#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
+#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/IR/CFG.h"
+#include "llvm/IR/ConstantRange.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/MDBuilder.h"
#include "llvm/IR/Metadata.h"
#include "llvm/IR/Module.h"
+#include "llvm/IR/NoFolder.h"
#include "llvm/IR/Operator.h"
+#include "llvm/IR/PatternMatch.h"
#include "llvm/IR/Type.h"
-#include "llvm/Support/CFG.h"
#include "llvm/Support/CommandLine.h"
-#include "llvm/Support/ConstantRange.h"
#include "llvm/Support/Debug.h"
-#include "llvm/Support/NoFolder.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include <algorithm>
#include <map>
#include <set>
using namespace llvm;
+using namespace PatternMatch;
static cl::opt<unsigned>
PHINodeFoldingThreshold("phi-node-folding-threshold", cl::Hidden, cl::init(1),
SinkCommon("simplifycfg-sink-common", cl::Hidden, cl::init(true),
cl::desc("Sink common instructions down to the end block"));
-static cl::opt<bool>
-HoistCondStores("simplifycfg-hoist-cond-stores", cl::Hidden, cl::init(true),
- cl::desc("Hoist conditional stores if an unconditional store preceeds"));
+static cl::opt<bool> HoistCondStores(
+ "simplifycfg-hoist-cond-stores", cl::Hidden, cl::init(true),
+ cl::desc("Hoist conditional stores if an unconditional store precedes"));
STATISTIC(NumBitMaps, "Number of switch instructions turned into bitmaps");
STATISTIC(NumLookupTables, "Number of switch instructions turned into lookup tables");
class SimplifyCFGOpt {
const TargetTransformInfo &TTI;
- const DataLayout *const TD;
-
+ const DataLayout *const DL;
Value *isValueEqualityComparison(TerminatorInst *TI);
BasicBlock *GetValueEqualityComparisonCases(TerminatorInst *TI,
std::vector<ValueEqualityComparisonCase> &Cases);
bool SimplifyCondBranch(BranchInst *BI, IRBuilder <>&Builder);
public:
- SimplifyCFGOpt(const TargetTransformInfo &TTI, const DataLayout *TD)
- : TTI(TTI), TD(TD) {}
+ SimplifyCFGOpt(const TargetTransformInfo &TTI, const DataLayout *DL)
+ : TTI(TTI), DL(DL) {}
bool run(BasicBlock *BB);
};
}
PN->addIncoming(PN->getIncomingValueForBlock(ExistPred), NewPred);
}
-
-/// GetIfCondition - Given a basic block (BB) with two predecessors (and at
-/// least one PHI node in it), check to see if the merge at this block is due
-/// to an "if condition". If so, return the boolean condition that determines
-/// which entry into BB will be taken. Also, return by references the block
-/// that will be entered from if the condition is true, and the block that will
-/// be entered if the condition is false.
-///
-/// This does no checking to see if the true/false blocks have large or unsavory
-/// instructions in them.
-static Value *GetIfCondition(BasicBlock *BB, BasicBlock *&IfTrue,
- BasicBlock *&IfFalse) {
- PHINode *SomePHI = cast<PHINode>(BB->begin());
- assert(SomePHI->getNumIncomingValues() == 2 &&
- "Function can only handle blocks with 2 predecessors!");
- BasicBlock *Pred1 = SomePHI->getIncomingBlock(0);
- BasicBlock *Pred2 = SomePHI->getIncomingBlock(1);
-
- // We can only handle branches. Other control flow will be lowered to
- // branches if possible anyway.
- BranchInst *Pred1Br = dyn_cast<BranchInst>(Pred1->getTerminator());
- BranchInst *Pred2Br = dyn_cast<BranchInst>(Pred2->getTerminator());
- if (Pred1Br == 0 || Pred2Br == 0)
- return 0;
-
- // Eliminate code duplication by ensuring that Pred1Br is conditional if
- // either are.
- if (Pred2Br->isConditional()) {
- // If both branches are conditional, we don't have an "if statement". In
- // reality, we could transform this case, but since the condition will be
- // required anyway, we stand no chance of eliminating it, so the xform is
- // probably not profitable.
- if (Pred1Br->isConditional())
- return 0;
-
- std::swap(Pred1, Pred2);
- std::swap(Pred1Br, Pred2Br);
- }
-
- if (Pred1Br->isConditional()) {
- // The only thing we have to watch out for here is to make sure that Pred2
- // doesn't have incoming edges from other blocks. If it does, the condition
- // doesn't dominate BB.
- if (Pred2->getSinglePredecessor() == 0)
- return 0;
-
- // If we found a conditional branch predecessor, make sure that it branches
- // to BB and Pred2Br. If it doesn't, this isn't an "if statement".
- if (Pred1Br->getSuccessor(0) == BB &&
- Pred1Br->getSuccessor(1) == Pred2) {
- IfTrue = Pred1;
- IfFalse = Pred2;
- } else if (Pred1Br->getSuccessor(0) == Pred2 &&
- Pred1Br->getSuccessor(1) == BB) {
- IfTrue = Pred2;
- IfFalse = Pred1;
- } else {
- // We know that one arm of the conditional goes to BB, so the other must
- // go somewhere unrelated, and this must not be an "if statement".
- return 0;
- }
-
- return Pred1Br->getCondition();
- }
-
- // Ok, if we got here, both predecessors end with an unconditional branch to
- // BB. Don't panic! If both blocks only have a single (identical)
- // predecessor, and THAT is a conditional branch, then we're all ok!
