• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 //===-- Analysis.cpp - CodeGen LLVM IR Analysis Utilities -----------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file defines several CodeGen-specific LLVM IR analysis utilities.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "llvm/CodeGen/Analysis.h"
15 #include "llvm/Analysis/ValueTracking.h"
16 #include "llvm/CodeGen/MachineFunction.h"
17 #include "llvm/CodeGen/TargetInstrInfo.h"
18 #include "llvm/CodeGen/TargetLowering.h"
19 #include "llvm/CodeGen/TargetSubtargetInfo.h"
20 #include "llvm/IR/DataLayout.h"
21 #include "llvm/IR/DerivedTypes.h"
22 #include "llvm/IR/Function.h"
23 #include "llvm/IR/Instructions.h"
24 #include "llvm/IR/IntrinsicInst.h"
25 #include "llvm/IR/LLVMContext.h"
26 #include "llvm/IR/Module.h"
27 #include "llvm/Support/ErrorHandling.h"
28 #include "llvm/Support/MathExtras.h"
29 #include "llvm/Transforms/Utils/GlobalStatus.h"
30 
31 using namespace llvm;
32 
33 /// Compute the linearized index of a member in a nested aggregate/struct/array
34 /// by recursing and accumulating CurIndex as long as there are indices in the
35 /// index list.
ComputeLinearIndex(Type * Ty,const unsigned * Indices,const unsigned * IndicesEnd,unsigned CurIndex)36 unsigned llvm::ComputeLinearIndex(Type *Ty,
37                                   const unsigned *Indices,
38                                   const unsigned *IndicesEnd,
39                                   unsigned CurIndex) {
40   // Base case: We're done.
41   if (Indices && Indices == IndicesEnd)
42     return CurIndex;
43 
44   // Given a struct type, recursively traverse the elements.
45   if (StructType *STy = dyn_cast<StructType>(Ty)) {
46     for (StructType::element_iterator EB = STy->element_begin(),
47                                       EI = EB,
48                                       EE = STy->element_end();
49         EI != EE; ++EI) {
50       if (Indices && *Indices == unsigned(EI - EB))
51         return ComputeLinearIndex(*EI, Indices+1, IndicesEnd, CurIndex);
52       CurIndex = ComputeLinearIndex(*EI, nullptr, nullptr, CurIndex);
53     }
54     assert(!Indices && "Unexpected out of bound");
55     return CurIndex;
56   }
57   // Given an array type, recursively traverse the elements.
58   else if (ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
59     Type *EltTy = ATy->getElementType();
60     unsigned NumElts = ATy->getNumElements();
61     // Compute the Linear offset when jumping one element of the array
62     unsigned EltLinearOffset = ComputeLinearIndex(EltTy, nullptr, nullptr, 0);
63     if (Indices) {
64       assert(*Indices < NumElts && "Unexpected out of bound");
65       // If the indice is inside the array, compute the index to the requested
66       // elt and recurse inside the element with the end of the indices list
67       CurIndex += EltLinearOffset* *Indices;
68       return ComputeLinearIndex(EltTy, Indices+1, IndicesEnd, CurIndex);
69     }
70     CurIndex += EltLinearOffset*NumElts;
71     return CurIndex;
72   }
73   // We haven't found the type we're looking for, so keep searching.
74   return CurIndex + 1;
75 }
76 
77 /// ComputeValueVTs - Given an LLVM IR type, compute a sequence of
78 /// EVTs that represent all the individual underlying
79 /// non-aggregate types that comprise it.
80 ///
81 /// If Offsets is non-null, it points to a vector to be filled in
82 /// with the in-memory offsets of each of the individual values.
83 ///
ComputeValueVTs(const TargetLowering & TLI,const DataLayout & DL,Type * Ty,SmallVectorImpl<EVT> & ValueVTs,SmallVectorImpl<uint64_t> * Offsets,uint64_t StartingOffset)84 void llvm::ComputeValueVTs(const TargetLowering &TLI, const DataLayout &DL,
85                            Type *Ty, SmallVectorImpl<EVT> &ValueVTs,
86                            SmallVectorImpl<uint64_t> *Offsets,
87                            uint64_t StartingOffset) {
88   // Given a struct type, recursively traverse the elements.
89   if (StructType *STy = dyn_cast<StructType>(Ty)) {
90     const StructLayout *SL = DL.getStructLayout(STy);
91     for (StructType::element_iterator EB = STy->element_begin(),
92                                       EI = EB,
93                                       EE = STy->element_end();
94          EI != EE; ++EI)
95       ComputeValueVTs(TLI, DL, *EI, ValueVTs, Offsets,
96                       StartingOffset + SL->getElementOffset(EI - EB));
97     return;
98   }
99   // Given an array type, recursively traverse the elements.
