C++ ops_set_halo_dirtybit3函数代码示例

//.........这里部分代码省略.........
    d_m[d] = args[7].dat->d_m[d];
#endif
  int base7 = 1 * 1 * (start[0] * args[7].stencil->stride[0] -
                       args[7].dat->base[0] - d_m[0]);
  base7 = base7 +
          args[7].dat->size[0] * 1 * (start[1] * args[7].stencil->stride[1] -
                                      args[7].dat->base[1] - d_m[1]);
  base7 = base7 +
          args[7].dat->size[0] * 1 * args[7].dat->size[1] * 1 *
              (start[2] * args[7].stencil->stride[2] - args[7].dat->base[2] -
               d_m[2]);

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

  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[118].mpi_time += t2 - t1;
  }

  if (globalWorkSize[0] > 0 && globalWorkSize[1] > 0 && globalWorkSize[2] > 0) {
    clSafeCall(clEnqueueWriteBuffer(
        OPS_opencl_core.command_queue, OPS_opencl_core.constant[7], CL_TRUE, 0,
        sizeof(field_type) * 1, (void *)&field, 0, NULL, NULL));
    clSafeCall(clFlush(OPS_opencl_core.command_queue));

    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 0, sizeof(cl_mem),
                              (void *)&arg0.data_d));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 1, sizeof(cl_mem),
                              (void *)&arg1.data_d));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 2, sizeof(cl_mem),
                              (void *)&arg2.data_d));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 3, sizeof(cl_mem),
                              (void *)&arg3.data_d));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 4, sizeof(cl_mem),
                              (void *)&arg4.data_d));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 5, sizeof(cl_mem),
                              (void *)&arg5.data_d));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 6, sizeof(cl_mem),
                              (void *)&arg6.data_d));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 7, sizeof(cl_mem),
                              (void *)&arg7.data_d));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 8, sizeof(cl_mem),
                              (void *)&OPS_opencl_core.constant[7]));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 9, sizeof(cl_int),
                              (void *)&base0));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 10, sizeof(cl_int),
                              (void *)&base1));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 11, sizeof(cl_int),
                              (void *)&base2));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 12, sizeof(cl_int),
                              (void *)&base3));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 13, sizeof(cl_int),
                              (void *)&base4));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 14, sizeof(cl_int),
                              (void *)&base5));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 15, sizeof(cl_int),
                              (void *)&base6));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 16, sizeof(cl_int),
                              (void *)&base7));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 17, sizeof(cl_int),
                              (void *)&x_size));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 18, sizeof(cl_int),
                              (void *)&y_size));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[118], 19, sizeof(cl_int),
                              (void *)&z_size));

    // call/enque opencl kernel wrapper function
    clSafeCall(clEnqueueNDRangeKernel(
        OPS_opencl_core.command_queue, OPS_opencl_core.kernel[118], 3, NULL,
        globalWorkSize, localWorkSize, 0, NULL, NULL));
  }
  if (OPS_diags > 1) {
    clSafeCall(clFinish(OPS_opencl_core.command_queue));
  }

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[118].time += t1 - t2;
  }

  ops_set_dirtybit_device(args, 8);
  ops_set_halo_dirtybit3(&args[6], range);
  ops_set_halo_dirtybit3(&args[7], range);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[118].mpi_time += t2 - t1;
    OPS_kernels[118].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[118].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[118].transfer += ops_compute_transfer(dim, start, end, &arg2);
    OPS_kernels[118].transfer += ops_compute_transfer(dim, start, end, &arg3);
    OPS_kernels[118].transfer += ops_compute_transfer(dim, start, end, &arg4);
    OPS_kernels[118].transfer += ops_compute_transfer(dim, start, end, &arg5);
    OPS_kernels[118].transfer += ops_compute_transfer(dim, start, end, &arg6);
    OPS_kernels[118].transfer += ops_compute_transfer(dim, start, end, &arg7);
  }
}

