#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/Statistic.h"
#include "llvm/Analysis/AliasAnalysis.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/GlobalVariable.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Metadata.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/SSAUpdater.h"
using namespace llvm;
using namespace PatternMatch;
STATISTIC(NumGVNInstr, "Number of instructions deleted");
STATISTIC(NumGVNLoad, "Number of loads deleted");
STATISTIC(NumGVNPRE, "Number of instructions PRE'd");
STATISTIC(NumGVNBlocks, "Number of blocks merged");
STATISTIC(NumGVNSimpl, "Number of instructions simplified");
STATISTIC(NumGVNEqProp, "Number of equalities propagated");
STATISTIC(NumPRELoad, "Number of loads PRE'd");
static cl::opt<bool> EnablePRE("enable-pre",
cl::init(true), cl::Hidden);
static cl::opt<bool> EnableLoadPRE("enable-load-pre", cl::init(true));
static cl::opt<uint32_t>
MaxRecurseDepth("max-recurse-depth", cl::Hidden, cl::init(1000), cl::ZeroOrMore,
cl::desc("Max recurse depth (default = 1000)"));
namespace {
struct Expression {
uint32_t opcode;
Type *type;
SmallVector<uint32_t, 4> varargs;
Expression(uint32_t o = ~2U) : opcode(o) { }
bool operator==(const Expression &other) const {
if (opcode != other.opcode)
return false;
if (opcode == ~0U || opcode == ~1U)
return true;
if (type != other.type)
return false;
if (varargs != other.varargs)
return false;
return true;
}
friend hash_code hash_value(const Expression &Value) {
return hash_combine(Value.opcode, Value.type,
hash_combine_range(Value.varargs.begin(),
Value.varargs.end()));
}
};
class ValueTable {
DenseMap<Value*, uint32_t> valueNumbering;
DenseMap<Expression, uint32_t> expressionNumbering;
AliasAnalysis *AA;
MemoryDependenceAnalysis *MD;
DominatorTree *DT;
uint32_t nextValueNumber;
Expression create_expression(Instruction* I);
Expression create_cmp_expression(unsigned Opcode,
CmpInst::Predicate Predicate,
Value *LHS, Value *RHS);
Expression create_extractvalue_expression(ExtractValueInst* EI);
uint32_t lookup_or_add_call(CallInst* C);
public:
ValueTable() : nextValueNumber(1) { }
uint32_t lookup_or_add(Value *V);
uint32_t lookup(Value *V) const;
uint32_t lookup_or_add_cmp(unsigned Opcode, CmpInst::Predicate Pred,
Value *LHS, Value *RHS);
void add(Value *V, uint32_t num);
void clear();
void erase(Value *v);
void setAliasAnalysis(AliasAnalysis* A) { AA = A; }
AliasAnalysis *getAliasAnalysis() const { return AA; }
void setMemDep(MemoryDependenceAnalysis* M) { MD = M; }
void setDomTree(DominatorTree* D) { DT = D; }
uint32_t getNextUnusedValueNumber() { return nextValueNumber; }
void verifyRemoved(const Value *) const;
};
}
namespace llvm {
template <> struct DenseMapInfo<Expression> {
static inline Expression getEmptyKey() {
return ~0U;
}
static inline Expression getTombstoneKey() {
return ~1U;
}
static unsigned getHashValue(const Expression e) {
using llvm::hash_value;
return static_cast<unsigned>(hash_value(e));
}
static bool isEqual(const Expression &LHS, const Expression &RHS) {
return LHS == RHS;
}
};
}
Expression ValueTable::create_expression(Instruction *I) {
Expression e;
e.type = I->getType();
e.opcode = I->getOpcode();
for (Instruction::op_iterator OI = I->op_begin(), OE = I->op_end();
OI != OE; ++OI)
e.varargs.push_back(lookup_or_add(*OI));
if (I->isCommutative()) {
assert(I->getNumOperands() == 2 && "Unsupported commutative instruction!");
if (e.varargs[0] > e.varargs[1])
std::swap(e.varargs[0], e.varargs[1]);
}
if (CmpInst *C = dyn_cast<CmpInst>(I)) {
CmpInst::Predicate Predicate = C->getPredicate();
if (e.varargs[0] > e.varargs[1]) {
std::swap(e.varargs[0], e.varargs[1]);
Predicate = CmpInst::getSwappedPredicate(Predicate);
}
e.opcode = (C->getOpcode() << 8) | Predicate;
} else if (InsertValueInst *E = dyn_cast<InsertValueInst>(I)) {
for (InsertValueInst::idx_iterator II = E->idx_begin(), IE = E->idx_end();
II != IE; ++II)
e.varargs.push_back(*II);
}
return e;
}
Expression ValueTable::create_cmp_expression(unsigned Opcode,
CmpInst::Predicate Predicate,
Value *LHS, Value *RHS) {
assert((Opcode == Instruction::ICmp || Opcode == Instruction::FCmp) &&
"Not a comparison!");
Expression e;
e.type = CmpInst::makeCmpResultType(LHS->getType());
e.varargs.push_back(lookup_or_add(LHS));
e.varargs.push_back(lookup_or_add(RHS));
if (e.varargs[0] > e.varargs[1]) {
std::swap(e.varargs[0], e.varargs[1]);
Predicate = CmpInst::getSwappedPredicate(Predicate);
}
e.opcode = (Opcode << 8) | Predicate;
return e;
}
Expression ValueTable::create_extractvalue_expression(ExtractValueInst *EI) {
assert(EI != 0 && "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 ) {
switch (I->getIntrinsicID()) {
case Intrinsic::sadd_with_overflow:
case Intrinsic::uadd_with_overflow:
e.opcode = Instruction::Add;
break;
case Intrinsic::ssub_with_overflow:
case Intrinsic::usub_with_overflow:
e.opcode = Instruction::Sub;
break;
case Intrinsic::smul_with_overflow:
case Intrinsic::umul_with_overflow:
e.opcode = Instruction::Mul;
break;
default:
break;
}
if (e.opcode != 0) {
assert(I->getNumArgOperands() == 2 &&
"Expect two args for recognised intrinsics.");
e.varargs.push_back(lookup_or_add(I->getArgOperand(0)));
e.varargs.push_back(lookup_or_add(I->getArgOperand(1)));
return e;
}
}
e.opcode = EI->getOpcode();
for (Instruction::op_iterator OI = EI->op_begin(), OE = EI->op_end();
OI != OE; ++OI)
e.varargs.push_back(lookup_or_add(*OI));
for (ExtractValueInst::idx_iterator II = EI->idx_begin(), IE = EI->idx_end();
II != IE; ++II)
e.varargs.push_back(*II);
return e;
}
void ValueTable::add(Value *V, uint32_t num) {
valueNumbering.insert(std::make_pair(V, num));
}
uint32_t ValueTable::lookup_or_add_call(CallInst *C) {
if (AA->doesNotAccessMemory(C)) {
Expression exp = create_expression(C);
uint32_t &e = expressionNumbering[exp];
if (!e) e = nextValueNumber++;
valueNumbering[C] = e;
return e;
} else if (AA->onlyReadsMemory(C)) {
Expression exp = create_expression(C);
uint32_t &e = expressionNumbering[exp];
if (!e) {
e = nextValueNumber++;
valueNumbering[C] = e;
return e;
}
if (!MD) {
e = nextValueNumber++;
valueNumbering[C] = e;
return e;
}
MemDepResult local_dep = MD->getDependency(C);
if (!local_dep.isDef() && !local_dep.isNonLocal()) {
valueNumbering[C] = nextValueNumber;
return nextValueNumber++;
}
if (local_dep.isDef()) {
CallInst* local_cdep = cast<CallInst>(local_dep.getInst());
if (local_cdep->getNumArgOperands() != C->getNumArgOperands()) {
valueNumbering[C] = nextValueNumber;
