-//===- BasicAliasAnalysis.cpp - Local Alias Analysis Impl -----------------===//
+//===- BasicAliasAnalysis.cpp - Stateless Alias Analysis Impl -------------===//
//
// The LLVM Compiler Infrastructure
//
//
//===----------------------------------------------------------------------===//
//
-// This file defines the default implementation of the Alias Analysis interface
-// that simply implements a few identities (two different globals cannot alias,
-// etc), but otherwise does no analysis.
+// This file defines the primary stateless implementation of the
+// Alias Analysis interface that implements identities (two different
+// globals cannot alias, etc), but does no stateful analysis.
//
//===----------------------------------------------------------------------===//
+#include "llvm/Analysis/Passes.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/SmallVector.h"
#include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Analysis/AssumptionCache.h"
+#include "llvm/Analysis/CFG.h"
#include "llvm/Analysis/CaptureTracking.h"
+#include "llvm/Analysis/InstructionSimplify.h"
+#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/MemoryBuiltins.h"
-#include "llvm/Analysis/Passes.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/Instructions.h"
-#include "llvm/IntrinsicInst.h"
-#include "llvm/Operator.h"
+#include "llvm/Analysis/TargetLibraryInfo.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Dominators.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GetElementPtrTypeIterator.h"
+#include "llvm/IR/GlobalAlias.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Operator.h"
#include "llvm/Pass.h"
-#include "llvm/Target/TargetData.h"
-#include "llvm/ADT/SmallSet.h"
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/GetElementPtrTypeIterator.h"
#include <algorithm>
using namespace llvm;
+/// Cutoff after which to stop analysing a set of phi nodes potentially involved
+/// in a cycle. Because we are analysing 'through' phi nodes we need to be
+/// careful with value equivalence. We use reachability to make sure a value
+/// cannot be involved in a cycle.
+const unsigned MaxNumPhiBBsValueReachabilityCheck = 20;
+
+// The max limit of the search depth in DecomposeGEPExpression() and
+// GetUnderlyingObject(), both functions need to use the same search
+// depth otherwise the algorithm in aliasGEP will assert.
+static const unsigned MaxLookupSearchDepth = 6;
+
//===----------------------------------------------------------------------===//
// Useful predicates
//===----------------------------------------------------------------------===//
-static const Value *GetGEPOperands(const Value *V,
- SmallVector<Value*, 16> &GEPOps) {
- assert(GEPOps.empty() && "Expect empty list to populate!");
- GEPOps.insert(GEPOps.end(), cast<User>(V)->op_begin()+1,
- cast<User>(V)->op_end());
-
- // Accumulate all of the chained indexes into the operand array
- V = cast<User>(V)->getOperand(0);
-
- while (const GEPOperator *G = dyn_cast<GEPOperator>(V)) {
- if (!isa<Constant>(GEPOps[0]) || isa<GlobalValue>(GEPOps[0]) ||
- !cast<Constant>(GEPOps[0])->isNullValue())
- break; // Don't handle folding arbitrary pointer offsets yet...
- GEPOps.erase(GEPOps.begin()); // Drop the zero index
- GEPOps.insert(GEPOps.begin(), G->op_begin()+1, G->op_end());
- V = G->getOperand(0);
- }
- return V;
-}
-
-/// isKnownNonNull - Return true if we know that the specified value is never
-/// null.
-static bool isKnownNonNull(const Value *V) {
- // Alloca never returns null, malloc might.
- if (isa<AllocaInst>(V)) return true;
-
- // A byval argument is never null.
- if (const Argument *A = dyn_cast<Argument>(V))
- return A->hasByValAttr();
-
- // Global values are not null unless extern weak.
- if (const GlobalValue *GV = dyn_cast<GlobalValue>(V))
- return !GV->hasExternalWeakLinkage();
- return false;
-}
-
/// isNonEscapingLocalObject - Return true if the pointer is to a function-local
/// object that never escapes from the function.
static bool isNonEscapingLocalObject(const Value *V) {
// If this is a local allocation, check to see if it escapes.
if (isa<AllocaInst>(V) || isNoAliasCall(V))
- return !PointerMayBeCaptured(V, false);
+ // Set StoreCaptures to True so that we can assume in our callers that the
+ // pointer is not the result of a load instruction. Currently
+ // PointerMayBeCaptured doesn't have any special analysis for the
+ // StoreCaptures=false case; if it did, our callers could be refined to be
+ // more precise.
+ return !PointerMayBeCaptured(V, false, /*StoreCaptures=*/true);
// If this is an argument that corresponds to a byval or noalias argument,
// then it has not escaped before entering the function. Check if it escapes
// inside the function.
if (const Argument *A = dyn_cast<Argument>(V))
- if (A->hasByValAttr() || A->hasNoAliasAttr()) {
- // Don't bother analyzing arguments already known not to escape.
- if (A->hasNoCaptureAttr())
- return true;
- return !PointerMayBeCaptured(V, false);
- }
+ if (A->hasByValAttr() || A->hasNoAliasAttr())
+ // Note even if the argument is marked nocapture we still need to check
+ // for copies made inside the function. The nocapture attribute only
+ // specifies that there are no copies made that outlive the function.
+ return !PointerMayBeCaptured(V, false, /*StoreCaptures=*/true);
+
return false;
}
+/// isEscapeSource - Return true if the pointer is one which would have
+/// been considered an escape by isNonEscapingLocalObject.
+static bool isEscapeSource(const Value *V) {
+ if (isa<CallInst>(V) || isa<InvokeInst>(V) || isa<Argument>(V))
+ return true;
+
+ // The load case works because isNonEscapingLocalObject considers all
+ // stores to be escapes (it passes true for the StoreCaptures argument
+ // to PointerMayBeCaptured).
+ if (isa<LoadInst>(V))
+ return true;
+
+ return false;
+}
+
+/// getObjectSize - Return the size of the object specified by V, or
+/// UnknownSize if unknown.
+static uint64_t getObjectSize(const Value *V, const DataLayout &DL,
+ const TargetLibraryInfo &TLI,
+ bool RoundToAlign = false) {
+ uint64_t Size;
+ if (getObjectSize(V, Size, DL, &TLI, RoundToAlign))
+ return Size;
+ return AliasAnalysis::UnknownSize;
+}
/// isObjectSmallerThan - Return true if we can prove that the object specified
/// by V is smaller than Size.
-static bool isObjectSmallerThan(const Value *V, unsigned Size,
- const TargetData &TD) {
- const Type *AccessTy;
- if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(V)) {
- AccessTy = GV->getType()->getElementType();
- } else if (const AllocaInst *AI = dyn_cast<AllocaInst>(V)) {
- if (!AI->isArrayAllocation())
- AccessTy = AI->getType()->getElementType();
- else
- return false;
- } else if (const CallInst* CI = extractMallocCall(V)) {
- if (!isArrayMalloc(V, &TD))
- // The size is the argument to the malloc call.
- if (const ConstantInt* C = dyn_cast<ConstantInt>(CI->getOperand(1)))
- return (C->getZExtValue() < Size);
- return false;
- } else if (const Argument *A = dyn_cast<Argument>(V)) {
- if (A->hasByValAttr())
- AccessTy = cast<PointerType>(A->getType())->getElementType();
- else
- return false;
- } else {
+static bool isObjectSmallerThan(const Value *V, uint64_t Size,
+ const DataLayout &DL,
+ const TargetLibraryInfo &TLI) {
+ // Note that the meanings of the "object" are slightly different in the
+ // following contexts:
+ // c1: llvm::getObjectSize()
+ // c2: llvm.objectsize() intrinsic
+ // c3: isObjectSmallerThan()
+ // c1 and c2 share the same meaning; however, the meaning of "object" in c3
+ // refers to the "entire object".
+ //
+ // Consider this example:
+ // char *p = (char*)malloc(100)
+ // char *q = p+80;
+ //
+ // In the context of c1 and c2, the "object" pointed by q refers to the
+ // stretch of memory of q[0:19]. So, getObjectSize(q) should return 20.
+ //
+ // However, in the context of c3, the "object" refers to the chunk of memory
+ // being allocated. So, the "object" has 100 bytes, and q points to the middle
+ // the "object". In case q is passed to isObjectSmallerThan() as the 1st
+ // parameter, before the llvm::getObjectSize() is called to get the size of
+ // entire object, we should:
+ // - either rewind the pointer q to the base-address of the object in
+ // question (in this case rewind to p), or
+ // - just give up. It is up to caller to make sure the pointer is pointing
+ // to the base address the object.
+ //
+ // We go for 2nd option for simplicity.
+ if (!isIdentifiedObject(V))
return false;
- }
-
- if (AccessTy->isSized())
- return TD.getTypeAllocSize(AccessTy) < Size;
- return false;
+
+ // This function needs to use the aligned object size because we allow
+ // reads a bit past the end given sufficient alignment.
+ uint64_t ObjectSize = getObjectSize(V, DL, TLI, /*RoundToAlign*/true);
+
+ return ObjectSize != AliasAnalysis::UnknownSize && ObjectSize < Size;
+}
+
+/// isObjectSize - Return true if we can prove that the object specified
+/// by V has size Size.
+static bool isObjectSize(const Value *V, uint64_t Size,
+ const DataLayout &DL, const TargetLibraryInfo &TLI) {
+ uint64_t ObjectSize = getObjectSize(V, DL, TLI);
+ return ObjectSize != AliasAnalysis::UnknownSize && ObjectSize == Size;
}
//===----------------------------------------------------------------------===//
-// NoAA Pass
+// GetElementPtr Instruction Decomposition and Analysis
//===----------------------------------------------------------------------===//
namespace {
- /// NoAA - This class implements the -no-aa pass, which always returns "I
- /// don't know" for alias queries. NoAA is unlike other alias analysis
- /// implementations, in that it does not chain to a previous analysis. As
- /// such it doesn't follow many of the rules that other alias analyses must.
- ///
- struct NoAA : public ImmutablePass, public AliasAnalysis {
- static char ID; // Class identification, replacement for typeinfo
- NoAA() : ImmutablePass(&ID) {}
- explicit NoAA(void *PID) : ImmutablePass(PID) { }
+ enum ExtensionKind {
+ EK_NotExtended,
+ EK_SignExt,
+ EK_ZeroExt
+ };
+
+ struct VariableGEPIndex {
+ const Value *V;
+ ExtensionKind Extension;
+ int64_t Scale;
- virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ bool operator==(const VariableGEPIndex &Other) const {
+ return V == Other.V && Extension == Other.Extension &&
+ Scale == Other.Scale;
}
- virtual void initializePass() {
- TD = getAnalysisIfAvailable<TargetData>();
+ bool operator!=(const VariableGEPIndex &Other) const {
+ return !operator==(Other);
}
+ };
+}
- virtual AliasResult alias(const Value *V1, unsigned V1Size,
- const Value *V2, unsigned V2Size) {
- return MayAlias;
+
+/// GetLinearExpression - Analyze the specified value as a linear expression:
+/// "A*V + B", where A and B are constant integers. Return the scale and offset
+/// values as APInts and return V as a Value*, and return whether we looked
+/// through any sign or zero extends. The incoming Value is known to have
+/// IntegerType and it may already be sign or zero extended.
+///
+/// Note that this looks through extends, so the high bits may not be
+/// represented in the result.
+static Value *GetLinearExpression(Value *V, APInt &Scale, APInt &Offset,
+ ExtensionKind &Extension,
+ const DataLayout &DL, unsigned Depth,
+ AssumptionCache *AC, DominatorTree *DT) {
+ assert(V->getType()->isIntegerTy() && "Not an integer value");
+
+ // Limit our recursion depth.
+ if (Depth == 6) {
+ Scale = 1;
+ Offset = 0;
+ return V;
+ }
+
+ if (ConstantInt *Const = dyn_cast<ConstantInt>(V)) {
+ // if it's a constant, just convert it to an offset
+ // and remove the variable.
+ Offset += Const->getValue();
+ assert(Scale == 0 && "Constant values don't have a scale");
+ return V;
+ }
+
+ if (BinaryOperator *BOp = dyn_cast<BinaryOperator>(V)) {
+ if (ConstantInt *RHSC = dyn_cast<ConstantInt>(BOp->getOperand(1))) {
+ switch (BOp->getOpcode()) {
+ default: break;
+ case Instruction::Or:
+ // X|C == X+C if all the bits in C are unset in X. Otherwise we can't
+ // analyze it.
+ if (!MaskedValueIsZero(BOp->getOperand(0), RHSC->getValue(), DL, 0, AC,
+ BOp, DT))
+ break;
+ // FALL THROUGH.
+ case Instruction::Add:
+ V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, Extension,
+ DL, Depth + 1, AC, DT);
+ Offset += RHSC->getValue();
+ return V;
+ case Instruction::Mul:
+ V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, Extension,
+ DL, Depth + 1, AC, DT);
+ Offset *= RHSC->getValue();
+ Scale *= RHSC->getValue();
+ return V;
+ case Instruction::Shl:
+ V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, Extension,
+ DL, Depth + 1, AC, DT);
+ Offset <<= RHSC->getValue().getLimitedValue();
+ Scale <<= RHSC->getValue().getLimitedValue();
+ return V;
+ }
}
+ }
+
+ // Since GEP indices are sign extended anyway, we don't care about the high
+ // bits of a sign or zero extended value - just scales and offsets. The
+ // extensions have to be consistent though.
