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 data 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 93 CeedCallBackend(CeedVectorDestroy(&impl->diag->elem_diag)); 94 CeedCallBackend(CeedVectorDestroy(&impl->diag->point_block_elem_diag)); 95 CeedCallBackend(CeedElemRestrictionDestroy(&impl->diag->diag_rstr)); 96 CeedCallBackend(CeedElemRestrictionDestroy(&impl->diag->point_block_diag_rstr)); 97 CeedCallBackend(CeedBasisDestroy(&impl->diag->basis_in)); 98 CeedCallBackend(CeedBasisDestroy(&impl->diag->basis_out)); 99 } 100 CeedCallBackend(CeedFree(&impl->diag)); 101 102 if (impl->asmb) { 103 CeedCallSycl(ceed, sycl_data->sycl_queue.wait_and_throw()); 104 CeedCallSycl(ceed, sycl::free(impl->asmb->d_B_in, sycl_data->sycl_context)); 105 CeedCallSycl(ceed, sycl::free(impl->asmb->d_B_out, sycl_data->sycl_context)); 106 } 107 CeedCallBackend(CeedFree(&impl->asmb)); 108 109 CeedCallBackend(CeedFree(&impl)); 110 return CEED_ERROR_SUCCESS; 111 } 112 113 //------------------------------------------------------------------------------ 114 // Setup infields or outfields 115 //------------------------------------------------------------------------------ 116 static int CeedOperatorSetupFields_Sycl(CeedQFunction qf, CeedOperator op, bool is_input, CeedVector *e_vecs, CeedVector *q_vecs, CeedInt start_e, 117 CeedInt num_fields, CeedInt Q, CeedInt num_elem) { 118 Ceed ceed; 119 CeedSize q_size; 120 bool is_strided, skip_restriction; 121 CeedInt size; 122 CeedOperatorField *op_fields; 123 CeedQFunctionField *qf_fields; 124 125 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 126 if (is_input) { 127 CeedCallBackend(CeedOperatorGetFields(op, NULL, &op_fields, NULL, NULL)); 128 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_fields, NULL, NULL)); 129 } else { 130 CeedCallBackend(CeedOperatorGetFields(op, NULL, NULL, NULL, &op_fields)); 131 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, NULL, NULL, &qf_fields)); 132 } 133 134 // Loop over fields 135 for (CeedInt i = 0; i < num_fields; i++) { 136 CeedEvalMode eval_mode; 137 CeedVector vec; 138 CeedElemRestriction elem_rstr; 139 140 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode)); 141 142 is_strided = false; 143 skip_restriction = false; 144 if (eval_mode != CEED_EVAL_WEIGHT) { 145 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_fields[i], &elem_rstr)); 146 147 // Check whether this field can skip the element restriction: 148 // must be passive input, with eval_mode NONE, and have a strided restriction with CEED_STRIDES_BACKEND. 149 150 // First, check whether the field is input or output: 151 if (is_input) { 152 // Check for passive input: 153 CeedCallBackend(CeedOperatorFieldGetVector(op_fields[i], &vec)); 154 if (vec != CEED_VECTOR_ACTIVE) { 155 // Check eval_mode 156 if (eval_mode == CEED_EVAL_NONE) { 157 // Check for is_strided restriction 158 CeedCallBackend(CeedElemRestrictionIsStrided(elem_rstr, &is_strided)); 159 if (is_strided) { 160 // Check if vector is already in preferred backend ordering 161 CeedCallBackend(CeedElemRestrictionHasBackendStrides(elem_rstr, &skip_restriction)); 162 } 163 } 164 } 165 CeedCallBackend(CeedVectorDestroy(&vec)); 166 } 167 if (skip_restriction) { 168 // We do not need an E-Vector, but will use the input field vector's data directly in the operator application 169 e_vecs[i + start_e] = NULL; 170 } else { 171 CeedCallBackend(CeedElemRestrictionCreateVector(elem_rstr, NULL, &e_vecs[i + start_e])); 172 } 173 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 174 } 175 176 switch (eval_mode) { 177 case CEED_EVAL_NONE: 178 CeedCallBackend(CeedQFunctionFieldGetSize(qf_fields[i], &size)); 179 q_size = (CeedSize)num_elem * Q * size; 180 CeedCallBackend(CeedVectorCreate(ceed, q_size, &q_vecs[i])); 181 break; 182 case CEED_EVAL_INTERP: 183 CeedCallBackend(CeedQFunctionFieldGetSize(qf_fields[i], &size)); 184 q_size = (CeedSize)num_elem * Q * size; 185 CeedCallBackend(CeedVectorCreate(ceed, q_size, &q_vecs[i])); 186 break; 187 case CEED_EVAL_GRAD: 188 CeedCallBackend(CeedQFunctionFieldGetSize(qf_fields[i], &size)); 189 q_size = (CeedSize)num_elem * Q * size; 190 CeedCallBackend(CeedVectorCreate(ceed, q_size, &q_vecs[i])); 191 break; 192 case CEED_EVAL_WEIGHT: { 193 CeedBasis basis; 194 195 // Note: only on input fields 196 q_size = (CeedSize)num_elem * Q; 197 CeedCallBackend(CeedVectorCreate(ceed, q_size, &q_vecs[i])); 198 CeedCallBackend(CeedOperatorFieldGetBasis(op_fields[i], &basis)); 199 CeedCallBackend(CeedBasisApply(basis, num_elem, CEED_NOTRANSPOSE, CEED_EVAL_WEIGHT, CEED_VECTOR_NONE, q_vecs[i])); 200 CeedCallBackend(CeedBasisDestroy(&basis)); 201 break; 202 } 203 case CEED_EVAL_DIV: 204 break; // TODO: Not implemented 205 case CEED_EVAL_CURL: 206 break; // TODO: Not implemented 207 } 208 } 209 return CEED_ERROR_SUCCESS; 210 } 211 212 //------------------------------------------------------------------------------ 213 // CeedOperator needs to connect all the named fields (be they active or 214 // passive) to the named inputs and outputs of its CeedQFunction. 215 //------------------------------------------------------------------------------ 216 static int CeedOperatorSetup_Sycl(CeedOperator op) { 217 Ceed ceed; 218 bool is_setup_done; 219 CeedInt Q, num_elem, num_input_fields, num_output_fields; 220 CeedQFunctionField *qf_input_fields, *qf_output_fields; 221 CeedQFunction qf; 222 CeedOperatorField *op_input_fields, *op_output_fields; 223 CeedOperator_Sycl *impl; 224 225 CeedCallBackend(CeedOperatorIsSetupDone(op, &is_setup_done)); 226 if (is_setup_done) return CEED_ERROR_SUCCESS; 227 228 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 229 CeedCallBackend(CeedOperatorGetData(op, &impl)); 230 CeedCallBackend(CeedOperatorGetQFunction(op, &qf)); 231 CeedCallBackend(CeedOperatorGetNumQuadraturePoints(op, &Q)); 232 CeedCallBackend(CeedOperatorGetNumElements(op, &num_elem)); 233 CeedCallBackend(CeedOperatorGetFields(op, &num_input_fields, &op_input_fields, &num_output_fields, &op_output_fields)); 234 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_input_fields, NULL, &qf_output_fields)); 235 236 // Allocate 237 CeedCallBackend(CeedCalloc(num_input_fields + num_output_fields, &impl->e_vecs)); 238 239 CeedCallBackend(CeedCalloc(CEED_FIELD_MAX, &impl->q_vecs_in)); 240 CeedCallBackend(CeedCalloc(CEED_FIELD_MAX, &impl->q_vecs_out)); 241 242 impl->num_e_in = num_input_fields; 243 impl->num_e_out = num_output_fields; 244 245 // Set up infield and outfield e_vecs and q_vecs 246 // Infields 247 CeedCallBackend(CeedOperatorSetupFields_Sycl(qf, op, true, impl->e_vecs, impl->q_vecs_in, 0, num_input_fields, Q, num_elem)); 248 // Outfields 249 CeedCallBackend(CeedOperatorSetupFields_Sycl(qf, op, false, impl->e_vecs, impl->q_vecs_out, num_input_fields, num_output_fields, Q, num_elem)); 250 251 CeedCallBackend(CeedOperatorSetSetupDone(op)); 252 return CEED_ERROR_SUCCESS; 253 } 254 255 //------------------------------------------------------------------------------ 256 // Setup Operator Inputs 257 //------------------------------------------------------------------------------ 258 static inline int CeedOperatorSetupInputs_Sycl(CeedInt num_input_fields, CeedQFunctionField *qf_input_fields, CeedOperatorField *op_input_fields, 259 CeedVector in_vec, const bool skip_active, CeedScalar *e_data[2 * CEED_FIELD_MAX], 260 CeedOperator_Sycl *impl, CeedRequest *request) { 261 for (CeedInt i = 0; i < num_input_fields; i++) { 262 bool is_active; 263 CeedEvalMode eval_mode; 264 CeedVector vec; 265 266 // Get input vector 267 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[i], &vec)); 