1 // Copyright (c) 2017-2026, 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 //------------------------------------------------------------------------------
CeedOperatorGetBasisPointer_Sycl(const CeedScalar ** basis_ptr,CeedEvalMode eval_mode,const CeedScalar * identity,const CeedScalar * interp,const CeedScalar * grad)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 //------------------------------------------------------------------------------
CeedOperatorDestroy_Sycl(CeedOperator op)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 CeedCallBackend(CeedDestroy(&ceed));
111 return CEED_ERROR_SUCCESS;
112 }
113
114 //------------------------------------------------------------------------------
115 // Setup infields or outfields
116 //------------------------------------------------------------------------------
CeedOperatorSetupFields_Sycl(CeedQFunction qf,CeedOperator op,bool is_input,CeedVector * e_vecs,CeedVector * q_vecs,CeedInt start_e,CeedInt num_fields,CeedInt Q,CeedInt num_elem)117 static int CeedOperatorSetupFields_Sycl(CeedQFunction qf, CeedOperator op, bool is_input, CeedVector *e_vecs, CeedVector *q_vecs, CeedInt start_e,
118 CeedInt num_fields, CeedInt Q, CeedInt num_elem) {
119 Ceed ceed;
120 CeedSize q_size;
121 bool is_strided, skip_restriction;
122 CeedInt size;
123 CeedOperatorField *op_fields;
124 CeedQFunctionField *qf_fields;
125
126 CeedCallBackend(CeedOperatorGetCeed(op, &ceed));
127 if (is_input) {
128 CeedCallBackend(CeedOperatorGetFields(op, NULL, &op_fields, NULL, NULL));
129 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_fields, NULL, NULL));
130 } else {
131 CeedCallBackend(CeedOperatorGetFields(op, NULL, NULL, NULL, &op_fields));
132 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, NULL, NULL, &qf_fields));
133 }
134
135 // Loop over fields
136 for (CeedInt i = 0; i < num_fields; i++) {
137 CeedEvalMode eval_mode;
138 CeedVector vec;
139 CeedElemRestriction elem_rstr;
140
141 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode));
142
143 is_strided = false;
144 skip_restriction = false;
145 if (eval_mode != CEED_EVAL_WEIGHT) {
146 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_fields[i], &elem_rstr));
147
148 // Check whether this field can skip the element restriction:
149 // must be passive input, with eval_mode NONE, and have a strided restriction with CEED_STRIDES_BACKEND.
150
151 // First, check whether the field is input or output:
152 if (is_input) {
153 // Check for passive input:
154 CeedCallBackend(CeedOperatorFieldGetVector(op_fields[i], &vec));
155 if (vec != CEED_VECTOR_ACTIVE) {
156 // Check eval_mode
157 if (eval_mode == CEED_EVAL_NONE) {
158 // Check for is_strided restriction
159 CeedCallBackend(CeedElemRestrictionIsStrided(elem_rstr, &is_strided));
160 if (is_strided) {
161 // Check if vector is already in preferred backend ordering
162 CeedCallBackend(CeedElemRestrictionHasBackendStrides(elem_rstr, &skip_restriction));
163 }
164 }
165 }
166 CeedCallBackend(CeedVectorDestroy(&vec));
167 }
168 if (skip_restriction) {
169 // We do not need an E-Vector, but will use the input field vector's data directly in the operator application
170 e_vecs[i + start_e] = NULL;
171 } else {
172 CeedCallBackend(CeedElemRestrictionCreateVector(elem_rstr, NULL, &e_vecs[i + start_e]));
173 }
174 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr));
175 }
176
177 switch (eval_mode) {
178 case CEED_EVAL_NONE:
179 CeedCallBackend(CeedQFunctionFieldGetSize(qf_fields[i], &size));
180 q_size = (CeedSize)num_elem * Q * size;
181 CeedCallBackend(CeedVectorCreate(ceed, q_size, &q_vecs[i]));
182 break;
183 case CEED_EVAL_INTERP:
184 CeedCallBackend(CeedQFunctionFieldGetSize(qf_fields[i], &size));
185 q_size = (CeedSize)num_elem * Q * size;
186 CeedCallBackend(CeedVectorCreate(ceed, q_size, &q_vecs[i]));
187 break;
188 case CEED_EVAL_GRAD:
189 CeedCallBackend(CeedQFunctionFieldGetSize(qf_fields[i], &size));
190 q_size = (CeedSize)num_elem * Q * size;
191 CeedCallBackend(CeedVectorCreate(ceed, q_size, &q_vecs[i]));
192 break;
193 case CEED_EVAL_WEIGHT: {
194 CeedBasis basis;
195
196 // Note: only on input fields
197 q_size = (CeedSize)num_elem * Q;
198 CeedCallBackend(CeedVectorCreate(ceed, q_size, &q_vecs[i]));
199 CeedCallBackend(CeedOperatorFieldGetBasis(op_fields[i], &basis));
200 CeedCallBackend(CeedBasisApply(basis, num_elem, CEED_NOTRANSPOSE, CEED_EVAL_WEIGHT, CEED_VECTOR_NONE, q_vecs[i]));
201 CeedCallBackend(CeedBasisDestroy(&basis));
202 break;
203 }
204 case CEED_EVAL_DIV:
205 break; // TODO: Not implemented
206 case CEED_EVAL_CURL:
207 break; // TODO: Not implemented
208 }
209 }
210 CeedCallBackend(CeedDestroy(&ceed));
211 return CEED_ERROR_SUCCESS;
212 }
213
214 //------------------------------------------------------------------------------
215 // CeedOperator needs to connect all the named fields (be they active or
216 // passive) to the named inputs and outputs of its CeedQFunction.
217 //------------------------------------------------------------------------------
CeedOperatorSetup_Sycl(CeedOperator op)218 static int CeedOperatorSetup_Sycl(CeedOperator op) {
219 bool is_setup_done;
220 CeedInt Q, num_elem, num_input_fields, num_output_fields;
221 CeedQFunctionField *qf_input_fields, *qf_output_fields;
222 CeedQFunction qf;
223 CeedOperatorField *op_input_fields, *op_output_fields;
224 CeedOperator_Sycl *impl;
225
226 CeedCallBackend(CeedOperatorIsSetupDone(op, &is_setup_done));
227 if (is_setup_done) return CEED_ERROR_SUCCESS;
228
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 CeedCallBackend(CeedQFunctionDestroy(&qf));
253 return CEED_ERROR_SUCCESS;
254 }
255
256 //------------------------------------------------------------------------------
257 // Setup Operator Inputs
258 //------------------------------------------------------------------------------
CeedOperatorSetupInputs_Sycl(CeedInt num_input_fields,CeedQFunctionField * qf_input_fields,CeedOperatorField * op_input_fields,CeedVector in_vec,const bool skip_active,CeedScalar * e_data[2* CEED_FIELD_MAX],CeedOperator_Sycl * impl,CeedRequest * request)259 static inline int CeedOperatorSetupInputs_Sycl(CeedInt num_input_fields, CeedQFunctionField *qf_input_fields, CeedOperatorField *op_input_fields,
260 CeedVector in_vec, const bool skip_active, CeedScalar *e_data[2 * CEED_FIELD_MAX],
261 CeedOperator_Sycl *impl, CeedRequest *request) {
262 for (CeedInt i = 0; i < num_input_fields; i++) {
263 bool is_active;
264 CeedEvalMode eval_mode;
265 CeedVector vec;
266
267 // Get input vector
268 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[i], &vec));
269 is_active = vec == CEED_VECTOR_ACTIVE;
270 if (is_active) {
271 if (skip_active) continue;
272 else vec = in_vec;
273 }
274
275 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode));
276 if (eval_mode == CEED_EVAL_WEIGHT) { // Skip
277 } else {
278 // Restrict, if necessary
279 if (!impl->e_vecs[i]) {
280 // No restriction for this field; read data directly from vec.