+ if ((isa<SExtInst>(V) && Extension != EK_ZeroExt) ||
+ (isa<ZExtInst>(V) && Extension != EK_SignExt)) {
+ Value *CastOp = cast<CastInst>(V)->getOperand(0);
+ unsigned OldWidth = Scale.getBitWidth();
+ unsigned SmallWidth = CastOp->getType()->getPrimitiveSizeInBits();
+ Scale = Scale.trunc(SmallWidth);
+ Offset = Offset.trunc(SmallWidth);
+ Extension = isa<SExtInst>(V) ? EK_SignExt : EK_ZeroExt;
+
+ Value *Result = GetLinearExpression(CastOp, Scale, Offset, Extension, DL,
+ Depth + 1, AC, DT);
+ Scale = Scale.zext(OldWidth);
+
+ // We have to sign-extend even if Extension == EK_ZeroExt as we can't
+ // decompose a sign extension (i.e. zext(x - 1) != zext(x) - zext(-1)).
+ Offset = Offset.sext(OldWidth);
+
+ return Result;
+ }
+
+ Scale = 1;
+ Offset = 0;
+ return V;
+}
- virtual void getArgumentAccesses(Function *F, CallSite CS,
- std::vector<PointerAccessInfo> &Info) {
- llvm_unreachable("This method may not be called on this function!");
+/// DecomposeGEPExpression - If V is a symbolic pointer expression, decompose it
+/// into a base pointer with a constant offset and a number of scaled symbolic
+/// offsets.
+///
+/// The scaled symbolic offsets (represented by pairs of a Value* and a scale in
+/// the VarIndices vector) are Value*'s that are known to be scaled by the
+/// specified amount, but which may have other unrepresented high bits. As such,
+/// the gep cannot necessarily be reconstructed from its decomposed form.
+///
+/// When DataLayout is around, this function is capable of analyzing everything
+/// that GetUnderlyingObject can look through. To be able to do that
+/// GetUnderlyingObject and DecomposeGEPExpression must use the same search
+/// depth (MaxLookupSearchDepth).
+/// When DataLayout not is around, it just looks through pointer casts.
+///
+static const Value *
+DecomposeGEPExpression(const Value *V, int64_t &BaseOffs,
+ SmallVectorImpl<VariableGEPIndex> &VarIndices,
+ bool &MaxLookupReached, const DataLayout &DL,
+ AssumptionCache *AC, DominatorTree *DT) {
+ // Limit recursion depth to limit compile time in crazy cases.
+ unsigned MaxLookup = MaxLookupSearchDepth;
+ MaxLookupReached = false;
+
+ BaseOffs = 0;
+ do {
+ // See if this is a bitcast or GEP.
+ const Operator *Op = dyn_cast<Operator>(V);
+ if (!Op) {
+ // The only non-operator case we can handle are GlobalAliases.
+ if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(V)) {
+ if (!GA->mayBeOverridden()) {
+ V = GA->getAliasee();
+ continue;
+ }
+ }
+ return V;
}
- virtual void getMustAliases(Value *P, std::vector<Value*> &RetVals) { }
- virtual bool pointsToConstantMemory(const Value *P) { return false; }
- virtual ModRefResult getModRefInfo(CallSite CS, Value *P, unsigned Size) {
- return ModRef;
+ if (Op->getOpcode() == Instruction::BitCast ||
+ Op->getOpcode() == Instruction::AddrSpaceCast) {
+ V = Op->getOperand(0);
+ continue;
}
- virtual ModRefResult getModRefInfo(CallSite CS1, CallSite CS2) {
- return ModRef;
+
+ const GEPOperator *GEPOp = dyn_cast<GEPOperator>(Op);
+ if (!GEPOp) {
+ // If it's not a GEP, hand it off to SimplifyInstruction to see if it
+ // can come up with something. This matches what GetUnderlyingObject does.
+ if (const Instruction *I = dyn_cast<Instruction>(V))
+ // TODO: Get a DominatorTree and AssumptionCache and use them here
+ // (these are both now available in this function, but this should be
+ // updated when GetUnderlyingObject is updated). TLI should be
+ // provided also.
+ if (const Value *Simplified =
+ SimplifyInstruction(const_cast<Instruction *>(I), DL)) {
+ V = Simplified;
+ continue;
+ }
+
+ return V;
}
- virtual bool hasNoModRefInfoForCalls() const { return true; }
- virtual void deleteValue(Value *V) {}
- virtual void copyValue(Value *From, Value *To) {}
- };
-} // End of anonymous namespace
+ // Don't attempt to analyze GEPs over unsized objects.
+ if (!GEPOp->getOperand(0)->getType()->getPointerElementType()->isSized())
+ return V;
+
+ unsigned AS = GEPOp->getPointerAddressSpace();
+ // Walk the indices of the GEP, accumulating them into BaseOff/VarIndices.
+ gep_type_iterator GTI = gep_type_begin(GEPOp);
+ for (User::const_op_iterator I = GEPOp->op_begin()+1,
+ E = GEPOp->op_end(); I != E; ++I) {
+ Value *Index = *I;
+ // Compute the (potentially symbolic) offset in bytes for this index.
+ if (StructType *STy = dyn_cast<StructType>(*GTI++)) {
+ // For a struct, add the member offset.
+ unsigned FieldNo = cast<ConstantInt>(Index)->getZExtValue();
+ if (FieldNo == 0) continue;
+
+ BaseOffs += DL.getStructLayout(STy)->getElementOffset(FieldNo);
+ continue;
+ }
-// Register this pass...
-char NoAA::ID = 0;
-static RegisterPass<NoAA>
-U("no-aa", "No Alias Analysis (always returns 'may' alias)", true, true);
+ // For an array/pointer, add the element offset, explicitly scaled.
+ if (ConstantInt *CIdx = dyn_cast<ConstantInt>(Index)) {
+ if (CIdx->isZero()) continue;
+ BaseOffs += DL.getTypeAllocSize(*GTI) * CIdx->getSExtValue();
+ continue;
+ }
-// Declare that we implement the AliasAnalysis interface
-static RegisterAnalysisGroup<AliasAnalysis> V(U);
+ uint64_t Scale = DL.getTypeAllocSize(*GTI);
+ ExtensionKind Extension = EK_NotExtended;
+
+ // If the integer type is smaller than the pointer size, it is implicitly
+ // sign extended to pointer size.
+ unsigned Width = Index->getType()->getIntegerBitWidth();
+ if (DL.getPointerSizeInBits(AS) > Width)
+ Extension = EK_SignExt;
+
+ // Use GetLinearExpression to decompose the index into a C1*V+C2 form.
+ APInt IndexScale(Width, 0), IndexOffset(Width, 0);
+ Index = GetLinearExpression(Index, IndexScale, IndexOffset, Extension, DL,
+ 0, AC, DT);
+
+ // The GEP index scale ("Scale") scales C1*V+C2, yielding (C1*V+C2)*Scale.
+ // This gives us an aggregate computation of (C1*Scale)*V + C2*Scale.
+ BaseOffs += IndexOffset.getSExtValue()*Scale;
+ Scale *= IndexScale.getSExtValue();
+
+ // If we already had an occurrence of this index variable, merge this
+ // scale into it. For example, we want to handle:
+ // A[x][x] -> x*16 + x*4 -> x*20
+ // This also ensures that 'x' only appears in the index list once.
+ for (unsigned i = 0, e = VarIndices.size(); i != e; ++i) {
+ if (VarIndices[i].V == Index &&
+ VarIndices[i].Extension == Extension) {
+ Scale += VarIndices[i].Scale;
+ VarIndices.erase(VarIndices.begin()+i);
+ break;
+ }
+ }
-ImmutablePass *llvm::createNoAAPass() { return new NoAA(); }
+ // Make sure that we have a scale that makes sense for this target's
+ // pointer size.
+ if (unsigned ShiftBits = 64 - DL.getPointerSizeInBits(AS)) {
+ Scale <<= ShiftBits;
+ Scale = (int64_t)Scale >> ShiftBits;
+ }
+
+ if (Scale) {
+ VariableGEPIndex Entry = {Index, Extension,
+ static_cast<int64_t>(Scale)};
+ VarIndices.push_back(Entry);
+ }
+ }
+
+ // Analyze the base pointer next.
+ V = GEPOp->getOperand(0);
+ } while (--MaxLookup);
+
+ // If the chain of expressions is too deep, just return early.
+ MaxLookupReached = true;
+ return V;
+}
//===----------------------------------------------------------------------===//
-// BasicAA Pass
+// BasicAliasAnalysis Pass
//===----------------------------------------------------------------------===//
+#ifndef NDEBUG
+static const Function *getParent(const Value *V) {
+ if (const Instruction *inst = dyn_cast<Instruction>(V))
+ return inst->getParent()->getParent();
+
+ if (const Argument *arg = dyn_cast<Argument>(V))
+ return arg->getParent();
+
+ return nullptr;
+}
+
+static bool notDifferentParent(const Value *O1, const Value *O2) {
+
+ const Function *F1 = getParent(O1);
+ const Function *F2 = getParent(O2);
+
+ return !F1 || !F2 || F1 == F2;
+}
+#endif
+
namespace {
- /// BasicAliasAnalysis - This is the default alias analysis implementation.
- /// Because it doesn't chain to a previous alias analysis (like -no-aa), it
- /// derives from the NoAA class.
- struct BasicAliasAnalysis : public NoAA {
+ /// BasicAliasAnalysis - This is the primary alias analysis implementation.
+ struct BasicAliasAnalysis : public ImmutablePass, public AliasAnalysis {
static char ID; // Class identification, replacement for typeinfo
- BasicAliasAnalysis() : NoAA(&ID) {}
- AliasResult alias(const Value *V1, unsigned V1Size,
- const Value *V2, unsigned V2Size) {
- assert(VisitedPHIs.empty() && "VisitedPHIs must be cleared after use!");
- AliasResult Alias = aliasCheck(V1, V1Size, V2, V2Size);
- VisitedPHIs.clear();
+ BasicAliasAnalysis() : ImmutablePass(ID) {
+ initializeBasicAliasAnalysisPass(*PassRegistry::getPassRegistry());
+ }
+
+ bool doInitialization(Module &M) override;
+
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
+ AU.addRequired<AliasAnalysis>();
+ AU.addRequired<AssumptionCacheTracker>();
+ AU.addRequired<TargetLibraryInfoWrapperPass>();
+ }
+
+ AliasResult alias(const Location &LocA, const Location &LocB) override {
+ assert(AliasCache.empty() && "AliasCache must be cleared after use!");
+ assert(notDifferentParent(LocA.Ptr, LocB.Ptr) &&
+ "BasicAliasAnalysis doesn't support interprocedural queries.");
+ AliasResult Alias = aliasCheck(LocA.Ptr, LocA.Size, LocA.AATags,
+ LocB.Ptr, LocB.Size, LocB.AATags);
+ // AliasCache rarely has more than 1 or 2 elements, always use
+ // shrink_and_clear so it quickly returns to the inline capacity of the
+ // SmallDenseMap if it ever grows larger.
+ // FIXME: This should really be shrink_to_inline_capacity_and_clear().
+ AliasCache.shrink_and_clear();
+ VisitedPhiBBs.clear();
return Alias;
}
- ModRefResult getModRefInfo(CallSite CS, Value *P, unsigned Size);
- ModRefResult getModRefInfo(CallSite CS1, CallSite CS2);
+ ModRefResult getModRefInfo(ImmutableCallSite CS,
+ const Location &Loc) override;
- /// hasNoModRefInfoForCalls - We can provide mod/ref information against
- /// non-escaping allocations.
- virtual bool hasNoModRefInfoForCalls() const { return false; }
+ ModRefResult getModRefInfo(ImmutableCallSite CS1,
+ ImmutableCallSite CS2) override;
/// pointsToConstantMemory - Chase pointers until we find a (constant
/// global) or not.
- bool pointsToConstantMemory(const Value *P);
+ bool pointsToConstantMemory(const Location &Loc, bool OrLocal) override;
+
+ /// Get the location associated with a pointer argument of a callsite.
+ Location getArgLocation(ImmutableCallSite CS, unsigned ArgIdx,
+ ModRefResult &Mask) override;
+
+ /// getModRefBehavior - Return the behavior when calling the given
+ /// call site.
+ ModRefBehavior getModRefBehavior(ImmutableCallSite CS) override;
+
+ /// getModRefBehavior - Return the behavior when calling the given function.
+ /// For use when the call site is not known.
+ ModRefBehavior getModRefBehavior(const Function *F) override;
+
+ /// getAdjustedAnalysisPointer - This method is used when a pass implements
+ /// an analysis interface through multiple inheritance. If needed, it
+ /// should override this to adjust the this pointer as needed for the
+ /// specified pass info.
+ void *getAdjustedAnalysisPointer(const void *ID) override {
+ if (ID == &AliasAnalysis::ID)
+ return (AliasAnalysis*)this;
+ return this;
+ }
private:
- // VisitedPHIs - Track PHI nodes visited by a aliasCheck() call.
- SmallPtrSet<const Value*, 16> VisitedPHIs;
-
- // aliasGEP - Provide a bunch of ad-hoc rules to disambiguate a GEP instruction
- // against another.
