SDValue visitSRA(SDNode *N);
SDValue visitSRL(SDNode *N);
SDValue visitRotate(SDNode *N);
+ SDValue visitBSWAP(SDNode *N);
SDValue visitCTLZ(SDNode *N);
SDValue visitCTLZ_ZERO_UNDEF(SDNode *N);
SDValue visitCTTZ(SDNode *N);
SDValue visitZERO_EXTEND(SDNode *N);
SDValue visitANY_EXTEND(SDNode *N);
SDValue visitSIGN_EXTEND_INREG(SDNode *N);
+ SDValue visitSIGN_EXTEND_VECTOR_INREG(SDNode *N);
SDValue visitTRUNCATE(SDNode *N);
SDValue visitBITCAST(SDNode *N);
SDValue visitBUILD_PAIR(SDNode *N);
unsigned SequenceNum;
};
+ /// This is a helper function for MergeStoresOfConstantsOrVecElts. Returns a
+ /// constant build_vector of the stored constant values in Stores.
+ SDValue getMergedConstantVectorStore(SelectionDAG &DAG,
+ SDLoc SL,
+ ArrayRef<MemOpLink> Stores,
+ EVT Ty) const;
+
/// This is a helper function for MergeConsecutiveStores. When the source
/// elements of the consecutive stores are all constants or all extracted
/// vector elements, try to merge them into one larger store.
EVT MemVT, unsigned NumElem,
bool IsConstantSrc, bool UseVector);
+ /// This is a helper function for MergeConsecutiveStores.
+ /// Stores that may be merged are placed in StoreNodes.
+ /// Loads that may alias with those stores are placed in AliasLoadNodes.
+ void getStoreMergeAndAliasCandidates(
+ StoreSDNode* St, SmallVectorImpl<MemOpLink> &StoreNodes,
+ SmallVectorImpl<LSBaseSDNode*> &AliasLoadNodes);
+
/// Merge consecutive store operations into a wide store.
/// This optimization uses wide integers or vectors when possible.
/// \return True if some memory operations were changed.
// fold (fneg (fsub 0, B)) -> B
if (ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(Op.getOperand(0)))
- if (N0CFP->getValueAPF().isZero())
+ if (N0CFP->isZero())
return Op.getOperand(1);
// fold (fneg (fsub A, B)) -> (fsub B, A)
case ISD::SRL: return visitSRL(N);
case ISD::ROTR:
case ISD::ROTL: return visitRotate(N);
+ case ISD::BSWAP: return visitBSWAP(N);
case ISD::CTLZ: return visitCTLZ(N);
case ISD::CTLZ_ZERO_UNDEF: return visitCTLZ_ZERO_UNDEF(N);
case ISD::CTTZ: return visitCTTZ(N);
case ISD::ZERO_EXTEND: return visitZERO_EXTEND(N);
case ISD::ANY_EXTEND: return visitANY_EXTEND(N);
case ISD::SIGN_EXTEND_INREG: return visitSIGN_EXTEND_INREG(N);
+ case ISD::SIGN_EXTEND_VECTOR_INREG: return visitSIGN_EXTEND_VECTOR_INREG(N);
case ISD::TRUNCATE: return visitTRUNCATE(N);
case ISD::BITCAST: return visitBITCAST(N);
case ISD::BUILD_PAIR: return visitBUILD_PAIR(N);
if (isa<ConstantSDNode>(N0) || !isa<ConstantSDNode>(N1)) {
SDValue Ops[] = {N1, N0};
SDNode *CSENode;
- if (const BinaryWithFlagsSDNode *BinNode =
- dyn_cast<BinaryWithFlagsSDNode>(N)) {
+ if (const auto *BinNode = dyn_cast<BinaryWithFlagsSDNode>(N)) {
CSENode = DAG.getNodeIfExists(N->getOpcode(), N->getVTList(), Ops,
- BinNode->Flags.hasNoUnsignedWrap(),
- BinNode->Flags.hasNoSignedWrap(),
- BinNode->Flags.hasExact());
+ &BinNode->Flags);
} else {
CSENode = DAG.getNodeIfExists(N->getOpcode(), N->getVTList(), Ops);
}
return Const != nullptr && Const->isNullValue();
}
+static bool isNullFPConstant(SDValue V) {
+ ConstantFPSDNode *Const = dyn_cast<ConstantFPSDNode>(V);
+ return Const != nullptr && Const->isZero() && !Const->isNegative();
+}
+
static bool isAllOnesConstant(SDValue V) {
ConstantSDNode *Const = dyn_cast<ConstantSDNode>(V);
return Const != nullptr && Const->isAllOnesValue();
return Const != nullptr && Const->isOne();
}
+/// If \p N is a ContantSDNode with isOpaque() == false return it casted to a
+/// ContantSDNode pointer else nullptr.
