-
Notifications
You must be signed in to change notification settings - Fork 11.8k
/
OpenMP.cpp
3171 lines (2811 loc) · 138 KB
/
OpenMP.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
//===-- OpenMP.cpp -- Open MP directive lowering --------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// Coding style: https://mlir.llvm.org/getting_started/DeveloperGuide/
//
//===----------------------------------------------------------------------===//
#include "flang/Lower/OpenMP.h"
#include "ClauseProcessor.h"
#include "Clauses.h"
#include "DataSharingProcessor.h"
#include "Decomposer.h"
#include "DirectivesCommon.h"
#include "ReductionProcessor.h"
#include "Utils.h"
#include "flang/Common/idioms.h"
#include "flang/Lower/Bridge.h"
#include "flang/Lower/ConvertExpr.h"
#include "flang/Lower/ConvertVariable.h"
#include "flang/Lower/StatementContext.h"
#include "flang/Lower/SymbolMap.h"
#include "flang/Optimizer/Builder/BoxValue.h"
#include "flang/Optimizer/Builder/FIRBuilder.h"
#include "flang/Optimizer/Builder/Todo.h"
#include "flang/Optimizer/Dialect/FIRType.h"
#include "flang/Optimizer/HLFIR/HLFIROps.h"
#include "flang/Parser/parse-tree.h"
#include "flang/Semantics/openmp-directive-sets.h"
#include "flang/Semantics/tools.h"
#include "mlir/Dialect/ControlFlow/IR/ControlFlowOps.h"
#include "mlir/Dialect/OpenMP/OpenMPDialect.h"
#include "mlir/Transforms/RegionUtils.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Frontend/OpenMP/OMPConstants.h"
using namespace Fortran::lower::omp;
//===----------------------------------------------------------------------===//
// Code generation helper functions
//===----------------------------------------------------------------------===//
namespace {
/// Structure holding the information needed to create and bind entry block
/// arguments associated to a single clause.
struct EntryBlockArgsEntry {
llvm::ArrayRef<const semantics::Symbol *> syms;
llvm::ArrayRef<mlir::Value> vars;
bool isValid() const {
// This check allows specifying a smaller number of symbols than values
// because in some case cases a single symbol generates multiple block
// arguments.
return syms.size() <= vars.size();
}
};
/// Structure holding the information needed to create and bind entry block
/// arguments associated to all clauses that can define them.
struct EntryBlockArgs {
EntryBlockArgsEntry inReduction;
EntryBlockArgsEntry map;
EntryBlockArgsEntry priv;
EntryBlockArgsEntry reduction;
EntryBlockArgsEntry taskReduction;
EntryBlockArgsEntry useDeviceAddr;
EntryBlockArgsEntry useDevicePtr;
bool isValid() const {
return inReduction.isValid() && map.isValid() && priv.isValid() &&
reduction.isValid() && taskReduction.isValid() &&
useDeviceAddr.isValid() && useDevicePtr.isValid();
}
};
} // namespace
static void genOMPDispatch(lower::AbstractConverter &converter,
lower::SymMap &symTable,
semantics::SemanticsContext &semaCtx,
lower::pft::Evaluation &eval, mlir::Location loc,
const ConstructQueue &queue,
ConstructQueue::const_iterator item);
/// Bind symbols to their corresponding entry block arguments.
///
/// The binding will be performed inside of the current block, which does not
/// necessarily have to be part of the operation for which the binding is done.
/// However, block arguments must be accessible. This enables controlling the
/// insertion point of any new MLIR operations related to the binding of
/// arguments of a loop wrapper operation.
///
/// \param [in] converter - PFT to MLIR conversion interface.
/// \param [in] op - owner operation of the block arguments to bind.
/// \param [in] args - entry block arguments information for the given
/// operation.
static void bindEntryBlockArgs(lower::AbstractConverter &converter,
mlir::omp::BlockArgOpenMPOpInterface op,
const EntryBlockArgs &args) {
assert(op != nullptr && "invalid block argument-defining operation");
assert(args.isValid() && "invalid args");
fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
auto bindSingleMapLike = [&converter,
&firOpBuilder](const semantics::Symbol &sym,
const mlir::BlockArgument &arg) {
// Clones the `bounds` placing them inside the entry block and returns
// them.
