blob: 1c62559739f9bd34c7e5b73fe36510cdd3f561ce [file] [log] [blame]
//===- SimplifyCFG.cpp - Code to perform CFG simplification ---------------===//
// The LLVM Compiler Infrastructure
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
// Peephole optimize the CFG.
#include "llvm/Transforms/Utils/Local.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SetVector.h"
#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/GlobalVariable.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/LLVMContext.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/CommandLine.h"
#include "llvm/Support/Debug.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;
#define DEBUG_TYPE "simplifycfg"
static cl::opt<unsigned>
PHINodeFoldingThreshold("phi-node-folding-threshold", cl::Hidden, cl::init(1),
cl::desc("Control the amount of phi node folding to perform (default = 1)"));
static cl::opt<bool>
DupRet("simplifycfg-dup-ret", cl::Hidden, cl::init(false),
cl::desc("Duplicate return instructions into unconditional branches"));
static cl::opt<bool>
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 precedes"));
STATISTIC(NumBitMaps, "Number of switch instructions turned into bitmaps");
STATISTIC(NumLookupTables, "Number of switch instructions turned into lookup tables");
STATISTIC(NumLookupTablesHoles, "Number of switch instructions turned into lookup tables (holes checked)");
STATISTIC(NumSinkCommons, "Number of common instructions sunk down to the end block");
STATISTIC(NumSpeculations, "Number of speculative executed instructions");
namespace {
/// ValueEqualityComparisonCase - Represents a case of a switch.
struct ValueEqualityComparisonCase {
ConstantInt *Value;
BasicBlock *Dest;
ValueEqualityComparisonCase(ConstantInt *Value, BasicBlock *Dest)
: Value(Value), Dest(Dest) {}
bool operator<(ValueEqualityComparisonCase RHS) const {
// Comparing pointers is ok as we only rely on the order for uniquing.
return Value < RHS.Value;
bool operator==(BasicBlock *RHSDest) const { return Dest == RHSDest; }
class SimplifyCFGOpt {
const TargetTransformInfo &TTI;
const DataLayout *const DL;
Value *isValueEqualityComparison(TerminatorInst *TI);
BasicBlock *GetValueEqualityComparisonCases(TerminatorInst *TI,
std::vector<ValueEqualityComparisonCase> &Cases);
bool SimplifyEqualityComparisonWithOnlyPredecessor(TerminatorInst *TI,
BasicBlock *Pred,
IRBuilder<> &Builder);
bool FoldValueComparisonIntoPredecessors(TerminatorInst *TI,
IRBuilder<> &Builder);
bool SimplifyReturn(ReturnInst *RI, IRBuilder<> &Builder);
bool SimplifyResume(ResumeInst *RI, IRBuilder<> &Builder);
bool SimplifyUnreachable(UnreachableInst *UI);
bool SimplifySwitch(SwitchInst *SI, IRBuilder<> &Builder);
bool SimplifyIndirectBr(IndirectBrInst *IBI);
bool SimplifyUncondBranch(BranchInst *BI, IRBuilder <> &Builder);
bool SimplifyCondBranch(BranchInst *BI, IRBuilder <>&Builder);
SimplifyCFGOpt(const TargetTransformInfo &TTI, const DataLayout *DL)
: TTI(TTI), DL(DL) {}
bool run(BasicBlock *BB);
/// SafeToMergeTerminators - Return true if it is safe to merge these two
/// terminator instructions together.
static bool SafeToMergeTerminators(TerminatorInst *SI1, TerminatorInst *SI2) {
if (SI1 == SI2) return false; // Can't merge with self!
// It is not safe to merge these two switch instructions if they have a common
// successor, and if that successor has a PHI node, and if *that* PHI node has
// conflicting incoming values from the two switch blocks.
BasicBlock *SI1BB = SI1->getParent();
BasicBlock *SI2BB = SI2->getParent();
SmallPtrSet<BasicBlock*, 16> SI1Succs(succ_begin(SI1BB), succ_end(SI1BB));
for (succ_iterator I = succ_begin(SI2BB), E = succ_end(SI2BB); I != E; ++I)
if (SI1Succs.count(*I))
for (BasicBlock::iterator BBI = (*I)->begin();
isa<PHINode>(BBI); ++BBI) {
PHINode *PN = cast<PHINode>(BBI);
if (PN->getIncomingValueForBlock(SI1BB) !=
return false;
return true;
/// isProfitableToFoldUnconditional - Return true if it is safe and profitable
/// to merge these two terminator instructions together, where SI1 is an
/// unconditional branch. PhiNodes will store all PHI nodes in common
/// successors.
static bool isProfitableToFoldUnconditional(BranchInst *SI1,
BranchInst *SI2,
Instruction *Cond,
SmallVectorImpl<PHINode*> &PhiNodes) {
if (SI1 == SI2) return false; // Can't merge with self!
assert(SI1->isUnconditional() && SI2->isConditional());
// We fold the unconditional branch if we can easily update all PHI nodes in
// common successors:
// 1> We have a constant incoming value for the conditional branch;
// 2> We have "Cond" as the incoming value for the unconditional branch;
// 3> SI2->getCondition() and Cond have same operands.
CmpInst *Ci2 = dyn_cast<CmpInst>(SI2->getCondition());
if (!Ci2) return false;
if (!(Cond->getOperand(0) == Ci2->getOperand(0) &&
Cond->getOperand(1) == Ci2->getOperand(1)) &&
!(Cond->getOperand(0) == Ci2->getOperand(1) &&
Cond->getOperand(1) == Ci2->getOperand(0)))
return false;
BasicBlock *SI1BB = SI1->getParent();
BasicBlock *SI2BB = SI2->getParent();
SmallPtrSet<BasicBlock*, 16> SI1Succs(succ_begin(SI1BB), succ_end(SI1BB));
for (succ_iterator I = succ_begin(SI2BB), E = succ_end(SI2BB); I != E; ++I)
if (SI1Succs.count(*I))
for (BasicBlock::iterator BBI = (*I)->begin();
isa<PHINode>(BBI); ++BBI) {
PHINode *PN = cast<PHINode>(BBI);
if (PN->getIncomingValueForBlock(SI1BB) != Cond ||
return false;
return true;
/// AddPredecessorToBlock - Update PHI nodes in Succ to indicate that there will
/// now be entries in it from the 'NewPred' block. The values that will be
/// flowing into the PHI nodes will be the same as those coming in from
/// ExistPred, an existing predecessor of Succ.
static void AddPredecessorToBlock(BasicBlock *Succ, BasicBlock *NewPred,
BasicBlock *ExistPred) {
if (!isa<PHINode>(Succ->begin())) return; // Quick exit if nothing to do
PHINode *PN;
for (BasicBlock::iterator I = Succ->begin();
(PN = dyn_cast<PHINode>(I)); ++I)
PN->addIncoming(PN->getIncomingValueForBlock(ExistPred), NewPred);
/// 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.
static unsigned ComputeSpeculationCost(const User *I, const DataLayout *DL) {
assert(isSafeToSpeculativelyExecute(I, DL) &&
"Instruction is not safe to speculatively execute!");
switch (Operator::getOpcode(I)) {
// In doubt, be conservative.
return UINT_MAX;
case Instruction::GetElementPtr:
// GEPs are cheap if all indices are constant.
if (!cast<GEPOperator>(I)->hasAllConstantIndices())
return UINT_MAX;
return 1;
case Instruction::ExtractValue:
case Instruction::Load:
case Instruction::Add:
case Instruction::Sub:
case Instruction::And:
case Instruction::Or:
case Instruction::Xor:
case Instruction::Shl:
case Instruction::LShr:
case Instruction::AShr:
case Instruction::ICmp:
case Instruction::Trunc:
case Instruction::ZExt:
case Instruction::SExt:
case Instruction::BitCast:
case Instruction::ExtractElement:
case Instruction::InsertElement:
return 1; // These are all cheap.
case Instruction::Call:
case Instruction::Select:
return 2;
/// DominatesMergePoint - If we have a merge point of an "if condition" as
/// accepted above, return true if the specified value dominates the block. We
/// don't handle the true generality of domination here, just a special case
/// which works well enough for us.
/// If AggressiveInsts is non-null, and if V does not dominate BB, we check to
/// see if V (which must be an instruction) and its recursive operands
/// that do not dominate BB have a combined cost lower than CostRemaining and
/// are non-trapping. If both are true, the instruction is inserted into the
/// set and true is returned.
/// The cost for most non-trapping instructions is defined as 1 except for
/// Select whose cost is 2.
/// After this function returns, CostRemaining is decreased by the cost of
/// V plus its non-dominating operands. If that cost is greater than
/// CostRemaining, false is returned and CostRemaining is undefined.
static bool DominatesMergePoint(Value *V, BasicBlock *BB,
SmallPtrSet<Instruction*, 4> *AggressiveInsts,
unsigned &CostRemaining,
const DataLayout *DL) {
Instruction *I = dyn_cast<Instruction>(V);
if (!I) {
// Non-instructions all dominate instructions, but not all constantexprs
// can be executed unconditionally.
if (ConstantExpr *C = dyn_cast<ConstantExpr>(V))
if (C->canTrap())
return false;
return true;
BasicBlock *PBB = I->getParent();
// We don't want to allow weird loops that might have the "if condition" in
// the bottom of this block.
if (PBB == BB) return false;
// If this instruction is defined in a block that contains an unconditional
// branch to BB, then it must be in the 'conditional' part of the "if
// statement". If not, it definitely dominates the region.
BranchInst *BI = dyn_cast<BranchInst>(PBB->getTerminator());
if (!BI || BI->isConditional() || BI->getSuccessor(0) != BB)
return true;
// If we aren't allowing aggressive promotion anymore, then don't consider
// instructions in the 'if region'.
if (!AggressiveInsts) return false;
// If we have seen this instruction before, don't count it again.
if (AggressiveInsts->count(I)) return true;
// Okay, it looks like the instruction IS in the "condition". Check to
// see if it's a cheap instruction to unconditionally compute, and if it
// only uses stuff defined outside of the condition. If so, hoist it out.
if (!isSafeToSpeculativelyExecute(I, DL))
return false;
unsigned Cost = ComputeSpeculationCost(I, DL);
if (Cost > CostRemaining)
return false;
CostRemaining -= Cost;
// Okay, we can only really hoist these out if their operands do
// not take us over the cost threshold.
for (User::op_iterator i = I->op_begin(), e = I->op_end(); i != e; ++i)
if (!DominatesMergePoint(*i, BB, AggressiveInsts, CostRemaining, DL))
return false;
// Okay, it's safe to do this! Remember this instruction.
return true;
/// 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 *DL) {
// Normal constant int.