- BasicBlock *CommonPred = Pred1->getSinglePredecessor();
- if (CommonPred == 0 || CommonPred != Pred2->getSinglePredecessor())
- return 0;
-
- // Otherwise, if this is a conditional branch, then we can use it!
- BranchInst *BI = dyn_cast<BranchInst>(CommonPred->getTerminator());
- if (BI == 0) return 0;
-
- assert(BI->isConditional() && "Two successors but not conditional?");
- if (BI->getSuccessor(0) == Pred1) {
- IfTrue = Pred1;
- IfFalse = Pred2;
- } else {
- IfTrue = Pred2;
- IfFalse = Pred1;
- }
- return BI->getCondition();
-}
-
/// ComputeSpeculationCost - Compute an abstract "cost" of speculating the
/// given instruction, which is assumed to be safe to speculate. 1 means
/// cheap, 2 means less cheap, and UINT_MAX means prohibitively expensive.
/// GetConstantInt - Extract ConstantInt from value, looking through IntToPtr
/// and PointerNullValue. Return NULL if value is not a constant int.
-static ConstantInt *GetConstantInt(Value *V, const DataLayout *TD) {
+static ConstantInt *GetConstantInt(Value *V, const DataLayout *DL) {
// Normal constant int.
ConstantInt *CI = dyn_cast<ConstantInt>(V);
- if (CI || !TD || !isa<Constant>(V) || !V->getType()->isPointerTy())
+ if (CI || !DL || !isa<Constant>(V) || !V->getType()->isPointerTy())
return CI;
// This is some kind of pointer constant. Turn it into a pointer-sized
// ConstantInt if possible.
- IntegerType *PtrTy = cast<IntegerType>(TD->getIntPtrType(V->getType()));
+ IntegerType *PtrTy = cast<IntegerType>(DL->getIntPtrType(V->getType()));
// Null pointer means 0, see SelectionDAGBuilder::getValue(const Value*).
if (isa<ConstantPointerNull>(V))
/// Values vector.
static Value *
GatherConstantCompares(Value *V, std::vector<ConstantInt*> &Vals, Value *&Extra,
- const DataLayout *TD, bool isEQ, unsigned &UsedICmps) {
+ const DataLayout *DL, bool isEQ, unsigned &UsedICmps) {
Instruction *I = dyn_cast<Instruction>(V);
if (I == 0) return 0;
// If this is an icmp against a constant, handle this as one of the cases.
if (ICmpInst *ICI = dyn_cast<ICmpInst>(I)) {
- if (ConstantInt *C = GetConstantInt(I->getOperand(1), TD)) {
+ if (ConstantInt *C = GetConstantInt(I->getOperand(1), DL)) {
+ Value *RHSVal;
+ ConstantInt *RHSC;
+
if (ICI->getPredicate() == (isEQ ? ICmpInst::ICMP_EQ:ICmpInst::ICMP_NE)) {
+ // (x & ~2^x) == y --> x == y || x == y|2^x
+ // This undoes a transformation done by instcombine to fuse 2 compares.
+ if (match(ICI->getOperand(0),
+ m_And(m_Value(RHSVal), m_ConstantInt(RHSC)))) {
+ APInt Not = ~RHSC->getValue();
+ if (Not.isPowerOf2()) {
+ Vals.push_back(C);
+ Vals.push_back(
+ ConstantInt::get(C->getContext(), C->getValue() | Not));
+ UsedICmps++;
+ return RHSVal;
+ }
+ }
+
UsedICmps++;
Vals.push_back(C);
return I->getOperand(0);
ConstantRange Span =
ConstantRange::makeICmpRegion(ICI->getPredicate(), C->getValue());
+ // Shift the range if the compare is fed by an add. This is the range
+ // compare idiom as emitted by instcombine.
+ bool hasAdd =
+ match(I->getOperand(0), m_Add(m_Value(RHSVal), m_ConstantInt(RHSC)));
+ if (hasAdd)
+ Span = Span.subtract(RHSC->getValue());
+
// If this is an and/!= check then we want to optimize "x ugt 2" into
// x != 0 && x != 1.
if (!isEQ)
for (APInt Tmp = Span.getLower(); Tmp != Span.getUpper(); ++Tmp)
Vals.push_back(ConstantInt::get(V->getContext(), Tmp));
UsedICmps++;
- return I->getOperand(0);
+ return hasAdd ? RHSVal : I->getOperand(0);
}
return 0;
}
unsigned NumValsBeforeLHS = Vals.size();
unsigned UsedICmpsBeforeLHS = UsedICmps;
- if (Value *LHS = GatherConstantCompares(I->getOperand(0), Vals, Extra, TD,
+ if (Value *LHS = GatherConstantCompares(I->getOperand(0), Vals, Extra, DL,
isEQ, UsedICmps)) {
unsigned NumVals = Vals.size();
unsigned UsedICmpsBeforeRHS = UsedICmps;
- if (Value *RHS = GatherConstantCompares(I->getOperand(1), Vals, Extra, TD,
+ if (Value *RHS = GatherConstantCompares(I->getOperand(1), Vals, Extra, DL,
isEQ, UsedICmps)) {
if (LHS == RHS)
return LHS;
if (Extra == 0 || Extra == I->getOperand(0)) {
Value *OldExtra = Extra;
Extra = I->getOperand(0);
- if (Value *RHS = GatherConstantCompares(I->getOperand(1), Vals, Extra, TD,
+ if (Value *RHS = GatherConstantCompares(I->getOperand(1), Vals, Extra, DL,
isEQ, UsedICmps))
return RHS;
assert(Vals.size() == NumValsBeforeLHS);
} else if (BranchInst *BI = dyn_cast<BranchInst>(TI))
if (BI->isConditional() && BI->getCondition()->hasOneUse())
if (ICmpInst *ICI = dyn_cast<ICmpInst>(BI->getCondition()))
- if (ICI->isEquality() && GetConstantInt(ICI->getOperand(1), TD))
+ if (ICI->isEquality() && GetConstantInt(ICI->getOperand(1), DL))
CV = ICI->getOperand(0);
// Unwrap any lossless ptrtoint cast.
