aboutsummaryrefslogtreecommitdiffstats
path: root/lib/Transforms/Scalar
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Transforms/Scalar')
-rw-r--r--lib/Transforms/Scalar/CodeGenPrepare.cpp8
-rw-r--r--lib/Transforms/Scalar/DeadStoreElimination.cpp2
-rw-r--r--lib/Transforms/Scalar/GVN.cpp32
-rw-r--r--lib/Transforms/Scalar/IndVarSimplify.cpp611
-rw-r--r--lib/Transforms/Scalar/LoopIdiomRecognize.cpp4
-rw-r--r--lib/Transforms/Scalar/LoopStrengthReduce.cpp74
-rw-r--r--lib/Transforms/Scalar/MemCpyOptimizer.cpp8
-rw-r--r--lib/Transforms/Scalar/ObjCARC.cpp42
-rw-r--r--lib/Transforms/Scalar/SCCP.cpp22
-rw-r--r--lib/Transforms/Scalar/ScalarReplAggregates.cpp192
-rw-r--r--lib/Transforms/Scalar/SimplifyLibCalls.cpp66
11 files changed, 693 insertions, 368 deletions
diff --git a/lib/Transforms/Scalar/CodeGenPrepare.cpp b/lib/Transforms/Scalar/CodeGenPrepare.cpp
index 0af14ed..17beeb5 100644
--- a/lib/Transforms/Scalar/CodeGenPrepare.cpp
+++ b/lib/Transforms/Scalar/CodeGenPrepare.cpp
@@ -104,7 +104,7 @@ namespace {
void EliminateMostlyEmptyBlock(BasicBlock *BB);
bool OptimizeBlock(BasicBlock &BB);
bool OptimizeInst(Instruction *I);
- bool OptimizeMemoryInst(Instruction *I, Value *Addr, const Type *AccessTy);
+ bool OptimizeMemoryInst(Instruction *I, Value *Addr, Type *AccessTy);
bool OptimizeInlineAsmInst(CallInst *CS);
bool OptimizeCallInst(CallInst *CI);
bool MoveExtToFormExtLoad(Instruction *I);
@@ -528,7 +528,7 @@ bool CodeGenPrepare::OptimizeCallInst(CallInst *CI) {
IntrinsicInst *II = dyn_cast<IntrinsicInst>(CI);
if (II && II->getIntrinsicID() == Intrinsic::objectsize) {
bool Min = (cast<ConstantInt>(II->getArgOperand(1))->getZExtValue() == 1);
- const Type *ReturnTy = CI->getType();
+ Type *ReturnTy = CI->getType();
Constant *RetVal = ConstantInt::get(ReturnTy, Min ? 0 : -1ULL);
// Substituting this can cause recursive simplifications, which can
@@ -724,7 +724,7 @@ static bool IsNonLocalValue(Value *V, BasicBlock *BB) {
/// This method is used to optimize both load/store and inline asms with memory
/// operands.
bool CodeGenPrepare::OptimizeMemoryInst(Instruction *MemoryInst, Value *Addr,
- const Type *AccessTy) {
+ Type *AccessTy) {
Value *Repl = Addr;
// Try to collapse single-value PHI nodes. This is necessary to undo
@@ -837,7 +837,7 @@ bool CodeGenPrepare::OptimizeMemoryInst(Instruction *MemoryInst, Value *Addr,
} else {
DEBUG(dbgs() << "CGP: SINKING nonlocal addrmode: " << AddrMode << " for "
<< *MemoryInst);
- const Type *IntPtrTy =
+ Type *IntPtrTy =
TLI->getTargetData()->getIntPtrType(AccessTy->getContext());
Value *Result = 0;
diff --git a/lib/Transforms/Scalar/DeadStoreElimination.cpp b/lib/Transforms/Scalar/DeadStoreElimination.cpp
index cb9b5be..e6089a9 100644
--- a/lib/Transforms/Scalar/DeadStoreElimination.cpp
+++ b/lib/Transforms/Scalar/DeadStoreElimination.cpp
@@ -264,7 +264,7 @@ static uint64_t getPointerSize(Value *V, AliasAnalysis &AA) {
}
assert(isa<Argument>(V) && "Expected AllocaInst or Argument!");
- const PointerType *PT = cast<PointerType>(V->getType());
+ PointerType *PT = cast<PointerType>(V->getType());
return TD->getTypeAllocSize(PT->getElementType());
}
diff --git a/lib/Transforms/Scalar/GVN.cpp b/lib/Transforms/Scalar/GVN.cpp
index 87b7317..b4d5667 100644
--- a/lib/Transforms/Scalar/GVN.cpp
+++ b/lib/Transforms/Scalar/GVN.cpp
@@ -63,7 +63,7 @@ static cl::opt<bool> EnableLoadPRE("enable-load-pre", cl::init(true));
namespace {
struct Expression {
uint32_t opcode;
- const Type *type;
+ Type *type;
SmallVector<uint32_t, 4> varargs;
Expression(uint32_t o = ~2U) : opcode(o) { }
@@ -655,7 +655,7 @@ SpeculationFailure:
/// CanCoerceMustAliasedValueToLoad - Return true if
/// CoerceAvailableValueToLoadType will succeed.
static bool CanCoerceMustAliasedValueToLoad(Value *StoredVal,
- const Type *LoadTy,
+ Type *LoadTy,
const TargetData &TD) {
// If the loaded or stored value is an first class array or struct, don't try
// to transform them. We need to be able to bitcast to integer.
@@ -680,14 +680,14 @@ static bool CanCoerceMustAliasedValueToLoad(Value *StoredVal,
///
/// If we can't do it, return null.
static Value *CoerceAvailableValueToLoadType(Value *StoredVal,
- const Type *LoadedTy,
+ Type *LoadedTy,
Instruction *InsertPt,
const TargetData &TD) {
if (!CanCoerceMustAliasedValueToLoad(StoredVal, LoadedTy, TD))
return 0;
// If this is already the right type, just return it.
- const Type *StoredValTy = StoredVal->getType();
+ Type *StoredValTy = StoredVal->getType();
uint64_t StoreSize = TD.getTypeStoreSizeInBits(StoredValTy);
uint64_t LoadSize = TD.getTypeStoreSizeInBits(LoadedTy);
@@ -704,7 +704,7 @@ static Value *CoerceAvailableValueToLoadType(Value *StoredVal,
StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt);
}
- const Type *TypeToCastTo = LoadedTy;
+ Type *TypeToCastTo = LoadedTy;
if (TypeToCastTo->isPointerTy())
TypeToCastTo = TD.getIntPtrType(StoredValTy->getContext());
@@ -743,7 +743,7 @@ static Value *CoerceAvailableValueToLoadType(Value *StoredVal,
}
// Truncate the integer to the right size now.
- const Type *NewIntTy = IntegerType::get(StoredValTy->getContext(), LoadSize);
+ Type *NewIntTy = IntegerType::get(StoredValTy->getContext(), LoadSize);
StoredVal = new TruncInst(StoredVal, NewIntTy, "trunc", InsertPt);
if (LoadedTy == NewIntTy)
@@ -765,7 +765,7 @@ static Value *CoerceAvailableValueToLoadType(Value *StoredVal,
/// Check this case to see if there is anything more we can do before we give
/// up. This returns -1 if we have to give up, or a byte number in the stored
/// value of the piece that feeds the load.
-static int AnalyzeLoadFromClobberingWrite(const Type *LoadTy, Value *LoadPtr,
+static int AnalyzeLoadFromClobberingWrite(Type *LoadTy, Value *LoadPtr,
Value *WritePtr,
uint64_t WriteSizeInBits,
const TargetData &TD) {
@@ -839,7 +839,7 @@ static int AnalyzeLoadFromClobberingWrite(const Type *LoadTy, Value *LoadPtr,
/// AnalyzeLoadFromClobberingStore - This function is called when we have a
/// memdep query of a load that ends up being a clobbering store.
-static int AnalyzeLoadFromClobberingStore(const Type *LoadTy, Value *LoadPtr,
+static int AnalyzeLoadFromClobberingStore(Type *LoadTy, Value *LoadPtr,
StoreInst *DepSI,
const TargetData &TD) {
// Cannot handle reading from store of first-class aggregate yet.
@@ -856,7 +856,7 @@ static int AnalyzeLoadFromClobberingStore(const Type *LoadTy, Value *LoadPtr,
/// AnalyzeLoadFromClobberingLoad - This function is called when we have a
/// memdep query of a load that ends up being clobbered by another load. See if
/// the other load can feed into the second load.
-static int AnalyzeLoadFromClobberingLoad(const Type *LoadTy, Value *LoadPtr,
+static int AnalyzeLoadFromClobberingLoad(Type *LoadTy, Value *LoadPtr,
LoadInst *DepLI, const TargetData &TD){
// Cannot handle reading from store of first-class aggregate yet.
if (DepLI->getType()->isStructTy() || DepLI->getType()->isArrayTy())
@@ -883,7 +883,7 @@ static int AnalyzeLoadFromClobberingLoad(const Type *LoadTy, Value *LoadPtr,
-static int AnalyzeLoadFromClobberingMemInst(const Type *LoadTy, Value *LoadPtr,
+static int AnalyzeLoadFromClobberingMemInst(Type *LoadTy, Value *LoadPtr,
MemIntrinsic *MI,
const TargetData &TD) {
// If the mem operation is a non-constant size, we can't handle it.
@@ -934,7 +934,7 @@ static int AnalyzeLoadFromClobberingMemInst(const Type *LoadTy, Value *LoadPtr,
/// mustalias. Check this case to see if there is anything more we can do
/// before we give up.
static Value *GetStoreValueForLoad(Value *SrcVal, unsigned Offset,
- const Type *LoadTy,
+ Type *LoadTy,
Instruction *InsertPt, const TargetData &TD){
LLVMContext &Ctx = SrcVal->getType()->getContext();
@@ -974,7 +974,7 @@ static Value *GetStoreValueForLoad(Value *SrcVal, unsigned Offset,
/// because the pointers don't mustalias. Check this case to see if there is
/// anything more we can do before we give up.
static Value *GetLoadValueForLoad(LoadInst *SrcVal, unsigned Offset,
- const Type *LoadTy, Instruction *InsertPt,
+ Type *LoadTy, Instruction *InsertPt,
GVN &gvn) {
const TargetData &TD = *gvn.getTargetData();
// If Offset+LoadTy exceeds the size of SrcVal, then we must be wanting to
@@ -996,7 +996,7 @@ static Value *GetLoadValueForLoad(LoadInst *SrcVal, unsigned Offset,
// memdep queries will find the new load. We can't easily remove the old
// load completely because it is already in the value numbering table.
IRBuilder<> Builder(SrcVal->getParent(), ++BasicBlock::iterator(SrcVal));
- const Type *DestPTy =
+ Type *DestPTy =
IntegerType::get(LoadTy->getContext(), NewLoadSize*8);
DestPTy = PointerType::get(DestPTy,
cast<PointerType>(PtrVal->getType())->getAddressSpace());
@@ -1034,7 +1034,7 @@ static Value *GetLoadValueForLoad(LoadInst *SrcVal, unsigned Offset,
/// GetMemInstValueForLoad - This function is called when we have a
/// memdep query of a load that ends up being a clobbering mem intrinsic.
static Value *GetMemInstValueForLoad(MemIntrinsic *SrcInst, unsigned Offset,
- const Type *LoadTy, Instruction *InsertPt,
+ Type *LoadTy, Instruction *InsertPt,
const TargetData &TD){
LLVMContext &Ctx = LoadTy->getContext();
uint64_t LoadSize = TD.getTypeSizeInBits(LoadTy)/8;
@@ -1154,7 +1154,7 @@ struct AvailableValueInBlock {
/// MaterializeAdjustedValue - Emit code into this block to adjust the value
/// defined here to the specified type. This handles various coercion cases.
- Value *MaterializeAdjustedValue(const Type *LoadTy, GVN &gvn) const {
+ Value *MaterializeAdjustedValue(Type *LoadTy, GVN &gvn) const {
Value *Res;
if (isSimpleValue()) {
Res = getSimpleValue();
@@ -1213,7 +1213,7 @@ static Value *ConstructSSAForLoadSet(LoadInst *LI,
SSAUpdater SSAUpdate(&NewPHIs);
SSAUpdate.Initialize(LI->getType(), LI->getName());
- const Type *LoadTy = LI->getType();
+ Type *LoadTy = LI->getType();
for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i) {
const AvailableValueInBlock &AV = ValuesPerBlock[i];
diff --git a/lib/Transforms/Scalar/IndVarSimplify.cpp b/lib/Transforms/Scalar/IndVarSimplify.cpp
index dee3d38..50140d9 100644
--- a/lib/Transforms/Scalar/IndVarSimplify.cpp
+++ b/lib/Transforms/Scalar/IndVarSimplify.cpp
@@ -79,6 +79,12 @@ static cl::opt<bool> DisableIVRewrite(
"disable-iv-rewrite", cl::Hidden,
cl::desc("Disable canonical induction variable rewriting"));
+// Temporary flag for use with -disable-iv-rewrite to force a canonical IV for
+// LFTR purposes.
+static cl::opt<bool> ForceLFTR(
+ "force-lftr", cl::Hidden,
+ cl::desc("Enable forced linear function test replacement"));
+
namespace {
class IndVarSimplify : public LoopPass {
IVUsers *IU;
@@ -140,9 +146,8 @@ namespace {
void RewriteIVExpressions(Loop *L, SCEVExpander &Rewriter);
- ICmpInst *LinearFunctionTestReplace(Loop *L, const SCEV *BackedgeTakenCount,
- PHINode *IndVar,
- SCEVExpander &Rewriter);
+ Value *LinearFunctionTestReplace(Loop *L, const SCEV *BackedgeTakenCount,
+ PHINode *IndVar, SCEVExpander &Rewriter);
void SinkUnusedInvariants(Loop *L);
};
@@ -211,6 +216,36 @@ bool IndVarSimplify::isValidRewrite(Value *FromVal, Value *ToVal) {
return true;
}
+/// Determine the insertion point for this user. By default, insert immediately
+/// before the user. SCEVExpander or LICM will hoist loop invariants out of the
+/// loop. For PHI nodes, there may be multiple uses, so compute the nearest
+/// common dominator for the incoming blocks.
+static Instruction *getInsertPointForUses(Instruction *User, Value *Def,
+ DominatorTree *DT) {
+ PHINode *PHI = dyn_cast<PHINode>(User);
+ if (!PHI)
+ return User;
+
+ Instruction *InsertPt = 0;
+ for (unsigned i = 0, e = PHI->getNumIncomingValues(); i != e; ++i) {
+ if (PHI->getIncomingValue(i) != Def)
+ continue;
+
+ BasicBlock *InsertBB = PHI->getIncomingBlock(i);
+ if (!InsertPt) {
+ InsertPt = InsertBB->getTerminator();
+ continue;
+ }
+ InsertBB = DT->findNearestCommonDominator(InsertPt->getParent(), InsertBB);
+ InsertPt = InsertBB->getTerminator();
+ }
+ assert(InsertPt && "Missing phi operand");
+ assert(!isa<Instruction>(Def) ||
+ DT->dominates(cast<Instruction>(Def), InsertPt) &&
+ "def does not dominate all uses");
+ return InsertPt;
+}
+
//===----------------------------------------------------------------------===//
// RewriteNonIntegerIVs and helpers. Prefer integer IVs.
//===----------------------------------------------------------------------===//
@@ -390,7 +425,7 @@ void IndVarSimplify::HandleFloatingPointIV(Loop *L, PHINode *PN) {
return;
}
- const IntegerType *Int32Ty = Type::getInt32Ty(PN->getContext());
+ IntegerType *Int32Ty = Type::getInt32Ty(PN->getContext());
// Insert new integer induction variable.
PHINode *NewPHI = PHINode::Create(Int32Ty, 2, PN->getName()+".int", PN);
@@ -665,7 +700,7 @@ void IndVarSimplify::RewriteIVExpressions(Loop *L, SCEVExpander &Rewriter) {
// of different sizes.
for (IVUsers::iterator UI = IU->begin(), E = IU->end(); UI != E; ++UI) {
Value *Op = UI->getOperandValToReplace();
- const Type *UseTy = Op->getType();
+ Type *UseTy = Op->getType();
Instruction *User = UI->getUser();
// Compute the final addrec to expand into code.
@@ -692,18 +727,7 @@ void IndVarSimplify::RewriteIVExpressions(Loop *L, SCEVExpander &Rewriter) {
// hoist loop invariants out of the loop. For PHI nodes, there may be
// multiple uses, so compute the nearest common dominator for the
// incoming blocks.
- Instruction *InsertPt = User;
- if (PHINode *PHI = dyn_cast<PHINode>(InsertPt))
- for (unsigned i = 0, e = PHI->getNumIncomingValues(); i != e; ++i)
- if (PHI->getIncomingValue(i) == Op) {
- if (InsertPt == User)
- InsertPt = PHI->getIncomingBlock(i)->getTerminator();
- else
- InsertPt =
- DT->findNearestCommonDominator(InsertPt->getParent(),
- PHI->getIncomingBlock(i))
- ->getTerminator();
- }
+ Instruction *InsertPt = getInsertPointForUses(User, Op, DT);
// Now expand it into actual Instructions and patch it into place.
