xref: /freebsd/contrib/llvm-project/llvm/lib/IR/Instruction.cpp (revision 1d386b48a555f61cb7325543adbbb5c3f3407a66)
1 //===-- Instruction.cpp - Implement the Instruction class -----------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file implements the Instruction class for the IR library.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "llvm/IR/Instruction.h"
14 #include "llvm/ADT/DenseSet.h"
15 #include "llvm/IR/Constants.h"
16 #include "llvm/IR/Instructions.h"
17 #include "llvm/IR/IntrinsicInst.h"
18 #include "llvm/IR/Intrinsics.h"
19 #include "llvm/IR/Operator.h"
20 #include "llvm/IR/ProfDataUtils.h"
21 #include "llvm/IR/Type.h"
22 using namespace llvm;
23 
24 Instruction::Instruction(Type *ty, unsigned it, Use *Ops, unsigned NumOps,
25                          Instruction *InsertBefore)
26   : User(ty, Value::InstructionVal + it, Ops, NumOps), Parent(nullptr) {
27 
28   // If requested, insert this instruction into a basic block...
29   if (InsertBefore) {
30     BasicBlock *BB = InsertBefore->getParent();
31     assert(BB && "Instruction to insert before is not in a basic block!");
32     insertInto(BB, InsertBefore->getIterator());
33   }
34 }
35 
36 Instruction::Instruction(Type *ty, unsigned it, Use *Ops, unsigned NumOps,
37                          BasicBlock *InsertAtEnd)
38   : User(ty, Value::InstructionVal + it, Ops, NumOps), Parent(nullptr) {
39 
40   // append this instruction into the basic block
41   assert(InsertAtEnd && "Basic block to append to may not be NULL!");
42   insertInto(InsertAtEnd, InsertAtEnd->end());
43 }
44 
45 Instruction::~Instruction() {
46   assert(!Parent && "Instruction still linked in the program!");
47 
48   // Replace any extant metadata uses of this instruction with undef to
49   // preserve debug info accuracy. Some alternatives include:
50   // - Treat Instruction like any other Value, and point its extant metadata
51   //   uses to an empty ValueAsMetadata node. This makes extant dbg.value uses
52   //   trivially dead (i.e. fair game for deletion in many passes), leading to
53   //   stale dbg.values being in effect for too long.
54   // - Call salvageDebugInfoOrMarkUndef. Not needed to make instruction removal
55   //   correct. OTOH results in wasted work in some common cases (e.g. when all
56   //   instructions in a BasicBlock are deleted).
57   if (isUsedByMetadata())
58     ValueAsMetadata::handleRAUW(this, UndefValue::get(getType()));
59 
60   // Explicitly remove DIAssignID metadata to clear up ID -> Instruction(s)
61   // mapping in LLVMContext.
62   setMetadata(LLVMContext::MD_DIAssignID, nullptr);
63 }
64 
65 
66 void Instruction::setParent(BasicBlock *P) {
67   Parent = P;
68 }
69 
70 const Module *Instruction::getModule() const {
71   return getParent()->getModule();
72 }
73 
74 const Function *Instruction::getFunction() const {
75   return getParent()->getParent();
76 }
77 
78 void Instruction::removeFromParent() {
79   getParent()->getInstList().remove(getIterator());
80 }
81 
82 iplist<Instruction>::iterator Instruction::eraseFromParent() {
83   return getParent()->getInstList().erase(getIterator());
84 }
85 
86 /// Insert an unlinked instruction into a basic block immediately before the
87 /// specified instruction.
88 void Instruction::insertBefore(Instruction *InsertPos) {
89   insertInto(InsertPos->getParent(), InsertPos->getIterator());
90 }
91 
92 /// Insert an unlinked instruction into a basic block immediately after the
93 /// specified instruction.
94 void Instruction::insertAfter(Instruction *InsertPos) {
95   insertInto(InsertPos->getParent(), std::next(InsertPos->getIterator()));
96 }
97 
98 BasicBlock::iterator Instruction::insertInto(BasicBlock *ParentBB,
99                                              BasicBlock::iterator It) {
100   assert(getParent() == nullptr && "Expected detached instruction");
101   assert((It == ParentBB->end() || It->getParent() == ParentBB) &&
102          "It not in ParentBB");
103   return ParentBB->getInstList().insert(It, this);
104 }
105 
106 /// Unlink this instruction from its current basic block and insert it into the
107 /// basic block that MovePos lives in, right before MovePos.
