// host stub function
void ops_par_loop_advec_mom_kernel1_z_nonvector(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) {

  // 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_timing_realloc(136, "advec_mom_kernel1_z_nonvector");
    OPS_kernels[136].count++;
    ops_timers_core(&c1, &t1);
  }

  // compute localy 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 // OPS_MPI

  int arg_idx[3];
  int arg_idx_base[3];
#ifdef OPS_MPI
  if (compute_ranges(args, 5, 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
  for (int n = 0; n < 3; n++) {
    arg_idx_base[n] = arg_idx[n];
  }

  int dat0 = args[0].dat->elem_size;
  int dat1 = args[1].dat->elem_size;
  int dat2 = args[2].dat->elem_size;
  int dat3 = args[3].dat->elem_size;
  int dat4 = args[4].dat->elem_size;

  // set up initial pointers
  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];
  base0 = base0 +
          (OPS_soa ? args[0].dat->type_size : args[0].dat->elem_size) *
              args[0].dat->size[0] * args[0].dat->size[1] * start[2] *
              args[0].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a0 = (double *)((char *)args[0].data_d + base0);
#else
  double *p_a0 = (double *)((char *)args[0].data + base0);
#endif

  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];
  base1 = base1 +
          (OPS_soa ? args[1].dat->type_size : args[1].dat->elem_size) *
              args[1].dat->size[0] * args[1].dat->size[1] * start[2] *
              args[1].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a1 = (double *)((char *)args[1].data_d + base1);
#else
  double *p_a1 = (double *)((char *)args[1].data + base1);
#endif

  int base2 = args[2].dat->base_offset +
              (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
                  start[0] * args[2].stencil->stride[0];
  base2 = base2 +
          (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
              args[2].dat->size[0] * start[1] * args[2].stencil->stride[1];
  base2 = base2 +
          (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
              args[2].dat->size[0] * args[2].dat->size[1] * start[2] *
              args[2].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a2 = (double *)((char *)args[2].data_d + base2);
#else
  double *p_a2 = (double *)((char *)args[2].data + base2);
#endif

  int base3 = args[3].dat->base_offset +
              (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size) *
                  start[0] * args[3].stencil->stride[0];
  base3 = base3 +
          (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size) *
              args[3].dat->size[0] * start[1] * args[3].stencil->stride[1];
  base3 = base3 +
          (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size) *
              args[3].dat->size[0] * args[3].dat->size[1] * start[2] *
              args[3].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a3 = (double *)((char *)args[3].data_d + base3);
#else
  double *p_a3 = (double *)((char *)args[3].data + base3);
#endif

  int base4 = args[4].dat->base_offset +
              (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size) *
                  start[0] * args[4].stencil->stride[0];
  base4 = base4 +
          (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size) *
              args[4].dat->size[0] * start[1] * args[4].stencil->stride[1];
  base4 = base4 +
          (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size) *
              args[4].dat->size[0] * args[4].dat->size[1] * start[2] *
              args[4].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a4 = (double *)((char *)args[4].data_d + base4);
#else
  double *p_a4 = (double *)((char *)args[4].data + base4);
#endif

  int x_size = MAX(0, end[0] - start[0]);
  int y_size = MAX(0, end[1] - start[1]);
  int z_size = MAX(0, end[2] - start[2]);

  // initialize global variable with the dimension of dats
  xdim0 = args[0].dat->size[0];
  ydim0 = args[0].dat->size[1];
  xdim1 = args[1].dat->size[0];
  ydim1 = args[1].dat->size[1];
  xdim2 = args[2].dat->size[0];
  ydim2 = args[2].dat->size[1];
  xdim3 = args[3].dat->size[0];
  ydim3 = args[3].dat->size[1];
  xdim4 = args[4].dat->size[0];
  ydim4 = args[4].dat->size[1];
  if (xdim0 != xdim0_advec_mom_kernel1_z_nonvector_h ||
      ydim0 != ydim0_advec_mom_kernel1_z_nonvector_h ||
      xdim1 != xdim1_advec_mom_kernel1_z_nonvector_h ||
      ydim1 != ydim1_advec_mom_kernel1_z_nonvector_h ||
      xdim2 != xdim2_advec_mom_kernel1_z_nonvector_h ||
      ydim2 != ydim2_advec_mom_kernel1_z_nonvector_h ||
      xdim3 != xdim3_advec_mom_kernel1_z_nonvector_h ||
      ydim3 != ydim3_advec_mom_kernel1_z_nonvector_h ||
      xdim4 != xdim4_advec_mom_kernel1_z_nonvector_h ||
      ydim4 != ydim4_advec_mom_kernel1_z_nonvector_h) {
    xdim0_advec_mom_kernel1_z_nonvector = xdim0;
    xdim0_advec_mom_kernel1_z_nonvector_h = xdim0;
    ydim0_advec_mom_kernel1_z_nonvector = ydim0;
    ydim0_advec_mom_kernel1_z_nonvector_h = ydim0;
    xdim1_advec_mom_kernel1_z_nonvector = xdim1;
    xdim1_advec_mom_kernel1_z_nonvector_h = xdim1;
    ydim1_advec_mom_kernel1_z_nonvector = ydim1;
    ydim1_advec_mom_kernel1_z_nonvector_h = ydim1;
    xdim2_advec_mom_kernel1_z_nonvector = xdim2;
    xdim2_advec_mom_kernel1_z_nonvector_h = xdim2;
    ydim2_advec_mom_kernel1_z_nonvector = ydim2;
    ydim2_advec_mom_kernel1_z_nonvector_h = ydim2;
    xdim3_advec_mom_kernel1_z_nonvector = xdim3;
    xdim3_advec_mom_kernel1_z_nonvector_h = xdim3;
    ydim3_advec_mom_kernel1_z_nonvector = ydim3;
    ydim3_advec_mom_kernel1_z_nonvector_h = ydim3;
    xdim4_advec_mom_kernel1_z_nonvector = xdim4;
    xdim4_advec_mom_kernel1_z_nonvector_h = xdim4;
    ydim4_advec_mom_kernel1_z_nonvector = ydim4;
    ydim4_advec_mom_kernel1_z_nonvector_h = ydim4;
  }

// Halo Exchanges

#ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 5);
#else
  ops_H_D_exchanges_host(args, 5);
#endif
  ops_halo_exchanges(args, 5, range);

#ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 5);
#else
  ops_H_D_exchanges_host(args, 5);
#endif
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[136].mpi_time += t2 - t1;
  }

  advec_mom_kernel1_z_nonvector_c_wrapper(p_a0, p_a1, p_a2, p_a3, p_a4, x_size,
                                          y_size, z_size);

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[136].time += t1 - t2;
  }
#ifdef OPS_GPU
  ops_set_dirtybit_device(args, 5);
#else
  ops_set_dirtybit_host(args, 5);
#endif
  ops_set_halo_dirtybit3(&args[2], range);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[136].mpi_time += t2 - t1;
    OPS_kernels[136].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[136].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[136].transfer += ops_compute_transfer(dim, start, end, &arg2);
    OPS_kernels[136].transfer += ops_compute_transfer(dim, start, end, &arg3);
    OPS_kernels[136].transfer += ops_compute_transfer(dim, start, end, &arg4);
  }
}
// host stub function
void ops_par_loop_updateRK3_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;
  ops_arg args[11] = {arg0, arg1, arg2, arg3, arg4, arg5,
                      arg6, arg7, arg8, arg9, arg10};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 11, range, 6))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(6, "updateRK3_kernel");
    OPS_kernels[6].count++;
    ops_timers_core(&c1, &t1);
  }

  // compute localy 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 // OPS_MPI

  int arg_idx[1];
  int arg_idx_base[1];
#ifdef OPS_MPI
  if (compute_ranges(args, 11, 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
  for (int n = 0; n < 1; n++) {
    arg_idx_base[n] = arg_idx[n];
  }

  int dat0 = args[0].dat->elem_size;
  int dat1 = args[1].dat->elem_size;
  int dat2 = args[2].dat->elem_size;
  int dat3 = args[3].dat->elem_size;
  int dat4 = args[4].dat->elem_size;
  int dat5 = args[5].dat->elem_size;
  int dat6 = args[6].dat->elem_size;
  int dat7 = args[7].dat->elem_size;
  int dat8 = args[8].dat->elem_size;

  // set up initial pointers
  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];
#ifdef OPS_GPU
  double *p_a0 = (double *)((char *)args[0].data_d + base0);
#else
  double *p_a0 = (double *)((char *)args[0].data + base0);
#endif

  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];
#ifdef OPS_GPU
  double *p_a1 = (double *)((char *)args[1].data_d + base1);
#else
  double *p_a1 = (double *)((char *)args[1].data + base1);
#endif

  int base2 = args[2].dat->base_offset +
              (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
                  start[0] * args[2].stencil->stride[0];
#ifdef OPS_GPU
  double *p_a2 = (double *)((char *)args[2].data_d + base2);
#else
  double *p_a2 = (double *)((char *)args[2].data + base2);
#endif

  int base3 = args[3].dat->base_offset +
              (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size) *
                  start[0] * args[3].stencil->stride[0];
#ifdef OPS_GPU
  double *p_a3 = (double *)((char *)args[3].data_d + base3);
#else
  double *p_a3 = (double *)((char *)args[3].data + base3);
#endif

  int base4 = args[4].dat->base_offset +
              (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size) *
                  start[0] * args[4].stencil->stride[0];
#ifdef OPS_GPU
  double *p_a4 = (double *)((char *)args[4].data_d + base4);
#else
  double *p_a4 = (double *)((char *)args[4].data + base4);
#endif

  int base5 = args[5].dat->base_offset +
              (OPS_soa ? args[5].dat->type_size : args[5].dat->elem_size) *
                  start[0] * args[5].stencil->stride[0];
#ifdef OPS_GPU
  double *p_a5 = (double *)((char *)args[5].data_d + base5);
#else
  double *p_a5 = (double *)((char *)args[5].data + base5);
#endif

  int base6 = args[6].dat->base_offset +
              (OPS_soa ? args[6].dat->type_size : args[6].dat->elem_size) *
                  start[0] * args[6].stencil->stride[0];
#ifdef OPS_GPU
  double *p_a6 = (double *)((char *)args[6].data_d + base6);
#else
  double *p_a6 = (double *)((char *)args[6].data + base6);
#endif

  int base7 = args[7].dat->base_offset +
              (OPS_soa ? args[7].dat->type_size : args[7].dat->elem_size) *
                  start[0] * args[7].stencil->stride[0];
#ifdef OPS_GPU
  double *p_a7 = (double *)((char *)args[7].data_d + base7);
#else
  double *p_a7 = (double *)((char *)args[7].data + base7);
#endif

  int base8 = args[8].dat->base_offset +
              (OPS_soa ? args[8].dat->type_size : args[8].dat->elem_size) *
                  start[0] * args[8].stencil->stride[0];
#ifdef OPS_GPU
  double *p_a8 = (double *)((char *)args[8].data_d + base8);
#else
  double *p_a8 = (double *)((char *)args[8].data + base8);
#endif

  double *p_a9 = (double *)args[9].data;
  double *p_a10 = (double *)args[10].data;

  int x_size = MAX(0, end[0] - start[0]);

  // initialize global variable with the dimension of dats
  xdim0 = args[0].dat->size[0];
  xdim1 = args[1].dat->size[0];
  xdim2 = args[2].dat->size[0];
  xdim3 = args[3].dat->size[0];
  xdim4 = args[4].dat->size[0];
  xdim5 = args[5].dat->size[0];
  xdim6 = args[6].dat->size[0];
  xdim7 = args[7].dat->size[0];
  xdim8 = args[8].dat->size[0];
  if (xdim0 != xdim0_updateRK3_kernel_h || xdim1 != xdim1_updateRK3_kernel_h ||
      xdim2 != xdim2_updateRK3_kernel_h || xdim3 != xdim3_updateRK3_kernel_h ||
      xdim4 != xdim4_updateRK3_kernel_h || xdim5 != xdim5_updateRK3_kernel_h ||
      xdim6 != xdim6_updateRK3_kernel_h || xdim7 != xdim7_updateRK3_kernel_h ||
      xdim8 != xdim8_updateRK3_kernel_h) {
    xdim0_updateRK3_kernel = xdim0;
    xdim0_updateRK3_kernel_h = xdim0;
    xdim1_updateRK3_kernel = xdim1;
    xdim1_updateRK3_kernel_h = xdim1;
    xdim2_updateRK3_kernel = xdim2;
    xdim2_updateRK3_kernel_h = xdim2;
    xdim3_updateRK3_kernel = xdim3;
    xdim3_updateRK3_kernel_h = xdim3;
    xdim4_updateRK3_kernel = xdim4;
    xdim4_updateRK3_kernel_h = xdim4;
    xdim5_updateRK3_kernel = xdim5;
    xdim5_updateRK3_kernel_h = xdim5;
    xdim6_updateRK3_kernel = xdim6;
    xdim6_updateRK3_kernel_h = xdim6;
    xdim7_updateRK3_kernel = xdim7;
    xdim7_updateRK3_kernel_h = xdim7;
    xdim8_updateRK3_kernel = xdim8;
    xdim8_updateRK3_kernel_h = xdim8;
  }

// Halo Exchanges

#ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 11);
#else
  ops_H_D_exchanges_host(args, 11);
#endif
  ops_halo_exchanges(args, 11, range);

#ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 11);
#else
  ops_H_D_exchanges_host(args, 11);
#endif
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[6].mpi_time += t2 - t1;
  }

  updateRK3_kernel_c_wrapper(p_a0, p_a1, p_a2, p_a3, p_a4, p_a5, p_a6, p_a7,
                             p_a8, *p_a9, *p_a10, x_size);

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[6].time += t1 - t2;
  }
#ifdef OPS_GPU
  ops_set_dirtybit_device(args, 11);
#else
  ops_set_dirtybit_host(args, 11);
#endif
  ops_set_halo_dirtybit3(&args[0], range);
  ops_set_halo_dirtybit3(&args[1], range);
  ops_set_halo_dirtybit3(&args[2], range);
  ops_set_halo_dirtybit3(&args[3], range);
  ops_set_halo_dirtybit3(&args[4], range);
  ops_set_halo_dirtybit3(&args[5], range);
  ops_set_halo_dirtybit3(&args[6], range);
  ops_set_halo_dirtybit3(&args[7], range);
  ops_set_halo_dirtybit3(&args[8], range);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[6].mpi_time += t2 - t1;
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg2);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg3);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg4);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg5);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg6);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg7);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg8);
  }
}
// host stub function
void ops_par_loop_left_bndcon(char const *name, ops_block block, int dim, int* range,
 ops_arg arg0, ops_arg arg1) {

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


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

  ops_timing_realloc(2,"left_bndcon");
  OPS_kernels[2].count++;

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

  int x_size = MAX(0,end[0]-start[0]);
  int y_size = MAX(0,end[1]-start[1]);

  int arg_idx[2];
  #ifdef OPS_MPI
  arg_idx[0] = sb->decomp_disp[0]+start[0];
  arg_idx[1] = sb->decomp_disp[1]+start[1];
  #else
  arg_idx[0] = start[0];
  arg_idx[1] = start[1];
  #endif

  xdim0 = args[0].dat->size[0];

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

  if (xdim0 != xdim0_left_bndcon_h) {
    xdim0_left_bndcon = xdim0;
    xdim0_left_bndcon_h = xdim0;
  }

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


  //set up initial pointers
  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
  for (int d = 0; d < dim; d++) d_m[d] = args[0].dat->d_m[d];
  #endif
  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]);
  double *p_a0 = (double *)((char *)args[0].data + base0);

  int *p_a1 = NULL;


  ops_H_D_exchanges_host(args, 2);
  ops_halo_exchanges(args,2,range);

  ops_timers_core(&c1,&t1);
  OPS_kernels[2].mpi_time += t1-t2;

  left_bndcon_c_wrapper(
    p_a0,
    p_a1,
    arg_idx[0], arg_idx[1],
    x_size, y_size);

  ops_timers_core(&c2,&t2);
  OPS_kernels[2].time += t2-t1;
  ops_set_dirtybit_host(args, 2);
  ops_set_halo_dirtybit3(&args[0],range);

  //Update kernel record
  OPS_kernels[2].transfer += ops_compute_transfer(dim, start, end, &arg0);
}
// host stub function
void ops_par_loop_update_halo_kernel2_zvel_minus_4_back(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;
  ops_timers_core(&c1,&t1);


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



  ops_timing_realloc(85,"update_halo_kernel2_zvel_minus_4_back");
  OPS_kernels[85].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_kernel2_zvel_minus_4_back");
  #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];



  int off0_0 = offs[0][0];
  int off0_1 = offs[0][1];
  int off0_2 = offs[0][2];
  int dat0 = args[0].dat->elem_size;
  int off1_0 = offs[1][0];
  int off1_1 = offs[1][1];
  int off1_2 = offs[1][2];
  int dat1 = args[1].dat->elem_size;


  #ifdef _OPENMP
  int nthreads = omp_get_max_threads( );
  #else
  int nthreads = 1;
  #endif
  xdim0 = args[0].dat->size[0]*args[0].dat->dim;
  ydim0 = args[0].dat->size[1];
  xdim1 = args[1].dat->size[0]*args[1].dat->dim;
  ydim1 = args[1].dat->size[1];

  ops_H_D_exchanges_host(args, 3);

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


  ops_timers_core(&c2,&t2);
  OPS_kernels[85].mpi_time += t2-t1;


  #pragma omp parallel for
  for ( int thr=0; thr<nthreads; thr++ ){

    int z_size = end[2]-start[2];
    char *p_a[3];

    int start_i = start[2] + ((z_size-1)/nthreads+1)*thr;
    int finish_i = start[2] + MIN(((z_size-1)/nthreads+1)*(thr+1),z_size);

    //get address per thread
    int start0 = start[0];
    int start1 = start[1];
    int start2 = start_i;

    //set up initial pointers 
    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 * 
    (start0 * args[0].stencil->stride[0] - args[0].dat->base[0] - d_m[0]);
    base0 = base0+ dat0 *
      args[0].dat->size[0] *
      (start1 * args[0].stencil->stride[1] - args[0].dat->base[1] - d_m[1]);
    base0 = base0+ dat0 *
      args[0].dat->size[0] *
      args[0].dat->size[1] *
      (start2 * args[0].stencil->stride[2] - args[0].dat->base[2] - d_m[2]);
    p_a[0] = (char *)args[0].data + base0;

    #ifdef OPS_MPI
    for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d] + OPS_sub_dat_list[args[1].dat->index]->d_im[d];
    #else //OPS_MPI
    for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d];
    #endif //OPS_MPI
    int base1 = dat1 * 1 * 
    (start0 * args[1].stencil->stride[0] - args[1].dat->base[0] - d_m[0]);
    base1 = base1+ dat1 *
      args[1].dat->size[0] *
      (start1 * args[1].stencil->stride[1] - args[1].dat->base[1] - d_m[1]);
    base1 = base1+ dat1 *
      args[1].dat->size[0] *
      args[1].dat->size[1] *
      (start2 * args[1].stencil->stride[2] - args[1].dat->base[2] - d_m[2]);
    p_a[1] = (char *)args[1].data + base1;

    p_a[2] = (char *)args[2].data;


    for ( int n_z=start_i; n_z<finish_i; n_z++ ){
      for ( int n_y=start[1]; n_y<end[1]; n_y++ ){
        for ( int n_x=start[0]; n_x<start[0]+(end[0]-start[0])/SIMD_VEC; n_x++ ){
          //call kernel function, passing in pointers to data -vectorised
          #pragma simd
          for ( int i=0; i<SIMD_VEC; i++ ){
            update_halo_kernel2_zvel_minus_4_back(  (double * )p_a[0]+ i*1, (double * )p_a[1]+ i*1, (int * )p_a[2] );

          }

          //shift pointers to data x direction
          p_a[0]= p_a[0] + (dat0 * off0_0)*SIMD_VEC;
          p_a[1]= p_a[1] + (dat1 * off1_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
          update_halo_kernel2_zvel_minus_4_back(  (double * )p_a[0], (double * )p_a[1], (int * )p_a[2] );


          //shift pointers to data x direction
          p_a[0]= p_a[0] + (dat0 * off0_0);
          p_a[1]= p_a[1] + (dat1 * off1_0);
        }

        //shift pointers to data y direction
        p_a[0]= p_a[0] + (dat0 * off0_1);
        p_a[1]= p_a[1] + (dat1 * off1_1);
      }
      //shift pointers to data z direction
      p_a[0]= p_a[0] + (dat0 * off0_2);
      p_a[1]= p_a[1] + (dat1 * off1_2);
    }
  }

  ops_timers_core(&c1,&t1);
  OPS_kernels[85].time += t1-t2;

  ops_set_dirtybit_host(args, 3);

  ops_set_halo_dirtybit3(&args[0],range);
  ops_set_halo_dirtybit3(&args[1],range);

  //Update kernel record
  ops_timers_core(&c2,&t2);
  OPS_kernels[85].mpi_time += t2-t1;
  OPS_kernels[85].transfer += ops_compute_transfer(dim, range, &arg0);
  OPS_kernels[85].transfer += ops_compute_transfer(dim, range, &arg1);
}
// host stub function
void ops_par_loop_initialise_chunk_kernel_cellz(char const *name, ops_block block, int dim, int* range,
 ops_arg arg0, ops_arg arg1, ops_arg arg2) {

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



  ops_timing_realloc(138,"initialise_chunk_kernel_cellz");
  OPS_kernels[138].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, "initialise_chunk_kernel_cellz");
  #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];



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

  int off0_0 = offs[0][0];
  int off0_1 = offs[0][1];
  int off0_2 = offs[0][2];
  int dat0 = args[0].dat->elem_size;
  int off1_0 = offs[1][0];
  int off1_1 = offs[1][1];
  int off1_2 = offs[1][2];
  int dat1 = args[1].dat->elem_size;
  int off2_0 = offs[2][0];
  int off2_1 = offs[2][1];
  int off2_2 = offs[2][2];
  int dat2 = args[2].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]);
  base0 = base0+ dat0 *
    args[0].dat->size[0] *
    args[0].dat->size[1] *
    (start[2] * args[0].stencil->stride[2] - args[0].dat->base[2] - d_m[2]);
  p_a[0] = (char *)args[0].data + base0;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d] + OPS_sub_dat_list[args[1].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d];
  #endif //OPS_MPI
  int base1 = dat1 * 1 * 
    (start[0] * args[1].stencil->stride[0] - args[1].dat->base[0] - d_m[0]);
  base1 = base1+ dat1 *
    args[1].dat->size[0] *
    (start[1] * args[1].stencil->stride[1] - args[1].dat->base[1] - d_m[1]);
  base1 = base1+ dat1 *
    args[1].dat->size[0] *
    args[1].dat->size[1] *
    (start[2] * args[1].stencil->stride[2] - args[1].dat->base[2] - d_m[2]);
  p_a[1] = (char *)args[1].data + base1;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[2].dat->d_m[d] + OPS_sub_dat_list[args[2].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[2].dat->d_m[d];
  #endif //OPS_MPI
  int base2 = dat2 * 1 * 
    (start[0] * args[2].stencil->stride[0] - args[2].dat->base[0] - d_m[0]);
  base2 = base2+ dat2 *
    args[2].dat->size[0] *
    (start[1] * args[2].stencil->stride[1] - args[2].dat->base[1] - d_m[1]);
  base2 = base2+ dat2 *
    args[2].dat->size[0] *
    args[2].dat->size[1] *
    (start[2] * args[2].stencil->stride[2] - args[2].dat->base[2] - d_m[2]);
  p_a[2] = (char *)args[2].data + base2;


  ops_H_D_exchanges_host(args, 3);
  ops_halo_exchanges(args,3,range);
  ops_H_D_exchanges_host(args, 3);

  ops_timers_core(&c1,&t1);
  OPS_kernels[138].mpi_time += t1-t2;

  xdim0 = args[0].dat->size[0]*args[0].dat->dim;
  ydim0 = args[0].dat->size[1];
  xdim1 = args[1].dat->size[0]*args[1].dat->dim;
  ydim1 = args[1].dat->size[1];
  xdim2 = args[2].dat->size[0]*args[2].dat->dim;
  ydim2 = args[2].dat->size[1];

  int n_x;
  for ( int n_z=start[2]; n_z<end[2]; n_z++ ){
    for ( int n_y=start[1]; n_y<end[1]; n_y++ ){
      #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
        #pragma simd
        for ( int i=0; i<SIMD_VEC; i++ ){
          initialise_chunk_kernel_cellz(  (double *)p_a[0]+ i*0, (double *)p_a[1]+ i*0, (double *)p_a[2]+ i*0 );

        }

        //shift pointers to data x direction
        p_a[0]= p_a[0] + (dat0 * off0_0)*SIMD_VEC;
        p_a[1]= p_a[1] + (dat1 * off1_0)*SIMD_VEC;
        p_a[2]= p_a[2] + (dat2 * off2_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
        initialise_chunk_kernel_cellz(  (double *)p_a[0], (double *)p_a[1], (double *)p_a[2] );


        //shift pointers to data x direction
        p_a[0]= p_a[0] + (dat0 * off0_0);
        p_a[1]= p_a[1] + (dat1 * off1_0);
        p_a[2]= p_a[2] + (dat2 * off2_0);
      }

      //shift pointers to data y direction
      p_a[0]= p_a[0] + (dat0 * off0_1);
      p_a[1]= p_a[1] + (dat1 * off1_1);
      p_a[2]= p_a[2] + (dat2 * off2_1);
    }
    //shift pointers to data z direction
    p_a[0]= p_a[0] + (dat0 * off0_2);
    p_a[1]= p_a[1] + (dat1 * off1_2);
    p_a[2]= p_a[2] + (dat2 * off2_2);
  }
  ops_timers_core(&c2,&t2);
  OPS_kernels[138].time += t2-t1;
  ops_set_dirtybit_host(args, 3);
  ops_set_halo_dirtybit3(&args[1],range);
  ops_set_halo_dirtybit3(&args[2],range);

  //Update kernel record
  OPS_kernels[138].transfer += ops_compute_transfer(dim, range, &arg0);
  OPS_kernels[138].transfer += ops_compute_transfer(dim, range, &arg1);
  OPS_kernels[138].transfer += ops_compute_transfer(dim, range, &arg2);
}
// host stub function
void ops_par_loop_update_halo_kernel2_zvel_plus_4_right(
    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;
  ops_arg args[3] = {arg0, arg1, arg2};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 3, range, 53))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(53, "update_halo_kernel2_zvel_plus_4_right");
    OPS_kernels[53].count++;
    ops_timers_core(&c1, &t1);
  }

