IR: Change the gep_type_iterator API to avoid always exposing the "cu…

…rrent" type.

Instead, expose whether the current type is an array or a struct, if an array
what the upper bound is, and if a struct the struct type itself. This is
in preparation for a later change which will make PointerType derive from
Type rather than SequentialType.

Differential Revision: https://reviews.llvm.org/D26594

llvm-svn: 288458

llvm/include/llvm/Analysis/TargetTransformInfoImpl.h

@@ -482,18 +482,20 @@ class TargetTransformInfoImplCRTPBase : public TargetTransformInfoImplBase {
     int64_t BaseOffset = 0;
     int64_t Scale = 0;
 
-    // Assumes the address space is 0 when Ptr is nullptr.
-    unsigned AS =
-        (Ptr == nullptr ? 0 : Ptr->getType()->getPointerAddressSpace());
-    auto GTI = gep_type_begin(PointeeType, AS, Operands);
+    auto GTI = gep_type_begin(PointeeType, Operands);
     for (auto I = Operands.begin(); I != Operands.end(); ++I, ++GTI) {
       // We assume that the cost of Scalar GEP with constant index and the
       // cost of Vector GEP with splat constant index are the same.
       const ConstantInt *ConstIdx = dyn_cast<ConstantInt>(*I);
       if (!ConstIdx)
         if (auto Splat = getSplatValue(*I))
           ConstIdx = dyn_cast<ConstantInt>(Splat);
-      if (isa<SequentialType>(*GTI)) {
+      if (StructType *STy = GTI.getStructTypeOrNull()) {
+        // For structures the index is always splat or scalar constant
+        assert(ConstIdx && "Unexpected GEP index");
+        uint64_t Field = ConstIdx->getZExtValue();
+        BaseOffset += DL.getStructLayout(STy)->getElementOffset(Field);
+      } else {
         int64_t ElementSize = DL.getTypeAllocSize(GTI.getIndexedType());
         if (ConstIdx)
           BaseOffset += ConstIdx->getSExtValue() * ElementSize;
@@ -504,17 +506,15 @@ class TargetTransformInfoImplCRTPBase : public TargetTransformInfoImplBase {
             return TTI::TCC_Basic;
           Scale = ElementSize;
         }
-      } else {
-        StructType *STy = cast<StructType>(*GTI);
-        // For structures the index is always splat or scalar constant
-        assert(ConstIdx && "Unexpected GEP index");
-        uint64_t Field = ConstIdx->getZExtValue();
-        BaseOffset += DL.getStructLayout(STy)->getElementOffset(Field);
       }
     }
 
+    // Assumes the address space is 0 when Ptr is nullptr.
+    unsigned AS =
+        (Ptr == nullptr ? 0 : Ptr->getType()->getPointerAddressSpace());
     if (static_cast<T *>(this)->isLegalAddressingMode(
-            PointerType::get(*GTI, AS), const_cast<GlobalValue *>(BaseGV),
+            PointerType::get(Type::getInt8Ty(PointeeType->getContext()), AS),
+            const_cast<GlobalValue *>(BaseGV),
             BaseOffset, HasBaseReg, Scale, AS)) {
       return TTI::TCC_Free;
     }

llvm/include/llvm/IR/GetElementPtrTypeIterator.h

@@ -16,7 +16,7 @@
 #define LLVM_IR_GETELEMENTPTRTYPEITERATOR_H
 
 #include "llvm/ADT/ArrayRef.h"
-#include "llvm/ADT/PointerIntPair.h"
+#include "llvm/ADT/PointerUnion.h"
 #include "llvm/IR/DerivedTypes.h"
 #include "llvm/IR/Operator.h"
 #include "llvm/IR/User.h"
@@ -33,18 +33,15 @@ namespace llvm {
                           Type *, ptrdiff_t> super;
 
     ItTy OpIt;
-    PointerIntPair<Type *, 1> CurTy;
-    unsigned AddrSpace;
-
+    PointerUnion<StructType *, Type *> CurTy;
+    enum { Unbounded = -1ull };
+    uint64_t NumElements = Unbounded;
     generic_gep_type_iterator() = default;
 
   public:
-    static generic_gep_type_iterator begin(Type *Ty, unsigned AddrSpace,
-                                           ItTy It) {
+    static generic_gep_type_iterator begin(Type *Ty, ItTy It) {
       generic_gep_type_iterator I;
-      I.CurTy.setPointer(Ty);
-      I.CurTy.setInt(true);
-      I.AddrSpace = AddrSpace;
+      I.CurTy = Ty;
       I.OpIt = It;
       return I;
     }
@@ -63,41 +60,68 @@ namespace llvm {
       return !operator==(x);
     }
 