return nextValueNumber++;
}
for (unsigned i = 0, e = C->getNumArgOperands(); i < e; ++i) {
uint32_t c_vn = lookup_or_add(C->getArgOperand(i));
uint32_t cd_vn = lookup_or_add(local_cdep->getArgOperand(i));
if (c_vn != cd_vn) {
valueNumbering[C] = nextValueNumber;
return nextValueNumber++;
}
}
uint32_t v = lookup_or_add(local_cdep);
valueNumbering[C] = v;
return v;
}
const MemoryDependenceAnalysis::NonLocalDepInfo &deps =
MD->getNonLocalCallDependency(CallSite(C));
CallInst* cdep = 0;
for (unsigned i = 0, e = deps.size(); i != e; ++i) {
const NonLocalDepEntry *I = &deps[i];
if (I->getResult().isNonLocal())
continue;
if (!I->getResult().isDef() || cdep != 0) {
cdep = 0;
break;
}
CallInst *NonLocalDepCall = dyn_cast<CallInst>(I->getResult().getInst());
if (NonLocalDepCall && DT->properlyDominates(I->getBB(), C->getParent())){
cdep = NonLocalDepCall;
continue;
}
cdep = 0;
break;
}
if (!cdep) {
valueNumbering[C] = nextValueNumber;
return nextValueNumber++;
}
if (cdep->getNumArgOperands() != C->getNumArgOperands()) {
valueNumbering[C] = nextValueNumber;
return nextValueNumber++;
}
for (unsigned i = 0, e = C->getNumArgOperands(); i < e; ++i) {
uint32_t c_vn = lookup_or_add(C->getArgOperand(i));
uint32_t cd_vn = lookup_or_add(cdep->getArgOperand(i));
if (c_vn != cd_vn) {
valueNumbering[C] = nextValueNumber;
return nextValueNumber++;
}
}
uint32_t v = lookup_or_add(cdep);
valueNumbering[C] = v;
return v;
} else {
valueNumbering[C] = nextValueNumber;
return nextValueNumber++;
}
}
uint32_t ValueTable::lookup_or_add(Value *V) {
DenseMap<Value*, uint32_t>::iterator VI = valueNumbering.find(V);
if (VI != valueNumbering.end())
return VI->second;
if (!isa<Instruction>(V)) {
valueNumbering[V] = nextValueNumber;
return nextValueNumber++;
}
Instruction* I = cast<Instruction>(V);
Expression exp;
switch (I->getOpcode()) {
case Instruction::Call:
return lookup_or_add_call(cast<CallInst>(I));
case Instruction::Add:
case Instruction::FAdd:
case Instruction::Sub:
case Instruction::FSub:
case Instruction::Mul:
case Instruction::FMul:
case Instruction::UDiv:
case Instruction::SDiv:
case Instruction::FDiv:
case Instruction::URem:
case Instruction::SRem:
case Instruction::FRem:
case Instruction::Shl:
case Instruction::LShr:
case Instruction::AShr:
case Instruction::And:
case Instruction::Or:
case Instruction::Xor:
case Instruction::ICmp:
case Instruction::FCmp:
case Instruction::Trunc:
case Instruction::ZExt:
case Instruction::SExt:
case Instruction::FPToUI:
case Instruction::FPToSI:
case Instruction::UIToFP:
case Instruction::SIToFP:
case Instruction::FPTrunc:
case Instruction::FPExt:
case Instruction::PtrToInt:
case Instruction::IntToPtr:
case Instruction::BitCast:
case Instruction::Select:
case Instruction::ExtractElement:
case Instruction::InsertElement:
case Instruction::ShuffleVector:
case Instruction::InsertValue:
case Instruction::GetElementPtr:
exp = create_expression(I);
break;
case Instruction::ExtractValue:
exp = create_extractvalue_expression(cast<ExtractValueInst>(I));
break;
default:
valueNumbering[V] = nextValueNumber;
return nextValueNumber++;
}
uint32_t& e = expressionNumbering[exp];
if (!e) e = nextValueNumber++;
valueNumbering[V] = e;
return e;
}
uint32_t ValueTable::lookup(Value *V) const {
DenseMap<Value*, uint32_t>::const_iterator VI = valueNumbering.find(V);
assert(VI != valueNumbering.end() && "Value not numbered?");
return VI->second;
}
uint32_t ValueTable::lookup_or_add_cmp(unsigned Opcode,
CmpInst::Predicate Predicate,
Value *LHS, Value *RHS) {
Expression exp = create_cmp_expression(Opcode, Predicate, LHS, RHS);
uint32_t& e = expressionNumbering[exp];
if (!e) e = nextValueNumber++;
return e;
}
void ValueTable::clear() {
valueNumbering.clear();
expressionNumbering.clear();
nextValueNumber = 1;
}
void ValueTable::erase(Value *V) {
valueNumbering.erase(V);
}
void ValueTable::verifyRemoved(const Value *V) const {
for (DenseMap<Value*, uint32_t>::const_iterator
I = valueNumbering.begin(), E = valueNumbering.end(); I != E; ++I) {
assert(I->first != V && "Inst still occurs in value numbering map!");
}
}
namespace {
class GVN : public FunctionPass {
bool NoLoads;
MemoryDependenceAnalysis *MD;
DominatorTree *DT;
const DataLayout *TD;
const TargetLibraryInfo *TLI;
ValueTable VN;
struct LeaderTableEntry {
Value *Val;
const BasicBlock *BB;
LeaderTableEntry *Next;
};
DenseMap<uint32_t, LeaderTableEntry> LeaderTable;
BumpPtrAllocator TableAllocator;
SmallVector<Instruction*, 8> InstrsToErase;
public:
static char ID; explicit GVN(bool noloads = false)
: FunctionPass(ID), NoLoads(noloads), MD(0) {
initializeGVNPass(*PassRegistry::getPassRegistry());
}
bool runOnFunction(Function &F);
void markInstructionForDeletion(Instruction *I) {
VN.erase(I);
InstrsToErase.push_back(I);
}
const DataLayout *getDataLayout() const { return TD; }
DominatorTree &getDominatorTree() const { return *DT; }
AliasAnalysis *getAliasAnalysis() const { return VN.getAliasAnalysis(); }
MemoryDependenceAnalysis &getMemDep() const { return *MD; }
private:
void addToLeaderTable(uint32_t N, Value *V, const BasicBlock *BB) {
LeaderTableEntry &Curr = LeaderTable[N];
if (!Curr.Val) {
Curr.Val = V;
Curr.BB = BB;
return;
}
LeaderTableEntry *Node = TableAllocator.Allocate<LeaderTableEntry>();
Node->Val = V;
Node->BB = BB;
Node->Next = Curr.Next;
Curr.Next = Node;
}
void removeFromLeaderTable(uint32_t N, Instruction *I, BasicBlock *BB) {
LeaderTableEntry* Prev = 0;
LeaderTableEntry* Curr = &LeaderTable[N];
while (Curr->Val != I || Curr->BB != BB) {
Prev = Curr;
Curr = Curr->Next;
}
if (Prev) {
Prev->Next = Curr->Next;
} else {
if (!Curr->Next) {
Curr->Val = 0;
Curr->BB = 0;
} else {
LeaderTableEntry* Next = Curr->Next;
Curr->Val = Next->Val;
Curr->BB = Next->BB;
Curr->Next = Next->Next;
}
}
}
SmallVector<std::pair<TerminatorInst*, unsigned>, 4> toSplit;
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
AU.addRequired<DominatorTree>();
AU.addRequired<TargetLibraryInfo>();
if (!NoLoads)
AU.addRequired<MemoryDependenceAnalysis>();
AU.addRequired<AliasAnalysis>();
AU.addPreserved<DominatorTree>();
AU.addPreserved<AliasAnalysis>();
}
bool processLoad(LoadInst *L);
bool processInstruction(Instruction *I);
bool processNonLocalLoad(LoadInst *L);
bool processBlock(BasicBlock *BB);
void dump(DenseMap<uint32_t, Value*> &d);
bool iterateOnFunction(Function &F);
bool performPRE(Function &F);
Value *findLeader(const BasicBlock *BB, uint32_t num);
void cleanupGlobalSets();
