forked from llvm/llvm-project
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathRISCVISelDAGToDAG.cpp
2140 lines (1906 loc) · 79.3 KB
/
RISCVISelDAGToDAG.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
//===-- RISCVISelDAGToDAG.cpp - A dag to dag inst selector for RISCV ------===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
// This file defines an instruction selector for the RISCV target.
//
//===----------------------------------------------------------------------===//
#include "RISCVISelDAGToDAG.h"
#include "MCTargetDesc/RISCVMCTargetDesc.h"
#include "MCTargetDesc/RISCVMatInt.h"
#include "RISCVISelLowering.h"
#include "RISCVMachineFunctionInfo.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/IR/IntrinsicsRISCV.h"
#include "llvm/Support/Alignment.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/KnownBits.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;
#define DEBUG_TYPE "riscv-isel"
namespace llvm {
namespace RISCV {
#define GET_RISCVVSSEGTable_IMPL
#define GET_RISCVVLSEGTable_IMPL
#define GET_RISCVVLXSEGTable_IMPL
#define GET_RISCVVSXSEGTable_IMPL
#define GET_RISCVVLETable_IMPL
#define GET_RISCVVSETable_IMPL
#define GET_RISCVVLXTable_IMPL
#define GET_RISCVVSXTable_IMPL
#include "RISCVGenSearchableTables.inc"
} // namespace RISCV
} // namespace llvm
void RISCVDAGToDAGISel::PreprocessISelDAG() {
for (SelectionDAG::allnodes_iterator I = CurDAG->allnodes_begin(),
E = CurDAG->allnodes_end();
I != E;) {
SDNode *N = &*I++; // Preincrement iterator to avoid invalidation issues.
// Lower SPLAT_VECTOR_SPLIT_I64 to two scalar stores and a stride 0 vector
// load. Done after lowering and combining so that we have a chance to
// optimize this to VMV_V_X_VL when the upper bits aren't needed.
if (N->getOpcode() != RISCVISD::SPLAT_VECTOR_SPLIT_I64_VL)
continue;
assert(N->getNumOperands() == 3 && "Unexpected number of operands");
MVT VT = N->getSimpleValueType(0);
SDValue Lo = N->getOperand(0);
SDValue Hi = N->getOperand(1);
SDValue VL = N->getOperand(2);
assert(VT.getVectorElementType() == MVT::i64 && VT.isScalableVector() &&
Lo.getValueType() == MVT::i32 && Hi.getValueType() == MVT::i32 &&
"Unexpected VTs!");
MachineFunction &MF = CurDAG->getMachineFunction();
RISCVMachineFunctionInfo *FuncInfo = MF.getInfo<RISCVMachineFunctionInfo>();
SDLoc DL(N);
// We use the same frame index we use for moving two i32s into 64-bit FPR.
// This is an analogous operation.
int FI = FuncInfo->getMoveF64FrameIndex(MF);
MachinePointerInfo MPI = MachinePointerInfo::getFixedStack(MF, FI);
const TargetLowering &TLI = CurDAG->getTargetLoweringInfo();
SDValue StackSlot =
CurDAG->getFrameIndex(FI, TLI.getPointerTy(CurDAG->getDataLayout()));
SDValue Chain = CurDAG->getEntryNode();
Lo = CurDAG->getStore(Chain, DL, Lo, StackSlot, MPI, Align(8));
SDValue OffsetSlot =
CurDAG->getMemBasePlusOffset(StackSlot, TypeSize::Fixed(4), DL);
Hi = CurDAG->getStore(Chain, DL, Hi, OffsetSlot, MPI.getWithOffset(4),
Align(8));
Chain = CurDAG->getNode(ISD::TokenFactor, DL, MVT::Other, Lo, Hi);
SDVTList VTs = CurDAG->getVTList({VT, MVT::Other});
SDValue IntID =
CurDAG->getTargetConstant(Intrinsic::riscv_vlse, DL, MVT::i64);
SDValue Ops[] = {Chain,
IntID,
CurDAG->getUNDEF(VT),
StackSlot,
CurDAG->getRegister(RISCV::X0, MVT::i64),
VL};
SDValue Result = CurDAG->getMemIntrinsicNode(
ISD::INTRINSIC_W_CHAIN, DL, VTs, Ops, MVT::i64, MPI, Align(8),
MachineMemOperand::MOLoad);
// We're about to replace all uses of the SPLAT_VECTOR_SPLIT_I64 with the
// vlse we created. This will cause general havok on the dag because
// anything below the conversion could be folded into other existing nodes.
// To avoid invalidating 'I', back it up to the convert node.
--I;
CurDAG->ReplaceAllUsesOfValueWith(SDValue(N, 0), Result);
// Now that we did that, the node is dead. Increment the iterator to the
// next node to process, then delete N.
++I;
CurDAG->DeleteNode(N);
}
}
void RISCVDAGToDAGISel::PostprocessISelDAG() {
SelectionDAG::allnodes_iterator Position = CurDAG->allnodes_end();
bool MadeChange = false;
while (Position != CurDAG->allnodes_begin()) {
SDNode *N = &*--Position;
// Skip dead nodes and any non-machine opcodes.
if (N->use_empty() || !N->isMachineOpcode())
continue;
MadeChange |= doPeepholeSExtW(N);
MadeChange |= doPeepholeLoadStoreADDI(N);
}
if (MadeChange)
CurDAG->RemoveDeadNodes();
}
static SDNode *selectImmWithConstantPool(SelectionDAG *CurDAG, const SDLoc &DL,
const MVT VT, int64_t Imm,
const RISCVSubtarget &Subtarget) {
assert(VT == MVT::i64 && "Expecting MVT::i64");
const RISCVTargetLowering *TLI = Subtarget.getTargetLowering();
ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(CurDAG->getConstantPool(
ConstantInt::get(EVT(VT).getTypeForEVT(*CurDAG->getContext()), Imm), VT));
SDValue Addr = TLI->getAddr(CP, *CurDAG);
SDValue Offset = CurDAG->getTargetConstant(0, DL, VT);
// Since there is no data race, the chain can be the entry node.
