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_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_1d : 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"; 531 532 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*" << (is_tensor && (dim >= 3) ? Q_name : "1") << "];\n"; 533 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", " << (P_1d > Q_1d ? P_name : Q_name) 534 << ">(data, r_e" << var_suffix << ", s_B" << var_suffix << ", r_q" << var_suffix << ");\n"; 535 } 536 break; 537 case CEED_EVAL_GRAD: 538 if (is_at_points) { 539 std::string function_name = (dim == 1 ? "Interp" : "InterpTensor") + std::to_string(dim) + "d"; 540 541 code << " CeedScalar r_c" << var_suffix << "[num_comp" << var_suffix << "*" << (dim >= 3 ? Q_name : "1") << "];\n"; 542 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", OP_T_1D>(data, r_e" << var_suffix 543 << ", s_B" << var_suffix << ", r_c" << var_suffix << ");\n"; 544 } else if (use_3d_slices) { 545 std::string function_name = (dim > 1 ? "InterpTensor" : "Interp") + std::to_string(dim) + "d"; 546 547 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*" << Q_name << "];\n"; 548 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", OP_T_1D>(data, r_e" << var_suffix 549 << ", s_B" << var_suffix << ", r_q" << var_suffix << ");\n"; 550 } else if (is_tensor) { 551 bool is_collocated = dim == 3 && Q_1d >= P_1d; 552 std::string function_name = (dim == 1 ? "Grad" : (is_collocated ? "GradTensorCollocated" : "GradTensor")) + std::to_string(dim) + "d" + 553 (is_all_tensor ? "" : "Flattened"); 554 555 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*dim" << var_suffix << "*" << (dim >= 3 ? Q_name : "1") 556 << "];\n"; 557 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", " << (P_1d > Q_1d ? P_name : Q_name) 558 << ">(data, r_e" << var_suffix << ", s_B" << var_suffix << ", s_G" << var_suffix << ", r_q" << var_suffix << ");\n"; 559 } else { 560 std::string function_name = "GradNonTensor"; 561 562 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*dim" << var_suffix << "];\n"; 563 code << " " << function_name << "<num_comp" << var_suffix << ", dim" << var_suffix << ", " << P_name << ", " << Q_name << ", " 564 << (P_1d > Q_1d ? P_name : Q_name) << ">(data, r_e" << var_suffix << ", s_G" << var_suffix << ", r_q" << var_suffix << ");\n"; 565 } 566 break; 567 case CEED_EVAL_WEIGHT: { 568 if (is_at_points) { 569 code << " // Nothing to do AtPoints\n"; 570 } else { 571 CeedBasis_Cuda_shared *basis_data; 572 std::string function_name = is_tensor 573 ? ((dim == 1 ? "Weight" : "WeightTensor") + std::to_string(dim) + "d" + (is_all_tensor ? "" : "Flattened")) 574 : "WeightNonTensor"; 575 576 code << " CeedScalar r_q" << var_suffix << "[" << (is_tensor && (dim >= 3) ? Q_name : "1") << "];\n"; 577 CeedCallBackend(CeedBasisGetData(basis, &basis_data)); 578 data->W = basis_data->d_q_weight_1d; 579 code << " " << function_name << "<" << P_name << ", " << Q_name << ">(data, W, r_q" << var_suffix << ");\n"; 580 } 581 break; 582 } 583 // LCOV_EXCL_START 584 case CEED_EVAL_DIV: 585 case CEED_EVAL_CURL: 586 break; // TODO: Not implemented 587 // LCOV_EXCL_STOP 588 } 589 } else { 590 switch (eval_mode) { 591 case CEED_EVAL_NONE: 592 code << " CeedScalar *r_e" << var_suffix << " = r_q" << var_suffix << ";\n"; 593 break; // No action 594 case CEED_EVAL_INTERP: 595 code << " CeedScalar *r_e" << var_suffix << " = r_e_scratch;\n"; 596 if (is_at_points) { 597 std::string function_name = (dim == 1 ? "InterpTranspose" : "InterpTransposeTensor") + std::to_string(dim) + "d"; 598 599 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", OP_T_1D>(data, r_c" << var_suffix 600 << ", s_B" << var_suffix << ", r_e" << var_suffix << ");\n"; 601 } else { 602 std::string function_name = 603 is_tensor ? ((dim == 1 ? "InterpTranspose" : "InterpTransposeTensor") + std::to_string(dim) + "d" + (is_all_tensor ? "" : "Flattened")) 604 : "InterpTransposeNonTensor"; 605 606 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", " << (P_1d > Q_1d ? P_name : Q_name) 607 << ">(data, r_q" << var_suffix << ", s_B" << var_suffix << ", r_e" << var_suffix << ");\n"; 608 } 609 break; 610 case CEED_EVAL_GRAD: 611 code << " CeedScalar *r_e" << var_suffix << " = r_e_scratch;\n"; 612 if (is_at_points) { 613 std::string function_name = (dim == 1 ? "InterpTranspose" : "InterpTransposeTensor") + std::to_string(dim) + "d"; 614 615 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", OP_T_1D>(data, r_c" << var_suffix 616 << ", s_B" << var_suffix << ", r_e" << var_suffix << ");\n"; 617 } else if (use_3d_slices) { 618 std::string function_name = (dim == 1 ? "InterpTranspose" : "InterpTransposeTensor") + std::to_string(dim) + "d"; 619 620 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", OP_T_1D>(data, r_q" << var_suffix 621 << ", s_B" << var_suffix << ", r_e" << var_suffix << ");\n"; 622 } else if (is_tensor) { 623 bool is_collocated = dim == 3 && Q_1d >= P_1d; 624 std::string function_name = (dim == 1 ? "GradTranspose" : (is_collocated ? "GradTransposeTensorCollocated" : "GradTransposeTensor")) + 625 std::to_string(dim) + "d" + (is_all_tensor ? "" : "Flattened"); 626 627 code << " " << function_name << "<num_comp" << var_suffix << ", " << P_name << ", " << Q_name << ", " << (P_1d > Q_1d ? P_name : Q_name) 628 << ">(data, r_q" << var_suffix << ", s_B" << var_suffix << ", s_G" << var_suffix << ", r_e" << var_suffix << ");\n"; 629 } else { 630 std::string function_name = "GradTransposeNonTensor"; 631 632 code << " " << function_name << "<num_comp" << var_suffix << ", dim" << var_suffix << ", " << P_name << ", " << Q_name << ", " 633 << (P_1d > Q_1d ? P_name : Q_name) << ">(data, r_q" << var_suffix << ", s_G" << var_suffix << ", r_e" << var_suffix << ");\n"; 634 } 635 break; 636 // LCOV_EXCL_START 637 case CEED_EVAL_WEIGHT: 638 break; // Should not occur 639 case CEED_EVAL_DIV: 640 case CEED_EVAL_CURL: 641 break; // TODO: Not implemented 642 // LCOV_EXCL_STOP 643 } 644 } 645 CeedCallBackend(CeedBasisDestroy(&basis)); 646 return CEED_ERROR_SUCCESS; 647 } 648 649 //------------------------------------------------------------------------------ 650 // QFunction 651 //------------------------------------------------------------------------------ 652 static int CeedOperatorBuildKernelQFunction_Cuda_gen(std::ostringstream &code, CeedOperator_Cuda_gen *data, CeedInt max_dim, CeedInt max_num_points, 653 CeedInt num_input_fields, CeedOperatorField *op_input_fields, 654 CeedQFunctionField *qf_input_fields, CeedInt num_output_fields, 655 CeedOperatorField *op_output_fields, CeedQFunctionField *qf_output_fields, 656 std::string qfunction_name, CeedInt Q_1d, bool is_all_tensor, bool is_at_points, 657 bool use_3d_slices) { 658 std::string Q_name = is_all_tensor ? "Q_1d" : "Q"; 659 CeedEvalMode eval_mode = CEED_EVAL_NONE; 660 CeedElemRestriction elem_rstr; 661 662 // Setup output arrays 663 code << "\n // -- Output field setup\n"; 664 for (CeedInt i = 0; i < num_output_fields; i++) { 665 const char *field_name; 666 std::string var_suffix = "_out_" + std::to_string(i); 667 668 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 669 code << " // ---- Output field " << i << ": " << field_name << "\n"; 670 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 671 switch (eval_mode) { 672 case CEED_EVAL_NONE: 673 if (is_at_points) { 674 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "];\n"; 675 } else { 676 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*" << (is_all_tensor && (max_dim >= 3) ? Q_name : "1") 677 << "];\n"; 678 } 679 break; 680 case CEED_EVAL_INTERP: 681 if (is_at_points) { 682 // Accumulator for point data 683 code << " CeedScalar r_c" << var_suffix << "[num_comp" << var_suffix << "*" << (max_dim >= 3 ? Q_name : "1") << "];\n"; 684 code << " for (CeedInt i = 0; i < num_comp" << var_suffix << "*" << (max_dim >= 3 ? Q_name : "1") << "; i++) {\n"; 685 code << " r_c" << var_suffix << "[i] = 0.0;\n"; 686 code << " }\n"; 687 } else { 688 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*" << (is_all_tensor && (max_dim >= 3) ? Q_name : "1") 689 << "];\n"; 690 } 691 break; 692 case CEED_EVAL_GRAD: 693 if (is_at_points) { 694 // Accumulator for point data 695 code << " CeedScalar r_c" << var_suffix << "[num_comp" << var_suffix << "*" << (max_dim >= 3 ? Q_name : "1") << "*dim" << var_suffix 696 << "];\n"; 697 code << " for (CeedInt i = 0; i < num_comp" << var_suffix << "*" << (max_dim >= 3 ? Q_name : "1") << "; i++) {\n"; 698 code << " r_c" << var_suffix << "[i] = 0.0;\n"; 699 code << " }\n"; 700 } else if (use_3d_slices) { 701 // Accumulator for gradient slices 702 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*" << Q_name << "];\n"; 703 code << " for (CeedInt i = 0; i < num_comp" << var_suffix << "*" << Q_name << "; i++) {\n"; 704 code << " r_q" << var_suffix << "[i] = 0.0;\n"; 705 code << " }\n"; 706 } else { 707 code << " CeedScalar r_q" << var_suffix << "[num_comp" << var_suffix << "*dim" << var_suffix << "*" 708 << (is_all_tensor && (max_dim >= 3) ? Q_name : "1") << "];\n"; 709 } 710 break; 711 case CEED_EVAL_WEIGHT: 712 break; 713 // LCOV_EXCL_START 714 case CEED_EVAL_DIV: 715 case CEED_EVAL_CURL: 716 break; // TODO: Not implemented 717 // LCOV_EXCL_STOP 718 } 719 } 720 721 if (is_at_points) { 722 // We need to handle batches of points 723 code << "\n // Note: Using batches of points\n"; 724 code << " const CeedInt point_loop_bound = (blockDim.x * blockDim.y) * ceil(1.0 * max_num_points / (blockDim.x * blockDim.y));\n\n"; 725 code << " #pragma unroll\n"; 726 code << " for (CeedInt i = threadIdx.x + threadIdx.y * blockDim.x; i < point_loop_bound; i += blockDim.x * blockDim.y) {\n"; 727 code << " const CeedInt p = i % max_num_points;\n\n"; 728 729 code << " // -- Coordinates\n"; 730 code << " CeedScalar r_x[max_dim];\n"; 731 code << " ReadPoint<max_dim, coords_comp_stride, max_num_points>(data, elem, p, max_num_points, points.indices, points.coords, r_x);\n\n"; 732 733 code << " // -- Input fields\n"; 734 for (CeedInt i = 0; i < num_input_fields; i++) { 735 const char *field_name; 736 std::string var_suffix = "_in_" + std::to_string(i); 737 std::string P_name = "P_1d" + var_suffix; 738 739 CeedCallBackend(CeedOperatorFieldGetName(op_input_fields[i], &field_name)); 740 code << " // ---- Input field " << i << ": " << field_name << "\n"; 741 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode)); 742 // Basis action 743 code << " // EvalMode: " << CeedEvalModes[eval_mode] << "\n"; 744 switch (eval_mode) { 745 case CEED_EVAL_NONE: 746 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 747 code << " ReadPoint<num_comp" << var_suffix << ", comp_stride" << var_suffix 748 << ", max_num_points>(data, elem, p, max_num_points, indices.inputs[" << i << "], d" << var_suffix << ", r_s" << var_suffix << ");\n"; 749 break; 750 case CEED_EVAL_INTERP: 751 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 752 