void verifyRemoved(const Instruction *I) const;
bool splitCriticalEdges();
unsigned replaceAllDominatedUsesWith(Value *From, Value *To,
const BasicBlockEdge &Root);
bool propagateEquality(Value *LHS, Value *RHS, const BasicBlockEdge &Root);
};
char GVN::ID = 0;
}
FunctionPass *llvm::createGVNPass(bool NoLoads) {
return new GVN(NoLoads);
}
INITIALIZE_PASS_BEGIN(GVN, "gvn", "Global Value Numbering", false, false)
INITIALIZE_PASS_DEPENDENCY(MemoryDependenceAnalysis)
INITIALIZE_PASS_DEPENDENCY(DominatorTree)
INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfo)
INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
INITIALIZE_PASS_END(GVN, "gvn", "Global Value Numbering", false, false)
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
void GVN::dump(DenseMap<uint32_t, Value*>& d) {
errs() << "{\n";
for (DenseMap<uint32_t, Value*>::iterator I = d.begin(),
E = d.end(); I != E; ++I) {
errs() << I->first << "\n";
I->second->dump();
}
errs() << "}\n";
}
#endif
static bool IsValueFullyAvailableInBlock(BasicBlock *BB,
DenseMap<BasicBlock*, char> &FullyAvailableBlocks,
uint32_t RecurseDepth) {
if (RecurseDepth > MaxRecurseDepth)
return false;
std::pair<DenseMap<BasicBlock*, char>::iterator, char> IV =
FullyAvailableBlocks.insert(std::make_pair(BB, 2));
if (!IV.second) {
if (IV.first->second == 2)
IV.first->second = 3;
return IV.first->second != 0;
}
pred_iterator PI = pred_begin(BB), PE = pred_end(BB);
if (PI == PE)
goto SpeculationFailure;
for (; PI != PE; ++PI)
if (!IsValueFullyAvailableInBlock(*PI, FullyAvailableBlocks,RecurseDepth+1))
goto SpeculationFailure;
return true;
SpeculationFailure:
char &BBVal = FullyAvailableBlocks[BB];
if (BBVal == 2) {
BBVal = 0;
return false;
}
SmallVector<BasicBlock*, 32> BBWorklist;
BBWorklist.push_back(BB);
do {
BasicBlock *Entry = BBWorklist.pop_back_val();
char &EntryVal = FullyAvailableBlocks[Entry];
if (EntryVal == 0) continue;
EntryVal = 0;
for (succ_iterator I = succ_begin(Entry), E = succ_end(Entry); I != E; ++I)
BBWorklist.push_back(*I);
} while (!BBWorklist.empty());
return false;
}
static bool CanCoerceMustAliasedValueToLoad(Value *StoredVal,
Type *LoadTy,
const DataLayout &TD) {
if (LoadTy->isStructTy() || LoadTy->isArrayTy() ||
StoredVal->getType()->isStructTy() ||
StoredVal->getType()->isArrayTy())
return false;
if (TD.getTypeSizeInBits(StoredVal->getType()) <
TD.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;
Type *StoredValTy = StoredVal->getType();
uint64_t StoreSize = TD.getTypeSizeInBits(StoredValTy);
uint64_t LoadSize = TD.getTypeSizeInBits(LoadedTy);
if (StoreSize == LoadSize) {
if (StoredValTy->getScalarType()->isPointerTy() &&
LoadedTy->getScalarType()->isPointerTy())
return new BitCastInst(StoredVal, LoadedTy, "", InsertPt);
if (StoredValTy->getScalarType()->isPointerTy()) {
StoredValTy = TD.getIntPtrType(StoredValTy);
StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt);
}
Type *TypeToCastTo = LoadedTy;
if (TypeToCastTo->getScalarType()->isPointerTy())
TypeToCastTo = TD.getIntPtrType(TypeToCastTo);
if (StoredValTy != TypeToCastTo)
StoredVal = new BitCastInst(StoredVal, TypeToCastTo, "", InsertPt);
if (LoadedTy->getScalarType()->isPointerTy())
StoredVal = new IntToPtrInst(StoredVal, LoadedTy, "", InsertPt);
return StoredVal;
}
assert(StoreSize >= LoadSize && "CanCoerceMustAliasedValueToLoad fail");
if (StoredValTy->getScalarType()->isPointerTy()) {
StoredValTy = TD.getIntPtrType(StoredValTy);
StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt);
}
if (!StoredValTy->isIntegerTy()) {
StoredValTy = IntegerType::get(StoredValTy->getContext(), StoreSize);
StoredVal = new BitCastInst(StoredVal, StoredValTy, "", InsertPt);
}
if (TD.isBigEndian()) {
Constant *Val = ConstantInt::get(StoredVal->getType(), StoreSize-LoadSize);
StoredVal = BinaryOperator::CreateLShr(StoredVal, Val, "tmp", InsertPt);
}
Type *NewIntTy = IntegerType::get(StoredValTy->getContext(), LoadSize);
StoredVal = new TruncInst(StoredVal, NewIntTy, "trunc", InsertPt);
if (LoadedTy == NewIntTy)
return StoredVal;
if (LoadedTy->getScalarType()->isPointerTy())
return new IntToPtrInst(StoredVal, LoadedTy, "inttoptr", InsertPt);
return new BitCastInst(StoredVal, LoadedTy, "bitcast", InsertPt);
}
static int AnalyzeLoadFromClobberingWrite(Type *LoadTy, Value *LoadPtr,
Value *WritePtr,
uint64_t WriteSizeInBits,
const DataLayout &TD) {
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);
if (StoreBase != LoadBase)
return -1;
#if 0
if (LoadOffset == StoreOffset) {
dbgs() << "STORE/LOAD DEP WITH COMMON POINTER MISSED:\n"
<< "Base = " << *StoreBase << "\n"
<< "Store Ptr = " << *WritePtr << "\n"
<< "Store Offs = " << StoreOffset << "\n"
<< "Load Ptr = " << *LoadPtr << "\n";
abort();
}
#endif
uint64_t LoadSize = TD.getTypeSizeInBits(LoadTy);
if ((WriteSizeInBits & 7) | (LoadSize & 7))
return -1;
uint64_t StoreSize = WriteSizeInBits >> 3; LoadSize >>= 3;
bool isAAFailure = false;
if (StoreOffset < LoadOffset)
isAAFailure = StoreOffset+int64_t(StoreSize) <= LoadOffset;
else
isAAFailure = LoadOffset+int64_t(LoadSize) <= StoreOffset;
if (isAAFailure) {
#if 0
dbgs() << "STORE LOAD DEP WITH COMMON BASE:\n"
<< "Base = " << *StoreBase << "\n"
<< "Store Ptr = " << *WritePtr << "\n"
<< "Store Offs = " << StoreOffset << "\n"
<< "Load Ptr = " << *LoadPtr << "\n";
abort();
#endif
return -1;
}
if (StoreOffset > LoadOffset ||
StoreOffset+StoreSize < LoadOffset+LoadSize)
return -1;
return LoadOffset-StoreOffset;
}
static int AnalyzeLoadFromClobberingStore(Type *LoadTy, Value *LoadPtr,
StoreInst *DepSI,
const DataLayout &TD) {
if (DepSI->getValueOperand()->getType()->isStructTy() ||
DepSI->getValueOperand()->getType()->isArrayTy())
return -1;
Value *StorePtr = DepSI->getPointerOperand();
uint64_t StoreSize =TD.getTypeSizeInBits(DepSI->getValueOperand()->getType());
return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr,
StorePtr, StoreSize, TD);
}
static int AnalyzeLoadFromClobberingLoad(Type *LoadTy, Value *LoadPtr,
LoadInst *DepLI, const DataLayout &TD){
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);
if (R != -1) return R;
int64_t LoadOffs = 0;
const Value *LoadBase =
GetPointerBaseWithConstantOffset(LoadPtr, LoadOffs, TD);
unsigned LoadSize = TD.getTypeStoreSize(LoadTy);
unsigned Size = MemoryDependenceAnalysis::
getLoadLoadClobberFullWidthSize(LoadBase, LoadOffs, LoadSize, DepLI, TD);
if (Size == 0) return -1;
return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, DepPtr, Size*8, TD);
}
