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 #define CEED_DEBUG_COLOR 12 9 10 #include <ceed.h> 11 #include <ceed/backend.h> 12 #include <ceed/jit-tools.h> 13 #include <cuda_runtime.h> 14 15 #include <iostream> 16 #include <sstream> 17 #include <string> 18 19 #include "../cuda-ref/ceed-cuda-ref.h" 20 #include "../cuda-shared/ceed-cuda-shared.h" 21 #include "../cuda/ceed-cuda-common.h" 22 #include "../cuda/ceed-cuda-compile.h" 23 #include "ceed-cuda-gen.h" 24 25 struct FieldReuse_Cuda { 26 CeedInt index; 27 bool is_input; 28 CeedEvalMode eval_mode; 29 }; 30 31 //------------------------------------------------------------------------------ 32 // Determine type of operator 33 //------------------------------------------------------------------------------ 34 static int CeedOperatorBuildKernelData_Cuda_gen(Ceed ceed, CeedInt num_input_fields, CeedOperatorField *op_input_fields, 35 CeedQFunctionField *qf_input_fields, CeedInt num_output_fields, CeedOperatorField *op_output_fields, 36 CeedQFunctionField *qf_output_fields, CeedInt *max_P, CeedInt *max_P_1d, CeedInt *Q, CeedInt *Q_1d, 37 CeedInt *max_dim, bool *is_all_tensor, bool *use_3d_slices) { 38 // Check if all are tensor 39 *is_all_tensor = true; 40 for (CeedInt i = 0; i < num_input_fields; i++) { 41 CeedBasis basis; 42 43 CeedCallBackend(CeedOperatorFieldGetBasis(op_input_fields[i], &basis)); 44 if (basis != CEED_BASIS_NONE) { 45 bool is_field_tensor; 46 47 CeedCallBackend(CeedBasisIsTensor(basis, &is_field_tensor)); 48 *is_all_tensor = *is_all_tensor && is_field_tensor; 49 } 50 CeedCallBackend(CeedBasisDestroy(&basis)); 51 } 52 for (CeedInt i = 0; i < num_output_fields; i++) { 53 CeedBasis basis; 54 55 CeedCallBackend(CeedOperatorFieldGetBasis(op_output_fields[i], &basis)); 56 if (basis != CEED_BASIS_NONE) { 57 bool is_field_tensor; 58 59 CeedCallBackend(CeedBasisIsTensor(basis, &is_field_tensor)); 60 *is_all_tensor = *is_all_tensor && is_field_tensor; 61 } 62 CeedCallBackend(CeedBasisDestroy(&basis)); 63 } 64 65 // Find max_P, max_P_1d, Q, and Q_1d 66 bool is_all_3d = true; 67 68 *max_P = 0; 69 *max_P_1d = 0; 70 *Q = 0; 71 *Q_1d = 0; 72 for (CeedInt i = 0; i < num_input_fields; i++) { 73 CeedBasis basis; 74 75 CeedCallBackend(CeedOperatorFieldGetBasis(op_input_fields[i], &basis)); 76 if (basis != CEED_BASIS_NONE) { 77 bool is_field_tensor; 78 CeedInt field_dim = 0, field_P = 0, field_P_1d = 0, field_Q = 0, field_Q_1d = 0; 79 80 // Check if 3D 81 CeedCallBackend(CeedBasisGetDimension(basis, &field_dim)); 82 is_all_3d = is_all_3d && (field_dim == 3); 83 *max_dim = CeedIntMax(*max_dim, field_dim); 84 85 // Collect P, P_1d, Q, and Q_1d 86 CeedCallBackend(CeedBasisGetNumNodes(basis, &field_P)); 87 *max_P = CeedIntMax(*max_P, field_P); 88 CeedCallBackend(CeedBasisIsTensor(basis, &is_field_tensor)); 89 if (is_field_tensor) { 90 CeedCallBackend(CeedBasisGetNumNodes1D(basis, &field_P_1d)); 91 *max_P_1d = CeedIntMax(*max_P_1d, field_P_1d); 92 } 93 CeedCallBackend(CeedBasisGetNumQuadraturePoints(basis, &field_Q)); 94 CeedCheck(*Q == 0 || field_Q == *Q, ceed, CEED_ERROR_BACKEND, "Quadrature spaces must be compatible"); 95 *Q = field_Q; 96 if (is_field_tensor) { 97 CeedCallBackend(CeedBasisGetNumQuadraturePoints1D(basis, &field_Q_1d)); 98 CeedCheck(*Q_1d == 0 || field_Q_1d == *Q_1d, ceed, CEED_ERROR_BACKEND, "Quadrature spaces must be compatible"); 99 *Q_1d = field_Q_1d; 100 } 101 } 102 CeedCallBackend(CeedBasisDestroy(&basis)); 103 } 104 for (CeedInt i = 0; i < num_output_fields; i++) { 105 CeedBasis basis; 106 107 CeedCallBackend(CeedOperatorFieldGetBasis(op_output_fields[i], &basis)); 108 if (basis != CEED_BASIS_NONE) { 109 bool is_field_tensor; 110 CeedInt field_dim = 0, field_P = 0, field_P_1d = 0, field_Q = 0, field_Q_1d = 0; 111 112 // Check if 3D 113 CeedCallBackend(CeedBasisGetDimension(basis, &field_dim)); 114 is_all_3d = is_all_3d && (field_dim == 3); 115 *max_dim = CeedIntMax(*max_dim, field_dim); 116 117 // Collect P, P_1d, Q, and Q_1d 118 CeedCallBackend(CeedBasisGetNumNodes(basis, &field_P)); 119 *max_P = CeedIntMax(*max_P, field_P); 120 CeedCallBackend(CeedBasisIsTensor(basis, &is_field_tensor)); 121 if (is_field_tensor) { 122 CeedCallBackend(CeedBasisGetNumNodes1D(basis, &field_P_1d)); 123 *max_P_1d = CeedIntMax(*max_P_1d, field_P_1d); 124 } 125 CeedCallBackend(CeedBasisGetNumQuadraturePoints(basis, &field_Q)); 126 CeedCheck(*Q == 0 || field_Q == *Q, ceed, CEED_ERROR_BACKEND, "Quadrature spaces must be compatible"); 127 *Q = field_Q; 128 if (is_field_tensor) { 129 CeedCallBackend(CeedBasisGetNumQuadraturePoints1D(basis, &field_Q_1d)); 130 CeedCheck(*Q_1d == 0 || field_Q_1d == *Q_1d, ceed, CEED_ERROR_BACKEND, "Quadrature spaces must be compatible"); 131 *Q_1d = field_Q_1d; 132 } 133 } 134 CeedCallBackend(CeedBasisDestroy(&basis)); 135 } 136 137 // Only use 3D collocated gradient parallelization strategy when gradient is computed 138 *use_3d_slices = false; 139 if (is_all_3d && *is_all_tensor) { 140 bool was_grad_found = false; 141 142 for (CeedInt i = 0; i < num_input_fields; i++) { 143 CeedEvalMode eval_mode; 144 145 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode)); 146 if (eval_mode == CEED_EVAL_GRAD) { 147 CeedBasis_Cuda_shared *basis_data; 148 CeedBasis basis; 149 150 CeedCallBackend(CeedOperatorFieldGetBasis(op_input_fields[i], &basis)); 151 CeedCallBackend(CeedBasisGetData(basis, &basis_data)); 152 *use_3d_slices = basis_data->d_collo_grad_1d && (was_grad_found ? *use_3d_slices : true); 153 was_grad_found = true; 154 CeedCallBackend(CeedBasisDestroy(&basis)); 155 } 156 } 157 for (CeedInt i = 0; i < num_output_fields; i++) { 158 CeedEvalMode eval_mode; 159 160 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 161 if (eval_mode == CEED_EVAL_GRAD) { 162 CeedBasis_Cuda_shared *basis_data; 163 CeedBasis basis; 164 165 CeedCallBackend(CeedOperatorFieldGetBasis(op_output_fields[i], &basis)); 166 CeedCallBackend(CeedBasisGetData(basis, &basis_data)); 167 *use_3d_slices = basis_data->d_collo_grad_1d && (was_grad_found ? *use_3d_slices : true); 168 was_grad_found = true; 169 CeedCallBackend(CeedBasisDestroy(&basis)); 170 } 171 } 172 } 173 return CEED_ERROR_SUCCESS; 174 } 175 176 //------------------------------------------------------------------------------ 177 // Setup fields 178 //------------------------------------------------------------------------------ 179 static int CeedOperatorBuildKernelFieldData_Cuda_gen(std::ostringstream &code, CeedOperator_Cuda_gen *data, CeedInt i, CeedOperatorField op_field, 180 CeedQFunctionField qf_field, FieldReuse_Cuda field_reuse, CeedInt Q, CeedInt Q_1d, bool is_input, 181 bool is_all_tensor, bool is_at_points, bool use_3d_slices) { 182 bool is_tensor = true; 183 CeedBasis basis; 184 CeedCallBackend(CeedOperatorFieldGetBasis(op_field, &basis)); 185 if (basis != CEED_BASIS_NONE) CeedCallBackend(CeedBasisIsTensor(basis, &is_tensor)); 186 187 const char *field_name; 188 std::string var_suffix = (is_input ? "_in_" : "_out_") + std::to_string(i); 189 std::string P_name = (is_tensor ? "P_1d" : "P") + var_suffix, Q_name = is_tensor ? "Q_1d" : "Q"; 190 std::string option_name = (is_input ? "inputs" : "outputs"); 191 CeedEvalMode eval_mode = CEED_EVAL_NONE; 192 CeedInt elem_size = 0, num_comp = 0, dim = 1, P_1d = 0; 193 CeedElemRestriction elem_rstr; 194 CeedBasis_Cuda_shared *basis_data; 195 196 // Field reuse info 197 bool use_previous_field = field_reuse.index != -1; 198 199 CeedCallBackend(CeedOperatorFieldGetName(op_field, &field_name)); 200 code << " // -- " << (is_input ? "Input" : "Output") << " field " << i << ": " << field_name << "\n"; 201 202 // Get field data 203 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_field, &elem_rstr)); 204 if (elem_rstr != CEED_ELEMRESTRICTION_NONE) { 205 CeedCallBackend(CeedElemRestrictionGetElementSize(elem_rstr, &elem_size)); 206 CeedCallBackend(CeedElemRestrictionGetNumComponents(elem_rstr, &num_comp)); 207 } 208 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 209 if (basis != CEED_BASIS_NONE) { 210 CeedCallBackend(CeedBasisGetData(basis, &basis_data)); 211 CeedCallBackend(CeedBasisGetDimension(basis, &dim)); 212 if (is_tensor) CeedCallBackend(CeedBasisGetNumNodes1D(basis, &P_1d)); 213 else CeedCallBackend(CeedBasisGetNumNodes(basis, &P_1d)); 214 } 215 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_field, &eval_mode)); 216 217 // Set field constants 218 code << " const CeedInt dim" << var_suffix << " = " << dim << ";\n"; 219 if (is_tensor && !is_all_tensor) { 220 CeedInt P = 0; 221 222 CeedCallBackend(CeedBasisGetNumNodes(basis, &P)); 223 code << " const CeedInt P" << var_suffix << " = " << (basis == CEED_BASIS_NONE ? Q : P) << ";\n"; 224 } 225 code << " const CeedInt " << P_name << " = " << (basis == CEED_BASIS_NONE ? Q_1d : P_1d) << ";\n"; 226 if (eval_mode != CEED_EVAL_WEIGHT) { 227 code << " const CeedInt num_comp" << var_suffix << " = " << num_comp << ";\n"; 228 } 229 230 // Load basis data 231 code << " // EvalMode: " << CeedEvalModes[eval_mode] << "\n"; 232 switch (eval_mode) { 233 case CEED_EVAL_NONE: 234 break; 235 case CEED_EVAL_INTERP: 236 if (is_at_points) { 237 // AtPoints 238 if (!basis_data->d_chebyshev_interp_1d) { 239 CeedSize interp_bytes; 240 CeedScalar *chebyshev_interp_1d; 241 242 interp_bytes = P_1d * Q_1d * sizeof(CeedScalar); 243 CeedCallBackend(CeedCalloc(P_1d * Q_1d, &chebyshev_interp_1d)); 244 CeedCallBackend(CeedBasisGetChebyshevInterp1D(basis, chebyshev_interp_1d)); 245 CeedCallCuda(CeedBasisReturnCeed(basis), cudaMalloc((void **)&basis_data->d_chebyshev_interp_1d, interp_bytes)); 246 CeedCallCuda(CeedBasisReturnCeed(basis), 247 cudaMemcpy(basis_data->d_chebyshev_interp_1d, chebyshev_interp_1d, interp_bytes, cudaMemcpyHostToDevice)); 248 CeedCallBackend(CeedFree(&chebyshev_interp_1d)); 249 } 250 if (is_input) data->B.inputs[i] = basis_data->d_chebyshev_interp_1d; 251 else data->B.outputs[i] = basis_data->d_chebyshev_interp_1d; 252 } else { 253 // Standard quadrature 254 if (is_input) data->B.inputs[i] = basis_data->d_interp_1d; 255 else data->B.outputs[i] = basis_data->d_interp_1d; 256 } 257 if (use_previous_field) { 258 std::string reuse_var = "s_B" + ((field_reuse.is_input ? "_in_" : "_out_") + std::to_string(field_reuse.index)); 259 260 code << " CeedScalar *s_B" << var_suffix << " = " << reuse_var << ";\n"; 261 } else { 262 code << " __shared__ CeedScalar s_B" << var_suffix << "[" << P_name << "*" << Q_name << "];\n"; 263 code << " LoadMatrix<" << P_name << ", " << Q_name << ">(data, B." << option_name << "[" << i << "], s_B" << var_suffix << ");\n"; 264 } 265 break; 266 case CEED_EVAL_GRAD: 267 if (is_at_points) { 268 // AtPoints 269 if (!basis_data->d_chebyshev_interp_1d) { 270 CeedSize interp_bytes; 271 CeedScalar *chebyshev_interp_1d; 272 273 interp_bytes = P_1d * Q_1d * sizeof(CeedScalar); 274 CeedCallBackend(CeedCalloc(P_1d * Q_1d, &chebyshev_interp_1d)); 275 CeedCallBackend(CeedBasisGetChebyshevInterp1D(basis, chebyshev_interp_1d)); 276 CeedCallCuda(CeedBasisReturnCeed(basis), cudaMalloc((void **)&basis_data->d_chebyshev_interp_1d, interp_bytes)); 277 CeedCallCuda(CeedBasisReturnCeed(basis), 278 cudaMemcpy(basis_data->d_chebyshev_interp_1d, chebyshev_interp_1d, interp_bytes, cudaMemcpyHostToDevice)); 279 CeedCallBackend(CeedFree(&chebyshev_interp_1d)); 280 } 281 if (is_input) data->B.inputs[i] = basis_data->d_chebyshev_interp_1d; 282 else data->B.outputs[i] = basis_data->d_chebyshev_interp_1d; 283 } else { 284 // Standard quadrature 285 if (is_input) data->B.inputs[i] = basis_data->d_interp_1d; 286 else data->B.outputs[i] = basis_data->d_interp_1d; 287 } 288 if (is_tensor) { 289 if (use_previous_field) { 290 std::string reuse_var = "s_B" + ((field_reuse.is_input ? "_in_" : "_out_") + std::to_string(field_reuse.index)); 291 292 code << " CeedScalar *s_B" << var_suffix << " = " << reuse_var << ";\n"; 293 } else { 294 code << " __shared__ CeedScalar s_B" << var_suffix << "[" << P_name << "*" << Q_name << "];\n"; 295 code << " LoadMatrix<" << P_name << ", " << Q_name << ">(data, B." << option_name << "[" << i << "], s_B" << var_suffix << ");\n"; 296 } 297 } 298 if (is_at_points) break; // No G mat for AtPoints 299 if (use_3d_slices) { 300 if (is_input) data->G.inputs[i] = basis_data->d_collo_grad_1d; 301 else data->G.outputs[i] = basis_data->d_collo_grad_1d; 302 if (use_previous_field && field_reuse.eval_mode == CEED_EVAL_GRAD) { 303 std::string reuse_var = "s_G" + ((field_reuse.is_input ? "_in_" : "_out_") + std::to_string(field_reuse.index)); 304 305 code << " CeedScalar *s_G" << var_suffix << " = " << reuse_var << ";\n"; 306 } else { 307 code << " __shared__ CeedScalar s_G" << var_suffix << "[" << Q_name << "*" << Q_name << "];\n"; 308 code << " LoadMatrix<" << Q_name << ", " << Q_name << ">(data, G." << option_name << "[" << i << "], s_G" << var_suffix << ");\n"; 309 } 310 } else { 311 bool has_collo_grad = basis_data->d_collo_grad_1d; 312 313 if (is_input) data->G.inputs[i] = has_collo_grad ? basis_data->d_collo_grad_1d : basis_data->d_grad_1d; 314 else data->G.outputs[i] = has_collo_grad ? basis_data->d_collo_grad_1d : basis_data->d_grad_1d; 315 if (has_collo_grad) { 316 if (use_previous_field && field_reuse.eval_mode == CEED_EVAL_GRAD) { 317 std::string reuse_var = "s_G" + ((field_reuse.is_input ? "_in_" : "_out_") + std::to_string(field_reuse.index)); 318 319 code << " CeedScalar *s_G" << var_suffix << " = " << reuse_var << ";\n"; 320 } else { 321 code << " __shared__ CeedScalar s_G" << var_suffix << "[" << Q_name << "*" << Q_name << "];\n"; 322 code << " LoadMatrix<" << Q_name << ", " << Q_name << ">(data, G." << option_name << "[" << i << "], s_G" << var_suffix << ");\n"; 323 } 324 } else { 325 if (use_previous_field && field_reuse.eval_mode == CEED_EVAL_GRAD) { 326 std::string reuse_var = "s_G" + ((field_reuse.is_input ? "_in_" : "_out_") + std::to_string(field_reuse.index)); 327 328 code << " CeedScalar *s_G" << var_suffix << " = " << reuse_var << ";\n"; 329 } else { 330 code << " __shared__ CeedScalar s_G" << var_suffix << "[" << P_name << "*" << Q_name << (is_tensor ? "" : "*dim") 331 << (is_tensor ? "" : var_suffix) << "];\n"; 332 code << " LoadMatrix<" << P_name << ", " << Q_name << (is_tensor ? "" : "*dim") << (is_tensor ? "" : var_suffix) << ">(data, G." 333 << option_name << "[" << i << "], s_G" << var_suffix << ");\n"; 334 } 335 } 336 } 337 break; 338 case CEED_EVAL_WEIGHT: 339 break; // No action 340 // LCOV_EXCL_START 341 case CEED_EVAL_DIV: 342 case CEED_EVAL_CURL: 343 break; // TODO: Not implemented 344 // LCOV_EXCL_STOP 345 } 346 CeedCallBackend(CeedBasisDestroy(&basis)); 347 return CEED_ERROR_SUCCESS; 348 } 349 350 //------------------------------------------------------------------------------ 351 // Restriction 352 //------------------------------------------------------------------------------ 353 static int CeedOperatorBuildKernelRestriction_Cuda_gen(std::ostringstream &code, CeedOperator_Cuda_gen *data, CeedInt i, CeedInt max_dim, 354 CeedInt field_input_buffer[], CeedOperatorField op_field, CeedQFunctionField qf_field, 355 CeedInt Q_1d, bool is_input, bool is_all_tensor, bool is_at_points, bool use_3d_slices) { 356 std::string var_suffix = (is_input ? "_in_" : "_out_") + std::to_string(i); 357 std::string P_name = (is_all_tensor ? "P_1d" : "P") + var_suffix; 358 CeedEvalMode eval_mode = CEED_EVAL_NONE; 359 CeedInt elem_size = 0, num_comp = 0; 360 CeedSize l_size; 361 CeedRestrictionType rstr_type = CEED_RESTRICTION_STANDARD; 362 CeedElemRestriction_Cuda *rstr_data; 363 CeedElemRestriction elem_rstr; 364 365 // Get field data 366 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_field, &elem_rstr)); 367 if (elem_rstr != CEED_ELEMRESTRICTION_NONE) { 368 CeedCallBackend(CeedElemRestrictionGetType(elem_rstr, &rstr_type)); 369 CeedCallBackend(CeedElemRestrictionGetElementSize(elem_rstr, &elem_size)); 370 CeedCallBackend(CeedElemRestrictionGetNumComponents(elem_rstr, &num_comp)); 371 CeedCallBackend(CeedElemRestrictionGetData(elem_rstr, &rstr_data)); 372 } 373 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_field, &eval_mode)); 374 375 // Restriction 376 if (is_input) { 377 // Input 378 if (field_input_buffer[i] != i) { 379 std::string buffer_name = "r_e_in_" + std::to_string(field_input_buffer[i]); 380 381 // Restriction was already done for previous input 382 code << " CeedScalar *r_e" << var_suffix << " = " << buffer_name << ";\n"; 383 } else if (eval_mode != CEED_EVAL_WEIGHT && !