- AliasResult aliasGEP(const Value *V1, unsigned V1Size,
- const Value *V2, unsigned V2Size);
-
- // aliasPHI - Provide a bunch of ad-hoc rules to disambiguate a PHI instruction
- // against another.
- AliasResult aliasPHI(const PHINode *PN, unsigned PNSize,
- const Value *V2, unsigned V2Size);
+ // AliasCache - Track alias queries to guard against recursion.
+ typedef std::pair<Location, Location> LocPair;
+ typedef SmallDenseMap<LocPair, AliasResult, 8> AliasCacheTy;
+ AliasCacheTy AliasCache;
+
+ /// \brief Track phi nodes we have visited. When interpret "Value" pointer
+ /// equality as value equality we need to make sure that the "Value" is not
+ /// part of a cycle. Otherwise, two uses could come from different
+ /// "iterations" of a cycle and see different values for the same "Value"
+ /// pointer.
+ /// The following example shows the problem:
+ /// %p = phi(%alloca1, %addr2)
+ /// %l = load %ptr
+ /// %addr1 = gep, %alloca2, 0, %l
+ /// %addr2 = gep %alloca2, 0, (%l + 1)
+ /// alias(%p, %addr1) -> MayAlias !
+ /// store %l, ...
+ SmallPtrSet<const BasicBlock*, 8> VisitedPhiBBs;
+
+ // Visited - Track instructions visited by pointsToConstantMemory.
+ SmallPtrSet<const Value*, 16> Visited;
+
+ /// \brief Check whether two Values can be considered equivalent.
+ ///
+ /// In addition to pointer equivalence of \p V1 and \p V2 this checks
+ /// whether they can not be part of a cycle in the value graph by looking at
+ /// all visited phi nodes an making sure that the phis cannot reach the
+ /// value. We have to do this because we are looking through phi nodes (That
+ /// is we say noalias(V, phi(VA, VB)) if noalias(V, VA) and noalias(V, VB).
+ bool isValueEqualInPotentialCycles(const Value *V1, const Value *V2);
+
+ /// \brief Dest and Src are the variable indices from two decomposed
+ /// GetElementPtr instructions GEP1 and GEP2 which have common base
+ /// pointers. Subtract the GEP2 indices from GEP1 to find the symbolic
+ /// difference between the two pointers.
+ void GetIndexDifference(SmallVectorImpl<VariableGEPIndex> &Dest,
+ const SmallVectorImpl<VariableGEPIndex> &Src);
+
+ // aliasGEP - Provide a bunch of ad-hoc rules to disambiguate a GEP
+ // instruction against another.
+ AliasResult aliasGEP(const GEPOperator *V1, uint64_t V1Size,
+ const AAMDNodes &V1AAInfo,
+ const Value *V2, uint64_t V2Size,
+ const AAMDNodes &V2AAInfo,
+ const Value *UnderlyingV1, const Value *UnderlyingV2);
+
+ // aliasPHI - Provide a bunch of ad-hoc rules to disambiguate a PHI
+ // instruction against another.
+ AliasResult aliasPHI(const PHINode *PN, uint64_t PNSize,
+ const AAMDNodes &PNAAInfo,
+ const Value *V2, uint64_t V2Size,
+ const AAMDNodes &V2AAInfo);
/// aliasSelect - Disambiguate a Select instruction against another value.
- AliasResult aliasSelect(const SelectInst *SI, unsigned SISize,
- const Value *V2, unsigned V2Size);
-
- AliasResult aliasCheck(const Value *V1, unsigned V1Size,
- const Value *V2, unsigned V2Size);
-
- // CheckGEPInstructions - Check two GEP instructions with known
- // must-aliasing base pointers. This checks to see if the index expressions
- // preclude the pointers from aliasing...
- AliasResult
- CheckGEPInstructions(const Type* BasePtr1Ty,
- Value **GEP1Ops, unsigned NumGEP1Ops, unsigned G1Size,
- const Type *BasePtr2Ty,
- Value **GEP2Ops, unsigned NumGEP2Ops, unsigned G2Size);
+ AliasResult aliasSelect(const SelectInst *SI, uint64_t SISize,
+ const AAMDNodes &SIAAInfo,
+ const Value *V2, uint64_t V2Size,
+ const AAMDNodes &V2AAInfo);
+
+ AliasResult aliasCheck(const Value *V1, uint64_t V1Size,
+ AAMDNodes V1AATag,
+ const Value *V2, uint64_t V2Size,
+ AAMDNodes V2AATag);
};
} // End of anonymous namespace
// Register this pass...
char BasicAliasAnalysis::ID = 0;
-static RegisterPass<BasicAliasAnalysis>
-X("basicaa", "Basic Alias Analysis (default AA impl)", false, true);
+INITIALIZE_AG_PASS_BEGIN(BasicAliasAnalysis, AliasAnalysis, "basicaa",
+ "Basic Alias Analysis (stateless AA impl)",
+ false, true, false)
+INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
+INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)
+INITIALIZE_AG_PASS_END(BasicAliasAnalysis, AliasAnalysis, "basicaa",
+ "Basic Alias Analysis (stateless AA impl)",
+ false, true, false)
-// Declare that we implement the AliasAnalysis interface
-static RegisterAnalysisGroup<AliasAnalysis, true> Y(X);
ImmutablePass *llvm::createBasicAliasAnalysisPass() {
return new BasicAliasAnalysis();
}
+/// pointsToConstantMemory - Returns whether the given pointer value
+/// points to memory that is local to the function, with global constants being
+/// considered local to all functions.
+bool
+BasicAliasAnalysis::pointsToConstantMemory(const Location &Loc, bool OrLocal) {
+ assert(Visited.empty() && "Visited must be cleared after use!");
+
+ unsigned MaxLookup = 8;
+ SmallVector<const Value *, 16> Worklist;
+ Worklist.push_back(Loc.Ptr);
+ do {
+ const Value *V = GetUnderlyingObject(Worklist.pop_back_val(), *DL);
+ if (!Visited.insert(V).second) {
+ Visited.clear();
+ return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal);
+ }
+
+ // An alloca instruction defines local memory.
+ if (OrLocal && isa<AllocaInst>(V))
+ continue;
+
+ // A global constant counts as local memory for our purposes.
+ if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(V)) {
+ // Note: this doesn't require GV to be "ODR" because it isn't legal for a
+ // global to be marked constant in some modules and non-constant in
+ // others. GV may even be a declaration, not a definition.
+ if (!GV->isConstant()) {
+ Visited.clear();
+ return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal);
+ }
+ continue;
+ }
+
+ // If both select values point to local memory, then so does the select.
+ if (const SelectInst *SI = dyn_cast<SelectInst>(V)) {
+ Worklist.push_back(SI->getTrueValue());
+ Worklist.push_back(SI->getFalseValue());
+ continue;
+ }
+
+ // If all values incoming to a phi node point to local memory, then so does
+ // the phi.
+ if (const PHINode *PN = dyn_cast<PHINode>(V)) {
+ // Don't bother inspecting phi nodes with many operands.
+ if (PN->getNumIncomingValues() > MaxLookup) {
+ Visited.clear();
+ return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal);
+ }
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
+ Worklist.push_back(PN->getIncomingValue(i));
+ continue;
+ }
+
+ // Otherwise be conservative.
+ Visited.clear();
+ return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal);
+
+ } while (!Worklist.empty() && --MaxLookup);
+
+ Visited.clear();
+ return Worklist.empty();
+}
+
+static bool isMemsetPattern16(const Function *MS,
+ const TargetLibraryInfo &TLI) {
+ if (TLI.has(LibFunc::memset_pattern16) &&
+ MS->getName() == "memset_pattern16") {
+ FunctionType *MemsetType = MS->getFunctionType();
+ if (!MemsetType->isVarArg() && MemsetType->getNumParams() == 3 &&
+ isa<PointerType>(MemsetType->getParamType(0)) &&
+ isa<PointerType>(MemsetType->getParamType(1)) &&
+ isa<IntegerType>(MemsetType->getParamType(2)))
+ return true;
+ }
+
+ return false;
+}
+
+/// getModRefBehavior - Return the behavior when calling the given call site.
+AliasAnalysis::ModRefBehavior
+BasicAliasAnalysis::getModRefBehavior(ImmutableCallSite CS) {
+ if (CS.doesNotAccessMemory())
+ // Can't do better than this.
+ return DoesNotAccessMemory;
+
+ ModRefBehavior Min = UnknownModRefBehavior;
+
+ // If the callsite knows it only reads memory, don't return worse
+ // than that.
+ if (CS.onlyReadsMemory())
+ Min = OnlyReadsMemory;
+
+ // The AliasAnalysis base class has some smarts, lets use them.
+ return ModRefBehavior(AliasAnalysis::getModRefBehavior(CS) & Min);
+}
+
+/// getModRefBehavior - Return the behavior when calling the given function.
+/// For use when the call site is not known.
+AliasAnalysis::ModRefBehavior
+BasicAliasAnalysis::getModRefBehavior(const Function *F) {
+ // If the function declares it doesn't access memory, we can't do better.
+ if (F->doesNotAccessMemory())
+ return DoesNotAccessMemory;
+
+ // For intrinsics, we can check the table.
+ if (unsigned iid = F->getIntrinsicID()) {
+#define GET_INTRINSIC_MODREF_BEHAVIOR
+#include "llvm/IR/Intrinsics.gen"
+#undef GET_INTRINSIC_MODREF_BEHAVIOR
+ }
+
+ ModRefBehavior Min = UnknownModRefBehavior;
+
+ // If the function declares it only reads memory, go with that.
+ if (F->onlyReadsMemory())
+ Min = OnlyReadsMemory;
+
+ const TargetLibraryInfo &TLI =
+ getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
+ if (isMemsetPattern16(F, TLI))
+ Min = OnlyAccessesArgumentPointees;
+
+ // Otherwise be conservative.
+ return ModRefBehavior(AliasAnalysis::getModRefBehavior(F) & Min);
+}
+
+AliasAnalysis::Location
+BasicAliasAnalysis::getArgLocation(ImmutableCallSite CS, unsigned ArgIdx,
+ ModRefResult &Mask) {
+ Location Loc = AliasAnalysis::getArgLocation(CS, ArgIdx, Mask);
+ const TargetLibraryInfo &TLI =
+ getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
+ const IntrinsicInst *II = dyn_cast<IntrinsicInst>(CS.getInstruction());
+ if (II != nullptr)
+ switch (II->getIntrinsicID()) {
+ default: break;
+ case Intrinsic::memset:
+ case Intrinsic::memcpy:
+ case Intrinsic::memmove: {
+ assert((ArgIdx == 0 || ArgIdx == 1) &&
+ "Invalid argument index for memory intrinsic");
+ if (ConstantInt *LenCI = dyn_cast<ConstantInt>(II->getArgOperand(2)))
+ Loc.Size = LenCI->getZExtValue();
+ assert(Loc.Ptr == II->getArgOperand(ArgIdx) &&
+ "Memory intrinsic location pointer not argument?");
+ Mask = ArgIdx ? Ref : Mod;
+ break;
+ }
+ case Intrinsic::lifetime_start:
+ case Intrinsic::lifetime_end:
+ case Intrinsic::invariant_start: {
+ assert(ArgIdx == 1 && "Invalid argument index");
+ assert(Loc.Ptr == II->getArgOperand(ArgIdx) &&
+ "Intrinsic location pointer not argument?");
+ Loc.Size = cast<ConstantInt>(II->getArgOperand(0))->getZExtValue();
+ break;
+ }
+ case Intrinsic::invariant_end: {
+ assert(ArgIdx == 2 && "Invalid argument index");
+ assert(Loc.Ptr == II->getArgOperand(ArgIdx) &&
+ "Intrinsic location pointer not argument?");
+ Loc.Size = cast<ConstantInt>(II->getArgOperand(1))->getZExtValue();
+ break;
+ }
+ case Intrinsic::arm_neon_vld1: {
+ assert(ArgIdx == 0 && "Invalid argument index");
+ assert(Loc.Ptr == II->getArgOperand(ArgIdx) &&
+ "Intrinsic location pointer not argument?");
+ // LLVM's vld1 and vst1 intrinsics currently only support a single
+ // vector register.
+ if (DL)
+ Loc.Size = DL->getTypeStoreSize(II->getType());
+ break;
+ }
+ case Intrinsic::arm_neon_vst1: {
+ assert(ArgIdx == 0 && "Invalid argument index");
+ assert(Loc.Ptr == II->getArgOperand(ArgIdx) &&
+ "Intrinsic location pointer not argument?");
+ if (DL)
+ Loc.Size = DL->getTypeStoreSize(II->getArgOperand(1)->getType());
+ break;
+ }
+ }
+
+ // We can bound the aliasing properties of memset_pattern16 just as we can
+ // for memcpy/memset. This is particularly important because the
+ // LoopIdiomRecognizer likes to turn loops into calls to memset_pattern16
+ // whenever possible.