static int AnalyzeLoadFromClobberingMemInst(Type *LoadTy, Value *LoadPtr,
MemIntrinsic *MI,
const DataLayout &TD) {
ConstantInt *SizeCst = dyn_cast<ConstantInt>(MI->getLength());
if (SizeCst == 0) return -1;
uint64_t MemSizeInBits = SizeCst->getZExtValue()*8;
if (MI->getIntrinsicID() == Intrinsic::memset)
return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, MI->getDest(),
MemSizeInBits, TD);
MemTransferInst *MTI = cast<MemTransferInst>(MI);
Constant *Src = dyn_cast<Constant>(MTI->getSource());
if (Src == 0) return -1;
GlobalVariable *GV = dyn_cast<GlobalVariable>(GetUnderlyingObject(Src, &TD));
if (GV == 0 || !GV->isConstant()) return -1;
int Offset = AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr,
MI->getDest(), MemSizeInBits, TD);
if (Offset == -1)
return Offset;
Src = ConstantExpr::getBitCast(Src,
llvm::Type::getInt8PtrTy(Src->getContext()));
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))
return Offset;
return -1;
}
static Value *GetStoreValueForLoad(Value *SrcVal, unsigned Offset,
Type *LoadTy,
Instruction *InsertPt, const DataLayout &TD){
LLVMContext &Ctx = SrcVal->getType()->getContext();
uint64_t StoreSize = (TD.getTypeSizeInBits(SrcVal->getType()) + 7) / 8;
uint64_t LoadSize = (TD.getTypeSizeInBits(LoadTy) + 7) / 8;
IRBuilder<> Builder(InsertPt->getParent(), InsertPt);
if (SrcVal->getType()->getScalarType()->isPointerTy())
SrcVal = Builder.CreatePtrToInt(SrcVal,
TD.getIntPtrType(SrcVal->getType()));
if (!SrcVal->getType()->isIntegerTy())
SrcVal = Builder.CreateBitCast(SrcVal, IntegerType::get(Ctx, StoreSize*8));
unsigned ShiftAmt;
if (TD.isLittleEndian())
ShiftAmt = Offset*8;
else
ShiftAmt = (StoreSize-LoadSize-Offset)*8;
if (ShiftAmt)
SrcVal = Builder.CreateLShr(SrcVal, ShiftAmt);
if (LoadSize != StoreSize)
SrcVal = Builder.CreateTrunc(SrcVal, IntegerType::get(Ctx, LoadSize*8));
return CoerceAvailableValueToLoadType(SrcVal, LoadTy, InsertPt, TD);
}
static Value *GetLoadValueForLoad(LoadInst *SrcVal, unsigned Offset,
Type *LoadTy, Instruction *InsertPt,
GVN &gvn) {
const DataLayout &TD = *gvn.getDataLayout();
unsigned SrcValSize = TD.getTypeStoreSize(SrcVal->getType());
unsigned LoadSize = TD.getTypeStoreSize(LoadTy);
if (Offset+LoadSize > SrcValSize) {
assert(SrcVal->isSimple() && "Cannot widen volatile/atomic load!");
assert(SrcVal->getType()->isIntegerTy() && "Can't widen non-integer load");
unsigned NewLoadSize = Offset+LoadSize;
if (!isPowerOf2_32(NewLoadSize))
NewLoadSize = NextPowerOf2(NewLoadSize);
Value *PtrVal = SrcVal->getPointerOperand();
IRBuilder<> Builder(SrcVal->getParent(), ++BasicBlock::iterator(SrcVal));
Type *DestPTy =
IntegerType::get(LoadTy->getContext(), NewLoadSize*8);
DestPTy = PointerType::get(DestPTy,
cast<PointerType>(PtrVal->getType())->getAddressSpace());
Builder.SetCurrentDebugLocation(SrcVal->getDebugLoc());
PtrVal = Builder.CreateBitCast(PtrVal, DestPTy);
LoadInst *NewLoad = Builder.CreateLoad(PtrVal);
NewLoad->takeName(SrcVal);
NewLoad->setAlignment(SrcVal->getAlignment());
DEBUG(dbgs() << "GVN WIDENED LOAD: " << *SrcVal << "\n");
DEBUG(dbgs() << "TO: " << *NewLoad << "\n");
Value *RV = NewLoad;
if (TD.isBigEndian())
RV = Builder.CreateLShr(RV,
NewLoadSize*8-SrcVal->getType()->getPrimitiveSizeInBits());
RV = Builder.CreateTrunc(RV, SrcVal->getType());
SrcVal->replaceAllUsesWith(RV);
gvn.getMemDep().removeInstruction(SrcVal);
SrcVal = NewLoad;
}
return GetStoreValueForLoad(SrcVal, Offset, LoadTy, InsertPt, TD);
}
static Value *GetMemInstValueForLoad(MemIntrinsic *SrcInst, unsigned Offset,
Type *LoadTy, Instruction *InsertPt,
const DataLayout &TD){
LLVMContext &Ctx = LoadTy->getContext();
uint64_t LoadSize = TD.getTypeSizeInBits(LoadTy)/8;
IRBuilder<> Builder(InsertPt->getParent(), InsertPt);
if (MemSetInst *MSI = dyn_cast<MemSetInst>(SrcInst)) {
Value *Val = MSI->getValue();
if (LoadSize != 1)
Val = Builder.CreateZExt(Val, IntegerType::get(Ctx, LoadSize*8));
Value *OneElt = Val;
for (unsigned NumBytesSet = 1; NumBytesSet != LoadSize; ) {
if (NumBytesSet*2 <= LoadSize) {
Value *ShVal = Builder.CreateShl(Val, NumBytesSet*8);
Val = Builder.CreateOr(Val, ShVal);
NumBytesSet <<= 1;
continue;
}
Value *ShVal = Builder.CreateShl(Val, 1*8);
Val = Builder.CreateOr(OneElt, ShVal);
++NumBytesSet;
}
return CoerceAvailableValueToLoadType(Val, LoadTy, InsertPt, TD);
}
MemTransferInst *MTI = cast<MemTransferInst>(SrcInst);
Constant *Src = cast<Constant>(MTI->getSource());
Src = ConstantExpr::getBitCast(Src,
llvm::Type::getInt8PtrTy(Src->getContext()));
Constant *OffsetCst =
ConstantInt::get(Type::getInt64Ty(Src->getContext()), (unsigned)Offset);
Src = ConstantExpr::getGetElementPtr(Src, OffsetCst);
Src = ConstantExpr::getBitCast(Src, PointerType::getUnqual(LoadTy));
return ConstantFoldLoadFromConstPtr(Src, &TD);
}
namespace {
struct AvailableValueInBlock {
BasicBlock *BB;
enum ValType {
SimpleVal, LoadVal, MemIntrin };
PointerIntPair<Value *, 2, ValType> Val;
unsigned Offset;
static AvailableValueInBlock get(BasicBlock *BB, Value *V,
unsigned Offset = 0) {
AvailableValueInBlock Res;
Res.BB = BB;
Res.Val.setPointer(V);
Res.Val.setInt(SimpleVal);
Res.Offset = Offset;
return Res;
}
static AvailableValueInBlock getMI(BasicBlock *BB, MemIntrinsic *MI,
unsigned Offset = 0) {
AvailableValueInBlock Res;
Res.BB = BB;
Res.Val.setPointer(MI);
Res.Val.setInt(MemIntrin);
Res.Offset = Offset;
return Res;
}
static AvailableValueInBlock getLoad(BasicBlock *BB, LoadInst *LI,
unsigned Offset = 0) {
AvailableValueInBlock Res;
Res.BB = BB;
Res.Val.setPointer(LI);
Res.Val.setInt(LoadVal);
Res.Offset = Offset;
return Res;
}
bool isSimpleValue() const { return Val.getInt() == SimpleVal; }
bool isCoercedLoadValue() const { return Val.getInt() == LoadVal; }
bool isMemIntrinValue() const { return Val.getInt() == MemIntrin; }
Value *getSimpleValue() const {
assert(isSimpleValue() && "Wrong accessor");
return Val.getPointer();
}
LoadInst *getCoercedLoadValue() const {
assert(isCoercedLoadValue() && "Wrong accessor");
return cast<LoadInst>(Val.getPointer());
}
MemIntrinsic *getMemIntrinValue() const {
assert(isMemIntrinValue() && "Wrong accessor");
return cast<MemIntrinsic>(Val.getPointer());
}
Value *MaterializeAdjustedValue(Type *LoadTy, GVN &gvn) const {
Value *Res;
if (isSimpleValue()) {
Res = getSimpleValue();
if (Res->getType() != LoadTy) {
const DataLayout *TD = gvn.getDataLayout();
assert(TD && "Need target data to handle type mismatch case");
Res = GetStoreValueForLoad(Res, Offset, LoadTy, BB->getTerminator(),
