//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "gvn"
#include "llvm/Transforms/Scalar.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/DepthFirstIterator.h"
#include "llvm/ADT/Hashing.h"
-#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/MapVector.h"
#include "llvm/ADT/SetVector.h"
+#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/CFG.h"
#include "llvm/Analysis/ConstantFolding.h"
-#include "llvm/Analysis/Dominators.h"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/Loads.h"
#include "llvm/Analysis/MemoryBuiltins.h"
#include "llvm/Analysis/MemoryDependenceAnalysis.h"
#include "llvm/Analysis/PHITransAddr.h"
#include "llvm/Analysis/ValueTracking.h"
-#include "llvm/Assembly/Writer.h"
#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Dominators.h"
#include "llvm/IR/GlobalVariable.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Metadata.h"
+#include "llvm/IR/PatternMatch.h"
#include "llvm/Support/Allocator.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
-#include "llvm/Support/PatternMatch.h"
#include "llvm/Target/TargetLibraryInfo.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
+#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Transforms/Utils/SSAUpdater.h"
#include <vector>
using namespace llvm;
using namespace PatternMatch;
+#define DEBUG_TYPE "gvn"
+
STATISTIC(NumGVNInstr, "Number of instructions deleted");
STATISTIC(NumGVNLoad, "Number of loads deleted");
STATISTIC(NumGVNPRE, "Number of instructions PRE'd");
}
Expression ValueTable::create_extractvalue_expression(ExtractValueInst *EI) {
- assert(EI != 0 && "Not an ExtractValueInst?");
+ assert(EI && "Not an ExtractValueInst?");
Expression e;
e.type = EI->getType();
e.opcode = 0;
IntrinsicInst *I = dyn_cast<IntrinsicInst>(EI->getAggregateOperand());
- if (I != 0 && EI->getNumIndices() == 1 && *EI->idx_begin() == 0 ) {
+ if (I != nullptr && EI->getNumIndices() == 1 && *EI->idx_begin() == 0 ) {
// EI might be an extract from one of our recognised intrinsics. If it
// is we'll synthesize a semantically equivalent expression instead on
// an extract value expression.
const MemoryDependenceAnalysis::NonLocalDepInfo &deps =
MD->getNonLocalCallDependency(CallSite(C));
// FIXME: Move the checking logic to MemDep!
- CallInst* cdep = 0;
+ CallInst* cdep = nullptr;
// Check to see if we have a single dominating call instruction that is
// identical to C.
// We don't handle non-definitions. If we already have a call, reject
// instruction dependencies.
- if (!I->getResult().isDef() || cdep != 0) {
- cdep = 0;
+ if (!I->getResult().isDef() || cdep != nullptr) {
+ cdep = nullptr;
break;
}
continue;
}
- cdep = 0;
+ cdep = nullptr;
break;
}
static AvailableValueInBlock getUndef(BasicBlock *BB) {
AvailableValueInBlock Res;
Res.BB = BB;
- Res.Val.setPointer(0);
+ Res.Val.setPointer(nullptr);
Res.Val.setInt(UndefVal);
Res.Offset = 0;
return Res;
bool NoLoads;
MemoryDependenceAnalysis *MD;
DominatorTree *DT;
- const DataLayout *TD;
+ const DataLayout *DL;
const TargetLibraryInfo *TLI;
SetVector<BasicBlock *> DeadBlocks;
public:
static char ID; // Pass identification, replacement for typeid
explicit GVN(bool noloads = false)
- : FunctionPass(ID), NoLoads(noloads), MD(0) {
+ : FunctionPass(ID), NoLoads(noloads), MD(nullptr) {
initializeGVNPass(*PassRegistry::getPassRegistry());
}
- bool runOnFunction(Function &F);
+ bool runOnFunction(Function &F) override;
/// markInstructionForDeletion - This removes the specified instruction from
/// our various maps and marks it for deletion.
InstrsToErase.push_back(I);
}
- const DataLayout *getDataLayout() const { return TD; }
+ const DataLayout *getDataLayout() const { return DL; }
DominatorTree &getDominatorTree() const { return *DT; }
AliasAnalysis *getAliasAnalysis() const { return VN.getAliasAnalysis(); }
MemoryDependenceAnalysis &getMemDep() const { return *MD; }
/// removeFromLeaderTable - Scan the list of values corresponding to a given
/// value number, and remove the given instruction if encountered.
void removeFromLeaderTable(uint32_t N, Instruction *I, BasicBlock *BB) {
- LeaderTableEntry* Prev = 0;
+ LeaderTableEntry* Prev = nullptr;
LeaderTableEntry* Curr = &LeaderTable[N];
while (Curr->Val != I || Curr->BB != BB) {
Prev->Next = Curr->Next;
} else {
if (!Curr->Next) {
- Curr->Val = 0;
- Curr->BB = 0;
+ Curr->Val = nullptr;
+ Curr->BB = nullptr;
} else {
LeaderTableEntry* Next = Curr->Next;
Curr->Val = Next->Val;
SmallVector<std::pair<TerminatorInst*, unsigned>, 4> toSplit;
// This transformation requires dominator postdominator info
- virtual void getAnalysisUsage(AnalysisUsage &AU) const {
- AU.addRequired<DominatorTree>();
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
+ AU.addRequired<DominatorTreeWrapperPass>();
AU.addRequired<TargetLibraryInfo>();
if (!NoLoads)
AU.addRequired<MemoryDependenceAnalysis>();
AU.addRequired<AliasAnalysis>();
- AU.addPreserved<DominatorTree>();
+ AU.addPreserved<DominatorTreeWrapperPass>();
AU.addPreserved<AliasAnalysis>();
}
INITIALIZE_PASS_BEGIN(GVN, "gvn", "Global Value Numbering", false, false)
INITIALIZE_PASS_DEPENDENCY(MemoryDependenceAnalysis)
-INITIALIZE_PASS_DEPENDENCY(DominatorTree)
+INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfo)
INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
INITIALIZE_PASS_END(GVN, "gvn", "Global Value Numbering", false, false)
// Mark as unavailable.