Value *NewVal = Rewriter.expandCodeFor(AR, UseTy, InsertPt);
@@ -747,7 +771,7 @@ namespace {
// extend operations. This information is recorded by CollectExtend and
// provides the input to WidenIV.
struct WideIVInfo {
- const Type *WidestNativeType; // Widest integer type created [sz]ext
+ Type *WidestNativeType; // Widest integer type created [sz]ext
bool IsSigned; // Was an sext user seen before a zext?
WideIVInfo() : WidestNativeType(0), IsSigned(false) {}
@@ -759,7 +783,7 @@ namespace {
/// the final width of the IV before actually widening it.
static void CollectExtend(CastInst *Cast, bool IsSigned, WideIVInfo &WI,
ScalarEvolution *SE, const TargetData *TD) {
- const Type *Ty = Cast->getType();
+ Type *Ty = Cast->getType();
uint64_t Width = SE->getTypeSizeInBits(Ty);
if (TD && !TD->isLegalInteger(Width))
return;
@@ -779,6 +803,21 @@ static void CollectExtend(CastInst *Cast, bool IsSigned, WideIVInfo &WI,
}
namespace {
+
+/// NarrowIVDefUse - Record a link in the Narrow IV def-use chain along with the
+/// WideIV that computes the same value as the Narrow IV def. This avoids
+/// caching Use* pointers.
+struct NarrowIVDefUse {
+ Instruction *NarrowDef;
+ Instruction *NarrowUse;
+ Instruction *WideDef;
+
+ NarrowIVDefUse(): NarrowDef(0), NarrowUse(0), WideDef(0) {}
+
+ NarrowIVDefUse(Instruction *ND, Instruction *NU, Instruction *WD):
+ NarrowDef(ND), NarrowUse(NU), WideDef(WD) {}
+};
+
/// WidenIV - The goal of this transform is to remove sign and zero extends
/// without creating any new induction variables. To do this, it creates a new
/// phi of the wider type and redirects all users, either removing extends or
@@ -787,7 +826,7 @@ namespace {
class WidenIV {
// Parameters
PHINode *OrigPhi;
- const Type *WideType;
+ Type *WideType;
bool IsSigned;
// Context
@@ -803,7 +842,7 @@ class WidenIV {
SmallVectorImpl<WeakVH> &DeadInsts;
SmallPtrSet<Instruction*,16> Widened;
- SmallVector<std::pair<Use *, Instruction *>, 8> NarrowIVUsers;
+ SmallVector<NarrowIVDefUse, 8> NarrowIVUsers;
public:
WidenIV(PHINode *PN, const WideIVInfo &WI, LoopInfo *LInfo,
@@ -826,20 +865,17 @@ public:
PHINode *CreateWideIV(SCEVExpander &Rewriter);
protected:
- Instruction *CloneIVUser(Instruction *NarrowUse,
- Instruction *NarrowDef,
- Instruction *WideDef);
+ Instruction *CloneIVUser(NarrowIVDefUse DU);
const SCEVAddRecExpr *GetWideRecurrence(Instruction *NarrowUse);
- Instruction *WidenIVUse(Use &NarrowDefUse, Instruction *NarrowDef,
- Instruction *WideDef);
+ Instruction *WidenIVUse(NarrowIVDefUse DU);
void pushNarrowIVUsers(Instruction *NarrowDef, Instruction *WideDef);
};
} // anonymous namespace
-static Value *getExtend( Value *NarrowOper, const Type *WideType,
+static Value *getExtend( Value *NarrowOper, Type *WideType,
bool IsSigned, IRBuilder<> &Builder) {
return IsSigned ? Builder.CreateSExt(NarrowOper, WideType) :
Builder.CreateZExt(NarrowOper, WideType);
@@ -848,10 +884,8 @@ static Value *getExtend( Value *NarrowOper, const Type *WideType,
/// CloneIVUser - Instantiate a wide operation to replace a narrow
/// operation. This only needs to handle operations that can evaluation to
/// SCEVAddRec. It can safely return 0 for any operation we decide not to clone.
-Instruction *WidenIV::CloneIVUser(Instruction *NarrowUse,
- Instruction *NarrowDef,
- Instruction *WideDef) {
- unsigned Opcode = NarrowUse->getOpcode();
+Instruction *WidenIV::CloneIVUser(NarrowIVDefUse DU) {
+ unsigned Opcode = DU.NarrowUse->getOpcode();
switch (Opcode) {
default:
return 0;
@@ -865,21 +899,21 @@ Instruction *WidenIV::CloneIVUser(Instruction *NarrowUse,
case Instruction::Shl:
case Instruction::LShr:
case Instruction::AShr:
- DEBUG(dbgs() << "Cloning IVUser: " << *NarrowUse << "\n");
+ DEBUG(dbgs() << "Cloning IVUser: " << *DU.NarrowUse << "\n");
- IRBuilder<> Builder(NarrowUse);
+ IRBuilder<> Builder(DU.NarrowUse);
// Replace NarrowDef operands with WideDef. Otherwise, we don't know
// anything about the narrow operand yet so must insert a [sz]ext. It is
// probably loop invariant and will be folded or hoisted. If it actually
// comes from a widened IV, it should be removed during a future call to
// WidenIVUse.
- Value *LHS = (NarrowUse->getOperand(0) == NarrowDef) ? WideDef :
- getExtend(NarrowUse->getOperand(0), WideType, IsSigned, Builder);
- Value *RHS = (NarrowUse->getOperand(1) == NarrowDef) ? WideDef :
- getExtend(NarrowUse->getOperand(1), WideType, IsSigned, Builder);
+ Value *LHS = (DU.NarrowUse->getOperand(0) == DU.NarrowDef) ? DU.WideDef :
+ getExtend(DU.NarrowUse->getOperand(0), WideType, IsSigned, Builder);
+ Value *RHS = (DU.NarrowUse->getOperand(1) == DU.NarrowDef) ? DU.WideDef :
+ getExtend(DU.NarrowUse->getOperand(1), WideType, IsSigned, Builder);
- BinaryOperator *NarrowBO = cast<BinaryOperator>(NarrowUse);
+ BinaryOperator *NarrowBO = cast<BinaryOperator>(DU.NarrowUse);
BinaryOperator *WideBO = BinaryOperator::Create(NarrowBO->getOpcode(),
LHS, RHS,
NarrowBO->getName());
@@ -957,41 +991,40 @@ const SCEVAddRecExpr *WidenIV::GetWideRecurrence(Instruction *NarrowUse) {
/// WidenIVUse - Determine whether an individual user of the narrow IV can be
/// widened. If so, return the wide clone of the user.
-Instruction *WidenIV::WidenIVUse(Use &NarrowDefUse, Instruction *NarrowDef,
- Instruction *WideDef) {
- Instruction *NarrowUse = cast<Instruction>(NarrowDefUse.getUser());
+Instruction *WidenIV::WidenIVUse(NarrowIVDefUse DU) {
// Stop traversing the def-use chain at inner-loop phis or post-loop phis.
- if (isa<PHINode>(NarrowUse) && LI->getLoopFor(NarrowUse->getParent()) != L)
+ if (isa<PHINode>(DU.NarrowUse) &&
+ LI->getLoopFor(DU.NarrowUse->getParent()) != L)
return 0;
// Our raison d'etre! Eliminate sign and zero extension.
- if (IsSigned ? isa<SExtInst>(NarrowUse) : isa<ZExtInst>(NarrowUse)) {
- Value *NewDef = WideDef;
- if (NarrowUse->getType() != WideType) {
- unsigned CastWidth = SE->getTypeSizeInBits(NarrowUse->getType());
+ if (IsSigned ? isa<SExtInst>(DU.NarrowUse) : isa<ZExtInst>(DU.NarrowUse)) {
+ Value *NewDef = DU.WideDef;
+ if (DU.NarrowUse->getType() != WideType) {
+ unsigned CastWidth = SE->getTypeSizeInBits(DU.NarrowUse->getType());
unsigned IVWidth = SE->getTypeSizeInBits(WideType);
if (CastWidth < IVWidth) {
// The cast isn't as wide as the IV, so insert a Trunc.
- IRBuilder<> Builder(NarrowDefUse);
- NewDef = Builder.CreateTrunc(WideDef, NarrowUse->getType());
+ IRBuilder<> Builder(DU.NarrowUse);
+ NewDef = Builder.CreateTrunc(DU.WideDef, DU.NarrowUse->getType());
}
else {
// A wider extend was hidden behind a narrower one. This may induce
// another round of IV widening in which the intermediate IV becomes
// dead. It should be very rare.
DEBUG(dbgs() << "INDVARS: New IV " << *WidePhi
- << " not wide enough to subsume " << *NarrowUse << "\n");
- NarrowUse->replaceUsesOfWith(NarrowDef, WideDef);
- NewDef = NarrowUse;
+ << " not wide enough to subsume " << *DU.NarrowUse << "\n");
+ DU.NarrowUse->replaceUsesOfWith(DU.NarrowDef, DU.WideDef);
+ NewDef = DU.NarrowUse;
}
}
- if (NewDef != NarrowUse) {
- DEBUG(dbgs() << "INDVARS: eliminating " << *NarrowUse
- << " replaced by " << *WideDef << "\n");
+ if (NewDef != DU.NarrowUse) {
+ DEBUG(dbgs() << "INDVARS: eliminating " << *DU.NarrowUse
+ << " replaced by " << *DU.WideDef << "\n");
++NumElimExt;
- NarrowUse->replaceAllUsesWith(NewDef);
- DeadInsts.push_back(NarrowUse);
+ DU.NarrowUse->replaceAllUsesWith(NewDef);
+ DeadInsts.push_back(DU.NarrowUse);
}
// Now that the extend is gone, we want to expose it's uses for potential
// further simplification. We don't need to directly inform SimplifyIVUsers
@@ -1004,29 +1037,29 @@ Instruction *WidenIV::WidenIVUse(Use &NarrowDefUse, Instruction *NarrowDef,
}
// Does this user itself evaluate to a recurrence after widening?
- const SCEVAddRecExpr *WideAddRec = GetWideRecurrence(NarrowUse);
+ const SCEVAddRecExpr *WideAddRec = GetWideRecurrence(DU.NarrowUse);
if (!WideAddRec) {
// This user does not evaluate to a recurence after widening, so don't
// follow it. Instead insert a Trunc to kill off the original use,
// eventually isolating the original narrow IV so it can be removed.
- IRBuilder<> Builder(NarrowDefUse);
- Value *Trunc = Builder.CreateTrunc(WideDef, NarrowDef->getType());
- NarrowUse->replaceUsesOfWith(NarrowDef, Trunc);
+ IRBuilder<> Builder(getInsertPointForUses(DU.NarrowUse, DU.NarrowDef, DT));
+ Value *Trunc = Builder.CreateTrunc(DU.WideDef, DU.NarrowDef->getType());
+ DU.NarrowUse->replaceUsesOfWith(DU.NarrowDef, Trunc);
return 0;
}
- // We assume that block terminators are not SCEVable. We wouldn't want to
+ // Assume block terminators cannot evaluate to a recurrence. We can't to
// insert a Trunc after a terminator if there happens to be a critical edge.
- assert(NarrowUse != NarrowUse->getParent()->getTerminator() &&
+ assert(DU.NarrowUse != DU.NarrowUse->getParent()->getTerminator() &&
"SCEV is not expected to evaluate a block terminator");
// Reuse the IV increment that SCEVExpander created as long as it dominates
// NarrowUse.
Instruction *WideUse = 0;
- if (WideAddRec == WideIncExpr && HoistStep(WideInc, NarrowUse, DT)) {
+ if (WideAddRec == WideIncExpr && HoistStep(WideInc, DU.NarrowUse, DT)) {
WideUse = WideInc;
}
else {
- WideUse = CloneIVUser(NarrowUse, NarrowDef, WideDef);
+ WideUse = CloneIVUser(DU);
if (!WideUse)
return 0;
}
@@ -1051,13 +1084,13 @@ Instruction *WidenIV::WidenIVUse(Use &NarrowDefUse, Instruction *NarrowDef,
void WidenIV::pushNarrowIVUsers(Instruction *NarrowDef, Instruction *WideDef) {
for (Value::use_iterator UI = NarrowDef->use_begin(),
UE = NarrowDef->use_end(); UI != UE; ++UI) {
- Use &U = UI.getUse();
+ Instruction *NarrowUse = cast<Instruction>(*UI);
// Handle data flow merges and bizarre phi cycles.
- if (!Widened.insert(cast<Instruction>(U.getUser())))
+ if (!Widened.insert(NarrowUse))
continue;
- NarrowIVUsers.push_back(std::make_pair(&UI.getUse(), WideDef));
+ NarrowIVUsers.push_back(NarrowIVDefUse(NarrowDef, NarrowUse, WideDef));
}
}
@@ -1124,23 +1157,19 @@ PHINode *WidenIV::CreateWideIV(SCEVExpander &Rewriter) {
pushNarrowIVUsers(OrigPhi, WidePhi);
while (!NarrowIVUsers.empty()) {
- Use *UsePtr;
- Instruction *WideDef;
- tie(UsePtr, WideDef) = NarrowIVUsers.pop_back_val();
- Use &NarrowDefUse = *UsePtr;
+ NarrowIVDefUse DU = NarrowIVUsers.pop_back_val();
// Process a def-use edge. This may replace the use, so don't hold a
// use_iterator across it.
- Instruction *NarrowDef = cast<Instruction>(NarrowDefUse.get());
- Instruction *WideUse = WidenIVUse(NarrowDefUse, NarrowDef, WideDef);
+ Instruction *WideUse = WidenIVUse(DU);
// Follow all def-use edges from the previous narrow use.
if (WideUse)
- pushNarrowIVUsers(cast<Instruction>(NarrowDefUse.getUser()), WideUse);
+ pushNarrowIVUsers(DU.NarrowUse, WideUse);
// WidenIVUse may have removed the def-use edge.
- if (NarrowDef->use_empty())
- DeadInsts.push_back(NarrowDef);
+ if (DU.NarrowDef->use_empty())
+ DeadInsts.push_back(DU.NarrowDef);
}
return WidePhi;
}
@@ -1302,10 +1331,6 @@ static bool isSimpleIVUser(Instruction *I, const Loop *L, ScalarEvolution *SE) {
// Get the symbolic expression for this instruction.
const SCEV *S = SE->getSCEV(I);
- // We assume that terminators are not SCEVable.
- assert((!S || I != I->getParent()->getTerminator()) &&
- "can't fold terminators");
-
// Only consider affine recurrences.
const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(S);
if (AR && AR->getLoop() == L)
@@ -1418,6 +1443,11 @@ void IndVarSimplify::SimplifyCongruentIVs(Loop *L) {
if (Inserted)
continue;
PHINode *OrigPhi = Pos->second;
+
+ // If one phi derives from the other via GEPs, types may differ.
+ if (OrigPhi->getType() != Phi->getType())
+ continue;
+
// Replacing the congruent phi is sufficient because acyclic redundancy
// elimination, CSE/GVN, should handle the rest. However, once SCEV proves
// that a phi is congruent, it's almost certain to be the head of an IV
@@ -1429,6 +1459,7 @@ void IndVarSimplify::SimplifyCongruentIVs(Loop *L) {
Instruction *IsomorphicInc =
cast<Instruction>(Phi->getIncomingValueForBlock(LatchBlock));
if (OrigInc != IsomorphicInc &&
+ OrigInc->getType() == IsomorphicInc->getType() &&
SE->getSCEV(OrigInc) == SE->getSCEV(IsomorphicInc) &&
HoistStep(OrigInc, IsomorphicInc, DT)) {
DEBUG(dbgs() << "INDVARS: Eliminated congruent iv.inc: "
@@ -1448,6 +1479,54 @@ void IndVarSimplify::SimplifyCongruentIVs(Loop *L) {
// LinearFunctionTestReplace and its kin. Rewrite the loop exit condition.
//===----------------------------------------------------------------------===//
+// Check for expressions that ScalarEvolution generates to compute
+// BackedgeTakenInfo. If these expressions have not been reduced, then expanding
+// them may incur additional cost (albeit in the loop preheader).
+static bool isHighCostExpansion(const SCEV *S, BranchInst *BI,
+ ScalarEvolution *SE) {
+ // If the backedge-taken count is a UDiv, it's very likely a UDiv that
+ // ScalarEvolution's HowFarToZero or HowManyLessThans produced to compute a
+ // precise expression, rather than a UDiv from the user's code. If we can't
+ // find a UDiv in the code with some simple searching, assume the former and
+ // forego rewriting the loop.
+ if (isa<SCEVUDivExpr>(S)) {
+ ICmpInst *OrigCond = dyn_cast<ICmpInst>(BI->getCondition());
+ if (!OrigCond) return true;
+ const SCEV *R = SE->getSCEV(OrigCond->getOperand(1));
+ R = SE->getMinusSCEV(R, SE->getConstant(R->getType(), 1));
+ if (R != S) {
+ const SCEV *L = SE->getSCEV(OrigCond->getOperand(0));
+ L = SE->getMinusSCEV(L, SE->getConstant(L->getType(), 1));
+ if (L != S)
+ return true;
+ }
+ }
+
+ if (!DisableIVRewrite || ForceLFTR)
+ return false;
+
+ // Recurse past add expressions, which commonly occur in the
+ // BackedgeTakenCount. They may already exist in program code, and if not,
+ // they are not too expensive rematerialize.