auto cloneBound = [&](mlir::Value bound) {
if (mlir::isMemoryEffectFree(bound.getDefiningOp())) {
mlir::Operation *clonedOp = firOpBuilder.clone(*bound.getDefiningOp());
return clonedOp->getResult(0);
}
TODO(converter.getCurrentLocation(),
"target map-like clause operand unsupported bound type");
};
auto cloneBounds = [cloneBound](llvm::ArrayRef<mlir::Value> bounds) {
llvm::SmallVector<mlir::Value> clonedBounds;
llvm::transform(bounds, std::back_inserter(clonedBounds),
[&](mlir::Value bound) { return cloneBound(bound); });
return clonedBounds;
};
fir::ExtendedValue extVal = converter.getSymbolExtendedValue(sym);
auto refType = mlir::dyn_cast<fir::ReferenceType>(arg.getType());
if (refType && fir::isa_builtin_cptr_type(refType.getElementType())) {
converter.bindSymbol(sym, arg);
} else {
extVal.match(
[&](const fir::BoxValue &v) {
converter.bindSymbol(sym,
fir::BoxValue(arg, cloneBounds(v.getLBounds()),
v.getExplicitParameters(),
v.getExplicitExtents()));
},
[&](const fir::MutableBoxValue &v) {
converter.bindSymbol(
sym, fir::MutableBoxValue(arg, cloneBounds(v.getLBounds()),
v.getMutableProperties()));
},
[&](const fir::ArrayBoxValue &v) {
converter.bindSymbol(
sym, fir::ArrayBoxValue(arg, cloneBounds(v.getExtents()),
cloneBounds(v.getLBounds()),
v.getSourceBox()));
},
[&](const fir::CharArrayBoxValue &v) {
converter.bindSymbol(
sym, fir::CharArrayBoxValue(arg, cloneBound(v.getLen()),
cloneBounds(v.getExtents()),
cloneBounds(v.getLBounds())));
},
[&](const fir::CharBoxValue &v) {
converter.bindSymbol(
sym, fir::CharBoxValue(arg, cloneBound(v.getLen())));
},
[&](const fir::UnboxedValue &v) { converter.bindSymbol(sym, arg); },
[&](const auto &) {
TODO(converter.getCurrentLocation(),
"target map clause operand unsupported type");
});
}
};
auto bindMapLike =
[&bindSingleMapLike](llvm::ArrayRef<const semantics::Symbol *> syms,
llvm::ArrayRef<mlir::BlockArgument> args) {
// Structure component symbols don't have bindings, and can only be
// explicitly mapped individually. If a member is captured implicitly
// we map the entirety of the derived type when we find its symbol.
llvm::SmallVector<const semantics::Symbol *> processedSyms;
llvm::copy_if(syms, std::back_inserter(processedSyms),
[](auto *sym) { return !sym->owner().IsDerivedType(); });
for (auto [sym, arg] : llvm::zip_equal(processedSyms, args))
bindSingleMapLike(*sym, arg);
};
auto bindPrivateLike = [&converter, &firOpBuilder](
llvm::ArrayRef<const semantics::Symbol *> syms,
llvm::ArrayRef<mlir::Value> vars,
llvm::ArrayRef<mlir::BlockArgument> args) {
llvm::SmallVector<const semantics::Symbol *> processedSyms;
for (auto *sym : syms) {
if (const auto *commonDet =
sym->detailsIf<semantics::CommonBlockDetails>()) {
llvm::transform(commonDet->objects(), std::back_inserter(processedSyms),
[&](const auto &mem) { return &*mem; });
} else {
processedSyms.push_back(sym);
}
}
for (auto [sym, var, arg] : llvm::zip_equal(processedSyms, vars, args))
converter.bindSymbol(
*sym,
hlfir::translateToExtendedValue(
var.getLoc(), firOpBuilder, hlfir::Entity{arg},
/*contiguousHint=*/
evaluate::IsSimplyContiguous(*sym, converter.getFoldingContext()))
.first);
};
// Process in clause name alphabetical order to match block arguments order.
bindPrivateLike(args.inReduction.syms, args.inReduction.vars,
op.getInReductionBlockArgs());
bindMapLike(args.map.syms, op.getMapBlockArgs());
bindPrivateLike(args.priv.syms, args.priv.vars, op.getPrivateBlockArgs());
bindPrivateLike(args.reduction.syms, args.reduction.vars,
op.getReductionBlockArgs());
bindPrivateLike(args.taskReduction.syms, args.taskReduction.vars,
op.getTaskReductionBlockArgs());
bindMapLike(args.useDeviceAddr.syms, op.getUseDeviceAddrBlockArgs());
bindMapLike(args.useDevicePtr.syms, op.getUseDevicePtrBlockArgs());
}
/// Get the list of base values that the specified map-like variables point to.
///
/// This function must be kept in sync with changes to the `createMapInfoOp`
/// utility function, since it must take into account the potential introduction
/// of levels of indirection (i.e. intermediate ops).
///
/// \param [in] vars - list of values passed to map-like clauses, returned
/// by an `omp.map.info` operation.
/// \param [out] baseOps - populated with the `var_ptr` values of the
/// corresponding defining operations.
static void
extractMappedBaseValues(llvm::ArrayRef<mlir::Value> vars,
llvm::SmallVectorImpl<mlir::Value> &baseOps) {
llvm::transform(vars, std::back_inserter(baseOps), [](mlir::Value map) {
auto mapInfo = map.getDefiningOp<mlir::omp::MapInfoOp>();
assert(mapInfo && "expected all map vars to be defined by omp.map.info");
mlir::Value varPtr = mapInfo.getVarPtr();
if (auto boxAddr = varPtr.getDefiningOp<fir::BoxAddrOp>())
return boxAddr.getVal();
return varPtr;
});
}
static lower::pft::Evaluation *
getCollapsedLoopEval(lower::pft::Evaluation &eval, int collapseValue) {
// Return the Evaluation of the innermost collapsed loop, or the current one
// if there was no COLLAPSE.
if (collapseValue == 0)
return &eval;
lower::pft::Evaluation *curEval = &eval.getFirstNestedEvaluation();
for (int i = 1; i < collapseValue; i++) {
// The nested evaluations should be DoConstructs (i.e. they should form
// a loop nest). Each DoConstruct is a tuple <NonLabelDoStmt, Block,
// EndDoStmt>.
