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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
|
//===-- SelectionDAGISel.cpp - Implement the SelectionDAGISel class -------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This implements the SelectionDAGISel class.
//
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "isel"
#include "llvm/CodeGen/SelectionDAGISel.h"
#include "SelectionDAGBuild.h"
#include "llvm/ADT/BitVector.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Constants.h"
#include "llvm/CallingConv.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Function.h"
#include "llvm/GlobalVariable.h"
#include "llvm/InlineAsm.h"
#include "llvm/Instructions.h"
#include "llvm/Intrinsics.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/CodeGen/FastISel.h"
#include "llvm/CodeGen/GCStrategy.h"
#include "llvm/CodeGen/GCMetadata.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/ScheduleDAG.h"
#include "llvm/CodeGen/SchedulerRegistry.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetFrameInfo.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/Timer.h"
#include <algorithm>
using namespace llvm;
static cl::opt<bool>
EnableValueProp("enable-value-prop", cl::Hidden);
static cl::opt<bool>
EnableLegalizeTypes("enable-legalize-types", cl::Hidden);
static cl::opt<bool>
EnableFastISelVerbose("fast-isel-verbose", cl::Hidden,
cl::desc("Enable verbose messages in the experimental \"fast\" "
"instruction selector"));
static cl::opt<bool>
EnableFastISelAbort("fast-isel-abort", cl::Hidden,
cl::desc("Enable abort calls when \"fast\" instruction fails"));
static cl::opt<bool>
SchedLiveInCopies("schedule-livein-copies",
cl::desc("Schedule copies of livein registers"),
cl::init(false));
#ifndef NDEBUG
static cl::opt<bool>
ViewDAGCombine1("view-dag-combine1-dags", cl::Hidden,
cl::desc("Pop up a window to show dags before the first "
"dag combine pass"));
static cl::opt<bool>
ViewLegalizeTypesDAGs("view-legalize-types-dags", cl::Hidden,
cl::desc("Pop up a window to show dags before legalize types"));
static cl::opt<bool>
ViewLegalizeDAGs("view-legalize-dags", cl::Hidden,
cl::desc("Pop up a window to show dags before legalize"));
static cl::opt<bool>
ViewDAGCombine2("view-dag-combine2-dags", cl::Hidden,
cl::desc("Pop up a window to show dags before the second "
"dag combine pass"));
static cl::opt<bool>
ViewISelDAGs("view-isel-dags", cl::Hidden,
cl::desc("Pop up a window to show isel dags as they are selected"));
static cl::opt<bool>
ViewSchedDAGs("view-sched-dags", cl::Hidden,
cl::desc("Pop up a window to show sched dags as they are processed"));
static cl::opt<bool>
ViewSUnitDAGs("view-sunit-dags", cl::Hidden,
cl::desc("Pop up a window to show SUnit dags after they are processed"));
#else
static const bool ViewDAGCombine1 = false,
ViewLegalizeTypesDAGs = false, ViewLegalizeDAGs = false,
ViewDAGCombine2 = false,
ViewISelDAGs = false, ViewSchedDAGs = false,
ViewSUnitDAGs = false;
#endif
//===---------------------------------------------------------------------===//
///
/// RegisterScheduler class - Track the registration of instruction schedulers.
///
//===---------------------------------------------------------------------===//
MachinePassRegistry RegisterScheduler::Registry;
//===---------------------------------------------------------------------===//
///
/// ISHeuristic command line option for instruction schedulers.
///
//===---------------------------------------------------------------------===//
static cl::opt<RegisterScheduler::FunctionPassCtor, false,
RegisterPassParser<RegisterScheduler> >
ISHeuristic("pre-RA-sched",
cl::init(&createDefaultScheduler),
cl::desc("Instruction schedulers available (before register"
" allocation):"));
static RegisterScheduler
defaultListDAGScheduler("default", " Best scheduler for the target",
createDefaultScheduler);
namespace llvm {
//===--------------------------------------------------------------------===//
/// createDefaultScheduler - This creates an instruction scheduler appropriate
/// for the target.
ScheduleDAG* createDefaultScheduler(SelectionDAGISel *IS,
SelectionDAG *DAG,
MachineBasicBlock *BB,
bool Fast) {
TargetLowering &TLI = IS->getTargetLowering();
if (TLI.getSchedulingPreference() == TargetLowering::SchedulingForLatency) {
return createTDListDAGScheduler(IS, DAG, BB, Fast);
} else {
assert(TLI.getSchedulingPreference() ==
TargetLowering::SchedulingForRegPressure && "Unknown sched type!");
return createBURRListDAGScheduler(IS, DAG, BB, Fast);
}
}
}
// EmitInstrWithCustomInserter - This method should be implemented by targets
// that mark instructions with the 'usesCustomDAGSchedInserter' flag. These
// instructions are special in various ways, which require special support to
// insert. The specified MachineInstr is created but not inserted into any
// basic blocks, and the scheduler passes ownership of it to this method.
MachineBasicBlock *TargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
MachineBasicBlock *MBB) {
cerr << "If a target marks an instruction with "
<< "'usesCustomDAGSchedInserter', it must implement "
<< "TargetLowering::EmitInstrWithCustomInserter!\n";
abort();
return 0;
}
/// EmitLiveInCopy - Emit a copy for a live in physical register. If the
/// physical register has only a single copy use, then coalesced the copy
/// if possible.
static void EmitLiveInCopy(MachineBasicBlock *MBB,
MachineBasicBlock::iterator &InsertPos,
unsigned VirtReg, unsigned PhysReg,
const TargetRegisterClass *RC,
DenseMap<MachineInstr*, unsigned> &CopyRegMap,
const MachineRegisterInfo &MRI,
const TargetRegisterInfo &TRI,
const TargetInstrInfo &TII) {
unsigned NumUses = 0;
MachineInstr *UseMI = NULL;
for (MachineRegisterInfo::use_iterator UI = MRI.use_begin(VirtReg),
UE = MRI.use_end(); UI != UE; ++UI) {
UseMI = &*UI;
if (++NumUses > 1)
break;
}
// If the number of uses is not one, or the use is not a move instruction,
// don't coalesce. Also, only coalesce away a virtual register to virtual
// register copy.
bool Coalesced = false;
unsigned SrcReg, DstReg;
if (NumUses == 1 &&
TII.isMoveInstr(*UseMI, SrcReg, DstReg) &&
TargetRegisterInfo::isVirtualRegister(DstReg)) {
VirtReg = DstReg;
Coalesced = true;
}
// Now find an ideal location to insert the copy.
