1 // Copyright (c) 2017-2024, Lawrence Livermore National Security, LLC and other 2 // CEED contributors. All Rights Reserved. See the top-level LICENSE and NOTICE 3 // files for details. 4 // 5 // SPDX-License-Identifier: BSD-2-Clause 6 // 7 // This file is part of CEED: http://github.com/ceed 8 9 #include <ceed/backend.h> 10 #include <ceed/ceed.h> 11 12 #include <cassert> 13 #include <string> 14 #include <sycl/sycl.hpp> 15 16 #include "../sycl/ceed-sycl-compile.hpp" 17 #include "ceed-sycl-ref.hpp" 18 19 class CeedOperatorSyclLinearDiagonal; 20 class CeedOperatorSyclLinearAssemble; 21 class CeedOperatorSyclLinearAssembleFallback; 22 23 //------------------------------------------------------------------------------ 24 // Get Basis Emode Pointer 25 //------------------------------------------------------------------------------ 26 void CeedOperatorGetBasisPointer_Sycl(const CeedScalar **basis_ptr, CeedEvalMode eval_mode, const CeedScalar *identity, const CeedScalar *interp, 27 const CeedScalar *grad) { 28 switch (eval_mode) { 29 case CEED_EVAL_NONE: 30 *basis_ptr = identity; 31 break; 32 case CEED_EVAL_INTERP: 33 *basis_ptr = interp; 34 break; 35 case CEED_EVAL_GRAD: 36 *basis_ptr = grad; 37 break; 38 case CEED_EVAL_WEIGHT: 39 case CEED_EVAL_DIV: 40 case CEED_EVAL_CURL: 41 break; // Caught by QF Assembly 42 } 43 } 44 45 //------------------------------------------------------------------------------ 46 // Destroy operator 47 //------------------------------------------------------------------------------ 48 static int CeedOperatorDestroy_Sycl(CeedOperator op) { 49 Ceed ceed; 50 Ceed_Sycl *sycl_data; 51 CeedOperator_Sycl *impl; 52 53 CeedCallBackend(CeedOperatorGetData(op, &impl)); 54 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 55 CeedCallBackend(CeedGetData(ceed, &sycl_data)); 56 57 // Apply data 58 for (CeedInt i = 0; i < impl->num_e_in + impl->num_e_out; i++) { 59 CeedCallBackend(CeedVectorDestroy(&impl->e_vecs[i])); 60 } 61 CeedCallBackend(CeedFree(&impl->e_vecs)); 62 63 for (CeedInt i = 0; i < impl->num_e_in; i++) { 64 CeedCallBackend(CeedVectorDestroy(&impl->q_vecs_in[i])); 65 } 66 CeedCallBackend(CeedFree(&impl->q_vecs_in)); 67 68 for (CeedInt i = 0; i < impl->num_e_out; i++) { 69 CeedCallBackend(CeedVectorDestroy(&impl->q_vecs_out[i])); 70 } 71 CeedCallBackend(CeedFree(&impl->q_vecs_out)); 72 73 // QFunction assembly dataf 74 for (CeedInt i = 0; i < impl->num_active_in; i++) { 75 CeedCallBackend(CeedVectorDestroy(&impl->qf_active_in[i])); 76 } 77 CeedCallBackend(CeedFree(&impl->qf_active_in)); 78 79 // Diag data 80 if (impl->diag) { 81 CeedCallBackend(CeedFree(&impl->diag->h_eval_mode_in)); 82 CeedCallBackend(CeedFree(&impl->diag->h_eval_mode_out)); 83 84 CeedCallSycl(ceed, sycl_data->sycl_queue.wait_and_throw()); 85 CeedCallSycl(ceed, sycl::free(impl->diag->d_eval_mode_in, sycl_data->sycl_context)); 86 CeedCallSycl(ceed, sycl::free(impl->diag->d_eval_mode_out, sycl_data->sycl_context)); 87 CeedCallSycl(ceed, sycl::free(impl->diag->d_identity, sycl_data->sycl_context)); 88 CeedCallSycl(ceed, sycl::free(impl->diag->d_interp_in, sycl_data->sycl_context)); 89 CeedCallSycl(ceed, sycl::free(impl->diag->d_interp_out, sycl_data->sycl_context)); 90 CeedCallSycl(ceed, sycl::free(impl->diag->d_grad_in, sycl_data->sycl_context)); 91 CeedCallSycl(ceed, sycl::free(impl->diag->d_grad_out, sycl_data->sycl_context)); 92 CeedCallBackend(CeedElemRestrictionDestroy(&impl->diag->point_block_diag_rstr)); 93 94 CeedCallBackend(CeedVectorDestroy(&impl->diag->elem_diag)); 95 CeedCallBackend(CeedVectorDestroy(&impl->diag->point_block_elem_diag)); 96 } 97 CeedCallBackend(CeedFree(&impl->diag)); 98 99 if (impl->asmb) { 100 CeedCallSycl(ceed, sycl_data->sycl_queue.wait_and_throw()); 101 CeedCallSycl(ceed, sycl::free(impl->asmb->d_B_in, sycl_data->sycl_context)); 102 CeedCallSycl(ceed, sycl::free(impl->asmb->d_B_out, sycl_data->sycl_context)); 103 } 104 CeedCallBackend(CeedFree(&impl->asmb)); 105 106 CeedCallBackend(CeedFree(&impl)); 107 return CEED_ERROR_SUCCESS; 108 } 109 110 //------------------------------------------------------------------------------ 111 // Setup infields or outfields 112 //------------------------------------------------------------------------------ 113 static int CeedOperatorSetupFields_Sycl(CeedQFunction qf, CeedOperator op, bool is_input, CeedVector *e_vecs, CeedVector *q_vecs, CeedInt start_e, 114 CeedInt num_fields, CeedInt Q, CeedInt num_elem) { 115 Ceed ceed; 116 CeedSize q_size; 117 bool is_strided, skip_restriction; 118 CeedInt dim, size; 119 CeedOperatorField *op_fields; 120 CeedQFunctionField *qf_fields; 121 122 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 123 if (is_input) { 124 CeedCallBackend(CeedOperatorGetFields(op, NULL, &op_fields, NULL, NULL)); 125 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_fields, NULL, NULL)); 126 } else { 127 CeedCallBackend(CeedOperatorGetFields(op, NULL, NULL, NULL, &op_fields)); 128 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, NULL, NULL, &qf_fields)); 129 } 130 131 // Loop over fields 132 for (CeedInt i = 0; i < num_fields; i++) { 133 CeedEvalMode eval_mode; 134 CeedVector vec; 135 CeedElemRestriction rstr; 136 CeedBasis basis; 137 138 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode)); 139 140 is_strided = false; 141 skip_restriction = false; 142 if (eval_mode != CEED_EVAL_WEIGHT) { 143 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_fields[i], &rstr)); 144 145 // Check whether this field can skip the element restriction: 146 // must be passive input, with eval_mode NONE, and have a strided restriction with CEED_STRIDES_BACKEND. 147 148 // First, check whether the field is input or output: 149 if (is_input) { 150 // Check for passive input: 151 CeedCallBackend(CeedOperatorFieldGetVector(op_fields[i], &vec)); 152 if (vec != CEED_VECTOR_ACTIVE) { 153 // Check eval_mode 154 if (eval_mode == CEED_EVAL_NONE) { 155 // Check for is_strided restriction 156 CeedCallBackend(CeedElemRestrictionIsStrided(rstr, &is_strided)); 157 if (is_strided) { 158 // Check if vector is already in preferred backend ordering 159 CeedCallBackend(CeedElemRestrictionHasBackendStrides(rstr, &skip_restriction)); 160 } 161 } 162 } 163 CeedCallBackend(CeedVectorDestroy(&vec)); 164 } 165 if (skip_restriction) { 166 // We do not need an E-Vector, but will use the input field vector's data directly in the operator application 167 e_vecs[i + start_e] = NULL; 168 } else { 169 CeedCallBackend(CeedElemRestrictionCreateVector(rstr, NULL, &e_vecs[i + start_e])); 170 } 171 CeedCallBackend(CeedElemRestrictionDestroy(&rstr)); 172 } 173 174 switch (eval_mode) { 175 case CEED_EVAL_NONE: 176 CeedCallBackend(CeedQFunctionFieldGetSize(qf_fields[i], &size)); 177 q_size = (CeedSize)num_elem * Q * size; 178 CeedCallBackend(CeedVectorCreate(ceed, q_size, &q_vecs[i])); 179 break; 180 case CEED_EVAL_INTERP: 181 CeedCallBackend(CeedQFunctionFieldGetSize(qf_fields[i], &size)); 182 q_size = (CeedSize)num_elem * Q * size; 183 CeedCallBackend(CeedVectorCreate(ceed, q_size, &q_vecs[i])); 184 break; 185 case CEED_EVAL_GRAD: 186 CeedCallBackend(CeedOperatorFieldGetBasis(op_fields[i], &basis)); 187 CeedCallBackend(CeedQFunctionFieldGetSize(qf_fields[i], &size)); 188 CeedCallBackend(CeedBasisGetDimension(basis, &dim)); 189 CeedCallBackend(CeedBasisDestroy(&basis)); 190 q_size = (CeedSize)num_elem * Q * size; 191 CeedCallBackend(CeedVectorCreate(ceed, q_size, &q_vecs[i])); 192 break; 193 case CEED_EVAL_WEIGHT: // Only on input fields 194 CeedCallBackend(CeedOperatorFieldGetBasis(op_fields[i], &basis)); 195 q_size = (CeedSize)num_elem * Q; 196 CeedCallBackend(CeedVectorCreate(ceed, q_size, &q_vecs[i])); 197 CeedCallBackend(CeedBasisApply(basis, num_elem, CEED_NOTRANSPOSE, CEED_EVAL_WEIGHT, CEED_VECTOR_NONE, q_vecs[i])); 198 CeedCallBackend(CeedBasisDestroy(&basis)); 199 break; 200 case CEED_EVAL_DIV: 201 break; // TODO: Not implemented 202 case CEED_EVAL_CURL: 203 break; // TODO: Not implemented 204 } 205 } 206 return CEED_ERROR_SUCCESS; 207 } 208 209 //------------------------------------------------------------------------------ 210 // CeedOperator needs to connect all the named fields (be they active or 211 // passive) to the named inputs and outputs of its CeedQFunction. 