268 is_active = vec == CEED_VECTOR_ACTIVE; 269 if (is_active) { 270 if (skip_active) continue; 271 else vec = in_vec; 272 } 273 274 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode)); 275 if (eval_mode == CEED_EVAL_WEIGHT) { // Skip 276 } else { 277 // Restrict, if necessary 278 if (!impl->e_vecs[i]) { 279 // No restriction for this field; read data directly from vec. 280 CeedCallBackend(CeedVectorGetArrayRead(vec, CEED_MEM_DEVICE, (const CeedScalar **)&e_data[i])); 281 } else { 282 CeedElemRestriction elem_rstr; 283 284 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_input_fields[i], &elem_rstr)); 285 CeedCallBackend(CeedElemRestrictionApply(elem_rstr, CEED_NOTRANSPOSE, vec, impl->e_vecs[i], request)); 286 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 287 CeedCallBackend(CeedVectorGetArrayRead(impl->e_vecs[i], CEED_MEM_DEVICE, (const CeedScalar **)&e_data[i])); 288 } 289 } 290 if (!is_active) CeedCallBackend(CeedVectorDestroy(&vec)); 291 } 292 return CEED_ERROR_SUCCESS; 293 } 294 295 //------------------------------------------------------------------------------ 296 // Input Basis Action 297 //------------------------------------------------------------------------------ 298 static inline int CeedOperatorInputBasis_Sycl(CeedInt num_elem, CeedQFunctionField *qf_input_fields, CeedOperatorField *op_input_fields, 299 CeedInt num_input_fields, const bool skip_active, CeedScalar *e_data[2 * CEED_FIELD_MAX], 300 CeedOperator_Sycl *impl) { 301 for (CeedInt i = 0; i < num_input_fields; i++) { 302 CeedEvalMode eval_mode; 303 304 // Skip active input 305 if (skip_active) { 306 bool is_active; 307 CeedVector vec; 308 309 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[i], &vec)); 310 is_active = vec == CEED_VECTOR_ACTIVE; 311 CeedCallBackend(CeedVectorDestroy(&vec)); 312 if (is_active) continue; 313 } 314 // Basis action 315 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode)); 316 switch (eval_mode) { 317 case CEED_EVAL_NONE: 318 CeedCallBackend(CeedVectorSetArray(impl->q_vecs_in[i], CEED_MEM_DEVICE, CEED_USE_POINTER, e_data[i])); 319 break; 320 case CEED_EVAL_INTERP: 321 case CEED_EVAL_GRAD: { 322 CeedBasis basis; 323 324 CeedCallBackend(CeedOperatorFieldGetBasis(op_input_fields[i], &basis)); 325 CeedCallBackend(CeedBasisApply(basis, num_elem, CEED_NOTRANSPOSE, eval_mode, impl->e_vecs[i], impl->q_vecs_in[i])); 326 CeedCallBackend(CeedBasisDestroy(&basis)); 327 break; 328 } 329 case CEED_EVAL_WEIGHT: 330 break; // No action 331 case CEED_EVAL_DIV: 332 break; // TODO: Not implemented 333 case CEED_EVAL_CURL: 334 break; // TODO: Not implemented 335 } 336 } 337 return CEED_ERROR_SUCCESS; 338 } 339 340 //------------------------------------------------------------------------------ 341 // Restore Input Vectors 342 //------------------------------------------------------------------------------ 343 static inline int CeedOperatorRestoreInputs_Sycl(CeedInt num_input_fields, CeedQFunctionField *qf_input_fields, CeedOperatorField *op_input_fields, 344 const bool skip_active, CeedScalar *e_data[2 * CEED_FIELD_MAX], CeedOperator_Sycl *impl) { 345 for (CeedInt i = 0; i < num_input_fields; i++) { 346 bool is_active; 347 CeedEvalMode eval_mode; 348 CeedVector vec; 349 350 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[i], &vec)); 351 is_active = vec == CEED_VECTOR_ACTIVE; 352 // Skip active input 353 if (skip_active) { 354 if (is_active) continue; 355 } 356 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode)); 357 if (eval_mode == CEED_EVAL_WEIGHT) { // Skip 358 } else { 359 if (!impl->e_vecs[i]) { // This was a skip_restriction case 360 CeedCallBackend(CeedVectorRestoreArrayRead(vec, (const CeedScalar **)&e_data[i])); 361 } else { 362 CeedCallBackend(CeedVectorRestoreArrayRead(impl->e_vecs[i], (const CeedScalar **)&e_data[i])); 363 } 364 } 365 if (!is_active) CeedCallBackend(CeedVectorDestroy(&vec)); 366 } 367 return CEED_ERROR_SUCCESS; 368 } 369 370 //------------------------------------------------------------------------------ 371 // Apply and add to output 372 //------------------------------------------------------------------------------ 373 static int CeedOperatorApplyAdd_Sycl(CeedOperator op, CeedVector in_vec, CeedVector out_vec, CeedRequest *request) { 374 CeedInt Q, num_elem, elem_size, num_input_fields, num_output_fields, size; 375 CeedEvalMode eval_mode; 376 CeedScalar *e_data[2 * CEED_FIELD_MAX] = {0}; 377 CeedQFunctionField *qf_input_fields, *qf_output_fields; 378 CeedQFunction qf; 379 CeedOperatorField *op_input_fields, *op_output_fields; 380 CeedOperator_Sycl *impl; 381 382 CeedCallBackend(CeedOperatorGetData(op, &impl)); 383 CeedCallBackend(CeedOperatorGetQFunction(op, &qf)); 384 CeedCallBackend(CeedOperatorGetNumQuadraturePoints(op, &Q)); 385 CeedCallBackend(CeedOperatorGetNumElements(op, &num_elem)); 386 CeedCallBackend(CeedOperatorGetFields(op, &num_input_fields, &op_input_fields, &num_output_fields, &op_output_fields)); 387 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_input_fields, NULL, &qf_output_fields)); 388 389 // Setup 390 CeedCallBackend(CeedOperatorSetup_Sycl(op)); 391 392 // Input Evecs and Restriction 393 CeedCallBackend(CeedOperatorSetupInputs_Sycl(num_input_fields, qf_input_fields, op_input_fields, in_vec, false, e_data, impl, request)); 394 395 // Input basis apply if needed 396 CeedCallBackend(CeedOperatorInputBasis_Sycl(num_elem, qf_input_fields, op_input_fields, num_input_fields, false, e_data, impl)); 397 398 // Output pointers, as necessary 399 for (CeedInt i = 0; i < num_output_fields; i++) { 400 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 401 if (eval_mode == CEED_EVAL_NONE) { 402 // Set the output Q-Vector to use the E-Vector data directly 403 CeedCallBackend(CeedVectorGetArrayWrite(impl->e_vecs[i + impl->num_e_in], CEED_MEM_DEVICE, &e_data[i + num_input_fields])); 404 CeedCallBackend(CeedVectorSetArray(impl->q_vecs_out[i], CEED_MEM_DEVICE, CEED_USE_POINTER, e_data[i + num_input_fields])); 405 } 406 } 407 408 // Q function 409 CeedCallBackend(CeedQFunctionApply(qf, num_elem * Q, impl->q_vecs_in, impl->q_vecs_out)); 410 411 // Output basis apply if needed 412 for (CeedInt i = 0; i < num_output_fields; i++) { 413 CeedElemRestriction elem_rstr; 414 415 // Get elem_size, eval_mode, size 416 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_output_fields[i], &elem_rstr)); 417 CeedCallBackend(CeedElemRestrictionGetElementSize(elem_rstr, &elem_size)); 418 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 419 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 420 CeedCallBackend(CeedQFunctionFieldGetSize(qf_output_fields[i], &size)); 421 // Basis action 422 switch (eval_mode) { 423 case CEED_EVAL_NONE: 424 break; 425 case CEED_EVAL_INTERP: 426 case CEED_EVAL_GRAD: { 427 CeedBasis basis; 428 429 CeedCallBackend(CeedOperatorFieldGetBasis(op_output_fields[i], &basis)); 430 CeedCallBackend(CeedBasisApply(basis, num_elem, CEED_TRANSPOSE, eval_mode, impl->q_vecs_out[i], impl->e_vecs[i + impl->num_e_in])); 431 CeedCallBackend(CeedBasisDestroy(&basis)); 432 break; 433 } 434 // LCOV_EXCL_START 435 case CEED_EVAL_WEIGHT: { 436 Ceed ceed; 437 438 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 439 return CeedError(ceed, CEED_ERROR_BACKEND, "CEED_EVAL_WEIGHT cannot be an output evaluation mode"); 440 break; // Should not occur 441 } 442 case CEED_EVAL_DIV: 443 case CEED_EVAL_CURL: { 444 Ceed ceed; 445 446 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 447 return CeedError(ceed, CEED_ERROR_BACKEND, "%s not supported", CeedEvalModes[eval_mode]); 448 