void deleteAndRecombine(SDNode *N);
bool recursivelyDeleteUnusedNodes(SDNode *N);
+ /// Replaces all uses of the results of one DAG node with new values.
SDValue CombineTo(SDNode *N, const SDValue *To, unsigned NumTo,
bool AddTo = true);
+ /// Replaces all uses of the results of one DAG node with new values.
SDValue CombineTo(SDNode *N, SDValue Res, bool AddTo = true) {
return CombineTo(N, &Res, 1, AddTo);
}
+ /// Replaces all uses of the results of one DAG node with new values.
SDValue CombineTo(SDNode *N, SDValue Res0, SDValue Res1,
bool AddTo = true) {
SDValue To[] = { Res0, Res1 };
SDValue visitADDE(SDNode *N);
SDValue visitSUBE(SDNode *N);
SDValue visitMUL(SDNode *N);
+ SDValue useDivRem(SDNode *N);
SDValue visitSDIV(SDNode *N);
SDValue visitUDIV(SDNode *N);
- SDValue visitSREM(SDNode *N);
- SDValue visitUREM(SDNode *N);
+ SDValue visitREM(SDNode *N);
SDValue visitMULHU(SDNode *N);
SDValue visitMULHS(SDNode *N);
SDValue visitSMUL_LOHI(SDNode *N);
SDValue visitUMUL_LOHI(SDNode *N);
SDValue visitSMULO(SDNode *N);
SDValue visitUMULO(SDNode *N);
- SDValue visitSDIVREM(SDNode *N);
- SDValue visitUDIVREM(SDNode *N);
+ SDValue visitIMINMAX(SDNode *N);
SDValue visitAND(SDNode *N);
SDValue visitANDLike(SDValue N0, SDValue N1, SDNode *LocReference);
SDValue visitOR(SDNode *N);
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 visitVSELECT(SDNode *N);
SDValue visitSELECT_CC(SDNode *N);
SDValue visitSETCC(SDNode *N);
+ SDValue visitSETCCE(SDNode *N);
SDValue visitSIGN_EXTEND(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);
SDValue visitBRCOND(SDNode *N);
SDValue visitBR_CC(SDNode *N);
SDValue visitLOAD(SDNode *N);
+
+ SDValue replaceStoreChain(StoreSDNode *ST, SDValue BetterChain);
+ SDValue replaceStoreOfFPConstant(StoreSDNode *ST);
+
SDValue visitSTORE(SDNode *N);
SDValue visitINSERT_VECTOR_ELT(SDNode *N);
SDValue visitEXTRACT_VECTOR_ELT(SDNode *N);
SDValue visitINSERT_SUBVECTOR(SDNode *N);
SDValue visitMLOAD(SDNode *N);
SDValue visitMSTORE(SDNode *N);
+ SDValue visitMGATHER(SDNode *N);
+ SDValue visitMSCATTER(SDNode *N);
SDValue visitFP_TO_FP16(SDNode *N);
+ SDValue visitFP16_TO_FP(SDNode *N);
+
+ SDValue visitFADDForFMACombine(SDNode *N);
+ SDValue visitFSUBForFMACombine(SDNode *N);
+ SDValue visitFMULForFMACombine(SDNode *N);
SDValue XformToShuffleWithZero(SDNode *N);
SDValue ReassociateOps(unsigned Opc, SDLoc DL, SDValue LHS, SDValue RHS);
unsigned HiOp);
SDValue CombineConsecutiveLoads(SDNode *N, EVT VT);
SDValue CombineExtLoad(SDNode *N);
+ SDValue combineRepeatedFPDivisors(SDNode *N);
SDValue ConstantFoldBITCASTofBUILD_VECTOR(SDNode *, EVT);
SDValue BuildSDIV(SDNode *N);
SDValue BuildSDIVPow2(SDNode *N);
SDValue BuildUDIV(SDNode *N);
- SDValue BuildReciprocalEstimate(SDValue Op);
- SDValue BuildRsqrtEstimate(SDValue Op);
- SDValue BuildRsqrtNROneConst(SDValue Op, SDValue Est, unsigned Iterations);
- SDValue BuildRsqrtNRTwoConst(SDValue Op, SDValue Est, unsigned Iterations);
+ SDValue BuildReciprocalEstimate(SDValue Op, SDNodeFlags *Flags);
+ SDValue BuildRsqrtEstimate(SDValue Op, SDNodeFlags *Flags);
+ SDValue BuildRsqrtNROneConst(SDValue Op, SDValue Est, unsigned Iterations,
+ SDNodeFlags *Flags);
+ SDValue BuildRsqrtNRTwoConst(SDValue Op, SDValue Est, unsigned Iterations,
+ SDNodeFlags *Flags);
SDValue MatchBSwapHWordLow(SDNode *N, SDValue N0, SDValue N1,
bool DemandHighBits = true);
SDValue MatchBSwapHWord(SDNode *N, SDValue N0, SDValue N1);
/// chain (aliasing node.)
SDValue FindBetterChain(SDNode *N, SDValue Chain);
+ /// Do FindBetterChain for a store and any possibly adjacent stores on
+ /// consecutive chains.
+ bool findBetterNeighborChains(StoreSDNode *St);
+
/// Holds a pointer to an LSBaseSDNode as well as information on where it
/// is located in a sequence of memory operations connected by a chain.
struct MemOpLink {
unsigned SequenceNum;
};
+ /// This is a helper function for visitMUL to check the profitability
+ /// of folding (mul (add x, c1), c2) -> (add (mul x, c2), c1*c2).
+ /// MulNode is the original multiply, AddNode is (add x, c1),
+ /// and ConstNode is c2.
+ bool isMulAddWithConstProfitable(SDNode *MulNode,
+ SDValue &AddNode,
+ SDValue &ConstNode);
+
+ /// 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,
+ SmallVectorImpl<SDValue> &Chains,
+ EVT Ty) const;
+
+ /// This is a helper function for visitAND and visitZERO_EXTEND. Returns
+ /// true if the (and (load x) c) pattern matches an extload. ExtVT returns
+ /// the type of the loaded value to be extended. LoadedVT returns the type
+ /// of the original loaded value. NarrowLoad returns whether the load would
+ /// need to be narrowed in order to match.
+ bool isAndLoadExtLoad(ConstantSDNode *AndC, LoadSDNode *LoadN,
+ EVT LoadResultTy, EVT &ExtVT, EVT &LoadedVT,
+ bool &NarrowLoad);
+
/// 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.
/// \return True if a merged store was created.
bool MergeStoresOfConstantsOrVecElts(SmallVectorImpl<MemOpLink> &StoreNodes,
- EVT MemVT, unsigned NumElem,
+ EVT MemVT, unsigned NumStores,
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.
DAGCombiner(SelectionDAG &D, AliasAnalysis &A, CodeGenOpt::Level OL)
: DAG(D), TLI(D.getTargetLoweringInfo()), Level(BeforeLegalizeTypes),
OptLevel(OL), LegalOperations(false), LegalTypes(false), AA(A) {
- auto *F = DAG.getMachineFunction().getFunction();
- ForCodeSize = F->hasFnAttribute(Attribute::OptimizeForSize) ||
- F->hasFnAttribute(Attribute::MinSize);
+ ForCodeSize = DAG.getMachineFunction().getFunction()->optForSize();
}
/// Runs the dag combiner on all nodes in the work list
assert(LHSTy.isInteger() && "Shift amount is not an integer type!");
if (LHSTy.isVector())
return LHSTy;
- return LegalTypes ? TLI.getScalarShiftAmountTy(LHSTy)
- : TLI.getPointerTy();
+ auto &DL = DAG.getDataLayout();
+ return LegalTypes ? TLI.getScalarShiftAmountTy(DL, LHSTy)
+ : TLI.getPointerTy(DL);
}
/// This method returns true if we are running before type legalization or
/// Convenience wrapper around TargetLowering::getSetCCResultType
EVT getSetCCResultType(EVT VT) const {
- return TLI.getSetCCResultType(*DAG.getContext(), VT);
+ return TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), VT);
}
};
}
assert(Op.hasOneUse() && "Unknown reuse!");
assert(Depth <= 6 && "GetNegatedExpression doesn't match isNegatibleForFree");
+
+ const SDNodeFlags *Flags = Op.getNode()->getFlags();
+
switch (Op.getOpcode()) {
default: llvm_unreachable("Unknown code");
case ISD::ConstantFP: {
APFloat V = cast<ConstantFPSDNode>(Op)->getValueAPF();
V.changeSign();
- return DAG.getConstantFP(V, Op.getValueType());
+ return DAG.getConstantFP(V, SDLoc(Op), Op.getValueType());
}
case ISD::FADD:
// FIXME: determine better conditions for this xform.
return DAG.getNode(ISD::FSUB, SDLoc(Op), Op.getValueType(),
GetNegatedExpression(Op.getOperand(0), DAG,
LegalOperations, Depth+1),
- Op.getOperand(1));
+ Op.getOperand(1), Flags);
// fold (fneg (fadd A, B)) -> (fsub (fneg B), A)
return DAG.getNode(ISD::FSUB, SDLoc(Op), Op.getValueType(),
GetNegatedExpression(Op.getOperand(1), DAG,
LegalOperations, Depth+1),
- Op.getOperand(0));
+ Op.getOperand(0), Flags);
case ISD::FSUB:
// We can't turn -(A-B) into B-A when we honor signed zeros.