212 //------------------------------------------------------------------------------ 213 static int CeedOperatorSetup_Sycl(CeedOperator op) { 214 Ceed ceed; 215 bool is_setup_done; 216 CeedInt Q, num_elem, num_input_fields, num_output_fields; 217 CeedQFunctionField *qf_input_fields, *qf_output_fields; 218 CeedQFunction qf; 219 CeedOperatorField *op_input_fields, *op_output_fields; 220 CeedOperator_Sycl *impl; 221 222 CeedCallBackend(CeedOperatorIsSetupDone(op, &is_setup_done)); 223 if (is_setup_done) return CEED_ERROR_SUCCESS; 224 225 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 226 CeedCallBackend(CeedOperatorGetData(op, &impl)); 227 CeedCallBackend(CeedOperatorGetQFunction(op, &qf)); 228 CeedCallBackend(CeedOperatorGetNumQuadraturePoints(op, &Q)); 229 CeedCallBackend(CeedOperatorGetNumElements(op, &num_elem)); 230 CeedCallBackend(CeedOperatorGetFields(op, &num_input_fields, &op_input_fields, &num_output_fields, &op_output_fields)); 231 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_input_fields, NULL, &qf_output_fields)); 232 233 // Allocate 234 CeedCallBackend(CeedCalloc(num_input_fields + num_output_fields, &impl->e_vecs)); 235 236 CeedCallBackend(CeedCalloc(CEED_FIELD_MAX, &impl->q_vecs_in)); 237 CeedCallBackend(CeedCalloc(CEED_FIELD_MAX, &impl->q_vecs_out)); 238 239 impl->num_e_in = num_input_fields; 240 impl->num_e_out = num_output_fields; 241 242 // Set up infield and outfield e_vecs and q_vecs 243 // Infields 244 CeedCallBackend(CeedOperatorSetupFields_Sycl(qf, op, true, impl->e_vecs, impl->q_vecs_in, 0, num_input_fields, Q, num_elem)); 245 // Outfields 246 CeedCallBackend(CeedOperatorSetupFields_Sycl(qf, op, false, impl->e_vecs, impl->q_vecs_out, num_input_fields, num_output_fields, Q, num_elem)); 247 248 CeedCallBackend(CeedOperatorSetSetupDone(op)); 249 return CEED_ERROR_SUCCESS; 250 } 251 252 //------------------------------------------------------------------------------ 253 // Setup Operator Inputs 254 //------------------------------------------------------------------------------ 255 static inline int CeedOperatorSetupInputs_Sycl(CeedInt num_input_fields, CeedQFunctionField *qf_input_fields, CeedOperatorField *op_input_fields, 256 CeedVector in_vec, const bool skip_active, CeedScalar *e_data[2 * CEED_FIELD_MAX], 257 CeedOperator_Sycl *impl, CeedRequest *request) { 258 for (CeedInt i = 0; i < num_input_fields; i++) { 259 bool is_active; 260 CeedEvalMode eval_mode; 261 CeedVector vec; 262 263 // Get input vector 264 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[i], &vec)); 265 is_active = vec == CEED_VECTOR_ACTIVE; 266 if (is_active) { 267 if (skip_active) continue; 268 else vec = in_vec; 269 } 270 271 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode)); 272 if (eval_mode == CEED_EVAL_WEIGHT) { // Skip 273 } else { 274 // Restrict, if necessary 275 if (!impl->e_vecs[i]) { 276 // No restriction for this field; read data directly from vec. 277 CeedCallBackend(CeedVectorGetArrayRead(vec, CEED_MEM_DEVICE, (const CeedScalar **)&e_data[i])); 278 } else { 279 CeedElemRestriction rstr; 280 281 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_input_fields[i], &rstr)); 282 CeedCallBackend(CeedElemRestrictionApply(rstr, CEED_NOTRANSPOSE, vec, impl->e_vecs[i], request)); 283 CeedCallBackend(CeedElemRestrictionDestroy(&rstr)); 284 CeedCallBackend(CeedVectorGetArrayRead(impl->e_vecs[i], CEED_MEM_DEVICE, (const CeedScalar **)&e_data[i])); 285 } 286 } 287 if (!is_active) CeedCallBackend(CeedVectorDestroy(&vec)); 288 } 289 return CEED_ERROR_SUCCESS; 290 } 291 292 //------------------------------------------------------------------------------ 293 // Input Basis Action 294 //------------------------------------------------------------------------------ 295 static inline int CeedOperatorInputBasis_Sycl(CeedInt num_elem, CeedQFunctionField *qf_input_fields, CeedOperatorField *op_input_fields, 296 CeedInt num_input_fields, const bool skip_active, CeedScalar *e_data[2 * CEED_FIELD_MAX], 297 CeedOperator_Sycl *impl) { 298 for (CeedInt i = 0; i < num_input_fields; i++) { 299 CeedEvalMode eval_mode; 300 CeedBasis basis; 301 302 // Skip active input 303 if (skip_active) { 304 bool is_active; 305 CeedVector vec; 306 307 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[i], &vec)); 308 is_active = vec == CEED_VECTOR_ACTIVE; 309 CeedCallBackend(CeedVectorDestroy(&vec)); 310 if (is_active) continue; 311 } 312 // Basis action 313 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode)); 314 switch (eval_mode) { 315 case CEED_EVAL_NONE: 316 CeedCallBackend(CeedVectorSetArray(impl->q_vecs_in[i], CEED_MEM_DEVICE, CEED_USE_POINTER, e_data[i])); 317 break; 318 case CEED_EVAL_INTERP: 319 case CEED_EVAL_GRAD: 320 CeedCallBackend(CeedOperatorFieldGetBasis(op_input_fields[i], &basis)); 321 CeedCallBackend(CeedBasisApply(basis, num_elem, CEED_NOTRANSPOSE, eval_mode, impl->e_vecs[i], impl->q_vecs_in[i])); 322 CeedCallBackend(CeedBasisDestroy(&basis)); 323 break; 324 case CEED_EVAL_WEIGHT: 325 break; // No action 326 case CEED_EVAL_DIV: 327 break; // TODO: Not implemented 328 case CEED_EVAL_CURL: 329 break; // TODO: Not implemented 330 } 331 } 332 return CEED_ERROR_SUCCESS; 333 } 334 335 //------------------------------------------------------------------------------ 336 // Restore Input Vectors 337 //------------------------------------------------------------------------------ 338 static inline int CeedOperatorRestoreInputs_Sycl(CeedInt num_input_fields, CeedQFunctionField *qf_input_fields, CeedOperatorField *op_input_fields, 339 const bool skip_active, CeedScalar *e_data[2 * CEED_FIELD_MAX], CeedOperator_Sycl *impl) { 340 for (CeedInt i = 0; i < num_input_fields; i++) { 341 bool is_active; 342 CeedEvalMode eval_mode; 343 CeedVector vec; 344 345 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[i], &vec)); 346 is_active = vec == CEED_VECTOR_ACTIVE; 347 // Skip active input 348 if (skip_active) { 349 if (is_active) continue; 350 } 351 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode)); 352 if (eval_mode == CEED_EVAL_WEIGHT) { // Skip 353 } else { 354 if (!impl->e_vecs[i]) { // This was a skip_restriction case 355 CeedCallBackend(CeedVectorRestoreArrayRead(vec, (const CeedScalar **)&e_data[i])); 356 } else { 357 CeedCallBackend(CeedVectorRestoreArrayRead(impl->e_vecs[i], (const CeedScalar **)&e_data[i])); 358 } 359 } 360 if (!is_active) CeedCallBackend(CeedVectorDestroy(&vec)); 361 } 362 return CEED_ERROR_SUCCESS; 363 } 364 365 //------------------------------------------------------------------------------ 366 // Apply and add to output 367 //------------------------------------------------------------------------------ 368 static int CeedOperatorApplyAdd_Sycl(CeedOperator op, CeedVector in_vec, CeedVector out_vec, CeedRequest *request) { 369 CeedInt Q, num_elem, elem_size, num_input_fields, num_output_fields, size; 370 CeedEvalMode eval_mode; 371 CeedScalar *e_data[2 * CEED_FIELD_MAX] = {0}; 372 CeedQFunctionField *qf_input_fields, *qf_output_fields; 373 CeedQFunction qf; 374 CeedOperatorField *op_input_fields, *op_output_fields; 375 CeedOperator_Sycl *impl; 376 377 CeedCallBackend(CeedOperatorGetData(op, &impl)); 378 CeedCallBackend(CeedOperatorGetQFunction(op, &qf)); 379 CeedCallBackend(CeedOperatorGetNumQuadraturePoints(op, &Q)); 380 CeedCallBackend(CeedOperatorGetNumElements(op, &num_elem)); 381 CeedCallBackend(CeedOperatorGetFields(op, &num_input_fields, &op_input_fields, &num_output_fields, &op_output_fields)); 