assert(curEval->isA<parser::DoConstruct>());
curEval = &*std::next(curEval->getNestedEvaluations().begin());
}
return curEval;
}
static void genNestedEvaluations(lower::AbstractConverter &converter,
lower::pft::Evaluation &eval,
int collapseValue = 0) {
lower::pft::Evaluation *curEval = getCollapsedLoopEval(eval, collapseValue);
for (lower::pft::Evaluation &e : curEval->getNestedEvaluations())
converter.genEval(e);
}
static fir::GlobalOp globalInitialization(lower::AbstractConverter &converter,
fir::FirOpBuilder &firOpBuilder,
const semantics::Symbol &sym,
const lower::pft::Variable &var,
mlir::Location currentLocation) {
mlir::Type ty = converter.genType(sym);
std::string globalName = converter.mangleName(sym);
mlir::StringAttr linkage = firOpBuilder.createInternalLinkage();
fir::GlobalOp global =
firOpBuilder.createGlobal(currentLocation, ty, globalName, linkage);
// Create default initialization for non-character scalar.
if (semantics::IsAllocatableOrObjectPointer(&sym)) {
mlir::Type baseAddrType = mlir::dyn_cast<fir::BoxType>(ty).getEleTy();
lower::createGlobalInitialization(
firOpBuilder, global, [&](fir::FirOpBuilder &b) {
mlir::Value nullAddr =
b.createNullConstant(currentLocation, baseAddrType);
mlir::Value box =
b.create<fir::EmboxOp>(currentLocation, ty, nullAddr);
b.create<fir::HasValueOp>(currentLocation, box);
});
} else {
lower::createGlobalInitialization(
firOpBuilder, global, [&](fir::FirOpBuilder &b) {
mlir::Value undef = b.create<fir::UndefOp>(currentLocation, ty);
b.create<fir::HasValueOp>(currentLocation, undef);
});
}
return global;
}
// Get the extended value for \p val by extracting additional variable
// information from \p base.
static fir::ExtendedValue getExtendedValue(fir::ExtendedValue base,
mlir::Value val) {
return base.match(
[&](const fir::MutableBoxValue &box) -> fir::ExtendedValue {
return fir::MutableBoxValue(val, box.nonDeferredLenParams(), {});
},
[&](const auto &) -> fir::ExtendedValue {
return fir::substBase(base, val);
});
}
#ifndef NDEBUG
static bool isThreadPrivate(lower::SymbolRef sym) {
if (const auto *details = sym->detailsIf<semantics::CommonBlockDetails>()) {
for (const auto &obj : details->objects())
if (!obj->test(semantics::Symbol::Flag::OmpThreadprivate))
return false;
return true;
}
return sym->test(semantics::Symbol::Flag::OmpThreadprivate);
}
#endif
static void threadPrivatizeVars(lower::AbstractConverter &converter,
lower::pft::Evaluation &eval) {
fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
mlir::Location currentLocation = converter.getCurrentLocation();
mlir::OpBuilder::InsertionGuard guard(firOpBuilder);
firOpBuilder.setInsertionPointToStart(firOpBuilder.getAllocaBlock());
// If the symbol corresponds to the original ThreadprivateOp, use the symbol
// value from that operation to create one ThreadprivateOp copy operation
// inside the parallel region.
// In some cases, however, the symbol will correspond to the original,
// non-threadprivate variable. This can happen, for instance, with a common
// block, declared in a separate module, used by a parent procedure and
// privatized in its child procedure.
auto genThreadprivateOp = [&](lower::SymbolRef sym) -> mlir::Value {
assert(isThreadPrivate(sym));
mlir::Value symValue = converter.getSymbolAddress(sym);
mlir::Operation *op = symValue.getDefiningOp();
if (auto declOp = mlir::dyn_cast<hlfir::DeclareOp>(op))
op = declOp.getMemref().getDefiningOp();
if (mlir::isa<mlir::omp::ThreadprivateOp>(op))
symValue = mlir::dyn_cast<mlir::omp::ThreadprivateOp>(op).getSymAddr();
return firOpBuilder.create<mlir::omp::ThreadprivateOp>(
currentLocation, symValue.getType(), symValue);
};
llvm::SetVector<const semantics::Symbol *> threadprivateSyms;
converter.collectSymbolSet(eval, threadprivateSyms,
semantics::Symbol::Flag::OmpThreadprivate,
/*collectSymbols=*/true,
/*collectHostAssociatedSymbols=*/true);
std::set<semantics::SourceName> threadprivateSymNames;
// For a COMMON block, the ThreadprivateOp is generated for itself instead of
// its members, so only bind the value of the new copied ThreadprivateOp
// inside the parallel region to the common block symbol only once for
// multiple members in one COMMON block.
llvm::SetVector<const semantics::Symbol *> commonSyms;
for (std::size_t i = 0; i < threadprivateSyms.size(); i++) {
const semantics::Symbol *sym = threadprivateSyms[i];
mlir::Value symThreadprivateValue;
// The variable may be used more than once, and each reference has one
// symbol with the same name. Only do once for references of one variable.