assert(Options.UnsafeFPMath);
// 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)
return DAG.getNode(ISD::FSUB, SDLoc(Op), Op.getValueType(),
- Op.getOperand(1), Op.getOperand(0));
+ Op.getOperand(1), Op.getOperand(0), Flags);
case ISD::FMUL:
case ISD::FDIV:
return DAG.getNode(Op.getOpcode(), SDLoc(Op), Op.getValueType(),
GetNegatedExpression(Op.getOperand(0), DAG,
LegalOperations, Depth+1),
- Op.getOperand(1));
+ Op.getOperand(1), Flags);
// fold (fneg (fmul X, Y)) -> (fmul X, (fneg Y))
return DAG.getNode(Op.getOpcode(), SDLoc(Op), Op.getValueType(),
Op.getOperand(0),
GetNegatedExpression(Op.getOperand(1), DAG,
- LegalOperations, Depth+1));
+ LegalOperations, Depth+1), Flags);
case ISD::FP_EXTEND:
case ISD::FSIN:
if (SDNode *L = isConstantIntBuildVectorOrConstantInt(N0.getOperand(1))) {
if (SDNode *R = isConstantIntBuildVectorOrConstantInt(N1)) {
// reassoc. (op (op x, c1), c2) -> (op x, (op c1, c2))
- if (SDValue OpNode = DAG.FoldConstantArithmetic(Opc, VT, L, R))
+ if (SDValue OpNode = DAG.FoldConstantArithmetic(Opc, DL, VT, L, R))
return DAG.getNode(Opc, DL, VT, N0.getOperand(0), OpNode);
return SDValue();
}
if (SDNode *R = isConstantIntBuildVectorOrConstantInt(N1.getOperand(1))) {
if (SDNode *L = isConstantIntBuildVectorOrConstantInt(N0)) {
// reassoc. (op c2, (op x, c1)) -> (op x, (op c1, c2))
- if (SDValue OpNode = DAG.FoldConstantArithmetic(Opc, VT, R, L))
+ if (SDValue OpNode = DAG.FoldConstantArithmetic(Opc, DL, VT, R, L))
return DAG.getNode(Opc, DL, VT, N1.getOperand(0), OpNode);
return SDValue();
}
continue;
if (N->use_empty()) {
- for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
- Nodes.insert(N->getOperand(i).getNode());
+ for (const SDValue &ChildN : N->op_values())
+ Nodes.insert(ChildN.getNode());
removeFromWorklist(N);
DAG.DeleteNode(N);
LegalTypes = Level >= AfterLegalizeTypes;
// Add all the dag nodes to the worklist.
- for (SelectionDAG::allnodes_iterator I = DAG.allnodes_begin(),
- E = DAG.allnodes_end(); I != E; ++I)
- AddToWorklist(I);
+ for (SDNode &Node : DAG.allnodes())
+ AddToWorklist(&Node);
// Create a dummy node (which is not added to allnodes), that adds a reference
// to the root node, preventing it from being deleted, and tracking any
// worklist as well. Because the worklist uniques things already, this
// won't repeatedly process the same operand.
CombinedNodes.insert(N);
- for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
- if (!CombinedNodes.count(N->getOperand(i).getNode()))
- AddToWorklist(N->getOperand(i).getNode());
+ for (const SDValue &ChildN : N->op_values())
+ if (!CombinedNodes.count(ChildN.getNode()))
+ AddToWorklist(ChildN.getNode());
SDValue RV = combine(N);
case ISD::MUL: return visitMUL(N);
case ISD::SDIV: return visitSDIV(N);
case ISD::UDIV: return visitUDIV(N);
- case ISD::SREM: return visitSREM(N);
- case ISD::UREM: return visitUREM(N);
+ case ISD::SREM:
+ case ISD::UREM: return visitREM(N);
case ISD::MULHU: return visitMULHU(N);
case ISD::MULHS: return visitMULHS(N);
case ISD::SMUL_LOHI: return visitSMUL_LOHI(N);
case ISD::UMUL_LOHI: return visitUMUL_LOHI(N);
case ISD::SMULO: return visitSMULO(N);
case ISD::UMULO: return visitUMULO(N);
- case ISD::SDIVREM: return visitSDIVREM(N);
- case ISD::UDIVREM: return visitUDIVREM(N);
+ case ISD::SMIN:
+ case ISD::SMAX:
+ case ISD::UMIN:
+ case ISD::UMAX: return visitIMINMAX(N);
case ISD::AND: return visitAND(N);
case ISD::OR: return visitOR(N);
case ISD::XOR: return visitXOR(N);
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::VSELECT: return visitVSELECT(N);
case ISD::SELECT_CC: return visitSELECT_CC(N);
case ISD::SETCC: return visitSETCC(N);
+ case ISD::SETCCE: return visitSETCCE(N);
case ISD::SIGN_EXTEND: return visitSIGN_EXTEND(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);
case ISD::VECTOR_SHUFFLE: return visitVECTOR_SHUFFLE(N);
case ISD::SCALAR_TO_VECTOR: return visitSCALAR_TO_VECTOR(N);
case ISD::INSERT_SUBVECTOR: return visitINSERT_SUBVECTOR(N);
+ case ISD::MGATHER: return visitMGATHER(N);
case ISD::MLOAD: return visitMLOAD(N);
+ case ISD::MSCATTER: return visitMSCATTER(N);
case ISD::MSTORE: return visitMSTORE(N);
case ISD::FP_TO_FP16: return visitFP_TO_FP16(N);
+ case ISD::FP16_TO_FP: return visitFP16_TO_FP(N);
}
return SDValue();
}
// Constant operands are canonicalized to RHS.
if (isa<ConstantSDNode>(N0) || !isa<ConstantSDNode>(N1)) {
SDValue Ops[] = {N1, N0};
- SDNode *CSENode;
- if (const BinaryWithFlagsSDNode *BinNode =
- dyn_cast<BinaryWithFlagsSDNode>(N)) {
- CSENode = DAG.getNodeIfExists(
- N->getOpcode(), N->getVTList(), Ops, BinNode->hasNoUnsignedWrap(),
- BinNode->hasNoSignedWrap(), BinNode->isExact());
- } else {
- CSENode = DAG.getNodeIfExists(N->getOpcode(), N->getVTList(), Ops);
- }
+ SDNode *CSENode = DAG.getNodeIfExists(N->getOpcode(), N->getVTList(), Ops,
+ N->getFlags());
if (CSENode)
return SDValue(CSENode, 0);
}
SDNode *TF = TFs[i];
// Check each of the operands.
- for (unsigned i = 0, ie = TF->getNumOperands(); i != ie; ++i) {
- SDValue Op = TF->getOperand(i);
+ for (const SDValue &Op : TF->op_values()) {
switch (Op.getOpcode()) {
case ISD::EntryToken:
return SDValue(N, 0); // Return N so it doesn't get rechecked!
}
+/// 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, VT, N0C, N1C);
+ return DAG.FoldConstantArithmetic(ISD::ADD, SDLoc(N), VT, N0C, N1C);
// canonicalize constant to RHS
if (isConstantIntBuildVectorOrConstantInt(N0) &&
!isConstantIntBuildVectorOrConstantInt(N1))
return DAG.getNode(ISD::ADD, SDLoc(N), VT, N1, N0);
// fold (add x, 0) -> x
- if (N1C && N1C->isNullValue())
+ if (isNullConstant(N1))
return N0;
// fold (add Sym, c) -> Sym+c
if (GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(N0))
(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)))
- return DAG.getNode(ISD::SUB, SDLoc(N), VT,
+ if (ConstantSDNode *N0C = getAsNonOpaqueConstant(N0.getOperand(0))) {
+ SDLoc DL(N);
+ return DAG.getNode(ISD::SUB, DL, VT,
DAG.getConstant(N1C->getAPIntValue()+
- N0C->getAPIntValue(), VT),
+ N0C->getAPIntValue(), DL, VT),
N0.getOperand(1));
+ }
// reassociate add
if (SDValue RADD = ReassociateOps(ISD::ADD, SDLoc(N), N0, N1))
return RADD;
// fold ((0-A) + B) -> B-A
- if (N0.getOpcode() == ISD::SUB && isa<ConstantSDNode>(N0.getOperand(0)) &&
- cast<ConstantSDNode>(N0.getOperand(0))->isNullValue())
+ if (N0.getOpcode() == ISD::SUB && isNullConstant(N0.getOperand(0)))
return DAG.getNode(ISD::SUB, SDLoc(N), VT, N1, N0.getOperand(1));
// fold (A + (0-B)) -> A-B
- if (N1.getOpcode() == ISD::SUB && isa<ConstantSDNode>(N1.getOperand(0)) &&
- cast<ConstantSDNode>(N1.getOperand(0))->isNullValue())
+ if (N1.getOpcode() == ISD::SUB && isNullConstant(N1.getOperand(0)))
return DAG.getNode(ISD::SUB, SDLoc(N), VT, N0, N1.getOperand(1));
// fold (A+(B-A)) -> B
if (N1.getOpcode() == ISD::SUB && N0 == N1.getOperand(1))
return SDValue(N, 0);
// fold (a+b) -> (a|b) iff a and b share no bits.
- if (VT.isInteger() && !VT.isVector()) {
- APInt LHSZero, LHSOne;
- APInt RHSZero, RHSOne;
- DAG.computeKnownBits(N0, LHSZero, LHSOne);
-
- if (LHSZero.getBoolValue()) {
- DAG.computeKnownBits(N1, RHSZero, RHSOne);
-
- // If all possibly-set bits on the LHS are clear on the RHS, return an OR.
- // If all possibly-set bits on the RHS are clear on the LHS, return an OR.