281 CeedCallBackend(CeedVectorGetArrayRead(vec, CEED_MEM_DEVICE, (const CeedScalar **)&e_data[i]));
282 } else {
283 CeedElemRestriction elem_rstr;
284
285 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_input_fields[i], &elem_rstr));
286 CeedCallBackend(CeedElemRestrictionApply(elem_rstr, CEED_NOTRANSPOSE, vec, impl->e_vecs[i], request));
287 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr));
288 CeedCallBackend(CeedVectorGetArrayRead(impl->e_vecs[i], CEED_MEM_DEVICE, (const CeedScalar **)&e_data[i]));
289 }
290 }
291 if (!is_active) CeedCallBackend(CeedVectorDestroy(&vec));
292 }
293 return CEED_ERROR_SUCCESS;
294 }
295
296 //------------------------------------------------------------------------------
297 // Input Basis Action
298 //------------------------------------------------------------------------------
CeedOperatorInputBasis_Sycl(CeedInt num_elem,CeedQFunctionField * qf_input_fields,CeedOperatorField * op_input_fields,CeedInt num_input_fields,const bool skip_active,CeedScalar * e_data[2* CEED_FIELD_MAX],CeedOperator_Sycl * impl)299 static inline int CeedOperatorInputBasis_Sycl(CeedInt num_elem, CeedQFunctionField *qf_input_fields, CeedOperatorField *op_input_fields,
300 CeedInt num_input_fields, const bool skip_active, CeedScalar *e_data[2 * CEED_FIELD_MAX],
301 CeedOperator_Sycl *impl) {
302 for (CeedInt i = 0; i < num_input_fields; i++) {
303 CeedEvalMode eval_mode;
304
305 // Skip active input
306 if (skip_active) {
307 bool is_active;
308 CeedVector vec;
309
310 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[i], &vec));
311 is_active = vec == CEED_VECTOR_ACTIVE;
312 CeedCallBackend(CeedVectorDestroy(&vec));
313 if (is_active) continue;
314 }
315 // Basis action
316 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode));
317 switch (eval_mode) {
318 case CEED_EVAL_NONE:
319 CeedCallBackend(CeedVectorSetArray(impl->q_vecs_in[i], CEED_MEM_DEVICE, CEED_USE_POINTER, e_data[i]));
320 break;
321 case CEED_EVAL_INTERP:
322 case CEED_EVAL_GRAD: {
323 CeedBasis basis;
324
325 CeedCallBackend(CeedOperatorFieldGetBasis(op_input_fields[i], &basis));
326 CeedCallBackend(CeedBasisApply(basis, num_elem, CEED_NOTRANSPOSE, eval_mode, impl->e_vecs[i], impl->q_vecs_in[i]));
327 CeedCallBackend(CeedBasisDestroy(&basis));
328 break;
329 }
330 case CEED_EVAL_WEIGHT:
331 break; // No action
332 case CEED_EVAL_DIV:
333 break; // TODO: Not implemented
334 case CEED_EVAL_CURL:
335 break; // TODO: Not implemented
336 }
337 }
338 return CEED_ERROR_SUCCESS;
339 }
340
341 //------------------------------------------------------------------------------
342 // Restore Input Vectors
343 //------------------------------------------------------------------------------
CeedOperatorRestoreInputs_Sycl(CeedInt num_input_fields,CeedQFunctionField * qf_input_fields,CeedOperatorField * op_input_fields,const bool skip_active,CeedScalar * e_data[2* CEED_FIELD_MAX],CeedOperator_Sycl * impl)344 static inline int CeedOperatorRestoreInputs_Sycl(CeedInt num_input_fields, CeedQFunctionField *qf_input_fields, CeedOperatorField *op_input_fields,
345 const bool skip_active, CeedScalar *e_data[2 * CEED_FIELD_MAX], CeedOperator_Sycl *impl) {
346 for (CeedInt i = 0; i < num_input_fields; i++) {
347 bool is_active;
348 CeedEvalMode eval_mode;
349 CeedVector vec;
350
351 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[i], &vec));
352 is_active = vec == CEED_VECTOR_ACTIVE;
353 // Skip active input
354 if (skip_active) {
355 if (is_active) continue;
356 }
357 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_input_fields[i], &eval_mode));
358 if (eval_mode == CEED_EVAL_WEIGHT) { // Skip
359 } else {
360 if (!impl->e_vecs[i]) { // This was a skip_restriction case
361 CeedCallBackend(CeedVectorRestoreArrayRead(vec, (const CeedScalar **)&e_data[i]));
362 } else {
363 CeedCallBackend(CeedVectorRestoreArrayRead(impl->e_vecs[i], (const CeedScalar **)&e_data[i]));
364 }
365 }
366 if (!is_active) CeedCallBackend(CeedVectorDestroy(&vec));
367 }
368 return CEED_ERROR_SUCCESS;
369 }
370
371 //------------------------------------------------------------------------------
372 // Apply and add to output
373 //------------------------------------------------------------------------------
CeedOperatorApplyAdd_Sycl(CeedOperator op,CeedVector in_vec,CeedVector out_vec,CeedRequest * request)374 static int CeedOperatorApplyAdd_Sycl(CeedOperator op, CeedVector in_vec, CeedVector out_vec, CeedRequest *request) {
375 CeedInt Q, num_elem, elem_size, num_input_fields, num_output_fields, size;
376 CeedEvalMode eval_mode;
377 CeedScalar *e_data[2 * CEED_FIELD_MAX] = {0};
378 CeedQFunctionField *qf_input_fields, *qf_output_fields;
379 CeedQFunction qf;
380 CeedOperatorField *op_input_fields, *op_output_fields;
381 CeedOperator_Sycl *impl;
382
383 CeedCallBackend(CeedOperatorGetData(op, &impl));
384 CeedCallBackend(CeedOperatorGetQFunction(op, &qf));
385 CeedCallBackend(CeedOperatorGetNumQuadraturePoints(op, &Q));
386 CeedCallBackend(CeedOperatorGetNumElements(op, &num_elem));
387 CeedCallBackend(CeedOperatorGetFields(op, &num_input_fields, &op_input_fields, &num_output_fields, &op_output_fields));
388 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_input_fields, NULL, &qf_output_fields));
389
390 // Setup
391 CeedCallBackend(CeedOperatorSetup_Sycl(op));
392
393 // Input Evecs and Restriction
394 CeedCallBackend(CeedOperatorSetupInputs_Sycl(num_input_fields, qf_input_fields, op_input_fields, in_vec, false, e_data, impl, request));
395
396 // Input basis apply if needed
397 CeedCallBackend(CeedOperatorInputBasis_Sycl(num_elem, qf_input_fields, op_input_fields, num_input_fields, false, e_data, impl));
398
399 // Output pointers, as necessary
400 for (CeedInt i = 0; i < num_output_fields; i++) {
401 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode));
402 if (eval_mode == CEED_EVAL_NONE) {
403 // Set the output Q-Vector to use the E-Vector data directly
404 CeedCallBackend(CeedVectorGetArrayWrite(impl->e_vecs[i + impl->num_e_in], CEED_MEM_DEVICE, &e_data[i + num_input_fields]));
405 CeedCallBackend(CeedVectorSetArray(impl->q_vecs_out[i], CEED_MEM_DEVICE, CEED_USE_POINTER, e_data[i + num_input_fields]));
406 }
407 }
408
409 // Q function
410 CeedCallBackend(CeedQFunctionApply(qf, num_elem * Q, impl->q_vecs_in, impl->q_vecs_out));
411
412 // Output basis apply if needed
413 for (CeedInt i = 0; i < num_output_fields; i++) {
414 CeedElemRestriction elem_rstr;
415
416 // Get elem_size, eval_mode, size
417 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_output_fields[i], &elem_rstr));
418 CeedCallBackend(CeedElemRestrictionGetElementSize(elem_rstr, &elem_size));
419 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr));
420 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode));
421 CeedCallBackend(CeedQFunctionFieldGetSize(qf_output_fields[i], &size));
422 // Basis action
423 switch (eval_mode) {
424 case CEED_EVAL_NONE:
425 break;
426 case CEED_EVAL_INTERP:
427 case CEED_EVAL_GRAD: {
428 CeedBasis basis;
429
430 CeedCallBackend(CeedOperatorFieldGetBasis(op_output_fields[i], &basis));
431 CeedCallBackend(CeedBasisApply(basis, num_elem, CEED_TRANSPOSE, eval_mode, impl->q_vecs_out[i], impl->e_vecs[i + impl->num_e_in]));
432 CeedCallBackend(CeedBasisDestroy(&basis));
433 break;
434 }
435 // LCOV_EXCL_START
436 case CEED_EVAL_WEIGHT:
437 return CeedError(CeedOperatorReturnCeed(op), CEED_ERROR_BACKEND, "CEED_EVAL_WEIGHT cannot be an output evaluation mode");
438 break; // Should not occur
439 case CEED_EVAL_DIV:
440 case CEED_EVAL_CURL:
441 return CeedError(CeedOperatorReturnCeed(op), CEED_ERROR_BACKEND, "%s not supported", CeedEvalModes[eval_mode]);
442 break; // Should not occur
443 // LCOV_EXCL_STOP
444 }
445 }
446
447 // Output restriction
448 for (CeedInt i = 0; i < num_output_fields; i++) {
449 bool is_active;
450 CeedEvalMode eval_mode;
451 CeedVector vec;
