS_ACC2(-1, 0, -1)] * post_vol[OPS_ACC1(-1, 0, -1)]);

        node_mass_pre[OPS_ACC3(0, 0, 0)] = node_mass_post[OPS_ACC0(0, 0, 0)] -
                                           node_flux[OPS_ACC4(0, 0, -1)] +
                                           node_flux[OPS_ACC4(0, 0, 0)];
      }
//.........这里部分代码省略.........

// host stub function
void ops_par_loop_advec_mom_kernel1_z_nonvector_execute(
    ops_kernel_descriptor *desc) {
  ops_block block = desc->block;
  int dim = desc->dim;
  int *range = desc->range;
  ops_arg arg0 = desc->args[0];
  ops_arg arg1 = desc->args[1];
  ops_arg arg2 = desc->args[2];
  ops_arg arg3 = desc->args[3];
  ops_arg arg4 = desc->args[4];

  // Timing
  double t1, t2, c1, c2;

  ops_arg args[5] = {arg0, arg1, arg2, arg3, arg4};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 5, range, 137))
    return;
#endif

  if (OPS_diags > 1) {
    OPS_kernels[137].count++;
    ops_timers_core(&c2, &t2);
  }

  // compute locally allocated range for the sub-block
  int start[3];
  int end[3];

  for (int n = 0; n < 3; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
  }

#ifdef OPS_DEBUG
  ops_register_args(args, "advec_mom_kernel1_z_nonvector");
#endif

  // set up initial pointers and exchange halos if necessary
  int base0 = args[0].dat->base_offset;
  const double *__restrict__ node_flux = (double *)(args[0].data + base0);

  int base1 = args[1].dat->base_offset;
  const double *__restrict__ node_mass_pre = (double *)(args[1].data + base1);

  int base2 = args[2].dat->base_offset;
  double *__restrict__ mom_flux = (double *)(args[2].data + base2);

  int base3 = args[3].dat->base_offset;
  const double *__restrict__ celldz = (double *)(args[3].data + base3);

  int base4 = args[4].dat->base_offset;
  const double *__restrict__ vel1 = (double *)(args[4].data + base4);

  // initialize global variable with the dimension of dats
  int xdim0_advec_mom_kernel1_z_nonvector = args[0].dat->size[0];
  int ydim0_advec_mom_kernel1_z_nonvector = args[0].dat->size[1];
  int xdim1_advec_mom_kernel1_z_nonvector = args[1].dat->size[0];
  int ydim1_advec_mom_kernel1_z_nonvector = args[1].dat->size[1];
  int xdim2_advec_mom_kernel1_z_nonvector = args[2].dat->size[0];
  int ydim2_advec_mom_kernel1_z_nonvector = args[2].dat->size[1];
  int xdim3_advec_mom_kernel1_z_nonvector = args[3].dat->size[0];
  int ydim3_advec_mom_kernel1_z_nonvector = args[3].dat->size[1];
  int xdim4_advec_mom_kernel1_z_nonvector = args[4].dat->size[0];
  int ydim4_advec_mom_kernel1_z_nonvector = args[4].dat->size[1];

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[137].mpi_time += t1 - t2;
  }

#pragma omp parallel for collapse(2)
  for (int n_z = start[2]; n_z < end[2]; n_z++) {
    for (int n_y = start[1]; n_y < end[1]; n_y++) {
#ifdef intel
#pragma loop_count(10000)
#pragma omp simd aligned(node_flux, node_mass_pre, mom_flux, celldz, vel1)
#else
#pragma simd
#endif
      for (int n_x = start[0]; n_x < end[0]; n_x++) {

        double sigma, wind, width;
        double vdiffuw, vdiffdw, auw, adw, limiter;
        int upwind, donor, downwind, dif;
        double advec_vel_temp;

        if ((node_flux[OPS_ACC0(0, 0, 0)]) < 0.0) {
          upwind = 2;
          donor = 1;
          downwind = 0;
          dif = donor;
        } else {
          upwind = -1;
          donor = 0;
          downwind = 1;
          dif = upwind;
        }
//.........这里部分代码省略.........