SDNode *Load = CurDAG->getMachineNode(RISCV::LD, DL, VT, Addr, Offset,
CurDAG->getEntryNode());
MachineFunction &MF = CurDAG->getMachineFunction();
MachineMemOperand *MemOp = MF.getMachineMemOperand(
MachinePointerInfo::getConstantPool(MF), MachineMemOperand::MOLoad,
LLT(VT), CP->getAlign());
CurDAG->setNodeMemRefs(cast<MachineSDNode>(Load), {MemOp});
return Load;
}
static SDNode *selectImm(SelectionDAG *CurDAG, const SDLoc &DL, const MVT VT,
int64_t Imm, const RISCVSubtarget &Subtarget) {
MVT XLenVT = Subtarget.getXLenVT();
RISCVMatInt::InstSeq Seq =
RISCVMatInt::generateInstSeq(Imm, Subtarget.getFeatureBits());
// If Imm is expensive to build, then we put it into constant pool.
if (Subtarget.useConstantPoolForLargeInts() &&
Seq.size() > Subtarget.getMaxBuildIntsCost())
return selectImmWithConstantPool(CurDAG, DL, VT, Imm, Subtarget);
SDNode *Result = nullptr;
SDValue SrcReg = CurDAG->getRegister(RISCV::X0, XLenVT);
for (RISCVMatInt::Inst &Inst : Seq) {
SDValue SDImm = CurDAG->getTargetConstant(Inst.Imm, DL, XLenVT);
if (Inst.Opc == RISCV::LUI)
Result = CurDAG->getMachineNode(RISCV::LUI, DL, XLenVT, SDImm);
else if (Inst.Opc == RISCV::ADD_UW)
Result = CurDAG->getMachineNode(RISCV::ADD_UW, DL, XLenVT, SrcReg,
CurDAG->getRegister(RISCV::X0, XLenVT));
else if (Inst.Opc == RISCV::SH1ADD || Inst.Opc == RISCV::SH2ADD ||
Inst.Opc == RISCV::SH3ADD)
Result = CurDAG->getMachineNode(Inst.Opc, DL, XLenVT, SrcReg, SrcReg);
else
Result = CurDAG->getMachineNode(Inst.Opc, DL, XLenVT, SrcReg, SDImm);
// Only the first instruction has X0 as its source.
SrcReg = SDValue(Result, 0);
}
return Result;
}
static SDValue createTupleImpl(SelectionDAG &CurDAG, ArrayRef<SDValue> Regs,
unsigned RegClassID, unsigned SubReg0) {
assert(Regs.size() >= 2 && Regs.size() <= 8);
SDLoc DL(Regs[0]);
SmallVector<SDValue, 8> Ops;
Ops.push_back(CurDAG.getTargetConstant(RegClassID, DL, MVT::i32));
for (unsigned I = 0; I < Regs.size(); ++I) {
Ops.push_back(Regs[I]);
Ops.push_back(CurDAG.getTargetConstant(SubReg0 + I, DL, MVT::i32));
}
SDNode *N =
CurDAG.getMachineNode(TargetOpcode::REG_SEQUENCE, DL, MVT::Untyped, Ops);
return SDValue(N, 0);
}
static SDValue createM1Tuple(SelectionDAG &CurDAG, ArrayRef<SDValue> Regs,
unsigned NF) {
static const unsigned RegClassIDs[] = {
RISCV::VRN2M1RegClassID, RISCV::VRN3M1RegClassID, RISCV::VRN4M1RegClassID,
RISCV::VRN5M1RegClassID, RISCV::VRN6M1RegClassID, RISCV::VRN7M1RegClassID,
RISCV::VRN8M1RegClassID};
return createTupleImpl(CurDAG, Regs, RegClassIDs[NF - 2], RISCV::sub_vrm1_0);
}
static SDValue createM2Tuple(SelectionDAG &CurDAG, ArrayRef<SDValue> Regs,
unsigned NF) {
static const unsigned RegClassIDs[] = {RISCV::VRN2M2RegClassID,
RISCV::VRN3M2RegClassID,
RISCV::VRN4M2RegClassID};
return createTupleImpl(CurDAG, Regs, RegClassIDs[NF - 2], RISCV::sub_vrm2_0);
}
static SDValue createM4Tuple(SelectionDAG &CurDAG, ArrayRef<SDValue> Regs,
unsigned NF) {
return createTupleImpl(CurDAG, Regs, RISCV::VRN2M4RegClassID,
RISCV::sub_vrm4_0);
}
static SDValue createTuple(SelectionDAG &CurDAG, ArrayRef<SDValue> Regs,
unsigned NF, RISCVII::VLMUL LMUL) {
switch (LMUL) {
default:
llvm_unreachable("Invalid LMUL.");
case RISCVII::VLMUL::LMUL_F8:
case RISCVII::VLMUL::LMUL_F4:
case RISCVII::VLMUL::LMUL_F2:
case RISCVII::VLMUL::LMUL_1:
return createM1Tuple(CurDAG, Regs, NF);
case RISCVII::VLMUL::LMUL_2:
return createM2Tuple(CurDAG, Regs, NF);
case RISCVII::VLMUL::LMUL_4:
return createM4Tuple(CurDAG, Regs, NF);
}
}
void RISCVDAGToDAGISel::addVectorLoadStoreOperands(
SDNode *Node, unsigned Log2SEW, const SDLoc &DL, unsigned CurOp,
bool IsMasked, bool IsStridedOrIndexed, SmallVectorImpl<SDValue> &Operands,
bool IsLoad, MVT *IndexVT) {
SDValue Chain = Node->getOperand(0);
SDValue Glue;
SDValue Base;
SelectBaseAddr(Node->getOperand(CurOp++), Base);
Operands.push_back(Base); // Base pointer.
if (IsStridedOrIndexed) {
Operands.push_back(Node->getOperand(CurOp++)); // Index.
if (IndexVT)
*IndexVT = Operands.back()->getSimpleValueType(0);
}
if (IsMasked) {
// Mask needs to be copied to V0.