//.........这里部分代码省略.........
  #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);
}

//.........这里部分代码省略.........
  consts_bytes = 0;
  arg2.data = OPS_consts_h + consts_bytes;
  arg2.data_d = OPS_consts_d + consts_bytes;
  for (int d = 0; d < NUM_FIELDS; d++)
    ((int *)arg2.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
  for (int d = 0; d < dim; d++)
    d_m[d] = args[0].dat->d_m[d];
#endif
  int base0 = 1 * 1 * (start[0] * args[0].stencil->stride[0] -
                       args[0].dat->base[0] - d_m[0]);
  base0 = base0 +
          args[0].dat->size[0] * 1 * (start[1] * args[0].stencil->stride[1] -
                                      args[0].dat->base[1] - d_m[1]);
  base0 = base0 +
          args[0].dat->size[0] * 1 * args[0].dat->size[1] * 1 *
              (start[2] * args[0].stencil->stride[2] - args[0].dat->base[2] -
               d_m[2]);

#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 = 1 * 1 * (start[0] * args[1].stencil->stride[0] -
                       args[1].dat->base[0] - d_m[0]);
  base1 = base1 +
          args[1].dat->size[0] * 1 * (start[1] * args[1].stencil->stride[1] -
                                      args[1].dat->base[1] - d_m[1]);
  base1 = base1 +
          args[1].dat->size[0] * 1 * args[1].dat->size[1] * 1 *
              (start[2] * args[1].stencil->stride[2] - args[1].dat->base[2] -
               d_m[2]);

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

  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[53].mpi_time += t2 - t1;
  }

  if (globalWorkSize[0] > 0 && globalWorkSize[1] > 0 && globalWorkSize[2] > 0) {

    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[53], 0, sizeof(cl_mem),
                              (void *)&arg0.data_d));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[53], 1, sizeof(cl_mem),
                              (void *)&arg1.data_d));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[53], 2, sizeof(cl_mem),
                              (void *)&arg2.data_d));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[53], 3, sizeof(cl_int),
                              (void *)&base0));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[53], 4, sizeof(cl_int),
                              (void *)&base1));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[53], 5, sizeof(cl_int),
                              (void *)&x_size));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[53], 6, sizeof(cl_int),
                              (void *)&y_size));
    clSafeCall(clSetKernelArg(OPS_opencl_core.kernel[53], 7, sizeof(cl_int),
                              (void *)&z_size));

    // call/enque opencl kernel wrapper function
    clSafeCall(clEnqueueNDRangeKernel(
        OPS_opencl_core.command_queue, OPS_opencl_core.kernel[53], 3, NULL,
        globalWorkSize, localWorkSize, 0, NULL, NULL));
  }
  if (OPS_diags > 1) {
    clSafeCall(clFinish(OPS_opencl_core.command_queue));
  }

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[53].time += t1 - t2;
  }

  ops_set_dirtybit_device(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(&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);
  }
}

//.........这里部分代码省略.........
    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
    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]);
    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
    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]);
    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_xvel_minus_4_right((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_kernel2_xvel_minus_4_right(
              (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(&c1, &t1);
    OPS_kernels[29].time += t1 - t2;
  }

  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(&c2, &t2);
    OPS_kernels[29].mpi_time += t2 - t1;
    OPS_kernels[29].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[29].transfer += ops_compute_transfer(dim, start, end, &arg1);
  }
}

//.........这里部分代码省略.........
    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_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[42].mpi_time += t1-t2;

  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, z_size);

  ops_timers_core(&c2,&t2);
  OPS_kernels[42].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[42].transfer += ops_compute_transfer(dim, range, &arg0);
  OPS_kernels[42].transfer += ops_compute_transfer(dim, range, &arg1);
  OPS_kernels[42].transfer += ops_compute_transfer(dim, range, &arg2);
  OPS_kernels[42].transfer += ops_compute_transfer(dim, range, &arg3);
  OPS_kernels[42].transfer += ops_compute_transfer(dim, range, &arg4);
  OPS_kernels[42].transfer += ops_compute_transfer(dim, range, &arg5);
  OPS_kernels[42].transfer += ops_compute_transfer(dim, range, &arg6);
}