100   if (ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
101     Type *EltTy = ATy->getElementType();
102     uint64_t EltSize = DL.getTypeAllocSize(EltTy);
103     for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
104       ComputeValueVTs(TLI, DL, EltTy, ValueVTs, Offsets,
105                       StartingOffset + i * EltSize);
106     return;
107   }
108   // Interpret void as zero return values.
109   if (Ty->isVoidTy())
110     return;
111   // Base case: we can get an EVT for this LLVM IR type.
112   ValueVTs.push_back(TLI.getValueType(DL, Ty));
113   if (Offsets)
114     Offsets->push_back(StartingOffset);
115 }
116 
117 /// ExtractTypeInfo - Returns the type info, possibly bitcast, encoded in V.
ExtractTypeInfo(Value * V)118 GlobalValue *llvm::ExtractTypeInfo(Value *V) {
119   V = V->stripPointerCasts();
120   GlobalValue *GV = dyn_cast<GlobalValue>(V);
121   GlobalVariable *Var = dyn_cast<GlobalVariable>(V);
122 
123   if (Var && Var->getName() == "llvm.eh.catch.all.value") {
124     assert(Var->hasInitializer() &&
125            "The EH catch-all value must have an initializer");
126     Value *Init = Var->getInitializer();
127     GV = dyn_cast<GlobalValue>(Init);
128     if (!GV) V = cast<ConstantPointerNull>(Init);
129   }
130 
131   assert((GV || isa<ConstantPointerNull>(V)) &&
132          "TypeInfo must be a global variable or NULL");
133   return GV;
134 }
135 
136 /// hasInlineAsmMemConstraint - Return true if the inline asm instruction being
137 /// processed uses a memory 'm' constraint.
138 bool
hasInlineAsmMemConstraint(InlineAsm::ConstraintInfoVector & CInfos,const TargetLowering & TLI)139 llvm::hasInlineAsmMemConstraint(InlineAsm::ConstraintInfoVector &CInfos,
140                                 const TargetLowering &TLI) {
141   for (unsigned i = 0, e = CInfos.size(); i != e; ++i) {
142     InlineAsm::ConstraintInfo &CI = CInfos[i];
143     for (unsigned j = 0, ee = CI.Codes.size(); j != ee; ++j) {
144       TargetLowering::ConstraintType CType = TLI.getConstraintType(CI.Codes[j]);
145       if (CType == TargetLowering::C_Memory)
146         return true;
147     }
148 
149     // Indirect operand accesses access memory.
150     if (CI.isIndirect)
151       return true;
152   }
153 
154   return false;
155 }
156 
157 /// getFCmpCondCode - Return the ISD condition code corresponding to
158 /// the given LLVM IR floating-point condition code.  This includes
159 /// consideration of global floating-point math flags.
160 ///
getFCmpCondCode(FCmpInst::Predicate Pred)161 ISD::CondCode llvm::getFCmpCondCode(FCmpInst::Predicate Pred) {
162   switch (Pred) {
163   case FCmpInst::FCMP_FALSE: return ISD::SETFALSE;
164   case FCmpInst::FCMP_OEQ:   return ISD::SETOEQ;
165   case FCmpInst::FCMP_OGT:   return ISD::SETOGT;
166   case FCmpInst::FCMP_OGE:   return ISD::SETOGE;
167   case FCmpInst::FCMP_OLT:   return ISD::SETOLT;
168   case FCmpInst::FCMP_OLE:   return ISD::SETOLE;
169   case FCmpInst::FCMP_ONE:   return ISD::SETONE;
170   case FCmpInst::FCMP_ORD:   return ISD::SETO;
171   case FCmpInst::FCMP_UNO:   return ISD::SETUO;
172   case FCmpInst::FCMP_UEQ:   return ISD::SETUEQ;
173   case FCmpInst::FCMP_UGT:   return ISD::SETUGT;
174   case FCmpInst::FCMP_UGE:   return ISD::SETUGE;
175   case FCmpInst::FCMP_ULT:   return ISD::SETULT;
176   case FCmpInst::FCMP_ULE:   return ISD::SETULE;
177   case FCmpInst::FCMP_UNE:   return ISD::SETUNE;
178   case FCmpInst::FCMP_TRUE:  return ISD::SETTRUE;
179   default: llvm_unreachable("Invalid FCmp predicate opcode!");
180   }
181 }
182 
getFCmpCodeWithoutNaN(ISD::CondCode CC)183 ISD::CondCode llvm::getFCmpCodeWithoutNaN(ISD::CondCode CC) {
184   switch (CC) {
185     case ISD::SETOEQ: case ISD::SETUEQ: return ISD::SETEQ;
186     case ISD::SETONE: case ISD::SETUNE: return ISD::SETNE;
187     case ISD::SETOLT: case ISD::SETULT: return ISD::SETLT;
188     case ISD::SETOLE: case ISD::SETULE: return ISD::SETLE;
189     case ISD::SETOGT: case ISD::SETUGT: return ISD::SETGT;
190     case ISD::SETOGE: case ISD::SETUGE: return ISD::SETGE;
191     default: return CC;
192   }
193 }
194 
195 /// getICmpCondCode - Return the ISD condition code corresponding to
196 /// the given LLVM IR integer condition code.