+ else if (CS.getCalledFunction() &&
+ isMemsetPattern16(CS.getCalledFunction(), TLI)) {
+ assert((ArgIdx == 0 || ArgIdx == 1) &&
+ "Invalid argument index for memset_pattern16");
+ if (ArgIdx == 1)
+ Loc.Size = 16;
+ else if (const ConstantInt *LenCI =
+ dyn_cast<ConstantInt>(CS.getArgument(2)))
+ Loc.Size = LenCI->getZExtValue();
+ assert(Loc.Ptr == CS.getArgument(ArgIdx) &&
+ "memset_pattern16 location pointer not argument?");
+ Mask = ArgIdx ? Ref : Mod;
+ }
+ // FIXME: Handle memset_pattern4 and memset_pattern8 also.
+
+ return Loc;
+}
+
+static bool isAssumeIntrinsic(ImmutableCallSite CS) {
+ const IntrinsicInst *II = dyn_cast<IntrinsicInst>(CS.getInstruction());
+ if (II && II->getIntrinsicID() == Intrinsic::assume)
+ return true;
-/// pointsToConstantMemory - Chase pointers until we find a (constant
-/// global) or not.
-bool BasicAliasAnalysis::pointsToConstantMemory(const Value *P) {
- if (const GlobalVariable *GV =
- dyn_cast<GlobalVariable>(P->getUnderlyingObject()))
- return GV->isConstant();
return false;
}
+bool BasicAliasAnalysis::doInitialization(Module &M) {
+ InitializeAliasAnalysis(this, &M.getDataLayout());
+ return true;
+}
-// getModRefInfo - Check to see if the specified callsite can clobber the
-// specified memory object. Since we only look at local properties of this
-// function, we really can't say much about this query. We do, however, use
-// simple "address taken" analysis on local objects.
-//
+/// getModRefInfo - Check to see if the specified callsite can clobber the
+/// specified memory object. Since we only look at local properties of this
+/// function, we really can't say much about this query. We do, however, use
+/// simple "address taken" analysis on local objects.
AliasAnalysis::ModRefResult
-BasicAliasAnalysis::getModRefInfo(CallSite CS, Value *P, unsigned Size) {
- if (!isa<Constant>(P)) {
- const Value *Object = P->getUnderlyingObject();
-
- // If this is a tail call and P points to a stack location, we know that
- // the tail call cannot access or modify the local stack.
- // We cannot exclude byval arguments here; these belong to the caller of
- // the current function not to the current function, and a tail callee
- // may reference them.
- if (isa<AllocaInst>(Object))
- if (CallInst *CI = dyn_cast<CallInst>(CS.getInstruction()))
- if (CI->isTailCall())
- return NoModRef;
-
- // If the pointer is to a locally allocated object that does not escape,
- // then the call can not mod/ref the pointer unless the call takes the
- // argument without capturing it.
- if (isNonEscapingLocalObject(Object) && CS.getInstruction() != Object) {
- bool passedAsArg = false;
- // TODO: Eventually only check 'nocapture' arguments.
- for (CallSite::arg_iterator CI = CS.arg_begin(), CE = CS.arg_end();
- CI != CE; ++CI)
- if (isa<PointerType>((*CI)->getType()) &&
- alias(cast<Value>(CI), ~0U, P, ~0U) != NoAlias)
- passedAsArg = true;
-
- if (!passedAsArg)
+BasicAliasAnalysis::getModRefInfo(ImmutableCallSite CS,
+ const Location &Loc) {
+ assert(notDifferentParent(CS.getInstruction(), Loc.Ptr) &&
+ "AliasAnalysis query involving multiple functions!");
+
+ const Value *Object = GetUnderlyingObject(Loc.Ptr, *DL);
+
+ // If this is a tail call and Loc.Ptr points to a stack location, we know that
+ // the tail call cannot access or modify the local stack.
+ // We cannot exclude byval arguments here; these belong to the caller of
+ // the current function not to the current function, and a tail callee
+ // may reference them.
+ if (isa<AllocaInst>(Object))
+ if (const CallInst *CI = dyn_cast<CallInst>(CS.getInstruction()))
+ if (CI->isTailCall())
return NoModRef;
- }
- if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(CS.getInstruction())) {
- switch (II->getIntrinsicID()) {
- default: break;
- case Intrinsic::memcpy:
- case Intrinsic::memmove: {
- unsigned Len = ~0U;
- if (ConstantInt *LenCI = dyn_cast<ConstantInt>(II->getOperand(3)))
- Len = LenCI->getZExtValue();
- Value *Dest = II->getOperand(1);
- Value *Src = II->getOperand(2);
- if (alias(Dest, Len, P, Size) == NoAlias) {
- if (alias(Src, Len, P, Size) == NoAlias)
- return NoModRef;
- return Ref;
- }
- }
- break;
- case Intrinsic::memset:
- if (ConstantInt *LenCI = dyn_cast<ConstantInt>(II->getOperand(3))) {
- unsigned Len = LenCI->getZExtValue();
- Value *Dest = II->getOperand(1);
- if (alias(Dest, Len, P, Size) == NoAlias)
- return NoModRef;
- }
+ // If the pointer is to a locally allocated object that does not escape,
+ // then the call can not mod/ref the pointer unless the call takes the pointer
+ // as an argument, and itself doesn't capture it.
+ if (!isa<Constant>(Object) && CS.getInstruction() != Object &&
+ isNonEscapingLocalObject(Object)) {
+ bool PassedAsArg = false;
+ unsigned ArgNo = 0;
+ for (ImmutableCallSite::arg_iterator CI = CS.arg_begin(), CE = CS.arg_end();
+ CI != CE; ++CI, ++ArgNo) {
+ // Only look at the no-capture or byval pointer arguments. If this
+ // pointer were passed to arguments that were neither of these, then it
+ // couldn't be no-capture.
+ if (!(*CI)->getType()->isPointerTy() ||
+ (!CS.doesNotCapture(ArgNo) && !CS.isByValArgument(ArgNo)))
+ continue;
+
+ // If this is a no-capture pointer argument, see if we can tell that it
+ // is impossible to alias the pointer we're checking. If not, we have to
+ // assume that the call could touch the pointer, even though it doesn't
+ // escape.
+ if (!isNoAlias(Location(*CI), Location(Object))) {
+ PassedAsArg = true;
break;
- case Intrinsic::atomic_cmp_swap:
- case Intrinsic::atomic_swap:
- case Intrinsic::atomic_load_add:
- case Intrinsic::atomic_load_sub:
- case Intrinsic::atomic_load_and:
- case Intrinsic::atomic_load_nand:
- case Intrinsic::atomic_load_or:
- case Intrinsic::atomic_load_xor:
- case Intrinsic::atomic_load_max:
- case Intrinsic::atomic_load_min:
- case Intrinsic::atomic_load_umax:
- case Intrinsic::atomic_load_umin:
- if (TD) {
- Value *Op1 = II->getOperand(1);
- unsigned Op1Size = TD->getTypeStoreSize(Op1->getType());
- if (alias(Op1, Op1Size, P, Size) == NoAlias)
- return NoModRef;
- }
- break;
- case Intrinsic::lifetime_start:
- case Intrinsic::lifetime_end:
- case Intrinsic::invariant_start: {
- unsigned PtrSize = cast<ConstantInt>(II->getOperand(1))->getZExtValue();
- if (alias(II->getOperand(2), PtrSize, P, Size) == NoAlias)
- return NoModRef;
- }
- break;
- case Intrinsic::invariant_end: {
- unsigned PtrSize = cast<ConstantInt>(II->getOperand(2))->getZExtValue();
- if (alias(II->getOperand(3), PtrSize, P, Size) == NoAlias)
- return NoModRef;
- }
- break;
}
}
+
+ if (!PassedAsArg)
+ return NoModRef;
}
+ // While the assume intrinsic is marked as arbitrarily writing so that
+ // proper control dependencies will be maintained, it never aliases any
+ // particular memory location.
+ if (isAssumeIntrinsic(CS))
+ return NoModRef;
+
// The AliasAnalysis base class has some smarts, lets use them.
- return AliasAnalysis::getModRefInfo(CS, P, Size);
+ return AliasAnalysis::getModRefInfo(CS, Loc);
}
+AliasAnalysis::ModRefResult
+BasicAliasAnalysis::getModRefInfo(ImmutableCallSite CS1,
+ ImmutableCallSite CS2) {
+ // While the assume intrinsic is marked as arbitrarily writing so that
+ // proper control dependencies will be maintained, it never aliases any
+ // particular memory location.
+ if (isAssumeIntrinsic(CS1) || isAssumeIntrinsic(CS2))
+ return NoModRef;
-AliasAnalysis::ModRefResult
-BasicAliasAnalysis::getModRefInfo(CallSite CS1, CallSite CS2) {
- // If CS1 or CS2 are readnone, they don't interact.
- ModRefBehavior CS1B = AliasAnalysis::getModRefBehavior(CS1);
- if (CS1B == DoesNotAccessMemory) return NoModRef;
-
- ModRefBehavior CS2B = AliasAnalysis::getModRefBehavior(CS2);
- if (CS2B == DoesNotAccessMemory) return NoModRef;
-
- // If they both only read from memory, just return ref.
- if (CS1B == OnlyReadsMemory && CS2B == OnlyReadsMemory)
- return Ref;
-
- // Otherwise, fall back to NoAA (mod+ref).
- return NoAA::getModRefInfo(CS1, CS2);
+ // The AliasAnalysis base class has some smarts, lets use them.
+ return AliasAnalysis::getModRefInfo(CS1, CS2);
}
-// aliasGEP - Provide a bunch of ad-hoc rules to disambiguate a GEP instruction
-// against another.
-//
-AliasAnalysis::AliasResult
-BasicAliasAnalysis::aliasGEP(const Value *V1, unsigned V1Size,
- const Value *V2, unsigned V2Size) {
- // If we have two gep instructions with must-alias'ing base pointers, figure
- // out if the indexes to the GEP tell us anything about the derived pointer.
- // Note that we also handle chains of getelementptr instructions as well as
- // constant expression getelementptrs here.
+/// \brief Provide ad-hoc rules to disambiguate accesses through two GEP
+/// operators, both having the exact same pointer operand.
+static AliasAnalysis::AliasResult
+aliasSameBasePointerGEPs(const GEPOperator *GEP1, uint64_t V1Size,
+ const GEPOperator *GEP2, uint64_t V2Size,
+ const DataLayout &DL) {
+
+ assert(GEP1->getPointerOperand() == GEP2->getPointerOperand() &&
+ "Expected GEPs with the same pointer operand");
+
+ // Try to determine whether GEP1 and GEP2 index through arrays, into structs,
+ // such that the struct field accesses provably cannot alias.
+ // We also need at least two indices (the pointer, and the struct field).
+ if (GEP1->getNumIndices() != GEP2->getNumIndices() ||
+ GEP1->getNumIndices() < 2)
+ return AliasAnalysis::MayAlias;
+
+ // If we don't know the size of the accesses through both GEPs, we can't
+ // determine whether the struct fields accessed can't alias.
+ if (V1Size == AliasAnalysis::UnknownSize ||
+ V2Size == AliasAnalysis::UnknownSize)
+ return AliasAnalysis::MayAlias;
+
+ ConstantInt *C1 =
+ dyn_cast<ConstantInt>(GEP1->getOperand(GEP1->getNumOperands() - 1));
+ ConstantInt *C2 =
+ dyn_cast<ConstantInt>(GEP2->getOperand(GEP2->getNumOperands() - 1));
+
+ // If the last (struct) indices aren't constants, we can't say anything.
+ // If they're identical, the other indices might be also be dynamically
+ // equal, so the GEPs can alias.
+ if (!C1 || !C2 || C1 == C2)
+ return AliasAnalysis::MayAlias;
+
+ // Find the last-indexed type of the GEP, i.e., the type you'd get if
+ // you stripped the last index.
+ // On the way, look at each indexed type. If there's something other
+ // than an array, different indices can lead to different final types.
+ SmallVector<Value *, 8> IntermediateIndices;
+
+ // Insert the first index; we don't need to check the type indexed
+ // through it as it only drops the pointer indirection.
+ assert(GEP1->getNumIndices() > 1 && "Not enough GEP indices to examine");
+ IntermediateIndices.push_back(GEP1->getOperand(1));
+
+ // Insert all the remaining indices but the last one.
+ // Also, check that they all index through arrays.
+ for (unsigned i = 1, e = GEP1->getNumIndices() - 1; i != e; ++i) {
+ if (!isa<ArrayType>(GetElementPtrInst::getIndexedType(
+ GEP1->getPointerOperandType(), IntermediateIndices)))
+ return AliasAnalysis::MayAlias;
+ IntermediateIndices.push_back(GEP1->getOperand(i + 1));
+ }
+
+ StructType *LastIndexedStruct =
+ dyn_cast<StructType>(GetElementPtrInst::getIndexedType(
+ GEP1->getPointerOperandType(), IntermediateIndices));
+
+ if (!LastIndexedStruct)
+ return AliasAnalysis::MayAlias;
+
+ // We know that:
+ // - both GEPs begin indexing from the exact same pointer;
+ // - the last indices in both GEPs are constants, indexing into a struct;
+ // - said indices are different, hence, the pointed-to fields are different;
+ // - both GEPs only index through arrays prior to that.
//
- if (isa<GEPOperator>(V1) && isa<GEPOperator>(V2)) {
- const User *GEP1 = cast<User>(V1);
- const User *GEP2 = cast<User>(V2);
-
- // If V1 and V2 are identical GEPs, just recurse down on both of them.
