bool BranchProbabilityAnalysis::calcZeroHeuristics(BasicBlock *BB) {
BranchInst * BI = dyn_cast<BranchInst>(BB->getTerminator());
if (!BI || !BI->isConditional())
return false;
Value *Cond = BI->getCondition();
ICmpInst *CI = dyn_cast<ICmpInst>(Cond);
if (!CI)
return false;
Value *RHS = CI->getOperand(1);
ConstantInt *CV = dyn_cast<ConstantInt>(RHS);
if (!CV)
return false;
bool isProb;
if (CV->isZero()) {
switch (CI->getPredicate()) {
case CmpInst::ICMP_EQ:
// X == 0 -> Unlikely
isProb = false;
break;
case CmpInst::ICMP_NE:
// X != 0 -> Likely
isProb = true;
break;
case CmpInst::ICMP_SLT:
// X < 0 -> Unlikely
isProb = false;
break;
case CmpInst::ICMP_SGT:
// X > 0 -> Likely
isProb = true;
break;
default:
return false;
}
} else if (CV->isOne() && CI->getPredicate() == CmpInst::ICMP_SLT) {
// InstCombine canonicalizes X <= 0 into X < 1.
// X <= 0 -> Unlikely
isProb = false;
} else if (CV->isAllOnesValue() && CI->getPredicate() == CmpInst::ICMP_SGT) {
// InstCombine canonicalizes X >= 0 into X > -1.
// X >= 0 -> Likely
isProb = true;
} else {
return false;
}
BasicBlock *Taken = BI->getSuccessor(0);
BasicBlock *NonTaken = BI->getSuccessor(1);
if (!isProb)
std::swap(Taken, NonTaken);
BP->setEdgeWeight(BB, Taken, ZH_TAKEN_WEIGHT);
BP->setEdgeWeight(BB, NonTaken, ZH_NONTAKEN_WEIGHT);
return true;
}
bool AMDGPUCodeGenPrepare::promoteUniformOpToI32(ICmpInst &I) const {
assert(needsPromotionToI32(I.getOperand(0)->getType()) &&
"I does not need promotion to i32");
IRBuilder<> Builder(&I);
Builder.SetCurrentDebugLocation(I.getDebugLoc());
Type *I32Ty = getI32Ty(Builder, I.getOperand(0)->getType());
Value *ExtOp0 = nullptr;
Value *ExtOp1 = nullptr;
Value *NewICmp = nullptr;
if (I.isSigned()) {
ExtOp0 = Builder.CreateSExt(I.getOperand(0), I32Ty);
ExtOp1 = Builder.CreateSExt(I.getOperand(1), I32Ty);
} else {
ExtOp0 = Builder.CreateZExt(I.getOperand(0), I32Ty);
ExtOp1 = Builder.CreateZExt(I.getOperand(1), I32Ty);
}
NewICmp = Builder.CreateICmp(I.getPredicate(), ExtOp0, ExtOp1);
I.replaceAllUsesWith(NewICmp);
I.eraseFromParent();
return true;
}
// Calculate Edge Weights using "Pointer Heuristics". Predict a comparsion
// between two pointer or pointer and NULL will fail.
bool BranchProbabilityInfo::calcPointerHeuristics(BasicBlock *BB) {
BranchInst * BI = dyn_cast<BranchInst>(BB->getTerminator());
if (!BI || !BI->isConditional())
return false;
Value *Cond = BI->getCondition();
ICmpInst *CI = dyn_cast<ICmpInst>(Cond);
if (!CI || !CI->isEquality())
return false;
Value *LHS = CI->getOperand(0);
if (!LHS->getType()->isPointerTy())
return false;
assert(CI->getOperand(1)->getType()->isPointerTy());
// p != 0 -> isProb = true
// p == 0 -> isProb = false
// p != q -> isProb = true
// p == q -> isProb = false;
unsigned TakenIdx = 0, NonTakenIdx = 1;
bool isProb = CI->getPredicate() == ICmpInst::ICMP_NE;
if (!isProb)
std::swap(TakenIdx, NonTakenIdx);
setEdgeWeight(BB, TakenIdx, PH_TAKEN_WEIGHT);
setEdgeWeight(BB, NonTakenIdx, PH_NONTAKEN_WEIGHT);
return true;
}
void IndVarSimplify::EliminateIVComparisons() {
// Look for ICmp users.
for (IVUsers::iterator I = IU->begin(), E = IU->end(); I != E; ++I) {
IVStrideUse &UI = *I;
ICmpInst *ICmp = dyn_cast<ICmpInst>(UI.getUser());
if (!ICmp) continue;
bool Swapped = UI.getOperandValToReplace() == ICmp->getOperand(1);
ICmpInst::Predicate Pred = ICmp->getPredicate();
if (Swapped) Pred = ICmpInst::getSwappedPredicate(Pred);
// Get the SCEVs for the ICmp operands.
const SCEV *S = IU->getReplacementExpr(UI);
const SCEV *X = SE->getSCEV(ICmp->getOperand(!Swapped));
// Simplify unnecessary loops away.
const Loop *ICmpLoop = LI->getLoopFor(ICmp->getParent());
S = SE->getSCEVAtScope(S, ICmpLoop);
X = SE->getSCEVAtScope(X, ICmpLoop);
// If the condition is always true or always false, replace it with
// a constant value.
if (SE->isKnownPredicate(Pred, S, X))
ICmp->replaceAllUsesWith(ConstantInt::getTrue(ICmp->getContext()));
else if (SE->isKnownPredicate(ICmpInst::getInversePredicate(Pred), S, X))
ICmp->replaceAllUsesWith(ConstantInt::getFalse(ICmp->getContext()));
else
continue;
DEBUG(dbgs() << "INDVARS: Eliminated comparison: " << *ICmp << '\n');
DeadInsts.push_back(ICmp);
}
}
bool visitICmpInst(ICmpInst& inst) {
// errs() << "got icmp instruction! " << inst << '\n';
bool changed = false;
if (inst.getPredicate() == CmpInst::ICMP_EQ) {
assert(inst.getNumOperands() == 2);
if (inst.getOperand(1) == processing) {
inst.swapOperands();
changed = true;
any_changes = true;
}
assert(dyn_cast<Instruction>(inst.getOperand(0)) == processing);
Value* other = inst.getOperand(1);
if (isa<ConstantPointerNull>(other)) {
if (VERBOSITY("opt") >= 2) {
errs() << inst << '\n';
errs() << "replacing with false!\n";
}
Value* new_value = ConstantInt::getFalse(other->getContext());
inst.replaceAllUsesWith(new_value);
inst.eraseFromParent();
changed = true;
any_changes = true;
}
}
return changed;
}
/// MatchSelectPattern - Pattern match integer [SU]MIN, [SU]MAX, and ABS idioms,
/// returning the kind and providing the out parameter results if we
/// successfully match.
static SelectPatternFlavor
MatchSelectPattern(Value *V, Value *&LHS, Value *&RHS) {
SelectInst *SI = dyn_cast<SelectInst>(V);
if (SI == 0) return SPF_UNKNOWN;
ICmpInst *ICI = dyn_cast<ICmpInst>(SI->getCondition());
if (ICI == 0) return SPF_UNKNOWN;
LHS = ICI->getOperand(0);
RHS = ICI->getOperand(1);
// (icmp X, Y) ? X : Y
if (SI->getTrueValue() == ICI->getOperand(0) &&
SI->getFalseValue() == ICI->getOperand(1)) {
switch (ICI->getPredicate()) {
default: return SPF_UNKNOWN; // Equality.
case ICmpInst::ICMP_UGT:
case ICmpInst::ICMP_UGE: return SPF_UMAX;
case ICmpInst::ICMP_SGT:
case ICmpInst::ICMP_SGE: return SPF_SMAX;
case ICmpInst::ICMP_ULT:
case ICmpInst::ICMP_ULE: return SPF_UMIN;
case ICmpInst::ICMP_SLT:
case ICmpInst::ICMP_SLE: return SPF_SMIN;
}
}
// (icmp X, Y) ? Y : X
if (SI->getTrueValue() == ICI->getOperand(1) &&
SI->getFalseValue() == ICI->getOperand(0)) {
switch (ICI->getPredicate()) {
default: return SPF_UNKNOWN; // Equality.
case ICmpInst::ICMP_UGT:
case ICmpInst::ICMP_UGE: return SPF_UMIN;
case ICmpInst::ICMP_SGT:
case ICmpInst::ICMP_SGE: return SPF_SMIN;
case ICmpInst::ICMP_ULT:
case ICmpInst::ICMP_ULE: return SPF_UMAX;
case ICmpInst::ICMP_SLT:
case ICmpInst::ICMP_SLE: return SPF_SMAX;
}
}
// TODO: (X > 4) ? X : 5 --> (X >= 5) ? X : 5 --> MAX(X, 5)
return SPF_UNKNOWN;
}
bool BranchProbabilityAnalysis::calcZeroHeuristics(BasicBlock *BB) {
BranchInst * BI = dyn_cast<BranchInst>(BB->getTerminator());
if (!BI || !BI->isConditional())
return false;
Value *Cond = BI->getCondition();
ICmpInst *CI = dyn_cast<ICmpInst>(Cond);
if (!CI)
return false;
Value *RHS = CI->getOperand(1);
ConstantInt *CV = dyn_cast<ConstantInt>(RHS);
if (!CV || !CV->isZero())
return false;
bool isProb;
switch (CI->getPredicate()) {
case CmpInst::ICMP_EQ:
// Equal to zero is not expected to be taken.
isProb = false;
break;
case CmpInst::ICMP_NE:
// Not equal to zero is expected.
isProb = true;
break;
case CmpInst::ICMP_SLT:
// Less or equal to zero is not expected.
// X < 0 -> Unlikely
isProb = false;
break;
case CmpInst::ICMP_UGT:
case CmpInst::ICMP_SGT:
// Greater or equal to zero is expected.