((eval_mode == CEED_EVAL_NONE) && use_3d_slices && is_at_points)) { 384 if (eval_mode == CEED_EVAL_NONE && rstr_type != CEED_RESTRICTION_POINTS) { 385 // No basis action, so r_e_in_* in also r_q_in_* and needs to be allocated 386 code << " CeedScalar r_e" << var_suffix << "[num_comp" << var_suffix << "*" << P_name << "];\n"; 387 } else if (rstr_type != CEED_RESTRICTION_POINTS) { 388 // Otherwise we're using the scratch space 389 code << " CeedScalar *r_e" << var_suffix << " = r_e_scratch;\n"; 390 } 391 switch (rstr_type) { 392 case CEED_RESTRICTION_STANDARD: { 393 CeedInt comp_stride; 394 395 CeedCallBackend(CeedElemRestrictionGetLVectorSize(elem_rstr, &l_size)); 396 code << " const CeedInt l_size" << var_suffix << " = " << l_size << ";\n"; 397 CeedCallBackend(CeedElemRestrictionGetCompStride(elem_rstr, &comp_stride)); 398 code << " // CompStride: " << comp_stride << "\n"; 399 data->indices.inputs[i] = (CeedInt *)rstr_data->d_offsets; 400 code << " ReadLVecStandard" << (is_all_tensor ? max_dim : 1) << "d<num_comp" << var_suffix << ", " << comp_stride << ", " << P_name 401 << ">(data, l_size" << var_suffix << ", elem, indices.inputs[" << i << "], d" << var_suffix << ", r_e" << var_suffix << ");\n"; 402 break; 403 } 404 case CEED_RESTRICTION_STRIDED: { 405 bool has_backend_strides; 406 CeedInt num_elem; 407 408 CeedCallBackend(CeedElemRestrictionHasBackendStrides(elem_rstr, &has_backend_strides)); 409 CeedCallBackend(CeedElemRestrictionGetNumElements(elem_rstr, &num_elem)); 410 CeedInt strides[3] = {1, elem_size * num_elem, elem_size}; 411 412 if (!has_backend_strides) { 413 CeedCallBackend(CeedElemRestrictionGetStrides(elem_rstr, strides)); 414 } 415 code << " // Strides: {" << strides[0] << ", " << strides[1] << ", " << strides[2] << "}\n"; 416 code << " ReadLVecStrided" << (is_all_tensor ? max_dim : 1) << "d<num_comp" << var_suffix << ", " << P_name << ", " << strides[0] << ", " 417 << strides[1] << ", " << strides[2] << ">(data, elem, d" << var_suffix << ", r_e" << var_suffix << ");\n"; 418 break; 419 } 420 case CEED_RESTRICTION_POINTS: { 421 CeedInt comp_stride; 422 423 CeedCallBackend(CeedElemRestrictionGetCompStride(elem_rstr, &comp_stride)); 424 code << " const CeedInt comp_stride" << var_suffix << " = " << comp_stride << ";\n"; 425 data->indices.inputs[i] = (CeedInt *)rstr_data->d_offsets; 426 break; 427 } 428 // LCOV_EXCL_START 429 case CEED_RESTRICTION_ORIENTED: 430 case CEED_RESTRICTION_CURL_ORIENTED: 431 break; // TODO: Not implemented 432 // LCOV_EXCL_STOP 433 } 434 } 435 } else { 436 // Output 437 switch (rstr_type) { 438 case CEED_RESTRICTION_STANDARD: { 439 CeedInt comp_stride; 440 441 CeedCallBackend(CeedElemRestrictionGetLVectorSize(elem_rstr, &l_size)); 442 code << " const CeedInt l_size" << var_suffix << " = " << l_size << ";\n"; 443 CeedCallBackend(CeedElemRestrictionGetCompStride(elem_rstr, &comp_stride)); 444 code << " // CompStride: " << comp_stride << "\n"; 445 data->indices.outputs[i] = (CeedInt *)rstr_data->d_offsets; 446 code << " WriteLVecStandard" << (is_all_tensor ? max_dim : 1) << "d<num_comp" << var_suffix << ", " << comp_stride << ", " << P_name 447 << ">(data, l_size" << var_suffix << ", elem, indices.outputs[" << i << "], r_e" << var_suffix << ", d" << var_suffix << ");\n"; 448 break; 449 } 450 case CEED_RESTRICTION_STRIDED: { 451 bool has_backend_strides; 452 CeedInt num_elem; 453 454 CeedCallBackend(CeedElemRestrictionHasBackendStrides(elem_rstr, &has_backend_strides)); 455 CeedCallBackend(CeedElemRestrictionGetNumElements(elem_rstr, &num_elem)); 456 CeedInt strides[3] = {1, elem_size * num_elem, elem_size}; 457 458 if (!has_backend_strides) { 459 CeedCallBackend(CeedElemRestrictionGetStrides(elem_rstr, strides)); 460 } 461 code << " // Strides: {" << strides[0] << ", " << strides[1] << ", " << strides[2] << "}\n"; 462 code << " WriteLVecStrided" << (is_all_tensor ? max_dim : 1) << "d<num_comp" << var_suffix << ", " << P_name << ", " << strides[0] << ", " 463 << strides[1] << ", " << strides[2] << ">(data, elem, r_e" << var_suffix << ", d" << var_suffix << ");\n"; 464 break; 465 } 466 case CEED_RESTRICTION_POINTS: 467 data->indices.outputs[i] = (CeedInt *)rstr_data->d_offsets; 468 break; 469 // LCOV_EXCL_START 470 case CEED_RESTRICTION_ORIENTED: 471 case CEED_RESTRICTION_CURL_ORIENTED: 472 break; // TODO: Not implemented 473 // LCOV_EXCL_STOP 474 } 475 } 476 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 477 return CEED_ERROR_SUCCESS; 478 } 479 480 //------------------------------------------------------------------------------ 481 // Basis 482 //------------------------------------------------------------------------------ 483 static int CeedOperatorBuildKernelBasis_Cuda_gen(std::ostringstream &code, CeedOperator_Cuda_gen *data, CeedInt i, CeedInt max_dim, 484 CeedOperatorField op_field, CeedQFunctionField qf_field, CeedInt Q_1d, bool is_input, 485 bool is_all_tensor, bool is_at_points, bool use_3d_slices) { 486 bool is_tensor = true; 487 CeedBasis basis; 488 CeedCallBackend(CeedOperatorFieldGetBasis(op_field, &basis)); 489 CeedCallBackend(CeedBasisIsTensor(basis, &is_tensor)); 490 491 std::string var_suffix = (is_input ? "_in_" : "_out_") + std::to_string(i); 492 std::string P_name = (is_tensor ? "P_1d" : "P") + var_suffix, Q_name = is_tensor ? "Q_1d" : "Q"; 493 CeedEvalMode eval_mode = CEED_EVAL_NONE; 494 CeedInt dim = max_dim, elem_size = 0, num_comp = 0, P_1d = 0; 495 CeedElemRestriction elem_rstr; 496 497 // Get field data 498 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_field, &elem_rstr)); 499 if (elem_rstr != CEED_ELEMRESTRICTION_NONE) { 500 CeedCallBackend(CeedElemRestrictionGetElementSize(elem_rstr, &elem_size)); 501 CeedCallBackend(CeedElemRestrictionGetNumComponents(elem_rstr, &num_comp)); 502 } 503 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 504 if (basis != CEED_BASIS_NONE) { 505 CeedCallBackend(CeedBasisGetDimension(basis, &dim)); 506 if (is_tensor) CeedCallBackend(CeedBasisGetNumNodes1D(basis, &P_1d)); 507 else CeedCallBackend(CeedBasisGetNumNodes(basis, &P_1d)); 508 } 509 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_field, &eval_mode)); 510 511 // Basis 512 code << " // EvalMode: " << CeedEvalModes[eval_mode] << "\n"; 513 if (is_input) { 514 switch (eval_mode) { 515 case CEED_EVAL_NONE: 516 if (!use_3d_slices && !is_at_points) { 517 code << " CeedScalar *r_q" << var_suffix << " = r_e" << var_suffix << ";\n"; 518 } 519 break; 520 case CEED_EVAL_INTERP: 521 if (is_at_points) { 522 std::string function_name = (dim == 1 ? "Interp" : "InterpTensor") + std::to_string(dim) + "d"; 523 524 code << " CeedScalar r_c" << var_suffix << "[num_comp" << var_suffix << "*" << (dim >= 3 ? Q_name : "1") << "];\n"; 525 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", OP_T_1D>(data, r_e" << var_suffix 526 << ", s_B" << var_suffix << ", r_c" << var_suffix << ");\n"; 527 } else { 528 std::string function_name = is_tensor 529 ? ((dim == 1 ? "Interp" : "InterpTensor") + std::to_string(dim) + "d" + (is_all_tensor ? "" : "Flattened")) 530 : "InterpNonTensor"; 531 std::string op_t_1d_name = (is_all_tensor || !is_tensor) ? "OP_T_1D" : (P_1d > Q_1d ? P_name : Q_name); 532 533 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*" << (is_tensor && (dim >= 3) ? Q_name : "1") << "];\n"; 534 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", " << op_t_1d_name << ">(data, r_e" 535 << var_suffix << ", s_B" << var_suffix << ", r_q" << var_suffix << ");\n"; 536 } 537 break; 538 case CEED_EVAL_GRAD: 539 if (is_at_points) { 540 std::string function_name = (dim == 1 ? "Interp" : "InterpTensor") + std::to_string(dim) + "d"; 541 542 code << " CeedScalar r_c" << var_suffix << "[num_comp" << var_suffix << "*" << (dim >= 3 ? Q_name : "1") << "];\n"; 543 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", OP_T_1D>(data, r_e" << var_suffix 544 << ", s_B" << var_suffix << ", r_c" << var_suffix << ");\n"; 545 } else if (use_3d_slices) { 546 std::string function_name = (dim > 1 ? "InterpTensor" : "Interp") + std::to_string(dim) + "d"; 547 548 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*" << Q_name << "];\n"; 549 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", OP_T_1D>(data, r_e" << var_suffix 550 << ", s_B" << var_suffix << ", r_q" << var_suffix << ");\n"; 551 } else if (is_tensor) { 552 bool is_collocated = dim == 3 && Q_1d >= P_1d; 553 std::string function_name = (dim == 1 ? "Grad" : (is_collocated ? "GradTensorCollocated" : "GradTensor")) + std::to_string(dim) + "d" + 554 (is_all_tensor ? "" : "Flattened"); 555 std::string op_t_1d_name = is_all_tensor ? "OP_T_1D" : (P_1d > Q_1d ? P_name : Q_name); 556 557 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*dim" << var_suffix << "*" << (dim >= 3 ? Q_name : "1") 558 << "];\n"; 559 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", " << op_t_1d_name << ">(data, r_e" 