  // compute localy 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 // OPS_MPI

  int arg_idx[3];
  int arg_idx_base[3];
#ifdef OPS_MPI
  if (compute_ranges(args, 3, 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
  for (int n = 0; n < 3; n++) {
    arg_idx_base[n] = arg_idx[n];
  }

  int dat0 = args[0].dat->elem_size;
  int dat1 = args[1].dat->elem_size;

  int *arg2h = (int *)arg2.data;
// Upload large globals
#ifdef OPS_GPU
  int consts_bytes = 0;
  consts_bytes += ROUND_UP(NUM_FIELDS * sizeof(int));
  reallocConstArrays(consts_bytes);
  consts_bytes = 0;
  args[2].data = OPS_consts_h + consts_bytes;
  args[2].data_d = OPS_consts_d + consts_bytes;
  for (int d = 0; d < NUM_FIELDS; d++)
    ((int *)args[2].data)[d] = arg2h[d];
  consts_bytes += ROUND_UP(NUM_FIELDS * sizeof(int));
  mvConstArraysToDevice(consts_bytes);
#endif // OPS_GPU

  // set up initial pointers
  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];
  base0 = base0 +
          (OPS_soa ? args[0].dat->type_size : args[0].dat->elem_size) *
              args[0].dat->size[0] * args[0].dat->size[1] * start[2] *
              args[0].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a0 = (double *)((char *)args[0].data_d + base0);
#else
  double *p_a0 = (double *)((char *)args[0].data + base0);
#endif

  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];
  base1 = base1 +
          (OPS_soa ? args[1].dat->type_size : args[1].dat->elem_size) *
              args[1].dat->size[0] * args[1].dat->size[1] * start[2] *
              args[1].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a1 = (double *)((char *)args[1].data_d + base1);
#else
  double *p_a1 = (double *)((char *)args[1].data + base1);
#endif

#ifdef OPS_GPU
  int *p_a2 = (int *)args[2].data_d;
#else
  int *p_a2 = arg2h;
#endif

  int x_size = MAX(0, end[0] - start[0]);
  int y_size = MAX(0, end[1] - start[1]);
  int z_size = MAX(0, end[2] - start[2]);

  // initialize global variable with the dimension of dats
  xdim0 = args[0].dat->size[0];
  ydim0 = args[0].dat->size[1];
  xdim1 = args[1].dat->size[0];
  ydim1 = args[1].dat->size[1];
  if (xdim0 != xdim0_update_halo_kernel2_zvel_plus_4_right_h ||
      ydim0 != ydim0_update_halo_kernel2_zvel_plus_4_right_h ||
      xdim1 != xdim1_update_halo_kernel2_zvel_plus_4_right_h ||
      ydim1 != ydim1_update_halo_kernel2_zvel_plus_4_right_h) {
    xdim0_update_halo_kernel2_zvel_plus_4_right = xdim0;
    xdim0_update_halo_kernel2_zvel_plus_4_right_h = xdim0;
    ydim0_update_halo_kernel2_zvel_plus_4_right = ydim0;
    ydim0_update_halo_kernel2_zvel_plus_4_right_h = ydim0;
    xdim1_update_halo_kernel2_zvel_plus_4_right = xdim1;
    xdim1_update_halo_kernel2_zvel_plus_4_right_h = xdim1;
    ydim1_update_halo_kernel2_zvel_plus_4_right = ydim1;
    ydim1_update_halo_kernel2_zvel_plus_4_right_h = ydim1;
  }

// Halo Exchanges

#ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 3);
#else
  ops_H_D_exchanges_host(args, 3);
#endif
  ops_halo_exchanges(args, 3, range);

#ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 3);
#else
  ops_H_D_exchanges_host(args, 3);
#endif
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[53].mpi_time += t2 - t1;
  }

  update_halo_kernel2_zvel_plus_4_right_c_wrapper(p_a0, p_a1, p_a2, x_size,
                                                  y_size, z_size);

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[53].time += t1 - t2;
  }
#ifdef OPS_GPU
  ops_set_dirtybit_device(args, 3);
#else
  ops_set_dirtybit_host(args, 3);
#endif
  ops_set_halo_dirtybit3(&args[0], range);
  ops_set_halo_dirtybit3(&args[1], range);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[53].mpi_time += t2 - t1;
    OPS_kernels[53].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[53].transfer += ops_compute_transfer(dim, start, end, &arg1);
  }
}
// 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],
      &end[0],args[8].dat->size, args[8].stencil->stride) - offs[8][0];
  offs[8][2] = off3D(2, &start[0],
      &end[0],args[8].dat->size, args[8].stencil->stride) - offs[8][1] - offs[8][0];

  offs[9][0] = args[9].stencil->stride[0]*1;  //unit step in x dimension
  offs[9][1] = off3D(1, &start[0],
      &end[0],args[9].dat->size, args[9].stencil->stride) - offs[9][0];
  offs[9][2] = off3D(2, &start[0],
      &end[0],args[9].dat->size, args[9].stencil->stride) - offs[9][1] - offs[9][0];

  offs[10][0] = args[10].stencil->stride[0]*1;  //unit step in x dimension
  offs[10][1] = off3D(1, &start[0],
      &end[0],args[10].dat->size, args[10].stencil->stride) - offs[10][0];
  offs[10][2] = off3D(2, &start[0],
      &end[0],args[10].dat->size, args[10].stencil->stride) - offs[10][1] - offs[10][0];

  offs[11][0] = args[11].stencil->stride[0]*1;  //unit step in x dimension
  offs[11][1] = off3D(1, &start[0],
      &end[0],args[11].dat->size, args[11].stencil->stride) - offs[11][0];
  offs[11][2] = off3D(2, &start[0],
      &end[0],args[11].dat->size, args[11].stencil->stride) - offs[11][1] - offs[11][0];

  offs[12][0] = args[12].stencil->stride[0]*1;  //unit step in x dimension
  offs[12][1] = off3D(1, &start[0],
      &end[0],args[12].dat->size, args[12].stencil->stride) - offs[12][0];
  offs[12][2] = off3D(2, &start[0],
      &end[0],args[12].dat->size, args[12].stencil->stride) - offs[12][1] - offs[12][0];

  offs[13][0] = args[13].stencil->stride[0]*1;  //unit step in x dimension
  offs[13][1] = off3D(1, &start[0],
      &end[0],args[13].dat->size, args[13].stencil->stride) - offs[13][0];
  offs[13][2] = off3D(2, &start[0],
      &end[0],args[13].dat->size, args[13].stencil->stride) - offs[13][1] - offs[13][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 off0_2 = offs[0][2];
  int dat0 = args[0].dat->elem_size;
  int off1_0 = offs[1][0];
  int off1_1 = offs[1][1];
  int off1_2 = offs[1][2];
  int dat1 = args[1].dat->elem_size;
  int off2_0 = offs[2][0];
  int off2_1 = offs[2][1];
  int off2_2 = offs[2][2];
  int dat2 = args[2].dat->elem_size;
  int off3_0 = offs[3][0];
  int off3_1 = offs[3][1];
  int off3_2 = offs[3][2];
  int dat3 = args[3].dat->elem_size;
  int off4_0 = offs[4][0];
  int off4_1 = offs[4][1];
  int off4_2 = offs[4][2];
  int dat4 = args[4].dat->elem_size;
  int off5_0 = offs[5][0];
  int off5_1 = offs[5][1];
  int off5_2 = offs[5][2];
  int dat5 = args[5].dat->elem_size;
  int off6_0 = offs[6][0];
  int off6_1 = offs[6][1];
  int off6_2 = offs[6][2];
  int dat6 = args[6].dat->elem_size;
  int off7_0 = offs[7][0];
  int off7_1 = offs[7][1];
  int off7_2 = offs[7][2];
  int dat7 = args[7].dat->elem_size;
  int off8_0 = offs[8][0];
  int off8_1 = offs[8][1];
  int off8_2 = offs[8][2];
  int dat8 = args[8].dat->elem_size;
  int off9_0 = offs[9][0];
  int off9_1 = offs[9][1];
  int off9_2 = offs[9][2];
  int dat9 = args[9].dat->elem_size;
  int off10_0 = offs[10][0];
  int off10_1 = offs[10][1];
  int off10_2 = offs[10][2];
  int dat10 = args[10].dat->elem_size;
  int off11_0 = offs[11][0];
  int off11_1 = offs[11][1];
  int off11_2 = offs[11][2];
  int dat11 = args[11].dat->elem_size;
  int off12_0 = offs[12][0];
  int off12_1 = offs[12][1];
  int off12_2 = offs[12][2];
  int dat12 = args[12].dat->elem_size;
  int off13_0 = offs[13][0];
  int off13_1 = offs[13][1];
  int off13_2 = offs[13][2];
  int dat13 = args[13].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]);
  base0 = base0+ dat0 *
    args[0].dat->size[0] *
    args[0].dat->size[1] *
    (start[2] * args[0].stencil->stride[2] - args[0].dat->base[2] - d_m[2]);
  p_a[0] = (char *)args[0].data + base0;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d] + OPS_sub_dat_list[args[1].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d];
  #endif //OPS_MPI
  int base1 = dat1 * 1 * 
    (start[0] * args[1].stencil->stride[0] - args[1].dat->base[0] - d_m[0]);
  base1 = base1+ dat1 *
    args[1].dat->size[0] *
    (start[1] * args[1].stencil->stride[1] - args[1].dat->base[1] - d_m[1]);
  base1 = base1+ dat1 *
    args[1].dat->size[0] *
    args[1].dat->size[1] *
    (start[2] * args[1].stencil->stride[2] - args[1].dat->base[2] - d_m[2]);
  p_a[1] = (char *)args[1].data + base1;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[2].dat->d_m[d] + OPS_sub_dat_list[args[2].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[2].dat->d_m[d];
  #endif //OPS_MPI
  int base2 = dat2 * 1 * 
    (start[0] * args[2].stencil->stride[0] - args[2].dat->base[0] - d_m[0]);
  base2 = base2+ dat2 *
    args[2].dat->size[0] *
    (start[1] * args[2].stencil->stride[1] - args[2].dat->base[1] - d_m[1]);
  base2 = base2+ dat2 *
    args[2].dat->size[0] *
    args[2].dat->size[1] *
    (start[2] * args[2].stencil->stride[2] - args[2].dat->base[2] - d_m[2]);
  p_a[2] = (char *)args[2].data + base2;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[3].dat->d_m[d] + OPS_sub_dat_list[args[3].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[3].dat->d_m[d];
  #endif //OPS_MPI
  int base3 = dat3 * 1 * 
    (start[0] * args[3].stencil->stride[0] - args[3].dat->base[0] - d_m[0]);
  base3 = base3+ dat3 *
    args[3].dat->size[0] *
    (start[1] * args[3].stencil->stride[1] - args[3].dat->base[1] - d_m[1]);
  base3 = base3+ dat3 *
    args[3].dat->size[0] *
    args[3].dat->size[1] *
    (start[2] * args[3].stencil->stride[2] - args[3].dat->base[2] - d_m[2]);
  p_a[3] = (char *)args[3].data + base3;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[4].dat->d_m[d] + OPS_sub_dat_list[args[4].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[4].dat->d_m[d];
  #endif //OPS_MPI
  int base4 = dat4 * 1 * 
    (start[0] * args[4].stencil->stride[0] - args[4].dat->base[0] - d_m[0]);
  base4 = base4+ dat4 *
    args[4].dat->size[0] *
    (start[1] * args[4].stencil->stride[1] - args[4].dat->base[1] - d_m[1]);
  base4 = base4+ dat4 *
    args[4].dat->size[0] *
    args[4].dat->size[1] *
    (start[2] * args[4].stencil->stride[2] - args[4].dat->base[2] - d_m[2]);
  p_a[4] = (char *)args[4].data + base4;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[5].dat->d_m[d] + OPS_sub_dat_list[args[5].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[5].dat->d_m[d];
  #endif //OPS_MPI
  int base5 = dat5 * 1 * 
    (start[0] * args[5].stencil->stride[0] - args[5].dat->base[0] - d_m[0]);
  base5 = base5+ dat5 *
    args[5].dat->size[0] *
    (start[1] * args[5].stencil->stride[1] - args[5].dat->base[1] - d_m[1]);
  base5 = base5+ dat5 *
    args[5].dat->size[0] *
    args[5].dat->size[1] *
    (start[2] * args[5].stencil->stride[2] - args[5].dat->base[2] - d_m[2]);
  p_a[5] = (char *)args[5].data + base5;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[6].dat->d_m[d] + OPS_sub_dat_list[args[6].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[6].dat->d_m[d];
  #endif //OPS_MPI
  int base6 = dat6 * 1 * 
    (start[0] * args[6].stencil->stride[0] - args[6].dat->base[0] - d_m[0]);
  base6 = base6+ dat6 *
    args[6].dat->size[0] *
    (start[1] * args[6].stencil->stride[1] - args[6].dat->base[1] - d_m[1]);
  base6 = base6+ dat6 *
    args[6].dat->size[0] *
    args[6].dat->size[1] *
    (start[2] * args[6].stencil->stride[2] - args[6].dat->base[2] - d_m[2]);
  p_a[6] = (char *)args[6].data + base6;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[7].dat->d_m[d] + OPS_sub_dat_list[args[7].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[7].dat->d_m[d];
  #endif //OPS_MPI
  int base7 = dat7 * 1 * 
    (start[0] * args[7].stencil->stride[0] - args[7].dat->base[0] - d_m[0]);
  base7 = base7+ dat7 *
    args[7].dat->size[0] *
    (start[1] * args[7].stencil->stride[1] - args[7].dat->base[1] - d_m[1]);
  base7 = base7+ dat7 *
    args[7].dat->size[0] *
    args[7].dat->size[1] *
    (start[2] * args[7].stencil->stride[2] - args[7].dat->base[2] - d_m[2]);
  p_a[7] = (char *)args[7].data + base7;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[8].dat->d_m[d] + OPS_sub_dat_list[args[8].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[8].dat->d_m[d];
  #endif //OPS_MPI
  int base8 = dat8 * 1 * 
    (start[0] * args[8].stencil->stride[0] - args[8].dat->base[0] - d_m[0]);
  base8 = base8+ dat8 *
    args[8].dat->size[0] *
    (start[1] * args[8].stencil->stride[1] - args[8].dat->base[1] - d_m[1]);
  base8 = base8+ dat8 *
    args[8].dat->size[0] *
    args[8].dat->size[1] *
    (start[2] * args[8].stencil->stride[2] - args[8].dat->base[2] - d_m[2]);
  p_a[8] = (char *)args[8].data + base8;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[9].dat->d_m[d] + OPS_sub_dat_list[args[9].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[9].dat->d_m[d];
  #endif //OPS_MPI
  int base9 = dat9 * 1 * 
    (start[0] * args[9].stencil->stride[0] - args[9].dat->base[0] - d_m[0]);
  base9 = base9+ dat9 *
    args[9].dat->size[0] *
    (start[1] * args[9].stencil->stride[1] - args[9].dat->base[1] - d_m[1]);
  base9 = base9+ dat9 *
    args[9].dat->size[0] *
    args[9].dat->size[1] *
    (start[2] * args[9].stencil->stride[2] - args[9].dat->base[2] - d_m[2]);
  p_a[9] = (char *)args[9].data + base9;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[10].dat->d_m[d] + OPS_sub_dat_list[args[10].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[10].dat->d_m[d];
  #endif //OPS_MPI
  int base10 = dat10 * 1 * 
    (start[0] * args[10].stencil->stride[0] - args[10].dat->base[0] - d_m[0]);
  base10 = base10+ dat10 *
    args[10].dat->size[0] *
    (start[1] * args[10].stencil->stride[1] - args[10].dat->base[1] - d_m[1]);
  base10 = base10+ dat10 *
    args[10].dat->size[0] *
    args[10].dat->size[1] *
    (start[2] * args[10].stencil->stride[2] - args[10].dat->base[2] - d_m[2]);
  p_a[10] = (char *)args[10].data + base10;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[11].dat->d_m[d] + OPS_sub_dat_list[args[11].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[11].dat->d_m[d];
  #endif //OPS_MPI
  int base11 = dat11 * 1 * 
    (start[0] * args[11].stencil->stride[0] - args[11].dat->base[0] - d_m[0]);
  base11 = base11+ dat11 *
    args[11].dat->size[0] *
    (start[1] * args[11].stencil->stride[1] - args[11].dat->base[1] - d_m[1]);
  base11 = base11+ dat11 *
    args[11].dat->size[0] *
    args[11].dat->size[1] *
    (start[2] * args[11].stencil->stride[2] - args[11].dat->base[2] - d_m[2]);
  p_a[11] = (char *)args[11].data + base11;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[12].dat->d_m[d] + OPS_sub_dat_list[args[12].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[12].dat->d_m[d];
  #endif //OPS_MPI
  int base12 = dat12 * 1 * 
    (start[0] * args[12].stencil->stride[0] - args[12].dat->base[0] - d_m[0]);
  base12 = base12+ dat12 *
    args[12].dat->size[0] *
    (start[1] * args[12].stencil->stride[1] - args[12].dat->base[1] - d_m[1]);
  base12 = base12+ dat12 *
    args[12].dat->size[0] *
    args[12].dat->size[1] *
    (start[2] * args[12].stencil->stride[2] - args[12].dat->base[2] - d_m[2]);
  p_a[12] = (char *)args[12].data + base12;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[13].dat->d_m[d] + OPS_sub_dat_list[args[13].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[13].dat->d_m[d];
  #endif //OPS_MPI
  int base13 = dat13 * 1 * 
    (start[0] * args[13].stencil->stride[0] - args[13].dat->base[0] - d_m[0]);
  base13 = base13+ dat13 *
    args[13].dat->size[0] *
    (start[1] * args[13].stencil->stride[1] - args[13].dat->base[1] - d_m[1]);
  base13 = base13+ dat13 *
    args[13].dat->size[0] *
    args[13].dat->size[1] *
    (start[2] * args[13].stencil->stride[2] - args[13].dat->base[2] - d_m[2]);
  p_a[13] = (char *)args[13].data + base13;


  ops_H_D_exchanges_host(args, 14);
  ops_halo_exchanges(args,14,range);
  ops_H_D_exchanges_host(args, 14);

  ops_timers_core(&c1,&t1);
  OPS_kernels[5].mpi_time += t1-t2;

  xdim0 = args[0].dat->size[0]*args[0].dat->dim;
  ydim0 = args[0].dat->size[1];
  xdim1 = args[1].dat->size[0]*args[1].dat->dim;
  ydim1 = args[1].dat->size[1];
  xdim2 = args[2].dat->size[0]*args[2].dat->dim;
  ydim2 = args[2].dat->size[1];
  xdim3 = args[3].dat->size[0]*args[3].dat->dim;
  ydim3 = args[3].dat->size[1];
  xdim4 = args[4].dat->size[0]*args[4].dat->dim;
  ydim4 = args[4].dat->size[1];
  xdim5 = args[5].dat->size[0]*args[5].dat->dim;
  ydim5 = args[5].dat->size[1];
  xdim6 = args[6].dat->size[0]*args[6].dat->dim;
  ydim6 = args[6].dat->size[1];
  xdim7 = args[7].dat->size[0]*args[7].dat->dim;
  ydim7 = args[7].dat->size[1];
  xdim8 = args[8].dat->size[0]*args[8].dat->dim;
  ydim8 = args[8].dat->size[1];
  xdim9 = args[9].dat->size[0]*args[9].dat->dim;
  ydim9 = args[9].dat->size[1];
  xdim10 = args[10].dat->size[0]*args[10].dat->dim;
  ydim10 = args[10].dat->size[1];
  xdim11 = args[11].dat->size[0]*args[11].dat->dim;
  ydim11 = args[11].dat->size[1];
  xdim12 = args[12].dat->size[0]*args[12].dat->dim;
  ydim12 = args[12].dat->size[1];
  xdim13 = args[13].dat->size[0]*args[13].dat->dim;
  ydim13 = args[13].dat->size[1];

  int n_x;
  for ( int n_z=start[2]; n_z<end[2]; n_z++ ){
    for ( int n_y=start[1]; n_y<end[1]; n_y++ ){
      #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
        #pragma simd
        for ( int i=0; i<SIMD_VEC; i++ ){
          PdV_kernel_predict(  (double *)p_a[0]+ i*1, (double *)p_a[1]+ i*1, (double *)p_a[2]+ i*1,
           (double *)p_a[3]+ i*1, (double *)p_a[4]+ i*1, (double *)p_a[5]+ i*1, (double *)p_a[6]+ i*1,
           (double *)p_a[7]+ i*1, (double *)p_a[8]+ i*1, (double *)p_a[9]+ i*1, (double *)p_a[10]+ i*1,
           (double *)p_a[11]+ i*1, (double *)p_a[12]+ i*1, (double *)p_a[13]+ i*1 );

        }

        //shift pointers to data x direction
        p_a[0]= p_a[0] + (dat0 * off0_0)*SIMD_VEC;
        p_a[1]= p_a[1] + (dat1 * off1_0)*SIMD_VEC;
        p_a[2]= p_a[2] + (dat2 * off2_0)*SIMD_VEC;
        p_a[3]= p_a[3] + (dat3 * off3_0)*SIMD_VEC;
        p_a[4]= p_a[4] + (dat4 * off4_0)*SIMD_VEC;
        p_a[5]= p_a[5] + (dat5 * off5_0)*SIMD_VEC;
        p_a[6]= p_a[6] + (dat6 * off6_0)*SIMD_VEC;
        p_a[7]= p_a[7] + (dat7 * off7_0)*SIMD_VEC;
        p_a[8]= p_a[8] + (dat8 * off8_0)*SIMD_VEC;
        p_a[9]= p_a[9] + (dat9 * off9_0)*SIMD_VEC;
        p_a[10]= p_a[10] + (dat10 * off10_0)*SIMD_VEC;
        p_a[11]= p_a[11] + (dat11 * off11_0)*SIMD_VEC;
        p_a[12]= p_a[12] + (dat12 * off12_0)*SIMD_VEC;
        p_a[13]= p_a[13] + (dat13 * off13_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
        PdV_kernel_predict(  (double *)p_a[0], (double *)p_a[1], (double *)p_a[2],
           (double *)p_a[3], (double *)p_a[4], (double *)p_a[5], (double *)p_a[6],
           (double *)p_a[7], (double *)p_a[8], (double *)p_a[9], (double *)p_a[10],
           (double *)p_a[11], (double *)p_a[12], (double *)p_a[13] );


        //shift pointers to data x direction
        p_a[0]= p_a[0] + (dat0 * off0_0);
        p_a[1]= p_a[1] + (dat1 * off1_0);
        p_a[2]= p_a[2] + (dat2 * off2_0);
        p_a[3]= p_a[3] + (dat3 * off3_0);
        p_a[4]= p_a[4] + (dat4 * off4_0);
        p_a[5]= p_a[5] + (dat5 * off5_0);
        p_a[6]= p_a[6] + (dat6 * off6_0);
        p_a[7]= p_a[7] + (dat7 * off7_0);
        p_a[8]= p_a[8] + (dat8 * off8_0);
        p_a[9]= p_a[9] + (dat9 * off9_0);
        p_a[10]= p_a[10] + (dat10 * off10_0);
        p_a[11]= p_a[11] + (dat11 * off11_0);
        p_a[12]= p_a[12] + (dat12 * off12_0);
        p_a[13]= p_a[13] + (dat13 * off13_0);
      }

      //shift pointers to data y direction
      p_a[0]= p_a[0] + (dat0 * off0_1);
      p_a[1]= p_a[1] + (dat1 * off1_1);
      p_a[2]= p_a[2] + (dat2 * off2_1);
      p_a[3]= p_a[3] + (dat3 * off3_1);
      p_a[4]= p_a[4] + (dat4 * off4_1);
      p_a[5]= p_a[5] + (dat5 * off5_1);
      p_a[6]= p_a[6] + (dat6 * off6_1);
      p_a[7]= p_a[7] + (dat7 * off7_1);
      p_a[8]= p_a[8] + (dat8 * off8_1);
      p_a[9]= p_a[9] + (dat9 * off9_1);
      p_a[10]= p_a[10] + (dat10 * off10_1);
      p_a[11]= p_a[11] + (dat11 * off11_1);
      p_a[12]= p_a[12] + (dat12 * off12_1);
      p_a[13]= p_a[13] + (dat13 * off13_1);
    }
    //shift pointers to data z direction
    p_a[0]= p_a[0] + (dat0 * off0_2);
    p_a[1]= p_a[1] + (dat1 * off1_2);
    p_a[2]= p_a[2] + (dat2 * off2_2);
    p_a[3]= p_a[3] + (dat3 * off3_2);
    p_a[4]= p_a[4] + (dat4 * off4_2);
    p_a[5]= p_a[5] + (dat5 * off5_2);
    p_a[6]= p_a[6] + (dat6 * off6_2);
    p_a[7]= p_a[7] + (dat7 * off7_2);
    p_a[8]= p_a[8] + (dat8 * off8_2);
    p_a[9]= p_a[9] + (dat9 * off9_2);
    p_a[10]= p_a[10] + (dat10 * off10_2);
    p_a[11]= p_a[11] + (dat11 * off11_2);
    p_a[12]= p_a[12] + (dat12 * off12_2);
    p_a[13]= p_a[13] + (dat13 * off13_2);
  }
  ops_timers_core(&c2,&t2);
  OPS_kernels[5].time += t2-t1;
  ops_set_dirtybit_host(args, 14);
  ops_set_halo_dirtybit3(&args[4],range);
  ops_set_halo_dirtybit3(&args[8],range);
  ops_set_halo_dirtybit3(&args[11],range);

  //Update kernel record
  OPS_kernels[5].transfer += ops_compute_transfer(dim, range, &arg0);
  OPS_kernels[5].transfer += ops_compute_transfer(dim, range, &arg1);
  OPS_kernels[5].transfer += ops_compute_transfer(dim, range, &arg2);
  OPS_kernels[5].transfer += ops_compute_transfer(dim, range, &arg3);
  OPS_kernels[5].transfer += ops_compute_transfer(dim, range, &arg4);
  OPS_kernels[5].transfer += ops_compute_transfer(dim, range, &arg5);
  OPS_kernels[5].transfer += ops_compute_transfer(dim, range, &arg6);
  OPS_kernels[5].transfer += ops_compute_transfer(dim, range, &arg7);
  OPS_kernels[5].transfer += ops_compute_transfer(dim, range, &arg8);
  OPS_kernels[5].transfer += ops_compute_transfer(dim, range, &arg9);
  OPS_kernels[5].transfer += ops_compute_transfer(dim, range, &arg10);
  OPS_kernels[5].transfer += ops_compute_transfer(dim, range, &arg11);
  OPS_kernels[5].transfer += ops_compute_transfer(dim, range, &arg12);
  OPS_kernels[5].transfer += ops_compute_transfer(dim, range, &arg13);
}
// host stub function
void ops_par_loop_limiter_kernel(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;
  ops_arg args[3] = {arg0, arg1, arg2};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 3, range, 8))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(8, "limiter_kernel");
    OPS_kernels[8].count++;
    ops_timers_core(&c1, &t1);
  }