ConstantInt *CI = dyn_cast<ConstantInt>(V);
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>(DL->getIntPtrType(V->getType()));
// Null pointer means 0, see SelectionDAGBuilder::getValue(const Value*).
if (isa<ConstantPointerNull>(V))
return ConstantInt::get(PtrTy, 0);
// IntToPtr const int.
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
if (CE->getOpcode() == Instruction::IntToPtr)
if (ConstantInt *CI = dyn_cast<ConstantInt>(CE->getOperand(0))) {
// The constant is very likely to have the right type already.
if (CI->getType() == PtrTy)
return CI;
return cast<ConstantInt>
(ConstantExpr::getIntegerCast(CI, PtrTy, /*isSigned=*/false));
return nullptr;
/// GatherConstantCompares - Given a potentially 'or'd or 'and'd together
/// collection of icmp eq/ne instructions that compare a value against a
/// constant, return the value being compared, and stick the constant into the
/// Values vector.
static Value *
GatherConstantCompares(Value *V, std::vector<ConstantInt*> &Vals, Value *&Extra,
const DataLayout *DL, bool isEQ, unsigned &UsedICmps) {
Instruction *I = dyn_cast<Instruction>(V);
if (!I) return nullptr;
// 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), 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()) {
ConstantInt::get(C->getContext(), C->getValue() | Not));
return RHSVal;
return I->getOperand(0);
// If we have "x ult 3" comparison, for example, then we can add 0,1,2 to
// the set.
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)
Span = Span.inverse();
// If there are a ton of values, we don't want to make a ginormous switch.
if (Span.getSetSize().ugt(8) || Span.isEmptySet())
return nullptr;
for (APInt Tmp = Span.getLower(); Tmp != Span.getUpper(); ++Tmp)
Vals.push_back(ConstantInt::get(V->getContext(), Tmp));
return hasAdd ? RHSVal : I->getOperand(0);
return nullptr;
// Otherwise, we can only handle an | or &, depending on isEQ.
if (I->getOpcode() != (isEQ ? Instruction::Or : Instruction::And))
return nullptr;
unsigned NumValsBeforeLHS = Vals.size();
unsigned UsedICmpsBeforeLHS = UsedICmps;
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, DL,
isEQ, UsedICmps)) {
if (LHS == RHS)
return LHS;
UsedICmps = UsedICmpsBeforeRHS;
// The RHS of the or/and can't be folded in and we haven't used "Extra" yet,
// set it and return success.
if (Extra == nullptr || Extra == I->getOperand(1)) {
Extra = I->getOperand(1);
return LHS;
UsedICmps = UsedICmpsBeforeLHS;
return nullptr;
// If the LHS can't be folded in, but Extra is available and RHS can, try to
// use LHS as Extra.
if (Extra == nullptr || Extra == I->getOperand(0)) {
Value *OldExtra = Extra;
Extra = I->getOperand(0);
if (Value *RHS = GatherConstantCompares(I->getOperand(1), Vals, Extra, DL,
isEQ, UsedICmps))
return RHS;
assert(Vals.size() == NumValsBeforeLHS);
Extra = OldExtra;
return nullptr;
static void EraseTerminatorInstAndDCECond(TerminatorInst *TI) {
Instruction *Cond = nullptr;
if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
Cond = dyn_cast<Instruction>(SI->getCondition());
} else if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
if (BI->isConditional())
Cond = dyn_cast<Instruction>(BI->getCondition());
} else if (IndirectBrInst *IBI = dyn_cast<IndirectBrInst>(TI)) {
Cond = dyn_cast<Instruction>(IBI->getAddress());
if (Cond) RecursivelyDeleteTriviallyDeadInstructions(Cond);
/// isValueEqualityComparison - Return true if the specified terminator checks
/// to see if a value is equal to constant integer value.
Value *SimplifyCFGOpt::isValueEqualityComparison(TerminatorInst *TI) {
Value *CV = nullptr;
if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
// Do not permit merging of large switch instructions into their
// predecessors unless there is only one predecessor.
if (SI->getNumSuccessors()*std::distance(pred_begin(SI->getParent()),
pred_end(SI->getParent())) <= 128)
CV = SI->getCondition();
} 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), DL))
CV = ICI->getOperand(0);
// Unwrap any lossless ptrtoint cast.
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;
/// GetValueEqualityComparisonCases - Given a value comparison instruction,
/// decode all of the 'cases' that it represents and return the 'default' block.
BasicBlock *SimplifyCFGOpt::
GetValueEqualityComparisonCases(TerminatorInst *TI,
&Cases) {
if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end(); i != e; ++i)
return SI->getDefaultDest();
BranchInst *BI = cast<BranchInst>(TI);
ICmpInst *ICI = cast<ICmpInst>(BI->getCondition());
BasicBlock *Succ = BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_NE);
return BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_EQ);
/// EliminateBlockCases - Given a vector of bb/value pairs, remove any entries
/// in the list that match the specified block.
static void EliminateBlockCases(BasicBlock *BB,
std::vector<ValueEqualityComparisonCase> &Cases) {
Cases.erase(std::remove(Cases.begin(), Cases.end(), BB), Cases.end());
/// ValuesOverlap - Return true if there are any keys in C1 that exist in C2 as
/// well.
static bool
ValuesOverlap(std::vector<ValueEqualityComparisonCase> &C1,
std::vector<ValueEqualityComparisonCase > &C2) {
std::vector<ValueEqualityComparisonCase> *V1 = &C1, *V2 = &C2;
// Make V1 be smaller than V2.
if (V1->size() > V2->size())
std::swap(V1, V2);
if (V1->size() == 0) return false;
if (V1->size() == 1) {
// Just scan V2.
ConstantInt *TheVal = (*V1)[0].Value;
for (unsigned i = 0, e = V2->size(); i != e; ++i)
if (TheVal == (*V2)[i].Value)
return true;
// Otherwise, just sort both lists and compare element by element.
array_pod_sort(V1->begin(), V1->end());
array_pod_sort(V2->begin(), V2->end());
unsigned i1 = 0, i2 = 0, e1 = V1->size(), e2 = V2->size();
while (i1 != e1 && i2 != e2) {
if ((*V1)[i1].Value == (*V2)[i2].Value)
return true;
if ((*V1)[i1].Value < (*V2)[i2].Value)
return false;
/// SimplifyEqualityComparisonWithOnlyPredecessor - If TI is known to be a
/// terminator instruction and its block is known to only have a single
/// predecessor block, check to see if that predecessor is also a value
/// comparison with the same value, and if that comparison determines the
/// outcome of this comparison. If so, simplify TI. This does a very limited
/// form of jump threading.
bool SimplifyCFGOpt::
SimplifyEqualityComparisonWithOnlyPredecessor(TerminatorInst *TI,
BasicBlock *Pred,
IRBuilder<> &Builder) {
Value *PredVal = isValueEqualityComparison(Pred->getTerminator());
if (!PredVal) return false; // Not a value comparison in predecessor.
Value *ThisVal = isValueEqualityComparison(TI);
assert(ThisVal && "This isn't a value comparison!!");
if (ThisVal != PredVal) return false; // Different predicates.
// TODO: Preserve branch weight metadata, similarly to how
// FoldValueComparisonIntoPredecessors preserves it.
// Find out information about when control will move from Pred to TI's block.
std::vector<ValueEqualityComparisonCase> PredCases;
BasicBlock *PredDef = GetValueEqualityComparisonCases(Pred->getTerminator(),
EliminateBlockCases(PredDef, PredCases); // Remove default from cases.
// Find information about how control leaves this block.
std::vector<ValueEqualityComparisonCase> ThisCases;
BasicBlock *ThisDef = GetValueEqualityComparisonCases(TI, ThisCases);
EliminateBlockCases(ThisDef, ThisCases); // Remove default from cases.
// If TI's block is the default block from Pred's comparison, potentially
// simplify TI based on this knowledge.
if (PredDef == TI->getParent()) {
// If we are here, we know that the value is none of those cases listed in
// PredCases. If there are any cases in ThisCases that are in PredCases, we
// can simplify TI.
if (!ValuesOverlap(PredCases, ThisCases))
return false;
if (isa<BranchInst>(TI)) {
// Okay, one of the successors of this condbr is dead. Convert it to a
// uncond br.
assert(ThisCases.size() == 1 && "Branch can only have one case!");
// Insert the new branch.
Instruction *NI = Builder.CreateBr(ThisDef);
(void) NI;
// Remove PHI node entries for the dead edge.
DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()
<< "Through successor TI: " << *TI << "Leaving: " << *NI << "\n");
return true;
SwitchInst *SI = cast<SwitchInst>(TI);
// Okay, TI has cases that are statically dead, prune them away.
SmallPtrSet<Constant*, 16> DeadCases;
for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()
<< "Through successor TI: " << *TI);
// Collect branch weights into a vector.
SmallVector<uint32_t, 8> Weights;
MDNode* MD = SI->getMetadata(LLVMContext::MD_prof);
bool HasWeight = MD && (MD->getNumOperands() == 2 + SI->getNumCases());
if (HasWeight)
for (unsigned MD_i = 1, MD_e = MD->getNumOperands(); MD_i < MD_e;
++MD_i) {
ConstantInt* CI = dyn_cast<ConstantInt>(MD->getOperand(MD_i));
for (SwitchInst::CaseIt i = SI->case_end(), e = SI->case_begin(); i != e;) {
if (DeadCases.count(i.getCaseValue())) {
if (HasWeight) {
std::swap(Weights[i.getCaseIndex()+1], Weights.back());
if (HasWeight && Weights.size() >= 2)
DEBUG(dbgs() << "Leaving: " << *TI << "\n");
return true;
// Otherwise, TI's block must correspond to some matched value. Find out
// which value (or set of values) this is.