382 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_input_fields, NULL, &qf_output_fields)); 383 384 // Setup 385 CeedCallBackend(CeedOperatorSetup_Sycl(op)); 386 387 // Input Evecs and Restriction 388 CeedCallBackend(CeedOperatorSetupInputs_Sycl(num_input_fields, qf_input_fields, op_input_fields, in_vec, false, e_data, impl, request)); 389 390 // Input basis apply if needed 391 CeedCallBackend(CeedOperatorInputBasis_Sycl(num_elem, qf_input_fields, op_input_fields, num_input_fields, false, e_data, impl)); 392 393 // Output pointers, as necessary 394 for (CeedInt i = 0; i < num_output_fields; i++) { 395 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 396 if (eval_mode == CEED_EVAL_NONE) { 397 // Set the output Q-Vector to use the E-Vector data directly 398 CeedCallBackend(CeedVectorGetArrayWrite(impl->e_vecs[i + impl->num_e_in], CEED_MEM_DEVICE, &e_data[i + num_input_fields])); 399 CeedCallBackend(CeedVectorSetArray(impl->q_vecs_out[i], CEED_MEM_DEVICE, CEED_USE_POINTER, e_data[i + num_input_fields])); 400 } 401 } 402 403 // Q function 404 CeedCallBackend(CeedQFunctionApply(qf, num_elem * Q, impl->q_vecs_in, impl->q_vecs_out)); 405 406 // Output basis apply if needed 407 for (CeedInt i = 0; i < num_output_fields; i++) { 408 CeedElemRestriction rstr; 409 CeedBasis basis; 410 411 // Get elem_size, eval_mode, size 412 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_output_fields[i], &rstr)); 413 CeedCallBackend(CeedElemRestrictionGetElementSize(rstr, &elem_size)); 414 CeedCallBackend(CeedElemRestrictionDestroy(&rstr)); 415 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 416 CeedCallBackend(CeedQFunctionFieldGetSize(qf_output_fields[i], &size)); 417 // Basis action 418 switch (eval_mode) { 419 case CEED_EVAL_NONE: 420 break; 421 case CEED_EVAL_INTERP: 422 case CEED_EVAL_GRAD: 423 CeedCallBackend(CeedOperatorFieldGetBasis(op_output_fields[i], &basis)); 424 CeedCallBackend(CeedBasisApply(basis, num_elem, CEED_TRANSPOSE, eval_mode, impl->q_vecs_out[i], impl->e_vecs[i + impl->num_e_in])); 425 CeedCallBackend(CeedBasisDestroy(&basis)); 426 break; 427 // LCOV_EXCL_START 428 case CEED_EVAL_WEIGHT: 429 Ceed ceed; 430 431 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 432 return CeedError(ceed, CEED_ERROR_BACKEND, "CEED_EVAL_WEIGHT cannot be an output evaluation mode"); 433 break; // Should not occur 434 case CEED_EVAL_DIV: 435 case CEED_EVAL_CURL: { 436 Ceed ceed; 437 438 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 439 return CeedError(ceed, CEED_ERROR_BACKEND, "%s not supported", CeedEvalModes[eval_mode]); 440 break; // Should not occur 441 } 442 // LCOV_EXCL_STOP 443 } 444 } 445 446 // Output restriction 447 for (CeedInt i = 0; i < num_output_fields; i++) { 448 bool is_active; 449 CeedEvalMode eval_mode; 450 CeedVector vec; 451 CeedElemRestriction rstr; 452 453 // Restore evec 454 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 455 if (eval_mode == CEED_EVAL_NONE) { 456 CeedCallBackend(CeedVectorRestoreArray(impl->e_vecs[i + impl->num_e_in], &e_data[i + num_input_fields])); 457 } 458 // Restrict 459 CeedCallBackend(CeedOperatorFieldGetVector(op_output_fields[i], &vec)); 460 is_active = vec == CEED_VECTOR_ACTIVE; 461 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_output_fields[i], &rstr)); 462 if (is_active) vec = out_vec; 463 CeedCallBackend(CeedElemRestrictionApply(rstr, CEED_TRANSPOSE, impl->e_vecs[i + impl->num_inputs], vec, request)); 464 if (!is_active) CeedCallBackend(CeedVectorDestroy(&vec)); 465 CeedCallBackend(CeedElemRestrictionDestroy(&rstr)); 466 } 467 468 // Restore input arrays 469 CeedCallBackend(CeedOperatorRestoreInputs_Sycl(num_input_fields, qf_input_fields, op_input_fields, false, e_data, impl)); 470 return CEED_ERROR_SUCCESS; 471 } 472 473 //------------------------------------------------------------------------------ 474 // Core code for assembling linear QFunction 475 //------------------------------------------------------------------------------ 476 static inline int CeedOperatorLinearAssembleQFunctionCore_Sycl(CeedOperator op, bool build_objects, CeedVector *assembled, CeedElemRestriction *rstr, 477 CeedRequest *request) { 478 Ceed ceed, ceed_parent; 479 CeedSize q_size; 480 CeedInt num_active_in, num_active_out, Q, num_elem, num_input_fields, num_output_fields, size; 481 CeedScalar *assembled_array, *e_data[2 * CEED_FIELD_MAX] = {NULL}; 482 CeedVector *active_in; 483 CeedQFunctionField *qf_input_fields, *qf_output_fields; 484 CeedQFunction qf; 485 CeedOperatorField *op_input_fields, *op_output_fields; 486 CeedOperator_Sycl *impl; 487 488 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 489 CeedCallBackend(CeedOperatorGetFallbackParentCeed(op, &ceed_parent)); 490 CeedCallBackend(CeedOperatorGetData(op, &impl)); 491 CeedCallBackend(CeedOperatorGetQFunction(op, &qf)); 492 CeedCallBackend(CeedOperatorGetNumQuadraturePoints(op, &Q)); 493 CeedCallBackend(CeedOperatorGetNumElements(op, &num_elem)); 494 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_input_fields, NULL, &qf_output_fields)); 495 CeedCallBackend(CeedOperatorGetFields(op, &num_input_fields, &op_input_fields, &num_output_fields, &op_output_fields)); 496 active_in = impl->qf_active_in; 497 num_active_in = impl->num_active_in, num_active_out = impl->num_active_out; 498 499 // Setup 500 CeedCallBackend(CeedOperatorSetup_Sycl(op)); 501 502 // Input Evecs and Restriction 503 CeedCallBackend(CeedOperatorSetupInputs_Sycl(num_input_fields, qf_input_fields, op_input_fields, NULL, true, e_data, impl, request)); 504 505 // Count number of active input fields 506 if (!num_active_in) { 507 for (CeedInt i = 0; i < num_input_fields; i++) { 508 CeedScalar *q_vec_array; 509 CeedVector vec; 510 511 // Check if active input 512 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[i], &vec)); 513 if (vec == CEED_VECTOR_ACTIVE) { 514 CeedCallBackend(CeedQFunctionFieldGetSize(qf_input_fields[i], &size)); 515 CeedCallBackend(CeedVectorSetValue(impl->q_vecs_in[i], 0.0)); 516 CeedCallBackend(CeedVectorGetArray(impl->q_vecs_in[i], CEED_MEM_DEVICE, &q_vec_array)); 517 CeedCallBackend(CeedRealloc(num_active_in + size, &active_in)); 518 for (CeedInt field = 0; field < size; field++) { 519 q_size = (CeedSize)Q * num_elem; 520 CeedCallBackend(CeedVectorCreate(ceed, q_size, &active_in[num_active_in + field])); 521 CeedCallBackend( 522 CeedVectorSetArray(active_in[num_active_in + field], CEED_MEM_DEVICE, CEED_USE_POINTER, &q_vec_array[field * Q * num_elem])); 523 } 524 num_active_in += size; 525 CeedCallBackend(CeedVectorRestoreArray(impl->q_vecs_in[i], &q_vec_array)); 526 } 527 CeedCallBackend(CeedVectorDestroy(&vec)); 528 } 529 impl->num_active_in = num_active_in; 530 impl->qf_active_in = active_in; 531 } 532 533 // Count number of active output fields 534 if (!num_active_out) { 535 for (CeedInt i = 0; i < num_output_fields; i++) { 536 CeedVector vec; 537 538 // Check if active output 539 CeedCallBackend(CeedOperatorFieldGetVector(op_output_fields[i], &vec)); 540 if (vec == CEED_VECTOR_ACTIVE) { 541 CeedCallBackend(CeedQFunctionFieldGetSize(qf_output_fields[i], &size)); 542 num_active_out += size; 543 } 544 CeedCallBackend(CeedVectorDestroy(&vec)); 545 } 546 impl->num_active_out = num_active_out; 547 } 548 549 // Check sizes 550 CeedCheck(num_active_in > 0 && num_active_out > 0, ceed, CEED_ERROR_BACKEND, "Cannot assemble QFunction without active inputs and outputs"); 551 552 // Build objects if needed 553 if (build_objects) { 554 CeedSize l_size = (CeedSize)num_elem * Q * num_active_in * num_active_out; 555 CeedInt strides[3] = {1, num_elem * Q, Q}; /* *NOPAD* */ 556 557 // Create output restriction 558 CeedCallBackend(CeedElemRestrictionCreateStrided(ceed_parent, num_elem, Q, num_active_in * num_active_out, l_size, strides, rstr)); 559 // Create assembled vector 560 