  // compute localy 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 // OPS_MPI

  int arg_idx[1];
  int arg_idx_base[1];
#ifdef OPS_MPI
  if (compute_ranges(args, 3, 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
  for (int n = 0; n < 1; n++) {
    arg_idx_base[n] = arg_idx[n];
  }

  int dat0 = args[0].dat->elem_size;
  int dat1 = args[1].dat->elem_size;
  int dat2 = args[2].dat->elem_size;

  // set up initial pointers
  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];
#ifdef OPS_GPU
  double *p_a0 = (double *)((char *)args[0].data_d + base0);
#else
  double *p_a0 = (double *)((char *)args[0].data + base0);
#endif

  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];
#ifdef OPS_GPU
  double *p_a1 = (double *)((char *)args[1].data_d + base1);
#else
  double *p_a1 = (double *)((char *)args[1].data + base1);
#endif

  int base2 = args[2].dat->base_offset +
              (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
                  start[0] * args[2].stencil->stride[0];
#ifdef OPS_GPU
  double *p_a2 = (double *)((char *)args[2].data_d + base2);
#else
  double *p_a2 = (double *)((char *)args[2].data + base2);
#endif

  int x_size = MAX(0, end[0] - start[0]);

  // initialize global variable with the dimension of dats
  xdim0 = args[0].dat->size[0];
  xdim1 = args[1].dat->size[0];
  xdim2 = args[2].dat->size[0];
  if (xdim0 != xdim0_limiter_kernel_h || xdim1 != xdim1_limiter_kernel_h ||
      xdim2 != xdim2_limiter_kernel_h) {
    xdim0_limiter_kernel = xdim0;
    xdim0_limiter_kernel_h = xdim0;
    xdim1_limiter_kernel = xdim1;
    xdim1_limiter_kernel_h = xdim1;
    xdim2_limiter_kernel = xdim2;
    xdim2_limiter_kernel_h = xdim2;
  }

// Halo Exchanges

#ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 3);
#else
  ops_H_D_exchanges_host(args, 3);
#endif
  ops_halo_exchanges(args, 3, range);

#ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 3);
#else
  ops_H_D_exchanges_host(args, 3);
#endif
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[8].mpi_time += t2 - t1;
  }

  limiter_kernel_c_wrapper(p_a0, p_a1, p_a2, x_size);

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[8].time += t1 - t2;
  }
#ifdef OPS_GPU
  ops_set_dirtybit_device(args, 3);
#else
  ops_set_dirtybit_host(args, 3);
#endif
  ops_set_halo_dirtybit3(&args[1], range);
  ops_set_halo_dirtybit3(&args[2], range);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[8].mpi_time += t2 - t1;
    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);
  }
}
// host stub function
void ops_par_loop_updateRK3_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][1];
  ops_arg args[11] = {arg0, arg1, arg2, arg3, arg4, arg5,
                      arg6, arg7, arg8, arg9, arg10};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 11, range, 6))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(6, "updateRK3_kernel");
    OPS_kernels[6].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[1];
  int end[1];
  int arg_idx[1];

#ifdef OPS_MPI
  if (!sb->owned)
    return;
  for (int n = 0; n < 1; 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 < 1; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
  }
#endif
#ifdef OPS_DEBUG
  ops_register_args(args, "updateRK3_kernel");
#endif

  offs[0][0] = args[0].stencil->stride[0] * 1; // unit step in x dimension

  offs[1][0] = args[1].stencil->stride[0] * 1; // unit step in x dimension

  offs[2][0] = args[2].stencil->stride[0] * 1; // unit step in x dimension

  offs[3][0] = args[3].stencil->stride[0] * 1; // unit step in x dimension

  offs[4][0] = args[4].stencil->stride[0] * 1; // unit step in x dimension

  offs[5][0] = args[5].stencil->stride[0] * 1; // unit step in x dimension

  offs[6][0] = args[6].stencil->stride[0] * 1; // unit step in x dimension

  offs[7][0] = args[7].stencil->stride[0] * 1; // unit step in x dimension

  offs[8][0] = args[8].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);
  int off1_0 = offs[1][0];
  int dat1 = (OPS_soa ? args[1].dat->type_size : args[1].dat->elem_size);
  int off2_0 = offs[2][0];
  int dat2 = (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size);
  int off3_0 = offs[3][0];
  int dat3 = (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size);
  int off4_0 = offs[4][0];
  int dat4 = (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size);
  int off5_0 = offs[5][0];
  int dat5 = (OPS_soa ? args[5].dat->type_size : args[5].dat->elem_size);
  int off6_0 = offs[6][0];
  int dat6 = (OPS_soa ? args[6].dat->type_size : args[6].dat->elem_size);
  int off7_0 = offs[7][0];
  int dat7 = (OPS_soa ? args[7].dat->type_size : args[7].dat->elem_size);
  int off8_0 = offs[8][0];
  int dat8 = (OPS_soa ? args[8].dat->type_size : args[8].dat->elem_size);

  // Halo Exchanges
  ops_H_D_exchanges_host(args, 11);
  ops_halo_exchanges(args, 11, range);
  ops_H_D_exchanges_host(args, 11);

#ifdef _OPENMP
  int nthreads = omp_get_max_threads();
#else
  int nthreads = 1;
#endif
  xdim0 = args[0].dat->size[0];
  xdim1 = args[1].dat->size[0];
  xdim2 = args[2].dat->size[0];
  xdim3 = args[3].dat->size[0];
  xdim4 = args[4].dat->size[0];
  xdim5 = args[5].dat->size[0];
  xdim6 = args[6].dat->size[0];
  xdim7 = args[7].dat->size[0];
  xdim8 = args[8].dat->size[0];

  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[6].mpi_time += t2 - t1;
  }

#pragma omp parallel for
  for (int thr = 0; thr < nthreads; thr++) {

    int x_size = end[0] - start[0];
    char *p_a[11];

    int start_i = start[0] + ((x_size - 1) / nthreads + 1) * thr;
    int finish_i =
        start[0] + MIN(((x_size - 1) / nthreads + 1) * (thr + 1), x_size);

    // get address per thread
    int start0 = start_i;

    // set up initial pointers
    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
    for (int d = 0; d < dim; d++)
      d_m[d] = args[0].dat->d_m[d];
#endif
    int base0 = dat0 * 1 * (start0 * args[0].stencil->stride[0] -
                            args[0].dat->base[0] - d_m[0]);
    p_a[0] = (char *)args[0].data + base0;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[1].dat->d_m[d] + OPS_sub_dat_list[args[1].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[1].dat->d_m[d];
#endif
    int base1 = dat1 * 1 * (start0 * args[1].stencil->stride[0] -
                            args[1].dat->base[0] - d_m[0]);
    p_a[1] = (char *)args[1].data + base1;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[2].dat->d_m[d] + OPS_sub_dat_list[args[2].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[2].dat->d_m[d];
#endif
    int base2 = dat2 * 1 * (start0 * args[2].stencil->stride[0] -
                            args[2].dat->base[0] - d_m[0]);
    p_a[2] = (char *)args[2].data + base2;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[3].dat->d_m[d] + OPS_sub_dat_list[args[3].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[3].dat->d_m[d];
#endif
    int base3 = dat3 * 1 * (start0 * args[3].stencil->stride[0] -
                            args[3].dat->base[0] - d_m[0]);
    p_a[3] = (char *)args[3].data + base3;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[4].dat->d_m[d] + OPS_sub_dat_list[args[4].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[4].dat->d_m[d];
#endif
    int base4 = dat4 * 1 * (start0 * args[4].stencil->stride[0] -
                            args[4].dat->base[0] - d_m[0]);
    p_a[4] = (char *)args[4].data + base4;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[5].dat->d_m[d] + OPS_sub_dat_list[args[5].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[5].dat->d_m[d];
#endif
    int base5 = dat5 * 1 * (start0 * args[5].stencil->stride[0] -
                            args[5].dat->base[0] - d_m[0]);
    p_a[5] = (char *)args[5].data + base5;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[6].dat->d_m[d] + OPS_sub_dat_list[args[6].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[6].dat->d_m[d];
#endif
    int base6 = dat6 * 1 * (start0 * args[6].stencil->stride[0] -
                            args[6].dat->base[0] - d_m[0]);
    p_a[6] = (char *)args[6].data + base6;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[7].dat->d_m[d] + OPS_sub_dat_list[args[7].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[7].dat->d_m[d];
#endif
    int base7 = dat7 * 1 * (start0 * args[7].stencil->stride[0] -
                            args[7].dat->base[0] - d_m[0]);
    p_a[7] = (char *)args[7].data + base7;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[8].dat->d_m[d] + OPS_sub_dat_list[args[8].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[8].dat->d_m[d];
#endif
    int base8 = dat8 * 1 * (start0 * args[8].stencil->stride[0] -
                            args[8].dat->base[0] - d_m[0]);
    p_a[8] = (char *)args[8].data + base8;

    p_a[9] = (char *)args[9].data;

    p_a[10] = (char *)args[10].data;

    for (int n_x = start_i; n_x < start_i + (finish_i - start_i) / SIMD_VEC;
         n_x++) {
// call kernel function, passing in pointers to data -vectorised
#pragma simd
      for (int i = 0; i < SIMD_VEC; i++) {
        updateRK3_kernel(
            (double *)p_a[0] + i * 1 * 1, (double *)p_a[1] + i * 1 * 1,
            (double *)p_a[2] + i * 1 * 1, (double *)p_a[3] + i * 1 * 1,
            (double *)p_a[4] + i * 1 * 1, (double *)p_a[5] + i * 1 * 1,
            (const double *)p_a[6] + i * 1 * 1,
            (const double *)p_a[7] + i * 1 * 1,
            (const double *)p_a[8] + i * 1 * 1, (double *)p_a[9],
            (double *)p_a[10]);
      }

      // shift pointers to data x direction
      p_a[0] = p_a[0] + (dat0 * off0_0) * SIMD_VEC;
      p_a[1] = p_a[1] + (dat1 * off1_0) * SIMD_VEC;
      p_a[2] = p_a[2] + (dat2 * off2_0) * SIMD_VEC;
      p_a[3] = p_a[3] + (dat3 * off3_0) * SIMD_VEC;
      p_a[4] = p_a[4] + (dat4 * off4_0) * SIMD_VEC;
      p_a[5] = p_a[5] + (dat5 * off5_0) * SIMD_VEC;
      p_a[6] = p_a[6] + (dat6 * off6_0) * SIMD_VEC;
      p_a[7] = p_a[7] + (dat7 * off7_0) * SIMD_VEC;
      p_a[8] = p_a[8] + (dat8 * off8_0) * SIMD_VEC;
    }

    for (int n_x = start_i + ((finish_i - start_i) / SIMD_VEC) * SIMD_VEC;
         n_x < finish_i; n_x++) {
      // call kernel function, passing in pointers to data - remainder
      updateRK3_kernel((double *)p_a[0], (double *)p_a[1], (double *)p_a[2],
                       (double *)p_a[3], (double *)p_a[4], (double *)p_a[5],
                       (const double *)p_a[6], (const double *)p_a[7],
                       (const double *)p_a[8], (double *)p_a[9],
                       (double *)p_a[10]);

      // shift pointers to data x direction
      p_a[0] = p_a[0] + (dat0 * off0_0);
      p_a[1] = p_a[1] + (dat1 * off1_0);
      p_a[2] = p_a[2] + (dat2 * off2_0);
      p_a[3] = p_a[3] + (dat3 * off3_0);
      p_a[4] = p_a[4] + (dat4 * off4_0);
      p_a[5] = p_a[5] + (dat5 * off5_0);
      p_a[6] = p_a[6] + (dat6 * off6_0);
      p_a[7] = p_a[7] + (dat7 * off7_0);
      p_a[8] = p_a[8] + (dat8 * off8_0);
    }
  }

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[6].time += t1 - t2;
  }

  ops_set_dirtybit_host(args, 11);

  ops_set_halo_dirtybit3(&args[0], range);
  ops_set_halo_dirtybit3(&args[1], range);
  ops_set_halo_dirtybit3(&args[2], range);
  ops_set_halo_dirtybit3(&args[3], range);
  ops_set_halo_dirtybit3(&args[4], range);
  ops_set_halo_dirtybit3(&args[5], range);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[6].mpi_time += t2 - t1;
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg2);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg3);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg4);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg5);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg6);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg7);
    OPS_kernels[6].transfer += ops_compute_transfer(dim, start, end, &arg8);
  }
}
// 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];

  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);
  int off2_0 = offs[2][0];
  int off2_1 = offs[2][1];
  int dat2 = (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size);
  int off3_0 = offs[3][0];
  int off3_1 = offs[3][1];
  int dat3 = (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size);
  int off4_0 = offs[4][0];
  int off4_1 = offs[4][1];
  int dat4 = (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size);
  int off5_0 = offs[5][0];
  int off5_1 = offs[5][1];
  int dat5 = (OPS_soa ? args[5].dat->type_size : args[5].dat->elem_size);

#ifdef OPS_MPI
  double *arg6h =
      (double *)(((ops_reduction)args[6].data)->data +
                 ((ops_reduction)args[6].data)->size * block->index);
#else
  double *arg6h = (double *)(((ops_reduction)args[6].data)->data);
#endif
#ifdef OPS_MPI
  double *arg7h =
      (double *)(((ops_reduction)args[7].data)->data +
                 ((ops_reduction)args[7].data)->size * block->index);
#else
  double *arg7h = (double *)(((ops_reduction)args[7].data)->data);
#endif
#ifdef OPS_MPI
  double *arg8h =
      (double *)(((ops_reduction)args[8].data)->data +
                 ((ops_reduction)args[8].data)->size * block->index);
#else
  double *arg8h = (double *)(((ops_reduction)args[8].data)->data);
#endif
#ifdef OPS_MPI
  double *arg9h =
      (double *)(((ops_reduction)args[9].data)->data +
                 ((ops_reduction)args[9].data)->size * block->index);
#else
  double *arg9h = (double *)(((ops_reduction)args[9].data)->data);
#endif
#ifdef OPS_MPI
  double *arg10h =
      (double *)(((ops_reduction)args[10].data)->data +
                 ((ops_reduction)args[10].data)->size * block->index);
#else
  double *arg10h = (double *)(((ops_reduction)args[10].data)->data);
#endif
  // Halo Exchanges
  ops_H_D_exchanges_host(args, 11);
  ops_halo_exchanges(args, 11, range);
  ops_H_D_exchanges_host(args, 11);

#ifdef _OPENMP
  int nthreads = omp_get_max_threads();
#else
  int nthreads = 1;
#endif
  // allocate and initialise arrays for global reduction
  // assumes a max of MAX_REDUCT_THREADS threads with a cacche line size of 64
  // bytes
  double arg_gbl6[MAX(1, 64) * MAX_REDUCT_THREADS];
  double arg_gbl7[MAX(1, 64) * MAX_REDUCT_THREADS];
  double arg_gbl8[MAX(1, 64) * MAX_REDUCT_THREADS];
  double arg_gbl9[MAX(1, 64) * MAX_REDUCT_THREADS];
  double arg_gbl10[MAX(1, 64) * MAX_REDUCT_THREADS];
  for (int thr = 0; thr < nthreads; thr++) {
    for (int d = 0; d < 1; d++) {
      arg_gbl6[d + 64 * thr] = ZERO_double;
    }
    for (int d = 0; d < 1; d++) {
      arg_gbl7[d + 64 * thr] = ZERO_double;
    }
    for (int d = 0; d < 1; d++) {
      arg_gbl8[d + 64 * thr] = ZERO_double;
    }
    for (int d = 0; d < 1; d++) {
      arg_gbl9[d + 64 * thr] = ZERO_double;
    }
    for (int d = 0; d < 1; d++) {
      arg_gbl10[d + 64 * thr] = ZERO_double;
    }
  }
  xdim0 = args[0].dat->size[0];
  xdim1 = args[1].dat->size[0];
  xdim2 = args[2].dat->size[0];
  xdim3 = args[3].dat->size[0];
  xdim4 = args[4].dat->size[0];
  xdim5 = args[5].dat->size[0];

  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[49].mpi_time += t2 - t1;
  }

#pragma omp parallel for
  for (int thr = 0; thr < nthreads; thr++) {

    int y_size = end[1] - start[1];
    char *p_a[11];

    int start_i = start[1] + ((y_size - 1) / nthreads + 1) * thr;
    int finish_i =
        start[1] + MIN(((y_size - 1) / nthreads + 1) * (thr + 1), y_size);

    // get address per thread
    int start0 = start[0];
    int start1 = start_i;

    // set up initial pointers
    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
    for (int d = 0; d < dim; d++)
      d_m[d] = args[0].dat->d_m[d];
#endif
    int base0 = dat0 * 1 * (start0 * args[0].stencil->stride[0] -
                            args[0].dat->base[0] - d_m[0]);
    base0 = base0 +
            dat0 * args[0].dat->size[0] * (start1 * args[0].stencil->stride[1] -
                                           args[0].dat->base[1] - d_m[1]);
    p_a[0] = (char *)args[0].data + base0;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[1].dat->d_m[d] + OPS_sub_dat_list[args[1].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[1].dat->d_m[d];
#endif
    int base1 = dat1 * 1 * (start0 * args[1].stencil->stride[0] -
                            args[1].dat->base[0] - d_m[0]);
    base1 = base1 +
            dat1 * args[1].dat->size[0] * (start1 * args[1].stencil->stride[1] -
                                           args[1].dat->base[1] - d_m[1]);
    p_a[1] = (char *)args[1].data + base1;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[2].dat->d_m[d] + OPS_sub_dat_list[args[2].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[2].dat->d_m[d];
#endif
    int base2 = dat2 * 1 * (start0 * args[2].stencil->stride[0] -
                            args[2].dat->base[0] - d_m[0]);
    base2 = base2 +
            dat2 * args[2].dat->size[0] * (start1 * args[2].stencil->stride[1] -
                                           args[2].dat->base[1] - d_m[1]);
    p_a[2] = (char *)args[2].data + base2;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[3].dat->d_m[d] + OPS_sub_dat_list[args[3].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[3].dat->d_m[d];
#endif
    int base3 = dat3 * 1 * (start0 * args[3].stencil->stride[0] -
                            args[3].dat->base[0] - d_m[0]);
    base3 = base3 +
            dat3 * args[3].dat->size[0] * (start1 * args[3].stencil->stride[1] -
                                           args[3].dat->base[1] - d_m[1]);
    p_a[3] = (char *)args[3].data + base3;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[4].dat->d_m[d] + OPS_sub_dat_list[args[4].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[4].dat->d_m[d];
#endif
    int base4 = dat4 * 1 * (start0 * args[4].stencil->stride[0] -
                            args[4].dat->base[0] - d_m[0]);
    base4 = base4 +
            dat4 * args[4].dat->size[0] * (start1 * args[4].stencil->stride[1] -
                                           args[4].dat->base[1] - d_m[1]);
    p_a[4] = (char *)args[4].data + base4;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[5].dat->d_m[d] + OPS_sub_dat_list[args[5].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[5].dat->d_m[d];
#endif
    int base5 = dat5 * 1 * (start0 * args[5].stencil->stride[0] -
                            args[5].dat->base[0] - d_m[0]);
    base5 = base5 +
            dat5 * args[5].dat->size[0] * (start1 * args[5].stencil->stride[1] -
                                           args[5].dat->base[1] - d_m[1]);
    p_a[5] = (char *)args[5].data + base5;

    p_a[6] = (char *)arg6h;

    p_a[7] = (char *)arg7h;

    p_a[8] = (char *)arg8h;

    p_a[9] = (char *)arg9h;

    p_a[10] = (char *)arg10h;

    for (int n_y = start_i; n_y < finish_i; n_y++) {
      for (int n_x = start[0]; n_x < start[0] + (end[0] - start[0]) / SIMD_VEC;
           n_x++) {
        // call kernel function, passing in pointers to data -vectorised
        for (int i = 0; i < SIMD_VEC; i++) {
          field_summary_kernel((const double *)p_a[0] + i * 1 * 1,
                               (const double *)p_a[1] + i * 1 * 1,
                               (const double *)p_a[2] + i * 1 * 1,
                               (const double *)p_a[3] + i * 1 * 1,
                               (const double *)p_a[4] + i * 1 * 1,
                               (const double *)p_a[5] + i * 1 * 1,
                               &arg_gbl6[64 * thr], &arg_gbl7[64 * thr],
                               &arg_gbl8[64 * thr], &arg_gbl9[64 * thr],
                               &arg_gbl10[64 * thr]);
        }

        // shift pointers to data x direction
        p_a[0] = p_a[0] + (dat0 * off0_0) * SIMD_VEC;
        p_a[1] = p_a[1] + (dat1 * off1_0) * SIMD_VEC;
        p_a[2] = p_a[2] + (dat2 * off2_0) * SIMD_VEC;
        p_a[3] = p_a[3] + (dat3 * off3_0) * SIMD_VEC;
        p_a[4] = p_a[4] + (dat4 * off4_0) * SIMD_VEC;
        p_a[5] = p_a[5] + (dat5 * off5_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
        field_summary_kernel((const double *)p_a[0], (const double *)p_a[1],
                             (const double *)p_a[2], (const double *)p_a[3],
                             (const double *)p_a[4], (const double *)p_a[5],
                             &arg_gbl6[64 * thr], &arg_gbl7[64 * thr],
                             &arg_gbl8[64 * thr], &arg_gbl9[64 * thr],
                             &arg_gbl10[64 * thr]);

        // shift pointers to data x direction
        p_a[0] = p_a[0] + (dat0 * off0_0);
        p_a[1] = p_a[1] + (dat1 * off1_0);
        p_a[2] = p_a[2] + (dat2 * off2_0);
        p_a[3] = p_a[3] + (dat3 * off3_0);
        p_a[4] = p_a[4] + (dat4 * off4_0);
        p_a[5] = p_a[5] + (dat5 * off5_0);
      }

      // shift pointers to data y direction
      p_a[0] = p_a[0] + (dat0 * off0_1);
      p_a[1] = p_a[1] + (dat1 * off1_1);
      p_a[2] = p_a[2] + (dat2 * off2_1);
      p_a[3] = p_a[3] + (dat3 * off3_1);
      p_a[4] = p_a[4] + (dat4 * off4_1);
      p_a[5] = p_a[5] + (dat5 * off5_1);
    }
  }

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[49].time += t1 - t2;
  }

  // combine reduction data
  for (int thr = 0; thr < nthreads; thr++) {
    for (int d = 0; d < 1; d++) {
      arg6h[d] += arg_gbl6[64 * thr + d];
    }
    for (int d = 0; d < 1; d++) {
      arg7h[d] += arg_gbl7[64 * thr + d];
    }
    for (int d = 0; d < 1; d++) {
      arg8h[d] += arg_gbl8[64 * thr + d];
    }
    for (int d = 0; d < 1; d++) {
      arg9h[d] += arg_gbl9[64 * thr + d];
    }
    for (int d = 0; d < 1; d++) {
      arg10h[d] += arg_gbl10[64 * thr + d];
    }
  }
  ops_set_dirtybit_host(args, 11);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[49].mpi_time += t2 - t1;
    OPS_kernels[49].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[49].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[49].transfer += ops_compute_transfer(dim, start, end, &arg2);
    OPS_kernels[49].transfer += ops_compute_transfer(dim, start, end, &arg3);
    OPS_kernels[49].transfer += ops_compute_transfer(dim, start, end, &arg4);
    OPS_kernels[49].transfer += ops_compute_transfer(dim, start, end, &arg5);
  }
}
// host stub function
void ops_par_loop_update_halo_kernel1_b1(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_arg args[8] = {arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 8, range, 10))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(10, "update_halo_kernel1_b1");
    OPS_kernels[10].count++;
    ops_timers_core(&c1, &t1);
  }

  // compute localy 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 // OPS_MPI

  int arg_idx[2];
  int arg_idx_base[2];
#ifdef OPS_MPI
  if (compute_ranges(args, 8, 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
  for (int n = 0; n < 2; n++) {
    arg_idx_base[n] = arg_idx[n];
  }

  int dat0 = args[0].dat->elem_size;
  int dat1 = args[1].dat->elem_size;
  int dat2 = args[2].dat->elem_size;
  int dat3 = args[3].dat->elem_size;
  int dat4 = args[4].dat->elem_size;
  int dat5 = args[5].dat->elem_size;
  int dat6 = args[6].dat->elem_size;

  int *arg7h = (int *)arg7.data;
// Upload large globals
#ifdef OPS_GPU
  int consts_bytes = 0;
  consts_bytes += ROUND_UP(NUM_FIELDS * sizeof(int));
  reallocConstArrays(consts_bytes);
  consts_bytes = 0;
  args[7].data = OPS_consts_h + consts_bytes;
  args[7].data_d = OPS_consts_d + consts_bytes;
  for (int d = 0; d < NUM_FIELDS; d++)
    ((int *)args[7].data)[d] = arg7h[d];
  consts_bytes += ROUND_UP(NUM_FIELDS * sizeof(int));
  mvConstArraysToDevice(consts_bytes);
#endif // OPS_GPU

  // set up initial pointers
  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];
#ifdef OPS_GPU
  double *p_a0 = (double *)((char *)args[0].data_d + base0);
#else
  double *p_a0 = (double *)((char *)args[0].data + base0);
#endif