ConstantInt *TIV = nullptr;
BasicBlock *TIBB = TI->getParent();
for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
if (PredCases[i].Dest == TIBB) {
if (TIV)
return false; // Cannot handle multiple values coming to this block.
TIV = PredCases[i].Value;
assert(TIV && "No edge from pred to succ?");
// Okay, we found the one constant that our value can be if we get into TI's
// BB. Find out which successor will unconditionally be branched to.
BasicBlock *TheRealDest = nullptr;
for (unsigned i = 0, e = ThisCases.size(); i != e; ++i)
if (ThisCases[i].Value == TIV) {
TheRealDest = ThisCases[i].Dest;
// If not handled by any explicit cases, it is handled by the default case.
if (!TheRealDest) TheRealDest = ThisDef;
// Remove PHI node entries for dead edges.
BasicBlock *CheckEdge = TheRealDest;
for (succ_iterator SI = succ_begin(TIBB), e = succ_end(TIBB); SI != e; ++SI)
if (*SI != CheckEdge)
CheckEdge = nullptr;
// Insert the new branch.
Instruction *NI = Builder.CreateBr(TheRealDest);
(void) NI;
DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()
<< "Through successor TI: " << *TI << "Leaving: " << *NI << "\n");
return true;
namespace {
/// ConstantIntOrdering - This class implements a stable ordering of constant
/// integers that does not depend on their address. This is important for
/// applications that sort ConstantInt's to ensure uniqueness.
struct ConstantIntOrdering {
bool operator()(const ConstantInt *LHS, const ConstantInt *RHS) const {
return LHS->getValue().ult(RHS->getValue());
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())
return 0;
return -1;
static inline bool HasBranchWeights(const Instruction* I) {
MDNode* ProfMD = I->getMetadata(LLVMContext::MD_prof);
if (ProfMD && ProfMD->getOperand(0))
if (MDString* MDS = dyn_cast<MDString>(ProfMD->getOperand(0)))
return MDS->getString().equals("branch_weights");
return false;
/// Get Weights of a given TerminatorInst, the default weight is at the front
/// of the vector. If TI is a conditional eq, we need to swap the branch-weight
/// metadata.
static void GetBranchWeights(TerminatorInst *TI,
SmallVectorImpl<uint64_t> &Weights) {
MDNode* MD = TI->getMetadata(LLVMContext::MD_prof);
for (unsigned i = 1, e = MD->getNumOperands(); i < e; ++i) {
ConstantInt *CI = cast<ConstantInt>(MD->getOperand(i));
// If TI is a conditional eq, the default case is the false case,
// and the corresponding branch-weight data is at index 2. We swap the
// default weight to be the first entry.
if (BranchInst* BI = dyn_cast<BranchInst>(TI)) {
assert(Weights.size() == 2);
ICmpInst *ICI = cast<ICmpInst>(BI->getCondition());
if (ICI->getPredicate() == ICmpInst::ICMP_EQ)
std::swap(Weights.front(), Weights.back());
/// Keep halving the weights until all can fit in uint32_t.
static void FitWeights(MutableArrayRef<uint64_t> Weights) {
uint64_t Max = *std::max_element(Weights.begin(), Weights.end());
if (Max > UINT_MAX) {
unsigned Offset = 32 - countLeadingZeros(Max);
for (uint64_t &I : Weights)
I >>= Offset;
/// FoldValueComparisonIntoPredecessors - The specified terminator is a value
/// equality comparison instruction (either a switch or a branch on "X == c").
/// See if any of the predecessors of the terminator block are value comparisons
/// on the same value. If so, and if safe to do so, fold them together.
bool SimplifyCFGOpt::FoldValueComparisonIntoPredecessors(TerminatorInst *TI,
IRBuilder<> &Builder) {
BasicBlock *BB = TI->getParent();
Value *CV = isValueEqualityComparison(TI); // CondVal
assert(CV && "Not a comparison?");
bool Changed = false;
SmallVector<BasicBlock*, 16> Preds(pred_begin(BB), pred_end(BB));
while (!Preds.empty()) {
BasicBlock *Pred = Preds.pop_back_val();
// See if the predecessor is a comparison with the same value.
TerminatorInst *PTI = Pred->getTerminator();
Value *PCV = isValueEqualityComparison(PTI); // PredCondVal
if (PCV == CV && SafeToMergeTerminators(TI, PTI)) {
// Figure out which 'cases' to copy from SI to PSI.
std::vector<ValueEqualityComparisonCase> BBCases;
BasicBlock *BBDefault = GetValueEqualityComparisonCases(TI, BBCases);
std::vector<ValueEqualityComparisonCase> PredCases;
BasicBlock *PredDefault = GetValueEqualityComparisonCases(PTI, PredCases);
// Based on whether the default edge from PTI goes to BB or not, fill in
// PredCases and PredDefault with the new switch cases we would like to
// build.
SmallVector<BasicBlock*, 8> NewSuccessors;
// Update the branch weight metadata along the way
SmallVector<uint64_t, 8> Weights;
bool PredHasWeights = HasBranchWeights(PTI);
bool SuccHasWeights = HasBranchWeights(TI);
if (PredHasWeights) {
GetBranchWeights(PTI, Weights);
// branch-weight metadata is inconsistent here.
if (Weights.size() != 1 + PredCases.size())
PredHasWeights = SuccHasWeights = false;
} else if (SuccHasWeights)
// If there are no predecessor weights but there are successor weights,
// populate Weights with 1, which will later be scaled to the sum of
// successor's weights
Weights.assign(1 + PredCases.size(), 1);
SmallVector<uint64_t, 8> SuccWeights;
if (SuccHasWeights) {
GetBranchWeights(TI, SuccWeights);
// branch-weight metadata is inconsistent here.
if (SuccWeights.size() != 1 + BBCases.size())
PredHasWeights = SuccHasWeights = false;
} else if (PredHasWeights)
SuccWeights.assign(1 + BBCases.size(), 1);
if (PredDefault == BB) {
// If this is the default destination from PTI, only the edges in TI
// that don't occur in PTI, or that branch to BB will be activated.
std::set<ConstantInt*, ConstantIntOrdering> PTIHandled;
for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
if (PredCases[i].Dest != BB)
else {
// The default destination is BB, we don't need explicit targets.
std::swap(PredCases[i], PredCases.back());
if (PredHasWeights || SuccHasWeights) {
// Increase weight for the default case.
Weights[0] += Weights[i+1];
std::swap(Weights[i+1], Weights.back());
--i; --e;
// Reconstruct the new switch statement we will be building.
if (PredDefault != BBDefault) {
PredDefault = BBDefault;
unsigned CasesFromPred = Weights.size();
uint64_t ValidTotalSuccWeight = 0;
for (unsigned i = 0, e = BBCases.size(); i != e; ++i)
if (!PTIHandled.count(BBCases[i].Value) &&
BBCases[i].Dest != BBDefault) {
if (SuccHasWeights || PredHasWeights) {
// The default weight is at index 0, so weight for the ith case
// should be at index i+1. Scale the cases from successor by
// PredDefaultWeight (Weights[0]).
Weights.push_back(Weights[0] * SuccWeights[i+1]);
ValidTotalSuccWeight += SuccWeights[i+1];
if (SuccHasWeights || PredHasWeights) {
ValidTotalSuccWeight += SuccWeights[0];
// Scale the cases from predecessor by ValidTotalSuccWeight.
for (unsigned i = 1; i < CasesFromPred; ++i)
Weights[i] *= ValidTotalSuccWeight;
// Scale the default weight by SuccDefaultWeight (SuccWeights[0]).
Weights[0] *= SuccWeights[0];
} else {
// If this is not the default destination from PSI, only the edges
// in SI that occur in PSI with a destination of BB will be
// activated.
std::set<ConstantInt*, ConstantIntOrdering> PTIHandled;
std::map<ConstantInt*, uint64_t> WeightsForHandled;
for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
if (PredCases[i].Dest == BB) {
if (PredHasWeights || SuccHasWeights) {
WeightsForHandled[PredCases[i].Value] = Weights[i+1];
std::swap(Weights[i+1], Weights.back());
std::swap(PredCases[i], PredCases.back());
--i; --e;
// Okay, now we know which constants were sent to BB from the
// predecessor. Figure out where they will all go now.
for (unsigned i = 0, e = BBCases.size(); i != e; ++i)
if (PTIHandled.count(BBCases[i].Value)) {
// If this is one we are capable of getting...
if (PredHasWeights || SuccHasWeights)
PTIHandled.erase(BBCases[i].Value);// This constant is taken care of
// If there are any constants vectored to BB that TI doesn't handle,
// they must go to the default destination of TI.
for (std::set<ConstantInt*, ConstantIntOrdering>::iterator I =
E = PTIHandled.end(); I != E; ++I) {
if (PredHasWeights || SuccHasWeights)
PredCases.push_back(ValueEqualityComparisonCase(*I, BBDefault));
// Okay, at this point, we know which new successor Pred will get. Make
// sure we update the number of entries in the PHI nodes for these
// successors.
for (unsigned i = 0, e = NewSuccessors.size(); i != e; ++i)
AddPredecessorToBlock(NewSuccessors[i], Pred, BB);
// Convert pointer to int before we switch.
if (CV->getType()->isPointerTy()) {
assert(DL && "Cannot switch on pointer without DataLayout");
CV = Builder.CreatePtrToInt(CV, DL->getIntPtrType(CV->getType()),
// Now that the successors are updated, create the new Switch instruction.
SwitchInst *NewSI = Builder.CreateSwitch(CV, PredDefault,
for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
NewSI->addCase(PredCases[i].Value, PredCases[i].Dest);
if (PredHasWeights || SuccHasWeights) {
// Halve the weights if any of them cannot fit in an uint32_t
SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end());
// Okay, last check. If BB is still a successor of PSI, then we must
// have an infinite loop case. If so, add an infinitely looping block
// to handle the case to preserve the behavior of the code.