197 ///
getICmpCondCode(ICmpInst::Predicate Pred)198 ISD::CondCode llvm::getICmpCondCode(ICmpInst::Predicate Pred) {
199   switch (Pred) {
200   case ICmpInst::ICMP_EQ:  return ISD::SETEQ;
201   case ICmpInst::ICMP_NE:  return ISD::SETNE;
202   case ICmpInst::ICMP_SLE: return ISD::SETLE;
203   case ICmpInst::ICMP_ULE: return ISD::SETULE;
204   case ICmpInst::ICMP_SGE: return ISD::SETGE;
205   case ICmpInst::ICMP_UGE: return ISD::SETUGE;
206   case ICmpInst::ICMP_SLT: return ISD::SETLT;
207   case ICmpInst::ICMP_ULT: return ISD::SETULT;
208   case ICmpInst::ICMP_SGT: return ISD::SETGT;
209   case ICmpInst::ICMP_UGT: return ISD::SETUGT;
210   default:
211     llvm_unreachable("Invalid ICmp predicate opcode!");
212   }
213 }
214 
isNoopBitcast(Type * T1,Type * T2,const TargetLoweringBase & TLI)215 static bool isNoopBitcast(Type *T1, Type *T2,
216                           const TargetLoweringBase& TLI) {
217   return T1 == T2 || (T1->isPointerTy() && T2->isPointerTy()) ||
218          (isa<VectorType>(T1) && isa<VectorType>(T2) &&
219           TLI.isTypeLegal(EVT::getEVT(T1)) && TLI.isTypeLegal(EVT::getEVT(T2)));
220 }
221 
222 /// Look through operations that will be free to find the earliest source of
223 /// this value.
224 ///
225 /// @param ValLoc If V has aggegate type, we will be interested in a particular
226 /// scalar component. This records its address; the reverse of this list gives a
227 /// sequence of indices appropriate for an extractvalue to locate the important
228 /// value. This value is updated during the function and on exit will indicate
229 /// similar information for the Value returned.
230 ///
231 /// @param DataBits If this function looks through truncate instructions, this
232 /// will record the smallest size attained.
getNoopInput(const Value * V,SmallVectorImpl<unsigned> & ValLoc,unsigned & DataBits,const TargetLoweringBase & TLI,const DataLayout & DL)233 static const Value *getNoopInput(const Value *V,
234                                  SmallVectorImpl<unsigned> &ValLoc,
235                                  unsigned &DataBits,
236                                  const TargetLoweringBase &TLI,
237                                  const DataLayout &DL) {
238   while (true) {
239     // Try to look through V1; if V1 is not an instruction, it can't be looked
240     // through.
241     const Instruction *I = dyn_cast<Instruction>(V);
242     if (!I || I->getNumOperands() == 0) return V;
243     const Value *NoopInput = nullptr;
244 
245     Value *Op = I->getOperand(0);
246     if (isa<BitCastInst>(I)) {
247       // Look through truly no-op bitcasts.
248       if (isNoopBitcast(Op->getType(), I->getType(), TLI))
249         NoopInput = Op;
250     } else if (isa<GetElementPtrInst>(I)) {
251       // Look through getelementptr
252       if (cast<GetElementPtrInst>(I)->hasAllZeroIndices())
253         NoopInput = Op;
254     } else if (isa<IntToPtrInst>(I)) {
255       // Look through inttoptr.
256       // Make sure this isn't a truncating or extending cast.  We could
257       // support this eventually, but don't bother for now.
258       if (!isa<VectorType>(I->getType()) &&
259           DL.getPointerSizeInBits() ==
260               cast<IntegerType>(Op->getType())->getBitWidth())
261         NoopInput = Op;
262     } else if (isa<PtrToIntInst>(I)) {
263       // Look through ptrtoint.