+ if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(S)) {
+ for (SCEVAddExpr::op_iterator I = Add->op_begin(), E = Add->op_end();
+ I != E; ++I) {
+ if (isHighCostExpansion(*I, BI, SE))
+ return true;
+ }
+ return false;
+ }
+
+ // HowManyLessThans uses a Max expression whenever the loop is not guarded by
+ // the exit condition.
+ if (isa<SCEVSMaxExpr>(S) || isa<SCEVUMaxExpr>(S))
+ return true;
+
+ // If we haven't recognized an expensive SCEV patter, assume its an expression
+ // produced by program code.
+ return false;
+}
+
/// canExpandBackedgeTakenCount - Return true if this loop's backedge taken
/// count expression can be safely and cheaply expanded into an instruction
/// sequence that can be used by LinearFunctionTestReplace.
@@ -1465,31 +1544,17 @@ static bool canExpandBackedgeTakenCount(Loop *L, ScalarEvolution *SE) {
if (!BI)
return false;
- // Special case: If the backedge-taken count is a UDiv, it's very likely a
- // UDiv that ScalarEvolution produced in order to compute a precise
- // expression, rather than a UDiv from the user's code. If we can't find a
- // UDiv in the code with some simple searching, assume the former and forego
- // rewriting the loop.
- if (isa<SCEVUDivExpr>(BackedgeTakenCount)) {
- ICmpInst *OrigCond = dyn_cast<ICmpInst>(BI->getCondition());
- if (!OrigCond) return false;
- const SCEV *R = SE->getSCEV(OrigCond->getOperand(1));
- R = SE->getMinusSCEV(R, SE->getConstant(R->getType(), 1));
- if (R != BackedgeTakenCount) {
- const SCEV *L = SE->getSCEV(OrigCond->getOperand(0));
- L = SE->getMinusSCEV(L, SE->getConstant(L->getType(), 1));
- if (L != BackedgeTakenCount)
- return false;
- }
- }
+ if (isHighCostExpansion(BackedgeTakenCount, BI, SE))
+ return false;
+
return true;
}
/// getBackedgeIVType - Get the widest type used by the loop test after peeking
/// through Truncs.
///
-/// TODO: Unnecessary if LFTR does not force a canonical IV.
-static const Type *getBackedgeIVType(Loop *L) {
+/// TODO: Unnecessary when ForceLFTR is removed.
+static Type *getBackedgeIVType(Loop *L) {
if (!L->getExitingBlock())
return 0;
@@ -1502,7 +1567,7 @@ static const Type *getBackedgeIVType(Loop *L) {
if (!Cond)
return 0;
- const Type *Ty = 0;
+ Type *Ty = 0;
for(User::op_iterator OI = Cond->op_begin(), OE = Cond->op_end();
OI != OE; ++OI) {
assert((!Ty || Ty == (*OI)->getType()) && "bad icmp operand types");
@@ -1515,12 +1580,198 @@ static const Type *getBackedgeIVType(Loop *L) {
return Ty;
}
+/// isLoopInvariant - Perform a quick domtree based check for loop invariance
+/// assuming that V is used within the loop. LoopInfo::isLoopInvariant() seems
+/// gratuitous for this purpose.
+static bool isLoopInvariant(Value *V, Loop *L, DominatorTree *DT) {
+ Instruction *Inst = dyn_cast<Instruction>(V);
+ if (!Inst)
+ return true;
+
+ return DT->properlyDominates(Inst->getParent(), L->getHeader());
+}
+
+/// getLoopPhiForCounter - Return the loop header phi IFF IncV adds a loop
+/// invariant value to the phi.
+static PHINode *getLoopPhiForCounter(Value *IncV, Loop *L, DominatorTree *DT) {
+ Instruction *IncI = dyn_cast<Instruction>(IncV);
+ if (!IncI)
+ return 0;
+
+ switch (IncI->getOpcode()) {
+ case Instruction::Add:
+ case Instruction::Sub:
+ break;
+ case Instruction::GetElementPtr:
+ // An IV counter must preserve its type.
+ if (IncI->getNumOperands() == 2)
+ break;
+ default:
+ return 0;
+ }
+
+ PHINode *Phi = dyn_cast<PHINode>(IncI->getOperand(0));
+ if (Phi && Phi->getParent() == L->getHeader()) {
+ if (isLoopInvariant(IncI->getOperand(1), L, DT))
+ return Phi;
+ return 0;
+ }
+ if (IncI->getOpcode() == Instruction::GetElementPtr)
+ return 0;
+
+ // Allow add/sub to be commuted.
+ Phi = dyn_cast<PHINode>(IncI->getOperand(1));
+ if (Phi && Phi->getParent() == L->getHeader()) {
+ if (isLoopInvariant(IncI->getOperand(0), L, DT))
+ return Phi;
+ }
+ return 0;
+}
+
+/// needsLFTR - LinearFunctionTestReplace policy. Return true unless we can show
+/// that the current exit test is already sufficiently canonical.
+static bool needsLFTR(Loop *L, DominatorTree *DT) {
+ assert(L->getExitingBlock() && "expected loop exit");
+
+ BasicBlock *LatchBlock = L->getLoopLatch();
+ // Don't bother with LFTR if the loop is not properly simplified.
+ if (!LatchBlock)
+ return false;
+
+ BranchInst *BI = dyn_cast<BranchInst>(L->getExitingBlock()->getTerminator());
+ assert(BI && "expected exit branch");
+
+ // Do LFTR to simplify the exit condition to an ICMP.
+ ICmpInst *Cond = dyn_cast<ICmpInst>(BI->getCondition());
+ if (!Cond)
+ return true;
+
+ // Do LFTR to simplify the exit ICMP to EQ/NE
+ ICmpInst::Predicate Pred = Cond->getPredicate();
+ if (Pred != ICmpInst::ICMP_NE && Pred != ICmpInst::ICMP_EQ)
+ return true;
+
+ // Look for a loop invariant RHS
+ Value *LHS = Cond->getOperand(0);
+ Value *RHS = Cond->getOperand(1);
+ if (!isLoopInvariant(RHS, L, DT)) {
+ if (!isLoopInvariant(LHS, L, DT))
+ return true;
+ std::swap(LHS, RHS);
+ }
+ // Look for a simple IV counter LHS
+ PHINode *Phi = dyn_cast<PHINode>(LHS);
+ if (!Phi)
+ Phi = getLoopPhiForCounter(LHS, L, DT);
+
+ if (!Phi)
+ return true;
+
+ // Do LFTR if the exit condition's IV is *not* a simple counter.
+ Value *IncV = Phi->getIncomingValueForBlock(L->getLoopLatch());
+ return Phi != getLoopPhiForCounter(IncV, L, DT);
+}
+
+/// AlmostDeadIV - Return true if this IV has any uses other than the (soon to
+/// be rewritten) loop exit test.
+static bool AlmostDeadIV(PHINode *Phi, BasicBlock *LatchBlock, Value *Cond) {
+ int LatchIdx = Phi->getBasicBlockIndex(LatchBlock);
+ Value *IncV = Phi->getIncomingValue(LatchIdx);
+
+ for (Value::use_iterator UI = Phi->use_begin(), UE = Phi->use_end();
+ UI != UE; ++UI) {
+ if (*UI != Cond && *UI != IncV) return false;
+ }
+
+ for (Value::use_iterator UI = IncV->use_begin(), UE = IncV->use_end();
+ UI != UE; ++UI) {
+ if (*UI != Cond && *UI != Phi) return false;
+ }
+ return true;
+}
+
+/// FindLoopCounter - Find an affine IV in canonical form.
+///
+/// FIXME: Accept -1 stride and set IVLimit = IVInit - BECount
+///
+/// FIXME: Accept non-unit stride as long as SCEV can reduce BECount * Stride.
+/// This is difficult in general for SCEV because of potential overflow. But we
+/// could at least handle constant BECounts.
+static PHINode *
+FindLoopCounter(Loop *L, const SCEV *BECount,
+ ScalarEvolution *SE, DominatorTree *DT, const TargetData *TD) {
+ // I'm not sure how BECount could be a pointer type, but we definitely don't
+ // want to LFTR that.
+ if (BECount->getType()->isPointerTy())
+ return 0;
+
+ uint64_t BCWidth = SE->getTypeSizeInBits(BECount->getType());
+
+ Value *Cond =
+ cast<BranchInst>(L->getExitingBlock()->getTerminator())->getCondition();
+
+ // Loop over all of the PHI nodes, looking for a simple counter.
+ PHINode *BestPhi = 0;
+ const SCEV *BestInit = 0;
+ BasicBlock *LatchBlock = L->getLoopLatch();
+ assert(LatchBlock && "needsLFTR should guarantee a loop latch");
+
+ for (BasicBlock::iterator I = L->getHeader()->begin(); isa<PHINode>(I); ++I) {
+ PHINode *Phi = cast<PHINode>(I);
+ if (!SE->isSCEVable(Phi->getType()))
+ continue;
+
+ const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(SE->getSCEV(Phi));
+ if (!AR || AR->getLoop() != L || !AR->isAffine())
+ continue;
+
+ // AR may be a pointer type, while BECount is an integer type.
+ // AR may be wider than BECount. With eq/ne tests overflow is immaterial.
+ // AR may not be a narrower type, or we may never exit.
+ uint64_t PhiWidth = SE->getTypeSizeInBits(AR->getType());
+ if (PhiWidth < BCWidth || (TD && !TD->isLegalInteger(PhiWidth)))
+ continue;
+
+ const SCEV *Step = dyn_cast<SCEVConstant>(AR->getStepRecurrence(*SE));
+ if (!Step || !Step->isOne())
+ continue;
+
+ int LatchIdx = Phi->getBasicBlockIndex(LatchBlock);
+ Value *IncV = Phi->getIncomingValue(LatchIdx);
+ if (getLoopPhiForCounter(IncV, L, DT) != Phi)
+ continue;
+
+ const SCEV *Init = AR->getStart();
+
+ if (BestPhi && !AlmostDeadIV(BestPhi, LatchBlock, Cond)) {
+ // Don't force a live loop counter if another IV can be used.
+ if (AlmostDeadIV(Phi, LatchBlock, Cond))
+ continue;
+
+ // Prefer to count-from-zero. This is a more "canonical" counter form. It
+ // also prefers integer to pointer IVs.
+ if (BestInit->isZero() != Init->isZero()) {
+ if (BestInit->isZero())
+ continue;
+ }
+ // If two IVs both count from zero or both count from nonzero then the
+ // narrower is likely a dead phi that has been widened. Use the wider phi
+ // to allow the other to be eliminated.
+ if (PhiWidth <= SE->getTypeSizeInBits(BestPhi->getType()))
+ continue;
+ }
+ BestPhi = Phi;
+ BestInit = Init;
+ }
+ return BestPhi;
+}
+
/// LinearFunctionTestReplace - This method rewrites the exit condition of the
/// loop to be a canonical != comparison against the incremented loop induction
/// variable. This pass is able to rewrite the exit tests of any loop where the
/// SCEV analysis can determine a loop-invariant trip count of the loop, which
/// is actually a much broader range than just linear tests.
-ICmpInst *IndVarSimplify::
+Value *IndVarSimplify::
LinearFunctionTestReplace(Loop *L,
const SCEV *BackedgeTakenCount,
PHINode *IndVar,
@@ -1528,62 +1779,118 @@ LinearFunctionTestReplace(Loop *L,
assert(canExpandBackedgeTakenCount(L, SE) && "precondition");
BranchInst *BI = cast<BranchInst>(L->getExitingBlock()->getTerminator());
+ // In DisableIVRewrite mode, IndVar is not necessarily a canonical IV. In this
+ // mode, LFTR can ignore IV overflow and truncate to the width of
+ // BECount. This avoids materializing the add(zext(add)) expression.
+ Type *CntTy = DisableIVRewrite ?
+ BackedgeTakenCount->getType() : IndVar->getType();
+
+ const SCEV *IVLimit = BackedgeTakenCount;
+
// If the exiting block is not the same as the backedge block, we must compare
// against the preincremented value, otherwise we prefer to compare against
// the post-incremented value.
Value *CmpIndVar;
- const SCEV *RHS = BackedgeTakenCount;
if (L->getExitingBlock() == L->getLoopLatch()) {
// Add one to the "backedge-taken" count to get the trip count.
// If this addition may overflow, we have to be more pessimistic and
// cast the induction variable before doing the add.
- const SCEV *Zero = SE->getConstant(BackedgeTakenCount->getType(), 0);
const SCEV *N =
- SE->getAddExpr(BackedgeTakenCount,
- SE->getConstant(BackedgeTakenCount->getType(), 1));
- if ((isa<SCEVConstant>(N) && !N->isZero()) ||
- SE->isLoopEntryGuardedByCond(L, ICmpInst::ICMP_NE, N, Zero)) {
- // No overflow. Cast the sum.
- RHS = SE->getTruncateOrZeroExtend(N, IndVar->getType());
- } else {
- // Potential overflow. Cast before doing the add.
- RHS = SE->getTruncateOrZeroExtend(BackedgeTakenCount,
- IndVar->getType());
- RHS = SE->getAddExpr(RHS,
- SE->getConstant(IndVar->getType(), 1));
+ SE->getAddExpr(IVLimit, SE->getConstant(IVLimit->getType(), 1));
+ if (CntTy == IVLimit->getType())
+ IVLimit = N;
+ else {
+ const SCEV *Zero = SE->getConstant(IVLimit->getType(), 0);
+ if ((isa<SCEVConstant>(N) && !N->isZero()) ||
+ SE->isLoopEntryGuardedByCond(L, ICmpInst::ICMP_NE, N, Zero)) {
+ // No overflow. Cast the sum.
+ IVLimit = SE->getTruncateOrZeroExtend(N, CntTy);
+ } else {
+ // Potential overflow. Cast before doing the add.
+ IVLimit = SE->getTruncateOrZeroExtend(IVLimit, CntTy);
+ IVLimit = SE->getAddExpr(IVLimit, SE->getConstant(CntTy, 1));
+ }
}
-
// The BackedgeTaken expression contains the number of times that the
// backedge branches to the loop header. This is one less than the
// number of times the loop executes, so use the incremented indvar.
CmpIndVar = IndVar->getIncomingValueForBlock(L->getExitingBlock());
} else {
// We have to use the preincremented value...
- RHS = SE->getTruncateOrZeroExtend(BackedgeTakenCount,
- IndVar->getType());
+ IVLimit = SE->getTruncateOrZeroExtend(IVLimit, CntTy);
CmpIndVar = IndVar;
}
+ // For unit stride, IVLimit = Start + BECount with 2's complement overflow.
+ // So for, non-zero start compute the IVLimit here.
+ bool isPtrIV = false;
+ Type *CmpTy = CntTy;
+ const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(SE->getSCEV(IndVar));
+ assert(AR && AR->getLoop() == L && AR->isAffine() && "bad loop counter");
+ if (!AR->getStart()->isZero()) {
+ assert(AR->getStepRecurrence(*SE)->isOne() && "only handles unit stride");
+ const SCEV *IVInit = AR->getStart();
+
+ // For pointer types, sign extend BECount in order to materialize a GEP.
+ // Note that for DisableIVRewrite, we never run SCEVExpander on a
+ // pointer type, because we must preserve the existing GEPs. Instead we
+ // directly generate a GEP later.
+ if (IVInit->getType()->isPointerTy()) {
+ isPtrIV = true;
+ CmpTy = SE->getEffectiveSCEVType(IVInit->getType());
+ IVLimit = SE->getTruncateOrSignExtend(IVLimit, CmpTy);
+ }
+ // For integer types, truncate the IV before computing IVInit + BECount.
+ else {
+ if (SE->getTypeSizeInBits(IVInit->getType())
+ > SE->getTypeSizeInBits(CmpTy))
+ IVInit = SE->getTruncateExpr(IVInit, CmpTy);
+
+ IVLimit = SE->getAddExpr(IVInit, IVLimit);
+ }
+ }
// Expand the code for the iteration count.
- assert(SE->isLoopInvariant(RHS, L) &&
+ IRBuilder<> Builder(BI);
+
+ assert(SE->isLoopInvariant(IVLimit, L) &&
"Computed iteration count is not loop invariant!");
- Value *ExitCnt = Rewriter.expandCodeFor(RHS, IndVar->getType(), BI);
+ Value *ExitCnt = Rewriter.expandCodeFor(IVLimit, CmpTy, BI);
+
+ // Create a gep for IVInit + IVLimit from on an existing pointer base.
+ assert(isPtrIV == IndVar->getType()->isPointerTy() &&
+ "IndVar type must match IVInit type");
+ if (isPtrIV) {
+ Value *IVStart = IndVar->getIncomingValueForBlock(L->getLoopPreheader());
+ assert(AR->getStart() == SE->getSCEV(IVStart) && "bad loop counter");
+ assert(SE->getSizeOfExpr(
+ cast<PointerType>(IVStart->getType())->getElementType())->isOne()
+ && "unit stride pointer IV must be i8*");
+
+ Builder.SetInsertPoint(L->getLoopPreheader()->getTerminator());
+ ExitCnt = Builder.CreateGEP(IVStart, ExitCnt, "lftr.limit");
+ Builder.SetInsertPoint(BI);
+ }
// Insert a new icmp_ne or icmp_eq instruction before the branch.