BasicBlock *InfLoopBlock = nullptr;
for (unsigned i = 0, e = NewSI->getNumSuccessors(); i != e; ++i)
if (NewSI->getSuccessor(i) == BB) {
if (!InfLoopBlock) {
// Insert it at the end of the function, because it's either code,
// or it won't matter if it's hot. :)
InfLoopBlock = BasicBlock::Create(BB->getContext(),
"infloop", BB->getParent());
BranchInst::Create(InfLoopBlock, InfLoopBlock);
NewSI->setSuccessor(i, InfLoopBlock);
Changed = true;
return Changed;
// isSafeToHoistInvoke - If we would need to insert a select that uses the
// value of this invoke (comments in HoistThenElseCodeToIf explain why we
// would need to do this), we can't hoist the invoke, as there is nowhere
// to put the select in this case.
static bool isSafeToHoistInvoke(BasicBlock *BB1, BasicBlock *BB2,
Instruction *I1, Instruction *I2) {
for (succ_iterator SI = succ_begin(BB1), E = succ_end(BB1); SI != E; ++SI) {
PHINode *PN;
for (BasicBlock::iterator BBI = SI->begin();
(PN = dyn_cast<PHINode>(BBI)); ++BBI) {
Value *BB1V = PN->getIncomingValueForBlock(BB1);
Value *BB2V = PN->getIncomingValueForBlock(BB2);
if (BB1V != BB2V && (BB1V==I1 || BB2V==I2)) {
return false;
return true;
/// HoistThenElseCodeToIf - Given a conditional branch that goes to BB1 and
/// BB2, hoist any common code in the two blocks up into the branch block. The
/// caller of this function guarantees that BI's block dominates BB1 and BB2.
static bool HoistThenElseCodeToIf(BranchInst *BI, const DataLayout *DL) {
// This does very trivial matching, with limited scanning, to find identical
// instructions in the two blocks. In particular, we don't want to get into
// O(M*N) situations here where M and N are the sizes of BB1 and BB2. As
// such, we currently just scan for obviously identical instructions in an
// identical order.
BasicBlock *BB1 = BI->getSuccessor(0); // The true destination.
BasicBlock *BB2 = BI->getSuccessor(1); // The false destination
BasicBlock::iterator BB1_Itr = BB1->begin();
BasicBlock::iterator BB2_Itr = BB2->begin();
Instruction *I1 = BB1_Itr++, *I2 = BB2_Itr++;
// Skip debug info if it is not identical.
DbgInfoIntrinsic *DBI1 = dyn_cast<DbgInfoIntrinsic>(I1);
DbgInfoIntrinsic *DBI2 = dyn_cast<DbgInfoIntrinsic>(I2);
if (!DBI1 || !DBI2 || !DBI1->isIdenticalToWhenDefined(DBI2)) {
while (isa<DbgInfoIntrinsic>(I1))
I1 = BB1_Itr++;
while (isa<DbgInfoIntrinsic>(I2))
I2 = BB2_Itr++;
if (isa<PHINode>(I1) || !I1->isIdenticalToWhenDefined(I2) ||
(isa<InvokeInst>(I1) && !isSafeToHoistInvoke(BB1, BB2, I1, I2)))
return false;
BasicBlock *BIParent = BI->getParent();
bool Changed = false;
do {
// If we are hoisting the terminator instruction, don't move one (making a
// broken BB), instead clone it, and remove BI.
if (isa<TerminatorInst>(I1))
goto HoistTerminator;
// For a normal instruction, we just move one to right before the branch,
// then replace all uses of the other with the first. Finally, we remove
// the now redundant second instruction.
BIParent->getInstList().splice(BI, BB1->getInstList(), I1);
if (!I2->use_empty())
Changed = true;
I1 = BB1_Itr++;
I2 = BB2_Itr++;
// Skip debug info if it is not identical.
DbgInfoIntrinsic *DBI1 = dyn_cast<DbgInfoIntrinsic>(I1);
DbgInfoIntrinsic *DBI2 = dyn_cast<DbgInfoIntrinsic>(I2);
if (!DBI1 || !DBI2 || !DBI1->isIdenticalToWhenDefined(DBI2)) {
while (isa<DbgInfoIntrinsic>(I1))
I1 = BB1_Itr++;
while (isa<DbgInfoIntrinsic>(I2))
I2 = BB2_Itr++;
} while (I1->isIdenticalToWhenDefined(I2));
return true;
// It may not be possible to hoist an invoke.
if (isa<InvokeInst>(I1) && !isSafeToHoistInvoke(BB1, BB2, I1, I2))
return Changed;
for (succ_iterator SI = succ_begin(BB1), E = succ_end(BB1); SI != E; ++SI) {
PHINode *PN;
for (BasicBlock::iterator BBI = SI->begin();
(PN = dyn_cast<PHINode>(BBI)); ++BBI) {
Value *BB1V = PN->getIncomingValueForBlock(BB1);
Value *BB2V = PN->getIncomingValueForBlock(BB2);
if (BB1V == BB2V)
if (isa<ConstantExpr>(BB1V) && !isSafeToSpeculativelyExecute(BB1V, DL))
return Changed;
if (isa<ConstantExpr>(BB2V) && !isSafeToSpeculativelyExecute(BB2V, DL))
return Changed;
// Okay, it is safe to hoist the terminator.
Instruction *NT = I1->clone();
BIParent->getInstList().insert(BI, NT);
if (!NT->getType()->isVoidTy()) {
IRBuilder<true, NoFolder> Builder(NT);
// Hoisting one of the terminators from our successor is a great thing.
// Unfortunately, the successors of the if/else blocks may have PHI nodes in
// them. If they do, all PHI entries for BB1/BB2 must agree for all PHI
// nodes, so we insert select instruction to compute the final result.
std::map<std::pair<Value*,Value*>, SelectInst*> InsertedSelects;
for (succ_iterator SI = succ_begin(BB1), E = succ_end(BB1); SI != E; ++SI) {
PHINode *PN;
for (BasicBlock::iterator BBI = SI->begin();
(PN = dyn_cast<PHINode>(BBI)); ++BBI) {
Value *BB1V = PN->getIncomingValueForBlock(BB1);
Value *BB2V = PN->getIncomingValueForBlock(BB2);
if (BB1V == BB2V) continue;
// These values do not agree. Insert a select instruction before NT
// that determines the right value.
SelectInst *&SI = InsertedSelects[std::make_pair(BB1V, BB2V)];
if (!SI)
SI = cast<SelectInst>
(Builder.CreateSelect(BI->getCondition(), BB1V, BB2V,
// Make the PHI node use the select for all incoming values for BB1/BB2
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
if (PN->getIncomingBlock(i) == BB1 || PN->getIncomingBlock(i) == BB2)
PN->setIncomingValue(i, SI);
// Update any PHI nodes in our new successors.
for (succ_iterator SI = succ_begin(BB1), E = succ_end(BB1); SI != E; ++SI)
AddPredecessorToBlock(*SI, BIParent, BB1);
return true;
/// SinkThenElseCodeToEnd - Given an unconditional branch that goes to BBEnd,
/// check whether BBEnd has only two predecessors and the other predecessor
/// ends with an unconditional branch. If it is true, sink any common code
/// in the two predecessors to BBEnd.
static bool SinkThenElseCodeToEnd(BranchInst *BI1) {
BasicBlock *BB1 = BI1->getParent();
BasicBlock *BBEnd = BI1->getSuccessor(0);
// Check that BBEnd has two predecessors and the other predecessor ends with
// an unconditional branch.
pred_iterator PI = pred_begin(BBEnd), PE = pred_end(BBEnd);
BasicBlock *Pred0 = *PI++;
if (PI == PE) // Only one predecessor.
return false;
BasicBlock *Pred1 = *PI++;
if (PI != PE) // More than two predecessors.
return false;
BasicBlock *BB2 = (Pred0 == BB1) ? Pred1 : Pred0;
BranchInst *BI2 = dyn_cast<BranchInst>(BB2->getTerminator());
if (!BI2 || !BI2->isUnconditional())
return false;
// Gather the PHI nodes in BBEnd.
std::map<Value*, std::pair<Value*, PHINode*> > MapValueFromBB1ToBB2;
Instruction *FirstNonPhiInBBEnd = nullptr;
for (BasicBlock::iterator I = BBEnd->begin(), E = BBEnd->end();
I != E; ++I) {
if (PHINode *PN = dyn_cast<PHINode>(I)) {
Value *BB1V = PN->getIncomingValueForBlock(BB1);
Value *BB2V = PN->getIncomingValueForBlock(BB2);
MapValueFromBB1ToBB2[BB1V] = std::make_pair(BB2V, PN);
} else {
FirstNonPhiInBBEnd = &*I;
if (!FirstNonPhiInBBEnd)
return false;
// This does very trivial matching, with limited scanning, to find identical
// instructions in the two blocks. We scan backward for obviously identical
// instructions in an identical order.
BasicBlock::InstListType::reverse_iterator RI1 = BB1->getInstList().rbegin(),
RE1 = BB1->getInstList().rend(), RI2 = BB2->getInstList().rbegin(),
RE2 = BB2->getInstList().rend();
// Skip debug info.
while (RI1 != RE1 && isa<DbgInfoIntrinsic>(&*RI1)) ++RI1;
if (RI1 == RE1)
return false;
while (RI2 != RE2 && isa<DbgInfoIntrinsic>(&*RI2)) ++RI2;
if (RI2 == RE2)
return false;
// Skip the unconditional branches.
bool Changed = false;
while (RI1 != RE1 && RI2 != RE2) {
// Skip debug info.
while (RI1 != RE1 && isa<DbgInfoIntrinsic>(&*RI1)) ++RI1;
if (RI1 == RE1)
return Changed;
while (RI2 != RE2 && isa<DbgInfoIntrinsic>(&*RI2)) ++RI2;
if (RI2 == RE2)
return Changed;
Instruction *I1 = &*RI1, *I2 = &*RI2;
// I1 and I2 should have a single use in the same PHI node, and they
// perform the same operation.