EntryVal = 0;
- for (succ_iterator I = succ_begin(Entry), E = succ_end(Entry); I != E; ++I)
- BBWorklist.push_back(*I);
+ BBWorklist.append(succ_begin(Entry), succ_end(Entry));
} while (!BBWorklist.empty());
return false;
/// CoerceAvailableValueToLoadType will succeed.
static bool CanCoerceMustAliasedValueToLoad(Value *StoredVal,
Type *LoadTy,
- const DataLayout &TD) {
+ const DataLayout &DL) {
// 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.
if (LoadTy->isStructTy() || LoadTy->isArrayTy() ||
return false;
// The store has to be at least as big as the load.
- if (TD.getTypeSizeInBits(StoredVal->getType()) <
- TD.getTypeSizeInBits(LoadTy))
+ if (DL.getTypeSizeInBits(StoredVal->getType()) <
+ DL.getTypeSizeInBits(LoadTy))
return false;
return true;
static Value *CoerceAvailableValueToLoadType(Value *StoredVal,
Type *LoadedTy,
Instruction *InsertPt,
- const DataLayout &TD) {
- if (!CanCoerceMustAliasedValueToLoad(StoredVal, LoadedTy, TD))
- return 0;
+ const DataLayout &DL) {
+ if (!CanCoerceMustAliasedValueToLoad(StoredVal, LoadedTy, DL))
+ return nullptr;
// If this is already the right type, just return it.
Type *StoredValTy = StoredVal->getType();
- uint64_t StoreSize = TD.getTypeSizeInBits(StoredValTy);
- uint64_t LoadSize = TD.getTypeSizeInBits(LoadedTy);
+ uint64_t StoreSize = DL.getTypeSizeInBits(StoredValTy);
+ uint64_t LoadSize = DL.getTypeSizeInBits(LoadedTy);
// If the store and reload are the same size, we can always reuse it.
if (StoreSize == LoadSize) {
// Convert source pointers to integers, which can be bitcast.
if (StoredValTy->getScalarType()->isPointerTy()) {
- StoredValTy = TD.getIntPtrType(StoredValTy);
+ StoredValTy = DL.getIntPtrType(StoredValTy);
StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt);
}
Type *TypeToCastTo = LoadedTy;
if (TypeToCastTo->getScalarType()->isPointerTy())
- TypeToCastTo = TD.getIntPtrType(TypeToCastTo);
+ TypeToCastTo = DL.getIntPtrType(TypeToCastTo);
if (StoredValTy != TypeToCastTo)
StoredVal = new BitCastInst(StoredVal, TypeToCastTo, "", InsertPt);
// Convert source pointers to integers, which can be manipulated.
if (StoredValTy->getScalarType()->isPointerTy()) {
- StoredValTy = TD.getIntPtrType(StoredValTy);
+ StoredValTy = DL.getIntPtrType(StoredValTy);
StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt);
}
// If this is a big-endian system, we need to shift the value down to the low
// bits so that a truncate will work.
- if (TD.isBigEndian()) {
+ if (DL.isBigEndian()) {
Constant *Val = ConstantInt::get(StoredVal->getType(), StoreSize-LoadSize);
StoredVal = BinaryOperator::CreateLShr(StoredVal, Val, "tmp", InsertPt);
}
static int AnalyzeLoadFromClobberingWrite(Type *LoadTy, Value *LoadPtr,
Value *WritePtr,
uint64_t WriteSizeInBits,
- const DataLayout &TD) {
+ const DataLayout &DL) {
// If the loaded or stored value is a first class array or struct, don't try
// to transform them. We need to be able to bitcast to integer.
if (LoadTy->isStructTy() || LoadTy->isArrayTy())
return -1;
int64_t StoreOffset = 0, LoadOffset = 0;
- Value *StoreBase = GetPointerBaseWithConstantOffset(WritePtr,StoreOffset,&TD);
- Value *LoadBase = GetPointerBaseWithConstantOffset(LoadPtr, LoadOffset, &TD);
+ Value *StoreBase = GetPointerBaseWithConstantOffset(WritePtr,StoreOffset,&DL);
+ Value *LoadBase = GetPointerBaseWithConstantOffset(LoadPtr, LoadOffset, &DL);
if (StoreBase != LoadBase)
return -1;
// If the load and store don't overlap at all, the store doesn't provide
// anything to the load. In this case, they really don't alias at all, AA
// must have gotten confused.