// host stub function
void ops_par_loop_initialise_chunk_kernel_cellx_execute(
    ops_kernel_descriptor *desc) {
  ops_block block = desc->block;
  int dim = desc->dim;
  int *range = desc->range;
  ops_arg arg0 = desc->args[0];
  ops_arg arg1 = desc->args[1];
  ops_arg arg2 = desc->args[2];

  // Timing
  double t1, t2, c1, c2;

  ops_arg args[3] = {arg0, arg1, arg2};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 3, range, 12))
    return;
#endif

  if (OPS_diags > 1) {
    OPS_kernels[12].count++;
    ops_timers_core(&c2, &t2);
  }

  // compute locally allocated range for the sub-block
  int start[2];
  int end[2];

  for (int n = 0; n < 2; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
  }

#ifdef OPS_DEBUG
  ops_register_args(args, "initialise_chunk_kernel_cellx");
#endif

  // set up initial pointers and exchange halos if necessary
  int base0 = args[0].dat->base_offset;
  const double *__restrict__ vertexx = (double *)(args[0].data + base0);

  int base1 = args[1].dat->base_offset;
  double *__restrict__ cellx = (double *)(args[1].data + base1);

  int base2 = args[2].dat->base_offset;
  double *__restrict__ celldx = (double *)(args[2].data + base2);

  // initialize global variable with the dimension of dats
  int xdim0_initialise_chunk_kernel_cellx = args[0].dat->size[0];
  int xdim1_initialise_chunk_kernel_cellx = args[1].dat->size[0];
  int xdim2_initialise_chunk_kernel_cellx = args[2].dat->size[0];

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[12].mpi_time += t1 - t2;
  }

#pragma omp parallel for
  for (int n_y = start[1]; n_y < end[1]; n_y++) {
#ifdef intel
#pragma loop_count(10000)
#pragma omp simd aligned(vertexx, cellx, celldx)
#else
#pragma simd
#endif
    for (int n_x = start[0]; n_x < end[0]; n_x++) {

      double d_x;
      d_x = (grid.xmax - grid.xmin) / (double)grid.x_cells;

      cellx[OPS_ACC1(0, 0)] =
          0.5 * (vertexx[OPS_ACC0(0, 0)] + vertexx[OPS_ACC0(1, 0)]);
      celldx[OPS_ACC2(0, 0)] = d_x;
    }
  }
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[12].time += t2 - t1;
  }

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c1, &t1);
    OPS_kernels[12].mpi_time += t1 - t2;
    OPS_kernels[12].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[12].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[12].transfer += ops_compute_transfer(dim, start, end, &arg2);
  }
}

// host stub function
void ops_par_loop_PdV_kernel_predict(char const *name, ops_block block, int dim, int* range,
 ops_arg arg0, ops_arg arg1, ops_arg arg2, ops_arg arg3,
 ops_arg arg4, ops_arg arg5, ops_arg arg6, ops_arg arg7,
 ops_arg arg8, ops_arg arg9, ops_arg arg10, ops_arg arg11,
 ops_arg arg12, ops_arg arg13) {

  char *p_a[14];
  int  offs[14][3];
  ops_arg args[14] = { arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7, arg8, arg9, arg10, arg11, arg12, arg13};



  ops_timing_realloc(5,"PdV_kernel_predict");
  OPS_kernels[5].count++;

  //compute locally allocated range for the sub-block
  int start[3];
  int end[3];

  #ifdef OPS_MPI
  sub_block_list sb = OPS_sub_block_list[block->index];
  if (!sb->owned) return;
  for ( int n=0; n<3; n++ ){
    start[n] = sb->decomp_disp[n];end[n] = sb->decomp_disp[n]+sb->decomp_size[n];
    if (start[n] >= range[2*n]) {
      start[n] = 0;
    }
    else {
      start[n] = range[2*n] - start[n];
    }
    if (sb->id_m[n]==MPI_PROC_NULL && range[2*n] < 0) start[n] = range[2*n];
    if (end[n] >= range[2*n+1]) {
      end[n] = range[2*n+1] - sb->decomp_disp[n];
    }
    else {
      end[n] = sb->decomp_size[n];
    }
    if (sb->id_p[n]==MPI_PROC_NULL && (range[2*n+1] > sb->decomp_disp[n]+sb->decomp_size[n]))
      end[n] += (range[2*n+1]-sb->decomp_disp[n]-sb->decomp_size[n]);
  }
  #else //OPS_MPI
  for ( int n=0; n<3; n++ ){
    start[n] = range[2*n];end[n] = range[2*n+1];
  }
  #endif //OPS_MPI
  #ifdef OPS_DEBUG
  ops_register_args(args, "PdV_kernel_predict");
  #endif

  offs[0][0] = args[0].stencil->stride[0]*1;  //unit step in x dimension
  offs[0][1] = off3D(1, &start[0],
      &end[0],args[0].dat->size, args[0].stencil->stride) - offs[0][0];
  offs[0][2] = off3D(2, &start[0],
      &end[0],args[0].dat->size, args[0].stencil->stride) - offs[0][1] - offs[0][0];

  offs[1][0] = args[1].stencil->stride[0]*1;  //unit step in x dimension
  offs[1][1] = off3D(1, &start[0],
      &end[0],args[1].dat->size, args[1].stencil->stride) - offs[1][0];
  offs[1][2] = off3D(2, &start[0],
      &end[0],args[1].dat->size, args[1].stencil->stride) - offs[1][1] - offs[1][0];

