1 // Copyright (c) 2017-2025, 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 tensor product basis with AtPoints evaluation 10 #include <ceed/types.h> 11 12 #include "cuda-shared-basis-read-write-templates.h" 13 #include "cuda-shared-basis-tensor-at-points-templates.h" 14 #include "cuda-shared-basis-tensor-templates.h" 15 16 //------------------------------------------------------------------------------ 17 // Tensor Basis Kernels AtPoints 18 //------------------------------------------------------------------------------ 19 20 //------------------------------------------------------------------------------ 21 // Interp 22 //------------------------------------------------------------------------------ 23 extern "C" __global__ void InterpAtPoints(const CeedInt num_elem, const CeedScalar *__restrict__ c_B, const CeedInt *__restrict__ points_per_elem, 24 const CeedScalar *__restrict__ d_X, const CeedScalar *__restrict__ d_U, CeedScalar *__restrict__ d_V) { 25 extern __shared__ CeedScalar slice[]; 26 27 SharedData_Cuda data; 28 data.t_id_x = threadIdx.x; 29 data.t_id_y = threadIdx.y; 30 data.t_id_z = threadIdx.z; 31 data.t_id = threadIdx.x + threadIdx.y * blockDim.x + threadIdx.z * blockDim.y * blockDim.x; 32 data.slice = slice + data.t_id_z * BASIS_T_1D * (BASIS_DIM > 1 ? BASIS_T_1D : 1); 33 34 CeedScalar r_X[BASIS_DIM]; 35 CeedScalar r_U[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_P_1D : 1)]; 36 CeedScalar r_C[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1)]; 37 CeedScalar r_V[BASIS_NUM_COMP]; 38 39 // load interp_1d into shared memory 40 __shared__ CeedScalar s_B[BASIS_P_1D * BASIS_Q_1D]; 41 LoadMatrix<BASIS_P_1D, BASIS_Q_1D>(data, c_B, s_B); 42 __syncthreads(); 43 44 // Apply basis element by element 45 for (CeedInt elem = blockIdx.x * blockDim.z + threadIdx.z; elem < num_elem; elem += gridDim.x * blockDim.z) { 46 // Map to coefficients 47 if (BASIS_DIM == 1) { 48 ReadElementStrided1d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * num_elem, BASIS_P_1D, d_U, r_U); 49 Interp1d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_U, s_B, r_C); 50 } else if (BASIS_DIM == 2) { 51 ReadElementStrided2d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * num_elem, BASIS_P_1D * BASIS_P_1D, d_U, r_U); 52 InterpTensor2d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_U, s_B, r_C); 53 } else if (BASIS_DIM == 3) { 54 ReadElementStrided3d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * BASIS_P_1D * num_elem, 55 BASIS_P_1D * BASIS_P_1D * BASIS_P_1D, d_U, r_U); 56 InterpTensor3d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_U, s_B, r_C); 57 } 58 59 // Map to points 60 const CeedInt point_loop_bound = (blockDim.x * blockDim.y) * ceil(1.0 * BASIS_NUM_PTS / (blockDim.x * blockDim.y)); 61 62 for (CeedInt i = threadIdx.x + threadIdx.y * blockDim.x; i < point_loop_bound; i += blockDim.x * blockDim.y) { 63 const CeedInt p = i % BASIS_NUM_PTS; 64 65 ReadPoint<BASIS_DIM, BASIS_NUM_PTS>(data, elem, p, BASIS_NUM_PTS, 1, num_elem * BASIS_NUM_PTS, BASIS_NUM_PTS, d_X, r_X); 66 if (BASIS_DIM == 1) { 67 InterpAtPoints1d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_C, r_X, r_V); 68 } else if (BASIS_DIM == 2) { 69 InterpAtPoints2d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_C, r_X, r_V); 70 } else if (BASIS_DIM == 3) { 71 InterpAtPoints3d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_C, r_X, r_V); 