MachineBasicBlock::iterator Pos = InsertPos;
while (Pos != MBB->begin()) {
MachineInstr *PrevMI = prior(Pos);
DenseMap<MachineInstr*, unsigned>::iterator RI = CopyRegMap.find(PrevMI);
// copyRegToReg might emit multiple instructions to do a copy.
unsigned CopyDstReg = (RI == CopyRegMap.end()) ? 0 : RI->second;
if (CopyDstReg && !TRI.regsOverlap(CopyDstReg, PhysReg))
// This is what the BB looks like right now:
// r1024 = mov r0
// ...
// r1 = mov r1024
//
// We want to insert "r1025 = mov r1". Inserting this copy below the
// move to r1024 makes it impossible for that move to be coalesced.
//
// r1025 = mov r1
// r1024 = mov r0
// ...
// r1 = mov 1024
// r2 = mov 1025
break; // Woot! Found a good location.
--Pos;
}
TII.copyRegToReg(*MBB, Pos, VirtReg, PhysReg, RC, RC);
CopyRegMap.insert(std::make_pair(prior(Pos), VirtReg));
if (Coalesced) {
if (&*InsertPos == UseMI) ++InsertPos;
MBB->erase(UseMI);
}
}
/// EmitLiveInCopies - If this is the first basic block in the function,
/// and if it has live ins that need to be copied into vregs, emit the
/// copies into the block.
static void EmitLiveInCopies(MachineBasicBlock *EntryMBB,
const MachineRegisterInfo &MRI,
const TargetRegisterInfo &TRI,
const TargetInstrInfo &TII) {
if (SchedLiveInCopies) {
// Emit the copies at a heuristically-determined location in the block.
DenseMap<MachineInstr*, unsigned> CopyRegMap;
MachineBasicBlock::iterator InsertPos = EntryMBB->begin();
for (MachineRegisterInfo::livein_iterator LI = MRI.livein_begin(),
E = MRI.livein_end(); LI != E; ++LI)
if (LI->second) {
const TargetRegisterClass *RC = MRI.getRegClass(LI->second);
EmitLiveInCopy(EntryMBB, InsertPos, LI->second, LI->first,
RC, CopyRegMap, MRI, TRI, TII);
}
} else {
// Emit the copies into the top of the block.
for (MachineRegisterInfo::livein_iterator LI = MRI.livein_begin(),
E = MRI.livein_end(); LI != E; ++LI)
if (LI->second) {
const TargetRegisterClass *RC = MRI.getRegClass(LI->second);
TII.copyRegToReg(*EntryMBB, EntryMBB->begin(),
LI->second, LI->first, RC, RC);
}
}
}
//===----------------------------------------------------------------------===//
// SelectionDAGISel code
//===----------------------------------------------------------------------===//
SelectionDAGISel::SelectionDAGISel(TargetLowering &tli, bool fast) :
FunctionPass(&ID), TLI(tli),
FuncInfo(new FunctionLoweringInfo(TLI)),
CurDAG(new SelectionDAG(TLI, *FuncInfo)),
SDL(new SelectionDAGLowering(*CurDAG, TLI, *FuncInfo)),
GFI(),
Fast(fast),
DAGSize(0)
{}
SelectionDAGISel::~SelectionDAGISel() {
delete SDL;
delete CurDAG;
delete FuncInfo;
}
unsigned SelectionDAGISel::MakeReg(MVT VT) {
return RegInfo->createVirtualRegister(TLI.getRegClassFor(VT));
}
void SelectionDAGISel::getAnalysisUsage(AnalysisUsage &AU) const {
AU.addRequired<AliasAnalysis>();
AU.addRequired<GCModuleInfo>();
AU.setPreservesAll();
}
bool SelectionDAGISel::runOnFunction(Function &Fn) {
// Do some sanity-checking on the command-line options.
assert((!EnableFastISelVerbose || EnableFastISel) &&
"-fast-isel-verbose requires -fast-isel");
assert((!EnableFastISelAbort || EnableFastISel) &&
"-fast-isel-abort requires -fast-isel");
// Get alias analysis for load/store combining.
AA = &getAnalysis<AliasAnalysis>();
TargetMachine &TM = TLI.getTargetMachine();
MachineFunction &MF = MachineFunction::construct(&Fn, TM);
const MachineRegisterInfo &MRI = MF.getRegInfo();
const TargetInstrInfo &TII = *TM.getInstrInfo();
const TargetRegisterInfo &TRI = *TM.getRegisterInfo();
if (MF.getFunction()->hasGC())
GFI = &getAnalysis<GCModuleInfo>().getFunctionInfo(*MF.getFunction());
else
GFI = 0;
RegInfo = &MF.getRegInfo();
DOUT << "\n\n\n=== " << Fn.getName() << "\n";
FuncInfo->set(Fn, MF, EnableFastISel);
MachineModuleInfo *MMI = getAnalysisToUpdate<MachineModuleInfo>();
CurDAG->init(MF, MMI);
SDL->init(GFI, *AA);
for (Function::iterator I = Fn.begin(), E = Fn.end(); I != E; ++I)
if (InvokeInst *Invoke = dyn_cast<InvokeInst>(I->getTerminator()))
// Mark landing pad.