- uint64_t LoadSize = TD.getTypeSizeInBits(LoadTy);
+ uint64_t LoadSize = DL.getTypeSizeInBits(LoadTy);
if ((WriteSizeInBits & 7) | (LoadSize & 7))
return -1;
/// memdep query of a load that ends up being a clobbering store.
static int AnalyzeLoadFromClobberingStore(Type *LoadTy, Value *LoadPtr,
StoreInst *DepSI,
- const DataLayout &TD) {
+ const DataLayout &DL) {
// Cannot handle reading from store of first-class aggregate yet.
if (DepSI->getValueOperand()->getType()->isStructTy() ||
DepSI->getValueOperand()->getType()->isArrayTy())
return -1;
Value *StorePtr = DepSI->getPointerOperand();
- uint64_t StoreSize =TD.getTypeSizeInBits(DepSI->getValueOperand()->getType());
+ uint64_t StoreSize =DL.getTypeSizeInBits(DepSI->getValueOperand()->getType());
return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr,
- StorePtr, StoreSize, TD);
+ StorePtr, StoreSize, DL);
}
/// 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(Type *LoadTy, Value *LoadPtr,
- LoadInst *DepLI, const DataLayout &TD){
+ LoadInst *DepLI, const DataLayout &DL){
// Cannot handle reading from store of first-class aggregate yet.
if (DepLI->getType()->isStructTy() || DepLI->getType()->isArrayTy())
return -1;
Value *DepPtr = DepLI->getPointerOperand();
- uint64_t DepSize = TD.getTypeSizeInBits(DepLI->getType());
- int R = AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, DepPtr, DepSize, TD);
+ uint64_t DepSize = DL.getTypeSizeInBits(DepLI->getType());
+ int R = AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, DepPtr, DepSize, DL);
if (R != -1) return R;
// If we have a load/load clobber an DepLI can be widened to cover this load,
// then we should widen it!
int64_t LoadOffs = 0;
const Value *LoadBase =
- GetPointerBaseWithConstantOffset(LoadPtr, LoadOffs, &TD);
- unsigned LoadSize = TD.getTypeStoreSize(LoadTy);
+ GetPointerBaseWithConstantOffset(LoadPtr, LoadOffs, &DL);
+ unsigned LoadSize = DL.getTypeStoreSize(LoadTy);
unsigned Size = MemoryDependenceAnalysis::
- getLoadLoadClobberFullWidthSize(LoadBase, LoadOffs, LoadSize, DepLI, TD);
+ getLoadLoadClobberFullWidthSize(LoadBase, LoadOffs, LoadSize, DepLI, DL);
if (Size == 0) return -1;
- return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, DepPtr, Size*8, TD);
+ return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, DepPtr, Size*8, DL);
}
static int AnalyzeLoadFromClobberingMemInst(Type *LoadTy, Value *LoadPtr,
MemIntrinsic *MI,
- const DataLayout &TD) {
+ const DataLayout &DL) {
// If the mem operation is a non-constant size, we can't handle it.
ConstantInt *SizeCst = dyn_cast<ConstantInt>(MI->getLength());
- if (SizeCst == 0) return -1;
+ if (!SizeCst) return -1;
uint64_t MemSizeInBits = SizeCst->getZExtValue()*8;
// If this is memset, we just need to see if the offset is valid in the size
// of the memset..
if (MI->getIntrinsicID() == Intrinsic::memset)
return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, MI->getDest(),
- MemSizeInBits, TD);
+ MemSizeInBits, DL);
// If we have a memcpy/memmove, the only case we can handle is if this is a
// copy from constant memory. In that case, we can read directly from the
MemTransferInst *MTI = cast<MemTransferInst>(MI);
Constant *Src = dyn_cast<Constant>(MTI->getSource());
- if (Src == 0) return -1;
+ if (!Src) return -1;
- GlobalVariable *GV = dyn_cast<GlobalVariable>(GetUnderlyingObject(Src, &TD));
- if (GV == 0 || !GV->isConstant()) return -1;
+ GlobalVariable *GV = dyn_cast<GlobalVariable>(GetUnderlyingObject(Src, &DL));
+ if (!GV || !GV->isConstant()) return -1;
// See if the access is within the bounds of the transfer.
int Offset = AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr,
- MI->getDest(), MemSizeInBits, TD);
+ MI->getDest(), MemSizeInBits, DL);
if (Offset == -1)
return Offset;
+ unsigned AS = Src->getType()->getPointerAddressSpace();
// Otherwise, see if we can constant fold a load from the constant with the
// offset applied as appropriate.
Src = ConstantExpr::getBitCast(Src,
- llvm::Type::getInt8PtrTy(Src->getContext()));
+ Type::getInt8PtrTy(Src->getContext(), AS));
Constant *OffsetCst =
ConstantInt::get(Type::getInt64Ty(Src->getContext()), (unsigned)Offset);
Src = ConstantExpr::getGetElementPtr(Src, OffsetCst);
- Src = ConstantExpr::getBitCast(Src, PointerType::getUnqual(LoadTy));
- if (ConstantFoldLoadFromConstPtr(Src, &TD))
+ Src = ConstantExpr::getBitCast(Src, PointerType::get(LoadTy, AS));
+ if (ConstantFoldLoadFromConstPtr(Src, &DL))
return Offset;
return -1;
}
/// before we give up.