- // This allows us to analyze things like:
- // P = gep A, 0, i, 1
- // Q = gep B, 0, i, 1
- // by just analyzing A and B. This is even safe for variable indices.
- if (GEP1->getType() == GEP2->getType() &&
- GEP1->getNumOperands() == GEP2->getNumOperands() &&
- GEP1->getOperand(0)->getType() == GEP2->getOperand(0)->getType() &&
- // All operands are the same, ignoring the base.
- std::equal(GEP1->op_begin()+1, GEP1->op_end(), GEP2->op_begin()+1))
- return aliasCheck(GEP1->getOperand(0), V1Size,
- GEP2->getOperand(0), V2Size);
-
- // Drill down into the first non-gep value, to test for must-aliasing of
- // the base pointers.
- while (isa<GEPOperator>(GEP1->getOperand(0)) &&
- GEP1->getOperand(1) ==
- Constant::getNullValue(GEP1->getOperand(1)->getType()))
- GEP1 = cast<User>(GEP1->getOperand(0));
- const Value *BasePtr1 = GEP1->getOperand(0);
-
- while (isa<GEPOperator>(GEP2->getOperand(0)) &&
- GEP2->getOperand(1) ==
- Constant::getNullValue(GEP2->getOperand(1)->getType()))
- GEP2 = cast<User>(GEP2->getOperand(0));
- const Value *BasePtr2 = GEP2->getOperand(0);
+ // This lets us determine that the struct that GEP1 indexes into and the
+ // struct that GEP2 indexes into must either precisely overlap or be
+ // completely disjoint. Because they cannot partially overlap, indexing into
+ // different non-overlapping fields of the struct will never alias.
+
+ // Therefore, the only remaining thing needed to show that both GEPs can't
+ // alias is that the fields are not overlapping.
+ const StructLayout *SL = DL.getStructLayout(LastIndexedStruct);
+ const uint64_t StructSize = SL->getSizeInBytes();
+ const uint64_t V1Off = SL->getElementOffset(C1->getZExtValue());
+ const uint64_t V2Off = SL->getElementOffset(C2->getZExtValue());
+
+ auto EltsDontOverlap = [StructSize](uint64_t V1Off, uint64_t V1Size,
+ uint64_t V2Off, uint64_t V2Size) {
+ return V1Off < V2Off && V1Off + V1Size <= V2Off &&
+ ((V2Off + V2Size <= StructSize) ||
+ (V2Off + V2Size - StructSize <= V1Off));
+ };
+ if (EltsDontOverlap(V1Off, V1Size, V2Off, V2Size) ||
+ EltsDontOverlap(V2Off, V2Size, V1Off, V1Size))
+ return AliasAnalysis::NoAlias;
+
+ return AliasAnalysis::MayAlias;
+}
+
+/// aliasGEP - Provide a bunch of ad-hoc rules to disambiguate a GEP instruction
+/// against another pointer. We know that V1 is a GEP, but we don't know
+/// anything about V2. UnderlyingV1 is GetUnderlyingObject(GEP1, DL),
+/// UnderlyingV2 is the same for V2.
+///
+AliasAnalysis::AliasResult
+BasicAliasAnalysis::aliasGEP(const GEPOperator *GEP1, uint64_t V1Size,
+ const AAMDNodes &V1AAInfo,
+ const Value *V2, uint64_t V2Size,
+ const AAMDNodes &V2AAInfo,
+ const Value *UnderlyingV1,
+ const Value *UnderlyingV2) {
+ int64_t GEP1BaseOffset;
+ bool GEP1MaxLookupReached;
+ SmallVector<VariableGEPIndex, 4> GEP1VariableIndices;
+
+ // We have to get two AssumptionCaches here because GEP1 and V2 may be from
+ // different functions.
+ // FIXME: This really doesn't make any sense. We get a dominator tree below
+ // that can only refer to a single function. But this function (aliasGEP) is
+ // a method on an immutable pass that can be called when there *isn't*
+ // a single function. The old pass management layer makes this "work", but
+ // this isn't really a clean solution.
+ AssumptionCacheTracker &ACT = getAnalysis<AssumptionCacheTracker>();
+ AssumptionCache *AC1 = nullptr, *AC2 = nullptr;
+ if (auto *GEP1I = dyn_cast<Instruction>(GEP1))
+ AC1 = &ACT.getAssumptionCache(
+ const_cast<Function &>(*GEP1I->getParent()->getParent()));
+ if (auto *I2 = dyn_cast<Instruction>(V2))
+ AC2 = &ACT.getAssumptionCache(
+ const_cast<Function &>(*I2->getParent()->getParent()));
+
+ DominatorTreeWrapperPass *DTWP =
+ getAnalysisIfAvailable<DominatorTreeWrapperPass>();
+ DominatorTree *DT = DTWP ? &DTWP->getDomTree() : nullptr;
+
+ // If we have two gep instructions with must-alias or not-alias'ing base
+ // pointers, figure out if the indexes to the GEP tell us anything about the
+ // derived pointer.
+ if (const GEPOperator *GEP2 = dyn_cast<GEPOperator>(V2)) {
// Do the base pointers alias?
- AliasResult BaseAlias = aliasCheck(BasePtr1, ~0U, BasePtr2, ~0U);
- if (BaseAlias == NoAlias) return NoAlias;
- if (BaseAlias == MustAlias) {
- // If the base pointers alias each other exactly, check to see if we can
- // figure out anything about the resultant pointers, to try to prove
- // non-aliasing.
-
- // Collect all of the chained GEP operands together into one simple place
- SmallVector<Value*, 16> GEP1Ops, GEP2Ops;
- BasePtr1 = GetGEPOperands(V1, GEP1Ops);
- BasePtr2 = GetGEPOperands(V2, GEP2Ops);
-
- // If GetGEPOperands were able to fold to the same must-aliased pointer,
- // do the comparison.
- if (BasePtr1 == BasePtr2) {
- AliasResult GAlias =
- CheckGEPInstructions(BasePtr1->getType(),
- &GEP1Ops[0], GEP1Ops.size(), V1Size,
- BasePtr2->getType(),
- &GEP2Ops[0], GEP2Ops.size(), V2Size);
- if (GAlias != MayAlias)
- return GAlias;
+ AliasResult BaseAlias = aliasCheck(UnderlyingV1, UnknownSize, AAMDNodes(),
+ UnderlyingV2, UnknownSize, AAMDNodes());
+
+ // Check for geps of non-aliasing underlying pointers where the offsets are
+ // identical.
+ if ((BaseAlias == MayAlias) && V1Size == V2Size) {
+ // Do the base pointers alias assuming type and size.
+ AliasResult PreciseBaseAlias = aliasCheck(UnderlyingV1, V1Size,
+ V1AAInfo, UnderlyingV2,
+ V2Size, V2AAInfo);
+ if (PreciseBaseAlias == NoAlias) {
+ // See if the computed offset from the common pointer tells us about the
+ // relation of the resulting pointer.
+ int64_t GEP2BaseOffset;
+ bool GEP2MaxLookupReached;
+ SmallVector<VariableGEPIndex, 4> GEP2VariableIndices;
+ const Value *GEP2BasePtr =
+ DecomposeGEPExpression(GEP2, GEP2BaseOffset, GEP2VariableIndices,
+ GEP2MaxLookupReached, *DL, AC2, DT);
+ const Value *GEP1BasePtr =
+ DecomposeGEPExpression(GEP1, GEP1BaseOffset, GEP1VariableIndices,
+ GEP1MaxLookupReached, *DL, AC1, DT);
+ // DecomposeGEPExpression and GetUnderlyingObject should return the
+ // same result except when DecomposeGEPExpression has no DataLayout.
+ if (GEP1BasePtr != UnderlyingV1 || GEP2BasePtr != UnderlyingV2) {
+ assert(!DL &&
+ "DecomposeGEPExpression and GetUnderlyingObject disagree!");
+ return MayAlias;
+ }
+ // If the max search depth is reached the result is undefined
+ if (GEP2MaxLookupReached || GEP1MaxLookupReached)
+ return MayAlias;
+
+ // Same offsets.
+ if (GEP1BaseOffset == GEP2BaseOffset &&
+ GEP1VariableIndices == GEP2VariableIndices)
+ return NoAlias;
+ GEP1VariableIndices.clear();
}
}
+
+ // If we get a No or May, then return it immediately, no amount of analysis
+ // will improve this situation.
+ if (BaseAlias != MustAlias) return BaseAlias;
+
+ // Otherwise, we have a MustAlias. Since the base pointers alias each other
+ // exactly, see if the computed offset from the common pointer tells us
+ // about the relation of the resulting pointer.
+ const Value *GEP1BasePtr =
+ DecomposeGEPExpression(GEP1, GEP1BaseOffset, GEP1VariableIndices,
+ GEP1MaxLookupReached, *DL, AC1, DT);
+
+ int64_t GEP2BaseOffset;
+ bool GEP2MaxLookupReached;
+ SmallVector<VariableGEPIndex, 4> GEP2VariableIndices;
+ const Value *GEP2BasePtr =
+ DecomposeGEPExpression(GEP2, GEP2BaseOffset, GEP2VariableIndices,
+ GEP2MaxLookupReached, *DL, AC2, DT);
+
+ // DecomposeGEPExpression and GetUnderlyingObject should return the
+ // same result except when DecomposeGEPExpression has no DataLayout.
+ if (GEP1BasePtr != UnderlyingV1 || GEP2BasePtr != UnderlyingV2) {
+ assert(!DL &&
+ "DecomposeGEPExpression and GetUnderlyingObject disagree!");
+ return MayAlias;
+ }
+
+ // If we know the two GEPs are based off of the exact same pointer (and not
+ // just the same underlying object), see if that tells us anything about
+ // the resulting pointers.
+ if (DL && GEP1->getPointerOperand() == GEP2->getPointerOperand()) {
+ AliasResult R = aliasSameBasePointerGEPs(GEP1, V1Size, GEP2, V2Size, *DL);
+ // If we couldn't find anything interesting, don't abandon just yet.
+ if (R != MayAlias)
+ return R;
+ }
+
+ // If the max search depth is reached the result is undefined
+ if (GEP2MaxLookupReached || GEP1MaxLookupReached)
+ return MayAlias;
+
+ // Subtract the GEP2 pointer from the GEP1 pointer to find out their
+ // symbolic difference.
+ GEP1BaseOffset -= GEP2BaseOffset;
+ GetIndexDifference(GEP1VariableIndices, GEP2VariableIndices);
+
+ } else {
+ // Check to see if these two pointers are related by the getelementptr
+ // instruction. If one pointer is a GEP with a non-zero index of the other
+ // pointer, we know they cannot alias.
+
+ // If both accesses are unknown size, we can't do anything useful here.
+ if (V1Size == UnknownSize && V2Size == UnknownSize)
+ return MayAlias;
+
+ AliasResult R = aliasCheck(UnderlyingV1, UnknownSize, AAMDNodes(),
+ V2, V2Size, V2AAInfo);
+ if (R != MustAlias)
+ // If V2 may alias GEP base pointer, conservatively returns MayAlias.
+ // If V2 is known not to alias GEP base pointer, then the two values
+ // cannot alias per GEP semantics: "A pointer value formed from a
+ // getelementptr instruction is associated with the addresses associated
+ // with the first operand of the getelementptr".
+ return R;
+
+ const Value *GEP1BasePtr =
+ DecomposeGEPExpression(GEP1, GEP1BaseOffset, GEP1VariableIndices,
+ GEP1MaxLookupReached, *DL, AC1, DT);
+
+ // DecomposeGEPExpression and GetUnderlyingObject should return the
+ // same result except when DecomposeGEPExpression has no DataLayout.
+ if (GEP1BasePtr != UnderlyingV1) {
+ assert(!DL &&
+ "DecomposeGEPExpression and GetUnderlyingObject disagree!");
+ return MayAlias;
+ }
+ // If the max search depth is reached the result is undefined
+ if (GEP1MaxLookupReached)
+ return MayAlias;
}
- // Check to see if these two pointers are related by a getelementptr
- // instruction. If one pointer is a GEP with a non-zero index of the other
- // pointer, we know they cannot alias.
+ // In the two GEP Case, if there is no difference in the offsets of the
+ // computed pointers, the resultant pointers are a must alias. This
+ // hapens when we have two lexically identical GEP's (for example).
//
- if (V1Size == ~0U || V2Size == ~0U)
- return MayAlias;
+ // In the other case, if we have getelementptr <ptr>, 0, 0, 0, 0, ... and V2
+ // must aliases the GEP, the end result is a must alias also.
+ if (GEP1BaseOffset == 0 && GEP1VariableIndices.empty())
+ return MustAlias;
- SmallVector<Value*, 16> GEPOperands;
- const Value *BasePtr = GetGEPOperands(V1, GEPOperands);
-
- AliasResult R = aliasCheck(BasePtr, ~0U, V2, V2Size);
- if (R != MustAlias)
- // If V2 may alias GEP base pointer, conservatively returns MayAlias.
- // If V2 is known not to alias GEP base pointer, then the two values
- // cannot alias per GEP semantics: "A pointer value formed from a
- // getelementptr instruction is associated with the addresses associated
- // with the first operand of the getelementptr".
- return R;
-
- // If there is at least one non-zero constant index, we know they cannot
- // alias.