- if (TD && CV && CV->getType() == TD->getIntPtrType(CV->getContext()))
- if (PtrToIntInst *PTII = dyn_cast<PtrToIntInst>(CV))
- CV = PTII->getOperand(0);
+ if (DL && CV) {
+ if (PtrToIntInst *PTII = dyn_cast<PtrToIntInst>(CV)) {
+ Value *Ptr = PTII->getPointerOperand();
+ if (PTII->getType() == DL->getIntPtrType(Ptr->getType()))
+ CV = Ptr;
+ }
+ }
return CV;
}
ICmpInst *ICI = cast<ICmpInst>(BI->getCondition());
BasicBlock *Succ = BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_NE);
Cases.push_back(ValueEqualityComparisonCase(GetConstantInt(ICI->getOperand(1),
- TD),
+ DL),
Succ));
return BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_EQ);
}
};
}
-static int ConstantIntSortPredicate(const void *P1, const void *P2) {
- const ConstantInt *LHS = *(const ConstantInt*const*)P1;
- const ConstantInt *RHS = *(const ConstantInt*const*)P2;
+static int ConstantIntSortPredicate(ConstantInt *const *P1,
+ ConstantInt *const *P2) {
+ const ConstantInt *LHS = *P1;
+ const ConstantInt *RHS = *P2;
if (LHS->getValue().ult(RHS->getValue()))
return 1;
if (LHS->getValue() == RHS->getValue())
}
}
-/// Sees if any of the weights are too big for a uint32_t, and halves all the
-/// weights if any are.
+/// Keep halving the weights until all can fit in uint32_t.
static void FitWeights(MutableArrayRef<uint64_t> Weights) {
- bool Halve = false;
- for (unsigned i = 0; i < Weights.size(); ++i)
- if (Weights[i] > UINT_MAX) {
- Halve = true;
- break;
- }
+ while (true) {
+ bool Halve = false;
+ for (unsigned i = 0; i < Weights.size(); ++i)
+ if (Weights[i] > UINT_MAX) {
+ Halve = true;
+ break;
+ }
- if (! Halve)
- return;
+ if (! Halve)
+ return;
- for (unsigned i = 0; i < Weights.size(); ++i)
- Weights[i] /= 2;
+ for (unsigned i = 0; i < Weights.size(); ++i)
+ Weights[i] /= 2;
+ }
}
/// FoldValueComparisonIntoPredecessors - The specified terminator is a value
Builder.SetInsertPoint(PTI);
// Convert pointer to int before we switch.
if (CV->getType()->isPointerTy()) {
- assert(TD && "Cannot switch on pointer without DataLayout");
- CV = Builder.CreatePtrToInt(CV, TD->getIntPtrType(CV->getContext()),
+ assert(DL && "Cannot switch on pointer without DataLayout");
+ CV = Builder.CreatePtrToInt(CV, DL->getIntPtrType(CV->getType()),
"magicptr");
}
Value *SpeculatedStoreValue = 0;
StoreInst *SpeculatedStore = 0;
for (BasicBlock::iterator BBI = ThenBB->begin(),
- BBE = llvm::prior(ThenBB->end());
+ BBE = std::prev(ThenBB->end());
BBI != BBE; ++BBI) {
Instruction *I = BBI;
// Skip debug info.
Value *OrigV = PN->getIncomingValueForBlock(BB);
Value *ThenV = PN->getIncomingValueForBlock(ThenBB);
+ // FIXME: Try to remove some of the duplication with HoistThenElseCodeToIf.
// Skip PHIs which are trivial.
if (ThenV == OrigV)
continue;
HaveRewritablePHIs = true;
- ConstantExpr *CE = dyn_cast<ConstantExpr>(ThenV);
- if (!CE)
+ ConstantExpr *OrigCE = dyn_cast<ConstantExpr>(OrigV);
+ ConstantExpr *ThenCE = dyn_cast<ConstantExpr>(ThenV);
+ if (!OrigCE && !ThenCE)
continue; // Known safe and cheap.
- if (!isSafeToSpeculativelyExecute(CE))
+ if ((ThenCE && !isSafeToSpeculativelyExecute(ThenCE)) ||
+ (OrigCE && !isSafeToSpeculativelyExecute(OrigCE)))
return false;
- if (ComputeSpeculationCost(CE) > PHINodeFoldingThreshold)
+ unsigned OrigCost = OrigCE ? ComputeSpeculationCost(OrigCE) : 0;
+ unsigned ThenCost = ThenCE ? ComputeSpeculationCost(ThenCE) : 0;
+ if (OrigCost + ThenCost > 2 * PHINodeFoldingThreshold)
return false;
// Account for the cost of an unfolded ConstantExpr which could end up
// Hoist the instructions.
BB->getInstList().splice(BI, ThenBB->getInstList(), ThenBB->begin(),
- llvm::prior(ThenBB->end()));
+ std::prev(ThenBB->end()));
// Insert selects and rewrite the PHI operands.
IRBuilder<true, NoFolder> Builder(BI);
return true;
}
+/// \returns True if this block contains a CallInst with the NoDuplicate
+/// attribute.