264       // Make sure this isn't a truncating or extending cast.  We could
265       // support this eventually, but don't bother for now.
266       if (!isa<VectorType>(I->getType()) &&
267           DL.getPointerSizeInBits() ==
268               cast<IntegerType>(I->getType())->getBitWidth())
269         NoopInput = Op;
270     } else if (isa<TruncInst>(I) &&
271                TLI.allowTruncateForTailCall(Op->getType(), I->getType())) {
272       DataBits = std::min(DataBits, I->getType()->getPrimitiveSizeInBits());
273       NoopInput = Op;
274     } else if (auto CS = ImmutableCallSite(I)) {
275       const Value *ReturnedOp = CS.getReturnedArgOperand();
276       if (ReturnedOp && isNoopBitcast(ReturnedOp->getType(), I->getType(), TLI))
277         NoopInput = ReturnedOp;
278     } else if (const InsertValueInst *IVI = dyn_cast<InsertValueInst>(V)) {
279       // Value may come from either the aggregate or the scalar
280       ArrayRef<unsigned> InsertLoc = IVI->getIndices();
281       if (ValLoc.size() >= InsertLoc.size() &&
282           std::equal(InsertLoc.begin(), InsertLoc.end(), ValLoc.rbegin())) {
283         // The type being inserted is a nested sub-type of the aggregate; we
284         // have to remove those initial indices to get the location we're
285         // interested in for the operand.
286         ValLoc.resize(ValLoc.size() - InsertLoc.size());
287         NoopInput = IVI->getInsertedValueOperand();
288       } else {
289         // The struct we're inserting into has the value we're interested in, no
290         // change of address.
291         NoopInput = Op;
292       }
293     } else if (const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(V)) {
294       // The part we're interested in will inevitably be some sub-section of the
295       // previous aggregate. Combine the two paths to obtain the true address of
296       // our element.
297       ArrayRef<unsigned> ExtractLoc = EVI->getIndices();
298       ValLoc.append(ExtractLoc.rbegin(), ExtractLoc.rend());
299       NoopInput = Op;
300     }
301     // Terminate if we couldn't find anything to look through.
302     if (!NoopInput)
303       return V;
304 
305     V = NoopInput;
306   }
307 }
308 
309 /// Return true if this scalar return value only has bits discarded on its path
310 /// from the "tail call" to the "ret". This includes the obvious noop
311 /// instructions handled by getNoopInput above as well as free truncations (or
312 /// extensions prior to the call).
slotOnlyDiscardsData(const Value * RetVal,const Value * CallVal,SmallVectorImpl<unsigned> & RetIndices,SmallVectorImpl<unsigned> & CallIndices,bool AllowDifferingSizes,const TargetLoweringBase & TLI,const DataLayout & DL)313 static bool slotOnlyDiscardsData(const Value *RetVal, const Value *CallVal,
314                                  SmallVectorImpl<unsigned> &RetIndices,
315                                  SmallVectorImpl<unsigned> &CallIndices,
316                                  bool AllowDifferingSizes,
317                                  const TargetLoweringBase &TLI,
318                                  const DataLayout &DL) {
319 
320   // Trace the sub-value needed by the return value as far back up the graph as
321   // possible, in the hope that it will intersect with the value produced by the
322   // call. In the simple case with no "returned" attribute, the hope is actually
323   // that we end up back at the tail call instruction itself.
324   unsigned BitsRequired = UINT_MAX;
325   RetVal = getNoopInput(RetVal, RetIndices, BitsRequired, TLI, DL);
326 
327   // If this slot in the value returned is undef, it doesn't matter what the
328   // call puts there, it'll be fine.
329   if (isa<UndefValue>(RetVal))
330     return true;
331 
332   // Now do a similar search up through the graph to find where the value
333   // actually returned by the "tail call" comes from. In the simple case without
334   // a "returned" attribute, the search will be blocked immediately and the loop
335   // a Noop.
336   unsigned BitsProvided = UINT_MAX;
337   CallVal = getNoopInput(CallVal, CallIndices, BitsProvided, TLI, DL);
338 
339   // There's no hope if we can't actually trace them to (the same part of!) the
340   // same value.
341   if (CallVal != RetVal || CallIndices != RetIndices)
342     return false;
343 
344   // However, intervening truncates may have made the call non-tail. Make sure
345   // all the bits that are needed by the "ret" have been provided by the "tail
346   // call". FIXME: with sufficiently cunning bit-tracking, we could look through
347   // extensions too.