- if ((RHSZero & ~LHSZero) == ~LHSZero || (LHSZero & ~RHSZero) == ~RHSZero){
- if (!LegalOperations || TLI.isOperationLegal(ISD::OR, VT))
- return DAG.getNode(ISD::OR, SDLoc(N), VT, N0, N1);
- }
- }
- }
+ if ((!LegalOperations || TLI.isOperationLegal(ISD::OR, VT)) &&
+ VT.isInteger() && !VT.isVector() && DAG.haveNoCommonBitsSet(N0, N1))
+ return DAG.getNode(ISD::OR, SDLoc(N), VT, N0, N1);
// fold (add x, shl(0 - y, n)) -> sub(x, shl(y, n))
- if (N1.getOpcode() == ISD::SHL &&
- N1.getOperand(0).getOpcode() == ISD::SUB)
- if (ConstantSDNode *C =
- dyn_cast<ConstantSDNode>(N1.getOperand(0).getOperand(0)))
- if (C->getAPIntValue() == 0)
- return DAG.getNode(ISD::SUB, SDLoc(N), VT, N0,
- DAG.getNode(ISD::SHL, SDLoc(N), VT,
- N1.getOperand(0).getOperand(1),
- N1.getOperand(1)));
- if (N0.getOpcode() == ISD::SHL &&
- N0.getOperand(0).getOpcode() == ISD::SUB)
- if (ConstantSDNode *C =
- dyn_cast<ConstantSDNode>(N0.getOperand(0).getOperand(0)))
- if (C->getAPIntValue() == 0)
- return DAG.getNode(ISD::SUB, SDLoc(N), VT, N1,
- DAG.getNode(ISD::SHL, SDLoc(N), VT,
- N0.getOperand(0).getOperand(1),
- N0.getOperand(1)));
+ if (N1.getOpcode() == ISD::SHL && N1.getOperand(0).getOpcode() == ISD::SUB &&
+ isNullConstant(N1.getOperand(0).getOperand(0)))
+ return DAG.getNode(ISD::SUB, SDLoc(N), VT, N0,
+ DAG.getNode(ISD::SHL, SDLoc(N), VT,
+ N1.getOperand(0).getOperand(1),
+ N1.getOperand(1)));
+ if (N0.getOpcode() == ISD::SHL && N0.getOperand(0).getOpcode() == ISD::SUB &&
+ isNullConstant(N0.getOperand(0).getOperand(0)))
+ return DAG.getNode(ISD::SUB, SDLoc(N), VT, N1,
+ DAG.getNode(ISD::SHL, SDLoc(N), VT,
+ N0.getOperand(0).getOperand(1),
+ N0.getOperand(1)));
if (N1.getOpcode() == ISD::AND) {
SDValue AndOp0 = N1.getOperand(0);
- ConstantSDNode *AndOp1 = dyn_cast<ConstantSDNode>(N1->getOperand(1));
unsigned NumSignBits = DAG.ComputeNumSignBits(AndOp0);
unsigned DestBits = VT.getScalarType().getSizeInBits();
// (add z, (and (sbbl x, x), 1)) -> (sub z, (sbbl x, x))
// and similar xforms where the inner op is either ~0 or 0.
- if (NumSignBits == DestBits && AndOp1 && AndOp1->isOne()) {
+ if (NumSignBits == DestBits && isOneConstant(N1->getOperand(1))) {
SDLoc DL(N);
return DAG.getNode(ISD::SUB, DL, VT, N->getOperand(0), AndOp0);
}
if (TN->getVT() == MVT::i1) {
SDLoc DL(N);
SDValue ZExt = DAG.getNode(ISD::AND, DL, VT, N1.getOperand(0),
- DAG.getConstant(1, VT));
+ DAG.getConstant(1, DL, VT));
return DAG.getNode(ISD::SUB, DL, VT, N0, ZExt);
}
}
return DAG.getNode(ISD::ADDC, SDLoc(N), N->getVTList(), N1, N0);
// fold (addc x, 0) -> x + no carry out
- if (N1C && N1C->isNullValue())
+ if (isNullConstant(N1))
return CombineTo(N, N0, DAG.getNode(ISD::CARRY_FALSE,
SDLoc(N), MVT::Glue));
SelectionDAG &DAG,
bool LegalOperations, bool LegalTypes) {
if (!VT.isVector())
- return DAG.getConstant(0, VT);
+ return DAG.getConstant(0, DL, VT);
if (!LegalOperations || TLI.isOperationLegal(ISD::BUILD_VECTOR, VT))
- return DAG.getConstant(0, VT);
+ return DAG.getConstant(0, DL, VT);
return SDValue();
}
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, VT, N0C, N1C);
+ return DAG.FoldConstantArithmetic(ISD::SUB, SDLoc(N), VT, N0C, N1C);
// fold (sub x, c) -> (add x, -c)
- if (N1C)
- return DAG.getNode(ISD::ADD, SDLoc(N), VT, N0,
- DAG.getConstant(-N1C->getAPIntValue(), VT));
+ if (N1C) {
+ SDLoc DL(N);
+ return DAG.getNode(ISD::ADD, DL, VT, N0,
+ DAG.getConstant(-N1C->getAPIntValue(), DL, VT));
+ }
// Canonicalize (sub -1, x) -> ~x, i.e. (xor x, -1)
- if (N0C && N0C->isAllOnesValue())
+ if (isAllOnesConstant(N0))
return DAG.getNode(ISD::XOR, SDLoc(N), VT, N1, N0);
// fold A-(A-B) -> B
if (N1.getOpcode() == ISD::SUB && N0 == N1.getOperand(0))
ConstantSDNode *N1C1 = N1.getOpcode() != ISD::ADD ? nullptr :
dyn_cast<ConstantSDNode>(N1.getOperand(1).getNode());
if (N1.getOpcode() == ISD::ADD && N0C && N1C1) {
+ SDLoc DL(N);
SDValue NewC = DAG.getConstant(N0C->getAPIntValue() - N1C1->getAPIntValue(),
- VT);
- return DAG.getNode(ISD::SUB, SDLoc(N), VT, NewC,
+ DL, VT);
+ return DAG.getNode(ISD::SUB, DL, VT, NewC,
N1.getOperand(0));
}
// fold ((A+(B+or-C))-B) -> A+or-C
if (GlobalAddressSDNode *GB = dyn_cast<GlobalAddressSDNode>(N1))
if (GA->getGlobal() == GB->getGlobal())
return DAG.getConstant((uint64_t)GA->getOffset() - GB->getOffset(),
- VT);
+ SDLoc(N), VT);
}
// sub X, (sextinreg Y i1) -> add X, (and Y 1)
if (TN->getVT() == MVT::i1) {
SDLoc DL(N);
SDValue ZExt = DAG.getNode(ISD::AND, DL, VT, N1.getOperand(0),
- DAG.getConstant(1, VT));
+ DAG.getConstant(1, DL, VT));
return DAG.getNode(ISD::ADD, DL, VT, N0, ZExt);
}
}
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
EVT VT = N0.getValueType();
+ SDLoc DL(N);
// If the flag result is dead, turn this into an SUB.
if (!N->hasAnyUseOfValue(1))
- return CombineTo(N, DAG.getNode(ISD::SUB, SDLoc(N), VT, N0, N1),
- DAG.getNode(ISD::CARRY_FALSE, SDLoc(N),
- MVT::Glue));
+ return CombineTo(N, DAG.getNode(ISD::SUB, DL, VT, N0, N1),
+ DAG.getNode(ISD::CARRY_FALSE, DL, MVT::Glue));
// fold (subc x, x) -> 0 + no borrow
if (N0 == N1)
- return CombineTo(N, DAG.getConstant(0, VT),
- DAG.getNode(ISD::CARRY_FALSE, SDLoc(N),
- MVT::Glue));
+ return CombineTo(N, DAG.getConstant(0, DL, VT),
+ DAG.getNode(ISD::CARRY_FALSE, DL, MVT::Glue));
// fold (subc x, 0) -> x + no borrow
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- if (N1C && N1C->isNullValue())
- return CombineTo(N, N0, DAG.getNode(ISD::CARRY_FALSE, SDLoc(N),
- MVT::Glue));
+ if (isNullConstant(N1))
+ return CombineTo(N, N0, DAG.getNode(ISD::CARRY_FALSE, DL, MVT::Glue));
// Canonicalize (sub -1, x) -> ~x, i.e. (xor x, -1) + no borrow
- if (N0C && N0C->isAllOnesValue())
- return CombineTo(N, DAG.getNode(ISD::XOR, SDLoc(N), VT, N1, N0),
- DAG.getNode(ISD::CARRY_FALSE, SDLoc(N),
- MVT::Glue));
+ if (isAllOnesConstant(N0))
+ return CombineTo(N, DAG.getNode(ISD::XOR, DL, VT, N1, N0),
+ DAG.getNode(ISD::CARRY_FALSE, DL, MVT::Glue));
return SDValue();
}
// fold (mul x, undef) -> 0
if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)
- return DAG.getConstant(0, VT);
+ return DAG.getConstant(0, SDLoc(N), VT);
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)
- return DAG.FoldConstantArithmetic(ISD::MUL, VT, N0.getNode(), N1.getNode());
+ if (N0IsConst && N1IsConst && !N0IsOpaqueConst && !N1IsOpaqueConst)
+ return DAG.FoldConstantArithmetic(ISD::MUL, SDLoc(N), VT,
+ N0.getNode(), N1.getNode());
// canonicalize constant to RHS (vector doesn't have to splat)
if (isConstantIntBuildVectorOrConstantInt(N0) &&
if (N1IsConst && ConstValue1 == 1 && IsFullSplat)
return N0;
// fold (mul x, -1) -> 0-x
- if (N1IsConst && ConstValue1.isAllOnesValue())
- return DAG.getNode(ISD::SUB, SDLoc(N), VT,
- DAG.getConstant(0, VT), N0);
+ if (N1IsConst && ConstValue1.isAllOnesValue()) {
+ SDLoc DL(N);
+ return DAG.getNode(ISD::SUB, DL, VT,
+ DAG.getConstant(0, DL, VT), N0);
+ }
// fold (mul x, (1 << c)) -> x << c
- if (N1IsConst && ConstValue1.isPowerOf2() && IsFullSplat)
- return DAG.getNode(ISD::SHL, SDLoc(N), VT, N0,
- DAG.getConstant(ConstValue1.logBase2(),
+ 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
// single-use add), we should put the negate there.