//.........这里部分代码省略.........
  (start[0] * args[1].stencil->stride[0] - args[1].dat->base[0] - d_m[0]);
  base1 = base1 + args[1].dat->size[0] *
  (start[1] * args[1].stencil->stride[1] - args[1].dat->base[1] - d_m[1]);

  #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 = 1 * 
  (start[0] * args[2].stencil->stride[0] - args[2].dat->base[0] - d_m[0]);
  base2 = base2 + args[2].dat->size[0] *
  (start[1] * args[2].stencil->stride[1] - args[2].dat->base[1] - d_m[1]);

  #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 = 1 * 
  (start[0] * args[3].stencil->stride[0] - args[3].dat->base[0] - d_m[0]);
  base3 = base3 + args[3].dat->size[0] *
  (start[1] * args[3].stencil->stride[1] - args[3].dat->base[1] - d_m[1]);

  #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 = 1 * 
  (start[0] * args[4].stencil->stride[0] - args[4].dat->base[0] - d_m[0]);
  base4 = base4 + args[4].dat->size[0] *
  (start[1] * args[4].stencil->stride[1] - args[4].dat->base[1] - d_m[1]);

  #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 = 1 * 
  (start[0] * args[5].stencil->stride[0] - args[5].dat->base[0] - d_m[0]);
  base5 = base5 + args[5].dat->size[0] *
  (start[1] * args[5].stencil->stride[1] - args[5].dat->base[1] - d_m[1]);

  #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 = 1 * 
  (start[0] * args[6].stencil->stride[0] - args[6].dat->base[0] - d_m[0]);
  base6 = base6 + args[6].dat->size[0] *
  (start[1] * args[6].stencil->stride[1] - args[6].dat->base[1] - d_m[1]);


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

  ops_timers_core(&c1,&t1);
  OPS_kernels[34].mpi_time += t1-t2;


  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 0, sizeof(cl_mem), (void*) &arg0.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 1, sizeof(cl_mem), (void*) &arg1.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 2, sizeof(cl_mem), (void*) &arg2.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 3, sizeof(cl_mem), (void*) &arg3.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 4, sizeof(cl_mem), (void*) &arg4.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 5, sizeof(cl_mem), (void*) &arg5.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 6, sizeof(cl_mem), (void*) &arg6.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 7, sizeof(cl_int), (void*) &base0 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 8, sizeof(cl_int), (void*) &base1 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 9, sizeof(cl_int), (void*) &base2 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 10, sizeof(cl_int), (void*) &base3 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 11, sizeof(cl_int), (void*) &base4 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 12, sizeof(cl_int), (void*) &base5 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 13, sizeof(cl_int), (void*) &base6 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 14, sizeof(cl_int), (void*) &x_size ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[34], 15, sizeof(cl_int), (void*) &y_size ));

  //call/enque opencl kernel wrapper function
  clSafeCall( clEnqueueNDRangeKernel(OPS_opencl_core.command_queue, OPS_opencl_core.kernel[34], 3, NULL, globalWorkSize, localWorkSize, 0, NULL, NULL) );
  if (OPS_diags>1) {
    clSafeCall( clFinish(OPS_opencl_core.command_queue) );
  }

  ops_set_dirtybit_device(args, 7);
  ops_set_halo_dirtybit3(&args[6],range);