+static ConstantSDNode *getAsNonOpaqueConstant(SDValue N) {
+ ConstantSDNode *Const = dyn_cast<ConstantSDNode>(N);
+ return Const != nullptr && !Const->isOpaque() ? Const : nullptr;
+}
+
SDValue DAGCombiner::visitADD(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
if (N1.getOpcode() == ISD::UNDEF)
return N1;
// fold (add c1, c2) -> c1+c2
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
+ ConstantSDNode *N0C = getAsNonOpaqueConstant(N0);
+ ConstantSDNode *N1C = getAsNonOpaqueConstant(N1);
if (N0C && N1C)
return DAG.FoldConstantArithmetic(ISD::ADD, SDLoc(N), VT, N0C, N1C);
// canonicalize constant to RHS
(uint64_t)N1C->getSExtValue());
// fold ((c1-A)+c2) -> (c1+c2)-A
if (N1C && N0.getOpcode() == ISD::SUB)
- if (ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0.getOperand(0))) {
+ if (ConstantSDNode *N0C = getAsNonOpaqueConstant(N0.getOperand(0))) {
SDLoc DL(N);
return DAG.getNode(ISD::SUB, DL, VT,
DAG.getConstant(N1C->getAPIntValue()+
if (N0 == N1)
return tryFoldToZero(SDLoc(N), TLI, VT, DAG, LegalOperations, LegalTypes);
// fold (sub c1, c2) -> c1-c2
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0.getNode());
- ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.getNode());
+ ConstantSDNode *N0C = getAsNonOpaqueConstant(N0);
+ ConstantSDNode *N1C = getAsNonOpaqueConstant(N1);
if (N0C && N1C)
return DAG.FoldConstantArithmetic(ISD::SUB, SDLoc(N), VT, N0C, N1C);
// fold (sub x, c) -> (add x, -c)
bool N0IsConst = false;
bool N1IsConst = false;
+ bool N1IsOpaqueConst = false;
+ bool N0IsOpaqueConst = false;
APInt ConstValue0, ConstValue1;
// fold vector ops
if (VT.isVector()) {
N1IsConst = isConstantSplatVector(N1.getNode(), ConstValue1);
} else {
N0IsConst = isa<ConstantSDNode>(N0);
- if (N0IsConst)
+ if (N0IsConst) {
ConstValue0 = cast<ConstantSDNode>(N0)->getAPIntValue();
+ N0IsOpaqueConst = cast<ConstantSDNode>(N0)->isOpaque();
+ }
N1IsConst = isa<ConstantSDNode>(N1);
- if (N1IsConst)
+ if (N1IsConst) {
ConstValue1 = cast<ConstantSDNode>(N1)->getAPIntValue();
+ N1IsOpaqueConst = cast<ConstantSDNode>(N1)->isOpaque();
+ }
}
// fold (mul c1, c2) -> c1*c2
- if (N0IsConst && N1IsConst)
+ if (N0IsConst && N1IsConst && !N0IsOpaqueConst && !N1IsOpaqueConst)
return DAG.FoldConstantArithmetic(ISD::MUL, SDLoc(N), VT,
N0.getNode(), N1.getNode());
DAG.getConstant(0, DL, VT), N0);
}
// fold (mul x, (1 << c)) -> x << c
- if (N1IsConst && ConstValue1.isPowerOf2() && IsFullSplat) {
+ if (N1IsConst && !N1IsOpaqueConst && ConstValue1.isPowerOf2() &&
+ IsFullSplat) {
SDLoc DL(N);
return DAG.getNode(ISD::SHL, DL, VT, N0,
DAG.getConstant(ConstValue1.logBase2(), DL,
getShiftAmountTy(N0.getValueType())));
}
// fold (mul x, -(1 << c)) -> -(x << c) or (-x) << c
- if (N1IsConst && (-ConstValue1).isPowerOf2() && IsFullSplat) {
+ if (N1IsConst && !N1IsOpaqueConst && (-ConstValue1).isPowerOf2() &&
+ IsFullSplat) {
unsigned Log2Val = (-ConstValue1).logBase2();
SDLoc DL(N);
// FIXME: If the input is something that is easily negated (e.g. a
// fold (sdiv c1, c2) -> c1/c2
ConstantSDNode *N0C = isConstOrConstSplat(N0);
ConstantSDNode *N1C = isConstOrConstSplat(N1);
- if (N0C && N1C && !N1C->isNullValue())
+ if (N0C && N1C && !N0C->isOpaque() && !N1C->isOpaque())
return DAG.FoldConstantArithmetic(ISD::SDIV, SDLoc(N), VT, N0C, N1C);
// fold (sdiv X, 1) -> X
if (N1C && N1C->isOne())
}
// fold (sdiv X, pow2) -> simple ops after legalize
- if (N1C && !N1C->isNullValue() && (N1C->getAPIntValue().isPowerOf2() ||
- (-N1C->getAPIntValue()).isPowerOf2())) {
+ if (N1C && !N1C->isNullValue() && !N1C->isOpaque() &&
+ (N1C->getAPIntValue().isPowerOf2() ||
+ (-N1C->getAPIntValue()).isPowerOf2())) {
// If dividing by powers of two is cheap, then don't perform the following
// fold.