108 void Instruction::moveBefore(Instruction *MovePos) {
109   moveBefore(*MovePos->getParent(), MovePos->getIterator());
110 }
111 
112 void Instruction::moveAfter(Instruction *MovePos) {
113   moveBefore(*MovePos->getParent(), ++MovePos->getIterator());
114 }
115 
116 void Instruction::moveBefore(BasicBlock &BB,
117                              SymbolTableList<Instruction>::iterator I) {
118   assert(I == BB.end() || I->getParent() == &BB);
119   BB.splice(I, getParent(), getIterator());
120 }
121 
122 bool Instruction::comesBefore(const Instruction *Other) const {
123   assert(Parent && Other->Parent &&
124          "instructions without BB parents have no order");
125   assert(Parent == Other->Parent && "cross-BB instruction order comparison");
126   if (!Parent->isInstrOrderValid())
127     Parent->renumberInstructions();
128   return Order < Other->Order;
129 }
130 
131 Instruction *Instruction::getInsertionPointAfterDef() {
132   assert(!getType()->isVoidTy() && "Instruction must define result");
133   BasicBlock *InsertBB;
134   BasicBlock::iterator InsertPt;
135   if (auto *PN = dyn_cast<PHINode>(this)) {
136     InsertBB = PN->getParent();
137     InsertPt = InsertBB->getFirstInsertionPt();
138   } else if (auto *II = dyn_cast<InvokeInst>(this)) {
139     InsertBB = II->getNormalDest();
140     InsertPt = InsertBB->getFirstInsertionPt();
141   } else if (auto *CB = dyn_cast<CallBrInst>(this)) {
142     InsertBB = CB->getDefaultDest();
143     InsertPt = InsertBB->getFirstInsertionPt();
144   } else {
145     assert(!isTerminator() && "Only invoke/callbr terminators return value");
146     InsertBB = getParent();
147     InsertPt = std::next(getIterator());
148   }
149 
150   // catchswitch blocks don't have any legal insertion point (because they
151   // are both an exception pad and a terminator).
152   if (InsertPt == InsertBB->end())
153     return nullptr;
154   return &*InsertPt;
155 }
156 
157 bool Instruction::isOnlyUserOfAnyOperand() {
158   return any_of(operands(), [](Value *V) { return V->hasOneUser(); });
159 }
160 
161 void Instruction::setHasNoUnsignedWrap(bool b) {
162   cast<OverflowingBinaryOperator>(this)->setHasNoUnsignedWrap(b);
163 }
164 
165 void Instruction::setHasNoSignedWrap(bool b) {
166   cast<OverflowingBinaryOperator>(this)->setHasNoSignedWrap(b);
167 }
168 
169 void Instruction::setIsExact(bool b) {
170   cast<PossiblyExactOperator>(this)->setIsExact(b);
171 }
172 
173 bool Instruction::hasNoUnsignedWrap() const {
174   return cast<OverflowingBinaryOperator>(this)->hasNoUnsignedWrap();
175 }
176 
177 bool Instruction::hasNoSignedWrap() const {
178   return cast<OverflowingBinaryOperator>(this)->hasNoSignedWrap();
179 }
180 
181 bool Instruction::hasPoisonGeneratingFlags() const {
182   return cast<Operator>(this)->hasPoisonGeneratingFlags();
183 }
184 
185 void Instruction::dropPoisonGeneratingFlags() {
186   switch (getOpcode()) {
187   case Instruction::Add:
188   case Instruction::Sub:
189   case Instruction::Mul:
190   case Instruction::Shl:
191     cast<OverflowingBinaryOperator>(this)->setHasNoUnsignedWrap(false);
192     cast<OverflowingBinaryOperator>(this)->setHasNoSignedWrap(false);
193     break;
194 
195   case Instruction::UDiv:
196   case Instruction::SDiv:
197   case Instruction::AShr:
198   case Instruction::LShr:
199     cast<PossiblyExactOperator>(this)->setIsExact(false);
200     break;
201 
202   case Instruction::GetElementPtr:
203     cast<GetElementPtrInst>(this)->setIsInBounds(false);
204     break;
205   }
206   if (isa<FPMathOperator>(this)) {
207     setHasNoNaNs(false);
208     setHasNoInfs(false);
209   }
210 
211   assert(!hasPoisonGeneratingFlags() && "must be kept in sync");
212 }
213 
214 bool Instruction::hasPoisonGeneratingMetadata() const {
215   return hasMetadata(LLVMContext::MD_range) ||
216          hasMetadata(LLVMContext::MD_nonnull) ||
217          hasMetadata(LLVMContext::MD_align);
218 }
219 
220 void Instruction::dropPoisonGeneratingMetadata() {
221   eraseMetadata(LLVMContext::MD_range);
222   eraseMetadata(LLVMContext::MD_nonnull);
223   eraseMetadata(LLVMContext::MD_align);
224 }
225 
226 void Instruction::dropUndefImplyingAttrsAndUnknownMetadata(
227     ArrayRef<unsigned> KnownIDs) {
228   dropUnknownNonDebugMetadata(KnownIDs);
229   auto *CB = dyn_cast<CallBase>(this);
230   if (!CB)
231     return;
232   // For call instructions, we also need to drop parameter and return attributes
233   // that are can cause UB if the call is moved to a location where the
234   // attribute is not valid.