// X > 0 -> Likely
isProb = true;
break;
default:
return false;
};
BasicBlock *Taken = BI->getSuccessor(0);
BasicBlock *NonTaken = BI->getSuccessor(1);
if (!isProb)
std::swap(Taken, NonTaken);
BP->setEdgeWeight(BB, Taken, ZH_TAKEN_WEIGHT);
BP->setEdgeWeight(BB, NonTaken, ZH_NONTAKEN_WEIGHT);
return true;
}
bool LowerExpectIntrinsic::HandleIfExpect(BranchInst *BI) {
if (BI->isUnconditional())
return false;
// Handle non-optimized IR code like:
// %expval = call i64 @llvm.expect.i64(i64 %conv1, i64 1)
// %tobool = icmp ne i64 %expval, 0
// br i1 %tobool, label %if.then, label %if.end
//
// Or the following simpler case:
// %expval = call i1 @llvm.expect.i1(i1 %cmp, i1 1)
// br i1 %expval, label %if.then, label %if.end
CallInst *CI;
ICmpInst *CmpI = dyn_cast<ICmpInst>(BI->getCondition());
if (!CmpI) {
CI = dyn_cast<CallInst>(BI->getCondition());
} else {
if (CmpI->getPredicate() != CmpInst::ICMP_NE)
return false;
CI = dyn_cast<CallInst>(CmpI->getOperand(0));
}
if (!CI)
return false;
Function *Fn = CI->getCalledFunction();
if (!Fn || Fn->getIntrinsicID() != Intrinsic::expect)
return false;
Value *ArgValue = CI->getArgOperand(0);
ConstantInt *ExpectedValue = dyn_cast<ConstantInt>(CI->getArgOperand(1));
if (!ExpectedValue)
return false;
MDBuilder MDB(CI->getContext());
MDNode *Node;
// If expect value is equal to 1 it means that we are more likely to take
// branch 0, in other case more likely is branch 1.
if (ExpectedValue->isOne())
Node = MDB.createBranchWeights(LikelyBranchWeight, UnlikelyBranchWeight);
else
Node = MDB.createBranchWeights(UnlikelyBranchWeight, LikelyBranchWeight);
BI->setMetadata(LLVMContext::MD_prof, Node);
if (CmpI)
CmpI->setOperand(0, ArgValue);
else
BI->setCondition(ArgValue);
return true;
}
void TempScopInfo::buildAffineCondition(Value &V, bool inverted,
Comparison **Comp,
TempScop &Scop) const {
Region &R = Scop.getMaxRegion();
ParamSetType &Params = Scop.getParamSet();
if (ConstantInt *C = dyn_cast<ConstantInt>(&V)) {
// If this is always true condition, we will create 1 >= 0,
// otherwise we will create 1 == 0.
SCEVAffFunc *AffLHS = new SCEVAffFunc(SE->getConstant(C->getType(), 0),
SCEVAffFunc::Eq, R, Params, LI, SE);
SCEVAffFunc *AffRHS = new SCEVAffFunc(SE->getConstant(C->getType(), 1),
SCEVAffFunc::Eq, R, Params, LI, SE);
if (C->isOne() == inverted)
*Comp = new Comparison(AffRHS, AffLHS, ICmpInst::ICMP_NE);
else
*Comp = new Comparison(AffLHS, AffLHS, ICmpInst::ICMP_EQ);
return;
}
ICmpInst *ICmp = dyn_cast<ICmpInst>(&V);
assert(ICmp && "Only ICmpInst of constant as condition supported!");
const SCEV *LHS = SE->getSCEV(ICmp->getOperand(0)),
*RHS = SE->getSCEV(ICmp->getOperand(1));
ICmpInst::Predicate Pred = ICmp->getPredicate();
// Invert the predicate if needed.
if (inverted)
Pred = ICmpInst::getInversePredicate(Pred);
SCEVAffFunc *AffLHS = new SCEVAffFunc(LHS, SCEVAffFunc::Eq, R, Params, LI,
SE);
SCEVAffFunc *AffRHS = new SCEVAffFunc(RHS, SCEVAffFunc::Eq, R, Params, LI,
SE);
switch (Pred) {
case ICmpInst::ICMP_UGT:
case ICmpInst::ICMP_UGE:
case ICmpInst::ICMP_ULT:
case ICmpInst::ICMP_ULE:
// TODO: At the moment we need to see everything as signed. This is an
// correctness issue that needs to be solved.
//AffLHS->setUnsigned();
//AffRHS->setUnsigned();
break;
default:
break;
}
*Comp = new Comparison(AffLHS, AffRHS, Pred);
}
void TempScopInfo::buildAffineCondition(Value &V, bool inverted,
Comparison **Comp) const {
if (ConstantInt *C = dyn_cast<ConstantInt>(&V)) {
// If this is always true condition, we will create 0 <= 1,
// otherwise we will create 0 >= 1.
const SCEV *LHS = SE->getConstant(C->getType(), 0);
const SCEV *RHS = SE->getConstant(C->getType(), 1);
if (C->isOne() == inverted)
*Comp = new Comparison(LHS, RHS, ICmpInst::ICMP_SLE);
else
*Comp = new Comparison(LHS, RHS, ICmpInst::ICMP_SGE);
return;
}
ICmpInst *ICmp = dyn_cast<ICmpInst>(&V);
assert(ICmp && "Only ICmpInst of constant as condition supported!");
Loop *L = LI->getLoopFor(ICmp->getParent());
const SCEV *LHS = SE->getSCEVAtScope(ICmp->getOperand(0), L);
const SCEV *RHS = SE->getSCEVAtScope(ICmp->getOperand(1), L);
ICmpInst::Predicate Pred = ICmp->getPredicate();
// Invert the predicate if needed.
if (inverted)
Pred = ICmpInst::getInversePredicate(Pred);
switch (Pred) {
case ICmpInst::ICMP_UGT:
case ICmpInst::ICMP_UGE:
case ICmpInst::ICMP_ULT:
case ICmpInst::ICMP_ULE:
// TODO: At the moment we need to see everything as signed. This is an
// correctness issue that needs to be solved.
// AffLHS->setUnsigned();
// AffRHS->setUnsigned();
break;
default:
break;
}
*Comp = new Comparison(LHS, RHS, Pred);
}
bool LowerExpectIntrinsic::HandleIfExpect(BranchInst *BI) {
if (BI->isUnconditional())
return false;
// Handle non-optimized IR code like:
// %expval = call i64 @llvm.expect.i64.i64(i64 %conv1, i64 1)
// %tobool = icmp ne i64 %expval, 0
// br i1 %tobool, label %if.then, label %if.end
ICmpInst *CmpI = dyn_cast<ICmpInst>(BI->getCondition());
if (!CmpI || CmpI->getPredicate() != CmpInst::ICMP_NE)
return false;
CallInst *CI = dyn_cast<CallInst>(CmpI->getOperand(0));
if (!CI)
return false;
Function *Fn = CI->getCalledFunction();
if (!Fn || Fn->getIntrinsicID() != Intrinsic::expect)
return false;
Value *ArgValue = CI->getArgOperand(0);
ConstantInt *ExpectedValue = dyn_cast<ConstantInt>(CI->getArgOperand(1));
if (!ExpectedValue)
return false;
LLVMContext &Context = CI->getContext();
const Type *Int32Ty = Type::getInt32Ty(Context);
bool Likely = ExpectedValue->isOne();
// If expect value is equal to 1 it means that we are more likely to take
// branch 0, in other case more likely is branch 1.
Value *Ops[] = {
MDString::get(Context, "branch_weights"),
ConstantInt::get(Int32Ty, Likely ? LikelyBranchWeight : UnlikelyBranchWeight),
ConstantInt::get(Int32Ty, Likely ? UnlikelyBranchWeight : LikelyBranchWeight)
};
MDNode *WeightsNode = MDNode::get(Context, Ops);
BI->setMetadata(LLVMContext::MD_prof, WeightsNode);
CmpI->setOperand(0, ArgValue);
return true;
}
void IndVarSimplify::EliminateIVComparisons() {
SmallVector<WeakVH, 16> DeadInsts;
// Look for ICmp users.
for (IVUsers::iterator I = IU->begin(), E = IU->end(); I != E; ++I) {
IVStrideUse &UI = *I;
ICmpInst *ICmp = dyn_cast<ICmpInst>(UI.getUser());
if (!ICmp) continue;
bool Swapped = UI.getOperandValToReplace() == ICmp->getOperand(1);
ICmpInst::Predicate Pred = ICmp->getPredicate();
if (Swapped) Pred = ICmpInst::getSwappedPredicate(Pred);
// Get the SCEVs for the ICmp operands.
const SCEV *S = IU->getReplacementExpr(UI);
const SCEV *X = SE->getSCEV(ICmp->getOperand(!Swapped));
// Simplify unnecessary loops away.
const Loop *ICmpLoop = LI->getLoopFor(ICmp->getParent());
S = SE->getSCEVAtScope(S, ICmpLoop);
X = SE->getSCEVAtScope(X, ICmpLoop);
// If the condition is always true or always false, replace it with
// a constant value.
if (SE->isKnownPredicate(Pred, S, X))
ICmp->replaceAllUsesWith(ConstantInt::getTrue(ICmp->getContext()));
else if (SE->isKnownPredicate(ICmpInst::getInversePredicate(Pred), S, X))
ICmp->replaceAllUsesWith(ConstantInt::getFalse(ICmp->getContext()));
else
continue;
DEBUG(dbgs() << "INDVARS: Eliminated comparison: " << *ICmp << '\n');
DeadInsts.push_back(ICmp);
}
// Now that we're done iterating through lists, clean up any instructions
// which are now dead.
while (!DeadInsts.empty())
if (Instruction *Inst =
dyn_cast_or_null<Instruction>(&*DeadInsts.pop_back_val()))
RecursivelyDeleteTriviallyDeadInstructions(Inst);
}
int BranchProbabilities::CheckPointerHeuristic()
{
// Heuristic fails if the last instruction is not a conditional branch
BranchInst *BI = dyn_cast<BranchInst>(_TI);
if ((!BI) || (BI->isUnconditional()))
return -1;
// All pointer comparisons are done with the icmp instruction
ICmpInst *icmp = dyn_cast<ICmpInst>(BI->getCondition());
if (!icmp)
return -1;
Value *v[2];
v[0] = icmp->getOperand(0);
v[1] = icmp->getOperand(1);
// Make sure we're comparing pointers
if (isa<PointerType>(v[0]->getType()))
{
assert(isa<PointerType>(v[1]->getType()) && "v[1] is not a pointer!");
// Choose the prefered branch depending on if this is an eq or neq comp
switch (icmp->getPredicate())
{
case ICmpInst::ICMP_EQ:
return 1;
case ICmpInst::ICMP_NE:
return 0;
default:
assert("non-equality comparison of pointers");
return -1;
}
}
return -1;
}
bool RelativeMinMax::getMinMax(Expr Ex, Expr &Min, Expr &Max) {
if (Ex.isConstant()) {
Min = Ex;
Max = Ex;
} else if (Ex.isSymbol()) {
// Bounds of induction variables have special treatment.
if (PHINode *Phi = dyn_cast<PHINode>(Ex.getSymbolValue())) {
if (Loop *L = LIE_->getLoopForInductionVariable(Phi)) {
Expr IndvarStart, IndvarEnd, IndvarStep;
LIE_->getLoopInfo(L, Phi, IndvarStart, IndvarEnd, IndvarStep);
BranchInst *BI = cast<BranchInst>(L->getExitingBlock()->getTerminator());
ICmpInst *ICI = cast<ICmpInst>(BI->getCondition());
Expr MinStart, MaxStart, MinEnd, MaxEnd;
if (!getMinMax(IndvarStart, MinStart, MaxStart) ||
!getMinMax(IndvarEnd, MinEnd, MaxEnd)) {
RMM_DEBUG(dbgs() << "RelativeMinMax: Could not infer min/max for "
<< IndvarStart << " and/or" << IndvarEnd << "\n");
return false;
}
// FIXME: we should wrap the loop in a conditional so that the following
// min/max assumptions always hold.