break; // Should not occur 449 } 450 // LCOV_EXCL_STOP 451 } 452 } 453 454 // Output restriction 455 for (CeedInt i = 0; i < num_output_fields; i++) { 456 bool is_active; 457 CeedEvalMode eval_mode; 458 CeedVector vec; 459 CeedElemRestriction elem_rstr; 460 461 // Restore evec 462 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode)); 463 if (eval_mode == CEED_EVAL_NONE) { 464 CeedCallBackend(CeedVectorRestoreArray(impl->e_vecs[i + impl->num_e_in], &e_data[i + num_input_fields])); 465 } 466 // Restrict 467 CeedCallBackend(CeedOperatorFieldGetVector(op_output_fields[i], &vec)); 468 is_active = vec == CEED_VECTOR_ACTIVE; 469 if (is_active) vec = out_vec; 470 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_output_fields[i], &elem_rstr)); 471 CeedCallBackend(CeedElemRestrictionApply(elem_rstr, CEED_TRANSPOSE, impl->e_vecs[i + impl->num_e_in], vec, request)); 472 if (!is_active) CeedCallBackend(CeedVectorDestroy(&vec)); 473 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 474 } 475 476 // Restore input arrays 477 CeedCallBackend(CeedOperatorRestoreInputs_Sycl(num_input_fields, qf_input_fields, op_input_fields, false, e_data, impl)); 478 return CEED_ERROR_SUCCESS; 479 } 480 481 //------------------------------------------------------------------------------ 482 // Core code for assembling linear QFunction 483 //------------------------------------------------------------------------------ 484 static inline int CeedOperatorLinearAssembleQFunctionCore_Sycl(CeedOperator op, bool build_objects, CeedVector *assembled, 485 CeedElemRestriction *elem_rstr, CeedRequest *request) { 486 Ceed ceed, ceed_parent; 487 CeedSize q_size; 488 CeedInt num_active_in, num_active_out, Q, num_elem, num_input_fields, num_output_fields, size; 489 CeedScalar *assembled_array, *e_data[2 * CEED_FIELD_MAX] = {NULL}; 490 CeedVector *active_in; 491 CeedQFunctionField *qf_input_fields, *qf_output_fields; 492 CeedQFunction qf; 493 CeedOperatorField *op_input_fields, *op_output_fields; 494 CeedOperator_Sycl *impl; 495 496 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 497 CeedCallBackend(CeedOperatorGetFallbackParentCeed(op, &ceed_parent)); 498 CeedCallBackend(CeedOperatorGetData(op, &impl)); 499 CeedCallBackend(CeedOperatorGetQFunction(op, &qf)); 500 CeedCallBackend(CeedOperatorGetNumQuadraturePoints(op, &Q)); 501 CeedCallBackend(CeedOperatorGetNumElements(op, &num_elem)); 502 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_input_fields, NULL, &qf_output_fields)); 503 CeedCallBackend(CeedOperatorGetFields(op, &num_input_fields, &op_input_fields, &num_output_fields, &op_output_fields)); 504 active_in = impl->qf_active_in; 505 num_active_in = impl->num_active_in, num_active_out = impl->num_active_out; 506 507 // Setup 508 CeedCallBackend(CeedOperatorSetup_Sycl(op)); 509 510 // Input Evecs and Restriction 511 CeedCallBackend(CeedOperatorSetupInputs_Sycl(num_input_fields, qf_input_fields, op_input_fields, NULL, true, e_data, impl, request)); 512 513 // Count number of active input fields 514 if (!num_active_in) { 515 for (CeedInt i = 0; i < num_input_fields; i++) { 516 CeedScalar *q_vec_array; 517 CeedVector vec; 518 519 // Check if active input 520 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[i], &vec)); 521 if (vec == CEED_VECTOR_ACTIVE) { 522 CeedCallBackend(CeedQFunctionFieldGetSize(qf_input_fields[i], &size)); 523 CeedCallBackend(CeedVectorSetValue(impl->q_vecs_in[i], 0.0)); 524 CeedCallBackend(CeedVectorGetArray(impl->q_vecs_in[i], CEED_MEM_DEVICE, &q_vec_array)); 525 CeedCallBackend(CeedRealloc(num_active_in + size, &active_in)); 526 for (CeedInt field = 0; field < size; field++) { 527 q_size = (CeedSize)Q * num_elem; 528 CeedCallBackend(CeedVectorCreate(ceed, q_size, &active_in[num_active_in + field])); 529 CeedCallBackend( 530 CeedVectorSetArray(active_in[num_active_in + field], CEED_MEM_DEVICE, CEED_USE_POINTER, &q_vec_array[field * Q * num_elem])); 531 } 532 num_active_in += size; 533 CeedCallBackend(CeedVectorRestoreArray(impl->q_vecs_in[i], &q_vec_array)); 534 } 535 CeedCallBackend(CeedVectorDestroy(&vec)); 536 } 537 impl->num_active_in = num_active_in; 538 impl->qf_active_in = active_in; 539 } 540 541 // Count number of active output fields 542 if (!num_active_out) { 543 for (CeedInt i = 0; i < num_output_fields; i++) { 544 CeedVector vec; 545 546 // Check if active output 547 CeedCallBackend(CeedOperatorFieldGetVector(op_output_fields[i], &vec)); 548 if (vec == CEED_VECTOR_ACTIVE) { 549 CeedCallBackend(CeedQFunctionFieldGetSize(qf_output_fields[i], &size)); 550 num_active_out += size; 551 } 552 CeedCallBackend(CeedVectorDestroy(&vec)); 553 } 554 impl->num_active_out = num_active_out; 555 } 556 557 // Check sizes 558 CeedCheck(num_active_in > 0 && num_active_out > 0, ceed, CEED_ERROR_BACKEND, "Cannot assemble QFunction without active inputs and outputs"); 559 560 // Build objects if needed 561 if (build_objects) { 562 CeedSize l_size = (CeedSize)num_elem * Q * num_active_in * num_active_out; 563 CeedInt strides[3] = {1, num_elem * Q, Q}; /* *NOPAD* */ 564 565 // Create output restriction 566 CeedCallBackend(CeedElemRestrictionCreateStrided(ceed_parent, num_elem, Q, num_active_in * num_active_out, l_size, strides, elem_rstr)); 567 // Create assembled vector 568 CeedCallBackend(CeedVectorCreate(ceed_parent, l_size, assembled)); 569 } 570 CeedCallBackend(CeedVectorSetValue(*assembled, 0.0)); 571 CeedCallBackend(CeedVectorGetArray(*assembled, CEED_MEM_DEVICE, &assembled_array)); 572 573 // Input basis apply 574 CeedCallBackend(CeedOperatorInputBasis_Sycl(num_elem, qf_input_fields, op_input_fields, num_input_fields, true, e_data, impl)); 575 576 // Assemble QFunction 577 for (CeedInt in = 0; in < num_active_in; in++) { 578 // Set Inputs 579 CeedCallBackend(CeedVectorSetValue(active_in[in], 1.0)); 580 if (num_active_in > 1) { 581 CeedCallBackend(CeedVectorSetValue(active_in[(in + num_active_in - 1) % num_active_in], 0.0)); 582 } 583 // Set Outputs 584 for (CeedInt out = 0; out < num_output_fields; out++) { 585 CeedVector vec; 586 587 // Check if active output 588 CeedCallBackend(CeedOperatorFieldGetVector(op_output_fields[out], &vec)); 589 if (vec == CEED_VECTOR_ACTIVE) { 590 CeedCallBackend(CeedVectorSetArray(impl->q_vecs_out[out], CEED_MEM_DEVICE, CEED_USE_POINTER, assembled_array)); 591 CeedCallBackend(CeedQFunctionFieldGetSize(qf_output_fields[out], &size)); 592 assembled_array += size * Q * num_elem; // Advance the pointer by the size of the output 593 } 594 CeedCallBackend(CeedVectorDestroy(&vec)); 595 } 596 // Apply QFunction 597 CeedCallBackend(CeedQFunctionApply(qf, Q * num_elem, impl->q_vecs_in, impl->q_vecs_out)); 598 } 599 600 // Un-set output Qvecs to prevent accidental overwrite of Assembled 601 for (CeedInt out = 0; out < num_output_fields; out++) { 602 CeedVector vec; 603 604 // Check if active output 605 CeedCallBackend(CeedOperatorFieldGetVector(op_output_fields[out], &vec)); 606 if (vec == CEED_VECTOR_ACTIVE) { 607 CeedCallBackend(CeedVectorTakeArray(impl->q_vecs_out[out], CEED_MEM_DEVICE, NULL)); 608 } 609 CeedCallBackend(CeedVectorDestroy(&vec)); 610 } 611 612 // Restore input arrays 613 CeedCallBackend(CeedOperatorRestoreInputs_Sycl(num_input_fields, qf_input_fields, op_input_fields, true, e_data, impl)); 614 615 // Restore output 616 CeedCallBackend(CeedVectorRestoreArray(*assembled, &assembled_array)); 617 return CEED_ERROR_SUCCESS; 618 } 619 620 //------------------------------------------------------------------------------ 621 // Assemble Linear QFunction 622 //------------------------------------------------------------------------------ 