348   if (BitsProvided < BitsRequired ||
349       (!AllowDifferingSizes && BitsProvided != BitsRequired))
350     return false;
351 
352   return true;
353 }
354 
355 /// For an aggregate type, determine whether a given index is within bounds or
356 /// not.
indexReallyValid(CompositeType * T,unsigned Idx)357 static bool indexReallyValid(CompositeType *T, unsigned Idx) {
358   if (ArrayType *AT = dyn_cast<ArrayType>(T))
359     return Idx < AT->getNumElements();
360 
361   return Idx < cast<StructType>(T)->getNumElements();
362 }
363 
364 /// Move the given iterators to the next leaf type in depth first traversal.
365 ///
366 /// Performs a depth-first traversal of the type as specified by its arguments,
367 /// stopping at the next leaf node (which may be a legitimate scalar type or an
368 /// empty struct or array).
369 ///
370 /// @param SubTypes List of the partial components making up the type from
371 /// outermost to innermost non-empty aggregate. The element currently
372 /// represented is SubTypes.back()->getTypeAtIndex(Path.back() - 1).
373 ///
374 /// @param Path Set of extractvalue indices leading from the outermost type
375 /// (SubTypes[0]) to the leaf node currently represented.
376 ///
377 /// @returns true if a new type was found, false otherwise. Calling this
378 /// function again on a finished iterator will repeatedly return
379 /// false. SubTypes.back()->getTypeAtIndex(Path.back()) is either an empty
380 /// aggregate or a non-aggregate
advanceToNextLeafType(SmallVectorImpl<CompositeType * > & SubTypes,SmallVectorImpl<unsigned> & Path)381 static bool advanceToNextLeafType(SmallVectorImpl<CompositeType *> &SubTypes,
382                                   SmallVectorImpl<unsigned> &Path) {
383   // First march back up the tree until we can successfully increment one of the
384   // coordinates in Path.
385   while (!Path.empty() && !indexReallyValid(SubTypes.back(), Path.back() + 1)) {
386     Path.pop_back();
387     SubTypes.pop_back();
388   }
389 
390   // If we reached the top, then the iterator is done.
391   if (Path.empty())
392     return false;
393 
394   // We know there's *some* valid leaf now, so march back down the tree picking
395   // out the left-most element at each node.
396   ++Path.back();
397   Type *DeeperType = SubTypes.back()->getTypeAtIndex(Path.back());
398   while (DeeperType->isAggregateType()) {
399     CompositeType *CT = cast<CompositeType>(DeeperType);
400     if (!indexReallyValid(CT, 0))
401       return true;
402 
403     SubTypes.push_back(CT);
404     Path.push_back(0);
405 
406     DeeperType = CT->getTypeAtIndex(0U);
407   }
408 
409   return true;
410 }
411 
412 /// Find the first non-empty, scalar-like type in Next and setup the iterator
413 /// components.
414 ///
415 /// Assuming Next is an aggregate of some kind, this function will traverse the
416 /// tree from left to right (i.e. depth-first) looking for the first
417 /// non-aggregate type which will play a role in function return.
418 ///
419 /// For example, if Next was {[0 x i64], {{}, i32, {}}, i32} then we would setup
420 /// Path as [1, 1] and SubTypes as [Next, {{}, i32, {}}] to represent the first
421 /// i32 in that type.
firstRealType(Type * Next,SmallVectorImpl<CompositeType * > & SubTypes,SmallVectorImpl<unsigned> & Path)422 static bool firstRealType(Type *Next,
423                           SmallVectorImpl<CompositeType *> &SubTypes,
424                           SmallVectorImpl<unsigned> &Path) {
425   // First initialise the iterator components to the first "leaf" node
426   // (i.e. node with no valid sub-type at any index, so {} does count as a leaf
427   // despite nominally being an aggregate).
428   while (Next->isAggregateType() &&
429          indexReallyValid(cast<CompositeType>(Next), 0)) {
430     SubTypes.push_back(cast<CompositeType>(Next));
431     Path.push_back(0);
432     Next = cast<CompositeType>(Next)->getTypeAtIndex(0U);
433   }
434 
435   // If there's no Path now, Next was originally scalar already (or empty
436   // leaf). We're done.
437   if (Path.empty())
438     return true;
439 
440   // Otherwise, use normal iteration to keep looking through the tree until we
441   // find a non-aggregate type.