CeedCallBackend(CeedVectorCreate(ceed_parent, l_size, assembled)); 561 } 562 CeedCallBackend(CeedVectorSetValue(*assembled, 0.0)); 563 CeedCallBackend(CeedVectorGetArray(*assembled, CEED_MEM_DEVICE, &assembled_array)); 564 565 // Input basis apply 566 CeedCallBackend(CeedOperatorInputBasis_Sycl(num_elem, qf_input_fields, op_input_fields, num_input_fields, true, e_data, impl)); 567 568 // Assemble QFunction 569 for (CeedInt in = 0; in < num_active_in; in++) { 570 // Set Inputs 571 CeedCallBackend(CeedVectorSetValue(active_in[in], 1.0)); 572 if (num_active_in > 1) { 573 CeedCallBackend(CeedVectorSetValue(active_in[(in + num_active_in - 1) % num_active_in], 0.0)); 574 } 575 // Set Outputs 576 for (CeedInt out = 0; out < num_output_fields; out++) { 577 CeedVector vec; 578 579 // Check if active output 580 CeedCallBackend(CeedOperatorFieldGetVector(op_output_fields[out], &vec)); 581 if (vec == CEED_VECTOR_ACTIVE) { 582 CeedCallBackend(CeedVectorSetArray(impl->q_vecs_out[out], CEED_MEM_DEVICE, CEED_USE_POINTER, assembled_array)); 583 CeedCallBackend(CeedQFunctionFieldGetSize(qf_output_fields[out], &size)); 584 assembled_array += size * Q * num_elem; // Advance the pointer by the size of the output 585 } 586 CeedCallBackend(CeedVectorDestroy(&vec)); 587 } 588 // Apply QFunction 589 CeedCallBackend(CeedQFunctionApply(qf, Q * num_elem, impl->q_vecs_in, impl->q_vecs_out)); 590 } 591 592 // Un-set output Qvecs to prevent accidental overwrite of Assembled 593 for (CeedInt out = 0; out < num_output_fields; out++) { 594 CeedVector vec; 595 596 // Check if active output 597 CeedCallBackend(CeedOperatorFieldGetVector(op_output_fields[out], &vec)); 598 if (vec == CEED_VECTOR_ACTIVE) { 599 CeedCallBackend(CeedVectorTakeArray(impl->q_vecs_out[out], CEED_MEM_DEVICE, NULL)); 600 } 601 CeedCallBackend(CeedVectorDestroy(&vec)); 602 } 603 604 // Restore input arrays 605 CeedCallBackend(CeedOperatorRestoreInputs_Sycl(num_input_fields, qf_input_fields, op_input_fields, true, e_data, impl)); 606 607 // Restore output 608 CeedCallBackend(CeedVectorRestoreArray(*assembled, &assembled_array)); 609 return CEED_ERROR_SUCCESS; 610 } 611 612 //------------------------------------------------------------------------------ 613 // Assemble Linear QFunction 614 //------------------------------------------------------------------------------ 615 static int CeedOperatorLinearAssembleQFunction_Sycl(CeedOperator op, CeedVector *assembled, CeedElemRestriction *rstr, CeedRequest *request) { 616 return CeedOperatorLinearAssembleQFunctionCore_Sycl(op, true, assembled, rstr, request); 617 } 618 619 //------------------------------------------------------------------------------ 620 // Update Assembled Linear QFunction 621 //------------------------------------------------------------------------------ 622 static int CeedOperatorLinearAssembleQFunctionUpdate_Sycl(CeedOperator op, CeedVector assembled, CeedElemRestriction rstr, CeedRequest *request) { 623 return CeedOperatorLinearAssembleQFunctionCore_Sycl(op, false, &assembled, &rstr, request); 624 } 625 626 //------------------------------------------------------------------------------ 627 // Assemble diagonal setup 628 //------------------------------------------------------------------------------ 629 static inline int CeedOperatorAssembleDiagonalSetup_Sycl(CeedOperator op) { 630 Ceed ceed; 631 Ceed_Sycl *sycl_data; 632 CeedInt num_input_fields, num_output_fields, num_eval_mode_in = 0, num_comp = 0, dim = 1, num_eval_mode_out = 0; 633 CeedEvalMode *eval_mode_in = NULL, *eval_mode_out = NULL; 634 CeedBasis basis_in = NULL, basis_out = NULL; 635 CeedQFunctionField *qf_fields; 636 CeedQFunction qf; 637 CeedOperatorField *op_fields; 638 CeedOperator_Sycl *impl; 639 640 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 641 CeedCallBackend(CeedOperatorGetQFunction(op, &qf)); 642 CeedCallBackend(CeedQFunctionGetNumArgs(qf, &num_input_fields, &num_output_fields)); 643 644 // Determine active input basis 645 CeedCallBackend(CeedOperatorGetFields(op, NULL, &op_fields, NULL, NULL)); 646 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_fields, NULL, NULL)); 647 for (CeedInt i = 0; i < num_input_fields; i++) { 648 CeedVector vec; 649 650 CeedCallBackend(CeedOperatorFieldGetVector(op_fields[i], &vec)); 651 if (vec == CEED_VECTOR_ACTIVE) { 652 CeedEvalMode eval_mode; 653 CeedBasis basis; 654 655 CeedCallBackend(CeedOperatorFieldGetBasis(op_fields[i], &basis)); 656 CeedCheck(!basis_in || basis_in == basis, ceed, CEED_ERROR_BACKEND, 657 "Backend does not implement operator diagonal assembly with multiple active bases"); 658 if (!basis_in) CeedCallBackend(CeedBasisReferenceCopy(basis, &basis_in)); 659 CeedCallBackend(CeedBasisDestroy(&basis)); 660 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode)); 661 switch (eval_mode) { 662 case CEED_EVAL_NONE: 663 case CEED_EVAL_INTERP: 664 CeedCallBackend(CeedRealloc(num_eval_mode_in + 1, &eval_mode_in)); 665 eval_mode_in[num_eval_mode_in] = eval_mode; 666 num_eval_mode_in += 1; 667 break; 668 case CEED_EVAL_GRAD: 669 CeedCallBackend(CeedRealloc(num_eval_mode_in + dim, &eval_mode_in)); 670 for (CeedInt d = 0; d < dim; d++) eval_mode_in[num_eval_mode_in + d] = eval_mode; 671 num_eval_mode_in += dim; 672 break; 673 case CEED_EVAL_WEIGHT: 674 case CEED_EVAL_DIV: 675 case CEED_EVAL_CURL: 676 break; // Caught by QF Assembly 677 } 678 } 679 CeedCallBackend(CeedVectorDestroy(&vec)); 680 } 681 682 // Determine active output basis 683 CeedCallBackend(CeedOperatorGetFields(op, NULL, NULL, NULL, &op_fields)); 684 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, NULL, NULL, &qf_fields)); 685 for (CeedInt i = 0; i < num_output_fields; i++) { 686 CeedVector vec; 687 688 CeedCallBackend(CeedOperatorFieldGetVector(op_fields[i], &vec)); 689 if (vec == CEED_VECTOR_ACTIVE) { 690 CeedEvalMode eval_mode; 691 CeedBasis basis; 692 693 CeedCallBackend(CeedOperatorFieldGetBasis(op_fields[i], &basis)); 694 CeedCheck(!basis_out || basis_out == basis, ceed, CEED_ERROR_BACKEND, 695 "Backend does not implement operator diagonal assembly with multiple active bases"); 696 if (!basis_out) CeedCallBackend(CeedBasisReferenceCopy(basis, &basis_out)); 697 CeedCallBackend(CeedBasisDestroy(&basis)); 698 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode)); 699 switch (eval_mode) { 700 case CEED_EVAL_NONE: 701 case CEED_EVAL_INTERP: 702 CeedCallBackend(CeedRealloc(num_eval_mode_in + 1, &eval_mode_in)); 703 eval_mode_in[num_eval_mode_in] = eval_mode; 704 num_eval_mode_in += 1; 705 break; 706 case CEED_EVAL_GRAD: 707 CeedCallBackend(CeedRealloc(num_eval_mode_in + dim, &eval_mode_in)); 708 for (CeedInt d = 0; d < dim; d++) eval_mode_in[num_eval_mode_in + d] = eval_mode; 709 num_eval_mode_in += dim; 710 break; 711 case CEED_EVAL_WEIGHT: 712 case CEED_EVAL_DIV: 713 case CEED_EVAL_CURL: 714 break; // Caught by QF Assembly 715 } 716 } 717 CeedCallBackend(CeedVectorDestroy(&vec)); 718 } 719 720 // Operator data struct 721 CeedCallBackend(CeedOperatorGetData(op, &impl)); 722 CeedCallBackend(CeedGetData(ceed, &sycl_data)); 723 CeedCallBackend(CeedCalloc(1, &impl->diag)); 724 CeedOperatorDiag_Sycl *diag = impl->diag; 725 726 diag->basis_in = basis_in; 727 diag->basis_out = basis_out; 728 diag->h_eval_mode_in = eval_mode_in; 729 diag->h_eval_mode_out = eval_mode_out; 730 diag->num_eval_mode_in = num_eval_mode_in; 731 diag->num_eval_mode_out = num_eval_mode_out; 732 733 // Kernel parameters 734 CeedInt num_nodes, num_qpts; 735 CeedCallBackend(CeedBasisGetNumNodes(basis_in, &num_nodes)); 736 CeedCallBackend(CeedBasisGetNumQuadraturePoints(basis_in, &num_qpts)); 737 diag->num_nodes = num_nodes; 738 diag->num_qpts = num_qpts; 739 diag->num_comp = num_comp; 740 741 // Basis matrices 742 const CeedInt i_len = num_qpts * num_nodes; 743 const CeedInt g_len = num_qpts * num_nodes * dim; 744 const CeedScalar *interp_in, *interp_out, *grad_in, *grad_out; 