+static bool HasNoDuplicateCall(const BasicBlock *BB) {
+ for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
+ const CallInst *CI = dyn_cast<CallInst>(I);
+ if (!CI)
+ continue;
+ if (CI->cannotDuplicate())
+ return true;
+ }
+ return false;
+}
+
/// BlockIsSimpleEnoughToThreadThrough - Return true if we can thread a branch
/// across this block.
static bool BlockIsSimpleEnoughToThreadThrough(BasicBlock *BB) {
/// that is defined in the same block as the branch and if any PHI entries are
/// constants, thread edges corresponding to that entry to be branches to their
/// ultimate destination.
-static bool FoldCondBranchOnPHI(BranchInst *BI, const DataLayout *TD) {
+static bool FoldCondBranchOnPHI(BranchInst *BI, const DataLayout *DL) {
BasicBlock *BB = BI->getParent();
PHINode *PN = dyn_cast<PHINode>(BI->getCondition());
// NOTE: we currently cannot transform this case if the PHI node is used
// Now we know that this block has multiple preds and two succs.
if (!BlockIsSimpleEnoughToThreadThrough(BB)) return false;
+ if (HasNoDuplicateCall(BB)) return false;
+
// Okay, this is a simple enough basic block. See if any phi values are
// constants.
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
}
// Check for trivial simplification.
- if (Value *V = SimplifyInstruction(N, TD)) {
+ if (Value *V = SimplifyInstruction(N, DL)) {
TranslateMap[BBI] = V;
delete N; // Instruction folded away, don't need actual inst
} else {
}
// Recurse, simplifying any other constants.
- return FoldCondBranchOnPHI(BI, TD) | true;
+ return FoldCondBranchOnPHI(BI, DL) | true;
}
return false;
/// FoldTwoEntryPHINode - Given a BB that starts with the specified two-entry
/// PHI node, see if we can eliminate it.
-static bool FoldTwoEntryPHINode(PHINode *PN, const DataLayout *TD) {
+static bool FoldTwoEntryPHINode(PHINode *PN, const DataLayout *DL) {
// Ok, this is a two entry PHI node. Check to see if this is a simple "if
// statement", which has a very simple dominance structure. Basically, we
// are trying to find the condition that is being branched on, which
for (BasicBlock::iterator II = BB->begin(); isa<PHINode>(II);) {
PHINode *PN = cast<PHINode>(II++);
- if (Value *V = SimplifyInstruction(PN, TD)) {
+ if (Value *V = SimplifyInstruction(PN, DL)) {
PN->replaceAllUsesWith(V);
PN->eraseFromParent();
continue;
// Ensure that any values used in the bonus instruction are also used
// by the terminator of the predecessor. This means that those values
// must already have been resolved, so we won't be inhibiting the
- // out-of-order core by speculating them earlier.
- if (BonusInst) {
+ // out-of-order core by speculating them earlier. We also allow
+ // instructions that are used by the terminator's condition because it
+ // exposes more merging opportunities.
+ bool UsedByBranch = (BonusInst && BonusInst->hasOneUse() &&
+ *BonusInst->use_begin() == Cond);
+
+ if (BonusInst && !UsedByBranch) {
// Collect the values used by the bonus inst
SmallPtrSet<Value*, 4> UsedValues;
for (Instruction::op_iterator OI = BonusInst->op_begin(),
OE = BonusInst->op_end(); OI != OE; ++OI) {
Value *V = *OI;
- if (!isa<Constant>(V))
+ if (!isa<Constant>(V) && !isa<Argument>(V))
UsedValues.insert(V);
}
Instruction *NewBonus = 0;
if (BonusInst) {
NewBonus = BonusInst->clone();
+
+ // If we moved a load, we cannot any longer claim any knowledge about
+ // its potential value. The previous information might have been valid
+ // only given the branch precondition.
+ // For an analogous reason, we must also drop all the metadata whose
+ // semantics we don't understand.
+ NewBonus->dropUnknownMetadata(LLVMContext::MD_dbg);
+
PredBlock->getInstList().insert(PBI, NewBonus);
NewBonus->takeName(BonusInst);
BonusInst->setName(BonusInst->getName()+".old");
/// the PHI, merging the third icmp into the switch.
static bool TryToSimplifyUncondBranchWithICmpInIt(
ICmpInst *ICI, IRBuilder<> &Builder, const TargetTransformInfo &TTI,
- const DataLayout *TD) {
+ const DataLayout *DL) {
BasicBlock *BB = ICI->getParent();
// If the block has any PHIs in it or the icmp has multiple uses, it is too
assert(VVal && "Should have a unique destination value");
ICI->setOperand(0, VVal);
- if (Value *V = SimplifyInstruction(ICI, TD)) {
+ if (Value *V = SimplifyInstruction(ICI, DL)) {
ICI->replaceAllUsesWith(V);
ICI->eraseFromParent();
}
// BB is now empty, so it is likely to simplify away.
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
}
// Ok, the block is reachable from the default dest. If the constant we're
ICI->replaceAllUsesWith(V);
ICI->eraseFromParent();
// BB is now empty, so it is likely to simplify away.
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
}
// The use of the icmp has to be in the 'end' block, by the only PHI node in
/// SimplifyBranchOnICmpChain - The specified branch is a conditional branch.
/// Check to see if it is branching on an or/and chain of icmp instructions, and
/// fold it into a switch instruction if so.