  //Update kernel record
  ops_timers_core(&c2,&t2);
  OPS_kernels[34].time += t2-t1;
  OPS_kernels[34].transfer += ops_compute_transfer(dim, range, &arg0);
  OPS_kernels[34].transfer += ops_compute_transfer(dim, range, &arg1);
  OPS_kernels[34].transfer += ops_compute_transfer(dim, range, &arg2);
  OPS_kernels[34].transfer += ops_compute_transfer(dim, range, &arg3);
  OPS_kernels[34].transfer += ops_compute_transfer(dim, range, &arg4);
  OPS_kernels[34].transfer += ops_compute_transfer(dim, range, &arg5);
  OPS_kernels[34].transfer += ops_compute_transfer(dim, range, &arg6);
}

//.........这里部分代码省略.........
    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);
  }
}

//.........这里部分代码省略.........
            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;

    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 * 1,
                                  (double *)p_a[4] + i * 1 * 1,
                                  (double *)p_a[5] + i * 1 * 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
      arg_idx[0] = start0;
#endif
      arg_idx[1]++;
    }
  }

  if (OPS_diags > 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);

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[0].mpi_time += t2 - t1;
    OPS_kernels[0].transfer += ops_compute_transfer(dim, start, end, &arg3);
    OPS_kernels[0].transfer += ops_compute_transfer(dim, start, end, &arg4);
    OPS_kernels[0].transfer += ops_compute_transfer(dim, start, end, &arg5);
  }
}

//.........这里部分代码省略.........
    #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 * 
    (start0 * args[6].stencil->stride[0] - args[6].dat->base[0] - d_m[0]);
    base6 = base6+ dat6 *
      args[6].dat->size[0] *
      (start1 * 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] *
      (start2 * args[6].stencil->stride[2] - args[6].dat->base[2] - d_m[2]);
    p_a[6] = (char *)args[6].data + base6;

    p_a[7] = (char *)args[7].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_kernel1_ba2(  (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_ba2(  (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);
      }
      //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);
    }
  }

  ops_timers_core(&c1,&t1);
  OPS_kernels[49].time += t1-t2;

  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_timers_core(&c2,&t2);
  OPS_kernels[49].mpi_time += t2-t1;
  OPS_kernels[49].transfer += ops_compute_transfer(dim, range, &arg0);
  OPS_kernels[49].transfer += ops_compute_transfer(dim, range, &arg1);
  OPS_kernels[49].transfer += ops_compute_transfer(dim, range, &arg2);
  OPS_kernels[49].transfer += ops_compute_transfer(dim, range, &arg3);
  OPS_kernels[49].transfer += ops_compute_transfer(dim, range, &arg4);
  OPS_kernels[49].transfer += ops_compute_transfer(dim, range, &arg5);
  OPS_kernels[49].transfer += ops_compute_transfer(dim, range, &arg6);
}

//.........这里部分代码省略.........
  #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);
}

//.........这里部分代码省略.........
  int nthreads = omp_get_max_threads( );
  #else
  int nthreads = 1;
  #endif
  xdim0 = args[0].dat->size[0]*args[0].dat->dim;

  ops_H_D_exchanges_host(args, 2);

  //Halo Exchanges
  ops_halo_exchanges(args,2,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[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;

    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];
    #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]);
    p_a[0] = (char *)args[0].data + base0;

    p_a[1] = (char *)arg_idx;


    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++ ){
          mblock_populate_kernel(  (double * )p_a[0]+ i*1, arg_idx );

          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
        mblock_populate_kernel(  (double * )p_a[0], arg_idx );


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

      //shift pointers to data y direction
      p_a[0]= p_a[0] + (dat0 * off0_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, 2);

  ops_set_halo_dirtybit3(&args[0],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, &arg0);
}

//.........这里部分代码省略.........