452 CeedElemRestriction elem_rstr;
453
454 // Restore evec
455 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_output_fields[i], &eval_mode));
456 if (eval_mode == CEED_EVAL_NONE) {
457 CeedCallBackend(CeedVectorRestoreArray(impl->e_vecs[i + impl->num_e_in], &e_data[i + num_input_fields]));
458 }
459 // Restrict
460 CeedCallBackend(CeedOperatorFieldGetVector(op_output_fields[i], &vec));
461 is_active = vec == CEED_VECTOR_ACTIVE;
462 if (is_active) vec = out_vec;
463 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_output_fields[i], &elem_rstr));
464 CeedCallBackend(CeedElemRestrictionApply(elem_rstr, CEED_TRANSPOSE, impl->e_vecs[i + impl->num_e_in], vec, request));
465 if (!is_active) CeedCallBackend(CeedVectorDestroy(&vec));
466 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr));
467 }
468
469 // Restore input arrays
470 CeedCallBackend(CeedOperatorRestoreInputs_Sycl(num_input_fields, qf_input_fields, op_input_fields, false, e_data, impl));
471 CeedCallBackend(CeedQFunctionDestroy(&qf));
472 return CEED_ERROR_SUCCESS;
473 }
474
475 //------------------------------------------------------------------------------
476 // Core code for assembling linear QFunction
477 //------------------------------------------------------------------------------
CeedOperatorLinearAssembleQFunctionCore_Sycl(CeedOperator op,bool build_objects,CeedVector * assembled,CeedElemRestriction * elem_rstr,CeedRequest * request)478 static inline int CeedOperatorLinearAssembleQFunctionCore_Sycl(CeedOperator op, bool build_objects, CeedVector *assembled,
479 CeedElemRestriction *elem_rstr, CeedRequest *request) {
480 Ceed ceed_parent;
481 CeedSize q_size;
482 CeedInt num_active_in, num_active_out, Q, num_elem, num_input_fields, num_output_fields, size;
483 CeedScalar *assembled_array, *e_data[2 * CEED_FIELD_MAX] = {NULL};
484 CeedVector *active_in;
485 CeedQFunctionField *qf_input_fields, *qf_output_fields;
486 CeedQFunction qf;
487 CeedOperatorField *op_input_fields, *op_output_fields;
488 CeedOperator_Sycl *impl;
489
490 CeedCallBackend(CeedOperatorGetFallbackParentCeed(op, &ceed_parent));
491 CeedCallBackend(CeedOperatorGetData(op, &impl));
492 CeedCallBackend(CeedOperatorGetQFunction(op, &qf));
493 CeedCallBackend(CeedOperatorGetNumQuadraturePoints(op, &Q));
494 CeedCallBackend(CeedOperatorGetNumElements(op, &num_elem));
495 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_input_fields, NULL, &qf_output_fields));
496 CeedCallBackend(CeedOperatorGetFields(op, &num_input_fields, &op_input_fields, &num_output_fields, &op_output_fields));
497 active_in = impl->qf_active_in;
498 num_active_in = impl->num_active_in, num_active_out = impl->num_active_out;
499
500 // Setup
501 CeedCallBackend(CeedOperatorSetup_Sycl(op));
502
503 // Input Evecs and Restriction
504 CeedCallBackend(CeedOperatorSetupInputs_Sycl(num_input_fields, qf_input_fields, op_input_fields, NULL, true, e_data, impl, request));
505
506 // Count number of active input fields
507 if (!num_active_in) {
508 for (CeedInt i = 0; i < num_input_fields; i++) {
509 CeedScalar *q_vec_array;
510 CeedVector vec;
511
512 // Check if active input
513 CeedCallBackend(CeedOperatorFieldGetVector(op_input_fields[i], &vec));
514 if (vec == CEED_VECTOR_ACTIVE) {
515 CeedCallBackend(CeedQFunctionFieldGetSize(qf_input_fields[i], &size));
516 CeedCallBackend(CeedVectorSetValue(impl->q_vecs_in[i], 0.0));
517 CeedCallBackend(CeedVectorGetArray(impl->q_vecs_in[i], CEED_MEM_DEVICE, &q_vec_array));
518 CeedCallBackend(CeedRealloc(num_active_in + size, &active_in));
519 for (CeedInt field = 0; field < size; field++) {
520 q_size = (CeedSize)Q * num_elem;
521 CeedCallBackend(CeedVectorCreate(ceed_parent, q_size, &active_in[num_active_in + field]));
522 CeedCallBackend(CeedVectorSetArray(active_in[num_active_in + field], CEED_MEM_DEVICE, CEED_USE_POINTER,
523 &q_vec_array[field * Q * num_elem]));
524 }
525 num_active_in += size;
526 CeedCallBackend(CeedVectorRestoreArray(impl->q_vecs_in[i], &q_vec_array));
527 }
528 CeedCallBackend(CeedVectorDestroy(&vec));
529 }
530 impl->num_active_in = num_active_in;
531 impl->qf_active_in = active_in;
532 }
533
534 // Count number of active output fields
535 if (!num_active_out) {
536 for (CeedInt i = 0; i < num_output_fields; i++) {
537 CeedVector vec;
538
539 // Check if active output
540 CeedCallBackend(CeedOperatorFieldGetVector(op_output_fields[i], &vec));
541 if (vec == CEED_VECTOR_ACTIVE) {
542 CeedCallBackend(CeedQFunctionFieldGetSize(qf_output_fields[i], &size));
543 num_active_out += size;
544 }
545 CeedCallBackend(CeedVectorDestroy(&vec));
546 }
547 impl->num_active_out = num_active_out;
548 }
549
550 // Check sizes
551 CeedCheck(num_active_in > 0 && num_active_out > 0, CeedOperatorReturnCeed(op), CEED_ERROR_BACKEND,
552 "Cannot assemble QFunction without active inputs and outputs");
553
554 // Build objects if needed
555 if (build_objects) {
556 CeedSize l_size = (CeedSize)num_elem * Q * num_active_in * num_active_out;
557 CeedInt strides[3] = {1, num_elem * Q, Q}; /* *NOPAD* */
558
559 // Create output restriction
560 CeedCallBackend(CeedElemRestrictionCreateStrided(ceed_parent, num_elem, Q, num_active_in * num_active_out, l_size, strides, elem_rstr));
561 // Create assembled vector
562 CeedCallBackend(CeedVectorCreate(ceed_parent, l_size, assembled));
563 }
564 CeedCallBackend(CeedVectorSetValue(*assembled, 0.0));
565 CeedCallBackend(CeedVectorGetArray(*assembled, CEED_MEM_DEVICE, &assembled_array));
566
567 // Input basis apply
568 CeedCallBackend(CeedOperatorInputBasis_Sycl(num_elem, qf_input_fields, op_input_fields, num_input_fields, true, e_data, impl));
569
570 // Assemble QFunction
571 for (CeedInt in = 0; in < num_active_in; in++) {
572 // Set Inputs
573 CeedCallBackend(CeedVectorSetValue(active_in[in], 1.0));
574 if (num_active_in > 1) {
575 CeedCallBackend(CeedVectorSetValue(active_in[(in + num_active_in - 1) % num_active_in], 0.0));
576 }
577 // Set Outputs
578 for (CeedInt out = 0; out < num_output_fields; out++) {
579 CeedVector vec;
580
581 // Check if active output
582 CeedCallBackend(CeedOperatorFieldGetVector(op_output_fields[out], &vec));
583 if (vec == CEED_VECTOR_ACTIVE) {
584 CeedCallBackend(CeedVectorSetArray(impl->q_vecs_out[out], CEED_MEM_DEVICE, CEED_USE_POINTER, assembled_array));
585 CeedCallBackend(CeedQFunctionFieldGetSize(qf_output_fields[out], &size));
586 assembled_array += size * Q * num_elem; // Advance the pointer by the size of the output
587 }
588 CeedCallBackend(CeedVectorDestroy(&vec));
589 }
590 // Apply QFunction
591 CeedCallBackend(CeedQFunctionApply(qf, Q * num_elem, impl->q_vecs_in, impl->q_vecs_out));
592 }
593
594 // Un-set output Qvecs to prevent accidental overwrite of Assembled
595 for (CeedInt out = 0; out < num_output_fields; out++) {
596 CeedVector vec;
597
598 // Check if active output
599 CeedCallBackend(CeedOperatorFieldGetVector(op_output_fields[out], &vec));
600 if (vec == CEED_VECTOR_ACTIVE) {
601 CeedCallBackend(CeedVectorTakeArray(impl->q_vecs_out[out], CEED_MEM_DEVICE, NULL));
602 }
603 CeedCallBackend(CeedVectorDestroy(&vec));
604 }
605
606 // Restore input arrays
607 CeedCallBackend(CeedOperatorRestoreInputs_Sycl(num_input_fields, qf_input_fields, op_input_fields, true, e_data, impl));
608
609 // Restore output
610 CeedCallBackend(CeedVectorRestoreArray(*assembled, &assembled_array));
611 CeedCallBackend(CeedDestroy(&ceed_parent));
612 CeedCallBackend(CeedQFunctionDestroy(&qf));
613 return CEED_ERROR_SUCCESS;
614 }
615
616 //------------------------------------------------------------------------------
617 // Assemble Linear QFunction
618 //------------------------------------------------------------------------------
CeedOperatorLinearAssembleQFunction_Sycl(CeedOperator op,CeedVector * assembled,CeedElemRestriction * elem_rstr,CeedRequest * request)619 static int CeedOperatorLinearAssembleQFunction_Sycl(CeedOperator op, CeedVector *assembled, CeedElemRestriction *elem_rstr, CeedRequest *request) {
