1 // Copyright (c) 2017-2024, Lawrence Livermore National Security, LLC and other CEED contributors. 2 // All Rights Reserved. See the top-level LICENSE and NOTICE files for details. 3 // 4 // SPDX-License-Identifier: BSD-2-Clause 5 // 6 // This file is part of CEED: http://github.com/ceed 7 8 /// @file 9 /// Internal header for CUDA offset element restriction kernels 10 11 #include <ceed.h> 12 13 //------------------------------------------------------------------------------ 14 // E-vector -> L-vector, standard (with offsets) 15 //------------------------------------------------------------------------------ 16 #if !USE_DETERMINISTIC 17 extern "C" __global__ void AtPointsTranspose(const CeedInt *__restrict__ indices, const CeedInt *__restrict__ points_per_elem, 18 const CeedScalar *__restrict__ u, CeedScalar *__restrict__ v) { 19 for (CeedInt node = blockIdx.x * blockDim.x + threadIdx.x; node < RSTR_NUM_ELEM * RSTR_ELEM_SIZE; node += blockDim.x * gridDim.x) { 20 const CeedInt ind = indices[node]; 21 const CeedInt loc_node = node % RSTR_ELEM_SIZE; 22 const CeedInt elem = node / RSTR_ELEM_SIZE; 23 24 if (loc_node >= points_per_elem[elem]) continue; 25 for (CeedInt comp = 0; comp < RSTR_NUM_COMP; comp++) { 26 atomicAdd(v + ind + comp * RSTR_COMP_STRIDE, u[loc_node + comp * RSTR_ELEM_SIZE * RSTR_NUM_ELEM + elem * RSTR_ELEM_SIZE]); 27 } 28 } 29 } 30 #else 31 extern "C" __global__ void AtPointsTranspose(const CeedInt *__restrict__ l_vec_indices, const CeedInt *__restrict__ t_indices, 32 const CeedInt *__restrict__ points_per_elem, const CeedInt *__restrict__ t_offsets, 33 const CeedScalar *__restrict__ u, CeedScalar *__restrict__ v) { 34 CeedScalar value[RSTR_NUM_COMP]; 35 36 for (CeedInt i = blockIdx.x * blockDim.x + threadIdx.x; i < RSTR_NUM_NODES; i += blockDim.x * gridDim.x) { 37 const CeedInt ind = l_vec_indices[i]; 38 const CeedInt range_1 = t_offsets[i]; 39 const CeedInt range_N = t_offsets[i + 1]; 40 41 for (CeedInt comp = 0; comp < RSTR_NUM_COMP; comp++) value[comp] = 0.0; 42 43 for (CeedInt j = range_1; j < range_N; j++) { 44 const CeedInt t_ind = t_indices[j]; 45 const CeedInt loc_node = t_ind % RSTR_ELEM_SIZE; 46 const CeedInt elem = t_ind / RSTR_ELEM_SIZE; 47 48 if (loc_node >= points_per_elem[elem]) continue; 49 for (CeedInt comp = 0; comp < RSTR_NUM_COMP; comp++) { 50 value[comp] += u[loc_node + comp * RSTR_ELEM_SIZE * RSTR_NUM_ELEM + elem * RSTR_ELEM_SIZE]; 51 } 52 } 53 54 for (CeedInt comp = 0; comp < RSTR_NUM_COMP; comp++) v[ind + comp * RSTR_COMP_STRIDE] += value[comp]; 55 } 56 } 57 #endif 58