72 } 73 WritePoint<BASIS_NUM_COMP, BASIS_NUM_PTS>(data, elem, p, BASIS_NUM_PTS, 1, num_elem * BASIS_NUM_PTS, BASIS_NUM_PTS, r_V, d_V); 74 } 75 } 76 } 77 78 extern "C" __global__ void InterpTransposeAtPoints(const CeedInt num_elem, const CeedScalar *__restrict__ c_B, 79 const CeedInt *__restrict__ points_per_elem, const CeedScalar *__restrict__ d_X, 80 const CeedScalar *__restrict__ d_U, CeedScalar *__restrict__ d_V) { 81 extern __shared__ CeedScalar slice[]; 82 83 SharedData_Cuda data; 84 data.t_id_x = threadIdx.x; 85 data.t_id_y = threadIdx.y; 86 data.t_id_z = threadIdx.z; 87 data.t_id = threadIdx.x + threadIdx.y * blockDim.x + threadIdx.z * blockDim.y * blockDim.x; 88 data.slice = slice + data.t_id_z * BASIS_T_1D * (BASIS_DIM > 1 ? BASIS_T_1D : 1); 89 90 CeedScalar r_X[BASIS_DIM]; 91 CeedScalar r_U[BASIS_NUM_COMP]; 92 CeedScalar r_C[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1)]; 93 CeedScalar r_V[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1)]; 94 95 // load interp_1d into shared memory 96 __shared__ CeedScalar s_B[BASIS_P_1D * BASIS_Q_1D]; 97 LoadMatrix<BASIS_P_1D, BASIS_Q_1D>(data, c_B, s_B); 98 __syncthreads(); 99 100 // Apply basis element by element 101 for (CeedInt elem = blockIdx.x * blockDim.z + threadIdx.z; elem < num_elem; elem += gridDim.x * blockDim.z) { 102 // Clear register 103 for (CeedInt i = 0; i < BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1); i++) r_C[i] = 0.0; 104 105 // Clear output vector 106 for (CeedInt i = 0; i < BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1); i++) r_V[i] = 0.0; 107 if (BASIS_DIM == 1) { 108 WriteElementStrided1d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * num_elem, BASIS_P_1D, r_V, d_V); 109 } else if (BASIS_DIM == 2) { 110 WriteElementStrided2d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * num_elem, BASIS_P_1D * BASIS_P_1D, r_V, d_V); 111 } else if (BASIS_DIM == 3) { 112 WriteElementStrided3d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * BASIS_P_1D * num_elem, 113 BASIS_P_1D * BASIS_P_1D * BASIS_P_1D, r_V, d_V); 114 } 115 116 // Map from points 117 const CeedInt point_loop_bound = (blockDim.x * blockDim.y) * ceil(1.0 * BASIS_NUM_PTS / (blockDim.x * blockDim.y)); 118 119 for (CeedInt i = threadIdx.x + threadIdx.y * blockDim.x; i < point_loop_bound; i += blockDim.x * blockDim.y) { 120 const CeedInt p = i % BASIS_NUM_PTS; 121 122 ReadPoint<BASIS_DIM, BASIS_NUM_PTS>(data, elem, p, BASIS_NUM_PTS, 1, num_elem * BASIS_NUM_PTS, BASIS_NUM_PTS, d_X, r_X); 123 ReadPoint<BASIS_NUM_COMP, BASIS_NUM_PTS>(data, elem, i, points_per_elem[elem], 1, num_elem * BASIS_NUM_PTS, BASIS_NUM_PTS, d_U, r_U); 124 if (BASIS_DIM == 1) { 125 InterpTransposeAtPoints1d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_U, r_X, r_C); 126 } else if (BASIS_DIM == 2) { 127 InterpTransposeAtPoints2d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_U, r_X, r_C); 128 } else if (BASIS_DIM == 3) { 129 InterpTransposeAtPoints3d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_U, r_X, r_C); 130 } 131 } 132 __syncthreads(); 133 134 // Map from coefficients 135 if (BASIS_DIM == 1) { 136 InterpTranspose1d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_C, s_B, r_V); 137 SumElementStrided1d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * num_elem, BASIS_P_1D, r_V, d_V); 138 } else if (BASIS_DIM == 2) { 139 InterpTransposeTensor2d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_C, s_B, r_V); 