FuncInfo->MBBMap[Invoke->getSuccessor(1)]->setIsLandingPad();
SelectAllBasicBlocks(Fn, MF, MMI);
// If the first basic block in the function has live ins that need to be
// copied into vregs, emit the copies into the top of the block before
// emitting the code for the block.
EmitLiveInCopies(MF.begin(), MRI, TRI, TII);
// Add function live-ins to entry block live-in set.
for (MachineRegisterInfo::livein_iterator I = RegInfo->livein_begin(),
E = RegInfo->livein_end(); I != E; ++I)
MF.begin()->addLiveIn(I->first);
#ifndef NDEBUG
assert(FuncInfo->CatchInfoFound.size() == FuncInfo->CatchInfoLost.size() &&
"Not all catch info was assigned to a landing pad!");
#endif
FuncInfo->clear();
return true;
}
static void copyCatchInfo(BasicBlock *SrcBB, BasicBlock *DestBB,
MachineModuleInfo *MMI, FunctionLoweringInfo &FLI) {
for (BasicBlock::iterator I = SrcBB->begin(), E = --SrcBB->end(); I != E; ++I)
if (EHSelectorInst *EHSel = dyn_cast<EHSelectorInst>(I)) {
// Apply the catch info to DestBB.
AddCatchInfo(*EHSel, MMI, FLI.MBBMap[DestBB]);
#ifndef NDEBUG
if (!FLI.MBBMap[SrcBB]->isLandingPad())
FLI.CatchInfoFound.insert(EHSel);
#endif
}
}
/// IsFixedFrameObjectWithPosOffset - Check if object is a fixed frame object and
/// whether object offset >= 0.
static bool
IsFixedFrameObjectWithPosOffset(MachineFrameInfo * MFI, SDValue Op) {
if (!isa<FrameIndexSDNode>(Op)) return false;
FrameIndexSDNode * FrameIdxNode = dyn_cast<FrameIndexSDNode>(Op);
int FrameIdx = FrameIdxNode->getIndex();
return MFI->isFixedObjectIndex(FrameIdx) &&
MFI->getObjectOffset(FrameIdx) >= 0;
}
/// IsPossiblyOverwrittenArgumentOfTailCall - Check if the operand could
/// possibly be overwritten when lowering the outgoing arguments in a tail
/// call. Currently the implementation of this call is very conservative and
/// assumes all arguments sourcing from FORMAL_ARGUMENTS or a CopyFromReg with
/// virtual registers would be overwritten by direct lowering.
static bool IsPossiblyOverwrittenArgumentOfTailCall(SDValue Op,
MachineFrameInfo * MFI) {
RegisterSDNode * OpReg = NULL;
if (Op.getOpcode() == ISD::FORMAL_ARGUMENTS ||
(Op.getOpcode()== ISD::CopyFromReg &&
(OpReg = dyn_cast<RegisterSDNode>(Op.getOperand(1))) &&
(OpReg->getReg() >= TargetRegisterInfo::FirstVirtualRegister)) ||
(Op.getOpcode() == ISD::LOAD &&
IsFixedFrameObjectWithPosOffset(MFI, Op.getOperand(1))) ||
(Op.getOpcode() == ISD::MERGE_VALUES &&
Op.getOperand(Op.getResNo()).getOpcode() == ISD::LOAD &&
IsFixedFrameObjectWithPosOffset(MFI, Op.getOperand(Op.getResNo()).
getOperand(1))))
return true;
return false;
}
/// CheckDAGForTailCallsAndFixThem - This Function looks for CALL nodes in the
/// DAG and fixes their tailcall attribute operand.
static void CheckDAGForTailCallsAndFixThem(SelectionDAG &DAG,
TargetLowering& TLI) {
SDNode * Ret = NULL;
SDValue Terminator = DAG.getRoot();
// Find RET node.
if (Terminator.getOpcode() == ISD::RET) {
Ret = Terminator.getNode();
}
// Fix tail call attribute of CALL nodes.
for (SelectionDAG::allnodes_iterator BE = DAG.allnodes_begin(),
BI = DAG.allnodes_end(); BI != BE; ) {
--BI;
if (CallSDNode *TheCall = dyn_cast<CallSDNode>(BI)) {
SDValue OpRet(Ret, 0);
SDValue OpCall(BI, 0);
bool isMarkedTailCall = TheCall->isTailCall();
// If CALL node has tail call attribute set to true and the call is not
// eligible (no RET or the target rejects) the attribute is fixed to
// false. The TargetLowering::IsEligibleForTailCallOptimization function
// must correctly identify tail call optimizable calls.
if (!isMarkedTailCall) continue;
if (Ret==NULL ||
!TLI.IsEligibleForTailCallOptimization(TheCall, OpRet, DAG)) {
// Not eligible. Mark CALL node as non tail call. Note that we
// can modify the call node in place since calls are not CSE'd.
TheCall->setNotTailCall();
} else {
// Look for tail call clobbered arguments. Emit a series of
// copyto/copyfrom virtual register nodes to protect them.
SmallVector<SDValue, 32> Ops;
SDValue Chain = TheCall->getChain(), InFlag;
Ops.push_back(Chain);
Ops.push_back(TheCall->getCallee());
for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; ++i) {
SDValue Arg = TheCall->getArg(i);
bool isByVal = TheCall->getArgFlags(i).isByVal();
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
if (!isByVal &&
IsPossiblyOverwrittenArgumentOfTailCall(Arg, MFI)) {
MVT VT = Arg.getValueType();
unsigned VReg = MF.getRegInfo().
createVirtualRegister(TLI.getRegClassFor(VT));
Chain = DAG.getCopyToReg(Chain, VReg, Arg, InFlag);
InFlag = Chain.getValue(1);
Arg = DAG.getCopyFromReg(Chain, VReg, VT, InFlag);
Chain = Arg.getValue(1);
InFlag = Arg.getValue(2);
}
Ops.push_back(Arg);
Ops.push_back(TheCall->getArgFlagsVal(i));
}
// Link in chain of CopyTo/CopyFromReg.