if (TLI.isPow2SDivCheap())
// fold (udiv c1, c2) -> c1/c2
ConstantSDNode *N0C = isConstOrConstSplat(N0);
ConstantSDNode *N1C = isConstOrConstSplat(N1);
- if (N0C && N1C && !N1C->isNullValue())
- return DAG.FoldConstantArithmetic(ISD::UDIV, SDLoc(N), VT, N0C, N1C);
+ if (N0C && N1C)
+ if (SDValue Folded = DAG.FoldConstantArithmetic(ISD::UDIV, SDLoc(N), VT,
+ N0C, N1C))
+ return Folded;
// fold (udiv x, (1 << c)) -> x >>u c
- if (N1C && N1C->getAPIntValue().isPowerOf2()) {
+ if (N1C && !N1C->isOpaque() && N1C->getAPIntValue().isPowerOf2()) {
SDLoc DL(N);
return DAG.getNode(ISD::SRL, DL, VT, N0,
DAG.getConstant(N1C->getAPIntValue().logBase2(), DL,
}
// fold (udiv x, (shl c, y)) -> x >>u (log2(c)+y) iff c is power of 2
if (N1.getOpcode() == ISD::SHL) {
- if (ConstantSDNode *SHC = dyn_cast<ConstantSDNode>(N1.getOperand(0))) {
+ if (ConstantSDNode *SHC = getAsNonOpaqueConstant(N1.getOperand(0))) {
if (SHC->getAPIntValue().isPowerOf2()) {
EVT ADDVT = N1.getOperand(1).getValueType();
SDLoc DL(N);
// fold (srem c1, c2) -> c1%c2
ConstantSDNode *N0C = isConstOrConstSplat(N0);
ConstantSDNode *N1C = isConstOrConstSplat(N1);
- if (N0C && N1C && !N1C->isNullValue())
- return DAG.FoldConstantArithmetic(ISD::SREM, SDLoc(N), VT, N0C, N1C);
+ if (N0C && N1C)
+ if (SDValue Folded = DAG.FoldConstantArithmetic(ISD::SREM, SDLoc(N), VT,
+ N0C, N1C))
+ return Folded;
// If we know the sign bits of both operands are zero, strength reduce to a
// urem instead. Handles (X & 0x0FFFFFFF) %s 16 -> X&15
if (!VT.isVector()) {
// fold (urem c1, c2) -> c1%c2
ConstantSDNode *N0C = isConstOrConstSplat(N0);
ConstantSDNode *N1C = isConstOrConstSplat(N1);
- if (N0C && N1C && !N1C->isNullValue())
- return DAG.FoldConstantArithmetic(ISD::UREM, SDLoc(N), VT, N0C, N1C);
+ if (N0C && N1C)
+ if (SDValue Folded = DAG.FoldConstantArithmetic(ISD::UREM, SDLoc(N), VT,
+ N0C, N1C))
+ return Folded;
// fold (urem x, pow2) -> (and x, pow2-1)
- if (N1C && !N1C->isNullValue() && N1C->getAPIntValue().isPowerOf2()) {
+ if (N1C && !N1C->isNullValue() && !N1C->isOpaque() &&
+ N1C->getAPIntValue().isPowerOf2()) {
SDLoc DL(N);
return DAG.getNode(ISD::AND, DL, VT, N0,
DAG.getConstant(N1C->getAPIntValue() - 1, DL, VT));
}
// fold (urem x, (shl pow2, y)) -> (and x, (add (shl pow2, y), -1))
if (N1.getOpcode() == ISD::SHL) {
- if (ConstantSDNode *SHC = dyn_cast<ConstantSDNode>(N1.getOperand(0))) {
+ if (ConstantSDNode *SHC = getAsNonOpaqueConstant(N1.getOperand(0))) {
if (SHC->getAPIntValue().isPowerOf2()) {
SDLoc DL(N);
SDValue Add =
}
// fold (and c1, c2) -> c1&c2
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
+ ConstantSDNode *N0C = getAsNonOpaqueConstant(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- if (N0C && N1C)
+ if (N0C && N1C && !N1C->isOpaque())
return DAG.FoldConstantArithmetic(ISD::AND, SDLoc(N), VT, N0C, N1C);
// canonicalize constant to RHS
if (isConstantIntBuildVectorOrConstantInt(N0) &&
}
// (or (and X, C1), (and Y, C2)) -> (and (or X, Y), C3) if possible.
- if (N0.getOpcode() == ISD::AND &&
- N1.getOpcode() == ISD::AND &&
- N0.getOperand(1).getOpcode() == ISD::Constant &&
- N1.getOperand(1).getOpcode() == ISD::Constant &&
+ if (N0.getOpcode() == ISD::AND && N1.getOpcode() == ISD::AND &&
// Don't increase # computations.
(N0.getNode()->hasOneUse() || N1.getNode()->hasOneUse())) {
// We can only do this xform if we know that bits from X that are set in C2
// but not in C1 are already zero. Likewise for Y.
- const APInt &LHSMask =
- cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
- const APInt &RHSMask =
- cast<ConstantSDNode>(N1.getOperand(1))->getAPIntValue();
-
- if (DAG.MaskedValueIsZero(N0.getOperand(0), RHSMask&~LHSMask) &&
- DAG.MaskedValueIsZero(N1.getOperand(0), LHSMask&~RHSMask)) {
- SDValue X = DAG.getNode(ISD::OR, SDLoc(N0), VT,
- N0.getOperand(0), N1.getOperand(0));
- SDLoc DL(LocReference);
- return DAG.getNode(ISD::AND, DL, VT, X,
- DAG.getConstant(LHSMask | RHSMask, DL, VT));
+ if (const ConstantSDNode *N0O1C =
+ getAsNonOpaqueConstant(N0.getOperand(1))) {
+ if (const ConstantSDNode *N1O1C =
+ getAsNonOpaqueConstant(N1.getOperand(1))) {
+ // We can only do this xform if we know that bits from X that are set in
+ // C2 but not in C1 are already zero. Likewise for Y.