if (threadprivateSymNames.find(sym->name()) != threadprivateSymNames.end())
continue;
threadprivateSymNames.insert(sym->name());
if (const semantics::Symbol *common =
semantics::FindCommonBlockContaining(sym->GetUltimate())) {
mlir::Value commonThreadprivateValue;
if (commonSyms.contains(common)) {
commonThreadprivateValue = converter.getSymbolAddress(*common);
} else {
commonThreadprivateValue = genThreadprivateOp(*common);
converter.bindSymbol(*common, commonThreadprivateValue);
commonSyms.insert(common);
}
symThreadprivateValue = lower::genCommonBlockMember(
converter, currentLocation, *sym, commonThreadprivateValue);
} else {
symThreadprivateValue = genThreadprivateOp(*sym);
}
fir::ExtendedValue sexv = converter.getSymbolExtendedValue(*sym);
fir::ExtendedValue symThreadprivateExv =
getExtendedValue(sexv, symThreadprivateValue);
converter.bindSymbol(*sym, symThreadprivateExv);
}
}
static mlir::Operation *
createAndSetPrivatizedLoopVar(lower::AbstractConverter &converter,
mlir::Location loc, mlir::Value indexVal,
const semantics::Symbol *sym) {
fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
mlir::OpBuilder::InsertPoint insPt = firOpBuilder.saveInsertionPoint();
firOpBuilder.setInsertionPointToStart(firOpBuilder.getAllocaBlock());
mlir::Type tempTy = converter.genType(*sym);
assert(converter.isPresentShallowLookup(*sym) &&
"Expected symbol to be in symbol table.");
firOpBuilder.restoreInsertionPoint(insPt);
mlir::Value cvtVal = firOpBuilder.createConvert(loc, tempTy, indexVal);
mlir::Operation *storeOp = firOpBuilder.create<fir::StoreOp>(
loc, cvtVal, converter.getSymbolAddress(*sym));
return storeOp;
}
// This helper function implements the functionality of "promoting" non-CPTR
// arguments of use_device_ptr to use_device_addr arguments (automagic
// conversion of use_device_ptr -> use_device_addr in these cases). The way we
// do so currently is through the shuffling of operands from the
// devicePtrOperands to deviceAddrOperands, as well as the types, locations and
// symbols.
//
// This effectively implements some deprecated OpenMP functionality that some
// legacy applications unfortunately depend on (deprecated in specification
// version 5.2):
//
// "If a list item in a use_device_ptr clause is not of type C_PTR, the behavior
// is as if the list item appeared in a use_device_addr clause. Support for
// such list items in a use_device_ptr clause is deprecated."
static void promoteNonCPtrUseDevicePtrArgsToUseDeviceAddr(
llvm::SmallVectorImpl<mlir::Value> &useDeviceAddrVars,
llvm::SmallVectorImpl<const semantics::Symbol *> &useDeviceAddrSyms,
llvm::SmallVectorImpl<mlir::Value> &useDevicePtrVars,
llvm::SmallVectorImpl<const semantics::Symbol *> &useDevicePtrSyms) {
// Iterate over our use_device_ptr list and shift all non-cptr arguments into
// use_device_addr.
auto *varIt = useDevicePtrVars.begin();
auto *symIt = useDevicePtrSyms.begin();
while (varIt != useDevicePtrVars.end()) {
if (fir::isa_builtin_cptr_type(fir::unwrapRefType(varIt->getType()))) {
++varIt;
++symIt;
continue;
}
useDeviceAddrVars.push_back(*varIt);
useDeviceAddrSyms.push_back(*symIt);
varIt = useDevicePtrVars.erase(varIt);
symIt = useDevicePtrSyms.erase(symIt);
}
}
/// Extract the list of function and variable symbols affected by the given
/// 'declare target' directive and return the intended device type for them.
static void getDeclareTargetInfo(
lower::AbstractConverter &converter, semantics::SemanticsContext &semaCtx,
lower::pft::Evaluation &eval,
const parser::OpenMPDeclareTargetConstruct &declareTargetConstruct,
mlir::omp::DeclareTargetOperands &clauseOps,
llvm::SmallVectorImpl<DeclareTargetCapturePair> &symbolAndClause) {
const auto &spec =
std::get<parser::OmpDeclareTargetSpecifier>(declareTargetConstruct.t);
if (const auto *objectList{parser::Unwrap<parser::OmpObjectList>(spec.u)}) {
ObjectList objects{makeObjects(*objectList, semaCtx)};
// Case: declare target(func, var1, var2)
gatherFuncAndVarSyms(objects, mlir::omp::DeclareTargetCaptureClause::to,
symbolAndClause);
} else if (const auto *clauseList{
parser::Unwrap<parser::OmpClauseList>(spec.u)}) {
List<Clause> clauses = makeClauses(*clauseList, semaCtx);
if (clauses.empty() &&
(!eval.getOwningProcedure()->isMainProgram() ||
eval.getOwningProcedure()->getMainProgramSymbol())) {
// Case: declare target, implicit capture of function
symbolAndClause.emplace_back(
mlir::omp::DeclareTargetCaptureClause::to,
eval.getOwningProcedure()->getSubprogramSymbol());
}
ClauseProcessor cp(converter, semaCtx, clauses);
cp.processDeviceType(clauseOps);
cp.processEnter(symbolAndClause);