442   while (SubTypes.back()->getTypeAtIndex(Path.back())->isAggregateType()) {
443     if (!advanceToNextLeafType(SubTypes, Path))
444       return false;
445   }
446 
447   return true;
448 }
449 
450 /// Set the iterator data-structures to the next non-empty, non-aggregate
451 /// subtype.
nextRealType(SmallVectorImpl<CompositeType * > & SubTypes,SmallVectorImpl<unsigned> & Path)452 static bool nextRealType(SmallVectorImpl<CompositeType *> &SubTypes,
453                          SmallVectorImpl<unsigned> &Path) {
454   do {
455     if (!advanceToNextLeafType(SubTypes, Path))
456       return false;
457 
458     assert(!Path.empty() && "found a leaf but didn't set the path?");
459   } while (SubTypes.back()->getTypeAtIndex(Path.back())->isAggregateType());
460 
461   return true;
462 }
463 
464 
465 /// Test if the given instruction is in a position to be optimized
466 /// with a tail-call. This roughly means that it's in a block with
467 /// a return and there's nothing that needs to be scheduled
468 /// between it and the return.
469 ///
470 /// This function only tests target-independent requirements.
isInTailCallPosition(ImmutableCallSite CS,const TargetMachine & TM)471 bool llvm::isInTailCallPosition(ImmutableCallSite CS, const TargetMachine &TM) {
472   const Instruction *I = CS.getInstruction();
473   const BasicBlock *ExitBB = I->getParent();
474   const TerminatorInst *Term = ExitBB->getTerminator();
475   const ReturnInst *Ret = dyn_cast<ReturnInst>(Term);
476 
477   // The block must end in a return statement or unreachable.
478   //
479   // FIXME: Decline tailcall if it's not guaranteed and if the block ends in
480   // an unreachable, for now. The way tailcall optimization is currently
481   // implemented means it will add an epilogue followed by a jump. That is
482   // not profitable. Also, if the callee is a special function (e.g.
483   // longjmp on x86), it can end up causing miscompilation that has not
484   // been fully understood.
485   if (!Ret &&
486       (!TM.Options.GuaranteedTailCallOpt || !isa<UnreachableInst>(Term)))
487     return false;
488 
489   // If I will have a chain, make sure no other instruction that will have a
490   // chain interposes between I and the return.
491   if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
492       !isSafeToSpeculativelyExecute(I))
493     for (BasicBlock::const_iterator BBI = std::prev(ExitBB->end(), 2);; --BBI) {
494       if (&*BBI == I)
495         break;
496       // Debug info intrinsics do not get in the way of tail call optimization.
497       if (isa<DbgInfoIntrinsic>(BBI))
498         continue;
499       if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
500           !isSafeToSpeculativelyExecute(&*BBI))
501         return false;
502     }
503 
504   const Function *F = ExitBB->getParent();
505   return returnTypeIsEligibleForTailCall(
506       F, I, Ret, *TM.getSubtargetImpl(*F)->getTargetLowering());
507 }
508 
attributesPermitTailCall(const Function * F,const Instruction * I,const ReturnInst * Ret,const TargetLoweringBase & TLI,bool * AllowDifferingSizes)509 bool llvm::attributesPermitTailCall(const Function *F, const Instruction *I,
510                                     const ReturnInst *Ret,
511                                     const TargetLoweringBase &TLI,
512                                     bool *AllowDifferingSizes) {
513   // ADS may be null, so don't write to it directly.
514   bool DummyADS;
515   bool &ADS = AllowDifferingSizes ? *AllowDifferingSizes : DummyADS;
516   ADS = true;
517 
518   AttrBuilder CallerAttrs(F->getAttributes(), AttributeList::ReturnIndex);
519   AttrBuilder CalleeAttrs(cast<CallInst>(I)->getAttributes(),
520                           AttributeList::ReturnIndex);
521 
522   // Noalias is completely benign as far as calling convention goes, it
523   // shouldn't affect whether the call is a tail call.
524   CallerAttrs.removeAttribute(Attribute::NoAlias);
525   CalleeAttrs.removeAttribute(Attribute::NoAlias);
526 
527   if (CallerAttrs.contains(Attribute::ZExt)) {
528     if (!CalleeAttrs.contains(Attribute::ZExt))
529       return false;
530 
531     ADS = false;
532     CallerAttrs.removeAttribute(Attribute::ZExt);
533     CalleeAttrs.removeAttribute(Attribute::ZExt);
534   } else if (CallerAttrs.contains(Attribute::SExt)) {
535     if (!CalleeAttrs.contains(Attribute::SExt))
536       return false;
537 
538     ADS = false;
539     CallerAttrs.removeAttribute(Attribute::SExt);
540     CalleeAttrs.removeAttribute(Attribute::SExt);
541   }
542 
543   // If they're still different, there's some facet we don't understand
544   // (currently only "inreg", but in future who knows). It may be OK but the
545   // only safe option is to reject the tail call.
