//===----------------------------------------------------------------------===//
#include "InstCombine.h"
-#include "llvm/IntrinsicInst.h"
+#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/Loads.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/IntrinsicInst.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/Local.h"
-#include "llvm/ADT/Statistic.h"
using namespace llvm;
STATISTIC(NumDeadStore, "Number of dead stores eliminated");
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(U)) {
// If the GEP has all zero indices, it doesn't offset the pointer. If it
// doesn't, it does.
- if (!isOnlyCopiedFromConstantGlobal(GEP, TheCopy, ToDelete,
- IsOffset || !GEP->hasAllZeroIndices()))
+ if (!isOnlyCopiedFromConstantGlobal(
+ GEP, TheCopy, ToDelete, IsOffset || !GEP->hasAllZeroIndices()))
return false;
continue;
}
return 0;
}
-/// getPointeeAlignment - Compute the minimum alignment of the value pointed
-/// to by the given pointer.
-static unsigned getPointeeAlignment(Value *V, const TargetData &TD) {
- if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
- if (CE->getOpcode() == Instruction::BitCast ||
- (CE->getOpcode() == Instruction::GetElementPtr &&
- cast<GEPOperator>(CE)->hasAllZeroIndices()))
- return getPointeeAlignment(CE->getOperand(0), TD);
-
- if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
- if (!GV->isDeclaration())
- return TD.getPreferredAlignment(GV);
-
- if (PointerType *PT = dyn_cast<PointerType>(V->getType()))
- return TD.getABITypeAlignment(PT->getElementType());
-
- return 0;
-}
-
Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
// Ensure that the alloca array size argument has type intptr_t, so that
// any casting is exposed early.
// Convert: alloca Ty, C - where C is a constant != 1 into: alloca [C x Ty], 1
if (AI.isArrayAllocation()) { // Check C != 1
if (const ConstantInt *C = dyn_cast<ConstantInt>(AI.getArraySize())) {
- Type *NewTy =
+ Type *NewTy =
ArrayType::get(AI.getAllocatedType(), C->getZExtValue());
AllocaInst *New = Builder->CreateAlloca(NewTy, 0, AI.getName());
New->setAlignment(AI.getAlignment());
}
}
- if (TD) {
+ if (AI.getAlignment()) {
// Check to see if this allocation is only modified by a memcpy/memmove from
// a constant global whose alignment is equal to or exceeds that of the
// allocation. If this is the case, we can change all users to use
// is only subsequently read.
SmallVector<Instruction *, 4> ToDelete;
if (MemTransferInst *Copy = isOnlyCopiedFromConstantGlobal(&AI, ToDelete)) {
- if (AI.getAlignment() <= getPointeeAlignment(Copy->getSource(), *TD)) {
+ unsigned SourceAlign = getOrEnforceKnownAlignment(Copy->getSource(),
+ AI.getAlignment(), TD);
+ if (AI.getAlignment() <= SourceAlign) {
DEBUG(dbgs() << "Found alloca equal to global: " << AI << '\n');
DEBUG(dbgs() << " memcpy = " << *Copy << '\n');
for (unsigned i = 0, e = ToDelete.size(); i != e; ++i)
/// InstCombineLoadCast - Fold 'load (cast P)' -> cast (load P)' when possible.
static Instruction *InstCombineLoadCast(InstCombiner &IC, LoadInst &LI,
- const TargetData *TD) {
+ const DataLayout *TD) {
User *CI = cast<User>(LI.getOperand(0));
Value *CastOp = CI->getOperand(0);
Type *SrcPTy = SrcTy->getElementType();
- if (DestPTy->isIntegerTy() || DestPTy->isPointerTy() ||
+ if (DestPTy->isIntegerTy() || DestPTy->isPointerTy() ||
DestPTy->isVectorTy()) {
// If the source is an array, the code below will not succeed. Check to
// see if a trivial 'gep P, 0, 0' will help matters. Only do this for
SrcPTy = SrcTy->getElementType();
}
- if (IC.getTargetData() &&
- (SrcPTy->isIntegerTy() || SrcPTy->isPointerTy() ||
+ if (IC.getDataLayout() &&
+ (SrcPTy->isIntegerTy() || SrcPTy->isPointerTy() ||
SrcPTy->isVectorTy()) &&
// Do not allow turning this into a load of an integer, which is then
// casted to a pointer, this pessimizes pointer analysis a lot.
(SrcPTy->isPointerTy() == LI.getType()->isPointerTy()) &&
- IC.getTargetData()->getTypeSizeInBits(SrcPTy) ==
- IC.getTargetData()->getTypeSizeInBits(DestPTy)) {
+ IC.getDataLayout()->getTypeSizeInBits(SrcPTy) ==
+ IC.getDataLayout()->getTypeSizeInBits(DestPTy)) {
// Okay, we are casting from one integer or pointer type to another of
// the same size. Instead of casting the pointer before the load, cast
// the result of the loaded value.