620 return CeedOperatorLinearAssembleQFunctionCore_Sycl(op, true, assembled, elem_rstr, request);
621 }
622
623 //------------------------------------------------------------------------------
624 // Update Assembled Linear QFunction
625 //------------------------------------------------------------------------------
CeedOperatorLinearAssembleQFunctionUpdate_Sycl(CeedOperator op,CeedVector assembled,CeedElemRestriction elem_rstr,CeedRequest * request)626 static int CeedOperatorLinearAssembleQFunctionUpdate_Sycl(CeedOperator op, CeedVector assembled, CeedElemRestriction elem_rstr,
627 CeedRequest *request) {
628 return CeedOperatorLinearAssembleQFunctionCore_Sycl(op, false, &assembled, &elem_rstr, request);
629 }
630
631 //------------------------------------------------------------------------------
632 // Assemble diagonal setup
633 //------------------------------------------------------------------------------
CeedOperatorAssembleDiagonalSetup_Sycl(CeedOperator op)634 static inline int CeedOperatorAssembleDiagonalSetup_Sycl(CeedOperator op) {
635 Ceed ceed;
636 Ceed_Sycl *sycl_data;
637 CeedInt num_input_fields, num_output_fields, num_eval_mode_in = 0, num_comp = 0, dim = 1, num_eval_mode_out = 0;
638 CeedEvalMode *eval_mode_in = NULL, *eval_mode_out = NULL;
639 CeedElemRestriction rstr_in = NULL, rstr_out = NULL;
640 CeedBasis basis_in = NULL, basis_out = NULL;
641 CeedQFunctionField *qf_fields;
642 CeedQFunction qf;
643 CeedOperatorField *op_fields;
644 CeedOperator_Sycl *impl;
645
646 CeedCallBackend(CeedOperatorGetCeed(op, &ceed));
647 CeedCallBackend(CeedOperatorGetQFunction(op, &qf));
648 CeedCallBackend(CeedQFunctionGetNumArgs(qf, &num_input_fields, &num_output_fields));
649
650 // Determine active input basis
651 CeedCallBackend(CeedOperatorGetFields(op, NULL, &op_fields, NULL, NULL));
652 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_fields, NULL, NULL));
653 for (CeedInt i = 0; i < num_input_fields; i++) {
654 CeedVector vec;
655
656 CeedCallBackend(CeedOperatorFieldGetVector(op_fields[i], &vec));
657 if (vec == CEED_VECTOR_ACTIVE) {
658 CeedEvalMode eval_mode;
659 CeedElemRestriction elem_rstr;
660 CeedBasis basis;
661
662 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_fields[i], &elem_rstr));
663 if (!rstr_in) CeedCallBackend(CeedElemRestrictionReferenceCopy(elem_rstr, &rstr_in));
664 CeedCheck(rstr_in == elem_rstr, ceed, CEED_ERROR_BACKEND, "Backend does not implement multi-field non-composite operator diagonal assembly");
665 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr));
666 CeedCallBackend(CeedOperatorFieldGetBasis(op_fields[i], &basis));
667 if (!basis_in) CeedCallBackend(CeedBasisReferenceCopy(basis, &basis_in));
668 CeedCheck(basis_in == basis, ceed, CEED_ERROR_BACKEND, "Backend does not implement operator diagonal assembly with multiple active bases");
669 CeedCallBackend(CeedBasisDestroy(&basis));
670 CeedCallBackend(CeedBasisGetDimension(basis_in, &dim));
671 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode));
672 switch (eval_mode) {
673 case CEED_EVAL_NONE:
674 case CEED_EVAL_INTERP:
675 CeedCallBackend(CeedRealloc(num_eval_mode_in + 1, &eval_mode_in));
676 eval_mode_in[num_eval_mode_in] = eval_mode;
677 num_eval_mode_in += 1;
678 break;
679 case CEED_EVAL_GRAD:
680 CeedCallBackend(CeedRealloc(num_eval_mode_in + dim, &eval_mode_in));
681 for (CeedInt d = 0; d < dim; d++) eval_mode_in[num_eval_mode_in + d] = eval_mode;
682 num_eval_mode_in += dim;
683 break;
684 case CEED_EVAL_WEIGHT:
685 case CEED_EVAL_DIV:
686 case CEED_EVAL_CURL:
687 break; // Caught by QF Assembly
688 }
689 }
690 CeedCallBackend(CeedVectorDestroy(&vec));
691 }
692 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_in));
693
694 // Determine active output basis
695 CeedCallBackend(CeedOperatorGetFields(op, NULL, NULL, NULL, &op_fields));
696 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, NULL, NULL, &qf_fields));
697 for (CeedInt i = 0; i < num_output_fields; i++) {
698 CeedVector vec;
699
700 CeedCallBackend(CeedOperatorFieldGetVector(op_fields[i], &vec));
701 if (vec == CEED_VECTOR_ACTIVE) {
702 CeedEvalMode eval_mode;
703 CeedElemRestriction elem_rstr;
704 CeedBasis basis;
705
706 CeedCallBackend(CeedOperatorFieldGetElemRestriction(op_fields[i], &elem_rstr));
707 if (!rstr_out) CeedCallBackend(CeedElemRestrictionReferenceCopy(elem_rstr, &rstr_out));
708 CeedCheck(rstr_out == elem_rstr, ceed, CEED_ERROR_BACKEND, "Backend does not implement multi-field non-composite operator diagonal assembly");
709 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr));
710 CeedCallBackend(CeedOperatorFieldGetBasis(op_fields[i], &basis));
711 if (!basis_out) CeedCallBackend(CeedBasisReferenceCopy(basis, &basis_out));
712 CeedCheck(basis_out == basis, ceed, CEED_ERROR_BACKEND, "Backend does not implement operator diagonal assembly with multiple active bases");
713 CeedCallBackend(CeedBasisDestroy(&basis));
714 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode));
715 switch (eval_mode) {
716 case CEED_EVAL_NONE:
717 case CEED_EVAL_INTERP:
718 CeedCallBackend(CeedRealloc(num_eval_mode_out + 1, &eval_mode_out));
719 eval_mode_out[num_eval_mode_out] = eval_mode;
720 num_eval_mode_out += 1;
721 break;
722 case CEED_EVAL_GRAD:
723 CeedCallBackend(CeedRealloc(num_eval_mode_out + dim, &eval_mode_out));
724 for (CeedInt d = 0; d < dim; d++) eval_mode_out[num_eval_mode_out + d] = eval_mode;
725 num_eval_mode_out += dim;
726 break;
727 case CEED_EVAL_WEIGHT:
728 case CEED_EVAL_DIV:
729 case CEED_EVAL_CURL:
730 break; // Caught by QF Assembly
731 }
732 }
733 CeedCallBackend(CeedVectorDestroy(&vec));
734 }
735
736 // Operator data struct
737 CeedCallBackend(CeedOperatorGetData(op, &impl));
738 CeedCallBackend(CeedGetData(ceed, &sycl_data));
739 CeedCallBackend(CeedCalloc(1, &impl->diag));
740 CeedOperatorDiag_Sycl *diag = impl->diag;
741
742 CeedCallBackend(CeedBasisReferenceCopy(basis_in, &diag->basis_in));
743 CeedCallBackend(CeedBasisReferenceCopy(basis_out, &diag->basis_out));
744 diag->h_eval_mode_in = eval_mode_in;
745 diag->h_eval_mode_out = eval_mode_out;
746 diag->num_eval_mode_in = num_eval_mode_in;
747 diag->num_eval_mode_out = num_eval_mode_out;
748
749 // Kernel parameters
750 CeedInt num_nodes, num_qpts;
751 CeedCallBackend(CeedBasisGetNumNodes(basis_in, &num_nodes));
752 CeedCallBackend(CeedBasisGetNumQuadraturePoints(basis_in, &num_qpts));
753 CeedCallBackend(CeedBasisGetNumComponents(basis_in, &num_comp));
754 diag->num_nodes = num_nodes;
755 diag->num_qpts = num_qpts;
756 diag->num_comp = num_comp;
757
758 // Basis matrices
759 const CeedInt i_len = num_qpts * num_nodes;
760 const CeedInt g_len = num_qpts * num_nodes * dim;
761 const CeedScalar *interp_in, *interp_out, *grad_in, *grad_out;
762
763 // CEED_EVAL_NONE
764 CeedScalar *identity = NULL;
765 bool has_eval_none = false;
766 for (CeedInt i = 0; i < num_eval_mode_in; i++) has_eval_none = has_eval_none || (eval_mode_in[i] == CEED_EVAL_NONE);
767 for (CeedInt i = 0; i < num_eval_mode_out; i++) has_eval_none = has_eval_none || (eval_mode_out[i] == CEED_EVAL_NONE);
768
769 std::vector<sycl::event> e;
770
771 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()};
772
773 std::vector<sycl::event> copy_events;
774
775 if (has_eval_none) {
776 CeedCallBackend(CeedCalloc(num_qpts * num_nodes, &identity));
777 for (CeedSize i = 0; i < (num_nodes < num_qpts ? num_nodes : num_qpts); i++) identity[i * num_nodes + i] = 1.0;
778 CeedCallSycl(ceed, diag->d_identity = sycl::malloc_device<CeedScalar>(i_len, sycl_data->sycl_device, sycl_data->sycl_context));
779 sycl::event identity_copy = sycl_data->sycl_queue.copy<CeedScalar>(identity, diag->d_identity, i_len, e);
780 copy_events.push_back(identity_copy);
781 }
782
783 // CEED_EVAL_INTERP
784 CeedCallBackend(CeedBasisGetInterp(basis_in, &interp_in));