560 << var_suffix << ", s_B" << var_suffix << ", s_G" << var_suffix << ", r_q" << var_suffix << ");\n"; 561 } else { 562 std::string function_name = "GradNonTensor"; 563 564 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*dim" << var_suffix << "];\n"; 565 code << " " << function_name << "<num_comp" << var_suffix << ", dim" << var_suffix << ", " << P_name << ", " << Q_name 566 << ", OP_T_1D>(data, r_e" << var_suffix << ", s_G" << var_suffix << ", r_q" << var_suffix << ");\n"; 567 } 568 break; 569 case CEED_EVAL_WEIGHT: { 570 if (is_at_points) { 571 code << " // Nothing to do AtPoints\n"; 572 } else { 573 CeedBasis_Cuda_shared *basis_data; 574 std::string function_name = is_tensor 575 ? ((dim == 1 ? "Weight" : "WeightTensor") + std::to_string(dim) + "d" + (is_all_tensor ? "" : "Flattened")) 576 : "WeightNonTensor"; 577 578 code << " CeedScalar r_q" << var_suffix << "[" << (is_tensor && (dim >= 3) ? Q_name : "1") << "];\n"; 579 CeedCallBackend(CeedBasisGetData(basis, &basis_data)); 580 data->W = basis_data->d_q_weight_1d; 581 code << " " << function_name << "<" << P_name << ", " << Q_name << ">(data, W, r_q" << var_suffix << ");\n"; 582 } 583 break; 584 } 585 // LCOV_EXCL_START 586 case CEED_EVAL_DIV: 587 case CEED_EVAL_CURL: 588 break; // TODO: Not implemented 589 // LCOV_EXCL_STOP 590 } 591 } else { 592 switch (eval_mode) { 593 case CEED_EVAL_NONE: 594 code << " CeedScalar *r_e" << var_suffix << " = r_q" << var_suffix << ";\n"; 595 break; // No action 596 case CEED_EVAL_INTERP: 597 code << " CeedScalar *r_e" << var_suffix << " = r_e_scratch;\n"; 598 if (is_at_points) { 599 std::string function_name = (dim == 1 ? "InterpTranspose" : "InterpTransposeTensor") + std::to_string(dim) + "d"; 600 601 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", OP_T_1D>(data, r_c" << var_suffix 602 << ", s_B" << var_suffix << ", r_e" << var_suffix << ");\n"; 603 } else { 604 std::string function_name = 605 is_tensor ? ((dim == 1 ? "InterpTranspose" : "InterpTransposeTensor") + std::to_string(dim) + "d" + (is_all_tensor ? "" : "Flattened")) 606 : "InterpTransposeNonTensor"; 607 std::string op_t_1d_name = (is_all_tensor || !is_tensor) ? "OP_T_1D" : (P_1d > Q_1d ? P_name : Q_name); 608 609 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", " << op_t_1d_name << ">(data, r_q" 610 << var_suffix << ", s_B" << var_suffix << ", r_e" << var_suffix << ");\n"; 611 } 612 break; 613 case CEED_EVAL_GRAD: 614 code << " CeedScalar *r_e" << var_suffix << " = r_e_scratch;\n"; 615 if (is_at_points) { 616 std::string function_name = (dim == 1 ? "InterpTranspose" : "InterpTransposeTensor") + std::to_string(dim) + "d"; 617 618 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", OP_T_1D>(data, r_c" << var_suffix 619 << ", s_B" << var_suffix << ", r_e" << var_suffix << ");\n"; 620 } else if (use_3d_slices) { 621 std::string function_name = (dim == 1 ? "InterpTranspose" : "InterpTransposeTensor") + std::to_string(dim) + "d"; 622 623 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", OP_T_1D>(data, r_q" << var_suffix 624 << ", s_B" << var_suffix << ", r_e" << var_suffix << ");\n"; 625 } else if (is_tensor) { 626 bool is_collocated = dim == 3 && Q_1d >= P_1d; 627 std::string function_name = (dim == 1 ? "GradTranspose" : (is_collocated ? "GradTransposeTensorCollocated" : "GradTransposeTensor")) + 628 std::to_string(dim) + "d" + (is_all_tensor ? "" : "Flattened"); 629 std::string op_t_1d_name = is_all_tensor ? "OP_T_1D" : (P_1d > Q_1d ? P_name : Q_name); 630 631 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", " << op_t_1d_name << ">(data, r_q" 632 << var_suffix << ", s_B" << var_suffix << ", s_G" << var_suffix << ", r_e" << var_suffix << ");\n"; 633 } else { 634 std::string function_name = "GradTransposeNonTensor"; 635 636 code << " " << function_name << "<num_comp" << var_suffix << ", dim" << var_suffix << ", " << P_name << ", " << Q_name 637 << ", OP_T_1D>(data, r_q" << var_suffix << ", s_G" << var_suffix << ", r_e" << var_suffix << ");\n"; 638 } 639 break; 640 // LCOV_EXCL_START 641 case CEED_EVAL_WEIGHT: 642 break; // Should not occur 643 case CEED_EVAL_DIV: 644 case CEED_EVAL_CURL: 645 break; // TODO: Not implemented 646 // LCOV_EXCL_STOP 647 } 648 } 649 CeedCallBackend(CeedBasisDestroy(&basis)); 650 return CEED_ERROR_SUCCESS; 651 } 652 653 //------------------------------------------------------------------------------ 654 // QFunction 655 //------------------------------------------------------------------------------ 656 static int CeedOperatorBuildKernelQFunction_Cuda_gen(std::ostringstream &code, CeedOperator_Cuda_gen *data, CeedInt max_dim, CeedInt max_num_points, 657 CeedInt num_input_fields, CeedOperatorField *op_input_fields, 658 CeedQFunctionField *qf_input_fields, CeedInt num_output_fields, 659 CeedOperatorField *op_output_fields, CeedQFunctionField *qf_output_fields, 660 std::string qfunction_name, CeedInt Q_1d, bool is_all_tensor, bool is_at_points, 661 bool use_3d_slices) { 662 std::string Q_name = is_all_tensor ? "Q_1d" : "Q"; 663 CeedEvalMode eval_mode = CEED_EVAL_NONE; 664 CeedElemRestriction elem_rstr; 665 666 // Setup output arrays 667 code << "\n // -- Output field setup\n"; 668 for (CeedInt i = 0; i < num_output_fields; i++) { 669 const char *field_name; 670 std::string var_suffix = "_out_" + std::to_string(i); 671 672 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 673 code << " // ---- Output field " << i << ": " << field_name << "\n"; 674 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 675 switch (eval_mode) { 676 case CEED_EVAL_NONE: 677 if (is_at_points) { 678 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "];\n"; 679 } else { 680 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*" << (is_all_tensor && (max_dim >= 3) ? Q_name : "1") 681 << "];\n"; 682 } 683 break; 684 case CEED_EVAL_INTERP: 685 if (is_at_points) { 686 // Accumulator for point data 687 code << " CeedScalar r_c" << var_suffix << "[num_comp" << var_suffix << "*" << (max_dim >= 3 ? Q_name : "1") << "];\n"; 688 code << " for (CeedInt i = 0; i < num_comp" << var_suffix << "*" << (max_dim >= 3 ? Q_name : "1") << "; i++) {\n"; 689 code << " r_c" << var_suffix << "[i] = 0.0;\n"; 690 code << " }\n"; 691 } else { 692 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*" << (is_all_tensor && (max_dim >= 3) ? Q_name : "1") 693 << "];\n"; 694 } 695 break; 696 case CEED_EVAL_GRAD: 697 if (is_at_points) { 698 // Accumulator for point data 699 code << " CeedScalar r_c" << var_suffix << "[num_comp" << var_suffix << "*" << (max_dim >= 3 ? Q_name : "1") << "*dim" << var_suffix 700 << "];\n"; 701 code << " for (CeedInt i = 0; i < num_comp" << var_suffix << "*" << (max_dim >= 3 ? Q_name : "1") << "; i++) {\n"; 702 code << " r_c" << var_suffix << "[i] = 0.0;\n"; 703 code << " }\n"; 704 } else if (use_3d_slices) { 705 // Accumulator for gradient slices 706 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*" << Q_name << "];\n"; 707 code << " for (CeedInt i = 0; i < num_comp" << var_suffix << "*" << Q_name << "; i++) {\n"; 708 code << " r_q" << var_suffix << "[i] = 0.0;\n"; 709 code << " }\n"; 710 } else { 711 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*dim" << var_suffix << "*" 712 << (is_all_tensor && (max_dim >= 3) ? Q_name : "1") << "];\n"; 713 } 714 break; 715 case CEED_EVAL_WEIGHT: 716 break; 717 // LCOV_EXCL_START 718 case CEED_EVAL_DIV: 719 case CEED_EVAL_CURL: 720 break; // TODO: Not implemented 721 // LCOV_EXCL_STOP 722 } 723 } 724 725 if (is_at_points) { 726 // We need to handle batches of points 727 code << "\n // Note: Using batches of points\n"; 728 code << " const CeedInt point_loop_bound = (blockDim.x * blockDim.y) * ceil(1.0 * max_num_points / (blockDim.x * blockDim.y));\n\n"; 729 code << " #pragma unroll\n"; 730 code << " for (CeedInt i = threadIdx.x + threadIdx.y * blockDim.x; i < point_loop_bound; i += blockDim.x * blockDim.y) {\n"; 731 code << " const CeedInt p = i % max_num_points;\n\n"; 732 733 code << " // -- Coordinates\n"; 734 code << " CeedScalar r_x[max_dim];\n"; 735 code << " ReadPoint<max_dim, coords_comp_stride, max_num_points>(data, elem, p, max_num_points, points.indices, points.coords, r_x);\n\n"; 736 737 code << " // -- Input fields\n"; 738 for (CeedInt i = 0; i < num_input_fields; i++) { 739 const char *field_name; 740 std::string var_suffix = "_in_" + std::to_string(i); 741 std::string P_name = "P_1d" + var_suffix; 742 743 