// Cannot move control-flow-involving, volatile loads, vaarg, etc.
if (isa<PHINode>(I1) || isa<PHINode>(I2) ||
isa<TerminatorInst>(I1) || isa<TerminatorInst>(I2) ||
isa<LandingPadInst>(I1) || isa<LandingPadInst>(I2) ||
isa<AllocaInst>(I1) || isa<AllocaInst>(I2) ||
I1->mayHaveSideEffects() || I2->mayHaveSideEffects() ||
I1->mayReadOrWriteMemory() || I2->mayReadOrWriteMemory() ||
!I1->hasOneUse() || !I2->hasOneUse() ||
MapValueFromBB1ToBB2.find(I1) == MapValueFromBB1ToBB2.end() ||
MapValueFromBB1ToBB2[I1].first != I2)
return Changed;
// Check whether we should swap the operands of ICmpInst.
ICmpInst *ICmp1 = dyn_cast<ICmpInst>(I1), *ICmp2 = dyn_cast<ICmpInst>(I2);
bool SwapOpnds = false;
if (ICmp1 && ICmp2 &&
ICmp1->getOperand(0) != ICmp2->getOperand(0) &&
ICmp1->getOperand(1) != ICmp2->getOperand(1) &&
(ICmp1->getOperand(0) == ICmp2->getOperand(1) ||
ICmp1->getOperand(1) == ICmp2->getOperand(0))) {
SwapOpnds = true;
if (!I1->isSameOperationAs(I2)) {
if (SwapOpnds)
return Changed;
// The operands should be either the same or they need to be generated
// with a PHI node after sinking. We only handle the case where there is
// a single pair of different operands.
Value *DifferentOp1 = nullptr, *DifferentOp2 = nullptr;
unsigned Op1Idx = 0;
for (unsigned I = 0, E = I1->getNumOperands(); I != E; ++I) {
if (I1->getOperand(I) == I2->getOperand(I))
// Early exit if we have more-than one pair of different operands or
// the different operand is already in MapValueFromBB1ToBB2.
// Early exit if we need a PHI node to replace a constant.
if (DifferentOp1 ||
MapValueFromBB1ToBB2.find(I1->getOperand(I)) !=
MapValueFromBB1ToBB2.end() ||
isa<Constant>(I1->getOperand(I)) ||
isa<Constant>(I2->getOperand(I))) {
// If we can't sink the instructions, undo the swapping.
if (SwapOpnds)
return Changed;
DifferentOp1 = I1->getOperand(I);
Op1Idx = I;
DifferentOp2 = I2->getOperand(I);
// We insert the pair of different operands to MapValueFromBB1ToBB2 and
// remove (I1, I2) from MapValueFromBB1ToBB2.
if (DifferentOp1) {
PHINode *NewPN = PHINode::Create(DifferentOp1->getType(), 2,
DifferentOp1->getName() + ".sink",
MapValueFromBB1ToBB2[DifferentOp1] = std::make_pair(DifferentOp2, NewPN);
// I1 should use NewPN instead of DifferentOp1.
I1->setOperand(Op1Idx, NewPN);
NewPN->addIncoming(DifferentOp1, BB1);
NewPN->addIncoming(DifferentOp2, BB2);
DEBUG(dbgs() << "Create PHI node " << *NewPN << "\n";);
PHINode *OldPN = MapValueFromBB1ToBB2[I1].second;
DEBUG(dbgs() << "SINK common instructions " << *I1 << "\n";);
DEBUG(dbgs() << " " << *I2 << "\n";);
// We need to update RE1 and RE2 if we are going to sink the first
// instruction in the basic block down.
bool UpdateRE1 = (I1 == BB1->begin()), UpdateRE2 = (I2 == BB2->begin());
// Sink the instruction.
BBEnd->getInstList().splice(FirstNonPhiInBBEnd, BB1->getInstList(), I1);
if (!OldPN->use_empty())
if (!I2->use_empty())
if (UpdateRE1)
RE1 = BB1->getInstList().rend();
if (UpdateRE2)
RE2 = BB2->getInstList().rend();
FirstNonPhiInBBEnd = I1;
Changed = true;
return Changed;
/// \brief Determine if we can hoist sink a sole store instruction out of a
/// conditional block.
/// We are looking for code like the following:
/// BrBB:
/// store i32 %add, i32* %arrayidx2
/// ... // No other stores or function calls (we could be calling a memory
/// ... // function).
/// %cmp = icmp ult %x, %y
/// br i1 %cmp, label %EndBB, label %ThenBB
/// ThenBB:
/// store i32 %add5, i32* %arrayidx2
/// br label EndBB
/// EndBB:
/// ...
/// We are going to transform this into:
/// BrBB:
/// store i32 %add, i32* %arrayidx2
/// ... //
/// %cmp = icmp ult %x, %y
/// %add.add5 = select i1 %cmp, i32 %add, %add5
/// store i32 %add.add5, i32* %arrayidx2
/// ...
/// \return The pointer to the value of the previous store if the store can be
/// hoisted into the predecessor block. 0 otherwise.
static Value *isSafeToSpeculateStore(Instruction *I, BasicBlock *BrBB,
BasicBlock *StoreBB, BasicBlock *EndBB) {
StoreInst *StoreToHoist = dyn_cast<StoreInst>(I);
if (!StoreToHoist)
return nullptr;
// Volatile or atomic.
if (!StoreToHoist->isSimple())
return nullptr;
Value *StorePtr = StoreToHoist->getPointerOperand();
// Look for a store to the same pointer in BrBB.
unsigned MaxNumInstToLookAt = 10;
for (BasicBlock::reverse_iterator RI = BrBB->rbegin(),
RE = BrBB->rend(); RI != RE && (--MaxNumInstToLookAt); ++RI) {
Instruction *CurI = &*RI;
// Could be calling an instruction that effects memory like free().
if (CurI->mayHaveSideEffects() && !isa<StoreInst>(CurI))
return nullptr;
StoreInst *SI = dyn_cast<StoreInst>(CurI);
// Found the previous store make sure it stores to the same location.
if (SI && SI->getPointerOperand() == StorePtr)
// Found the previous store, return its value operand.
return SI->getValueOperand();
else if (SI)
return nullptr; // Unknown store.
return nullptr;
/// \brief Speculate a conditional basic block flattening the CFG.
/// Note that this is a very risky transform currently. Speculating
/// instructions like this is most often not desirable. Instead, there is an MI
/// pass which can do it with full awareness of the resource constraints.
/// However, some cases are "obvious" and we should do directly. An example of
/// this is speculating a single, reasonably cheap instruction.
/// There is only one distinct advantage to flattening the CFG at the IR level:
/// it makes very common but simplistic optimizations such as are common in
/// instcombine and the DAG combiner more powerful by removing CFG edges and
/// modeling their effects with easier to reason about SSA value graphs.
/// An illustration of this transform is turning this IR:
/// \code
/// BB:
/// %cmp = icmp ult %x, %y
/// br i1 %cmp, label %EndBB, label %ThenBB
/// ThenBB:
/// %sub = sub %x, %y
/// br label BB2
/// EndBB:
/// %phi = phi [ %sub, %ThenBB ], [ 0, %EndBB ]
/// ...
/// \endcode
/// Into this IR:
/// \code
/// BB:
/// %cmp = icmp ult %x, %y
/// %sub = sub %x, %y
/// %cond = select i1 %cmp, 0, %sub
/// ...
/// \endcode
/// \returns true if the conditional block is removed.
static bool SpeculativelyExecuteBB(BranchInst *BI, BasicBlock *ThenBB,
const DataLayout *DL) {
// Be conservative for now. FP select instruction can often be expensive.
Value *BrCond = BI->getCondition();
if (isa<FCmpInst>(BrCond))
return false;
BasicBlock *BB = BI->getParent();
BasicBlock *EndBB = ThenBB->getTerminator()->getSuccessor(0);
// If ThenBB is actually on the false edge of the conditional branch, remember
// to swap the select operands later.
bool Invert = false;
if (ThenBB != BI->getSuccessor(0)) {
assert(ThenBB == BI->getSuccessor(1) && "No edge from 'if' block?");
Invert = true;
assert(EndBB == BI->getSuccessor(!Invert) && "No edge from to end block");
// Keep a count of how many times instructions are used within CondBB when
// they are candidates for sinking into CondBB. Specifically:
// - They are defined in BB, and
// - They have no side effects, and
// - All of their uses are in CondBB.
SmallDenseMap<Instruction *, unsigned, 4> SinkCandidateUseCounts;
unsigned SpeculationCost = 0;
Value *SpeculatedStoreValue = nullptr;
StoreInst *SpeculatedStore = nullptr;
for (BasicBlock::iterator BBI = ThenBB->begin(),
BBE = std::prev(ThenBB->end());
BBI != BBE; ++BBI) {
Instruction *I = BBI;
// Skip debug info.
if (isa<DbgInfoIntrinsic>(I))
// Only speculatively execution a single instruction (not counting the
// terminator) for now.
if (SpeculationCost > 1)
return false;
// Don't hoist the instruction if it's unsafe or expensive.
if (!isSafeToSpeculativelyExecute(I, DL) &&
!(HoistCondStores &&
(SpeculatedStoreValue = isSafeToSpeculateStore(I, BB, ThenBB,
return false;
if (!SpeculatedStoreValue &&
ComputeSpeculationCost(I, DL) > PHINodeFoldingThreshold)
return false;
// Store the store speculation candidate.
if (SpeculatedStoreValue)
SpeculatedStore = cast<StoreInst>(I);
// Do not hoist the instruction if any of its operands are defined but not
// used in BB. The transformation will prevent the operand from
// being sunk into the use block.
for (User::op_iterator i = I->op_begin(), e = I->op_end();
i != e; ++i) {
Instruction *OpI = dyn_cast<Instruction>(*i);
if (!OpI || OpI->getParent() != BB ||
continue; // Not a candidate for sinking.
