C++ ops_timers_core函数代码示例

OGeek|极客世界-中国程序员成长平台 › 门户 › 编程› C++›C++教程

原作者: [db:作者] 来自: [db:来源] 收藏邀请

本文整理汇总了C++中ops_timers_core函数的典型用法代码示例。如果您正苦于以下问题：C++ ops_timers_core函数的具体用法？C++ ops_timers_core怎么用？C++ ops_timers_core使用的例子？那么恭喜您, 这里精选的函数代码示例或许可以为您提供帮助。

在下文中一共展示了ops_timers_core函数的20个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于我们的系统推荐出更棒的C++代码示例。

示例1: ops_par_loop_update_halo_kernel2_yvel_minus_4_top

// host stub function
void ops_par_loop_update_halo_kernel2_yvel_minus_4_top(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(67,"update_halo_kernel2_yvel_minus_4_top");
  OPS_kernels[67].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_kernel2_yvel_minus_4_top_h || ydim0 != ydim0_update_halo_kernel2_yvel_minus_4_top_h || xdim1 != xdim1_update_halo_kernel2_yvel_minus_4_top_h || ydim1 != ydim1_update_halo_kernel2_yvel_minus_4_top_h) {
    xdim0_update_halo_kernel2_yvel_minus_4_top = xdim0;
    xdim0_update_halo_kernel2_yvel_minus_4_top_h = xdim0;
    ydim0_update_halo_kernel2_yvel_minus_4_top = ydim0;
    ydim0_update_halo_kernel2_yvel_minus_4_top_h = ydim0;
    xdim1_update_halo_kernel2_yvel_minus_4_top = xdim1;
    xdim1_update_halo_kernel2_yvel_minus_4_top_h = xdim1;
    ydim1_update_halo_kernel2_yvel_minus_4_top = ydim1;
    ydim1_update_halo_kernel2_yvel_minus_4_top_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);
//.........这里部分代码省略.........

开发者ID:satyajammy，项目名称:OPS，代码行数:101，代码来源:update_halo_kernel2_yvel_minus_4_top_openacc_kernel.cpp

示例2: ops_par_loop_update_halo_kernel3_minus_2_a

// host stub function
void ops_par_loop_update_halo_kernel3_minus_2_a(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(94,"update_halo_kernel3_minus_2_a");
  OPS_kernels[94].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

  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]);


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

  //build opencl kernel if not already built

  buildOpenCLKernels_update_halo_kernel3_minus_2_a(
  xdim0,ydim0,xdim1,ydim1);

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

  //set up OpenCL thread blocks
  size_t globalWorkSize[3] = {((x_size-1)/OPS_block_size_x+ 1)*OPS_block_size_x, ((y_size-1)/OPS_block_size_y + 1)*OPS_block_size_y, MAX(1,end[2]-start[2])};
  size_t localWorkSize[3] =  {OPS_block_size_x,OPS_block_size_y,1};


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

  int consts_bytes = 0;

  consts_bytes += ROUND_UP(NUM_FIELDS*sizeof(int));

  reallocConstArrays(consts_bytes);

  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);
  int dat0 = args[0].dat->elem_size;
  int dat1 = args[1].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 = 1 * 
  (start[0] * args[0].stencil->stride[0] - args[0].dat->base[0] - d_m[0]);
  base0 = base0 + args[0].dat->size[0] *
  (start[1] * args[0].stencil->stride[1] - args[0].dat->base[1] - d_m[1]);
  base0 = base0 + 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_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 = 1 * 
//.........这里部分代码省略.........

开发者ID:satyajammy，项目名称:OPS，代码行数:101，代码来源:update_halo_kernel3_minus_2_a_opencl_kernel.cpp

示例3: ops_par_loop_advec_cell_kernel4_xdir

// 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
//.........这里部分代码省略.........