switch (ICI->getPredicate()) {
case CmpInst::ICMP_SLT:
case CmpInst::ICMP_ULT:
case CmpInst::ICMP_SLE:
case CmpInst::ICMP_ULE:
Min = MinStart;
Max = MaxEnd;
break;
case CmpInst::ICMP_SGT:
case CmpInst::ICMP_UGT:
case CmpInst::ICMP_UGE:
case CmpInst::ICMP_SGE:
Min = MaxStart;
Max = MinEnd;
break;
default:
llvm_unreachable("Invalid comparison predicate");
}
RMM_DEBUG(dbgs() << "RelativeMinMax: min/max for induction variable "
<< *Phi << ": " << Min << ", " << Max << "\n");
return true;
}
}
Min = Ex;
Max = Ex;
} else if (Ex.isAdd()) {
for (auto SubEx : Ex) {
Expr TmpMin, TmpMax;
if (!getMinMax(SubEx, TmpMin, TmpMax)) {
RMM_DEBUG(dbgs() << "RelativeMinMax: Could not infer min/max for "
<< SubEx << "\n");
return false;
}
addMinMax(TmpMin, TmpMax, Min, Max, Min, Max);
}
} else if (Ex.isMul()) {
Min = Expr::InvalidExpr();
for (auto SubEx : Ex) {
Expr TmpMin, TmpMax;
if (!getMinMax(SubEx, TmpMin, TmpMax)) {
RMM_DEBUG(dbgs() << "RelativeMinMax: Could not infer min/max for "
<< SubEx << "\n");
return false;
}
if (!Min.isValid()) {
Min = TmpMin;
Max = TmpMax;
} else {
mulMinMax(TmpMin, TmpMax, Min, Max, Min, Max);
}
}
} else if (Ex.isPow()) {
if (!Ex.getPowExp().isConstant()) {
RMM_DEBUG(dbgs() << "RelativeMinMax: non-constant exponent\n");
return false;
}
Expr BaseMin, BaseMax;
if (!getMinMax(Ex.getPowBase(), BaseMin, BaseMax)) {
RMM_DEBUG(dbgs() << "RelativeMinMax: Could not infer min/max for "
<< Ex.getPowBase() << "\n");
return false;
}
if (Ex.getPowExp().isPositive()) {
Min = BaseMin ^ Ex.getPowExp();
Max = BaseMax ^ Ex.getPowExp();
} else {
Min = BaseMax ^ Ex.getPowExp();
Max = BaseMin ^ Ex.getPowExp();
}
} else if (Ex.isMin()) {
Expr MinFirst, MinSecond, Bogus;
getMinMax(Ex.at(0), MinFirst, Bogus);
getMinMax(Ex.at(1), MinSecond, Bogus);
Min = Max = MinFirst.min(MinSecond);
} else if (Ex.isMax()) {
Expr MaxFirst, MaxSecond, Bogus;
getMinMax(Ex.at(0), MaxFirst, Bogus);
getMinMax(Ex.at(1), MaxSecond, Bogus);
Min = Max = MaxFirst.max(MaxSecond);
} else {
RMM_DEBUG(dbgs() << "RelativeMinMax: unhandled expression: " << Ex << "\n");
return false;
}
return true;
}
bool CallAnalyzer::visitICmp(ICmpInst &I) {
Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);
// First try to handle simplified comparisons.
if (!isa<Constant>(LHS))
if (Constant *SimpleLHS = SimplifiedValues.lookup(LHS))
LHS = SimpleLHS;
if (!isa<Constant>(RHS))
if (Constant *SimpleRHS = SimplifiedValues.lookup(RHS))
RHS = SimpleRHS;
if (Constant *CLHS = dyn_cast<Constant>(LHS))
if (Constant *CRHS = dyn_cast<Constant>(RHS))
if (Constant *C = ConstantExpr::getICmp(I.getPredicate(), CLHS, CRHS)) {
SimplifiedValues[&I] = C;
return true;
}
// Otherwise look for a comparison between constant offset pointers with
// a common base.
Value *LHSBase, *RHSBase;
APInt LHSOffset, RHSOffset;
llvm::tie(LHSBase, LHSOffset) = ConstantOffsetPtrs.lookup(LHS);
if (LHSBase) {
llvm::tie(RHSBase, RHSOffset) = ConstantOffsetPtrs.lookup(RHS);
if (RHSBase && LHSBase == RHSBase) {
// We have common bases, fold the icmp to a constant based on the
// offsets.
Constant *CLHS = ConstantInt::get(LHS->getContext(), LHSOffset);
Constant *CRHS = ConstantInt::get(RHS->getContext(), RHSOffset);
if (Constant *C = ConstantExpr::getICmp(I.getPredicate(), CLHS, CRHS)) {
SimplifiedValues[&I] = C;
++NumConstantPtrCmps;
return true;
}
}
}
// If the comparison is an equality comparison with null, we can simplify it
// for any alloca-derived argument.
if (I.isEquality() && isa<ConstantPointerNull>(I.getOperand(1)))
if (isAllocaDerivedArg(I.getOperand(0))) {
// We can actually predict the result of comparisons between an
// alloca-derived value and null. Note that this fires regardless of
// SROA firing.
bool IsNotEqual = I.getPredicate() == CmpInst::ICMP_NE;
SimplifiedValues[&I] = IsNotEqual ? ConstantInt::getTrue(I.getType())
: ConstantInt::getFalse(I.getType());
return true;
}
// Finally check for SROA candidates in comparisons.
Value *SROAArg;
DenseMap<Value *, int>::iterator CostIt;
if (lookupSROAArgAndCost(I.getOperand(0), SROAArg, CostIt)) {
if (isa<ConstantPointerNull>(I.getOperand(1))) {
accumulateSROACost(CostIt, InlineConstants::InstrCost);
return true;
}
disableSROA(CostIt);
}
return false;
}
bool BranchProbabilityInfo::calcZeroHeuristics(BasicBlock *BB) {
BranchInst * BI = dyn_cast<BranchInst>(BB->getTerminator());
if (!BI || !BI->isConditional())
return false;
Value *Cond = BI->getCondition();
ICmpInst *CI = dyn_cast<ICmpInst>(Cond);
if (!CI)
return false;
Value *RHS = CI->getOperand(1);
ConstantInt *CV = dyn_cast<ConstantInt>(RHS);
if (!CV)
return false;
// If the LHS is the result of AND'ing a value with a single bit bitmask,
// we don't have information about probabilities.
if (Instruction *LHS = dyn_cast<Instruction>(CI->getOperand(0)))
if (LHS->getOpcode() == Instruction::And)
if (ConstantInt *AndRHS = dyn_cast<ConstantInt>(LHS->getOperand(1)))
if (AndRHS->getUniqueInteger().isPowerOf2())
return false;
bool isProb;
if (CV->isZero()) {
switch (CI->getPredicate()) {
case CmpInst::ICMP_EQ:
// X == 0 -> Unlikely
isProb = false;
break;
case CmpInst::ICMP_NE:
// X != 0 -> Likely
isProb = true;
break;
case CmpInst::ICMP_SLT:
// X < 0 -> Unlikely
isProb = false;
break;
case CmpInst::ICMP_SGT:
// X > 0 -> Likely
isProb = true;
break;
default:
return false;
}
} else if (CV->isOne() && CI->getPredicate() == CmpInst::ICMP_SLT) {
// InstCombine canonicalizes X <= 0 into X < 1.
// X <= 0 -> Unlikely
isProb = false;
} else if (CV->isAllOnesValue()) {
switch (CI->getPredicate()) {
case CmpInst::ICMP_EQ:
// X == -1 -> Unlikely
isProb = false;
break;
case CmpInst::ICMP_NE:
// X != -1 -> Likely
isProb = true;
break;
case CmpInst::ICMP_SGT:
// InstCombine canonicalizes X >= 0 into X > -1.
// X >= 0 -> Likely
isProb = true;
break;
default:
return false;
}
} else {
return false;
}
unsigned TakenIdx = 0, NonTakenIdx = 1;
if (!isProb)
std::swap(TakenIdx, NonTakenIdx);
BranchProbability TakenProb(ZH_TAKEN_WEIGHT,
ZH_TAKEN_WEIGHT + ZH_NONTAKEN_WEIGHT);
setEdgeProbability(BB, TakenIdx, TakenProb);
setEdgeProbability(BB, NonTakenIdx, TakenProb.getCompl());
return true;
}
int BranchProbabilities::CheckIntegerHeuristic()
{
// Heuristic fails if the last instruction is not a conditional branch
BranchInst *BI = dyn_cast<BranchInst>(_TI);
if ((!BI) || (BI->isUnconditional()))
return -1;
// All integer comparisons are done with the icmp instruction
ICmpInst *icmp = dyn_cast<ICmpInst>(BI->getCondition());
if (!icmp)
return -1;
Value *v[2];
v[0] = icmp->getOperand(0);
v[1] = icmp->getOperand(1);
// If neither is a constant, nothing to do
if (!isa<ConstantInt>(v[0]) && !isa<ConstantInt>(v[1]))
return -1;
// If we're dealing with something other than ints, nothing to do
if (!isa<IntegerType>(v[0]->getType()))
return -1;
// Get comparison
ICmpInst::Predicate pred = icmp->getPredicate();
// Eq and Not Eq are easy cases
if (pred == ICmpInst::ICMP_EQ)
return 1;
else if (pred == ICmpInst::ICMP_NE)
return 0;
ConstantInt *CI = dyn_cast<ConstantInt>(v[1]);
// If the right side isn't a constant, swap the predicate so we can pretend
if (!CI)
{
pred = icmp->getSwappedPredicate();
CI = cast<ConstantInt>(v[0]);
}
// Choose the appropriate branch depending on the const val and predicate
if (CI->isZero())
{
switch (pred)
{
case ICmpInst::ICMP_UGE:
assert("UGE zero always returns true");
return -1;
case ICmpInst::ICMP_ULT:
assert("ULT zero always returns false");
return -1;
case ICmpInst::ICMP_UGT:
case ICmpInst::ICMP_SGT:
case ICmpInst::ICMP_SGE:
return 0;
case ICmpInst::ICMP_ULE:
case ICmpInst::ICMP_SLT:
case ICmpInst::ICMP_SLE:
return 1;
default:
return -1;
}
}
else if (CI->isOne())
{
switch (pred)
{
case ICmpInst::ICMP_UGE:
case ICmpInst::ICMP_SGE:
return 0;
case ICmpInst::ICMP_ULT:
case ICmpInst::ICMP_SLT:
return 1;
default:
return -1;
}
}
else if (CI->isAllOnesValue())
{
switch (pred)
{
case ICmpInst::ICMP_SGT:
return 0;
case ICmpInst::ICMP_SLE:
return 1;
default:
return -1;
}
}
return -1;
}
bool LoopIndexSplit::splitLoop() {
SplitCondition = NULL;
if (ExitCondition->getPredicate() == ICmpInst::ICMP_NE
|| ExitCondition->getPredicate() == ICmpInst::ICMP_EQ)
return false;
BasicBlock *Header = L->getHeader();
BasicBlock *Latch = L->getLoopLatch();
BranchInst *SBR = NULL; // Split Condition Branch
BranchInst *EBR = cast<BranchInst>(ExitCondition->getParent()->getTerminator());
// If Exiting block includes loop variant instructions then this
// loop may not be split safely.