- return DAG.getNode(ISD::SUB, SDLoc(N), VT,
- DAG.getConstant(0, VT),
- DAG.getNode(ISD::SHL, SDLoc(N), VT, N0,
- DAG.getConstant(Log2Val,
+ return DAG.getNode(ISD::SUB, DL, VT,
+ DAG.getConstant(0, DL, VT),
+ DAG.getNode(ISD::SHL, DL, VT, N0,
+ DAG.getConstant(Log2Val, DL,
getShiftAmountTy(N0.getValueType()))));
}
}
// fold (mul (add x, c1), c2) -> (add (mul x, c2), c1*c2)
- if (N1IsConst && N0.getOpcode() == ISD::ADD && N0.getNode()->hasOneUse() &&
- (isConstantSplatVector(N0.getOperand(1).getNode(), Val) ||
- isa<ConstantSDNode>(N0.getOperand(1))))
- return DAG.getNode(ISD::ADD, SDLoc(N), VT,
- DAG.getNode(ISD::MUL, SDLoc(N0), VT,
- N0.getOperand(0), N1),
- DAG.getNode(ISD::MUL, SDLoc(N1), VT,
- N0.getOperand(1), N1));
+ if (isConstantIntBuildVectorOrConstantInt(N1) &&
+ N0.getOpcode() == ISD::ADD &&
+ isConstantIntBuildVectorOrConstantInt(N0.getOperand(1)) &&
+ isMulAddWithConstProfitable(N, N0, N1))
+ return DAG.getNode(ISD::ADD, SDLoc(N), VT,
+ DAG.getNode(ISD::MUL, SDLoc(N0), VT,
+ N0.getOperand(0), N1),
+ DAG.getNode(ISD::MUL, SDLoc(N1), VT,
+ N0.getOperand(1), N1));
// reassociate mul
if (SDValue RMUL = ReassociateOps(ISD::MUL, SDLoc(N), N0, N1))
return SDValue();
}
+/// Return true if divmod libcall is available.
+static bool isDivRemLibcallAvailable(SDNode *Node, bool isSigned,
+ const TargetLowering &TLI) {
+ RTLIB::Libcall LC;
+ switch (Node->getSimpleValueType(0).SimpleTy) {
+ default: return false; // No libcall for vector types.
+ case MVT::i8: LC= isSigned ? RTLIB::SDIVREM_I8 : RTLIB::UDIVREM_I8; break;
+ case MVT::i16: LC= isSigned ? RTLIB::SDIVREM_I16 : RTLIB::UDIVREM_I16; break;
+ case MVT::i32: LC= isSigned ? RTLIB::SDIVREM_I32 : RTLIB::UDIVREM_I32; break;
+ case MVT::i64: LC= isSigned ? RTLIB::SDIVREM_I64 : RTLIB::UDIVREM_I64; break;
+ case MVT::i128: LC= isSigned ? RTLIB::SDIVREM_I128:RTLIB::UDIVREM_I128; break;
+ }
+
+ return TLI.getLibcallName(LC) != nullptr;
+}
+
+/// Issue divrem if both quotient and remainder are needed.
+SDValue DAGCombiner::useDivRem(SDNode *Node) {
+ if (Node->use_empty())
+ return SDValue(); // This is a dead node, leave it alone.
+
+ EVT VT = Node->getValueType(0);
+ if (!TLI.isTypeLegal(VT))
+ return SDValue();
+
+ unsigned Opcode = Node->getOpcode();
+ bool isSigned = (Opcode == ISD::SDIV) || (Opcode == ISD::SREM);
+
+ unsigned DivRemOpc = isSigned ? ISD::SDIVREM : ISD::UDIVREM;
+ // If DIVREM is going to get expanded into a libcall,
+ // but there is no libcall available, then don't combine.
+ if (!TLI.isOperationLegalOrCustom(DivRemOpc, VT) &&
+ !isDivRemLibcallAvailable(Node, isSigned, TLI))
+ return SDValue();
+
+ // If div is legal, it's better to do the normal expansion
+ unsigned OtherOpcode = 0;
+ if ((Opcode == ISD::SDIV) || (Opcode == ISD::UDIV)) {
+ OtherOpcode = isSigned ? ISD::SREM : ISD::UREM;
+ if (TLI.isOperationLegalOrCustom(Opcode, VT))
+ return SDValue();
+ } else {
+ OtherOpcode = isSigned ? ISD::SDIV : ISD::UDIV;
+ if (TLI.isOperationLegalOrCustom(OtherOpcode, VT))
+ return SDValue();
+ }
+
+ SDValue Op0 = Node->getOperand(0);
+ SDValue Op1 = Node->getOperand(1);
+ SDValue combined;
+ for (SDNode::use_iterator UI = Op0.getNode()->use_begin(),
+ UE = Op0.getNode()->use_end(); UI != UE; ++UI) {
+ SDNode *User = *UI;
+ if (User == Node || User->use_empty())
+ continue;
+ // Convert the other matching node(s), too;
+ // otherwise, the DIVREM may get target-legalized into something
+ // target-specific that we won't be able to recognize.
+ unsigned UserOpc = User->getOpcode();
+ if ((UserOpc == Opcode || UserOpc == OtherOpcode || UserOpc == DivRemOpc) &&
+ User->getOperand(0) == Op0 &&
+ User->getOperand(1) == Op1) {
+ if (!combined) {
+ if (UserOpc == OtherOpcode) {
+ SDVTList VTs = DAG.getVTList(VT, VT);
+ combined = DAG.getNode(DivRemOpc, SDLoc(Node), VTs, Op0, Op1);
+ } else if (UserOpc == DivRemOpc) {
+ combined = SDValue(User, 0);
+ } else {
+ assert(UserOpc == Opcode);
+ continue;
+ }
+ }
+ if (UserOpc == ISD::SDIV || UserOpc == ISD::UDIV)
+ CombineTo(User, combined);
+ else if (UserOpc == ISD::SREM || UserOpc == ISD::UREM)
+ CombineTo(User, combined.getValue(1));
+ }
+ }
+ return combined;
+}
+
SDValue DAGCombiner::visitSDIV(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
if (SDValue FoldedVOp = SimplifyVBinOp(N))
return FoldedVOp;
+ SDLoc DL(N);
+
// fold (sdiv c1, c2) -> c1/c2
ConstantSDNode *N0C = isConstOrConstSplat(N0);
ConstantSDNode *N1C = isConstOrConstSplat(N1);
- if (N0C && N1C && !N1C->isNullValue())
- return DAG.FoldConstantArithmetic(ISD::SDIV, VT, N0C, N1C);
+ if (N0C && N1C && !N0C->isOpaque() && !N1C->isOpaque())
+ return DAG.FoldConstantArithmetic(ISD::SDIV, DL, VT, N0C, N1C);
// fold (sdiv X, 1) -> X
- if (N1C && N1C->getAPIntValue() == 1LL)
+ if (N1C && N1C->isOne())
return N0;
// fold (sdiv X, -1) -> 0-X
if (N1C && N1C->isAllOnesValue())
- return DAG.getNode(ISD::SUB, SDLoc(N), VT,
- DAG.getConstant(0, VT), N0);
+ return DAG.getNode(ISD::SUB, DL, VT,
+ DAG.getConstant(0, DL, VT), N0);
+
// If we know the sign bits of both operands are zero, strength reduce to a
// udiv instead. Handles (X&15) /s 4 -> X&15 >> 2
if (!VT.isVector()) {
if (DAG.SignBitIsZero(N1) && DAG.SignBitIsZero(N0))
- return DAG.getNode(ISD::UDIV, SDLoc(N), N1.getValueType(),
- N0, N1);
+ return DAG.getNode(ISD::UDIV, DL, N1.getValueType(), N0, N1);
}
// fold (sdiv X, pow2) -> simple ops after legalize
- if (N1C && !N1C->isNullValue() && (N1C->getAPIntValue().isPowerOf2() ||
- (-N1C->getAPIntValue()).isPowerOf2())) {
- // If dividing by powers of two is cheap, then don't perform the following
- // fold.
- if (TLI.isPow2SDivCheap())
- return SDValue();
-
+ // FIXME: We check for the exact bit here because the generic lowering gives
+ // better results in that case. The target-specific lowering should learn how
+ // to handle exact sdivs efficiently.
+ if (N1C && !N1C->isNullValue() && !N1C->isOpaque() &&
+ !cast<BinaryWithFlagsSDNode>(N)->Flags.hasExact() &&
+ (N1C->getAPIntValue().isPowerOf2() ||
+ (-N1C->getAPIntValue()).isPowerOf2())) {
// Target-specific implementation of sdiv x, pow2.