-static bool SimplifyBranchOnICmpChain(BranchInst *BI, const DataLayout *TD,
+static bool SimplifyBranchOnICmpChain(BranchInst *BI, const DataLayout *DL,
IRBuilder<> &Builder) {
Instruction *Cond = dyn_cast<Instruction>(BI->getCondition());
if (Cond == 0) return false;
unsigned UsedICmps = 0;
if (Cond->getOpcode() == Instruction::Or) {
- CompVal = GatherConstantCompares(Cond, Values, ExtraCase, TD, true,
+ CompVal = GatherConstantCompares(Cond, Values, ExtraCase, DL, true,
UsedICmps);
} else if (Cond->getOpcode() == Instruction::And) {
- CompVal = GatherConstantCompares(Cond, Values, ExtraCase, TD, false,
+ CompVal = GatherConstantCompares(Cond, Values, ExtraCase, DL, false,
UsedICmps);
TrueWhenEqual = false;
}
Builder.SetInsertPoint(BI);
// Convert pointer to int before we switch.
if (CompVal->getType()->isPointerTy()) {
- assert(TD && "Cannot switch on pointer without DataLayout");
+ assert(DL && "Cannot switch on pointer without DataLayout");
CompVal = Builder.CreatePtrToInt(CompVal,
- TD->getIntPtrType(CompVal->getContext()),
+ DL->getIntPtrType(CompVal->getType()),
"magicptr");
}
/// and use it to remove dead cases.
static bool EliminateDeadSwitchCases(SwitchInst *SI) {
Value *Cond = SI->getCondition();
- unsigned Bits = cast<IntegerType>(Cond->getType())->getBitWidth();
+ unsigned Bits = Cond->getType()->getIntegerBitWidth();
APInt KnownZero(Bits, 0), KnownOne(Bits, 0);
ComputeMaskedBits(Cond, KnownZero, KnownOne);
Case.getCaseSuccessor()->removePredecessor(SI->getParent());
SI->removeCase(Case);
}
- if (HasWeight) {
+ if (HasWeight && Weights.size() >= 2) {
SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end());
SI->setMetadata(LLVMContext::MD_prof,
MDBuilder(SI->getParent()->getContext()).
for (ForwardingNodesMap::iterator I = ForwardingNodes.begin(),
E = ForwardingNodes.end(); I != E; ++I) {
PHINode *Phi = I->first;
- SmallVector<int,4> &Indexes = I->second;
+ SmallVectorImpl<int> &Indexes = I->second;
if (Indexes.size() < 2) continue;
/// simple instructions such as binary operations where both operands are
/// constant or can be replaced by constants from the ConstantPool. Returns the
/// resulting constant on success, 0 otherwise.
-static Constant *ConstantFold(Instruction *I,
- const SmallDenseMap<Value*, Constant*>& ConstantPool) {
- if (BinaryOperator *BO = dyn_cast<BinaryOperator>(I)) {
- Constant *A = LookupConstant(BO->getOperand(0), ConstantPool);
- if (!A)
- return 0;
- Constant *B = LookupConstant(BO->getOperand(1), ConstantPool);
- if (!B)
- return 0;
- return ConstantExpr::get(BO->getOpcode(), A, B);
- }
-
- if (CmpInst *Cmp = dyn_cast<CmpInst>(I)) {
- Constant *A = LookupConstant(I->getOperand(0), ConstantPool);
- if (!A)
- return 0;
- Constant *B = LookupConstant(I->getOperand(1), ConstantPool);
- if (!B)
- return 0;
- return ConstantExpr::getCompare(Cmp->getPredicate(), A, B);
- }
-
+static Constant *
+ConstantFold(Instruction *I,
+ const SmallDenseMap<Value *, Constant *> &ConstantPool,
+ const DataLayout *DL) {
if (SelectInst *Select = dyn_cast<SelectInst>(I)) {
Constant *A = LookupConstant(Select->getCondition(), ConstantPool);
if (!A)
return 0;
}
- if (CastInst *Cast = dyn_cast<CastInst>(I)) {
- Constant *A = LookupConstant(I->getOperand(0), ConstantPool);
- if (!A)
+ SmallVector<Constant *, 4> COps;
+ for (unsigned N = 0, E = I->getNumOperands(); N != E; ++N) {
+ if (Constant *A = LookupConstant(I->getOperand(N), ConstantPool))
+ COps.push_back(A);
+ else
return 0;
- return ConstantExpr::getCast(Cast->getOpcode(), A, Cast->getDestTy());
}
- return 0;
+ if (CmpInst *Cmp = dyn_cast<CmpInst>(I))
+ return ConstantFoldCompareInstOperands(Cmp->getPredicate(), COps[0],
+ COps[1], DL);
+
+ return ConstantFoldInstOperands(I->getOpcode(), I->getType(), COps, DL);
}
/// GetCaseResults - Try to determine the resulting constant values in phi nodes
/// at the common destination basic block, *CommonDest, for one of the case
/// destionations CaseDest corresponding to value CaseVal (0 for the default
/// case), of a switch instruction SI.
-static bool GetCaseResults(SwitchInst *SI,
- ConstantInt *CaseVal,
- BasicBlock *CaseDest,
- BasicBlock **CommonDest,
- SmallVector<std::pair<PHINode*,Constant*>, 4> &Res) {
+static bool
+GetCaseResults(SwitchInst *SI,
+ ConstantInt *CaseVal,
+ BasicBlock *CaseDest,
+ BasicBlock **CommonDest,
+ SmallVectorImpl<std::pair<PHINode *, Constant *> > &Res,
+ const DataLayout *DL) {
// The block from which we enter the common destination.