BasicBlock *ExitingBlock = ExitCondition->getParent();
if (!cleanBlock(ExitingBlock)) return false;
LLVMContext &Context = Header->getContext();
for (Loop::block_iterator I = L->block_begin(), E = L->block_end();
I != E; ++I) {
BranchInst *BR = dyn_cast<BranchInst>((*I)->getTerminator());
if (!BR || BR->isUnconditional()) continue;
ICmpInst *CI = dyn_cast<ICmpInst>(BR->getCondition());
if (!CI || CI == ExitCondition
|| CI->getPredicate() == ICmpInst::ICMP_NE
|| CI->getPredicate() == ICmpInst::ICMP_EQ)
continue;
// Unable to handle triangle loops at the moment.
// In triangle loop, split condition is in header and one of the
// the split destination is loop latch. If split condition is EQ
// then such loops are already handle in processOneIterationLoop().
if (Header == (*I)
&& (Latch == BR->getSuccessor(0) || Latch == BR->getSuccessor(1)))
continue;
// If the block does not dominate the latch then this is not a diamond.
// Such loop may not benefit from index split.
if (!DT->dominates((*I), Latch))
continue;
// If split condition branches heads do not have single predecessor,
// SplitCondBlock, then is not possible to remove inactive branch.
if (!BR->getSuccessor(0)->getSinglePredecessor()
|| !BR->getSuccessor(1)->getSinglePredecessor())
return false;
// If the merge point for BR is not loop latch then skip this condition.
if (BR->getSuccessor(0) != Latch) {
DominanceFrontier::iterator DF0 = DF->find(BR->getSuccessor(0));
assert (DF0 != DF->end() && "Unable to find dominance frontier");
if (!DF0->second.count(Latch))
continue;
}
if (BR->getSuccessor(1) != Latch) {
DominanceFrontier::iterator DF1 = DF->find(BR->getSuccessor(1));
assert (DF1 != DF->end() && "Unable to find dominance frontier");
if (!DF1->second.count(Latch))
continue;
}
SplitCondition = CI;
SBR = BR;
break;
}
if (!SplitCondition)
return false;
// If the predicate sign does not match then skip.
if (ExitCondition->isSigned() != SplitCondition->isSigned())
return false;
unsigned EVOpNum = (ExitCondition->getOperand(1) == IVExitValue);
unsigned SVOpNum = IVBasedValues.count(SplitCondition->getOperand(0));
Value *SplitValue = SplitCondition->getOperand(SVOpNum);
if (!L->isLoopInvariant(SplitValue))
return false;
if (!IVBasedValues.count(SplitCondition->getOperand(!SVOpNum)))
return false;
// Normalize loop conditions so that it is easier to calculate new loop
// bounds.
if (IVisGT(*ExitCondition) || IVisGE(*ExitCondition)) {
ExitCondition->setPredicate(ExitCondition->getInversePredicate());
BasicBlock *T = EBR->getSuccessor(0);
EBR->setSuccessor(0, EBR->getSuccessor(1));
EBR->setSuccessor(1, T);
}
if (IVisGT(*SplitCondition) || IVisGE(*SplitCondition)) {
SplitCondition->setPredicate(SplitCondition->getInversePredicate());
BasicBlock *T = SBR->getSuccessor(0);
SBR->setSuccessor(0, SBR->getSuccessor(1));
SBR->setSuccessor(1, T);
}
//[*] Calculate new loop bounds.
Value *AEV = SplitValue;
Value *BSV = SplitValue;
bool Sign = SplitCondition->isSigned();
Instruction *PHTerm = L->getLoopPreheader()->getTerminator();
if (IVisLT(*ExitCondition)) {
if (IVisLT(*SplitCondition)) {
/* Do nothing */
}
else if (IVisLE(*SplitCondition)) {
AEV = getPlusOne(SplitValue, Sign, PHTerm, Context);
BSV = getPlusOne(SplitValue, Sign, PHTerm, Context);
} else {
assert (0 && "Unexpected split condition!");
}
}
else if (IVisLE(*ExitCondition)) {
if (IVisLT(*SplitCondition)) {
AEV = getMinusOne(SplitValue, Sign, PHTerm, Context);
}
else if (IVisLE(*SplitCondition)) {
BSV = getPlusOne(SplitValue, Sign, PHTerm, Context);
} else {
assert (0 && "Unexpected split condition!");
}
} else {
assert (0 && "Unexpected exit condition!");
}
AEV = getMin(AEV, IVExitValue, Sign, PHTerm);
BSV = getMax(BSV, IVStartValue, Sign, PHTerm);
// [*] Clone Loop
DenseMap<const Value *, Value *> ValueMap;
Loop *BLoop = CloneLoop(L, LPM, LI, ValueMap, this);
Loop *ALoop = L;
// [*] ALoop's exiting edge enters BLoop's header.
// ALoop's original exit block becomes BLoop's exit block.
PHINode *B_IndVar = cast<PHINode>(ValueMap[IndVar]);
BasicBlock *A_ExitingBlock = ExitCondition->getParent();
BranchInst *A_ExitInsn =
dyn_cast<BranchInst>(A_ExitingBlock->getTerminator());
assert (A_ExitInsn && "Unable to find suitable loop exit branch");
BasicBlock *B_ExitBlock = A_ExitInsn->getSuccessor(1);
BasicBlock *B_Header = BLoop->getHeader();
if (ALoop->contains(B_ExitBlock)) {
B_ExitBlock = A_ExitInsn->getSuccessor(0);
A_ExitInsn->setSuccessor(0, B_Header);
} else
A_ExitInsn->setSuccessor(1, B_Header);
// [*] Update ALoop's exit value using new exit value.
ExitCondition->setOperand(EVOpNum, AEV);
// [*] Update BLoop's header phi nodes. Remove incoming PHINode's from
// original loop's preheader. Add incoming PHINode values from
// ALoop's exiting block. Update BLoop header's domiantor info.
// Collect inverse map of Header PHINodes.
DenseMap<Value *, Value *> InverseMap;
for (BasicBlock::iterator BI = ALoop->getHeader()->begin(),
BE = ALoop->getHeader()->end(); BI != BE; ++BI) {
if (PHINode *PN = dyn_cast<PHINode>(BI)) {
PHINode *PNClone = cast<PHINode>(ValueMap[PN]);
InverseMap[PNClone] = PN;
} else
break;
}
BasicBlock *A_Preheader = ALoop->getLoopPreheader();
for (BasicBlock::iterator BI = B_Header->begin(), BE = B_Header->end();
BI != BE; ++BI) {
if (PHINode *PN = dyn_cast<PHINode>(BI)) {
// Remove incoming value from original preheader.
PN->removeIncomingValue(A_Preheader);
// Add incoming value from A_ExitingBlock.
if (PN == B_IndVar)
PN->addIncoming(BSV, A_ExitingBlock);
else {
PHINode *OrigPN = cast<PHINode>(InverseMap[PN]);
Value *V2 = NULL;
// If loop header is also loop exiting block then
// OrigPN is incoming value for B loop header.
if (A_ExitingBlock == ALoop->getHeader())
V2 = OrigPN;
else
V2 = OrigPN->getIncomingValueForBlock(A_ExitingBlock);
PN->addIncoming(V2, A_ExitingBlock);
}
} else
break;
}
DT->changeImmediateDominator(B_Header, A_ExitingBlock);
DF->changeImmediateDominator(B_Header, A_ExitingBlock, DT);
// [*] Update BLoop's exit block. Its new predecessor is BLoop's exit
// block. Remove incoming PHINode values from ALoop's exiting block.
// Add new incoming values from BLoop's incoming exiting value.
// Update BLoop exit block's dominator info..
BasicBlock *B_ExitingBlock = cast<BasicBlock>(ValueMap[A_ExitingBlock]);
for (BasicBlock::iterator BI = B_ExitBlock->begin(), BE = B_ExitBlock->end();
BI != BE; ++BI) {
if (PHINode *PN = dyn_cast<PHINode>(BI)) {
PN->addIncoming(ValueMap[PN->getIncomingValueForBlock(A_ExitingBlock)],
B_ExitingBlock);
PN->removeIncomingValue(A_ExitingBlock);
} else
break;
}
DT->changeImmediateDominator(B_ExitBlock, B_ExitingBlock);
DF->changeImmediateDominator(B_ExitBlock, B_ExitingBlock, DT);
//[*] Split ALoop's exit edge. This creates a new block which
// serves two purposes. First one is to hold PHINode defnitions
// to ensure that ALoop's LCSSA form. Second use it to act
// as a preheader for BLoop.
BasicBlock *A_ExitBlock = SplitEdge(A_ExitingBlock, B_Header, this);
//[*] Preserve ALoop's LCSSA form. Create new forwarding PHINodes
// in A_ExitBlock to redefine outgoing PHI definitions from ALoop.
for(BasicBlock::iterator BI = B_Header->begin(), BE = B_Header->end();
BI != BE; ++BI) {
if (PHINode *PN = dyn_cast<PHINode>(BI)) {
Value *V1 = PN->getIncomingValueForBlock(A_ExitBlock);
PHINode *newPHI = PHINode::Create(PN->getType(), PN->getName());
newPHI->addIncoming(V1, A_ExitingBlock);
A_ExitBlock->getInstList().push_front(newPHI);
PN->removeIncomingValue(A_ExitBlock);
PN->addIncoming(newPHI, A_ExitBlock);
} else
break;
}
//[*] Eliminate split condition's inactive branch from ALoop.
BasicBlock *A_SplitCondBlock = SplitCondition->getParent();
BranchInst *A_BR = cast<BranchInst>(A_SplitCondBlock->getTerminator());
BasicBlock *A_InactiveBranch = NULL;
BasicBlock *A_ActiveBranch = NULL;
A_ActiveBranch = A_BR->getSuccessor(0);
A_InactiveBranch = A_BR->getSuccessor(1);
A_BR->setUnconditionalDest(A_ActiveBranch);
removeBlocks(A_InactiveBranch, L, A_ActiveBranch);
//[*] Eliminate split condition's inactive branch in from BLoop.
BasicBlock *B_SplitCondBlock = cast<BasicBlock>(ValueMap[A_SplitCondBlock]);
BranchInst *B_BR = cast<BranchInst>(B_SplitCondBlock->getTerminator());
BasicBlock *B_InactiveBranch = NULL;
BasicBlock *B_ActiveBranch = NULL;
B_ActiveBranch = B_BR->getSuccessor(1);
B_InactiveBranch = B_BR->getSuccessor(0);
B_BR->setUnconditionalDest(B_ActiveBranch);
removeBlocks(B_InactiveBranch, BLoop, B_ActiveBranch);
BasicBlock *A_Header = ALoop->getHeader();
if (A_ExitingBlock == A_Header)
return true;
//[*] Move exit condition into split condition block to avoid
// executing dead loop iteration.