140 SumElementStrided2d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * num_elem, BASIS_P_1D * BASIS_P_1D, r_V, d_V); 141 } else if (BASIS_DIM == 3) { 142 InterpTransposeTensor3d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_C, s_B, r_V); 143 SumElementStrided3d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * BASIS_P_1D * num_elem, 144 BASIS_P_1D * BASIS_P_1D * BASIS_P_1D, r_V, d_V); 145 } 146 } 147 } 148 149 extern "C" __global__ void InterpTransposeAddAtPoints(const CeedInt num_elem, const CeedScalar *__restrict__ c_B, 150 const CeedInt *__restrict__ points_per_elem, const CeedScalar *__restrict__ d_X, 151 const CeedScalar *__restrict__ d_U, CeedScalar *__restrict__ d_V) { 152 extern __shared__ CeedScalar slice[]; 153 154 SharedData_Cuda data; 155 data.t_id_x = threadIdx.x; 156 data.t_id_y = threadIdx.y; 157 data.t_id_z = threadIdx.z; 158 data.t_id = threadIdx.x + threadIdx.y * blockDim.x + threadIdx.z * blockDim.y * blockDim.x; 159 data.slice = slice + data.t_id_z * BASIS_T_1D * (BASIS_DIM > 1 ? BASIS_T_1D : 1); 160 161 CeedScalar r_X[BASIS_DIM]; 162 CeedScalar r_U[BASIS_NUM_COMP]; 163 CeedScalar r_C[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1)]; 164 CeedScalar r_V[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1)]; 165 166 // load interp_1d into shared memory 167 __shared__ CeedScalar s_B[BASIS_P_1D * BASIS_Q_1D]; 168 LoadMatrix<BASIS_P_1D, BASIS_Q_1D>(data, c_B, s_B); 169 __syncthreads(); 170 171 // Apply basis element by element 172 for (CeedInt elem = blockIdx.x * blockDim.z + threadIdx.z; elem < num_elem; elem += gridDim.x * blockDim.z) { 173 // Clear register 174 for (CeedInt i = 0; i < BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1); i++) r_C[i] = 0.0; 175 176 // Map from points 177 const CeedInt point_loop_bound = (blockDim.x * blockDim.y) * ceil(1.0 * BASIS_NUM_PTS / (blockDim.x * blockDim.y)); 178 179 for (CeedInt i = threadIdx.x + threadIdx.y * blockDim.x; i < point_loop_bound; i += blockDim.x * blockDim.y) { 180 const CeedInt p = i % BASIS_NUM_PTS; 181 182 ReadPoint<BASIS_DIM, BASIS_NUM_PTS>(data, elem, p, BASIS_NUM_PTS, 1, num_elem * BASIS_NUM_PTS, BASIS_NUM_PTS, d_X, r_X); 183 ReadPoint<BASIS_NUM_COMP, BASIS_NUM_PTS>(data, elem, i, points_per_elem[elem], 1, num_elem * BASIS_NUM_PTS, BASIS_NUM_PTS, d_U, r_U); 184 if (BASIS_DIM == 1) { 185 InterpTransposeAtPoints1d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_U, r_X, r_C); 186 } else if (BASIS_DIM == 2) { 187 InterpTransposeAtPoints2d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_U, r_X, r_C); 188 } else if (BASIS_DIM == 3) { 189 InterpTransposeAtPoints3d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_U, r_X, r_C); 190 } 191 } 192 __syncthreads(); 193 194 // Map from coefficients 195 if (BASIS_DIM == 1) { 196 InterpTranspose1d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_C, s_B, r_V); 197 SumElementStrided1d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * num_elem, BASIS_P_1D, r_V, d_V); 198 } else if (BASIS_DIM == 2) { 199 InterpTransposeTensor2d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_C, s_B, r_V); 200 SumElementStrided2d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * num_elem, BASIS_P_1D * BASIS_P_1D, r_V, d_V); 201 } else if (BASIS_DIM == 3) { 202 InterpTransposeTensor3d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_C, s_B, r_V); 