开发者ID:gihanmudalige，项目名称:OPS，代码行数:101，代码来源:advec_cell_kernel4_xdir_seq_kernel.cpp

示例4: ops_par_loop_update_halo_kernel2_zvel_minus_2_back

// host stub function
void ops_par_loop_update_halo_kernel2_zvel_minus_2_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_arg args[3] = {arg0, arg1, arg2};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 3, range, 57))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(57, "update_halo_kernel2_zvel_minus_2_back");
    OPS_kernels[57].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
//.........这里部分代码省略.........

开发者ID:gihanmudalige，项目名称:OPS，代码行数:101，代码来源:update_halo_kernel2_zvel_minus_2_back_openacc_kernel.cpp

示例5: ops_par_loop_update_halo_kernel2_yvel_plus_2_right

// host stub function
void ops_par_loop_update_halo_kernel2_yvel_plus_2_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;

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

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 3, range, 43))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(43, "update_halo_kernel2_yvel_plus_2_right");
    OPS_kernels[43].count++;
    ops_timers_core(&c1, &t1);
  }

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

  // compute locally allocated range for the sub-block

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

#ifdef OPS_MPI
  if (!sb->owned)
    return;
  for (int n = 0; n < 3; n++) {
    start[n] = sb->decomp_disp[n];
    end[n] = sb->decomp_disp[n] + sb->decomp_size[n];
    if (start[n] >= range[2 * n]) {
      start[n] = 0;
    } else {
      start[n] = range[2 * n] - start[n];
    }
    if (sb->id_m[n] == MPI_PROC_NULL && range[2 * n] < 0)
      start[n] = range[2 * n];
    if (end[n] >= range[2 * n + 1]) {
      end[n] = range[2 * n + 1] - sb->decomp_disp[n];
    } else {
      end[n] = sb->decomp_size[n];
    }
    if (sb->id_p[n] == MPI_PROC_NULL &&
        (range[2 * n + 1] > sb->decomp_disp[n] + sb->decomp_size[n]))
      end[n] += (range[2 * n + 1] - sb->decomp_disp[n] - sb->decomp_size[n]);
    if (end[n] < start[n])
      end[n] = start[n];
  }
#else
  for (int n = 0; n < 3; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
  }
#endif
#ifdef OPS_DEBUG
  ops_register_args(args, "update_halo_kernel2_yvel_plus_2_right");
#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 = (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);

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

#ifdef _OPENMP
  int nthreads = omp_get_max_threads();
#else
  int nthreads = 1;
//.........这里部分代码省略.........

开发者ID:gihanmudalige，项目名称:OPS，代码行数:101，代码来源:update_halo_kernel2_yvel_plus_2_right_omp_kernel.cpp

示例6: ops_par_loop_advec_mom_kernel_post_pre_advec_y_execute

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

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

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

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 5, range, 132))
    return;
#endif

  if (OPS_diags > 1) {
    OPS_kernels[132].count++;
    ops_timers_core(&c2, &t2);
  }

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

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

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

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

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

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

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

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

  // initialize global variable with the dimension of dats
  int xdim0_advec_mom_kernel_post_pre_advec_y = args[0].dat->size[0];
  int ydim0_advec_mom_kernel_post_pre_advec_y = args[0].dat->size[1];
  int xdim1_advec_mom_kernel_post_pre_advec_y = args[1].dat->size[0];
  int ydim1_advec_mom_kernel_post_pre_advec_y = args[1].dat->size[1];
  int xdim2_advec_mom_kernel_post_pre_advec_y = args[2].dat->size[0];
  int ydim2_advec_mom_kernel_post_pre_advec_y = args[2].dat->size[1];
  int xdim3_advec_mom_kernel_post_pre_advec_y = args[3].dat->size[0];
  int ydim3_advec_mom_kernel_post_pre_advec_y = args[3].dat->size[1];
  int xdim4_advec_mom_kernel_post_pre_advec_y = args[4].dat->size[0];
  int ydim4_advec_mom_kernel_post_pre_advec_y = args[4].dat->size[1];