static Value *GetStoreValueForLoad(Value *SrcVal, unsigned Offset,
Type *LoadTy,
- Instruction *InsertPt, const DataLayout &TD){
+ Instruction *InsertPt, const DataLayout &DL){
LLVMContext &Ctx = SrcVal->getType()->getContext();
- uint64_t StoreSize = (TD.getTypeSizeInBits(SrcVal->getType()) + 7) / 8;
- uint64_t LoadSize = (TD.getTypeSizeInBits(LoadTy) + 7) / 8;
+ uint64_t StoreSize = (DL.getTypeSizeInBits(SrcVal->getType()) + 7) / 8;
+ uint64_t LoadSize = (DL.getTypeSizeInBits(LoadTy) + 7) / 8;
IRBuilder<> Builder(InsertPt->getParent(), InsertPt);
// to an integer type to start with.
if (SrcVal->getType()->getScalarType()->isPointerTy())
SrcVal = Builder.CreatePtrToInt(SrcVal,
- TD.getIntPtrType(SrcVal->getType()));
+ DL.getIntPtrType(SrcVal->getType()));
if (!SrcVal->getType()->isIntegerTy())
SrcVal = Builder.CreateBitCast(SrcVal, IntegerType::get(Ctx, StoreSize*8));
// Shift the bits to the least significant depending on endianness.
unsigned ShiftAmt;
- if (TD.isLittleEndian())
+ if (DL.isLittleEndian())
ShiftAmt = Offset*8;
else
ShiftAmt = (StoreSize-LoadSize-Offset)*8;
if (LoadSize != StoreSize)
SrcVal = Builder.CreateTrunc(SrcVal, IntegerType::get(Ctx, LoadSize*8));
- return CoerceAvailableValueToLoadType(SrcVal, LoadTy, InsertPt, TD);
+ return CoerceAvailableValueToLoadType(SrcVal, LoadTy, InsertPt, DL);
}
/// GetLoadValueForLoad - This function is called when we have a
static Value *GetLoadValueForLoad(LoadInst *SrcVal, unsigned Offset,
Type *LoadTy, Instruction *InsertPt,
GVN &gvn) {
- const DataLayout &TD = *gvn.getDataLayout();
+ const DataLayout &DL = *gvn.getDataLayout();
// If Offset+LoadTy exceeds the size of SrcVal, then we must be wanting to
// widen SrcVal out to a larger load.
- unsigned SrcValSize = TD.getTypeStoreSize(SrcVal->getType());
- unsigned LoadSize = TD.getTypeStoreSize(LoadTy);
+ unsigned SrcValSize = DL.getTypeStoreSize(SrcVal->getType());
+ unsigned LoadSize = DL.getTypeStoreSize(LoadTy);
if (Offset+LoadSize > SrcValSize) {
assert(SrcVal->isSimple() && "Cannot widen volatile/atomic load!");
assert(SrcVal->getType()->isIntegerTy() && "Can't widen non-integer load");
Type *DestPTy =
IntegerType::get(LoadTy->getContext(), NewLoadSize*8);
DestPTy = PointerType::get(DestPTy,
- cast<PointerType>(PtrVal->getType())->getAddressSpace());
+ PtrVal->getType()->getPointerAddressSpace());
Builder.SetCurrentDebugLocation(SrcVal->getDebugLoc());
PtrVal = Builder.CreateBitCast(PtrVal, DestPTy);
LoadInst *NewLoad = Builder.CreateLoad(PtrVal);
// Replace uses of the original load with the wider load. On a big endian
// system, we need to shift down to get the relevant bits.
Value *RV = NewLoad;
- if (TD.isBigEndian())
+ if (DL.isBigEndian())
RV = Builder.CreateLShr(RV,
NewLoadSize*8-SrcVal->getType()->getPrimitiveSizeInBits());
RV = Builder.CreateTrunc(RV, SrcVal->getType());
SrcVal = NewLoad;
}
- return GetStoreValueForLoad(SrcVal, Offset, LoadTy, InsertPt, TD);
+ return GetStoreValueForLoad(SrcVal, Offset, LoadTy, InsertPt, DL);
}
/// memdep query of a load that ends up being a clobbering mem intrinsic.
static Value *GetMemInstValueForLoad(MemIntrinsic *SrcInst, unsigned Offset,
Type *LoadTy, Instruction *InsertPt,
- const DataLayout &TD){
+ const DataLayout &DL){
LLVMContext &Ctx = LoadTy->getContext();
- uint64_t LoadSize = TD.getTypeSizeInBits(LoadTy)/8;
+ uint64_t LoadSize = DL.getTypeSizeInBits(LoadTy)/8;
IRBuilder<> Builder(InsertPt->getParent(), InsertPt);
++NumBytesSet;
}
- return CoerceAvailableValueToLoadType(Val, LoadTy, InsertPt, TD);
+ return CoerceAvailableValueToLoadType(Val, LoadTy, InsertPt, DL);
}
// Otherwise, this is a memcpy/memmove from a constant global.
MemTransferInst *MTI = cast<MemTransferInst>(SrcInst);
Constant *Src = cast<Constant>(MTI->getSource());
+ unsigned AS = Src->getType()->getPointerAddressSpace();
// Otherwise, see if we can constant fold a load from the constant with the
// offset applied as appropriate.