SDValue Mask = Node->getOperand(CurOp++);
Chain = CurDAG->getCopyToReg(Chain, DL, RISCV::V0, Mask, SDValue());
Glue = Chain.getValue(1);
Operands.push_back(CurDAG->getRegister(RISCV::V0, Mask.getValueType()));
}
SDValue VL;
selectVLOp(Node->getOperand(CurOp++), VL);
Operands.push_back(VL);
MVT XLenVT = Subtarget->getXLenVT();
SDValue SEWOp = CurDAG->getTargetConstant(Log2SEW, DL, XLenVT);
Operands.push_back(SEWOp);
// Masked load has the tail policy argument.
if (IsMasked && IsLoad) {
// Policy must be a constant.
uint64_t Policy = Node->getConstantOperandVal(CurOp++);
SDValue PolicyOp = CurDAG->getTargetConstant(Policy, DL, XLenVT);
Operands.push_back(PolicyOp);
}
Operands.push_back(Chain); // Chain.
if (Glue)
Operands.push_back(Glue);
}
void RISCVDAGToDAGISel::selectVLSEG(SDNode *Node, bool IsMasked,
bool IsStrided) {
SDLoc DL(Node);
unsigned NF = Node->getNumValues() - 1;
MVT VT = Node->getSimpleValueType(0);
unsigned Log2SEW = Log2_32(VT.getScalarSizeInBits());
RISCVII::VLMUL LMUL = RISCVTargetLowering::getLMUL(VT);
unsigned CurOp = 2;
SmallVector<SDValue, 8> Operands;
if (IsMasked) {
SmallVector<SDValue, 8> Regs(Node->op_begin() + CurOp,
Node->op_begin() + CurOp + NF);
SDValue MaskedOff = createTuple(*CurDAG, Regs, NF, LMUL);
Operands.push_back(MaskedOff);
CurOp += NF;
}
addVectorLoadStoreOperands(Node, Log2SEW, DL, CurOp, IsMasked, IsStrided,
Operands, /*IsLoad=*/true);
const RISCV::VLSEGPseudo *P =
RISCV::getVLSEGPseudo(NF, IsMasked, IsStrided, /*FF*/ false, Log2SEW,
static_cast<unsigned>(LMUL));
MachineSDNode *Load =
CurDAG->getMachineNode(P->Pseudo, DL, MVT::Untyped, MVT::Other, Operands);
if (auto *MemOp = dyn_cast<MemSDNode>(Node))
CurDAG->setNodeMemRefs(Load, {MemOp->getMemOperand()});
SDValue SuperReg = SDValue(Load, 0);
for (unsigned I = 0; I < NF; ++I) {
unsigned SubRegIdx = RISCVTargetLowering::getSubregIndexByMVT(VT, I);
ReplaceUses(SDValue(Node, I),
CurDAG->getTargetExtractSubreg(SubRegIdx, DL, VT, SuperReg));
}
ReplaceUses(SDValue(Node, NF), SDValue(Load, 1));
CurDAG->RemoveDeadNode(Node);
}
void RISCVDAGToDAGISel::selectVLSEGFF(SDNode *Node, bool IsMasked) {
SDLoc DL(Node);
unsigned NF = Node->getNumValues() - 2; // Do not count VL and Chain.
MVT VT = Node->getSimpleValueType(0);
MVT XLenVT = Subtarget->getXLenVT();
unsigned Log2SEW = Log2_32(VT.getScalarSizeInBits());
RISCVII::VLMUL LMUL = RISCVTargetLowering::getLMUL(VT);
unsigned CurOp = 2;
SmallVector<SDValue, 7> Operands;
if (IsMasked) {
SmallVector<SDValue, 8> Regs(Node->op_begin() + CurOp,
Node->op_begin() + CurOp + NF);
SDValue MaskedOff = createTuple(*CurDAG, Regs, NF, LMUL);
Operands.push_back(MaskedOff);
CurOp += NF;
}
addVectorLoadStoreOperands(Node, Log2SEW, DL, CurOp, IsMasked,
/*IsStridedOrIndexed*/ false, Operands,
/*IsLoad=*/true);
const RISCV::VLSEGPseudo *P =
RISCV::getVLSEGPseudo(NF, IsMasked, /*Strided*/ false, /*FF*/ true,
Log2SEW, static_cast<unsigned>(LMUL));
MachineSDNode *Load = CurDAG->getMachineNode(P->Pseudo, DL, MVT::Untyped,
MVT::Other, MVT::Glue, Operands);
SDNode *ReadVL = CurDAG->getMachineNode(RISCV::PseudoReadVL, DL, XLenVT,
/*Glue*/ SDValue(Load, 2));
if (auto *MemOp = dyn_cast<MemSDNode>(Node))
CurDAG->setNodeMemRefs(Load, {MemOp->getMemOperand()});
SDValue SuperReg = SDValue(Load, 0);
for (unsigned I = 0; I < NF; ++I) {
unsigned SubRegIdx = RISCVTargetLowering::getSubregIndexByMVT(VT, I);
ReplaceUses(SDValue(Node, I),
CurDAG->getTargetExtractSubreg(SubRegIdx, DL, VT, SuperReg));
}
ReplaceUses(SDValue(Node, NF), SDValue(ReadVL, 0)); // VL
ReplaceUses(SDValue(Node, NF + 1), SDValue(Load, 1)); // Chain
CurDAG->RemoveDeadNode(Node);
}
void RISCVDAGToDAGISel::selectVLXSEG(SDNode *Node, bool IsMasked,
bool IsOrdered) {
SDLoc DL(Node);
unsigned NF = Node->getNumValues() - 1;
MVT VT = Node->getSimpleValueType(0);
unsigned Log2SEW = Log2_32(VT.getScalarSizeInBits());
RISCVII::VLMUL LMUL = RISCVTargetLowering::getLMUL(VT);