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[132].mpi_time += t1 - t2;
  }

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

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

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

开发者ID:gihanmudalige，项目名称:OPS，代码行数:101，代码来源:advec_mom_kernel_post_pre_advec_y_seq_kernel.cpp

示例7: ops_par_loop_advec_mom_kernel_post_pre_advec_y

// host stub function
void ops_par_loop_advec_mom_kernel_post_pre_advec_y(
    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, 132))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(132, "advec_mom_kernel_post_pre_advec_y");
    OPS_kernels[132].count++;
    ops_timers_core(&c1, &t1);
  }

  // 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
  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]);

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

  // build opencl kernel if not already built

  buildOpenCLKernels_advec_mom_kernel_post_pre_advec_y(
      xdim0, ydim0, xdim1, ydim1, xdim2, ydim2, xdim3, ydim3, xdim4, ydim4);

  // set up OpenCL thread blocks
  size_t globalWorkSize[3] = {
      ((x_size - 1) / OPS_block_size_x + 1) * OPS_block_size_x,
      ((y_size - 1) / OPS_block_size_y + 1) * OPS_block_size_y,
      ((z_size - 1) / OPS_block_size_z + 1) * OPS_block_size_z};
  size_t localWorkSize[3] = {OPS_block_size_x, OPS_block_size_y,
                             OPS_block_size_z};

  // 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] -
//.........这里部分代码省略.........

开发者ID:gihanmudalige，项目名称:OPS，代码行数:101，代码来源:advec_mom_kernel_post_pre_advec_y_opencl_kernel.cpp

示例8: ops_par_loop_update_halo_kernel5_minus_4_back_execute

// host stub function
void ops_par_loop_update_halo_kernel5_minus_4_back_execute(
    ops_kernel_descriptor *desc) {
  ops_block block = desc->block;
  int dim = desc->dim;
  int *range = desc->range;
  ops_arg arg0 = desc->args[0];
  ops_arg arg1 = desc->args[1];
  ops_arg arg2 = desc->args[2];

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

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

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 3, range, 92))
    return;
#endif

  if (OPS_diags > 1) {
    OPS_kernels[92].count++;
    ops_timers_core(&c2, &t2);
  }

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

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

#ifdef OPS_DEBUG
  ops_register_args(args, "update_halo_kernel5_minus_4_back");
#endif

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

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

  const int *__restrict__ fields = (int *)args[2].data;

  // initialize global variable with the dimension of dats
  int xdim0_update_halo_kernel5_minus_4_back = args[0].dat->size[0];
  int ydim0_update_halo_kernel5_minus_4_back = args[0].dat->size[1];
  int xdim1_update_halo_kernel5_minus_4_back = args[1].dat->size[0];
  int ydim1_update_halo_kernel5_minus_4_back = args[1].dat->size[1];

  if (OPS_diags > 1) {
    ops_timers_core(&c1, &t1);
    OPS_kernels[92].mpi_time += t1 - t2;
  }

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

        if (fields[FIELD_VOL_FLUX_Z] == 1)
          vol_flux_z[OPS_ACC0(0, 0, 0)] = -vol_flux_z[OPS_ACC0(0, 0, 4)];
        if (fields[FIELD_MASS_FLUX_Z] == 1)
          mass_flux_z[OPS_ACC1(0, 0, 0)] = -mass_flux_z[OPS_ACC1(0, 0, 4)];
      }
    }
  }
  if (OPS_diags > 1) {
    ops_timers_core(&c2, &t2);
    OPS_kernels[92].time += t2 - t1;
  }

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c1, &t1);
    OPS_kernels[92].mpi_time += t1 - t2;
    OPS_kernels[92].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[92].transfer += ops_compute_transfer(dim, start, end, &arg1);
  }
}

开发者ID:gihanmudalige，项目名称:OPS，代码行数:89，代码来源:update_halo_kernel5_minus_4_back_seq_kernel.cpp