Src = ConstantExpr::getBitCast(Src,
- llvm::Type::getInt8PtrTy(Src->getContext()));
+ Type::getInt8PtrTy(Src->getContext(), AS));
Constant *OffsetCst =
- ConstantInt::get(Type::getInt64Ty(Src->getContext()), (unsigned)Offset);
+ ConstantInt::get(Type::getInt64Ty(Src->getContext()), (unsigned)Offset);
Src = ConstantExpr::getGetElementPtr(Src, OffsetCst);
- Src = ConstantExpr::getBitCast(Src, PointerType::getUnqual(LoadTy));
- return ConstantFoldLoadFromConstPtr(Src, &TD);
+ Src = ConstantExpr::getBitCast(Src, PointerType::get(LoadTy, AS));
+ return ConstantFoldLoadFromConstPtr(Src, &DL);
}
if (isSimpleValue()) {
Res = getSimpleValue();
if (Res->getType() != LoadTy) {
- const DataLayout *TD = gvn.getDataLayout();
- assert(TD && "Need target data to handle type mismatch case");
+ const DataLayout *DL = gvn.getDataLayout();
+ assert(DL && "Need target data to handle type mismatch case");
Res = GetStoreValueForLoad(Res, Offset, LoadTy, BB->getTerminator(),
- *TD);
+ *DL);
DEBUG(dbgs() << "GVN COERCED NONLOCAL VAL:\nOffset: " << Offset << " "
<< *getSimpleValue() << '\n'
<< *Res << '\n' << "\n\n\n");
}
} else if (isMemIntrinValue()) {
- const DataLayout *TD = gvn.getDataLayout();
- assert(TD && "Need target data to handle type mismatch case");
+ const DataLayout *DL = gvn.getDataLayout();
+ assert(DL && "Need target data to handle type mismatch case");
Res = GetMemInstValueForLoad(getMemIntrinValue(), Offset,
- LoadTy, BB->getTerminator(), *TD);
+ LoadTy, BB->getTerminator(), *DL);
DEBUG(dbgs() << "GVN COERCED NONLOCAL MEM INTRIN:\nOffset: " << Offset
<< " " << *getMemIntrinValue() << '\n'
<< *Res << '\n' << "\n\n\n");
// read by the load, we can extract the bits we need for the load from the
// stored value.
if (StoreInst *DepSI = dyn_cast<StoreInst>(DepInfo.getInst())) {
- if (TD && Address) {
+ if (DL && Address) {
int Offset = AnalyzeLoadFromClobberingStore(LI->getType(), Address,
- DepSI, *TD);
+ DepSI, *DL);
if (Offset != -1) {
ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB,
DepSI->getValueOperand(),
if (LoadInst *DepLI = dyn_cast<LoadInst>(DepInfo.getInst())) {
// If this is a clobber and L is the first instruction in its block, then
// we have the first instruction in the entry block.
- if (DepLI != LI && Address && TD) {
- int Offset = AnalyzeLoadFromClobberingLoad(LI->getType(),
- LI->getPointerOperand(),
- DepLI, *TD);
+ if (DepLI != LI && Address && DL) {
+ int Offset = AnalyzeLoadFromClobberingLoad(LI->getType(), Address,
+ DepLI, *DL);
if (Offset != -1) {
ValuesPerBlock.push_back(AvailableValueInBlock::getLoad(DepBB,DepLI,
// If the clobbering value is a memset/memcpy/memmove, see if we can
// forward a value on from it.
if (MemIntrinsic *DepMI = dyn_cast<MemIntrinsic>(DepInfo.getInst())) {
- if (TD && Address) {
+ if (DL && Address) {
int Offset = AnalyzeLoadFromClobberingMemInst(LI->getType(), Address,
- DepMI, *TD);
+ DepMI, *DL);
if (Offset != -1) {
ValuesPerBlock.push_back(AvailableValueInBlock::getMI(DepBB, DepMI,
Offset));
continue;
}
+ // Loading from calloc (which zero initializes memory) -> zero
+ if (isCallocLikeFn(DepInst, TLI)) {
+ ValuesPerBlock.push_back(AvailableValueInBlock::get(
+ DepBB, Constant::getNullValue(LI->getType())));
+ continue;
+ }
+
if (StoreInst *S = dyn_cast<StoreInst>(DepInst)) {
// Reject loads and stores that are to the same address but are of
// different types if we have to.
if (S->getValueOperand()->getType() != LI->getType()) {
// If the stored value is larger or equal to the loaded value, we can
// reuse it.
- if (TD == 0 || !CanCoerceMustAliasedValueToLoad(S->getValueOperand(),
- LI->getType(), *TD)) {
+ if (!DL || !CanCoerceMustAliasedValueToLoad(S->getValueOperand(),
+ LI->getType(), *DL)) {
UnavailableBlocks.push_back(DepBB);
continue;
}
if (LD->getType() != LI->getType()) {
// If the stored value is larger or equal to the loaded value, we can
// reuse it.
- if (TD == 0 || !CanCoerceMustAliasedValueToLoad(LD, LI->getType(),*TD)){
+ if (!DL || !CanCoerceMustAliasedValueToLoad(LD, LI->getType(),*DL)) {
UnavailableBlocks.push_back(DepBB);
continue;
}
// Check to see how many predecessors have the loaded value fully
// available.
- DenseMap<BasicBlock*, Value*> PredLoads;
+ MapVector<BasicBlock *, Value *> PredLoads;
DenseMap<BasicBlock*, char> FullyAvailableBlocks;
for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i)
FullyAvailableBlocks[ValuesPerBlock[i].BB] = true;
if (IsValueFullyAvailableInBlock(Pred, FullyAvailableBlocks, 0)) {
continue;
}
- PredLoads[Pred] = 0;
if (Pred->getTerminator()->getNumSuccessors() != 1) {
if (isa<IndirectBrInst>(Pred->getTerminator())) {
}
CriticalEdgePred.push_back(Pred);
+ } else {
+ // Only add the predecessors that will not be split for now.