// Consider any sink candidates which are only used in CondBB as costs for
// speculation. Note, while we iterate over a DenseMap here, we are summing
// and so iteration order isn't significant.
for (SmallDenseMap<Instruction *, unsigned, 4>::iterator I =
SinkCandidateUseCounts.begin(), E = SinkCandidateUseCounts.end();
I != E; ++I)
if (I->first->getNumUses() == I->second) {
if (SpeculationCost > 1)
return false;
// Check that the PHI nodes can be converted to selects.
bool HaveRewritablePHIs = false;
for (BasicBlock::iterator I = EndBB->begin();
PHINode *PN = dyn_cast<PHINode>(I); ++I) {
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)
HaveRewritablePHIs = true;
ConstantExpr *OrigCE = dyn_cast<ConstantExpr>(OrigV);
ConstantExpr *ThenCE = dyn_cast<ConstantExpr>(ThenV);
if (!OrigCE && !ThenCE)
continue; // Known safe and cheap.
if ((ThenCE && !isSafeToSpeculativelyExecute(ThenCE, DL)) ||
(OrigCE && !isSafeToSpeculativelyExecute(OrigCE, DL)))
return false;
unsigned OrigCost = OrigCE ? ComputeSpeculationCost(OrigCE, DL) : 0;
unsigned ThenCost = ThenCE ? ComputeSpeculationCost(ThenCE, DL) : 0;
if (OrigCost + ThenCost > 2 * PHINodeFoldingThreshold)
return false;
// Account for the cost of an unfolded ConstantExpr which could end up
// getting expanded into Instructions.
// FIXME: This doesn't account for how many operations are combined in the
// constant expression.
if (SpeculationCost > 1)
return false;
// If there are no PHIs to process, bail early. This helps ensure idempotence
// as well.
if (!HaveRewritablePHIs && !(HoistCondStores && SpeculatedStoreValue))
return false;
// If we get here, we can hoist the instruction and if-convert.
DEBUG(dbgs() << "SPECULATIVELY EXECUTING BB" << *ThenBB << "\n";);
// Insert a select of the value of the speculated store.
if (SpeculatedStoreValue) {
IRBuilder<true, NoFolder> Builder(BI);
Value *TrueV = SpeculatedStore->getValueOperand();
Value *FalseV = SpeculatedStoreValue;
if (Invert)
std::swap(TrueV, FalseV);
Value *S = Builder.CreateSelect(BrCond, TrueV, FalseV, TrueV->getName() +
"." + FalseV->getName());
SpeculatedStore->setOperand(0, S);
// Hoist the instructions.
BB->getInstList().splice(BI, ThenBB->getInstList(), ThenBB->begin(),
// Insert selects and rewrite the PHI operands.
IRBuilder<true, NoFolder> Builder(BI);
for (BasicBlock::iterator I = EndBB->begin();
PHINode *PN = dyn_cast<PHINode>(I); ++I) {
unsigned OrigI = PN->getBasicBlockIndex(BB);
unsigned ThenI = PN->getBasicBlockIndex(ThenBB);
Value *OrigV = PN->getIncomingValue(OrigI);
Value *ThenV = PN->getIncomingValue(ThenI);
// Skip PHIs which are trivial.
if (OrigV == ThenV)
// Create a select whose true value is the speculatively executed value and
// false value is the preexisting value. Swap them if the branch
// destinations were inverted.
Value *TrueV = ThenV, *FalseV = OrigV;
if (Invert)
std::swap(TrueV, FalseV);
Value *V = Builder.CreateSelect(BrCond, TrueV, FalseV,
TrueV->getName() + "." + FalseV->getName());
PN->setIncomingValue(OrigI, V);
PN->setIncomingValue(ThenI, V);
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)
if (CI->cannotDuplicate())
return true;
return false;
/// BlockIsSimpleEnoughToThreadThrough - Return true if we can thread a branch
/// across this block.
static bool BlockIsSimpleEnoughToThreadThrough(BasicBlock *BB) {
BranchInst *BI = cast<BranchInst>(BB->getTerminator());
unsigned Size = 0;
for (BasicBlock::iterator BBI = BB->begin(); &*BBI != BI; ++BBI) {
if (isa<DbgInfoIntrinsic>(BBI))
if (Size > 10) return false; // Don't clone large BB's.
// We can only support instructions that do not define values that are
// live outside of the current basic block.
for (User *U : BBI->users()) {
Instruction *UI = cast<Instruction>(U);
if (UI->getParent() != BB || isa<PHINode>(UI)) return false;
// Looks ok, continue checking.
return true;
/// FoldCondBranchOnPHI - If we have a conditional branch on a PHI node value
/// 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 *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
// outside of the block.
if (!PN || PN->getParent() != BB || !PN->hasOneUse())
return false;
// Degenerate case of a single entry PHI.
if (PN->getNumIncomingValues() == 1) {
return true;
// 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) {
ConstantInt *CB = dyn_cast<ConstantInt>(PN->getIncomingValue(i));
if (!CB || !CB->getType()->isIntegerTy(1)) continue;
// Okay, we now know that all edges from PredBB should be revectored to
// branch to RealDest.
BasicBlock *PredBB = PN->getIncomingBlock(i);
BasicBlock *RealDest = BI->getSuccessor(!CB->getZExtValue());
if (RealDest == BB) continue; // Skip self loops.
// Skip if the predecessor's terminator is an indirect branch.
if (isa<IndirectBrInst>(PredBB->getTerminator())) continue;
// The dest block might have PHI nodes, other predecessors and other
// difficult cases. Instead of being smart about this, just insert a new
// block that jumps to the destination block, effectively splitting
// the edge we are about to create.
BasicBlock *EdgeBB = BasicBlock::Create(BB->getContext(),
RealDest->getParent(), RealDest);
BranchInst::Create(RealDest, EdgeBB);
// Update PHI nodes.
AddPredecessorToBlock(RealDest, EdgeBB, BB);
// BB may have instructions that are being threaded over. Clone these
// instructions into EdgeBB. We know that there will be no uses of the
// cloned instructions outside of EdgeBB.
BasicBlock::iterator InsertPt = EdgeBB->begin();
DenseMap<Value*, Value*> TranslateMap; // Track translated values.
for (BasicBlock::iterator BBI = BB->begin(); &*BBI != BI; ++BBI) {
if (PHINode *PN = dyn_cast<PHINode>(BBI)) {
TranslateMap[PN] = PN->getIncomingValueForBlock(PredBB);
// Clone the instruction.
Instruction *N = BBI->clone();
if (BBI->hasName()) N->setName(BBI->getName()+".c");
// Update operands due to translation.
for (User::op_iterator i = N->op_begin(), e = N->op_end();
i != e; ++i) {
DenseMap<Value*, Value*>::iterator PI = TranslateMap.find(*i);
if (PI != TranslateMap.end())
*i = PI->second;
// Check for trivial simplification.
if (Value *V = SimplifyInstruction(N, DL)) {
TranslateMap[BBI] = V;
delete N; // Instruction folded away, don't need actual inst
} else {
// Insert the new instruction into its new home.
EdgeBB->getInstList().insert(InsertPt, N);
if (!BBI->use_empty())
TranslateMap[BBI] = N;
// Loop over all of the edges from PredBB to BB, changing them to branch
// to EdgeBB instead.
TerminatorInst *PredBBTI = PredBB->getTerminator();
for (unsigned i = 0, e = PredBBTI->getNumSuccessors(); i != e; ++i)
if (PredBBTI->getSuccessor(i) == BB) {
PredBBTI->setSuccessor(i, EdgeBB);
// Recurse, simplifying any other constants.
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 *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
// subsequently causes this merge to happen. We really want control
// dependence information for this check, but simplifycfg can't keep it up
// to date, and this catches most of the cases we care about anyway.
BasicBlock *BB = PN->getParent();
BasicBlock *IfTrue, *IfFalse;
Value *IfCond = GetIfCondition(BB, IfTrue, IfFalse);
if (!IfCond ||
// Don't bother if the branch will be constant folded trivially.
return false;
// Okay, we found that we can merge this two-entry phi node into a select.
// Doing so would require us to fold *all* two entry phi nodes in this block.
// At some point this becomes non-profitable (particularly if the target
// doesn't support cmov's). Only do this transformation if there are two or
// fewer PHI nodes in this block.
unsigned NumPhis = 0;
for (BasicBlock::iterator I = BB->begin(); isa<PHINode>(I); ++NumPhis, ++I)
if (NumPhis > 2)
return false;
// Loop over the PHI's seeing if we can promote them all to select
// instructions. While we are at it, keep track of the instructions
// that need to be moved to the dominating block.
SmallPtrSet<Instruction*, 4> AggressiveInsts;
unsigned MaxCostVal0 = PHINodeFoldingThreshold,
MaxCostVal1 = PHINodeFoldingThreshold;
for (BasicBlock::iterator II = BB->begin(); isa<PHINode>(II);) {
PHINode *PN = cast<PHINode>(II++);
if (Value *V = SimplifyInstruction(PN, DL)) {
if (!DominatesMergePoint(PN->getIncomingValue(0), BB, &AggressiveInsts,
MaxCostVal0, DL) ||
!DominatesMergePoint(PN->getIncomingValue(1), BB, &AggressiveInsts,
MaxCostVal1, DL))
return false;
// If we folded the first phi, PN dangles at this point. Refresh it. If
// we ran out of PHIs then we simplified them all.
PN = dyn_cast<PHINode>(BB->begin());
if (!PN) return true;
// Don't fold i1 branches on PHIs which contain binary operators. These can
// often be turned into switches and other things.
if (PN->getType()->isIntegerTy(1) &&
(isa<BinaryOperator>(PN->getIncomingValue(0)) ||
isa<BinaryOperator>(PN->getIncomingValue(1)) ||
return false;
// If we all PHI nodes are promotable, check to make sure that all
// instructions in the predecessor blocks can be promoted as well. If
// not, we won't be able to get rid of the control flow, so it's not
// worth promoting to select instructions.
BasicBlock *DomBlock = nullptr;
BasicBlock *IfBlock1 = PN->getIncomingBlock(0);
BasicBlock *IfBlock2 = PN->getIncomingBlock(1);
if (cast<BranchInst>(IfBlock1->getTerminator())->isConditional()) {
IfBlock1 = nullptr;
} else {
DomBlock = *pred_begin(IfBlock1);
for (BasicBlock::iterator I = IfBlock1->begin();!isa<TerminatorInst>(I);++I)
if (!AggressiveInsts.count(I) && !isa<DbgInfoIntrinsic>(I)) {
// This is not an aggressive instruction that we can promote.