203 SumElementStrided3d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * BASIS_P_1D * num_elem, 204 BASIS_P_1D * BASIS_P_1D * BASIS_P_1D, r_V, d_V); 205 } 206 } 207 } 208 209 //------------------------------------------------------------------------------ 210 // Grad 211 //------------------------------------------------------------------------------ 212 extern "C" __global__ void GradAtPoints(const CeedInt num_elem, const CeedScalar *__restrict__ c_B, const CeedInt *__restrict__ points_per_elem, 213 const CeedScalar *__restrict__ d_X, const CeedScalar *__restrict__ d_U, CeedScalar *__restrict__ d_V) { 214 extern __shared__ CeedScalar slice[]; 215 216 SharedData_Cuda data; 217 data.t_id_x = threadIdx.x; 218 data.t_id_y = threadIdx.y; 219 data.t_id_z = threadIdx.z; 220 data.t_id = threadIdx.x + threadIdx.y * blockDim.x + threadIdx.z * blockDim.y * blockDim.x; 221 data.slice = slice + data.t_id_z * BASIS_T_1D * (BASIS_DIM > 1 ? BASIS_T_1D : 1); 222 223 CeedScalar r_X[BASIS_DIM]; 224 CeedScalar r_U[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_P_1D : 1)]; 225 CeedScalar r_C[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1)]; 226 CeedScalar r_V[BASIS_NUM_COMP * BASIS_DIM]; 227 228 // load interp_1d into shared memory 229 __shared__ CeedScalar s_B[BASIS_P_1D * BASIS_Q_1D]; 230 LoadMatrix<BASIS_P_1D, BASIS_Q_1D>(data, c_B, s_B); 231 __syncthreads(); 232 233 // Apply basis element by element 234 for (CeedInt elem = blockIdx.x * blockDim.z + threadIdx.z; elem < num_elem; elem += gridDim.x * blockDim.z) { 235 // Map to coefficients 236 if (BASIS_DIM == 1) { 237 ReadElementStrided1d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * num_elem, BASIS_P_1D, d_U, r_U); 238 Interp1d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_U, s_B, r_C); 239 } else if (BASIS_DIM == 2) { 240 ReadElementStrided2d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * num_elem, BASIS_P_1D * BASIS_P_1D, d_U, r_U); 241 InterpTensor2d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_U, s_B, r_C); 242 } else if (BASIS_DIM == 3) { 243 ReadElementStrided3d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * BASIS_P_1D * num_elem, 244 BASIS_P_1D * BASIS_P_1D * BASIS_P_1D, d_U, r_U); 245 InterpTensor3d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_U, s_B, r_C); 246 } 247 248 // Map to points 249 const CeedInt point_loop_bound = (blockDim.x * blockDim.y) * ceil(1.0 * BASIS_NUM_PTS / (blockDim.x * blockDim.y)); 250 251 for (CeedInt i = threadIdx.x + threadIdx.y * blockDim.x; i < point_loop_bound; i += blockDim.x * blockDim.y) { 252 const CeedInt p = i % BASIS_NUM_PTS; 253 254 ReadPoint<BASIS_DIM, BASIS_NUM_PTS>(data, elem, p, BASIS_NUM_PTS, 1, num_elem * BASIS_NUM_PTS, BASIS_NUM_PTS, d_X, r_X); 255 if (BASIS_DIM == 1) { 256 GradAtPoints1d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_C, r_X, r_V); 257 } else if (BASIS_DIM == 2) { 258 GradAtPoints2d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_C, r_X, r_V); 259 } else if (BASIS_DIM == 3) { 260 GradAtPoints3d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_C, r_X, r_V); 261 } 262 WritePoint<BASIS_NUM_COMP * BASIS_DIM, BASIS_NUM_PTS>(data, elem, p, BASIS_NUM_PTS, 1, num_elem * BASIS_NUM_PTS, BASIS_NUM_PTS, r_V, d_V); 263 } 264 } 265 } 266 267 extern "C" __global__ void GradTransposeAtPoints(const CeedInt num_elem, const CeedScalar *__restrict__ c_B, 268 const