745 746 // CEED_EVAL_NONE 747 CeedScalar *identity = NULL; 748 bool has_eval_none = false; 749 for (CeedInt i = 0; i < num_eval_mode_in; i++) has_eval_none = has_eval_none || (eval_mode_in[i] == CEED_EVAL_NONE); 750 for (CeedInt i = 0; i < num_eval_mode_out; i++) has_eval_none = has_eval_none || (eval_mode_out[i] == CEED_EVAL_NONE); 751 752 std::vector<sycl::event> e; 753 754 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()}; 755 756 std::vector<sycl::event> copy_events; 757 758 if (has_eval_none) { 759 CeedCallBackend(CeedCalloc(num_qpts * num_nodes, &identity)); 760 for (CeedSize i = 0; i < (num_nodes < num_qpts ? num_nodes : num_qpts); i++) identity[i * num_nodes + i] = 1.0; 761 CeedCallSycl(ceed, diag->d_identity = sycl::malloc_device<CeedScalar>(i_len, sycl_data->sycl_device, sycl_data->sycl_context)); 762 sycl::event identity_copy = sycl_data->sycl_queue.copy<CeedScalar>(identity, diag->d_identity, i_len, e); 763 copy_events.push_back(identity_copy); 764 } 765 766 // CEED_EVAL_INTERP 767 CeedCallBackend(CeedBasisGetInterp(basis_in, &interp_in)); 768 CeedCallSycl(ceed, diag->d_interp_in = sycl::malloc_device<CeedScalar>(i_len, sycl_data->sycl_device, sycl_data->sycl_context)); 769 sycl::event interp_in_copy = sycl_data->sycl_queue.copy<CeedScalar>(interp_in, diag->d_interp_in, i_len, e); 770 copy_events.push_back(interp_in_copy); 771 772 CeedCallBackend(CeedBasisGetInterp(basis_out, &interp_out)); 773 CeedCallSycl(ceed, diag->d_interp_out = sycl::malloc_device<CeedScalar>(i_len, sycl_data->sycl_device, sycl_data->sycl_context)); 774 sycl::event interp_out_copy = sycl_data->sycl_queue.copy<CeedScalar>(interp_out, diag->d_interp_out, i_len, e); 775 copy_events.push_back(interp_out_copy); 776 777 // CEED_EVAL_GRAD 778 CeedCallBackend(CeedBasisGetGrad(basis_in, &grad_in)); 779 CeedCallSycl(ceed, diag->d_grad_in = sycl::malloc_device<CeedScalar>(g_len, sycl_data->sycl_device, sycl_data->sycl_context)); 780 sycl::event grad_in_copy = sycl_data->sycl_queue.copy<CeedScalar>(grad_in, diag->d_grad_in, g_len, e); 781 copy_events.push_back(grad_in_copy); 782 783 CeedCallBackend(CeedBasisGetGrad(basis_out, &grad_out)); 784 CeedCallSycl(ceed, diag->d_grad_out = sycl::malloc_device<CeedScalar>(g_len, sycl_data->sycl_device, sycl_data->sycl_context)); 785 sycl::event grad_out_copy = sycl_data->sycl_queue.copy<CeedScalar>(grad_out, diag->d_grad_out, g_len, e); 786 copy_events.push_back(grad_out_copy); 787 788 // Arrays of eval_modes 789 CeedCallSycl(ceed, diag->d_eval_mode_in = sycl::malloc_device<CeedEvalMode>(num_eval_mode_in, sycl_data->sycl_device, sycl_data->sycl_context)); 790 sycl::event eval_mode_in_copy = sycl_data->sycl_queue.copy<CeedEvalMode>(eval_mode_in, diag->d_eval_mode_in, num_eval_mode_in, e); 791 copy_events.push_back(eval_mode_in_copy); 792 793 CeedCallSycl(ceed, diag->d_eval_mode_out = sycl::malloc_device<CeedEvalMode>(num_eval_mode_out, sycl_data->sycl_device, sycl_data->sycl_context)); 794 sycl::event eval_mode_out_copy = sycl_data->sycl_queue.copy<CeedEvalMode>(eval_mode_out, diag->d_eval_mode_out, num_eval_mode_out, e); 795 copy_events.push_back(eval_mode_out_copy); 796 797 // Restriction 798 { 799 CeedElemRestriction rstr_out; 800 801 CeedCallBackend(CeedOperatorGetActiveElemRestrictions(op, NULL, &rstr_out)); 802 diag->diag_rstr = rstr_out; 803 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_out)); 804 } 805 806 // Wait for all copies to complete and handle exceptions 807 CeedCallSycl(ceed, sycl::event::wait_and_throw(copy_events)); 808 return CEED_ERROR_SUCCESS; 809 } 810 811 //------------------------------------------------------------------------------ 812 // Kernel for diagonal assembly 813 //------------------------------------------------------------------------------ 814 static int CeedOperatorLinearDiagonal_Sycl(sycl::queue &sycl_queue, const bool is_point_block, const CeedInt num_elem, 815 const CeedOperatorDiag_Sycl *diag, const CeedScalar *assembled_qf_array, CeedScalar *elem_diag_array) { 816 const CeedSize num_nodes = diag->num_nodes; 817 const CeedSize num_qpts = diag->num_qpts; 818 const CeedSize num_comp = diag->num_comp; 819 const CeedSize num_eval_mode_in = diag->num_eval_mode_in; 820 const CeedSize num_eval_mode_out = diag->num_eval_mode_out; 821 const CeedScalar *identity = diag->d_identity; 822 const CeedScalar *interp_in = diag->d_interp_in; 823 const CeedScalar *grad_in = diag->d_grad_in; 824 const CeedScalar *interp_out = diag->d_interp_out; 825 const CeedScalar *grad_out = diag->d_grad_out; 826 const CeedEvalMode *eval_mode_in = diag->d_eval_mode_in; 827 const CeedEvalMode *eval_mode_out = diag->d_eval_mode_out; 828 829 sycl::range<1> kernel_range(num_elem * num_nodes); 830 831 std::vector<sycl::event> e; 832 833 if (!sycl_queue.is_in_order()) e = {sycl_queue.ext_oneapi_submit_barrier()}; 834 835 sycl_queue.parallel_for<CeedOperatorSyclLinearDiagonal>(kernel_range, e, [=](sycl::id<1> idx) { 836 const CeedInt tid = idx % num_nodes; 837 const CeedInt e = idx / num_nodes; 838 839 // Compute the diagonal of B^T D B 840 // Each element 841 CeedInt d_out = -1; 842 // Each basis eval mode pair 843 for (CeedSize e_out = 0; e_out < num_eval_mode_out; e_out++) { 844 const CeedScalar *bt = NULL; 845 846 if (eval_mode_out[e_out] == CEED_EVAL_GRAD) ++d_out; 847 CeedOperatorGetBasisPointer_Sycl(&bt, eval_mode_out[e_out], identity, interp_out, &grad_out[d_out * num_qpts * num_nodes]); 848 CeedInt d_in = -1; 849 850 for (CeedSize e_in = 0; e_in < num_eval_mode_in; e_in++) { 851 const CeedScalar *b = NULL; 852 853 if (eval_mode_in[e_in] == CEED_EVAL_GRAD) ++d_in; 854 CeedOperatorGetBasisPointer_Sycl(&b, eval_mode_in[e_in], identity, interp_in, &grad_in[d_in * num_qpts * num_nodes]); 855 // Each component 856 for (CeedSize comp_out = 0; comp_out < num_comp; comp_out++) { 857 // Each qpoint/node pair 858 if (is_point_block) { 859 // Point Block Diagonal 860 for (CeedInt comp_in = 0; comp_in < num_comp; comp_in++) { 861 CeedScalar e_value = 0.0; 862 863 for (CeedSize q = 0; q < num_qpts; q++) { 864 const CeedScalar qf_value = 865 assembled_qf_array[((((e_in * num_comp + comp_in) * num_eval_mode_out + e_out) * num_comp + comp_out) * num_elem + e) * num_qpts + 866 q]; 867 868 e_value += bt[q * num_nodes + tid] * qf_value * b[q * num_nodes + tid]; 869 } 870 elem_diag_array[((comp_out * num_comp + comp_in) * num_elem + e) * num_nodes + tid] += e_value; 871 } 872 } else { 873 // Diagonal Only 874 CeedScalar e_value = 0.0; 875 876 for (CeedSize q = 0; q < num_qpts; q++) { 877 const CeedScalar qf_value = 878 assembled_qf_array[((((e_in * num_comp + comp_out) * num_eval_mode_out + e_out) * num_comp + comp_out) * num_elem + e) * num_qpts + 879 q]; 880 e_value += bt[q * num_nodes + tid] * qf_value * b[q * num_nodes + tid]; 881 } 882 elem_diag_array[(comp_out * num_elem + e) * num_nodes + tid] += e_value; 883 } 884 } 885 } 886 } 887 }); 888 return CEED_ERROR_SUCCESS; 889 } 890 891 //------------------------------------------------------------------------------ 892 // Assemble diagonal common code 893 //------------------------------------------------------------------------------ 894 static inline int CeedOperatorAssembleDiagonalCore_Sycl(CeedOperator op, CeedVector assembled, CeedRequest *request, const bool is_point_block) { 895 Ceed ceed; 896 Ceed_Sycl *sycl_data; 897 CeedInt num_elem; 898 CeedScalar *elem_diag_array; 899 const CeedScalar *assembled_qf_array; 900 CeedVector assembled_qf = NULL; 901 CeedElemRestriction rstr = NULL; 902 CeedOperator_Sycl *impl; 903 904 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 905 CeedCallBackend(CeedOperatorGetData(op, &impl)); 906 CeedCallBackend(CeedGetData(ceed, &sycl_data)); 907 908 // Assemble QFunction 909 CeedCallBackend(CeedOperatorLinearAssembleQFunctionBuildOrUpdate(op, &assembled_qf, &rstr, request)); 910 