785 CeedCallSycl(ceed, diag->d_interp_in = sycl::malloc_device<CeedScalar>(i_len, sycl_data->sycl_device, sycl_data->sycl_context));
786 sycl::event interp_in_copy = sycl_data->sycl_queue.copy<CeedScalar>(interp_in, diag->d_interp_in, i_len, e);
787 copy_events.push_back(interp_in_copy);
788
789 CeedCallBackend(CeedBasisGetInterp(basis_out, &interp_out));
790 CeedCallSycl(ceed, diag->d_interp_out = sycl::malloc_device<CeedScalar>(i_len, sycl_data->sycl_device, sycl_data->sycl_context));
791 sycl::event interp_out_copy = sycl_data->sycl_queue.copy<CeedScalar>(interp_out, diag->d_interp_out, i_len, e);
792 copy_events.push_back(interp_out_copy);
793
794 // CEED_EVAL_GRAD
795 CeedCallBackend(CeedBasisGetGrad(basis_in, &grad_in));
796 CeedCallSycl(ceed, diag->d_grad_in = sycl::malloc_device<CeedScalar>(g_len, sycl_data->sycl_device, sycl_data->sycl_context));
797 sycl::event grad_in_copy = sycl_data->sycl_queue.copy<CeedScalar>(grad_in, diag->d_grad_in, g_len, e);
798 copy_events.push_back(grad_in_copy);
799
800 CeedCallBackend(CeedBasisGetGrad(basis_out, &grad_out));
801 CeedCallSycl(ceed, diag->d_grad_out = sycl::malloc_device<CeedScalar>(g_len, sycl_data->sycl_device, sycl_data->sycl_context));
802 sycl::event grad_out_copy = sycl_data->sycl_queue.copy<CeedScalar>(grad_out, diag->d_grad_out, g_len, e);
803 copy_events.push_back(grad_out_copy);
804
805 // Arrays of eval_modes
806 CeedCallSycl(ceed, diag->d_eval_mode_in = sycl::malloc_device<CeedEvalMode>(num_eval_mode_in, sycl_data->sycl_device, sycl_data->sycl_context));
807 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);
808 copy_events.push_back(eval_mode_in_copy);
809
810 CeedCallSycl(ceed, diag->d_eval_mode_out = sycl::malloc_device<CeedEvalMode>(num_eval_mode_out, sycl_data->sycl_device, sycl_data->sycl_context));
811 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);
812 copy_events.push_back(eval_mode_out_copy);
813
814 // Restriction
815 CeedCallBackend(CeedElemRestrictionReferenceCopy(rstr_out, &diag->diag_rstr));
816 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_out));
817
818 // Wait for all copies to complete and handle exceptions
819 CeedCallSycl(ceed, sycl::event::wait_and_throw(copy_events));
820
821 // Cleanup
822 CeedCallBackend(CeedDestroy(&ceed));
823 CeedCallBackend(CeedBasisDestroy(&basis_in));
824 CeedCallBackend(CeedBasisDestroy(&basis_out));
825 CeedCallBackend(CeedQFunctionDestroy(&qf));
826 return CEED_ERROR_SUCCESS;
827 }
828
829 //------------------------------------------------------------------------------
830 // Kernel for diagonal assembly
831 //------------------------------------------------------------------------------
CeedOperatorLinearDiagonal_Sycl(sycl::queue & sycl_queue,const bool is_point_block,const CeedInt num_elem,const CeedOperatorDiag_Sycl * diag,const CeedScalar * assembled_qf_array,CeedScalar * elem_diag_array)832 static int CeedOperatorLinearDiagonal_Sycl(sycl::queue &sycl_queue, const bool is_point_block, const CeedInt num_elem,
833 const CeedOperatorDiag_Sycl *diag, const CeedScalar *assembled_qf_array, CeedScalar *elem_diag_array) {
834 const CeedSize num_nodes = diag->num_nodes;
835 const CeedSize num_qpts = diag->num_qpts;
836 const CeedSize num_comp = diag->num_comp;
837 const CeedSize num_eval_mode_in = diag->num_eval_mode_in;
838 const CeedSize num_eval_mode_out = diag->num_eval_mode_out;
839 const CeedScalar *identity = diag->d_identity;
840 const CeedScalar *interp_in = diag->d_interp_in;
841 const CeedScalar *grad_in = diag->d_grad_in;
842 const CeedScalar *interp_out = diag->d_interp_out;
843 const CeedScalar *grad_out = diag->d_grad_out;
844 const CeedEvalMode *eval_mode_in = diag->d_eval_mode_in;
845 const CeedEvalMode *eval_mode_out = diag->d_eval_mode_out;
846
847 sycl::range<1> kernel_range(num_elem * num_nodes);
848
849 std::vector<sycl::event> e;
850
851 if (!sycl_queue.is_in_order()) e = {sycl_queue.ext_oneapi_submit_barrier()};
852
853 sycl_queue.parallel_for<CeedOperatorSyclLinearDiagonal>(kernel_range, e, [=](sycl::id<1> idx) {
854 const CeedInt tid = idx % num_nodes;
855 const CeedInt e = idx / num_nodes;
856
857 // Compute the diagonal of B^T D B
858 // Each element
859 CeedInt d_out = -1;
860 // Each basis eval mode pair
861 for (CeedSize e_out = 0; e_out < num_eval_mode_out; e_out++) {
862 const CeedScalar *bt = NULL;
863
864 if (eval_mode_out[e_out] == CEED_EVAL_GRAD) ++d_out;
865 CeedOperatorGetBasisPointer_Sycl(&bt, eval_mode_out[e_out], identity, interp_out, &grad_out[d_out * num_qpts * num_nodes]);
866 CeedInt d_in = -1;
867
868 for (CeedSize e_in = 0; e_in < num_eval_mode_in; e_in++) {
869 const CeedScalar *b = NULL;
870
871 if (eval_mode_in[e_in] == CEED_EVAL_GRAD) ++d_in;
872 CeedOperatorGetBasisPointer_Sycl(&b, eval_mode_in[e_in], identity, interp_in, &grad_in[d_in * num_qpts * num_nodes]);
873 // Each component
874 for (CeedSize comp_out = 0; comp_out < num_comp; comp_out++) {
875 // Each qpoint/node pair
876 if (is_point_block) {
877 // Point Block Diagonal
878 for (CeedInt comp_in = 0; comp_in < num_comp; comp_in++) {
879 CeedScalar e_value = 0.0;
880
881 for (CeedSize q = 0; q < num_qpts; q++) {
882 const CeedScalar qf_value =
883 assembled_qf_array[((((e_in * num_comp + comp_in) * num_eval_mode_out + e_out) * num_comp + comp_out) * num_elem + e) * num_qpts +
884 q];
885
886 e_value += bt[q * num_nodes + tid] * qf_value * b[q * num_nodes + tid];
887 }
888 elem_diag_array[((comp_out * num_comp + comp_in) * num_elem + e) * num_nodes + tid] += e_value;
889 }
890 } else {
891 // Diagonal Only
892 CeedScalar e_value = 0.0;
893
894 for (CeedSize q = 0; q < num_qpts; q++) {
895 const CeedScalar qf_value =
896 assembled_qf_array[((((e_in * num_comp + comp_out) * num_eval_mode_out + e_out) * num_comp + comp_out) * num_elem + e) * num_qpts +
897 q];
898 e_value += bt[q * num_nodes + tid] * qf_value * b[q * num_nodes + tid];
899 }
900 elem_diag_array[(comp_out * num_elem + e) * num_nodes + tid] += e_value;
901 }
902 }
903 }
904 }
905 });
906 return CEED_ERROR_SUCCESS;
907 }
908
909 //------------------------------------------------------------------------------
910 // Assemble diagonal common code
911 //------------------------------------------------------------------------------
CeedOperatorAssembleDiagonalCore_Sycl(CeedOperator op,CeedVector assembled,CeedRequest * request,const bool is_point_block)912 static inline int CeedOperatorAssembleDiagonalCore_Sycl(CeedOperator op, CeedVector assembled, CeedRequest *request, const bool is_point_block) {
913 Ceed ceed;
914 Ceed_Sycl *sycl_data;
915 CeedInt num_elem;
916 CeedScalar *elem_diag_array;
917 const CeedScalar *assembled_qf_array;
918 CeedVector assembled_qf = NULL;
919 CeedOperator_Sycl *impl;
920
921 CeedCallBackend(CeedOperatorGetCeed(op, &ceed));
922 CeedCallBackend(CeedGetData(ceed, &sycl_data));
923 CeedCallBackend(CeedDestroy(&ceed));
924 CeedCallBackend(CeedOperatorGetData(op, &impl));
925
926 // Assemble QFunction
927 {
928 CeedElemRestriction elem_rstr = NULL;
929
930 CeedCallBackend(CeedOperatorLinearAssembleQFunctionBuildOrUpdate(op, &assembled_qf, &elem_rstr, request));
931 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr));
932 }
933
934 // Setup
935 if (!impl->diag) {
936 CeedCallBackend(CeedOperatorAssembleDiagonalSetup_Sycl(op));
937 }
938 CeedOperatorDiag_Sycl *diag = impl->diag;
939
940 assert(diag != NULL);
941
942 // Restriction
943 if (is_point_block && !diag->point_block_diag_rstr) {
944 CeedCallBackend(CeedOperatorCreateActivePointBlockRestriction(diag->diag_rstr, &diag->point_block_diag_rstr));
945 }
946 CeedElemRestriction diag_rstr = is_point_block ? diag->point_block_diag_rstr : diag->diag_rstr;
947
948 // Create diagonal vector
949 CeedVector elem_diag = is_point_block ? diag->point_block_elem_diag : diag->elem_diag;
950
951 if (!elem_diag) {