- ICmpInst::Predicate Opcode;
+ ICmpInst::Predicate P;
if (L->contains(BI->getSuccessor(0)))
- Opcode = ICmpInst::ICMP_NE;
+ P = ICmpInst::ICMP_NE;
else
- Opcode = ICmpInst::ICMP_EQ;
+ P = ICmpInst::ICMP_EQ;
DEBUG(dbgs() << "INDVARS: Rewriting loop exit condition to:\n"
<< " LHS:" << *CmpIndVar << '\n'
<< " op:\t"
- << (Opcode == ICmpInst::ICMP_NE ? "!=" : "==") << "\n"
- << " RHS:\t" << *RHS << "\n");
+ << (P == ICmpInst::ICMP_NE ? "!=" : "==") << "\n"
+ << " RHS:\t" << *ExitCnt << "\n"
+ << " Expr:\t" << *IVLimit << "\n");
+
+ if (SE->getTypeSizeInBits(CmpIndVar->getType())
+ > SE->getTypeSizeInBits(CmpTy)) {
+ CmpIndVar = Builder.CreateTrunc(CmpIndVar, CmpTy, "lftr.wideiv");
+ }
- ICmpInst *Cond = new ICmpInst(BI, Opcode, CmpIndVar, ExitCnt, "exitcond");
- Cond->setDebugLoc(BI->getDebugLoc());
+ Value *Cond = Builder.CreateICmp(P, CmpIndVar, ExitCnt, "exitcond");
Value *OrigCond = BI->getCondition();
// It's tempting to use replaceAllUsesWith here to fully replace the old
// comparison, but that's not immediately safe, since users of the old
@@ -1748,15 +2055,16 @@ bool IndVarSimplify::runOnLoop(Loop *L, LPPassManager &LPM) {
// Compute the type of the largest recurrence expression, and decide whether
// a canonical induction variable should be inserted.
- const Type *LargestType = 0;
+ Type *LargestType = 0;
bool NeedCannIV = false;
+ bool ReuseIVForExit = DisableIVRewrite && !ForceLFTR;
bool ExpandBECount = canExpandBackedgeTakenCount(L, SE);
- if (ExpandBECount) {
+ if (ExpandBECount && !ReuseIVForExit) {
// If we have a known trip count and a single exit block, we'll be
// rewriting the loop exit test condition below, which requires a
// canonical induction variable.
NeedCannIV = true;
- const Type *Ty = BackedgeTakenCount->getType();
+ Type *Ty = BackedgeTakenCount->getType();
if (DisableIVRewrite) {
// In this mode, SimplifyIVUsers may have already widened the IV used by
// the backedge test and inserted a Trunc on the compare's operand. Get
@@ -1772,7 +2080,7 @@ bool IndVarSimplify::runOnLoop(Loop *L, LPPassManager &LPM) {
if (!DisableIVRewrite) {
for (IVUsers::const_iterator I = IU->begin(), E = IU->end(); I != E; ++I) {
NeedCannIV = true;
- const Type *Ty =
+ Type *Ty =
SE->getEffectiveSCEVType(I->getOperandValToReplace()->getType());
if (!LargestType ||
SE->getTypeSizeInBits(Ty) >
@@ -1814,15 +2122,13 @@ bool IndVarSimplify::runOnLoop(Loop *L, LPPassManager &LPM) {
OldCannIV->insertBefore(L->getHeader()->getFirstNonPHI());
}
}
-
+ else if (ExpandBECount && ReuseIVForExit && needsLFTR(L, DT)) {
+ IndVar = FindLoopCounter(L, BackedgeTakenCount, SE, DT, TD);
+ }
// If we have a trip count expression, rewrite the loop's exit condition
// using it. We can currently only handle loops with a single exit.
- ICmpInst *NewICmp = 0;
- if (ExpandBECount) {
- assert(canExpandBackedgeTakenCount(L, SE) &&
- "canonical IV disrupted BackedgeTaken expansion");
- assert(NeedCannIV &&
- "LinearFunctionTestReplace requires a canonical induction variable");
+ Value *NewICmp = 0;
+ if (ExpandBECount && IndVar) {
// Check preconditions for proper SCEVExpander operation. SCEV does not
// express SCEVExpander's dependencies, such as LoopSimplify. Instead any
// pass that uses the SCEVExpander must do it. This does not work well for
@@ -1860,12 +2166,33 @@ bool IndVarSimplify::runOnLoop(Loop *L, LPPassManager &LPM) {
// For completeness, inform IVUsers of the IV use in the newly-created
// loop exit test instruction.
- if (NewICmp && IU)
- IU->AddUsersIfInteresting(cast<Instruction>(NewICmp->getOperand(0)));
-
+ if (IU && NewICmp) {
+ ICmpInst *NewICmpInst = dyn_cast<ICmpInst>(NewICmp);
+ if (NewICmpInst)
+ IU->AddUsersIfInteresting(cast<Instruction>(NewICmpInst->getOperand(0)));
+ }
// Clean up dead instructions.
Changed |= DeleteDeadPHIs(L->getHeader());
// Check a post-condition.
- assert(L->isLCSSAForm(*DT) && "Indvars did not leave the loop in lcssa form!");
+ assert(L->isLCSSAForm(*DT) &&
+ "Indvars did not leave the loop in lcssa form!");
+
+ // Verify that LFTR, and any other change have not interfered with SCEV's
+ // ability to compute trip count.
+#ifndef NDEBUG
+ if (DisableIVRewrite && !isa<SCEVCouldNotCompute>(BackedgeTakenCount)) {
+ SE->forgetLoop(L);
+ const SCEV *NewBECount = SE->getBackedgeTakenCount(L);
+ if (SE->getTypeSizeInBits(BackedgeTakenCount->getType()) <
+ SE->getTypeSizeInBits(NewBECount->getType()))
+ NewBECount = SE->getTruncateOrNoop(NewBECount,
+ BackedgeTakenCount->getType());
+ else
+ BackedgeTakenCount = SE->getTruncateOrNoop(BackedgeTakenCount,
+ NewBECount->getType());
+ assert(BackedgeTakenCount == NewBECount && "indvars must preserve SCEV");
+ }
+#endif
+
return Changed;
}
diff --git a/lib/Transforms/Scalar/LoopIdiomRecognize.cpp b/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
index a0e41d9..ea4c515 100644
--- a/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
+++ b/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
@@ -498,7 +498,7 @@ processLoopStridedStore(Value *DestPtr, unsigned StoreSize,
// The # stored bytes is (BECount+1)*Size. Expand the trip count out to
// pointer size if it isn't already.
- const Type *IntPtr = TD->getIntPtrType(DestPtr->getContext());
+ Type *IntPtr = TD->getIntPtrType(DestPtr->getContext());
BECount = SE->getTruncateOrZeroExtend(BECount, IntPtr);
const SCEV *NumBytesS = SE->getAddExpr(BECount, SE->getConstant(IntPtr, 1),
@@ -604,7 +604,7 @@ processLoopStoreOfLoopLoad(StoreInst *SI, unsigned StoreSize,
// The # stored bytes is (BECount+1)*Size. Expand the trip count out to
// pointer size if it isn't already.
- const Type *IntPtr = TD->getIntPtrType(SI->getContext());
+ Type *IntPtr = TD->getIntPtrType(SI->getContext());
BECount = SE->getTruncateOrZeroExtend(BECount, IntPtr);
const SCEV *NumBytesS = SE->getAddExpr(BECount, SE->getConstant(IntPtr, 1),
diff --git a/lib/Transforms/Scalar/LoopStrengthReduce.cpp b/lib/Transforms/Scalar/LoopStrengthReduce.cpp
index 509d026..e90b5bc 100644
--- a/lib/Transforms/Scalar/LoopStrengthReduce.cpp
+++ b/lib/Transforms/Scalar/LoopStrengthReduce.cpp
@@ -219,7 +219,7 @@ struct Formula {
void InitialMatch(const SCEV *S, Loop *L, ScalarEvolution &SE);
unsigned getNumRegs() const;
- const Type *getType() const;
+ Type *getType() const;
void DeleteBaseReg(const SCEV *&S);
@@ -319,7 +319,7 @@ unsigned Formula::getNumRegs() const {
/// getType - Return the type of this formula, if it has one, or null
/// otherwise. This type is meaningless except for the bit size.
-const Type *Formula::getType() const {
+Type *Formula::getType() const {
return !BaseRegs.empty() ? BaseRegs.front()->getType() :
ScaledReg ? ScaledReg->getType() :
AM.BaseGV ? AM.BaseGV->getType() :
@@ -397,7 +397,7 @@ void Formula::dump() const {
/// isAddRecSExtable - Return true if the given addrec can be sign-extended
/// without changing its value.
static bool isAddRecSExtable(const SCEVAddRecExpr *AR, ScalarEvolution &SE) {
- const Type *WideTy =
+ Type *WideTy =
IntegerType::get(SE.getContext(), SE.getTypeSizeInBits(AR->getType()) + 1);
return isa<SCEVAddRecExpr>(SE.getSignExtendExpr(AR, WideTy));
}
@@ -405,7 +405,7 @@ static bool isAddRecSExtable(const SCEVAddRecExpr *AR, ScalarEvolution &SE) {
/// isAddSExtable - Return true if the given add can be sign-extended
/// without changing its value.
static bool isAddSExtable(const SCEVAddExpr *A, ScalarEvolution &SE) {
- const Type *WideTy =
+ Type *WideTy =
IntegerType::get(SE.getContext(), SE.getTypeSizeInBits(A->getType()) + 1);
return isa<SCEVAddExpr>(SE.getSignExtendExpr(A, WideTy));
}
@@ -413,7 +413,7 @@ static bool isAddSExtable(const SCEVAddExpr *A, ScalarEvolution &SE) {
/// isMulSExtable - Return true if the given mul can be sign-extended
/// without changing its value.
static bool isMulSExtable(const SCEVMulExpr *M, ScalarEvolution &SE) {
- const Type *WideTy =
+ Type *WideTy =
IntegerType::get(SE.getContext(),
SE.getTypeSizeInBits(M->getType()) * M->getNumOperands());
return isa<SCEVMulExpr>(SE.getSignExtendExpr(M, WideTy));
@@ -594,8 +594,8 @@ static bool isAddressUse(Instruction *Inst, Value *OperandVal) {
}
/// getAccessType - Return the type of the memory being accessed.
-static const Type *getAccessType(const Instruction *Inst) {
- const Type *AccessTy = Inst->getType();
+static Type *getAccessType(const Instruction *Inst) {
+ Type *AccessTy = Inst->getType();
if (const StoreInst *SI = dyn_cast<StoreInst>(Inst))
AccessTy = SI->getOperand(0)->getType();
else if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst)) {
@@ -614,7 +614,7 @@ static const Type *getAccessType(const Instruction *Inst) {
// All pointers have the same requirements, so canonicalize them to an
// arbitrary pointer type to minimize variation.
- if (const PointerType *PTy = dyn_cast<PointerType>(AccessTy))
+ if (PointerType *PTy = dyn_cast<PointerType>(AccessTy))
AccessTy = PointerType::get(IntegerType::get(PTy->getContext(), 1),
PTy->getAddressSpace());
@@ -980,7 +980,7 @@ public:
};
KindType Kind;
- const Type *AccessTy;
+ Type *AccessTy;
SmallVector<int64_t, 8> Offsets;
int64_t MinOffset;
@@ -995,7 +995,7 @@ public:
/// this LSRUse. FindUseWithSimilarFormula can't consider uses with different
/// max fixup widths to be equivalent, because the narrower one may be relying
/// on the implicit truncation to truncate away bogus bits.
- const Type *WidestFixupType;
+ Type *WidestFixupType;
/// Formulae - A list of ways to build a value that can satisfy this user.
/// After the list is populated, one of these is selected heuristically and
@@ -1005,7 +1005,7 @@ public:
/// Regs - The set of register candidates used by all formulae in this LSRUse.
SmallPtrSet<const SCEV *, 4> Regs;
- LSRUse(KindType K, const Type *T) : Kind(K), AccessTy(T),
+ LSRUse(KindType K, Type *T) : Kind(K), AccessTy(T),
MinOffset(INT64_MAX),
MaxOffset(INT64_MIN),
AllFixupsOutsideLoop(true),
@@ -1127,7 +1127,7 @@ void LSRUse::dump() const {
/// be completely folded into the user instruction at isel time. This includes
/// address-mode folding and special icmp tricks.
static bool isLegalUse(const TargetLowering::AddrMode &AM,
- LSRUse::KindType Kind, const Type *AccessTy,
+ LSRUse::KindType Kind, Type *AccessTy,
const TargetLowering *TLI) {
switch (Kind) {
case LSRUse::Address:
@@ -1176,7 +1176,7 @@ static bool isLegalUse(const TargetLowering::AddrMode &AM,
static bool isLegalUse(TargetLowering::AddrMode AM,
int64_t MinOffset, int64_t MaxOffset,
- LSRUse::KindType Kind, const Type *AccessTy,
+ LSRUse::KindType Kind, Type *AccessTy,
const TargetLowering *TLI) {
// Check for overflow.
if (((int64_t)((uint64_t)AM.BaseOffs + MinOffset) > AM.BaseOffs) !=
@@ -1198,7 +1198,7 @@ static bool isLegalUse(TargetLowering::AddrMode AM,
static bool isAlwaysFoldable(int64_t BaseOffs,
GlobalValue *BaseGV,
bool HasBaseReg,
- LSRUse::KindType Kind, const Type *AccessTy,
+ LSRUse::KindType Kind, Type *AccessTy,
const TargetLowering *TLI) {
// Fast-path: zero is always foldable.
if (BaseOffs == 0 && !BaseGV) return true;
@@ -1224,7 +1224,7 @@ static bool isAlwaysFoldable(int64_t BaseOffs,
static bool isAlwaysFoldable(const SCEV *S,
int64_t MinOffset, int64_t MaxOffset,
bool HasBaseReg,
- LSRUse::KindType Kind, const Type *AccessTy,
+ LSRUse::KindType Kind, Type *AccessTy,
const TargetLowering *TLI,
ScalarEvolution &SE) {
// Fast-path: zero is always foldable.
@@ -1299,7 +1299,7 @@ class LSRInstance {
SmallSetVector<int64_t, 8> Factors;
/// Types - Interesting use types, to facilitate truncation reuse.
- SmallSetVector<const Type *, 4> Types;
+ SmallSetVector<Type *, 4> Types;
/// Fixups - The list of operands which are to be replaced.
SmallVector<LSRFixup, 16> Fixups;
@@ -1330,11 +1330,11 @@ class LSRInstance {
UseMapTy UseMap;
bool reconcileNewOffset(LSRUse &LU, int64_t NewOffset, bool HasBaseReg,
- LSRUse::KindType Kind, const Type *AccessTy);
+ LSRUse::KindType Kind, Type *AccessTy);
std::pair<size_t, int64_t> getUse(const SCEV *&Expr,
LSRUse::KindType Kind,
- const Type *AccessTy);
+ Type *AccessTy);
void DeleteUse(LSRUse &LU, size_t LUIdx);
@@ -1426,7 +1426,7 @@ void LSRInstance::OptimizeShadowIV() {
IVUsers::const_iterator CandidateUI = UI;
++UI;
Instruction *ShadowUse = CandidateUI->getUser();
- const Type *DestTy = NULL;
+ Type *DestTy = NULL;
/* If shadow use is a int->float cast then insert a second IV
to eliminate this cast.
@@ -1457,7 +1457,7 @@ void LSRInstance::OptimizeShadowIV() {
if (!PH) continue;
if (PH->getNumIncomingValues() != 2) continue;
- const Type *SrcTy = PH->getType();
+ Type *SrcTy = PH->getType();
int Mantissa = DestTy->getFPMantissaWidth();
if (Mantissa == -1) continue;
if ((int)SE.getTypeSizeInBits(SrcTy) > Mantissa)
@@ -1776,7 +1776,7 @@ LSRInstance::OptimizeLoopTermCond() {
if (!TLI)
goto decline_post_inc;
// Check for possible scaled-address reuse.
