C++ ops_set_dirtybit_host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_set_dirtybit_host函数的典型用法代码示例。如果您正苦于以下问题：C++ ops_set_dirtybit_host函数的具体用法？C++ ops_set_dirtybit_host怎么用？C++ ops_set_dirtybit_host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_set_dirtybit_host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_advec_mom_kernel1_z_nonvector


//.........这里部分代码省略.........
          (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size) *
              args[4].dat->size[0] * args[4].dat->size[1] * start[2] *
              args[4].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a4 = (double *)((char *)args[4].data_d + base4);
#else
  double *p_a4 = (double *)((char *)args[4].data + base4);
#endif

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

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

// Halo Exchanges

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

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

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

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

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[136].mpi_time += t2 - t1;
    OPS_kernels[136].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[136].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[136].transfer += ops_compute_transfer(dim, start, end, &arg2);
    OPS_kernels[136].transfer += ops_compute_transfer(dim, start, end, &arg3);
    OPS_kernels[136].transfer += ops_compute_transfer(dim, start, end, &arg4);
  }
}

开发者ID:gihanmudalige，项目名称:OPS，代码行数:101，代码来源:advec_mom_kernel1_z_nonvector_openacc_kernel.cpp

示例2: ops_par_loop_updateRK3_kernel


//.........这里部分代码省略.........
#else
  double *p_a8 = (double *)((char *)args[8].data + base8);
#endif

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

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

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

// Halo Exchanges

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

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

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

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

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

开发者ID:gihanmudalige，项目名称:OPS，代码行数:101，代码来源:updateRK3_kernel_openacc_kernel.cpp

示例3: ops_par_loop_left_bndcon


//.........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);
}

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

示例4: ops_par_loop_update_halo_kernel2_zvel_minus_4_back


//.........这里部分代码省略.........
  #pragma omp parallel for
  for ( int thr=0; thr<nthreads; thr++ ){

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

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

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

    //set up initial pointers 
    int d_m[OPS_MAX_DIM];
    #ifdef OPS_MPI
    for (int d = 0; d < dim; d++) d_m[d] = args[0].dat->d_m[d] + OPS_sub_dat_list[args[0].dat->index]->d_im[d];
    #else //OPS_MPI
    for (int d = 0; d < dim; d++) d_m[d] = args[0].dat->d_m[d];
    #endif //OPS_MPI
    int base0 = dat0 * 1 * 
    (start0 * args[0].stencil->stride[0] - args[0].dat->base[0] - d_m[0]);
    base0 = base0+ dat0 *
      args[0].dat->size[0] *
      (start1 * args[0].stencil->stride[1] - args[0].dat->base[1] - d_m[1]);
    base0 = base0+ dat0 *
      args[0].dat->size[0] *
      args[0].dat->size[1] *
      (start2 * args[0].stencil->stride[2] - args[0].dat->base[2] - d_m[2]);
    p_a[0] = (char *)args[0].data + base0;

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

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


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

          }

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

        for ( int n_x=start[0]+((end[0]-start[0])/SIMD_VEC)*SIMD_VEC; n_x<end[0]; n_x++ ){
          //call kernel function, passing in pointers to data - remainder
          update_halo_kernel2_zvel_minus_4_back(  (double * )p_a[0], (double * )p_a[1], (int * )p_a[2] );


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

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

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

  ops_set_dirtybit_host(args, 3);

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

  //Update kernel record
  ops_timers_core(&c2,&t2);
  OPS_kernels[85].mpi_time += t2-t1;
  OPS_kernels[85].transfer += ops_compute_transfer(dim, range, &arg0);
  OPS_kernels[85].transfer += ops_compute_transfer(dim, range, &arg1);
}

开发者ID:satyajammy，项目名称:OPS，代码行数:101，代码来源:update_halo_kernel2_zvel_minus_4_back_omp_kernel.cpp