示例9: ops_par_loop_accelerate_kernel_execute

// host stub function
void ops_par_loop_accelerate_kernel_execute(ops_kernel_descriptor *desc) {
  ops_block block = desc->block;
  int dim = desc->dim;
  int *range = desc->range;
  ops_arg arg0 = desc->args[0];
  ops_arg arg1 = desc->args[1];
  ops_arg arg2 = desc->args[2];
  ops_arg arg3 = desc->args[3];
  ops_arg arg4 = desc->args[4];
  ops_arg arg5 = desc->args[5];
  ops_arg arg6 = desc->args[6];
  ops_arg arg7 = desc->args[7];
  ops_arg arg8 = desc->args[8];
  ops_arg arg9 = desc->args[9];
  ops_arg arg10 = desc->args[10];
  ops_arg arg11 = desc->args[11];
  ops_arg arg12 = desc->args[12];
  ops_arg arg13 = desc->args[13];

  // 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, 105))
    return;
#endif

  if (OPS_diags > 1) {
    OPS_kernels[105].count++;
    ops_timers_core(&c2, &t2);
  }

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

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

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

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

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

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

  int base3 = args[3].dat->base_offset;
  const double *__restrict__ xvel0 = (double *)(args[3].data + base3);

  int base4 = args[4].dat->base_offset;
  double *__restrict__ xvel1 = (double *)(args[4].data + base4);

  int base5 = args[5].dat->base_offset;
  const double *__restrict__ xarea = (double *)(args[5].data + base5);

  int base6 = args[6].dat->base_offset;
  const double *__restrict__ pressure = (double *)(args[6].data + base6);

  int base7 = args[7].dat->base_offset;
  const double *__restrict__ yvel0 = (double *)(args[7].data + base7);

  int base8 = args[8].dat->base_offset;
  double *__restrict__ yvel1 = (double *)(args[8].data + base8);

  int base9 = args[9].dat->base_offset;
  const double *__restrict__ yarea = (double *)(args[9].data + base9);

  int base10 = args[10].dat->base_offset;
  const double *__restrict__ viscosity = (double *)(args[10].data + base10);

  int base11 = args[11].dat->base_offset;
  const double *__restrict__ zvel0 = (double *)(args[11].data + base11);

  int base12 = args[12].dat->base_offset;
  double *__restrict__ zvel1 = (double *)(args[12].data + base12);

  int base13 = args[13].dat->base_offset;
  const double *__restrict__ zarea = (double *)(args[13].data + base13);

  // initialize global variable with the dimension of dats
  int xdim0_accelerate_kernel = args[0].dat->size[0];
  int ydim0_accelerate_kernel = args[0].dat->size[1];
  int xdim1_accelerate_kernel = args[1].dat->size[0];
  int ydim1_accelerate_kernel = args[1].dat->size[1];
  int xdim2_accelerate_kernel = args[2].dat->size[0];
  int ydim2_accelerate_kernel = args[2].dat->size[1];
  int xdim3_accelerate_kernel = args[3].dat->size[0];
//.........这里部分代码省略.........

开发者ID:gihanmudalige，项目名称:OPS，代码行数:101，代码来源:accelerate_kernel_seq_kernel.cpp

示例10: ops_par_loop_reset_field_kernel1

// host stub function
void ops_par_loop_reset_field_kernel1(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};


  ops_timing_realloc(3,"reset_field_kernel1");
  OPS_kernels[3].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];