BasicBlock *Pred = SI->getParent();
} else if (isa<DbgInfoIntrinsic>(I)) {
// Skip debug intrinsic.
continue;
- } else if (Constant *C = ConstantFold(I, ConstantPool)) {
+ } else if (Constant *C = ConstantFold(I, ConstantPool, DL)) {
// Instruction is side-effect free and constant.
ConstantPool.insert(std::make_pair(I, C));
} else {
Res.push_back(std::make_pair(PHI, ConstVal));
}
- return true;
+ return Res.size() > 0;
}
namespace {
SwitchLookupTable(Module &M,
uint64_t TableSize,
ConstantInt *Offset,
- const SmallVector<std::pair<ConstantInt*, Constant*>, 4>& Values,
+ const SmallVectorImpl<std::pair<ConstantInt*, Constant*> >& Values,
Constant *DefaultValue,
- const DataLayout *TD);
+ const DataLayout *DL);
/// BuildLookup - Build instructions with Builder to retrieve the value at
/// the position given by Index in the lookup table.
/// WouldFitInRegister - Return true if a table with TableSize elements of
/// type ElementType would fit in a target-legal register.
- static bool WouldFitInRegister(const DataLayout *TD,
+ static bool WouldFitInRegister(const DataLayout *DL,
uint64_t TableSize,
const Type *ElementType);
SwitchLookupTable::SwitchLookupTable(Module &M,
uint64_t TableSize,
ConstantInt *Offset,
- const SmallVector<std::pair<ConstantInt*, Constant*>, 4>& Values,
+ const SmallVectorImpl<std::pair<ConstantInt*, Constant*> >& Values,
Constant *DefaultValue,
- const DataLayout *TD)
+ const DataLayout *DL)
: SingleValue(0), BitMap(0), BitMapElementTy(0), Array(0) {
assert(Values.size() && "Can't build lookup table without values!");
assert(TableSize >= Values.size() && "Can't fit values in table!");
// If all values in the table are equal, this is that value.
SingleValue = Values.begin()->second;
+ Type *ValueType = Values.begin()->second->getType();
+
// Build up the table contents.
SmallVector<Constant*, 64> TableContents(TableSize);
for (size_t I = 0, E = Values.size(); I != E; ++I) {
ConstantInt *CaseVal = Values[I].first;
Constant *CaseRes = Values[I].second;
- assert(CaseRes->getType() == DefaultValue->getType());
+ assert(CaseRes->getType() == ValueType);
uint64_t Idx = (CaseVal->getValue() - Offset->getValue())
.getLimitedValue();
// Fill in any holes in the table with the default result.
if (Values.size() < TableSize) {
+ assert(DefaultValue && "Need a default value to fill the lookup table holes.");
+ assert(DefaultValue->getType() == ValueType);
for (uint64_t I = 0; I < TableSize; ++I) {
if (!TableContents[I])
TableContents[I] = DefaultValue;
}
// If the type is integer and the table fits in a register, build a bitmap.
- if (WouldFitInRegister(TD, TableSize, DefaultValue->getType())) {
- IntegerType *IT = cast<IntegerType>(DefaultValue->getType());
+ if (WouldFitInRegister(DL, TableSize, ValueType)) {
+ IntegerType *IT = cast<IntegerType>(ValueType);
APInt TableInt(TableSize * IT->getBitWidth(), 0);
for (uint64_t I = TableSize; I > 0; --I) {
TableInt <<= IT->getBitWidth();
}
// Store the table in an array.
- ArrayType *ArrayTy = ArrayType::get(DefaultValue->getType(), TableSize);
+ ArrayType *ArrayTy = ArrayType::get(ValueType, TableSize);
Constant *Initializer = ConstantArray::get(ArrayTy, TableContents);
Array = new GlobalVariable(M, ArrayTy, /*constant=*/ true,
llvm_unreachable("Unknown lookup table kind!");
}
-bool SwitchLookupTable::WouldFitInRegister(const DataLayout *TD,
+bool SwitchLookupTable::WouldFitInRegister(const DataLayout *DL,
uint64_t TableSize,
const Type *ElementType) {
- if (!TD)
+ if (!DL)
return false;
const IntegerType *IT = dyn_cast<IntegerType>(ElementType);
if (!IT)
// Avoid overflow, fitsInLegalInteger uses unsigned int for the width.
if (TableSize >= UINT_MAX/IT->getBitWidth())
return false;
- return TD->fitsInLegalInteger(TableSize * IT->getBitWidth());
+ return DL->fitsInLegalInteger(TableSize * IT->getBitWidth());
}
/// ShouldBuildLookupTable - Determine whether a lookup table should be built
static bool ShouldBuildLookupTable(SwitchInst *SI,
uint64_t TableSize,
const TargetTransformInfo &TTI,
- const DataLayout *TD,
+ const DataLayout *DL,
const SmallDenseMap<PHINode*, Type*>& ResultTypes) {
if (SI->getNumCases() > TableSize || TableSize >= UINT64_MAX / 10)
return false; // TableSize overflowed, or mul below might overflow.
// Saturate this flag to false.
AllTablesFitInRegister = AllTablesFitInRegister &&
- SwitchLookupTable::WouldFitInRegister(TD, TableSize, Ty);
+ SwitchLookupTable::WouldFitInRegister(DL, TableSize, Ty);
// If both flags saturate, we're done. NOTE: This *only* works with
// saturating flags, and all flags have to saturate first due to the
static bool SwitchToLookupTable(SwitchInst *SI,
IRBuilder<> &Builder,
const TargetTransformInfo &TTI,
- const DataLayout* TD) {
+ const DataLayout* DL) {
assert(SI->getNumCases() > 1 && "Degenerate switch?");
// Only build lookup table when we have a target that supports it.