cp.processLink(symbolAndClause);
cp.processTo(symbolAndClause);
cp.processTODO<clause::Indirect>(converter.getCurrentLocation(),
llvm::omp::Directive::OMPD_declare_target);
}
}
static void collectDeferredDeclareTargets(
lower::AbstractConverter &converter, semantics::SemanticsContext &semaCtx,
lower::pft::Evaluation &eval,
const parser::OpenMPDeclareTargetConstruct &declareTargetConstruct,
llvm::SmallVectorImpl<lower::OMPDeferredDeclareTargetInfo>
&deferredDeclareTarget) {
mlir::omp::DeclareTargetOperands clauseOps;
llvm::SmallVector<DeclareTargetCapturePair> symbolAndClause;
getDeclareTargetInfo(converter, semaCtx, eval, declareTargetConstruct,
clauseOps, symbolAndClause);
// Return the device type only if at least one of the targets for the
// directive is a function or subroutine
mlir::ModuleOp mod = converter.getFirOpBuilder().getModule();
for (const DeclareTargetCapturePair &symClause : symbolAndClause) {
mlir::Operation *op = mod.lookupSymbol(
converter.mangleName(std::get<const semantics::Symbol &>(symClause)));
if (!op) {
deferredDeclareTarget.push_back({std::get<0>(symClause),
clauseOps.deviceType,
std::get<1>(symClause)});
}
}
}
static std::optional<mlir::omp::DeclareTargetDeviceType>
getDeclareTargetFunctionDevice(
lower::AbstractConverter &converter, semantics::SemanticsContext &semaCtx,
lower::pft::Evaluation &eval,
const parser::OpenMPDeclareTargetConstruct &declareTargetConstruct) {
mlir::omp::DeclareTargetOperands clauseOps;
llvm::SmallVector<DeclareTargetCapturePair> symbolAndClause;
getDeclareTargetInfo(converter, semaCtx, eval, declareTargetConstruct,
clauseOps, symbolAndClause);
// Return the device type only if at least one of the targets for the
// directive is a function or subroutine
mlir::ModuleOp mod = converter.getFirOpBuilder().getModule();
for (const DeclareTargetCapturePair &symClause : symbolAndClause) {
mlir::Operation *op = mod.lookupSymbol(
converter.mangleName(std::get<const semantics::Symbol &>(symClause)));
if (mlir::isa_and_nonnull<mlir::func::FuncOp>(op))
return clauseOps.deviceType;
}
return std::nullopt;
}
/// Set up the entry block of the given `omp.loop_nest` operation, adding a
/// block argument for each loop induction variable and allocating and
/// initializing a private value to hold each of them.
///
/// This function can also bind the symbols of any variables that should match
/// block arguments on parent loop wrapper operations attached to the same
/// loop. This allows the introduction of any necessary `hlfir.declare`
/// operations inside of the entry block of the `omp.loop_nest` operation and
/// not directly under any of the wrappers, which would invalidate them.
///
/// \param [in] op - the loop nest operation.
/// \param [in] converter - PFT to MLIR conversion interface.
/// \param [in] loc - location.
/// \param [in] args - symbols of induction variables.
/// \param [in] wrapperArgs - list of parent loop wrappers and their associated
/// entry block arguments.
static void genLoopVars(
mlir::Operation *op, lower::AbstractConverter &converter,
mlir::Location &loc, llvm::ArrayRef<const semantics::Symbol *> args,
llvm::ArrayRef<
std::pair<mlir::omp::BlockArgOpenMPOpInterface, const EntryBlockArgs &>>
wrapperArgs = {}) {
fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
auto ®ion = op->getRegion(0);
std::size_t loopVarTypeSize = 0;
for (const semantics::Symbol *arg : args)
loopVarTypeSize = std::max(loopVarTypeSize, arg->GetUltimate().size());
mlir::Type loopVarType = getLoopVarType(converter, loopVarTypeSize);
llvm::SmallVector<mlir::Type> tiv(args.size(), loopVarType);
llvm::SmallVector<mlir::Location> locs(args.size(), loc);
firOpBuilder.createBlock(®ion, {}, tiv, locs);
// Bind the entry block arguments of parent wrappers to the corresponding
// symbols.
for (auto [argGeneratingOp, args] : wrapperArgs)
bindEntryBlockArgs(converter, argGeneratingOp, args);
// The argument is not currently in memory, so make a temporary for the
// argument, and store it there, then bind that location to the argument.
mlir::Operation *storeOp = nullptr;
for (auto [argIndex, argSymbol] : llvm::enumerate(args)) {
mlir::Value indexVal = fir::getBase(region.front().getArgument(argIndex));
storeOp =
createAndSetPrivatizedLoopVar(converter, loc, indexVal, argSymbol);
}
firOpBuilder.setInsertionPointAfter(storeOp);
}
/// Create an entry block for the given region, including the clause-defined
/// arguments specified.
///
/// \param [in] converter - PFT to MLIR conversion interface.
/// \param [in] args - entry block arguments information for the given
/// operation.