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

  if (xdim0 != xdim0_reset_field_kernel1_h || ydim0 != ydim0_reset_field_kernel1_h || xdim1 != xdim1_reset_field_kernel1_h || ydim1 != ydim1_reset_field_kernel1_h || xdim2 != xdim2_reset_field_kernel1_h || ydim2 != ydim2_reset_field_kernel1_h || xdim3 != xdim3_reset_field_kernel1_h || ydim3 != ydim3_reset_field_kernel1_h) {
    xdim0_reset_field_kernel1 = xdim0;
    xdim0_reset_field_kernel1_h = xdim0;
    ydim0_reset_field_kernel1 = ydim0;
    ydim0_reset_field_kernel1_h = ydim0;
    xdim1_reset_field_kernel1 = xdim1;
    xdim1_reset_field_kernel1_h = xdim1;
    ydim1_reset_field_kernel1 = ydim1;
    ydim1_reset_field_kernel1_h = ydim1;
    xdim2_reset_field_kernel1 = xdim2;
    xdim2_reset_field_kernel1_h = xdim2;
    ydim2_reset_field_kernel1 = ydim2;
    ydim2_reset_field_kernel1_h = ydim2;
    xdim3_reset_field_kernel1 = xdim3;
    xdim3_reset_field_kernel1_h = xdim3;
    ydim3_reset_field_kernel1 = ydim3;
    ydim3_reset_field_kernel1_h = ydim3;
  }

  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;


  //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
//.........这里部分代码省略.........

开发者ID:satyajammy，项目名称:OPS，代码行数:101，代码来源:reset_field_kernel1_openacc_kernel.cpp

示例11: ops_par_loop_update_halo_kernel2_xvel_minus_4_right

// host stub function
void ops_par_loop_update_halo_kernel2_xvel_minus_4_right(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(59,"update_halo_kernel2_xvel_minus_4_right");
  OPS_kernels[59].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_xvel_minus_4_right");
  #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];



  //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;

  //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 *
//.........这里部分代码省略.........

开发者ID:satyajammy，项目名称:OPS，代码行数:101，代码来源:update_halo_kernel2_xvel_minus_4_right_seq_kernel.cpp

示例12: ops_par_loop_tea_leaf_common_init_kernel

// host stub function
void ops_par_loop_tea_leaf_common_init_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;
  ops_arg args[7] = {arg0, arg1, arg2, arg3, arg4, arg5, arg6};

#ifdef CHECKPOINTING
  if (!ops_checkpointing_before(args, 7, range, 36))
    return;
#endif

  if (OPS_diags > 1) {
    ops_timing_realloc(36, "tea_leaf_common_init_kernel");
    OPS_kernels[36].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, 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
  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;

  // 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
//.........这里部分代码省略.........

开发者ID:gihanmudalige，项目名称:OPS，代码行数:101，代码来源:tea_leaf_common_init_kernel_openacc_kernel.cpp

示例13: ops_timers

void ops_timers(double * cpu, double * et)
{
    ops_timers_core(cpu,et);
}

开发者ID:satyajammy，项目名称:OPS，代码行数:4，代码来源:ops_seq.c

示例14: main


//.........这里部分代码省略.........
  ops_halo *halos = (ops_halo *)malloc(2*(ngrid_x*(ngrid_y-1)+(ngrid_x-1)*ngrid_y)*sizeof(ops_halo *));
  int off = 0;
  for (int j = 0; j < ngrid_y; j++) {
    for (int i = 0; i < ngrid_x; i++) {
      if (i > 0) {
        int halo_iter[] = {1,sizes[2*(i+ngrid_x*j)+1]};
        int base_from[] = {sizes[2*(i-1+ngrid_x*j)]-1,0};
        int base_to[] = {-1,0};
        int dir[] = {1,2};
        halos[off++] = ops_decl_halo(u[i-1+ngrid_x*j], u[i+ngrid_x*j], halo_iter, base_from, base_to, dir, dir);
        base_from[0] = 0; base_to[0] = sizes[2*(i+ngrid_x*j)];
        halos[off++] = ops_decl_halo(u[i+ngrid_x*j], u[i-1+ngrid_x*j], halo_iter, base_from, base_to, dir, dir);
      }
      if (j > 0) {
        int halo_iter[] = {sizes[2*(i+ngrid_x*j)],1};
        int base_from[] = {0,sizes[2*(i+ngrid_x*(j-1))+1]-1};
        int base_to[] = {0,-1};
        int dir[] = {1,2};
        halos[off++] = ops_decl_halo(u[i+ngrid_x*(j-1)], u[i+ngrid_x*j], halo_iter, base_from, base_to, dir, dir);
        base_from[1] = 0; base_to[1] = sizes[2*(i+ngrid_x*j)+1];
        halos[off++] = ops_decl_halo(u[i+ngrid_x*j], u[i+ngrid_x*(j-1)], halo_iter, base_from, base_to, dir, dir);
      }
    }
  }
  if (off != 2*(ngrid_x*(ngrid_y-1)+(ngrid_x-1)*ngrid_y)) printf("Something is not right\n");
  ops_halo_group u_halos = ops_decl_halo_group(off,halos);