CeedCallBackend(CeedElemRestrictionDestroy(&rstr)); 911 912 // Setup 913 if (!impl->diag) { 914 CeedCallBackend(CeedOperatorAssembleDiagonalSetup_Sycl(op)); 915 } 916 CeedOperatorDiag_Sycl *diag = impl->diag; 917 918 assert(diag != NULL); 919 920 // Restriction 921 if (is_point_block && !diag->point_block_diag_rstr) { 922 CeedCallBackend(CeedOperatorCreateActivePointBlockRestriction(diag->diag_rstr, &diag->point_block_diag_rstr)); 923 } 924 CeedElemRestriction diag_rstr = is_point_block ? diag->point_block_diag_rstr : diag->diag_rstr; 925 926 // Create diagonal vector 927 CeedVector elem_diag = is_point_block ? diag->point_block_elem_diag : diag->elem_diag; 928 929 if (!elem_diag) { 930 CeedCallBackend(CeedElemRestrictionCreateVector(diag_rstr, NULL, &elem_diag)); 931 if (is_point_block) diag->point_block_elem_diag = elem_diag; 932 else diag->elem_diag = elem_diag; 933 } 934 CeedCallBackend(CeedVectorSetValue(elem_diag, 0.0)); 935 936 // Assemble element operator diagonals 937 CeedCallBackend(CeedVectorGetArray(elem_diag, CEED_MEM_DEVICE, &elem_diag_array)); 938 CeedCallBackend(CeedVectorGetArrayRead(assembled_qf, CEED_MEM_DEVICE, &assembled_qf_array)); 939 CeedCallBackend(CeedElemRestrictionGetNumElements(diag_rstr, &num_elem)); 940 941 // Compute the diagonal of B^T D B 942 CeedCallBackend(CeedOperatorLinearDiagonal_Sycl(sycl_data->sycl_queue, is_point_block, num_elem, diag, assembled_qf_array, elem_diag_array)); 943 944 // Wait for queue to complete and handle exceptions 945 sycl_data->sycl_queue.wait_and_throw(); 946 947 // Restore arrays 948 CeedCallBackend(CeedVectorRestoreArray(elem_diag, &elem_diag_array)); 949 CeedCallBackend(CeedVectorRestoreArrayRead(assembled_qf, &assembled_qf_array)); 950 951 // Assemble local operator diagonal 952 CeedCallBackend(CeedElemRestrictionApply(diag_rstr, CEED_TRANSPOSE, elem_diag, assembled, request)); 953 954 // Cleanup 955 CeedCallBackend(CeedVectorDestroy(&assembled_qf)); 956 return CEED_ERROR_SUCCESS; 957 } 958 959 //------------------------------------------------------------------------------ 960 // Assemble Linear Diagonal 961 //------------------------------------------------------------------------------ 962 static int CeedOperatorLinearAssembleAddDiagonal_Sycl(CeedOperator op, CeedVector assembled, CeedRequest *request) { 963 CeedCallBackend(CeedOperatorAssembleDiagonalCore_Sycl(op, assembled, request, false)); 964 return CEED_ERROR_SUCCESS; 965 } 966 967 //------------------------------------------------------------------------------ 968 // Assemble Linear Point Block Diagonal 969 //------------------------------------------------------------------------------ 970 static int CeedOperatorLinearAssembleAddPointBlockDiagonal_Sycl(CeedOperator op, CeedVector assembled, CeedRequest *request) { 971 CeedCallBackend(CeedOperatorAssembleDiagonalCore_Sycl(op, assembled, request, true)); 972 return CEED_ERROR_SUCCESS; 973 } 974 975 //------------------------------------------------------------------------------ 976 // Single operator assembly setup 977 //------------------------------------------------------------------------------ 978 static int CeedSingleOperatorAssembleSetup_Sycl(CeedOperator op) { 979 Ceed ceed; 980 CeedInt num_input_fields, num_output_fields, num_eval_mode_in = 0, dim = 1, num_B_in_mats_to_load = 0, size_B_in = 0, num_eval_mode_out = 0, 981 num_B_out_mats_to_load = 0, size_B_out = 0, num_qpts = 0, elem_size = 0, num_elem, num_comp, 982 mat_start = 0; 983 CeedEvalMode *eval_mode_in = NULL, *eval_mode_out = NULL; 984 const CeedScalar *interp_in, *grad_in; 985 CeedElemRestriction rstr_in = NULL, rstr_out = NULL; 986 CeedBasis basis_in = NULL, basis_out = NULL; 987 CeedQFunctionField *qf_fields; 988 CeedQFunction qf; 989 CeedOperatorField *input_fields, *output_fields; 990 CeedOperator_Sycl *impl; 991 992 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 993 CeedCallBackend(CeedOperatorGetData(op, &impl)); 994 995 // Get input and output fields 996 CeedCallBackend(CeedOperatorGetFields(op, &num_input_fields, &input_fields, &num_output_fields, &output_fields)); 997 998 // Determine active input basis eval mode 999 CeedCallBackend(CeedOperatorGetQFunction(op, &qf)); 1000 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_fields, NULL, NULL)); 1001 // Note that the kernel will treat each dimension of a gradient action separately; 1002 // i.e., when an active input has a CEED_EVAL_GRAD mode, num_ eval_mode_in will increment by dim. 1003 // However, for the purposes of loading the B matrices, it will be treated as one mode, and we will load/copy the entire gradient matrix at once, 1004 // so num_B_in_mats_to_load will be incremented by 1. 1005 for (CeedInt i = 0; i < num_input_fields; i++) { 1006 CeedVector vec; 1007 1008 CeedCallBackend(CeedOperatorFieldGetVector(input_fields[i], &vec)); 1009 if (vec == CEED_VECTOR_ACTIVE) { 1010 CeedEvalMode eval_mode; 1011 CeedElemRestriction elem_rstr; 1012 CeedBasis basis; 1013 1014 CeedCallBackend(CeedOperatorFieldGetBasis(input_fields[i], &basis)); 1015 CeedCheck(!basis_in || basis_in == basis, ceed, CEED_ERROR_BACKEND, "Backend does not implement operator assembly with multiple active bases"); 1016 if (!basis_in) CeedCallBackend(CeedBasisReferenceCopy(basis, &basis_in)); 1017 CeedCallBackend(CeedBasisDestroy(&basis)); 1018 CeedCallBackend(CeedBasisGetDimension(basis_in, &dim)); 1019 CeedCallBackend(CeedBasisGetNumQuadraturePoints(basis_in, &num_qpts)); 1020 CeedCallBackend(CeedOperatorFieldGetElemRestriction(input_fields[i], &elem_rstr)); 1021 if (!rstr_in) CeedCallBackend(CeedElemRestrictionReferenceCopy(elem_rstr, &rstr_in)); 1022 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 1023 CeedCallBackend(CeedElemRestrictionGetElementSize(rstr_in, &elem_size)); 1024 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode)); 1025 if (eval_mode != CEED_EVAL_NONE) { 1026 CeedCallBackend(CeedRealloc(num_B_in_mats_to_load + 1, &eval_mode_in)); 1027 eval_mode_in[num_B_in_mats_to_load] = eval_mode; 1028 num_B_in_mats_to_load += 1; 1029 if (eval_mode == CEED_EVAL_GRAD) { 1030 num_eval_mode_in += dim; 1031 size_B_in += dim * elem_size * num_qpts; 1032 } else { 1033 num_eval_mode_in += 1; 1034 size_B_in += elem_size * num_qpts; 1035 } 1036 } 1037 } 1038 CeedCallBackend(CeedVectorDestroy(&vec)); 1039 } 1040 1041 // Determine active output basis; basis_out and rstr_out only used if same as input, TODO 1042 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, NULL, NULL, &qf_fields)); 1043 for (CeedInt i = 0; i < num_output_fields; i++) { 1044 CeedVector vec; 1045 1046 CeedCallBackend(CeedOperatorFieldGetVector(output_fields[i], &vec)); 1047 if (vec == CEED_VECTOR_ACTIVE) { 1048 CeedEvalMode eval_mode; 1049 CeedElemRestriction elem_rstr; 1050 CeedBasis basis; 1051 1052 CeedCallBackend(CeedOperatorFieldGetBasis(output_fields[i], &basis)); 1053 CeedCheck(!basis_out || basis_out == basis, ceed, CEED_ERROR_BACKEND, 1054 "Backend does not implement operator assembly with multiple active bases"); 1055 if (!basis_out) CeedCallBackend(CeedBasisReferenceCopy(basis, &basis_out)); 1056 CeedCallBackend(CeedBasisDestroy(&basis)); 1057 CeedCallBackend(CeedOperatorFieldGetElemRestriction(output_fields[i], &elem_rstr)); 1058 if (!rstr_out) CeedCallBackend(CeedElemRestrictionReferenceCopy(elem_rstr, &rstr_out)); 1059 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 1060 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode)); 1061 if (eval_mode != CEED_EVAL_NONE) { 1062 CeedCallBackend(CeedRealloc(num_B_out_mats_to_load + 1, &eval_mode_out)); 1063 eval_mode_out[num_B_out_mats_to_load] = eval_mode; 1064 num_B_out_mats_to_load += 1; 1065 if (eval_mode == CEED_EVAL_GRAD) { 1066 num_eval_mode_out += dim; 1067 size_B_out += dim * elem_size * num_qpts; 1068 } else { 1069 num_eval_mode_out += 1; 1070 size_B_out += elem_size * num_qpts; 1071 } 1072 } 1073 } 1074 CeedCallBackend(CeedVectorDestroy(&vec)); 1075 } 1076 CeedCheck(num_eval_mode_in > 0 && num_eval_mode_out > 0, ceed, CEED_ERROR_UNSUPPORTED, "Cannot assemble operator without inputs/outputs"); 1077 1078 CeedCallBackend(CeedElemRestrictionGetNumElements(rstr_in, &num_elem)); 1079 CeedCallBackend(CeedElemRestrictionGetNumComponents(rstr_in, &num_comp)); 1080 1081 CeedCallBackend(CeedCalloc(1, &impl->asmb)); 1082 CeedOperatorAssemble_Sycl *asmb = impl->asmb; 1083 asmb->num_elem = num_elem; 1084 1085 Ceed_Sycl *sycl_data; 1086 CeedCallBackend(CeedGetData(ceed, &sycl_data)); 1087 1088 // Kernel setup 1089 int elems_per_block = 1; 1090 asmb->elems_per_block = elems_per_block; 1091 asmb->block_size_x = elem_size; 1092 asmb->block_size_y = elem_size; 1093 asmb->num_eval_mode_in = num_eval_mode_in; 1094 asmb->num_eval_mode_out = num_eval_mode_out; 1095 asmb->num_qpts = num_qpts; 1096 asmb->num_nodes = elem_size; 1097 asmb->block_size = elem_size * elem_size * elems_per_block; 1098 asmb->num_comp = num_comp; 1099 1100 // Build 'full' B matrices (not 1D arrays used for tensor-product matrices 1101 CeedCallBackend(CeedBasisGetInterp(basis_in, &interp_in)); 1102 CeedCallBackend(CeedBasisGetGrad(basis_in, &grad_in)); 1103 1104 // Load into B_in, in order that they will be used in eval_mode 1105 CeedCallSycl(ceed, asmb->d_B_in = sycl::malloc_device<CeedScalar>(size_B_in, sycl_data->sycl_device, sycl_data->sycl_context)); 1106 for (int i = 0; i < num_B_in_mats_to_load; i++) { 1107 CeedEvalMode eval_mode = eval_mode_in[i]; 1108 1109 if (eval_mode == CEED_EVAL_INTERP) { 1110 std::vector<sycl::event> e; 1111 1112 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()}; 1113 sycl_data->sycl_queue.copy<CeedScalar>(interp_in, &asmb->d_B_in[mat_start], elem_size * num_qpts, e); 1114 mat_start += elem_size * num_qpts; 1115 } else if (eval_mode == CEED_EVAL_GRAD) { 1116 std::vector<sycl::event> e; 1117 1118 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()}; 1119 sycl_data->sycl_queue.copy<CeedScalar>(grad_in, &asmb->d_B_in[mat_start], dim * elem_size * num_qpts, e); 1120 mat_start += dim * elem_size * num_qpts; 1121 } 1122 } 1123 1124 const CeedScalar *interp_out, *grad_out; 1125 // Note that this function currently assumes 1 basis, so this should always be true 1126 // for now 1127 if (basis_out == basis_in) { 1128 interp_out = interp_in; 1129 grad_out = grad_in; 1130 } else { 1131 CeedCallBackend(CeedBasisGetInterp(basis_out, &interp_out)); 1132 CeedCallBackend(CeedBasisGetGrad(basis_out, &grad_out)); 1133 } 1134 1135 // Load into B_out, in order that they will be used in eval_mode 1136 mat_start = 0; 1137 CeedCallSycl(ceed, asmb->d_B_out = sycl::malloc_device<CeedScalar>(size_B_out, sycl_data->sycl_device, sycl_data->sycl_context)); 1138 for (int i = 0; i < num_B_out_mats_to_load; i++) { 1139 CeedEvalMode eval_mode = eval_mode_out[i]; 1140 1141 if (eval_mode == CEED_EVAL_INTERP) { 1142 std::vector<sycl::event> e; 1143 1144 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()}; 1145 sycl_data->sycl_queue.copy<CeedScalar>(interp_out, &asmb->d_B_out[mat_start], elem_size * num_qpts, e); 1146 mat_start += elem_size * num_qpts; 1147 } else if (eval_mode == CEED_EVAL_GRAD) { 1148 std::vector<sycl::event> e; 1149 1150 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()}; 1151 sycl_data->sycl_queue.copy<CeedScalar>(grad_out, &asmb->d_B_out[mat_start], dim * elem_size * num_qpts, e); 1152 mat_start += dim * elem_size * num_qpts; 1153 } 1154 } 1155 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_in)); 1156 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_out)); 1157 CeedCallBackend(CeedBasisDestroy(&basis_in)); 1158 CeedCallBackend(CeedBasisDestroy(&basis_out)); 1159 return CEED_ERROR_SUCCESS; 1160 } 1161 1162 //------------------------------------------------------------------------------ 1163 // Matrix assembly kernel for low-order elements (3D thread block) 1164 //------------------------------------------------------------------------------ 1165 static int CeedOperatorLinearAssemble_Sycl(sycl::queue &sycl_queue, const CeedOperator_Sycl *impl, const CeedScalar *qf_array, 1166 CeedScalar *values_array) { 1167 // This kernels assumes B_in and B_out have the same number of quadrature points and basis points. 1168 // TODO: expand to more general cases 1169 CeedOperatorAssemble_Sycl *asmb = impl->asmb; 1170 const CeedInt num_elem = asmb->num_elem; 1171 const CeedSize num_nodes = asmb->num_nodes; 1172 const CeedSize num_comp = asmb->num_comp; 1173 const CeedSize num_qpts = asmb->num_qpts; 1174 const CeedSize num_eval_mode_in = asmb->num_eval_mode_in; 1175 const CeedSize num_eval_mode_out = asmb->num_eval_mode_out; 1176 1177 // Strides for final output ordering, determined by the reference (inference) implementation of the symbolic assembly, slowest --> fastest: element, 1178 // comp_in, comp_out, node_row, node_col 1179 const CeedSize comp_out_stride = num_nodes * num_nodes; 1180 const CeedSize comp_in_stride = comp_out_stride * num_comp; 1181 const CeedSize e_stride = comp_in_stride * num_comp; 1182 // Strides for QF array, slowest --> fastest: eval_mode_in, comp_in, eval_mode_out, comp_out, elem, qpt 1183 const CeedSize q_e_stride = num_qpts; 1184 const CeedSize q_comp_out_stride = num_elem * q_e_stride; 1185 const CeedSize q_eval_mode_out_stride = q_comp_out_stride * num_comp; 1186 const CeedSize q_comp_in_stride = q_eval_mode_out_stride * num_eval_mode_out; 1187 const CeedSize q_eval_mode_in_stride = q_comp_in_stride * num_comp; 1188 1189 CeedScalar *B_in, *B_out; 1190 B_in = asmb->d_B_in; 1191 B_out = asmb->d_B_out; 1192 const CeedInt block_size_x = asmb->block_size_x; 1193 const CeedInt block_size_y = asmb->block_size_y; 1194 1195 sycl::range<3> kernel_range(num_elem, block_size_y, block_size_x); 1196 1197 std::vector<sycl::event> e; 1198 1199 if (!sycl_queue.is_in_order()) e = {sycl_queue.ext_oneapi_submit_barrier()}; 1200 sycl_queue.parallel_for<CeedOperatorSyclLinearAssemble>(kernel_range, e, [=](sycl::id<3> idx) { 1201 const int e = idx.get(0); // Element index 1202 const int l = idx.get(1); // The output column index of each B^TDB operation 1203 const int i = idx.get(2); // The output row index of each B^TDB operation 1204 // such that we have (Bout^T)_ij D_jk Bin_kl = C_il 1205 for (CeedSize comp_in = 0; comp_in < num_comp; comp_in++) { 1206 for (CeedSize comp_out = 0; comp_out < num_comp; comp_out++) { 1207 CeedScalar result = 0.0; 1208 CeedSize qf_index_comp = q_comp_in_stride * comp_in + q_comp_out_stride * comp_out + q_e_stride * e; 1209 1210 for (CeedSize eval_mode_in = 0; eval_mode_in < num_eval_mode_in; eval_mode_in++) { 1211 CeedSize b_in_index = eval_mode_in * num_qpts * num_nodes; 1212 1213 for (CeedSize eval_mode_out = 0; eval_mode_out < num_eval_mode_out; eval_mode_out++) { 1214 CeedSize b_out_index = eval_mode_out * num_qpts * num_nodes; 1215 CeedSize qf_index = qf_index_comp + q_eval_mode_out_stride * eval_mode_out + q_eval_mode_in_stride * eval_mode_in; 1216 1217 // Perform the B^T D B operation for this 'chunk' of D (the qf_array) 1218 for (CeedSize j = 0; j < num_qpts; j++) { 1219 result += B_out[b_out_index + j * num_nodes + i] * qf_array[qf_index + j] * B_in[b_in_index + j * num_nodes + l]; 1220 } 1221 } // end of eval_mode_out 1222 } // end of eval_mode_in 1223 CeedSize val_index = comp_in_stride * comp_in + comp_out_stride * comp_out + e_stride * e + num_nodes * i + l; 1224 1225 values_array[val_index] = result; 1226 } // end of out component 1227 } // end of in component 1228 }); 1229 return CEED_ERROR_SUCCESS; 1230 } 1231 1232 //------------------------------------------------------------------------------ 1233 // Fallback kernel for larger orders (1D thread block) 1234 //------------------------------------------------------------------------------ 1235 /* 1236 static int CeedOperatorLinearAssembleFallback_Sycl(sycl::queue &sycl_queue, const CeedOperator_Sycl *impl, const CeedScalar *qf_array, 1237 CeedScalar *values_array) { 1238 // This kernel assumes B_in and B_out have the same number of quadrature points and basis points. 