- const Type *AccessTy = getAccessType(UI->getUser());
+ Type *AccessTy = getAccessType(UI->getUser());
TargetLowering::AddrMode AM;
AM.Scale = C->getSExtValue();
if (TLI->isLegalAddressingMode(AM, AccessTy))
@@ -1840,10 +1840,10 @@ LSRInstance::OptimizeLoopTermCond() {
/// return true.
bool
LSRInstance::reconcileNewOffset(LSRUse &LU, int64_t NewOffset, bool HasBaseReg,
- LSRUse::KindType Kind, const Type *AccessTy) {
+ LSRUse::KindType Kind, Type *AccessTy) {
int64_t NewMinOffset = LU.MinOffset;
int64_t NewMaxOffset = LU.MaxOffset;
- const Type *NewAccessTy = AccessTy;
+ Type *NewAccessTy = AccessTy;
// Check for a mismatched kind. It's tempting to collapse mismatched kinds to
// something conservative, however this can pessimize in the case that one of
@@ -1882,7 +1882,7 @@ LSRInstance::reconcileNewOffset(LSRUse &LU, int64_t NewOffset, bool HasBaseReg,
/// Either reuse an existing use or create a new one, as needed.
std::pair<size_t, int64_t>
LSRInstance::getUse(const SCEV *&Expr,
- LSRUse::KindType Kind, const Type *AccessTy) {
+ LSRUse::KindType Kind, Type *AccessTy) {
const SCEV *Copy = Expr;
int64_t Offset = ExtractImmediate(Expr, SE);
@@ -2044,7 +2044,7 @@ void LSRInstance::CollectFixupsAndInitialFormulae() {
LF.PostIncLoops = UI->getPostIncLoops();
LSRUse::KindType Kind = LSRUse::Basic;
- const Type *AccessTy = 0;
+ Type *AccessTy = 0;
if (isAddressUse(LF.UserInst, LF.OperandValToReplace)) {
Kind = LSRUse::Address;
AccessTy = getAccessType(LF.UserInst);
@@ -2464,7 +2464,7 @@ void LSRInstance::GenerateICmpZeroScales(LSRUse &LU, unsigned LUIdx,
if (LU.Kind != LSRUse::ICmpZero) return;
// Determine the integer type for the base formula.
- const Type *IntTy = Base.getType();
+ Type *IntTy = Base.getType();
if (!IntTy) return;
if (SE.getTypeSizeInBits(IntTy) > 64) return;
@@ -2538,7 +2538,7 @@ void LSRInstance::GenerateICmpZeroScales(LSRUse &LU, unsigned LUIdx,
/// scaled-offset address modes, for example.
void LSRInstance::GenerateScales(LSRUse &LU, unsigned LUIdx, Formula Base) {
// Determine the integer type for the base formula.
- const Type *IntTy = Base.getType();
+ Type *IntTy = Base.getType();
if (!IntTy) return;
// If this Formula already has a scaled register, we can't add another one.
@@ -2598,13 +2598,13 @@ void LSRInstance::GenerateTruncates(LSRUse &LU, unsigned LUIdx, Formula Base) {
if (Base.AM.BaseGV) return;
// Determine the integer type for the base formula.
- const Type *DstTy = Base.getType();
+ Type *DstTy = Base.getType();
if (!DstTy) return;
DstTy = SE.getEffectiveSCEVType(DstTy);
- for (SmallSetVector<const Type *, 4>::const_iterator
+ for (SmallSetVector<Type *, 4>::const_iterator
I = Types.begin(), E = Types.end(); I != E; ++I) {
- const Type *SrcTy = *I;
+ Type *SrcTy = *I;
if (SrcTy != DstTy && TLI->isTruncateFree(SrcTy, DstTy)) {
Formula F = Base;
@@ -2741,7 +2741,7 @@ void LSRInstance::GenerateCrossUseConstantOffsets() {
int64_t Imm = WI.Imm;
const SCEV *OrigReg = WI.OrigReg;
- const Type *IntTy = SE.getEffectiveSCEVType(OrigReg->getType());
+ Type *IntTy = SE.getEffectiveSCEVType(OrigReg->getType());
const SCEV *NegImmS = SE.getSCEV(ConstantInt::get(IntTy, -(uint64_t)Imm));
unsigned BitWidth = SE.getTypeSizeInBits(IntTy);
@@ -3440,9 +3440,9 @@ Value *LSRInstance::Expand(const LSRFixup &LF,
Rewriter.setPostInc(LF.PostIncLoops);
// This is the type that the user actually needs.
- const Type *OpTy = LF.OperandValToReplace->getType();
+ Type *OpTy = LF.OperandValToReplace->getType();
// This will be the type that we'll initially expand to.
- const Type *Ty = F.getType();
+ Type *Ty = F.getType();
if (!Ty)
// No type known; just expand directly to the ultimate type.
Ty = OpTy;
@@ -3450,7 +3450,7 @@ Value *LSRInstance::Expand(const LSRFixup &LF,
// Expand directly to the ultimate type if it's the right size.
Ty = OpTy;
// This is the type to do integer arithmetic in.
- const Type *IntTy = SE.getEffectiveSCEVType(Ty);
+ Type *IntTy = SE.getEffectiveSCEVType(Ty);
// Build up a list of operands to add together to form the full base.
SmallVector<const SCEV *, 8> Ops;
@@ -3637,7 +3637,7 @@ void LSRInstance::RewriteForPHI(PHINode *PN,
Value *FullV = Expand(LF, F, BB->getTerminator(), Rewriter, DeadInsts);
// If this is reuse-by-noop-cast, insert the noop cast.
- const Type *OpTy = LF.OperandValToReplace->getType();
+ Type *OpTy = LF.OperandValToReplace->getType();
if (FullV->getType() != OpTy)
FullV =
CastInst::Create(CastInst::getCastOpcode(FullV, false,
@@ -3667,7 +3667,7 @@ void LSRInstance::Rewrite(const LSRFixup &LF,
Value *FullV = Expand(LF, F, LF.UserInst, Rewriter, DeadInsts);
// If this is reuse-by-noop-cast, insert the noop cast.
- const Type *OpTy = LF.OperandValToReplace->getType();
+ Type *OpTy = LF.OperandValToReplace->getType();
if (FullV->getType() != OpTy) {
Instruction *Cast =
CastInst::Create(CastInst::getCastOpcode(FullV, false, OpTy, false),
@@ -3793,7 +3793,7 @@ void LSRInstance::print_factors_and_types(raw_ostream &OS) const {
OS << '*' << *I;
}
- for (SmallSetVector<const Type *, 4>::const_iterator
+ for (SmallSetVector<Type *, 4>::const_iterator
I = Types.begin(), E = Types.end(); I != E; ++I) {
if (!First) OS << ", ";
First = false;
diff --git a/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
index 7ed3db6..ba5ee68 100644
--- a/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -54,7 +54,7 @@ static int64_t GetOffsetFromIndex(const GetElementPtrInst *GEP, unsigned Idx,
if (OpC->isZero()) continue; // No offset.
// Handle struct indices, which add their field offset to the pointer.
- if (const StructType *STy = dyn_cast<StructType>(*GTI)) {
+ if (StructType *STy = dyn_cast<StructType>(*GTI)) {
Offset += TD.getStructLayout(STy)->getElementOffset(OpC->getZExtValue());
continue;
}
@@ -448,7 +448,7 @@ Instruction *MemCpyOpt::tryMergingIntoMemset(Instruction *StartInst,
// Determine alignment
unsigned Alignment = Range.Alignment;
if (Alignment == 0) {
- const Type *EltType =
+ Type *EltType =
cast<PointerType>(StartPtr->getType())->getElementType();
Alignment = TD->getABITypeAlignment(EltType);
}
@@ -616,7 +616,7 @@ bool MemCpyOpt::performCallSlotOptzn(Instruction *cpy,
if (!A->hasStructRetAttr())
return false;
- const Type *StructTy = cast<PointerType>(A->getType())->getElementType();
+ Type *StructTy = cast<PointerType>(A->getType())->getElementType();
uint64_t destSize = TD->getTypeAllocSize(StructTy);
if (destSize < srcSize)
@@ -860,7 +860,7 @@ bool MemCpyOpt::processByValArgument(CallSite CS, unsigned ArgNo) {
// Find out what feeds this byval argument.
Value *ByValArg = CS.getArgument(ArgNo);
- const Type *ByValTy =cast<PointerType>(ByValArg->getType())->getElementType();
+ Type *ByValTy =cast<PointerType>(ByValArg->getType())->getElementType();
uint64_t ByValSize = TD->getTypeAllocSize(ByValTy);
MemDepResult DepInfo =
MD->getPointerDependencyFrom(AliasAnalysis::Location(ByValArg, ByValSize),
diff --git a/lib/Transforms/Scalar/ObjCARC.cpp b/lib/Transforms/Scalar/ObjCARC.cpp
index ee132d3..f2e5ff9 100644
--- a/lib/Transforms/Scalar/ObjCARC.cpp
+++ b/lib/Transforms/Scalar/ObjCARC.cpp
@@ -180,7 +180,7 @@ static bool IsPotentialUse(const Value *Op) {
Arg->hasStructRetAttr())
return false;
// Only consider values with pointer types, and not function pointers.
- const PointerType *Ty = dyn_cast<PointerType>(Op->getType());
+ PointerType *Ty = dyn_cast<PointerType>(Op->getType());
if (!Ty || isa<FunctionType>(Ty->getElementType()))
return false;
// Conservatively assume anything else is a potential use.
@@ -213,8 +213,8 @@ static InstructionClass GetFunctionClass(const Function *F) {
const Argument *A0 = AI++;
if (AI == AE)
// Argument is a pointer.
- if (const PointerType *PTy = dyn_cast<PointerType>(A0->getType())) {
- const Type *ETy = PTy->getElementType();
+ if (PointerType *PTy = dyn_cast<PointerType>(A0->getType())) {
+ Type *ETy = PTy->getElementType();
// Argument is i8*.
if (ETy->isIntegerTy(8))
return StringSwitch<InstructionClass>(F->getName())
@@ -234,7 +234,7 @@ static InstructionClass GetFunctionClass(const Function *F) {
.Default(IC_CallOrUser);
// Argument is i8**
- if (const PointerType *Pte = dyn_cast<PointerType>(ETy))
+ if (PointerType *Pte = dyn_cast<PointerType>(ETy))
if (Pte->getElementType()->isIntegerTy(8))
return StringSwitch<InstructionClass>(F->getName())
.Case("objc_loadWeakRetained", IC_LoadWeakRetained)
@@ -246,11 +246,11 @@ static InstructionClass GetFunctionClass(const Function *F) {
// Two arguments, first is i8**.
const Argument *A1 = AI++;
if (AI == AE)
- if (const PointerType *PTy = dyn_cast<PointerType>(A0->getType()))
- if (const PointerType *Pte = dyn_cast<PointerType>(PTy->getElementType()))
+ if (PointerType *PTy = dyn_cast<PointerType>(A0->getType()))
+ if (PointerType *Pte = dyn_cast<PointerType>(PTy->getElementType()))
if (Pte->getElementType()->isIntegerTy(8))
- if (const PointerType *PTy1 = dyn_cast<PointerType>(A1->getType())) {
- const Type *ETy1 = PTy1->getElementType();
+ if (PointerType *PTy1 = dyn_cast<PointerType>(A1->getType())) {
+ Type *ETy1 = PTy1->getElementType();
// Second argument is i8*
if (ETy1->isIntegerTy(8))
return StringSwitch<InstructionClass>(F->getName())
@@ -258,7 +258,7 @@ static InstructionClass GetFunctionClass(const Function *F) {
.Case("objc_initWeak", IC_InitWeak)
.Default(IC_CallOrUser);
// Second argument is i8**.
- if (const PointerType *Pte1 = dyn_cast<PointerType>(ETy1))
+ if (PointerType *Pte1 = dyn_cast<PointerType>(ETy1))
if (Pte1->getElementType()->isIntegerTy(8))
return StringSwitch<InstructionClass>(F->getName())
.Case("objc_moveWeak", IC_MoveWeak)
@@ -1501,7 +1501,7 @@ Constant *ObjCARCOpt::getRetainRVCallee(Module *M) {
Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
std::vector<Type *> Params;
Params.push_back(I8X);
- const FunctionType *FTy =
+ FunctionType *FTy =
FunctionType::get(I8X, Params, /*isVarArg=*/false);
AttrListPtr Attributes;
Attributes.addAttr(~0u, Attribute::NoUnwind);
@@ -1518,7 +1518,7 @@ Constant *ObjCARCOpt::getAutoreleaseRVCallee(Module *M) {
Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
std::vector<Type *> Params;
Params.push_back(I8X);
- const FunctionType *FTy =
+ FunctionType *FTy =
FunctionType::get(I8X, Params, /*isVarArg=*/false);
AttrListPtr Attributes;
Attributes.addAttr(~0u, Attribute::NoUnwind);
@@ -1953,7 +1953,7 @@ void ObjCARCOpt::OptimizeIndividualCalls(Function &F) {
case IC_DestroyWeak: {
CallInst *CI = cast<CallInst>(Inst);
if (isNullOrUndef(CI->getArgOperand(0))) {
- const Type *Ty = CI->getArgOperand(0)->getType();
+ Type *Ty = CI->getArgOperand(0)->getType();
new StoreInst(UndefValue::get(cast<PointerType>(Ty)->getElementType()),
Constant::getNullValue(Ty),
CI);
@@ -1968,7 +1968,7 @@ void ObjCARCOpt::OptimizeIndividualCalls(Function &F) {
CallInst *CI = cast<CallInst>(Inst);
if (isNullOrUndef(CI->getArgOperand(0)) ||
isNullOrUndef(CI->getArgOperand(1))) {
- const Type *Ty = CI->getArgOperand(0)->getType();
+ Type *Ty = CI->getArgOperand(0)->getType();
new StoreInst(UndefValue::get(cast<PointerType>(Ty)->getElementType()),
Constant::getNullValue(Ty),
CI);
@@ -2090,7 +2090,7 @@ void ObjCARCOpt::OptimizeIndividualCalls(Function &F) {
++NumPartialNoops;
// Clone the call into each predecessor that has a non-null value.
CallInst *CInst = cast<CallInst>(Inst);
- const Type *ParamTy = CInst->getArgOperand(0)->getType();
+ Type *ParamTy = CInst->getArgOperand(0)->getType();
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
Value *Incoming =
StripPointerCastsAndObjCCalls(PN->getIncomingValue(i));
@@ -2566,8 +2566,8 @@ void ObjCARCOpt::MoveCalls(Value *Arg,
MapVector<Value *, RRInfo> &Retains,
DenseMap<Value *, RRInfo> &Releases,
SmallVectorImpl<Instruction *> &DeadInsts) {
- const Type *ArgTy = Arg->getType();
- const Type *ParamTy =
+ Type *ArgTy = Arg->getType();
+ Type *ParamTy =
(RetainRVFunc ? RetainRVFunc :
RetainFunc ? RetainFunc :
RetainBlockFunc)->arg_begin()->getType();
@@ -3294,7 +3294,7 @@ Constant *ObjCARCContract::getRetainAutoreleaseCallee(Module *M) {
Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
std::vector<Type *> Params;
Params.push_back(I8X);
- const FunctionType *FTy =
+ FunctionType *FTy =
FunctionType::get(I8X, Params, /*isVarArg=*/false);
AttrListPtr Attributes;
Attributes.addAttr(~0u, Attribute::NoUnwind);
@@ -3310,7 +3310,7 @@ Constant *ObjCARCContract::getRetainAutoreleaseRVCallee(Module *M) {
Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
std::vector<Type *> Params;
Params.push_back(I8X);
- const FunctionType *FTy =
+ FunctionType *FTy =
FunctionType::get(I8X, Params, /*isVarArg=*/false);
AttrListPtr Attributes;
Attributes.addAttr(~0u, Attribute::NoUnwind);
@@ -3411,8 +3411,8 @@ void ObjCARCContract::ContractRelease(Instruction *Release,
++NumStoreStrongs;
LLVMContext &C = Release->getContext();
- const Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
- const Type *I8XX = PointerType::getUnqual(I8X);
+ Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
+ Type *I8XX = PointerType::getUnqual(I8X);
Value *Args[] = { Load->getPointerOperand(), New };
if (Args[0]->getType() != I8XX)
@@ -3548,7 +3548,7 @@ bool ObjCARCContract::runOnFunction(Function &F) {
if (Inst != UserInst && DT->dominates(Inst, UserInst)) {
Changed = true;
Instruction *Replacement = Inst;
- const Type *UseTy = U.get()->getType();
+ Type *UseTy = U.get()->getType();
if (PHINode *PHI = dyn_cast<PHINode>(UserInst)) {
// For PHI nodes, insert the bitcast in the predecessor block.
unsigned ValNo =
diff --git a/lib/Transforms/Scalar/SCCP.cpp b/lib/Transforms/Scalar/SCCP.cpp
index 083412e..5b12c92 100644
--- a/lib/Transforms/Scalar/SCCP.cpp
+++ b/lib/Transforms/Scalar/SCCP.cpp
@@ -241,7 +241,7 @@ public:
/// this method must be called.
void AddTrackedFunction(Function *F) {
// Add an entry, F -> undef.
- if (const StructType *STy = dyn_cast<StructType>(F->getReturnType())) {
+ if (StructType *STy = dyn_cast<StructType>(F->getReturnType())) {
MRVFunctionsTracked.insert(F);
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
TrackedMultipleRetVals.insert(std::make_pair(std::make_pair(F, i),
@@ -302,7 +302,7 @@ public:
/// markAnythingOverdefined - Mark the specified value overdefined. This
/// works with both scalars and structs.
void markAnythingOverdefined(Value *V) {
- if (const StructType *STy = dyn_cast<StructType>(V->getType()))
+ if (StructType *STy = dyn_cast<StructType>(V->getType()))
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
markOverdefined(getStructValueState(V, i), V);
else
@@ -417,7 +417,7 @@ private:
else if (ConstantStruct *CS = dyn_cast<ConstantStruct>(C))
LV.markConstant(CS->getOperand(i)); // Constants are constant.
else if (isa<ConstantAggregateZero>(C)) {
- const Type *FieldTy = cast<StructType>(V->getType())->getElementType(i);
+ Type *FieldTy = cast<StructType>(V->getType())->getElementType(i);
LV.markConstant(Constant::getNullValue(FieldTy));
} else
LV.markOverdefined(); // Unknown sort of constant.