- bool ConstantFound = false;
- bool AllZerosFound = true;
- for (unsigned i = 0, e = GEPOperands.size(); i != e; ++i)
- if (const Constant *C = dyn_cast<Constant>(GEPOperands[i])) {
- if (!C->isNullValue()) {
- ConstantFound = true;
- AllZerosFound = false;
- break;
+ // If there is a constant difference between the pointers, but the difference
+ // is less than the size of the associated memory object, then we know
+ // that the objects are partially overlapping. If the difference is
+ // greater, we know they do not overlap.
+ if (GEP1BaseOffset != 0 && GEP1VariableIndices.empty()) {
+ if (GEP1BaseOffset >= 0) {
+ if (V2Size != UnknownSize) {
+ if ((uint64_t)GEP1BaseOffset < V2Size)
+ return PartialAlias;
+ return NoAlias;
}
} else {
- AllZerosFound = false;
+ // We have the situation where:
+ // + +
+ // | BaseOffset |
+ // ---------------->|
+ // |-->V1Size |-------> V2Size
+ // GEP1 V2
+ // We need to know that V2Size is not unknown, otherwise we might have
+ // stripped a gep with negative index ('gep <ptr>, -1, ...).
+ if (V1Size != UnknownSize && V2Size != UnknownSize) {
+ if (-(uint64_t)GEP1BaseOffset < V1Size)
+ return PartialAlias;
+ return NoAlias;
+ }
+ }
+ }
+
+ if (!GEP1VariableIndices.empty()) {
+ uint64_t Modulo = 0;
+ bool AllPositive = true;
+ for (unsigned i = 0, e = GEP1VariableIndices.size(); i != e; ++i) {
+
+ // Try to distinguish something like &A[i][1] against &A[42][0].
+ // Grab the least significant bit set in any of the scales. We
+ // don't need std::abs here (even if the scale's negative) as we'll
+ // be ^'ing Modulo with itself later.
+ Modulo |= (uint64_t) GEP1VariableIndices[i].Scale;
+
+ if (AllPositive) {
+ // If the Value could change between cycles, then any reasoning about
+ // the Value this cycle may not hold in the next cycle. We'll just
+ // give up if we can't determine conditions that hold for every cycle:
+ const Value *V = GEP1VariableIndices[i].V;
+
+ bool SignKnownZero, SignKnownOne;
+ ComputeSignBit(const_cast<Value *>(V), SignKnownZero, SignKnownOne, *DL,
+ 0, AC1, nullptr, DT);
+
+ // Zero-extension widens the variable, and so forces the sign
+ // bit to zero.
+ bool IsZExt = GEP1VariableIndices[i].Extension == EK_ZeroExt;
+ SignKnownZero |= IsZExt;
+ SignKnownOne &= !IsZExt;
+
+ // If the variable begins with a zero then we know it's
+ // positive, regardless of whether the value is signed or
+ // unsigned.
+ int64_t Scale = GEP1VariableIndices[i].Scale;
+ AllPositive =
+ (SignKnownZero && Scale >= 0) ||
+ (SignKnownOne && Scale < 0);
+ }
}
- // If we have getelementptr <ptr>, 0, 0, 0, 0, ... and V2 must aliases
- // the ptr, the end result is a must alias also.
- if (AllZerosFound)
- return MustAlias;
+ Modulo = Modulo ^ (Modulo & (Modulo - 1));
- if (ConstantFound) {
- if (V2Size <= 1 && V1Size <= 1) // Just pointer check?
+ // We can compute the difference between the two addresses
+ // mod Modulo. Check whether that difference guarantees that the
+ // two locations do not alias.
+ uint64_t ModOffset = (uint64_t)GEP1BaseOffset & (Modulo - 1);
+ if (V1Size != UnknownSize && V2Size != UnknownSize &&
+ ModOffset >= V2Size && V1Size <= Modulo - ModOffset)
return NoAlias;
- // Otherwise we have to check to see that the distance is more than
- // the size of the argument... build an index vector that is equal to
- // the arguments provided, except substitute 0's for any variable
- // indexes we find...
- if (TD &&
- cast<PointerType>(BasePtr->getType())->getElementType()->isSized()) {
- for (unsigned i = 0; i != GEPOperands.size(); ++i)
- if (!isa<ConstantInt>(GEPOperands[i]))
- GEPOperands[i] = Constant::getNullValue(GEPOperands[i]->getType());
- int64_t Offset = TD->getIndexedOffset(BasePtr->getType(),
- &GEPOperands[0],
- GEPOperands.size());
-
- if (Offset >= (int64_t)V2Size || Offset <= -(int64_t)V1Size)
- return NoAlias;
- }
+ // If we know all the variables are positive, then GEP1 >= GEP1BasePtr.
+ // If GEP1BasePtr > V2 (GEP1BaseOffset > 0) then we know the pointers
+ // don't alias if V2Size can fit in the gap between V2 and GEP1BasePtr.
+ if (AllPositive && GEP1BaseOffset > 0 && V2Size <= (uint64_t) GEP1BaseOffset)
+ return NoAlias;
}
- return MayAlias;
+ // Statically, we can see that the base objects are the same, but the
+ // pointers have dynamic offsets which we can't resolve. And none of our
+ // little tricks above worked.
+ //
+ // TODO: Returning PartialAlias instead of MayAlias is a mild hack; the
+ // practical effect of this is protecting TBAA in the case of dynamic
+ // indices into arrays of unions or malloc'd memory.
+ return PartialAlias;
}
-// aliasSelect - Provide a bunch of ad-hoc rules to disambiguate a Select instruction
-// against another.
+static AliasAnalysis::AliasResult
+MergeAliasResults(AliasAnalysis::AliasResult A, AliasAnalysis::AliasResult B) {
+ // If the results agree, take it.
+ if (A == B)
+ return A;
+ // A mix of PartialAlias and MustAlias is PartialAlias.
+ if ((A == AliasAnalysis::PartialAlias && B == AliasAnalysis::MustAlias) ||
+ (B == AliasAnalysis::PartialAlias && A == AliasAnalysis::MustAlias))
+ return AliasAnalysis::PartialAlias;
+ // Otherwise, we don't know anything.
+ return AliasAnalysis::MayAlias;
+}
+
+/// aliasSelect - Provide a bunch of ad-hoc rules to disambiguate a Select
+/// instruction against another.
AliasAnalysis::AliasResult
-BasicAliasAnalysis::aliasSelect(const SelectInst *SI, unsigned SISize,
- const Value *V2, unsigned V2Size) {
+BasicAliasAnalysis::aliasSelect(const SelectInst *SI, uint64_t SISize,
+ const AAMDNodes &SIAAInfo,
+ const Value *V2, uint64_t V2Size,
+ const AAMDNodes &V2AAInfo) {
// If the values are Selects with the same condition, we can do a more precise
// check: just check for aliases between the values on corresponding arms.
if (const SelectInst *SI2 = dyn_cast<SelectInst>(V2))
if (SI->getCondition() == SI2->getCondition()) {
AliasResult Alias =
- aliasCheck(SI->getTrueValue(), SISize,
- SI2->getTrueValue(), V2Size);
+ aliasCheck(SI->getTrueValue(), SISize, SIAAInfo,
+ SI2->getTrueValue(), V2Size, V2AAInfo);
if (Alias == MayAlias)
return MayAlias;
AliasResult ThisAlias =
- aliasCheck(SI->getFalseValue(), SISize,
- SI2->getFalseValue(), V2Size);
- if (ThisAlias != Alias)
- return MayAlias;
- return Alias;
+ aliasCheck(SI->getFalseValue(), SISize, SIAAInfo,
+ SI2->getFalseValue(), V2Size, V2AAInfo);
+ return MergeAliasResults(ThisAlias, Alias);
}
// If both arms of the Select node NoAlias or MustAlias V2, then returns
// NoAlias / MustAlias. Otherwise, returns MayAlias.
AliasResult Alias =
- aliasCheck(SI->getTrueValue(), SISize, V2, V2Size);
+ aliasCheck(V2, V2Size, V2AAInfo, SI->getTrueValue(), SISize, SIAAInfo);
if (Alias == MayAlias)
return MayAlias;
+
AliasResult ThisAlias =
- aliasCheck(SI->getFalseValue(), SISize, V2, V2Size);
- if (ThisAlias != Alias)
- return MayAlias;
- return Alias;
+ aliasCheck(V2, V2Size, V2AAInfo, SI->getFalseValue(), SISize, SIAAInfo);
+ return MergeAliasResults(ThisAlias, Alias);
}
// aliasPHI - Provide a bunch of ad-hoc rules to disambiguate a PHI instruction
// against another.
AliasAnalysis::AliasResult
-BasicAliasAnalysis::aliasPHI(const PHINode *PN, unsigned PNSize,
- const Value *V2, unsigned V2Size) {
- // The PHI node has already been visited, avoid recursion any further.
- if (!VisitedPHIs.insert(PN))
- return MayAlias;
+BasicAliasAnalysis::aliasPHI(const PHINode *PN, uint64_t PNSize,
+ const AAMDNodes &PNAAInfo,
+ const Value *V2, uint64_t V2Size,
+ const AAMDNodes &V2AAInfo) {
+ // Track phi nodes we have visited. We use this information when we determine
+ // value equivalence.
+ VisitedPhiBBs.insert(PN->getParent());
// If the values are PHIs in the same block, we can do a more precise
// as well as efficient check: just check for aliases between the values
// on corresponding edges.
if (const PHINode *PN2 = dyn_cast<PHINode>(V2))
if (PN2->getParent() == PN->getParent()) {
- AliasResult Alias =
- aliasCheck(PN->getIncomingValue(0), PNSize,
- PN2->getIncomingValueForBlock(PN->getIncomingBlock(0)),
- V2Size);
- if (Alias == MayAlias)
- return MayAlias;
- for (unsigned i = 1, e = PN->getNumIncomingValues(); i != e; ++i) {
+ LocPair Locs(Location(PN, PNSize, PNAAInfo),
+ Location(V2, V2Size, V2AAInfo));
+ if (PN > V2)
+ std::swap(Locs.first, Locs.second);
+ // Analyse the PHIs' inputs under the assumption that the PHIs are
+ // NoAlias.
+ // If the PHIs are May/MustAlias there must be (recursively) an input
+ // operand from outside the PHIs' cycle that is MayAlias/MustAlias or
+ // there must be an operation on the PHIs within the PHIs' value cycle
+ // that causes a MayAlias.
+ // Pretend the phis do not alias.
+ AliasResult Alias = NoAlias;
+ assert(AliasCache.count(Locs) &&
+ "There must exist an entry for the phi node");
+ AliasResult OrigAliasResult = AliasCache[Locs];
+ AliasCache[Locs] = NoAlias;
+
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
AliasResult ThisAlias =
- aliasCheck(PN->getIncomingValue(i), PNSize,
+ aliasCheck(PN->getIncomingValue(i), PNSize, PNAAInfo,
PN2->getIncomingValueForBlock(PN->getIncomingBlock(i)),
- V2Size);
- if (ThisAlias != Alias)
- return MayAlias;
+ V2Size, V2AAInfo);
+ Alias = MergeAliasResults(ThisAlias, Alias);
+ if (Alias == MayAlias)
+ break;
}
+
+ // Reset if speculation failed.
+ if (Alias != NoAlias)
+ AliasCache[Locs] = OrigAliasResult;
+
return Alias;
}
// sides are PHI nodes. In which case, this is O(m x n) time where 'm'
// and 'n' are the number of PHI sources.
return MayAlias;
- if (UniqueSrc.insert(PV1))
+ if (UniqueSrc.insert(PV1).second)
V1Srcs.push_back(PV1);
}
- AliasResult Alias = aliasCheck(V2, V2Size, V1Srcs[0], PNSize);
+ AliasResult Alias = aliasCheck(V2, V2Size, V2AAInfo,
+ V1Srcs[0], PNSize, PNAAInfo);
// Early exit if the check of the first PHI source against V2 is MayAlias.
// Other results are not possible.
if (Alias == MayAlias)
for (unsigned i = 1, e = V1Srcs.size(); i != e; ++i) {
Value *V = V1Srcs[i];
- // If V2 is a PHI, the recursive case will have been caught in the
- // above aliasCheck call, so these subsequent calls to aliasCheck
- // don't need to assume that V2 is being visited recursively.
- VisitedPHIs.erase(V2);
-
- AliasResult ThisAlias = aliasCheck(V2, V2Size, V, PNSize);
- if (ThisAlias != Alias || ThisAlias == MayAlias)
- return MayAlias;
+ AliasResult ThisAlias = aliasCheck(V2, V2Size, V2AAInfo,
+ V, PNSize, PNAAInfo);
+ Alias = MergeAliasResults(ThisAlias, Alias);
+ if (Alias == MayAlias)
+ break;
}
return Alias;
// such as array references.
//
AliasAnalysis::AliasResult
-BasicAliasAnalysis::aliasCheck(const Value *V1, unsigned V1Size,
- const Value *V2, unsigned V2Size) {
+BasicAliasAnalysis::aliasCheck(const Value *V1, uint64_t V1Size,
+ AAMDNodes V1AAInfo,
+ const Value *V2, uint64_t V2Size,
+ AAMDNodes V2AAInfo) {
+ // If either of the memory references is empty, it doesn't matter what the
+ // pointer values are.
+ if (V1Size == 0 || V2Size == 0)
+ return NoAlias;
+
// Strip off any casts if they exist.