+ const APInt &LHSMask = N0O1C->getAPIntValue();
+ const APInt &RHSMask = N1O1C->getAPIntValue();
+
+ if (DAG.MaskedValueIsZero(N0.getOperand(0), RHSMask&~LHSMask) &&
+ DAG.MaskedValueIsZero(N1.getOperand(0), LHSMask&~RHSMask)) {
+ SDValue X = DAG.getNode(ISD::OR, SDLoc(N0), VT,
+ N0.getOperand(0), N1.getOperand(0));
+ SDLoc DL(LocReference);
+ return DAG.getNode(ISD::AND, DL, VT, X,
+ DAG.getConstant(LHSMask | RHSMask, DL, VT));
+ }
+ }
}
}
}
// fold (or c1, c2) -> c1|c2
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
+ ConstantSDNode *N0C = getAsNonOpaqueConstant(N0);
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- if (N0C && N1C)
+ if (N0C && N1C && !N1C->isOpaque())
return DAG.FoldConstantArithmetic(ISD::OR, SDLoc(N), VT, N0C, N1C);
// canonicalize constant to RHS
if (isConstantIntBuildVectorOrConstantInt(N0) &&
if (N1.getOpcode() == ISD::UNDEF)
return N1;
// fold (xor c1, c2) -> c1^c2
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
+ ConstantSDNode *N0C = getAsNonOpaqueConstant(N0);
+ ConstantSDNode *N1C = getAsNonOpaqueConstant(N1);
if (N0C && N1C)
return DAG.FoldConstantArithmetic(ISD::XOR, SDLoc(N), VT, N0C, N1C);
// canonicalize constant to RHS
}
// fold (xor (xor x, c1), c2) -> (xor x, (xor c1, c2))
if (N1C && N0.getOpcode() == ISD::XOR) {
- ConstantSDNode *N00C = dyn_cast<ConstantSDNode>(N0.getOperand(0));
- ConstantSDNode *N01C = dyn_cast<ConstantSDNode>(N0.getOperand(1));
- if (N00C) {
+ if (const ConstantSDNode *N00C = getAsNonOpaqueConstant(N0.getOperand(0))) {
SDLoc DL(N);
return DAG.getNode(ISD::XOR, DL, VT, N0.getOperand(1),
DAG.getConstant(N1C->getAPIntValue() ^
N00C->getAPIntValue(), DL, VT));
}
- if (N01C) {
+ if (const ConstantSDNode *N01C = getAsNonOpaqueConstant(N0.getOperand(1))) {
SDLoc DL(N);
return DAG.getNode(ISD::XOR, DL, VT, N0.getOperand(0),
DAG.getConstant(N1C->getAPIntValue() ^
/// Handle transforms common to the three shifts, when the shift amount is a
/// constant.
SDValue DAGCombiner::visitShiftByConstant(SDNode *N, ConstantSDNode *Amt) {
- // We can't and shouldn't fold opaque constants.
- if (Amt->isOpaque())
- return SDValue();
-
SDNode *LHS = N->getOperand(0).getNode();
if (!LHS->hasOneUse()) return SDValue();
}
// We require the RHS of the binop to be a constant and not opaque as well.
- ConstantSDNode *BinOpCst = dyn_cast<ConstantSDNode>(LHS->getOperand(1));
- if (!BinOpCst || BinOpCst->isOpaque()) return SDValue();
+ ConstantSDNode *BinOpCst = getAsNonOpaqueConstant(LHS->getOperand(1));
+ if (!BinOpCst) return SDValue();
// FIXME: disable this unless the input to the binop is a shift by a constant.
// If it is not a shift, it pessimizes some common cases like:
SDValue N01 = N->getOperand(0).getOperand(1);
if (ConstantSDNode *N01C = isConstOrConstSplat(N01)) {
- EVT TruncVT = N->getValueType(0);
- SDValue N00 = N->getOperand(0).getOperand(0);
- APInt TruncC = N01C->getAPIntValue();
- TruncC = TruncC.trunc(TruncVT.getScalarSizeInBits());
- SDLoc DL(N);
+ if (!N01C->isOpaque()) {
+ EVT TruncVT = N->getValueType(0);
+ SDValue N00 = N->getOperand(0).getOperand(0);
+ APInt TruncC = N01C->getAPIntValue();
+ TruncC = TruncC.trunc(TruncVT.getScalarSizeInBits());
+ SDLoc DL(N);
- return DAG.getNode(ISD::AND, DL, TruncVT,
- DAG.getNode(ISD::TRUNCATE, DL, TruncVT, N00),
- DAG.getConstant(TruncC, DL, TruncVT));
+ return DAG.getNode(ISD::AND, DL, TruncVT,
+ DAG.getNode(ISD::TRUNCATE, DL, TruncVT, N00),
+ DAG.getConstant(TruncC, DL, TruncVT));
+ }
}
}
}
// fold (shl c1, c2) -> c1<<c2
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- if (N0C && N1C)
+ ConstantSDNode *N0C = getAsNonOpaqueConstant(N0);
+ if (N0C && N1C && !N1C->isOpaque())
return DAG.FoldConstantArithmetic(ISD::SHL, SDLoc(N), VT, N0C, N1C);
// fold (shl 0, x) -> 0
if (isNullConstant(N0))
return DAG.getNode(ISD::ADD, SDLoc(N), VT, Shl0, Shl1);
}
- if (N1C) {
+ if (N1C && !N1C->isOpaque()) {
SDValue NewSHL = visitShiftByConstant(N, N1C);
if (NewSHL.getNode())
return NewSHL;
}
// fold (sra c1, c2) -> (sra c1, c2)
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- if (N0C && N1C)
+ ConstantSDNode *N0C = getAsNonOpaqueConstant(N0);
+ if (N0C && N1C && !N1C->isOpaque())
return DAG.FoldConstantArithmetic(ISD::SRA, SDLoc(N), VT, N0C, N1C);
// fold (sra 0, x) -> 0
if (isNullConstant(N0))
if (DAG.SignBitIsZero(N0))
return DAG.getNode(ISD::SRL, SDLoc(N), VT, N0, N1);
- if (N1C) {
+ if (N1C && !N1C->isOpaque()) {
SDValue NewSRA = visitShiftByConstant(N, N1C);
if (NewSRA.getNode())
return NewSRA;
}
// fold (srl c1, c2) -> c1 >>u c2
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- if (N0C && N1C)