CeedCallBackend(CeedOperatorFieldGetName(op_input_fields[i], &field_name)); 744 code << " // ---- Input field " << i << ": " << field_name << "\n"; 745 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode)); 746 // Basis action 747 code << " // EvalMode: " << CeedEvalModes[eval_mode] << "\n"; 748 switch (eval_mode) { 749 case CEED_EVAL_NONE: 750 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 751 code << " ReadPoint<num_comp" << var_suffix << ", comp_stride" << var_suffix 752 << ", max_num_points>(data, elem, p, max_num_points, indices.inputs[" << i << "], d" << var_suffix << ", r_s" << var_suffix << ");\n"; 753 break; 754 case CEED_EVAL_INTERP: 755 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 756 code << " InterpAtPoints" << max_dim << "d<num_comp" << var_suffix << ", max_num_points, " << P_name << ", " << Q_name 757 << ">(data, i, r_c" << var_suffix << ", r_x, r_s" << var_suffix << ");\n"; 758 break; 759 case CEED_EVAL_GRAD: 760 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "*dim" << var_suffix << "];\n"; 761 code << " GradAtPoints" << max_dim << "d<num_comp" << var_suffix << ", max_num_points, " << P_name << ", " << Q_name 762 << ">(data, i, r_c" << var_suffix << ", r_x, r_s" << var_suffix << ");\n"; 763 break; 764 case CEED_EVAL_WEIGHT: 765 code << " CeedScalar r_s" << var_suffix << "[1];\n"; 766 code << " r_s" << var_suffix << "[0] = 1.0;\n"; 767 break; 768 // LCOV_EXCL_START 769 case CEED_EVAL_DIV: 770 case CEED_EVAL_CURL: 771 break; // TODO: Not implemented 772 // LCOV_EXCL_STOP 773 } 774 } 775 code << "\n // -- Output fields\n"; 776 for (CeedInt i = 0; i < num_output_fields; i++) { 777 const char *field_name; 778 std::string var_suffix = "_out_" + std::to_string(i); 779 780 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 781 code << " // ---- Output field " << i << ": " << field_name << "\n"; 782 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 783 // Basis action 784 switch (eval_mode) { 785 case CEED_EVAL_NONE: 786 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 787 break; 788 case CEED_EVAL_INTERP: 789 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 790 break; 791 case CEED_EVAL_GRAD: 792 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "*dim" << var_suffix << "];\n"; 793 break; 794 // LCOV_EXCL_START 795 case CEED_EVAL_WEIGHT: 796 break; // Should not occur 797 case CEED_EVAL_DIV: 798 case CEED_EVAL_CURL: 799 break; // TODO: Not implemented 800 // LCOV_EXCL_STOP 801 } 802 } 803 804 } else if (use_3d_slices) { 805 // We treat quadrature points per slice in 3d to save registers 806 code << "\n // Note: Using planes of 3D elements\n"; 807 code << " #pragma unroll\n"; 808 code << " for (CeedInt q = 0; q < " << Q_name << "; q++) {\n"; 809 code << " // -- Input fields\n"; 810 for (CeedInt i = 0; i < num_input_fields; i++) { 811 const char *field_name; 812 std::string var_suffix = "_in_" + std::to_string(i); 813 814 CeedCallBackend(CeedOperatorFieldGetName(op_input_fields[i], &field_name)); 815 code << " // ---- Input field " << i << ": " << field_name << "\n"; 816 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode)); 817 // Basis action 818 code << " // EvalMode: " << CeedEvalModes[eval_mode] << "\n"; 819 switch (eval_mode) { 820 case CEED_EVAL_NONE: 821 bool is_strided; 822 823 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 824 825 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_input_fields[i], &elem_rstr)); 826 CeedCallBackend(CeedElemRestrictionIsStrided(elem_rstr, &is_strided)); 827 if (is_strided) { 828 bool has_backend_strides; 829 CeedInt num_elem, elem_size; 830 831 CeedCallBackend(CeedElemRestrictionGetElementSize(elem_rstr, &elem_size)); 832 CeedCallBackend(CeedElemRestrictionHasBackendStrides(elem_rstr, &has_backend_strides)); 833 CeedCallBackend(CeedElemRestrictionGetNumElements(elem_rstr, &num_elem)); 834 CeedInt strides[3] = {1, elem_size * num_elem, elem_size}; 835 836 if (!has_backend_strides) { 837 CeedCallBackend(CeedElemRestrictionGetStrides(elem_rstr, strides)); 838 } 839 code << " // Strides: {" << strides[0] << ", " << strides[1] << ", " << strides[2] << "}\n"; 840 code << " ReadEVecSliceStrided3d<num_comp" << var_suffix << ", " << Q_name << ", " << strides[0] << ", " << strides[1] << ", " 841 << strides[2] << ">(data, elem, q, d" << var_suffix << ", r_s" << var_suffix << ");\n"; 842 } else { 843 CeedSize l_size = 0; 844 CeedInt comp_stride; 845 CeedElemRestriction_Cuda *rstr_data; 846 847 CeedCallBackend(CeedElemRestrictionGetLVectorSize(elem_rstr, &l_size)); 848 code << " const CeedInt l_size" << var_suffix << " = " << l_size << ";\n"; 849 CeedCallBackend(CeedElemRestrictionGetCompStride(elem_rstr, &comp_stride)); 850 code << " // CompStride: " << comp_stride << "\n"; 851 CeedCallBackend(CeedElemRestrictionGetData(elem_rstr, &rstr_data)); 852 data->indices.inputs[i] = (CeedInt *)rstr_data->d_offsets; 853 code << " ReadEVecSliceStandard3d<num_comp" << var_suffix << ", " << comp_stride << ", " << Q_name << ">(data, l_size" << var_suffix 854 << ", elem, q, indices.inputs[" << i << "], d" << var_suffix << ", r_s" << var_suffix << ");\n"; 855 } 856 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 857 break; 858 case CEED_EVAL_INTERP: 859 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 860 code << " for (CeedInt j = 0; j < num_comp" << var_suffix << "; j++) {\n"; 861 code << " r_s" << var_suffix << "[j] = r_q" << var_suffix << "[q + j*" << Q_name << "];\n"; 862 code << " }\n"; 863 break; 864 case CEED_EVAL_GRAD: 865 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "*dim" << var_suffix << "];\n"; 866 code << " GradColloSlice3d<num_comp" << var_suffix << ", " << Q_name << ", OP_T_1D>(data, q, r_q" << var_suffix << ", s_G" 867 << var_suffix << ", r_s" << var_suffix << ");\n"; 868 break; 869 case CEED_EVAL_WEIGHT: 870 code << " CeedScalar r_s" << var_suffix << "[1];\n"; 871 code << " r_s" << var_suffix << "[0] = r_q" << var_suffix << "[q];\n"; 872 break; 873 // LCOV_EXCL_START 874 case CEED_EVAL_DIV: 875 case CEED_EVAL_CURL: 876 break; // TODO: Not implemented 877 // LCOV_EXCL_STOP 878 } 879 } 880 code << "\n // -- Output fields\n"; 881 for (CeedInt i = 0; i < num_output_fields; i++) { 882 const char *field_name; 883 std::string var_suffix = "_out_" + std::to_string(i); 884 885 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 886 code << " // ---- Output field " << i << ": " << field_name << "\n"; 887 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 888 // Basis action 889 switch (eval_mode) { 890 case CEED_EVAL_NONE: 891 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 892 break; 893 case CEED_EVAL_INTERP: 894 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 895 break; 896 case CEED_EVAL_GRAD: 897 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "*dim" << var_suffix << "];\n"; 898 break; 899 // LCOV_EXCL_START 900 case CEED_EVAL_WEIGHT: 901 break; // Should not occur 902 case CEED_EVAL_DIV: 903 case CEED_EVAL_CURL: 904 break; // TODO: Not implemented 905 // LCOV_EXCL_STOP 906 } 907 } 908 } else { 909 code << "\n // Note: Using full elements\n"; 910 code << " {\n"; 911 code << " // -- Input fields\n"; 912 for (CeedInt i = 0; i < num_input_fields; i++) { 913 const char *field_name; 914 915 CeedCallBackend(CeedOperatorFieldGetName(op_input_fields[i], &field_name)); 916 code << " // ---- Input field " << i << ": " << field_name << "\n"; 917 code << " CeedScalar *r_s_in_" << i << " = r_q_in_" << i << ";\n"; 918 } 919 code << " // -- Output fields\n"; 920 for (CeedInt i = 0; i < num_output_fields; i++) { 921 const char *field_name; 922 923 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 924 code << " // ---- Output field " << i << ": " << field_name << "\n"; 925 code << " CeedScalar *r_s_out_" << i << " = r_q_out_" << i << ";\n"; 926 } 927 } 928 929 // Input and output buffers 930 code << "\n // -- QFunction inputs and outputs\n"; 931 code << " // ---- Inputs\n"; 932 code << " CeedScalar *inputs[" << CeedIntMax(num_input_fields, 1) << "];\n"; 933 for (CeedInt i = 0; i < num_input_fields; i++) { 934 const char *field_name; 935 936 CeedCallBackend(CeedOperatorFieldGetName(op_input_fields[i], &field_name)); 937 code << " // ------ Input field " << i << ": " << field_name << "\n"; 938 code << " inputs[" << i << "] = r_s_in_" << i << ";\n"; 939 } 940 code << " // ---- Outputs\n"; 941 code << " CeedScalar *outputs[" << CeedIntMax(num_output_fields, 1) << "];\n"; 