unsigned CurOp = 2;
SmallVector<SDValue, 8> Operands;
if (IsMasked) {
SmallVector<SDValue, 8> Regs(Node->op_begin() + CurOp,
Node->op_begin() + CurOp + NF);
SDValue MaskedOff = createTuple(*CurDAG, Regs, NF, LMUL);
Operands.push_back(MaskedOff);
CurOp += NF;
}
MVT IndexVT;
addVectorLoadStoreOperands(Node, Log2SEW, DL, CurOp, IsMasked,
/*IsStridedOrIndexed*/ true, Operands,
/*IsLoad=*/true, &IndexVT);
assert(VT.getVectorElementCount() == IndexVT.getVectorElementCount() &&
"Element count mismatch");
RISCVII::VLMUL IndexLMUL = RISCVTargetLowering::getLMUL(IndexVT);
unsigned IndexLog2EEW = Log2_32(IndexVT.getScalarSizeInBits());
if (IndexLog2EEW == 6 && !Subtarget->is64Bit()) {
report_fatal_error("The V extension does not support EEW=64 for index "
"values when XLEN=32");
}
const RISCV::VLXSEGPseudo *P = RISCV::getVLXSEGPseudo(
NF, IsMasked, IsOrdered, IndexLog2EEW, static_cast<unsigned>(LMUL),
static_cast<unsigned>(IndexLMUL));
MachineSDNode *Load =
CurDAG->getMachineNode(P->Pseudo, DL, MVT::Untyped, MVT::Other, Operands);
if (auto *MemOp = dyn_cast<MemSDNode>(Node))
CurDAG->setNodeMemRefs(Load, {MemOp->getMemOperand()});
SDValue SuperReg = SDValue(Load, 0);
for (unsigned I = 0; I < NF; ++I) {
unsigned SubRegIdx = RISCVTargetLowering::getSubregIndexByMVT(VT, I);
ReplaceUses(SDValue(Node, I),
CurDAG->getTargetExtractSubreg(SubRegIdx, DL, VT, SuperReg));
}
ReplaceUses(SDValue(Node, NF), SDValue(Load, 1));
CurDAG->RemoveDeadNode(Node);
}
void RISCVDAGToDAGISel::selectVSSEG(SDNode *Node, bool IsMasked,
bool IsStrided) {
SDLoc DL(Node);
unsigned NF = Node->getNumOperands() - 4;
if (IsStrided)
NF--;
if (IsMasked)
NF--;
MVT VT = Node->getOperand(2)->getSimpleValueType(0);
unsigned Log2SEW = Log2_32(VT.getScalarSizeInBits());
RISCVII::VLMUL LMUL = RISCVTargetLowering::getLMUL(VT);
SmallVector<SDValue, 8> Regs(Node->op_begin() + 2, Node->op_begin() + 2 + NF);
SDValue StoreVal = createTuple(*CurDAG, Regs, NF, LMUL);
SmallVector<SDValue, 8> Operands;
Operands.push_back(StoreVal);
unsigned CurOp = 2 + NF;
addVectorLoadStoreOperands(Node, Log2SEW, DL, CurOp, IsMasked, IsStrided,
Operands);
const RISCV::VSSEGPseudo *P = RISCV::getVSSEGPseudo(
NF, IsMasked, IsStrided, Log2SEW, static_cast<unsigned>(LMUL));
MachineSDNode *Store =
CurDAG->getMachineNode(P->Pseudo, DL, Node->getValueType(0), Operands);
if (auto *MemOp = dyn_cast<MemSDNode>(Node))
CurDAG->setNodeMemRefs(Store, {MemOp->getMemOperand()});
ReplaceNode(Node, Store);
}
void RISCVDAGToDAGISel::selectVSXSEG(SDNode *Node, bool IsMasked,
bool IsOrdered) {
SDLoc DL(Node);
unsigned NF = Node->getNumOperands() - 5;
if (IsMasked)
--NF;
MVT VT = Node->getOperand(2)->getSimpleValueType(0);
unsigned Log2SEW = Log2_32(VT.getScalarSizeInBits());
RISCVII::VLMUL LMUL = RISCVTargetLowering::getLMUL(VT);
SmallVector<SDValue, 8> Regs(Node->op_begin() + 2, Node->op_begin() + 2 + NF);
SDValue StoreVal = createTuple(*CurDAG, Regs, NF, LMUL);
SmallVector<SDValue, 8> Operands;
Operands.push_back(StoreVal);
unsigned CurOp = 2 + NF;
MVT IndexVT;
addVectorLoadStoreOperands(Node, Log2SEW, DL, CurOp, IsMasked,
/*IsStridedOrIndexed*/ true, Operands,
/*IsLoad=*/false, &IndexVT);
assert(VT.getVectorElementCount() == IndexVT.getVectorElementCount() &&
"Element count mismatch");
RISCVII::VLMUL IndexLMUL = RISCVTargetLowering::getLMUL(IndexVT);
unsigned IndexLog2EEW = Log2_32(IndexVT.getScalarSizeInBits());
if (IndexLog2EEW == 6 && !Subtarget->is64Bit()) {
report_fatal_error("The V extension does not support EEW=64 for index "
"values when XLEN=32");
}
const RISCV::VSXSEGPseudo *P = RISCV::getVSXSEGPseudo(
NF, IsMasked, IsOrdered, IndexLog2EEW, static_cast<unsigned>(LMUL),
static_cast<unsigned>(IndexLMUL));
MachineSDNode *Store =
CurDAG->getMachineNode(P->Pseudo, DL, Node->getValueType(0), Operands);
if (auto *MemOp = dyn_cast<MemSDNode>(Node))
CurDAG->setNodeMemRefs(Store, {MemOp->getMemOperand()});