-    Type *operator*() const {
-      if (CurTy.getInt())
-        return CurTy.getPointer()->getPointerTo(AddrSpace);
-      return CurTy.getPointer();
-    }
-
+    // FIXME: Make this the iterator's operator*() after the 4.0 release.
+    // operator*() had a different meaning in earlier releases, so we're
+    // temporarily not giving this iterator an operator*() to avoid a subtle
+    // semantics break.
     Type *getIndexedType() const {
-      if (CurTy.getInt())
-        return CurTy.getPointer();
-      CompositeType *CT = cast<CompositeType>(CurTy.getPointer());
-      return CT->getTypeAtIndex(getOperand());
+      if (auto *T = CurTy.dyn_cast<Type *>())
+        return T;
+      return CurTy.get<StructType *>()->getTypeAtIndex(getOperand());
     }
 
-    // This is a non-standard operator->.  It allows you to call methods on the
-    // current type directly.
-    Type *operator->() const { return operator*(); }
-
     Value *getOperand() const { return const_cast<Value *>(&**OpIt); }
 
     generic_gep_type_iterator& operator++() {   // Preincrement
-      if (CurTy.getInt()) {
-        CurTy.setInt(false);
-      } else if (CompositeType *CT =
-                     dyn_cast<CompositeType>(CurTy.getPointer())) {
-        CurTy.setPointer(CT->getTypeAtIndex(getOperand()));
-      } else {
-        CurTy.setPointer(nullptr);
-      }
+      Type *Ty = getIndexedType();
+      if (auto *ATy = dyn_cast<ArrayType>(Ty)) {
+        CurTy = ATy->getElementType();
+        NumElements = ATy->getNumElements();
+      } else if (auto *VTy = dyn_cast<VectorType>(Ty)) {
+        CurTy = VTy->getElementType();
+        NumElements = VTy->getNumElements();
+      } else
+        CurTy = dyn_cast<StructType>(Ty);
       ++OpIt;
       return *this;
     }
 
     generic_gep_type_iterator operator++(int) { // Postincrement
       generic_gep_type_iterator tmp = *this; ++*this; return tmp;
     }
+
+    // All of the below API is for querying properties of the "outer type", i.e.
+    // the type that contains the indexed type. Most of the time this is just
+    // the type that was visited immediately prior to the indexed type, but for
+    // the first element this is an unbounded array of the GEP's source element
+    // type, for which there is no clearly corresponding IR type (we've
+    // historically used a pointer type as the outer type in this case, but
+    // pointers will soon lose their element type).
+    //
+    // FIXME: Most current users of this class are just interested in byte
+    // offsets (a few need to know whether the outer type is a struct because
+    // they are trying to replace a constant with a variable, which is only
+    // legal for arrays, e.g. canReplaceOperandWithVariable in SimplifyCFG.cpp);
+    // we should provide a more minimal API here that exposes not much more than
+    // that.
+
+    bool isStruct() const { return CurTy.is<StructType *>(); }
+    bool isSequential() const { return CurTy.is<Type *>(); }
+
+    StructType *getStructType() const { return CurTy.get<StructType *>(); }
+
+    StructType *getStructTypeOrNull() const {
+      return CurTy.dyn_cast<StructType *>();
+    }
+
+    bool isBoundedSequential() const {
+      return isSequential() && NumElements != Unbounded;
+    }
+
+    uint64_t getSequentialNumElements() const {
+      assert(isBoundedSequential());
+      return NumElements;
+    }
   };
 
   typedef generic_gep_type_iterator<> gep_type_iterator;
@@ -106,8 +130,6 @@ namespace llvm {
     auto *GEPOp = cast<GEPOperator>(GEP);
     return gep_type_iterator::begin(
         GEPOp->getSourceElementType(),
-        cast<PointerType>(GEPOp->getPointerOperandType()->getScalarType())
-            ->getAddressSpace(),
         GEP->op_begin() + 1);
   }
 
@@ -119,8 +141,6 @@ namespace llvm {
     auto &GEPOp = cast<GEPOperator>(GEP);
     return gep_type_iterator::begin(
         GEPOp.getSourceElementType(),
-        cast<PointerType>(GEPOp.getPointerOperandType()->getScalarType())
-            ->getAddressSpace(),
         GEP.op_begin() + 1);
   }
 