  ops_partition("");
  ops_checkpointing_init("check.h5", 5.0);
  /**-------------------------- Computations --------------------------**/


  double ct0, ct1, et0, et1;
  ops_timers_core(&ct0, &et0);

  //populate forcing, reference solution and boundary conditions
  for (int j = 0; j < ngrid_y; j++) {
    for (int i = 0; i < ngrid_x; i++) {
      int iter_range[] = {-1,sizes[2*(i+ngrid_x*j)]+1,-1,sizes[2*(i+ngrid_x*j)+1]+1};
      ops_par_loop(poisson_kernel_populate, "poisson_kernel_populate", blocks[i+ngrid_x*j], 2, iter_range,
               ops_arg_gbl(&disps[2*(i+ngrid_x*j)], 1, "int", OPS_READ),
               ops_arg_gbl(&a

鲜花

握手

雷人

路过

鸡蛋

该文章已有0人参与评论

请发表评论

全部评论

专题导读

More+

10-27 六六分期app的软件客服如何联系？(六六分期

11-06 可心卡盟:win10系统火狐flash插件崩溃怎么

11-06 亲亲特价:怎么删除回收站图标

11-06 济南大学虚拟社区:鲁大师节能降温的具体办

11-06 xlueops.exe:无线网络安装向导

11-06 女斗合众国:win7系统cf与主机连接不稳定怎

11-06 0xc000022-[cf烟雾头]cf怎么调烟雾头

11-06 qizideyouhuo:应用程序无法正常启动0xc0000

11-06 ipz-185:win7系统vcf文件怎么打开

11-06 傻哥蹦迪:win10系统s4怎么打开usb调试

11-06 八神浩树gtaste:回收站清空了怎么恢复

11-06 妖尾之黑色守护:win10系统电脑没有1440x900

11-06 校园至尊魔王小说:win7系统浏览网页时字体

11-06 女斗合众国:win10系统访问共享文件夹提示请

11-06 tokyo hot n0654:恢复win7系统默认字体一招

11-06 雨酷仙境:设置win7系统转移临时文件夹腾出

11-06 阿穆纳伊之杖:win7系统开始菜单在右边还原

11-06 tunespotting:win10系统火狐flash插件总是

11-06 甘尔葛分析师：计谋网站seo关键词暴涨有什

11-06 蔡贵霖: 计谋网站seo关键词暴涨有什么秘密

11-06 博益网首页:ao3网页版进入不了解决方法

11-06 漏斗子专栏: 网站数据分析小白易懂精华篇

11-06 见证双虹怎么做:win7系统开启telnet命令的

11-06 颾狐蝶蜋:系统资源不足无法完成请求的服务

11-06 国光中学校歌:提交网站到alexa查询详细步骤

11-06 西安有情天:静态网页和动态网页的区别

11-06 红木雅尚斋:外部链接构造对网站的好处

11-06 前官礼遇：防止域名劫持–增强域安全性的10

11-06 密传二转答案: 中文分词算法有哪些

11-06 金泉家园邮编:百度快照劫持的表现及应对方

C++ ops_timing_realloc函数代码示例发布时间：2022-05-31

C++ ops_set_dirtybit_host函数代码示例发布时间：2022-05-31

剪的笔顺,诠释剪的笔画,认识剪的部首

1 六六分期app的软件客服如何联系？(六六分期

六六分期app的软件客服如何联系？不知道吗？加qq群【895510560】即可！标题：六六分期

阅读：18169|2023-10-27

2 可心卡盟:win10系统火狐flash插件崩溃怎么

今天小编告诉大家如何处理win10系统火狐flash插件总是崩溃的问题，可能很多用户都不知