235   AttributeList AL = CB->getAttributes();
236   if (AL.isEmpty())
237     return;
238   AttributeMask UBImplyingAttributes =
239       AttributeFuncs::getUBImplyingAttributes();
240   for (unsigned ArgNo = 0; ArgNo < CB->arg_size(); ArgNo++)
241     CB->removeParamAttrs(ArgNo, UBImplyingAttributes);
242   CB->removeRetAttrs(UBImplyingAttributes);
243 }
244 
245 bool Instruction::isExact() const {
246   return cast<PossiblyExactOperator>(this)->isExact();
247 }
248 
249 void Instruction::setFast(bool B) {
250   assert(isa<FPMathOperator>(this) && "setting fast-math flag on invalid op");
251   cast<FPMathOperator>(this)->setFast(B);
252 }
253 
254 void Instruction::setHasAllowReassoc(bool B) {
255   assert(isa<FPMathOperator>(this) && "setting fast-math flag on invalid op");
256   cast<FPMathOperator>(this)->setHasAllowReassoc(B);
257 }
258 
259 void Instruction::setHasNoNaNs(bool B) {
260   assert(isa<FPMathOperator>(this) && "setting fast-math flag on invalid op");
261   cast<FPMathOperator>(this)->setHasNoNaNs(B);
262 }
263 
264 void Instruction::setHasNoInfs(bool B) {
265   assert(isa<FPMathOperator>(this) && "setting fast-math flag on invalid op");
266   cast<FPMathOperator>(this)->setHasNoInfs(B);
267 }
268 
269 void Instruction::setHasNoSignedZeros(bool B) {
270   assert(isa<FPMathOperator>(this) && "setting fast-math flag on invalid op");
271   cast<FPMathOperator>(this)->setHasNoSignedZeros(B);
272 }
273 
274 void Instruction::setHasAllowReciprocal(bool B) {
275   assert(isa<FPMathOperator>(this) && "setting fast-math flag on invalid op");
276   cast<FPMathOperator>(this)->setHasAllowReciprocal(B);
277 }
278 
279 void Instruction::setHasAllowContract(bool B) {
280   assert(isa<FPMathOperator>(this) && "setting fast-math flag on invalid op");
281   cast<FPMathOperator>(this)->setHasAllowContract(B);
282 }
283 
284 void Instruction::setHasApproxFunc(bool B) {
285   assert(isa<FPMathOperator>(this) && "setting fast-math flag on invalid op");
286   cast<FPMathOperator>(this)->setHasApproxFunc(B);
287 }
288 
289 void Instruction::setFastMathFlags(FastMathFlags FMF) {
290   assert(isa<FPMathOperator>(this) && "setting fast-math flag on invalid op");
291   cast<FPMathOperator>(this)->setFastMathFlags(FMF);
292 }
293 
294 void Instruction::copyFastMathFlags(FastMathFlags FMF) {
295   assert(isa<FPMathOperator>(this) && "copying fast-math flag on invalid op");
296   cast<FPMathOperator>(this)->copyFastMathFlags(FMF);
297 }
298 
299 bool Instruction::isFast() const {
300   assert(isa<FPMathOperator>(this) && "getting fast-math flag on invalid op");
301   return cast<FPMathOperator>(this)->isFast();
302 }
303 
304 bool Instruction::hasAllowReassoc() const {
305   assert(isa<FPMathOperator>(this) && "getting fast-math flag on invalid op");
306   return cast<FPMathOperator>(this)->hasAllowReassoc();
307 }
308 
309 bool Instruction::hasNoNaNs() const {
310   assert(isa<FPMathOperator>(this) && "getting fast-math flag on invalid op");
311   return cast<FPMathOperator>(this)->hasNoNaNs();
312 }
313 
314 bool Instruction::hasNoInfs() const {
315   assert(isa<FPMathOperator>(this) && "getting fast-math flag on invalid op");
316   return cast<FPMathOperator>(this)->hasNoInfs();
317 }
318 
319 bool Instruction::hasNoSignedZeros() const {
320   assert(isa<FPMathOperator>(this) && "getting fast-math flag on invalid op");
321   return cast<FPMathOperator>(this)->hasNoSignedZeros();
322 }
323 
324 bool Instruction::hasAllowReciprocal() const {
325   assert(isa<FPMathOperator>(this) && "getting fast-math flag on invalid op");
326   return cast<FPMathOperator>(this)->hasAllowReciprocal();
327 }
328 
329 bool Instruction::hasAllowContract() const {
330   assert(isa<FPMathOperator>(this) && "getting fast-math flag on invalid op");
331   return cast<FPMathOperator>(this)->hasAllowContract();
332 }
333 
334 bool Instruction::hasApproxFunc() const {
335   assert(isa<FPMathOperator>(this) && "getting fast-math flag on invalid op");
336   return cast<FPMathOperator>(this)->hasApproxFunc();
337 }
338 
339 FastMathFlags Instruction::getFastMathFlags() const {
340   assert(isa<FPMathOperator>(this) && "getting fast-math flag on invalid op");
341   return cast<FPMathOperator>(this)->getFastMathFlags();
342 }
343 
344 void Instruction::copyFastMathFlags(const Instruction *I) {
345   copyFastMathFlags(I->getFastMathFlags());
346 }
347 
348 void Instruction::copyIRFlags(const Value *V, bool IncludeWrapFlags) {
349   // Copy the wrapping flags.