Ops[0] = Chain;
DAG.UpdateNodeOperands(OpCall, Ops.begin(), Ops.size());
}
}
}
}
void SelectionDAGISel::SelectBasicBlock(BasicBlock *LLVMBB,
BasicBlock::iterator Begin,
BasicBlock::iterator End) {
SDL->setCurrentBasicBlock(BB);
MachineModuleInfo *MMI = CurDAG->getMachineModuleInfo();
if (MMI && BB->isLandingPad()) {
// Add a label to mark the beginning of the landing pad. Deletion of the
// landing pad can thus be detected via the MachineModuleInfo.
unsigned LabelID = MMI->addLandingPad(BB);
CurDAG->setRoot(CurDAG->getLabel(ISD::EH_LABEL,
CurDAG->getEntryNode(), LabelID));
// Mark exception register as live in.
unsigned Reg = TLI.getExceptionAddressRegister();
if (Reg) BB->addLiveIn(Reg);
// Mark exception selector register as live in.
Reg = TLI.getExceptionSelectorRegister();
if (Reg) BB->addLiveIn(Reg);
// FIXME: Hack around an exception handling flaw (PR1508): the personality
// function and list of typeids logically belong to the invoke (or, if you
// like, the basic block containing the invoke), and need to be associated
// with it in the dwarf exception handling tables. Currently however the
// information is provided by an intrinsic (eh.selector) that can be moved
// to unexpected places by the optimizers: if the unwind edge is critical,
// then breaking it can result in the intrinsics being in the successor of
// the landing pad, not the landing pad itself. This results in exceptions
// not being caught because no typeids are associated with the invoke.
// This may not be the only way things can go wrong, but it is the only way
// we try to work around for the moment.
BranchInst *Br = dyn_cast<BranchInst>(LLVMBB->getTerminator());
if (Br && Br->isUnconditional()) { // Critical edge?
BasicBlock::iterator I, E;
for (I = LLVMBB->begin(), E = --LLVMBB->end(); I != E; ++I)
if (isa<EHSelectorInst>(I))
break;
if (I == E)
// No catch info found - try to extract some from the successor.
copyCatchInfo(Br->getSuccessor(0), LLVMBB, MMI, *FuncInfo);
}
}
// Lower all of the non-terminator instructions.
for (BasicBlock::iterator I = Begin; I != End; ++I)
if (!isa<TerminatorInst>(I))
SDL->visit(*I);
// Ensure that all instructions which are used outside of their defining
// blocks are available as virtual registers. Invoke is handled elsewhere.
for (BasicBlock::iterator I = Begin; I != End; ++I)
if (!I->use_empty() && !isa<PHINode>(I) && !isa<InvokeInst>(I)) {
DenseMap<const Value*,unsigned>::iterator VMI =FuncInfo->ValueMap.find(I);
if (VMI != FuncInfo->ValueMap.end())
SDL->CopyValueToVirtualRegister(I, VMI->second);
}
// Handle PHI nodes in successor blocks.
if (End == LLVMBB->end()) {
HandlePHINodesInSuccessorBlocks(LLVMBB);
// Lower the terminator after the copies are emitted.
SDL->visit(*LLVMBB->getTerminator());
}
// Make sure the root of the DAG is up-to-date.
CurDAG->setRoot(SDL->getControlRoot());
// Check whether calls in this block are real tail calls. Fix up CALL nodes
// with correct tailcall attribute so that the target can rely on the tailcall
// attribute indicating whether the call is really eligible for tail call
// optimization.
if (PerformTailCallOpt)
CheckDAGForTailCallsAndFixThem(*CurDAG, TLI);
// Final step, emit the lowered DAG as machine code.
CodeGenAndEmitDAG();
SDL->clear();
}
void SelectionDAGISel::ComputeLiveOutVRegInfo() {
SmallPtrSet<SDNode*, 128> VisitedNodes;
SmallVector<SDNode*, 128> Worklist;
Worklist.push_back(CurDAG->getRoot().getNode());
APInt Mask;
APInt KnownZero;
APInt KnownOne;
while (!Worklist.empty()) {
SDNode *N = Worklist.back();
Worklist.pop_back();
// If we've already seen this node, ignore it.
if (!VisitedNodes.insert(N))
continue;
// Otherwise, add all chain operands to the worklist.
for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
if (N->getOperand(i).getValueType() == MVT::Other)
Worklist.push_back(N->getOperand(i).getNode());
// If this is a CopyToReg with a vreg dest, process it.
if (N->getOpcode() != ISD::CopyToReg)
continue;
unsigned DestReg = cast<RegisterSDNode>(N->getOperand(1))->getReg();
if (!TargetRegisterInfo::isVirtualRegister(DestReg))
continue;
// Ignore non-scalar or non-integer values.