546   return CallerAttrs == CalleeAttrs;
547 }
548 
returnTypeIsEligibleForTailCall(const Function * F,const Instruction * I,const ReturnInst * Ret,const TargetLoweringBase & TLI)549 bool llvm::returnTypeIsEligibleForTailCall(const Function *F,
550                                            const Instruction *I,
551                                            const ReturnInst *Ret,
552                                            const TargetLoweringBase &TLI) {
553   // If the block ends with a void return or unreachable, it doesn't matter
554   // what the call's return type is.
555   if (!Ret || Ret->getNumOperands() == 0) return true;
556 
557   // If the return value is undef, it doesn't matter what the call's
558   // return type is.
559   if (isa<UndefValue>(Ret->getOperand(0))) return true;
560 
561   // Make sure the attributes attached to each return are compatible.
562   bool AllowDifferingSizes;
563   if (!attributesPermitTailCall(F, I, Ret, TLI, &AllowDifferingSizes))
564     return false;
565 
566   const Value *RetVal = Ret->getOperand(0), *CallVal = I;
567   // Intrinsic like llvm.memcpy has no return value, but the expanded
568   // libcall may or may not have return value. On most platforms, it
569   // will be expanded as memcpy in libc, which returns the first
570   // argument. On other platforms like arm-none-eabi, memcpy may be
571   // expanded as library call without return value, like __aeabi_memcpy.
572   const CallInst *Call = cast<CallInst>(I);
573   if (Function *F = Call->getCalledFunction()) {
574     Intrinsic::ID IID = F->getIntrinsicID();
575     if (((IID == Intrinsic::memcpy &&
576           TLI.getLibcallName(RTLIB::MEMCPY) == StringRef("memcpy")) ||
577          (IID == Intrinsic::memmove &&
578           TLI.getLibcallName(RTLIB::MEMMOVE) == StringRef("memmove")) ||
579          (IID == Intrinsic::memset &&
580           TLI.getLibcallName(RTLIB::MEMSET) == StringRef("memset"))) &&
581         RetVal == Call->getArgOperand(0))
582       return true;
583   }
584 
585   SmallVector<unsigned, 4> RetPath, CallPath;
586   SmallVector<CompositeType *, 4> RetSubTypes, CallSubTypes;
587 
588   bool RetEmpty = !firstRealType(RetVal->getType(), RetSubTypes, RetPath);
589   bool CallEmpty = !firstRealType(CallVal->getType(), CallSubTypes, CallPath);
590 
591   // Nothing's actually returned, it doesn't matter what the callee put there
592   // it's a valid tail call.
593   if (RetEmpty)
594     return true;
595 
596   // Iterate pairwise through each of the value types making up the tail call
597   // and the corresponding return. For each one we want to know whether it's
598   // essentially going directly from the tail call to the ret, via operations
599   // that end up not generating any code.
600   //
601   // We allow a certain amount of covariance here. For example it's permitted
602   // for the tail call to define more bits than the ret actually cares about
603   // (e.g. via a truncate).
604   do {
605     if (CallEmpty) {
606       // We've exhausted the values produced by the tail call instruction, the
607       // rest are essentially undef. The type doesn't really matter, but we need
608       // *something*.
609       Type *SlotType = RetSubTypes.back()->getTypeAtIndex(RetPath.back());
610       CallVal = UndefValue::get(SlotType);
611     }
612 
613     // The manipulations performed when we're looking through an insertvalue or
614     // an extractvalue would happen at the front of the RetPath list, so since
615     // we have to copy it anyway it's more efficient to create a reversed copy.
616     SmallVector<unsigned, 4> TmpRetPath(RetPath.rbegin(), RetPath.rend());
617     SmallVector<unsigned, 4> TmpCallPath(CallPath.rbegin(), CallPath.rend());
618 
619     // Finally, we can check whether the value produced by the tail call at this
620     // index is compatible with the value we return.
621     if (!slotOnlyDiscardsData(RetVal, CallVal, TmpRetPath, TmpCallPath,
622                               AllowDifferingSizes, TLI,
623                               F->getParent()->getDataLayout()))
624       return false;
625 
626     CallEmpty  = !nextRealType(CallSubTypes, CallPath);
627   } while(nextRealType(RetSubTypes, RetPath));
628 
629   return true;
630 }
631 
collectEHScopeMembers(DenseMap<const MachineBasicBlock *,int> & EHScopeMembership,int EHScope,const MachineBasicBlock * MBB)632 static void collectEHScopeMembers(
633     DenseMap<const MachineBasicBlock *, int> &EHScopeMembership, int EHScope,
634     const MachineBasicBlock *MBB) {
635   SmallVector<const MachineBasicBlock *, 16> Worklist = {MBB};
636   while (!Worklist.empty()) {
637     const MachineBasicBlock *Visiting = Worklist.pop_back_val();
638     // Don't follow blocks which start new scopes.