ICmpInst *B_ExitCondition = cast<ICmpInst>(ValueMap[ExitCondition]);
Instruction *B_IndVarIncrement = cast<Instruction>(ValueMap[IVIncrement]);
ICmpInst *B_SplitCondition = cast<ICmpInst>(ValueMap[SplitCondition]);
moveExitCondition(A_SplitCondBlock, A_ActiveBranch, A_ExitBlock, ExitCondition,
cast<ICmpInst>(SplitCondition), IndVar, IVIncrement,
ALoop, EVOpNum);
moveExitCondition(B_SplitCondBlock, B_ActiveBranch,
B_ExitBlock, B_ExitCondition,
B_SplitCondition, B_IndVar, B_IndVarIncrement,
BLoop, EVOpNum);
NumIndexSplit++;
return true;
}
/// MatchSelectPattern - Pattern match integer [SU]MIN, [SU]MAX, and ABS idioms,
/// returning the kind and providing the out parameter results if we
/// successfully match.
static SelectPatternFlavor
MatchSelectPattern(Value *V, Value *&LHS, Value *&RHS) {
SelectInst *SI = dyn_cast<SelectInst>(V);
if (!SI) return SPF_UNKNOWN;
ICmpInst *ICI = dyn_cast<ICmpInst>(SI->getCondition());
if (!ICI) return SPF_UNKNOWN;
ICmpInst::Predicate Pred = ICI->getPredicate();
Value *CmpLHS = ICI->getOperand(0);
Value *CmpRHS = ICI->getOperand(1);
Value *TrueVal = SI->getTrueValue();
Value *FalseVal = SI->getFalseValue();
LHS = CmpLHS;
RHS = CmpRHS;
// (icmp X, Y) ? X : Y
if (TrueVal == CmpLHS && FalseVal == CmpRHS) {
switch (Pred) {
default: return SPF_UNKNOWN; // Equality.
case ICmpInst::ICMP_UGT:
case ICmpInst::ICMP_UGE: return SPF_UMAX;
case ICmpInst::ICMP_SGT:
case ICmpInst::ICMP_SGE: return SPF_SMAX;
case ICmpInst::ICMP_ULT:
case ICmpInst::ICMP_ULE: return SPF_UMIN;
case ICmpInst::ICMP_SLT:
case ICmpInst::ICMP_SLE: return SPF_SMIN;
}
}
// (icmp X, Y) ? Y : X
if (TrueVal == CmpRHS && FalseVal == CmpLHS) {
switch (Pred) {
default: return SPF_UNKNOWN; // Equality.
case ICmpInst::ICMP_UGT:
case ICmpInst::ICMP_UGE: return SPF_UMIN;
case ICmpInst::ICMP_SGT:
case ICmpInst::ICMP_SGE: return SPF_SMIN;
case ICmpInst::ICMP_ULT:
case ICmpInst::ICMP_ULE: return SPF_UMAX;
case ICmpInst::ICMP_SLT:
case ICmpInst::ICMP_SLE: return SPF_SMAX;
}
}
if (ConstantInt *C1 = dyn_cast<ConstantInt>(CmpRHS)) {
if ((CmpLHS == TrueVal && match(FalseVal, m_Neg(m_Specific(CmpLHS)))) ||
(CmpLHS == FalseVal && match(TrueVal, m_Neg(m_Specific(CmpLHS))))) {
// ABS(X) ==> (X >s 0) ? X : -X and (X >s -1) ? X : -X
// NABS(X) ==> (X >s 0) ? -X : X and (X >s -1) ? -X : X
if (Pred == ICmpInst::ICMP_SGT && (C1->isZero() || C1->isMinusOne())) {
return (CmpLHS == TrueVal) ? SPF_ABS : SPF_NABS;
}
// ABS(X) ==> (X <s 0) ? -X : X and (X <s 1) ? -X : X
// NABS(X) ==> (X <s 0) ? X : -X and (X <s 1) ? X : -X
if (Pred == ICmpInst::ICMP_SLT && (C1->isZero() || C1->isOne())) {
return (CmpLHS == FalseVal) ? SPF_ABS : SPF_NABS;
}
}
}
// TODO: (X > 4) ? X : 5 --> (X >= 5) ? X : 5 --> MAX(X, 5)
return SPF_UNKNOWN;
}
Value* LoopTripCount::insertTripCount(Loop* L, Instruction* InsertPos)
{
// inspired from Loop::getCanonicalInductionVariable
BasicBlock *H = L->getHeader();
BasicBlock* LoopPred = L->getLoopPredecessor();
BasicBlock* startBB = NULL;//which basicblock stores start value
int OneStep = 0;// the extra add or plus step for calc
Assert(LoopPred, "Require Loop has a Pred");
DEBUG(errs()<<"loop depth:"<<L->getLoopDepth()<<"\n");
/** whats difference on use of predecessor and preheader??*/
//RET_ON_FAIL(self->getLoopLatch()&&self->getLoopPreheader());
//assert(self->getLoopLatch() && self->getLoopPreheader() && "need loop simplify form" );
ret_null_fail(L->getLoopLatch(), "need loop simplify form");
BasicBlock* TE = NULL;//True Exit
SmallVector<BasicBlock*,4> Exits;
L->getExitingBlocks(Exits);
if(Exits.size()==1) TE = Exits.front();
else{
if(std::find(Exits.begin(),Exits.end(),L->getLoopLatch())!=Exits.end()) TE = L->getLoopLatch();
else{
SmallVector<llvm::Loop::Edge,4> ExitEdges;
L->getExitEdges(ExitEdges);
//stl 用法,先把所有满足条件的元素(出口的结束符是不可到达)移动到数组的末尾,再统一删除
ExitEdges.erase(std::remove_if(ExitEdges.begin(), ExitEdges.end(),
[](llvm::Loop::Edge& I){
return isa<UnreachableInst>(I.second->getTerminator());
}), ExitEdges.end());
if(ExitEdges.size()==1) TE = const_cast<BasicBlock*>(ExitEdges.front().first);
}
}
//process true exit
ret_null_fail(TE, "need have a true exit");
Instruction* IndOrNext = NULL;
Value* END = NULL;
//终止块的终止指令：分情况讨论branchinst,switchinst;
//跳转指令br bool a1,a2;condition<-->bool
if(isa<BranchInst>(TE->getTerminator())){
const BranchInst* EBR = cast<BranchInst>(TE->getTerminator());
Assert(EBR->isConditional(), "end branch is not conditional");
ICmpInst* EC = dyn_cast<ICmpInst>(EBR->getCondition());
if(EC->getPredicate() == EC->ICMP_SGT){
Assert(!L->contains(EBR->getSuccessor(0)), *EBR<<":abnormal exit with great than");//终止块的终止指令---->跳出执行循环外的指令
OneStep += 1;
} else if(EC->getPredicate() == EC->ICMP_EQ)
Assert(!L->contains(EBR->getSuccessor(0)), *EBR<<":abnormal exit with great than");
else if(EC->getPredicate() == EC->ICMP_SLT) {
ret_null_fail(!L->contains(EBR->getSuccessor(1)), *EBR<<":abnormal exit with less than");
} else {
ret_null_fail(0, *EC<<" unknow combination of end condition");
}
IndOrNext = dyn_cast<Instruction>(castoff(EC->getOperand(0)));//去掉类型转化
END = EC->getOperand(1);
DEBUG(errs()<<"end value:"<<*EC<<"\n");
}else if(isa<SwitchInst>(TE->getTerminator())){
SwitchInst* ESW = const_cast<SwitchInst*>(cast<SwitchInst>(TE->getTerminator()));
IndOrNext = dyn_cast<Instruction>(castoff(ESW->getCondition()));
for(auto I = ESW->case_begin(),E = ESW->case_end();I!=E;++I){
if(!L->contains(I.getCaseSuccessor())){
ret_null_fail(!END,"");
assert(!END && "shouldn't have two ends");
END = I.getCaseValue();
}
}
DEBUG(errs()<<"end value:"<<*ESW<<"\n");
}else{
assert(0 && "unknow terminator type");
}
ret_null_fail(L->isLoopInvariant(END), "end value should be loop invariant");//至此得END值
Value* start = NULL;
Value* ind = NULL;
Instruction* next = NULL;
bool addfirst = false;//add before icmp ed
DISABLE(errs()<<*IndOrNext<<"\n");
if(isa<LoadInst>(IndOrNext)){
//memory depend analysis
Value* PSi = IndOrNext->getOperand(0);//point type Step.i
int SICount[2] = {0};//store in predecessor count,store in loop body count
for( auto I = PSi->use_begin(),E = PSi->use_end();I!=E;++I){
DISABLE(errs()<<**I<<"\n");
StoreInst* SI = dyn_cast<StoreInst>(*I);
if(!SI || SI->getOperand(1) != PSi) continue;
if(!start&&L->isLoopInvariant(SI->getOperand(0))) {
if(SI->getParent() != LoopPred)
if(std::find(pred_begin(LoopPred),pred_end(LoopPred),SI->getParent()) == pred_end(LoopPred)) continue;
start = SI->getOperand(0);
startBB = SI->getParent();
++SICount[0];
}
Instruction* SI0 = dyn_cast<Instruction>(SI->getOperand(0));
if(L->contains(SI) && SI0 && SI0->getOpcode() == Instruction::Add){
next = SI0;
++SICount[1];
}
}
Assert(SICount[0]==1 && SICount[1]==1, "");
ind = IndOrNext;
}else{
if(isa<PHINode>(IndOrNext)){
PHINode* PHI = cast<PHINode>(IndOrNext);
ind = IndOrNext;
if(castoff(PHI->getIncomingValue(0)) == castoff(PHI->getIncomingValue(1)) && PHI->getParent() != H)
ind = castoff(PHI->getIncomingValue(0));
addfirst = false;
}else if(IndOrNext->getOpcode() == Instruction::Add){
next = IndOrNext;
addfirst = true;
}else{
Assert(0 ,"unknow how to analysis");
}
for(auto I = H->begin();isa<PHINode>(I);++I){
PHINode* P = cast<PHINode>(I);
if(ind && P == ind){
//start = P->getIncomingValueForBlock(L->getLoopPredecessor());
start = tryFindStart(P, L, startBB);
next = dyn_cast<Instruction>(P->getIncomingValueForBlock(L->getLoopLatch()));
}else if(next && P->getIncomingValueForBlock(L->getLoopLatch()) == next){
//start = P->getIncomingValueForBlock(L->getLoopPredecessor());
start = tryFindStart(P, L, startBB);
ind = P;
}
}
}
Assert(start ,"couldn't find a start value");
//process complex loops later
//DEBUG(if(L->getLoopDepth()>1 || !L->getSubLoops().empty()) return NULL);
DEBUG(errs()<<"start value:"<<*start<<"\n");
DEBUG(errs()<<"ind value:"<<*ind<<"\n");
DEBUG(errs()<<"next value:"<<*next<<"\n");
//process non add later
unsigned next_phi_idx = 0;
ConstantInt* Step = NULL,*PrevStep = NULL;/*only used if next is phi node*/
ret_null_fail(next, "");
PHINode* next_phi = dyn_cast<PHINode>(next);
do{
if(next_phi) {
next = dyn_cast<Instruction>(next_phi->getIncomingValue(next_phi_idx));
ret_null_fail(next, "");
DEBUG(errs()<<"next phi "<<next_phi_idx<<":"<<*next<<"\n");
if(Step&&PrevStep){
Assert(Step->getSExtValue() == PrevStep->getSExtValue(),"");
}
PrevStep = Step;
}
Assert(next->getOpcode() == Instruction::Add , "why induction increment is not Add");
Assert(next->getOperand(0) == ind ,"why induction increment is not add it self");
Step = dyn_cast<ConstantInt>(next->getOperand(1));
Assert(Step,"");
}while(next_phi && ++next_phi_idx<next_phi->getNumIncomingValues());
//RET_ON_FAIL(Step->equalsInt(1));
//assert(VERBOSE(Step->equalsInt(1),Step) && "why induction increment number is not 1");
Value* RES = NULL;
//if there are no predecessor, we can insert code into start value basicblock
IRBuilder<> Builder(InsertPos);
Assert(start->getType()->isIntegerTy() && END->getType()->isIntegerTy() , " why increment is not integer type");
if(start->getType() != END->getType()){
start = Builder.CreateCast(CastInst::getCastOpcode(start, false,
END->getType(), false),start,END->getType());
}
if(Step->getType() != END->getType()){
//Because Step is a Constant, so it casted is constant
Step = dyn_cast<ConstantInt>(Builder.CreateCast(CastInst::getCastOpcode(Step, false,
END->getType(), false),Step,END->getType()));
AssertRuntime(Step);
}
if(Step->isMinusOne())
RES = Builder.CreateSub(start,END);
else//Step Couldn't be zero
RES = Builder.CreateSub(END, start);
if(addfirst) OneStep -= 1;
if(Step->isMinusOne()) OneStep*=-1;
assert(OneStep<=1 && OneStep>=-1);
RES = (OneStep==1)?Builder.CreateAdd(RES,Step):(OneStep==-1)?Builder.CreateSub(RES, Step):RES;
if(!Step->isMinusOne()&&!Step->isOne())
RES = Builder.CreateSDiv(RES, Step);
RES->setName(H->getName()+".tc");
return RES;
}
bool AlignmentFromAssumptions::extractAlignmentInfo(CallInst *I,
Value *&AAPtr, const SCEV *&AlignSCEV,
const SCEV *&OffSCEV) {
// An alignment assume must be a statement about the least-significant
// bits of the pointer being zero, possibly with some offset.
