FoldCmpLoadFromIndexedGlobal(GetElementPtrInst *GEP, GlobalVariable *GV,
CmpInst &ICI, ConstantInt *AndCst) {
// We need TD information to know the pointer size unless this is inbounds.
- if (!GEP->isInBounds() && TD == 0) return 0;
+ if (!GEP->isInBounds() && TD == 0)
+ return 0;
Constant *Init = GV->getInitializer();
if (!isa<ConstantArray>(Init) && !isa<ConstantDataArray>(Init))
}
}
+
+
// Okay, we know we have a single variable index, which must be a
// pointer/array/vector index. If there is no offset, life is simple, return
// the index.
- unsigned IntPtrWidth = TD.getPointerSizeInBits();
+ Type *IntPtrTy = TD.getIntPtrType(GEP->getOperand(0)->getType());
+ unsigned IntPtrWidth = IntPtrTy->getIntegerBitWidth();
if (Offset == 0) {
// Cast to intptrty in case a truncation occurs. If an extension is needed,
// we don't need to bother extending: the extension won't affect where the
// computation crosses zero.
if (VariableIdx->getType()->getPrimitiveSizeInBits() > IntPtrWidth) {
- Type *IntPtrTy = TD.getIntPtrType(VariableIdx->getContext());
VariableIdx = IC.Builder->CreateTrunc(VariableIdx, IntPtrTy);
}
return VariableIdx;
return 0;
// Okay, we can do this evaluation. Start by converting the index to intptr.
- Type *IntPtrTy = TD.getIntPtrType(VariableIdx->getContext());
if (VariableIdx->getType() != IntPtrTy)
VariableIdx = IC.Builder->CreateIntCast(VariableIdx, IntPtrTy,
true /*Signed*/);
}
/// FoldICmpAddOpCst - Fold "icmp pred (X+CI), X".
-Instruction *InstCombiner::FoldICmpAddOpCst(ICmpInst &ICI,
+Instruction *InstCombiner::FoldICmpAddOpCst(Instruction &ICI,
Value *X, ConstantInt *CI,
- ICmpInst::Predicate Pred,
- Value *TheAdd) {
+ ICmpInst::Predicate Pred) {
// If we have X+0, exit early (simplifying logic below) and let it get folded
// elsewhere. icmp X+0, X -> icmp X, X
if (CI->isZero()) {
}
+/// \brief Check if the order of \p Op0 and \p Op1 as operand in an ICmpInst
+/// should be swapped.
+/// The descision is based on how many times these two operands are reused
+/// as subtract operands and their positions in those instructions.
+/// The rational is that several architectures use the same instruction for
+/// both subtract and cmp, thus it is better if the order of those operands
+/// match.
+/// \return true if Op0 and Op1 should be swapped.
+static bool swapMayExposeCSEOpportunities(const Value * Op0,
+ const Value * Op1) {
+ // Filter out pointer value as those cannot appears directly in subtract.
+ // FIXME: we may want to go through inttoptrs or bitcasts.
+ if (Op0->getType()->isPointerTy())
+ return false;
+ // Count every uses of both Op0 and Op1 in a subtract.
+ // Each time Op0 is the first operand, count -1: swapping is bad, the
+ // subtract has already the same layout as the compare.
+ // Each time Op0 is the second operand, count +1: swapping is good, the
+ // subtract has a diffrent layout as the compare.
+ // At the end, if the benefit is greater than 0, Op0 should come second to
+ // expose more CSE opportunities.
+ int GlobalSwapBenefits = 0;
+ for (Value::const_use_iterator UI = Op0->use_begin(), UIEnd = Op0->use_end(); UI != UIEnd; ++UI) {
+ const BinaryOperator *BinOp = dyn_cast<BinaryOperator>(*UI);
+ if (!BinOp || BinOp->getOpcode() != Instruction::Sub)
+ continue;
+ // If Op0 is the first argument, this is not beneficial to swap the
+ // arguments.
+ int LocalSwapBenefits = -1;
+ unsigned Op1Idx = 1;
+ if (BinOp->getOperand(Op1Idx) == Op0) {
+ Op1Idx = 0;
+ LocalSwapBenefits = 1;
+ }
+ if (BinOp->getOperand(Op1Idx) != Op1)
+ continue;
+ GlobalSwapBenefits += LocalSwapBenefits;
+ }
+ return GlobalSwapBenefits > 0;
+}
+
Instruction *InstCombiner::visitICmpInst(ICmpInst &I) {
bool Changed = false;
Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
+ unsigned Op0Cplxity = getComplexity(Op0);
+ unsigned Op1Cplxity = getComplexity(Op1);
/// Orders the operands of the compare so that they are listed from most
/// complex to least complex. This puts constants before unary operators,
/// before binary operators.
- if (getComplexity(Op0) < getComplexity(Op1)) {
+ if (Op0Cplxity < Op1Cplxity ||
+ (Op0Cplxity == Op1Cplxity &&
+ swapMayExposeCSEOpportunities(Op0, Op1))) {
I.swapOperands();
std::swap(Op0, Op1);
Changed = true;
case Instruction::IntToPtr:
// icmp pred inttoptr(X), null -> icmp pred X, 0
if (RHSC->isNullValue() && TD &&
- TD->getIntPtrType(RHSC->getContext()) ==
+ TD->getIntPtrType(RHSC->getType()) ==
LHSI->getOperand(0)->getType())
return new ICmpInst(I.getPredicate(), LHSI->getOperand(0),
Constant::getNullValue(LHSI->getOperand(0)->getType()));
Value *X; ConstantInt *Cst;
// icmp X+Cst, X
if (match(Op0, m_Add(m_Value(X), m_ConstantInt(Cst))) && Op1 == X)
- return FoldICmpAddOpCst(I, X, Cst, I.getPredicate(), Op0);
+ return FoldICmpAddOpCst(I, X, Cst, I.getPredicate());
// icmp X, X+Cst
if (match(Op1, m_Add(m_Value(X), m_ConstantInt(Cst))) && Op0 == X)
- return FoldICmpAddOpCst(I, X, Cst, I.getSwappedPredicate(), Op1);
+ return FoldICmpAddOpCst(I, X, Cst, I.getSwappedPredicate());
}
return Changed ? &I : 0;
}
-
-
-
-
-
/// FoldFCmp_IntToFP_Cst - Fold fcmp ([us]itofp x, cst) if possible.
///
Instruction *InstCombiner::FoldFCmp_IntToFP_Cst(FCmpInst &I,