350   if (IncludeWrapFlags && isa<OverflowingBinaryOperator>(this)) {
351     if (auto *OB = dyn_cast<OverflowingBinaryOperator>(V)) {
352       setHasNoSignedWrap(OB->hasNoSignedWrap());
353       setHasNoUnsignedWrap(OB->hasNoUnsignedWrap());
354     }
355   }
356 
357   // Copy the exact flag.
358   if (auto *PE = dyn_cast<PossiblyExactOperator>(V))
359     if (isa<PossiblyExactOperator>(this))
360       setIsExact(PE->isExact());
361 
362   // Copy the fast-math flags.
363   if (auto *FP = dyn_cast<FPMathOperator>(V))
364     if (isa<FPMathOperator>(this))
365       copyFastMathFlags(FP->getFastMathFlags());
366 
367   if (auto *SrcGEP = dyn_cast<GetElementPtrInst>(V))
368     if (auto *DestGEP = dyn_cast<GetElementPtrInst>(this))
369       DestGEP->setIsInBounds(SrcGEP->isInBounds() || DestGEP->isInBounds());
370 }
371 
372 void Instruction::andIRFlags(const Value *V) {
373   if (auto *OB = dyn_cast<OverflowingBinaryOperator>(V)) {
374     if (isa<OverflowingBinaryOperator>(this)) {
375       setHasNoSignedWrap(hasNoSignedWrap() && OB->hasNoSignedWrap());
376       setHasNoUnsignedWrap(hasNoUnsignedWrap() && OB->hasNoUnsignedWrap());
377     }
378   }
379 
380   if (auto *PE = dyn_cast<PossiblyExactOperator>(V))
381     if (isa<PossiblyExactOperator>(this))
382       setIsExact(isExact() && PE->isExact());
383 
384   if (auto *FP = dyn_cast<FPMathOperator>(V)) {
385     if (isa<FPMathOperator>(this)) {
386       FastMathFlags FM = getFastMathFlags();
387       FM &= FP->getFastMathFlags();
388       copyFastMathFlags(FM);
389     }
390   }
391 
392   if (auto *SrcGEP = dyn_cast<GetElementPtrInst>(V))
393     if (auto *DestGEP = dyn_cast<GetElementPtrInst>(this))
394       DestGEP->setIsInBounds(SrcGEP->isInBounds() && DestGEP->isInBounds());
395 }
396 
397 const char *Instruction::getOpcodeName(unsigned OpCode) {
398   switch (OpCode) {
399   // Terminators
400   case Ret:    return "ret";
401   case Br:     return "br";
402   case Switch: return "switch";
403   case IndirectBr: return "indirectbr";
404   case Invoke: return "invoke";
405   case Resume: return "resume";
406   case Unreachable: return "unreachable";
407   case CleanupRet: return "cleanupret";
408   case CatchRet: return "catchret";
409   case CatchPad: return "catchpad";
410   case CatchSwitch: return "catchswitch";
411   case CallBr: return "callbr";
412 
413   // Standard unary operators...
414   case FNeg: return "fneg";
415 
416   // Standard binary operators...
417   case Add: return "add";
418   case FAdd: return "fadd";
419   case Sub: return "sub";
420   case FSub: return "fsub";
421   case Mul: return "mul";
422   case FMul: return "fmul";
423   case UDiv: return "udiv";
424   case SDiv: return "sdiv";
425   case FDiv: return "fdiv";
426   case URem: return "urem";
427   case SRem: return "srem";
428   case FRem: return "frem";
429 
430   // Logical operators...
431   case And: return "and";
432   case Or : return "or";
433   case Xor: return "xor";
434 
435   // Memory instructions...
436   case Alloca:        return "alloca";
437   case Load:          return "load";
438   case Store:         return "store";
439   case AtomicCmpXchg: return "cmpxchg";
440   case AtomicRMW:     return "atomicrmw";
441   case Fence:         return "fence";
442   case GetElementPtr: return "getelementptr";
443 
444   // Convert instructions...
445   case Trunc:         return "trunc";
446   case ZExt:          return "zext";
447   case SExt:          return "sext";
448   case FPTrunc:       return "fptrunc";
449   case FPExt:         return "fpext";
450   case FPToUI:        return "fptoui";
451   case FPToSI:        return "fptosi";
452   case UIToFP:        return "uitofp";
453   case SIToFP:        return "sitofp";
454   case IntToPtr:      return "inttoptr";
455   case PtrToInt:      return "ptrtoint";
456   case BitCast:       return "bitcast";
457   case AddrSpaceCast: return "addrspacecast";
458 
459   // Other instructions...