ICmpInst *ICI = dyn_cast<ICmpInst>(I->getArgOperand(0));
if (!ICI)
return false;
// This must be an expression of the form: x & m == 0.
if (ICI->getPredicate() != ICmpInst::ICMP_EQ)
return false;
// Swap things around so that the RHS is 0.
Value *CmpLHS = ICI->getOperand(0);
Value *CmpRHS = ICI->getOperand(1);
const SCEV *CmpLHSSCEV = SE->getSCEV(CmpLHS);
const SCEV *CmpRHSSCEV = SE->getSCEV(CmpRHS);
if (CmpLHSSCEV->isZero())
std::swap(CmpLHS, CmpRHS);
else if (!CmpRHSSCEV->isZero())
return false;
BinaryOperator *CmpBO = dyn_cast<BinaryOperator>(CmpLHS);
if (!CmpBO || CmpBO->getOpcode() != Instruction::And)
return false;
// Swap things around so that the right operand of the and is a constant
// (the mask); we cannot deal with variable masks.
Value *AndLHS = CmpBO->getOperand(0);
Value *AndRHS = CmpBO->getOperand(1);
const SCEV *AndLHSSCEV = SE->getSCEV(AndLHS);
const SCEV *AndRHSSCEV = SE->getSCEV(AndRHS);
if (isa<SCEVConstant>(AndLHSSCEV)) {
std::swap(AndLHS, AndRHS);
std::swap(AndLHSSCEV, AndRHSSCEV);
}
const SCEVConstant *MaskSCEV = dyn_cast<SCEVConstant>(AndRHSSCEV);
if (!MaskSCEV)
return false;
// The mask must have some trailing ones (otherwise the condition is
// trivial and tells us nothing about the alignment of the left operand).
unsigned TrailingOnes = MaskSCEV->getAPInt().countTrailingOnes();
if (!TrailingOnes)
return false;
// Cap the alignment at the maximum with which LLVM can deal (and make sure
// we don't overflow the shift).
uint64_t Alignment;
TrailingOnes = std::min(TrailingOnes,
unsigned(sizeof(unsigned) * CHAR_BIT - 1));
Alignment = std::min(1u << TrailingOnes, +Value::MaximumAlignment);
Type *Int64Ty = Type::getInt64Ty(I->getParent()->getParent()->getContext());
AlignSCEV = SE->getConstant(Int64Ty, Alignment);
// The LHS might be a ptrtoint instruction, or it might be the pointer
// with an offset.
AAPtr = nullptr;
OffSCEV = nullptr;
if (PtrToIntInst *PToI = dyn_cast<PtrToIntInst>(AndLHS)) {
AAPtr = PToI->getPointerOperand();
OffSCEV = SE->getZero(Int64Ty);
} else if (const SCEVAddExpr* AndLHSAddSCEV =
dyn_cast<SCEVAddExpr>(AndLHSSCEV)) {
// Try to find the ptrtoint; subtract it and the rest is the offset.
for (SCEVAddExpr::op_iterator J = AndLHSAddSCEV->op_begin(),
JE = AndLHSAddSCEV->op_end(); J != JE; ++J)
if (const SCEVUnknown *OpUnk = dyn_cast<SCEVUnknown>(*J))
if (PtrToIntInst *PToI = dyn_cast<PtrToIntInst>(OpUnk->getValue())) {
AAPtr = PToI->getPointerOperand();
OffSCEV = SE->getMinusSCEV(AndLHSAddSCEV, *J);
break;
}
}
if (!AAPtr)
return false;
// Sign extend the offset to 64 bits (so that it is like all of the other
// expressions).
unsigned OffSCEVBits = OffSCEV->getType()->getPrimitiveSizeInBits();
if (OffSCEVBits < 64)
OffSCEV = SE->getSignExtendExpr(OffSCEV, Int64Ty);
else if (OffSCEVBits > 64)
return false;
AAPtr = AAPtr->stripPointerCasts();
return true;
}
bool AtomicExpandLoadLinked::expandAtomicCmpXchg(AtomicCmpXchgInst *CI) {
AtomicOrdering SuccessOrder = CI->getSuccessOrdering();
AtomicOrdering FailureOrder = CI->getFailureOrdering();
Value *Addr = CI->getPointerOperand();
BasicBlock *BB = CI->getParent();
Function *F = BB->getParent();
LLVMContext &Ctx = F->getContext();
// Given: cmpxchg some_op iN* %addr, iN %desired, iN %new success_ord fail_ord
//
// The full expansion we produce is:
// [...]
// fence?
// cmpxchg.start:
// %loaded = @load.linked(%addr)
// %should_store = icmp eq %loaded, %desired
// br i1 %should_store, label %cmpxchg.trystore,
// label %cmpxchg.end/%cmpxchg.barrier
// cmpxchg.trystore:
// %stored = @store_conditional(%new, %addr)
// %try_again = icmp i32 ne %stored, 0
// br i1 %try_again, label %loop, label %cmpxchg.end
// cmpxchg.barrier:
// fence?
// br label %cmpxchg.end
// cmpxchg.end:
// [...]
BasicBlock *ExitBB = BB->splitBasicBlock(CI, "cmpxchg.end");
auto BarrierBB = BasicBlock::Create(Ctx, "cmpxchg.barrier", F, ExitBB);
auto TryStoreBB = BasicBlock::Create(Ctx, "cmpxchg.trystore", F, BarrierBB);
auto LoopBB = BasicBlock::Create(Ctx, "cmpxchg.start", F, TryStoreBB);
// This grabs the DebugLoc from CI
IRBuilder<> Builder(CI);
// The split call above "helpfully" added a branch at the end of BB (to the
// wrong place), but we might want a fence too. It's easiest to just remove
// the branch entirely.
std::prev(BB->end())->eraseFromParent();
Builder.SetInsertPoint(BB);
AtomicOrdering MemOpOrder = insertLeadingFence(Builder, SuccessOrder);
Builder.CreateBr(LoopBB);
// Start the main loop block now that we've taken care of the preliminaries.
Builder.SetInsertPoint(LoopBB);
Value *Loaded = TLI->emitLoadLinked(Builder, Addr, MemOpOrder);
Value *ShouldStore =
Builder.CreateICmpEQ(Loaded, CI->getCompareOperand(), "should_store");
// If the the cmpxchg doesn't actually need any ordering when it fails, we can
// jump straight past that fence instruction (if it exists).
BasicBlock *FailureBB = FailureOrder == Monotonic ? ExitBB : BarrierBB;
Builder.CreateCondBr(ShouldStore, TryStoreBB, FailureBB);
Builder.SetInsertPoint(TryStoreBB);
Value *StoreSuccess = TLI->emitStoreConditional(
Builder, CI->getNewValOperand(), Addr, MemOpOrder);
Value *TryAgain = Builder.CreateICmpNE(
StoreSuccess, ConstantInt::get(Type::getInt32Ty(Ctx), 0), "success");
Builder.CreateCondBr(TryAgain, LoopBB, BarrierBB);
// Make sure later instructions don't get reordered with a fence if necessary.
Builder.SetInsertPoint(BarrierBB);
insertTrailingFence(Builder, SuccessOrder);
Builder.CreateBr(ExitBB);
// Finally, we have control-flow based knowledge of whether the cmpxchg
// succeeded or not. We expose this to later passes by converting any
// subsequent "icmp eq/ne %loaded, %oldval" into a use of an appropriate PHI.
// Setup the builder so we can create any PHIs we need.