code << " InterpAtPoints" << max_dim << "d<num_comp" << var_suffix << ", max_num_points, " << P_name << ", " << Q_name 753 << ">(data, i, r_c" << var_suffix << ", r_x, r_s" << var_suffix << ");\n"; 754 break; 755 case CEED_EVAL_GRAD: 756 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "*dim" << var_suffix << "];\n"; 757 code << " GradAtPoints" << max_dim << "d<num_comp" << var_suffix << ", max_num_points, " << P_name << ", " << Q_name 758 << ">(data, i, r_c" << var_suffix << ", r_x, r_s" << var_suffix << ");\n"; 759 break; 760 case CEED_EVAL_WEIGHT: 761 code << " CeedScalar r_s" << var_suffix << "[1];\n"; 762 code << " r_s" << var_suffix << "[0] = 1.0;\n"; 763 break; 764 // LCOV_EXCL_START 765 case CEED_EVAL_DIV: 766 case CEED_EVAL_CURL: 767 break; // TODO: Not implemented 768 // LCOV_EXCL_STOP 769 } 770 } 771 code << "\n // -- Output fields\n"; 772 for (CeedInt i = 0; i < num_output_fields; i++) { 773 const char *field_name; 774 std::string var_suffix = "_out_" + std::to_string(i); 775 776 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 777 code << " // ---- Output field " << i << ": " << field_name << "\n"; 778 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 779 // Basis action 780 switch (eval_mode) { 781 case CEED_EVAL_NONE: 782 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 783 break; 784 case CEED_EVAL_INTERP: 785 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 786 break; 787 case CEED_EVAL_GRAD: 788 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "*dim" << var_suffix << "];\n"; 789 break; 790 // LCOV_EXCL_START 791 case CEED_EVAL_WEIGHT: 792 break; // Should not occur 793 case CEED_EVAL_DIV: 794 case CEED_EVAL_CURL: 795 break; // TODO: Not implemented 796 // LCOV_EXCL_STOP 797 } 798 } 799 800 } else if (use_3d_slices) { 801 // We treat quadrature points per slice in 3d to save registers 802 code << "\n // Note: Using planes of 3D elements\n"; 803 code << " #pragma unroll\n"; 804 code << " for (CeedInt q = 0; q < " << Q_name << "; q++) {\n"; 805 code << " // -- Input fields\n"; 806 for (CeedInt i = 0; i < num_input_fields; i++) { 807 const char *field_name; 808 std::string var_suffix = "_in_" + std::to_string(i); 809 810 CeedCallBackend(CeedOperatorFieldGetName(op_input_fields[i], &field_name)); 811 code << " // ---- Input field " << i << ": " << field_name << "\n"; 812 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode)); 813 // Basis action 814 code << " // EvalMode: " << CeedEvalModes[eval_mode] << "\n"; 815 switch (eval_mode) { 816 case CEED_EVAL_NONE: 817 bool is_strided; 818 819 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 820 821 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_input_fields[i], &elem_rstr)); 822 CeedCallBackend(CeedElemRestrictionIsStrided(elem_rstr, &is_strided)); 823 if (is_strided) { 824 bool has_backend_strides; 825 CeedInt num_elem, elem_size; 826 827 CeedCallBackend(CeedElemRestrictionGetElementSize(elem_rstr, &elem_size)); 828 CeedCallBackend(CeedElemRestrictionHasBackendStrides(elem_rstr, &has_backend_strides)); 829 CeedCallBackend(CeedElemRestrictionGetNumElements(elem_rstr, &num_elem)); 830 CeedInt strides[3] = {1, elem_size * num_elem, elem_size}; 831 832 if (!has_backend_strides) { 833 CeedCallBackend(CeedElemRestrictionGetStrides(elem_rstr, strides)); 834 } 835 code << " // Strides: {" << strides[0] << ", " << strides[1] << ", " << strides[2] << "}\n"; 836 code << " ReadEVecSliceStrided3d<num_comp" << var_suffix << ", " << Q_name << ", " << strides[0] << ", " << strides[1] << ", " 837 << strides[2] << ">(data, elem, q, d" << var_suffix << ", r_s" << var_suffix << ");\n"; 838 } else { 839 CeedSize l_size = 0; 840 CeedInt comp_stride; 841 CeedElemRestriction_Cuda *rstr_data; 842 843 CeedCallBackend(CeedElemRestrictionGetLVectorSize(elem_rstr, &l_size)); 844 code << " const CeedInt l_size" << var_suffix << " = " << l_size << ";\n"; 845 CeedCallBackend(CeedElemRestrictionGetCompStride(elem_rstr, &comp_stride)); 846 code << " // CompStride: " << comp_stride << "\n"; 847 CeedCallBackend(CeedElemRestrictionGetData(elem_rstr, &rstr_data)); 848 data->indices.inputs[i] = (CeedInt *)rstr_data->d_offsets; 849 code << " ReadEVecSliceStandard3d<num_comp" << var_suffix << ", " << comp_stride << ", " << Q_name << ">(data, l_size" << var_suffix 850 << ", elem, q, indices.inputs[" << i << "], d" << var_suffix << ", r_s" << var_suffix << ");\n"; 851 } 852 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 853 break; 854 case CEED_EVAL_INTERP: 855 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 856 code << " for (CeedInt j = 0; j < num_comp" << var_suffix << "; j++) {\n"; 857 code << " r_s" << var_suffix << "[j] = r_q" << var_suffix << "[q + j*" << Q_name << "];\n"; 858 code << " }\n"; 859 break; 860 case CEED_EVAL_GRAD: 861 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "*dim" << var_suffix << "];\n"; 862 code << " GradColloSlice3d<num_comp" << var_suffix << ", " << Q_name << ", OP_T_1D>(data, q, r_q" << var_suffix << ", s_G" 863 << var_suffix << ", r_s" << var_suffix << ");\n"; 864 break; 865 case CEED_EVAL_WEIGHT: 866 code << " CeedScalar r_s" << var_suffix << "[1];\n"; 867 code << " r_s" << var_suffix << "[0] = r_q" << var_suffix << "[q];\n"; 868 break; 869 // LCOV_EXCL_START 870 case CEED_EVAL_DIV: 871 case CEED_EVAL_CURL: 872 break; // TODO: Not implemented 873 // LCOV_EXCL_STOP 874 } 875 } 876 code << "\n // -- Output fields\n"; 877 for (CeedInt i = 0; i < num_output_fields; i++) { 878 const char *field_name; 879 std::string var_suffix = "_out_" + std::to_string(i); 880 881 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 882 code << " // ---- Output field " << i << ": " << field_name << "\n"; 883 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 884 // Basis action 885 switch (eval_mode) { 886 case CEED_EVAL_NONE: 887 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 888 break; 889 case CEED_EVAL_INTERP: 890 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "];\n"; 891 break; 892 case CEED_EVAL_GRAD: 893 code << " CeedScalar r_s" << var_suffix << "[num_comp" << var_suffix << "*dim" << var_suffix << "];\n"; 894 break; 895 // LCOV_EXCL_START 896 case CEED_EVAL_WEIGHT: 897 break; // Should not occur 898 case CEED_EVAL_DIV: 899 case CEED_EVAL_CURL: 900 break; // TODO: Not implemented 901 // LCOV_EXCL_STOP 902 } 903 } 904 } else { 905 code << "\n // Note: Using full elements\n"; 906 code << " {\n"; 907 code << " // -- Input fields\n"; 908 for (CeedInt i = 0; i < num_input_fields; i++) { 909 const char *field_name; 910 911 CeedCallBackend(CeedOperatorFieldGetName(op_input_fields[i], &field_name)); 912 code << " // ---- Input field " << i << ": " << field_name << "\n"; 913 code << " CeedScalar *r_s_in_" << i << " = r_q_in_" << i << ";\n"; 914 } 915 code << " // -- Output fields\n"; 916 for (CeedInt i = 0; i < num_output_fields; i++) { 917 const char *field_name; 918 919 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 920 code << " // ---- Output field " << i << ": " << field_name << "\n"; 921 code << " CeedScalar *r_s_out_" << i << " = r_q_out_" << i << ";\n"; 922 } 923 } 924 925 // Input and output buffers 926 code << "\n // -- QFunction inputs and outputs\n"; 927 code << " // ---- Inputs\n"; 928 code << " CeedScalar *inputs[" << CeedIntMax(num_input_fields, 1) << "];\n"; 929 for (CeedInt i = 0; i < num_input_fields; i++) { 930 const char *field_name; 931 932 CeedCallBackend(CeedOperatorFieldGetName(op_input_fields[i], &field_name)); 933 code << " // ------ Input field " << i << ": " << field_name << "\n"; 934 code << " inputs[" << i << "] = r_s_in_" << i << ";\n"; 935 } 936 code << " // ---- Outputs\n"; 937 code << " CeedScalar *outputs[" << CeedIntMax(num_output_fields, 1) << "];\n"; 938 for (CeedInt i = 0; i < num_output_fields; i++) { 939 const char *field_name; 940 941 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 942 code << " // ------ Output field " << i << ": " << field_name << "\n"; 943 code << " outputs[" << i << "] = r_s_out_" << i << ";\n"; 944 } 945 946 // Apply QFunction 947 code << "\n // -- Apply QFunction\n"; 948 code << " " << qfunction_name << "(ctx, "; 949 if (max_dim != 3 || is_at_points || use_3d_slices || !is_all_tensor) { 950 code << "1"; 951 } else { 952 code << Q_name; 953 } 954 code << ", inputs, outputs);\n"; 955 956 if (is_at_points) { 957 // Map back to coefficients 958 code << "\n // -- Output fields\n"; 959 for (CeedInt i = 0; i < num_output_fields; i++) { 960 const char *field_name; 961 std::string var_suffix = "_out_" + std::to_string(i); 962 std::string P_name = "P_1d" + var_suffix; 963 964 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 965 code << " // ---- Output field " << i << ": " << field_name << "\n"; 966 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 967 // Basis action 968 code << " // EvalMode: " << CeedEvalModes[eval_mode] << "\n"; 969 switch (eval_mode) { 970 case CEED_EVAL_NONE: { 971 CeedInt comp_stride; 972 CeedElemRestriction elem_rstr; 973 974 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_output_fields[i], &elem_rstr)); 975 CeedCallBackend(CeedElemRestrictionGetCompStride(elem_rstr, &comp_stride)); 976 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 977 code << " const CeedInt comp_stride" << var_suffix << " = " << comp_stride << ";\n"; 978 code << " WritePoint<num_comp" << var_suffix << ", comp_stride" << var_suffix 979 << ", max_num_points>(data, elem, i, points.num_per_elem[elem], indices.outputs[" << i << "]" 980 << ", r_s" << var_suffix << ", d" << var_suffix << ");\n"; 981 break; 982 } 983 case CEED_EVAL_INTERP: 984 code << " if (i >= points.num_per_elem[elem]) {\n"; 985 code << " for (CeedInt j = 0; j < num_comp" << var_suffix << "; j++) r_s" << var_suffix << "[j] = 0.0;\n"; 986 code << " }\n"; 987 code << " InterpTransposeAtPoints" << max_dim << "d<num_comp" << var_suffix << ", max_num_points, " << P_name << ", " << Q_name 988 << ">(data, i, r_s" << var_suffix << ", r_x, r_c" << var_suffix << ");\n"; 989 break; 990 case CEED_EVAL_GRAD: 991 code << " if (i >= points.num_per_elem[elem]) {\n"; 992 code << " for (CeedInt j = 0; j < num_comp" << var_suffix << "*dim" << var_suffix << "; j++) r_s" << var_suffix << "[j] = 0.0;\n"; 993 code << " }\n"; 994 code << " GradTransposeAtPoints" << max_dim << "d<num_comp" << var_suffix << ", max_num_points, " << P_name << ", " << Q_name 995 << ">(data, i, r_s" << var_suffix << ", r_x, r_c" << var_suffix << ");\n"; 996 break; 997 // LCOV_EXCL_START 998 case CEED_EVAL_WEIGHT: 999 break; // Should not occur 1000 case CEED_EVAL_DIV: 1001 case CEED_EVAL_CURL: 1002 break; // TODO: Not implemented 1003 // LCOV_EXCL_STOP 1004 } 1005 } 1006 } else if (use_3d_slices) { 1007 // Copy or apply transpose grad, if needed 1008 code << "\n // -- Output fields\n"; 1009 for (CeedInt i = 0; i < num_output_fields; i++) { 1010 const char *field_name; 1011 std::string var_suffix = "_out_" + std::to_string(i); 1012 std::string P_name = "P_1d" + var_suffix; 1013 1014 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 1015 code << " // ---- Output field " << i << ": " << field_name << "\n"; 1016 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 1017 // Basis action 1018 code << " // EvalMode: " << CeedEvalModes[eval_mode] << "\n"; 1019 switch (eval_mode) { 1020 case CEED_EVAL_NONE: 1021 code << " for (CeedInt j = 0; j < num_comp" << var_suffix << " ; j++) {\n"; 1022 code << " r_q" << var_suffix << "[q + j*" << Q_name << "] = r_s" << var_suffix << "[j];\n"; 1023 code << " }\n"; 1024 break; 1025 case CEED_EVAL_INTERP: 1026 code << " for (CeedInt j = 0; j < num_comp" << var_suffix << " ; j++) {\n"; 1027 code << " r_q" << var_suffix << "[q + j*" << Q_name << "] = r_s" << var_suffix << "[j];\n"; 1028 code << " }\n"; 1029 break; 1030 case CEED_EVAL_GRAD: 1031 code << " GradColloSliceTranspose3d<num_comp" << var_suffix << ", " << Q_name << ", OP_T_1D>(data, q, r_s" << var_suffix << ", s_G" 1032 << var_suffix << ", r_q" << var_suffix << ");\n"; 1033 break; 1034 // LCOV_EXCL_START 1035 case CEED_EVAL_WEIGHT: 1036 break; // Should not occur 1037 case CEED_EVAL_DIV: 1038 case CEED_EVAL_CURL: 1039 break; // TODO: Not implemented 1040 // LCOV_EXCL_STOP 1041 } 1042 } 1043 } 1044 code << " }\n"; 1045 return CEED_ERROR_SUCCESS; 1046 } 1047 1048 //------------------------------------------------------------------------------ 1049 // Build single operator kernel 1050 //------------------------------------------------------------------------------ 1051 extern "C" int CeedOperatorBuildKernel_Cuda_gen(CeedOperator op, bool *is_good_build) { 1052 bool is_all_tensor = true, is_all_nontensor = true, is_at_points = false, use_3d_slices = false; 1053 Ceed ceed; 1054 CeedInt Q, Q_1d, num_input_fields, num_output_fields, max_dim = 1, max_num_points = 0, coords_comp_stride = 0; 1055 CeedQFunctionField *qf_input_fields, *qf_output_fields; 1056 CeedQFunction_Cuda_gen *qf_data; 1057 CeedQFunction qf; 1058 CeedOperatorField *op_input_fields, *op_output_fields; 1059 CeedOperator_Cuda_gen *data; 1060 std::ostringstream code; 1061 1062 CeedCallBackend(CeedOperatorGetData(op, &data)); 1063 { 1064 bool is_setup_done; 1065 1066 CeedCallBackend(CeedOperatorIsSetupDone(op, &is_setup_done)); 1067 if (is_setup_done) { 1068 *is_good_build = !data->use_fallback; 1069 return CEED_ERROR_SUCCESS; 1070 } 1071 } 1072 1073 // Check field compatibility 1074 CeedCallBackend(CeedOperatorGetFields(op, &num_input_fields, &op_input_fields, &num_output_fields, &op_output_fields)); 1075 { 1076 bool has_shared_bases = true; 1077 1078 for (CeedInt i = 0; i < num_input_fields; i++) { 1079 CeedBasis basis; 1080 1081 CeedCallBackend(CeedOperatorFieldGetBasis(op_input_fields[i], &basis)); 1082 if (basis != CEED_BASIS_NONE) { 1083 bool is_tensor = true; 1084 const char *resource; 1085 char *resource_root; 1086 Ceed basis_ceed; 1087 1088 CeedCallBackend(CeedBasisIsTensor(basis, &is_tensor)); 1089 is_all_tensor = is_all_tensor && is_tensor; 1090 is_all_nontensor = is_all_nontensor && !is_tensor; 1091 CeedCallBackend(CeedBasisGetCeed(basis, &basis_ceed)); 1092 CeedCallBackend(CeedGetResource(basis_ceed, &resource)); 1093 CeedCallBackend(CeedGetResourceRoot(basis_ceed, resource, ":", &resource_root)); 1094 has_shared_bases = has_shared_bases && !strcmp(resource_root, "/gpu/cuda/shared"); 1095 CeedCallBackend(CeedFree(&resource_root)); 1096 CeedCallBackend(CeedDestroy(&basis_ceed)); 1097 } 1098 CeedCallBackend(CeedBasisDestroy(&basis)); 1099 } 1100 1101 for (CeedInt i = 0; i < num_output_fields; i++) { 1102 CeedBasis basis; 1103 1104 CeedCallBackend(CeedOperatorFieldGetBasis(op_output_fields[i], &basis)); 1105 if (basis != CEED_BASIS_NONE) { 1106 bool is_tensor = true; 1107 const char *resource; 1108 char *resource_root; 1109 Ceed basis_ceed; 1110 1111 CeedCallBackend(CeedBasisIsTensor(basis, &is_tensor)); 1112 is_all_tensor = is_all_tensor && is_tensor; 1113 is_all_nontensor = is_all_nontensor && !is_tensor; 1114 1115 CeedCallBackend(CeedBasisGetCeed(basis, &basis_ceed)); 1116 CeedCallBackend(CeedGetResource(basis_ceed, &resource)); 1117 CeedCallBackend(CeedGetResourceRoot(basis_ceed, resource, ":", &resource_root)); 1118 has_shared_bases = has_shared_bases && !strcmp(resource_root, "/gpu/cuda/shared"); 1119 CeedCallBackend(CeedFree(&resource_root)); 1120 CeedCallBackend(CeedDestroy(&basis_ceed)); 1121 } 1122 CeedCallBackend(CeedBasisDestroy(&basis)); 1123 } 1124 // -- Fallback to ref if not all bases are shared 1125 if (!has_shared_bases) { 1126 *is_good_build = false; 1127 return CEED_ERROR_SUCCESS; 1128 } 1129 } 1130 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 1131 CeedCallBackend(CeedOperatorGetQFunction(op, &qf)); 1132 CeedCallBackend(CeedQFunctionGetData(qf, &qf_data)); 1133 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_input_fields, NULL, &qf_output_fields)); 1134 1135 // Get operator data 1136 CeedCallBackend(CeedOperatorIsAtPoints(op, &is_at_points)); 1137 { 1138 CeedInt max_P, max_P_1d; 1139 1140 CeedCallBackend(CeedOperatorBuildKernelData_Cuda_gen(ceed, num_input_fields, op_input_fields, qf_input_fields, num_output_fields, 1141 op_output_fields, qf_output_fields, &max_P, &max_P_1d, &Q, &Q_1d, &max_dim, &is_all_tensor, 1142 &use_3d_slices)); 1143 data->max_P_1d = is_all_tensor ? max_P_1d : max_P; 1144 } 1145 if (max_dim == 0) max_dim = 1; 1146 data->dim = max_dim; 1147 if (is_at_points) { 1148 CeedElemRestriction_Cuda *rstr_data; 1149 CeedElemRestriction rstr_points = NULL; 1150 1151 CeedCallBackend(CeedOperatorAtPointsGetPoints(op, &rstr_points, NULL)); 1152 CeedCallBackend(CeedElemRestrictionGetMaxPointsInElement(rstr_points, &max_num_points)); 1153 CeedCallBackend(CeedElemRestrictionGetCompStride(rstr_points, &coords_comp_stride)); 1154 CeedCallBackend(CeedElemRestrictionGetData(rstr_points, &rstr_data)); 1155 data->points.indices = (CeedInt *)rstr_data->d_offsets; 1156 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_points)); 1157 } 1158 if (is_at_points) use_3d_slices = false; 1159 if (Q_1d == 0) { 1160 if (is_at_points) Q_1d = max_num_points; 1161 else CeedCallBackend(CeedOperatorGetNumQuadraturePoints(op, &Q_1d)); 1162 } 1163 data->Q_1d = Q_1d; 1164 1165 // Check for restriction only identity operator 1166 { 1167 bool is_identity_qf; 1168 1169 CeedCallBackend(CeedQFunctionIsIdentity(qf, &is_identity_qf)); 1170 if (is_identity_qf) { 1171 CeedEvalMode eval_mode_in, eval_mode_out; 1172 1173 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[0], &eval_mode_in)); 1174 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[0], &eval_mode_out)); 1175 CeedCheck(eval_mode_in != CEED_EVAL_NONE || eval_mode_out != CEED_EVAL_NONE, ceed, CEED_ERROR_BACKEND, 1176 "Backend does not implement restriction only identity operators"); 1177 } 1178 } 1179 1180 // Add atomicAdd function for old NVidia architectures 1181 { 1182 Ceed_Cuda *ceed_data; 1183 struct cudaDeviceProp prop; 1184 1185 CeedCallBackend(CeedGetData(ceed, &ceed_data)); 1186 CeedCallBackend(cudaGetDeviceProperties(&prop, ceed_data->device_id)); 1187 if ((prop.major < 6) && (CEED_SCALAR_TYPE != CEED_SCALAR_FP32)) { 1188 code << "// AtomicAdd fallback source\n"; 1189 code << "#include <ceed/jit-source/cuda/cuda-atomic-add-fallback.h>\n\n"; 1190 } 1191 } 1192 1193 // Load basis source files 1194 if (!is_all_nontensor) { 1195 code << "// Tensor basis source\n"; 1196 code << "#include <ceed/jit-source/cuda/cuda-shared-basis-tensor-templates.h>\n\n"; 1197 } 1198 if (!is_all_tensor) { 1199 code << "// Non-tensor basis source\n"; 1200 code << "#include <ceed/jit-source/cuda/cuda-shared-basis-nontensor-templates.h>\n\n"; 1201 } 1202 if (is_at_points) { 1203 code << "// AtPoints basis source\n"; 1204 code << "#include <ceed/jit-source/cuda/cuda-shared-basis-tensor-at-points-templates.h>\n\n"; 1205 } 1206 code << "// CodeGen operator source\n"; 1207 code << "#include <ceed/jit-source/cuda/cuda-gen-templates.h>\n\n"; 1208 1209 // Get QFunction name 1210 std::string qfunction_name(qf_data->qfunction_name); 1211 std::string operator_name; 1212 1213 operator_name = "CeedKernelCudaGenOperator_" + qfunction_name; 1214 1215 // Define CEED_Q_VLA 1216 code << "\n#undef CEED_Q_VLA\n"; 1217 if (max_dim != 3 || is_at_points || use_3d_slices || !is_all_tensor) { 1218 code << "#define CEED_Q_VLA 1\n\n"; 1219 } else { 1220 code << "#define CEED_Q_VLA " << Q_1d << "\n\n"; 1221 } 1222 1223 // Add user QFunction source 1224 { 1225 const char *source_path; 1226 1227 CeedCallBackend(CeedQFunctionGetSourcePath(qf, &source_path)); 1228 CeedCheck(source_path, ceed, CEED_ERROR_UNSUPPORTED, "/gpu/cuda/gen backend requires QFunction source code file"); 1229 1230 code << "// User QFunction source\n"; 1231 code << "#include \"" << source_path << "\"\n\n"; 1232 } 1233 1234 // Setup 1235 code << "\n// -----------------------------------------------------------------------------\n"; 1236 code << "// Operator Kernel\n"; 1237 code << "// \n"; 1238 code << "// d_[in,out]_i: CeedVector device array\n"; 1239 code << "// r_[in,out]_e_i: Element vector register\n"; 1240 code << "// r_[in,out]_q_i: Quadrature space vector register\n"; 1241 code << "// r_[in,out]_c_i: AtPoints Chebyshev coefficients register\n"; 1242 code << "// r_[in,out]_s_i: Quadrature space slice vector register\n"; 1243 code << "// \n"; 1244 code << "// s_B_[in,out]_i: Interpolation matrix, shared memory\n"; 1245 code << "// s_G_[in,out]_i: Gradient matrix, shared memory\n"; 1246 code << "// -----------------------------------------------------------------------------\n"; 1247 code << "extern \"C\" __global__ void " << operator_name 1248 << "(CeedInt num_elem, void* ctx, FieldsInt_Cuda indices, Fields_Cuda fields, Fields_Cuda B, Fields_Cuda G, CeedScalar *W, Points_Cuda " 1249 "points) {\n"; 1250 1251 // Scratch buffers 1252 for (CeedInt i = 0; i < num_input_fields; i++) { 1253 CeedEvalMode eval_mode; 1254 1255 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode)); 1256 if (eval_mode != CEED_EVAL_WEIGHT) { // Skip CEED_EVAL_WEIGHT 1257 code << " const CeedScalar *__restrict__ d_in_" << i << " = fields.inputs[" << i << "];\n"; 1258 } 1259 } 1260 for (CeedInt i = 0; i < num_output_fields; i++) { 1261 code << " CeedScalar *__restrict__ d_out_" << i << " = fields.outputs[" << i << "];\n"; 1262 } 1263 1264 code << " const CeedInt max_dim = " << max_dim << ";\n"; 1265 if (!is_all_tensor) { 1266 code << " const CeedInt Q = " << Q << ";\n"; 1267 } 1268 if (!is_all_nontensor) { 1269 code << " const CeedInt Q_1d = " << Q_1d << ";\n"; 1270 } 1271 if (is_at_points) { 1272 code << " const CeedInt max_num_points = " << max_num_points << ";\n"; 1273 code << " const CeedInt coords_comp_stride = " << coords_comp_stride << ";\n"; 1274 } 1275 1276 // Shared data 1277 code << " extern __shared__ CeedScalar slice[];\n"; 1278 code << " SharedData_Cuda data;\n"; 1279 