V1 = V1->stripPointerCasts();
V2 = V2->stripPointerCasts();
// Are we checking for alias of the same value?
- if (V1 == V2) return MustAlias;
+ // Because we look 'through' phi nodes we could look at "Value" pointers from
+ // different iterations. We must therefore make sure that this is not the
+ // case. The function isValueEqualInPotentialCycles ensures that this cannot
+ // happen by looking at the visited phi nodes and making sure they cannot
+ // reach the value.
+ if (isValueEqualInPotentialCycles(V1, V2))
+ return MustAlias;
- if (!isa<PointerType>(V1->getType()) || !isa<PointerType>(V2->getType()))
+ if (!V1->getType()->isPointerTy() || !V2->getType()->isPointerTy())
return NoAlias; // Scalars cannot alias each other
// Figure out what objects these things are pointing to if we can.
- const Value *O1 = V1->getUnderlyingObject();
- const Value *O2 = V2->getUnderlyingObject();
+ const Value *O1 = GetUnderlyingObject(V1, *DL, MaxLookupSearchDepth);
+ const Value *O2 = GetUnderlyingObject(V2, *DL, MaxLookupSearchDepth);
// Null values in the default address space don't point to any object, so they
// don't alias any other pointer.
// If V1/V2 point to two different objects we know that we have no alias.
if (isIdentifiedObject(O1) && isIdentifiedObject(O2))
return NoAlias;
-
- // Arguments can't alias with local allocations or noalias calls.
- if ((isa<Argument>(O1) && (isa<AllocaInst>(O2) || isNoAliasCall(O2))) ||
- (isa<Argument>(O2) && (isa<AllocaInst>(O1) || isNoAliasCall(O1))))
+
+ // Constant pointers can't alias with non-const isIdentifiedObject objects.
+ if ((isa<Constant>(O1) && isIdentifiedObject(O2) && !isa<Constant>(O2)) ||
+ (isa<Constant>(O2) && isIdentifiedObject(O1) && !isa<Constant>(O1)))
+ return NoAlias;
+
+ // Function arguments can't alias with things that are known to be
+ // unambigously identified at the function level.
+ if ((isa<Argument>(O1) && isIdentifiedFunctionLocal(O2)) ||
+ (isa<Argument>(O2) && isIdentifiedFunctionLocal(O1)))
return NoAlias;
// Most objects can't alias null.
- if ((isa<ConstantPointerNull>(V2) && isKnownNonNull(O1)) ||
- (isa<ConstantPointerNull>(V1) && isKnownNonNull(O2)))
+ if ((isa<ConstantPointerNull>(O2) && isKnownNonNull(O1)) ||
+ (isa<ConstantPointerNull>(O1) && isKnownNonNull(O2)))
+ return NoAlias;
+
+ // If one pointer is the result of a call/invoke or load and the other is a
+ // non-escaping local object within the same function, then we know the
+ // object couldn't escape to a point where the call could return it.
+ //
+ // Note that if the pointers are in different functions, there are a
+ // variety of complications. A call with a nocapture argument may still
+ // temporary store the nocapture argument's value in a temporary memory
+ // location if that memory location doesn't escape. Or it may pass a
+ // nocapture value to other functions as long as they don't capture it.
+ if (isEscapeSource(O1) && isNonEscapingLocalObject(O2))
+ return NoAlias;
+ if (isEscapeSource(O2) && isNonEscapingLocalObject(O1))
return NoAlias;
}
-
+
// If the size of one access is larger than the entire object on the other
// side, then we know such behavior is undefined and can assume no alias.
- if (TD)
- if ((V1Size != ~0U && isObjectSmallerThan(O2, V1Size, *TD)) ||
- (V2Size != ~0U && isObjectSmallerThan(O1, V2Size, *TD)))
+ if (DL)
+ if ((V1Size != UnknownSize && isObjectSmallerThan(O2, V1Size, *DL, *TLI)) ||
+ (V2Size != UnknownSize && isObjectSmallerThan(O1, V2Size, *DL, *TLI)))
return NoAlias;
-
- // If one pointer is the result of a call/invoke and the other is a
- // non-escaping local object, then we know the object couldn't escape to a
- // point where the call could return it.
- if ((isa<CallInst>(O1) || isa<InvokeInst>(O1)) &&
- isNonEscapingLocalObject(O2) && O1 != O2)
- return NoAlias;
- if ((isa<CallInst>(O2) || isa<InvokeInst>(O2)) &&
- isNonEscapingLocalObject(O1) && O1 != O2)
- return NoAlias;
+ // Check the cache before climbing up use-def chains. This also terminates
+ // otherwise infinitely recursive queries.
+ LocPair Locs(Location(V1, V1Size, V1AAInfo),
+ Location(V2, V2Size, V2AAInfo));
+ if (V1 > V2)
+ std::swap(Locs.first, Locs.second);
+ std::pair<AliasCacheTy::iterator, bool> Pair =
+ AliasCache.insert(std::make_pair(Locs, MayAlias));
+ if (!Pair.second)
+ return Pair.first->second;
+
+ // FIXME: This isn't aggressively handling alias(GEP, PHI) for example: if the
+ // GEP can't simplify, we don't even look at the PHI cases.
if (!isa<GEPOperator>(V1) && isa<GEPOperator>(V2)) {
std::swap(V1, V2);
std::swap(V1Size, V2Size);
+ std::swap(O1, O2);
+ std::swap(V1AAInfo, V2AAInfo);
+ }
+ if (const GEPOperator *GV1 = dyn_cast<GEPOperator>(V1)) {
+ AliasResult Result = aliasGEP(GV1, V1Size, V1AAInfo, V2, V2Size, V2AAInfo, O1, O2);
+ if (Result != MayAlias) return AliasCache[Locs] = Result;
}
- if (isa<GEPOperator>(V1))
- return aliasGEP(V1, V1Size, V2, V2Size);
if (isa<PHINode>(V2) && !isa<PHINode>(V1)) {
std::swap(V1, V2);
std::swap(V1Size, V2Size);
+ std::swap(V1AAInfo, V2AAInfo);
+ }
+ if (const PHINode *PN = dyn_cast<PHINode>(V1)) {
+ AliasResult Result = aliasPHI(PN, V1Size, V1AAInfo,
+ V2, V2Size, V2AAInfo);
+ if (Result != MayAlias) return AliasCache[Locs] = Result;
}
- if (const PHINode *PN = dyn_cast<PHINode>(V1))
- return aliasPHI(PN, V1Size, V2, V2Size);
if (isa<SelectInst>(V2) && !isa<SelectInst>(V1)) {
std::swap(V1, V2);
std::swap(V1Size, V2Size);
+ std::swap(V1AAInfo, V2AAInfo);
}
- if (const SelectInst *S1 = dyn_cast<SelectInst>(V1))
- return aliasSelect(S1, V1Size, V2, V2Size);
-
- return MayAlias;
-}
-
-// This function is used to determine if the indices of two GEP instructions are
-// equal. V1 and V2 are the indices.
-static bool IndexOperandsEqual(Value *V1, Value *V2) {
- if (V1->getType() == V2->getType())
- return V1 == V2;
- if (Constant *C1 = dyn_cast<Constant>(V1))
- if (Constant *C2 = dyn_cast<Constant>(V2)) {
- // Sign extend the constants to long types, if necessary
- if (C1->getType() != Type::getInt64Ty(C1->getContext()))
- C1 = ConstantExpr::getSExt(C1, Type::getInt64Ty(C1->getContext()));
- if (C2->getType() != Type::getInt64Ty(C1->getContext()))
- C2 = ConstantExpr::getSExt(C2, Type::getInt64Ty(C1->getContext()));
- return C1 == C2;
- }
- return false;
-}
-
-/// CheckGEPInstructions - Check two GEP instructions with known must-aliasing
-/// base pointers. This checks to see if the index expressions preclude the
-/// pointers from aliasing...
-AliasAnalysis::AliasResult
-BasicAliasAnalysis::CheckGEPInstructions(
- const Type* BasePtr1Ty, Value **GEP1Ops, unsigned NumGEP1Ops, unsigned G1S,
- const Type *BasePtr2Ty, Value **GEP2Ops, unsigned NumGEP2Ops, unsigned G2S) {
- // We currently can't handle the case when the base pointers have different
- // primitive types. Since this is uncommon anyway, we are happy being
- // extremely conservative.
- if (BasePtr1Ty != BasePtr2Ty)
- return MayAlias;
-
- const PointerType *GEPPointerTy = cast<PointerType>(BasePtr1Ty);
-
- // Find the (possibly empty) initial sequence of equal values... which are not
- // necessarily constants.
- unsigned NumGEP1Operands = NumGEP1Ops, NumGEP2Operands = NumGEP2Ops;
- unsigned MinOperands = std::min(NumGEP1Operands, NumGEP2Operands);
- unsigned MaxOperands = std::max(NumGEP1Operands, NumGEP2Operands);
- unsigned UnequalOper = 0;
- while (UnequalOper != MinOperands &&
- IndexOperandsEqual(GEP1Ops[UnequalOper], GEP2Ops[UnequalOper])) {
- // Advance through the type as we go...
- ++UnequalOper;
- if (const CompositeType *CT = dyn_cast<CompositeType>(BasePtr1Ty))
- BasePtr1Ty = CT->getTypeAtIndex(GEP1Ops[UnequalOper-1]);
- else {
- // If all operands equal each other, then the derived pointers must
- // alias each other...
- BasePtr1Ty = 0;
- assert(UnequalOper == NumGEP1Operands && UnequalOper == NumGEP2Operands &&
- "Ran out of type nesting, but not out of operands?");
- return MustAlias;
- }
- }
-
- // If we have seen all constant operands, and run out of indexes on one of the
- // getelementptrs, check to see if the tail of the leftover one is all zeros.
- // If so, return mustalias.
- if (UnequalOper == MinOperands) {
- if (NumGEP1Ops < NumGEP2Ops) {
- std::swap(GEP1Ops, GEP2Ops);
- std::swap(NumGEP1Ops, NumGEP2Ops);
- }
-
- bool AllAreZeros = true;
- for (unsigned i = UnequalOper; i != MaxOperands; ++i)
- if (!isa<Constant>(GEP1Ops[i]) ||
- !cast<Constant>(GEP1Ops[i])->isNullValue()) {
- AllAreZeros = false;
- break;
- }
- if (AllAreZeros) return MustAlias;
- }
-
-
- // So now we know that the indexes derived from the base pointers,
- // which are known to alias, are different. We can still determine a
- // no-alias result if there are differing constant pairs in the index
- // chain. For example:
- // A[i][0] != A[j][1] iff (&A[0][1]-&A[0][0] >= std::max(G1S, G2S))
- //
- // We have to be careful here about array accesses. In particular, consider:
- // A[1][0] vs A[0][i]
- // In this case, we don't *know* that the array will be accessed in bounds:
- // the index could even be negative. Because of this, we have to
- // conservatively *give up* and return may alias. We disregard differing
- // array subscripts that are followed by a variable index without going
- // through a struct.
- //
- unsigned SizeMax = std::max(G1S, G2S);
- if (SizeMax == ~0U) return MayAlias; // Avoid frivolous work.
-
- // Scan for the first operand that is constant and unequal in the
- // two getelementptrs...
- unsigned FirstConstantOper = UnequalOper;
- for (; FirstConstantOper != MinOperands; ++FirstConstantOper) {
- const Value *G1Oper = GEP1Ops[FirstConstantOper];
- const Value *G2Oper = GEP2Ops[FirstConstantOper];
-
- if (G1Oper != G2Oper) // Found non-equal constant indexes...
- if (Constant *G1OC = dyn_cast<ConstantInt>(const_cast<Value*>(G1Oper)))
- if (Constant *G2OC = dyn_cast<ConstantInt>(const_cast<Value*>(G2Oper))){
- if (G1OC->getType() != G2OC->getType()) {
- // Sign extend both operands to long.
- const Type *Int64Ty = Type::getInt64Ty(G1OC->getContext());
- if (G1OC->getType() != Int64Ty)
- G1OC = ConstantExpr::getSExt(G1OC, Int64Ty);
- if (G2OC->getType() != Int64Ty)
- G2OC = ConstantExpr::getSExt(G2OC, Int64Ty);
- GEP1Ops[FirstConstantOper] = G1OC;
- GEP2Ops[FirstConstantOper] = G2OC;
- }
-
- if (G1OC != G2OC) {
- // Handle the "be careful" case above: if this is an array/vector
- // subscript, scan for a subsequent variable array index.
- if (const SequentialType *STy =
- dyn_cast<SequentialType>(BasePtr1Ty)) {
- const Type *NextTy = STy;
- bool isBadCase = false;
-
- for (unsigned Idx = FirstConstantOper;
- Idx != MinOperands && isa<SequentialType>(NextTy); ++Idx) {
- const Value *V1 = GEP1Ops[Idx], *V2 = GEP2Ops[Idx];
- if (!isa<Constant>(V1) || !isa<Constant>(V2)) {
- isBadCase = true;
- break;
- }
- // If the array is indexed beyond the bounds of the static type
- // at this level, it will also fall into the "be careful" case.
- // It would theoretically be possible to analyze these cases,
- // but for now just be conservatively correct.