623 static int CeedOperatorLinearAssembleQFunction_Sycl(CeedOperator op, CeedVector *assembled, CeedElemRestriction *elem_rstr, CeedRequest *request) { 624 return CeedOperatorLinearAssembleQFunctionCore_Sycl(op, true, assembled, elem_rstr, request); 625 } 626 627 //------------------------------------------------------------------------------ 628 // Update Assembled Linear QFunction 629 //------------------------------------------------------------------------------ 630 static int CeedOperatorLinearAssembleQFunctionUpdate_Sycl(CeedOperator op, CeedVector assembled, CeedElemRestriction elem_rstr, 631 CeedRequest *request) { 632 return CeedOperatorLinearAssembleQFunctionCore_Sycl(op, false, &assembled, &elem_rstr, request); 633 } 634 635 //------------------------------------------------------------------------------ 636 // Assemble diagonal setup 637 //------------------------------------------------------------------------------ 638 static inline int CeedOperatorAssembleDiagonalSetup_Sycl(CeedOperator op) { 639 Ceed ceed; 640 Ceed_Sycl *sycl_data; 641 CeedInt num_input_fields, num_output_fields, num_eval_mode_in = 0, num_comp = 0, dim = 1, num_eval_mode_out = 0; 642 CeedEvalMode *eval_mode_in = NULL, *eval_mode_out = NULL; 643 CeedElemRestriction rstr_in = NULL, rstr_out = NULL; 644 CeedBasis basis_in = NULL, basis_out = NULL; 645 CeedQFunctionField *qf_fields; 646 CeedQFunction qf; 647 CeedOperatorField *op_fields; 648 CeedOperator_Sycl *impl; 649 650 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 651 CeedCallBackend(CeedOperatorGetQFunction(op, &qf)); 652 CeedCallBackend(CeedQFunctionGetNumArgs(qf, &num_input_fields, &num_output_fields)); 653 654 // Determine active input basis 655 CeedCallBackend(CeedOperatorGetFields(op, NULL, &op_fields, NULL, NULL)); 656 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_fields, NULL, NULL)); 657 for (CeedInt i = 0; i < num_input_fields; i++) { 658 CeedVector vec; 659 660 CeedCallBackend(CeedOperatorFieldGetVector(op_fields[i], &vec)); 661 if (vec == CEED_VECTOR_ACTIVE) { 662 CeedEvalMode eval_mode; 663 CeedElemRestriction elem_rstr; 664 CeedBasis basis; 665 666 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_fields[i], &elem_rstr)); 667 if (!rstr_in) CeedCallBackend(CeedElemRestrictionReferenceCopy(elem_rstr, &rstr_in)); 668 CeedCheck(rstr_in == elem_rstr, ceed, CEED_ERROR_BACKEND, "Backend does not implement multi-field non-composite operator diagonal assembly"); 669 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 670 CeedCallBackend(CeedOperatorFieldGetBasis(op_fields[i], &basis)); 671 if (!basis_in) CeedCallBackend(CeedBasisReferenceCopy(basis, &basis_in)); 672 CeedCheck(basis_in == basis, ceed, CEED_ERROR_BACKEND, "Backend does not implement operator diagonal assembly with multiple active bases"); 673 CeedCallBackend(CeedBasisDestroy(&basis)); 674 CeedCallBackend(CeedBasisGetDimension(basis_in, &dim)); 675 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode)); 676 switch (eval_mode) { 677 case CEED_EVAL_NONE: 678 case CEED_EVAL_INTERP: 679 CeedCallBackend(CeedRealloc(num_eval_mode_in + 1, &eval_mode_in)); 680 eval_mode_in[num_eval_mode_in] = eval_mode; 681 num_eval_mode_in += 1; 682 break; 683 case CEED_EVAL_GRAD: 684 CeedCallBackend(CeedRealloc(num_eval_mode_in + dim, &eval_mode_in)); 685 for (CeedInt d = 0; d < dim; d++) eval_mode_in[num_eval_mode_in + d] = eval_mode; 686 num_eval_mode_in += dim; 687 break; 688 case CEED_EVAL_WEIGHT: 689 case CEED_EVAL_DIV: 690 case CEED_EVAL_CURL: 691 break; // Caught by QF Assembly 692 } 693 } 694 CeedCallBackend(CeedVectorDestroy(&vec)); 695 } 696 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_in)); 697 698 // Determine active output basis 699 CeedCallBackend(CeedOperatorGetFields(op, NULL, NULL, NULL, &op_fields)); 700 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, NULL, NULL, &qf_fields)); 701 for (CeedInt i = 0; i < num_output_fields; i++) { 702 CeedVector vec; 703 704 CeedCallBackend(CeedOperatorFieldGetVector(op_fields[i], &vec)); 705 if (vec == CEED_VECTOR_ACTIVE) { 706 CeedEvalMode eval_mode; 707 CeedElemRestriction elem_rstr; 708 CeedBasis basis; 709 710 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_fields[i], &elem_rstr)); 711 if (!rstr_out) CeedCallBackend(CeedElemRestrictionReferenceCopy(elem_rstr, &rstr_out)); 712 CeedCheck(rstr_out == elem_rstr, ceed, CEED_ERROR_BACKEND, "Backend does not implement multi-field non-composite operator diagonal assembly"); 713 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 714 CeedCallBackend(CeedOperatorFieldGetBasis(op_fields[i], &basis)); 715 if (!basis_out) CeedCallBackend(CeedBasisReferenceCopy(basis, &basis_out)); 716 CeedCheck(basis_out == basis, ceed, CEED_ERROR_BACKEND, "Backend does not implement operator diagonal assembly with multiple active bases"); 717 CeedCallBackend(CeedBasisDestroy(&basis)); 718 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode)); 719 switch (eval_mode) { 720 case CEED_EVAL_NONE: 721 case CEED_EVAL_INTERP: 722 CeedCallBackend(CeedRealloc(num_eval_mode_out + 1, &eval_mode_out)); 723 eval_mode_out[num_eval_mode_out] = eval_mode; 724 num_eval_mode_out += 1; 725 break; 726 case CEED_EVAL_GRAD: 727 CeedCallBackend(CeedRealloc(num_eval_mode_out + dim, &eval_mode_out)); 728 for (CeedInt d = 0; d < dim; d++) eval_mode_out[num_eval_mode_out + d] = eval_mode; 729 num_eval_mode_out += dim; 730 break; 731 case CEED_EVAL_WEIGHT: 732 case CEED_EVAL_DIV: 733 case CEED_EVAL_CURL: 734 break; // Caught by QF Assembly 735 } 736 } 737 CeedCallBackend(CeedVectorDestroy(&vec)); 738 } 739 740 // Operator data struct 741 CeedCallBackend(CeedOperatorGetData(op, &impl)); 742 CeedCallBackend(CeedGetData(ceed, &sycl_data)); 743 CeedCallBackend(CeedCalloc(1, &impl->diag)); 744 CeedOperatorDiag_Sycl *diag = impl->diag; 745 746 CeedCallBackend(CeedBasisReferenceCopy(basis_in, &diag->basis_in)); 747 CeedCallBackend(CeedBasisReferenceCopy(basis_out, &diag->basis_out)); 748 diag->h_eval_mode_in = eval_mode_in; 749 diag->h_eval_mode_out = eval_mode_out; 750 diag->num_eval_mode_in = num_eval_mode_in; 751 diag->num_eval_mode_out = num_eval_mode_out; 752 753 // Kernel parameters 754 CeedInt num_nodes, num_qpts; 755 CeedCallBackend(CeedBasisGetNumNodes(basis_in, &num_nodes)); 756 CeedCallBackend(CeedBasisGetNumQuadraturePoints(basis_in, &num_qpts)); 757 CeedCallBackend(CeedBasisGetNumComponents(basis_in, &num_comp)); 758 diag->num_nodes = num_nodes; 759 diag->num_qpts = num_qpts; 760 diag->num_comp = num_comp; 761 762 // Basis matrices 763 const CeedInt i_len = num_qpts * num_nodes; 764 const CeedInt g_len = num_qpts * num_nodes * dim; 765 const CeedScalar *interp_in, *interp_out, *grad_in, *grad_out; 766 767 // CEED_EVAL_NONE 768 CeedScalar *identity = NULL; 769 bool has_eval_none = false; 770 for (CeedInt i = 0; i < num_eval_mode_in; i++) has_eval_none = has_eval_none || (eval_mode_in[i] == CEED_EVAL_NONE); 771 for (CeedInt i = 0; i < num_eval_mode_out; i++) has_eval_none = has_eval_none || (eval_mode_out[i] == CEED_EVAL_NONE); 772 773 std::vector<sycl::event> e; 774 775 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()}; 776 777 std::vector<sycl::event> copy_events; 778 779 if (has_eval_none) { 780 CeedCallBackend(CeedCalloc(num_qpts * num_nodes, &identity)); 781 for (CeedSize i = 0; i < (num_nodes < num_qpts ? num_nodes : num_qpts); i++) identity[i * num_nodes + i] = 1.0; 782 CeedCallSycl(ceed, diag->d_identity = sycl::malloc_device<CeedScalar>(i_len, sycl_data->sycl_device, sycl_data->sycl_context)); 783 sycl::event identity_copy = sycl_data->sycl_queue.copy<CeedScalar>(identity, diag->d_identity, i_len, e); 784 copy_events.push_back(identity_copy); 785 } 786 787 // CEED_EVAL_INTERP 788 CeedCallBackend(CeedBasisGetInterp(basis_in, &interp_in)); 789 CeedCallSycl(ceed, diag->d_interp_in = sycl::malloc_device<CeedScalar>(i_len, sycl_data->sycl_device, sycl_data->sycl_context)); 790 sycl::event interp_in_copy = sycl_data->sycl_queue.copy<CeedScalar>(interp_in, diag->d_interp_in, i_len, e); 791 copy_events.push_back(interp_in_copy); 792 793 CeedCallBackend(CeedBasisGetInterp(basis_out, &interp_out)); 794 CeedCallSycl(ceed, diag->d_interp_out = sycl::malloc_device<CeedScalar>(i_len, sycl_data->sycl_device, sycl_data->sycl_context)); 795 sycl::event interp_out_copy = sycl_data->sycl_queue.copy<CeedScalar>(interp_out, diag->d_interp_out, i_len, e); 796 copy_events.push_back(interp_out_copy); 797 798 // CEED_EVAL_GRAD 799 CeedCallBackend(CeedBasisGetGrad(basis_in, &grad_in)); 800 CeedCallSycl(ceed, diag->d_grad_in = sycl::malloc_device<CeedScalar>(g_len, sycl_data->sycl_device, sycl_data->sycl_context)); 801 sycl::event grad_in_copy = sycl_data->sycl_queue.copy<CeedScalar>(grad_in, diag->d_grad_in, g_len, e); 802 copy_events.push_back(grad_in_copy); 803 804 CeedCallBackend(CeedBasisGetGrad(basis_out, &grad_out)); 805 CeedCallSycl(ceed, diag->d_grad_out = sycl::malloc_device<CeedScalar>(g_len, sycl_data->sycl_device, sycl_data->sycl_context)); 806 sycl::event grad_out_copy = sycl_data->sycl_queue.copy<CeedScalar>(grad_out, diag->d_grad_out, g_len, e); 807 copy_events.push_back(grad_out_copy); 808 809 // Arrays of eval_modes 810 CeedCallSycl(ceed, diag->d_eval_mode_in = sycl::malloc_device<CeedEvalMode>(num_eval_mode_in, sycl_data->sycl_device, sycl_data->sycl_context)); 811 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); 812 copy_events.push_back(eval_mode_in_copy); 813 814 CeedCallSycl(ceed, diag->d_eval_mode_out = sycl::malloc_device<CeedEvalMode>(num_eval_mode_out, sycl_data->sycl_device, sycl_data->sycl_context)); 815 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); 816 copy_events.push_back(eval_mode_out_copy); 817 818 // Restriction 819 CeedCallBackend(CeedElemRestrictionReferenceCopy(rstr_out, &diag->diag_rstr)); 820 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_out)); 821 822 // Cleanup 823 CeedCallBackend(CeedBasisDestroy(&basis_in)); 824 CeedCallBackend(CeedBasisDestroy(&basis_out)); 825 826 // Wait for all copies to complete and handle exceptions 827 CeedCallSycl(ceed, sycl::event::wait_and_throw(copy_events)); 828 return CEED_ERROR_SUCCESS; 829 } 830 831 //------------------------------------------------------------------------------ 832 // Kernel for diagonal assembly 833 //------------------------------------------------------------------------------ 834 static int CeedOperatorLinearDiagonal_Sycl(sycl::queue &sycl_queue, const bool is_point_block, const CeedInt num_elem, 835 const CeedOperatorDiag_Sycl *diag, const CeedScalar *assembled_qf_array, CeedScalar *elem_diag_array) { 836 const CeedSize num_nodes = diag->num_nodes; 837 const CeedSize num_qpts = diag->num_qpts; 838 const CeedSize num_comp = diag->num_comp; 839 const CeedSize num_eval_mode_in = diag->num_eval_mode_in; 840 const CeedSize num_eval_mode_out = diag->num_eval_mode_out; 841 const CeedScalar *identity = diag->d_identity; 842 const CeedScalar *interp_in = diag->d_interp_in; 843 const CeedScalar *grad_in = diag->d_grad_in; 844 const CeedScalar *interp_out = diag->d_interp_out; 845 const CeedScalar *grad_out = diag->d_grad_out; 846 const CeedEvalMode *eval_mode_in = diag->d_eval_mode_in; 847 const CeedEvalMode *eval_mode_out = diag->d_eval_mode_out; 848 849 sycl::range<1> kernel_range(num_elem * num_nodes); 850 851 std::vector<sycl::event> e; 852 853 if (!sycl_queue.is_in_order()) e = {sycl_queue.ext_oneapi_submit_barrier()}; 854 855 sycl_queue.parallel_for<CeedOperatorSyclLinearDiagonal>(kernel_range, e, [=](sycl::id<1> idx) { 856 const CeedInt tid = idx % num_nodes; 857 const CeedInt e = idx / num_nodes; 858 859 // Compute the diagonal of B^T D B 860 // Each element 861 CeedInt d_out = -1; 862 // Each basis eval mode pair 863 for (CeedSize e_out = 0; e_out < num_eval_mode_out; e_out++) { 864 const CeedScalar *bt = NULL; 865 866 if (eval_mode_out[e_out] == CEED_EVAL_GRAD) ++d_out; 867 CeedOperatorGetBasisPointer_Sycl(&bt, eval_mode_out[e_out], identity, interp_out, &grad_out[d_out * num_qpts * num_nodes]); 868 CeedInt d_in = -1; 869 870 for (CeedSize e_in = 0; e_in < num_eval_mode_in; e_in++) { 871 const CeedScalar *b = NULL; 872 873 if (eval_mode_in[e_in] == CEED_EVAL_GRAD) ++d_in; 874 CeedOperatorGetBasisPointer_Sycl(&b, eval_mode_in[e_in], identity, interp_in, &grad_in[d_in * num_qpts * num_nodes]); 875 // Each component 876 for (CeedSize comp_out = 0; comp_out < num_comp; comp_out++) { 877 // Each qpoint/node pair 878 if (is_point_block) { 879 // Point Block Diagonal 880 for (CeedInt comp_in = 0; comp_in < num_comp; comp_in++) { 881 CeedScalar e_value = 0.0; 882 883 for (CeedSize q = 0; q < num_qpts; q++) { 884 const CeedScalar qf_value = 885 assembled_qf_array[((((e_in * num_comp + comp_in) * num_eval_mode_out + e_out) * num_comp + comp_out) * num_elem + e) * num_qpts + 886 q]; 887 888 e_value += bt[q * num_nodes + tid] * qf_value * b[q * num_nodes + tid]; 889 } 890 elem_diag_array[((comp_out * num_comp + comp_in) * num_elem + e) * num_nodes + tid] += e_value; 891 } 892 } else { 893 // Diagonal Only 894 CeedScalar e_value = 0.0; 895 896 for (CeedSize q = 0; q < num_qpts; q++) { 897 const CeedScalar qf_value = 898 assembled_qf_array[((((e_in * num_comp + comp_out) * num_eval_mode_out + e_out) * num_comp + comp_out) * num_elem + e) * num_qpts + 899 q]; 900 e_value += bt[q * num_nodes + tid] * qf_value * b[q * num_nodes + tid]; 901 } 902 elem_diag_array[(comp_out * num_elem + e) * num_nodes + tid] += e_value; 903 } 904 } 905 } 906 } 907 }); 908 return CEED_ERROR_SUCCESS; 909 } 910 911 //------------------------------------------------------------------------------ 912 // Assemble diagonal common code 913 //------------------------------------------------------------------------------ 914 static inline int CeedOperatorAssembleDiagonalCore_Sycl(CeedOperator op, CeedVector assembled, CeedRequest *request, const bool is_point_block) { 915 Ceed ceed; 916 Ceed_Sycl *sycl_data; 917 CeedInt num_elem; 918 CeedScalar *elem_diag_array; 919 const CeedScalar *assembled_qf_array; 920 CeedVector assembled_qf = NULL; 921 CeedOperator_Sycl *impl; 922 923 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 924 CeedCallBackend(CeedOperatorGetData(op, &impl)); 925 CeedCallBackend(CeedGetData(ceed, &sycl_data)); 926 927 // Assemble QFunction 928 { 929 CeedElemRestriction elem_rstr = NULL; 930 931 CeedCallBackend(CeedOperatorLinearAssembleQFunctionBuildOrUpdate(op, &assembled_qf, &elem_rstr, request)); 932 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 933 } 934 935 // Setup 936 if (!impl->diag) { 937 CeedCallBackend(CeedOperatorAssembleDiagonalSetup_Sycl(op)); 938 } 939 CeedOperatorDiag_Sycl *diag = impl->diag; 940 941 assert(diag != NULL); 942 943 // Restriction 944 if (is_point_block && !diag->point_block_diag_rstr) { 945 CeedCallBackend(CeedOperatorCreateActivePointBlockRestriction(diag->diag_rstr, &diag->point_block_diag_rstr)); 946 } 947 CeedElemRestriction diag_rstr = is_point_block ? diag->point_block_diag_rstr : diag->diag_rstr; 948 949 // Create diagonal vector 950 CeedVector elem_diag = is_point_block ? diag->point_block_elem_diag : diag->elem_diag; 951 952 if (!elem_diag) { 953 CeedCallBackend(CeedElemRestrictionCreateVector(diag_rstr, NULL, &elem_diag)); 