ReplaceNode(Node, Store);
}
void RISCVDAGToDAGISel::selectVSETVLI(SDNode *Node) {
if (!Subtarget->hasVInstructions())
return;
assert((Node->getOpcode() == ISD::INTRINSIC_W_CHAIN ||
Node->getOpcode() == ISD::INTRINSIC_WO_CHAIN) &&
"Unexpected opcode");
SDLoc DL(Node);
MVT XLenVT = Subtarget->getXLenVT();
bool HasChain = Node->getOpcode() == ISD::INTRINSIC_W_CHAIN;
unsigned IntNoOffset = HasChain ? 1 : 0;
unsigned IntNo = Node->getConstantOperandVal(IntNoOffset);
assert((IntNo == Intrinsic::riscv_vsetvli ||
IntNo == Intrinsic::riscv_vsetvlimax ||
IntNo == Intrinsic::riscv_vsetvli_opt ||
IntNo == Intrinsic::riscv_vsetvlimax_opt) &&
"Unexpected vsetvli intrinsic");
bool VLMax = IntNo == Intrinsic::riscv_vsetvlimax ||
IntNo == Intrinsic::riscv_vsetvlimax_opt;
unsigned Offset = IntNoOffset + (VLMax ? 1 : 2);
assert(Node->getNumOperands() == Offset + 2 &&
"Unexpected number of operands");
unsigned SEW =
RISCVVType::decodeVSEW(Node->getConstantOperandVal(Offset) & 0x7);
RISCVII::VLMUL VLMul = static_cast<RISCVII::VLMUL>(
Node->getConstantOperandVal(Offset + 1) & 0x7);
unsigned VTypeI = RISCVVType::encodeVTYPE(VLMul, SEW, /*TailAgnostic*/ true,
/*MaskAgnostic*/ false);
SDValue VTypeIOp = CurDAG->getTargetConstant(VTypeI, DL, XLenVT);
SmallVector<EVT, 2> VTs = {XLenVT};
if (HasChain)
VTs.push_back(MVT::Other);
SDValue VLOperand;
unsigned Opcode = RISCV::PseudoVSETVLI;
if (VLMax) {
VLOperand = CurDAG->getRegister(RISCV::X0, XLenVT);
Opcode = RISCV::PseudoVSETVLIX0;
} else {
VLOperand = Node->getOperand(IntNoOffset + 1);
if (auto *C = dyn_cast<ConstantSDNode>(VLOperand)) {
uint64_t AVL = C->getZExtValue();
if (isUInt<5>(AVL)) {
SDValue VLImm = CurDAG->getTargetConstant(AVL, DL, XLenVT);
SmallVector<SDValue, 3> Ops = {VLImm, VTypeIOp};
if (HasChain)
Ops.push_back(Node->getOperand(0));
ReplaceNode(
Node, CurDAG->getMachineNode(RISCV::PseudoVSETIVLI, DL, VTs, Ops));
return;
}
}
}
SmallVector<SDValue, 3> Ops = {VLOperand, VTypeIOp};
if (HasChain)
Ops.push_back(Node->getOperand(0));
ReplaceNode(Node, CurDAG->getMachineNode(Opcode, DL, VTs, Ops));
}
void RISCVDAGToDAGISel::Select(SDNode *Node) {
// If we have a custom node, we have already selected.
if (Node->isMachineOpcode()) {
LLVM_DEBUG(dbgs() << "== "; Node->dump(CurDAG); dbgs() << "\n");
Node->setNodeId(-1);
return;
}
// Instruction Selection not handled by the auto-generated tablegen selection
// should be handled here.
unsigned Opcode = Node->getOpcode();
MVT XLenVT = Subtarget->getXLenVT();
SDLoc DL(Node);
MVT VT = Node->getSimpleValueType(0);
switch (Opcode) {
case ISD::Constant: {
auto *ConstNode = cast<ConstantSDNode>(Node);
if (VT == XLenVT && ConstNode->isZero()) {
SDValue New =
CurDAG->getCopyFromReg(CurDAG->getEntryNode(), DL, RISCV::X0, XLenVT);
ReplaceNode(Node, New.getNode());
return;
}
int64_t Imm = ConstNode->getSExtValue();
// If the upper XLen-16 bits are not used, try to convert this to a simm12
// by sign extending bit 15.
if (isUInt<16>(Imm) && isInt<12>(SignExtend64(Imm, 16)) &&
hasAllHUsers(Node))
Imm = SignExtend64(Imm, 16);
// If the upper 32-bits are not used try to convert this into a simm32 by
// sign extending bit 32.
if (!isInt<32>(Imm) && isUInt<32>(Imm) && hasAllWUsers(Node))
Imm = SignExtend64(Imm, 32);
ReplaceNode(Node, selectImm(CurDAG, DL, VT, Imm, *Subtarget));
return;
}
case ISD::FrameIndex: {
SDValue Imm = CurDAG->getTargetConstant(0, DL, XLenVT);
int FI = cast<FrameIndexSDNode>(Node)->getIndex();
SDValue TFI = CurDAG->getTargetFrameIndex(FI, VT);
ReplaceNode(Node, CurDAG->getMachineNode(RISCV::ADDI, DL, VT, TFI, Imm));
return;
}
case ISD::SRL: {
// Optimize (srl (and X, C2), C) ->
// (srli (slli X, (XLen-C3), (XLen-C3) + C)
// Where C2 is a mask with C3 trailing ones.
// Taking into account that the C2 may have had lower bits unset by
// SimplifyDemandedBits. This avoids materializing the C2 immediate.