  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];
#ifdef OPS_GPU
  double *p_a1 = (double *)((char *)args[1].data_d + base1);
#else
  double *p_a1 = (double *)((char *)args[1].data + base1);
#endif

  int base2 = args[2].dat->base_offset +
              (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
                  start[0] * args[2].stencil->stride[0];
  base2 = base2 +
          (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
              args[2].dat->size[0] * start[1] * args[2].stencil->stride[1];
#ifdef OPS_GPU
  double *p_a2 = (double *)((char *)args[2].data_d + base2);
#else
  double *p_a2 = (double *)((char *)args[2].data + base2);
#endif

  int base3 = args[3].dat->base_offset +
              (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size) *
                  start[0] * args[3].stencil->stride[0];
  base3 = base3 +
          (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size) *
              args[3].dat->size[0] * start[1] * args[3].stencil->stride[1];
#ifdef OPS_GPU
  double *p_a3 = (double *)((char *)args[3].data_d + base3);
#else
  double *p_a3 = (double *)((char *)args[3].data + base3);
#endif

  int base4 = args[4].dat->base_offset +
              (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size) *
                  start[0] * args[4].stencil->stride[0];
  base4 = base4 +
          (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size) *
              args[4].dat->size[0] * start[1] * args[4].stencil->stride[1];
#ifdef OPS_GPU
  double *p_a4 = (double *)((char *)args[4].data_d + base4);
#else
  double *p_a4 = (double *)((char *)args[4].data + base4);
#endif

  int base5 = args[5].dat->base_offset +
              (OPS_soa ? args[5].dat->type_size : args[5].dat->elem_size) *
                  start[0] * args[5].stencil->stride[0];
  base5 = base5 +
          (OPS_soa ? args[5].dat->type_size : args[5].dat->elem_size) *
              args[5].dat->size[0] * start[1] * args[5].stencil->stride[1];
#ifdef OPS_GPU
  double *p_a5 = (double *)((char *)args[5].data_d + base5);
#else
  double *p_a5 = (double *)((char *)args[5].data + base5);
#endif

  int base6 = args[6].dat->base_offset +
              (OPS_soa ? args[6].dat->type_size : args[6].dat->elem_size) *
                  start[0] * args[6].stencil->stride[0];
  base6 = base6 +
          (OPS_soa ? args[6].dat->type_size : args[6].dat->elem_size) *
              args[6].dat->size[0] * start[1] * args[6].stencil->stride[1];
#ifdef OPS_GPU
  double *p_a6 = (double *)((char *)args[6].data_d + base6);
#else
  double *p_a6 = (double *)((char *)args[6].data + base6);
#endif

#ifdef OPS_GPU
  int *p_a7 = (int *)args[7].data_d;
#else
  int *p_a7 = arg7h;
#endif

  int x_size = MAX(0, end[0] - start[0]);
  int y_size = MAX(0, end[1] - start[1]);

  // initialize global variable with the dimension of dats
  xdim0 = args[0].dat->size[0];
  xdim1 = args[1].dat->size[0];
  xdim2 = args[2].dat->size[0];
  xdim3 = args[3].dat->size[0];
  xdim4 = args[4].dat->size[0];
  xdim5 = args[5].dat->size[0];
  xdim6 = args[6].dat->size[0];
  if (xdim0 != xdim0_update_halo_kernel1_b1_h ||
      xdim1 != xdim1_update_halo_kernel1_b1_h ||
      xdim2 != xdim2_update_halo_kernel1_b1_h ||
      xdim3 != xdim3_update_halo_kernel1_b1_h ||
      xdim4 != xdim4_update_halo_kernel1_b1_h ||
      xdim5 != xdim5_update_halo_kernel1_b1_h ||
      xdim6 != xdim6_update_halo_kernel1_b1_h) {
    xdim0_update_halo_kernel1_b1 = xdim0;
    xdim0_update_halo_kernel1_b1_h = xdim0;
    xdim1_update_halo_kernel1_b1 = xdim1;
    xdim1_update_halo_kernel1_b1_h = xdim1;
    xdim2_update_halo_kernel1_b1 = xdim2;
    xdim2_update_halo_kernel1_b1_h = xdim2;
    xdim3_update_halo_kernel1_b1 = xdim3;
    xdim3_update_halo_kernel1_b1_h = xdim3;
    xdim4_update_halo_kernel1_b1 = xdim4;
    xdim4_update_halo_kernel1_b1_h = xdim4;
    xdim5_update_halo_kernel1_b1 = xdim5;
    xdim5_update_halo_kernel1_b1_h = xdim5;
    xdim6_update_halo_kernel1_b1 = xdim6;
    xdim6_update_halo_kernel1_b1_h = xdim6;
  }

// Halo Exchanges

#ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 8);
#else
  ops_H_D_exchanges_host(args, 8);
#endif
  ops_halo_exchanges(args, 8, range);

#ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 8);
#else
  ops_H_D_exchanges_host(args, 8);
#endif
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[10].mpi_time += t2 - t1;
  }

  update_halo_kernel1_b1_c_wrapper(p_a0, p_a1, p_a2, p_a3, p_a4, p_a5, p_a6,
                                   p_a7, x_size, y_size);

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[10].time += t1 - t2;
  }
#ifdef OPS_GPU
  ops_set_dirtybit_device(args, 8);
#else
  ops_set_dirtybit_host(args, 8);
#endif
  ops_set_halo_dirtybit3(&args[0], range);
  ops_set_halo_dirtybit3(&args[1], range);
  ops_set_halo_dirtybit3(&args[2], range);
  ops_set_halo_dirtybit3(&args[3], range);
  ops_set_halo_dirtybit3(&args[4], range);
  ops_set_halo_dirtybit3(&args[5], range);
  ops_set_halo_dirtybit3(&args[6], range);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[10].mpi_time += t2 - t1;
    OPS_kernels[10].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[10].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[10].transfer += ops_compute_transfer(dim, start, end, &arg2);
    OPS_kernels[10].transfer += ops_compute_transfer(dim, start, end, &arg3);
    OPS_kernels[10].transfer += ops_compute_transfer(dim, start, end, &arg4);
    OPS_kernels[10].transfer += ops_compute_transfer(dim, start, end, &arg5);
    OPS_kernels[10].transfer += ops_compute_transfer(dim, start, end, &arg6);
  }
}
// host stub function
void ops_par_loop_update_halo_kernel1_b2(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) {

  char *p_a[8];
  int  offs[8][2];
  ops_arg args[8] = { arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7};



  #ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args,8,range,43)) return;
  #endif

  ops_timing_realloc(43,"update_halo_kernel1_b2");
  OPS_kernels[43].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, "update_halo_kernel1_b2");
  #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];

  offs[6][0] = args[6].stencil->stride[0]*1;  //unit step in x dimension
  offs[6][1] = off2D(1, &start[0],
      &end[0],args[6].dat->size, args[6].stencil->stride) - offs[6][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;
  int off4_0 = offs[4][0];
  int off4_1 = offs[4][1];
  int dat4 = args[4].dat->elem_size;
  int off5_0 = offs[5][0];
  int off5_1 = offs[5][1];
  int dat5 = args[5].dat->elem_size;
  int off6_0 = offs[6][0];
  int off6_1 = offs[6][1];
  int dat6 = args[6].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;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d] + OPS_sub_dat_list[args[1].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d];
  #endif //OPS_MPI
  int base1 = dat1 * 1 * 
    (start[0] * args[1].stencil->stride[0] - args[1].dat->base[0] - d_m[0]);
  base1 = base1+ dat1 *
    args[1].dat->size[0] *
    (start[1] * args[1].stencil->stride[1] - args[1].dat->base[1] - d_m[1]);
  p_a[1] = (char *)args[1].data + base1;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[2].dat->d_m[d] + OPS_sub_dat_list[args[2].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[2].dat->d_m[d];
  #endif //OPS_MPI
  int base2 = dat2 * 1 * 
    (start[0] * args[2].stencil->stride[0] - args[2].dat->base[0] - d_m[0]);
  base2 = base2+ dat2 *
    args[2].dat->size[0] *
    (start[1] * args[2].stencil->stride[1] - args[2].dat->base[1] - d_m[1]);
  p_a[2] = (char *)args[2].data + base2;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[3].dat->d_m[d] + OPS_sub_dat_list[args[3].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[3].dat->d_m[d];
  #endif //OPS_MPI
  int base3 = dat3 * 1 * 
    (start[0] * args[3].stencil->stride[0] - args[3].dat->base[0] - d_m[0]);
  base3 = base3+ dat3 *
    args[3].dat->size[0] *
    (start[1] * args[3].stencil->stride[1] - args[3].dat->base[1] - d_m[1]);
  p_a[3] = (char *)args[3].data + base3;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[4].dat->d_m[d] + OPS_sub_dat_list[args[4].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[4].dat->d_m[d];
  #endif //OPS_MPI
  int base4 = dat4 * 1 * 
    (start[0] * args[4].stencil->stride[0] - args[4].dat->base[0] - d_m[0]);
  base4 = base4+ dat4 *
    args[4].dat->size[0] *
    (start[1] * args[4].stencil->stride[1] - args[4].dat->base[1] - d_m[1]);
  p_a[4] = (char *)args[4].data + base4;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[5].dat->d_m[d] + OPS_sub_dat_list[args[5].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[5].dat->d_m[d];
  #endif //OPS_MPI
  int base5 = dat5 * 1 * 
    (start[0] * args[5].stencil->stride[0] - args[5].dat->base[0] - d_m[0]);
  base5 = base5+ dat5 *
    args[5].dat->size[0] *
    (start[1] * args[5].stencil->stride[1] - args[5].dat->base[1] - d_m[1]);
  p_a[5] = (char *)args[5].data + base5;

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[6].dat->d_m[d] + OPS_sub_dat_list[args[6].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[6].dat->d_m[d];
  #endif //OPS_MPI
  int base6 = dat6 * 1 * 
    (start[0] * args[6].stencil->stride[0] - args[6].dat->base[0] - d_m[0]);
  base6 = base6+ dat6 *
    args[6].dat->size[0] *
    (start[1] * args[6].stencil->stride[1] - args[6].dat->base[1] - d_m[1]);
  p_a[6] = (char *)args[6].data + base6;

  p_a[7] = args[7].data;



  ops_H_D_exchanges_host(args, 8);
  ops_halo_exchanges(args,8,range);
  ops_H_D_exchanges_host(args, 8);

  ops_timers_core(&c1,&t1);
  OPS_kernels[43].mpi_time += t1-t2;

  xdim0 = args[0].dat->size[0]*args[0].dat->dim;
  xdim1 = args[1].dat->size[0]*args[1].dat->dim;
  xdim2 = args[2].dat->size[0]*args[2].dat->dim;
  xdim3 = args[3].dat->size[0]*args[3].dat->dim;
  xdim4 = args[4].dat->size[0]*args[4].dat->dim;
  xdim5 = args[5].dat->size[0]*args[5].dat->dim;
  xdim6 = args[6].dat->size[0]*args[6].dat->dim;

  int n_x;
  for ( int n_y=start[1]; n_y<end[1]; n_y++ ){
    #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
      #pragma simd
      for ( int i=0; i<SIMD_VEC; i++ ){
        update_halo_kernel1_b2(  (double *)p_a[0]+ i*1, (double *)p_a[1]+ i*1, (double *)p_a[2]+ i*1,
           (double *)p_a[3]+ i*1, (double *)p_a[4]+ i*1, (double *)p_a[5]+ i*1, (double *)p_a[6]+ i*1,
           (int *)p_a[7] );

      }

      //shift pointers to data x direction
      p_a[0]= p_a[0] + (dat0 * off0_0)*SIMD_VEC;
      p_a[1]= p_a[1] + (dat1 * off1_0)*SIMD_VEC;
      p_a[2]= p_a[2] + (dat2 * off2_0)*SIMD_VEC;
      p_a[3]= p_a[3] + (dat3 * off3_0)*SIMD_VEC;
      p_a[4]= p_a[4] + (dat4 * off4_0)*SIMD_VEC;
      p_a[5]= p_a[5] + (dat5 * off5_0)*SIMD_VEC;
      p_a[6]= p_a[6] + (dat6 * off6_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
      update_halo_kernel1_b2(  (double *)p_a[0], (double *)p_a[1], (double *)p_a[2],
           (double *)p_a[3], (double *)p_a[4], (double *)p_a[5], (double *)p_a[6],
           (int *)p_a[7] );


      //shift pointers to data x direction
      p_a[0]= p_a[0] + (dat0 * off0_0);
      p_a[1]= p_a[1] + (dat1 * off1_0);
      p_a[2]= p_a[2] + (dat2 * off2_0);
      p_a[3]= p_a[3] + (dat3 * off3_0);
      p_a[4]= p_a[4] + (dat4 * off4_0);
      p_a[5]= p_a[5] + (dat5 * off5_0);
      p_a[6]= p_a[6] + (dat6 * off6_0);
    }

    //shift pointers to data y direction
    p_a[0]= p_a[0] + (dat0 * off0_1);
    p_a[1]= p_a[1] + (dat1 * off1_1);
    p_a[2]= p_a[2] + (dat2 * off2_1);
    p_a[3]= p_a[3] + (dat3 * off3_1);
    p_a[4]= p_a[4] + (dat4 * off4_1);
    p_a[5]= p_a[5] + (dat5 * off5_1);
    p_a[6]= p_a[6] + (dat6 * off6_1);
  }
  ops_timers_core(&c2,&t2);
  OPS_kernels[43].time += t2-t1;
  ops_set_dirtybit_host(args, 8);
  ops_set_halo_dirtybit3(&args[0],range);
  ops_set_halo_dirtybit3(&args[1],range);
  ops_set_halo_dirtybit3(&args[2],range);
  ops_set_halo_dirtybit3(&args[3],range);
  ops_set_halo_dirtybit3(&args[4],range);
  ops_set_halo_dirtybit3(&args[5],range);
  ops_set_halo_dirtybit3(&args[6],range);

  //Update kernel record
  OPS_kernels[43].transfer += ops_compute_transfer(dim, range, &arg0);
  OPS_kernels[43].transfer += ops_compute_transfer(dim, range, &arg1);
  OPS_kernels[43].transfer += ops_compute_transfer(dim, range, &arg2);
  OPS_kernels[43].transfer += ops_compute_transfer(dim, range, &arg3);
  OPS_kernels[43].transfer += ops_compute_transfer(dim, range, &arg4);
  OPS_kernels[43].transfer += ops_compute_transfer(dim, range, &arg5);
  OPS_kernels[43].transfer += ops_compute_transfer(dim, range, &arg6);
}
// host stub function
void ops_par_loop_tea_leaf_cg_calc_w_reduce_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) {

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

  char *p_a[7];
  int offs[7][2];
  ops_arg args[7] = {arg0, arg1, arg2, arg3, arg4, arg5, arg6};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 7, range, 19))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(19, "tea_leaf_cg_calc_w_reduce_kernel");
    OPS_kernels[19].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, "tea_leaf_cg_calc_w_reduce_kernel");
#endif

  int arg_idx[2];
  int arg_idx_base[2];
#ifdef OPS_MPI
  if (compute_ranges(args, 7, 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];

  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];

  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);
  int off2_0 = offs[2][0];
  int off2_1 = offs[2][1];
  int dat2 = (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size);
  int off3_0 = offs[3][0];
  int off3_1 = offs[3][1];
  int dat3 = (OPS_soa ? args[3].dat->type_size : args[3].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;

  int base2 = args[2].dat->base_offset +
              (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
                  start[0] * args[2].stencil->stride[0];
  base2 = base2 +
          (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
              args[2].dat->size[0] * start[1] * args[2].stencil->stride[1];
  p_a[2] = (char *)args[2].data + base2;

  int base3 = args[3].dat->base_offset +
              (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size) *
                  start[0] * args[3].stencil->stride[0];
  base3 = base3 +
          (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size) *
              args[3].dat->size[0] * start[1] * args[3].stencil->stride[1];
  p_a[3] = (char *)args[3].data + base3;

  p_a[4] = args[4].data;

  p_a[5] = args[5].data;

#ifdef OPS_MPI
  p_a[6] = ((ops_reduction)args[6].data)->data +
           ((ops_reduction)args[6].data)->size * block->index;
#else
  p_a[6] = ((ops_reduction)args[6].data)->data;
#endif

  // initialize global variable with the dimension of dats
  xdim0 = args[0].dat->size[0];
  xdim1 = args[1].dat->size[0];
  xdim2 = args[2].dat->size[0];
  xdim3 = args[3].dat->size[0];

  // Halo Exchanges
  ops_H_D_exchanges_host(args, 7);
  ops_halo_exchanges(args, 7, range);
  ops_H_D_exchanges_host(args, 7);

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[19].mpi_time += t1 - t2;
  }

  int n_x;
  for (int n_y = start[1]; n_y < end[1]; n_y++) {
#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++) {
        tea_leaf_cg_calc_w_reduce_kernel(
            (double *)p_a[0] + i * 1 * 1, (double *)p_a[1] + i * 1 * 1,
            (double *)p_a[2] + i * 1 * 1, (double *)p_a[3] + i * 1 * 1,
            (double *)p_a[4], (double *)p_a[5], (double *)p_a[6]);
      }

      // shift pointers to data x direction
      p_a[0] = p_a[0] + (dat0 * off0_0) * SIMD_VEC;
      p_a[1] = p_a[1] + (dat1 * off1_0) * SIMD_VEC;
      p_a[2] = p_a[2] + (dat2 * off2_0) * SIMD_VEC;
      p_a[3] = p_a[3] + (dat3 * off3_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
      tea_leaf_cg_calc_w_reduce_kernel((double *)p_a[0], (double *)p_a[1],
                                       (double *)p_a[2], (double *)p_a[3],
                                       (double *)p_a[4], (double *)p_a[5],
                                       (double *)p_a[6]);

      // shift pointers to data x direction
      p_a[0] = p_a[0] + (dat0 * off0_0);
      p_a[1] = p_a[1] + (dat1 * off1_0);
      p_a[2] = p_a[2] + (dat2 * off2_0);
      p_a[3] = p_a[3] + (dat3 * off3_0);
    }

    // shift pointers to data y direction
    p_a[0] = p_a[0] + (dat0 * off0_1);
    p_a[1] = p_a[1] + (dat1 * off1_1);
    p_a[2] = p_a[2] + (dat2 * off2_1);
    p_a[3] = p_a[3] + (dat3 * off3_1);
  }
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[19].time += t2 - t1;
  }
  ops_set_dirtybit_host(args, 7);
  ops_set_halo_dirtybit3(&args[0], range);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c1, &t1);
    OPS_kernels[19].mpi_time += t1 - t2;
    OPS_kernels[19].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[19].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[19].transfer += ops_compute_transfer(dim, start, end, &arg2);
    OPS_kernels[19].transfer += ops_compute_transfer(dim, start, end, &arg3);
  }
}
Ejemplo n.º 14
0
// host stub function
void ops_par_loop_gridgen_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, 0))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(0, "gridgen_kernel");
    OPS_kernels[0].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, "gridgen_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;

  p_a[1] = (char *)arg_idx;

  // 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[0].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++) {
      gridgen_kernel((double *)p_a[0] + i * 1 * 1, (int *)p_a[1]);

      arg_idx[0]++;
    }

    // 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
    gridgen_kernel((double *)p_a[0], (int *)p_a[1]);

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

  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[0].time += t2 - t1;
  }
  ops_set_dirtybit_host(args, 2);
  ops_set_halo_dirtybit3(&args[0], range);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c1, &t1);
    OPS_kernels[0].mpi_time += t1 - t2;
    OPS_kernels[0].transfer += ops_compute_transfer(dim, start, end, &arg0);
  }
}
// host stub function
void ops_par_loop_calc_dt_kernel_min(char const *name, ops_block block, int dim,
                                     int *range, ops_arg arg0, ops_arg arg1) {

  // Timing
  double t1, t2, c1, c2;
  ops_arg args[2] = {arg0, arg1};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 2, range, 52))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(52, "calc_dt_kernel_min");
    OPS_kernels[52].count++;
    ops_timers_core(&c1, &t1);
  }

  // compute localy 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 // OPS_MPI

  int arg_idx[2];
  int arg_idx_base[2];
#ifdef OPS_MPI
  if (compute_ranges(args, 2, 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
  for (int n = 0; n < 2; n++) {
    arg_idx_base[n] = arg_idx[n];
  }

  int dat0 = args[0].dat->elem_size;

#ifdef OPS_MPI
  double *arg1h =
      (double *)(((ops_reduction)args[1].data)->data +
                 ((ops_reduction)args[1].data)->size * block->index);
#else
  double *arg1h = (double *)(((ops_reduction)args[1].data)->data);
#endif

  // set up initial pointers
  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];
#ifdef OPS_GPU
  double *p_a0 = (double *)((char *)args[0].data_d + base0);
#else
  double *p_a0 = (double *)((char *)args[0].data + base0);
#endif

  double *p_a1 = arg1h;

  int x_size = MAX(0, end[0] - start[0]);
  int y_size = MAX(0, end[1] - start[1]);

  // initialize global variable with the dimension of dats
  xdim0 = args[0].dat->size[0];
  if (xdim0 != xdim0_calc_dt_kernel_min_h) {
    xdim0_calc_dt_kernel_min = xdim0;
    xdim0_calc_dt_kernel_min_h = xdim0;
  }

// Halo Exchanges

#ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 2);
#else
  ops_H_D_exchanges_host(args, 2);
#endif
  ops_halo_exchanges(args, 2, range);

#ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 2);
#else
  ops_H_D_exchanges_host(args, 2);
#endif
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[52].mpi_time += t2 - t1;
  }

  calc_dt_kernel_min_c_wrapper(p_a0, p_a1, x_size, y_size);

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[52].time += t1 - t2;
  }
#ifdef OPS_GPU
  ops_set_dirtybit_device(args, 2);
#else
  ops_set_dirtybit_host(args, 2);
#endif

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[52].mpi_time += t2 - t1;
    OPS_kernels[52].transfer += ops_compute_transfer(dim, start, end, &arg0);
  }
}
// host stub function
void ops_par_loop_poisson_kernel_populate(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) {

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


  int  offs[6][2];
  ops_arg args[6] = { arg0, arg1, arg2, arg3, arg4, arg5};



  ops_timing_realloc(0,"poisson_kernel_populate");
  OPS_kernels[0].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, "poisson_kernel_populate");
  #endif

  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];



  int off3_0 = offs[3][0];
  int off3_1 = offs[3][1];
  int dat3 = args[3].dat->elem_size;
  int off4_0 = offs[4][0];
  int off4_1 = offs[4][1];
  int dat4 = args[4].dat->elem_size;
  int off5_0 = offs[5][0];
  int off5_1 = offs[5][1];
  int dat5 = args[5].dat->elem_size;


  #ifdef _OPENMP
  int nthreads = omp_get_max_threads( );
  #else
  int nthreads = 1;
  #endif
  xdim3 = args[3].dat->size[0]*args[3].dat->dim;
  xdim4 = args[4].dat->size[0]*args[4].dat->dim;
  xdim5 = args[5].dat->size[0]*args[5].dat->dim;

  ops_H_D_exchanges_host(args, 6);

  //Halo Exchanges
  ops_halo_exchanges(args,6,range);


  ops_timers_core(&c2,&t2);
  OPS_kernels[0].mpi_time += t2-t1;


  #pragma omp parallel for
  for ( int thr=0; thr<nthreads; thr++ ){

    int y_size = end[1]-start[1];
    char *p_a[6];

    int start_i = start[1] + ((y_size-1)/nthreads+1)*thr;
    int finish_i = start[1] + MIN(((y_size-1)/nthreads+1)*(thr+1),y_size);

    //get address per thread
    int start0 = start[0];
    int start1 = start_i;

    int arg_idx[2];
    #ifdef OPS_MPI
    arg_idx[0] = sb->decomp_disp[0]+start0;
    arg_idx[1] = sb->decomp_disp[1]+start1;
    #else //OPS_MPI
    arg_idx[0] = start0;
    arg_idx[1] = start1;
    #endif //OPS_MPI
    //set up initial pointers 
    int d_m[OPS_MAX_DIM];
    p_a[0] = (char *)args[0].data;

    p_a[1] = (char *)args[1].data;

    p_a[2] = (char *)arg_idx;

    #ifdef OPS_MPI
    for (int d = 0; d < dim; d++) d_m[d] = args[3].dat->d_m[d] + OPS_sub_dat_list[args[3].dat->index]->d_im[d];
    #else //OPS_MPI
    for (int d = 0; d < dim; d++) d_m[d] = args[3].dat->d_m[d];
    #endif //OPS_MPI
    int base3 = dat3 * 1 * 
    (start0 * args[3].stencil->stride[0] - args[3].dat->base[0] - d_m[0]);
    base3 = base3+ dat3 *
      args[3].dat->size[0] *
      (start1 * args[3].stencil->stride[1] - args[3].dat->base[1] - d_m[1]);
    p_a[3] = (char *)args[3].data + base3;

    #ifdef OPS_MPI
    for (int d = 0; d < dim; d++) d_m[d] = args[4].dat->d_m[d] + OPS_sub_dat_list[args[4].dat->index]->d_im[d];
    #else //OPS_MPI
    for (int d = 0; d < dim; d++) d_m[d] = args[4].dat->d_m[d];
    #endif //OPS_MPI
    int base4 = dat4 * 1 * 
    (start0 * args[4].stencil->stride[0] - args[4].dat->base[0] - d_m[0]);
    base4 = base4+ dat4 *
      args[4].dat->size[0] *
      (start1 * args[4].stencil->stride[1] - args[4].dat->base[1] - d_m[1]);
    p_a[4] = (char *)args[4].data + base4;

    #ifdef OPS_MPI
    for (int d = 0; d < dim; d++) d_m[d] = args[5].dat->d_m[d] + OPS_sub_dat_list[args[5].dat->index]->d_im[d];
    #else //OPS_MPI
    for (int d = 0; d < dim; d++) d_m[d] = args[5].dat->d_m[d];
    #endif //OPS_MPI
    int base5 = dat5 * 1 * 
    (start0 * args[5].stencil->stride[0] - args[5].dat->base[0] - d_m[0]);
    base5 = base5+ dat5 *
      args[5].dat->size[0] *
      (start1 * args[5].stencil->stride[1] - args[5].dat->base[1] - d_m[1]);
    p_a[5] = (char *)args[5].data + base5;


    for ( int n_y=start_i; n_y<finish_i; n_y++ ){
      for ( int n_x=start[0]; n_x<start[0]+(end[0]-start[0])/SIMD_VEC; n_x++ ){
        //call kernel function, passing in pointers to data -vectorised
        for ( int i=0; i<SIMD_VEC; i++ ){
          poisson_kernel_populate(  (int * )p_a[0], (int * )p_a[1], arg_idx,
           (double * )p_a[3]+ i*1, (double * )p_a[4]+ i*1, (double * )p_a[5]+ i*1 );

          arg_idx[0]++;
        }

        //shift pointers to data x direction
        p_a[3]= p_a[3] + (dat3 * off3_0)*SIMD_VEC;
        p_a[4]= p_a[4] + (dat4 * off4_0)*SIMD_VEC;
        p_a[5]= p_a[5] + (dat5 * off5_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
        poisson_kernel_populate(  (int * )p_a[0], (int * )p_a[1], arg_idx,
           (double * )p_a[3], (double * )p_a[4], (double * )p_a[5] );


        //shift pointers to data x direction
        p_a[3]= p_a[3] + (dat3 * off3_0);
        p_a[4]= p_a[4] + (dat4 * off4_0);
        p_a[5]= p_a[5] + (dat5 * off5_0);
        arg_idx[0]++;
      }