942 for (CeedInt i = 0; i < num_output_fields; i++) { 943 const char *field_name; 944 945 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 946 code << " // ------ Output field " << i << ": " << field_name << "\n"; 947 code << " outputs[" << i << "] = r_s_out_" << i << ";\n"; 948 } 949 950 // Apply QFunction 951 code << "\n // -- Apply QFunction\n"; 952 code << " " << qfunction_name << "(ctx, "; 953 if (max_dim != 3 || is_at_points || use_3d_slices || !is_all_tensor) { 954 code << "1"; 955 } else { 956 code << Q_name; 957 } 958 code << ", inputs, outputs);\n"; 959 960 if (is_at_points) { 961 // Map back to coefficients 962 code << "\n // -- Output fields\n"; 963 for (CeedInt i = 0; i < num_output_fields; i++) { 964 const char *field_name; 965 std::string var_suffix = "_out_" + std::to_string(i); 966 std::string P_name = "P_1d" + var_suffix; 967 968 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 969 code << " // ---- Output field " << i << ": " << field_name << "\n"; 970 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 971 // Basis action 972 code << " // EvalMode: " << CeedEvalModes[eval_mode] << "\n"; 973 switch (eval_mode) { 974 case CEED_EVAL_NONE: { 975 CeedInt comp_stride; 976 CeedElemRestriction elem_rstr; 977 978 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_output_fields[i], &elem_rstr)); 979 CeedCallBackend(CeedElemRestrictionGetCompStride(elem_rstr, &comp_stride)); 980 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 981 code << " const CeedInt comp_stride" << var_suffix << " = " << comp_stride << ";\n"; 982 code << " WritePoint<num_comp" << var_suffix << ", comp_stride" << var_suffix 983 << ", max_num_points>(data, elem, i, points.num_per_elem[elem], indices.outputs[" << i << "]" 984 << ", r_s" << var_suffix << ", d" << var_suffix << ");\n"; 985 break; 986 } 987 case CEED_EVAL_INTERP: 988 code << " if (i >= points.num_per_elem[elem]) {\n"; 989 code << " for (CeedInt j = 0; j < num_comp" << var_suffix << "; j++) r_s" << var_suffix << "[j] = 0.0;\n"; 990 code << " }\n"; 991 code << " InterpTransposeAtPoints" << max_dim << "d<num_comp" << var_suffix << ", max_num_points, " << P_name << ", " << Q_name 992 << ">(data, i, r_s" << var_suffix << ", r_x, r_c" << var_suffix << ");\n"; 993 break; 994 case CEED_EVAL_GRAD: 995 code << " if (i >= points.num_per_elem[elem]) {\n"; 996 code << " for (CeedInt j = 0; j < num_comp" << var_suffix << "*dim" << var_suffix << "; j++) r_s" << var_suffix << "[j] = 0.0;\n"; 997 code << " }\n"; 998 code << " GradTransposeAtPoints" << max_dim << "d<num_comp" << var_suffix << ", max_num_points, " << P_name << ", " << Q_name 999 << ">(data, i, r_s" << var_suffix << ", r_x, r_c" << var_suffix << ");\n"; 1000 break; 1001 // LCOV_EXCL_START 1002 case CEED_EVAL_WEIGHT: 1003 break; // Should not occur 1004 case CEED_EVAL_DIV: 1005 case CEED_EVAL_CURL: 1006 break; // TODO: Not implemented 1007 // LCOV_EXCL_STOP 1008 } 1009 } 1010 } else if (use_3d_slices) { 1011 // Copy or apply transpose grad, if needed 1012 code << "\n // -- Output fields\n"; 1013 for (CeedInt i = 0; i < num_output_fields; i++) { 1014 const char *field_name; 1015 std::string var_suffix = "_out_" + std::to_string(i); 1016 std::string P_name = "P_1d" + var_suffix; 1017 1018 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 1019 code << " // ---- Output field " << i << ": " << field_name << "\n"; 1020 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 1021 // Basis action 1022 code << " // EvalMode: " << CeedEvalModes[eval_mode] << "\n"; 1023 switch (eval_mode) { 1024 case CEED_EVAL_NONE: 1025 code << " for (CeedInt j = 0; j < num_comp" << var_suffix << " ; j++) {\n"; 1026 code << " r_q" << var_suffix << "[q + j*" << Q_name << "] = r_s" << var_suffix << "[j];\n"; 1027 code << " }\n"; 1028 break; 1029 case CEED_EVAL_INTERP: 1030 code << " for (CeedInt j = 0; j < num_comp" << var_suffix << " ; j++) {\n"; 1031 code << " r_q" << var_suffix << "[q + j*" << Q_name << "] = r_s" << var_suffix << "[j];\n"; 1032 code << " }\n"; 1033 break; 1034 case CEED_EVAL_GRAD: 1035 code << " GradColloSliceTranspose3d<num_comp" << var_suffix << ", " << Q_name << ", OP_T_1D>(data, q, r_s" << var_suffix << ", s_G" 1036 << var_suffix << ", r_q" << var_suffix << ");\n"; 1037 break; 1038 // LCOV_EXCL_START 1039 case CEED_EVAL_WEIGHT: 1040 break; // Should not occur 1041 case CEED_EVAL_DIV: 1042 case CEED_EVAL_CURL: 1043 break; // TODO: Not implemented 1044 // LCOV_EXCL_STOP 1045 } 1046 } 1047 } 1048 code << " }\n"; 1049 return CEED_ERROR_SUCCESS; 1050 } 1051 1052 //------------------------------------------------------------------------------ 1053 // Build single operator kernel 1054 //------------------------------------------------------------------------------ 1055 extern "C" int CeedOperatorBuildKernel_Cuda_gen(CeedOperator op, bool *is_good_build) { 1056 bool is_all_tensor = true, is_all_nontensor = true, is_at_points = false, use_3d_slices = false; 1057 Ceed ceed; 1058 CeedInt Q, Q_1d, num_input_fields, num_output_fields, max_dim = 1, max_num_points = 0, coords_comp_stride = 0; 1059 CeedQFunctionField *qf_input_fields, *qf_output_fields; 1060 CeedQFunction_Cuda_gen *qf_data; 1061 CeedQFunction qf; 1062 CeedOperatorField *op_input_fields, *op_output_fields; 1063 CeedOperator_Cuda_gen *data; 1064 std::ostringstream code; 1065 1066 CeedCallBackend(CeedOperatorGetData(op, &data)); 1067 { 1068 bool is_setup_done; 1069 1070 CeedCallBackend(CeedOperatorIsSetupDone(op, &is_setup_done)); 1071 if (is_setup_done) { 1072 *is_good_build = !data->use_fallback; 1073 return CEED_ERROR_SUCCESS; 1074 } 1075 } 1076 1077 // Check field compatibility 1078 CeedCallBackend(CeedOperatorGetFields(op, &num_input_fields, &op_input_fields, &num_output_fields, &op_output_fields)); 1079 { 1080 bool has_shared_bases = true; 1081 1082 for (CeedInt i = 0; i < num_input_fields; i++) { 1083 CeedBasis basis; 1084 1085 CeedCallBackend(CeedOperatorFieldGetBasis(op_input_fields[i], &basis)); 1086 if (basis != CEED_BASIS_NONE) { 1087 bool is_tensor = true; 1088 const char *resource; 1089 char *resource_root; 1090 Ceed basis_ceed; 1091 1092 CeedCallBackend(CeedBasisIsTensor(basis, &is_tensor)); 1093 is_all_tensor = is_all_tensor && is_tensor; 1094 is_all_nontensor = is_all_nontensor && !is_tensor; 1095 CeedCallBackend(CeedBasisGetCeed(basis, &basis_ceed)); 1096 CeedCallBackend(CeedGetResource(basis_ceed, &resource)); 1097 CeedCallBackend(CeedGetResourceRoot(basis_ceed, resource, ":", &resource_root)); 1098 has_shared_bases = has_shared_bases && !strcmp(resource_root, "/gpu/cuda/shared"); 1099 CeedCallBackend(CeedFree(&resource_root)); 1100 CeedCallBackend(CeedDestroy(&basis_ceed)); 1101 } 1102 CeedCallBackend(CeedBasisDestroy(&basis)); 1103 } 1104 1105 for (CeedInt i = 0; i < num_output_fields; i++) { 1106 CeedBasis basis; 1107 1108 CeedCallBackend(CeedOperatorFieldGetBasis(op_output_fields[i], &basis)); 1109 if (basis != CEED_BASIS_NONE) { 1110 bool is_tensor = true; 1111 const char *resource; 1112 char *resource_root; 1113 Ceed basis_ceed; 1114 1115 CeedCallBackend(CeedBasisIsTensor(basis, &is_tensor)); 1116 is_all_tensor = is_all_tensor && is_tensor; 1117 is_all_nontensor = is_all_nontensor && !is_tensor; 1118 1119 CeedCallBackend(CeedBasisGetCeed(basis, &basis_ceed)); 1120 CeedCallBackend(CeedGetResource(basis_ceed, &resource)); 1121 CeedCallBackend(CeedGetResourceRoot(basis_ceed, resource, ":", &resource_root)); 1122 has_shared_bases = has_shared_bases && !strcmp(resource_root, "/gpu/cuda/shared"); 1123 CeedCallBackend(CeedFree(&resource_root)); 1124 CeedCallBackend(CeedDestroy(&basis_ceed)); 1125 } 1126 CeedCallBackend(CeedBasisDestroy(&basis)); 1127 } 1128 // -- Fallback to ref if not all bases are shared 1129 if (!has_shared_bases) { 1130 *is_good_build = false; 1131 return CEED_ERROR_SUCCESS; 1132 } 1133 } 1134 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 1135 CeedCallBackend(CeedOperatorGetQFunction(op, &qf)); 1136 CeedCallBackend(CeedQFunctionGetData(qf, &qf_data)); 1137 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_input_fields, NULL, &qf_output_fields)); 1138 1139 // Get operator data 1140 CeedCallBackend(CeedOperatorIsAtPoints(op, &is_at_points)); 1141 { 1142 CeedInt max_P, max_P_1d; 1143 1144 CeedCallBackend(CeedOperatorBuildKernelData_Cuda_gen(ceed, num_input_fields, op_input_fields, qf_input_fields, num_output_fields, 1145 op_output_fields, qf_output_fields, &max_P, &max_P_1d, &Q, &Q_1d, &max_dim, &is_all_tensor, 