SDValue Src = N->getOperand(2);
MVT SrcVT = Src.getValueType();
if (!SrcVT.isInteger() || SrcVT.isVector())
continue;
unsigned NumSignBits = CurDAG->ComputeNumSignBits(Src);
Mask = APInt::getAllOnesValue(SrcVT.getSizeInBits());
CurDAG->ComputeMaskedBits(Src, Mask, KnownZero, KnownOne);
// Only install this information if it tells us something.
if (NumSignBits != 1 || KnownZero != 0 || KnownOne != 0) {
DestReg -= TargetRegisterInfo::FirstVirtualRegister;
FunctionLoweringInfo &FLI = CurDAG->getFunctionLoweringInfo();
if (DestReg >= FLI.LiveOutRegInfo.size())
FLI.LiveOutRegInfo.resize(DestReg+1);
FunctionLoweringInfo::LiveOutInfo &LOI = FLI.LiveOutRegInfo[DestReg];
LOI.NumSignBits = NumSignBits;
LOI.KnownOne = NumSignBits;
LOI.KnownZero = NumSignBits;
}
}
}
void SelectionDAGISel::CodeGenAndEmitDAG() {
std::string GroupName;
if (TimePassesIsEnabled)
GroupName = "Instruction Selection and Scheduling";
std::string BlockName;
if (ViewDAGCombine1 || ViewLegalizeTypesDAGs || ViewLegalizeDAGs ||
ViewDAGCombine2 || ViewISelDAGs || ViewSchedDAGs || ViewSUnitDAGs)
BlockName = CurDAG->getMachineFunction().getFunction()->getName() + ':' +
BB->getBasicBlock()->getName();
DOUT << "Initial selection DAG:\n";
DEBUG(CurDAG->dump());
if (ViewDAGCombine1) CurDAG->viewGraph("dag-combine1 input for " + BlockName);
// Run the DAG combiner in pre-legalize mode.
if (TimePassesIsEnabled) {
NamedRegionTimer T("DAG Combining 1", GroupName);
CurDAG->Combine(false, *AA, Fast);
} else {
CurDAG->Combine(false, *AA, Fast);
}
DOUT << "Optimized lowered selection DAG:\n";
DEBUG(CurDAG->dump());
// Second step, hack on the DAG until it only uses operations and types that
// the target supports.
if (EnableLegalizeTypes) {// Enable this some day.
if (ViewLegalizeTypesDAGs) CurDAG->viewGraph("legalize-types input for " +
BlockName);
if (TimePassesIsEnabled) {
NamedRegionTimer T("Type Legalization", GroupName);
CurDAG->LegalizeTypes();
} else {
CurDAG->LegalizeTypes();
}
DOUT << "Type-legalized selection DAG:\n";
DEBUG(CurDAG->dump());
// TODO: enable a dag combine pass here.
}
if (ViewLegalizeDAGs) CurDAG->viewGraph("legalize input for " + BlockName);
if (TimePassesIsEnabled) {
NamedRegionTimer T("DAG Legalization", GroupName);
CurDAG->Legalize();
} else {
CurDAG->Legalize();
}
DOUT << "Legalized selection DAG:\n";
DEBUG(CurDAG->dump());
if (ViewDAGCombine2) CurDAG->viewGraph("dag-combine2 input for " + BlockName);
// Run the DAG combiner in post-legalize mode.
if (TimePassesIsEnabled) {
NamedRegionTimer T("DAG Combining 2", GroupName);
CurDAG->Combine(true, *AA, Fast);
} else {
CurDAG->Combine(true, *AA, Fast);
}
DOUT << "Optimized legalized selection DAG:\n";
DEBUG(CurDAG->dump());
if (ViewISelDAGs) CurDAG->viewGraph("isel input for " + BlockName);
if (!Fast && EnableValueProp)
ComputeLiveOutVRegInfo();
// Third, instruction select all of the operations to machine code, adding the
// code to the MachineBasicBlock.
if (TimePassesIsEnabled) {
NamedRegionTimer T("Instruction Selection", GroupName);
InstructionSelect();
} else {
InstructionSelect();
}
DOUT << "Selected selection DAG:\n";
DEBUG(CurDAG->dump());
if (ViewSchedDAGs) CurDAG->viewGraph("scheduler input for " + BlockName);
// Schedule machine code.
ScheduleDAG *Scheduler;
if (TimePassesIsEnabled) {
NamedRegionTimer T("Instruction Scheduling", GroupName);
Scheduler = Schedule();
} else {
Scheduler = Schedule();
}
if (ViewSUnitDAGs) Scheduler->viewGraph();
// Emit machine code to BB. This can change 'BB' to the last block being
// inserted into.
if (TimePassesIsEnabled) {
NamedRegionTimer T("Instruction Creation", GroupName);
BB = Scheduler->EmitSchedule();
} else {
BB = Scheduler->EmitSchedule();
}
// Free the scheduler state.
if (TimePassesIsEnabled) {
NamedRegionTimer T("Instruction Scheduling Cleanup", GroupName);
delete Scheduler;
} else {
delete Scheduler;
}
DOUT << "Selected machine code:\n";
DEBUG(BB->dump());
}
void SelectionDAGISel::SelectAllBasicBlocks(Function &Fn, MachineFunction &MF,
MachineModuleInfo *MMI) {
// Initialize the Fast-ISel state, if needed.
FastISel *FastIS = 0;
if (EnableFastISel)
FastIS = TLI.createFastISel(*FuncInfo->MF, MMI,
FuncInfo->ValueMap,
FuncInfo->MBBMap,
FuncInfo->StaticAllocaMap);
// Iterate over all basic blocks in the function.
for (Function::iterator I = Fn.begin(), E = Fn.end(); I != E; ++I) {
BasicBlock *LLVMBB = &*I;
BB = FuncInfo->MBBMap[LLVMBB];
BasicBlock::iterator const Begin = LLVMBB->begin();
BasicBlock::iterator const End = LLVMBB->end();
BasicBlock::iterator BI = Begin;
// Lower any arguments needed in this block if this is the entry block.
bool SuppressFastISel = false;
if (LLVMBB == &Fn.getEntryBlock()) {
LowerArguments(LLVMBB);
// If any of the arguments has the byval attribute, forgo
// fast-isel in the entry block.
if (FastIS) {
unsigned j = 1;
for (Function::arg_iterator I = Fn.arg_begin(), E = Fn.arg_end();
I != E; ++I, ++j)
if (Fn.paramHasAttr(j, Attribute::ByVal)) {
if (EnableFastISelVerbose || EnableFastISelAbort)
cerr << "FastISel skips entry block due to byval argument\n";
SuppressFastISel = true;
break;
}
}
}
// Before doing SelectionDAG ISel, see if FastISel has been requested.