code << " data.t_id_x = threadIdx.x;\n"; 1280 code << " data.t_id_y = threadIdx.y;\n"; 1281 code << " data.t_id_z = threadIdx.z;\n"; 1282 code << " data.t_id = threadIdx.x + threadIdx.y*blockDim.x + threadIdx.z*blockDim.y*blockDim.x;\n"; 1283 code << " data.slice = slice + data.t_id_z*OP_T_1D" << ((!is_all_tensor || max_dim == 1) ? "" : "*OP_T_1D") << ";\n"; 1284 1285 // -- Determine input mat reuse 1286 FieldReuse_Cuda input_matrix_reuse[CEED_FIELD_MAX]; 1287 1288 for (CeedInt i = 0; i < num_input_fields; i++) { 1289 input_matrix_reuse[i].index = -1; 1290 } 1291 for (CeedInt i = 0; i < num_input_fields; i++) { 1292 bool is_tensor = true; 1293 CeedEvalMode eval_mode_i; 1294 CeedBasis basis_i; 1295 1296 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode_i)); 1297 if (eval_mode_i == CEED_EVAL_WEIGHT) continue; 1298 CeedCallBackend(CeedOperatorFieldGetBasis(op_input_fields[i], &basis_i)); 1299 CeedCallBackend(CeedBasisIsTensor(basis_i, &is_tensor)); 1300 for (CeedInt j = 0; (input_matrix_reuse[i].index == -1) && (j < i); j++) { 1301 CeedEvalMode eval_mode_j; 1302 CeedBasis basis_j; 1303 1304 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[j], &eval_mode_j)); 1305 if (eval_mode_j == CEED_EVAL_WEIGHT) continue; 1306 CeedCallBackend(CeedOperatorFieldGetBasis(op_input_fields[j], &basis_j)); 1307 if (basis_i == basis_j) { 1308 if (is_tensor) { 1309 input_matrix_reuse[i].index = j; 1310 input_matrix_reuse[i].is_input = true; 1311 input_matrix_reuse[i].eval_mode = eval_mode_j; 1312 } else { 1313 // For non-tensor can only re-use with the same eval mode 1314 if (eval_mode_i == eval_mode_j) { 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 } 1319 } 1320 } 1321 CeedCallBackend(CeedBasisDestroy(&basis_j)); 1322 } 1323 CeedCallBackend(CeedBasisDestroy(&basis_i)); 1324 } 1325 1326 // -- Determine output mat reuse 1327 FieldReuse_Cuda output_matrix_reuse[CEED_FIELD_MAX]; 1328 1329 for (CeedInt i = 0; i < num_output_fields; i++) { 1330 output_matrix_reuse[i].index = -1; 1331 } 1332 for (CeedInt i = 0; i < num_output_fields; i++) { 1333 bool is_tensor = true; 1334 CeedEvalMode eval_mode_i; 1335 CeedBasis basis_i; 1336 1337 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode_i)); 1338 CeedCallBackend(CeedOperatorFieldGetBasis(op_output_fields[i], &basis_i)); 1339 CeedCallBackend(CeedBasisIsTensor(basis_i, &is_tensor)); 1340 for (CeedInt j = 0; (output_matrix_reuse[i].index == -1) && (j < num_input_fields); j++) { 1341 CeedEvalMode eval_mode_j; 1342 CeedBasis basis_j; 1343 1344 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[j], &eval_mode_j)); 1345 if (eval_mode_j == CEED_EVAL_WEIGHT) continue; 1346 CeedCallBackend(CeedOperatorFieldGetBasis(op_input_fields[j], &basis_j)); 1347 if (basis_i == basis_j) { 1348 if (is_tensor) { 1349 output_matrix_reuse[i].index = j; 1350 output_matrix_reuse[i].is_input = true; 1351 output_matrix_reuse[i].eval_mode = eval_mode_j; 1352 } else { 1353 // For non-tensor can only re-use with the same eval mode 1354 if (eval_mode_i == eval_mode_j) { 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 } 1359 } 1360 } 1361 CeedCallBackend(CeedBasisDestroy(&basis_j)); 1362 } 1363 for (CeedInt j = 0; (output_matrix_reuse[i].index == -1) && (j < i); j++) { 1364 CeedEvalMode eval_mode_j; 1365 CeedBasis basis_j; 1366 1367 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[j], &eval_mode_j)); 1368 if (eval_mode_j == CEED_EVAL_WEIGHT) continue; 1369 CeedCallBackend(CeedOperatorFieldGetBasis(op_output_fields[j], &basis_j)); 1370 if (basis_i == basis_j) { 1371 if (is_tensor) { 1372 output_matrix_reuse[i].index = j; 1373 output_matrix_reuse[i].is_input = false; 1374 output_matrix_reuse[i].eval_mode = eval_mode_j; 1375 } else { 1376 // For non-tensor can only re-use with the same eval mode 1377 if (eval_mode_i == eval_mode_j) { 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 } 1382 } 1383 } 1384 CeedCallBackend(CeedBasisDestroy(&basis_j)); 1385 } 1386 CeedCallBackend(CeedBasisDestroy(&basis_i)); 1387 } 1388 1389 // Initialize constants, and matrices B and G 1390 code << "\n // Input field constants and basis data\n"; 1391 for (CeedInt i = 0; i < num_input_fields; i++) { 1392 CeedCallBackend(CeedOperatorBuildKernelFieldData_Cuda_gen(code, data, i, op_input_fields[i], qf_input_fields[i], input_matrix_reuse[i], Q_1d, 1393 true, is_all_tensor, is_at_points, use_3d_slices)); 1394 } 1395 code << "\n // Output field constants and basis data\n"; 1396 for (CeedInt i = 0; i < num_output_fields; i++) { 1397 CeedCallBackend(CeedOperatorBuildKernelFieldData_Cuda_gen(code, data, i, op_output_fields[i], qf_output_fields[i], output_matrix_reuse[i], Q_1d, 1398 false, is_all_tensor, is_at_points, use_3d_slices)); 1399 } 1400 1401 // Loop over all elements 1402 code << "\n // Element loop\n"; 1403 code << " __syncthreads();\n"; 1404 code << " for (CeedInt elem = blockIdx.x*blockDim.z + threadIdx.z; elem < num_elem; elem += gridDim.x*blockDim.z) {\n"; 1405 1406 // -- Compute minimum buffer space needed 1407 CeedInt max_rstr_buffer_size = 1; 1408 1409 for (CeedInt i = 0; i < num_input_fields; i++) { 1410 CeedInt num_comp, elem_size; 1411 CeedElemRestriction elem_rstr; 1412 1413 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_input_fields[i], &elem_rstr)); 1414 CeedCallBackend(CeedElemRestrictionGetNumComponents(elem_rstr, &num_comp)); 1415 CeedCallBackend(CeedElemRestrictionGetElementSize(elem_rstr, &elem_size)); 1416 max_rstr_buffer_size = CeedIntMax(max_rstr_buffer_size, num_comp * (is_all_tensor && (max_dim >= 3) ? elem_size : 1)); 1417 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 1418 } 1419 for (CeedInt i = 0; i < num_output_fields; i++) { 1420 CeedInt num_comp, elem_size; 1421 CeedElemRestriction elem_rstr; 1422 1423 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_output_fields[i], &elem_rstr)); 1424 CeedCallBackend(CeedElemRestrictionGetNumComponents(elem_rstr, &num_comp)); 1425 CeedCallBackend(CeedElemRestrictionGetElementSize(elem_rstr, &elem_size)); 1426 max_rstr_buffer_size = CeedIntMax(max_rstr_buffer_size, num_comp * (is_all_tensor && (max_dim >= 3) ? elem_size : 1)); 1427 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 1428 } 1429 code << " // Scratch restriction buffer space\n"; 1430 code << " CeedScalar r_e_scratch[" << max_rstr_buffer_size << "];\n"; 1431 1432 // -- Determine best input field processing order 1433 CeedInt field_rstr_in_buffer[CEED_FIELD_MAX], input_field_order[CEED_FIELD_MAX]; 1434 1435 for (CeedInt i = 0; i < num_input_fields; i++) { 1436 field_rstr_in_buffer[i] = -1; 1437 input_field_order[i] = -1; 1438 } 1439 { 1440 bool is_ordered[CEED_FIELD_MAX]; 1441 CeedInt curr_index = 0; 1442 1443 for (CeedInt i = 0; i < num_input_fields; i++) is_ordered[i] = false; 1444 for (CeedInt i = 0; i < num_input_fields; i++) { 1445 CeedVector vec_i; 1446 CeedElemRestriction rstr_i; 1447 1448 if (is_ordered[i]) continue; 1449 field_rstr_in_buffer[i] = i; 1450 is_ordered[i] = true; 1451 input_field_order[curr_index] = i; 1452 curr_index++; 1453 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[i], &vec_i)); 1454 if (vec_i == CEED_VECTOR_NONE) continue; // CEED_EVAL_WEIGHT 1455 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_input_fields[i], &rstr_i)); 1456 for (CeedInt j = i + 1; j < num_input_fields; j++) { 1457 CeedVector vec_j; 1458 CeedElemRestriction rstr_j; 1459 1460 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[j], &vec_j)); 1461 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_input_fields[j], &rstr_j)); 1462 if (rstr_i == rstr_j && vec_i == vec_j) { 1463 field_rstr_in_buffer[j] = i; 1464 is_ordered[j] = true; 1465 input_field_order[curr_index] = j; 1466 curr_index++; 1467 } 1468 CeedCallBackend(CeedVectorDestroy(&vec_j)); 1469 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_j)); 1470 } 1471 CeedCallBackend(CeedVectorDestroy(&vec_i)); 1472 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_i)); 1473 } 1474 } 1475 1476 // -- Input restriction and basis 1477 code << "\n // -- Input field restrictions and basis actions\n"; 1478 for (CeedInt i = 0; i < num_input_fields; i++) { 1479 const char *field_name; 1480 const CeedInt f = input_field_order[i]; 1481 1482 CeedCallBackend(CeedOperatorFieldGetName(op_input_fields[f], &field_name)); 1483 code << " // ---- Input field " << f << ": " << field_name << "\n"; 1484 1485 // ---- Restriction 1486 CeedCallBackend(CeedOperatorBuildKernelRestriction_Cuda_gen(code, data, f, max_dim, field_rstr_in_buffer, op_input_fields[f], qf_input_fields[f], 1487 Q_1d, true, is_all_tensor, is_at_points, use_3d_slices)); 1488 1489 // ---- Basis action 1490 CeedCallBackend(CeedOperatorBuildKernelBasis_Cuda_gen(code, data, f, max_dim, op_input_fields[f], qf_input_fields[f], Q_1d, true, is_all_tensor, 1491 is_at_points, use_3d_slices)); 1492 } 1493 1494 // -- Q function 1495 CeedCallBackend(CeedOperatorBuildKernelQFunction_Cuda_gen(code, data, max_dim, max_num_points, num_input_fields, op_input_fields, qf_input_fields, 1496 num_output_fields, op_output_fields, qf_output_fields, qfunction_name, Q_1d, 1497 is_all_tensor, is_at_points, use_3d_slices)); 1498 1499 // -- Output basis and restriction 1500 code << "\n // -- Output field basis action and restrictions\n"; 1501 for (CeedInt i = 0; i < num_output_fields; i++) { 1502 const char *field_name; 1503 1504 CeedCallBackend(CeedOperatorFieldGetName(op_output_fields[i], &field_name)); 1505 code << " // ---- Output field " << i << ": " << field_name << "\n"; 1506 1507 // ---- Basis action 1508 CeedCallBackend(CeedOperatorBuildKernelBasis_Cuda_gen(code, data, i, max_dim, op_output_fields[i], qf_output_fields[i], Q_1d, false, 1509 is_all_tensor, is_at_points, use_3d_slices)); 1510 1511 // ---- Restriction 1512 CeedCallBackend(CeedOperatorBuildKernelRestriction_Cuda_gen(code, data, i, max_dim, NULL, op_output_fields[i], qf_output_fields[i], Q_1d, false, 1513 is_all_tensor, is_at_points, use_3d_slices)); 1514 } 1515 1516 // Close loop and function 1517 code << " }\n"; 1518 code << "}\n"; 1519 code << "// -----------------------------------------------------------------------------\n\n"; 1520 1521 // Compile 1522 { 1523 bool is_compile_good = false; 1524 const CeedInt T_1d = CeedIntMax(is_all_tensor ? Q_1d : Q, data->max_P_1d); 1525 1526 CeedCallBackend(CeedTryCompile_Cuda(ceed, code.str().c_str(), &is_compile_good, &data->module, 1, "OP_T_1D", T_1d)); 1527 if (is_compile_good) { 1528 *is_good_build = true; 1529 CeedCallBackend(CeedGetKernel_Cuda(ceed, data->module, operator_name.c_str(), &data->op)); 1530 } else { 1531 *is_good_build = false; 1532 data->use_fallback = true; 1533 } 1534 } 1535 CeedCallBackend(CeedOperatorSetSetupDone(op)); 1536 CeedCallBackend(CeedDestroy(&ceed)); 1537 CeedCallBackend(CeedQFunctionDestroy(&qf)); 1538 return CEED_ERROR_SUCCESS; 1539 } 1540 1541 //------------------------------------------------------------------------------ 1542