952 CeedCallBackend(CeedElemRestrictionCreateVector(diag_rstr, NULL, &elem_diag));
953 if (is_point_block) diag->point_block_elem_diag = elem_diag;
954 else diag->elem_diag = elem_diag;
955 }
956 CeedCallBackend(CeedVectorSetValue(elem_diag, 0.0));
957
958 // Assemble element operator diagonals
959 CeedCallBackend(CeedVectorGetArray(elem_diag, CEED_MEM_DEVICE, &elem_diag_array));
960 CeedCallBackend(CeedVectorGetArrayRead(assembled_qf, CEED_MEM_DEVICE, &assembled_qf_array));
961 CeedCallBackend(CeedElemRestrictionGetNumElements(diag_rstr, &num_elem));
962
963 // Compute the diagonal of B^T D B
964 CeedCallBackend(CeedOperatorLinearDiagonal_Sycl(sycl_data->sycl_queue, is_point_block, num_elem, diag, assembled_qf_array, elem_diag_array));
965
966 // Wait for queue to complete and handle exceptions
967 sycl_data->sycl_queue.wait_and_throw();
968
969 // Restore arrays
970 CeedCallBackend(CeedVectorRestoreArray(elem_diag, &elem_diag_array));
971 CeedCallBackend(CeedVectorRestoreArrayRead(assembled_qf, &assembled_qf_array));
972
973 // Assemble local operator diagonal
974 CeedCallBackend(CeedElemRestrictionApply(diag_rstr, CEED_TRANSPOSE, elem_diag, assembled, request));
975
976 // Cleanup
977 CeedCallBackend(CeedVectorDestroy(&assembled_qf));
978 return CEED_ERROR_SUCCESS;
979 }
980
981 //------------------------------------------------------------------------------
982 // Assemble Linear Diagonal
983 //------------------------------------------------------------------------------
CeedOperatorLinearAssembleAddDiagonal_Sycl(CeedOperator op,CeedVector assembled,CeedRequest * request)984 static int CeedOperatorLinearAssembleAddDiagonal_Sycl(CeedOperator op, CeedVector assembled, CeedRequest *request) {
985 CeedCallBackend(CeedOperatorAssembleDiagonalCore_Sycl(op, assembled, request, false));
986 return CEED_ERROR_SUCCESS;
987 }
988
989 //------------------------------------------------------------------------------
990 // Assemble Linear Point Block Diagonal
991 //------------------------------------------------------------------------------
CeedOperatorLinearAssembleAddPointBlockDiagonal_Sycl(CeedOperator op,CeedVector assembled,CeedRequest * request)992 static int CeedOperatorLinearAssembleAddPointBlockDiagonal_Sycl(CeedOperator op, CeedVector assembled, CeedRequest *request) {
993 CeedCallBackend(CeedOperatorAssembleDiagonalCore_Sycl(op, assembled, request, true));
994 return CEED_ERROR_SUCCESS;
995 }
996
997 //------------------------------------------------------------------------------
998 // Single operator assembly setup
999 //------------------------------------------------------------------------------
CeedOperatorAssembleSingleSetup_Sycl(CeedOperator op)1000 static int CeedOperatorAssembleSingleSetup_Sycl(CeedOperator op) {
1001 Ceed ceed;
1002 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,
1003 num_B_out_mats_to_load = 0, size_B_out = 0, num_qpts = 0, elem_size = 0, num_elem, num_comp,
1004 mat_start = 0;
1005 CeedEvalMode *eval_mode_in = NULL, *eval_mode_out = NULL;
1006 const CeedScalar *interp_in, *grad_in;
1007 CeedElemRestriction rstr_in = NULL, rstr_out = NULL;
1008 CeedBasis basis_in = NULL, basis_out = NULL;
1009 CeedQFunctionField *qf_fields;
1010 CeedQFunction qf;
1011 CeedOperatorField *input_fields, *output_fields;
1012 CeedOperator_Sycl *impl;
1013
1014 CeedCallBackend(CeedOperatorGetCeed(op, &ceed));
1015 CeedCallBackend(CeedOperatorGetData(op, &impl));
1016
1017 // Get input and output fields
1018 CeedCallBackend(CeedOperatorGetFields(op, &num_input_fields, &input_fields, &num_output_fields, &output_fields));
1019
1020 // Determine active input basis eval mode
1021 CeedCallBackend(CeedOperatorGetQFunction(op, &qf));
1022 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, &qf_fields, NULL, NULL));
1023 // Note that the kernel will treat each dimension of a gradient action separately;
1024 // i.e., when an active input has a CEED_EVAL_GRAD mode, num_ eval_mode_in will increment by dim.
1025 // 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,
1026 // so num_B_in_mats_to_load will be incremented by 1.
1027 for (CeedInt i = 0; i < num_input_fields; i++) {
1028 CeedVector vec;
1029
1030 CeedCallBackend(CeedOperatorFieldGetVector(input_fields[i], &vec));
1031 if (vec == CEED_VECTOR_ACTIVE) {
1032 CeedEvalMode eval_mode;
1033 CeedElemRestriction elem_rstr;
1034 CeedBasis basis;
1035
1036 CeedCallBackend(CeedOperatorFieldGetElemRestriction(input_fields[i], &elem_rstr));
1037 if (!rstr_in) CeedCallBackend(CeedElemRestrictionReferenceCopy(elem_rstr, &rstr_in));
1038 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr));
1039 CeedCallBackend(CeedElemRestrictionGetElementSize(rstr_in, &elem_size));
1040 CeedCallBackend(CeedOperatorFieldGetBasis(input_fields[i], &basis));
1041 if (!basis_in) CeedCallBackend(CeedBasisReferenceCopy(basis, &basis_in));
1042 CeedCheck(basis_in == basis, ceed, CEED_ERROR_BACKEND, "Backend does not implement operator assembly with multiple active bases");
1043 CeedCallBackend(CeedBasisDestroy(&basis));
1044 CeedCallBackend(CeedBasisGetDimension(basis_in, &dim));
1045 CeedCallBackend(CeedBasisGetNumQuadraturePoints(basis_in, &num_qpts));
1046 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode));
1047 if (eval_mode != CEED_EVAL_NONE) {
1048 CeedCallBackend(CeedRealloc(num_B_in_mats_to_load + 1, &eval_mode_in));
1049 eval_mode_in[num_B_in_mats_to_load] = eval_mode;
1050 num_B_in_mats_to_load += 1;
1051 if (eval_mode == CEED_EVAL_GRAD) {
1052 num_eval_mode_in += dim;
1053 size_B_in += dim * elem_size * num_qpts;
1054 } else {
1055 num_eval_mode_in += 1;
1056 size_B_in += elem_size * num_qpts;
1057 }
1058 }
1059 }
1060 CeedCallBackend(CeedVectorDestroy(&vec));
1061 }
1062
1063 // Determine active output basis; basis_out and rstr_out only used if same as input, TODO
1064 CeedCallBackend(CeedQFunctionGetFields(qf, NULL, NULL, NULL, &qf_fields));
1065 for (CeedInt i = 0; i < num_output_fields; i++) {
1066 CeedVector vec;
1067
1068 CeedCallBackend(CeedOperatorFieldGetVector(output_fields[i], &vec));
1069 if (vec == CEED_VECTOR_ACTIVE) {
1070 CeedEvalMode eval_mode;
1071 CeedElemRestriction elem_rstr;
1072 CeedBasis basis;
1073
1074 CeedCallBackend(CeedOperatorFieldGetElemRestriction(output_fields[i], &elem_rstr));
1075 if (!rstr_out) CeedCallBackend(CeedElemRestrictionReferenceCopy(elem_rstr, &rstr_out));
1076 CeedCheck(rstr_out == rstr_in, ceed, CEED_ERROR_BACKEND, "Backend does not implement multi-field non-composite operator assembly");
1077 CeedCallBackend(CeedElemRestrictionDestroy(&elem_rstr));
1078 CeedCallBackend(CeedOperatorFieldGetBasis(output_fields[i], &basis));
1079 if (!basis_out) CeedCallBackend(CeedBasisReferenceCopy(basis, &basis_out));
1080 CeedCheck(basis_out == basis, ceed, CEED_ERROR_BACKEND, "Backend does not implement operator assembly with multiple active bases");
1081 CeedCallBackend(CeedBasisDestroy(&basis));
1082 CeedCallBackend(CeedQFunctionFieldGetEvalMode(qf_fields[i], &eval_mode));
1083 if (eval_mode != CEED_EVAL_NONE) {
1084 CeedCallBackend(CeedRealloc(num_B_out_mats_to_load + 1, &eval_mode_out));
1085 eval_mode_out[num_B_out_mats_to_load] = eval_mode;
1086 num_B_out_mats_to_load += 1;
1087 if (eval_mode == CEED_EVAL_GRAD) {
1088 num_eval_mode_out += dim;
1089 size_B_out += dim * elem_size * num_qpts;
1090 } else {
1091 num_eval_mode_out += 1;
1092 size_B_out += elem_size * num_qpts;
1093 }
1094 }
1095 }
1096 CeedCallBackend(CeedVectorDestroy(&vec));
1097 }
1098 CeedCheck(num_eval_mode_in > 0 && num_eval_mode_out > 0, ceed, CEED_ERROR_UNSUPPORTED, "Cannot assemble operator without inputs/outputs");
1099
1100 CeedCallBackend(CeedElemRestrictionGetNumElements(rstr_in, &num_elem));
1101 CeedCallBackend(CeedElemRestrictionGetNumComponents(rstr_in, &num_comp));
1102
1103 CeedCallBackend(CeedCalloc(1, &impl->asmb));
1104 CeedOperatorAssemble_Sycl *asmb = impl->asmb;
1105 asmb->num_elem = num_elem;