1239 // TODO: expand to more general cases 1240 CeedOperatorAssemble_Sycl *asmb = impl->asmb; 1241 const CeedInt num_elem = asmb->num_elem; 1242 const CeedInt num_nodes = asmb->num_nodes; 1243 const CeedInt num_comp = asmb->num_comp; 1244 const CeedInt num_qpts = asmb->num_qpts; 1245 const CeedInt num_eval_mode_in = asmb->num_eval_mode_in; 1246 const CeedInt num_eval_mode_out = asmb->num_eval_mode_out; 1247 1248 // Strides for final output ordering, determined by the reference (interface) implementation of the symbolic assembly, slowest --> fastest: elememt, 1249 // comp_in, comp_out, node_row, node_col 1250 const CeedInt comp_out_stride = num_nodes * num_nodes; 1251 const CeedInt comp_in_stride = comp_out_stride * num_comp; 1252 const CeedInt e_stride = comp_in_stride * num_comp; 1253 // Strides for QF array, slowest --> fastest: eval_mode_in, comp_in, eval_mode_out, comp_out, elem, qpt 1254 const CeedInt q_e_stride = num_qpts; 1255 const CeedInt q_comp_out_stride = num_elem * q_e_stride; 1256 const CeedInt q_eval_mode_out_stride = q_comp_out_stride * num_comp; 1257 const CeedInt q_comp_in_stride = q_eval_mode_out_stride * num_eval_mode_out; 1258 const CeedInt q_eval_mode_in_stride = q_comp_in_stride * num_comp; 1259 1260 CeedScalar *B_in, *B_out; 1261 B_in = asmb->d_B_in; 1262 B_out = asmb->d_B_out; 1263 const CeedInt elems_per_block = asmb->elems_per_block; 1264 const CeedInt block_size_x = asmb->block_size_x; 1265 const CeedInt block_size_y = asmb->block_size_y; // This will be 1 for the fallback kernel 1266 1267 const CeedInt grid = num_elem / elems_per_block + ((num_elem / elems_per_block * elems_per_block < num_elem) ? 1 : 0); 1268 sycl::range<3> local_range(block_size_x, block_size_y, elems_per_block); 1269 sycl::range<3> global_range(grid * block_size_x, block_size_y, elems_per_block); 1270 sycl::nd_range<3> kernel_range(global_range, local_range); 1271 1272 sycl_queue.parallel_for<CeedOperatorSyclLinearAssembleFallback>(kernel_range, [=](sycl::nd_item<3> work_item) { 1273 const CeedInt blockIdx = work_item.get_group(0); 1274 const CeedInt gridDimx = work_item.get_group_range(0); 1275 const CeedInt threadIdx = work_item.get_local_id(0); 1276 const CeedInt threadIdz = work_item.get_local_id(2); 1277 const CeedInt blockDimz = work_item.get_local_range(2); 1278 1279 const int l = threadIdx; // The output column index of each B^TDB operation 1280 // such that we have (Bout^T)_ij D_jk Bin_kl = C_il 1281 for (CeedInt e = blockIdx * blockDimz + threadIdz; e < num_elem; e += gridDimx * blockDimz) { 1282 for (CeedInt comp_in = 0; comp_in < num_comp; comp_in++) { 1283 for (CeedInt comp_out = 0; comp_out < num_comp; comp_out++) { 1284 for (CeedInt i = 0; i < num_nodes; i++) { 1285 CeedScalar result = 0.0; 1286 CeedInt qf_index_comp = q_comp_in_stride * comp_in + q_comp_out_stride * comp_out + q_e_stride * e; 1287 for (CeedInt eval_mode_in = 0; eval_mode_in < num_eval_mode_in; eval_mode_in++) { 1288 CeedInt b_in_index = eval_mode_in * num_qpts * num_nodes; 1289 for (CeedInt eval_mode_out = 0; eval_mode_out < num_eval_mode_out; eval_mode_out++) { 1290 CeedInt b_out_index = eval_mode_out * num_qpts * num_nodes; 1291 CeedInt qf_index = qf_index_comp + q_eval_mode_out_stride * eval_mode_out + q_eval_mode_in_stride * eval_mode_in; 1292 // Perform the B^T D B operation for this 'chunk' of D (the qf_array) 1293 for (CeedInt j = 0; j < num_qpts; j++) { 1294 result += B_out[b_out_index + j * num_nodes + i] * qf_array[qf_index + j] * B_in[b_in_index + j * num_nodes + l]; 1295 } 1296 } // end of eval_mode_out 1297 } // end of eval_mode_in 1298 CeedInt val_index = comp_in_stride * comp_in + comp_out_stride * comp_out + e_stride * e + num_nodes * i + l; 1299 values_array[val_index] = result; 1300 } // end of loop over element node index, i 1301 } // end of out component 1302 } // end of in component 1303 } // end of element loop 1304 }); 1305 return CEED_ERROR_SUCCESS; 1306 }*/ 1307 1308 //------------------------------------------------------------------------------ 1309 // Assemble matrix data for COO matrix of assembled operator. 1310 // The sparsity pattern is set by CeedOperatorLinearAssembleSymbolic. 1311 // 1312 // Note that this (and other assembly routines) currently assume only one active 1313 // input restriction/basis per operator (could have multiple basis eval modes). 1314 // TODO: allow multiple active input restrictions/basis objects 1315 //------------------------------------------------------------------------------ 1316 static int CeedSingleOperatorAssemble_Sycl(CeedOperator op, CeedInt offset, CeedVector values) { 1317 Ceed ceed; 1318 Ceed_Sycl *sycl_data; 1319 CeedScalar *values_array; 1320 const CeedScalar *qf_array; 1321 CeedVector assembled_qf = NULL; 1322 CeedElemRestriction rstr_q = NULL; 1323 CeedOperator_Sycl *impl; 1324 1325 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 1326 CeedCallBackend(CeedOperatorGetData(op, &impl)); 1327 CeedCallBackend(CeedGetData(ceed, &sycl_data)); 1328 1329 // Setup 1330 if (!impl->asmb) { 1331 CeedCallBackend(CeedSingleOperatorAssembleSetup_Sycl(op)); 1332 assert(impl->asmb != NULL); 1333 } 1334 1335 // Assemble QFunction 1336 CeedCallBackend(CeedOperatorLinearAssembleQFunctionBuildOrUpdate(op, &assembled_qf, &rstr_q, CEED_REQUEST_IMMEDIATE)); 1337 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_q)); 1338 CeedCallBackend(CeedVectorGetArrayWrite(values, CEED_MEM_DEVICE, &values_array)); 1339 values_array += offset; 1340 CeedCallBackend(CeedVectorGetArrayRead(assembled_qf, CEED_MEM_DEVICE, &qf_array)); 1341 1342 // Compute B^T D B 1343 CeedCallBackend(CeedOperatorLinearAssemble_Sycl(sycl_data->sycl_queue, impl, qf_array, values_array)); 1344 1345 // Wait for kernels to be completed 1346 // Kris: Review if this is necessary -- enqueing an async barrier may be sufficient 1347 sycl_data->sycl_queue.wait_and_throw(); 1348 1349 // Restore arrays 1350 CeedCallBackend(CeedVectorRestoreArray(values, &values_array)); 1351 CeedCallBackend(CeedVectorRestoreArrayRead(assembled_qf, &qf_array)); 1352 1353 // Cleanup 1354 CeedCallBackend(CeedVectorDestroy(&assembled_qf)); 1355 return CEED_ERROR_SUCCESS; 1356 } 1357 1358 //------------------------------------------------------------------------------ 1359 // Create operator 1360 //------------------------------------------------------------------------------ 1361 int CeedOperatorCreate_Sycl(CeedOperator op) { 1362 Ceed ceed; 1363 CeedOperator_Sycl *impl; 1364 1365 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 1366 1367 CeedCallBackend(CeedCalloc(1, &impl)); 1368 CeedCallBackend(CeedOperatorSetData(op, impl)); 1369 1370 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleQFunction", CeedOperatorLinearAssembleQFunction_Sycl)); 1371 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleQFunctionUpdate", CeedOperatorLinearAssembleQFunctionUpdate_Sycl)); 1372 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleAddDiagonal", CeedOperatorLinearAssembleAddDiagonal_Sycl)); 1373 CeedCallBackend( 1374 CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleAddPointBlockDiagonal", CeedOperatorLinearAssembleAddPointBlockDiagonal_Sycl)); 1375 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleSingle", CeedSingleOperatorAssemble_Sycl)); 1376 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "ApplyAdd", CeedOperatorApplyAdd_Sycl)); 1377 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "Destroy", CeedOperatorDestroy_Sycl)); 1378 return CEED_ERROR_SUCCESS; 1379 } 1380 1381 //------------------------------------------------------------------------------ 1382