// This pattern occurs when type legalizing right shifts for types with
// less than XLen bits.
auto *N1C = dyn_cast<ConstantSDNode>(Node->getOperand(1));
if (!N1C)
break;
SDValue N0 = Node->getOperand(0);
if (N0.getOpcode() != ISD::AND || !N0.hasOneUse() ||
!isa<ConstantSDNode>(N0.getOperand(1)))
break;
unsigned ShAmt = N1C->getZExtValue();
uint64_t Mask = N0.getConstantOperandVal(1);
Mask |= maskTrailingOnes<uint64_t>(ShAmt);
if (!isMask_64(Mask))
break;
unsigned TrailingOnes = countTrailingOnes(Mask);
// 32 trailing ones should use srliw via tablegen pattern.
if (TrailingOnes == 32 || ShAmt >= TrailingOnes)
break;
unsigned LShAmt = Subtarget->getXLen() - TrailingOnes;
SDNode *SLLI =
CurDAG->getMachineNode(RISCV::SLLI, DL, VT, N0->getOperand(0),
CurDAG->getTargetConstant(LShAmt, DL, VT));
SDNode *SRLI = CurDAG->getMachineNode(
RISCV::SRLI, DL, VT, SDValue(SLLI, 0),
CurDAG->getTargetConstant(LShAmt + ShAmt, DL, VT));
ReplaceNode(Node, SRLI);
return;
}
case ISD::SRA: {
// Optimize (sra (sext_inreg X, i16), C) ->
// (srai (slli X, (XLen-16), (XLen-16) + C)
// And (sra (sext_inreg X, i8), C) ->
// (srai (slli X, (XLen-8), (XLen-8) + C)
// This can occur when Zbb is enabled, which makes sext_inreg i16/i8 legal.
// This transform matches the code we get without Zbb. The shifts are more
// compressible, and this can help expose CSE opportunities in the sdiv by
// constant optimization.
auto *N1C = dyn_cast<ConstantSDNode>(Node->getOperand(1));
if (!N1C)
break;
SDValue N0 = Node->getOperand(0);
if (N0.getOpcode() != ISD::SIGN_EXTEND_INREG || !N0.hasOneUse())
break;
unsigned ShAmt = N1C->getZExtValue();
unsigned ExtSize =
cast<VTSDNode>(N0.getOperand(1))->getVT().getSizeInBits();
// ExtSize of 32 should use sraiw via tablegen pattern.
if (ExtSize >= 32 || ShAmt >= ExtSize)
break;
unsigned LShAmt = Subtarget->getXLen() - ExtSize;
SDNode *SLLI =
CurDAG->getMachineNode(RISCV::SLLI, DL, VT, N0->getOperand(0),
CurDAG->getTargetConstant(LShAmt, DL, VT));
SDNode *SRAI = CurDAG->getMachineNode(
RISCV::SRAI, DL, VT, SDValue(SLLI, 0),
CurDAG->getTargetConstant(LShAmt + ShAmt, DL, VT));
ReplaceNode(Node, SRAI);
return;
}
case ISD::AND: {
auto *N1C = dyn_cast<ConstantSDNode>(Node->getOperand(1));
if (!N1C)
break;
SDValue N0 = Node->getOperand(0);
bool LeftShift = N0.getOpcode() == ISD::SHL;
if (!LeftShift && N0.getOpcode() != ISD::SRL)
break;
auto *C = dyn_cast<ConstantSDNode>(N0.getOperand(1));
if (!C)
break;
uint64_t C2 = C->getZExtValue();
unsigned XLen = Subtarget->getXLen();
if (!C2 || C2 >= XLen)
break;
uint64_t C1 = N1C->getZExtValue();
// Keep track of whether this is an andi.
bool IsANDI = isInt<12>(N1C->getSExtValue());
// Clear irrelevant bits in the mask.
if (LeftShift)
C1 &= maskTrailingZeros<uint64_t>(C2);
else
C1 &= maskTrailingOnes<uint64_t>(XLen - C2);
// Some transforms should only be done if the shift has a single use or
// the AND would become (srli (slli X, 32), 32)
bool OneUseOrZExtW = N0.hasOneUse() || C1 == UINT64_C(0xFFFFFFFF);
SDValue X = N0.getOperand(0);
// Turn (and (srl x, c2) c1) -> (srli (slli x, c3-c2), c3) if c1 is a mask
// with c3 leading zeros.
if (!LeftShift && isMask_64(C1)) {
uint64_t C3 = XLen - (64 - countLeadingZeros(C1));
if (C2 < C3) {
// If the number of leading zeros is C2+32 this can be SRLIW.
if (C2 + 32 == C3) {
SDNode *SRLIW =
CurDAG->getMachineNode(RISCV::SRLIW, DL, XLenVT, X,
CurDAG->getTargetConstant(C2, DL, XLenVT));
ReplaceNode(Node, SRLIW);
return;
}
// (and (srl (sexti32 Y), c2), c1) -> (srliw (sraiw Y, 31), c3 - 32) if
// c1 is a mask with c3 leading zeros and c2 >= 32 and c3-c2==1.
//
// This pattern occurs when (i32 (srl (sra 31), c3 - 32)) is type
// legalized and goes through DAG combine.
if (C2 >= 32 && (C3 - C2) == 1 && N0.hasOneUse() &&
X.getOpcode() == ISD::SIGN_EXTEND_INREG &&
cast<VTSDNode>(X.getOperand(1))->getVT() == MVT::i32) {
SDNode *SRAIW =
CurDAG->getMachineNode(RISCV::SRAIW, DL, XLenVT, X.getOperand(0),
CurDAG->getTargetConstant(31, DL, XLenVT));
SDNode *SRLIW = CurDAG->getMachineNode(
RISCV::SRLIW, DL, XLenVT, SDValue(SRAIW, 0),
CurDAG->getTargetConstant(C3 - 32, DL, XLenVT));
ReplaceNode(Node, SRLIW);
return;
}
// (srli (slli x, c3-c2), c3).