@@ -130,13 +150,13 @@ namespace llvm {
 
   template<typename T>
   inline generic_gep_type_iterator<const T *>
-  gep_type_begin(Type *Op0, unsigned AS, ArrayRef<T> A) {
-    return generic_gep_type_iterator<const T *>::begin(Op0, AS, A.begin());
+  gep_type_begin(Type *Op0, ArrayRef<T> A) {
+    return generic_gep_type_iterator<const T *>::begin(Op0, A.begin());
   }
 
   template<typename T>
   inline generic_gep_type_iterator<const T *>
-  gep_type_end(Type * /*Op0*/, unsigned /*AS*/, ArrayRef<T> A) {
+  gep_type_end(Type * /*Op0*/, ArrayRef<T> A) {
     return generic_gep_type_iterator<const T *>::end(A.end());
   }
 

llvm/include/llvm/Transforms/Utils/Local.h

@@ -217,7 +217,7 @@ Value *EmitGEPOffset(IRBuilderTy *Builder, const DataLayout &DL, User *GEP,
         continue;
 
       // Handle a struct index, which adds its field offset to the pointer.
-      if (StructType *STy = dyn_cast<StructType>(*GTI)) {
+      if (StructType *STy = GTI.getStructTypeOrNull()) {
         if (OpC->getType()->isVectorTy())
           OpC = OpC->getSplatValue();
 

llvm/lib/Analysis/BasicAliasAnalysis.cpp

@@ -412,10 +412,10 @@ bool BasicAAResult::DecomposeGEPExpression(const Value *V,
     // Assume all GEP operands are constants until proven otherwise.
     bool GepHasConstantOffset = true;
     for (User::const_op_iterator I = GEPOp->op_begin() + 1, E = GEPOp->op_end();
-         I != E; ++I) {
+         I != E; ++I, ++GTI) {
       const Value *Index = *I;
       // Compute the (potentially symbolic) offset in bytes for this index.
-      if (StructType *STy = dyn_cast<StructType>(*GTI++)) {
+      if (StructType *STy = GTI.getStructTypeOrNull()) {
         // For a struct, add the member offset.
         unsigned FieldNo = cast<ConstantInt>(Index)->getZExtValue();
         if (FieldNo == 0)
@@ -431,13 +431,13 @@ bool BasicAAResult::DecomposeGEPExpression(const Value *V,
         if (CIdx->isZero())
           continue;
         Decomposed.OtherOffset +=
-          DL.getTypeAllocSize(*GTI) * CIdx->getSExtValue();
+          DL.getTypeAllocSize(GTI.getIndexedType()) * CIdx->getSExtValue();
         continue;
       }
 
       GepHasConstantOffset = false;
 
-      uint64_t Scale = DL.getTypeAllocSize(*GTI);
+      uint64_t Scale = DL.getTypeAllocSize(GTI.getIndexedType());
       unsigned ZExtBits = 0, SExtBits = 0;
 
       // If the integer type is smaller than the pointer size, it is implicitly

llvm/lib/Analysis/InlineCost.cpp

@@ -318,7 +318,7 @@ bool CallAnalyzer::accumulateGEPOffset(GEPOperator &GEP, APInt &Offset) {
       continue;
 
     // Handle a struct index, which adds its field offset to the pointer.
-    if (StructType *STy = dyn_cast<StructType>(*GTI)) {
+    if (StructType *STy = GTI.getStructTypeOrNull()) {
       unsigned ElementIdx = OpC->getZExtValue();
       const StructLayout *SL = DL.getStructLayout(STy);
       Offset += APInt(IntPtrWidth, SL->getElementOffset(ElementIdx));

llvm/lib/Analysis/ValueTracking.cpp

@@ -1231,7 +1231,7 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
     gep_type_iterator GTI = gep_type_begin(I);
     for (unsigned i = 1, e = I->getNumOperands(); i != e; ++i, ++GTI) {
       Value *Index = I->getOperand(i);
-      if (StructType *STy = dyn_cast<StructType>(*GTI)) {
+      if (StructType *STy = GTI.getStructTypeOrNull()) {
         // Handle struct member offset arithmetic.
 