示例5: ops_par_loop_initialise_chunk_kernel_cellz


//.........这里部分代码省略.........
    (start[2] * args[0].stencil->stride[2] - args[0].dat->base[2] - d_m[2]);
  p_a[0] = (char *)args[0].data + base0;

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

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


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

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

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

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

        }

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

      for ( int n_x=start[0]+((end[0]-start[0])/SIMD_VEC)*SIMD_VEC; n_x<end[0]; n_x++ ){
        //call kernel function, passing in pointers to data - remainder
        initialise_chunk_kernel_cellz(  (double *)p_a[0], (double *)p_a[1], (double *)p_a[2] );


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

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

  //Update kernel record
  OPS_kernels[138].transfer += ops_compute_transfer(dim, range, &arg0);
  OPS_kernels[138].transfer += ops_compute_transfer(dim, range, &arg1);
  OPS_kernels[138].transfer += ops_compute_transfer(dim, range, &arg2);
}

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

示例6: ops_par_loop_update_halo_kernel2_zvel_plus_4_right


//.........这里部分代码省略.........
  base0 = base0 +
          (OPS_soa ? args[0].dat->type_size : args[0].dat->elem_size) *
              args[0].dat->size[0] * start[1] * args[0].stencil->stride[1];
  base0 = base0 +
          (OPS_soa ? args[0].dat->type_size : args[0].dat->elem_size) *
              args[0].dat->size[0] * args[0].dat->size[1] * start[2] *
              args[0].stencil->stride[2];
#ifdef OPS_GPU
  double *p_a0 = (double *)((char *)args[0].data_d + base0);
#else
  double *p_a0 = (double *)((char *)args[0].data + base0);
#endif

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

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

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

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

// Halo Exchanges

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

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

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

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

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[53].mpi_time += t2 - t1;
    OPS_kernels[53].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[53].transfer += ops_compute_transfer(dim, start, end, &arg1);
  }
}

开发者ID:gihanmudalige，项目名称:OPS，代码行数:101，代码来源:update_halo_kernel2_zvel_plus_4_right_openacc_kernel.cpp

示例7: ops_par_loop_PdV_kernel_predict


//.........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);
}

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

示例8: ops_par_loop_limiter_kernel


//.........这里部分代码省略.........
  for (int n = 0; n < 1; n++) {
    start[n] = range[2 * n];
    end[n] = range[2 * n + 1];
    arg_idx[n] = start[n];
  }
#endif
  for (int n = 0; n < 1; n++) {
    arg_idx_base[n] = arg_idx[n];
  }

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

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

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

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

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

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

// Halo Exchanges

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

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

  limiter_kernel_c_wrapper(p_a0, p_a1, p_a2, x_size);

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

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[8].mpi_time += t2 - t1;
    OPS_kernels[8].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[8].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[8].transfer += ops_compute_transfer(dim, start, end, &arg2);
  }
}

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

示例9: ops_par_loop_updateRK3_kernel


//.........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);
  }
}

开发者ID:gihanmudalige，项目名称:OPS，代码行数:101，代码来源:updateRK3_kernel_omp_kernel.cpp

示例10: ops_par_loop_field_summary_kernel


//.........这里部分代码省略.........

    p_a[7] = (char *)arg7h;

    p_a[8] = (char *)arg8h;

    p_a[9] = (char *)arg9h;

    p_a[10] = (char *)arg10h;

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

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

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

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

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

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

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

  if (OPS_diags > 1) {
    // Update kernel record
    ops_timers_core(&c2, &t2);
    OPS_kernels[49].mpi_time += t2 - t1;
    OPS_kernels[49].transfer += ops_compute_transfer(dim, start, end, &arg0);
    OPS_kernels[49].transfer += ops_compute_transfer(dim, start, end, &arg1);
    OPS_kernels[49].transfer += ops_compute_transfer(dim, start, end, &arg2);
    OPS_kernels[49].transfer += ops_compute_transfer(dim, start, end, &arg3);
    OPS_kernels[49].transfer += ops_compute_transfer(dim, start, end, &arg4);
    OPS_kernels[49].transfer += ops_compute_transfer(dim, start, end, &arg5);
  }
}