      //shift pointers to data y direction
      p_a[3]= p_a[3] + (dat3 * off3_1);
      p_a[4]= p_a[4] + (dat4 * off4_1);
      p_a[5]= p_a[5] + (dat5 * off5_1);
      #ifdef OPS_MPI
      arg_idx[0] = sb->decomp_disp[0]+start0;
      #else //OPS_MPI
      arg_idx[0] = start0;
      #endif //OPS_MPI
      arg_idx[1]++;
    }
  }

  ops_timers_core(&c1,&t1);
  OPS_kernels[0].time += t1-t2;

  ops_set_dirtybit_host(args, 6);

  ops_set_halo_dirtybit3(&args[3],range);
  ops_set_halo_dirtybit3(&args[4],range);
  ops_set_halo_dirtybit3(&args[5],range);

  //Update kernel record
  ops_timers_core(&c2,&t2);
  OPS_kernels[0].mpi_time += t2-t1;
  OPS_kernels[0].transfer += ops_compute_transfer(dim, range, &arg3);
  OPS_kernels[0].transfer += ops_compute_transfer(dim, range, &arg4);
  OPS_kernels[0].transfer += ops_compute_transfer(dim, range, &arg5);
}
// host stub function
void ops_par_loop_advec_cell_kernel4_xdir(
    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;

  char *p_a[11];
  int offs[11][3];
  ops_arg args[11] = {arg0, arg1, arg2, arg3, arg4, arg5,
                      arg6, arg7, arg8, arg9, arg10};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 11, range, 111))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(111, "advec_cell_kernel4_xdir");
    OPS_kernels[111].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, "advec_cell_kernel4_xdir");
#endif

  int arg_idx[3];
  int arg_idx_base[3];
#ifdef OPS_MPI
  if (compute_ranges(args, 11, 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];

  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], &end[0], args[8].dat->size, args[8].stencil->stride) -
      offs[8][0];
  offs[8][2] =
      off3D(2, &start[0], &end[0], args[8].dat->size, args[8].stencil->stride) -
      offs[8][1] - offs[8][0];

  offs[9][0] = args[9].stencil->stride[0] * 1; // unit step in x dimension
  offs[9][1] =
      off3D(1, &start[0], &end[0], args[9].dat->size, args[9].stencil->stride) -
      offs[9][0];
  offs[9][2] =
      off3D(2, &start[0], &end[0], args[9].dat->size, args[9].stencil->stride) -
      offs[9][1] - offs[9][0];

  offs[10][0] = args[10].stencil->stride[0] * 1; // unit step in x dimension
  offs[10][1] = off3D(1, &start[0], &end[0], args[10].dat->size,
                      args[10].stencil->stride) -
                offs[10][0];
  offs[10][2] = off3D(2, &start[0], &end[0], args[10].dat->size,
                      args[10].stencil->stride) -
                offs[10][1] - offs[10][0];

  int off0_0 = offs[0][0];
  int off0_1 = offs[0][1];
  int off0_2 = offs[0][2];
  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 off1_2 = offs[1][2];
  int dat1 = (OPS_soa ? args[1].dat->type_size : args[1].dat->elem_size);
  int off2_0 = offs[2][0];
  int off2_1 = offs[2][1];
  int off2_2 = offs[2][2];
  int dat2 = (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size);
  int off3_0 = offs[3][0];
  int off3_1 = offs[3][1];
  int off3_2 = offs[3][2];
  int dat3 = (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size);
  int off4_0 = offs[4][0];
  int off4_1 = offs[4][1];
  int off4_2 = offs[4][2];
  int dat4 = (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size);
  int off5_0 = offs[5][0];
  int off5_1 = offs[5][1];
  int off5_2 = offs[5][2];
  int dat5 = (OPS_soa ? args[5].dat->type_size : args[5].dat->elem_size);
  int off6_0 = offs[6][0];
  int off6_1 = offs[6][1];
  int off6_2 = offs[6][2];
  int dat6 = (OPS_soa ? args[6].dat->type_size : args[6].dat->elem_size);
  int off7_0 = offs[7][0];
  int off7_1 = offs[7][1];
  int off7_2 = offs[7][2];
  int dat7 = (OPS_soa ? args[7].dat->type_size : args[7].dat->elem_size);
  int off8_0 = offs[8][0];
  int off8_1 = offs[8][1];
  int off8_2 = offs[8][2];
  int dat8 = (OPS_soa ? args[8].dat->type_size : args[8].dat->elem_size);
  int off9_0 = offs[9][0];
  int off9_1 = offs[9][1];
  int off9_2 = offs[9][2];
  int dat9 = (OPS_soa ? args[9].dat->type_size : args[9].dat->elem_size);
  int off10_0 = offs[10][0];
  int off10_1 = offs[10][1];
  int off10_2 = offs[10][2];
  int dat10 = (OPS_soa ? args[10].dat->type_size : args[10].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];
  base0 = base0 +
          (OPS_soa ? args[0].dat->type_size : args[0].dat->elem_size) *
              args[0].dat->size[0] * args[0].dat->size[1] * start[2] *
              args[0].stencil->stride[2];
  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];
  base1 = base1 +
          (OPS_soa ? args[1].dat->type_size : args[1].dat->elem_size) *
              args[1].dat->size[0] * args[1].dat->size[1] * start[2] *
              args[1].stencil->stride[2];
  p_a[1] = (char *)args[1].data + base1;

  int base2 = args[2].dat->base_offset +
              (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
                  start[0] * args[2].stencil->stride[0];
  base2 = base2 +
          (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
              args[2].dat->size[0] * start[1] * args[2].stencil->stride[1];
  base2 = base2 +
          (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
              args[2].dat->size[0] * args[2].dat->size[1] * start[2] *
              args[2].stencil->stride[2];
  p_a[2] = (char *)args[2].data + base2;

  int base3 = args[3].dat->base_offset +
              (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size) *
                  start[0] * args[3].stencil->stride[0];
  base3 = base3 +
          (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size) *
              args[3].dat->size[0] * start[1] * args[3].stencil->stride[1];
  base3 = base3 +
          (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size) *
              args[3].dat->size[0] * args[3].dat->size[1] * start[2] *
              args[3].stencil->stride[2];
  p_a[3] = (char *)args[3].data + base3;

  int base4 = args[4].dat->base_offset +
              (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size) *
                  start[0] * args[4].stencil->stride[0];
  base4 = base4 +
          (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size) *
              args[4].dat->size[0] * start[1] * args[4].stencil->stride[1];
  base4 = base4 +
          (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size) *
              args[4].dat->size[0] * args[4].dat->size[1] * start[2] *
              args[4].stencil->stride[2];
  p_a[4] = (char *)args[4].data + base4;

  int base5 = args[5].dat->base_offset +
              (OPS_soa ? args[5].dat->type_size : args[5].dat->elem_size) *
                  start[0] * args[5].stencil->stride[0];
  base5 = base5 +
          (OPS_soa ? args[5].dat->type_size : args[5].dat->elem_size) *
              args[5].dat->size[0] * start[1] * args[5].stencil->stride[1];
  base5 = base5 +
          (OPS_soa ? args[5].dat->type_size : args[5].dat->elem_size) *
              args[5].dat->size[0] * args[5].dat->size[1] * start[2] *
              args[5].stencil->stride[2];
  p_a[5] = (char *)args[5].data + base5;

  int base6 = args[6].dat->base_offset +
              (OPS_soa ? args[6].dat->type_size : args[6].dat->elem_size) *
                  start[0] * args[6].stencil->stride[0];
  base6 = base6 +
          (OPS_soa ? args[6].dat->type_size : args[6].dat->elem_size) *
              args[6].dat->size[0] * start[1] * args[6].stencil->stride[1];
  base6 = base6 +
          (OPS_soa ? args[6].dat->type_size : args[6].dat->elem_size) *
              args[6].dat->size[0] * args[6].dat->size[1] * start[2] *
              args[6].stencil->stride[2];
  p_a[6] = (char *)args[6].data + base6;

  int base7 = args[7].dat->base_offset +
              (OPS_soa ? args[7].dat->type_size : args[7].dat->elem_size) *
                  start[0] * args[7].stencil->stride[0];
  base7 = base7 +
          (OPS_soa ? args[7].dat->type_size : args[7].dat->elem_size) *
              args[7].dat->size[0] * start[1] * args[7].stencil->stride[1];
  base7 = base7 +
          (OPS_soa ? args[7].dat->type_size : args[7].dat->elem_size) *
              args[7].dat->size[0] * args[7].dat->size[1] * start[2] *
              args[7].stencil->stride[2];
  p_a[7] = (char *)args[7].data + base7;

  int base8 = args[8].dat->base_offset +
              (OPS_soa ? args[8].dat->type_size : args[8].dat->elem_size) *
                  start[0] * args[8].stencil->stride[0];
  base8 = base8 +
          (OPS_soa ? args[8].dat->type_size : args[8].dat->elem_size) *
              args[8].dat->size[0] * start[1] * args[8].stencil->stride[1];
  base8 = base8 +
          (OPS_soa ? args[8].dat->type_size : args[8].dat->elem_size) *
              args[8].dat->size[0] * args[8].dat->size[1] * start[2] *
              args[8].stencil->stride[2];
  p_a[8] = (char *)args[8].data + base8;

  int base9 = args[9].dat->base_offset +
              (OPS_soa ? args[9].dat->type_size : args[9].dat->elem_size) *
                  start[0] * args[9].stencil->stride[0];
  base9 = base9 +
          (OPS_soa ? args[9].dat->type_size : args[9].dat->elem_size) *
              args[9].dat->size[0] * start[1] * args[9].stencil->stride[1];
  base9 = base9 +
          (OPS_soa ? args[9].dat->type_size : args[9].dat->elem_size) *
              args[9].dat->size[0] * args[9].dat->size[1] * start[2] *
              args[9].stencil->stride[2];
  p_a[9] = (char *)args[9].data + base9;

  int base10 = args[10].dat->base_offset +
               (OPS_soa ? args[10].dat->type_size : args[10].dat->elem_size) *
                   start[0] * args[10].stencil->stride[0];
  base10 = base10 +
           (OPS_soa ? args[10].dat->type_size : args[10].dat->elem_size) *
               args[10].dat->size[0] * start[1] * args[10].stencil->stride[1];
  base10 = base10 +
           (OPS_soa ? args[10].dat->type_size : args[10].dat->elem_size) *
               args[10].dat->size[0] * args[10].dat->size[1] * start[2] *
               args[10].stencil->stride[2];
  p_a[10] = (char *)args[10].data + base10;

  // initialize global variable with the dimension of dats
  xdim0 = args[0].dat->size[0];
  ydim0 = args[0].dat->size[1];
  xdim1 = args[1].dat->size[0];
  ydim1 = args[1].dat->size[1];
  xdim2 = args[2].dat->size[0];
  ydim2 = args[2].dat->size[1];
  xdim3 = args[3].dat->size[0];
  ydim3 = args[3].dat->size[1];
  xdim4 = args[4].dat->size[0];
  ydim4 = args[4].dat->size[1];
  xdim5 = args[5].dat->size[0];
  ydim5 = args[5].dat->size[1];
  xdim6 = args[6].dat->size[0];
  ydim6 = args[6].dat->size[1];
  xdim7 = args[7].dat->size[0];
  ydim7 = args[7].dat->size[1];
  xdim8 = args[8].dat->size[0];
  ydim8 = args[8].dat->size[1];
  xdim9 = args[9].dat->size[0];
  ydim9 = args[9].dat->size[1];
  xdim10 = args[10].dat->size[0];
  ydim10 = args[10].dat->size[1];

  // Halo Exchanges
  ops_H_D_exchanges_host(args, 11);
  ops_halo_exchanges(args, 11, range);
  ops_H_D_exchanges_host(args, 11);

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[111].mpi_time += t1 - t2;
  }

  int n_x;
  for (int n_z = start[2]; n_z < end[2]; n_z++) {
    for (int n_y = start[1]; n_y < end[1]; n_y++) {
#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
#pragma simd
        for (int i = 0; i < SIMD_VEC; i++) {
          advec_cell_kernel4_xdir(
              (double *)p_a[0] + i * 1 * 1, (double *)p_a[1] + i * 1 * 1,
              (double *)p_a[2] + i * 1 * 1, (double *)p_a[3] + i * 1 * 1,
              (double *)p_a[4] + i * 1 * 1, (double *)p_a[5] + i * 1 * 1,
              (double *)p_a[6] + i * 1 * 1, (double *)p_a[7] + i * 1 * 1,
              (double *)p_a[8] + i * 1 * 1, (double *)p_a[9] + i * 1 * 1,
              (double *)p_a[10] + i * 1 * 1);
        }

        // shift pointers to data x direction
        p_a[0] = p_a[0] + (dat0 * off0_0) * SIMD_VEC;
        p_a[1] = p_a[1] + (dat1 * off1_0) * SIMD_VEC;
        p_a[2] = p_a[2] + (dat2 * off2_0) * SIMD_VEC;
        p_a[3] = p_a[3] + (dat3 * off3_0) * SIMD_VEC;
        p_a[4] = p_a[4] + (dat4 * off4_0) * SIMD_VEC;
        p_a[5] = p_a[5] + (dat5 * off5_0) * SIMD_VEC;
        p_a[6] = p_a[6] + (dat6 * off6_0) * SIMD_VEC;
        p_a[7] = p_a[7] + (dat7 * off7_0) * SIMD_VEC;
        p_a[8] = p_a[8] + (dat8 * off8_0) * SIMD_VEC;
        p_a[9] = p_a[9] + (dat9 * off9_0) * SIMD_VEC;
        p_a[10] = p_a[10] + (dat10 * off10_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
        advec_cell_kernel4_xdir(
            (double *)p_a[0], (double *)p_a[1], (double *)p_a[2],
            (double *)p_a[3], (double *)p_a[4], (double *)p_a[5],
            (double *)p_a[6], (double *)p_a[7], (double *)p_a[8],
            (double *)p_a[9], (double *)p_a[10]);

        // shift pointers to data x direction
        p_a[0] = p_a[0] + (dat0 * off0_0);
        p_a[1] = p_a[1] + (dat1 * off1_0);
        p_a[2] = p_a[2] + (dat2 * off2_0);
        p_a[3] = p_a[3] + (dat3 * off3_0);
        p_a[4] = p_a[4] + (dat4 * off4_0);
        p_a[5] = p_a[5] + (dat5 * off5_0);
        p_a[6] = p_a[6] + (dat6 * off6_0);
        p_a[7] = p_a[7] + (dat7 * off7_0);
        p_a[8] = p_a[8] + (dat8 * off8_0);
        p_a[9] = p_a[9] + (dat9 * off9_0);
        p_a[10] = p_a[10] + (dat10 * off10_0);
      }

      // shift pointers to data y direction
      p_a[0] = p_a[0] + (dat0 * off0_1);
      p_a[1] = p_a[1] + (dat1 * off1_1);
      p_a[2] = p_a[2] + (dat2 * off2_1);
      p_a[3] = p_a[3] + (dat3 * off3_1);
      p_a[4] = p_a[4] + (dat4 * off4_1);
      p_a[5] = p_a[5] + (dat5 * off5_1);
      p_a[6] = p_a[6] + (dat6 * off6_1);
      p_a[7] = p_a[7] + (dat7 * off7_1);
      p_a[8] = p_a[8] + (dat8 * off8_1);
      p_a[9] = p_a[9] + (dat9 * off9_1);
      p_a[10] = p_a[10] + (dat10 * off10_1);
    }
    // shift pointers to data z direction
    p_a[0] = p_a[0] + (dat0 * off0_2);
    p_a[1] = p_a[1] + (dat1 * off1_2);
    p_a[2] = p_a[2] + (dat2 * off2_2);
    p_a[3] = p_a[3] + (dat3 * off3_2);
    p_a[4] = p_a[4] + (dat4 * off4_2);
    p_a[5] = p_a[5] + (dat5 * off5_2);
    p_a[6] = p_a[6] + (dat6 * off6_2);
    p_a[7] = p_a[7] + (dat7 * off7_2);
    p_a[8] = p_a[8] + (dat8 * off8_2);
    p_a[9] = p_a[9] + (dat9 * off9_2);
    p_a[10] = p_a[10] + (dat10 * off10_2);
  }
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[111].time += t2 - t1;
  }
  ops_set_dirtybit_host(args, 11);
  ops_set_halo_dirtybit3(&args[0], range);
  ops_set_halo_dirtybit3(&args[1], range);
  ops_set_halo_dirtybit3(&args[6], range);
  ops_set_halo_dirtybit3(&args[7], range);
  ops_set_halo_dirtybit3(&args[8], range);
  ops_set_halo_dirtybit3(&args[9], range);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c1, &t1);
    OPS_kernels[111].mpi_time += t1 - t2;
    OPS_kernels[111].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[111].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[111].transfer += ops_compute_transfer(dim, start, end, &arg2);
    OPS_kernels[111].transfer += ops_compute_transfer(dim, start, end, &arg3);
    OPS_kernels[111].transfer += ops_compute_transfer(dim, start, end, &arg4);
    OPS_kernels[111].transfer += ops_compute_transfer(dim, start, end, &arg5);
    OPS_kernels[111].transfer += ops_compute_transfer(dim, start, end, &arg6);
    OPS_kernels[111].transfer += ops_compute_transfer(dim, start, end, &arg7);
    OPS_kernels[111].transfer += ops_compute_transfer(dim, start, end, &arg8);
    OPS_kernels[111].transfer += ops_compute_transfer(dim, start, end, &arg9);
    OPS_kernels[111].transfer += ops_compute_transfer(dim, start, end, &arg10);
  }
}
// host stub function
void ops_par_loop_tea_leaf_norm2_kernel(char const *name, ops_block block,
                                        int dim, int *range, ops_arg arg0,
                                        ops_arg arg1) {

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

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

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 2, range, 39))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(39, "tea_leaf_norm2_kernel");
    OPS_kernels[39].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, "tea_leaf_norm2_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];

  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);

#ifdef OPS_MPI
  double *arg1h =
      (double *)(((ops_reduction)args[1].data)->data +
                 ((ops_reduction)args[1].data)->size * block->index);
#else
  double *arg1h = (double *)(((ops_reduction)args[1].data)->data);
#endif
  // Halo Exchanges
  ops_H_D_exchanges_host(args, 2);
  ops_halo_exchanges(args, 2, range);
  ops_H_D_exchanges_host(args, 2);

#ifdef _OPENMP
  int nthreads = omp_get_max_threads();
#else
  int nthreads = 1;
#endif
  // allocate and initialise arrays for global reduction
  // assumes a max of MAX_REDUCT_THREADS threads with a cacche line size of 64
  // bytes
  double arg_gbl1[MAX(1, 64) * MAX_REDUCT_THREADS];
  for (int thr = 0; thr < nthreads; thr++) {
    for (int d = 0; d < 1; d++) {
      arg_gbl1[d + 64 * thr] = ZERO_double;
    }
  }
  xdim0 = args[0].dat->size[0];

  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[39].mpi_time += t2 - t1;
  }

#pragma omp parallel for
  for (int thr = 0; thr < nthreads; thr++) {

    int y_size = end[1] - start[1];
    char *p_a[2];

    int start_i = start[1] + ((y_size - 1) / nthreads + 1) * thr;
    int finish_i =
        start[1] + MIN(((y_size - 1) / nthreads + 1) * (thr + 1), y_size);

    // get address per thread
    int start0 = start[0];
    int start1 = start_i;

    // set up initial pointers
    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
    for (int d = 0; d < dim; d++)
      d_m[d] = args[0].dat->d_m[d];
#endif
    int base0 = dat0 * 1 * (start0 * args[0].stencil->stride[0] -
                            args[0].dat->base[0] - d_m[0]);
    base0 = base0 +
            dat0 * args[0].dat->size[0] * (start1 * args[0].stencil->stride[1] -
                                           args[0].dat->base[1] - d_m[1]);
    p_a[0] = (char *)args[0].data + base0;

    p_a[1] = (char *)arg1h;

    for (int n_y = start_i; n_y < finish_i; n_y++) {
      for (int n_x = start[0]; n_x < start[0] + (end[0] - start[0]) / SIMD_VEC;
           n_x++) {
        // call kernel function, passing in pointers to data -vectorised
        for (int i = 0; i < SIMD_VEC; i++) {
          tea_leaf_norm2_kernel((const double *)p_a[0] + i * 1 * 1,
                                &arg_gbl1[64 * thr]);
        }

        // 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
        tea_leaf_norm2_kernel((const double *)p_a[0], &arg_gbl1[64 * thr]);

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

      // shift pointers to data y direction
      p_a[0] = p_a[0] + (dat0 * off0_1);
    }
  }

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[39].time += t1 - t2;
  }

  // combine reduction data
  for (int thr = 0; thr < nthreads; thr++) {
    for (int d = 0; d < 1; d++) {
      arg1h[d] += arg_gbl1[64 * thr + d];
    }
  }
  ops_set_dirtybit_host(args, 2);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[39].mpi_time += t2 - t1;
    OPS_kernels[39].transfer += ops_compute_transfer(dim, start, end, &arg0);
  }
}
Ejemplo n.º 19
0
// host stub function
void ops_par_loop_copy(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][2];
  ops_arg args[2] = { arg0, arg1};



  #ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args,2,range,5)) return;
  #endif

  if (OPS_diags > 1) {
    ops_timing_realloc(5,"copy");
    OPS_kernels[5].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];
  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
  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, "copy");
  #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];



  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;


  //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, 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[5].mpi_time += t1-t2;
  }

  int n_x;
  for ( int n_y=start[1]; n_y<end[1]; n_y++ ){
    #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
      #pragma simd
      for ( int i=0; i<SIMD_VEC; i++ ){
        copy(  (double *)p_a[0]+ i*1*1, (double *)p_a[1]+ i*1*1 );

      }

      //shift pointers to data x direction
      p_a[0]= p_a[0] + (dat0 * off0_0)*SIMD_VEC;
      p_a[1]= p_a[1] + (dat1 * off1_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
      copy(  (double *)p_a[0], (double *)p_a[1] );


      //shift pointers to data x direction
      p_a[0]= p_a[0] + (dat0 * off0_0);
      p_a[1]= p_a[1] + (dat1 * off1_0);
    }

    //shift pointers to data y direction
    p_a[0]= p_a[0] + (dat0 * off0_1);
    p_a[1]= p_a[1] + (dat1 * off1_1);
  }
  if (OPS_diags > 1) {
    ops_timers_core(&c2,&t2);
    OPS_kernels[5].time += t2-t1;
  }
  ops_set_dirtybit_host(args, 2);
  ops_set_halo_dirtybit3(&args[0],range);

  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);
  }
}
// host stub function
void ops_par_loop_update_halo_kernel3_plus_2_b(char const *name, ops_block Block, int dim, int* range,
 ops_arg arg0, ops_arg arg1, ops_arg arg2) {

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


  ops_timing_realloc(92,"update_halo_kernel3_plus_2_b");
  OPS_kernels[92].count++;

  //compute localy 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

  int x_size = MAX(0,end[0]-start[0]);
  int y_size = MAX(0,end[1]-start[1]);
  int z_size = MAX(0,end[2]-start[2]);


  xdim0 = args[0].dat->size[0]*args[0].dat->dim;
  ydim0 = args[0].dat->size[1];
  xdim1 = args[1].dat->size[0]*args[1].dat->dim;
  ydim1 = args[1].dat->size[1];

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

  if (xdim0 != xdim0_update_halo_kernel3_plus_2_b_h || ydim0 != ydim0_update_halo_kernel3_plus_2_b_h || xdim1 != xdim1_update_halo_kernel3_plus_2_b_h || ydim1 != ydim1_update_halo_kernel3_plus_2_b_h) {
    xdim0_update_halo_kernel3_plus_2_b = xdim0;
    xdim0_update_halo_kernel3_plus_2_b_h = xdim0;
    ydim0_update_halo_kernel3_plus_2_b = ydim0;
    ydim0_update_halo_kernel3_plus_2_b_h = ydim0;
    xdim1_update_halo_kernel3_plus_2_b = xdim1;
    xdim1_update_halo_kernel3_plus_2_b_h = xdim1;
    ydim1_update_halo_kernel3_plus_2_b = ydim1;
    ydim1_update_halo_kernel3_plus_2_b_h = ydim1;
  }

  int dat0 = args[0].dat->elem_size;
  int dat1 = args[1].dat->elem_size;

  int *arg2h = (int *)arg2.data;
  //Upload large globals
  int consts_bytes = 0;
  consts_bytes += ROUND_UP(NUM_FIELDS*sizeof(int));
  reallocConstArrays(consts_bytes);
  consts_bytes = 0;
  args[2].data = OPS_consts_h + consts_bytes;
  args[2].data_d = OPS_consts_d + consts_bytes;
  for (int d=0; d<NUM_FIELDS; d++) ((int *)args[2].data)[d] = arg2h[d];
  consts_bytes += ROUND_UP(NUM_FIELDS*sizeof(int));
  mvConstArraysToDevice(consts_bytes);

  //set up initial pointers
  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]);
  base0 = base0+ dat0 *
    args[0].dat->size[0] *
    args[0].dat->size[1] *
    (start[2] * args[0].stencil->stride[2] - args[0].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a0 = (double *)((char *)args[0].data_d + base0);
  #else
  double *p_a0 = (double *)((char *)args[0].data + base0);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d] + OPS_sub_dat_list[args[1].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d];
  #endif //OPS_MPI
  int base1 = dat1 * 1 * 
    (start[0] * args[1].stencil->stride[0] - args[1].dat->base[0] - d_m[0]);
  base1 = base1+ dat1 *
    args[1].dat->size[0] *
    (start[1] * args[1].stencil->stride[1] - args[1].dat->base[1] - d_m[1]);
  base1 = base1+ dat1 *
    args[1].dat->size[0] *
    args[1].dat->size[1] *
    (start[2] * args[1].stencil->stride[2] - args[1].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a1 = (double *)((char *)args[1].data_d + base1);
  #else
  double *p_a1 = (double *)((char *)args[1].data + base1);
  #endif

  #ifdef OPS_GPU
  int *p_a2 = (int *)args[2].data_d;
  #else
  int *p_a2 = arg2h;
  #endif

  #ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 3);
  #else
  ops_H_D_exchanges_host(args, 3);
  #endif
  ops_halo_exchanges(args,3,range);

  ops_timers_core(&c1,&t1);
  OPS_kernels[92].mpi_time += t1-t2;

  update_halo_kernel3_plus_2_b_c_wrapper(
    p_a0,
    p_a1,
    p_a2,
    x_size, y_size, z_size);

  ops_timers_core(&c2,&t2);
  OPS_kernels[92].time += t2-t1;
  #ifdef OPS_GPU
  ops_set_dirtybit_device(args, 3);
  #else
  ops_set_dirtybit_host(args, 3);
  #endif
  ops_set_halo_dirtybit3(&args[0],range);
  ops_set_halo_dirtybit3(&args[1],range);