1146 &use_3d_slices)); 1147 data->max_P_1d = is_all_tensor ? max_P_1d : max_P; 1148 } 1149 if (max_dim == 0) max_dim = 1; 1150 data->dim = max_dim; 1151 if (is_at_points) { 1152 CeedElemRestriction_Cuda *rstr_data; 1153 CeedElemRestriction rstr_points = NULL; 1154 1155 CeedCallBackend(CeedOperatorAtPointsGetPoints(op, &rstr_points, NULL)); 1156 CeedCallBackend(CeedElemRestrictionGetMaxPointsInElement(rstr_points, &max_num_points)); 1157 CeedCallBackend(CeedElemRestrictionGetCompStride(rstr_points, &coords_comp_stride)); 1158 CeedCallBackend(CeedElemRestrictionGetData(rstr_points, &rstr_data)); 1159 data->points.indices = (CeedInt *)rstr_data->d_offsets; 1160 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_points)); 1161 } 1162 if (is_at_points) use_3d_slices = false; 1163 if (Q_1d == 0) { 1164 if (is_at_points) Q_1d = max_num_points; 1165 else CeedCallBackend(CeedOperatorGetNumQuadraturePoints(op, &Q_1d)); 1166 } 1167 if (Q == 0) Q = Q_1d; 1168 data->Q = Q; 1169 data->Q_1d = Q_1d; 1170 1171 // Check for restriction only identity operator 1172 { 1173 bool is_identity_qf; 1174 1175 CeedCallBackend(CeedQFunctionIsIdentity(qf, &is_identity_qf)); 1176 if (is_identity_qf) { 1177 CeedEvalMode eval_mode_in, eval_mode_out; 1178 1179 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[0], &eval_mode_in)); 1180 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[0], &eval_mode_out)); 1181 CeedCheck(eval_mode_in != CEED_EVAL_NONE || eval_mode_out != CEED_EVAL_NONE, ceed, CEED_ERROR_BACKEND, 1182 "Backend does not implement restriction only identity operators"); 1183 } 1184 } 1185 1186 // Add atomicAdd function for old NVidia architectures 1187 { 1188 Ceed_Cuda *ceed_data; 1189 struct cudaDeviceProp prop; 1190 1191 CeedCallBackend(CeedGetData(ceed, &ceed_data)); 1192 CeedCallBackend(cudaGetDeviceProperties(&prop, ceed_data->device_id)); 1193 if ((prop.major < 6) && (CEED_SCALAR_TYPE != CEED_SCALAR_FP32)) { 1194 code << "// AtomicAdd fallback source\n"; 1195 code << "#include <ceed/jit-source/cuda/cuda-atomic-add-fallback.h>\n\n"; 1196 } 1197 } 1198 1199 // Load basis source files 1200 if (!is_all_nontensor) { 1201 code << "// Tensor basis source\n"; 1202 code << "#include <ceed/jit-source/cuda/cuda-shared-basis-tensor-templates.h>\n\n"; 1203 } 1204 if (!is_all_tensor) { 1205 code << "// Non-tensor basis source\n"; 1206 code << "#include <ceed/jit-source/cuda/cuda-shared-basis-nontensor-templates.h>\n\n"; 1207 } 1208 if (is_at_points) { 1209 code << "// AtPoints basis source\n"; 1210 code << "#include <ceed/jit-source/cuda/cuda-shared-basis-tensor-at-points-templates.h>\n\n"; 1211 } 1212 code << "// CodeGen operator source\n"; 1213 code << "#include <ceed/jit-source/cuda/cuda-gen-templates.h>\n\n"; 1214 1215 // Get QFunction name 1216 std::string qfunction_name(qf_data->qfunction_name); 1217 std::string operator_name; 1218 1219 operator_name = "CeedKernelCudaGenOperator_" + qfunction_name; 1220 1221 // Define CEED_Q_VLA 1222 code << "\n#undef CEED_Q_VLA\n"; 1223 if (max_dim != 3 || is_at_points || use_3d_slices || !is_all_tensor) { 1224 code << "#define CEED_Q_VLA 1\n\n"; 1225 } else { 1226 code << "#define CEED_Q_VLA " << Q_1d << "\n\n"; 1227 } 1228 1229 // Add user QFunction source 1230 { 1231 const char *source_path; 1232 1233 CeedCallBackend(CeedQFunctionGetSourcePath(qf, &source_path)); 1234 CeedCheck(source_path, ceed, CEED_ERROR_UNSUPPORTED, "/gpu/cuda/gen backend requires QFunction source code file"); 1235 1236 code << "// User QFunction source\n"; 1237 code << "#include \"" << source_path << "\"\n\n"; 1238 } 1239 1240 // Setup 1241 code << "\n// -----------------------------------------------------------------------------\n"; 1242 code << "// Operator Kernel\n"; 1243 code << "// \n"; 1244 code << "// d_[in,out]_i: CeedVector device array\n"; 1245 code << "// r_[in,out]_e_i: Element vector register\n"; 1246 code << "// r_[in,out]_q_i: Quadrature space vector register\n"; 1247 code << "// r_[in,out]_c_i: AtPoints Chebyshev coefficients register\n"; 1248 code << "// r_[in,out]_s_i: Quadrature space slice vector register\n"; 1249 code << "// \n"; 1250 code << "// s_B_[in,out]_i: Interpolation matrix, shared memory\n"; 1251 code << "// s_G_[in,out]_i: Gradient matrix, shared memory\n"; 1252 code << "// -----------------------------------------------------------------------------\n"; 1253 code << "extern \"C\" __global__ void " << operator_name 1254 << "(CeedInt num_elem, void* ctx, FieldsInt_Cuda indices, Fields_Cuda fields, Fields_Cuda B, Fields_Cuda G, CeedScalar *W, Points_Cuda " 1255 "points) {\n"; 1256 1257 // Scratch buffers 1258 for (CeedInt i = 0; i < num_input_fields; i++) { 1259 CeedEvalMode eval_mode; 1260 1261 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode)); 1262 if (eval_mode != CEED_EVAL_WEIGHT) { // Skip CEED_EVAL_WEIGHT 1263 code << " const CeedScalar *__restrict__ d_in_" << i << " = fields.inputs[" << i << "];\n"; 1264 } 1265 } 1266 for (CeedInt i = 0; i < num_output_fields; i++) { 1267 code << " CeedScalar *__restrict__ d_out_" << i << " = fields.outputs[" << i << "];\n"; 1268 } 1269 1270 code << " const CeedInt max_dim = " << max_dim << ";\n"; 1271 if (!is_all_tensor) { 1272 code << " const CeedInt Q = " << Q << ";\n"; 1273 } 1274 if (!is_all_nontensor) { 1275 code << " const CeedInt Q_1d = " << Q_1d << ";\n"; 1276 } 1277 if (is_at_points) { 1278 code << " const CeedInt max_num_points = " << max_num_points << ";\n"; 1279 code << " const CeedInt coords_comp_stride = " << coords_comp_stride << ";\n"; 1280 } 1281 1282 // Shared data 1283 code << " extern __shared__ CeedScalar slice[];\n"; 1284 code << " SharedData_Cuda data;\n"; 1285 code << " data.t_id_x = threadIdx.x;\n"; 1286 code << " data.t_id_y = threadIdx.y;\n"; 1287 code << " data.t_id_z = threadIdx.z;\n"; 1288 code << " data.t_id = threadIdx.x + threadIdx.y*blockDim.x + threadIdx.z*blockDim.y*blockDim.x;\n"; 1289 code << " data.slice = slice + data.t_id_z*OP_T_1D" << ((!is_all_tensor || max_dim == 1) ? "" : "*OP_T_1D") << ";\n"; 1290 1291 // -- Determine input mat reuse 1292 FieldReuse_Cuda input_matrix_reuse[CEED_FIELD_MAX]; 1293 1294 for (CeedInt i = 0; i < num_input_fields; i++) { 1295 input_matrix_reuse[i].index = -1; 1296 } 1297 for (CeedInt i = 0; i < num_input_fields; i++) { 1298 bool is_tensor = true; 1299 CeedEvalMode eval_mode_i; 1300 CeedBasis basis_i; 1301 1302 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode_i)); 1303 if (eval_mode_i == CEED_EVAL_WEIGHT) continue; 1304 CeedCallBackend(CeedOperatorFieldGetBasis(op_input_fields[i], &basis_i)); 1305 CeedCallBackend(CeedBasisIsTensor(basis_i, &is_tensor)); 1306 for (CeedInt j = 0; (input_matrix_reuse[i].index == -1) && (j < i); j++) { 1307 CeedEvalMode eval_mode_j; 1308 CeedBasis basis_j; 1309 1310 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[j], &eval_mode_j)); 1311 if (eval_mode_j == CEED_EVAL_WEIGHT) continue; 1312 CeedCallBackend(CeedOperatorFieldGetBasis(op_input_fields[j], &basis_j)); 1313 if (basis_i == basis_j) { 1314 if (is_tensor) { 1315 input_matrix_reuse[i].index = j; 1316 input_matrix_reuse[i].is_input = true; 1317 input_matrix_reuse[i].eval_mode = eval_mode_j; 1318 } else { 1319 // For non-tensor can only re-use with the same eval mode 1320 if (eval_mode_i == eval_mode_j) { 1321 input_matrix_reuse[i].index = j; 1322 input_matrix_reuse[i].is_input = true; 1323 input_matrix_reuse[i].eval_mode = eval_mode_j; 1324 } 1325 } 1326 } 1327 CeedCallBackend(CeedBasisDestroy(&basis_j)); 1328 } 1329 CeedCallBackend(CeedBasisDestroy(&basis_i)); 1330 } 1331 1332 // -- Determine output mat reuse 1333 FieldReuse_Cuda output_matrix_reuse[CEED_FIELD_MAX]; 1334 1335 for (CeedInt i = 0; i < num_output_fields; i++) { 1336 output_matrix_reuse[i].index = -1; 1337 } 1338 for (CeedInt i = 0; i < num_output_fields; i++) { 1339 bool is_tensor = true; 1340 CeedEvalMode eval_mode_i; 1341 CeedBasis basis_i; 1342 1343 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode_i)); 1344 CeedCallBackend(CeedOperatorFieldGetBasis(op_output_fields[i], &basis_i)); 1345 CeedCallBackend(CeedBasisIsTensor(basis_i, &is_tensor)); 1346 for (CeedInt j = 0; (output_matrix_reuse[i].index == -1) && (j < num_input_fields); j++) { 1347 CeedEvalMode eval_mode_j; 1348 CeedBasis basis_j; 1349 1350 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[j], &eval_mode_j)); 1351 if (eval_mode_j == CEED_EVAL_WEIGHT) continue; 