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

示例11: ops_par_loop_update_halo_kernel1_b1


//.........这里部分代码省略.........
          (OPS_soa ? args[6].dat->type_size : args[6].dat->elem_size) *
              args[6].dat->size[0] * start[1] * args[6].stencil->stride[1];
#ifdef OPS_GPU
  double *p_a6 = (double *)((char *)args[6].data_d + base6);
#else
  double *p_a6 = (double *)((char *)args[6].data + base6);
#endif

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

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

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

// Halo Exchanges

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

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

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

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

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

开发者ID:gihanmudalige，项目名称:OPS，代码行数:101，代码来源:update_halo_kernel1_b1_openacc_kernel.cpp

示例12: ops_par_loop_update_halo_kernel1_b2


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

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



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

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

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

  int n_x;
  for ( int n_y=start[1]; n_y<end[1]; n_y++ ){
    #pragma novector
    for( n_x=start[0]; n_x<start[0]+((end[0]-start[0])/SIMD_VEC)*SIMD_VEC; n_x+=SIMD_VEC ) {
      //call kernel function, passing in pointers to data -vectorised
      #pragma simd
      for ( int i=0; i<SIMD_VEC; i++ ){
        update_halo_kernel1_b2(  (double *)p_a[0]+ i*1, (double *)p_a[1]+ i*1, (double *)p_a[2]+ i*1,
           (double *)p_a[3]+ i*1, (double *)p_a[4]+ i*1, (double *)p_a[5]+ i*1, (double *)p_a[6]+ i*1,
           (int *)p_a[7] );

      }

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

    for ( int n_x=start[0]+((end[0]-start[0])/SIMD_VEC)*SIMD_VEC; n_x<end[0]; n_x++ ){
      //call kernel function, passing in pointers to data - remainder
      update_halo_kernel1_b2(  (double *)p_a[0], (double *)p_a[1], (double *)p_a[2],
           (double *)p_a[3], (double *)p_a[4], (double *)p_a[5], (double *)p_a[6],
           (int *)p_a[7] );


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

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

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

开发者ID:satyajammy，项目名称:OPS，代码行数:101，代码来源:update_halo_kernel1_b2_seq_kernel.cpp

示例13: ops_par_loop_tea_leaf_cg_calc_w_reduce_kernel


//.........这里部分代码省略.........
                  start[0] * args[2].stencil->stride[0];
  base2 = base2 +
          (OPS_soa ? args[2].dat->type_size : args[2].dat->elem_size) *
              args[2].dat->size[0] * start[1] * args[2].stencil->stride[1];
  p_a[2] = (char *)args[2].data + base2;

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

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_timers_core函数代码示例发布时间：2022-05-31

C++ op_timers_core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_set_dirtybit_host函数代码示例

示例1: ops_par_loop_advec_mom_kernel1_z_nonvector

示例2: ops_par_loop_updateRK3_kernel

示例3: ops_par_loop_left_bndcon

示例4: ops_par_loop_update_halo_kernel2_zvel_minus_4_back

示例5: ops_par_loop_initialise_chunk_kernel_cellz

示例6: ops_par_loop_update_halo_kernel2_zvel_plus_4_right

示例7: ops_par_loop_PdV_kernel_predict

示例8: ops_par_loop_limiter_kernel

示例9: ops_par_loop_updateRK3_kernel

示例10: ops_par_loop_field_summary_kernel

示例11: ops_par_loop_update_halo_kernel1_b1

示例12: ops_par_loop_update_halo_kernel1_b2

示例13: ops_par_loop_tea_leaf_cg_calc_w_reduce_kernel

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