460   case ICmp:           return "icmp";
461   case FCmp:           return "fcmp";
462   case PHI:            return "phi";
463   case Select:         return "select";
464   case Call:           return "call";
465   case Shl:            return "shl";
466   case LShr:           return "lshr";
467   case AShr:           return "ashr";
468   case VAArg:          return "va_arg";
469   case ExtractElement: return "extractelement";
470   case InsertElement:  return "insertelement";
471   case ShuffleVector:  return "shufflevector";
472   case ExtractValue:   return "extractvalue";
473   case InsertValue:    return "insertvalue";
474   case LandingPad:     return "landingpad";
475   case CleanupPad:     return "cleanuppad";
476   case Freeze:         return "freeze";
477 
478   default: return "<Invalid operator> ";
479   }
480 }
481 
482 /// Return true if both instructions have the same special state. This must be
483 /// kept in sync with FunctionComparator::cmpOperations in
484 /// lib/Transforms/IPO/MergeFunctions.cpp.
485 static bool haveSameSpecialState(const Instruction *I1, const Instruction *I2,
486                                  bool IgnoreAlignment = false) {
487   assert(I1->getOpcode() == I2->getOpcode() &&
488          "Can not compare special state of different instructions");
489 
490   if (const AllocaInst *AI = dyn_cast<AllocaInst>(I1))
491     return AI->getAllocatedType() == cast<AllocaInst>(I2)->getAllocatedType() &&
492            (AI->getAlign() == cast<AllocaInst>(I2)->getAlign() ||
493             IgnoreAlignment);
494   if (const LoadInst *LI = dyn_cast<LoadInst>(I1))
495     return LI->isVolatile() == cast<LoadInst>(I2)->isVolatile() &&
496            (LI->getAlign() == cast<LoadInst>(I2)->getAlign() ||
497             IgnoreAlignment) &&
498            LI->getOrdering() == cast<LoadInst>(I2)->getOrdering() &&
499            LI->getSyncScopeID() == cast<LoadInst>(I2)->getSyncScopeID();
500   if (const StoreInst *SI = dyn_cast<StoreInst>(I1))
501     return SI->isVolatile() == cast<StoreInst>(I2)->isVolatile() &&
502            (SI->getAlign() == cast<StoreInst>(I2)->getAlign() ||
503             IgnoreAlignment) &&
504            SI->getOrdering() == cast<StoreInst>(I2)->getOrdering() &&
505            SI->getSyncScopeID() == cast<StoreInst>(I2)->getSyncScopeID();
506   if (const CmpInst *CI = dyn_cast<CmpInst>(I1))
507     return CI->getPredicate() == cast<CmpInst>(I2)->getPredicate();
508   if (const CallInst *CI = dyn_cast<CallInst>(I1))
509     return CI->isTailCall() == cast<CallInst>(I2)->isTailCall() &&
510            CI->getCallingConv() == cast<CallInst>(I2)->getCallingConv() &&
511            CI->getAttributes() == cast<CallInst>(I2)->getAttributes() &&
512            CI->hasIdenticalOperandBundleSchema(*cast<CallInst>(I2));
513   if (const InvokeInst *CI = dyn_cast<InvokeInst>(I1))
514     return CI->getCallingConv() == cast<InvokeInst>(I2)->getCallingConv() &&
515            CI->getAttributes() == cast<InvokeInst>(I2)->getAttributes() &&
516            CI->hasIdenticalOperandBundleSchema(*cast<InvokeInst>(I2));
517   if (const CallBrInst *CI = dyn_cast<CallBrInst>(I1))
518     return CI->getCallingConv() == cast<CallBrInst>(I2)->getCallingConv() &&
519            CI->getAttributes() == cast<CallBrInst>(I2)->getAttributes() &&
520            CI->hasIdenticalOperandBundleSchema(*cast<CallBrInst>(I2));
521   if (const InsertValueInst *IVI = dyn_cast<InsertValueInst>(I1))
522     return IVI->getIndices() == cast<InsertValueInst>(I2)->getIndices();
523   if (const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(I1))
524     return EVI->getIndices() == cast<ExtractValueInst>(I2)->getIndices();
525   if (const FenceInst *FI = dyn_cast<FenceInst>(I1))
526     return FI->getOrdering() == cast<FenceInst>(I2)->getOrdering() &&
527            FI->getSyncScopeID() == cast<FenceInst>(I2)->getSyncScopeID();
528   if (const AtomicCmpXchgInst *CXI = dyn_cast<AtomicCmpXchgInst>(I1))
529     return CXI->isVolatile() == cast<AtomicCmpXchgInst>(I2)->isVolatile() &&
530            CXI->isWeak() == cast<AtomicCmpXchgInst>(I2)->isWeak() &&
531            CXI->getSuccessOrdering() ==
532                cast<AtomicCmpXchgInst>(I2)->getSuccessOrdering() &&
533            CXI->getFailureOrdering() ==
534                cast<AtomicCmpXchgInst>(I2)->getFailureOrdering() &&
535            CXI->getSyncScopeID() ==
536                cast<AtomicCmpXchgInst>(I2)->getSyncScopeID();
537   if (const AtomicRMWInst *RMWI = dyn_cast<AtomicRMWInst>(I1))
538     return RMWI->getOperation() == cast<AtomicRMWInst>(I2)->getOperation() &&
539            RMWI->isVolatile() == cast<AtomicRMWInst>(I2)->isVolatile() &&
540            RMWI->getOrdering() == cast<AtomicRMWInst>(I2)->getOrdering() &&
541            RMWI->getSyncScopeID() == cast<AtomicRMWInst>(I2)->getSyncScopeID();
542   if (const ShuffleVectorInst *SVI = dyn_cast<ShuffleVectorInst>(I1))
543     return SVI->getShuffleMask() ==
544            cast<ShuffleVectorInst>(I2)->getShuffleMask();
545   if (const GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I1))
546     return GEP->getSourceElementType() ==
547            cast<GetElementPtrInst>(I2)->getSourceElementType();
548 
549   return true;
550 }
551 
552 bool Instruction::isIdenticalTo(const Instruction *I) const {
553   return isIdenticalToWhenDefined(I) &&
554          SubclassOptionalData == I->SubclassOptionalData;
555 }
556 
557 bool Instruction::isIdenticalToWhenDefined(const Instruction *I) const {
558   if (getOpcode() != I->getOpcode() ||
559       getNumOperands() != I->getNumOperands() ||
560       getType() != I->getType())
561     return false;
562 
563   // If both instructions have no operands, they are identical.