*TD);
DEBUG(dbgs() << "GVN COERCED NONLOCAL VAL:\nOffset: " << Offset << " "
<< *getSimpleValue() << '\n'
<< *Res << '\n' << "\n\n\n");
}
} else if (isCoercedLoadValue()) {
LoadInst *Load = getCoercedLoadValue();
if (Load->getType() == LoadTy && Offset == 0) {
Res = Load;
} else {
Res = GetLoadValueForLoad(Load, Offset, LoadTy, BB->getTerminator(),
gvn);
DEBUG(dbgs() << "GVN COERCED NONLOCAL LOAD:\nOffset: " << Offset << " "
<< *getCoercedLoadValue() << '\n'
<< *Res << '\n' << "\n\n\n");
}
} else {
const DataLayout *TD = gvn.getDataLayout();
assert(TD && "Need target data to handle type mismatch case");
Res = GetMemInstValueForLoad(getMemIntrinValue(), Offset,
LoadTy, BB->getTerminator(), *TD);
DEBUG(dbgs() << "GVN COERCED NONLOCAL MEM INTRIN:\nOffset: " << Offset
<< " " << *getMemIntrinValue() << '\n'
<< *Res << '\n' << "\n\n\n");
}
return Res;
}
};
}
static Value *ConstructSSAForLoadSet(LoadInst *LI,
SmallVectorImpl<AvailableValueInBlock> &ValuesPerBlock,
GVN &gvn) {
if (ValuesPerBlock.size() == 1 &&
gvn.getDominatorTree().properlyDominates(ValuesPerBlock[0].BB,
LI->getParent()))
return ValuesPerBlock[0].MaterializeAdjustedValue(LI->getType(), gvn);
SmallVector<PHINode*, 8> NewPHIs;
SSAUpdater SSAUpdate(&NewPHIs);
SSAUpdate.Initialize(LI->getType(), LI->getName());
Type *LoadTy = LI->getType();
for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i) {
const AvailableValueInBlock &AV = ValuesPerBlock[i];
BasicBlock *BB = AV.BB;
if (SSAUpdate.HasValueForBlock(BB))
continue;
SSAUpdate.AddAvailableValue(BB, AV.MaterializeAdjustedValue(LoadTy, gvn));
}
Value *V = SSAUpdate.GetValueInMiddleOfBlock(LI->getParent());
if (V->getType()->getScalarType()->isPointerTy()) {
AliasAnalysis *AA = gvn.getAliasAnalysis();
for (unsigned i = 0, e = NewPHIs.size(); i != e; ++i)
AA->copyValue(LI, NewPHIs[i]);
for (unsigned i = 0, e = NewPHIs.size(); i != e; ++i) {
PHINode *P = NewPHIs[i];
for (unsigned ii = 0, ee = P->getNumIncomingValues(); ii != ee; ++ii) {
unsigned jj = PHINode::getOperandNumForIncomingValue(ii);
AA->addEscapingUse(P->getOperandUse(jj));
}
}
}
return V;
}
static bool isLifetimeStart(const Instruction *Inst) {
if (const IntrinsicInst* II = dyn_cast<IntrinsicInst>(Inst))
return II->getIntrinsicID() == Intrinsic::lifetime_start;
return false;
}
bool GVN::processNonLocalLoad(LoadInst *LI) {
SmallVector<NonLocalDepResult, 64> Deps;
AliasAnalysis::Location Loc = VN.getAliasAnalysis()->getLocation(LI);
MD->getNonLocalPointerDependency(Loc, true, LI->getParent(), Deps);
unsigned NumDeps = Deps.size();
if (NumDeps > 100)
return false;
if (NumDeps == 1 &&
!Deps[0].getResult().isDef() && !Deps[0].getResult().isClobber()) {
DEBUG(
dbgs() << "GVN: non-local load ";
WriteAsOperand(dbgs(), LI);
dbgs() << " has unknown dependencies\n";
);
return false;
}
SmallVector<AvailableValueInBlock, 64> ValuesPerBlock;
SmallVector<BasicBlock*, 64> UnavailableBlocks;
for (unsigned i = 0, e = NumDeps; i != e; ++i) {
BasicBlock *DepBB = Deps[i].getBB();
MemDepResult DepInfo = Deps[i].getResult();
if (!DepInfo.isDef() && !DepInfo.isClobber()) {
UnavailableBlocks.push_back(DepBB);
continue;
}
if (DepInfo.isClobber()) {
Value *Address = Deps[i].getAddress();
if (StoreInst *DepSI = dyn_cast<StoreInst>(DepInfo.getInst())) {
if (TD && Address) {
int Offset = AnalyzeLoadFromClobberingStore(LI->getType(), Address,
DepSI, *TD);
if (Offset != -1) {
ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB,
DepSI->getValueOperand(),
Offset));
continue;
}
}
}
if (LoadInst *DepLI = dyn_cast<LoadInst>(DepInfo.getInst())) {
if (DepLI != LI && Address && TD) {
int Offset = AnalyzeLoadFromClobberingLoad(LI->getType(),
LI->getPointerOperand(),
DepLI, *TD);
if (Offset != -1) {
ValuesPerBlock.push_back(AvailableValueInBlock::getLoad(DepBB,DepLI,
Offset));
continue;
}
}
}
if (MemIntrinsic *DepMI = dyn_cast<MemIntrinsic>(DepInfo.getInst())) {
if (TD && Address) {
int Offset = AnalyzeLoadFromClobberingMemInst(LI->getType(), Address,
DepMI, *TD);
if (Offset != -1) {
ValuesPerBlock.push_back(AvailableValueInBlock::getMI(DepBB, DepMI,
Offset));
continue;
}
}
}
UnavailableBlocks.push_back(DepBB);
continue;
}
Instruction *DepInst = DepInfo.getInst();
if (isa<AllocaInst>(DepInst) || isMallocLikeFn(DepInst, TLI) ||
isLifetimeStart(DepInst)) {
ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB,
UndefValue::get(LI->getType())));
continue;
}
if (StoreInst *S = dyn_cast<StoreInst>(DepInst)) {
if (S->getValueOperand()->getType() != LI->getType()) {
if (TD == 0 || !CanCoerceMustAliasedValueToLoad(S->getValueOperand(),
LI->getType(), *TD)) {
UnavailableBlocks.push_back(DepBB);
continue;
}
}
ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB,
S->getValueOperand()));
continue;
}
if (LoadInst *LD = dyn_cast<LoadInst>(DepInst)) {
if (LD->getType() != LI->getType()) {
if (TD == 0 || !CanCoerceMustAliasedValueToLoad(LD, LI->getType(),*TD)){
UnavailableBlocks.push_back(DepBB);
continue;
}
}
ValuesPerBlock.push_back(AvailableValueInBlock::getLoad(DepBB, LD));
continue;
}
UnavailableBlocks.push_back(DepBB);
continue;
}
if (ValuesPerBlock.empty()) return false;
if (UnavailableBlocks.empty()) {
DEBUG(dbgs() << "GVN REMOVING NONLOCAL LOAD: " << *LI << '\n');
Value *V = ConstructSSAForLoadSet(LI, ValuesPerBlock, *this);
LI->replaceAllUsesWith(V);
if (isa<PHINode>(V))
V->takeName(LI);
if (V->getType()->getScalarType()->isPointerTy())
MD->invalidateCachedPointerInfo(V);
markInstructionForDeletion(LI);
++NumGVNLoad;
return true;
}
if (!EnablePRE || !EnableLoadPRE)
return false;
SmallPtrSet<BasicBlock *, 4> Blockers;
for (unsigned i = 0, e = UnavailableBlocks.size(); i != e; ++i)
Blockers.insert(UnavailableBlocks[i]);
BasicBlock *LoadBB = LI->getParent();
BasicBlock *TmpBB = LoadBB;
bool isSinglePred = false;
bool allSingleSucc = true;
while (TmpBB->getSinglePredecessor()) {
isSinglePred = true;
TmpBB = TmpBB->getSinglePredecessor();
if (TmpBB == LoadBB) return false;
if (Blockers.count(TmpBB))
return false;
if (TmpBB->getTerminator()->getNumSuccessors() != 1)
return false;
}
assert(TmpBB);
LoadBB = TmpBB;
if (isSinglePred) {
bool isHot = false;
for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i) {
const AvailableValueInBlock &AV = ValuesPerBlock[i];
if (AV.isSimpleValue())
if (Instruction *I = dyn_cast<Instruction>(AV.getSimpleValue()))
if (DT->dominates(LI, I)) {
isHot = true;
break;
}
}
if (!isHot)
return false;
}
DenseMap<BasicBlock*, Value*> PredLoads;
DenseMap<BasicBlock*, char> FullyAvailableBlocks;