1352 CeedCallBackend(CeedOperatorFieldGetBasis(op_input_fields[j], &basis_j)); 1353 if (basis_i == basis_j) { 1354 if (is_tensor) { 1355 output_matrix_reuse[i].index = j; 1356 output_matrix_reuse[i].is_input = true; 1357 output_matrix_reuse[i].eval_mode = eval_mode_j; 1358 } else { 1359 // For non-tensor can only re-use with the same eval mode 1360 if (eval_mode_i == eval_mode_j) { 1361 output_matrix_reuse[i].index = j; 1362 output_matrix_reuse[i].is_input = true; 1363 output_matrix_reuse[i].eval_mode = eval_mode_j; 1364 } 1365 } 1366 } 1367 CeedCallBackend(CeedBasisDestroy(&basis_j)); 1368 } 1369 for (CeedInt j = 0; (output_matrix_reuse[i].index == -1) && (j < i); j++) { 1370 CeedEvalMode eval_mode_j; 1371 CeedBasis basis_j; 1372 1373 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[j], &eval_mode_j)); 1374 if (eval_mode_j == CEED_EVAL_WEIGHT) continue; 1375 CeedCallBackend(CeedOperatorFieldGetBasis(op_output_fields[j], &basis_j)); 1376 if (basis_i == basis_j) { 1377 if (is_tensor) { 1378 output_matrix_reuse[i].index = j; 1379 output_matrix_reuse[i].is_input = false; 1380 output_matrix_reuse[i].eval_mode = eval_mode_j; 1381 } else { 1382 // For non-tensor can only re-use with the same eval mode 1383 if (eval_mode_i == eval_mode_j) { 1384 output_matrix_reuse[i].index = j; 1385 output_matrix_reuse[i].is_input = false; 1386 output_matrix_reuse[i].eval_mode = eval_mode_j; 1387 } 1388 } 1389 } 1390 CeedCallBackend(CeedBasisDestroy(&basis_j)); 1391 } 1392 CeedCallBackend(CeedBasisDestroy(&basis_i)); 1393 } 1394 1395 // Initialize constants, and matrices B and G 1396 code << "\n // Input field constants and basis data\n"; 1397 for (CeedInt i = 0; i < num_input_fields; i++) { 1398 CeedCallBackend(CeedOperatorBuildKernelFieldData_Cuda_gen(code, data, i, op_input_fields[i], qf_input_fields[i], input_matrix_reuse[i], Q, Q_1d, 1399 true, is_all_tensor, is_at_points, use_3d_slices)); 1400 } 1401 code << "\n // Output field constants and basis data\n"; 1402 for (CeedInt i = 0; i < num_output_fields; i++) { 1403 CeedCallBackend(CeedOperatorBuildKernelFieldData_Cuda_gen(code, data, i, op_output_fields[i], qf_output_fields[i], output_matrix_reuse[i], Q, 1404 Q_1d, false, is_all_tensor, is_at_points, use_3d_slices)); 1405 } 1406 1407 // Loop over all elements 1408 code << "\n // Element loop\n"; 1409 code << " __syncthreads();\n"; 1410 code << " for (CeedInt elem = blockIdx.x*blockDim.z + threadIdx.z; elem < num_elem; elem += gridDim.x*blockDim.z) {\n"; 1411 1412 // -- Compute minimum buffer space needed 1413 CeedInt max_rstr_buffer_size = 1; 1414 1415 for (CeedInt i = 0; i < num_input_fields; i++) { 1416 CeedInt num_comp, elem_size; 1417 CeedElemRestriction elem_rstr; 1418 1419 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_input_fields[i], &elem_rstr)); 1420 CeedCallBackend(CeedElemRestrictionGetNumComponents(elem_rstr, &num_comp)); 1421 CeedCallBackend(CeedElemRestrictionGetElementSize(elem_rstr, &elem_size)); 1422 max_rstr_buffer_size = CeedIntMax(max_rstr_buffer_size, num_comp * (is_all_tensor && (max_dim >= 3) ? elem_size : 1)); 1423 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 1424 } 1425 for (CeedInt i = 0; i < num_output_fields; i++) { 1426 CeedInt num_comp, elem_size; 1427 CeedElemRestriction elem_rstr; 1428 1429 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_output_fields[i], &elem_rstr)); 1430 CeedCallBackend(CeedElemRestrictionGetNumComponents(elem_rstr, &num_comp)); 1431 CeedCallBackend(CeedElemRestrictionGetElementSize(elem_rstr, &elem_size)); 1432 max_rstr_buffer_size = CeedIntMax(max_rstr_buffer_size, num_comp * (is_all_tensor && (max_dim >= 3) ? elem_size : 1)); 1433 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 1434 } 1435 code << " // Scratch restriction buffer space\n"; 1436 code << " CeedScalar r_e_scratch[" << max_rstr_buffer_size << "];\n"; 1437 1438 // -- Determine best input field processing order 1439 CeedInt field_rstr_in_buffer[CEED_FIELD_MAX], input_field_order[CEED_FIELD_MAX]; 1440 1441 for (CeedInt i = 0; i < num_input_fields; i++) { 1442 field_rstr_in_buffer[i] = -1; 1443 input_field_order[i] = -1; 1444 } 1445 { 1446 bool is_ordered[CEED_FIELD_MAX]; 1447 CeedInt curr_index = 0; 1448 1449 for (CeedInt i = 0; i < num_input_fields; i++) is_ordered[i] = false; 1450 for (CeedInt i = 0; i < num_input_fields; i++) { 1451 CeedVector vec_i; 1452 CeedElemRestriction rstr_i; 1453 1454 if (is_ordered[i]) continue; 1455 field_rstr_in_buffer[i] = i; 1456 is_ordered[i] = true; 1457 input_field_order[curr_index] = i; 1458 curr_index++; 1459 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[i], &vec_i)); 1460 if (vec_i == CEED_VECTOR_NONE) continue; // CEED_EVAL_WEIGHT 1461 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_input_fields[i], &rstr_i)); 1462 for (CeedInt j = i + 1; j < num_input_fields; j++) { 1463 CeedVector vec_j; 1464 CeedElemRestriction rstr_j; 1465 1466 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[j], &vec_j)); 1467 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_input_fields[j], &rstr_j)); 1468 if (rstr_i == rstr_j && vec_i == vec_j) { 1469 field_rstr_in_buffer[j] = i; 1470 is_ordered[j] = true; 1471 input_field_order[curr_index] = j; 1472 curr_index++; 1473 } 1474 CeedCallBackend(CeedVectorDestroy(&vec_j)); 1475 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_j)); 1476 } 1477 CeedCallBackend(CeedVectorDestroy(&vec_i)); 1478 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_i)); 1479 } 1480 } 1481 1482 // -- Input restriction and basis 1483 code << "\n // -- Input field restrictions and basis actions\n"; 1484 for (CeedInt i = 0; i < num_input_fields; i++) { 1485 const char *field_name; 1486 const CeedInt f = input_field_order[i]; 1487 1488 CeedCallBackend(CeedOperatorFieldGetName(op_input_fields[f], &field_name)); 1489 code << " // ---- Input field " << f << ": " << field_name << "\n"; 1490 1491 // ---- Restriction 1492 CeedCallBackend(CeedOperatorBuildKernelRestriction_Cuda_gen(code, data, f, max_dim, field_rstr_in_buffer, op_input_fields[f], qf_input_fields[f], 1493 Q_1d, true, is_all_tensor, is_at_points, use_3d_slices)); 1494 1495 // ---- Basis action 1496 CeedCallBackend(CeedOperatorBuildKernelBasis_Cuda_gen(code, data, f, max_dim, op_input_fields[f], qf_input_fields[f], Q_1d, true, is_all_tensor, 1497 is_at_points, use_3d_slices)); 1498 } 1499 1500 // -- Q function 1501 CeedCallBackend(CeedOperatorBuildKernelQFunction_Cuda_gen(code, data, max_dim, max_num_points, num_input_fields, op_input_fields, qf_input_fields, 1502 num_output_fields, op_output_fields, qf_output_fields, qfunction_name, Q_1d, 1503 is_all_tensor, is_at_points, use_3d_slices)); 1504 1505 // -- Output basis and restriction 1506 code << "\n // -- Output field basis action and restrictions\n"; 1507 for (CeedInt i = 0; i < num_output_fields; i++) { 1508 const char *field_name; 1509 1510 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 1511 code << " // ---- Output field " << i << ": " << field_name << "\n"; 1512 1513 // ---- Basis action 1514 CeedCallBackend(CeedOperatorBuildKernelBasis_Cuda_gen(code, data, i, max_dim, op_output_fields[i], qf_output_fields[i], Q_1d, false, 1515 is_all_tensor, is_at_points, use_3d_slices)); 1516 1517 // ---- Restriction 1518 CeedCallBackend(CeedOperatorBuildKernelRestriction_Cuda_gen(code, data, i, max_dim, NULL, op_output_fields[i], qf_output_fields[i], Q_1d, false, 1519 is_all_tensor, is_at_points, use_3d_slices)); 1520 } 1521 1522 // Close loop and function 1523 code << " }\n"; 1524 code << "}\n"; 1525 code << "// -----------------------------------------------------------------------------\n\n"; 1526 1527 // Compile 1528 { 1529 bool is_compile_good = false; 1530 const CeedInt T_1d = CeedIntMax(is_all_tensor ? Q_1d : Q, data->max_P_1d); 1531 1532 CeedCallBackend(CeedTryCompile_Cuda(ceed, code.str().c_str(), &is_compile_good, &data->module, 1, "OP_T_1D", T_1d)); 1533 if (is_compile_good) { 1534 *is_good_build = true; 1535 CeedCallBackend(CeedGetKernel_Cuda(ceed, data->module, operator_name.c_str(), &data->op)); 1536 } else { 1537 *is_good_build = false; 1538 data->use_fallback = true; 1539 } 1540 } 1541 CeedCallBackend(CeedOperatorSetSetupDone(op)); 1542 CeedCallBackend(CeedDestroy(&ceed)); 1543 CeedCallBackend(CeedQFunctionDestroy(&qf)); 1544 return CEED_ERROR_SUCCESS; 1545 } 1546 1547 //------------------------------------------------------------------------------ 1548