564   if (getNumOperands() == 0 && I->getNumOperands() == 0)
565     return haveSameSpecialState(this, I);
566 
567   // We have two instructions of identical opcode and #operands.  Check to see
568   // if all operands are the same.
569   if (!std::equal(op_begin(), op_end(), I->op_begin()))
570     return false;
571 
572   // WARNING: this logic must be kept in sync with EliminateDuplicatePHINodes()!
573   if (const PHINode *thisPHI = dyn_cast<PHINode>(this)) {
574     const PHINode *otherPHI = cast<PHINode>(I);
575     return std::equal(thisPHI->block_begin(), thisPHI->block_end(),
576                       otherPHI->block_begin());
577   }
578 
579   return haveSameSpecialState(this, I);
580 }
581 
582 // Keep this in sync with FunctionComparator::cmpOperations in
583 // lib/Transforms/IPO/MergeFunctions.cpp.
584 bool Instruction::isSameOperationAs(const Instruction *I,
585                                     unsigned flags) const {
586   bool IgnoreAlignment = flags & CompareIgnoringAlignment;
587   bool UseScalarTypes  = flags & CompareUsingScalarTypes;
588 
589   if (getOpcode() != I->getOpcode() ||
590       getNumOperands() != I->getNumOperands() ||
591       (UseScalarTypes ?
592        getType()->getScalarType() != I->getType()->getScalarType() :
593        getType() != I->getType()))
594     return false;
595 
596   // We have two instructions of identical opcode and #operands.  Check to see
597   // if all operands are the same type
598   for (unsigned i = 0, e = getNumOperands(); i != e; ++i)
599     if (UseScalarTypes ?
600         getOperand(i)->getType()->getScalarType() !=
601           I->getOperand(i)->getType()->getScalarType() :
602         getOperand(i)->getType() != I->getOperand(i)->getType())
603       return false;
604 
605   return haveSameSpecialState(this, I, IgnoreAlignment);
606 }
607 
608 bool Instruction::isUsedOutsideOfBlock(const BasicBlock *BB) const {
609   for (const Use &U : uses()) {
610     // PHI nodes uses values in the corresponding predecessor block.  For other
611     // instructions, just check to see whether the parent of the use matches up.
612     const Instruction *I = cast<Instruction>(U.getUser());
613     const PHINode *PN = dyn_cast<PHINode>(I);
614     if (!PN) {
615       if (I->getParent() != BB)
616         return true;
617       continue;
618     }
619 
620     if (PN->getIncomingBlock(U) != BB)
621       return true;
622   }
623   return false;
624 }
625 
626 bool Instruction::mayReadFromMemory() const {
627   switch (getOpcode()) {
628   default: return false;
629   case Instruction::VAArg:
630   case Instruction::Load:
631   case Instruction::Fence: // FIXME: refine definition of mayReadFromMemory
632   case Instruction::AtomicCmpXchg:
633   case Instruction::AtomicRMW:
634   case Instruction::CatchPad:
635   case Instruction::CatchRet:
636     return true;
637   case Instruction::Call:
638   case Instruction::Invoke:
639   case Instruction::CallBr:
640     return !cast<CallBase>(this)->onlyWritesMemory();
641   case Instruction::Store:
642     return !cast<StoreInst>(this)->isUnordered();
643   }
644 }
645 
646 bool Instruction::mayWriteToMemory() const {
647   switch (getOpcode()) {
648   default: return false;
649   case Instruction::Fence: // FIXME: refine definition of mayWriteToMemory
650   case Instruction::Store:
651   case Instruction::VAArg:
652   case Instruction::AtomicCmpXchg:
653   case Instruction::AtomicRMW:
654   case Instruction::CatchPad:
655   case Instruction::CatchRet:
656     return true;
657   case Instruction::Call:
658   case Instruction::Invoke:
659   case Instruction::CallBr:
660     return !cast<CallBase>(this)->onlyReadsMemory();
661   case Instruction::Load:
662     return !cast<LoadInst>(this)->isUnordered();
663   }
664 }
665 
666 bool Instruction::isAtomic() const {
667   switch (getOpcode()) {
668   default:
669     return false;
670   case Instruction::AtomicCmpXchg:
671   case Instruction::AtomicRMW:
672   case Instruction::Fence:
673     return true;
674   case Instruction::Load:
675     return cast<LoadInst>(this)->getOrdering() != AtomicOrdering::NotAtomic;
676   case Instruction::Store:
677     return cast<StoreInst>(this)->getOrdering() != AtomicOrdering::NotAtomic;
678   }
679 }
680 
681 bool Instruction::hasAtomicLoad() const {
682   assert(isAtomic());
683   switch (getOpcode()) {
684   default:
685     return false;
686   case Instruction::AtomicCmpXchg:
687   case Instruction::AtomicRMW:
688   case Instruction::Load:
689     return true;
690   }
691 }
692 
693 bool Instruction::hasAtomicStore() const {
694   assert(isAtomic());
695   switch (getOpcode()) {
696   default:
697     return false;
698   case Instruction::AtomicCmpXchg:
699   case Instruction::AtomicRMW:
700   case Instruction::Store:
701     return true;
702   }
703 }
704 
705 bool Instruction::isVolatile() const {
706   switch (getOpcode()) {
707   default:
708     return false;
709   case Instruction::AtomicRMW:
710     return cast<AtomicRMWInst>(this)->isVolatile();
711   case Instruction::Store:
712     return cast<StoreInst>(this)->isVolatile();
713   case Instruction::Load:
714     return cast<LoadInst>(this)->isVolatile();
715   case Instruction::AtomicCmpXchg:
716     return cast<AtomicCmpXchgInst>(this)->isVolatile();
717   case Instruction::Call:
718   case Instruction::Invoke:
719     // There are a very limited number of intrinsics with volatile flags.
720     if (auto *II = dyn_cast<IntrinsicInst>(this)) {
721       if (auto *MI = dyn_cast<MemIntrinsic>(II))
722         return MI->isVolatile();
723       switch (II->getIntrinsicID()) {
724       default: break;
725       case Intrinsic::matrix_column_major_load:
726         return cast<ConstantInt>(II->getArgOperand(2))->isOne();
727       case Intrinsic::matrix_column_major_store:
728         return cast<ConstantInt>(II->getArgOperand(3))->isOne();
729       }
730     }
731     return false;
732   }
733 }
734 
735 bool Instruction::mayThrow() const {
736   if (const CallInst *CI = dyn_cast<CallInst>(this))
737     return !CI->doesNotThrow();
738   if (const auto *CRI = dyn_cast<CleanupReturnInst>(this))
739     return CRI->unwindsToCaller();
740   if (const auto *CatchSwitch = dyn_cast<CatchSwitchInst>(this))
741     return CatchSwitch->unwindsToCaller();
742   return isa<ResumeInst>(this);
743 }
744 
745 bool Instruction::mayHaveSideEffects() const {
746   return mayWriteToMemory() || mayThrow() || !willReturn();
747 }
748 
749 bool Instruction::isSafeToRemove() const {
750   return (!isa<CallInst>(this) || !this->mayHaveSideEffects()) &&
751          !this->isTerminator() && !this->isEHPad();
752 }
753 
754 bool Instruction::willReturn() const {
755   // Volatile store isn't guaranteed to return; see LangRef.
756   if (auto *SI = dyn_cast<StoreInst>(this))
757     return !SI->isVolatile();
758 
759   if (const auto *CB = dyn_cast<CallBase>(this))
760     return CB->hasFnAttr(Attribute::WillReturn);
761   return true;
762 }
763 
764 bool Instruction::isLifetimeStartOrEnd() const {
765   auto *II = dyn_cast<IntrinsicInst>(this);
766   if (!II)
767     return false;
768   Intrinsic::ID ID = II->getIntrinsicID();
769   return ID == Intrinsic::lifetime_start || ID == Intrinsic::lifetime_end;
770 }
771 
772 bool Instruction::isLaunderOrStripInvariantGroup() const {
773   auto *II = dyn_cast<IntrinsicInst>(this);
774   if (!II)
775     return false;
776   Intrinsic::ID ID = II->getIntrinsicID();
777   return ID == Intrinsic::launder_invariant_group ||
778          ID == Intrinsic::strip_invariant_group;
779 }
780 
781 bool Instruction::isDebugOrPseudoInst() const {
782   return isa<DbgInfoIntrinsic>(this) || isa<PseudoProbeInst>(this);
783 }
784 
785 const Instruction *
786 Instruction::getNextNonDebugInstruction(bool SkipPseudoOp) const {
787   for (const Instruction *I = getNextNode(); I; I = I->getNextNode())
788     if (!isa<DbgInfoIntrinsic>(I) && !(SkipPseudoOp && isa<PseudoProbeInst>(I)))
789       return I;
790   return nullptr;
791 }
792 
793 const Instruction *
794 Instruction::getPrevNonDebugInstruction(bool SkipPseudoOp) const {
795   for (const Instruction *I = getPrevNode(); I; I = I->getPrevNode())
796     if (!isa<DbgInfoIntrinsic>(I) && !(SkipPseudoOp && isa<PseudoProbeInst>(I)))
797       return I;
798   return nullptr;
799 }
800 
801 bool Instruction::isAssociative() const {
802   unsigned Opcode = getOpcode();
803   if (isAssociative(Opcode))
804     return true;
805 
806   switch (Opcode) {
807   case FMul:
808   case FAdd:
809     return cast<FPMathOperator>(this)->hasAllowReassoc() &&
810            cast<FPMathOperator>(this)->hasNoSignedZeros();
811   default:
812     return false;
813   }
814 }
815 
816 bool Instruction::isCommutative() const {
817   if (auto *II = dyn_cast<IntrinsicInst>(this))
818     return II->isCommutative();
819   // TODO: Should allow icmp/fcmp?