Builder.SetInsertPoint(FailureBB, FailureBB->begin());
BasicBlock *SuccessBB = FailureOrder == Monotonic ? BarrierBB : TryStoreBB;
PHINode *Success = 0, *Failure = 0;
// Look for any users of the cmpxchg that are just comparing the loaded value
// against the desired one, and replace them with the CFG-derived version.
for (auto User : CI->users()) {
ICmpInst *ICmp = dyn_cast<ICmpInst>(User);
if (!ICmp)
continue;
// Because we know ICmp uses CI, we only need one operand to be the old
// value.
if (ICmp->getOperand(0) != CI->getCompareOperand() &&
ICmp->getOperand(1) != CI->getCompareOperand())
continue;
if (ICmp->getPredicate() == CmpInst::ICMP_EQ) {
if (!Success) {
Success = Builder.CreatePHI(Type::getInt1Ty(Ctx), 2);
Success->addIncoming(ConstantInt::getTrue(Ctx), SuccessBB);
Success->addIncoming(ConstantInt::getFalse(Ctx), LoopBB);
}
ICmp->replaceAllUsesWith(Success);
} else if (ICmp->getPredicate() == CmpInst::ICMP_NE) {
if (!Failure) {
Failure = Builder.CreatePHI(Type::getInt1Ty(Ctx), 2);
Failure->addIncoming(ConstantInt::getFalse(Ctx), SuccessBB);
Failure->addIncoming(ConstantInt::getTrue(Ctx), LoopBB);
}
ICmp->replaceAllUsesWith(Failure);
}
}
CI->replaceAllUsesWith(Loaded);
CI->eraseFromParent();
return true;
}
/// getEdgeValue - This method attempts to infer more complex
bool LazyValueInfoCache::getEdgeValue(Value *Val, BasicBlock *BBFrom,
BasicBlock *BBTo, LVILatticeVal &Result) {
// If already a constant, there is nothing to compute.
if (Constant *VC = dyn_cast<Constant>(Val)) {
Result = LVILatticeVal::get(VC);
return true;
}
// TODO: Handle more complex conditionals. If (v == 0 || v2 < 1) is false, we
// know that v != 0.
if (BranchInst *BI = dyn_cast<BranchInst>(BBFrom->getTerminator())) {
// If this is a conditional branch and only one successor goes to BBTo, then
// we maybe able to infer something from the condition.
if (BI->isConditional() &&
BI->getSuccessor(0) != BI->getSuccessor(1)) {
bool isTrueDest = BI->getSuccessor(0) == BBTo;
assert(BI->getSuccessor(!isTrueDest) == BBTo &&
"BBTo isn't a successor of BBFrom");
// If V is the condition of the branch itself, then we know exactly what
// it is.
if (BI->getCondition() == Val) {
Result = LVILatticeVal::get(ConstantInt::get(
Type::getInt1Ty(Val->getContext()), isTrueDest));
return true;
}
// If the condition of the branch is an equality comparison, we may be
// able to infer the value.
ICmpInst *ICI = dyn_cast<ICmpInst>(BI->getCondition());
if (ICI && ICI->getOperand(0) == Val &&
isa<Constant>(ICI->getOperand(1))) {
if (ICI->isEquality()) {
// We know that V has the RHS constant if this is a true SETEQ or
// false SETNE.
if (isTrueDest == (ICI->getPredicate() == ICmpInst::ICMP_EQ))
Result = LVILatticeVal::get(cast<Constant>(ICI->getOperand(1)));
else
Result = LVILatticeVal::getNot(cast<Constant>(ICI->getOperand(1)));
return true;
}
if (ConstantInt *CI = dyn_cast<ConstantInt>(ICI->getOperand(1))) {
// Calculate the range of values that would satisfy the comparison.
ConstantRange CmpRange(CI->getValue(), CI->getValue()+1);
ConstantRange TrueValues =
ConstantRange::makeICmpRegion(ICI->getPredicate(), CmpRange);
// If we're interested in the false dest, invert the condition.
if (!isTrueDest) TrueValues = TrueValues.inverse();
// Figure out the possible values of the query BEFORE this branch.
if (!hasBlockValue(Val, BBFrom)) {
BlockValueStack.push(std::make_pair(BBFrom, Val));
return false;
}
LVILatticeVal InBlock = getBlockValue(Val, BBFrom);
if (!InBlock.isConstantRange()) {
Result = LVILatticeVal::getRange(TrueValues);
return true;
}
// Find all potential values that satisfy both the input and output
// conditions.
ConstantRange PossibleValues =
TrueValues.intersectWith(InBlock.getConstantRange());
Result = LVILatticeVal::getRange(PossibleValues);
return true;
}
}
}
}
// If the edge was formed by a switch on the value, then we may know exactly
// what it is.
if (SwitchInst *SI = dyn_cast<SwitchInst>(BBFrom->getTerminator())) {
if (SI->getCondition() == Val) {
// We don't know anything in the default case.
if (SI->getDefaultDest() == BBTo) {
Result.markOverdefined();
return true;
}
// We only know something if there is exactly one value that goes from
// BBFrom to BBTo.
unsigned NumEdges = 0;
ConstantInt *EdgeVal = 0;
for (unsigned i = 1, e = SI->getNumSuccessors(); i != e; ++i) {
if (SI->getSuccessor(i) != BBTo) continue;
if (NumEdges++) break;
EdgeVal = SI->getCaseValue(i);
}
assert(EdgeVal && "Missing successor?");
if (NumEdges == 1) {
Result = LVILatticeVal::get(EdgeVal);
return true;
}
}
}
// Otherwise see if the value is known in the block.
if (hasBlockValue(Val, BBFrom)) {
Result = getBlockValue(Val, BBFrom);
return true;
}
BlockValueStack.push(std::make_pair(BBFrom, Val));
return false;
}
bool BranchProbabilityInfo::calcZeroHeuristics(const BasicBlock *BB,
const TargetLibraryInfo *TLI) {
const BranchInst *BI = dyn_cast<BranchInst>(BB->getTerminator());
if (!BI || !BI->isConditional())
return false;
Value *Cond = BI->getCondition();
ICmpInst *CI = dyn_cast<ICmpInst>(Cond);
if (!CI)
return false;
Value *RHS = CI->getOperand(1);
ConstantInt *CV = dyn_cast<ConstantInt>(RHS);
if (!CV)
return false;
// If the LHS is the result of AND'ing a value with a single bit bitmask,
// we don't have information about probabilities.
if (Instruction *LHS = dyn_cast<Instruction>(CI->getOperand(0)))
if (LHS->getOpcode() == Instruction::And)
if (ConstantInt *AndRHS = dyn_cast<ConstantInt>(LHS->getOperand(1)))
if (AndRHS->getValue().isPowerOf2())
return false;
// Check if the LHS is the return value of a library function
LibFunc Func = NumLibFuncs;
if (TLI)
if (CallInst *Call = dyn_cast<CallInst>(CI->getOperand(0)))
if (Function *CalledFn = Call->getCalledFunction())
TLI->getLibFunc(*CalledFn, Func);
bool isProb;
if (Func == LibFunc_strcasecmp ||
Func == LibFunc_strcmp ||
Func == LibFunc_strncasecmp ||
Func == LibFunc_strncmp ||
Func == LibFunc_memcmp) {
// strcmp and similar functions return zero, negative, or positive, if the
// first string is equal, less, or greater than the second. We consider it
// likely that the strings are not equal, so a comparison with zero is
// probably false, but also a comparison with any other number is also
// probably false given that what exactly is returned for nonzero values is
// not specified. Any kind of comparison other than equality we know
// nothing about.
switch (CI->getPredicate()) {
case CmpInst::ICMP_EQ:
isProb = false;
break;
case CmpInst::ICMP_NE:
isProb = true;
break;
default:
return false;
}
} else if (CV->isZero()) {
switch (CI->getPredicate()) {
case CmpInst::ICMP_EQ:
// X == 0 -> Unlikely
isProb = false;
break;
case CmpInst::ICMP_NE:
// X != 0 -> Likely
isProb = true;
break;
case CmpInst::ICMP_SLT:
// X < 0 -> Unlikely
isProb = false;
break;
case CmpInst::ICMP_SGT:
// X > 0 -> Likely
isProb = true;
break;
default:
return false;
}
} else if (CV->isOne() && CI->getPredicate() == CmpInst::ICMP_SLT) {
// InstCombine canonicalizes X <= 0 into X < 1.
// X <= 0 -> Unlikely
isProb = false;
} else if (CV->isMinusOne()) {
switch (CI->getPredicate()) {
case CmpInst::ICMP_EQ:
// X == -1 -> Unlikely
isProb = false;
break;
case CmpInst::ICMP_NE:
// X != -1 -> Likely
isProb = true;
break;
case CmpInst::ICMP_SGT:
// InstCombine canonicalizes X >= 0 into X > -1.
// X >= 0 -> Likely
isProb = true;
break;
default:
return false;
}
} else {
return false;
}
unsigned TakenIdx = 0, NonTakenIdx = 1;
if (!isProb)
std::swap(TakenIdx, NonTakenIdx);
BranchProbability TakenProb(ZH_TAKEN_WEIGHT,
ZH_TAKEN_WEIGHT + ZH_NONTAKEN_WEIGHT);
setEdgeProbability(BB, TakenIdx, TakenProb);
setEdgeProbability(BB, NonTakenIdx, TakenProb.getCompl());
return true;
}
/// \brief Compute the value of Val on the edge BBFrom -> BBTo. Returns false if
/// Val is not constrained on the edge.
static bool getEdgeValueLocal(Value *Val, BasicBlock *BBFrom,
BasicBlock *BBTo, LVILatticeVal &Result) {
// TODO: Handle more complex conditionals. If (v == 0 || v2 < 1) is false, we
// know that v != 0.
if (BranchInst *BI = dyn_cast<BranchInst>(BBFrom->getTerminator())) {
// If this is a conditional branch and only one successor goes to BBTo, then
// we maybe able to infer something from the condition.
if (BI->isConditional() &&
BI->getSuccessor(0) != BI->getSuccessor(1)) {
bool isTrueDest = BI->getSuccessor(0) == BBTo;
assert(BI->getSuccessor(!isTrueDest) == BBTo &&
"BBTo isn't a successor of BBFrom");
// If V is the condition of the branch itself, then we know exactly what
// it is.
if (BI->getCondition() == Val) {
Result = LVILatticeVal::get(ConstantInt::get(
Type::getInt1Ty(Val->getContext()), isTrueDest));
return true;
}
// If the condition of the branch is an equality comparison, we may be
// able to infer the value.
ICmpInst *ICI = dyn_cast<ICmpInst>(BI->getCondition());
if (ICI && isa<Constant>(ICI->getOperand(1))) {
if (ICI->isEquality() && ICI->getOperand(0) == Val) {
// We know that V has the RHS constant if this is a true SETEQ or
// false SETNE.
if (isTrueDest == (ICI->getPredicate() == ICmpInst::ICMP_EQ))
Result = LVILatticeVal::get(cast<Constant>(ICI->getOperand(1)));
else
Result = LVILatticeVal::getNot(cast<Constant>(ICI->getOperand(1)));
return true;
}
// Recognize the range checking idiom that InstCombine produces.
// (X-C1) u< C2 --> [C1, C1+C2)
ConstantInt *NegOffset = 0;
if (ICI->getPredicate() == ICmpInst::ICMP_ULT)
match(ICI->getOperand(0), m_Add(m_Specific(Val),
m_ConstantInt(NegOffset)));
ConstantInt *CI = dyn_cast<ConstantInt>(ICI->getOperand(1));
if (CI && (ICI->getOperand(0) == Val || NegOffset)) {
// Calculate the range of values that would satisfy the comparison.