@@ -772,7 +772,7 @@ void SCCPSolver::visitReturnInst(ReturnInst &I) {
// Handle functions that return multiple values.
if (!TrackedMultipleRetVals.empty()) {
- if (const StructType *STy = dyn_cast<StructType>(ResultOp->getType()))
+ if (StructType *STy = dyn_cast<StructType>(ResultOp->getType()))
if (MRVFunctionsTracked.count(F))
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
mergeInValue(TrackedMultipleRetVals[std::make_pair(F, i)], F,
@@ -825,7 +825,7 @@ void SCCPSolver::visitExtractValueInst(ExtractValueInst &EVI) {
}
void SCCPSolver::visitInsertValueInst(InsertValueInst &IVI) {
- const StructType *STy = dyn_cast<StructType>(IVI.getType());
+ StructType *STy = dyn_cast<StructType>(IVI.getType());
if (STy == 0)
return markOverdefined(&IVI);
@@ -925,7 +925,7 @@ void SCCPSolver::visitBinaryOperator(Instruction &I) {
// Could annihilate value.
if (I.getOpcode() == Instruction::And)
markConstant(IV, &I, Constant::getNullValue(I.getType()));
- else if (const VectorType *PT = dyn_cast<VectorType>(I.getType()))
+ else if (VectorType *PT = dyn_cast<VectorType>(I.getType()))
markConstant(IV, &I, Constant::getAllOnesValue(PT));
else
markConstant(IV, &I,
@@ -1278,7 +1278,7 @@ CallOverdefined:
// If we can constant fold this, mark the result of the call as a
// constant.
- if (Constant *C = ConstantFoldCall(F, Operands.data(), Operands.size()))
+ if (Constant *C = ConstantFoldCall(F, Operands))
return markConstant(I, C);
}
@@ -1303,7 +1303,7 @@ CallOverdefined:
continue;
}
- if (const StructType *STy = dyn_cast<StructType>(AI->getType())) {
+ if (StructType *STy = dyn_cast<StructType>(AI->getType())) {
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
LatticeVal CallArg = getStructValueState(*CAI, i);
mergeInValue(getStructValueState(AI, i), AI, CallArg);
@@ -1315,7 +1315,7 @@ CallOverdefined:
}
// If this is a single/zero retval case, see if we're tracking the function.
- if (const StructType *STy = dyn_cast<StructType>(F->getReturnType())) {
+ if (StructType *STy = dyn_cast<StructType>(F->getReturnType())) {
if (!MRVFunctionsTracked.count(F))
goto CallOverdefined; // Not tracking this callee.
@@ -1419,7 +1419,7 @@ bool SCCPSolver::ResolvedUndefsIn(Function &F) {
// Look for instructions which produce undef values.
if (I->getType()->isVoidTy()) continue;
- if (const StructType *STy = dyn_cast<StructType>(I->getType())) {
+ if (StructType *STy = dyn_cast<StructType>(I->getType())) {
// Only a few things that can be structs matter for undef. Just send
// all their results to overdefined. We could be more precise than this
// but it isn't worth bothering.
@@ -1457,7 +1457,7 @@ bool SCCPSolver::ResolvedUndefsIn(Function &F) {
// If this is an instructions whose result is defined even if the input is
// not fully defined, propagate the information.
- const Type *ITy = I->getType();
+ Type *ITy = I->getType();
switch (I->getOpcode()) {
default: break; // Leave the instruction as an undef.
case Instruction::ZExt:
diff --git a/lib/Transforms/Scalar/ScalarReplAggregates.cpp b/lib/Transforms/Scalar/ScalarReplAggregates.cpp
index 7d6349c..fbf3092 100644
--- a/lib/Transforms/Scalar/ScalarReplAggregates.cpp
+++ b/lib/Transforms/Scalar/ScalarReplAggregates.cpp
@@ -129,11 +129,11 @@ namespace {
AllocaInfo &Info);
void isSafeGEP(GetElementPtrInst *GEPI, uint64_t &Offset, AllocaInfo &Info);
void isSafeMemAccess(uint64_t Offset, uint64_t MemSize,
- const Type *MemOpType, bool isStore, AllocaInfo &Info,
+ Type *MemOpType, bool isStore, AllocaInfo &Info,
Instruction *TheAccess, bool AllowWholeAccess);
- bool TypeHasComponent(const Type *T, uint64_t Offset, uint64_t Size);
- uint64_t FindElementAndOffset(const Type *&T, uint64_t &Offset,
- const Type *&IdxTy);
+ bool TypeHasComponent(Type *T, uint64_t Offset, uint64_t Size);
+ uint64_t FindElementAndOffset(Type *&T, uint64_t &Offset,
+ Type *&IdxTy);
void DoScalarReplacement(AllocaInst *AI,
std::vector<AllocaInst*> &WorkList);
@@ -253,7 +253,7 @@ class ConvertToScalarInfo {
/// VectorTy - This tracks the type that we should promote the vector to if
/// it is possible to turn it into a vector. This starts out null, and if it
/// isn't possible to turn into a vector type, it gets set to VoidTy.
- const VectorType *VectorTy;
+ VectorType *VectorTy;
/// HadNonMemTransferAccess - True if there is at least one access to the
/// alloca that is not a MemTransferInst. We don't want to turn structs into
@@ -269,11 +269,11 @@ public:
private:
bool CanConvertToScalar(Value *V, uint64_t Offset);
- void MergeInTypeForLoadOrStore(const Type *In, uint64_t Offset);
- bool MergeInVectorType(const VectorType *VInTy, uint64_t Offset);
+ void MergeInTypeForLoadOrStore(Type *In, uint64_t Offset);
+ bool MergeInVectorType(VectorType *VInTy, uint64_t Offset);
void ConvertUsesToScalar(Value *Ptr, AllocaInst *NewAI, uint64_t Offset);
- Value *ConvertScalar_ExtractValue(Value *NV, const Type *ToType,
+ Value *ConvertScalar_ExtractValue(Value *NV, Type *ToType,
uint64_t Offset, IRBuilder<> &Builder);
Value *ConvertScalar_InsertValue(Value *StoredVal, Value *ExistingVal,
uint64_t Offset, IRBuilder<> &Builder);
@@ -306,7 +306,7 @@ AllocaInst *ConvertToScalarInfo::TryConvert(AllocaInst *AI) {
// random stuff that doesn't use vectors (e.g. <9 x double>) because then
// we just get a lot of insert/extracts. If at least one vector is
// involved, then we probably really do have a union of vector/array.
- const Type *NewTy;
+ Type *NewTy;
if (ScalarKind == Vector) {
assert(VectorTy && "Missing type for vector scalar.");
DEBUG(dbgs() << "CONVERT TO VECTOR: " << *AI << "\n TYPE = "
@@ -344,7 +344,7 @@ AllocaInst *ConvertToScalarInfo::TryConvert(AllocaInst *AI) {
/// large) integer type with extract and insert operations where the loads
/// and stores would mutate the memory. We mark this by setting VectorTy
/// to VoidTy.
-void ConvertToScalarInfo::MergeInTypeForLoadOrStore(const Type *In,
+void ConvertToScalarInfo::MergeInTypeForLoadOrStore(Type *In,
uint64_t Offset) {
// If we already decided to turn this into a blob of integer memory, there is
// nothing to be done.
@@ -355,7 +355,7 @@ void ConvertToScalarInfo::MergeInTypeForLoadOrStore(const Type *In,
// If the In type is a vector that is the same size as the alloca, see if it
// matches the existing VecTy.
- if (const VectorType *VInTy = dyn_cast<VectorType>(In)) {
+ if (VectorType *VInTy = dyn_cast<VectorType>(In)) {
if (MergeInVectorType(VInTy, Offset))
return;
} else if (In->isFloatTy() || In->isDoubleTy() ||
@@ -395,7 +395,7 @@ void ConvertToScalarInfo::MergeInTypeForLoadOrStore(const Type *In,
/// MergeInVectorType - Handles the vector case of MergeInTypeForLoadOrStore,
/// returning true if the type was successfully merged and false otherwise.
-bool ConvertToScalarInfo::MergeInVectorType(const VectorType *VInTy,
+bool ConvertToScalarInfo::MergeInVectorType(VectorType *VInTy,
uint64_t Offset) {
// TODO: Support nonzero offsets?
if (Offset != 0)
@@ -422,8 +422,8 @@ bool ConvertToScalarInfo::MergeInVectorType(const VectorType *VInTy,
return true;
}
- const Type *ElementTy = VectorTy->getElementType();
- const Type *InElementTy = VInTy->getElementType();
+ Type *ElementTy = VectorTy->getElementType();
+ Type *InElementTy = VInTy->getElementType();
// Do not allow mixed integer and floating-point accesses from vectors of
// different sizes.
@@ -516,7 +516,7 @@ bool ConvertToScalarInfo::CanConvertToScalar(Value *V, uint64_t Offset) {
// Compute the offset that this GEP adds to the pointer.
SmallVector<Value*, 8> Indices(GEP->op_begin()+1, GEP->op_end());
uint64_t GEPOffset = TD.getIndexedOffset(GEP->getPointerOperandType(),
- &Indices[0], Indices.size());
+ Indices);
// See if all uses can be converted.
if (!CanConvertToScalar(GEP, Offset+GEPOffset))
return false;
@@ -589,7 +589,7 @@ void ConvertToScalarInfo::ConvertUsesToScalar(Value *Ptr, AllocaInst *NewAI,
// Compute the offset that this GEP adds to the pointer.
SmallVector<Value*, 8> Indices(GEP->op_begin()+1, GEP->op_end());
uint64_t GEPOffset = TD.getIndexedOffset(GEP->getPointerOperandType(),
- &Indices[0], Indices.size());
+ Indices);
ConvertUsesToScalar(GEP, NewAI, Offset+GEPOffset*8);
GEP->eraseFromParent();
continue;
@@ -668,8 +668,8 @@ void ConvertToScalarInfo::ConvertUsesToScalar(Value *Ptr, AllocaInst *NewAI,
// pointer (bitcasted), then a store to our new alloca.
assert(MTI->getRawDest() == Ptr && "Neither use is of pointer?");
Value *SrcPtr = MTI->getSource();
- const PointerType* SPTy = cast<PointerType>(SrcPtr->getType());
- const PointerType* AIPTy = cast<PointerType>(NewAI->getType());
+ PointerType* SPTy = cast<PointerType>(SrcPtr->getType());
+ PointerType* AIPTy = cast<PointerType>(NewAI->getType());
if (SPTy->getAddressSpace() != AIPTy->getAddressSpace()) {
AIPTy = PointerType::get(AIPTy->getElementType(),
SPTy->getAddressSpace());
@@ -685,8 +685,8 @@ void ConvertToScalarInfo::ConvertUsesToScalar(Value *Ptr, AllocaInst *NewAI,
assert(MTI->getRawSource() == Ptr && "Neither use is of pointer?");
LoadInst *SrcVal = Builder.CreateLoad(NewAI, "srcval");
- const PointerType* DPTy = cast<PointerType>(MTI->getDest()->getType());
- const PointerType* AIPTy = cast<PointerType>(NewAI->getType());
+ PointerType* DPTy = cast<PointerType>(MTI->getDest()->getType());
+ PointerType* AIPTy = cast<PointerType>(NewAI->getType());
if (DPTy->getAddressSpace() != AIPTy->getAddressSpace()) {
AIPTy = PointerType::get(AIPTy->getElementType(),
DPTy->getAddressSpace());
@@ -711,7 +711,7 @@ void ConvertToScalarInfo::ConvertUsesToScalar(Value *Ptr, AllocaInst *NewAI,
/// access of an alloca. The input types must be integer or floating-point
/// scalar or vector types, and the resulting type is an integer, float or
/// double.
-static const Type *getScaledElementType(const Type *Ty1, const Type *Ty2,
+static Type *getScaledElementType(Type *Ty1, Type *Ty2,
unsigned NewBitWidth) {
bool IsFP1 = Ty1->isFloatingPointTy() ||
(Ty1->isVectorTy() &&
@@ -737,11 +737,11 @@ static const Type *getScaledElementType(const Type *Ty1, const Type *Ty2,
/// CreateShuffleVectorCast - Creates a shuffle vector to convert one vector
/// to another vector of the same element type which has the same allocation
/// size but different primitive sizes (e.g. <3 x i32> and <4 x i32>).
-static Value *CreateShuffleVectorCast(Value *FromVal, const Type *ToType,
+static Value *CreateShuffleVectorCast(Value *FromVal, Type *ToType,
IRBuilder<> &Builder) {
- const Type *FromType = FromVal->getType();
- const VectorType *FromVTy = cast<VectorType>(FromType);
- const VectorType *ToVTy = cast<VectorType>(ToType);
+ Type *FromType = FromVal->getType();
+ VectorType *FromVTy = cast<VectorType>(FromType);
+ VectorType *ToVTy = cast<VectorType>(ToType);
assert((ToVTy->getElementType() == FromVTy->getElementType()) &&
"Vectors must have the same element type");
Value *UnV = UndefValue::get(FromType);
@@ -749,7 +749,7 @@ static Value *CreateShuffleVectorCast(Value *FromVal, const Type *ToType,
unsigned numEltsTo = ToVTy->getNumElements();
SmallVector<Constant*, 3> Args;
- const Type* Int32Ty = Builder.getInt32Ty();
+ Type* Int32Ty = Builder.getInt32Ty();
unsigned minNumElts = std::min(numEltsFrom, numEltsTo);
unsigned i;
for (i=0; i != minNumElts; ++i)
@@ -775,16 +775,16 @@ static Value *CreateShuffleVectorCast(Value *FromVal, const Type *ToType,
/// Offset is an offset from the original alloca, in bits that need to be
/// shifted to the right.
Value *ConvertToScalarInfo::
-ConvertScalar_ExtractValue(Value *FromVal, const Type *ToType,
+ConvertScalar_ExtractValue(Value *FromVal, Type *ToType,
uint64_t Offset, IRBuilder<> &Builder) {
// If the load is of the whole new alloca, no conversion is needed.
- const Type *FromType = FromVal->getType();
+ Type *FromType = FromVal->getType();
if (FromType == ToType && Offset == 0)
return FromVal;
// If the result alloca is a vector type, this is either an element
// access or a bitcast to another vector type of the same size.
- if (const VectorType *VTy = dyn_cast<VectorType>(FromType)) {
+ if (VectorType *VTy = dyn_cast<VectorType>(FromType)) {
unsigned FromTypeSize = TD.getTypeAllocSize(FromType);
unsigned ToTypeSize = TD.getTypeAllocSize(ToType);
if (FromTypeSize == ToTypeSize) {
@@ -803,12 +803,12 @@ ConvertScalar_ExtractValue(Value *FromVal, const Type *ToType,
assert(!(ToType->isVectorTy() && Offset != 0) && "Can't extract a value "
"of a smaller vector type at a nonzero offset.");
- const Type *CastElementTy = getScaledElementType(FromType, ToType,
+ Type *CastElementTy = getScaledElementType(FromType, ToType,
ToTypeSize * 8);
unsigned NumCastVectorElements = FromTypeSize / ToTypeSize;
LLVMContext &Context = FromVal->getContext();
- const Type *CastTy = VectorType::get(CastElementTy,
+ Type *CastTy = VectorType::get(CastElementTy,
NumCastVectorElements);
Value *Cast = Builder.CreateBitCast(FromVal, CastTy, "tmp");
@@ -837,7 +837,7 @@ ConvertScalar_ExtractValue(Value *FromVal, const Type *ToType,
// If ToType is a first class aggregate, extract out each of the pieces and
// use insertvalue's to form the FCA.
- if (const StructType *ST = dyn_cast<StructType>(ToType)) {
+ if (StructType *ST = dyn_cast<StructType>(ToType)) {
const StructLayout &Layout = *TD.getStructLayout(ST);
Value *Res = UndefValue::get(ST);
for (unsigned i = 0, e = ST->getNumElements(); i != e; ++i) {
@@ -849,7 +849,7 @@ ConvertScalar_ExtractValue(Value *FromVal, const Type *ToType,
return Res;
}
- if (const ArrayType *AT = dyn_cast<ArrayType>(ToType)) {
+ if (ArrayType *AT = dyn_cast<ArrayType>(ToType)) {
uint64_t EltSize = TD.getTypeAllocSizeInBits(AT->getElementType());
Value *Res = UndefValue::get(AT);
for (unsigned i = 0, e = AT->getNumElements(); i != e; ++i) {
@@ -861,7 +861,7 @@ ConvertScalar_ExtractValue(Value *FromVal, const Type *ToType,
}
// Otherwise, this must be a union that was converted to an integer value.
- const IntegerType *NTy = cast<IntegerType>(FromVal->getType());
+ IntegerType *NTy = cast<IntegerType>(FromVal->getType());
// If this is a big-endian system and the load is narrower than the
// full alloca type, we need to do a shift to get the right bits.