if (OneUseOrZExtW && !IsANDI) {
SDNode *SLLI = CurDAG->getMachineNode(
RISCV::SLLI, DL, XLenVT, X,
CurDAG->getTargetConstant(C3 - C2, DL, XLenVT));
SDNode *SRLI =
CurDAG->getMachineNode(RISCV::SRLI, DL, XLenVT, SDValue(SLLI, 0),
CurDAG->getTargetConstant(C3, DL, XLenVT));
ReplaceNode(Node, SRLI);
return;
}
}
}
// Turn (and (shl x, c2), c1) -> (srli (slli c2+c3), c3) if c1 is a mask
// shifted by c2 bits with c3 leading zeros.
if (LeftShift && isShiftedMask_64(C1)) {
uint64_t C3 = XLen - (64 - countLeadingZeros(C1));
if (C2 + C3 < XLen &&
C1 == (maskTrailingOnes<uint64_t>(XLen - (C2 + C3)) << C2)) {
// Use slli.uw when possible.
if ((XLen - (C2 + C3)) == 32 && Subtarget->hasStdExtZba()) {
SDNode *SLLI_UW =
CurDAG->getMachineNode(RISCV::SLLI_UW, DL, XLenVT, X,
CurDAG->getTargetConstant(C2, DL, XLenVT));
ReplaceNode(Node, SLLI_UW);
return;
}
// (srli (slli c2+c3), c3)
if (OneUseOrZExtW && !IsANDI) {
SDNode *SLLI = CurDAG->getMachineNode(
RISCV::SLLI, DL, XLenVT, X,
CurDAG->getTargetConstant(C2 + C3, DL, XLenVT));
SDNode *SRLI =
CurDAG->getMachineNode(RISCV::SRLI, DL, XLenVT, SDValue(SLLI, 0),
CurDAG->getTargetConstant(C3, DL, XLenVT));
ReplaceNode(Node, SRLI);
return;
}
}
}
// Turn (and (shr x, c2), c1) -> (slli (srli x, c2+c3), c3) if c1 is a
// shifted mask with c2 leading zeros and c3 trailing zeros.
if (!LeftShift && isShiftedMask_64(C1)) {
uint64_t Leading = XLen - (64 - countLeadingZeros(C1));
uint64_t C3 = countTrailingZeros(C1);
if (Leading == C2 && C2 + C3 < XLen && OneUseOrZExtW && !IsANDI) {
SDNode *SRLI = CurDAG->getMachineNode(
RISCV::SRLI, DL, XLenVT, X,
CurDAG->getTargetConstant(C2 + C3, DL, XLenVT));
SDNode *SLLI =
CurDAG->getMachineNode(RISCV::SLLI, DL, XLenVT, SDValue(SRLI, 0),
CurDAG->getTargetConstant(C3, DL, XLenVT));
ReplaceNode(Node, SLLI);
return;
}
// If the leading zero count is C2+32, we can use SRLIW instead of SRLI.
if (Leading > 32 && (Leading - 32) == C2 && C2 + C3 < 32 &&
OneUseOrZExtW && !IsANDI) {
SDNode *SRLIW = CurDAG->getMachineNode(
RISCV::SRLIW, DL, XLenVT, X,
CurDAG->getTargetConstant(C2 + C3, DL, XLenVT));
SDNode *SLLI =
CurDAG->getMachineNode(RISCV::SLLI, DL, XLenVT, SDValue(SRLIW, 0),
CurDAG->getTargetConstant(C3, DL, XLenVT));
ReplaceNode(Node, SLLI);
return;
}
}
// Turn (and (shl x, c2), c1) -> (slli (srli x, c3-c2), c3) if c1 is a
// shifted mask with no leading zeros and c3 trailing zeros.
if (LeftShift && isShiftedMask_64(C1)) {
uint64_t Leading = XLen - (64 - countLeadingZeros(C1));
uint64_t C3 = countTrailingZeros(C1);
if (Leading == 0 && C2 < C3 && OneUseOrZExtW && !IsANDI) {
SDNode *SRLI = CurDAG->getMachineNode(
RISCV::SRLI, DL, XLenVT, X,
CurDAG->getTargetConstant(C3 - C2, DL, XLenVT));
SDNode *SLLI =
CurDAG->getMachineNode(RISCV::SLLI, DL, XLenVT, SDValue(SRLI, 0),
CurDAG->getTargetConstant(C3, DL, XLenVT));
ReplaceNode(Node, SLLI);
return;
}
// If we have (32-C2) leading zeros, we can use SRLIW instead of SRLI.
if (C2 < C3 && Leading + C2 == 32 && OneUseOrZExtW && !IsANDI) {
SDNode *SRLIW = CurDAG->getMachineNode(
RISCV::SRLIW, DL, XLenVT, X,
CurDAG->getTargetConstant(C3 - C2, DL, XLenVT));
SDNode *SLLI =
CurDAG->getMachineNode(RISCV::SLLI, DL, XLenVT, SDValue(SRLIW, 0),
CurDAG->getTargetConstant(C3, DL, XLenVT));
ReplaceNode(Node, SLLI);
return;
}
}
break;
}
case ISD::MUL: {
// Special case for calculating (mul (and X, C2), C1) where the full product
// fits in XLen bits. We can shift X left by the number of leading zeros in
// C2 and shift C1 left by XLen-lzcnt(C2). This will ensure the final
// product has XLen trailing zeros, putting it in the output of MULHU. This
// can avoid materializing a constant in a register for C2.
// RHS should be a constant.
auto *N1C = dyn_cast<ConstantSDNode>(Node->getOperand(1));
if (!N1C || !N1C->hasOneUse())
break;
// LHS should be an AND with constant.
SDValue N0 = Node->getOperand(0);
if (N0.getOpcode() != ISD::AND || !isa<ConstantSDNode>(N0.getOperand(1)))
break;
uint64_t C2 = cast<ConstantSDNode>(N0.getOperand(1))->getZExtValue();
// Constant should be a mask.
if (!isMask_64(C2))
break;
// This should be the only use of the AND unless we will use
// (SRLI (SLLI X, 32), 32). We don't use a shift pair for other AND
// constants.
if (!N0.hasOneUse() && C2 != UINT64_C(0xFFFFFFFF))
break;
// If this can be an ANDI, ZEXT.H or ZEXT.W we don't need to do this
// optimization.
if (isInt<12>(C2) ||
(C2 == UINT64_C(0xFFFF) &&
(Subtarget->hasStdExtZbb() || Subtarget->hasStdExtZbp())) ||
(C2 == UINT64_C(0xFFFFFFFF) && Subtarget->hasStdExtZba()))
break;
// We need to shift left the AND input and C1 by a total of XLen bits.