// GEP needs a runtime relocation in PIC code. We should just build one big
// string and lookup indices into that.
- // Ignore the switch if the number of cases is too small.
- // This is similar to the check when building jump tables in
- // SelectionDAGBuilder::handleJTSwitchCase.
- // FIXME: Determine the best cut-off.
- if (SI->getNumCases() < 4)
+ // Ignore switches with less than three cases. Lookup tables will not make them
+ // faster, so we don't analyze them.
+ if (SI->getNumCases() < 3)
return false;
// Figure out the corresponding result for each case value and phi node in the
typedef SmallVector<std::pair<PHINode*, Constant*>, 4> ResultsTy;
ResultsTy Results;
if (!GetCaseResults(SI, CaseVal, CI.getCaseSuccessor(), &CommonDest,
- Results))
+ Results, DL))
return false;
// Append the result from this case to the list for each phi.
}
}
- // Get the resulting values for the default case.
+ // Keep track of the result types.
+ for (size_t I = 0, E = PHIs.size(); I != E; ++I) {
+ PHINode *PHI = PHIs[I];
+ ResultTypes[PHI] = ResultLists[PHI][0].second->getType();
+ }
+
+ uint64_t NumResults = ResultLists[PHIs[0]].size();
+ APInt RangeSpread = MaxCaseVal->getValue() - MinCaseVal->getValue();
+ uint64_t TableSize = RangeSpread.getLimitedValue() + 1;
+ bool TableHasHoles = (NumResults < TableSize);
+
+ // If the table has holes, we need a constant result for the default case.
SmallVector<std::pair<PHINode*, Constant*>, 4> DefaultResultsList;
- if (!GetCaseResults(SI, 0, SI->getDefaultDest(), &CommonDest,
- DefaultResultsList))
+ if (TableHasHoles && !GetCaseResults(SI, 0, SI->getDefaultDest(), &CommonDest,
+ DefaultResultsList, DL))
return false;
+
for (size_t I = 0, E = DefaultResultsList.size(); I != E; ++I) {
PHINode *PHI = DefaultResultsList[I].first;
Constant *Result = DefaultResultsList[I].second;
DefaultResults[PHI] = Result;
- ResultTypes[PHI] = Result->getType();
}
- APInt RangeSpread = MaxCaseVal->getValue() - MinCaseVal->getValue();
- uint64_t TableSize = RangeSpread.getLimitedValue() + 1;
- if (!ShouldBuildLookupTable(SI, TableSize, TTI, TD, ResultTypes))
+ if (!ShouldBuildLookupTable(SI, TableSize, TTI, DL, ResultTypes))
return false;
// Create the BB that does the lookups.
CommonDest->getParent(),
CommonDest);
- // Check whether the condition value is within the case range, and branch to
- // the new BB.
+ // Compute the table index value.
Builder.SetInsertPoint(SI);
Value *TableIndex = Builder.CreateSub(SI->getCondition(), MinCaseVal,
"switch.tableidx");
- Value *Cmp = Builder.CreateICmpULT(TableIndex, ConstantInt::get(
- MinCaseVal->getType(), TableSize));
- Builder.CreateCondBr(Cmp, LookupBB, SI->getDefaultDest());
+
+ // Compute the maximum table size representable by the integer type we are
+ // switching upon.
+ unsigned CaseSize = MinCaseVal->getType()->getPrimitiveSizeInBits();
+ uint64_t MaxTableSize = CaseSize > 63 ? UINT64_MAX : 1ULL << CaseSize;
+ assert(MaxTableSize >= TableSize &&
+ "It is impossible for a switch to have more entries than the max "
+ "representable value of its input integer type's size.");
+
+ // If we have a fully covered lookup table, unconditionally branch to the
+ // lookup table BB. Otherwise, check if the condition value is within the case
+ // range. If it is so, branch to the new BB. Otherwise branch to SI's default
+ // destination.
+ const bool GeneratingCoveredLookupTable = MaxTableSize == TableSize;
+ if (GeneratingCoveredLookupTable) {
+ Builder.CreateBr(LookupBB);
+ SI->getDefaultDest()->removePredecessor(SI->getParent());
+ } else {
+ Value *Cmp = Builder.CreateICmpULT(TableIndex, ConstantInt::get(
+ MinCaseVal->getType(), TableSize));
+ Builder.CreateCondBr(Cmp, LookupBB, SI->getDefaultDest());
+ }
// Populate the BB that does the lookups.
Builder.SetInsertPoint(LookupBB);
PHINode *PHI = PHIs[I];
SwitchLookupTable Table(Mod, TableSize, MinCaseVal, ResultLists[PHI],
- DefaultResults[PHI], TD);
+ DefaultResults[PHI], DL);
Value *Result = Table.BuildLookup(TableIndex, Builder);
Builder.CreateBr(CommonDest);
// Remove the switch.
- for (unsigned i = 0; i < SI->getNumSuccessors(); ++i) {
+ for (unsigned i = 0, e = SI->getNumSuccessors(); i < e; ++i) {
BasicBlock *Succ = SI->getSuccessor(i);
- if (Succ == SI->getDefaultDest()) continue;
+
+ if (Succ == SI->getDefaultDest())
+ continue;
Succ->removePredecessor(SI->getParent());
}
SI->eraseFromParent();
// see if that predecessor totally determines the outcome of this switch.
if (BasicBlock *OnlyPred = BB->getSinglePredecessor())
if (SimplifyEqualityComparisonWithOnlyPredecessor(SI, OnlyPred, Builder))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
Value *Cond = SI->getCondition();
if (SelectInst *Select = dyn_cast<SelectInst>(Cond))
if (SimplifySwitchOnSelect(SI, Select))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
// If the block only contains the switch, see if we can fold the block
// away into any preds.
