void poisson_kernel_populate_c_wrapper( int p_a0, int p_a1, int *p_a2, double *p_a3, double *p_a4, double *p_a5, int arg_idx0, int arg_idx1, int x_size, int y_size) { #ifdef OPS_GPU #pragma acc parallel deviceptr(p_a3,p_a4,p_a5) #pragma acc loop #endif for ( int n_y=0; n_y<y_size; n_y++ ){ #ifdef OPS_GPU #pragma acc loop #endif for ( int n_x=0; n_x<x_size; n_x++ ){ int arg_idx[] = {arg_idx0+n_x, arg_idx1+n_y}; poisson_kernel_populate( &p_a0, &p_a1,arg_idx, p_a3 + n_x*1 + n_y*xdim3_poisson_kernel_populate*1, p_a4 + n_x*1 + n_y*xdim4_poisson_kernel_populate*1, p_a5 + n_x*1 + n_y*xdim5_poisson_kernel_populate*1 ); } } }
// host stub function void ops_par_loop_poisson_kernel_populate(char const *name, ops_block block, int dim, int *range, ops_arg arg0, ops_arg arg1, ops_arg arg2, ops_arg arg3, ops_arg arg4, ops_arg arg5) { // Timing double t1, t2, c1, c2; int offs[6][2]; ops_arg args[6] = {arg0, arg1, arg2, arg3, arg4, arg5}; #ifdef CHECKPOINTING if (!ops_checkpointing_before(args, 6, range, 0)) return; #endif if (OPS_diags > 1) { ops_timing_realloc(0, "poisson_kernel_populate"); OPS_kernels[0].count++; ops_timers_core(&c1, &t1); } #ifdef OPS_MPI sub_block_list sb = OPS_sub_block_list[block->index]; #endif // compute locally allocated range for the sub-block int start[2]; int end[2]; int arg_idx[2]; #ifdef OPS_MPI if (!sb->owned) return; for (int n = 0; n < 2; n++) { start[n] = sb->decomp_disp[n]; end[n] = sb->decomp_disp[n] + sb->decomp_size[n]; if (start[n] >= range[2 * n]) { start[n] = 0; } else { start[n] = range[2 * n] - start[n]; } if (sb->id_m[n] == MPI_PROC_NULL && range[2 * n] < 0) start[n] = range[2 * n]; if (end[n] >= range[2 * n + 1]) { end[n] = range[2 * n + 1] - sb->decomp_disp[n]; } else { end[n] = sb->decomp_size[n]; } if (sb->id_p[n] == MPI_PROC_NULL && (range[2 * n + 1] > sb->decomp_disp[n] + sb->decomp_size[n])) end[n] += (range[2 * n + 1] - sb->decomp_disp[n] - sb->decomp_size[n]); if (end[n] < start[n]) end[n] = start[n]; } #else for (int n = 0; n < 2; n++) { start[n] = range[2 * n]; end[n] = range[2 * n + 1]; } #endif #ifdef OPS_DEBUG ops_register_args(args, "poisson_kernel_populate"); #endif offs[3][0] = args[3].stencil->stride[0] * 1; // unit step in x dimension offs[3][1] = off2D(1, &start[0], &end[0], args[3].dat->size, args[3].stencil->stride) - offs[3][0]; offs[4][0] = args[4].stencil->stride[0] * 1; // unit step in x dimension offs[4][1] = off2D(1, &start[0], &end[0], args[4].dat->size, args[4].stencil->stride) - offs[4][0]; offs[5][0] = args[5].stencil->stride[0] * 1; // unit step in x dimension offs[5][1] = off2D(1, &start[0], &end[0], args[5].dat->size, args[5].stencil->stride) - offs[5][0]; int off3_0 = offs[3][0]; int off3_1 = offs[3][1]; int dat3 = (OPS_soa ? args[3].dat->type_size : args[3].dat->elem_size); int off4_0 = offs[4][0]; int off4_1 = offs[4][1]; int dat4 = (OPS_soa ? args[4].dat->type_size : args[4].dat->elem_size); int off5_0 = offs[5][0]; int off5_1 = offs[5][1]; int dat5 = (OPS_soa ? args[5].dat->type_size : args[5].dat->elem_size); // Halo Exchanges ops_H_D_exchanges_host(args, 6); ops_halo_exchanges(args, 6, range); ops_H_D_exchanges_host(args, 6); #ifdef _OPENMP int nthreads = omp_get_max_threads(); #else int nthreads = 1; #endif xdim3 = args[3].dat->size[0]; xdim4 = args[4].dat->size[0]; xdim5 = args[5].dat->size[0]; if (OPS_diags > 1) { ops_timers_core(&c2, &t2); OPS_kernels[0].mpi_time += t2 - t1; } #pragma omp parallel for for (int thr = 0; thr < nthreads; thr++) { int y_size = end[1] - start[1]; char *p_a[6]; int start_i = start[1] + ((y_size - 1) / nthreads + 1) * thr; int finish_i = start[1] + MIN(((y_size - 1) / nthreads + 1) * (thr + 1), y_size); // get address per thread int start0 = start[0]; int start1 = start_i; int arg_idx[2]; #ifdef OPS_MPI arg_idx[0] = sb->decomp_disp[0] + start0; arg_idx[1] = sb->decomp_disp[1] + start1; #else arg_idx[0] = start0; arg_idx[1] = start1; #endif // set up initial pointers int d_m[OPS_MAX_DIM]; p_a[0] = (char *)args[0].data; p_a[1] = (char *)args[1].data; p_a[2] = (char *)arg_idx; #ifdef OPS_MPI for (int d = 0; d < dim; d++) d_m[d] = args[3].dat->d_m[d] + OPS_sub_dat_list[args[3].dat->index]->d_im[d]; #else for (int d = 0; d < dim; d++) d_m[d] = args[3].dat->d_m[d]; #endif int base3 = dat3 * 1 * (start0 * args[3].stencil->stride[0] - args[3].dat->base[0] - d_m[0]); base3 = base3 + dat3 * args[3].dat->size[0] * (start1 * args[3].stencil->stride[1] - args[3].dat->base[1] - d_m[1]); p_a[3] = (char *)args[3].data + base3; #ifdef OPS_MPI for (int d = 0; d < dim; d++) d_m[d] = args[4].dat->d_m[d] + OPS_sub_dat_list[args[4].dat->index]->d_im[d]; #else for (int d = 0; d < dim; d++) d_m[d] = args[4].dat->d_m[d]; #endif int base4 = dat4 * 1 * (start0 * args[4].stencil->stride[0] - args[4].dat->base[0] - d_m[0]); base4 = base4 + dat4 * args[4].dat->size[0] * (start1 * args[4].stencil->stride[1] - args[4].dat->base[1] - d_m[1]); p_a[4] = (char *)args[4].data + base4; #ifdef OPS_MPI for (int d = 0; d < dim; d++) d_m[d] = args[5].dat->d_m[d] + OPS_sub_dat_list[args[5].dat->index]->d_im[d]; #else for (int d = 0; d < dim; d++) d_m[d] = args[5].dat->d_m[d]; #endif int base5 = dat5 * 1 * (start0 * args[5].stencil->stride[0] - args[5].dat->base[0] - d_m[0]); base5 = base5 + dat5 * args[5].dat->size[0] * (start1 * args[5].stencil->stride[1] - args[5].dat->base[1] - d_m[1]); p_a[5] = (char *)args[5].data + base5; for (int n_y = start_i; n_y < finish_i; n_y++) { for (int n_x = start[0]; n_x < start[0] + (end[0] - start[0]) / SIMD_VEC; n_x++) { // call kernel function, passing in pointers to data -vectorised for (int i = 0; i < SIMD_VEC; i++) { poisson_kernel_populate((int *)p_a[0], (int *)p_a[1], arg_idx, (double *)p_a[3] + i * 1 * 1, (double *)p_a[4] + i * 1 * 1, (double *)p_a[5] + i * 1 * 1); arg_idx[0]++; } // shift pointers to data x direction p_a[3] = p_a[3] + (dat3 * off3_0) * SIMD_VEC; p_a[4] = p_a[4] + (dat4 * off4_0) * SIMD_VEC; p_a[5] = p_a[5] + (dat5 * off5_0) * SIMD_VEC; } for (int n_x = start[0] + ((end[0] - start[0]) / SIMD_VEC) * SIMD_VEC; n_x < end[0]; n_x++) { // call kernel function, passing in pointers to data - remainder poisson_kernel_populate((int *)p_a[0], (int *)p_a[1], arg_idx, (double *)p_a[3], (double *)p_a[4], (double *)p_a[5]); // shift pointers to data x direction p_a[3] = p_a[3] + (dat3 * off3_0); p_a[4] = p_a[4] + (dat4 * off4_0); p_a[5] = p_a[5] + (dat5 * off5_0); arg_idx[0]++; } // shift pointers to data y direction p_a[3] = p_a[3] + (dat3 * off3_1); p_a[4] = p_a[4] + (dat4 * off4_1); p_a[5] = p_a[5] + (dat5 * off5_1); #ifdef OPS_MPI arg_idx[0] = sb->decomp_disp[0] + start0; #else arg_idx[0] = start0; #endif arg_idx[1]++; } } if (OPS_diags > 1) { ops_timers_core(&c1, &t1); OPS_kernels[0].time += t1 - t2; } ops_set_dirtybit_host(args, 6); ops_set_halo_dirtybit3(&args[3], range); ops_set_halo_dirtybit3(&args[4], range); ops_set_halo_dirtybit3(&args[5], range); if (OPS_diags > 1) { // Update kernel record ops_timers_core(&c2, &t2); OPS_kernels[0].mpi_time += t2 - t1; OPS_kernels[0].transfer += ops_compute_transfer(dim, start, end, &arg3); OPS_kernels[0].transfer += ops_compute_transfer(dim, start, end, &arg4); OPS_kernels[0].transfer += ops_compute_transfer(dim, start, end, &arg5); } }