  //Update kernel record
  OPS_kernels[92].transfer += ops_compute_transfer(dim, range, &arg0);
  OPS_kernels[92].transfer += ops_compute_transfer(dim, range, &arg1);
}
// host stub function
void ops_par_loop_tea_leaf_ppcg_reduce_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 args[4] = {arg0, arg1, arg2, arg3};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 4, range, 48))
    return;
#endif

  ops_timing_realloc(48, "tea_leaf_ppcg_reduce_kernel");
  OPS_kernels[48].count++;

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

  int x_size = MAX(0, end[0] - start[0]);
  int y_size = MAX(0, end[1] - start[1]);

  xdim0 = args[0].dat->size[0];
  xdim1 = args[1].dat->size[0];
  xdim2 = args[2].dat->size[0];

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

  if (xdim0 != xdim0_tea_leaf_ppcg_reduce_kernel_h ||
      xdim1 != xdim1_tea_leaf_ppcg_reduce_kernel_h ||
      xdim2 != xdim2_tea_leaf_ppcg_reduce_kernel_h) {
    xdim0_tea_leaf_ppcg_reduce_kernel = xdim0;
    xdim0_tea_leaf_ppcg_reduce_kernel_h = xdim0;
    xdim1_tea_leaf_ppcg_reduce_kernel = xdim1;
    xdim1_tea_leaf_ppcg_reduce_kernel_h = xdim1;
    xdim2_tea_leaf_ppcg_reduce_kernel = xdim2;
    xdim2_tea_leaf_ppcg_reduce_kernel_h = xdim2;
  }

  int dat0 = (OPS_soa ? args[0].dat->type_size : args[0].dat->elem_size);
  int dat1 = (OPS_soa ? args[1].dat->type_size : args[1].dat->elem_size);
  int dat2 = (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size);

#ifdef OPS_MPI
  double *arg3h =
      (double *)(((ops_reduction)args[3].data)->data +
                 ((ops_reduction)args[3].data)->size * block->index);
#else
  double *arg3h = (double *)(((ops_reduction)args[3].data)->data);
#endif

  // set up initial pointers
  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
  for (int d = 0; d < dim; d++)
    d_m[d] = args[0].dat->d_m[d];
#endif
  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]);
  double *p_a0 = (double *)((char *)args[0].data + base0);

#ifdef OPS_MPI
  for (int d = 0; d < dim; d++)
    d_m[d] =
        args[1].dat->d_m[d] + OPS_sub_dat_list[args[1].dat->index]->d_im[d];
#else
  for (int d = 0; d < dim; d++)
    d_m[d] = args[1].dat->d_m[d];
#endif
  int base1 = dat1 * 1 * (start[0] * args[1].stencil->stride[0] -
                          args[1].dat->base[0] - d_m[0]);
  base1 = base1 +
          dat1 * args[1].dat->size[0] * (start[1] * args[1].stencil->stride[1] -
                                         args[1].dat->base[1] - d_m[1]);
  double *p_a1 = (double *)((char *)args[1].data + base1);

#ifdef OPS_MPI
  for (int d = 0; d < dim; d++)
    d_m[d] =
        args[2].dat->d_m[d] + OPS_sub_dat_list[args[2].dat->index]->d_im[d];
#else
  for (int d = 0; d < dim; d++)
    d_m[d] = args[2].dat->d_m[d];
#endif
  int base2 = dat2 * 1 * (start[0] * args[2].stencil->stride[0] -
                          args[2].dat->base[0] - d_m[0]);
  base2 = base2 +
          dat2 * args[2].dat->size[0] * (start[1] * args[2].stencil->stride[1] -
                                         args[2].dat->base[1] - d_m[1]);
  double *p_a2 = (double *)((char *)args[2].data + base2);

  double *p_a3 = arg3h;

  ops_H_D_exchanges_host(args, 4);
  ops_halo_exchanges(args, 4, range);

  ops_timers_core(&c1, &t1);
  OPS_kernels[48].mpi_time += t1 - t2;

  tea_leaf_ppcg_reduce_kernel_c_wrapper(p_a0, p_a1, p_a2, p_a3, x_size, y_size);

  ops_timers_core(&c2, &t2);
  OPS_kernels[48].time += t2 - t1;
  ops_set_dirtybit_host(args, 4);

  // Update kernel record
  OPS_kernels[48].transfer += ops_compute_transfer(dim, start, end, &arg0);
  OPS_kernels[48].transfer += ops_compute_transfer(dim, start, end, &arg1);
  OPS_kernels[48].transfer += ops_compute_transfer(dim, start, end, &arg2);
}
// host stub function
void ops_par_loop_generate_chunk_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 args[7] = {arg0, arg1, arg2, arg3, arg4, arg5, arg6};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 7, range, 1))
    return;
#endif

  ops_timing_realloc(1, "generate_chunk_kernel");
  OPS_kernels[1].count++;

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

  int x_size = MAX(0, end[0] - start[0]);
  int y_size = MAX(0, end[1] - start[1]);

  xdim0 = args[0].dat->size[0];
  xdim1 = args[1].dat->size[0];
  xdim2 = args[2].dat->size[0];
  xdim3 = args[3].dat->size[0];
  xdim4 = args[4].dat->size[0];
  xdim5 = args[5].dat->size[0];
  xdim6 = args[6].dat->size[0];

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

  if (xdim0 != xdim0_generate_chunk_kernel_h ||
      xdim1 != xdim1_generate_chunk_kernel_h ||
      xdim2 != xdim2_generate_chunk_kernel_h ||
      xdim3 != xdim3_generate_chunk_kernel_h ||
      xdim4 != xdim4_generate_chunk_kernel_h ||
      xdim5 != xdim5_generate_chunk_kernel_h ||
      xdim6 != xdim6_generate_chunk_kernel_h) {
    xdim0_generate_chunk_kernel = xdim0;
    xdim0_generate_chunk_kernel_h = xdim0;
    xdim1_generate_chunk_kernel = xdim1;
    xdim1_generate_chunk_kernel_h = xdim1;
    xdim2_generate_chunk_kernel = xdim2;
    xdim2_generate_chunk_kernel_h = xdim2;
    xdim3_generate_chunk_kernel = xdim3;
    xdim3_generate_chunk_kernel_h = xdim3;
    xdim4_generate_chunk_kernel = xdim4;
    xdim4_generate_chunk_kernel_h = xdim4;
    xdim5_generate_chunk_kernel = xdim5;
    xdim5_generate_chunk_kernel_h = xdim5;
    xdim6_generate_chunk_kernel = xdim6;
    xdim6_generate_chunk_kernel_h = xdim6;
  }

  int dat0 = (OPS_soa ? args[0].dat->type_size : args[0].dat->elem_size);
  int dat1 = (OPS_soa ? args[1].dat->type_size : args[1].dat->elem_size);
  int dat2 = (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size);
  int dat3 = (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size);
  int dat4 = (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size);
  int dat5 = (OPS_soa ? args[5].dat->type_size : args[5].dat->elem_size);
  int dat6 = (OPS_soa ? args[6].dat->type_size : args[6].dat->elem_size);

  // set up initial pointers
  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
  for (int d = 0; d < dim; d++)
    d_m[d] = args[0].dat->d_m[d];
#endif
  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]);
  double *p_a0 = (double *)((char *)args[0].data + base0);

#ifdef OPS_MPI
  for (int d = 0; d < dim; d++)
    d_m[d] =
        args[1].dat->d_m[d] + OPS_sub_dat_list[args[1].dat->index]->d_im[d];
#else
  for (int d = 0; d < dim; d++)
    d_m[d] = args[1].dat->d_m[d];
#endif
  int base1 = dat1 * 1 * (start[0] * args[1].stencil->stride[0] -
                          args[1].dat->base[0] - d_m[0]);
  base1 = base1 +
          dat1 * args[1].dat->size[0] * (start[1] * args[1].stencil->stride[1] -
                                         args[1].dat->base[1] - d_m[1]);
  double *p_a1 = (double *)((char *)args[1].data + base1);

#ifdef OPS_MPI
  for (int d = 0; d < dim; d++)
    d_m[d] =
        args[2].dat->d_m[d] + OPS_sub_dat_list[args[2].dat->index]->d_im[d];
#else
  for (int d = 0; d < dim; d++)
    d_m[d] = args[2].dat->d_m[d];
#endif
  int base2 = dat2 * 1 * (start[0] * args[2].stencil->stride[0] -
                          args[2].dat->base[0] - d_m[0]);
  base2 = base2 +
          dat2 * args[2].dat->size[0] * (start[1] * args[2].stencil->stride[1] -
                                         args[2].dat->base[1] - d_m[1]);
  double *p_a2 = (double *)((char *)args[2].data + base2);

#ifdef OPS_MPI
  for (int d = 0; d < dim; d++)
    d_m[d] =
        args[3].dat->d_m[d] + OPS_sub_dat_list[args[3].dat->index]->d_im[d];
#else
  for (int d = 0; d < dim; d++)
    d_m[d] = args[3].dat->d_m[d];
#endif
  int base3 = dat3 * 1 * (start[0] * args[3].stencil->stride[0] -
                          args[3].dat->base[0] - d_m[0]);
  base3 = base3 +
          dat3 * args[3].dat->size[0] * (start[1] * args[3].stencil->stride[1] -
                                         args[3].dat->base[1] - d_m[1]);
  double *p_a3 = (double *)((char *)args[3].data + base3);

#ifdef OPS_MPI
  for (int d = 0; d < dim; d++)
    d_m[d] =
        args[4].dat->d_m[d] + OPS_sub_dat_list[args[4].dat->index]->d_im[d];
#else
  for (int d = 0; d < dim; d++)
    d_m[d] = args[4].dat->d_m[d];
#endif
  int base4 = dat4 * 1 * (start[0] * args[4].stencil->stride[0] -
                          args[4].dat->base[0] - d_m[0]);
  base4 = base4 +
          dat4 * args[4].dat->size[0] * (start[1] * args[4].stencil->stride[1] -
                                         args[4].dat->base[1] - d_m[1]);
  double *p_a4 = (double *)((char *)args[4].data + base4);

#ifdef OPS_MPI
  for (int d = 0; d < dim; d++)
    d_m[d] =
        args[5].dat->d_m[d] + OPS_sub_dat_list[args[5].dat->index]->d_im[d];
#else
  for (int d = 0; d < dim; d++)
    d_m[d] = args[5].dat->d_m[d];
#endif
  int base5 = dat5 * 1 * (start[0] * args[5].stencil->stride[0] -
                          args[5].dat->base[0] - d_m[0]);
  base5 = base5 +
          dat5 * args[5].dat->size[0] * (start[1] * args[5].stencil->stride[1] -
                                         args[5].dat->base[1] - d_m[1]);
  double *p_a5 = (double *)((char *)args[5].data + base5);

#ifdef OPS_MPI
  for (int d = 0; d < dim; d++)
    d_m[d] =
        args[6].dat->d_m[d] + OPS_sub_dat_list[args[6].dat->index]->d_im[d];
#else
  for (int d = 0; d < dim; d++)
    d_m[d] = args[6].dat->d_m[d];
#endif
  int base6 = dat6 * 1 * (start[0] * args[6].stencil->stride[0] -
                          args[6].dat->base[0] - d_m[0]);
  base6 = base6 +
          dat6 * args[6].dat->size[0] * (start[1] * args[6].stencil->stride[1] -
                                         args[6].dat->base[1] - d_m[1]);
  double *p_a6 = (double *)((char *)args[6].data + base6);

  ops_H_D_exchanges_host(args, 7);
  ops_halo_exchanges(args, 7, range);

  ops_timers_core(&c1, &t1);
  OPS_kernels[1].mpi_time += t1 - t2;

  generate_chunk_kernel_c_wrapper(p_a0, p_a1, p_a2, p_a3, p_a4, p_a5, p_a6,
                                  x_size, y_size);

  ops_timers_core(&c2, &t2);
  OPS_kernels[1].time += t2 - t1;
  ops_set_dirtybit_host(args, 7);
  ops_set_halo_dirtybit3(&args[2], range);
  ops_set_halo_dirtybit3(&args[3], range);
  ops_set_halo_dirtybit3(&args[4], range);

  // Update kernel record
  OPS_kernels[1].transfer += ops_compute_transfer(dim, start, end, &arg0);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, start, end, &arg1);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, start, end, &arg2);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, start, end, &arg3);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, start, end, &arg4);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, start, end, &arg5);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, start, end, &arg6);
}
// host stub function
void ops_par_loop_update_halo_kernel4_plus_2_b(char const *name,
                                               ops_block block, int dim,
                                               int *range, ops_arg arg0,
                                               ops_arg arg1, ops_arg arg2) {

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

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 3, range, 79))
    return;
#endif

  ops_timing_realloc(79, "update_halo_kernel4_plus_2_b");
  OPS_kernels[79].count++;

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

  int x_size = MAX(0, end[0] - start[0]);
  int y_size = MAX(0, end[1] - start[1]);
  int z_size = MAX(0, end[2] - start[2]);

  xdim0 = args[0].dat->size[0];
  ydim0 = args[0].dat->size[1];
  xdim1 = args[1].dat->size[0];
  ydim1 = args[1].dat->size[1];

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

  if (xdim0 != xdim0_update_halo_kernel4_plus_2_b_h ||
      ydim0 != ydim0_update_halo_kernel4_plus_2_b_h ||
      xdim1 != xdim1_update_halo_kernel4_plus_2_b_h ||
      ydim1 != ydim1_update_halo_kernel4_plus_2_b_h) {
    xdim0_update_halo_kernel4_plus_2_b = xdim0;
    xdim0_update_halo_kernel4_plus_2_b_h = xdim0;
    ydim0_update_halo_kernel4_plus_2_b = ydim0;
    ydim0_update_halo_kernel4_plus_2_b_h = ydim0;
    xdim1_update_halo_kernel4_plus_2_b = xdim1;
    xdim1_update_halo_kernel4_plus_2_b_h = xdim1;
    ydim1_update_halo_kernel4_plus_2_b = ydim1;
    ydim1_update_halo_kernel4_plus_2_b_h = ydim1;
  }

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

  int *arg2h = (int *)arg2.data;

  // set up initial pointers
  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
  for (int d = 0; d < dim; d++)
    d_m[d] = args[0].dat->d_m[d];
#endif
  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]);
  base0 = base0 +
          dat0 * args[0].dat->size[0] * args[0].dat->size[1] *
              (start[2] * args[0].stencil->stride[2] - args[0].dat->base[2] -
               d_m[2]);
  double *p_a0 = (double *)((char *)args[0].data + base0);

#ifdef OPS_MPI
  for (int d = 0; d < dim; d++)
    d_m[d] =
        args[1].dat->d_m[d] + OPS_sub_dat_list[args[1].dat->index]->d_im[d];
#else
  for (int d = 0; d < dim; d++)
    d_m[d] = args[1].dat->d_m[d];
#endif
  int base1 = dat1 * 1 * (start[0] * args[1].stencil->stride[0] -
                          args[1].dat->base[0] - d_m[0]);
  base1 = base1 +
          dat1 * args[1].dat->size[0] * (start[1] * args[1].stencil->stride[1] -
                                         args[1].dat->base[1] - d_m[1]);
  base1 = base1 +
          dat1 * args[1].dat->size[0] * args[1].dat->size[1] *
              (start[2] * args[1].stencil->stride[2] - args[1].dat->base[2] -
               d_m[2]);
  double *p_a1 = (double *)((char *)args[1].data + base1);

  int *p_a2 = arg2h;

  ops_H_D_exchanges_host(args, 3);
  ops_halo_exchanges(args, 3, range);

  ops_timers_core(&c1, &t1);
  OPS_kernels[79].mpi_time += t1 - t2;

  update_halo_kernel4_plus_2_b_c_wrapper(p_a0, p_a1, p_a2, x_size, y_size,
                                         z_size);

  ops_timers_core(&c2, &t2);
  OPS_kernels[79].time += t2 - t1;
  ops_set_dirtybit_host(args, 3);
  ops_set_halo_dirtybit3(&args[0], range);
  ops_set_halo_dirtybit3(&args[1], range);

  // Update kernel record
  OPS_kernels[79].transfer += ops_compute_transfer(dim, start, end, &arg0);
  OPS_kernels[79].transfer += ops_compute_transfer(dim, start, end, &arg1);
}
// host stub function
void ops_par_loop_update_halo_kernel4_minus_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][3];
  ops_arg args[3] = {arg0, arg1, arg2};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 3, range, 72))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(72, "update_halo_kernel4_minus_4_a");
    OPS_kernels[72].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_kernel4_minus_4_a");
#endif

  int arg_idx[3];
  int arg_idx_base[3];
#ifdef OPS_MPI
  if (compute_ranges(args, 3, 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];

  int off0_0 = offs[0][0];
  int off0_1 = offs[0][1];
  int off0_2 = offs[0][2];
  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 off1_2 = offs[1][2];
  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];
  base0 = base0 +
          (OPS_soa ? args[0].dat->type_size : args[0].dat->elem_size) *
              args[0].dat->size[0] * args[0].dat->size[1] * start[2] *
              args[0].stencil->stride[2];
  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];
  base1 = base1 +
          (OPS_soa ? args[1].dat->type_size : args[1].dat->elem_size) *
              args[1].dat->size[0] * args[1].dat->size[1] * start[2] *
              args[1].stencil->stride[2];
  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];
  ydim0 = args[0].dat->size[1];
  xdim1 = args[1].dat->size[0];
  ydim1 = args[1].dat->size[1];

  // 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[72].mpi_time += t1 - t2;
  }

  int n_x;
  for (int n_z = start[2]; n_z < end[2]; n_z++) {
    for (int n_y = start[1]; n_y < end[1]; n_y++) {
#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
#pragma simd
        for (int i = 0; i < SIMD_VEC; i++) {
          update_halo_kernel4_minus_4_a((double *)p_a[0] + i * 1 * 1,
                                        (double *)p_a[1] + i * 1 * 1,
                                        (int *)p_a[2]);
        }

        // shift pointers to data x direction
        p_a[0] = p_a[0] + (dat0 * off0_0) * SIMD_VEC;
        p_a[1] = p_a[1] + (dat1 * off1_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
        update_halo_kernel4_minus_4_a((double *)p_a[0], (double *)p_a[1],
                                      (int *)p_a[2]);

        // shift pointers to data x direction
        p_a[0] = p_a[0] + (dat0 * off0_0);
        p_a[1] = p_a[1] + (dat1 * off1_0);
      }

      // shift pointers to data y direction
      p_a[0] = p_a[0] + (dat0 * off0_1);
      p_a[1] = p_a[1] + (dat1 * off1_1);
    }
    // shift pointers to data z direction
    p_a[0] = p_a[0] + (dat0 * off0_2);
    p_a[1] = p_a[1] + (dat1 * off1_2);
  }
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[72].time += t2 - t1;
  }
  ops_set_dirtybit_host(args, 3);
  ops_set_halo_dirtybit3(&args[0], range);
  ops_set_halo_dirtybit3(&args[1], range);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c1, &t1);
    OPS_kernels[72].mpi_time += t1 - t2;
    OPS_kernels[72].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[72].transfer += ops_compute_transfer(dim, start, end, &arg1);
  }
}
// host stub function
void ops_par_loop_accelerate_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, ops_arg arg11, ops_arg arg12, ops_arg arg13) {

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


  ops_timing_realloc(1,"accelerate_kernel");
  OPS_kernels[1].count++;

  //compute localy 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

  int x_size = MAX(0,end[0]-start[0]);
  int y_size = MAX(0,end[1]-start[1]);
  int z_size = MAX(0,end[2]-start[2]);


  xdim0 = args[0].dat->size[0]*args[0].dat->dim;
  ydim0 = args[0].dat->size[1];
  xdim1 = args[1].dat->size[0]*args[1].dat->dim;
  ydim1 = args[1].dat->size[1];
  xdim2 = args[2].dat->size[0]*args[2].dat->dim;
  ydim2 = args[2].dat->size[1];
  xdim3 = args[3].dat->size[0]*args[3].dat->dim;
  ydim3 = args[3].dat->size[1];
  xdim4 = args[4].dat->size[0]*args[4].dat->dim;
  ydim4 = args[4].dat->size[1];
  xdim5 = args[5].dat->size[0]*args[5].dat->dim;
  ydim5 = args[5].dat->size[1];
  xdim6 = args[6].dat->size[0]*args[6].dat->dim;
  ydim6 = args[6].dat->size[1];
  xdim7 = args[7].dat->size[0]*args[7].dat->dim;
  ydim7 = args[7].dat->size[1];
  xdim8 = args[8].dat->size[0]*args[8].dat->dim;
  ydim8 = args[8].dat->size[1];
  xdim9 = args[9].dat->size[0]*args[9].dat->dim;
  ydim9 = args[9].dat->size[1];
  xdim10 = args[10].dat->size[0]*args[10].dat->dim;
  ydim10 = args[10].dat->size[1];
  xdim11 = args[11].dat->size[0]*args[11].dat->dim;
  ydim11 = args[11].dat->size[1];
  xdim12 = args[12].dat->size[0]*args[12].dat->dim;
  ydim12 = args[12].dat->size[1];
  xdim13 = args[13].dat->size[0]*args[13].dat->dim;
  ydim13 = args[13].dat->size[1];

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

  if (xdim0 != xdim0_accelerate_kernel_h || ydim0 != ydim0_accelerate_kernel_h || xdim1 != xdim1_accelerate_kernel_h || ydim1 != ydim1_accelerate_kernel_h || xdim2 != xdim2_accelerate_kernel_h || ydim2 != ydim2_accelerate_kernel_h || xdim3 != xdim3_accelerate_kernel_h || ydim3 != ydim3_accelerate_kernel_h || xdim4 != xdim4_accelerate_kernel_h || ydim4 != ydim4_accelerate_kernel_h || xdim5 != xdim5_accelerate_kernel_h || ydim5 != ydim5_accelerate_kernel_h || xdim6 != xdim6_accelerate_kernel_h || ydim6 != ydim6_accelerate_kernel_h || xdim7 != xdim7_accelerate_kernel_h || ydim7 != ydim7_accelerate_kernel_h || xdim8 != xdim8_accelerate_kernel_h || ydim8 != ydim8_accelerate_kernel_h || xdim9 != xdim9_accelerate_kernel_h || ydim9 != ydim9_accelerate_kernel_h || xdim10 != xdim10_accelerate_kernel_h || ydim10 != ydim10_accelerate_kernel_h || xdim11 != xdim11_accelerate_kernel_h || ydim11 != ydim11_accelerate_kernel_h || xdim12 != xdim12_accelerate_kernel_h || ydim12 != ydim12_accelerate_kernel_h || xdim13 != xdim13_accelerate_kernel_h || ydim13 != ydim13_accelerate_kernel_h) {
    xdim0_accelerate_kernel = xdim0;
    xdim0_accelerate_kernel_h = xdim0;
    ydim0_accelerate_kernel = ydim0;
    ydim0_accelerate_kernel_h = ydim0;
    xdim1_accelerate_kernel = xdim1;
    xdim1_accelerate_kernel_h = xdim1;
    ydim1_accelerate_kernel = ydim1;
    ydim1_accelerate_kernel_h = ydim1;
    xdim2_accelerate_kernel = xdim2;
    xdim2_accelerate_kernel_h = xdim2;
    ydim2_accelerate_kernel = ydim2;
    ydim2_accelerate_kernel_h = ydim2;
    xdim3_accelerate_kernel = xdim3;
    xdim3_accelerate_kernel_h = xdim3;
    ydim3_accelerate_kernel = ydim3;
    ydim3_accelerate_kernel_h = ydim3;
    xdim4_accelerate_kernel = xdim4;
    xdim4_accelerate_kernel_h = xdim4;
    ydim4_accelerate_kernel = ydim4;
    ydim4_accelerate_kernel_h = ydim4;
    xdim5_accelerate_kernel = xdim5;
    xdim5_accelerate_kernel_h = xdim5;
    ydim5_accelerate_kernel = ydim5;
    ydim5_accelerate_kernel_h = ydim5;
    xdim6_accelerate_kernel = xdim6;
    xdim6_accelerate_kernel_h = xdim6;
    ydim6_accelerate_kernel = ydim6;
    ydim6_accelerate_kernel_h = ydim6;
    xdim7_accelerate_kernel = xdim7;
    xdim7_accelerate_kernel_h = xdim7;
    ydim7_accelerate_kernel = ydim7;
    ydim7_accelerate_kernel_h = ydim7;
    xdim8_accelerate_kernel = xdim8;
    xdim8_accelerate_kernel_h = xdim8;
    ydim8_accelerate_kernel = ydim8;
    ydim8_accelerate_kernel_h = ydim8;
    xdim9_accelerate_kernel = xdim9;
    xdim9_accelerate_kernel_h = xdim9;
    ydim9_accelerate_kernel = ydim9;
    ydim9_accelerate_kernel_h = ydim9;
    xdim10_accelerate_kernel = xdim10;
    xdim10_accelerate_kernel_h = xdim10;
    ydim10_accelerate_kernel = ydim10;
    ydim10_accelerate_kernel_h = ydim10;
    xdim11_accelerate_kernel = xdim11;
    xdim11_accelerate_kernel_h = xdim11;
    ydim11_accelerate_kernel = ydim11;
    ydim11_accelerate_kernel_h = ydim11;
    xdim12_accelerate_kernel = xdim12;
    xdim12_accelerate_kernel_h = xdim12;
    ydim12_accelerate_kernel = ydim12;
    ydim12_accelerate_kernel_h = ydim12;
    xdim13_accelerate_kernel = xdim13;
    xdim13_accelerate_kernel_h = xdim13;
    ydim13_accelerate_kernel = ydim13;
    ydim13_accelerate_kernel_h = ydim13;
  }

  int dat0 = args[0].dat->elem_size;
  int dat1 = args[1].dat->elem_size;
  int dat2 = args[2].dat->elem_size;
  int dat3 = args[3].dat->elem_size;
  int dat4 = args[4].dat->elem_size;
  int dat5 = args[5].dat->elem_size;
  int dat6 = args[6].dat->elem_size;
  int dat7 = args[7].dat->elem_size;
  int dat8 = args[8].dat->elem_size;
  int dat9 = args[9].dat->elem_size;
  int dat10 = args[10].dat->elem_size;
  int dat11 = args[11].dat->elem_size;
  int dat12 = args[12].dat->elem_size;
  int dat13 = args[13].dat->elem_size;