- SDValue Res = BuildSDIVPow2(N);
- if (Res.getNode())
+ if (SDValue Res = BuildSDIVPow2(N))
return Res;
unsigned lg2 = N1C->getAPIntValue().countTrailingZeros();
// Splat the sign bit into the register
SDValue SGN =
- DAG.getNode(ISD::SRA, SDLoc(N), VT, N0,
- DAG.getConstant(VT.getScalarSizeInBits() - 1,
+ DAG.getNode(ISD::SRA, DL, VT, N0,
+ DAG.getConstant(VT.getScalarSizeInBits() - 1, DL,
getShiftAmountTy(N0.getValueType())));
AddToWorklist(SGN.getNode());
// Add (N0 < 0) ? abs2 - 1 : 0;
SDValue SRL =
- DAG.getNode(ISD::SRL, SDLoc(N), VT, SGN,
- DAG.getConstant(VT.getScalarSizeInBits() - lg2,
+ DAG.getNode(ISD::SRL, DL, VT, SGN,
+ DAG.getConstant(VT.getScalarSizeInBits() - lg2, DL,
getShiftAmountTy(SGN.getValueType())));
- SDValue ADD = DAG.getNode(ISD::ADD, SDLoc(N), VT, N0, SRL);
+ SDValue ADD = DAG.getNode(ISD::ADD, DL, VT, N0, SRL);
AddToWorklist(SRL.getNode());
AddToWorklist(ADD.getNode()); // Divide by pow2
- SDValue SRA = DAG.getNode(ISD::SRA, SDLoc(N), VT, ADD,
- DAG.getConstant(lg2, getShiftAmountTy(ADD.getValueType())));
+ SDValue SRA = DAG.getNode(ISD::SRA, DL, VT, ADD,
+ DAG.getConstant(lg2, DL,
+ getShiftAmountTy(ADD.getValueType())));
// If we're dividing by a positive value, we're done. Otherwise, we must
// negate the result.
return SRA;
AddToWorklist(SRA.getNode());
- return DAG.getNode(ISD::SUB, SDLoc(N), VT, DAG.getConstant(0, VT), SRA);
+ return DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, DL, VT), SRA);
}
// If integer divide is expensive and we satisfy the requirements, emit an
- // alternate sequence.
- if (N1C && !TLI.isIntDivCheap()) {
- SDValue Op = BuildSDIV(N);
- if (Op.getNode()) return Op;
- }
+ // alternate sequence. Targets may check function attributes for size/speed
+ // trade-offs.
+ AttributeSet Attr = DAG.getMachineFunction().getFunction()->getAttributes();
+ if (N1C && !TLI.isIntDivCheap(N->getValueType(0), Attr))
+ if (SDValue Op = BuildSDIV(N))
+ return Op;
+
+ // sdiv, srem -> sdivrem
+ // If the divisor is constant, then return DIVREM only if isIntDivCheap() is true.
+ // Otherwise, we break the simplification logic in visitREM().
+ if (!N1C || TLI.isIntDivCheap(N->getValueType(0), Attr))
+ if (SDValue DivRem = useDivRem(N))
+ return DivRem;
// undef / X -> 0
if (N0.getOpcode() == ISD::UNDEF)
- return DAG.getConstant(0, VT);
+ return DAG.getConstant(0, DL, VT);
// X / undef -> undef
if (N1.getOpcode() == ISD::UNDEF)
return N1;
if (SDValue FoldedVOp = SimplifyVBinOp(N))
return FoldedVOp;
+ SDLoc DL(N);
+
// fold (udiv c1, c2) -> c1/c2
ConstantSDNode *N0C = isConstOrConstSplat(N0);
ConstantSDNode *N1C = isConstOrConstSplat(N1);
- if (N0C && N1C && !N1C->isNullValue())
- return DAG.FoldConstantArithmetic(ISD::UDIV, VT, N0C, N1C);
+ if (N0C && N1C)
+ if (SDValue Folded = DAG.FoldConstantArithmetic(ISD::UDIV, DL, VT,
+ N0C, N1C))
+ return Folded;
// fold (udiv x, (1 << c)) -> x >>u c
- if (N1C && N1C->getAPIntValue().isPowerOf2())
- return DAG.getNode(ISD::SRL, SDLoc(N), VT, N0,
- DAG.getConstant(N1C->getAPIntValue().logBase2(),
+ if (N1C && !N1C->isOpaque() && N1C->getAPIntValue().isPowerOf2())
+ return DAG.getNode(ISD::SRL, DL, VT, N0,
+ DAG.getConstant(N1C->getAPIntValue().logBase2(), DL,
getShiftAmountTy(N0.getValueType())));
+
// 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();
- SDValue Add = DAG.getNode(ISD::ADD, SDLoc(N), ADDVT,
+ SDValue Add = DAG.getNode(ISD::ADD, DL, ADDVT,
N1.getOperand(1),
DAG.getConstant(SHC->getAPIntValue()
.logBase2(),
- ADDVT));
+ DL, ADDVT));
AddToWorklist(Add.getNode());
- return DAG.getNode(ISD::SRL, SDLoc(N), VT, N0, Add);
+ return DAG.getNode(ISD::SRL, DL, VT, N0, Add);
}
}
}
+
// fold (udiv x, c) -> alternate
- if (N1C && !TLI.isIntDivCheap()) {
- SDValue Op = BuildUDIV(N);
- if (Op.getNode()) return Op;
- }
+ AttributeSet Attr = DAG.getMachineFunction().getFunction()->getAttributes();
+ if (N1C && !TLI.isIntDivCheap(N->getValueType(0), Attr))
+ if (SDValue Op = BuildUDIV(N))
+ return Op;
+
+ // sdiv, srem -> sdivrem
+ // If the divisor is constant, then return DIVREM only if isIntDivCheap() is true.
+ // Otherwise, we break the simplification logic in visitREM().
+ if (!N1C || TLI.isIntDivCheap(N->getValueType(0), Attr))
+ if (SDValue DivRem = useDivRem(N))
+ return DivRem;
// undef / X -> 0
if (N0.getOpcode() == ISD::UNDEF)
- return DAG.getConstant(0, VT);
+ return DAG.getConstant(0, DL, VT);
// X / undef -> undef
if (N1.getOpcode() == ISD::UNDEF)
return N1;
return SDValue();
}
-SDValue DAGCombiner::visitSREM(SDNode *N) {
+// handles ISD::SREM and ISD::UREM
+SDValue DAGCombiner::visitREM(SDNode *N) {
+ unsigned Opcode = N->getOpcode();
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
EVT VT = N->getValueType(0);
+ bool isSigned = (Opcode == ISD::SREM);
+ SDLoc DL(N);
- // fold (srem c1, c2) -> c1%c2
+ // fold (rem c1, c2) -> c1%c2
ConstantSDNode *N0C = isConstOrConstSplat(N0);
ConstantSDNode *N1C = isConstOrConstSplat(N1);
- if (N0C && N1C && !N1C->isNullValue())
- return DAG.FoldConstantArithmetic(ISD::SREM, VT, N0C, N1C);
- // 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()) {
- if (DAG.SignBitIsZero(N1) && DAG.SignBitIsZero(N0))
- return DAG.getNode(ISD::UREM, SDLoc(N), VT, N0, N1);
- }
+ if (N0C && N1C)
+ if (SDValue Folded = DAG.FoldConstantArithmetic(Opcode, DL, VT, N0C, N1C))
+ return Folded;
- // If X/C can be simplified by the division-by-constant logic, lower
- // X%C to the equivalent of X-X/C*C.
- if (N1C && !N1C->isNullValue()) {
- SDValue Div = DAG.getNode(ISD::SDIV, SDLoc(N), VT, N0, N1);
- AddToWorklist(Div.getNode());
- SDValue OptimizedDiv = combine(Div.getNode());
- if (OptimizedDiv.getNode() && OptimizedDiv.getNode() != Div.getNode()) {
- SDValue Mul = DAG.getNode(ISD::MUL, SDLoc(N), VT,
- OptimizedDiv, N1);
- SDValue Sub = DAG.getNode(ISD::SUB, SDLoc(N), VT, N0, Mul);
- AddToWorklist(Mul.getNode());
- return Sub;
+ if (isSigned) {
+ // 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()) {
+ if (DAG.SignBitIsZero(N1) && DAG.SignBitIsZero(N0))
+ return DAG.getNode(ISD::UREM, DL, VT, N0, N1);
}
- }
-
- // undef % X -> 0
- if (N0.getOpcode() == ISD::UNDEF)
- return DAG.getConstant(0, VT);
- // X % undef -> undef
- if (N1.getOpcode() == ISD::UNDEF)
- return N1;
-
- return SDValue();
-}
-
-SDValue DAGCombiner::visitUREM(SDNode *N) {
- SDValue N0 = N->getOperand(0);
- SDValue N1 = N->getOperand(1);
- EVT VT = N->getValueType(0);
-
- // fold (urem c1, c2) -> c1%c2
- ConstantSDNode *N0C = isConstOrConstSplat(N0);
- ConstantSDNode *N1C = isConstOrConstSplat(N1);
- if (N0C && N1C && !N1C->isNullValue())
- return DAG.FoldConstantArithmetic(ISD::UREM, VT, N0C, N1C);
- // fold (urem x, pow2) -> (and x, pow2-1)
- if (N1C && !N1C->isNullValue() && N1C->getAPIntValue().isPowerOf2())
- return DAG.getNode(ISD::AND, SDLoc(N), VT, N0,
- DAG.getConstant(N1C->getAPIntValue()-1,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 (SHC->getAPIntValue().isPowerOf2()) {
- SDValue Add =
- DAG.getNode(ISD::ADD, SDLoc(N), VT, N1,
- DAG.getConstant(APInt::getAllOnesValue(VT.getSizeInBits()),
+ } else {
+ // fold (urem x, pow2) -> (and x, pow2-1)
+ if (N1C && !N1C->isNullValue() && !N1C->isOpaque() &&
+ N1C->getAPIntValue().isPowerOf2()) {
+ 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 = getAsNonOpaqueConstant(N1.getOperand(0))) {
+ if (SHC->getAPIntValue().isPowerOf2()) {
+ SDValue Add =
+ DAG.getNode(ISD::ADD, DL, VT, N1,
+ DAG.getConstant(APInt::getAllOnesValue(VT.getSizeInBits()), DL,
VT));
- AddToWorklist(Add.getNode());
- return DAG.getNode(ISD::AND, SDLoc(N), VT, N0, Add);
+ AddToWorklist(Add.getNode());
+ return DAG.getNode(ISD::AND, DL, VT, N0, Add);
+ }
}
}
}
+ AttributeSet Attr = DAG.getMachineFunction().getFunction()->getAttributes();
+
// If X/C can be simplified by the division-by-constant logic, lower
// X%C to the equivalent of X-X/C*C.