CeedInt *__restrict__ points_per_elem, const CeedScalar *__restrict__ d_X, 269 const CeedScalar *__restrict__ d_U, CeedScalar *__restrict__ d_V) { 270 extern __shared__ CeedScalar slice[]; 271 272 SharedData_Cuda data; 273 data.t_id_x = threadIdx.x; 274 data.t_id_y = threadIdx.y; 275 data.t_id_z = threadIdx.z; 276 data.t_id = threadIdx.x + threadIdx.y * blockDim.x + threadIdx.z * blockDim.y * blockDim.x; 277 data.slice = slice + data.t_id_z * BASIS_T_1D * (BASIS_DIM > 1 ? BASIS_T_1D : 1); 278 279 CeedScalar r_X[BASIS_DIM]; 280 CeedScalar r_U[BASIS_NUM_COMP * BASIS_DIM]; 281 CeedScalar r_C[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1)]; 282 CeedScalar r_V[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1)]; 283 284 // load interp_1d into shared memory 285 __shared__ CeedScalar s_B[BASIS_P_1D * BASIS_Q_1D]; 286 LoadMatrix<BASIS_P_1D, BASIS_Q_1D>(data, c_B, s_B); 287 __syncthreads(); 288 289 // Apply basis element by element 290 for (CeedInt elem = blockIdx.x * blockDim.z + threadIdx.z; elem < num_elem; elem += gridDim.x * blockDim.z) { 291 // Clear register 292 for (CeedInt i = 0; i < BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1); i++) r_C[i] = 0.0; 293 294 // Clear output vector 295 for (CeedInt i = 0; i < BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1); i++) r_V[i] = 0.0; 296 if (BASIS_DIM == 1) { 297 WriteElementStrided1d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * num_elem, BASIS_P_1D, r_V, d_V); 298 } else if (BASIS_DIM == 2) { 299 WriteElementStrided2d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * num_elem, BASIS_P_1D * BASIS_P_1D, r_V, d_V); 300 } else if (BASIS_DIM == 3) { 301 WriteElementStrided3d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * BASIS_P_1D * num_elem, 302 BASIS_P_1D * BASIS_P_1D * BASIS_P_1D, r_V, d_V); 303 } 304 305 // Map from points 306 const CeedInt point_loop_bound = (blockDim.x * blockDim.y) * ceil(1.0 * BASIS_NUM_PTS / (blockDim.x * blockDim.y)); 307 308 for (CeedInt i = threadIdx.x + threadIdx.y * blockDim.x; i < point_loop_bound; i += blockDim.x * blockDim.y) { 309 const CeedInt p = i % BASIS_NUM_PTS; 310 311 ReadPoint<BASIS_DIM, BASIS_NUM_PTS>(data, elem, p, BASIS_NUM_PTS, 1, num_elem * BASIS_NUM_PTS, BASIS_NUM_PTS, d_X, r_X); 312 ReadPoint<BASIS_NUM_COMP * BASIS_DIM, BASIS_NUM_PTS>(data, elem, i, points_per_elem[elem], 1, num_elem * BASIS_NUM_PTS, BASIS_NUM_PTS, d_U, 313 r_U); 314 if (BASIS_DIM == 1) { 315 GradTransposeAtPoints1d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_U, r_X, r_C); 316 } else if (BASIS_DIM == 2) { 317 GradTransposeAtPoints2d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_U, r_X, r_C); 318 } else if (BASIS_DIM == 3) { 319 GradTransposeAtPoints3d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_U, r_X, r_C); 320 } 321 } 322 __syncthreads(); 323 324 // Map from coefficients 325 if (BASIS_DIM == 1) { 326 InterpTranspose1d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_C, s_B, r_V); 327 SumElementStrided1d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * num_elem, BASIS_P_1D, r_V, d_V); 328 } else if (BASIS_DIM == 2) { 329 InterpTransposeTensor2d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_C, s_B, r_V); 330 SumElementStrided2d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * num_elem, BASIS_P_1D * BASIS_P_1D, r_V, d_V); 331 } else if (BASIS_DIM == 3) { 332 