+ ConstantSDNode *N0C = getAsNonOpaqueConstant(N0);
+ if (N0C && N1C && !N1C->isOpaque())
return DAG.FoldConstantArithmetic(ISD::SRL, SDLoc(N), VT, N0C, N1C);
// fold (srl 0, x) -> 0
if (isNullConstant(N0))
if (N1C && SimplifyDemandedBits(SDValue(N, 0)))
return SDValue(N, 0);
- if (N1C) {
+ if (N1C && !N1C->isOpaque()) {
SDValue NewSRL = visitShiftByConstant(N, N1C);
if (NewSRL.getNode())
return NewSRL;
return SDValue();
}
+SDValue DAGCombiner::visitBSWAP(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+ EVT VT = N->getValueType(0);
+
+ // fold (bswap c1) -> c2
+ if (isConstantIntBuildVectorOrConstantInt(N0))
+ return DAG.getNode(ISD::BSWAP, SDLoc(N), VT, N0);
+ // fold (bswap (bswap x)) -> x
+ if (N0.getOpcode() == ISD::BSWAP)
+ return N0->getOperand(0);
+ return SDValue();
+}
+
SDValue DAGCombiner::visitCTLZ(SDNode *N) {
SDValue N0 = N->getOperand(0);
EVT VT = N->getValueType(0);
// fold (ctlz c1) -> c2
- if (isa<ConstantSDNode>(N0))
+ if (isConstantIntBuildVectorOrConstantInt(N0))
return DAG.getNode(ISD::CTLZ, SDLoc(N), VT, N0);
return SDValue();
}
EVT VT = N->getValueType(0);
// fold (ctlz_zero_undef c1) -> c2
- if (isa<ConstantSDNode>(N0))
+ if (isConstantIntBuildVectorOrConstantInt(N0))
return DAG.getNode(ISD::CTLZ_ZERO_UNDEF, SDLoc(N), VT, N0);
return SDValue();
}
EVT VT = N->getValueType(0);
// fold (cttz c1) -> c2
- if (isa<ConstantSDNode>(N0))
+ if (isConstantIntBuildVectorOrConstantInt(N0))
return DAG.getNode(ISD::CTTZ, SDLoc(N), VT, N0);
return SDValue();
}
EVT VT = N->getValueType(0);
// fold (cttz_zero_undef c1) -> c2
- if (isa<ConstantSDNode>(N0))
+ if (isConstantIntBuildVectorOrConstantInt(N0))
return DAG.getNode(ISD::CTTZ_ZERO_UNDEF, SDLoc(N), VT, N0);
return SDValue();
}
EVT VT = N->getValueType(0);
// fold (ctpop c1) -> c2
- if (isa<ConstantSDNode>(N0))
+ if (isConstantIntBuildVectorOrConstantInt(N0))
return DAG.getNode(ISD::CTPOP, SDLoc(N), VT, N0);
return SDValue();
}
std::tie(IndexLo, IndexHi) = DAG.SplitVector(MSC->getIndex(), DL);
MachineMemOperand *MMO = DAG.getMachineFunction().
- getMachineMemOperand(MSC->getPointerInfo(),
+ getMachineMemOperand(MSC->getPointerInfo(),
MachineMemOperand::MOStore, LoMemVT.getStoreSize(),
Alignment, MSC->getAAInfo(), MSC->getRanges());
std::tie(IndexLo, IndexHi) = DAG.SplitVector(Index, DL);
MachineMemOperand *MMO = DAG.getMachineFunction().
- getMachineMemOperand(MGT->getPointerInfo(),
+ getMachineMemOperand(MGT->getPointerInfo(),
MachineMemOperand::MOLoad, LoMemVT.getStoreSize(),
Alignment, MGT->getAAInfo(), MGT->getRanges());
EVT VT = N->getValueType(0);
assert((Opcode == ISD::SIGN_EXTEND || Opcode == ISD::ZERO_EXTEND ||
- Opcode == ISD::ANY_EXTEND) && "Expected EXTEND dag node in input!");
+ Opcode == ISD::ANY_EXTEND || Opcode == ISD::SIGN_EXTEND_VECTOR_INREG)
+ && "Expected EXTEND dag node in input!");
// fold (sext c1) -> c1
// fold (zext c1) -> c1
unsigned EVTBits = N0->getValueType(0).getScalarType().getSizeInBits();
unsigned ShAmt = VTBits - EVTBits;
SmallVector<SDValue, 8> Elts;
- unsigned NumElts = N0->getNumOperands();
+ unsigned NumElts = VT.getVectorNumElements();
SDLoc DL(N);
for (unsigned i=0; i != NumElts; ++i) {
SDLoc DL(Op);
ConstantSDNode *CurrentND = cast<ConstantSDNode>(Op);
const APInt &C = APInt(VTBits, CurrentND->getAPIntValue().getZExtValue());
- if (Opcode == ISD::SIGN_EXTEND)
+ if (Opcode == ISD::SIGN_EXTEND || Opcode == ISD::SIGN_EXTEND_VECTOR_INREG)
Elts.push_back(DAG.getConstant(C.shl(ShAmt).ashr(ShAmt).getZExtValue(),
DL, SVT));
else
// Only look at single-use SRLs.
if (!V.getNode()->hasOneUse())
break;
- if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(V.getOperand(1))) {
+ if (ConstantSDNode *RHSC = getAsNonOpaqueConstant(V.getOperand(1))) {
// See if we can recursively simplify the LHS.
unsigned Amt = RHSC->getZExtValue();
return SDValue();
}
+SDValue DAGCombiner::visitSIGN_EXTEND_VECTOR_INREG(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+ EVT VT = N->getValueType(0);
+
+ if (N0.getOpcode() == ISD::UNDEF)
+ return DAG.getUNDEF(VT);
+
+ if (SDNode *Res = tryToFoldExtendOfConstant(N, TLI, DAG, LegalTypes,
+ LegalOperations))
+ return SDValue(Res, 0);
+
+ return SDValue();
+}
+
SDValue DAGCombiner::visitTRUNCATE(SDNode *N) {
SDValue N0 = N->getOperand(0);
EVT VT = N->getValueType(0);
bool AllowNewConst = (Level < AfterLegalizeDAG);
// fold (fadd A, 0) -> A
- if (N1CFP && N1CFP->getValueAPF().isZero())
+ if (N1CFP && N1CFP->isZero())
return N0;
// fold (fadd (fadd x, c1), c2) -> (fadd x, (fadd c1, c2))
// If 'unsafe math' is enabled, fold lots of things.