- if (const ArrayType *ATy = dyn_cast<ArrayType>(STy))
- if (cast<ConstantInt>(G1OC)->getZExtValue() >=
- ATy->getNumElements() ||
- cast<ConstantInt>(G2OC)->getZExtValue() >=
- ATy->getNumElements()) {
- isBadCase = true;
- break;
- }
- if (const VectorType *VTy = dyn_cast<VectorType>(STy))
- if (cast<ConstantInt>(G1OC)->getZExtValue() >=
- VTy->getNumElements() ||
- cast<ConstantInt>(G2OC)->getZExtValue() >=
- VTy->getNumElements()) {
- isBadCase = true;
- break;
- }
- STy = cast<SequentialType>(NextTy);
- NextTy = cast<SequentialType>(NextTy)->getElementType();
- }
-
- if (isBadCase) G1OC = 0;
- }
-
- // Make sure they are comparable (ie, not constant expressions), and
- // make sure the GEP with the smaller leading constant is GEP1.
- if (G1OC) {
- Constant *Compare = ConstantExpr::getICmp(ICmpInst::ICMP_SGT,
- G1OC, G2OC);
- if (ConstantInt *CV = dyn_cast<ConstantInt>(Compare)) {
- if (CV->getZExtValue()) { // If they are comparable and G2 > G1
- std::swap(GEP1Ops, GEP2Ops); // Make GEP1 < GEP2
- std::swap(NumGEP1Ops, NumGEP2Ops);
- }
- break;
- }
- }
- }
- }
- BasePtr1Ty = cast<CompositeType>(BasePtr1Ty)->getTypeAtIndex(G1Oper);
+ if (const SelectInst *S1 = dyn_cast<SelectInst>(V1)) {
+ AliasResult Result = aliasSelect(S1, V1Size, V1AAInfo,
+ V2, V2Size, V2AAInfo);
+ if (Result != MayAlias) return AliasCache[Locs] = Result;
}
- // No shared constant operands, and we ran out of common operands. At this
- // point, the GEP instructions have run through all of their operands, and we
- // haven't found evidence that there are any deltas between the GEP's.
- // However, one GEP may have more operands than the other. If this is the
- // case, there may still be hope. Check this now.
- if (FirstConstantOper == MinOperands) {
- // Without TargetData, we won't know what the offsets are.
- if (!TD)
- return MayAlias;
-
- // Make GEP1Ops be the longer one if there is a longer one.
- if (NumGEP1Ops < NumGEP2Ops) {
- std::swap(GEP1Ops, GEP2Ops);
- std::swap(NumGEP1Ops, NumGEP2Ops);
- }
-
- // Is there anything to check?
- if (NumGEP1Ops > MinOperands) {
- for (unsigned i = FirstConstantOper; i != MaxOperands; ++i)
- if (isa<ConstantInt>(GEP1Ops[i]) &&
- !cast<ConstantInt>(GEP1Ops[i])->isZero()) {
- // Yup, there's a constant in the tail. Set all variables to
- // constants in the GEP instruction to make it suitable for
- // TargetData::getIndexedOffset.
- for (i = 0; i != MaxOperands; ++i)
- if (!isa<ConstantInt>(GEP1Ops[i]))
- GEP1Ops[i] = Constant::getNullValue(GEP1Ops[i]->getType());
- // Okay, now get the offset. This is the relative offset for the full
- // instruction.
- int64_t Offset1 = TD->getIndexedOffset(GEPPointerTy, GEP1Ops,
- NumGEP1Ops);
-
- // Now check without any constants at the end.
- int64_t Offset2 = TD->getIndexedOffset(GEPPointerTy, GEP1Ops,
- MinOperands);
-
- // Make sure we compare the absolute difference.
- if (Offset1 > Offset2)
- std::swap(Offset1, Offset2);
-
- // If the tail provided a bit enough offset, return noalias!
- if ((uint64_t)(Offset2-Offset1) >= SizeMax)
- return NoAlias;
- // Otherwise break - we don't look for another constant in the tail.
- break;
- }
- }
+ // If both pointers are pointing into the same object and one of them
+ // accesses is accessing the entire object, then the accesses must
+ // overlap in some way.
+ if (DL && O1 == O2)
+ if ((V1Size != UnknownSize && isObjectSize(O1, V1Size, *DL, *TLI)) ||
+ (V2Size != UnknownSize && isObjectSize(O2, V2Size, *DL, *TLI)))
+ return AliasCache[Locs] = PartialAlias;
+
+ AliasResult Result =
+ AliasAnalysis::alias(Location(V1, V1Size, V1AAInfo),
+ Location(V2, V2Size, V2AAInfo));
+ return AliasCache[Locs] = Result;
+}
- // Couldn't find anything useful.
- return MayAlias;
- }
+bool BasicAliasAnalysis::isValueEqualInPotentialCycles(const Value *V,
+ const Value *V2) {
+ if (V != V2)
+ return false;
- // If there are non-equal constants arguments, then we can figure
- // out a minimum known delta between the two index expressions... at
- // this point we know that the first constant index of GEP1 is less
- // than the first constant index of GEP2.
-
- // Advance BasePtr[12]Ty over this first differing constant operand.
- BasePtr2Ty = cast<CompositeType>(BasePtr1Ty)->
- getTypeAtIndex(GEP2Ops[FirstConstantOper]);
- BasePtr1Ty = cast<CompositeType>(BasePtr1Ty)->
- getTypeAtIndex(GEP1Ops[FirstConstantOper]);
-
- // We are going to be using TargetData::getIndexedOffset to determine the
- // offset that each of the GEP's is reaching. To do this, we have to convert
- // all variable references to constant references. To do this, we convert the
- // initial sequence of array subscripts into constant zeros to start with.
- const Type *ZeroIdxTy = GEPPointerTy;
- for (unsigned i = 0; i != FirstConstantOper; ++i) {
- if (!isa<StructType>(ZeroIdxTy))
- GEP1Ops[i] = GEP2Ops[i] =
- Constant::getNullValue(Type::getInt32Ty(ZeroIdxTy->getContext()));
-
- if (const CompositeType *CT = dyn_cast<CompositeType>(ZeroIdxTy))
- ZeroIdxTy = CT->getTypeAtIndex(GEP1Ops[i]);
- }
+ const Instruction *Inst = dyn_cast<Instruction>(V);
+ if (!Inst)
+ return true;
- // We know that GEP1Ops[FirstConstantOper] & GEP2Ops[FirstConstantOper] are ok
-
- // Loop over the rest of the operands...
- for (unsigned i = FirstConstantOper+1; i != MaxOperands; ++i) {
- const Value *Op1 = i < NumGEP1Ops ? GEP1Ops[i] : 0;
- const Value *Op2 = i < NumGEP2Ops ? GEP2Ops[i] : 0;
- // If they are equal, use a zero index...
- if (Op1 == Op2 && BasePtr1Ty == BasePtr2Ty) {
- if (!isa<ConstantInt>(Op1))
- GEP1Ops[i] = GEP2Ops[i] = Constant::getNullValue(Op1->getType());
- // Otherwise, just keep the constants we have.
- } else {
- if (Op1) {
- if (const ConstantInt *Op1C = dyn_cast<ConstantInt>(Op1)) {
- // If this is an array index, make sure the array element is in range.
- if (const ArrayType *AT = dyn_cast<ArrayType>(BasePtr1Ty)) {
- if (Op1C->getZExtValue() >= AT->getNumElements())
- return MayAlias; // Be conservative with out-of-range accesses
- } else if (const VectorType *VT = dyn_cast<VectorType>(BasePtr1Ty)) {
- if (Op1C->getZExtValue() >= VT->getNumElements())
- return MayAlias; // Be conservative with out-of-range accesses
- }
-
- } else {
- // GEP1 is known to produce a value less than GEP2. To be
- // conservatively correct, we must assume the largest possible
- // constant is used in this position. This cannot be the initial
- // index to the GEP instructions (because we know we have at least one
- // element before this one with the different constant arguments), so
- // we know that the current index must be into either a struct or
- // array. Because we know it's not constant, this cannot be a
- // structure index. Because of this, we can calculate the maximum
- // value possible.
- //
- if (const ArrayType *AT = dyn_cast<ArrayType>(BasePtr1Ty))
- GEP1Ops[i] =
- ConstantInt::get(Type::getInt64Ty(AT->getContext()),
- AT->getNumElements()-1);
- else if (const VectorType *VT = dyn_cast<VectorType>(BasePtr1Ty))
- GEP1Ops[i] =
- ConstantInt::get(Type::getInt64Ty(VT->getContext()),
- VT->getNumElements()-1);
- }
- }
+ if (VisitedPhiBBs.size() > MaxNumPhiBBsValueReachabilityCheck)
+ return false;
- if (Op2) {
- if (const ConstantInt *Op2C = dyn_cast<ConstantInt>(Op2)) {
- // If this is an array index, make sure the array element is in range.
- if (const ArrayType *AT = dyn_cast<ArrayType>(BasePtr2Ty)) {
- if (Op2C->getZExtValue() >= AT->getNumElements())
- return MayAlias; // Be conservative with out-of-range accesses
- } else if (const VectorType *VT = dyn_cast<VectorType>(BasePtr2Ty)) {
- if (Op2C->getZExtValue() >= VT->getNumElements())
- return MayAlias; // Be conservative with out-of-range accesses
- }
- } else { // Conservatively assume the minimum value for this index
- GEP2Ops[i] = Constant::getNullValue(Op2->getType());
- }
- }
- }
+ // Use dominance or loop info if available.
+ DominatorTreeWrapperPass *DTWP =
+ getAnalysisIfAvailable<DominatorTreeWrapperPass>();
+ DominatorTree *DT = DTWP ? &DTWP->getDomTree() : nullptr;
+ auto *LIWP = getAnalysisIfAvailable<LoopInfoWrapperPass>();
+ LoopInfo *LI = LIWP ? &LIWP->getLoopInfo() : nullptr;
+
+ // Make sure that the visited phis cannot reach the Value. This ensures that
+ // the Values cannot come from different iterations of a potential cycle the
+ // phi nodes could be involved in.
+ for (auto *P : VisitedPhiBBs)
+ if (isPotentiallyReachable(P->begin(), Inst, DT, LI))
+ return false;
- if (BasePtr1Ty && Op1) {
- if (const CompositeType *CT = dyn_cast<CompositeType>(BasePtr1Ty))
- BasePtr1Ty = CT->getTypeAtIndex(GEP1Ops[i]);
- else
- BasePtr1Ty = 0;
- }
+ return true;
+}
- if (BasePtr2Ty && Op2) {
- if (const CompositeType *CT = dyn_cast<CompositeType>(BasePtr2Ty))
- BasePtr2Ty = CT->getTypeAtIndex(GEP2Ops[i]);
+/// GetIndexDifference - Dest and Src are the variable indices from two
+/// decomposed GetElementPtr instructions GEP1 and GEP2 which have common base
+/// pointers. Subtract the GEP2 indices from GEP1 to find the symbolic
+/// difference between the two pointers.
+void BasicAliasAnalysis::GetIndexDifference(
+ SmallVectorImpl<VariableGEPIndex> &Dest,
+ const SmallVectorImpl<VariableGEPIndex> &Src) {
+ if (Src.empty())
+ return;
+
+ for (unsigned i = 0, e = Src.size(); i != e; ++i) {
+ const Value *V = Src[i].V;
+ ExtensionKind Extension = Src[i].Extension;
+ int64_t Scale = Src[i].Scale;
+
+ // Find V in Dest. This is N^2, but pointer indices almost never have more
+ // than a few variable indexes.
+ for (unsigned j = 0, e = Dest.size(); j != e; ++j) {
+ if (!isValueEqualInPotentialCycles(Dest[j].V, V) ||
+ Dest[j].Extension != Extension)
+ continue;
+
+ // If we found it, subtract off Scale V's from the entry in Dest. If it
+ // goes to zero, remove the entry.
+ if (Dest[j].Scale != Scale)
+ Dest[j].Scale -= Scale;
else
- BasePtr2Ty = 0;
+ Dest.erase(Dest.begin() + j);
+ Scale = 0;
+ break;
}
- }
- if (TD && GEPPointerTy->getElementType()->isSized()) {
- int64_t Offset1 =
- TD->getIndexedOffset(GEPPointerTy, GEP1Ops, NumGEP1Ops);
- int64_t Offset2 =
- TD->getIndexedOffset(GEPPointerTy, GEP2Ops, NumGEP2Ops);
- assert(Offset1 != Offset2 &&
- "There is at least one different constant here!");
-
- // Make sure we compare the absolute difference.
- if (Offset1 > Offset2)
- std::swap(Offset1, Offset2);
-
- if ((uint64_t)(Offset2-Offset1) >= SizeMax) {
- //cerr << "Determined that these two GEP's don't alias ["
- // << SizeMax << " bytes]: \n" << *GEP1 << *GEP2;
- return NoAlias;
+ // If we didn't consume this entry, add it to the end of the Dest list.
+ if (Scale) {
+ VariableGEPIndex Entry = { V, Extension, -Scale };
+ Dest.push_back(Entry);
}
}
- return MayAlias;
}
-
-// Make sure that anything that uses AliasAnalysis pulls in this file...
-DEFINING_FILE_FOR(BasicAliasAnalysis)
projects / oota-llvm.git / blobdiff