1106
1107 Ceed_Sycl *sycl_data;
1108 CeedCallBackend(CeedGetData(ceed, &sycl_data));
1109
1110 // Kernel setup
1111 int elems_per_block = 1;
1112 asmb->elems_per_block = elems_per_block;
1113 asmb->block_size_x = elem_size;
1114 asmb->block_size_y = elem_size;
1115 asmb->num_eval_mode_in = num_eval_mode_in;
1116 asmb->num_eval_mode_out = num_eval_mode_out;
1117 asmb->num_qpts = num_qpts;
1118 asmb->num_nodes = elem_size;
1119 asmb->block_size = elem_size * elem_size * elems_per_block;
1120 asmb->num_comp = num_comp;
1121
1122 // Build 'full' B matrices (not 1D arrays used for tensor-product matrices
1123 CeedCallBackend(CeedBasisGetInterp(basis_in, &interp_in));
1124 CeedCallBackend(CeedBasisGetGrad(basis_in, &grad_in));
1125
1126 // Load into B_in, in order that they will be used in eval_mode
1127 CeedCallSycl(ceed, asmb->d_B_in = sycl::malloc_device<CeedScalar>(size_B_in, sycl_data->sycl_device, sycl_data->sycl_context));
1128 for (int i = 0; i < num_B_in_mats_to_load; i++) {
1129 CeedEvalMode eval_mode = eval_mode_in[i];
1130
1131 if (eval_mode == CEED_EVAL_INTERP) {
1132 std::vector<sycl::event> e;
1133
1134 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()};
1135 sycl_data->sycl_queue.copy<CeedScalar>(interp_in, &asmb->d_B_in[mat_start], elem_size * num_qpts, e);
1136 mat_start += elem_size * num_qpts;
1137 } else if (eval_mode == CEED_EVAL_GRAD) {
1138 std::vector<sycl::event> e;
1139
1140 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()};
1141 sycl_data->sycl_queue.copy<CeedScalar>(grad_in, &asmb->d_B_in[mat_start], dim * elem_size * num_qpts, e);
1142 mat_start += dim * elem_size * num_qpts;
1143 }
1144 }
1145
1146 const CeedScalar *interp_out, *grad_out;
1147 // Note that this function currently assumes 1 basis, so this should always be true
1148 // for now
1149 if (basis_out == basis_in) {
1150 interp_out = interp_in;
1151 grad_out = grad_in;
1152 } else {
1153 CeedCallBackend(CeedBasisGetInterp(basis_out, &interp_out));
1154 CeedCallBackend(CeedBasisGetGrad(basis_out, &grad_out));
1155 }
1156
1157 // Load into B_out, in order that they will be used in eval_mode
1158 mat_start = 0;
1159 CeedCallSycl(ceed, asmb->d_B_out = sycl::malloc_device<CeedScalar>(size_B_out, sycl_data->sycl_device, sycl_data->sycl_context));
1160 for (int i = 0; i < num_B_out_mats_to_load; i++) {
1161 CeedEvalMode eval_mode = eval_mode_out[i];
1162
1163 if (eval_mode == CEED_EVAL_INTERP) {
1164 std::vector<sycl::event> e;
1165
1166 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()};
1167 sycl_data->sycl_queue.copy<CeedScalar>(interp_out, &asmb->d_B_out[mat_start], elem_size * num_qpts, e);
1168 mat_start += elem_size * num_qpts;
1169 } else if (eval_mode == CEED_EVAL_GRAD) {
1170 std::vector<sycl::event> e;
1171
1172 if (!sycl_data->sycl_queue.is_in_order()) e = {sycl_data->sycl_queue.ext_oneapi_submit_barrier()};
1173 sycl_data->sycl_queue.copy<CeedScalar>(grad_out, &asmb->d_B_out[mat_start], dim * elem_size * num_qpts, e);
1174 mat_start += dim * elem_size * num_qpts;
1175 }
1176 }
1177 CeedCallBackend(CeedDestroy(&ceed));
1178 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_in));
1179 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_out));
1180 CeedCallBackend(CeedBasisDestroy(&basis_in));
1181 CeedCallBackend(CeedBasisDestroy(&basis_out));
1182 CeedCallBackend(CeedQFunctionDestroy(&qf));
1183 return CEED_ERROR_SUCCESS;
1184 }
1185
1186 //------------------------------------------------------------------------------
1187 // Matrix assembly kernel for low-order elements (3D thread block)
1188 //------------------------------------------------------------------------------
CeedOperatorLinearAssemble_Sycl(sycl::queue & sycl_queue,const CeedOperator_Sycl * impl,const CeedScalar * qf_array,CeedScalar * values_array)1189 static int CeedOperatorLinearAssemble_Sycl(sycl::queue &sycl_queue, const CeedOperator_Sycl *impl, const CeedScalar *qf_array,
1190 CeedScalar *values_array) {
1191 // This kernels assumes B_in and B_out have the same number of quadrature points and basis points.
1192 // TODO: expand to more general cases
1193 CeedOperatorAssemble_Sycl *asmb = impl->asmb;
1194 const CeedInt num_elem = asmb->num_elem;
1195 const CeedSize num_nodes = asmb->num_nodes;
1196 const CeedSize num_comp = asmb->num_comp;
1197 const CeedSize num_qpts = asmb->num_qpts;
1198 const CeedSize num_eval_mode_in = asmb->num_eval_mode_in;
1199 const CeedSize num_eval_mode_out = asmb->num_eval_mode_out;
1200
1201 // Strides for final output ordering, determined by the reference (inference) implementation of the symbolic assembly, slowest --> fastest: element,
1202 // comp_in, comp_out, node_row, node_col
1203 const CeedSize comp_out_stride = num_nodes * num_nodes;
1204 const CeedSize comp_in_stride = comp_out_stride * num_comp;
1205 const CeedSize e_stride = comp_in_stride * num_comp;
1206 // Strides for QF array, slowest --> fastest: eval_mode_in, comp_in, eval_mode_out, comp_out, elem, qpt
1207 const CeedSize q_e_stride = num_qpts;
1208 const CeedSize q_comp_out_stride = num_elem * q_e_stride;
1209 const CeedSize q_eval_mode_out_stride = q_comp_out_stride * num_comp;
1210 const CeedSize q_comp_in_stride = q_eval_mode_out_stride * num_eval_mode_out;
1211 const CeedSize q_eval_mode_in_stride = q_comp_in_stride * num_comp;
1212
1213 CeedScalar *B_in, *B_out;
1214 B_in = asmb->d_B_in;
1215 B_out = asmb->d_B_out;
1216 const CeedInt block_size_x = asmb->block_size_x;
1217 const CeedInt block_size_y = asmb->block_size_y;
1218
1219 sycl::range<3> kernel_range(num_elem, block_size_y, block_size_x);
1220
1221 std::vector<sycl::event> e;
1222
1223 if (!sycl_queue.is_in_order()) e = {sycl_queue.ext_oneapi_submit_barrier()};
1224 sycl_queue.parallel_for<CeedOperatorSyclLinearAssemble>(kernel_range, e, [=](sycl::id<3> idx) {
1225 const int e = idx.get(0); // Element index
1226 const int l = idx.get(1); // The output column index of each B^TDB operation
1227 const int i = idx.get(2); // The output row index of each B^TDB operation
1228 // such that we have (Bout^T)_ij D_jk Bin_kl = C_il
1229 for (CeedSize comp_in = 0; comp_in < num_comp; comp_in++) {
1230 for (CeedSize comp_out = 0; comp_out < num_comp; comp_out++) {
1231 CeedScalar result = 0.0;
1232 CeedSize qf_index_comp = q_comp_in_stride * comp_in + q_comp_out_stride * comp_out + q_e_stride * e;
1233
1234 for (CeedSize eval_mode_in = 0; eval_mode_in < num_eval_mode_in; eval_mode_in++) {
1235 CeedSize b_in_index = eval_mode_in * num_qpts * num_nodes;
1236
1237 for (CeedSize eval_mode_out = 0; eval_mode_out < num_eval_mode_out; eval_mode_out++) {
1238 CeedSize b_out_index = eval_mode_out * num_qpts * num_nodes;
1239 CeedSize qf_index = qf_index_comp + q_eval_mode_out_stride * eval_mode_out + q_eval_mode_in_stride * eval_mode_in;
1240
1241 // Perform the B^T D B operation for this 'chunk' of D (the qf_array)
1242 for (CeedSize j = 0; j < num_qpts; j++) {
1243 result += B_out[b_out_index + j * num_nodes + i] * qf_array[qf_index + j] * B_in[b_in_index + j * num_nodes + l];
1244 }
1245 } // end of eval_mode_out
1246 } // end of eval_mode_in
1247 CeedSize val_index = comp_in_stride * comp_in + comp_out_stride * comp_out + e_stride * e + num_nodes * i + l;
1248
1249 values_array[val_index] = result;
1250 } // end of out component
1251 } // end of in component
1252 });
1253 return CEED_ERROR_SUCCESS;
1254 }
1255
1256 //------------------------------------------------------------------------------
1257 // Fallback kernel for larger orders (1D thread block)
1258 //------------------------------------------------------------------------------
1259 /*
1260 static int CeedOperatorLinearAssembleFallback_Sycl(sycl::queue &sycl_queue, const CeedOperator_Sycl *impl, const CeedScalar *qf_array,
1261 CeedScalar *values_array) {
1262 // This kernel assumes B_in and B_out have the same number of quadrature points and basis points.