// Because of this, we won't be able to get rid of the control
// flow, so the xform is not worth it.
return false;
if (cast<BranchInst>(IfBlock2->getTerminator())->isConditional()) {
IfBlock2 = nullptr;
} else {
DomBlock = *pred_begin(IfBlock2);
for (BasicBlock::iterator I = IfBlock2->begin();!isa<TerminatorInst>(I);++I)
if (!AggressiveInsts.count(I) && !isa<DbgInfoIntrinsic>(I)) {
// This is not an aggressive instruction that we can promote.
// Because of this, we won't be able to get rid of the control
// flow, so the xform is not worth it.
return false;
DEBUG(dbgs() << "FOUND IF CONDITION! " << *IfCond << " T: "
<< IfTrue->getName() << " F: " << IfFalse->getName() << "\n");
// If we can still promote the PHI nodes after this gauntlet of tests,
// do all of the PHI's now.
Instruction *InsertPt = DomBlock->getTerminator();
IRBuilder<true, NoFolder> Builder(InsertPt);
// Move all 'aggressive' instructions, which are defined in the
// conditional parts of the if's up to the dominating block.
if (IfBlock1)
IfBlock1->getInstList(), IfBlock1->begin(),
if (IfBlock2)
IfBlock2->getInstList(), IfBlock2->begin(),
while (PHINode *PN = dyn_cast<PHINode>(BB->begin())) {
// Change the PHI node into a select instruction.
Value *TrueVal = PN->getIncomingValue(PN->getIncomingBlock(0) == IfFalse);
Value *FalseVal = PN->getIncomingValue(PN->getIncomingBlock(0) == IfTrue);
SelectInst *NV =
cast<SelectInst>(Builder.CreateSelect(IfCond, TrueVal, FalseVal, ""));
// At this point, IfBlock1 and IfBlock2 are both empty, so our if statement
// has been flattened. Change DomBlock to jump directly to our new block to
// avoid other simplifycfg's kicking in on the diamond.
TerminatorInst *OldTI = DomBlock->getTerminator();
return true;
/// SimplifyCondBranchToTwoReturns - If we found a conditional branch that goes
/// to two returning blocks, try to merge them together into one return,
/// introducing a select if the return values disagree.
static bool SimplifyCondBranchToTwoReturns(BranchInst *BI,
IRBuilder<> &Builder) {
assert(BI->isConditional() && "Must be a conditional branch");
BasicBlock *TrueSucc = BI->getSuccessor(0);
BasicBlock *FalseSucc = BI->getSuccessor(1);
ReturnInst *TrueRet = cast<ReturnInst>(TrueSucc->getTerminator());
ReturnInst *FalseRet = cast<ReturnInst>(FalseSucc->getTerminator());
// Check to ensure both blocks are empty (just a return) or optionally empty
// with PHI nodes. If there are other instructions, merging would cause extra
// computation on one path or the other.
if (!TrueSucc->getFirstNonPHIOrDbg()->isTerminator())
return false;
if (!FalseSucc->getFirstNonPHIOrDbg()->isTerminator())
return false;
// Okay, we found a branch that is going to two return nodes. If
// there is no return value for this function, just change the
// branch into a return.
if (FalseRet->getNumOperands() == 0) {
return true;
// Otherwise, figure out what the true and false return values are
// so we can insert a new select instruction.
Value *TrueValue = TrueRet->getReturnValue();
Value *FalseValue = FalseRet->getReturnValue();
// Unwrap any PHI nodes in the return blocks.
if (PHINode *TVPN = dyn_cast_or_null<PHINode>(TrueValue))
if (TVPN->getParent() == TrueSucc)
TrueValue = TVPN->getIncomingValueForBlock(BI->getParent());
if (PHINode *FVPN = dyn_cast_or_null<PHINode>(FalseValue))
if (FVPN->getParent() == FalseSucc)
FalseValue = FVPN->getIncomingValueForBlock(BI->getParent());
// In order for this transformation to be safe, we must be able to
// unconditionally execute both operands to the return. This is
// normally the case, but we could have a potentially-trapping
// constant expression that prevents this transformation from being
// safe.
if (ConstantExpr *TCV = dyn_cast_or_null<ConstantExpr>(TrueValue))
if (TCV->canTrap())
return false;
if (ConstantExpr *FCV = dyn_cast_or_null<ConstantExpr>(FalseValue))
if (FCV->canTrap())
return false;
// Okay, we collected all the mapped values and checked them for sanity, and
// defined to really do this transformation. First, update the CFG.
// Insert select instructions where needed.
Value *BrCond = BI->getCondition();
if (TrueValue) {
// Insert a select if the results differ.
if (TrueValue == FalseValue || isa<UndefValue>(FalseValue)) {
} else if (isa<UndefValue>(TrueValue)) {
TrueValue = FalseValue;
} else {
TrueValue = Builder.CreateSelect(BrCond, TrueValue,
FalseValue, "retval");
Value *RI = !TrueValue ?
Builder.CreateRetVoid() : Builder.CreateRet(TrueValue);
(void) RI;
<< "\n " << *BI << "NewRet = " << *RI
<< "TRUEBLOCK: " << *TrueSucc << "FALSEBLOCK: "<< *FalseSucc);
return true;
/// ExtractBranchMetadata - Given a conditional BranchInstruction, retrieve the
/// probabilities of the branch taking each edge. Fills in the two APInt
/// parameters and return true, or returns false if no or invalid metadata was
/// found.
static bool ExtractBranchMetadata(BranchInst *BI,
uint64_t &ProbTrue, uint64_t &ProbFalse) {
assert(BI->isConditional() &&
"Looking for probabilities on unconditional branch?");
MDNode *ProfileData = BI->getMetadata(LLVMContext::MD_prof);
if (!ProfileData || ProfileData->getNumOperands() != 3) return false;
ConstantInt *CITrue = dyn_cast<ConstantInt>(ProfileData->getOperand(1));
ConstantInt *CIFalse = dyn_cast<ConstantInt>(ProfileData->getOperand(2));
if (!CITrue || !CIFalse) return false;
ProbTrue = CITrue->getValue().getZExtValue();
ProbFalse = CIFalse->getValue().getZExtValue();
return true;
/// checkCSEInPredecessor - Return true if the given instruction is available
/// in its predecessor block. If yes, the instruction will be removed.
static bool checkCSEInPredecessor(Instruction *Inst, BasicBlock *PB) {
if (!isa<BinaryOperator>(Inst) && !isa<CmpInst>(Inst))
return false;
for (BasicBlock::iterator I = PB->begin(), E = PB->end(); I != E; I++) {
Instruction *PBI = &*I;
// Check whether Inst and PBI generate the same value.
if (Inst->isIdenticalTo(PBI)) {
return true;
return false;
/// FoldBranchToCommonDest - If this basic block is simple enough, and if a
/// predecessor branches to us and one of our successors, fold the block into
/// the predecessor and use logical operations to pick the right destination.
bool llvm::FoldBranchToCommonDest(BranchInst *BI, const DataLayout *DL) {
BasicBlock *BB = BI->getParent();
Instruction *Cond = nullptr;
if (BI->isConditional())
Cond = dyn_cast<Instruction>(BI->getCondition());
else {
// For unconditional branch, check for a simple CFG pattern, where
// BB has a single predecessor and BB's successor is also its predecessor's
// successor. If such pattern exisits, check for CSE between BB and its
// predecessor.
if (BasicBlock *PB = BB->getSinglePredecessor())
if (BranchInst *PBI = dyn_cast<BranchInst>(PB->getTerminator()))
if (PBI->isConditional() &&
(BI->getSuccessor(0) == PBI->getSuccessor(0) ||
BI->getSuccessor(0) == PBI->getSuccessor(1))) {
for (BasicBlock::iterator I = BB->begin(), E = BB->end();
I != E; ) {
Instruction *Curr = I++;
if (isa<CmpInst>(Curr)) {
Cond = Curr;
// Quit if we can't remove this instruction.
if (!checkCSEInPredecessor(Curr, PB))
return false;
if (!Cond)
return false;
if (!Cond || (!isa<CmpInst>(Cond) && !isa<BinaryOperator>(Cond)) ||
Cond->getParent() != BB || !Cond->hasOneUse())
return false;
// Only allow this if the condition is a simple instruction that can be
// executed unconditionally. It must be in the same block as the branch, and
// must be at the front of the block.
BasicBlock::iterator FrontIt = BB->front();
// Ignore dbg intrinsics.
while (isa<DbgInfoIntrinsic>(FrontIt)) ++FrontIt;
// Allow a single instruction to be hoisted in addition to the compare
// that feeds the branch. We later ensure that any values that _it_ uses
// were also live in the predecessor, so that we don't unnecessarily create
// register pressure or inhibit out-of-order execution.
Instruction *BonusInst = nullptr;
if (&*FrontIt != Cond &&
FrontIt->hasOneUse() && FrontIt->user_back() == Cond &&
isSafeToSpeculativelyExecute(FrontIt, DL)) {
BonusInst = &*FrontIt;
// Ignore dbg intrinsics.
while (isa<DbgInfoIntrinsic>(FrontIt)) ++FrontIt;
// Only a single bonus inst is allowed.
if (&*FrontIt != Cond)
return false;
// Make sure the instruction after the condition is the cond branch.
BasicBlock::iterator CondIt = Cond; ++CondIt;
// Ignore dbg intrinsics.
while (isa<DbgInfoIntrinsic>(CondIt)) ++CondIt;
if (&*CondIt != BI)
return false;
// Cond is known to be a compare or binary operator. Check to make sure that
// neither operand is a potentially-trapping constant expression.
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Cond->getOperand(0)))
if (CE->canTrap())
return false;
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Cond->getOperand(1)))
if (CE->canTrap())
return false;
// Finally, don't infinitely unroll conditional loops.
BasicBlock *TrueDest = BI->getSuccessor(0);
BasicBlock *FalseDest = (BI->isConditional()) ? BI->getSuccessor(1) : nullptr;
if (TrueDest == BB || FalseDest == BB)
return false;
for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
BasicBlock *PredBlock = *PI;
BranchInst *PBI = dyn_cast<BranchInst>(PredBlock->getTerminator());
// Check that we have two conditional branches. If there is a PHI node in
// the common successor, verify that the same value flows in from both
// blocks.