- LoadInst *NewLoad =
+ LoadInst *NewLoad =
IC.Builder->CreateLoad(CastOp, LI.isVolatile(), CI->getName());
NewLoad->setAlignment(LI.getAlignment());
NewLoad->setAtomic(LI.getOrdering(), LI.getSynchScope());
// None of the following transforms are legal for volatile/atomic loads.
// FIXME: Some of it is okay for atomic loads; needs refactoring.
if (!LI.isSimple()) return 0;
-
+
// Do really simple store-to-load forwarding and load CSE, to catch cases
// where there are several consecutive memory accesses to the same location,
// separated by a few arithmetic operations.
Constant::getNullValue(Op->getType()), &LI);
return ReplaceInstUsesWith(LI, UndefValue::get(LI.getType()));
}
- }
+ }
// load null/undef -> unreachable
// TODO: Consider a target hook for valid address spaces for this xform.
if (CE->isCast())
if (Instruction *Res = InstCombineLoadCast(*this, LI, TD))
return Res;
-
+
if (Op->hasOneUse()) {
// Change select and PHI nodes to select values instead of addresses: this
// helps alias analysis out a lot, allows many others simplifications, and
Type *DestPTy = cast<PointerType>(CI->getType())->getElementType();
PointerType *SrcTy = dyn_cast<PointerType>(CastOp->getType());
if (SrcTy == 0) return 0;
-
+
Type *SrcPTy = SrcTy->getElementType();
if (!DestPTy->isIntegerTy() && !DestPTy->isPointerTy())
return 0;
-
+
/// NewGEPIndices - If SrcPTy is an aggregate type, we can emit a "noop gep"
/// to its first element. This allows us to handle things like:
/// store i32 xxx, (bitcast {foo*, float}* %P to i32*)
/// on 32-bit hosts.
SmallVector<Value*, 4> NewGEPIndices;
-
+
// If the source is an array, the code below will not succeed. Check to
// see if a trivial 'gep P, 0, 0' will help matters. Only do this for
// constants.
// Index through pointer.
Constant *Zero = Constant::getNullValue(Type::getInt32Ty(SI.getContext()));
NewGEPIndices.push_back(Zero);
-
+
while (1) {
if (StructType *STy = dyn_cast<StructType>(SrcPTy)) {
if (!STy->getNumElements()) /* Struct can be empty {} */
break;
}
}
-
+
SrcTy = PointerType::get(SrcPTy, SrcTy->getAddressSpace());
}
if (!SrcPTy->isIntegerTy() && !SrcPTy->isPointerTy())
return 0;
-
+
// If the pointers point into different address spaces or if they point to
// values with different sizes, we can't do the transformation.
- if (!IC.getTargetData() ||
- SrcTy->getAddressSpace() !=
+ if (!IC.getDataLayout() ||
+ SrcTy->getAddressSpace() !=
cast<PointerType>(CI->getType())->getAddressSpace() ||
- IC.getTargetData()->getTypeSizeInBits(SrcPTy) !=
- IC.getTargetData()->getTypeSizeInBits(DestPTy))
+ IC.getDataLayout()->getTypeSizeInBits(SrcPTy) !=
+ IC.getDataLayout()->getTypeSizeInBits(DestPTy))
return 0;
// Okay, we are casting from one integer or pointer type to another of
- // the same size. Instead of casting the pointer before
+ // the same size. Instead of casting the pointer before
// the store, cast the value to be stored.
Value *NewCast;
Value *SIOp0 = SI.getOperand(0);
if (SIOp0->getType()->isPointerTy())
opcode = Instruction::PtrToInt;
}
-
+
// SIOp0 is a pointer to aggregate and this is a store to the first field,
// emit a GEP to index into its first field.
if (!NewGEPIndices.empty())
CastOp = IC.Builder->CreateInBoundsGEP(CastOp, NewGEPIndices);
-
+
NewCast = IC.Builder->CreateCast(opcode, SIOp0, CastDstTy,
SIOp0->getName()+".c");
SI.setOperand(0, NewCast);
static bool equivalentAddressValues(Value *A, Value *B) {
// Test if the values are trivially equivalent.
if (A == B) return true;
-
+
// Test if the values come form identical arithmetic instructions.