@@ -927,10 +927,10 @@ ConvertScalar_InsertValue(Value *SV, Value *Old,
uint64_t Offset, IRBuilder<> &Builder) {
// Convert the stored type to the actual type, shift it left to insert
// then 'or' into place.
- const Type *AllocaType = Old->getType();
+ Type *AllocaType = Old->getType();
LLVMContext &Context = Old->getContext();
- if (const VectorType *VTy = dyn_cast<VectorType>(AllocaType)) {
+ if (VectorType *VTy = dyn_cast<VectorType>(AllocaType)) {
uint64_t VecSize = TD.getTypeAllocSizeInBits(VTy);
uint64_t ValSize = TD.getTypeAllocSizeInBits(SV->getType());
@@ -952,12 +952,12 @@ ConvertScalar_InsertValue(Value *SV, Value *Old,
assert(!(SV->getType()->isVectorTy() && Offset != 0) && "Can't insert a "
"value of a smaller vector type at a nonzero offset.");
- const Type *CastElementTy = getScaledElementType(VTy, SV->getType(),
+ Type *CastElementTy = getScaledElementType(VTy, SV->getType(),
ValSize);
unsigned NumCastVectorElements = VecSize / ValSize;
LLVMContext &Context = SV->getContext();
- const Type *OldCastTy = VectorType::get(CastElementTy,
+ Type *OldCastTy = VectorType::get(CastElementTy,
NumCastVectorElements);
Value *OldCast = Builder.CreateBitCast(Old, OldCastTy, "tmp");
@@ -982,7 +982,7 @@ ConvertScalar_InsertValue(Value *SV, Value *Old,
}
// If SV is a first-class aggregate value, insert each value recursively.
- if (const StructType *ST = dyn_cast<StructType>(SV->getType())) {
+ if (StructType *ST = dyn_cast<StructType>(SV->getType())) {
const StructLayout &Layout = *TD.getStructLayout(ST);
for (unsigned i = 0, e = ST->getNumElements(); i != e; ++i) {
Value *Elt = Builder.CreateExtractValue(SV, i, "tmp");
@@ -993,7 +993,7 @@ ConvertScalar_InsertValue(Value *SV, Value *Old,
return Old;
}
- if (const ArrayType *AT = dyn_cast<ArrayType>(SV->getType())) {
+ if (ArrayType *AT = dyn_cast<ArrayType>(SV->getType())) {
uint64_t EltSize = TD.getTypeAllocSizeInBits(AT->getElementType());
for (unsigned i = 0, e = AT->getNumElements(); i != e; ++i) {
Value *Elt = Builder.CreateExtractValue(SV, i, "tmp");
@@ -1393,7 +1393,7 @@ static bool tryToMakeAllocaBePromotable(AllocaInst *AI, const TargetData *TD) {
continue;
}
- const Type *LoadTy = cast<PointerType>(PN->getType())->getElementType();
+ Type *LoadTy = cast<PointerType>(PN->getType())->getElementType();
PHINode *NewPN = PHINode::Create(LoadTy, PN->getNumIncomingValues(),
PN->getName()+".ld", PN);
@@ -1483,13 +1483,13 @@ bool SROA::performPromotion(Function &F) {
/// ShouldAttemptScalarRepl - Decide if an alloca is a good candidate for
/// SROA. It must be a struct or array type with a small number of elements.
static bool ShouldAttemptScalarRepl(AllocaInst *AI) {
- const Type *T = AI->getAllocatedType();
+ Type *T = AI->getAllocatedType();
// Do not promote any struct into more than 32 separate vars.
- if (const StructType *ST = dyn_cast<StructType>(T))
+ if (StructType *ST = dyn_cast<StructType>(T))
return ST->getNumElements() <= 32;
// Arrays are much less likely to be safe for SROA; only consider
// them if they are very small.
- if (const ArrayType *AT = dyn_cast<ArrayType>(T))
+ if (ArrayType *AT = dyn_cast<ArrayType>(T))
return AT->getNumElements() <= 8;
return false;
}
@@ -1594,7 +1594,7 @@ void SROA::DoScalarReplacement(AllocaInst *AI,
std::vector<AllocaInst*> &WorkList) {
DEBUG(dbgs() << "Found inst to SROA: " << *AI << '\n');
SmallVector<AllocaInst*, 32> ElementAllocas;
- if (const StructType *ST = dyn_cast<StructType>(AI->getAllocatedType())) {
+ if (StructType *ST = dyn_cast<StructType>(AI->getAllocatedType())) {
ElementAllocas.reserve(ST->getNumContainedTypes());
for (unsigned i = 0, e = ST->getNumContainedTypes(); i != e; ++i) {
AllocaInst *NA = new AllocaInst(ST->getContainedType(i), 0,
@@ -1604,9 +1604,9 @@ void SROA::DoScalarReplacement(AllocaInst *AI,
WorkList.push_back(NA); // Add to worklist for recursive processing
}
} else {
- const ArrayType *AT = cast<ArrayType>(AI->getAllocatedType());
+ ArrayType *AT = cast<ArrayType>(AI->getAllocatedType());
ElementAllocas.reserve(AT->getNumElements());
- const Type *ElTy = AT->getElementType();
+ Type *ElTy = AT->getElementType();
for (unsigned i = 0, e = AT->getNumElements(); i != e; ++i) {
AllocaInst *NA = new AllocaInst(ElTy, 0, AI->getAlignment(),
AI->getName() + "." + Twine(i), AI);
@@ -1672,7 +1672,7 @@ void SROA::isSafeForScalarRepl(Instruction *I, uint64_t Offset,
} else if (LoadInst *LI = dyn_cast<LoadInst>(User)) {
if (LI->isVolatile())
return MarkUnsafe(Info, User);
- const Type *LIType = LI->getType();
+ Type *LIType = LI->getType();
isSafeMemAccess(Offset, TD->getTypeAllocSize(LIType),
LIType, false, Info, LI, true /*AllowWholeAccess*/);
Info.hasALoadOrStore = true;
@@ -1682,7 +1682,7 @@ void SROA::isSafeForScalarRepl(Instruction *I, uint64_t Offset,
if (SI->isVolatile() || SI->getOperand(0) == I)
return MarkUnsafe(Info, User);
- const Type *SIType = SI->getOperand(0)->getType();
+ Type *SIType = SI->getOperand(0)->getType();
isSafeMemAccess(Offset, TD->getTypeAllocSize(SIType),
SIType, true, Info, SI, true /*AllowWholeAccess*/);
Info.hasALoadOrStore = true;
@@ -1727,7 +1727,7 @@ void SROA::isSafePHISelectUseForScalarRepl(Instruction *I, uint64_t Offset,
} else if (LoadInst *LI = dyn_cast<LoadInst>(User)) {
if (LI->isVolatile())
return MarkUnsafe(Info, User);
- const Type *LIType = LI->getType();
+ Type *LIType = LI->getType();
isSafeMemAccess(Offset, TD->getTypeAllocSize(LIType),
LIType, false, Info, LI, false /*AllowWholeAccess*/);
Info.hasALoadOrStore = true;
@@ -1737,7 +1737,7 @@ void SROA::isSafePHISelectUseForScalarRepl(Instruction *I, uint64_t Offset,
if (SI->isVolatile() || SI->getOperand(0) == I)
return MarkUnsafe(Info, User);
- const Type *SIType = SI->getOperand(0)->getType();
+ Type *SIType = SI->getOperand(0)->getType();
isSafeMemAccess(Offset, TD->getTypeAllocSize(SIType),
SIType, true, Info, SI, false /*AllowWholeAccess*/);
Info.hasALoadOrStore = true;
@@ -1776,8 +1776,7 @@ void SROA::isSafeGEP(GetElementPtrInst *GEPI,
// Compute the offset due to this GEP and check if the alloca has a
// component element at that offset.
SmallVector<Value*, 8> Indices(GEPI->op_begin() + 1, GEPI->op_end());
- Offset += TD->getIndexedOffset(GEPI->getPointerOperandType(),
- &Indices[0], Indices.size());
+ Offset += TD->getIndexedOffset(GEPI->getPointerOperandType(), Indices);
if (!TypeHasComponent(Info.AI->getAllocatedType(), Offset, 0))
MarkUnsafe(Info, GEPI);
}
@@ -1786,14 +1785,14 @@ void SROA::isSafeGEP(GetElementPtrInst *GEPI,
/// elements of the same type (which is always true for arrays). If so,
/// return true with NumElts and EltTy set to the number of elements and the
/// element type, respectively.
-static bool isHomogeneousAggregate(const Type *T, unsigned &NumElts,
- const Type *&EltTy) {
- if (const ArrayType *AT = dyn_cast<ArrayType>(T)) {
+static bool isHomogeneousAggregate(Type *T, unsigned &NumElts,
+ Type *&EltTy) {
+ if (ArrayType *AT = dyn_cast<ArrayType>(T)) {
NumElts = AT->getNumElements();
EltTy = (NumElts == 0 ? 0 : AT->getElementType());
return true;
}
- if (const StructType *ST = dyn_cast<StructType>(T)) {
+ if (StructType *ST = dyn_cast<StructType>(T)) {
NumElts = ST->getNumContainedTypes();
EltTy = (NumElts == 0 ? 0 : ST->getContainedType(0));
for (unsigned n = 1; n < NumElts; ++n) {
@@ -1807,12 +1806,12 @@ static bool isHomogeneousAggregate(const Type *T, unsigned &NumElts,
/// isCompatibleAggregate - Check if T1 and T2 are either the same type or are
/// "homogeneous" aggregates with the same element type and number of elements.
-static bool isCompatibleAggregate(const Type *T1, const Type *T2) {
+static bool isCompatibleAggregate(Type *T1, Type *T2) {
if (T1 == T2)
return true;
unsigned NumElts1, NumElts2;
- const Type *EltTy1, *EltTy2;
+ Type *EltTy1, *EltTy2;
if (isHomogeneousAggregate(T1, NumElts1, EltTy1) &&
isHomogeneousAggregate(T2, NumElts2, EltTy2) &&
NumElts1 == NumElts2 &&
@@ -1830,7 +1829,7 @@ static bool isCompatibleAggregate(const Type *T1, const Type *T2) {
/// If AllowWholeAccess is true, then this allows uses of the entire alloca as a
/// unit. If false, it only allows accesses known to be in a single element.
void SROA::isSafeMemAccess(uint64_t Offset, uint64_t MemSize,
- const Type *MemOpType, bool isStore,
+ Type *MemOpType, bool isStore,
AllocaInfo &Info, Instruction *TheAccess,
bool AllowWholeAccess) {
// Check if this is a load/store of the entire alloca.
@@ -1857,7 +1856,7 @@ void SROA::isSafeMemAccess(uint64_t Offset, uint64_t MemSize,
}
}
// Check if the offset/size correspond to a component within the alloca type.
- const Type *T = Info.AI->getAllocatedType();
+ Type *T = Info.AI->getAllocatedType();
if (TypeHasComponent(T, Offset, MemSize)) {
Info.hasSubelementAccess = true;
return;
@@ -1868,16 +1867,16 @@ void SROA::isSafeMemAccess(uint64_t Offset, uint64_t MemSize,
/// TypeHasComponent - Return true if T has a component type with the
/// specified offset and size. If Size is zero, do not check the size.
-bool SROA::TypeHasComponent(const Type *T, uint64_t Offset, uint64_t Size) {
- const Type *EltTy;
+bool SROA::TypeHasComponent(Type *T, uint64_t Offset, uint64_t Size) {
+ Type *EltTy;
uint64_t EltSize;
- if (const StructType *ST = dyn_cast<StructType>(T)) {
+ if (StructType *ST = dyn_cast<StructType>(T)) {
const StructLayout *Layout = TD->getStructLayout(ST);
unsigned EltIdx = Layout->getElementContainingOffset(Offset);
EltTy = ST->getContainedType(EltIdx);
EltSize = TD->getTypeAllocSize(EltTy);
Offset -= Layout->getElementOffset(EltIdx);
- } else if (const ArrayType *AT = dyn_cast<ArrayType>(T)) {
+ } else if (ArrayType *AT = dyn_cast<ArrayType>(T)) {
EltTy = AT->getElementType();
EltSize = TD->getTypeAllocSize(EltTy);
if (Offset >= AT->getNumElements() * EltSize)
@@ -1926,7 +1925,7 @@ void SROA::RewriteForScalarRepl(Instruction *I, AllocaInst *AI, uint64_t Offset,
}
if (LoadInst *LI = dyn_cast<LoadInst>(User)) {
- const Type *LIType = LI->getType();
+ Type *LIType = LI->getType();
if (isCompatibleAggregate(LIType, AI->getAllocatedType())) {
// Replace:
@@ -1956,7 +1955,7 @@ void SROA::RewriteForScalarRepl(Instruction *I, AllocaInst *AI, uint64_t Offset,
if (StoreInst *SI = dyn_cast<StoreInst>(User)) {
Value *Val = SI->getOperand(0);
- const Type *SIType = Val->getType();
+ Type *SIType = Val->getType();
if (isCompatibleAggregate(SIType, AI->getAllocatedType())) {
// Replace:
// store { i32, i32 } %val, { i32, i32 }* %alloc
@@ -2026,10 +2025,10 @@ void SROA::RewriteBitCast(BitCastInst *BC, AllocaInst *AI, uint64_t Offset,
/// Sets T to the type of the element and Offset to the offset within that
/// element. IdxTy is set to the type of the index result to be used in a
/// GEP instruction.
-uint64_t SROA::FindElementAndOffset(const Type *&T, uint64_t &Offset,
- const Type *&IdxTy) {
+uint64_t SROA::FindElementAndOffset(Type *&T, uint64_t &Offset,
+ Type *&IdxTy) {
uint64_t Idx = 0;
- if (const StructType *ST = dyn_cast<StructType>(T)) {
+ if (StructType *ST = dyn_cast<StructType>(T)) {
const StructLayout *Layout = TD->getStructLayout(ST);
Idx = Layout->getElementContainingOffset(Offset);
T = ST->getContainedType(Idx);
@@ -2037,7 +2036,7 @@ uint64_t SROA::FindElementAndOffset(const Type *&T, uint64_t &Offset,
IdxTy = Type::getInt32Ty(T->getContext());
return Idx;
}
- const ArrayType *AT = cast<ArrayType>(T);
+ ArrayType *AT = cast<ArrayType>(T);
T = AT->getElementType();
uint64_t EltSize = TD->getTypeAllocSize(T);
Idx = Offset / EltSize;
@@ -2053,13 +2052,12 @@ void SROA::RewriteGEP(GetElementPtrInst *GEPI, AllocaInst *AI, uint64_t Offset,
SmallVector<AllocaInst*, 32> &NewElts) {
uint64_t OldOffset = Offset;
SmallVector<Value*, 8> Indices(GEPI->op_begin() + 1, GEPI->op_end());
- Offset += TD->getIndexedOffset(GEPI->getPointerOperandType(),
- &Indices[0], Indices.size());
+ Offset += TD->getIndexedOffset(GEPI->getPointerOperandType(), Indices);
RewriteForScalarRepl(GEPI, AI, Offset, NewElts);
- const Type *T = AI->getAllocatedType();
- const Type *IdxTy;
+ Type *T = AI->getAllocatedType();
+ Type *IdxTy;
uint64_t OldIdx = FindElementAndOffset(T, OldOffset, IdxTy);
if (GEPI->getOperand(0) == AI)
OldIdx = ~0ULL; // Force the GEP to be rewritten.
@@ -2073,7 +2071,7 @@ void SROA::RewriteGEP(GetElementPtrInst *GEPI, AllocaInst *AI, uint64_t Offset,
if (Idx == OldIdx)
return;
- const Type *i32Ty = Type::getInt32Ty(AI->getContext());
+ Type *i32Ty = Type::getInt32Ty(AI->getContext());
SmallVector<Value*, 8> NewArgs;
NewArgs.push_back(Constant::getNullValue(i32Ty));
while (EltOffset != 0) {
@@ -2139,7 +2137,7 @@ void SROA::RewriteMemIntrinUserOfAlloca(MemIntrinsic *MI, Instruction *Inst,
// If the pointer is not the right type, insert a bitcast to the right
// type.
- const Type *NewTy =
+ Type *NewTy =
PointerType::get(AI->getType()->getElementType(), AddrSpace);
if (OtherPtr->getType() != NewTy)
@@ -2163,12 +2161,12 @@ void SROA::RewriteMemIntrinUserOfAlloca(MemIntrinsic *MI, Instruction *Inst,
OtherPtr->getName()+"."+Twine(i),
MI);
uint64_t EltOffset;
- const PointerType *OtherPtrTy = cast<PointerType>(OtherPtr->getType());
- const Type *OtherTy = OtherPtrTy->getElementType();
- if (const StructType *ST = dyn_cast<StructType>(OtherTy)) {
+ PointerType *OtherPtrTy = cast<PointerType>(OtherPtr->getType());
+ Type *OtherTy = OtherPtrTy->getElementType();
+ if (StructType *ST = dyn_cast<StructType>(OtherTy)) {
EltOffset = TD->getStructLayout(ST)->getElementOffset(i);
} else {
- const Type *EltTy = cast<SequentialType>(OtherTy)->getElementType();
+ Type *EltTy = cast<SequentialType>(OtherTy)->getElementType();
EltOffset = TD->getTypeAllocSize(EltTy)*i;
}
@@ -2181,7 +2179,7 @@ void SROA::RewriteMemIntrinUserOfAlloca(MemIntrinsic *MI, Instruction *Inst,
}
Value *EltPtr = NewElts[i];
- const Type *EltTy = cast<PointerType>(EltPtr->getType())->getElementType();
+ Type *EltTy = cast<PointerType>(EltPtr->getType())->getElementType();
// If we got down to a scalar, insert a load or store as appropriate.
if (EltTy->isSingleValueType()) {
@@ -2207,7 +2205,7 @@ void SROA::RewriteMemIntrinUserOfAlloca(MemIntrinsic *MI, Instruction *Inst,
StoreVal = Constant::getNullValue(EltTy); // 0.0, null, 0, <0,0>
} else {
// If EltTy is a vector type, get the element type.