  //set up initial pointers
  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]);
  base0 = base0+ dat0 *
    args[0].dat->size[0] *
    args[0].dat->size[1] *
    (start[2] * args[0].stencil->stride[2] - args[0].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a0 = (double *)((char *)args[0].data_d + base0);
  #else
  double *p_a0 = (double *)((char *)args[0].data + base0);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d] + OPS_sub_dat_list[args[1].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d];
  #endif //OPS_MPI
  int base1 = dat1 * 1 * 
    (start[0] * args[1].stencil->stride[0] - args[1].dat->base[0] - d_m[0]);
  base1 = base1+ dat1 *
    args[1].dat->size[0] *
    (start[1] * args[1].stencil->stride[1] - args[1].dat->base[1] - d_m[1]);
  base1 = base1+ dat1 *
    args[1].dat->size[0] *
    args[1].dat->size[1] *
    (start[2] * args[1].stencil->stride[2] - args[1].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a1 = (double *)((char *)args[1].data_d + base1);
  #else
  double *p_a1 = (double *)((char *)args[1].data + base1);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[2].dat->d_m[d] + OPS_sub_dat_list[args[2].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[2].dat->d_m[d];
  #endif //OPS_MPI
  int base2 = dat2 * 1 * 
    (start[0] * args[2].stencil->stride[0] - args[2].dat->base[0] - d_m[0]);
  base2 = base2+ dat2 *
    args[2].dat->size[0] *
    (start[1] * args[2].stencil->stride[1] - args[2].dat->base[1] - d_m[1]);
  base2 = base2+ dat2 *
    args[2].dat->size[0] *
    args[2].dat->size[1] *
    (start[2] * args[2].stencil->stride[2] - args[2].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a2 = (double *)((char *)args[2].data_d + base2);
  #else
  double *p_a2 = (double *)((char *)args[2].data + base2);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[3].dat->d_m[d] + OPS_sub_dat_list[args[3].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[3].dat->d_m[d];
  #endif //OPS_MPI
  int base3 = dat3 * 1 * 
    (start[0] * args[3].stencil->stride[0] - args[3].dat->base[0] - d_m[0]);
  base3 = base3+ dat3 *
    args[3].dat->size[0] *
    (start[1] * args[3].stencil->stride[1] - args[3].dat->base[1] - d_m[1]);
  base3 = base3+ dat3 *
    args[3].dat->size[0] *
    args[3].dat->size[1] *
    (start[2] * args[3].stencil->stride[2] - args[3].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a3 = (double *)((char *)args[3].data_d + base3);
  #else
  double *p_a3 = (double *)((char *)args[3].data + base3);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[4].dat->d_m[d] + OPS_sub_dat_list[args[4].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[4].dat->d_m[d];
  #endif //OPS_MPI
  int base4 = dat4 * 1 * 
    (start[0] * args[4].stencil->stride[0] - args[4].dat->base[0] - d_m[0]);
  base4 = base4+ dat4 *
    args[4].dat->size[0] *
    (start[1] * args[4].stencil->stride[1] - args[4].dat->base[1] - d_m[1]);
  base4 = base4+ dat4 *
    args[4].dat->size[0] *
    args[4].dat->size[1] *
    (start[2] * args[4].stencil->stride[2] - args[4].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a4 = (double *)((char *)args[4].data_d + base4);
  #else
  double *p_a4 = (double *)((char *)args[4].data + base4);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[5].dat->d_m[d] + OPS_sub_dat_list[args[5].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[5].dat->d_m[d];
  #endif //OPS_MPI
  int base5 = dat5 * 1 * 
    (start[0] * args[5].stencil->stride[0] - args[5].dat->base[0] - d_m[0]);
  base5 = base5+ dat5 *
    args[5].dat->size[0] *
    (start[1] * args[5].stencil->stride[1] - args[5].dat->base[1] - d_m[1]);
  base5 = base5+ dat5 *
    args[5].dat->size[0] *
    args[5].dat->size[1] *
    (start[2] * args[5].stencil->stride[2] - args[5].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a5 = (double *)((char *)args[5].data_d + base5);
  #else
  double *p_a5 = (double *)((char *)args[5].data + base5);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[6].dat->d_m[d] + OPS_sub_dat_list[args[6].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[6].dat->d_m[d];
  #endif //OPS_MPI
  int base6 = dat6 * 1 * 
    (start[0] * args[6].stencil->stride[0] - args[6].dat->base[0] - d_m[0]);
  base6 = base6+ dat6 *
    args[6].dat->size[0] *
    (start[1] * args[6].stencil->stride[1] - args[6].dat->base[1] - d_m[1]);
  base6 = base6+ dat6 *
    args[6].dat->size[0] *
    args[6].dat->size[1] *
    (start[2] * args[6].stencil->stride[2] - args[6].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a6 = (double *)((char *)args[6].data_d + base6);
  #else
  double *p_a6 = (double *)((char *)args[6].data + base6);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[7].dat->d_m[d] + OPS_sub_dat_list[args[7].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[7].dat->d_m[d];
  #endif //OPS_MPI
  int base7 = dat7 * 1 * 
    (start[0] * args[7].stencil->stride[0] - args[7].dat->base[0] - d_m[0]);
  base7 = base7+ dat7 *
    args[7].dat->size[0] *
    (start[1] * args[7].stencil->stride[1] - args[7].dat->base[1] - d_m[1]);
  base7 = base7+ dat7 *
    args[7].dat->size[0] *
    args[7].dat->size[1] *
    (start[2] * args[7].stencil->stride[2] - args[7].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a7 = (double *)((char *)args[7].data_d + base7);
  #else
  double *p_a7 = (double *)((char *)args[7].data + base7);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[8].dat->d_m[d] + OPS_sub_dat_list[args[8].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[8].dat->d_m[d];
  #endif //OPS_MPI
  int base8 = dat8 * 1 * 
    (start[0] * args[8].stencil->stride[0] - args[8].dat->base[0] - d_m[0]);
  base8 = base8+ dat8 *
    args[8].dat->size[0] *
    (start[1] * args[8].stencil->stride[1] - args[8].dat->base[1] - d_m[1]);
  base8 = base8+ dat8 *
    args[8].dat->size[0] *
    args[8].dat->size[1] *
    (start[2] * args[8].stencil->stride[2] - args[8].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a8 = (double *)((char *)args[8].data_d + base8);
  #else
  double *p_a8 = (double *)((char *)args[8].data + base8);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[9].dat->d_m[d] + OPS_sub_dat_list[args[9].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[9].dat->d_m[d];
  #endif //OPS_MPI
  int base9 = dat9 * 1 * 
    (start[0] * args[9].stencil->stride[0] - args[9].dat->base[0] - d_m[0]);
  base9 = base9+ dat9 *
    args[9].dat->size[0] *
    (start[1] * args[9].stencil->stride[1] - args[9].dat->base[1] - d_m[1]);
  base9 = base9+ dat9 *
    args[9].dat->size[0] *
    args[9].dat->size[1] *
    (start[2] * args[9].stencil->stride[2] - args[9].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a9 = (double *)((char *)args[9].data_d + base9);
  #else
  double *p_a9 = (double *)((char *)args[9].data + base9);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[10].dat->d_m[d] + OPS_sub_dat_list[args[10].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[10].dat->d_m[d];
  #endif //OPS_MPI
  int base10 = dat10 * 1 * 
    (start[0] * args[10].stencil->stride[0] - args[10].dat->base[0] - d_m[0]);
  base10 = base10+ dat10 *
    args[10].dat->size[0] *
    (start[1] * args[10].stencil->stride[1] - args[10].dat->base[1] - d_m[1]);
  base10 = base10+ dat10 *
    args[10].dat->size[0] *
    args[10].dat->size[1] *
    (start[2] * args[10].stencil->stride[2] - args[10].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a10 = (double *)((char *)args[10].data_d + base10);
  #else
  double *p_a10 = (double *)((char *)args[10].data + base10);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[11].dat->d_m[d] + OPS_sub_dat_list[args[11].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[11].dat->d_m[d];
  #endif //OPS_MPI
  int base11 = dat11 * 1 * 
    (start[0] * args[11].stencil->stride[0] - args[11].dat->base[0] - d_m[0]);
  base11 = base11+ dat11 *
    args[11].dat->size[0] *
    (start[1] * args[11].stencil->stride[1] - args[11].dat->base[1] - d_m[1]);
  base11 = base11+ dat11 *
    args[11].dat->size[0] *
    args[11].dat->size[1] *
    (start[2] * args[11].stencil->stride[2] - args[11].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a11 = (double *)((char *)args[11].data_d + base11);
  #else
  double *p_a11 = (double *)((char *)args[11].data + base11);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[12].dat->d_m[d] + OPS_sub_dat_list[args[12].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[12].dat->d_m[d];
  #endif //OPS_MPI
  int base12 = dat12 * 1 * 
    (start[0] * args[12].stencil->stride[0] - args[12].dat->base[0] - d_m[0]);
  base12 = base12+ dat12 *
    args[12].dat->size[0] *
    (start[1] * args[12].stencil->stride[1] - args[12].dat->base[1] - d_m[1]);
  base12 = base12+ dat12 *
    args[12].dat->size[0] *
    args[12].dat->size[1] *
    (start[2] * args[12].stencil->stride[2] - args[12].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a12 = (double *)((char *)args[12].data_d + base12);
  #else
  double *p_a12 = (double *)((char *)args[12].data + base12);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[13].dat->d_m[d] + OPS_sub_dat_list[args[13].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[13].dat->d_m[d];
  #endif //OPS_MPI
  int base13 = dat13 * 1 * 
    (start[0] * args[13].stencil->stride[0] - args[13].dat->base[0] - d_m[0]);
  base13 = base13+ dat13 *
    args[13].dat->size[0] *
    (start[1] * args[13].stencil->stride[1] - args[13].dat->base[1] - d_m[1]);
  base13 = base13+ dat13 *
    args[13].dat->size[0] *
    args[13].dat->size[1] *
    (start[2] * args[13].stencil->stride[2] - args[13].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a13 = (double *)((char *)args[13].data_d + base13);
  #else
  double *p_a13 = (double *)((char *)args[13].data + base13);
  #endif


  #ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 14);
  #else
  ops_H_D_exchanges_host(args, 14);
  #endif
  ops_halo_exchanges(args,14,range);

  ops_timers_core(&c1,&t1);
  OPS_kernels[1].mpi_time += t1-t2;

  accelerate_kernel_c_wrapper(
    p_a0,
    p_a1,
    p_a2,
    p_a3,
    p_a4,
    p_a5,
    p_a6,
    p_a7,
    p_a8,
    p_a9,
    p_a10,
    p_a11,
    p_a12,
    p_a13,
    x_size, y_size, z_size);

  ops_timers_core(&c2,&t2);
  OPS_kernels[1].time += t2-t1;
  #ifdef OPS_GPU
  ops_set_dirtybit_device(args, 14);
  #else
  ops_set_dirtybit_host(args, 14);
  #endif
  ops_set_halo_dirtybit3(&args[2],range);
  ops_set_halo_dirtybit3(&args[4],range);
  ops_set_halo_dirtybit3(&args[8],range);
  ops_set_halo_dirtybit3(&args[12],range);

  //Update kernel record
  OPS_kernels[1].transfer += ops_compute_transfer(dim, range, &arg0);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, range, &arg1);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, range, &arg2);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, range, &arg3);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, range, &arg4);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, range, &arg5);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, range, &arg6);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, range, &arg7);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, range, &arg8);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, range, &arg9);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, range, &arg10);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, range, &arg11);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, range, &arg12);
  OPS_kernels[1].transfer += ops_compute_transfer(dim, range, &arg13);
}
// host stub function
void ops_par_loop_advec_mom_kernel_post_pre_advec_x(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 args[5] = { arg0, arg1, arg2, arg3, arg4};


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

  //compute localy 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

  int x_size = MAX(0,end[0]-start[0]);
  int y_size = MAX(0,end[1]-start[1]);
  int z_size = MAX(0,end[2]-start[2]);


  xdim0 = args[0].dat->size[0]*args[0].dat->dim;
  ydim0 = args[0].dat->size[1];
  xdim1 = args[1].dat->size[0]*args[1].dat->dim;
  ydim1 = args[1].dat->size[1];
  xdim2 = args[2].dat->size[0]*args[2].dat->dim;
  ydim2 = args[2].dat->size[1];
  xdim3 = args[3].dat->size[0]*args[3].dat->dim;
  ydim3 = args[3].dat->size[1];
  xdim4 = args[4].dat->size[0]*args[4].dat->dim;
  ydim4 = args[4].dat->size[1];

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

  if (xdim0 != xdim0_advec_mom_kernel_post_pre_advec_x_h || ydim0 != ydim0_advec_mom_kernel_post_pre_advec_x_h || xdim1 != xdim1_advec_mom_kernel_post_pre_advec_x_h || ydim1 != ydim1_advec_mom_kernel_post_pre_advec_x_h || xdim2 != xdim2_advec_mom_kernel_post_pre_advec_x_h || ydim2 != ydim2_advec_mom_kernel_post_pre_advec_x_h || xdim3 != xdim3_advec_mom_kernel_post_pre_advec_x_h || ydim3 != ydim3_advec_mom_kernel_post_pre_advec_x_h || xdim4 != xdim4_advec_mom_kernel_post_pre_advec_x_h || ydim4 != ydim4_advec_mom_kernel_post_pre_advec_x_h) {
    xdim0_advec_mom_kernel_post_pre_advec_x = xdim0;
    xdim0_advec_mom_kernel_post_pre_advec_x_h = xdim0;
    ydim0_advec_mom_kernel_post_pre_advec_x = ydim0;
    ydim0_advec_mom_kernel_post_pre_advec_x_h = ydim0;
    xdim1_advec_mom_kernel_post_pre_advec_x = xdim1;
    xdim1_advec_mom_kernel_post_pre_advec_x_h = xdim1;
    ydim1_advec_mom_kernel_post_pre_advec_x = ydim1;
    ydim1_advec_mom_kernel_post_pre_advec_x_h = ydim1;
    xdim2_advec_mom_kernel_post_pre_advec_x = xdim2;
    xdim2_advec_mom_kernel_post_pre_advec_x_h = xdim2;
    ydim2_advec_mom_kernel_post_pre_advec_x = ydim2;
    ydim2_advec_mom_kernel_post_pre_advec_x_h = ydim2;
    xdim3_advec_mom_kernel_post_pre_advec_x = xdim3;
    xdim3_advec_mom_kernel_post_pre_advec_x_h = xdim3;
    ydim3_advec_mom_kernel_post_pre_advec_x = ydim3;
    ydim3_advec_mom_kernel_post_pre_advec_x_h = ydim3;
    xdim4_advec_mom_kernel_post_pre_advec_x = xdim4;
    xdim4_advec_mom_kernel_post_pre_advec_x_h = xdim4;
    ydim4_advec_mom_kernel_post_pre_advec_x = ydim4;
    ydim4_advec_mom_kernel_post_pre_advec_x_h = ydim4;
  }

  int dat0 = args[0].dat->elem_size;
  int dat1 = args[1].dat->elem_size;
  int dat2 = args[2].dat->elem_size;
  int dat3 = args[3].dat->elem_size;
  int dat4 = args[4].dat->elem_size;


  //set up initial pointers
  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]);
  base0 = base0+ dat0 *
    args[0].dat->size[0] *
    args[0].dat->size[1] *
    (start[2] * args[0].stencil->stride[2] - args[0].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a0 = (double *)((char *)args[0].data_d + base0);
  #else
  double *p_a0 = (double *)((char *)args[0].data + base0);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d] + OPS_sub_dat_list[args[1].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d];
  #endif //OPS_MPI
  int base1 = dat1 * 1 * 
    (start[0] * args[1].stencil->stride[0] - args[1].dat->base[0] - d_m[0]);
  base1 = base1+ dat1 *
    args[1].dat->size[0] *
    (start[1] * args[1].stencil->stride[1] - args[1].dat->base[1] - d_m[1]);
  base1 = base1+ dat1 *
    args[1].dat->size[0] *
    args[1].dat->size[1] *
    (start[2] * args[1].stencil->stride[2] - args[1].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a1 = (double *)((char *)args[1].data_d + base1);
  #else
  double *p_a1 = (double *)((char *)args[1].data + base1);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[2].dat->d_m[d] + OPS_sub_dat_list[args[2].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[2].dat->d_m[d];
  #endif //OPS_MPI
  int base2 = dat2 * 1 * 
    (start[0] * args[2].stencil->stride[0] - args[2].dat->base[0] - d_m[0]);
  base2 = base2+ dat2 *
    args[2].dat->size[0] *
    (start[1] * args[2].stencil->stride[1] - args[2].dat->base[1] - d_m[1]);
  base2 = base2+ dat2 *
    args[2].dat->size[0] *
    args[2].dat->size[1] *
    (start[2] * args[2].stencil->stride[2] - args[2].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a2 = (double *)((char *)args[2].data_d + base2);
  #else
  double *p_a2 = (double *)((char *)args[2].data + base2);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[3].dat->d_m[d] + OPS_sub_dat_list[args[3].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[3].dat->d_m[d];
  #endif //OPS_MPI
  int base3 = dat3 * 1 * 
    (start[0] * args[3].stencil->stride[0] - args[3].dat->base[0] - d_m[0]);
  base3 = base3+ dat3 *
    args[3].dat->size[0] *
    (start[1] * args[3].stencil->stride[1] - args[3].dat->base[1] - d_m[1]);
  base3 = base3+ dat3 *
    args[3].dat->size[0] *
    args[3].dat->size[1] *
    (start[2] * args[3].stencil->stride[2] - args[3].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a3 = (double *)((char *)args[3].data_d + base3);
  #else
  double *p_a3 = (double *)((char *)args[3].data + base3);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[4].dat->d_m[d] + OPS_sub_dat_list[args[4].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[4].dat->d_m[d];
  #endif //OPS_MPI
  int base4 = dat4 * 1 * 
    (start[0] * args[4].stencil->stride[0] - args[4].dat->base[0] - d_m[0]);
  base4 = base4+ dat4 *
    args[4].dat->size[0] *
    (start[1] * args[4].stencil->stride[1] - args[4].dat->base[1] - d_m[1]);
  base4 = base4+ dat4 *
    args[4].dat->size[0] *
    args[4].dat->size[1] *
    (start[2] * args[4].stencil->stride[2] - args[4].dat->base[2] - d_m[2]);
  #ifdef OPS_GPU
  double *p_a4 = (double *)((char *)args[4].data_d + base4);
  #else
  double *p_a4 = (double *)((char *)args[4].data + base4);
  #endif


  #ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 5);
  #else
  ops_H_D_exchanges_host(args, 5);
  #endif
  ops_halo_exchanges(args,5,range);

  ops_timers_core(&c1,&t1);
  OPS_kernels[18].mpi_time += t1-t2;

  advec_mom_kernel_post_pre_advec_x_c_wrapper(
    p_a0,
    p_a1,
    p_a2,
    p_a3,
    p_a4,
    x_size, y_size, z_size);

  ops_timers_core(&c2,&t2);
  OPS_kernels[18].time += t2-t1;
  #ifdef OPS_GPU
  ops_set_dirtybit_device(args, 5);
  #else
  ops_set_dirtybit_host(args, 5);
  #endif
  ops_set_halo_dirtybit3(&args[0],range);
  ops_set_halo_dirtybit3(&args[3],range);

  //Update kernel record
  OPS_kernels[18].transfer += ops_compute_transfer(dim, range, &arg0);
  OPS_kernels[18].transfer += ops_compute_transfer(dim, range, &arg1);
  OPS_kernels[18].transfer += ops_compute_transfer(dim, range, &arg2);
  OPS_kernels[18].transfer += ops_compute_transfer(dim, range, &arg3);
  OPS_kernels[18].transfer += ops_compute_transfer(dim, range, &arg4);
}
// 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) {

  // Timing
  double t1, t2, c1, c2;
  ops_arg args[14] = {arg0, arg1, arg2, arg3,  arg4,  arg5,  arg6,
                      arg7, arg8, arg9, arg10, arg11, arg12, arg13};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 14, range, 101))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(101, "PdV_kernel_predict");
    OPS_kernels[101].count++;
    ops_timers_core(&c1, &t1);
  }

  // compute localy 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 // OPS_MPI

  int arg_idx[3];
  int arg_idx_base[3];
#ifdef OPS_MPI
  if (compute_ranges(args, 14, 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
  for (int n = 0; n < 3; n++) {
    arg_idx_base[n] = arg_idx[n];
  }

  int dat0 = args[0].dat->elem_size;
  int dat1 = args[1].dat->elem_size;
  int dat2 = args[2].dat->elem_size;
  int dat3 = args[3].dat->elem_size;
  int dat4 = args[4].dat->elem_size;
  int dat5 = args[5].dat->elem_size;
  int dat6 = args[6].dat->elem_size;
  int dat7 = args[7].dat->elem_size;
  int dat8 = args[8].dat->elem_size;
  int dat9 = args[9].dat->elem_size;
  int dat10 = args[10].dat->elem_size;
  int dat11 = args[11].dat->elem_size;
  int dat12 = args[12].dat->elem_size;
  int dat13 = args[13].dat->elem_size;

  // set up initial pointers
  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];
  base0 = base0 +
          (OPS_soa ? args[0].dat->type_size : args[0].dat->elem_size) *
              args[0].dat->size[0] * args[0].dat->size[1] * start[2] *
              args[0].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a0 = (double *)((char *)args[0].data_d + base0);
#else
  double *p_a0 = (double *)((char *)args[0].data + base0);
#endif

  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];
  base1 = base1 +
          (OPS_soa ? args[1].dat->type_size : args[1].dat->elem_size) *
              args[1].dat->size[0] * args[1].dat->size[1] * start[2] *
              args[1].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a1 = (double *)((char *)args[1].data_d + base1);
#else
  double *p_a1 = (double *)((char *)args[1].data + base1);
#endif

  int base2 = args[2].dat->base_offset +
              (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
                  start[0] * args[2].stencil->stride[0];
  base2 = base2 +
          (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
              args[2].dat->size[0] * start[1] * args[2].stencil->stride[1];
  base2 = base2 +
          (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
              args[2].dat->size[0] * args[2].dat->size[1] * start[2] *
              args[2].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a2 = (double *)((char *)args[2].data_d + base2);
#else
  double *p_a2 = (double *)((char *)args[2].data + base2);
#endif

  int base3 = args[3].dat->base_offset +
              (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size) *
                  start[0] * args[3].stencil->stride[0];
  base3 = base3 +
          (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size) *
              args[3].dat->size[0] * start[1] * args[3].stencil->stride[1];
  base3 = base3 +
          (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size) *
              args[3].dat->size[0] * args[3].dat->size[1] * start[2] *
              args[3].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a3 = (double *)((char *)args[3].data_d + base3);
#else
  double *p_a3 = (double *)((char *)args[3].data + base3);
#endif

  int base4 = args[4].dat->base_offset +
              (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size) *
                  start[0] * args[4].stencil->stride[0];
  base4 = base4 +
          (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size) *
              args[4].dat->size[0] * start[1] * args[4].stencil->stride[1];
  base4 = base4 +
          (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size) *
              args[4].dat->size[0] * args[4].dat->size[1] * start[2] *
              args[4].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a4 = (double *)((char *)args[4].data_d + base4);
#else
  double *p_a4 = (double *)((char *)args[4].data + base4);
#endif

  int base5 = args[5].dat->base_offset +
              (OPS_soa ? args[5].dat->type_size : args[5].dat->elem_size) *
                  start[0] * args[5].stencil->stride[0];
  base5 = base5 +
          (OPS_soa ? args[5].dat->type_size : args[5].dat->elem_size) *
              args[5].dat->size[0] * start[1] * args[5].stencil->stride[1];
  base5 = base5 +
          (OPS_soa ? args[5].dat->type_size : args[5].dat->elem_size) *
              args[5].dat->size[0] * args[5].dat->size[1] * start[2] *
              args[5].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a5 = (double *)((char *)args[5].data_d + base5);
#else
  double *p_a5 = (double *)((char *)args[5].data + base5);
#endif

  int base6 = args[6].dat->base_offset +
              (OPS_soa ? args[6].dat->type_size : args[6].dat->elem_size) *
                  start[0] * args[6].stencil->stride[0];
  base6 = base6 +
          (OPS_soa ? args[6].dat->type_size : args[6].dat->elem_size) *
              args[6].dat->size[0] * start[1] * args[6].stencil->stride[1];
  base6 = base6 +
          (OPS_soa ? args[6].dat->type_size : args[6].dat->elem_size) *
              args[6].dat->size[0] * args[6].dat->size[1] * start[2] *
              args[6].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a6 = (double *)((char *)args[6].data_d + base6);
#else
  double *p_a6 = (double *)((char *)args[6].data + base6);
#endif

  int base7 = args[7].dat->base_offset +
              (OPS_soa ? args[7].dat->type_size : args[7].dat->elem_size) *
                  start[0] * args[7].stencil->stride[0];
  base7 = base7 +
          (OPS_soa ? args[7].dat->type_size : args[7].dat->elem_size) *
              args[7].dat->size[0] * start[1] * args[7].stencil->stride[1];
  base7 = base7 +
          (OPS_soa ? args[7].dat->type_size : args[7].dat->elem_size) *
              args[7].dat->size[0] * args[7].dat->size[1] * start[2] *
              args[7].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a7 = (double *)((char *)args[7].data_d + base7);
#else
  double *p_a7 = (double *)((char *)args[7].data + base7);
#endif

  int base8 = args[8].dat->base_offset +
              (OPS_soa ? args[8].dat->type_size : args[8].dat->elem_size) *
                  start[0] * args[8].stencil->stride[0];
  base8 = base8 +
          (OPS_soa ? args[8].dat->type_size : args[8].dat->elem_size) *
              args[8].dat->size[0] * start[1] * args[8].stencil->stride[1];
  base8 = base8 +
          (OPS_soa ? args[8].dat->type_size : args[8].dat->elem_size) *
              args[8].dat->size[0] * args[8].dat->size[1] * start[2] *
              args[8].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a8 = (double *)((char *)args[8].data_d + base8);
#else
  double *p_a8 = (double *)((char *)args[8].data + base8);
#endif

  int base9 = args[9].dat->base_offset +
              (OPS_soa ? args[9].dat->type_size : args[9].dat->elem_size) *
                  start[0] * args[9].stencil->stride[0];
  base9 = base9 +
          (OPS_soa ? args[9].dat->type_size : args[9].dat->elem_size) *
              args[9].dat->size[0] * start[1] * args[9].stencil->stride[1];
  base9 = base9 +
          (OPS_soa ? args[9].dat->type_size : args[9].dat->elem_size) *
              args[9].dat->size[0] * args[9].dat->size[1] * start[2] *
              args[9].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a9 = (double *)((char *)args[9].data_d + base9);
#else
  double *p_a9 = (double *)((char *)args[9].data + base9);
#endif

  int base10 = args[10].dat->base_offset +
               (OPS_soa ? args[10].dat->type_size : args[10].dat->elem_size) *
                   start[0] * args[10].stencil->stride[0];
  base10 = base10 +
           (OPS_soa ? args[10].dat->type_size : args[10].dat->elem_size) *
               args[10].dat->size[0] * start[1] * args[10].stencil->stride[1];
  base10 = base10 +
           (OPS_soa ? args[10].dat->type_size : args[10].dat->elem_size) *
               args[10].dat->size[0] * args[10].dat->size[1] * start[2] *
               args[10].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a10 = (double *)((char *)args[10].data_d + base10);
#else
  double *p_a10 = (double *)((char *)args[10].data + base10);
#endif

  int base11 = args[11].dat->base_offset +
               (OPS_soa ? args[11].dat->type_size : args[11].dat->elem_size) *
                   start[0] * args[11].stencil->stride[0];
  base11 = base11 +
           (OPS_soa ? args[11].dat->type_size : args[11].dat->elem_size) *
               args[11].dat->size[0] * start[1] * args[11].stencil->stride[1];
  base11 = base11 +
           (OPS_soa ? args[11].dat->type_size : args[11].dat->elem_size) *
               args[11].dat->size[0] * args[11].dat->size[1] * start[2] *
               args[11].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a11 = (double *)((char *)args[11].data_d + base11);
#else
  double *p_a11 = (double *)((char *)args[11].data + base11);
#endif

  int base12 = args[12].dat->base_offset +
               (OPS_soa ? args[12].dat->type_size : args[12].dat->elem_size) *
                   start[0] * args[12].stencil->stride[0];
  base12 = base12 +
           (OPS_soa ? args[12].dat->type_size : args[12].dat->elem_size) *
               args[12].dat->size[0] * start[1] * args[12].stencil->stride[1];
  base12 = base12 +
           (OPS_soa ? args[12].dat->type_size : args[12].dat->elem_size) *
               args[12].dat->size[0] * args[12].dat->size[1] * start[2] *
               args[12].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a12 = (double *)((char *)args[12].data_d + base12);
#else
  double *p_a12 = (double *)((char *)args[12].data + base12);
#endif

  int base13 = args[13].dat->base_offset +
               (OPS_soa ? args[13].dat->type_size : args[13].dat->elem_size) *
                   start[0] * args[13].stencil->stride[0];
  base13 = base13 +
           (OPS_soa ? args[13].dat->type_size : args[13].dat->elem_size) *
               args[13].dat->size[0] * start[1] * args[13].stencil->stride[1];
  base13 = base13 +
           (OPS_soa ? args[13].dat->type_size : args[13].dat->elem_size) *
               args[13].dat->size[0] * args[13].dat->size[1] * start[2] *
               args[13].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a13 = (double *)((char *)args[13].data_d + base13);
#else
  double *p_a13 = (double *)((char *)args[13].data + base13);
#endif

  int x_size = MAX(0, end[0] - start[0]);
  int y_size = MAX(0, end[1] - start[1]);
  int z_size = MAX(0, end[2] - start[2]);