+ PredLoads[Pred] = nullptr;
}
}
// Decide whether PRE is profitable for this load.
- unsigned NumUnavailablePreds = PredLoads.size();
+ unsigned NumUnavailablePreds = PredLoads.size() + CriticalEdgePred.size();
assert(NumUnavailablePreds != 0 &&
"Fully available value should already be eliminated!");
return false;
// Split critical edges, and update the unavailable predecessors accordingly.
- for (SmallVectorImpl<BasicBlock *>::iterator I = CriticalEdgePred.begin(),
- E = CriticalEdgePred.end(); I != E; I++) {
- BasicBlock *OrigPred = *I;
+ for (BasicBlock *OrigPred : CriticalEdgePred) {
BasicBlock *NewPred = splitCriticalEdges(OrigPred, LoadBB);
- PredLoads.erase(OrigPred);
- PredLoads[NewPred] = 0;
+ assert(!PredLoads.count(OrigPred) && "Split edges shouldn't be in map!");
+ PredLoads[NewPred] = nullptr;
DEBUG(dbgs() << "Split critical edge " << OrigPred->getName() << "->"
<< LoadBB->getName() << '\n');
}
// Check if the load can safely be moved to all the unavailable predecessors.
bool CanDoPRE = true;
SmallVector<Instruction*, 8> NewInsts;
- for (DenseMap<BasicBlock*, Value*>::iterator I = PredLoads.begin(),
- E = PredLoads.end(); I != E; ++I) {
- BasicBlock *UnavailablePred = I->first;
+ for (auto &PredLoad : PredLoads) {
+ BasicBlock *UnavailablePred = PredLoad.first;
// Do PHI translation to get its value in the predecessor if necessary. The
// returned pointer (if non-null) is guaranteed to dominate UnavailablePred.
// If all preds have a single successor, then we know it is safe to insert
// the load on the pred (?!?), so we can insert code to materialize the
// pointer if it is not available.
- PHITransAddr Address(LI->getPointerOperand(), TD);
- Value *LoadPtr = 0;
+ PHITransAddr Address(LI->getPointerOperand(), DL);
+ Value *LoadPtr = nullptr;
LoadPtr = Address.PHITranslateWithInsertion(LoadBB, UnavailablePred,
*DT, NewInsts);
// If we couldn't find or insert a computation of this phi translated value,
// we fail PRE.
- if (LoadPtr == 0) {
+ if (!LoadPtr) {
DEBUG(dbgs() << "COULDN'T INSERT PHI TRANSLATED VALUE OF: "
<< *LI->getPointerOperand() << "\n");
CanDoPRE = false;
break;
}
- I->second = LoadPtr;
+ PredLoad.second = LoadPtr;
}
if (!CanDoPRE) {
if (MD) MD->removeInstruction(I);
I->eraseFromParent();
}
- // HINT:Don't revert the edge-splitting as following transformation may
- // also need to split these critial edges.
+ // HINT: Don't revert the edge-splitting as following transformation may
+ // also need to split these critical edges.
return !CriticalEdgePred.empty();
}
VN.lookup_or_add(NewInsts[i]);
}
- for (DenseMap<BasicBlock*, Value*>::iterator I = PredLoads.begin(),
- E = PredLoads.end(); I != E; ++I) {
- BasicBlock *UnavailablePred = I->first;
- Value *LoadPtr = I->second;
+ for (const auto &PredLoad : PredLoads) {
+ BasicBlock *UnavailablePred = PredLoad.first;
+ Value *LoadPtr = PredLoad.second;
Instruction *NewLoad = new LoadInst(LoadPtr, LI->getName()+".pre", false,
LI->getAlignment(),
UnavailablePred->getTerminator());
- // Transfer the old load's TBAA tag to the new load.
- if (MDNode *Tag = LI->getMetadata(LLVMContext::MD_tbaa))
- NewLoad->setMetadata(LLVMContext::MD_tbaa, Tag);
+ // Transfer the old load's AA tags to the new load.
+ AAMDNodes Tags;
+ LI->getAAMetadata(Tags);
+ if (Tags)
+ NewLoad->setAAMetadata(Tags);
// Transfer DebugLoc.