// FastISel doesn't support EH landing pads, which require special handling.
if (FastIS && !SuppressFastISel && !BB->isLandingPad()) {
// Emit code for any incoming arguments. This must happen before
// beginning FastISel on the entry block.
if (LLVMBB == &Fn.getEntryBlock()) {
CurDAG->setRoot(SDL->getControlRoot());
CodeGenAndEmitDAG();
SDL->clear();
}
FastIS->startNewBlock(BB);
// Do FastISel on as many instructions as possible.
for (; BI != End; ++BI) {
// Just before the terminator instruction, insert instructions to
// feed PHI nodes in successor blocks.
if (isa<TerminatorInst>(BI))
if (!HandlePHINodesInSuccessorBlocksFast(LLVMBB, FastIS)) {
if (EnableFastISelVerbose || EnableFastISelAbort) {
cerr << "FastISel miss: ";
BI->dump();
}
if (EnableFastISelAbort)
assert(0 && "FastISel didn't handle a PHI in a successor");
break;
}
// First try normal tablegen-generated "fast" selection.
if (FastIS->SelectInstruction(BI))
continue;
// Next, try calling the target to attempt to handle the instruction.
if (FastIS->TargetSelectInstruction(BI))
continue;
// Then handle certain instructions as single-LLVM-Instruction blocks.
if (isa<CallInst>(BI)) {
if (EnableFastISelVerbose || EnableFastISelAbort) {
cerr << "FastISel missed call: ";
BI->dump();
}
if (BI->getType() != Type::VoidTy) {
unsigned &R = FuncInfo->ValueMap[BI];
if (!R)
R = FuncInfo->CreateRegForValue(BI);
}
SelectBasicBlock(LLVMBB, BI, next(BI));
// If the instruction was codegen'd with multiple blocks,
// inform the FastISel object where to resume inserting.
FastIS->setCurrentBlock(BB);
continue;
}
// Otherwise, give up on FastISel for the rest of the block.
// For now, be a little lenient about non-branch terminators.
if (!isa<TerminatorInst>(BI) || isa<BranchInst>(BI)) {
if (EnableFastISelVerbose || EnableFastISelAbort) {
cerr << "FastISel miss: ";
BI->dump();
}
if (EnableFastISelAbort)
// The "fast" selector couldn't handle something and bailed.
// For the purpose of debugging, just abort.
assert(0 && "FastISel didn't select the entire block");
}
break;
}
}
// Run SelectionDAG instruction selection on the remainder of the block
// not handled by FastISel. If FastISel is not run, this is the entire
// block.
if (BI != End)
SelectBasicBlock(LLVMBB, BI, End);
FinishBasicBlock();
}
delete FastIS;
}
void
SelectionDAGISel::FinishBasicBlock() {
// Perform target specific isel post processing.
InstructionSelectPostProcessing();
DOUT << "Target-post-processed machine code:\n";
DEBUG(BB->dump());
DOUT << "Total amount of phi nodes to update: "
<< SDL->PHINodesToUpdate.size() << "\n";
DEBUG(for (unsigned i = 0, e = SDL->PHINodesToUpdate.size(); i != e; ++i)
DOUT << "Node " << i << " : (" << SDL->PHINodesToUpdate[i].first
<< ", " << SDL->PHINodesToUpdate[i].second << ")\n";);
// Next, now that we know what the last MBB the LLVM BB expanded is, update
// PHI nodes in successors.
if (SDL->SwitchCases.empty() &&
SDL->JTCases.empty() &&
SDL->BitTestCases.empty()) {
for (unsigned i = 0, e = SDL->PHINodesToUpdate.size(); i != e; ++i) {
MachineInstr *PHI = SDL->PHINodesToUpdate[i].first;
assert(PHI->getOpcode() == TargetInstrInfo::PHI &&
"This is not a machine PHI node that we are updating!");
PHI->addOperand(MachineOperand::CreateReg(SDL->PHINodesToUpdate[i].second,
false));
PHI->addOperand(MachineOperand::CreateMBB(BB));
}
SDL->PHINodesToUpdate.clear();
return;
}
for (unsigned i = 0, e = SDL->BitTestCases.size(); i != e; ++i) {
// Lower header first, if it wasn't already lowered
if (!SDL->BitTestCases[i].Emitted) {
// Set the current basic block to the mbb we wish to insert the code into
BB = SDL->BitTestCases[i].Parent;
SDL->setCurrentBasicBlock(BB);
// Emit the code
SDL->visitBitTestHeader(SDL->BitTestCases[i]);
CurDAG->setRoot(SDL->getRoot());
CodeGenAndEmitDAG();
SDL->clear();
}
for (unsigned j = 0, ej = SDL->BitTestCases[i].Cases.size(); j != ej; ++j) {
// Set the current basic block to the mbb we wish to insert the code into
BB = SDL->BitTestCases[i].Cases[j].ThisBB;
SDL->setCurrentBasicBlock(BB);
// Emit the code
if (j+1 != ej)
SDL->visitBitTestCase(SDL->BitTestCases[i].Cases[j+1].ThisBB,
SDL->BitTestCases[i].Reg,
SDL->BitTestCases[i].Cases[j]);
else
SDL->visitBitTestCase(SDL->BitTestCases[i].Default,
SDL->BitTestCases[i].Reg,
SDL->BitTestCases[i].Cases[j]);
CurDAG->setRoot(SDL->getRoot());
CodeGenAndEmitDAG();
SDL->clear();
}
// Update PHI Nodes
for (unsigned pi = 0, pe = SDL->PHINodesToUpdate.size(); pi != pe; ++pi) {
MachineInstr *PHI = SDL->PHINodesToUpdate[pi].first;
MachineBasicBlock *PHIBB = PHI->getParent();
assert(PHI->getOpcode() == TargetInstrInfo::PHI &&
"This is not a machine PHI node that we are updating!");
// This is "default" BB. We have two jumps to it. From "header" BB and
// from last "case" BB.