- if (N1C && !N1C->isNullValue()) {
- SDValue Div = DAG.getNode(ISD::UDIV, SDLoc(N), VT, N0, N1);
+ // To avoid mangling nodes, this simplification requires that the combine()
+ // call for the speculative DIV must not cause a DIVREM conversion. We guard
+ // against this by skipping the simplification if isIntDivCheap(). When
+ // div is not cheap, combine will not return a DIVREM. Regardless,
+ // checking cheapness here makes sense since the simplification results in
+ // fatter code.
+ if (N1C && !N1C->isNullValue() && !TLI.isIntDivCheap(VT, Attr)) {
+ unsigned DivOpcode = isSigned ? ISD::SDIV : ISD::UDIV;
+ SDValue Div = DAG.getNode(DivOpcode, DL, VT, N0, N1);
AddToWorklist(Div.getNode());
SDValue OptimizedDiv = combine(Div.getNode());
if (OptimizedDiv.getNode() && OptimizedDiv.getNode() != Div.getNode()) {
- SDValue Mul = DAG.getNode(ISD::MUL, SDLoc(N), VT,
- OptimizedDiv, N1);
- SDValue Sub = DAG.getNode(ISD::SUB, SDLoc(N), VT, N0, Mul);
+ assert((OptimizedDiv.getOpcode() != ISD::UDIVREM) &&
+ (OptimizedDiv.getOpcode() != ISD::SDIVREM));
+ SDValue Mul = DAG.getNode(ISD::MUL, DL, VT, OptimizedDiv, N1);
+ SDValue Sub = DAG.getNode(ISD::SUB, DL, VT, N0, Mul);
AddToWorklist(Mul.getNode());
return Sub;
}
}
+ // sdiv, srem -> sdivrem
+ if (SDValue DivRem = useDivRem(N))
+ return DivRem.getValue(1);
+
// undef % X -> 0
if (N0.getOpcode() == ISD::UNDEF)
- return DAG.getConstant(0, VT);
+ return DAG.getConstant(0, DL, VT);
// X % undef -> undef
if (N1.getOpcode() == ISD::UNDEF)
return N1;
SDValue DAGCombiner::visitMULHS(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
- ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
EVT VT = N->getValueType(0);
SDLoc DL(N);
// fold (mulhs x, 0) -> 0
- if (N1C && N1C->isNullValue())
+ if (isNullConstant(N1))
return N1;
// fold (mulhs x, 1) -> (sra x, size(x)-1)
- if (N1C && N1C->getAPIntValue() == 1)
- return DAG.getNode(ISD::SRA, SDLoc(N), N0.getValueType(), N0,
+ if (isOneConstant(N1)) {
+ SDLoc DL(N);
+ return DAG.getNode(ISD::SRA, DL, N0.getValueType(), N0,
DAG.getConstant(N0.getValueType().getSizeInBits() - 1,
+ DL,
getShiftAmountTy(N0.getValueType())));
+ }
// fold (mulhs x, undef) -> 0
if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)
- return DAG.getConstant(0, VT);
+ return DAG.getConstant(0, SDLoc(N), VT);
// If the type twice as wide is legal, transform the mulhs to a wider multiply
// plus a shift.
N1 = DAG.getNode(ISD::SIGN_EXTEND, DL, NewVT, N1);
N1 = DAG.getNode(ISD::MUL, DL, NewVT, N0, N1);
N1 = DAG.getNode(ISD::SRL, DL, NewVT, N1,
- DAG.getConstant(SimpleSize, getShiftAmountTy(N1.getValueType())));
+ DAG.getConstant(SimpleSize, DL,
+ getShiftAmountTy(N1.getValueType())));
return DAG.getNode(ISD::TRUNCATE, DL, VT, N1);
}
}
SDValue DAGCombiner::visitMULHU(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
- ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
EVT VT = N->getValueType(0);
SDLoc DL(N);
// fold (mulhu x, 0) -> 0
- if (N1C && N1C->isNullValue())
+ if (isNullConstant(N1))
return N1;
// fold (mulhu x, 1) -> 0
- if (N1C && N1C->getAPIntValue() == 1)
- return DAG.getConstant(0, N0.getValueType());
+ if (isOneConstant(N1))
+ return DAG.getConstant(0, DL, N0.getValueType());
// fold (mulhu x, undef) -> 0
if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)
- return DAG.getConstant(0, VT);
+ return DAG.getConstant(0, DL, VT);
// If the type twice as wide is legal, transform the mulhu to a wider multiply
// plus a shift.
N1 = DAG.getNode(ISD::ZERO_EXTEND, DL, NewVT, N1);
N1 = DAG.getNode(ISD::MUL, DL, NewVT, N0, N1);
N1 = DAG.getNode(ISD::SRL, DL, NewVT, N1,
- DAG.getConstant(SimpleSize, getShiftAmountTy(N1.getValueType())));
+ DAG.getConstant(SimpleSize, DL,
+ getShiftAmountTy(N1.getValueType())));
return DAG.getNode(ISD::TRUNCATE, DL, VT, N1);
}
}
}
SDValue DAGCombiner::visitSMUL_LOHI(SDNode *N) {
- SDValue Res = SimplifyNodeWithTwoResults(N, ISD::MUL, ISD::MULHS);
- if (Res.getNode()) return Res;
+ if (SDValue Res = SimplifyNodeWithTwoResults(N, ISD::MUL, ISD::MULHS))
+ return Res;
EVT VT = N->getValueType(0);
SDLoc DL(N);
Lo = DAG.getNode(ISD::MUL, DL, NewVT, Lo, Hi);
// Compute the high part as N1.
Hi = DAG.getNode(ISD::SRL, DL, NewVT, Lo,
- DAG.getConstant(SimpleSize, getShiftAmountTy(Lo.getValueType())));
+ DAG.getConstant(SimpleSize, DL,
+ getShiftAmountTy(Lo.getValueType())));
Hi = DAG.getNode(ISD::TRUNCATE, DL, VT, Hi);
// Compute the low part as N0.
Lo = DAG.getNode(ISD::TRUNCATE, DL, VT, Lo);
}
SDValue DAGCombiner::visitUMUL_LOHI(SDNode *N) {
- SDValue Res = SimplifyNodeWithTwoResults(N, ISD::MUL, ISD::MULHU);
- if (Res.getNode()) return Res;
+ if (SDValue Res = SimplifyNodeWithTwoResults(N, ISD::MUL, ISD::MULHU))
+ return Res;
EVT VT = N->getValueType(0);
SDLoc DL(N);
Lo = DAG.getNode(ISD::MUL, DL, NewVT, Lo, Hi);
// Compute the high part as N1.
Hi = DAG.getNode(ISD::SRL, DL, NewVT, Lo,
- DAG.getConstant(SimpleSize, getShiftAmountTy(Lo.getValueType())));
+ DAG.getConstant(SimpleSize, DL,
+ getShiftAmountTy(Lo.getValueType())));
Hi = DAG.getNode(ISD::TRUNCATE, DL, VT, Hi);
// Compute the low part as N0.
Lo = DAG.getNode(ISD::TRUNCATE, DL, VT, Lo);
return SDValue();
}
-SDValue DAGCombiner::visitSDIVREM(SDNode *N) {
- SDValue Res = SimplifyNodeWithTwoResults(N, ISD::SDIV, ISD::SREM);
- if (Res.getNode()) return Res;
+SDValue DAGCombiner::visitIMINMAX(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+ SDValue N1 = N->getOperand(1);
+ EVT VT = N0.getValueType();
- return SDValue();
-}
+ // fold vector ops
+ if (VT.isVector())
+ if (SDValue FoldedVOp = SimplifyVBinOp(N))
+ return FoldedVOp;
+
+ // fold (add c1, c2) -> c1+c2
+ ConstantSDNode *N0C = getAsNonOpaqueConstant(N0);
+ ConstantSDNode *N1C = getAsNonOpaqueConstant(N1);
+ if (N0C && N1C)
+ return DAG.FoldConstantArithmetic(N->getOpcode(), SDLoc(N), VT, N0C, N1C);
-SDValue DAGCombiner::visitUDIVREM(SDNode *N) {
- SDValue Res = SimplifyNodeWithTwoResults(N, ISD::UDIV, ISD::UREM);
- if (Res.getNode()) return Res;
+ // canonicalize constant to RHS
+ if (isConstantIntBuildVectorOrConstantInt(N0) &&
+ !isConstantIntBuildVectorOrConstantInt(N1))
+ return DAG.getNode(N->getOpcode(), SDLoc(N), VT, N1, N0);
return SDValue();
}
// build vector of all zeros that might be illegal at this stage.