         // Handle case when index is vector zeroinitializer
@@ -1730,7 +1730,7 @@ static bool isGEPKnownNonNull(const GEPOperator *GEP, unsigned Depth,
   for (gep_type_iterator GTI = gep_type_begin(GEP), GTE = gep_type_end(GEP);
        GTI != GTE; ++GTI) {
     // Struct types are easy -- they must always be indexed by a constant.
-    if (StructType *STy = dyn_cast<StructType>(*GTI)) {
+    if (StructType *STy = GTI.getStructTypeOrNull()) {
       ConstantInt *OpC = cast<ConstantInt>(GTI.getOperand());
       unsigned ElementIdx = OpC->getZExtValue();
       const StructLayout *SL = Q.DL.getStructLayout(STy);

llvm/lib/Analysis/VectorUtils.cpp

@@ -107,11 +107,11 @@ unsigned llvm::getGEPInductionOperand(const GetElementPtrInst *Gep) {
   while (LastOperand > 1 && match(Gep->getOperand(LastOperand), m_Zero())) {
     // Find the type we're currently indexing into.
     gep_type_iterator GEPTI = gep_type_begin(Gep);
-    std::advance(GEPTI, LastOperand - 1);
+    std::advance(GEPTI, LastOperand - 2);
 
     // If it's a type with the same allocation size as the result of the GEP we
     // can peel off the zero index.
-    if (DL.getTypeAllocSize(*GEPTI) != GEPAllocSize)
+    if (DL.getTypeAllocSize(GEPTI.getIndexedType()) != GEPAllocSize)
       break;
     --LastOperand;
   }

llvm/lib/CodeGen/CodeGenPrepare.cpp

@@ -3261,7 +3261,7 @@ bool AddressingModeMatcher::matchOperationAddr(User *AddrInst, unsigned Opcode,
     int64_t ConstantOffset = 0;
     gep_type_iterator GTI = gep_type_begin(AddrInst);
     for (unsigned i = 1, e = AddrInst->getNumOperands(); i != e; ++i, ++GTI) {
-      if (StructType *STy = dyn_cast<StructType>(*GTI)) {
+      if (StructType *STy = GTI.getStructTypeOrNull()) {
         const StructLayout *SL = DL.getStructLayout(STy);
         unsigned Idx =
           cast<ConstantInt>(AddrInst->getOperand(i))->getZExtValue();

llvm/lib/CodeGen/SelectionDAG/FastISel.cpp

@@ -488,7 +488,7 @@ bool FastISel::selectGetElementPtr(const User *I) {
   for (gep_type_iterator GTI = gep_type_begin(I), E = gep_type_end(I);
        GTI != E; ++GTI) {
     const Value *Idx = GTI.getOperand();
-    if (auto *StTy = dyn_cast<StructType>(*GTI)) {
+    if (StructType *StTy = GTI.getStructTypeOrNull()) {
       uint64_t Field = cast<ConstantInt>(Idx)->getZExtValue();
       if (Field) {
         // N = N + Offset

llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

@@ -3274,7 +3274,7 @@ void SelectionDAGBuilder::visitGetElementPtr(const User &I) {
   for (gep_type_iterator GTI = gep_type_begin(&I), E = gep_type_end(&I);
        GTI != E; ++GTI) {
     const Value *Idx = GTI.getOperand();
-    if (StructType *StTy = dyn_cast<StructType>(*GTI)) {
+    if (StructType *StTy = GTI.getStructTypeOrNull()) {
       unsigned Field = cast<Constant>(Idx)->getUniqueInteger().getZExtValue();
       if (Field) {
         // N = N + Offset

llvm/lib/ExecutionEngine/Interpreter/Execution.cpp

@@ -999,15 +999,14 @@ GenericValue Interpreter::executeGEPOperation(Value *Ptr, gep_type_iterator I,
   uint64_t Total = 0;
 
   for (; I != E; ++I) {
-    if (StructType *STy = dyn_cast<StructType>(*I)) {
+    if (StructType *STy = I.getStructTypeOrNull()) {
       const StructLayout *SLO = getDataLayout().getStructLayout(STy);
 
       const ConstantInt *CPU = cast<ConstantInt>(I.getOperand());
       unsigned Index = unsigned(CPU->getZExtValue());
 
       Total += SLO->getElementOffset(Index);
     } else {
-      SequentialType *ST = cast<SequentialType>(*I);
       // Get the index number for the array... which must be long type...
       GenericValue IdxGV = getOperandValue(I.getOperand(), SF);
 
@@ -1020,7 +1019,7 @@ GenericValue Interpreter::executeGEPOperation(Value *Ptr, gep_type_iterator I,
         assert(BitWidth == 64 && "Invalid index type for getelementptr");
         Idx = (int64_t)IdxGV.IntVal.getZExtValue();
       }
-      Total += getDataLayout().getTypeAllocSize(ST->getElementType()) * Idx;
+      Total += getDataLayout().getTypeAllocSize(I.getIndexedType()) * Idx;
     }
   }