for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i)
FullyAvailableBlocks[ValuesPerBlock[i].BB] = true;
for (unsigned i = 0, e = UnavailableBlocks.size(); i != e; ++i)
FullyAvailableBlocks[UnavailableBlocks[i]] = false;
SmallVector<std::pair<TerminatorInst*, unsigned>, 4> NeedToSplit;
for (pred_iterator PI = pred_begin(LoadBB), E = pred_end(LoadBB);
PI != E; ++PI) {
BasicBlock *Pred = *PI;
if (IsValueFullyAvailableInBlock(Pred, FullyAvailableBlocks, 0)) {
continue;
}
PredLoads[Pred] = 0;
if (Pred->getTerminator()->getNumSuccessors() != 1) {
if (isa<IndirectBrInst>(Pred->getTerminator())) {
DEBUG(dbgs() << "COULD NOT PRE LOAD BECAUSE OF INDBR CRITICAL EDGE '"
<< Pred->getName() << "': " << *LI << '\n');
return false;
}
if (LoadBB->isLandingPad()) {
DEBUG(dbgs()
<< "COULD NOT PRE LOAD BECAUSE OF LANDING PAD CRITICAL EDGE '"
<< Pred->getName() << "': " << *LI << '\n');
return false;
}
unsigned SuccNum = GetSuccessorNumber(Pred, LoadBB);
NeedToSplit.push_back(std::make_pair(Pred->getTerminator(), SuccNum));
}
}
if (!NeedToSplit.empty()) {
toSplit.append(NeedToSplit.begin(), NeedToSplit.end());
return false;
}
unsigned NumUnavailablePreds = PredLoads.size();
assert(NumUnavailablePreds != 0 &&
"Fully available value should be eliminated above!");
if (NumUnavailablePreds != 1)
return false;
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;
PHITransAddr Address(LI->getPointerOperand(), TD);
Value *LoadPtr = 0;
if (allSingleSucc) {
LoadPtr = Address.PHITranslateWithInsertion(LoadBB, UnavailablePred,
*DT, NewInsts);
} else {
Address.PHITranslateValue(LoadBB, UnavailablePred, DT);
LoadPtr = Address.getAddr();
}
if (LoadPtr == 0) {
DEBUG(dbgs() << "COULDN'T INSERT PHI TRANSLATED VALUE OF: "
<< *LI->getPointerOperand() << "\n");
CanDoPRE = false;
break;
}
if (!allSingleSucc &&
!isSafeToLoadUnconditionally(LoadPtr,
UnavailablePred->getTerminator(),
LI->getAlignment(), TD)) {
CanDoPRE = false;
break;
}
I->second = LoadPtr;
}
if (!CanDoPRE) {
while (!NewInsts.empty()) {
Instruction *I = NewInsts.pop_back_val();
if (MD) MD->removeInstruction(I);
I->eraseFromParent();
}
return false;
}
DEBUG(dbgs() << "GVN REMOVING PRE LOAD: " << *LI << '\n');
DEBUG(if (!NewInsts.empty())
dbgs() << "INSERTED " << NewInsts.size() << " INSTS: "
<< *NewInsts.back() << '\n');
for (unsigned i = 0, e = NewInsts.size(); i != e; ++i) {
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;
Instruction *NewLoad = new LoadInst(LoadPtr, LI->getName()+".pre", false,
LI->getAlignment(),
UnavailablePred->getTerminator());
if (MDNode *Tag = LI->getMetadata(LLVMContext::MD_tbaa))
NewLoad->setMetadata(LLVMContext::MD_tbaa, Tag);
NewLoad->setDebugLoc(LI->getDebugLoc());
ValuesPerBlock.push_back(AvailableValueInBlock::get(UnavailablePred,
NewLoad));
MD->invalidateCachedPointerInfo(LoadPtr);
DEBUG(dbgs() << "GVN INSERTED " << *NewLoad << '\n');
}
Value *V = ConstructSSAForLoadSet(LI, ValuesPerBlock, *this);
LI->replaceAllUsesWith(V);
if (isa<PHINode>(V))
V->takeName(LI);
if (V->getType()->getScalarType()->isPointerTy())
MD->invalidateCachedPointerInfo(V);
markInstructionForDeletion(LI);
++NumPRELoad;
return true;
}
static void patchReplacementInstruction(Value *Repl, Instruction *I) {
BinaryOperator *Op = dyn_cast<BinaryOperator>(I);
BinaryOperator *ReplOp = dyn_cast<BinaryOperator>(Repl);
if (Op && ReplOp && isa<OverflowingBinaryOperator>(Op) &&
isa<OverflowingBinaryOperator>(ReplOp)) {
if (ReplOp->hasNoSignedWrap() && !Op->hasNoSignedWrap())
ReplOp->setHasNoSignedWrap(false);
if (ReplOp->hasNoUnsignedWrap() && !Op->hasNoUnsignedWrap())
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); 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;
}
}
}
}
static void patchAndReplaceAllUsesWith(Value *Repl, Instruction *I) {
patchReplacementInstruction(Repl, I);
I->replaceAllUsesWith(Repl);
}
bool GVN::processLoad(LoadInst *L) {
if (!MD)
return false;
if (!L->isSimple())
return false;
if (L->use_empty()) {
markInstructionForDeletion(L);
return true;
}
MemDepResult Dep = MD->getDependency(L);
if (Dep.isClobber() && TD) {
Value *AvailVal = 0;
if (StoreInst *DepSI = dyn_cast<StoreInst>(Dep.getInst())) {
int Offset = AnalyzeLoadFromClobberingStore(L->getType(),
L->getPointerOperand(),
DepSI, *TD);
if (Offset != -1)
AvailVal = GetStoreValueForLoad(DepSI->getValueOperand(), Offset,
L->getType(), L, *TD);
}
if (LoadInst *DepLI = dyn_cast<LoadInst>(Dep.getInst())) {
if (DepLI == L)
return false;
int Offset = AnalyzeLoadFromClobberingLoad(L->getType(),
L->getPointerOperand(),
DepLI, *TD);
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);
if (Offset != -1)
AvailVal = GetMemInstValueForLoad(DepMI, Offset, L->getType(), L, *TD);
}
if (AvailVal) {
DEBUG(dbgs() << "GVN COERCED INST:\n" << *Dep.getInst() << '\n'
<< *AvailVal << '\n' << *L << "\n\n\n");
L->replaceAllUsesWith(AvailVal);
if (AvailVal->getType()->getScalarType()->isPointerTy())
MD->invalidateCachedPointerInfo(AvailVal);
markInstructionForDeletion(L);
++NumGVNLoad;
return true;
}
}
if (Dep.isClobber()) {
DEBUG(
dbgs() << "GVN: load ";
WriteAsOperand(dbgs(), L);
Instruction *I = Dep.getInst();
dbgs() << " is clobbered by " << *I << '\n';
);
return false;
}
if (Dep.isNonLocal())
return processNonLocalLoad(L);
if (!Dep.isDef()) {
DEBUG(
dbgs() << "GVN: load ";
WriteAsOperand(dbgs(), L);
dbgs() << " has unknown dependence\n";
);
return false;
}
Instruction *DepInst = Dep.getInst();
if (StoreInst *DepSI = dyn_cast<StoreInst>(DepInst)) {
Value *StoredVal = DepSI->getValueOperand();
if (StoredVal->getType() != L->getType()) {
if (TD) {
StoredVal = CoerceAvailableValueToLoadType(StoredVal, L->getType(),
L, *TD);
if (StoredVal == 0)
return false;
DEBUG(dbgs() << "GVN COERCED STORE:\n" << *DepSI << '\n' << *StoredVal
<< '\n' << *L << "\n\n\n");
}
else
return false;
}
L->replaceAllUsesWith(StoredVal);
if (StoredVal->getType()->getScalarType()->isPointerTy())
MD->invalidateCachedPointerInfo(StoredVal);
markInstructionForDeletion(L);
++NumGVNLoad;
return true;
}
if (LoadInst *DepLI = dyn_cast<LoadInst>(DepInst)) {
Value *AvailableVal = DepLI;
if (DepLI->getType() != L->getType()) {
if (TD) {
AvailableVal = CoerceAvailableValueToLoadType(DepLI, L->getType(),
L, *TD);
if (AvailableVal == 0)
return false;
DEBUG(dbgs() << "GVN COERCED LOAD:\n" << *DepLI << "\n" << *AvailableVal
<< "\n" << *L << "\n\n\n");
}
else
return false;
}
patchAndReplaceAllUsesWith(AvailableVal, L);
if (DepLI->getType()->getScalarType()->isPointerTy())