if (N->getOpcode() == ISD::XOR && ShOp.getOpcode() != ISD::UNDEF) {
if (!LegalTypes)
- ShOp = DAG.getConstant(0, VT);
+ ShOp = DAG.getConstant(0, SDLoc(N), VT);
else
ShOp = SDValue();
}
ShOp = N0->getOperand(0);
if (N->getOpcode() == ISD::XOR && ShOp.getOpcode() != ISD::UNDEF) {
if (!LegalTypes)
- ShOp = DAG.getConstant(0, VT);
+ ShOp = DAG.getConstant(0, SDLoc(N), VT);
else
ShOp = SDValue();
}
// fold (and x, undef) -> 0
if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)
- return DAG.getConstant(0, VT);
+ return DAG.getConstant(0, SDLoc(LocReference), VT);
// fold (and (setcc x), (setcc y)) -> (setcc (and x, y))
SDValue LL, LR, RL, RR, CC0, CC1;
if (isSetCCEquivalent(N0, LL, LR, CC0) && isSetCCEquivalent(N1, RL, RR, CC1)){
if (LR == RR && isa<ConstantSDNode>(LR) && Op0 == Op1 &&
LL.getValueType().isInteger()) {
// fold (and (seteq X, 0), (seteq Y, 0)) -> (seteq (or X, Y), 0)
- if (cast<ConstantSDNode>(LR)->isNullValue() && Op1 == ISD::SETEQ) {
+ if (isNullConstant(LR) && Op1 == ISD::SETEQ) {
SDValue ORNode = DAG.getNode(ISD::OR, SDLoc(N0),
LR.getValueType(), LL, RL);
AddToWorklist(ORNode.getNode());
return DAG.getSetCC(SDLoc(LocReference), VT, ORNode, LR, Op1);
}
- // fold (and (seteq X, -1), (seteq Y, -1)) -> (seteq (and X, Y), -1)
- if (cast<ConstantSDNode>(LR)->isAllOnesValue() && Op1 == ISD::SETEQ) {
- SDValue ANDNode = DAG.getNode(ISD::AND, SDLoc(N0),
- LR.getValueType(), LL, RL);
- AddToWorklist(ANDNode.getNode());
- return DAG.getSetCC(SDLoc(LocReference), VT, ANDNode, LR, Op1);
- }
- // fold (and (setgt X, -1), (setgt Y, -1)) -> (setgt (or X, Y), -1)
- if (cast<ConstantSDNode>(LR)->isAllOnesValue() && Op1 == ISD::SETGT) {
- SDValue ORNode = DAG.getNode(ISD::OR, SDLoc(N0),
- LR.getValueType(), LL, RL);
- AddToWorklist(ORNode.getNode());
- return DAG.getSetCC(SDLoc(LocReference), VT, ORNode, LR, Op1);
+ if (isAllOnesConstant(LR)) {
+ // fold (and (seteq X, -1), (seteq Y, -1)) -> (seteq (and X, Y), -1)
+ if (Op1 == ISD::SETEQ) {
+ SDValue ANDNode = DAG.getNode(ISD::AND, SDLoc(N0),
+ LR.getValueType(), LL, RL);
+ AddToWorklist(ANDNode.getNode());
+ return DAG.getSetCC(SDLoc(LocReference), VT, ANDNode, LR, Op1);
+ }
+ // fold (and (setgt X, -1), (setgt Y, -1)) -> (setgt (or X, Y), -1)
+ if (Op1 == ISD::SETGT) {
+ SDValue ORNode = DAG.getNode(ISD::OR, SDLoc(N0),
+ LR.getValueType(), LL, RL);
+ AddToWorklist(ORNode.getNode());
+ return DAG.getSetCC(SDLoc(LocReference), VT, ORNode, LR, Op1);
+ }
}
}
// Simplify (and (setne X, 0), (setne X, -1)) -> (setuge (add X, 1), 2)
if (LL == RL && isa<ConstantSDNode>(LR) && isa<ConstantSDNode>(RR) &&
Op0 == Op1 && LL.getValueType().isInteger() &&
- Op0 == ISD::SETNE && ((cast<ConstantSDNode>(LR)->isNullValue() &&
- cast<ConstantSDNode>(RR)->isAllOnesValue()) ||
- (cast<ConstantSDNode>(LR)->isAllOnesValue() &&
- cast<ConstantSDNode>(RR)->isNullValue()))) {
- SDValue ADDNode = DAG.getNode(ISD::ADD, SDLoc(N0), LL.getValueType(),
- LL, DAG.getConstant(1, LL.getValueType()));
+ Op0 == ISD::SETNE && ((isNullConstant(LR) && isAllOnesConstant(RR)) ||
+ (isAllOnesConstant(LR) && isNullConstant(RR)))) {
+ SDLoc DL(N0);
+ SDValue ADDNode = DAG.getNode(ISD::ADD, DL, LL.getValueType(),
+ LL, DAG.getConstant(1, DL,
+ LL.getValueType()));
AddToWorklist(ADDNode.getNode());
return DAG.getSetCC(SDLoc(LocReference), VT, ADDNode,
- DAG.getConstant(2, LL.getValueType()), ISD::SETUGE);
+ DAG.getConstant(2, DL, LL.getValueType()),
+ ISD::SETUGE);
}
// canonicalize equivalent to ll == rl
if (LL == RR && LR == RL) {
if (Result != ISD::SETCC_INVALID &&
(!LegalOperations ||
(TLI.isCondCodeLegal(Result, LL.getSimpleValueType()) &&
- TLI.isOperationLegal(ISD::SETCC,
- getSetCCResultType(N0.getSimpleValueType())))))
- return DAG.getSetCC(SDLoc(LocReference), N0.getValueType(),
- LL, LR, Result);
+ TLI.isOperationLegal(ISD::SETCC, LL.getValueType())))) {
+ EVT CCVT = getSetCCResultType(LL.getValueType());
+ if (N0.getValueType() == CCVT ||
+ (!LegalOperations && N0.getValueType() == MVT::i1))
+ return DAG.getSetCC(SDLoc(LocReference), N0.getValueType(),
+ LL, LR, Result);
+ }
}
}
if (DAG.MaskedValueIsZero(N0.getOperand(1), Mask)) {
ADDC |= Mask;
if (TLI.isLegalAddImmediate(ADDC.getSExtValue())) {
+ SDLoc DL(N0);
SDValue NewAdd =
- DAG.getNode(ISD::ADD, SDLoc(N0), VT,
- N0.getOperand(0), DAG.getConstant(ADDC, VT));
+ DAG.getNode(ISD::ADD, DL, VT,
+ N0.getOperand(0), DAG.getConstant(ADDC, DL, VT));
CombineTo(N0.getNode(), NewAdd);
// Return N so it doesn't get rechecked!
return SDValue(LocReference, 0);
return SDValue();
}
+bool DAGCombiner::isAndLoadExtLoad(ConstantSDNode *AndC, LoadSDNode *LoadN,
+ EVT LoadResultTy, EVT &ExtVT, EVT &LoadedVT,
+ bool &NarrowLoad) {
+ uint32_t ActiveBits = AndC->getAPIntValue().getActiveBits();
+
+ if (ActiveBits == 0 || !APIntOps::isMask(ActiveBits, AndC->getAPIntValue()))
+ return false;
+
+ ExtVT = EVT::getIntegerVT(*DAG.getContext(), ActiveBits);
+ LoadedVT = LoadN->getMemoryVT();
+
+ if (ExtVT == LoadedVT &&
+ (!LegalOperations ||
+ TLI.isLoadExtLegal(ISD::ZEXTLOAD, LoadResultTy, ExtVT))) {
+ // ZEXTLOAD will match without needing to change the size of the value being
+ // loaded.
+ NarrowLoad = false;
+ return true;
+ }
+
+ // Do not change the width of a volatile load.
+ if (LoadN->isVolatile())
+ return false;
+
+ // Do not generate loads of non-round integer types since these can
+ // be expensive (and would be wrong if the type is not byte sized).
+ if (!LoadedVT.bitsGT(ExtVT) || !ExtVT.isRound())
+ return false;
+
+ if (LegalOperations &&
+ !TLI.isLoadExtLegal(ISD::ZEXTLOAD, LoadResultTy, ExtVT))
+ return false;
+
+ if (!TLI.shouldReduceLoadWidth(LoadN, ISD::ZEXTLOAD, ExtVT))
+ return false;
+
+ NarrowLoad = true;
+ return true;
+}
+
SDValue DAGCombiner::visitAND(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
return DAG.getConstant(
APInt::getNullValue(
N0.getValueType().getScalarType().getSizeInBits()),
- N0.getValueType());
+ SDLoc(N), N0.getValueType());
if (ISD::isBuildVectorAllZeros(N1.getNode()))
// do not return N1, because undef node may exist in N1
return DAG.getConstant(
APInt::getNullValue(
N1.getValueType().getScalarType().getSizeInBits()),
- N1.getValueType());
+ SDLoc(N), N1.getValueType());
// fold (and x, -1) -> x, vector edition
if (ISD::isBuildVectorAllOnes(N0.getNode()))
}
// 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)
- return DAG.FoldConstantArithmetic(ISD::AND, VT, N0C, N1C);
+ if (N0C && N1C && !N1C->isOpaque())
+ return DAG.FoldConstantArithmetic(ISD::AND, SDLoc(N), VT, N0C, N1C);
// canonicalize constant to RHS
if (isConstantIntBuildVectorOrConstantInt(N0) &&
!isConstantIntBuildVectorOrConstantInt(N1))
return DAG.getNode(ISD::AND, SDLoc(N), VT, N1, N0);
// fold (and x, -1) -> x
- if (N1C && N1C->isAllOnesValue())
+ if (isAllOnesConstant(N1))
return N0;
// if (and x, c) is known to be zero, return 0
unsigned BitWidth = VT.getScalarType().getSizeInBits();
if (N1C && DAG.MaskedValueIsZero(SDValue(N, 0),
APInt::getAllOnesValue(BitWidth)))
- return DAG.getConstant(0, VT);
+ return DAG.getConstant(0, SDLoc(N), VT);
// reassociate and
if (SDValue RAND = ReassociateOps(ISD::AND, SDLoc(N), N0, N1))
return RAND;
: cast<LoadSDNode>(N0);
if (LN0->getExtensionType() != ISD::SEXTLOAD &&
LN0->isUnindexed() && N0.hasOneUse() && SDValue(LN0, 0).hasOneUse()) {
- uint32_t ActiveBits = N1C->getAPIntValue().getActiveBits();
- if (ActiveBits > 0 && APIntOps::isMask(ActiveBits, N1C->getAPIntValue())){
- EVT ExtVT = EVT::getIntegerVT(*DAG.getContext(), ActiveBits);
- EVT LoadedVT = LN0->getMemoryVT();
- EVT LoadResultTy = HasAnyExt ? LN0->getValueType(0) : VT;
-
- if (ExtVT == LoadedVT &&
- (!LegalOperations || TLI.isLoadExtLegal(ISD::ZEXTLOAD, LoadResultTy,
- ExtVT))) {
-
+ auto NarrowLoad = false;
+ EVT LoadResultTy = HasAnyExt ? LN0->getValueType(0) : VT;
+ EVT ExtVT, LoadedVT;
+ if (isAndLoadExtLoad(N1C, LN0, LoadResultTy, ExtVT, LoadedVT,
+ NarrowLoad)) {
+ if (!NarrowLoad) {
SDValue NewLoad =
DAG.getExtLoad(ISD::ZEXTLOAD, SDLoc(LN0), LoadResultTy,
LN0->getChain(), LN0->getBasePtr(), ExtVT,
AddToWorklist(N);
CombineTo(LN0, NewLoad, NewLoad.getValue(1));
return SDValue(N, 0); // Return N so it doesn't get rechecked!