// This uses isIdenticalToWhenDefined instead of isIdenticalTo because
// its only used to compare two uses within the same basic block, which
if (Instruction *BI = dyn_cast<Instruction>(B))
if (cast<Instruction>(A)->isIdenticalToWhenDefined(BI))
return true;
-
+
// Otherwise they may not be equivalent.
return false;
}
// If the RHS is an alloca with a single use, zapify the store, making the
// alloca dead.
if (Ptr->hasOneUse()) {
- if (isa<AllocaInst>(Ptr))
+ if (isa<AllocaInst>(Ptr))
return EraseInstFromFunction(SI);
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Ptr)) {
if (isa<AllocaInst>(GEP->getOperand(0))) {
(isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy())) {
ScanInsts++;
continue;
- }
-
+ }
+
if (StoreInst *PrevSI = dyn_cast<StoreInst>(BBI)) {
// Prev store isn't volatile, and stores to the same location?
if (PrevSI->isSimple() && equivalentAddressValues(PrevSI->getOperand(1),
}
break;
}
-
+
// If this is a load, we have to stop. However, if the loaded value is from
// the pointer we're loading and is producing the pointer we're storing,
// then *this* store is dead (X = load P; store X -> P).
if (LI == Val && equivalentAddressValues(LI->getOperand(0), Ptr) &&
LI->isSimple())
return EraseInstFromFunction(SI);
-
+
// Otherwise, this is a load from some other location. Stores before it
// may not be dead.
break;
}
-
+
// Don't skip over loads or things that can modify memory.
if (BBI->mayWriteToMemory() || BBI->mayReadFromMemory())
break;
if (Instruction *Res = InstCombineStoreToCast(*this, SI))
return Res;
-
+
// If this store is the last instruction in the basic block (possibly
// excepting debug info instructions), and if the block ends with an
// unconditional branch, try to move it to the successor block.
- BBI = &SI;
+ BBI = &SI;
do {
++BBI;
} while (isa<DbgInfoIntrinsic>(BBI) ||
if (BI->isUnconditional())
if (SimplifyStoreAtEndOfBlock(SI))
return 0; // xform done!
-
+
return 0;
}
///
bool InstCombiner::SimplifyStoreAtEndOfBlock(StoreInst &SI) {
BasicBlock *StoreBB = SI.getParent();
-
+
// Check to see if the successor block has exactly two incoming edges. If
// so, see if the other predecessor contains a store to the same location.
// if so, insert a PHI node (if needed) and move the stores down.
BasicBlock *DestBB = StoreBB->getTerminator()->getSuccessor(0);
-
+
// Determine whether Dest has exactly two predecessors and, if so, compute
// the other predecessor.
pred_iterator PI = pred_begin(DestBB);
if (++PI == pred_end(DestBB))
return false;
-
+
P = *PI;
if (P != StoreBB) {
if (OtherBB)
BranchInst *OtherBr = dyn_cast<BranchInst>(BBI);
if (!OtherBr || BBI == OtherBB->begin())
return false;
-
+
// If the other block ends in an unconditional branch, check for the 'if then
// else' case. there is an instruction before the branch.
StoreInst *OtherStore = 0;
} else {
// Otherwise, the other block ended with a conditional branch. If one of the
// destinations is StoreBB, then we have the if/then case.
- if (OtherBr->getSuccessor(0) != StoreBB &&
+ if (OtherBr->getSuccessor(0) != StoreBB &&
OtherBr->getSuccessor(1) != StoreBB)
return false;
-
+
// Okay, we know that OtherBr now goes to Dest and StoreBB, so this is an
// if/then triangle. See if there is a store to the same ptr as SI that
// lives in OtherBB.
BBI == OtherBB->begin())
return false;
}
-
+
// In order to eliminate the store in OtherBr, we have to
// make sure nothing reads or overwrites the stored value in
// StoreBB.
return false;
}
}
-
+
// Insert a PHI node now if we need it.
Value *MergedVal = OtherStore->getOperand(0);
if (MergedVal != SI.getOperand(0)) {
PN->addIncoming(OtherStore->getOperand(0), OtherBB);
MergedVal = InsertNewInstBefore(PN, DestBB->front());
}
-
+
// Advance to a place where it is safe to insert the new store and
// insert it.
BBI = DestBB->getFirstInsertionPt();
SI.getOrdering(),
SI.getSynchScope());
InsertNewInstBefore(NewSI, *BBI);
- NewSI->setDebugLoc(OtherStore->getDebugLoc());
+ NewSI->setDebugLoc(OtherStore->getDebugLoc());
+
+ // If the two stores had the same TBAA tag, preserve it.
+ if (MDNode *TBAATag = SI.getMetadata(LLVMContext::MD_tbaa))
+ if ((TBAATag = MDNode::getMostGenericTBAA(TBAATag,
+ OtherStore->getMetadata(LLVMContext::MD_tbaa))))
+ NewSI->setMetadata(LLVMContext::MD_tbaa, TBAATag);
+
// Nuke the old stores.
EraseInstFromFunction(SI);