  offs[2][0] = args[2].stencil->stride[0]*1;  //unit step in x dimension
  offs[2][1] = off3D(1, &start[0],
      &end[0],args[2].dat->size, args[2].stencil->stride) - offs[2][0];
  offs[2][2] = off3D(2, &start[0],
      &end[0],args[2].dat->size, args[2].stencil->stride) - offs[2][1] - offs[2][0];

  offs[3][0] = args[3].stencil->stride[0]*1;  //unit step in x dimension
  offs[3][1] = off3D(1, &start[0],
      &end[0],args[3].dat->size, args[3].stencil->stride) - offs[3][0];
  offs[3][2] = off3D(2, &start[0],
      &end[0],args[3].dat->size, args[3].stencil->stride) - offs[3][1] - offs[3][0];

  offs[4][0] = args[4].stencil->stride[0]*1;  //unit step in x dimension
  offs[4][1] = off3D(1, &start[0],
      &end[0],args[4].dat->size, args[4].stencil->stride) - offs[4][0];
  offs[4][2] = off3D(2, &start[0],
      &end[0],args[4].dat->size, args[4].stencil->stride) - offs[4][1] - offs[4][0];

  offs[5][0] = args[5].stencil->stride[0]*1;  //unit step in x dimension
  offs[5][1] = off3D(1, &start[0],
      &end[0],args[5].dat->size, args[5].stencil->stride) - offs[5][0];
  offs[5][2] = off3D(2, &start[0],
      &end[0],args[5].dat->size, args[5].stencil->stride) - offs[5][1] - offs[5][0];

  offs[6][0] = args[6].stencil->stride[0]*1;  //unit step in x dimension
  offs[6][1] = off3D(1, &start[0],
      &end[0],args[6].dat->size, args[6].stencil->stride) - offs[6][0];
  offs[6][2] = off3D(2, &start[0],
      &end[0],args[6].dat->size, args[6].stencil->stride) - offs[6][1] - offs[6][0];

  offs[7][0] = args[7].stencil->stride[0]*1;  //unit step in x dimension
  offs[7][1] = off3D(1, &start[0],
      &end[0],args[7].dat->size, args[7].stencil->stride) - offs[7][0];
  offs[7][2] = off3D(2, &start[0],
      &end[0],args[7].dat->size, args[7].stencil->stride) - offs[7][1] - offs[7][0];

  offs[8][0] = args[8].stencil->stride[0]*1;  //unit step in x dimension
  offs[8][1] = off3D(1, &start[0],
//.........这里部分代码省略.........

// host stub function
void ops_par_loop_field_summary_kernel(char const *name, ops_block block,
                                       int dim, int *range, ops_arg arg0,
                                       ops_arg arg1, ops_arg arg2, ops_arg arg3,
                                       ops_arg arg4, ops_arg arg5, ops_arg arg6,
                                       ops_arg arg7, ops_arg arg8, ops_arg arg9,
                                       ops_arg arg10) {

  // Timing
  double t1, t2, c1, c2;

  int offs[11][2];
  ops_arg args[11] = {arg0, arg1, arg2, arg3, arg4, arg5,
                      arg6, arg7, arg8, arg9, arg10};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 11, range, 49))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(49, "field_summary_kernel");
    OPS_kernels[49].count++;
    ops_timers_core(&c1, &t1);
  }

#ifdef OPS_MPI
  sub_block_list sb = OPS_sub_block_list[block->index];
#endif

  // compute locally allocated range for the sub-block

  int start[2];
  int end[2];
  int arg_idx[2];

#ifdef OPS_MPI
  if (!sb->owned)
    return;
  for (int n = 0; n < 2; n++) {
    start[n] = sb->decomp_disp[n];
    end[n] = sb->decomp_disp[n] + sb->decomp_size[n];
    if (start[n] >= range[2 * n]) {
      start[n] = 0;
    } else {
      start[n] = range[2 * n] - start[n];
    }
    if (sb->id_m[n] == MPI_PROC_NULL && range[2 * n] < 0)
      start[n] = range[2 * n];
    if (end[n] >= range[2 * n + 1]) {
      end[n] = range[2 * n + 1] - sb->decomp_disp[n];
    } else {
      end[n] = sb->decomp_size[n];
    }
    if (sb->id_p[n] == MPI_PROC_NULL &&
        (range[2 * n + 1] > sb->decomp_disp[n] + sb->decomp_size[n]))
      end[n] += (range[2 * n + 1] - sb->decomp_disp[n] - sb->decomp_size[n]);
    if (end[n] < start[n])
      end[n] = start[n];
  }
#else
  for (int n = 0; n < 2; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
  }
#endif
#ifdef OPS_DEBUG
  ops_register_args(args, "field_summary_kernel");
#endif