        }

        //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);
}

//.........这里部分代码省略.........
      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);
  }
}

//.........这里部分代码省略.........
  }
  #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, "left_bndcon");
  #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 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

  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);

  //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;

  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[2].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++ ){
        left_bndcon(  (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
      left_bndcon(  (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]++;
    }

    //shift pointers to data y direction
    p_a[0]= p_a[0] + (dat0 * off0_1);
    #ifdef OPS_MPI
    arg_idx[0] = sb->decomp_disp[0]+start[0];
    #else
    arg_idx[0] = start[0];
    #endif
    arg_idx[1]++;
  }
  if (OPS_diags > 1) {
    ops_timers_core(&c2,&t2);
    OPS_kernels[2].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[2].mpi_time += t1-t2;
    OPS_kernels[2].transfer += ops_compute_transfer(dim, start, end, &arg0);
  }
}

//.........这里部分代码省略.........
  base11 = base11 + args[11].dat->size[0] *
  (start[1] * args[11].stencil->stride[1] - args[11].dat->base[1] - d_m[1]);
  base11 = base11 + 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_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 = 1 * 
  (start[0] * args[12].stencil->stride[0] - args[12].dat->base[0] - d_m[0]);
  base12 = base12 + args[12].dat->size[0] *
  (start[1] * args[12].stencil->stride[1] - args[12].dat->base[1] - d_m[1]);
  base12 = base12 + 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_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 = 1 * 
  (start[0] * args[13].stencil->stride[0] - args[13].dat->base[0] - d_m[0]);
  base13 = base13 + args[13].dat->size[0] *
  (start[1] * args[13].stencil->stride[1] - args[13].dat->base[1] - d_m[1]);
  base13 = base13 + 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]);


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

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


  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 0, sizeof(cl_mem), (void*) &arg0.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 1, sizeof(cl_mem), (void*) &arg1.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 2, sizeof(cl_mem), (void*) &arg2.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 3, sizeof(cl_mem), (void*) &arg3.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 4, sizeof(cl_mem), (void*) &arg4.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 5, sizeof(cl_mem), (void*) &arg5.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 6, sizeof(cl_mem), (void*) &arg6.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 7, sizeof(cl_mem), (void*) &arg7.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 8, sizeof(cl_mem), (void*) &arg8.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 9, sizeof(cl_mem), (void*) &arg9.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 10, sizeof(cl_mem), (void*) &arg10.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 11, sizeof(cl_mem), (void*) &arg11.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 12, sizeof(cl_mem), (void*) &arg12.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 13, sizeof(cl_mem), (void*) &arg13.data_d ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 14, sizeof(cl_double), (void*) &dt ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 15, sizeof(cl_int), (void*) &base0 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 16, sizeof(cl_int), (void*) &base1 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 17, sizeof(cl_int), (void*) &base2 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 18, sizeof(cl_int), (void*) &base3 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 19, sizeof(cl_int), (void*) &base4 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 20, sizeof(cl_int), (void*) &base5 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 21, sizeof(cl_int), (void*) &base6 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 22, sizeof(cl_int), (void*) &base7 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 23, sizeof(cl_int), (void*) &base8 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 24, sizeof(cl_int), (void*) &base9 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 25, sizeof(cl_int), (void*) &base10 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 26, sizeof(cl_int), (void*) &base11 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 27, sizeof(cl_int), (void*) &base12 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 28, sizeof(cl_int), (void*) &base13 ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 29, sizeof(cl_int), (void*) &x_size ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 30, sizeof(cl_int), (void*) &y_size ));
  clSafeCall( clSetKernelArg(OPS_opencl_core.kernel[5], 31, sizeof(cl_int), (void*) &z_size ));

  //call/enque opencl kernel wrapper function
  clSafeCall( clEnqueueNDRangeKernel(OPS_opencl_core.command_queue, OPS_opencl_core.kernel[5], 3, NULL, globalWorkSize, localWorkSize, 0, NULL, NULL) );
  if (OPS_diags>1) {
    clSafeCall( clFinish(OPS_opencl_core.command_queue) );
  }

  ops_set_dirtybit_device(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_timers_core(&c2,&t2);
  OPS_kernels[5].time += t2-t1;
  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);
}