// How far left do we need to shift the AND input?
unsigned XLen = Subtarget->getXLen();
unsigned LeadingZeros = XLen - (64 - countLeadingZeros(C2));
// The constant gets shifted by the remaining amount unless that would
// shift bits out.
uint64_t C1 = N1C->getZExtValue();
unsigned ConstantShift = XLen - LeadingZeros;
if (ConstantShift > (XLen - (64 - countLeadingZeros(C1))))
break;
uint64_t ShiftedC1 = C1 << ConstantShift;
// If this RV32, we need to sign extend the constant.
if (XLen == 32)
ShiftedC1 = SignExtend64(ShiftedC1, 32);
// Create (mulhu (slli X, lzcnt(C2)), C1 << (XLen - lzcnt(C2))).
SDNode *Imm = selectImm(CurDAG, DL, VT, ShiftedC1, *Subtarget);
SDNode *SLLI =
CurDAG->getMachineNode(RISCV::SLLI, DL, VT, N0.getOperand(0),
CurDAG->getTargetConstant(LeadingZeros, DL, VT));
SDNode *MULHU = CurDAG->getMachineNode(RISCV::MULHU, DL, VT,
SDValue(SLLI, 0), SDValue(Imm, 0));
ReplaceNode(Node, MULHU);
return;
}
case ISD::INTRINSIC_WO_CHAIN: {
unsigned IntNo = Node->getConstantOperandVal(0);
switch (IntNo) {
// By default we do not custom select any intrinsic.
default:
break;
case Intrinsic::riscv_vmsgeu:
case Intrinsic::riscv_vmsge: {
SDValue Src1 = Node->getOperand(1);
SDValue Src2 = Node->getOperand(2);
bool IsUnsigned = IntNo == Intrinsic::riscv_vmsgeu;
bool IsCmpUnsignedZero = false;
// Only custom select scalar second operand.
if (Src2.getValueType() != XLenVT)
break;
// Small constants are handled with patterns.
if (auto *C = dyn_cast<ConstantSDNode>(Src2)) {
int64_t CVal = C->getSExtValue();
if (CVal >= -15 && CVal <= 16) {
if (!IsUnsigned || CVal != 0)
break;
IsCmpUnsignedZero = true;
}
}
MVT Src1VT = Src1.getSimpleValueType();
unsigned VMSLTOpcode, VMNANDOpcode, VMSetOpcode;
switch (RISCVTargetLowering::getLMUL(Src1VT)) {
default:
llvm_unreachable("Unexpected LMUL!");
#define CASE_VMSLT_VMNAND_VMSET_OPCODES(lmulenum, suffix, suffix_b) \
case RISCVII::VLMUL::lmulenum: \
VMSLTOpcode = IsUnsigned ? RISCV::PseudoVMSLTU_VX_##suffix \
: RISCV::PseudoVMSLT_VX_##suffix; \
VMNANDOpcode = RISCV::PseudoVMNAND_MM_##suffix; \
VMSetOpcode = RISCV::PseudoVMSET_M_##suffix_b; \
break;
CASE_VMSLT_VMNAND_VMSET_OPCODES(LMUL_F8, MF8, B1)
CASE_VMSLT_VMNAND_VMSET_OPCODES(LMUL_F4, MF4, B2)
CASE_VMSLT_VMNAND_VMSET_OPCODES(LMUL_F2, MF2, B4)
CASE_VMSLT_VMNAND_VMSET_OPCODES(LMUL_1, M1, B8)
CASE_VMSLT_VMNAND_VMSET_OPCODES(LMUL_2, M2, B16)
CASE_VMSLT_VMNAND_VMSET_OPCODES(LMUL_4, M4, B32)
CASE_VMSLT_VMNAND_VMSET_OPCODES(LMUL_8, M8, B64)
#undef CASE_VMSLT_VMNAND_VMSET_OPCODES
}
SDValue SEW = CurDAG->getTargetConstant(
Log2_32(Src1VT.getScalarSizeInBits()), DL, XLenVT);
SDValue VL;
selectVLOp(Node->getOperand(3), VL);
// If vmsgeu with 0 immediate, expand it to vmset.
if (IsCmpUnsignedZero) {
ReplaceNode(Node, CurDAG->getMachineNode(VMSetOpcode, DL, VT, VL, SEW));
return;
}
// Expand to
// vmslt{u}.vx vd, va, x; vmnand.mm vd, vd, vd
SDValue Cmp = SDValue(
CurDAG->getMachineNode(VMSLTOpcode, DL, VT, {Src1, Src2, VL, SEW}),
0);
ReplaceNode(Node, CurDAG->getMachineNode(VMNANDOpcode, DL, VT,
{Cmp, Cmp, VL, SEW}));
return;
}
case Intrinsic::riscv_vmsgeu_mask:
case Intrinsic::riscv_vmsge_mask: {
SDValue Src1 = Node->getOperand(2);
SDValue Src2 = Node->getOperand(3);
bool IsUnsigned = IntNo == Intrinsic::riscv_vmsgeu_mask;
bool IsCmpUnsignedZero = false;
// Only custom select scalar second operand.
if (Src2.getValueType() != XLenVT)
break;
// Small constants are handled with patterns.
if (auto *C = dyn_cast<ConstantSDNode>(Src2)) {
int64_t CVal = C->getSExtValue();
if (CVal >= -15 && CVal <= 16) {
if (!IsUnsigned || CVal != 0)
break;
IsCmpUnsignedZero = true;
}
}