if (Options.UnsafeFPMath) {
// (fsub A, 0) -> A
- if (N1CFP && N1CFP->getValueAPF().isZero())
+ if (N1CFP && N1CFP->isZero())
return N0;
// (fsub 0, B) -> -B
- if (N0CFP && N0CFP->getValueAPF().isZero()) {
+ if (N0CFP && N0CFP->isZero()) {
if (isNegatibleForFree(N1, LegalOperations, TLI, &Options))
return GetNegatedExpression(N1, DAG, LegalOperations);
if (!LegalOperations || TLI.isOperationLegal(ISD::FNEG, VT))
if (Options.UnsafeFPMath) {
// fold (fmul A, 0) -> 0
- if (N1CFP && N1CFP->getValueAPF().isZero())
+ if (N1CFP && N1CFP->isZero())
return N1;
// fold (fmul (fmul x, c1), c2) -> (fmul x, (fmul c1, c2))
auto *BV1 = dyn_cast<BuildVectorSDNode>(N1);
auto *BV00 = dyn_cast<BuildVectorSDNode>(N00);
auto *BV01 = dyn_cast<BuildVectorSDNode>(N01);
-
+
// Check 1: Make sure that the first operand of the inner multiply is NOT
// a constant. Otherwise, we may induce infinite looping.
if (!(isConstOrConstSplatFP(N00) || (BV00 && BV00->isConstant()))) {
SmallVector<SDNode *, 4> Users;
// Find all FDIV users of the same divisor.
- for (auto U : N1->uses()) {
+ for (auto *U : N1->uses()) {
if (U->getOpcode() == ISD::FDIV && U->getOperand(1) == N1)
Users.push_back(U);
}
SDValue Reciprocal = DAG.getNode(ISD::FDIV, DL, VT, FPOne, N1);
// Dividend / Divisor -> Dividend * Reciprocal
- for (auto U : Users) {
+ for (auto *U : Users) {
SDValue Dividend = U->getOperand(0);
if (Dividend != FPOne) {
SDValue NewNode = DAG.getNode(ISD::FMUL, SDLoc(U), VT, Dividend,
}
// (fneg (fmul c, x)) -> (fmul -c, x)
- if (N0.getOpcode() == ISD::FMUL) {
+ if (N0.getOpcode() == ISD::FMUL &&
+ (N0.getNode()->hasOneUse() || !TLI.isFNegFree(VT))) {
ConstantFPSDNode *CFP1 = dyn_cast<ConstantFPSDNode>(N0.getOperand(1));
if (CFP1) {
APFloat CVal = CFP1->getValueAPF();
SelectionDAG &DAG,
const TargetLowering &TLI) {
EVT VT;
+ unsigned AS;
+
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(Use)) {
if (LD->isIndexed() || LD->getBasePtr().getNode() != N)
return false;
VT = LD->getMemoryVT();
+ AS = LD->getAddressSpace();
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(Use)) {
if (ST->isIndexed() || ST->getBasePtr().getNode() != N)
return false;
VT = ST->getMemoryVT();
+ AS = ST->getAddressSpace();
} else
return false;
} else
return false;
- return TLI.isLegalAddressingMode(AM, VT.getTypeForEVT(*DAG.getContext()));
+ return TLI.isLegalAddressingMode(AM, VT.getTypeForEVT(*DAG.getContext()), AS);
}
/// Try turning a load/store into a pre-indexed load/store when the base
};
} // namespace
+SDValue DAGCombiner::getMergedConstantVectorStore(SelectionDAG &DAG,
+ SDLoc SL,
+ ArrayRef<MemOpLink> Stores,
+ EVT Ty) const {
+ SmallVector<SDValue, 8> BuildVector;
+
+ for (unsigned I = 0, E = Ty.getVectorNumElements(); I != E; ++I)
+ BuildVector.push_back(cast<StoreSDNode>(Stores[I].MemNode)->getValue());
+
+ return DAG.getNode(ISD::BUILD_VECTOR, SL, Ty, BuildVector);
+}
+
bool DAGCombiner::MergeStoresOfConstantsOrVecElts(
SmallVectorImpl<MemOpLink> &StoreNodes, EVT MemVT,
unsigned NumElem, bool IsConstantSrc, bool UseVector) {
EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT, NumElem);
assert(TLI.isTypeLegal(Ty) && "Illegal vector store");
if (IsConstantSrc) {
- // A vector store with a constant source implies that the constant is
- // zero; we only handle merging stores of constant zeros because the zero
- // can be materialized without a load.
- // It may be beneficial to loosen this restriction to allow non-zero
- // store merging.