  // initialize global variable with the dimension of dats
  xdim0 = args[0].dat->size[0];
  ydim0 = args[0].dat->size[1];
  xdim1 = args[1].dat->size[0];
  ydim1 = args[1].dat->size[1];
  xdim2 = args[2].dat->size[0];
  ydim2 = args[2].dat->size[1];
  xdim3 = args[3].dat->size[0];
  ydim3 = args[3].dat->size[1];
  xdim4 = args[4].dat->size[0];
  ydim4 = args[4].dat->size[1];
  xdim5 = args[5].dat->size[0];
  ydim5 = args[5].dat->size[1];
  xdim6 = args[6].dat->size[0];
  ydim6 = args[6].dat->size[1];
  xdim7 = args[7].dat->size[0];
  ydim7 = args[7].dat->size[1];
  xdim8 = args[8].dat->size[0];
  ydim8 = args[8].dat->size[1];
  xdim9 = args[9].dat->size[0];
  ydim9 = args[9].dat->size[1];
  xdim10 = args[10].dat->size[0];
  ydim10 = args[10].dat->size[1];
  xdim11 = args[11].dat->size[0];
  ydim11 = args[11].dat->size[1];
  xdim12 = args[12].dat->size[0];
  ydim12 = args[12].dat->size[1];
  xdim13 = args[13].dat->size[0];
  ydim13 = args[13].dat->size[1];
  if (xdim0 != xdim0_PdV_kernel_predict_h ||
      ydim0 != ydim0_PdV_kernel_predict_h ||
      xdim1 != xdim1_PdV_kernel_predict_h ||
      ydim1 != ydim1_PdV_kernel_predict_h ||
      xdim2 != xdim2_PdV_kernel_predict_h ||
      ydim2 != ydim2_PdV_kernel_predict_h ||
      xdim3 != xdim3_PdV_kernel_predict_h ||
      ydim3 != ydim3_PdV_kernel_predict_h ||
      xdim4 != xdim4_PdV_kernel_predict_h ||
      ydim4 != ydim4_PdV_kernel_predict_h ||
      xdim5 != xdim5_PdV_kernel_predict_h ||
      ydim5 != ydim5_PdV_kernel_predict_h ||
      xdim6 != xdim6_PdV_kernel_predict_h ||
      ydim6 != ydim6_PdV_kernel_predict_h ||
      xdim7 != xdim7_PdV_kernel_predict_h ||
      ydim7 != ydim7_PdV_kernel_predict_h ||
      xdim8 != xdim8_PdV_kernel_predict_h ||
      ydim8 != ydim8_PdV_kernel_predict_h ||
      xdim9 != xdim9_PdV_kernel_predict_h ||
      ydim9 != ydim9_PdV_kernel_predict_h ||
      xdim10 != xdim10_PdV_kernel_predict_h ||
      ydim10 != ydim10_PdV_kernel_predict_h ||
      xdim11 != xdim11_PdV_kernel_predict_h ||
      ydim11 != ydim11_PdV_kernel_predict_h ||
      xdim12 != xdim12_PdV_kernel_predict_h ||
      ydim12 != ydim12_PdV_kernel_predict_h ||
      xdim13 != xdim13_PdV_kernel_predict_h ||
      ydim13 != ydim13_PdV_kernel_predict_h) {
    xdim0_PdV_kernel_predict = xdim0;
    xdim0_PdV_kernel_predict_h = xdim0;
    ydim0_PdV_kernel_predict = ydim0;
    ydim0_PdV_kernel_predict_h = ydim0;
    xdim1_PdV_kernel_predict = xdim1;
    xdim1_PdV_kernel_predict_h = xdim1;
    ydim1_PdV_kernel_predict = ydim1;
    ydim1_PdV_kernel_predict_h = ydim1;
    xdim2_PdV_kernel_predict = xdim2;
    xdim2_PdV_kernel_predict_h = xdim2;
    ydim2_PdV_kernel_predict = ydim2;
    ydim2_PdV_kernel_predict_h = ydim2;
    xdim3_PdV_kernel_predict = xdim3;
    xdim3_PdV_kernel_predict_h = xdim3;
    ydim3_PdV_kernel_predict = ydim3;
    ydim3_PdV_kernel_predict_h = ydim3;
    xdim4_PdV_kernel_predict = xdim4;
    xdim4_PdV_kernel_predict_h = xdim4;
    ydim4_PdV_kernel_predict = ydim4;
    ydim4_PdV_kernel_predict_h = ydim4;
    xdim5_PdV_kernel_predict = xdim5;
    xdim5_PdV_kernel_predict_h = xdim5;
    ydim5_PdV_kernel_predict = ydim5;
    ydim5_PdV_kernel_predict_h = ydim5;
    xdim6_PdV_kernel_predict = xdim6;
    xdim6_PdV_kernel_predict_h = xdim6;
    ydim6_PdV_kernel_predict = ydim6;
    ydim6_PdV_kernel_predict_h = ydim6;
    xdim7_PdV_kernel_predict = xdim7;
    xdim7_PdV_kernel_predict_h = xdim7;
    ydim7_PdV_kernel_predict = ydim7;
    ydim7_PdV_kernel_predict_h = ydim7;
    xdim8_PdV_kernel_predict = xdim8;
    xdim8_PdV_kernel_predict_h = xdim8;
    ydim8_PdV_kernel_predict = ydim8;
    ydim8_PdV_kernel_predict_h = ydim8;
    xdim9_PdV_kernel_predict = xdim9;
    xdim9_PdV_kernel_predict_h = xdim9;
    ydim9_PdV_kernel_predict = ydim9;
    ydim9_PdV_kernel_predict_h = ydim9;
    xdim10_PdV_kernel_predict = xdim10;
    xdim10_PdV_kernel_predict_h = xdim10;
    ydim10_PdV_kernel_predict = ydim10;
    ydim10_PdV_kernel_predict_h = ydim10;
    xdim11_PdV_kernel_predict = xdim11;
    xdim11_PdV_kernel_predict_h = xdim11;
    ydim11_PdV_kernel_predict = ydim11;
    ydim11_PdV_kernel_predict_h = ydim11;
    xdim12_PdV_kernel_predict = xdim12;
    xdim12_PdV_kernel_predict_h = xdim12;
    ydim12_PdV_kernel_predict = ydim12;
    ydim12_PdV_kernel_predict_h = ydim12;
    xdim13_PdV_kernel_predict = xdim13;
    xdim13_PdV_kernel_predict_h = xdim13;
    ydim13_PdV_kernel_predict = ydim13;
    ydim13_PdV_kernel_predict_h = ydim13;
  }

// Halo Exchanges

#ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 14);
#else
  ops_H_D_exchanges_host(args, 14);
#endif
  ops_halo_exchanges(args, 14, range);

#ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 14);
#else
  ops_H_D_exchanges_host(args, 14);
#endif
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[101].mpi_time += t2 - t1;
  }

  PdV_kernel_predict_c_wrapper(p_a0, p_a1, p_a2, p_a3, p_a4, p_a5, p_a6, p_a7,
                               p_a8, p_a9, p_a10, p_a11, p_a12, p_a13, x_size,
                               y_size, z_size);

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[101].time += t1 - t2;
  }
#ifdef OPS_GPU
  ops_set_dirtybit_device(args, 14);
#else
  ops_set_dirtybit_host(args, 14);
#endif
  ops_set_halo_dirtybit3(&args[4], range);
  ops_set_halo_dirtybit3(&args[8], range);
  ops_set_halo_dirtybit3(&args[11], range);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[101].mpi_time += t2 - t1;
    OPS_kernels[101].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[101].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[101].transfer += ops_compute_transfer(dim, start, end, &arg2);
    OPS_kernels[101].transfer += ops_compute_transfer(dim, start, end, &arg3);
    OPS_kernels[101].transfer += ops_compute_transfer(dim, start, end, &arg4);
    OPS_kernels[101].transfer += ops_compute_transfer(dim, start, end, &arg5);
    OPS_kernels[101].transfer += ops_compute_transfer(dim, start, end, &arg6);
    OPS_kernels[101].transfer += ops_compute_transfer(dim, start, end, &arg7);
    OPS_kernels[101].transfer += ops_compute_transfer(dim, start, end, &arg8);
    OPS_kernels[101].transfer += ops_compute_transfer(dim, start, end, &arg9);
    OPS_kernels[101].transfer += ops_compute_transfer(dim, start, end, &arg10);
    OPS_kernels[101].transfer += ops_compute_transfer(dim, start, end, &arg11);
    OPS_kernels[101].transfer += ops_compute_transfer(dim, start, end, &arg12);
    OPS_kernels[101].transfer += ops_compute_transfer(dim, start, end, &arg13);
  }
}
// host stub function
void ops_par_loop_advec_mom_kernel2_y(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][2];
  ops_arg args[4] = {arg0, arg1, arg2, arg3};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 4, range, 80))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(80, "advec_mom_kernel2_y");
    OPS_kernels[80].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, "advec_mom_kernel2_y");
#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];

  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);
  int off2_0 = offs[2][0];
  int off2_1 = offs[2][1];
  int dat2 = (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size);
  int off3_0 = offs[3][0];
  int off3_1 = offs[3][1];
  int dat3 = (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size);

  // Halo Exchanges
  ops_H_D_exchanges_host(args, 4);
  ops_halo_exchanges(args, 4, range);
  ops_H_D_exchanges_host(args, 4);

#ifdef _OPENMP
  int nthreads = omp_get_max_threads();
#else
  int nthreads = 1;
#endif
  xdim0 = args[0].dat->size[0];
  xdim1 = args[1].dat->size[0];
  xdim2 = args[2].dat->size[0];
  xdim3 = args[3].dat->size[0];

  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[80].mpi_time += t2 - t1;
  }

#pragma omp parallel for
  for (int thr = 0; thr < nthreads; thr++) {

    int y_size = end[1] - start[1];
    char *p_a[4];

    int start_i = start[1] + ((y_size - 1) / nthreads + 1) * thr;
    int finish_i =
        start[1] + MIN(((y_size - 1) / nthreads + 1) * (thr + 1), y_size);

    // get address per thread
    int start0 = start[0];
    int start1 = start_i;

    // set up initial pointers
    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
    for (int d = 0; d < dim; d++)
      d_m[d] = args[0].dat->d_m[d];
#endif
    int base0 = dat0 * 1 * (start0 * args[0].stencil->stride[0] -
                            args[0].dat->base[0] - d_m[0]);
    base0 = base0 +
            dat0 * args[0].dat->size[0] * (start1 * args[0].stencil->stride[1] -
                                           args[0].dat->base[1] - d_m[1]);
    p_a[0] = (char *)args[0].data + base0;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[1].dat->d_m[d] + OPS_sub_dat_list[args[1].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[1].dat->d_m[d];
#endif
    int base1 = dat1 * 1 * (start0 * args[1].stencil->stride[0] -
                            args[1].dat->base[0] - d_m[0]);
    base1 = base1 +
            dat1 * args[1].dat->size[0] * (start1 * args[1].stencil->stride[1] -
                                           args[1].dat->base[1] - d_m[1]);
    p_a[1] = (char *)args[1].data + base1;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[2].dat->d_m[d] + OPS_sub_dat_list[args[2].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[2].dat->d_m[d];
#endif
    int base2 = dat2 * 1 * (start0 * args[2].stencil->stride[0] -
                            args[2].dat->base[0] - d_m[0]);
    base2 = base2 +
            dat2 * args[2].dat->size[0] * (start1 * args[2].stencil->stride[1] -
                                           args[2].dat->base[1] - d_m[1]);
    p_a[2] = (char *)args[2].data + base2;

#ifdef OPS_MPI
    for (int d = 0; d < dim; d++)
      d_m[d] =
          args[3].dat->d_m[d] + OPS_sub_dat_list[args[3].dat->index]->d_im[d];
#else
    for (int d = 0; d < dim; d++)
      d_m[d] = args[3].dat->d_m[d];
#endif
    int base3 = dat3 * 1 * (start0 * args[3].stencil->stride[0] -
                            args[3].dat->base[0] - d_m[0]);
    base3 = base3 +
            dat3 * args[3].dat->size[0] * (start1 * args[3].stencil->stride[1] -
                                           args[3].dat->base[1] - d_m[1]);
    p_a[3] = (char *)args[3].data + base3;

    for (int n_y = start_i; n_y < finish_i; n_y++) {
      for (int n_x = start[0]; n_x < start[0] + (end[0] - start[0]) / SIMD_VEC;
           n_x++) {
// call kernel function, passing in pointers to data -vectorised
#pragma simd
        for (int i = 0; i < SIMD_VEC; i++) {
          advec_mom_kernel2_y((double *)p_a[0] + i * 1 * 1,
                              (const double *)p_a[1] + i * 1 * 1,
                              (const double *)p_a[2] + i * 1 * 1,
                              (const double *)p_a[3] + i * 1 * 1);
        }

        // shift pointers to data x direction
        p_a[0] = p_a[0] + (dat0 * off0_0) * SIMD_VEC;
        p_a[1] = p_a[1] + (dat1 * off1_0) * SIMD_VEC;
        p_a[2] = p_a[2] + (dat2 * off2_0) * SIMD_VEC;
        p_a[3] = p_a[3] + (dat3 * off3_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
        advec_mom_kernel2_y((double *)p_a[0], (const double *)p_a[1],
                            (const double *)p_a[2], (const double *)p_a[3]);

        // shift pointers to data x direction
        p_a[0] = p_a[0] + (dat0 * off0_0);
        p_a[1] = p_a[1] + (dat1 * off1_0);
        p_a[2] = p_a[2] + (dat2 * off2_0);
        p_a[3] = p_a[3] + (dat3 * off3_0);
      }

      // shift pointers to data y direction
      p_a[0] = p_a[0] + (dat0 * off0_1);
      p_a[1] = p_a[1] + (dat1 * off1_1);
      p_a[2] = p_a[2] + (dat2 * off2_1);
      p_a[3] = p_a[3] + (dat3 * off3_1);
    }
  }

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[80].time += t1 - t2;
  }

  ops_set_dirtybit_host(args, 4);

  ops_set_halo_dirtybit3(&args[0], range);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[80].mpi_time += t2 - t1;
    OPS_kernels[80].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[80].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[80].transfer += ops_compute_transfer(dim, start, end, &arg2);
    OPS_kernels[80].transfer += ops_compute_transfer(dim, start, end, &arg3);
  }
}
// host stub function
void ops_par_loop_update_halo_kernel1_r1(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 args[8] = { arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7};


  #ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args,8,range,50)) return;
  #endif

  ops_timing_realloc(50,"update_halo_kernel1_r1");
  OPS_kernels[50].count++;

  //compute localy 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

  int x_size = MAX(0,end[0]-start[0]);
  int y_size = MAX(0,end[1]-start[1]);


  xdim0 = args[0].dat->size[0]*args[0].dat->dim;
  xdim1 = args[1].dat->size[0]*args[1].dat->dim;
  xdim2 = args[2].dat->size[0]*args[2].dat->dim;
  xdim3 = args[3].dat->size[0]*args[3].dat->dim;
  xdim4 = args[4].dat->size[0]*args[4].dat->dim;
  xdim5 = args[5].dat->size[0]*args[5].dat->dim;
  xdim6 = args[6].dat->size[0]*args[6].dat->dim;

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

  if (xdim0 != xdim0_update_halo_kernel1_r1_h || xdim1 != xdim1_update_halo_kernel1_r1_h || xdim2 != xdim2_update_halo_kernel1_r1_h || xdim3 != xdim3_update_halo_kernel1_r1_h || xdim4 != xdim4_update_halo_kernel1_r1_h || xdim5 != xdim5_update_halo_kernel1_r1_h || xdim6 != xdim6_update_halo_kernel1_r1_h) {
    xdim0_update_halo_kernel1_r1 = xdim0;
    xdim0_update_halo_kernel1_r1_h = xdim0;
    xdim1_update_halo_kernel1_r1 = xdim1;
    xdim1_update_halo_kernel1_r1_h = xdim1;
    xdim2_update_halo_kernel1_r1 = xdim2;
    xdim2_update_halo_kernel1_r1_h = xdim2;
    xdim3_update_halo_kernel1_r1 = xdim3;
    xdim3_update_halo_kernel1_r1_h = xdim3;
    xdim4_update_halo_kernel1_r1 = xdim4;
    xdim4_update_halo_kernel1_r1_h = xdim4;
    xdim5_update_halo_kernel1_r1 = xdim5;
    xdim5_update_halo_kernel1_r1_h = xdim5;
    xdim6_update_halo_kernel1_r1 = xdim6;
    xdim6_update_halo_kernel1_r1_h = xdim6;
  }

  int dat0 = args[0].dat->elem_size;
  int dat1 = args[1].dat->elem_size;
  int dat2 = args[2].dat->elem_size;
  int dat3 = args[3].dat->elem_size;
  int dat4 = args[4].dat->elem_size;
  int dat5 = args[5].dat->elem_size;
  int dat6 = args[6].dat->elem_size;

  int *arg7h = (int *)arg7.data;
  //Upload large globals
  int consts_bytes = 0;
  consts_bytes += ROUND_UP(NUM_FIELDS*sizeof(int));
  reallocConstArrays(consts_bytes);
  consts_bytes = 0;
  args[7].data = OPS_consts_h + consts_bytes;
  args[7].data_d = OPS_consts_d + consts_bytes;
  for (int d=0; d<NUM_FIELDS; d++) ((int *)args[7].data)[d] = arg7h[d];
  consts_bytes += ROUND_UP(NUM_FIELDS*sizeof(int));
  mvConstArraysToDevice(consts_bytes);

  //set up initial pointers
  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]);
  #ifdef OPS_GPU
  double *p_a0 = (double *)((char *)args[0].data_d + base0);
  #else
  double *p_a0 = (double *)((char *)args[0].data + base0);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d] + OPS_sub_dat_list[args[1].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[1].dat->d_m[d];
  #endif //OPS_MPI
  int base1 = dat1 * 1 * 
    (start[0] * args[1].stencil->stride[0] - args[1].dat->base[0] - d_m[0]);
  base1 = base1+ dat1 *
    args[1].dat->size[0] *
    (start[1] * args[1].stencil->stride[1] - args[1].dat->base[1] - d_m[1]);
  #ifdef OPS_GPU
  double *p_a1 = (double *)((char *)args[1].data_d + base1);
  #else
  double *p_a1 = (double *)((char *)args[1].data + base1);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[2].dat->d_m[d] + OPS_sub_dat_list[args[2].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[2].dat->d_m[d];
  #endif //OPS_MPI
  int base2 = dat2 * 1 * 
    (start[0] * args[2].stencil->stride[0] - args[2].dat->base[0] - d_m[0]);
  base2 = base2+ dat2 *
    args[2].dat->size[0] *
    (start[1] * args[2].stencil->stride[1] - args[2].dat->base[1] - d_m[1]);
  #ifdef OPS_GPU
  double *p_a2 = (double *)((char *)args[2].data_d + base2);
  #else
  double *p_a2 = (double *)((char *)args[2].data + base2);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[3].dat->d_m[d] + OPS_sub_dat_list[args[3].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[3].dat->d_m[d];
  #endif //OPS_MPI
  int base3 = dat3 * 1 * 
    (start[0] * args[3].stencil->stride[0] - args[3].dat->base[0] - d_m[0]);
  base3 = base3+ dat3 *
    args[3].dat->size[0] *
    (start[1] * args[3].stencil->stride[1] - args[3].dat->base[1] - d_m[1]);
  #ifdef OPS_GPU
  double *p_a3 = (double *)((char *)args[3].data_d + base3);
  #else
  double *p_a3 = (double *)((char *)args[3].data + base3);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[4].dat->d_m[d] + OPS_sub_dat_list[args[4].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[4].dat->d_m[d];
  #endif //OPS_MPI
  int base4 = dat4 * 1 * 
    (start[0] * args[4].stencil->stride[0] - args[4].dat->base[0] - d_m[0]);
  base4 = base4+ dat4 *
    args[4].dat->size[0] *
    (start[1] * args[4].stencil->stride[1] - args[4].dat->base[1] - d_m[1]);
  #ifdef OPS_GPU
  double *p_a4 = (double *)((char *)args[4].data_d + base4);
  #else
  double *p_a4 = (double *)((char *)args[4].data + base4);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[5].dat->d_m[d] + OPS_sub_dat_list[args[5].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[5].dat->d_m[d];
  #endif //OPS_MPI
  int base5 = dat5 * 1 * 
    (start[0] * args[5].stencil->stride[0] - args[5].dat->base[0] - d_m[0]);
  base5 = base5+ dat5 *
    args[5].dat->size[0] *
    (start[1] * args[5].stencil->stride[1] - args[5].dat->base[1] - d_m[1]);
  #ifdef OPS_GPU
  double *p_a5 = (double *)((char *)args[5].data_d + base5);
  #else
  double *p_a5 = (double *)((char *)args[5].data + base5);
  #endif

  #ifdef OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[6].dat->d_m[d] + OPS_sub_dat_list[args[6].dat->index]->d_im[d];
  #else //OPS_MPI
  for (int d = 0; d < dim; d++) d_m[d] = args[6].dat->d_m[d];
  #endif //OPS_MPI
  int base6 = dat6 * 1 * 
    (start[0] * args[6].stencil->stride[0] - args[6].dat->base[0] - d_m[0]);
  base6 = base6+ dat6 *
    args[6].dat->size[0] *
    (start[1] * args[6].stencil->stride[1] - args[6].dat->base[1] - d_m[1]);
  #ifdef OPS_GPU
  double *p_a6 = (double *)((char *)args[6].data_d + base6);
  #else
  double *p_a6 = (double *)((char *)args[6].data + base6);
  #endif

  #ifdef OPS_GPU
  int *p_a7 = (int *)args[7].data_d;
  #else
  int *p_a7 = arg7h;
  #endif

  #ifdef OPS_GPU
  ops_H_D_exchanges_device(args, 8);
  #else
  ops_H_D_exchanges_host(args, 8);
  #endif
  ops_halo_exchanges(args,8,range);

  ops_timers_core(&c1,&t1);
  OPS_kernels[50].mpi_time += t1-t2;

  update_halo_kernel1_r1_c_wrapper(
    p_a0,
    p_a1,
    p_a2,
    p_a3,
    p_a4,
    p_a5,
    p_a6,
    p_a7,
    x_size, y_size);

  ops_timers_core(&c2,&t2);
  OPS_kernels[50].time += t2-t1;
  #ifdef OPS_GPU
  ops_set_dirtybit_device(args, 8);
  #else
  ops_set_dirtybit_host(args, 8);
  #endif
  ops_set_halo_dirtybit3(&args[0],range);
  ops_set_halo_dirtybit3(&args[1],range);
  ops_set_halo_dirtybit3(&args[2],range);
  ops_set_halo_dirtybit3(&args[3],range);
  ops_set_halo_dirtybit3(&args[4],range);
  ops_set_halo_dirtybit3(&args[5],range);
  ops_set_halo_dirtybit3(&args[6],range);

  //Update kernel record
  OPS_kernels[50].transfer += ops_compute_transfer(dim, range, &arg0);
  OPS_kernels[50].transfer += ops_compute_transfer(dim, range, &arg1);
  OPS_kernels[50].transfer += ops_compute_transfer(dim, range, &arg2);
  OPS_kernels[50].transfer += ops_compute_transfer(dim, range, &arg3);
  OPS_kernels[50].transfer += ops_compute_transfer(dim, range, &arg4);
  OPS_kernels[50].transfer += ops_compute_transfer(dim, range, &arg5);
  OPS_kernels[50].transfer += ops_compute_transfer(dim, range, &arg6);
}