if (PHIBB == SDL->BitTestCases[i].Default) {
PHI->addOperand(MachineOperand::CreateReg(SDL->PHINodesToUpdate[pi].second,
false));
PHI->addOperand(MachineOperand::CreateMBB(SDL->BitTestCases[i].Parent));
PHI->addOperand(MachineOperand::CreateReg(SDL->PHINodesToUpdate[pi].second,
false));
PHI->addOperand(MachineOperand::CreateMBB(SDL->BitTestCases[i].Cases.
back().ThisBB));
}
// One of "cases" BB.
for (unsigned j = 0, ej = SDL->BitTestCases[i].Cases.size();
j != ej; ++j) {
MachineBasicBlock* cBB = SDL->BitTestCases[i].Cases[j].ThisBB;
if (cBB->succ_end() !=
std::find(cBB->succ_begin(),cBB->succ_end(), PHIBB)) {
PHI->addOperand(MachineOperand::CreateReg(SDL->PHINodesToUpdate[pi].second,
false));
PHI->addOperand(MachineOperand::CreateMBB(cBB));
}
}
}
}
SDL->BitTestCases.clear();
// If the JumpTable record is filled in, then we need to emit a jump table.
// Updating the PHI nodes is tricky in this case, since we need to determine
// whether the PHI is a successor of the range check MBB or the jump table MBB
for (unsigned i = 0, e = SDL->JTCases.size(); i != e; ++i) {
// Lower header first, if it wasn't already lowered
if (!SDL->JTCases[i].first.Emitted) {
// Set the current basic block to the mbb we wish to insert the code into
BB = SDL->JTCases[i].first.HeaderBB;
SDL->setCurrentBasicBlock(BB);
// Emit the code
SDL->visitJumpTableHeader(SDL->JTCases[i].second, SDL->JTCases[i].first);
CurDAG->setRoot(SDL->getRoot());
CodeGenAndEmitDAG();
SDL->clear();
}
// Set the current basic block to the mbb we wish to insert the code into
BB = SDL->JTCases[i].second.MBB;
SDL->setCurrentBasicBlock(BB);
// Emit the code
SDL->visitJumpTable(SDL->JTCases[i].second);
CurDAG->setRoot(SDL->getRoot());
CodeGenAndEmitDAG();
SDL->clear();
// Update PHI Nodes
for (unsigned pi = 0, pe = SDL->PHINodesToUpdate.size(); pi != pe; ++pi) {
MachineInstr *PHI = SDL->PHINodesToUpdate[pi].first;
MachineBasicBlock *PHIBB = PHI->getParent();
assert(PHI->getOpcode() == TargetInstrInfo::PHI &&
"This is not a machine PHI node that we are updating!");
// "default" BB. We can go there only from header BB.
if (PHIBB == SDL->JTCases[i].second.Default) {
PHI->addOperand(MachineOperand::CreateReg(SDL->PHINodesToUpdate[pi].second,
false));
PHI->addOperand(MachineOperand::CreateMBB(SDL->JTCases[i].first.HeaderBB));
}
// JT BB. Just iterate over successors here
if (BB->succ_end() != std::find(BB->succ_begin(),BB->succ_end(), PHIBB)) {
PHI->addOperand(MachineOperand::CreateReg(SDL->PHINodesToUpdate[pi].second,
false));
PHI->addOperand(MachineOperand::CreateMBB(BB));
}
}
}
SDL->JTCases.clear();
// If the switch block involved a branch to one of the actual successors, we
// need to update PHI nodes in that block.
for (unsigned i = 0, e = SDL->PHINodesToUpdate.size(); i != e; ++i) {
MachineInstr *PHI = SDL->PHINodesToUpdate[i].first;
assert(PHI->getOpcode() == TargetInstrInfo::PHI &&
"This is not a machine PHI node that we are updating!");
if (BB->isSuccessor(PHI->getParent())) {
PHI->addOperand(MachineOperand::CreateReg(SDL->PHINodesToUpdate[i].second,
false));
PHI->addOperand(MachineOperand::CreateMBB(BB));
}
}
// If we generated any switch lowering information, build and codegen any
// additional DAGs necessary.
for (unsigned i = 0, e = SDL->SwitchCases.size(); i != e; ++i) {
// Set the current basic block to the mbb we wish to insert the code into
BB = SDL->SwitchCases[i].ThisBB;
SDL->setCurrentBasicBlock(BB);
// Emit the code
SDL->visitSwitchCase(SDL->SwitchCases[i]);
CurDAG->setRoot(SDL->getRoot());
CodeGenAndEmitDAG();
SDL->clear();
// Handle any PHI nodes in successors of this chunk, as if we were coming
// from the original BB before switch expansion. Note that PHI nodes can
// occur multiple times in PHINodesToUpdate. We have to be very careful to
// handle them the right number of times.
while ((BB = SDL->SwitchCases[i].TrueBB)) { // Handle LHS and RHS.
for (MachineBasicBlock::iterator Phi = BB->begin();
Phi != BB->end() && Phi->getOpcode() == TargetInstrInfo::PHI; ++Phi){
// This value for this PHI node is recorded in PHINodesToUpdate, get it.
for (unsigned pn = 0; ; ++pn) {
assert(pn != SDL->PHINodesToUpdate.size() &&
"Didn't find PHI entry!");
if (SDL->PHINodesToUpdate[pn].first == Phi) {
Phi->addOperand(MachineOperand::CreateReg(SDL->PHINodesToUpdate[pn].
second, false));
Phi->addOperand(MachineOperand::CreateMBB(SDL->SwitchCases[i].ThisBB));
break;
}
}
}
// Don't process RHS if same block as LHS.
if (BB == SDL->SwitchCases[i].FalseBB)
SDL->SwitchCases[i].FalseBB = 0;
// If we haven't handled the RHS, do so now. Otherwise, we're done.