820   return isCommutative(getOpcode());
821 }
822 
823 unsigned Instruction::getNumSuccessors() const {
824   switch (getOpcode()) {
825 #define HANDLE_TERM_INST(N, OPC, CLASS)                                        \
826   case Instruction::OPC:                                                       \
827     return static_cast<const CLASS *>(this)->getNumSuccessors();
828 #include "llvm/IR/Instruction.def"
829   default:
830     break;
831   }
832   llvm_unreachable("not a terminator");
833 }
834 
835 BasicBlock *Instruction::getSuccessor(unsigned idx) const {
836   switch (getOpcode()) {
837 #define HANDLE_TERM_INST(N, OPC, CLASS)                                        \
838   case Instruction::OPC:                                                       \
839     return static_cast<const CLASS *>(this)->getSuccessor(idx);
840 #include "llvm/IR/Instruction.def"
841   default:
842     break;
843   }
844   llvm_unreachable("not a terminator");
845 }
846 
847 void Instruction::setSuccessor(unsigned idx, BasicBlock *B) {
848   switch (getOpcode()) {
849 #define HANDLE_TERM_INST(N, OPC, CLASS)                                        \
850   case Instruction::OPC:                                                       \
851     return static_cast<CLASS *>(this)->setSuccessor(idx, B);
852 #include "llvm/IR/Instruction.def"
853   default:
854     break;
855   }
856   llvm_unreachable("not a terminator");
857 }
858 
859 void Instruction::replaceSuccessorWith(BasicBlock *OldBB, BasicBlock *NewBB) {
860   for (unsigned Idx = 0, NumSuccessors = Instruction::getNumSuccessors();
861        Idx != NumSuccessors; ++Idx)
862     if (getSuccessor(Idx) == OldBB)
863       setSuccessor(Idx, NewBB);
864 }
865 
866 Instruction *Instruction::cloneImpl() const {
867   llvm_unreachable("Subclass of Instruction failed to implement cloneImpl");
868 }
869 
870 void Instruction::swapProfMetadata() {
871   MDNode *ProfileData = getBranchWeightMDNode(*this);
872   if (!ProfileData || ProfileData->getNumOperands() != 3)
873     return;
874 
875   // The first operand is the name. Fetch them backwards and build a new one.
876   Metadata *Ops[] = {ProfileData->getOperand(0), ProfileData->getOperand(2),
877                      ProfileData->getOperand(1)};
878   setMetadata(LLVMContext::MD_prof,
879               MDNode::get(ProfileData->getContext(), Ops));
880 }
881 
882 void Instruction::copyMetadata(const Instruction &SrcInst,
883                                ArrayRef<unsigned> WL) {
884   if (!SrcInst.hasMetadata())
885     return;
886 
887   DenseSet<unsigned> WLS;
888   for (unsigned M : WL)
889     WLS.insert(M);
890 
891   // Otherwise, enumerate and copy over metadata from the old instruction to the
892   // new one.
893   SmallVector<std::pair<unsigned, MDNode *>, 4> TheMDs;
894   SrcInst.getAllMetadataOtherThanDebugLoc(TheMDs);
895   for (const auto &MD : TheMDs) {
896     if (WL.empty() || WLS.count(MD.first))
897       setMetadata(MD.first, MD.second);
898   }
899   if (WL.empty() || WLS.count(LLVMContext::MD_dbg))
900     setDebugLoc(SrcInst.getDebugLoc());
901 }
902 
903 Instruction *Instruction::clone() const {
904   Instruction *New = nullptr;
905   switch (getOpcode()) {
906   default:
907     llvm_unreachable("Unhandled Opcode.");
908 #define HANDLE_INST(num, opc, clas)                                            \
909   case Instruction::opc:                                                       \
910     New = cast<clas>(this)->cloneImpl();                                       \
911     break;
912 #include "llvm/IR/Instruction.def"
913 #undef HANDLE_INST
914   }
915 
916   New->SubclassOptionalData = SubclassOptionalData;
917   New->copyMetadata(*this);
918   return New;
919 }
920