- StoredVal = DAG.getConstant(0, DL, Ty);
+ StoredVal = getMergedConstantVectorStore(DAG, DL, StoreNodes, Ty);
} else {
SmallVector<SDValue, 8> Ops;
for (unsigned i = 0; i < NumElem ; ++i) {
// elements, so this path implies a store of constants.
assert(IsConstantSrc && "Merged vector elements should use vector store");
- unsigned StoreBW = NumElem * ElementSizeBytes * 8;
- APInt StoreInt(StoreBW, 0);
+ unsigned SizeInBits = NumElem * ElementSizeBytes * 8;
+ APInt StoreInt(SizeInBits, 0);
// Construct a single integer constant which is made of the smaller
// constant inputs.
unsigned Idx = IsLE ? (NumElem - 1 - i) : i;
StoreSDNode *St = cast<StoreSDNode>(StoreNodes[Idx].MemNode);
SDValue Val = St->getValue();
- StoreInt <<= ElementSizeBytes*8;
+ StoreInt <<= ElementSizeBytes * 8;
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Val)) {
- StoreInt |= C->getAPIntValue().zext(StoreBW);
+ StoreInt |= C->getAPIntValue().zext(SizeInBits);
} else if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(Val)) {
- StoreInt |= C->getValueAPF().bitcastToAPInt().zext(StoreBW);
+ StoreInt |= C->getValueAPF().bitcastToAPInt().zext(SizeInBits);
} else {
llvm_unreachable("Invalid constant element type");
}
}
// Create the new Load and Store operations.
- EVT StoreTy = EVT::getIntegerVT(*DAG.getContext(), StoreBW);
+ EVT StoreTy = EVT::getIntegerVT(*DAG.getContext(), SizeInBits);
StoredVal = DAG.getConstant(StoreInt, DL, StoreTy);
}
return (Align >= ABIAlignment);
}
-bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) {
- if (OptLevel == CodeGenOpt::None)
- return false;
-
- EVT MemVT = St->getMemoryVT();
- int64_t ElementSizeBytes = MemVT.getSizeInBits()/8;
- bool NoVectors = DAG.getMachineFunction().getFunction()->hasFnAttribute(
- Attribute::NoImplicitFloat);
-
- // This function cannot currently deal with non-byte-sized memory sizes.
- if (ElementSizeBytes * 8 != MemVT.getSizeInBits())
- return false;
-
- // Don't merge vectors into wider inputs.
- if (MemVT.isVector() || !MemVT.isSimple())
- return false;
-
- // Perform an early exit check. Do not bother looking at stored values that
- // are not constants, loads, or extracted vector elements.
- SDValue StoredVal = St->getValue();
- bool IsLoadSrc = isa<LoadSDNode>(StoredVal);
- bool IsConstantSrc = isa<ConstantSDNode>(StoredVal) ||
- isa<ConstantFPSDNode>(StoredVal);
- bool IsExtractVecEltSrc = (StoredVal.getOpcode() == ISD::EXTRACT_VECTOR_ELT);
-
- if (!IsConstantSrc && !IsLoadSrc && !IsExtractVecEltSrc)
- return false;
-
- // Only look at ends of store sequences.
- SDValue Chain = SDValue(St, 0);
- if (Chain->hasOneUse() && Chain->use_begin()->getOpcode() == ISD::STORE)
- return false;
-
+void DAGCombiner::getStoreMergeAndAliasCandidates(
+ StoreSDNode* St, SmallVectorImpl<MemOpLink> &StoreNodes,
+ SmallVectorImpl<LSBaseSDNode*> &AliasLoadNodes) {
// This holds the base pointer, index, and the offset in bytes from the base
// pointer.
BaseIndexOffset BasePtr = BaseIndexOffset::match(St->getBasePtr());
// We must have a base and an offset.
if (!BasePtr.Base.getNode())
- return false;
+ return;
// Do not handle stores to undef base pointers.
if (BasePtr.Base.getOpcode() == ISD::UNDEF)
- return false;
-
- // Save the LoadSDNodes that we find in the chain.
- // We need to make sure that these nodes do not interfere with
- // any of the store nodes.
- SmallVector<LSBaseSDNode*, 8> AliasLoadNodes;
-
- // Save the StoreSDNodes that we find in the chain.
- SmallVector<MemOpLink, 8> StoreNodes;
+ return;
// Walk up the chain and look for nodes with offsets from the same
// base pointer. Stop when reaching an instruction with a different kind
// or instruction which has a different base pointer.
+ EVT MemVT = St->getMemoryVT();
unsigned Seq = 0;
StoreSDNode *Index = St;
while (Index) {
}
}
}
+}
+bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) {
+ if (OptLevel == CodeGenOpt::None)
+ return false;
+
+ EVT MemVT = St->getMemoryVT();
+ int64_t ElementSizeBytes = MemVT.getSizeInBits() / 8;
+ bool NoVectors = DAG.getMachineFunction().getFunction()->hasFnAttribute(
+ Attribute::NoImplicitFloat);
+
+ // This function cannot currently deal with non-byte-sized memory sizes.
+ if (ElementSizeBytes * 8 != MemVT.getSizeInBits())
+ return false;
+
+ // Don't merge vectors into wider inputs.
+ if (MemVT.isVector() || !MemVT.isSimple())
+ return false;
+
+ // Perform an early exit check. Do not bother looking at stored values that
+ // are not constants, loads, or extracted vector elements.
+ SDValue StoredVal = St->getValue();
+ bool IsLoadSrc = isa<LoadSDNode>(StoredVal);
+ bool IsConstantSrc = isa<ConstantSDNode>(StoredVal) ||
+ isa<ConstantFPSDNode>(StoredVal);
+ bool IsExtractVecEltSrc = (StoredVal.getOpcode() == ISD::EXTRACT_VECTOR_ELT);
+
+ if (!IsConstantSrc && !IsLoadSrc && !IsExtractVecEltSrc)
+ return false;
+
+ // Only look at ends of store sequences.
+ SDValue Chain = SDValue(St, 0);
+ if (Chain->hasOneUse() && Chain->use_begin()->getOpcode() == ISD::STORE)
+ return false;
+
+ // Save the LoadSDNodes that we find in the chain.
+ // We need to make sure that these nodes do not interfere with
+ // any of the store nodes.
+ SmallVector<LSBaseSDNode*, 8> AliasLoadNodes;
+
+ // Save the StoreSDNodes that we find in the chain.