/// \param [in] region - Empty region in which to create the entry block.
static mlir::Block *genEntryBlock(lower::AbstractConverter &converter,
const EntryBlockArgs &args,
mlir::Region ®ion) {
assert(args.isValid() && "invalid args");
assert(region.empty() && "non-empty region");
fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
llvm::SmallVector<mlir::Type> types;
llvm::SmallVector<mlir::Location> locs;
unsigned numVars = args.inReduction.vars.size() + args.map.vars.size() +
args.priv.vars.size() + args.reduction.vars.size() +
args.taskReduction.vars.size() +
args.useDeviceAddr.vars.size();
types.reserve(numVars);
locs.reserve(numVars);
auto extractTypeLoc = [&types, &locs](llvm::ArrayRef<mlir::Value> vals) {
llvm::transform(vals, std::back_inserter(types),
[](mlir::Value v) { return v.getType(); });
llvm::transform(vals, std::back_inserter(locs),
[](mlir::Value v) { return v.getLoc(); });
};
// Populate block arguments in clause name alphabetical order to match
// expected order by the BlockArgOpenMPOpInterface.
extractTypeLoc(args.inReduction.vars);
extractTypeLoc(args.map.vars);
extractTypeLoc(args.priv.vars);
extractTypeLoc(args.reduction.vars);
extractTypeLoc(args.taskReduction.vars);
extractTypeLoc(args.useDeviceAddr.vars);
extractTypeLoc(args.useDevicePtr.vars);
return firOpBuilder.createBlock(®ion, {}, types, locs);
}
static void
markDeclareTarget(mlir::Operation *op, lower::AbstractConverter &converter,
mlir::omp::DeclareTargetCaptureClause captureClause,
mlir::omp::DeclareTargetDeviceType deviceType) {
// TODO: Add support for program local variables with declare target applied
auto declareTargetOp = llvm::dyn_cast<mlir::omp::DeclareTargetInterface>(op);
if (!declareTargetOp)
fir::emitFatalError(
converter.getCurrentLocation(),
"Attempt to apply declare target on unsupported operation");
// The function or global already has a declare target applied to it, very
// likely through implicit capture (usage in another declare target
// function/subroutine). It should be marked as any if it has been assigned
// both host and nohost, else we skip, as there is no change
if (declareTargetOp.isDeclareTarget()) {
if (declareTargetOp.getDeclareTargetDeviceType() != deviceType)
declareTargetOp.setDeclareTarget(mlir::omp::DeclareTargetDeviceType::any,
captureClause);
return;
}
declareTargetOp.setDeclareTarget(deviceType, captureClause);
}
//===----------------------------------------------------------------------===//
// Op body generation helper structures and functions
//===----------------------------------------------------------------------===//
struct OpWithBodyGenInfo {
/// A type for a code-gen callback function. This takes as argument the op for
/// which the code is being generated and returns the arguments of the op's
/// region.
using GenOMPRegionEntryCBFn =
std::function<llvm::SmallVector<const semantics::Symbol *>(
mlir::Operation *)>;
OpWithBodyGenInfo(lower::AbstractConverter &converter,
lower::SymMap &symTable,
semantics::SemanticsContext &semaCtx, mlir::Location loc,
lower::pft::Evaluation &eval, llvm::omp::Directive dir)
: converter(converter), symTable(symTable), semaCtx(semaCtx), loc(loc),
eval(eval), dir(dir) {}
OpWithBodyGenInfo &setClauses(const List<Clause> *value) {
clauses = value;
return *this;
}
OpWithBodyGenInfo &setDataSharingProcessor(DataSharingProcessor *value) {
dsp = value;
return *this;
}
OpWithBodyGenInfo &setGenRegionEntryCb(GenOMPRegionEntryCBFn value) {
genRegionEntryCB = value;
return *this;
}
OpWithBodyGenInfo &setGenSkeletonOnly(bool value) {
genSkeletonOnly = value;
return *this;
}
/// [inout] converter to use for the clauses.
lower::AbstractConverter &converter;
/// [in] Symbol table
lower::SymMap &symTable;
/// [in] Semantics context
semantics::SemanticsContext &semaCtx;
/// [in] location in source code.
mlir::Location loc;
/// [in] current PFT node/evaluation.
lower::pft::Evaluation &eval;
/// [in] leaf directive for which to generate the op body.
llvm::omp::Directive dir;
/// [in] list of clauses to process.
const List<Clause> *clauses = nullptr;
/// [in] if provided, processes the construct's data-sharing attributes.
DataSharingProcessor *dsp = nullptr;
/// [in] if provided, emits the op's region entry. Otherwise, an emtpy block
/// is created in the region.
GenOMPRegionEntryCBFn genRegionEntryCB = nullptr;
/// [in] if set to `true`, skip generating nested evaluations and dispatching
/// any further leaf constructs.
bool genSkeletonOnly = false;
};
/// Create the body (block) for an OpenMP Operation.
///
/// \param [in] op - the operation the body belongs to.
/// \param [in] info - options controlling code-gen for the construction.
/// \param [in] queue - work queue with nested constructs.
/// \param [in] item - item in the queue to generate body for.
static void createBodyOfOp(mlir::Operation &op, const OpWithBodyGenInfo &info,
const ConstructQueue &queue,
ConstructQueue::const_iterator item) {
fir::FirOpBuilder &firOpBuilder = info.converter.getFirOpBuilder();
auto insertMarker = [](fir::FirOpBuilder &builder) {
mlir::Value undef = builder.create<fir::UndefOp>(builder.getUnknownLoc(),
builder.getIndexType());
return undef.getDefiningOp();
};
// If an argument for the region is provided then create the block with that
// argument. Also update the symbol's address with the mlir argument value.