954 if (is_point_block) diag->point_block_elem_diag = elem_diag; 955 else diag->elem_diag = elem_diag; 956 } 957 CeedCallBackend(CeedVectorSetValue(elem_diag, 0.0)); 958 959 // Assemble element operator diagonals 960 CeedCallBackend(CeedVectorGetArray(elem_diag, CEED_MEM_DEVICE, &elem_diag_array)); 961 CeedCallBackend(CeedVectorGetArrayRead(assembled_qf, CEED_MEM_DEVICE, &assembled_qf_array)); 962 CeedCallBackend(CeedElemRestrictionGetNumElements(diag_rstr, &num_elem)); 963 964 // Compute the diagonal of B^T D B 965 CeedCallBackend(CeedOperatorLinearDiagonal_Sycl(sycl_data->sycl_queue, is_point_block, num_elem, diag, assembled_qf_array, elem_diag_array)); 966 967 // Wait for queue to complete and handle exceptions 968 sycl_data->sycl_queue.wait_and_throw(); 969 970 // Restore arrays 971 CeedCallBackend(CeedVectorRestoreArray(elem_diag, &elem_diag_array)); 972 CeedCallBackend(CeedVectorRestoreArrayRead(assembled_qf, &assembled_qf_array)); 973 974 // Assemble local operator diagonal 975 CeedCallBackend(CeedElemRestrictionApply(diag_rstr, CEED_TRANSPOSE, elem_diag, assembled, request)); 976 977 // Cleanup 978 CeedCallBackend(CeedVectorDestroy(&assembled_qf)); 979 return CEED_ERROR_SUCCESS; 980 } 981 982 //------------------------------------------------------------------------------ 983 // Assemble Linear Diagonal 984 //------------------------------------------------------------------------------ 985 static int CeedOperatorLinearAssembleAddDiagonal_Sycl(CeedOperator op, CeedVector assembled, CeedRequest *request) { 986 CeedCallBackend(CeedOperatorAssembleDiagonalCore_Sycl(op, assembled, request, false)); 987 return CEED_ERROR_SUCCESS; 988 } 989 990 //------------------------------------------------------------------------------ 991 // Assemble Linear Point Block Diagonal 992 //------------------------------------------------------------------------------ 993 static int CeedOperatorLinearAssembleAddPointBlockDiagonal_Sycl(CeedOperator op, CeedVector assembled, CeedRequest *request) { 994 CeedCallBackend(CeedOperatorAssembleDiagonalCore_Sycl(op, assembled, request, true)); 995 return CEED_ERROR_SUCCESS; 996 } 997 998 //------------------------------------------------------------------------------ 999 // Single operator assembly setup 1000 //------------------------------------------------------------------------------ 1001 static int CeedSingleOperatorAssembleSetup_Sycl(CeedOperator op) { 1002 Ceed ceed; 1003 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, 1004 num_B_out_mats_to_load = 0, size_B_out = 0, num_qpts = 0, elem_size = 0, num_elem, num_comp, 1005 mat_start = 0; 1006 CeedEvalMode *eval_mode_in = NULL, *eval_mode_out = NULL; 1007 const CeedScalar *interp_in, *grad_in; 1008 CeedElemRestriction rstr_in = NULL, rstr_out = NULL; 1009 CeedBasis basis_in = NULL, basis_out = NULL; 1010 CeedQFunctionField *qf_fields; 1011 CeedQFunction qf; 1012 CeedOperatorField *input_fields, *output_fields; 1013 CeedOperator_Sycl *impl; 1014 1015 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 1016 CeedCallBackend(CeedOperatorGetData(op, &impl)); 1017 1018 // Get input and output fields 1019 CeedCallBackend(CeedOperatorGetFields(op, &num_input_fields, &input_fields, &num_output_fields, &output_fields)); 1020 1021 // Determine active input basis eval mode 1022 CeedCallBackend(CeedOperatorGetQFunction(op, &qf)); 1023 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_fields, NULL, NULL)); 1024 // Note that the kernel will treat each dimension of a gradient action separately; 1025 // i.e., when an active input has a CEED_EVAL_GRAD mode, num_ eval_mode_in will increment by dim. 1026 // 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, 1027 // so num_B_in_mats_to_load will be incremented by 1. 1028 for (CeedInt i = 0; i < num_input_fields; i++) { 1029 CeedVector vec; 1030 1031 CeedCallBackend(CeedOperatorFieldGetVector(input_fields[i], &vec)); 1032 if (vec == CEED_VECTOR_ACTIVE) { 1033 CeedEvalMode eval_mode; 1034 CeedElemRestriction elem_rstr; 1035 CeedBasis basis; 1036 1037 CeedCallBackend(CeedOperatorFieldGetElemRestriction(input_fields[i], &elem_rstr)); 1038 if (!rstr_in) CeedCallBackend(CeedElemRestrictionReferenceCopy(elem_rstr, &rstr_in)); 1039 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 1040 CeedCallBackend(CeedElemRestrictionGetElementSize(rstr_in, &elem_size)); 1041 CeedCallBackend(CeedOperatorFieldGetBasis(input_fields[i], &basis)); 1042 if (!basis_in) CeedCallBackend(CeedBasisReferenceCopy(basis, &basis_in)); 1043 CeedCheck(basis_in == basis, ceed, CEED_ERROR_BACKEND, "Backend does not implement operator assembly with multiple active bases"); 1044 CeedCallBackend(CeedBasisDestroy(&basis)); 1045 CeedCallBackend(CeedBasisGetDimension(basis_in, &dim)); 1046 CeedCallBackend(CeedBasisGetNumQuadraturePoints(basis_in, &num_qpts)); 1047 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode)); 1048 if (eval_mode != CEED_EVAL_NONE) { 1049 CeedCallBackend(CeedRealloc(num_B_in_mats_to_load + 1, &eval_mode_in)); 1050 eval_mode_in[num_B_in_mats_to_load] = eval_mode; 1051 num_B_in_mats_to_load += 1; 1052 if (eval_mode == CEED_EVAL_GRAD) { 1053 num_eval_mode_in += dim; 1054 size_B_in += dim * elem_size * num_qpts; 1055 } else { 1056 num_eval_mode_in += 1; 1057 size_B_in += elem_size * num_qpts; 1058 } 1059 } 1060 } 1061 CeedCallBackend(CeedVectorDestroy(&vec)); 1062 } 1063 1064 // Determine active output basis; basis_out and rstr_out only used if same as input, TODO 1065 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, NULL, NULL, &qf_fields)); 1066 for (CeedInt i = 0; i < num_output_fields; i++) { 1067 CeedVector vec; 1068 1069 CeedCallBackend(CeedOperatorFieldGetVector(output_fields[i], &vec)); 1070 if (vec == CEED_VECTOR_ACTIVE) { 1071 CeedEvalMode eval_mode; 1072 CeedElemRestriction elem_rstr; 1073 CeedBasis basis; 1074 1075 CeedCallBackend(CeedOperatorFieldGetElemRestriction(output_fields[i], &elem_rstr)); 1076 if (!rstr_out) CeedCallBackend(CeedElemRestrictionReferenceCopy(elem_rstr, &rstr_out)); 1077 CeedCheck(rstr_out == rstr_in, ceed, CEED_ERROR_BACKEND, "Backend does not implement multi-field non-composite operator assembly"); 1078 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr)); 1079 CeedCallBackend(CeedOperatorFieldGetBasis(output_fields[i], &basis)); 1080 if (!basis_out) CeedCallBackend(CeedBasisReferenceCopy(basis, &basis_out)); 1081 CeedCheck(basis_out == basis, ceed, CEED_ERROR_BACKEND, "Backend does not implement operator assembly with multiple active bases"); 1082 CeedCallBackend(CeedBasisDestroy(&basis)); 1083 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode)); 1084 if (eval_mode != CEED_EVAL_NONE) { 1085 CeedCallBackend(CeedRealloc(num_B_out_mats_to_load + 1, &eval_mode_out)); 1086 eval_mode_out[num_B_out_mats_to_load] = eval_mode; 1087 num_B_out_mats_to_load += 1; 1088 if (eval_mode == CEED_EVAL_GRAD) { 1089 num_eval_mode_out += dim; 1090 size_B_out += dim * elem_size * num_qpts; 1091 } else { 1092 num_eval_mode_out += 1; 1093 size_B_out += elem_size * num_qpts; 1094 } 1095 } 1096 } 1097 CeedCallBackend(CeedVectorDestroy(&vec)); 1098 } 1099 CeedCheck(num_eval_mode_in > 0 && num_eval_mode_out > 0, ceed, CEED_ERROR_UNSUPPORTED, "Cannot assemble operator without inputs/outputs"); 1100 1101 CeedCallBackend(CeedElemRestrictionGetNumElements(rstr_in, &num_elem)); 1102 CeedCallBackend(CeedElemRestrictionGetNumComponents(rstr_in, &num_comp)); 1103 1104 CeedCallBackend(CeedCalloc(1, &impl->asmb)); 1105 CeedOperatorAssemble_Sycl *asmb = impl->asmb; 1106 asmb->num_elem = num_elem; 