++BBI;
if (SI == &*BBI)
if (FoldValueComparisonIntoPredecessors(SI, Builder))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
}
// Try to transform the switch into an icmp and a branch.
if (TurnSwitchRangeIntoICmp(SI, Builder))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
// Remove unreachable cases.
if (EliminateDeadSwitchCases(SI))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
if (ForwardSwitchConditionToPHI(SI))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
- if (SwitchToLookupTable(SI, Builder, TTI, TD))
- return SimplifyCFG(BB, TTI, TD) | true;
+ if (SwitchToLookupTable(SI, Builder, TTI, DL))
+ return SimplifyCFG(BB, TTI, DL) | true;
return false;
}
if (SelectInst *SI = dyn_cast<SelectInst>(IBI->getAddress())) {
if (SimplifyIndirectBrOnSelect(IBI, SI))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
}
return Changed;
}
for (++I; isa<DbgInfoIntrinsic>(I); ++I)
;
if (I->isTerminator() &&
- TryToSimplifyUncondBranchWithICmpInIt(ICI, Builder, TTI, TD))
+ TryToSimplifyUncondBranchWithICmpInIt(ICI, Builder, TTI, DL))
return true;
}
// predecessor and use logical operations to update the incoming value
// for PHI nodes in common successor.
if (FoldBranchToCommonDest(BI))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
return false;
}
// switch.
if (BasicBlock *OnlyPred = BB->getSinglePredecessor())
if (SimplifyEqualityComparisonWithOnlyPredecessor(BI, OnlyPred, Builder))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
// This block must be empty, except for the setcond inst, if it exists.
// Ignore dbg intrinsics.
++I;
if (&*I == BI) {
if (FoldValueComparisonIntoPredecessors(BI, Builder))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
} else if (&*I == cast<Instruction>(BI->getCondition())){
++I;
// Ignore dbg intrinsics.
while (isa<DbgInfoIntrinsic>(I))
++I;
if (&*I == BI && FoldValueComparisonIntoPredecessors(BI, Builder))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
}
}
// Try to turn "br (X == 0 | X == 1), T, F" into a switch instruction.
- if (SimplifyBranchOnICmpChain(BI, TD, Builder))
+ if (SimplifyBranchOnICmpChain(BI, DL, Builder))
return true;
// If this basic block is ONLY a compare and a branch, and if a predecessor
// branches to us and one of our successors, fold the comparison into the
// predecessor and use logical operations to pick the right destination.
if (FoldBranchToCommonDest(BI))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
// We have a conditional branch to two blocks that are only reachable
// from BI. We know that the condbr dominates the two blocks, so see if
if (BI->getSuccessor(0)->getSinglePredecessor() != 0) {
if (BI->getSuccessor(1)->getSinglePredecessor() != 0) {
if (HoistThenElseCodeToIf(BI))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
} else {
// If Successor #1 has multiple preds, we may be able to conditionally
// execute Successor #0 if it branches to successor #1.
if (Succ0TI->getNumSuccessors() == 1 &&
Succ0TI->getSuccessor(0) == BI->getSuccessor(1))
if (SpeculativelyExecuteBB(BI, BI->getSuccessor(0)))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
}
} else if (BI->getSuccessor(1)->getSinglePredecessor() != 0) {
// If Successor #0 has multiple preds, we may be able to conditionally
if (Succ1TI->getNumSuccessors() == 1 &&
Succ1TI->getSuccessor(0) == BI->getSuccessor(0))
if (SpeculativelyExecuteBB(BI, BI->getSuccessor(1)))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
}
// If this is a branch on a phi node in the current block, thread control
// through this block if any PHI node entries are constants.
if (PHINode *PN = dyn_cast<PHINode>(BI->getCondition()))
if (PN->getParent() == BI->getParent())
- if (FoldCondBranchOnPHI(BI, TD))
- return SimplifyCFG(BB, TTI, TD) | true;
+ if (FoldCondBranchOnPHI(BI, DL))
+ return SimplifyCFG(BB, TTI, DL) | true;
// Scan predecessor blocks for conditional branches.
for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI)
if (BranchInst *PBI = dyn_cast<BranchInst>((*PI)->getTerminator()))
if (PBI != BI && PBI->isConditional())
if (SimplifyCondBranchToCondBranch(PBI, BI))
- return SimplifyCFG(BB, TTI, TD) | true;
+ return SimplifyCFG(BB, TTI, DL) | true;
return false;
}
// eliminate it, do so now.
if (PHINode *PN = dyn_cast<PHINode>(BB->begin()))
if (PN->getNumIncomingValues() == 2)
- Changed |= FoldTwoEntryPHINode(PN, TD);
+ Changed |= FoldTwoEntryPHINode(PN, DL);
Builder.SetInsertPoint(BB->getTerminator());
if (BranchInst *BI = dyn_cast<BranchInst>(BB->getTerminator())) {
/// of the CFG. It returns true if a modification was made.
///
bool llvm::SimplifyCFG(BasicBlock *BB, const TargetTransformInfo &TTI,
- const DataLayout *TD) {
- return SimplifyCFGOpt(TTI, TD).run(BB);
+ const DataLayout *DL) {
+ return SimplifyCFGOpt(TTI, DL).run(BB);
}