ConstantRange CmpRange(CI->getValue());
ConstantRange TrueValues =
ConstantRange::makeICmpRegion(ICI->getPredicate(), CmpRange);
if (NegOffset) // Apply the offset from above.
TrueValues = TrueValues.subtract(NegOffset->getValue());
// If we're interested in the false dest, invert the condition.
if (!isTrueDest) TrueValues = TrueValues.inverse();
Result = LVILatticeVal::getRange(TrueValues);
return true;
}
}
}
}
// If the edge was formed by a switch on the value, then we may know exactly
// what it is.
if (SwitchInst *SI = dyn_cast<SwitchInst>(BBFrom->getTerminator())) {
if (SI->getCondition() != Val)
return false;
bool DefaultCase = SI->getDefaultDest() == BBTo;
unsigned BitWidth = Val->getType()->getIntegerBitWidth();
ConstantRange EdgesVals(BitWidth, DefaultCase/*isFullSet*/);
for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
i != e; ++i) {
ConstantRange EdgeVal(i.getCaseValue()->getValue());
if (DefaultCase) {
// It is possible that the default destination is the destination of
// some cases. There is no need to perform difference for those cases.
if (i.getCaseSuccessor() != BBTo)
EdgesVals = EdgesVals.difference(EdgeVal);
} else if (i.getCaseSuccessor() == BBTo)
EdgesVals = EdgesVals.unionWith(EdgeVal);
}
Result = LVILatticeVal::getRange(EdgesVals);
return true;
}
return false;
}
bool RippleElimination::runOnFunction(Function &F) {
std::vector<Instruction*> **RippleLists;
unsigned ListID = 0;
unsigned NumofLists = 0;
RippleLists = new std::vector<Instruction*>*;
RippleLists[ListID] = new std::vector<Instruction*>;
RippleLists[ListID]->clear();
//errs() << "RippleLists[0] size:" << RippleLists[0]->size() << "\n";
for (Function::iterator BB = F.begin(), BE = F.end(); BB != BE; ++BB) {
BasicBlock *BC = dyn_cast<BasicBlock>(BB); //==> how to get an identifier of a BB?
for (BasicBlock::iterator IB = BB->begin(), IE = BB->end(); IB != IE; ++IB) {
// The dyn_cast<> operator is a "checking cast" operation. It checks to
// see if the operand is of the specified type, and if so, returns a
// pointer to it (this operator does not work with references). If the
// operand is not of the correct type, a null pointer is returned.
Instruction *IT = dyn_cast<Instruction>(IB);
Instruction *BackIT;
ICmpInst * cmpinst = dyn_cast<ICmpInst>(IT);
if(cmpinst!=NULL) { // If not a NULL pointer
errs()<< "!!!! In CMP Oh yeah!!!!"<<"\n";
switch(cmpinst->getPredicate()){
case ICmpInst::ICMP_SLE:
{
errs()<<"QQQQ SLE\n";
}
break;
case ICmpInst::ICMP_SLT:
{
Value *op0 = IT->getOperand(0);
Instruction * op0_inst = dyn_cast<Instruction> (op0);
Value *op0_op1 = op0_inst->getOperand(1);
ConstantInt *op0_op1_const = dyn_cast<ConstantInt>(op0_op1);
APInt b = op0_op1_const->getValue();
ConstantInt *op1 = dyn_cast<ConstantInt>(IT->getOperand(1));
APInt a = op1->getValue();
errs() << "CMP Value0 ::" << b.getLimitedValue() << "\n";
errs() << "CMP Value1 ::" << a.getLimitedValue() << "\n";
int aa = a.getLimitedValue();
int bb = b.getLimitedValue();
// errs()<<a.getBitWidth()<<"\n";
errs()<<"The flag adder should be "<<(int)ceil(log(aa)/log(2))<<" bit adder\n";
errs()<<"QQQQ SLT\n";
}
break;
default:
break;
}
}
// Get Binary Operator Instruction::Add
if(dyn_cast<BinaryOperator>(IT)) { // If not a NULL pointer
BinaryOperator* Inst = dyn_cast<BinaryOperator>(IT);
// errs()<<"Zhuangh put: "<<Inst->getOpcode()<<"\n";
if (Inst->getOpcode() == Instruction::Add) {
Value *Op0 = IT->getOperand(0);
Value *Op1 = IT->getOperand(1);
//errs() << "Add Value0 ::" << *Op0 << "\n";
//errs() << "Add Value1 ::" << *Op1 << "\n";
if(RippleLists[0]->empty()) {
RippleLists[0]->push_back(IT);
}
else {
BackIT = dyn_cast<Instruction>(RippleLists[0]->back());
//errs() << "BackIT :" << *BackIT << "\n";
if(BackIT->getName().equals(Op0->getName()) || BackIT->getName().equals(Op1->getName())) {
RippleLists[0]->push_back(IT);
}
}
}
}
} // End Instruction iterator
if(RippleLists[0]->size() < 2) RippleLists[0]->clear(); // No ripple detected
for(std::vector<Instruction*>::iterator RplIT = RippleLists[0]->begin(); RplIT != RippleLists[0]->end(); ++RplIT){
Instruction *Inst = dyn_cast<Instruction>(*RplIT);
errs() << "Ripple Operations:" << *Inst << "\n";
//errs() << "Name:" << Inst->getName().str() << " ";
//errs() << "Op0:" << Inst->getOperand(0)->getName().str() << " ";
//errs() << "Op1:" << Inst->getOperand(1)->getName().str() << "\n";
//It->eraseFromParent();
}
//errs() << "RippleLists[0] size:" << RippleLists[0]->size() << "\n";
errs() << "\n";
RippleLists[0]->clear();
} // End basic block iterator
return true;
}
// addEdgesFor
// Creates a node for I and inserts edges from the created node to the
// appropriate node of other values.
void IneqGraph::addEdgesFor(Instruction *I) {
if (I->getType()->isPointerTy())
return;
Range RI = RA->getRange(I);
if (!RI.getLower().isMinSignedValue())
addMayEdge(AlfaConst, I, -RI.getLower().getSExtValue());
if (!RI.getUpper().isMaxSignedValue())
addMayEdge(I, AlfaConst, RI.getUpper().getSExtValue());
// TODO: Handle multiplication, remainder and division instructions.
switch (I->getOpcode()) {
case Instruction::SExt:
case Instruction::ZExt:
case Instruction::Trunc:
case Instruction::BitCast:
addMayEdge(I, I->getOperand(0), 0);
addMayEdge(I->getOperand(0), I, 0);
break;
case Instruction::Add:
// a = b + c
// ==> a <= b + sup(c)
// ==> a <= c + sup(b)
// ==> b <= a - inf(c)
// ==> c <= a - inf(b)
{
Value *A = I->getOperand(0);
Value *B = I->getOperand(1);
Range AR = RA->getRange(A);
Range BR = RA->getRange(B);
if (!isa<ConstantInt>(B) && !AR.getUpper().isMaxSignedValue())
addMayEdge(I, B, AR.getUpper());
if (!isa<ConstantInt>(A) && !BR.getUpper().isMaxSignedValue())
addMayEdge(I, A, BR.getUpper());
if (!isa<ConstantInt>(A) && !BR.getLower().isMinSignedValue())
addMayEdge(A, I, -BR.getUpper());
if (!isa<ConstantInt>(B) && !AR.getLower().isMinSignedValue())
addMayEdge(B, I, -AR.getUpper());
break;
}
case Instruction::Sub:
// a = b - c
// ==> a <= b - inf(c)
{
Value *A = I->getOperand(0);
Value *B = I->getOperand(1);
Range AR = RA->getRange(A);
Range BR = RA->getRange(B);
if (!isa<ConstantInt>(A) && !BR.getLower().isMinSignedValue())
addMayEdge(I, A, -BR.getLower());
break;
}
case Instruction::Br:
// if (a > b) {
// a1 = sigma(a)
// b1 = sigma(b)
{
BranchInst *BI = cast<BranchInst>(I);
ICmpInst *Cmp = dyn_cast<ICmpInst>(I->getOperand(0));
if (!Cmp)
break;
Value *L = Cmp->getOperand(0);
DEBUG(dbgs() << "IneqGraph: L: " << *L << "\n");
Value *R = Cmp->getOperand(1);
DEBUG(dbgs() << "IneqGraph: R: " << *R << "\n");
Value *LSigma = VS->findSigma(L, BI->getSuccessor(0), BI->getParent());
DEBUG(dbgs() << "IneqGraph: LSigma: " << *LSigma << "\n");
Value *RSigma = VS->findSigma(R, BI->getSuccessor(0), BI->getParent());
DEBUG(dbgs() << "IneqGraph: RSigma: " << *RSigma << "\n");
Value *LSExtSigma = VS->findSExtSigma(L, BI->getSuccessor(0), BI->getParent());
DEBUG(dbgs() << "IneqGraph: LSExtSigma: " << *LSExtSigma << "\n");
Value *RSExtSigma = VS->findSExtSigma(R, BI->getSuccessor(0), BI->getParent());
DEBUG(dbgs() << "IneqGraph: RSExtSigma: " << *RSExtSigma << "\n");
switch (Cmp->getPredicate()) {
case ICmpInst::ICMP_SLT:
DEBUG(dbgs() << "IneqGraph: SLT:\n");
if (!isa<ConstantInt>(R) && LSigma) {
if (RSigma)
addMayEdge(LSigma, RSigma, -1);
if (RSExtSigma)
addMayEdge(LSigma, RSExtSigma, -1);
}
if (!isa<ConstantInt>(R) && LSExtSigma && LSExtSigma != LSigma) {
if (RSigma)
addMayEdge(LSExtSigma, RSigma, -1);
if (RSExtSigma)
addMayEdge(LSExtSigma, RSExtSigma, -1);
}
break;
case ICmpInst::ICMP_SLE:
DEBUG(dbgs() << "IneqGraph: SLE:\n");
if (!isa<ConstantInt>(R) && LSigma && RSigma)
addMayEdge(LSigma, RSigma, 0);
if (!isa<ConstantInt>(R) && (LSExtSigma != LSigma || RSExtSigma != RSigma))
addMayEdge(LSExtSigma, RSExtSigma, 0);
break;
default:
break;
}
break;
}
case Instruction::PHI:
{
PHINode *Phi = cast<PHINode>(I);
for (unsigned Idx = 0; Idx < Phi->getNumIncomingValues(); ++Idx) {
addMustEdge(Phi, Phi->getIncomingValue(Idx), 0);
}
break;
}
case Instruction::Call:
{
CallInst *CI = cast<CallInst>(I);
if (Function *F = CI->getCalledFunction()) {
unsigned Idx = 0;
for (Function::arg_iterator AI = F->arg_begin(), AE = F->arg_end();
AI != AE; ++AI, ++Idx) {
addMustEdge(&(*AI), CI->getArgOperand(Idx), 0);
}
}
break;
}
}
}