- const Type *ValTy = EltTy->getScalarType();
+ Type *ValTy = EltTy->getScalarType();
// Construct an integer with the right value.
unsigned EltSize = TD->getTypeSizeInBits(ValTy);
@@ -2271,7 +2269,7 @@ void SROA::RewriteStoreUserOfWholeAlloca(StoreInst *SI, AllocaInst *AI,
// Extract each element out of the integer according to its structure offset
// and store the element value to the individual alloca.
Value *SrcVal = SI->getOperand(0);
- const Type *AllocaEltTy = AI->getAllocatedType();
+ Type *AllocaEltTy = AI->getAllocatedType();
uint64_t AllocaSizeBits = TD->getTypeAllocSizeInBits(AllocaEltTy);
IRBuilder<> Builder(SI);
@@ -2286,12 +2284,12 @@ void SROA::RewriteStoreUserOfWholeAlloca(StoreInst *SI, AllocaInst *AI,
// There are two forms here: AI could be an array or struct. Both cases
// have different ways to compute the element offset.
- if (const StructType *EltSTy = dyn_cast<StructType>(AllocaEltTy)) {
+ if (StructType *EltSTy = dyn_cast<StructType>(AllocaEltTy)) {
const StructLayout *Layout = TD->getStructLayout(EltSTy);
for (unsigned i = 0, e = NewElts.size(); i != e; ++i) {
// Get the number of bits to shift SrcVal to get the value.
- const Type *FieldTy = EltSTy->getElementType(i);
+ Type *FieldTy = EltSTy->getElementType(i);
uint64_t Shift = Layout->getElementOffsetInBits(i);
if (TD->isBigEndian())
@@ -2327,8 +2325,8 @@ void SROA::RewriteStoreUserOfWholeAlloca(StoreInst *SI, AllocaInst *AI,
}
} else {
- const ArrayType *ATy = cast<ArrayType>(AllocaEltTy);
- const Type *ArrayEltTy = ATy->getElementType();
+ ArrayType *ATy = cast<ArrayType>(AllocaEltTy);
+ Type *ArrayEltTy = ATy->getElementType();
uint64_t ElementOffset = TD->getTypeAllocSizeInBits(ArrayEltTy);
uint64_t ElementSizeBits = TD->getTypeSizeInBits(ArrayEltTy);
@@ -2384,7 +2382,7 @@ void SROA::RewriteLoadUserOfWholeAlloca(LoadInst *LI, AllocaInst *AI,
SmallVector<AllocaInst*, 32> &NewElts) {
// Extract each element out of the NewElts according to its structure offset
// and form the result value.
- const Type *AllocaEltTy = AI->getAllocatedType();
+ Type *AllocaEltTy = AI->getAllocatedType();
uint64_t AllocaSizeBits = TD->getTypeAllocSizeInBits(AllocaEltTy);
DEBUG(dbgs() << "PROMOTING LOAD OF WHOLE ALLOCA: " << *AI << '\n' << *LI
@@ -2394,10 +2392,10 @@ void SROA::RewriteLoadUserOfWholeAlloca(LoadInst *LI, AllocaInst *AI,
// have different ways to compute the element offset.
const StructLayout *Layout = 0;
uint64_t ArrayEltBitOffset = 0;
- if (const StructType *EltSTy = dyn_cast<StructType>(AllocaEltTy)) {
+ if (StructType *EltSTy = dyn_cast<StructType>(AllocaEltTy)) {
Layout = TD->getStructLayout(EltSTy);
} else {
- const Type *ArrayEltTy = cast<ArrayType>(AllocaEltTy)->getElementType();
+ Type *ArrayEltTy = cast<ArrayType>(AllocaEltTy)->getElementType();
ArrayEltBitOffset = TD->getTypeAllocSizeInBits(ArrayEltTy);
}
@@ -2408,14 +2406,14 @@ void SROA::RewriteLoadUserOfWholeAlloca(LoadInst *LI, AllocaInst *AI,
// Load the value from the alloca. If the NewElt is an aggregate, cast
// the pointer to an integer of the same size before doing the load.
Value *SrcField = NewElts[i];
- const Type *FieldTy =
+ Type *FieldTy =
cast<PointerType>(SrcField->getType())->getElementType();
uint64_t FieldSizeBits = TD->getTypeSizeInBits(FieldTy);
// Ignore zero sized fields like {}, they obviously contain no data.
if (FieldSizeBits == 0) continue;
- const IntegerType *FieldIntTy = IntegerType::get(LI->getContext(),
+ IntegerType *FieldIntTy = IntegerType::get(LI->getContext(),
FieldSizeBits);
if (!FieldTy->isIntegerTy() && !FieldTy->isFloatingPointTy() &&
!FieldTy->isVectorTy())
@@ -2468,14 +2466,14 @@ void SROA::RewriteLoadUserOfWholeAlloca(LoadInst *LI, AllocaInst *AI,
/// HasPadding - Return true if the specified type has any structure or
/// alignment padding in between the elements that would be split apart
/// by SROA; return false otherwise.
-static bool HasPadding(const Type *Ty, const TargetData &TD) {
- if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
+static bool HasPadding(Type *Ty, const TargetData &TD) {
+ if (ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
Ty = ATy->getElementType();
return TD.getTypeSizeInBits(Ty) != TD.getTypeAllocSizeInBits(Ty);
}
// SROA currently handles only Arrays and Structs.
- const StructType *STy = cast<StructType>(Ty);
+ StructType *STy = cast<StructType>(Ty);
const StructLayout *SL = TD.getStructLayout(STy);
unsigned PrevFieldBitOffset = 0;
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
@@ -2530,7 +2528,7 @@ bool SROA::isSafeAllocaToScalarRepl(AllocaInst *AI) {
// and fusion code.
if (!Info.hasSubelementAccess && Info.hasALoadOrStore) {
// If the struct/array just has one element, use basic SRoA.
- if (const StructType *ST = dyn_cast<StructType>(AI->getAllocatedType())) {
+ if (StructType *ST = dyn_cast<StructType>(AI->getAllocatedType())) {
if (ST->getNumElements() > 1) return false;
} else {
if (cast<ArrayType>(AI->getAllocatedType())->getNumElements() > 1)
diff --git a/lib/Transforms/Scalar/SimplifyLibCalls.cpp b/lib/Transforms/Scalar/SimplifyLibCalls.cpp
index 7c415e5..ad52417 100644
--- a/lib/Transforms/Scalar/SimplifyLibCalls.cpp
+++ b/lib/Transforms/Scalar/SimplifyLibCalls.cpp
@@ -134,7 +134,7 @@ namespace {
struct StrCatOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
// Verify the "strcat" function prototype.
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 2 ||
FT->getReturnType() != B.getInt8PtrTy() ||
FT->getParamType(0) != FT->getReturnType() ||
@@ -184,7 +184,7 @@ struct StrCatOpt : public LibCallOptimization {
struct StrNCatOpt : public StrCatOpt {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
// Verify the "strncat" function prototype.
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 3 ||
FT->getReturnType() != B.getInt8PtrTy() ||
FT->getParamType(0) != FT->getReturnType() ||
@@ -232,7 +232,7 @@ struct StrNCatOpt : public StrCatOpt {
struct StrChrOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
// Verify the "strchr" function prototype.
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 2 ||
FT->getReturnType() != B.getInt8PtrTy() ||
FT->getParamType(0) != FT->getReturnType() ||
@@ -282,7 +282,7 @@ struct StrChrOpt : public LibCallOptimization {
struct StrRChrOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
// Verify the "strrchr" function prototype.
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 2 ||
FT->getReturnType() != B.getInt8PtrTy() ||
FT->getParamType(0) != FT->getReturnType() ||
@@ -323,7 +323,7 @@ struct StrRChrOpt : public LibCallOptimization {
struct StrCmpOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
// Verify the "strcmp" function prototype.
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 2 ||
!FT->getReturnType()->isIntegerTy(32) ||
FT->getParamType(0) != FT->getParamType(1) ||
@@ -371,7 +371,7 @@ struct StrCmpOpt : public LibCallOptimization {
struct StrNCmpOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
// Verify the "strncmp" function prototype.
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 3 ||
!FT->getReturnType()->isIntegerTy(32) ||
FT->getParamType(0) != FT->getParamType(1) ||
@@ -426,7 +426,7 @@ struct StrCpyOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
// Verify the "strcpy" function prototype.
unsigned NumParams = OptChkCall ? 3 : 2;
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != NumParams ||
FT->getReturnType() != FT->getParamType(0) ||
FT->getParamType(0) != FT->getParamType(1) ||
@@ -462,7 +462,7 @@ struct StrCpyOpt : public LibCallOptimization {
struct StrNCpyOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 3 || FT->getReturnType() != FT->getParamType(0) ||
FT->getParamType(0) != FT->getParamType(1) ||
FT->getParamType(0) != B.getInt8PtrTy() ||
@@ -511,7 +511,7 @@ struct StrNCpyOpt : public LibCallOptimization {
struct StrLenOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 1 ||
FT->getParamType(0) != B.getInt8PtrTy() ||
!FT->getReturnType()->isIntegerTy())
@@ -537,7 +537,7 @@ struct StrLenOpt : public LibCallOptimization {
struct StrPBrkOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 2 ||
FT->getParamType(0) != B.getInt8PtrTy() ||
FT->getParamType(1) != FT->getParamType(0) ||
@@ -575,7 +575,7 @@ struct StrPBrkOpt : public LibCallOptimization {
struct StrToOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if ((FT->getNumParams() != 2 && FT->getNumParams() != 3) ||
!FT->getParamType(0)->isPointerTy() ||
!FT->getParamType(1)->isPointerTy())
@@ -597,7 +597,7 @@ struct StrToOpt : public LibCallOptimization {
struct StrSpnOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 2 ||
FT->getParamType(0) != B.getInt8PtrTy() ||
FT->getParamType(1) != FT->getParamType(0) ||
@@ -626,7 +626,7 @@ struct StrSpnOpt : public LibCallOptimization {
struct StrCSpnOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 2 ||
FT->getParamType(0) != B.getInt8PtrTy() ||
FT->getParamType(1) != FT->getParamType(0) ||
@@ -658,7 +658,7 @@ struct StrCSpnOpt : public LibCallOptimization {
struct StrStrOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 2 ||
!FT->getParamType(0)->isPointerTy() ||
!FT->getParamType(1)->isPointerTy() ||
@@ -722,7 +722,7 @@ struct StrStrOpt : public LibCallOptimization {
struct MemCmpOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 3 || !FT->getParamType(0)->isPointerTy() ||
!FT->getParamType(1)->isPointerTy() ||
!FT->getReturnType()->isIntegerTy(32))
@@ -773,7 +773,7 @@ struct MemCpyOpt : public LibCallOptimization {
// These optimizations require TargetData.
if (!TD) return 0;
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 3 || FT->getReturnType() != FT->getParamType(0) ||
!FT->getParamType(0)->isPointerTy() ||
!FT->getParamType(1)->isPointerTy() ||
@@ -795,7 +795,7 @@ struct MemMoveOpt : public LibCallOptimization {
// These optimizations require TargetData.
if (!TD) return 0;
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 3 || FT->getReturnType() != FT->getParamType(0) ||
!FT->getParamType(0)->isPointerTy() ||
!FT->getParamType(1)->isPointerTy() ||
@@ -817,7 +817,7 @@ struct MemSetOpt : public LibCallOptimization {
// These optimizations require TargetData.
if (!TD) return 0;
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 3 || FT->getReturnType() != FT->getParamType(0) ||
!FT->getParamType(0)->isPointerTy() ||
!FT->getParamType(1)->isIntegerTy() ||
@@ -840,7 +840,7 @@ struct MemSetOpt : public LibCallOptimization {
struct PowOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
// Just make sure this has 2 arguments of the same FP type, which match the
// result type.
if (FT->getNumParams() != 2 || FT->getReturnType() != FT->getParamType(0) ||
@@ -895,7 +895,7 @@ struct PowOpt : public LibCallOptimization {
struct Exp2Opt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
// Just make sure this has 1 argument of FP type, which matches the
// result type.
if (FT->getNumParams() != 1 || FT->getReturnType() != FT->getParamType(0) ||
@@ -946,7 +946,7 @@ struct Exp2Opt : public LibCallOptimization {
struct UnaryDoubleFPOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 1 || !FT->getReturnType()->isDoubleTy() ||
!FT->getParamType(0)->isDoubleTy())
return 0;
@@ -973,7 +973,7 @@ struct UnaryDoubleFPOpt : public LibCallOptimization {
struct FFSOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
// Just make sure this has 2 arguments of the same FP type, which match the
// result type.
if (FT->getNumParams() != 1 ||
@@ -1009,7 +1009,7 @@ struct FFSOpt : public LibCallOptimization {
struct IsDigitOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
// We require integer(i32)
if (FT->getNumParams() != 1 || !FT->getReturnType()->isIntegerTy() ||
!FT->getParamType(0)->isIntegerTy(32))
@@ -1028,7 +1028,7 @@ struct IsDigitOpt : public LibCallOptimization {
struct IsAsciiOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
// We require integer(i32)
if (FT->getNumParams() != 1 || !FT->getReturnType()->isIntegerTy() ||
!FT->getParamType(0)->isIntegerTy(32))
@@ -1046,7 +1046,7 @@ struct IsAsciiOpt : public LibCallOptimization {
struct AbsOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
// We require integer(integer) where the types agree.
if (FT->getNumParams() != 1 || !FT->getReturnType()->isIntegerTy() ||
FT->getParamType(0) != FT->getReturnType())
@@ -1067,7 +1067,7 @@ struct AbsOpt : public LibCallOptimization {
struct ToAsciiOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
// We require i32(i32)
if (FT->getNumParams() != 1 || FT->getReturnType() != FT->getParamType(0) ||
!FT->getParamType(0)->isIntegerTy(32))
@@ -1147,7 +1147,7 @@ struct PrintFOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
// Require one fixed pointer argument and an integer/void result.
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() < 1 || !FT->getParamType(0)->isPointerTy() ||
!(FT->getReturnType()->isIntegerTy() ||
FT->getReturnType()->isVoidTy()))
@@ -1241,7 +1241,7 @@ struct SPrintFOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
// Require two fixed pointer arguments and an integer result.
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 2 || !FT->getParamType(0)->isPointerTy() ||
!FT->getParamType(1)->isPointerTy() ||
!FT->getReturnType()->isIntegerTy())
@@ -1272,7 +1272,7 @@ struct SPrintFOpt : public LibCallOptimization {
struct FWriteOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
// Require a pointer, an integer, an integer, a pointer, returning integer.
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 4 || !FT->getParamType(0)->isPointerTy() ||
!FT->getParamType(1)->isIntegerTy() ||
!FT->getParamType(2)->isIntegerTy() ||
@@ -1310,7 +1310,7 @@ struct FPutsOpt : public LibCallOptimization {
if (!TD) return 0;
// Require two pointers. Also, we can't optimize if return value is used.
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 2 || !FT->getParamType(0)->isPointerTy() ||
!FT->getParamType(1)->isPointerTy() ||
!CI->use_empty())
@@ -1379,7 +1379,7 @@ struct FPrintFOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
// Require two fixed paramters as pointers and integer result.
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() != 2 || !FT->getParamType(0)->isPointerTy() ||
!FT->getParamType(1)->isPointerTy() ||
!FT->getReturnType()->isIntegerTy())
@@ -1410,7 +1410,7 @@ struct FPrintFOpt : public LibCallOptimization {
struct PutsOpt : public LibCallOptimization {
virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
// Require one fixed pointer argument and an integer/void result.
- const FunctionType *FT = Callee->getFunctionType();
+ FunctionType *FT = Callee->getFunctionType();
if (FT->getNumParams() < 1 || !FT->getParamType(0)->isPointerTy() ||
!(FT->getReturnType()->isIntegerTy() ||
FT->getReturnType()->isVoidTy()))
@@ -1685,7 +1685,7 @@ void SimplifyLibCalls::setDoesNotAlias(Function &F, unsigned n) {
void SimplifyLibCalls::inferPrototypeAttributes(Function &F) {
- const FunctionType *FTy = F.getFunctionType();
+ FunctionType *FTy = F.getFunctionType();
StringRef Name = F.getName();
switch (Name[0]) {
generated by cgit v1.1 at 2025-08-15 18:37:11 +0000