1107 1108 Ceed_Sycl *sycl_data; 1109 CeedCallBackend(CeedGetData(ceed, &sycl_data)); 1110 1111 // Kernel setup 1112 int elems_per_block = 1; 1113 asmb->elems_per_block = elems_per_block; 1114 asmb->block_size_x = elem_size; 1115 asmb->block_size_y = elem_size; 1116 asmb->num_eval_mode_in = num_eval_mode_in; 1117 asmb->num_eval_mode_out = num_eval_mode_out; 1118 asmb->num_qpts = num_qpts; 1119 asmb->num_nodes = elem_size; 1120 asmb->block_size = elem_size * elem_size * elems_per_block; 1121 asmb->num_comp = num_comp; 1122 1123 // Build 'full' B matrices (not 1D arrays used for tensor-product matrices 1124 CeedCallBackend(CeedBasisGetInterp(basis_in, &interp_in)); 1125 CeedCallBackend(CeedBasisGetGrad(basis_in, &grad_in)); 1126 1127 // Load into B_in, in order that they will be used in eval_mode 1128 CeedCallSycl(ceed, asmb->d_B_in = sycl::malloc_device<CeedScalar>(size_B_in, sycl_data->sycl_device, sycl_data->sycl_context)); 1129 for (int i = 0; i < num_B_in_mats_to_load; i++) { 1130 CeedEvalMode eval_mode = eval_mode_in[i]; 1131 1132 if (eval_mode == CEED_EVAL_INTERP) { 1133 std::vector<sycl::event> e; 1134 1135 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()}; 1136 sycl_data->sycl_queue.copy<CeedScalar>(interp_in, &asmb->d_B_in[mat_start], elem_size * num_qpts, e); 1137 mat_start += elem_size * num_qpts; 1138 } else if (eval_mode == CEED_EVAL_GRAD) { 1139 std::vector<sycl::event> e; 1140 1141 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()}; 1142 sycl_data->sycl_queue.copy<CeedScalar>(grad_in, &asmb->d_B_in[mat_start], dim * elem_size * num_qpts, e); 1143 mat_start += dim * elem_size * num_qpts; 1144 } 1145 } 1146 1147 const CeedScalar *interp_out, *grad_out; 1148 // Note that this function currently assumes 1 basis, so this should always be true 1149 // for now 1150 if (basis_out == basis_in) { 1151 interp_out = interp_in; 1152 grad_out = grad_in; 1153 } else { 1154 CeedCallBackend(CeedBasisGetInterp(basis_out, &interp_out)); 1155 CeedCallBackend(CeedBasisGetGrad(basis_out, &grad_out)); 1156 } 1157 1158 // Load into B_out, in order that they will be used in eval_mode 1159 mat_start = 0; 1160 CeedCallSycl(ceed, asmb->d_B_out = sycl::malloc_device<CeedScalar>(size_B_out, sycl_data->sycl_device, sycl_data->sycl_context)); 1161 for (int i = 0; i < num_B_out_mats_to_load; i++) { 1162 CeedEvalMode eval_mode = eval_mode_out[i]; 1163 1164 if (eval_mode == CEED_EVAL_INTERP) { 1165 std::vector<sycl::event> e; 1166 1167 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()}; 1168 sycl_data->sycl_queue.copy<CeedScalar>(interp_out, &asmb->d_B_out[mat_start], elem_size * num_qpts, e); 1169 mat_start += elem_size * num_qpts; 1170 } else if (eval_mode == CEED_EVAL_GRAD) { 1171 std::vector<sycl::event> e; 1172 1173 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()}; 1174 sycl_data->sycl_queue.copy<CeedScalar>(grad_out, &asmb->d_B_out[mat_start], dim * elem_size * num_qpts, e); 1175 mat_start += dim * elem_size * num_qpts; 1176 } 1177 } 1178 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_in)); 1179 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_out)); 1180 CeedCallBackend(CeedBasisDestroy(&basis_in)); 1181 CeedCallBackend(CeedBasisDestroy(&basis_out)); 1182 return CEED_ERROR_SUCCESS; 1183 } 1184 1185 //------------------------------------------------------------------------------ 1186 // Matrix assembly kernel for low-order elements (3D thread block) 1187 //------------------------------------------------------------------------------ 1188 static int CeedOperatorLinearAssemble_Sycl(sycl::queue &sycl_queue, const CeedOperator_Sycl *impl, const CeedScalar *qf_array, 1189 CeedScalar *values_array) { 1190 // This kernels assumes B_in and B_out have the same number of quadrature points and basis points. 1191 // TODO: expand to more general cases 1192 CeedOperatorAssemble_Sycl *asmb = impl->asmb; 1193 const CeedInt num_elem = asmb->num_elem; 1194 const CeedSize num_nodes = asmb->num_nodes; 1195 const CeedSize num_comp = asmb->num_comp; 1196 const CeedSize num_qpts = asmb->num_qpts; 1197 const CeedSize num_eval_mode_in = asmb->num_eval_mode_in; 1198 const CeedSize num_eval_mode_out = asmb->num_eval_mode_out; 1199 1200 // Strides for final output ordering, determined by the reference (inference) implementation of the symbolic assembly, slowest --> fastest: element, 1201 // comp_in, comp_out, node_row, node_col 1202 const CeedSize comp_out_stride = num_nodes * num_nodes; 1203 const CeedSize comp_in_stride = comp_out_stride * num_comp; 1204 const CeedSize e_stride = comp_in_stride * num_comp; 1205 // Strides for QF array, slowest --> fastest: eval_mode_in, comp_in, eval_mode_out, comp_out, elem, qpt 1206 const CeedSize q_e_stride = num_qpts; 1207 const CeedSize q_comp_out_stride = num_elem * q_e_stride; 1208 const CeedSize q_eval_mode_out_stride = q_comp_out_stride * num_comp; 1209 const CeedSize q_comp_in_stride = q_eval_mode_out_stride * num_eval_mode_out; 1210 const CeedSize q_eval_mode_in_stride = q_comp_in_stride * num_comp; 1211 1212 CeedScalar *B_in, *B_out; 1213 B_in = asmb->d_B_in; 1214 B_out = asmb->d_B_out; 1215 const CeedInt block_size_x = asmb->block_size_x; 1216 const CeedInt block_size_y = asmb->block_size_y; 1217 1218 sycl::range<3> kernel_range(num_elem, block_size_y, block_size_x); 1219 1220 std::vector<sycl::event> e; 1221 1222 if (!sycl_queue.is_in_order()) e = {sycl_queue.ext_oneapi_submit_barrier()}; 1223 sycl_queue.parallel_for<CeedOperatorSyclLinearAssemble>(kernel_range, e, [=](sycl::id<3> idx) { 1224 const int e = idx.get(0); // Element index 1225 const int l = idx.get(1); // The output column index of each B^TDB operation 1226 const int i = idx.get(2); // The output row index of each B^TDB operation 1227 // such that we have (Bout^T)_ij D_jk Bin_kl = C_il 1228 for (CeedSize comp_in = 0; comp_in < num_comp; comp_in++) { 1229 for (CeedSize comp_out = 0; comp_out < num_comp; comp_out++) { 1230 CeedScalar result = 0.0; 1231 CeedSize qf_index_comp = q_comp_in_stride * comp_in + q_comp_out_stride * comp_out + q_e_stride * e; 1232 1233 for (CeedSize eval_mode_in = 0; eval_mode_in < num_eval_mode_in; eval_mode_in++) { 1234 CeedSize b_in_index = eval_mode_in * num_qpts * num_nodes; 1235 1236 for (CeedSize eval_mode_out = 0; eval_mode_out < num_eval_mode_out; eval_mode_out++) { 1237 CeedSize b_out_index = eval_mode_out * num_qpts * num_nodes; 1238 CeedSize qf_index = qf_index_comp + q_eval_mode_out_stride * eval_mode_out + q_eval_mode_in_stride * eval_mode_in; 1239 1240 // Perform the B^T D B operation for this 'chunk' of D (the qf_array) 1241 for (CeedSize j = 0; j < num_qpts; j++) { 1242 result += B_out[b_out_index + j * num_nodes + i] * qf_array[qf_index + j] * B_in[b_in_index + j * num_nodes + l]; 1243 } 1244 } // end of eval_mode_out 1245 } // end of eval_mode_in 1246 CeedSize val_index = comp_in_stride * comp_in + comp_out_stride * comp_out + e_stride * e + num_nodes * i + l; 1247 1248 values_array[val_index] = result; 1249 } // end of out component 1250 } // end of in component 1251 }); 1252 return CEED_ERROR_SUCCESS; 1253 } 1254 1255 //------------------------------------------------------------------------------ 1256 // Fallback kernel for larger orders (1D thread block) 1257 //------------------------------------------------------------------------------ 1258 /* 1259 static int CeedOperatorLinearAssembleFallback_Sycl(sycl::queue &sycl_queue, const CeedOperator_Sycl *impl, const CeedScalar *qf_array, 1260 CeedScalar *values_array) { 1261 // This kernel assumes B_in and B_out have the same number of quadrature points and basis points. 