MD->invalidateCachedPointerInfo(DepLI);
markInstructionForDeletion(L);
++NumGVNLoad;
return true;
}
if (isa<AllocaInst>(DepInst) || isMallocLikeFn(DepInst, TLI)) {
L->replaceAllUsesWith(UndefValue::get(L->getType()));
markInstructionForDeletion(L);
++NumGVNLoad;
return true;
}
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(DepInst)) {
if (II->getIntrinsicID() == Intrinsic::lifetime_start) {
L->replaceAllUsesWith(UndefValue::get(L->getType()));
markInstructionForDeletion(L);
++NumGVNLoad;
return true;
}
}
return false;
}
Value *GVN::findLeader(const BasicBlock *BB, uint32_t num) {
LeaderTableEntry Vals = LeaderTable[num];
if (!Vals.Val) return 0;
Value *Val = 0;
if (DT->dominates(Vals.BB, BB)) {
Val = Vals.Val;
if (isa<Constant>(Val)) return Val;
}
LeaderTableEntry* Next = Vals.Next;
while (Next) {
if (DT->dominates(Next->BB, BB)) {
if (isa<Constant>(Next->Val)) return Next->Val;
if (!Val) Val = Next->Val;
}
Next = Next->Next;
}
return Val;
}
unsigned GVN::replaceAllDominatedUsesWith(Value *From, Value *To,
const BasicBlockEdge &Root) {
unsigned Count = 0;
for (Value::use_iterator UI = From->use_begin(), UE = From->use_end();
UI != UE; ) {
Use &U = (UI++).getUse();
if (DT->dominates(Root, U)) {
U.set(To);
++Count;
}
}
return Count;
}
static bool isOnlyReachableViaThisEdge(const BasicBlockEdge &E,
DominatorTree *DT) {
const BasicBlock *Pred = E.getEnd()->getSinglePredecessor();
const BasicBlock *Src = E.getStart();
assert((!Pred || Pred == Src) && "No edge between these basic blocks!");
(void)Src;
return Pred != 0;
}
bool GVN::propagateEquality(Value *LHS, Value *RHS,
const BasicBlockEdge &Root) {
SmallVector<std::pair<Value*, Value*>, 4> Worklist;
Worklist.push_back(std::make_pair(LHS, RHS));
bool Changed = false;
bool RootDominatesEnd = isOnlyReachableViaThisEdge(Root, DT);
while (!Worklist.empty()) {
std::pair<Value*, Value*> Item = Worklist.pop_back_val();
LHS = Item.first; RHS = Item.second;
if (LHS == RHS) continue;
assert(LHS->getType() == RHS->getType() && "Equality but unequal types!");
if (isa<Constant>(LHS) && isa<Constant>(RHS)) continue;
if (isa<Constant>(LHS) || (isa<Argument>(LHS) && !isa<Constant>(RHS)))
std::swap(LHS, RHS);
assert((isa<Argument>(LHS) || isa<Instruction>(LHS)) && "Unexpected value!");
uint32_t LVN = VN.lookup_or_add(LHS);
if ((isa<Argument>(LHS) && isa<Argument>(RHS)) ||
(isa<Instruction>(LHS) && isa<Instruction>(RHS))) {
uint32_t RVN = VN.lookup_or_add(RHS);
if (LVN < RVN) {
std::swap(LHS, RHS);
LVN = RVN;
}
}
if (RootDominatesEnd && !isa<Instruction>(RHS))
addToLeaderTable(LVN, RHS, Root.getEnd());
if (!LHS->hasOneUse()) {
unsigned NumReplacements = replaceAllDominatedUsesWith(LHS, RHS, Root);
Changed |= NumReplacements > 0;
NumGVNEqProp += NumReplacements;
}
if (!RHS->getType()->isIntegerTy(1))
continue;
ConstantInt *CI = dyn_cast<ConstantInt>(RHS);
if (!CI)
continue;
bool isKnownTrue = CI->isAllOnesValue();
bool isKnownFalse = !isKnownTrue;
Value *A, *B;
if ((isKnownTrue && match(LHS, m_And(m_Value(A), m_Value(B)))) ||
(isKnownFalse && match(LHS, m_Or(m_Value(A), m_Value(B))))) {
Worklist.push_back(std::make_pair(A, RHS));
Worklist.push_back(std::make_pair(B, RHS));
continue;
}
if (ICmpInst *Cmp = dyn_cast<ICmpInst>(LHS)) {
Value *Op0 = Cmp->getOperand(0), *Op1 = Cmp->getOperand(1);
if ((isKnownTrue && Cmp->getPredicate() == CmpInst::ICMP_EQ) ||
(isKnownFalse && Cmp->getPredicate() == CmpInst::ICMP_NE))
Worklist.push_back(std::make_pair(Op0, Op1));
CmpInst::Predicate NotPred = Cmp->getInversePredicate();
Constant *NotVal = ConstantInt::get(Cmp->getType(), isKnownFalse);
uint32_t NextNum = VN.getNextUnusedValueNumber();
uint32_t Num = VN.lookup_or_add_cmp(Cmp->getOpcode(), NotPred, Op0, Op1);
if (Num < NextNum) {
Value *NotCmp = findLeader(Root.getEnd(), Num);
if (NotCmp && isa<Instruction>(NotCmp)) {
unsigned NumReplacements =
replaceAllDominatedUsesWith(NotCmp, NotVal, Root);
Changed |= NumReplacements > 0;
NumGVNEqProp += NumReplacements;
}
}
if (RootDominatesEnd)
addToLeaderTable(Num, NotVal, Root.getEnd());
continue;
}
}
return Changed;
}
bool GVN::processInstruction(Instruction *I) {
if (isa<DbgInfoIntrinsic>(I))
return false;
if (Value *V = SimplifyInstruction(I, TD, TLI, DT)) {
I->replaceAllUsesWith(V);
if (MD && V->getType()->getScalarType()->isPointerTy())
MD->invalidateCachedPointerInfo(V);
markInstructionForDeletion(I);
++NumGVNSimpl;
return true;
}
if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
if (processLoad(LI))
return true;
unsigned Num = VN.lookup_or_add(LI);
addToLeaderTable(Num, LI, LI->getParent());
return false;
}
if (BranchInst *BI = dyn_cast<BranchInst>(I)) {
if (!BI->isConditional() || isa<Constant>(BI->getCondition()))
return false;
Value *BranchCond = BI->getCondition();
BasicBlock *TrueSucc = BI->getSuccessor(0);
BasicBlock *FalseSucc = BI->getSuccessor(1);
if (TrueSucc == FalseSucc)
return false;
BasicBlock *Parent = BI->getParent();
bool Changed = false;
Value *TrueVal = ConstantInt::getTrue(TrueSucc->getContext());
BasicBlockEdge TrueE(Parent, TrueSucc);
Changed |= propagateEquality(BranchCond, TrueVal, TrueE);
Value *FalseVal = ConstantInt::getFalse(FalseSucc->getContext());
BasicBlockEdge FalseE(Parent, FalseSucc);
Changed |= propagateEquality(BranchCond, FalseVal, FalseE);
return Changed;
}
if (SwitchInst *SI = dyn_cast<SwitchInst>(I)) {
Value *SwitchCond = SI->getCondition();
BasicBlock *Parent = SI->getParent();
bool Changed = false;
SmallDenseMap<BasicBlock *, unsigned, 16> SwitchEdges;
for (unsigned i = 0, n = SI->getNumSuccessors(); i != n; ++i)
++SwitchEdges[SI->getSuccessor(i)];
for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
i != e; ++i) {
BasicBlock *Dst = i.getCaseSuccessor();
if (SwitchEdges.lookup(Dst) == 1) {
BasicBlockEdge E(Parent, Dst);
Changed |= propagateEquality(SwitchCond, i.getCaseValue(), E);
}
}
return Changed;
}
if (I->getType()->isVoidTy()) return false;
uint32_t NextNum = VN.getNextUnusedValueNumber();
unsigned Num = VN.lookup_or_add(I);
if (isa<AllocaInst>(I) || isa<TerminatorInst>(I) || isa<PHINode>(I)) {
addToLeaderTable(Num, I, I->getParent());
return false;
}
if (Num >= NextNum) {
addToLeaderTable(Num, I, I->getParent());
return false;
}
Value *repl = findLeader(I->getParent(), Num);
if (repl == 0) {
addToLeaderTable(Num, I, I->getParent());
return false;
}
patchAndReplaceAllUsesWith(repl, I);
if (MD && repl->getType()->getScalarType()->isPointerTy())
MD->invalidateCachedPointerInfo(repl);