+ SmallVector<MemOpLink, 8> StoreNodes;
+
+ getStoreMergeAndAliasCandidates(St, StoreNodes, AliasLoadNodes);
+
// Check if there is anything to merge.
if (StoreNodes.size() < 2)
return false;
}
// Find a legal type for the constant store.
- unsigned StoreBW = (i+1) * ElementSizeBytes * 8;
- EVT StoreTy = EVT::getIntegerVT(*DAG.getContext(), StoreBW);
+ unsigned SizeInBits = (i+1) * ElementSizeBytes * 8;
+ EVT StoreTy = EVT::getIntegerVT(*DAG.getContext(), SizeInBits);
if (TLI.isTypeLegal(StoreTy) &&
allowableAlignment(DAG, TLI, StoreTy, FirstStoreAS,
FirstStoreAlign)) {
}
}
- // We only use vectors if the constant is known to be zero and the
- // function is not marked with the noimplicitfloat attribute.
- if (NonZero || NoVectors)
+
+ // We only use vectors if the constant is known to be zero or the target
+ // allows it and the function is not marked with the noimplicitfloat
+ // attribute.
+ if (NoVectors) {
LastLegalVectorType = 0;
+ } else if (NonZero && !TLI.storeOfVectorConstantIsCheap(MemVT,
+ LastLegalVectorType,
+ FirstStoreAS)) {
+ LastLegalVectorType = 0;
+ }
// Check if we found a legal integer type to store.
if (LastLegalType == 0 && LastLegalVectorType == 0)
}
// Find a legal type for the integer store.
- unsigned StoreBW = (i+1) * ElementSizeBytes * 8;
- StoreTy = EVT::getIntegerVT(*DAG.getContext(), StoreBW);
+ unsigned SizeInBits = (i+1) * ElementSizeBytes * 8;
+ StoreTy = EVT::getIntegerVT(*DAG.getContext(), SizeInBits);
if (TLI.isTypeLegal(StoreTy) &&
allowableAlignment(DAG, TLI, StoreTy, FirstStoreAS, FirstStoreAlign) &&
allowableAlignment(DAG, TLI, StoreTy, FirstLoadAS, FirstLoadAlign))
if (UseVectorTy) {
JointMemOpVT = EVT::getVectorVT(*DAG.getContext(), MemVT, NumElem);
} else {
- unsigned StoreBW = NumElem * ElementSizeBytes * 8;
- JointMemOpVT = EVT::getIntegerVT(*DAG.getContext(), StoreBW);
+ unsigned SizeInBits = NumElem * ElementSizeBytes * 8;
+ JointMemOpVT = EVT::getIntegerVT(*DAG.getContext(), SizeInBits);
}
SDLoc LoadDL(LoadNodes[0].MemNode);
if (Op.getOpcode() == ISD::UNDEF) continue;
// See if we can combine this build_vector into a blend with a zero vector.
- if (!VecIn2.getNode() && (isNullConstant(Op) ||
- (Op.getOpcode() == ISD::ConstantFP &&
- cast<ConstantFPSDNode>(Op.getNode())->getValueAPF().isZero()))) {
+ if (!VecIn2.getNode() && (isNullConstant(Op) || isNullFPConstant(Op))) {
UsesZeroVector = true;
continue;
}
}
// If any of the operands is a floating point scalar bitcast to a vector,
- // use floating point types throughout, and bitcast everything.
+ // use floating point types throughout, and bitcast everything.
// Replace UNDEFs by another scalar UNDEF node, of the final desired type.
if (AnyFP) {
SVT = EVT::getFloatingPointVT(OpVT.getSizeInBits());
SDValue RHS = N->getOperand(1);
SDLoc dl(N);
- // Make sure we're not running after operation legalization where it
+ // Make sure we're not running after operation legalization where it
// may have custom lowered the vector shuffles.
if (LegalOperations)
return SDValue();
if (N->getOpcode() == ISD::SDIV || N->getOpcode() == ISD::UDIV ||
N->getOpcode() == ISD::FDIV) {
if (isNullConstant(RHSOp) || (RHSOp.getOpcode() == ISD::ConstantFP &&
- cast<ConstantFPSDNode>(RHSOp.getNode())->getValueAPF().isZero()))
+ cast<ConstantFPSDNode>(RHSOp.getNode())->isZero()))
break;
}
// Check to see if we can simplify the select into an fabs node
if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(N1)) {
// Allow either -0.0 or 0.0
- if (CFP->getValueAPF().isZero()) {
+ if (CFP->isZero()) {
// select (setg[te] X, +/-0.0), X, fneg(X) -> fabs
if ((CC == ISD::SETGE || CC == ISD::SETGT) &&
N0 == N2 && N3.getOpcode() == ISD::FNEG &&
int64_t Overlap2 = (Op1->getMemoryVT().getSizeInBits() >> 3) +
Op1->getSrcValueOffset() - MinOffset;
AliasAnalysis::AliasResult AAResult =
- AA.alias(AliasAnalysis::Location(Op0->getMemOperand()->getValue(),
- Overlap1,
- UseTBAA ? Op0->getAAInfo() : AAMDNodes()),
- AliasAnalysis::Location(Op1->getMemOperand()->getValue(),
- Overlap2,
- UseTBAA ? Op1->getAAInfo() : AAMDNodes()));
+ AA.alias(MemoryLocation(Op0->getMemOperand()->getValue(), Overlap1,
+ UseTBAA ? Op0->getAAInfo() : AAMDNodes()),
+ MemoryLocation(Op1->getMemOperand()->getValue(), Overlap2,
+ UseTBAA ? Op1->getAAInfo() : AAMDNodes()));
if (AAResult == AliasAnalysis::NoAlias)
return false;
}