  offs[0][0] = args[0].stencil->stride[0] * 1; // unit step in x dimension
  offs[0][1] =
      off2D(1, &start[0], &end[0], args[0].dat->size, args[0].stencil->stride) -
      offs[0][0];

  offs[1][0] = args[1].stencil->stride[0] * 1; // unit step in x dimension
  offs[1][1] =
      off2D(1, &start[0], &end[0], args[1].dat->size, args[1].stencil->stride) -
      offs[1][0];

  offs[2][0] = args[2].stencil->stride[0] * 1; // unit step in x dimension
  offs[2][1] =
      off2D(1, &start[0], &end[0], args[2].dat->size, args[2].stencil->stride) -
      offs[2][0];

  offs[3][0] = args[3].stencil->stride[0] * 1; // unit step in x dimension
  offs[3][1] =
      off2D(1, &start[0], &end[0], args[3].dat->size, args[3].stencil->stride) -
      offs[3][0];

  offs[4][0] = args[4].stencil->stride[0] * 1; // unit step in x dimension
  offs[4][1] =
      off2D(1, &start[0], &end[0], args[4].dat->size, args[4].stencil->stride) -
      offs[4][0];

  offs[5][0] = args[5].stencil->stride[0] * 1; // unit step in x dimension
  offs[5][1] =
      off2D(1, &start[0], &end[0], args[5].dat->size, args[5].stencil->stride) -
      offs[5][0];

//.........这里部分代码省略.........

// host stub function
void ops_par_loop_test_kernel(char const *name, ops_block block, int dim,
                              int *range, ops_arg arg0, ops_arg arg1) {

  // Timing
  double t1, t2, c1, c2;

  char *p_a[2];
  int offs[2][1];
  ops_arg args[2] = {arg0, arg1};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 2, range, 14))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(14, "test_kernel");
    OPS_kernels[14].count++;
    ops_timers_core(&c2, &t2);
  }

  // compute locally allocated range for the sub-block
  int start[1];
  int end[1];

#ifdef OPS_MPI
  sub_block_list sb = OPS_sub_block_list[block->index];
#endif
#ifdef OPS_DEBUG
  ops_register_args(args, "test_kernel");
#endif

  int arg_idx[1];
  int arg_idx_base[1];
#ifdef OPS_MPI
  if (compute_ranges(args, 2, block, range, start, end, arg_idx) < 0)
    return;
#else  // OPS_MPI
  for (int n = 0; n < 1; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
    arg_idx[n] = start[n];
  }
#endif // OPS_MPI
  for (int n = 0; n < 1; n++) {
    arg_idx_base[n] = arg_idx[n];
  }
  offs[0][0] = args[0].stencil->stride[0] * 1; // unit step in x dimension

  int off0_0 = offs[0][0];
  int dat0 = (OPS_soa ? args[0].dat->type_size : args[0].dat->elem_size);

  // set up initial pointers and exchange halos if necessary
  int base0 = args[0].dat->base_offset +
              (OPS_soa ? args[0].dat->type_size : args[0].dat->elem_size) *
                  start[0] * args[0].stencil->stride[0];
  p_a[0] = (char *)args[0].data + base0;

#ifdef OPS_MPI
  p_a[1] = ((ops_reduction)args[1].data)->data +
           ((ops_reduction)args[1].data)->size * block->index;
#else
  p_a[1] = ((ops_reduction)args[1].data)->data;
#endif

  // initialize global variable with the dimension of dats
  xdim0 = args[0].dat->size[0];

  // Halo Exchanges
  ops_H_D_exchanges_host(args, 2);
  ops_halo_exchanges(args, 2, range);
  ops_H_D_exchanges_host(args, 2);

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[14].mpi_time += t1 - t2;
  }

  int n_x;
#pragma novector
  for (n_x = start[0];
       n_x < start[0] + ((end[0] - start[0]) / SIMD_VEC) * SIMD_VEC;
       n_x += SIMD_VEC) {
    // call kernel function, passing in pointers to data -vectorised
    for (int i = 0; i < SIMD_VEC; i++) {
      test_kernel((double *)p_a[0] + i * 1 * 1, (double *)p_a[1]);
    }

    // shift pointers to data x direction
    p_a[0] = p_a[0] + (dat0 * off0_0) * SIMD_VEC;
  }

  for (int n_x = start[0] + ((end[0] - start[0]) / SIMD_VEC) * SIMD_VEC;
       n_x < end[0]; n_x++) {
    // call kernel function, passing in pointers to data - remainder
    test_kernel((double *)p_a[0], (double *)p_a[1]);

    // shift pointers to data x direction
    p_a[0] = p_a[0] + (dat0 * off0_0);
//.........这里部分代码省略.........