markInstructionForDeletion(I);
return true;
}
bool GVN::runOnFunction(Function& F) {
if (!NoLoads)
MD = &getAnalysis<MemoryDependenceAnalysis>();
DT = &getAnalysis<DominatorTree>();
TD = getAnalysisIfAvailable<DataLayout>();
TLI = &getAnalysis<TargetLibraryInfo>();
VN.setAliasAnalysis(&getAnalysis<AliasAnalysis>());
VN.setMemDep(MD);
VN.setDomTree(DT);
bool Changed = false;
bool ShouldContinue = true;
for (Function::iterator FI = F.begin(), FE = F.end(); FI != FE; ) {
BasicBlock *BB = FI++;
bool removedBlock = MergeBlockIntoPredecessor(BB, this);
if (removedBlock) ++NumGVNBlocks;
Changed |= removedBlock;
}
unsigned Iteration = 0;
while (ShouldContinue) {
DEBUG(dbgs() << "GVN iteration: " << Iteration << "\n");
ShouldContinue = iterateOnFunction(F);
if (splitCriticalEdges())
ShouldContinue = true;
Changed |= ShouldContinue;
++Iteration;
}
if (EnablePRE) {
bool PREChanged = true;
while (PREChanged) {
PREChanged = performPRE(F);
Changed |= PREChanged;
}
}
cleanupGlobalSets();
return Changed;
}
bool GVN::processBlock(BasicBlock *BB) {
assert(InstrsToErase.empty() &&
"We expect InstrsToErase to be empty across iterations");
bool ChangedFunction = false;
for (BasicBlock::iterator BI = BB->begin(), BE = BB->end();
BI != BE;) {
ChangedFunction |= processInstruction(BI);
if (InstrsToErase.empty()) {
++BI;
continue;
}
NumGVNInstr += InstrsToErase.size();
bool AtStart = BI == BB->begin();
if (!AtStart)
--BI;
for (SmallVector<Instruction*, 4>::iterator I = InstrsToErase.begin(),
E = InstrsToErase.end(); I != E; ++I) {
DEBUG(dbgs() << "GVN removed: " << **I << '\n');
if (MD) MD->removeInstruction(*I);
(*I)->eraseFromParent();
DEBUG(verifyRemoved(*I));
}
InstrsToErase.clear();
if (AtStart)
BI = BB->begin();
else
++BI;
}
return ChangedFunction;
}
bool GVN::performPRE(Function &F) {
bool Changed = false;
DenseMap<BasicBlock*, Value*> predMap;
for (df_iterator<BasicBlock*> DI = df_begin(&F.getEntryBlock()),
DE = df_end(&F.getEntryBlock()); DI != DE; ++DI) {
BasicBlock *CurrentBlock = *DI;
if (CurrentBlock == &F.getEntryBlock()) continue;
if (CurrentBlock->isLandingPad()) continue;
for (BasicBlock::iterator BI = CurrentBlock->begin(),
BE = CurrentBlock->end(); BI != BE; ) {
Instruction *CurInst = BI++;
if (isa<AllocaInst>(CurInst) ||
isa<TerminatorInst>(CurInst) || isa<PHINode>(CurInst) ||
CurInst->getType()->isVoidTy() ||
CurInst->mayReadFromMemory() || CurInst->mayHaveSideEffects() ||
isa<DbgInfoIntrinsic>(CurInst))
continue;
if (isa<CmpInst>(CurInst))
continue;
if (CallInst *CallI = dyn_cast<CallInst>(CurInst))
if (CallI->isInlineAsm())
continue;
uint32_t ValNo = VN.lookup(CurInst);
unsigned NumWith = 0;
unsigned NumWithout = 0;
BasicBlock *PREPred = 0;
predMap.clear();
for (pred_iterator PI = pred_begin(CurrentBlock),
PE = pred_end(CurrentBlock); PI != PE; ++PI) {
BasicBlock *P = *PI;
if (P == CurrentBlock) {
NumWithout = 2;
break;
} else if (!DT->dominates(&F.getEntryBlock(), P)) {
NumWithout = 2;
break;
}
Value* predV = findLeader(P, ValNo);
if (predV == 0) {
PREPred = P;
++NumWithout;
} else if (predV == CurInst) {
NumWithout = 2;
} else {
predMap[P] = predV;
++NumWith;
}
}
if (NumWithout != 1 || NumWith == 0)
continue;
if (isa<IndirectBrInst>(PREPred->getTerminator()))
continue;
unsigned SuccNum = GetSuccessorNumber(PREPred, CurrentBlock);
if (isCriticalEdge(PREPred->getTerminator(), SuccNum)) {
toSplit.push_back(std::make_pair(PREPred->getTerminator(), SuccNum));
continue;
}
Instruction *PREInstr = CurInst->clone();
bool success = true;
for (unsigned i = 0, e = CurInst->getNumOperands(); i != e; ++i) {
Value *Op = PREInstr->getOperand(i);
if (isa<Argument>(Op) || isa<Constant>(Op) || isa<GlobalValue>(Op))
continue;
if (Value *V = findLeader(PREPred, VN.lookup(Op))) {
PREInstr->setOperand(i, V);
} else {
success = false;
break;
}
}
if (!success) {
delete PREInstr;
DEBUG(verifyRemoved(PREInstr));
continue;
}
PREInstr->insertBefore(PREPred->getTerminator());
PREInstr->setName(CurInst->getName() + ".pre");
PREInstr->setDebugLoc(CurInst->getDebugLoc());
predMap[PREPred] = PREInstr;
VN.add(PREInstr, ValNo);
++NumGVNPRE;
addToLeaderTable(ValNo, PREInstr, PREPred);
pred_iterator PB = pred_begin(CurrentBlock), PE = pred_end(CurrentBlock);
PHINode* Phi = PHINode::Create(CurInst->getType(), std::distance(PB, PE),
CurInst->getName() + ".pre-phi",
CurrentBlock->begin());
for (pred_iterator PI = PB; PI != PE; ++PI) {
BasicBlock *P = *PI;
Phi->addIncoming(predMap[P], P);
}
VN.add(Phi, ValNo);
addToLeaderTable(ValNo, Phi, CurrentBlock);
Phi->setDebugLoc(CurInst->getDebugLoc());
CurInst->replaceAllUsesWith(Phi);
if (Phi->getType()->getScalarType()->isPointerTy()) {
for (unsigned ii = 0, ee = Phi->getNumIncomingValues(); ii != ee;
++ii) {
unsigned jj = PHINode::getOperandNumForIncomingValue(ii);
VN.getAliasAnalysis()->addEscapingUse(Phi->getOperandUse(jj));
}
if (MD)
MD->invalidateCachedPointerInfo(Phi);
}
VN.erase(CurInst);
removeFromLeaderTable(ValNo, CurInst, CurrentBlock);
DEBUG(dbgs() << "GVN PRE removed: " << *CurInst << '\n');
if (MD) MD->removeInstruction(CurInst);
CurInst->eraseFromParent();
DEBUG(verifyRemoved(CurInst));
Changed = true;
}
}
if (splitCriticalEdges())
Changed = true;
return Changed;
}
bool GVN::splitCriticalEdges() {
if (toSplit.empty())
return false;
do {
std::pair<TerminatorInst*, unsigned> Edge = toSplit.pop_back_val();
SplitCriticalEdge(Edge.first, Edge.second, this);
} while (!toSplit.empty());
if (MD) MD->invalidateCachedPredecessors();
return true;
}
bool GVN::iterateOnFunction(Function &F) {
cleanupGlobalSets();
bool Changed = false;
#if 0
ReversePostOrderTraversal<Function*> RPOT(&F);
for (ReversePostOrderTraversal<Function*>::rpo_iterator RI = RPOT.begin(),
RE = RPOT.end(); RI != RE; ++RI)
Changed |= processBlock(*RI);
#else
for (df_iterator<DomTreeNode*> DI = df_begin(DT->getRootNode()),
DE = df_end(DT->getRootNode()); DI != DE; ++DI)
Changed |= processBlock(DI->getBlock());
#endif
return Changed;
}
void GVN::cleanupGlobalSets() {
VN.clear();
LeaderTable.clear();
TableAllocator.Reset();
}
void GVN::verifyRemoved(const Instruction *Inst) const {
VN.verifyRemoved(Inst);
for (DenseMap<uint32_t, LeaderTableEntry>::const_iterator
I = LeaderTable.begin(), E = LeaderTable.end(); I != E; ++I) {
const LeaderTableEntry *Node = &I->second;
assert(Node->Val != Inst && "Inst still in value numbering scope!");
while (Node->Next) {
Node = Node->Next;
assert(Node->Val != Inst && "Inst still in value numbering scope!");
}
}
}