1263 // TODO: expand to more general cases
1264 CeedOperatorAssemble_Sycl *asmb = impl->asmb;
1265 const CeedInt num_elem = asmb->num_elem;
1266 const CeedInt num_nodes = asmb->num_nodes;
1267 const CeedInt num_comp = asmb->num_comp;
1268 const CeedInt num_qpts = asmb->num_qpts;
1269 const CeedInt num_eval_mode_in = asmb->num_eval_mode_in;
1270 const CeedInt num_eval_mode_out = asmb->num_eval_mode_out;
1271
1272 // Strides for final output ordering, determined by the reference (interface) implementation of the symbolic assembly, slowest --> fastest: elememt,
1273 // comp_in, comp_out, node_row, node_col
1274 const CeedInt comp_out_stride = num_nodes * num_nodes;
1275 const CeedInt comp_in_stride = comp_out_stride * num_comp;
1276 const CeedInt e_stride = comp_in_stride * num_comp;
1277 // Strides for QF array, slowest --> fastest: eval_mode_in, comp_in, eval_mode_out, comp_out, elem, qpt
1278 const CeedInt q_e_stride = num_qpts;
1279 const CeedInt q_comp_out_stride = num_elem * q_e_stride;
1280 const CeedInt q_eval_mode_out_stride = q_comp_out_stride * num_comp;
1281 const CeedInt q_comp_in_stride = q_eval_mode_out_stride * num_eval_mode_out;
1282 const CeedInt q_eval_mode_in_stride = q_comp_in_stride * num_comp;
1283
1284 CeedScalar *B_in, *B_out;
1285 B_in = asmb->d_B_in;
1286 B_out = asmb->d_B_out;
1287 const CeedInt elems_per_block = asmb->elems_per_block;
1288 const CeedInt block_size_x = asmb->block_size_x;
1289 const CeedInt block_size_y = asmb->block_size_y; // This will be 1 for the fallback kernel
1290
1291 const CeedInt grid = num_elem / elems_per_block + ((num_elem / elems_per_block * elems_per_block < num_elem) ? 1 : 0);
1292 sycl::range<3> local_range(block_size_x, block_size_y, elems_per_block);
1293 sycl::range<3> global_range(grid * block_size_x, block_size_y, elems_per_block);
1294 sycl::nd_range<3> kernel_range(global_range, local_range);
1295
1296 sycl_queue.parallel_for<CeedOperatorSyclLinearAssembleFallback>(kernel_range, [=](sycl::nd_item<3> work_item) {
1297 const CeedInt blockIdx = work_item.get_group(0);
1298 const CeedInt gridDimx = work_item.get_group_range(0);
1299 const CeedInt threadIdx = work_item.get_local_id(0);
1300 const CeedInt threadIdz = work_item.get_local_id(2);
1301 const CeedInt blockDimz = work_item.get_local_range(2);
1302
1303 const int l = threadIdx; // The output column index of each B^TDB operation
1304 // such that we have (Bout^T)_ij D_jk Bin_kl = C_il
1305 for (CeedInt e = blockIdx * blockDimz + threadIdz; e < num_elem; e += gridDimx * blockDimz) {
1306 for (CeedInt comp_in = 0; comp_in < num_comp; comp_in++) {
1307 for (CeedInt comp_out = 0; comp_out < num_comp; comp_out++) {
1308 for (CeedInt i = 0; i < num_nodes; i++) {
1309 CeedScalar result = 0.0;
1310 CeedInt qf_index_comp = q_comp_in_stride * comp_in + q_comp_out_stride * comp_out + q_e_stride * e;
1311 for (CeedInt eval_mode_in = 0; eval_mode_in < num_eval_mode_in; eval_mode_in++) {
1312 CeedInt b_in_index = eval_mode_in * num_qpts * num_nodes;
1313 for (CeedInt eval_mode_out = 0; eval_mode_out < num_eval_mode_out; eval_mode_out++) {
1314 CeedInt b_out_index = eval_mode_out * num_qpts * num_nodes;
1315 CeedInt qf_index = qf_index_comp + q_eval_mode_out_stride * eval_mode_out + q_eval_mode_in_stride * eval_mode_in;
1316 // Perform the B^T D B operation for this 'chunk' of D (the qf_array)
1317 for (CeedInt j = 0; j < num_qpts; j++) {
1318 result += B_out[b_out_index + j * num_nodes + i] * qf_array[qf_index + j] * B_in[b_in_index + j * num_nodes + l];
1319 }
1320 } // end of eval_mode_out
1321 } // end of eval_mode_in
1322 CeedInt val_index = comp_in_stride * comp_in + comp_out_stride * comp_out + e_stride * e + num_nodes * i + l;
1323 values_array[val_index] = result;
1324 } // end of loop over element node index, i
1325 } // end of out component
1326 } // end of in component
1327 } // end of element loop
1328 });
1329 return CEED_ERROR_SUCCESS;
1330 }*/
1331
1332 //------------------------------------------------------------------------------
1333 // Assemble matrix data for COO matrix of assembled operator.
1334 // The sparsity pattern is set by CeedOperatorLinearAssembleSymbolic.
1335 //
1336 // Note that this (and other assembly routines) currently assume only one active
1337 // input restriction/basis per operator (could have multiple basis eval modes).
1338 // TODO: allow multiple active input restrictions/basis objects
1339 //------------------------------------------------------------------------------
CeedOperatorAssembleSingle_Sycl(CeedOperator op,CeedInt offset,CeedVector values)1340 static int CeedOperatorAssembleSingle_Sycl(CeedOperator op, CeedInt offset, CeedVector values) {
1341 Ceed ceed;
1342 Ceed_Sycl *sycl_data;
1343 CeedScalar *values_array;
1344 const CeedScalar *qf_array;
1345 CeedVector assembled_qf = NULL;
1346 CeedElemRestriction rstr_q = NULL;
1347 CeedOperator_Sycl *impl;
1348
1349 CeedCallBackend(CeedOperatorGetCeed(op, &ceed));
1350 CeedCallBackend(CeedGetData(ceed, &sycl_data));
1351 CeedCallBackend(CeedDestroy(&ceed));
1352 CeedCallBackend(CeedOperatorGetData(op, &impl));
1353
1354 // Setup
1355 if (!impl->asmb) {
1356 CeedCallBackend(CeedOperatorAssembleSingleSetup_Sycl(op));
1357 assert(impl->asmb != NULL);
1358 }
1359
1360 // Assemble QFunction
1361 CeedCallBackend(CeedOperatorLinearAssembleQFunctionBuildOrUpdate(op, &assembled_qf, &rstr_q, CEED_REQUEST_IMMEDIATE));
1362 CeedCallBackend(CeedElemRestrictionDestroy(&rstr_q));
1363 CeedCallBackend(CeedVectorGetArrayWrite(values, CEED_MEM_DEVICE, &values_array));
1364 values_array += offset;
1365 CeedCallBackend(CeedVectorGetArrayRead(assembled_qf, CEED_MEM_DEVICE, &qf_array));
1366
1367 // Compute B^T D B
1368 CeedCallBackend(CeedOperatorLinearAssemble_Sycl(sycl_data->sycl_queue, impl, qf_array, values_array));
1369
1370 // Wait for kernels to be completed
1371 // Kris: Review if this is necessary -- enqueing an async barrier may be sufficient
1372 sycl_data->sycl_queue.wait_and_throw();
1373
1374 // Restore arrays
1375 CeedCallBackend(CeedVectorRestoreArray(values, &values_array));
1376 CeedCallBackend(CeedVectorRestoreArrayRead(assembled_qf, &qf_array));
1377
1378 // Cleanup
1379 CeedCallBackend(CeedVectorDestroy(&assembled_qf));
1380 return CEED_ERROR_SUCCESS;
1381 }
1382
1383 //------------------------------------------------------------------------------
1384 // Create operator
1385 //------------------------------------------------------------------------------
CeedOperatorCreate_Sycl(CeedOperator op)1386 int CeedOperatorCreate_Sycl(CeedOperator op) {
1387 Ceed ceed;
1388 CeedOperator_Sycl *impl;
1389
1390 CeedCallBackend(CeedOperatorGetCeed(op, &ceed));
1391
1392 CeedCallBackend(CeedCalloc(1, &impl));
1393 CeedCallBackend(CeedOperatorSetData(op, impl));
1394
1395 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleQFunction", CeedOperatorLinearAssembleQFunction_Sycl));
1396 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleQFunctionUpdate", CeedOperatorLinearAssembleQFunctionUpdate_Sycl));
1397 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleAddDiagonal", CeedOperatorLinearAssembleAddDiagonal_Sycl));
1398 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleAddPointBlockDiagonal",
1399 CeedOperatorLinearAssembleAddPointBlockDiagonal_Sycl));
1400 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "LinearAssembleSingle", CeedOperatorAssembleSingle_Sycl));
1401 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "ApplyAdd", CeedOperatorApplyAdd_Sycl));
1402 CeedCallBackend(CeedSetBackendFunctionCpp(ceed, "Operator", op, "Destroy", CeedOperatorDestroy_Sycl));
1403 CeedCallBackend(CeedDestroy(&ceed));
1404 return CEED_ERROR_SUCCESS;
1405 }
1406
1407 //------------------------------------------------------------------------------
1408