1262 // TODO: expand to more general cases 1263 CeedOperatorAssemble_Sycl *asmb = impl->asmb; 1264 const CeedInt num_elem = asmb->num_elem; 1265 const CeedInt num_nodes = asmb->num_nodes; 1266 const CeedInt num_comp = asmb->num_comp; 1267 const CeedInt num_qpts = asmb->num_qpts; 1268 const CeedInt num_eval_mode_in = asmb->num_eval_mode_in; 1269 const CeedInt num_eval_mode_out = asmb->num_eval_mode_out; 1270 1271 // Strides for final output ordering, determined by the reference (interface) implementation of the symbolic assembly, slowest --> fastest: elememt, 1272 // comp_in, comp_out, node_row, node_col 1273 const CeedInt comp_out_stride = num_nodes * num_nodes; 1274 const CeedInt comp_in_stride = comp_out_stride * num_comp; 1275 const CeedInt e_stride = comp_in_stride * num_comp; 1276 // Strides for QF array, slowest --> fastest: eval_mode_in, comp_in, eval_mode_out, comp_out, elem, qpt 1277 const CeedInt q_e_stride = num_qpts; 1278 const CeedInt q_comp_out_stride = num_elem * q_e_stride; 1279 const CeedInt q_eval_mode_out_stride = q_comp_out_stride * num_comp; 1280 const CeedInt q_comp_in_stride = q_eval_mode_out_stride * num_eval_mode_out; 1281 const CeedInt q_eval_mode_in_stride = q_comp_in_stride * num_comp; 1282 1283 CeedScalar *B_in, *B_out; 1284 B_in = asmb->d_B_in; 1285 B_out = asmb->d_B_out; 1286 const CeedInt elems_per_block = asmb->elems_per_block; 1287 const CeedInt block_size_x = asmb->block_size_x; 1288 const CeedInt block_size_y = asmb->block_size_y; // This will be 1 for the fallback kernel 1289 1290 const CeedInt grid = num_elem / elems_per_block + ((num_elem / elems_per_block * elems_per_block < num_elem) ? 1 : 0); 1291 sycl::range<3> local_range(block_size_x, block_size_y, elems_per_block); 1292 sycl::range<3> global_range(grid * block_size_x, block_size_y, elems_per_block); 1293 sycl::nd_range<3> kernel_range(global_range, local_range); 1294 1295 sycl_queue.parallel_for<CeedOperatorSyclLinearAssembleFallback>(kernel_range, [=](sycl::nd_item<3> work_item) { 1296 const CeedInt blockIdx = work_item.get_group(0); 1297 const CeedInt gridDimx = work_item.get_group_range(0); 1298 const CeedInt threadIdx = work_item.get_local_id(0); 1299 const CeedInt threadIdz = work_item.get_local_id(2); 1300 const CeedInt blockDimz = work_item.get_local_range(2); 1301 1302 const int l = threadIdx; // The output column index of each B^TDB operation 1303 // such that we have (Bout^T)_ij D_jk Bin_kl = C_il 1304 for (CeedInt e = blockIdx * blockDimz + threadIdz; e < num_elem; e += gridDimx * blockDimz) { 1305 for (CeedInt comp_in = 0; comp_in < num_comp; comp_in++) { 1306 for (CeedInt comp_out = 0; comp_out < num_comp; comp_out++) { 1307 for (CeedInt i = 0; i < num_nodes; i++) { 1308 CeedScalar result = 0.0; 1309 CeedInt qf_index_comp = q_comp_in_stride * comp_in + q_comp_out_stride * comp_out + q_e_stride * e; 1310 for (CeedInt eval_mode_in = 0; eval_mode_in < num_eval_mode_in; eval_mode_in++) { 1311 CeedInt b_in_index = eval_mode_in * num_qpts * num_nodes; 1312 for (CeedInt eval_mode_out = 0; eval_mode_out < num_eval_mode_out; eval_mode_out++) { 1313 CeedInt b_out_index = eval_mode_out * num_qpts * num_nodes; 1314 CeedInt qf_index = qf_index_comp + q_eval_mode_out_stride * eval_mode_out + q_eval_mode_in_stride * eval_mode_in; 1315 // Perform the B^T D B operation for this 'chunk' of D (the qf_array) 1316 for (CeedInt j = 0; j < num_qpts; j++) { 1317 result += B_out[b_out_index + j * num_nodes + i] * qf_array[qf_index + j] * B_in[b_in_index + j * num_nodes + l]; 1318 } 1319 } // end of eval_mode_out 1320 } // end of eval_mode_in 1321 CeedInt val_index = comp_in_stride * comp_in + comp_out_stride * comp_out + e_stride * e + num_nodes * i + l; 1322 values_array[val_index] = result; 1323 } // end of loop over element node index, i 1324 } // end of out component 1325 } // end of in component 1326 } // end of element loop 1327 }); 1328 return CEED_ERROR_SUCCESS; 1329 }*/ 1330 1331 //------------------------------------------------------------------------------ 1332 // Assemble matrix data for COO matrix of assembled operator. 1333 // The sparsity pattern is set by CeedOperatorLinearAssembleSymbolic. 1334 // 1335 // Note that this (and other assembly routines) currently assume only one active 1336 // input restriction/basis per operator (could have multiple basis eval modes). 1337 // TODO: allow multiple active input restrictions/basis objects 1338 //------------------------------------------------------------------------------ 1339 static int CeedSingleOperatorAssemble_Sycl(CeedOperator op, CeedInt offset, CeedVector values) { 1340 Ceed ceed; 1341 Ceed_Sycl *sycl_data; 1342 CeedScalar *values_array; 1343 const CeedScalar *qf_array; 1344 CeedVector assembled_qf = NULL; 1345 CeedElemRestriction rstr_q = NULL; 1346 CeedOperator_Sycl *impl; 1347 1348 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 1349 CeedCallBackend(CeedOperatorGetData(op, &impl)); 1350 CeedCallBackend(CeedGetData(ceed, &sycl_data)); 1351 1352 // Setup 1353 if (!impl->asmb) { 1354 CeedCallBackend(CeedSingleOperatorAssembleSetup_Sycl(op)); 1355 assert(impl->asmb != NULL); 1356 } 1357 1358 // Assemble QFunction 1359 CeedCallBackend(CeedOperatorLinearAssembleQFunctionBuildOrUpdate(op, &assembled_qf, &rstr_q, CEED_REQUEST_IMMEDIATE)); 1360 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_q)); 1361 CeedCallBackend(CeedVectorGetArrayWrite(values, CEED_MEM_DEVICE, &values_array)); 1362 values_array += offset; 1363 CeedCallBackend(CeedVectorGetArrayRead(assembled_qf, CEED_MEM_DEVICE, &qf_array)); 1364 1365 // Compute B^T D B 1366 CeedCallBackend(CeedOperatorLinearAssemble_Sycl(sycl_data->sycl_queue, impl, qf_array, values_array)); 1367 1368 // Wait for kernels to be completed 1369 // Kris: Review if this is necessary -- enqueing an async barrier may be sufficient 1370 sycl_data->sycl_queue.wait_and_throw(); 1371 1372 // Restore arrays 1373 CeedCallBackend(CeedVectorRestoreArray(values, &values_array)); 1374 CeedCallBackend(CeedVectorRestoreArrayRead(assembled_qf, &qf_array)); 1375 1376 // Cleanup 1377 CeedCallBackend(CeedVectorDestroy(&assembled_qf)); 1378 return CEED_ERROR_SUCCESS; 1379 } 1380 1381 //------------------------------------------------------------------------------ 1382 // Create operator 1383 //------------------------------------------------------------------------------ 1384 int CeedOperatorCreate_Sycl(CeedOperator op) { 1385 Ceed ceed; 1386 CeedOperator_Sycl *impl; 1387 1388 CeedCallBackend(CeedOperatorGetCeed(op, &ceed)); 1389 1390 CeedCallBackend(CeedCalloc(1, &impl)); 1391 CeedCallBackend(CeedOperatorSetData(op, impl)); 1392 1393 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleQFunction", CeedOperatorLinearAssembleQFunction_Sycl)); 1394 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleQFunctionUpdate", CeedOperatorLinearAssembleQFunctionUpdate_Sycl)); 1395 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleAddDiagonal", CeedOperatorLinearAssembleAddDiagonal_Sycl)); 1396 CeedCallBackend( 1397 CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleAddPointBlockDiagonal", CeedOperatorLinearAssembleAddPointBlockDiagonal_Sycl)); 1398 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleSingle", CeedSingleOperatorAssemble_Sycl)); 1399 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "ApplyAdd", CeedOperatorApplyAdd_Sycl)); 1400 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "Destroy", CeedOperatorDestroy_Sycl)); 1401 return CEED_ERROR_SUCCESS; 1402 } 1403 1404 //------------------------------------------------------------------------------ 1405