NewLoad->setDebugLoc(LI->getDebugLoc());
!Deps[0].getResult().isDef() && !Deps[0].getResult().isClobber()) {
DEBUG(
dbgs() << "GVN: non-local load ";
- WriteAsOperand(dbgs(), LI);
+ LI->printAsOperand(dbgs());
dbgs() << " has unknown dependencies\n";
);
return false;
ReplOp->setHasNoUnsignedWrap(false);
}
if (Instruction *ReplInst = dyn_cast<Instruction>(Repl)) {
- SmallVector<std::pair<unsigned, MDNode*>, 4> Metadata;
- ReplInst->getAllMetadataOtherThanDebugLoc(Metadata);
- for (int i = 0, n = Metadata.size(); i < n; ++i) {
- unsigned Kind = Metadata[i].first;
- MDNode *IMD = I->getMetadata(Kind);
- MDNode *ReplMD = Metadata[i].second;
- switch(Kind) {
- default:
- ReplInst->setMetadata(Kind, NULL); // Remove unknown metadata
- break;
- case LLVMContext::MD_dbg:
- llvm_unreachable("getAllMetadataOtherThanDebugLoc returned a MD_dbg");
- case LLVMContext::MD_tbaa:
- ReplInst->setMetadata(Kind, MDNode::getMostGenericTBAA(IMD, ReplMD));
- break;
- case LLVMContext::MD_range:
- ReplInst->setMetadata(Kind, MDNode::getMostGenericRange(IMD, ReplMD));
- break;
- case LLVMContext::MD_prof:
- llvm_unreachable("MD_prof in a non terminator instruction");
- break;
- case LLVMContext::MD_fpmath:
- ReplInst->setMetadata(Kind, MDNode::getMostGenericFPMath(IMD, ReplMD));
- break;
- }
- }
+ // FIXME: If both the original and replacement value are part of the
+ // same control-flow region (meaning that the execution of one
+ // guarentees the executation of the other), then we can combine the
+ // noalias scopes here and do better than the general conservative
+ // answer used in combineMetadata().
+
+ // In general, GVN unifies expressions over different control-flow
+ // regions, and so we need a conservative combination of the noalias
+ // scopes.
+ unsigned KnownIDs[] = {
+ LLVMContext::MD_tbaa,
+ LLVMContext::MD_alias_scope,
+ LLVMContext::MD_noalias,
+ LLVMContext::MD_range,
+ LLVMContext::MD_fpmath,
+ LLVMContext::MD_invariant_load,
+ };
+ combineMetadata(ReplInst, I, KnownIDs);
}
}
// If we have a clobber and target data is around, see if this is a clobber
// that we can fix up through code synthesis.
- if (Dep.isClobber() && TD) {
+ if (Dep.isClobber() && DL) {
// Check to see if we have something like this:
// store i32 123, i32* %P
// %A = bitcast i32* %P to i8*
// a common base + constant offset, and if the previous store (or memset)
// completely covers this load. This sort of thing can happen in bitfield
// access code.
- Value *AvailVal = 0;
+ Value *AvailVal = nullptr;
if (StoreInst *DepSI = dyn_cast<StoreInst>(Dep.getInst())) {
int Offset = AnalyzeLoadFromClobberingStore(L->getType(),
L->getPointerOperand(),
- DepSI, *TD);
+ DepSI, *DL);
if (Offset != -1)
AvailVal = GetStoreValueForLoad(DepSI->getValueOperand(), Offset,
- L->getType(), L, *TD);
+ L->getType(), L, *DL);
}
// Check to see if we have something like this:
int Offset = AnalyzeLoadFromClobberingLoad(L->getType(),
L->getPointerOperand(),
- DepLI, *TD);
+ DepLI, *DL);
if (Offset != -1)
AvailVal = GetLoadValueForLoad(DepLI, Offset, L->getType(), L, *this);
}
if (MemIntrinsic *DepMI = dyn_cast<MemIntrinsic>(Dep.getInst())) {
int Offset = AnalyzeLoadFromClobberingMemInst(L->getType(),
L->getPointerOperand(),
- DepMI, *TD);
+ DepMI, *DL);
if (Offset != -1)
- AvailVal = GetMemInstValueForLoad(DepMI, Offset, L->getType(), L, *TD);
+ AvailVal = GetMemInstValueForLoad(DepMI, Offset, L->getType(), L, *DL);
}
if (AvailVal) {
DEBUG(
// fast print dep, using operator<< on instruction is too slow.
dbgs() << "GVN: load ";
- WriteAsOperand(dbgs(), L);
+ L->printAsOperand(dbgs());
Instruction *I = Dep.getInst();
dbgs() << " is clobbered by " << *I << '\n';
);
DEBUG(
// fast print dep, using operator<< on instruction is too slow.
dbgs() << "GVN: load ";
- WriteAsOperand(dbgs(), L);
+ L->printAsOperand(dbgs());
dbgs() << " has unknown dependence\n";
);
return false;
// actually have the same type. See if we know how to reuse the stored
// value (depending on its type).
if (StoredVal->getType() != L->getType()) {
- if (TD) {
+ if (DL) {
StoredVal = CoerceAvailableValueToLoadType(StoredVal, L->getType(),
- L, *TD);
- if (StoredVal == 0)
+ L, *DL);
+ if (!StoredVal)
return false;
DEBUG(dbgs() << "GVN COERCED STORE:\n" << *DepSI << '\n' << *StoredVal
// the same type. See if we know how to reuse the previously loaded value
// (depending on its type).
if (DepLI->getType() != L->getType()) {
- if (TD) {
+ if (DL) {
AvailableVal = CoerceAvailableValueToLoadType(DepLI, L->getType(),
- L, *TD);
- if (AvailableVal == 0)
+ L, *DL);
+ if (!AvailableVal)
return false;
DEBUG(dbgs() << "GVN COERCED LOAD:\n" << *DepLI << "\n" << *AvailableVal
}
}
+ // If this load follows a calloc (which zero initializes memory),
+ // then the loaded value is zero
+ if (isCallocLikeFn(DepInst, TLI)) {
+ L->replaceAllUsesWith(Constant::getNullValue(L->getType()));
+ markInstructionForDeletion(L);
+ ++NumGVNLoad;
+ return true;
+ }
+
return false;
}
// a few comparisons of DFS numbers.