SDL->SwitchCases[i].TrueBB = SDL->SwitchCases[i].FalseBB;
SDL->SwitchCases[i].FalseBB = 0;
}
assert(SDL->SwitchCases[i].TrueBB == 0 && SDL->SwitchCases[i].FalseBB == 0);
}
SDL->SwitchCases.clear();
SDL->PHINodesToUpdate.clear();
}
/// Schedule - Pick a safe ordering for instructions for each
/// target node in the graph.
///
ScheduleDAG *SelectionDAGISel::Schedule() {
RegisterScheduler::FunctionPassCtor Ctor = RegisterScheduler::getDefault();
if (!Ctor) {
Ctor = ISHeuristic;
RegisterScheduler::setDefault(Ctor);
}
ScheduleDAG *Scheduler = Ctor(this, CurDAG, BB, Fast);
Scheduler->Run();
return Scheduler;
}
HazardRecognizer *SelectionDAGISel::CreateTargetHazardRecognizer() {
return new HazardRecognizer();
}
//===----------------------------------------------------------------------===//
// Helper functions used by the generated instruction selector.
//===----------------------------------------------------------------------===//
// Calls to these methods are generated by tblgen.
/// CheckAndMask - The isel is trying to match something like (and X, 255). If
/// the dag combiner simplified the 255, we still want to match. RHS is the
/// actual value in the DAG on the RHS of an AND, and DesiredMaskS is the value
/// specified in the .td file (e.g. 255).
bool SelectionDAGISel::CheckAndMask(SDValue LHS, ConstantSDNode *RHS,
int64_t DesiredMaskS) const {
const APInt &ActualMask = RHS->getAPIntValue();
const APInt &DesiredMask = APInt(LHS.getValueSizeInBits(), DesiredMaskS);
// If the actual mask exactly matches, success!
if (ActualMask == DesiredMask)
return true;
// If the actual AND mask is allowing unallowed bits, this doesn't match.
if (ActualMask.intersects(~DesiredMask))
return false;
// Otherwise, the DAG Combiner may have proven that the value coming in is
// either already zero or is not demanded. Check for known zero input bits.
APInt NeededMask = DesiredMask & ~ActualMask;
if (CurDAG->MaskedValueIsZero(LHS, NeededMask))
return true;
// TODO: check to see if missing bits are just not demanded.
// Otherwise, this pattern doesn't match.
return false;
}
/// CheckOrMask - The isel is trying to match something like (or X, 255). If
/// the dag combiner simplified the 255, we still want to match. RHS is the
/// actual value in the DAG on the RHS of an OR, and DesiredMaskS is the value
/// specified in the .td file (e.g. 255).
bool SelectionDAGISel::CheckOrMask(SDValue LHS, ConstantSDNode *RHS,
int64_t DesiredMaskS) const {
const APInt &ActualMask = RHS->getAPIntValue();
const APInt &DesiredMask = APInt(LHS.getValueSizeInBits(), DesiredMaskS);
// If the actual mask exactly matches, success!
if (ActualMask == DesiredMask)
return true;
// If the actual AND mask is allowing unallowed bits, this doesn't match.
if (ActualMask.intersects(~DesiredMask))
return false;
// Otherwise, the DAG Combiner may have proven that the value coming in is
// either already zero or is not demanded. Check for known zero input bits.
APInt NeededMask = DesiredMask & ~ActualMask;
APInt KnownZero, KnownOne;
CurDAG->ComputeMaskedBits(LHS, NeededMask, KnownZero, KnownOne);
// If all the missing bits in the or are already known to be set, match!
if ((NeededMask & KnownOne) == NeededMask)
return true;
// TODO: check to see if missing bits are just not demanded.
// Otherwise, this pattern doesn't match.
return false;
}
/// SelectInlineAsmMemoryOperands - Calls to this are automatically generated
/// by tblgen. Others should not call it.
void SelectionDAGISel::
SelectInlineAsmMemoryOperands(std::vector<SDValue> &Ops) {
std::vector<SDValue> InOps;
std::swap(InOps, Ops);
Ops.push_back(InOps[0]); // input chain.
Ops.push_back(InOps[1]); // input asm string.
unsigned i = 2, e = InOps.size();
if (InOps[e-1].getValueType() == MVT::Flag)
--e; // Don't process a flag operand if it is here.
while (i != e) {
unsigned Flags = cast<ConstantSDNode>(InOps[i])->getZExtValue();
if ((Flags & 7) != 4 /*MEM*/) {
// Just skip over this operand, copying the operands verbatim.
Ops.insert(Ops.end(), InOps.begin()+i, InOps.begin()+i+(Flags >> 3) + 1);
i += (Flags >> 3) + 1;
} else {
assert((Flags >> 3) == 1 && "Memory operand with multiple values?");
// Otherwise, this is a memory operand. Ask the target to select it.
std::vector<SDValue> SelOps;
if (SelectInlineAsmMemoryOperand(InOps[i+1], 'm', SelOps)) {
cerr << "Could not match memory address. Inline asm failure!\n";
exit(1);
}
// Add this to the output node.
MVT IntPtrTy = CurDAG->getTargetLoweringInfo().getPointerTy();
Ops.push_back(CurDAG->getTargetConstant(4/*MEM*/ | (SelOps.size()<< 3),
IntPtrTy));
Ops.insert(Ops.end(), SelOps.begin(), SelOps.end());
i += 2;
}
}
// Add the flag input back if present.
if (e != InOps.size())
Ops.push_back(InOps.back());
}
char SelectionDAGISel::ID = 0;
|