阅读：9640|2022-11-06

3 亲亲特价:怎么删除回收站图标

今天小编告诉大家如何对win10系统删除桌面回收站图标进行设置，可能很多用户都不知道

阅读：8166|2022-11-06

4 济南大学虚拟社区:鲁大师节能降温的具体办

今天小编告诉大家如何对win10系统电脑设置节能降温的设置方法，想必大家都遇到过需要

阅读：8542|2022-11-06

5 xlueops.exe:无线网络安装向导

我们在使用xp系统的过程中,经常需要对xp系统无线网络安装向导设置进行设置，可能很多

阅读：8444|2022-11-06

6 女斗合众国:win7系统cf与主机连接不稳定怎

今天小编告诉大家如何处理win7系统玩cf老是与主机连接不稳定的问题，可能很多用户都不

阅读：9366|2022-11-06

7 0xc000022-[cf烟雾头]cf怎么调烟雾头

电脑对日常生活的重要性小编就不多说了，可是一旦碰到win7系统设置cf烟雾头的问题，很

阅读：8415|2022-11-06

8 qizideyouhuo:应用程序无法正常启动0xc0000

我们在日常使用电脑的时候，有的小伙伴们可能在打开应用的时候会遇见提示应用程序无法

阅读：7852|2022-11-06

9 ipz-185:win7系统vcf文件怎么打开

今天小编告诉大家如何对win7系统打开vcf文件进行设置，可能很多用户都不知道怎么对win

阅读：8398|2022-11-06

10 傻哥蹦迪:win10系统s4怎么打开usb调试

今天小编告诉大家如何对win10系统s4开启USB调试模式进行设置，可能很多用户都不知道怎

阅读：7390|2022-11-06

客服电话

电子邮件

C++ ops_timers_core函数代码示例

示例1: ops_par_loop_update_halo_kernel2_yvel_minus_4_top

示例2: ops_par_loop_update_halo_kernel3_minus_2_a

示例3: ops_par_loop_advec_cell_kernel4_xdir

示例4: ops_par_loop_update_halo_kernel2_zvel_minus_2_back

示例5: ops_par_loop_update_halo_kernel2_yvel_plus_2_right

示例6: ops_par_loop_advec_mom_kernel_post_pre_advec_y_execute

示例7: ops_par_loop_advec_mom_kernel_post_pre_advec_y

示例8: ops_par_loop_update_halo_kernel5_minus_4_back_execute

示例9: ops_par_loop_accelerate_kernel_execute

示例10: ops_par_loop_reset_field_kernel1

示例11: ops_par_loop_update_halo_kernel2_xvel_minus_4_right

示例12: ops_par_loop_tea_leaf_common_init_kernel

示例13: ops_timers

示例14: main

请发表评论

全部评论

上一篇：

下一篇：

CVE-2022-31513

librespeed/speedtest: Self-hosted Speedt

ozzieperez/packtpub-library-downloader:

avehtari/BDA_m_demos: Bayesian Data Anal

四维彩超怎么看性别？四维看男孩女孩诀窍

剪的笔顺,诠释剪的笔画,认识剪的部首

六六分期app的软件客服如何联系？(六六分期

florent37/ViewAnimator: A fluent Android

florent37/Shrine-MaterialDesign2: implem

CVE-2020-36276

SimpleSoftwareIO/simple-sms: Send and re

关于我们

产品与服务

解决方案

139-2527-9053