Value *GVN::findLeader(const BasicBlock *BB, uint32_t num) {
LeaderTableEntry Vals = LeaderTable[num];
- if (!Vals.Val) return 0;
+ if (!Vals.Val) return nullptr;
- Value *Val = 0;
+ Value *Val = nullptr;
if (DT->dominates(Vals.BB, BB)) {
Val = Vals.Val;
if (isa<Constant>(Val)) return Val;
unsigned Count = 0;
for (Value::use_iterator UI = From->use_begin(), UE = From->use_end();
UI != UE; ) {
- Use &U = (UI++).getUse();
+ Use &U = *UI++;
if (DT->dominates(Root, U)) {
U.set(To);
const BasicBlock *Src = E.getStart();
assert((!Pred || Pred == Src) && "No edge between these basic blocks!");
(void)Src;
- return Pred != 0;
+ return Pred != nullptr;
}
/// propagateEquality - The given values are known to be equal in every block
// to value numbering it. Value numbering often exposes redundancies, for
// example if it determines that %y is equal to %x then the instruction
// "%z = and i32 %x, %y" becomes "%z = and i32 %x, %x" which we now simplify.
- if (Value *V = SimplifyInstruction(I, TD, TLI, DT)) {
+ if (Value *V = SimplifyInstruction(I, DL, TLI, DT)) {
I->replaceAllUsesWith(V);
if (MD && V->getType()->getScalarType()->isPointerTy())
MD->invalidateCachedPointerInfo(V);
// Perform fast-path value-number based elimination of values inherited from
// dominators.
Value *repl = findLeader(I->getParent(), Num);
- if (repl == 0) {
+ if (!repl) {
// Failure, just remember this instance for future use.
addToLeaderTable(Num, I, I->getParent());
return false;
/// runOnFunction - This is the main transformation entry point for a function.
bool GVN::runOnFunction(Function& F) {
+ if (skipOptnoneFunction(F))
+ return false;
+
if (!NoLoads)
MD = &getAnalysis<MemoryDependenceAnalysis>();
- DT = &getAnalysis<DominatorTree>();
- TD = getAnalysisIfAvailable<DataLayout>();
+ DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
+ DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
+ DL = DLP ? &DLP->getDataLayout() : nullptr;
TLI = &getAnalysis<TargetLibraryInfo>();
VN.setAliasAnalysis(&getAnalysis<AliasAnalysis>());
VN.setMemDep(MD);
bool GVN::performPRE(Function &F) {
bool Changed = false;
SmallVector<std::pair<Value*, BasicBlock*>, 8> predMap;
- for (df_iterator<BasicBlock*> DI = df_begin(&F.getEntryBlock()),
- DE = df_end(&F.getEntryBlock()); DI != DE; ++DI) {
- BasicBlock *CurrentBlock = *DI;
-
+ for (BasicBlock *CurrentBlock : depth_first(&F.getEntryBlock())) {
// Nothing to PRE in the entry block.
if (CurrentBlock == &F.getEntryBlock()) continue;
// more complicated to get right.
unsigned NumWith = 0;
unsigned NumWithout = 0;
- BasicBlock *PREPred = 0;
+ BasicBlock *PREPred = nullptr;
predMap.clear();
for (pred_iterator PI = pred_begin(CurrentBlock),
}
Value* predV = findLeader(P, ValNo);
- if (predV == 0) {
- predMap.push_back(std::make_pair(static_cast<Value *>(0), P));
+ if (!predV) {
+ predMap.push_back(std::make_pair(static_cast<Value *>(nullptr), P));
PREPred = P;
++NumWithout;
} else if (predV == CurInst) {
//
std::vector<BasicBlock *> BBVect;
BBVect.reserve(256);
- for (df_iterator<DomTreeNode*> DI = df_begin(DT->getRootNode()),
- DE = df_end(DT->getRootNode()); DI != DE; ++DI)
- BBVect.push_back(DI->getBlock());
+ for (DomTreeNode *x : depth_first(DT->getRootNode()))
+ BBVect.push_back(x->getBlock());
for (std::vector<BasicBlock *>::iterator I = BBVect.begin(), E = BBVect.end();
I != E; I++)
if (DeadBlocks.count(B))
continue;
- for (pred_iterator PI = pred_begin(B), PE = pred_end(B); PI != PE; PI++) {
+ SmallVector<BasicBlock *, 4> Preds(pred_begin(B), pred_end(B));
+ for (SmallVectorImpl<BasicBlock *>::iterator PI = Preds.begin(),
+ PE = Preds.end(); PI != PE; PI++) {
BasicBlock *P = *PI;
+
if (!DeadBlocks.count(P))
continue;
+
+ if (isCriticalEdge(P->getTerminator(), GetSuccessorNumber(P, B))) {
+ if (BasicBlock *S = splitCriticalEdges(P, B))
+ DeadBlocks.insert(P = S);
+ }
+
for (BasicBlock::iterator II = B->begin(); isa<PHINode>(II); ++II) {
PHINode &Phi = cast<PHINode>(*II);
Phi.setIncomingValue(Phi.getBasicBlockIndex(P),
return true;
}
-// performPRE() will trigger assert if it come across an instruciton without
+// performPRE() will trigger assert if it comes across an instruction without
// associated val-num. As it normally has far more live instructions than dead
// instructions, it makes more sense just to "fabricate" a val-number for the
// dead code than checking if instruction involved is dead or not.