// e.g. For loops the argument is the induction variable. And all further
// uses of the induction variable should use this mlir value.
auto regionArgs = [&]() -> llvm::SmallVector<const semantics::Symbol *> {
if (info.genRegionEntryCB != nullptr) {
return info.genRegionEntryCB(&op);
}
firOpBuilder.createBlock(&op.getRegion(0));
return {};
}();
// Mark the earliest insertion point.
mlir::Operation *marker = insertMarker(firOpBuilder);
// If it is an unstructured region, create empty blocks for all evaluations.
if (lower::omp::isLastItemInQueue(item, queue) &&
info.eval.lowerAsUnstructured()) {
lower::createEmptyRegionBlocks<mlir::omp::TerminatorOp, mlir::omp::YieldOp>(
firOpBuilder, info.eval.getNestedEvaluations());
}
// Start with privatization, so that the lowering of the nested
// code will use the right symbols.
bool isLoop = llvm::omp::getDirectiveAssociation(info.dir) ==
llvm::omp::Association::Loop;
bool privatize = info.clauses;
firOpBuilder.setInsertionPoint(marker);
std::optional<DataSharingProcessor> tempDsp;
if (privatize && !info.dsp) {
tempDsp.emplace(info.converter, info.semaCtx, *info.clauses, info.eval,
Fortran::lower::omp::isLastItemInQueue(item, queue));
tempDsp->processStep1();
}
if (info.dir == llvm::omp::Directive::OMPD_parallel) {
threadPrivatizeVars(info.converter, info.eval);
if (info.clauses) {
firOpBuilder.setInsertionPoint(marker);
ClauseProcessor(info.converter, info.semaCtx, *info.clauses)
.processCopyin();
}
}
if (!info.genSkeletonOnly) {
if (ConstructQueue::const_iterator next = std::next(item);
next != queue.end()) {
genOMPDispatch(info.converter, info.symTable, info.semaCtx, info.eval,
info.loc, queue, next);
} else {
// genFIR(Evaluation&) tries to patch up unterminated blocks, causing
// a lot of complications for our approach if the terminator generation
// is delayed past this point. Insert a temporary terminator here, then
// delete it.
firOpBuilder.setInsertionPointToEnd(&op.getRegion(0).back());
auto *temp = lower::genOpenMPTerminator(firOpBuilder, &op, info.loc);
firOpBuilder.setInsertionPointAfter(marker);
genNestedEvaluations(info.converter, info.eval);
temp->erase();
}
}
// Get or create a unique exiting block from the given region, or
// return nullptr if there is no exiting block.
auto getUniqueExit = [&](mlir::Region ®ion) -> mlir::Block * {
// Find the blocks where the OMP terminator should go. In simple cases
// it is the single block in the operation's region. When the region
// is more complicated, especially with unstructured control flow, there
// may be multiple blocks, and some of them may have non-OMP terminators
// resulting from lowering of the code contained within the operation.
// All the remaining blocks are potential exit points from the op's region.
//
// Explicit control flow cannot exit any OpenMP region (other than via
// STOP), and that is enforced by semantic checks prior to lowering. STOP
// statements are lowered to a function call.
// Collect unterminated blocks.
llvm::SmallVector<mlir::Block *> exits;
for (mlir::Block &b : region) {
if (b.empty() || !b.back().hasTrait<mlir::OpTrait::IsTerminator>())
exits.push_back(&b);
}
if (exits.empty())
return nullptr;
// If there already is a unique exiting block, do not create another one.
// Additionally, some ops (e.g. omp.sections) require only 1 block in
// its region.
if (exits.size() == 1)
return exits[0];
mlir::Block *exit = firOpBuilder.createBlock(®ion);
for (mlir::Block *b : exits) {
firOpBuilder.setInsertionPointToEnd(b);
firOpBuilder.create<mlir::cf::BranchOp>(info.loc, exit);
}
return exit;
};
if (auto *exitBlock = getUniqueExit(op.getRegion(0))) {
firOpBuilder.setInsertionPointToEnd(exitBlock);
auto *term = lower::genOpenMPTerminator(firOpBuilder, &op, info.loc);
// Only insert lastprivate code when there actually is an exit block.
// Such a block may not exist if the nested code produced an infinite
// loop (this may not make sense in production code, but a user could
// write that and we should handle it).
firOpBuilder.setInsertionPoint(term);
if (privatize) {
// DataSharingProcessor::processStep2() may create operations before/after
// the one passed as argument. We need to treat loop wrappers and their
// nested loop as a unit, so we need to pass the top level wrapper (if
// present). Otherwise, these operations will be inserted within a
// wrapper region.
mlir::Operation *privatizationTopLevelOp = &op;
if (auto loopNest = llvm::dyn_cast<mlir::omp::LoopNestOp>(op)) {
llvm::SmallVector<mlir::omp::LoopWrapperInterface> wrappers;
loopNest.gatherWrappers(wrappers);
if (!wrappers.empty())
privatizationTopLevelOp = &*wrappers.back();
}
if (!info.dsp) {
assert(tempDsp.has_value());
tempDsp->processStep2(privatizationTopLevelOp, isLoop);
} else {
if (isLoop && regionArgs.size() > 0) {
for (const auto ®ionArg : regionArgs) {
info.dsp->pushLoopIV(info.converter.getSymbolAddress(*regionArg));
}
}
info.dsp->processStep2(privatizationTopLevelOp, isLoop);
}
}
}
firOpBuilder.setInsertionPointAfter(marker);
marker->erase();
}
static void genBodyOfTargetDataOp(
lower::AbstractConverter &converter, lower::SymMap &symTable,
semantics::SemanticsContext &semaCtx, lower::pft::Evaluation &eval,
mlir::omp::TargetDataOp &dataOp, const EntryBlockArgs &args,
const mlir::Location ¤tLocation, const ConstructQueue &queue,
ConstructQueue::const_iterator item) {
fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
genEntryBlock(converter, args, dataOp.getRegion());
bindEntryBlockArgs(converter, dataOp, args);
// Insert dummy instruction to remember the insertion position. The
// marker will be deleted by clean up passes since there are no uses.