SmallVector<PHINode*, 4> PHIs;
if (!PBI || PBI->isUnconditional() ||
(BI->isConditional() &&
!SafeToMergeTerminators(BI, PBI)) ||
(!BI->isConditional() &&
!isProfitableToFoldUnconditional(BI, PBI, Cond, PHIs)))
// Determine if the two branches share a common destination.
Instruction::BinaryOps Opc = Instruction::BinaryOpsEnd;
bool InvertPredCond = false;
if (BI->isConditional()) {
if (PBI->getSuccessor(0) == TrueDest)
Opc = Instruction::Or;
else if (PBI->getSuccessor(1) == FalseDest)
Opc = Instruction::And;
else if (PBI->getSuccessor(0) == FalseDest)
Opc = Instruction::And, InvertPredCond = true;
else if (PBI->getSuccessor(1) == TrueDest)
Opc = Instruction::Or, InvertPredCond = true;
} else {
if (PBI->getSuccessor(0) != TrueDest && PBI->getSuccessor(1) != TrueDest)
// 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. We also allow
// instructions that are used by the terminator's condition because it
// exposes more merging opportunities.
bool UsedByBranch = (BonusInst && BonusInst->hasOneUse() &&
BonusInst->user_back() == 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) && !isa<Argument>(V))
SmallVector<std::pair<Value*, unsigned>, 4> Worklist;
Worklist.push_back(std::make_pair(PBI->getOperand(0), 0));
// Walk up to four levels back up the use-def chain of the predecessor's
// terminator to see if all those values were used. The choice of four
// levels is arbitrary, to provide a compile-time-cost bound.
while (!Worklist.empty()) {
std::pair<Value*, unsigned> Pair = Worklist.back();
if (Pair.second >= 4) continue;
if (UsedValues.empty()) break;
if (Instruction *I = dyn_cast<Instruction>(Pair.first)) {
for (Instruction::op_iterator OI = I->op_begin(), OE = I->op_end();
OI != OE; ++OI)
Worklist.push_back(std::make_pair(OI->get(), Pair.second+1));
if (!UsedValues.empty()) return false;
IRBuilder<> Builder(PBI);
// If we need to invert the condition in the pred block to match, do so now.
if (InvertPredCond) {
Value *NewCond = PBI->getCondition();
if (NewCond->hasOneUse() && isa<CmpInst>(NewCond)) {
CmpInst *CI = cast<CmpInst>(NewCond);
} else {
NewCond = Builder.CreateNot(NewCond,
// If we have a bonus inst, clone it into the predecessor block.
Instruction *NewBonus = nullptr;
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.
PredBlock->getInstList().insert(PBI, NewBonus);
// Clone Cond into the predecessor basic block, and or/and the
// two conditions together.
Instruction *New = Cond->clone();
if (BonusInst) New->replaceUsesOfWith(BonusInst, NewBonus);
PredBlock->getInstList().insert(PBI, New);
if (BI->isConditional()) {
Instruction *NewCond =
cast<Instruction>(Builder.CreateBinOp(Opc, PBI->getCondition(),
New, "or.cond"));
uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight;
bool PredHasWeights = ExtractBranchMetadata(PBI, PredTrueWeight,
bool SuccHasWeights = ExtractBranchMetadata(BI, SuccTrueWeight,
SmallVector<uint64_t, 8> NewWeights;
if (PBI->getSuccessor(0) == BB) {
if (PredHasWeights && SuccHasWeights) {
// PBI: br i1 %x, BB, FalseDest
// BI: br i1 %y, TrueDest, FalseDest
//TrueWeight is TrueWeight for PBI * TrueWeight for BI.
NewWeights.push_back(PredTrueWeight * SuccTrueWeight);
//FalseWeight is FalseWeight for PBI * TotalWeight for BI +
// TrueWeight for PBI * FalseWeight for BI.
// We assume that total weights of a BranchInst can fit into 32 bits.
// Therefore, we will not have overflow using 64-bit arithmetic.
NewWeights.push_back(PredFalseWeight * (SuccFalseWeight +
SuccTrueWeight) + PredTrueWeight * SuccFalseWeight);
AddPredecessorToBlock(TrueDest, PredBlock, BB);
PBI->setSuccessor(0, TrueDest);
if (PBI->getSuccessor(1) == BB) {
if (PredHasWeights && SuccHasWeights) {
// PBI: br i1 %x, TrueDest, BB
// BI: br i1 %y, TrueDest, FalseDest
//TrueWeight is TrueWeight for PBI * TotalWeight for BI +
// FalseWeight for PBI * TrueWeight for BI.
NewWeights.push_back(PredTrueWeight * (SuccFalseWeight +
SuccTrueWeight) + PredFalseWeight * SuccTrueWeight);
//FalseWeight is FalseWeight for PBI * FalseWeight for BI.
NewWeights.push_back(PredFalseWeight * SuccFalseWeight);
AddPredecessorToBlock(FalseDest, PredBlock, BB);
PBI->setSuccessor(1, FalseDest);
if (NewWeights.size() == 2) {
// Halve the weights if any of them cannot fit in an uint32_t
SmallVector<uint32_t, 8> MDWeights(NewWeights.begin(),NewWeights.end());
} else
PBI->setMetadata(LLVMContext::MD_prof, nullptr);
} else {
// Update PHI nodes in the common successors.
for (unsigned i = 0, e = PHIs.size(); i != e; ++i) {
ConstantInt *PBI_C = cast<ConstantInt>(
Instruction *MergedCond = nullptr;
if (PBI->getSuccessor(0) == TrueDest) {
// Create (PBI_Cond and PBI_C) or (!PBI_Cond and BI_Value)
// PBI_C is true: PBI_Cond or (!PBI_Cond and BI_Value)
// is false: !PBI_Cond and BI_Value
Instruction *NotCond =
MergedCond =
NotCond, New,
if (PBI_C->isOne())
MergedCond =
PBI->getCondition(), MergedCond,
} else {
// Create (PBI_Cond and BI_Value) or (!PBI_Cond and PBI_C)
// PBI_C is true: (PBI_Cond and BI_Value) or (!PBI_Cond)
// is false: PBI_Cond and BI_Value
MergedCond =
PBI->getCondition(), New,
if (PBI_C->isOne()) {
Instruction *NotCond =
MergedCond =
NotCond, MergedCond,
// Update PHI Node.
// Change PBI from Conditional to Unconditional.
BranchInst *New_PBI = BranchInst::Create(TrueDest, PBI);
PBI = New_PBI;
// TODO: If BB is reachable from all paths through PredBlock, then we
// could replace PBI's branch probabilities with BI's.
// Copy any debug value intrinsics into the end of PredBlock.
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
if (isa<DbgInfoIntrinsic>(*I))
return true;
return false;
/// SimplifyCondBranchToCondBranch - If we have a conditional branch as a
/// predecessor of another block, this function tries to simplify it. We know
/// that PBI and BI are both conditional branches, and BI is in one of the
/// successor blocks of PBI - PBI branches to BI.
static bool SimplifyCondBranchToCondBranch(BranchInst *PBI, BranchInst *BI) {
assert(PBI->isConditional() && BI->isConditional());
BasicBlock *BB = BI->getParent();
// If this block ends with a branch instruction, and if there is a
// predecessor that ends on a branch of the same condition, make
// this conditional branch redundant.
if (PBI->getCondition() == BI->getCondition() &&
PBI->getSuccessor(0) != PBI->getSuccessor(1)) {
// Okay, the outcome of this conditional branch is statically
// knowable. If this block had a single pred, handle specially.
if (BB->getSinglePredecessor()) {
// Turn this into a branch on constant.
bool CondIsTrue = PBI->getSuccessor(0) == BB;
return true; // Nuke the branch on constant.
// Otherwise, if there are multiple predecessors, insert a PHI that merges
// in the constant and simplify the block result. Subsequent passes of
// simplifycfg will thread the block.
if (BlockIsSimpleEnoughToThreadThrough(BB)) {
pred_iterator PB = pred_begin(BB), PE = pred_end(BB);
PHINode *NewPN = PHINode::Create(Type::getInt1Ty(BB->getContext()),
std::distance(PB, PE),
BI->getCondition()->getName() + ".pr",
// Okay, we're going to insert the PHI node. Since PBI is not the only
// predecessor, compute the PHI'd conditional value for all of the preds.
// Any predecessor where the condition is not computable we keep symbolic.
for (pred_iterator PI = PB; PI != PE; ++PI) {
BasicBlock *P = *PI;
if ((PBI = dyn_cast<BranchInst>(P->getTerminator())) &&
PBI != BI && PBI->isConditional() &&
PBI->getCondition() == BI->getCondition() &&
PBI->getSuccessor(0) != PBI->getSuccessor(1)) {
bool CondIsTrue = PBI->getSuccessor(0) == BB;
CondIsTrue), P);
} else {
NewPN->addIncoming(BI->getCondition(), P);
return true;
// If this is a conditional branch in an empty block, and if any
// predecessors are a conditional branch to one of our destinations,
// fold the conditions into logical ops and one cond br.
BasicBlock::iterator BBI = BB->begin();
// Ignore dbg intrinsics.
while (isa<DbgInfoIntrinsic>(BBI))
if (&*BBI != BI)
return false;
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(BI->getCondition()))
if (CE->canTrap())
return false;
int PBIOp, BIOp;
if (PBI->getSuccessor(0) == BI->getSuccessor(0))
PBIOp = BIOp = 0;
else if (PBI->getSuccessor(0) == BI->getSuccessor(1))
PBIOp = 0, BIOp = 1;
else if (PBI->getSuccessor(1) == BI->getSuccessor(0))
PBIOp = 1, BIOp = 0;
else if (PBI->getSuccessor(1) == BI->getSuccessor(1))
PBIOp = BIOp = 1;
return false;
// Check to make sure that the other destination of this branch
// isn't BB itself. If so, this is an infinite loop that will
// keep getting unwound.
if (PBI->getSuccessor(PBIOp) == BB)
return false;
// Do not perform this transformation if it would require
// insertion of a large number of select instructions. For targets
// without predication/cmovs, this is a big pessimization.
// Also do not perform this transformation if any phi node in the common
// destination block can trap when reached by BB or PBB (PR17073). In that