639     if (Visiting->isEHPad() && Visiting != MBB)
640       continue;
641 
642     // Add this MBB to our scope.
643     auto P = EHScopeMembership.insert(std::make_pair(Visiting, EHScope));
644 
645     // Don't revisit blocks.
646     if (!P.second) {
647       assert(P.first->second == EHScope && "MBB is part of two scopes!");
648       continue;
649     }
650 
651     // Returns are boundaries where scope transfer can occur, don't follow
652     // successors.
653     if (Visiting->isReturnBlock())
654       continue;
655 
656     for (const MachineBasicBlock *Succ : Visiting->successors())
657       Worklist.push_back(Succ);
658   }
659 }
660 
661 DenseMap<const MachineBasicBlock *, int>
getEHScopeMembership(const MachineFunction & MF)662 llvm::getEHScopeMembership(const MachineFunction &MF) {
663   DenseMap<const MachineBasicBlock *, int> EHScopeMembership;
664 
665   // We don't have anything to do if there aren't any EH pads.
666   if (!MF.hasEHScopes())
667     return EHScopeMembership;
668 
669   int EntryBBNumber = MF.front().getNumber();
670   bool IsSEH = isAsynchronousEHPersonality(
671       classifyEHPersonality(MF.getFunction().getPersonalityFn()));
672 
673   const TargetInstrInfo *TII = MF.getSubtarget().getInstrInfo();
674   SmallVector<const MachineBasicBlock *, 16> EHScopeBlocks;
675   SmallVector<const MachineBasicBlock *, 16> UnreachableBlocks;
676   SmallVector<const MachineBasicBlock *, 16> SEHCatchPads;
677   SmallVector<std::pair<const MachineBasicBlock *, int>, 16> CatchRetSuccessors;
678   for (const MachineBasicBlock &MBB : MF) {
679     if (MBB.isEHScopeEntry()) {
680       EHScopeBlocks.push_back(&MBB);
681     } else if (IsSEH && MBB.isEHPad()) {
682       SEHCatchPads.push_back(&MBB);
683     } else if (MBB.pred_empty()) {
684       UnreachableBlocks.push_back(&MBB);
685     }
686 
687     MachineBasicBlock::const_iterator MBBI = MBB.getFirstTerminator();
688 
689     // CatchPads are not scopes for SEH so do not consider CatchRet to
690     // transfer control to another scope.
691     if (MBBI == MBB.end() || MBBI->getOpcode() != TII->getCatchReturnOpcode())
692       continue;
693 
694     // FIXME: SEH CatchPads are not necessarily in the parent function:
695     // they could be inside a finally block.
696     const MachineBasicBlock *Successor = MBBI->getOperand(0).getMBB();
697     const MachineBasicBlock *SuccessorColor = MBBI->getOperand(1).getMBB();
698     CatchRetSuccessors.push_back(
699         {Successor, IsSEH ? EntryBBNumber : SuccessorColor->getNumber()});
700   }
701 
702   // We don't have anything to do if there aren't any EH pads.
703   if (EHScopeBlocks.empty())
704     return EHScopeMembership;
705 
706   // Identify all the basic blocks reachable from the function entry.
707   collectEHScopeMembers(EHScopeMembership, EntryBBNumber, &MF.front());
708   // All blocks not part of a scope are in the parent function.
709   for (const MachineBasicBlock *MBB : UnreachableBlocks)
710     collectEHScopeMembers(EHScopeMembership, EntryBBNumber, MBB);
711   // Next, identify all the blocks inside the scopes.
712   for (const MachineBasicBlock *MBB : EHScopeBlocks)
713     collectEHScopeMembers(EHScopeMembership, MBB->getNumber(), MBB);
714   // SEH CatchPads aren't really scopes, handle them separately.
715   for (const MachineBasicBlock *MBB : SEHCatchPads)
716     collectEHScopeMembers(EHScopeMembership, EntryBBNumber, MBB);
717   // Finally, identify all the targets of a catchret.
718   for (std::pair<const MachineBasicBlock *, int> CatchRetPair :
719        CatchRetSuccessors)
720     collectEHScopeMembers(EHScopeMembership, CatchRetPair.second,
721                           CatchRetPair.first);
722   return EHScopeMembership;
723 }
724