// Remembering the position for further insertion is important since
// there are hlfir.declares inserted above while setting block arguments
// and new code from the body should be inserted after that.
mlir::Value undefMarker = firOpBuilder.create<fir::UndefOp>(
dataOp.getLoc(), firOpBuilder.getIndexType());
// Create blocks for unstructured regions. This has to be done since
// blocks are initially allocated with the function as the parent region.
if (eval.lowerAsUnstructured()) {
lower::createEmptyRegionBlocks<mlir::omp::TerminatorOp, mlir::omp::YieldOp>(
firOpBuilder, eval.getNestedEvaluations());
}
firOpBuilder.create<mlir::omp::TerminatorOp>(currentLocation);
// Set the insertion point after the marker.
firOpBuilder.setInsertionPointAfter(undefMarker.getDefiningOp());
if (ConstructQueue::const_iterator next = std::next(item);
next != queue.end()) {
genOMPDispatch(converter, symTable, semaCtx, eval, currentLocation, queue,
next);
} else {
genNestedEvaluations(converter, eval);
}
}
// This generates intermediate common block member accesses within a region
// and then rebinds the members symbol to the intermediate accessors we have
// generated so that subsequent code generation will utilise these instead.
//
// When the scope changes, the bindings to the intermediate accessors should
// be dropped in place of the original symbol bindings.
//
// This is for utilisation with TargetOp.
static void genIntermediateCommonBlockAccessors(
Fortran::lower::AbstractConverter &converter,
const mlir::Location ¤tLocation,
llvm::ArrayRef<mlir::BlockArgument> mapBlockArgs,
llvm::ArrayRef<const Fortran::semantics::Symbol *> mapSyms) {
// Iterate over the symbol list, which will be shorter than the list of
// arguments if new entry block arguments were introduced to implicitly map
// outside values used by the bounds cloned into the target region. In that
// case, the additional block arguments do not need processing here.
for (auto [mapSym, mapArg] : llvm::zip_first(mapSyms, mapBlockArgs)) {
auto *details = mapSym->detailsIf<Fortran::semantics::CommonBlockDetails>();
if (!details)
continue;
for (auto obj : details->objects()) {
auto targetCBMemberBind = Fortran::lower::genCommonBlockMember(
converter, currentLocation, *obj, mapArg);
fir::ExtendedValue sexv = converter.getSymbolExtendedValue(*obj);
fir::ExtendedValue targetCBExv =
getExtendedValue(sexv, targetCBMemberBind);
converter.bindSymbol(*obj, targetCBExv);
}
}
}
// This functions creates a block for the body of the targetOp's region. It adds
// all the symbols present in mapSymbols as block arguments to this block.
static void genBodyOfTargetOp(
lower::AbstractConverter &converter, lower::SymMap &symTable,
semantics::SemanticsContext &semaCtx, lower::pft::Evaluation &eval,
mlir::omp::TargetOp &targetOp, const EntryBlockArgs &args,
const mlir::Location ¤tLocation, const ConstructQueue &queue,
ConstructQueue::const_iterator item, DataSharingProcessor &dsp) {
fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
auto argIface = llvm::cast<mlir::omp::BlockArgOpenMPOpInterface>(*targetOp);
mlir::Region ®ion = targetOp.getRegion();
mlir::Block *entryBlock = genEntryBlock(converter, args, region);
bindEntryBlockArgs(converter, targetOp, args);
// Check if cloning the bounds introduced any dependency on the outer region.
// If so, then either clone them as well if they are MemoryEffectFree, or else
// copy them to a new temporary and add them to the map and block_argument
// lists and replace their uses with the new temporary.
llvm::SetVector<mlir::Value> valuesDefinedAbove;
mlir::getUsedValuesDefinedAbove(region, valuesDefinedAbove);
while (!valuesDefinedAbove.empty()) {
for (mlir::Value val : valuesDefinedAbove) {
mlir::Operation *valOp = val.getDefiningOp();
if (mlir::isMemoryEffectFree(valOp)) {
mlir::Operation *clonedOp = valOp->clone();
entryBlock->push_front(clonedOp);
val.replaceUsesWithIf(clonedOp->getResult(0),
[entryBlock](mlir::OpOperand &use) {
return use.getOwner()->getBlock() == entryBlock;
});
} else {
auto savedIP = firOpBuilder.getInsertionPoint();
firOpBuilder.setInsertionPointAfter(valOp);
auto copyVal =
firOpBuilder.createTemporary(val.getLoc(), val.getType());
firOpBuilder.createStoreWithConvert(copyVal.getLoc(), val, copyVal);
llvm::SmallVector<mlir::Value> bounds;
std::stringstream name;
firOpBuilder.setInsertionPoint(targetOp);
mlir::Value mapOp = createMapInfoOp(
firOpBuilder, copyVal.getLoc(), copyVal,
/*varPtrPtr=*/mlir::Value{}, name.str(), bounds,
/*members=*/llvm::SmallVector<mlir::Value>{},