InterpTransposeTensor3d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_C, s_B, r_V); 333 SumElementStrided3d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * BASIS_P_1D * num_elem, 334 BASIS_P_1D * BASIS_P_1D * BASIS_P_1D, r_V, d_V); 335 } 336 } 337 } 338 339 extern "C" __global__ void GradTransposeAddAtPoints(const CeedInt num_elem, const CeedScalar *__restrict__ c_B, 340 const CeedInt *__restrict__ points_per_elem, const CeedScalar *__restrict__ d_X, 341 const CeedScalar *__restrict__ d_U, CeedScalar *__restrict__ d_V) { 342 extern __shared__ CeedScalar slice[]; 343 344 SharedData_Cuda data; 345 data.t_id_x = threadIdx.x; 346 data.t_id_y = threadIdx.y; 347 data.t_id_z = threadIdx.z; 348 data.t_id = threadIdx.x + threadIdx.y * blockDim.x + threadIdx.z * blockDim.y * blockDim.x; 349 data.slice = slice + data.t_id_z * BASIS_T_1D * (BASIS_DIM > 1 ? BASIS_T_1D : 1); 350 351 CeedScalar r_X[BASIS_DIM]; 352 CeedScalar r_U[BASIS_NUM_COMP * BASIS_DIM]; 353 CeedScalar r_C[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1)]; 354 CeedScalar r_V[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1)]; 355 356 // load interp_1d into shared memory 357 __shared__ CeedScalar s_B[BASIS_P_1D * BASIS_Q_1D]; 358 LoadMatrix<BASIS_P_1D, BASIS_Q_1D>(data, c_B, s_B); 359 __syncthreads(); 360 361 // Apply basis element by element 362 for (CeedInt elem = blockIdx.x * blockDim.z + threadIdx.z; elem < num_elem; elem += gridDim.x * blockDim.z) { 363 // Clear register 364 for (CeedInt i = 0; i < BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1); i++) r_C[i] = 0.0; 365 366 // Map from points 367 const CeedInt point_loop_bound = (blockDim.x * blockDim.y) * ceil(1.0 * BASIS_NUM_PTS / (blockDim.x * blockDim.y)); 368 369 for (CeedInt i = threadIdx.x + threadIdx.y * blockDim.x; i < point_loop_bound; i += blockDim.x * blockDim.y) { 370 const CeedInt p = i % BASIS_NUM_PTS; 371 372 ReadPoint<BASIS_DIM, BASIS_NUM_PTS>(data, elem, p, BASIS_NUM_PTS, 1, num_elem * BASIS_NUM_PTS, BASIS_NUM_PTS, d_X, r_X); 373 ReadPoint<BASIS_NUM_COMP * BASIS_DIM, BASIS_NUM_PTS>(data, elem, i, points_per_elem[elem], 1, num_elem * BASIS_NUM_PTS, BASIS_NUM_PTS, d_U, 374 r_U); 375 if (BASIS_DIM == 1) { 376 GradTransposeAtPoints1d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_U, r_X, r_C); 377 } else if (BASIS_DIM == 2) { 378 GradTransposeAtPoints2d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_U, r_X, r_C); 379 } else if (BASIS_DIM == 3) { 380 GradTransposeAtPoints3d<BASIS_NUM_COMP, BASIS_NUM_PTS, BASIS_P_1D, BASIS_Q_1D>(data, i, r_U, r_X, r_C); 381 } 382 } 383 __syncthreads(); 384 385 // Map from coefficients 386 if (BASIS_DIM == 1) { 387 InterpTranspose1d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_C, s_B, r_V); 388 SumElementStrided1d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * num_elem, BASIS_P_1D, r_V, d_V); 389 } else if (BASIS_DIM == 2) { 390 InterpTransposeTensor2d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_C, s_B, r_V); 391 SumElementStrided2d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * num_elem, BASIS_P_1D * BASIS_P_1D, r_V, d_V); 392 } else if (BASIS_DIM == 3) { 393 InterpTransposeTensor3d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D, BASIS_T_1D>(data, r_C, s_B, r_V); 394 SumElementStrided3d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * BASIS_P_1D * num_elem, 395 BASIS_P_1D * BASIS_P_1D * BASIS_P_1D, r_V, d_V); 396 } 397 } 398 } 399