- }
-
- // Do not change the width of a volatile load.
- // Do not generate loads of non-round integer types since these can
- // be expensive (and would be wrong if the type is not byte sized).
- if (!LN0->isVolatile() && LoadedVT.bitsGT(ExtVT) && ExtVT.isRound() &&
- (!LegalOperations || TLI.isLoadExtLegal(ISD::ZEXTLOAD, LoadResultTy,
- ExtVT))) {
+ } else {
EVT PtrType = LN0->getOperand(1).getValueType();
unsigned Alignment = LN0->getAlignment();
// For big endian targets, we need to add an offset to the pointer
// to load the correct bytes. For little endian systems, we merely
// need to read fewer bytes from the same pointer.
- if (TLI.isBigEndian()) {
+ if (DAG.getDataLayout().isBigEndian()) {
unsigned LVTStoreBytes = LoadedVT.getStoreSize();
unsigned EVTStoreBytes = ExtVT.getStoreSize();
unsigned PtrOff = LVTStoreBytes - EVTStoreBytes;
- NewPtr = DAG.getNode(ISD::ADD, SDLoc(LN0), PtrType,
- NewPtr, DAG.getConstant(PtrOff, PtrType));
+ SDLoc DL(LN0);
+ NewPtr = DAG.getNode(ISD::ADD, DL, PtrType,
+ NewPtr, DAG.getConstant(PtrOff, DL, PtrType));
Alignment = MinAlign(Alignment, PtrOff);
}
return Combined;
// Simplify: (and (op x...), (op y...)) -> (op (and x, y))
- if (N0.getOpcode() == N1.getOpcode()) {
- SDValue Tmp = SimplifyBinOpWithSameOpcodeHands(N);
- if (Tmp.getNode()) return Tmp;
- }
+ if (N0.getOpcode() == N1.getOpcode())
+ if (SDValue Tmp = SimplifyBinOpWithSameOpcodeHands(N))
+ return Tmp;
// fold (and (sign_extend_inreg x, i16 to i32), 1) -> (and x, 1)
// fold (and (sra)) -> (and (srl)) when possible.
}
SDValue Res = DAG.getNode(ISD::BSWAP, SDLoc(N), VT, N00);
- if (OpSizeInBits > 16)
- Res = DAG.getNode(ISD::SRL, SDLoc(N), VT, Res,
- DAG.getConstant(OpSizeInBits-16, getShiftAmountTy(VT)));
+ if (OpSizeInBits > 16) {
+ SDLoc DL(N);
+ Res = DAG.getNode(ISD::SRL, DL, VT, Res,
+ DAG.getConstant(OpSizeInBits - 16, DL,
+ getShiftAmountTy(VT)));
+ }
return Res;
}
if (Parts[0] != Parts[1] || Parts[0] != Parts[2] || Parts[0] != Parts[3])
return SDValue();
- SDValue BSwap = DAG.getNode(ISD::BSWAP, SDLoc(N), VT,
- SDValue(Parts[0],0));
+ SDLoc DL(N);
+ SDValue BSwap = DAG.getNode(ISD::BSWAP, DL, VT,
+ SDValue(Parts[0], 0));
// Result of the bswap should be rotated by 16. If it's not legal, then
// do (x << 16) | (x >> 16).
- SDValue ShAmt = DAG.getConstant(16, getShiftAmountTy(VT));
+ SDValue ShAmt = DAG.getConstant(16, DL, getShiftAmountTy(VT));
if (TLI.isOperationLegalOrCustom(ISD::ROTL, VT))
- return DAG.getNode(ISD::ROTL, SDLoc(N), VT, BSwap, ShAmt);
+ return DAG.getNode(ISD::ROTL, DL, VT, BSwap, ShAmt);
if (TLI.isOperationLegalOrCustom(ISD::ROTR, VT))
- return DAG.getNode(ISD::ROTR, SDLoc(N), VT, BSwap, ShAmt);
- return DAG.getNode(ISD::OR, SDLoc(N), VT,
- DAG.getNode(ISD::SHL, SDLoc(N), VT, BSwap, ShAmt),
- DAG.getNode(ISD::SRL, SDLoc(N), VT, BSwap, ShAmt));
+ return DAG.getNode(ISD::ROTR, DL, VT, BSwap, ShAmt);
+ return DAG.getNode(ISD::OR, DL, VT,
+ DAG.getNode(ISD::SHL, DL, VT, BSwap, ShAmt),
+ DAG.getNode(ISD::SRL, DL, VT, BSwap, ShAmt));
}
/// This contains all DAGCombine rules which reduce two values combined by
if (!LegalOperations &&
(N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)) {
EVT EltVT = VT.isVector() ? VT.getVectorElementType() : VT;
- return DAG.getConstant(APInt::getAllOnesValue(EltVT.getSizeInBits()), VT);
+ return DAG.getConstant(APInt::getAllOnesValue(EltVT.getSizeInBits()),
+ SDLoc(LocReference), VT);
}
// fold (or (setcc x), (setcc y)) -> (setcc (or x, y))
SDValue LL, LR, RL, RR, CC0, CC1;
ISD::CondCode Op0 = cast<CondCodeSDNode>(CC0)->get();
ISD::CondCode Op1 = cast<CondCodeSDNode>(CC1)->get();
- if (LR == RR && isa<ConstantSDNode>(LR) && Op0 == Op1 &&
- LL.getValueType().isInteger()) {
+ if (LR == RR && Op0 == Op1 && LL.getValueType().isInteger()) {
// fold (or (setne X, 0), (setne Y, 0)) -> (setne (or X, Y), 0)
// fold (or (setlt X, 0), (setlt Y, 0)) -> (setne (or X, Y), 0)
- if (cast<ConstantSDNode>(LR)->isNullValue() &&
- (Op1 == ISD::SETNE || Op1 == ISD::SETLT)) {
+ if (isNullConstant(LR) && (Op1 == ISD::SETNE || Op1 == ISD::SETLT)) {
SDValue ORNode = DAG.getNode(ISD::OR, SDLoc(LR),
LR.getValueType(), LL, RL);
AddToWorklist(ORNode.getNode());
}
// fold (or (setne X, -1), (setne Y, -1)) -> (setne (and X, Y), -1)
// fold (or (setgt X, -1), (setgt Y -1)) -> (setgt (and X, Y), -1)
- if (cast<ConstantSDNode>(LR)->isAllOnesValue() &&
- (Op1 == ISD::SETNE || Op1 == ISD::SETGT)) {
+ if (isAllOnesConstant(LR) && (Op1 == ISD::SETNE || Op1 == ISD::SETGT)) {
SDValue ANDNode = DAG.getNode(ISD::AND, SDLoc(LR),
LR.getValueType(), LL, RL);
AddToWorklist(ANDNode.getNode());
if (Result != ISD::SETCC_INVALID &&
(!LegalOperations ||
(TLI.isCondCodeLegal(Result, LL.getSimpleValueType()) &&
- TLI.isOperationLegal(ISD::SETCC,
- getSetCCResultType(N0.getValueType())))))
- return DAG.getSetCC(SDLoc(LocReference), N0.getValueType(),
- LL, LR, Result);
+ TLI.isOperationLegal(ISD::SETCC, LL.getValueType())))) {
+ EVT CCVT = getSetCCResultType(LL.getValueType());
+ if (N0.getValueType() == CCVT ||
+ (!LegalOperations && N0.getValueType() == MVT::i1))
+ return DAG.getSetCC(SDLoc(LocReference), N0.getValueType(),
+ LL, LR, Result);
+ }
}
}
// (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));
- return DAG.getNode(ISD::AND, SDLoc(LocReference), VT, X,
- DAG.getConstant(LHSMask | RHSMask, 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));
+ }
+ }
}
}
return DAG.getConstant(
APInt::getAllOnesValue(
N0.getValueType().getScalarType().getSizeInBits()),
- N0.getValueType());
+ SDLoc(N), N0.getValueType());
if (ISD::isBuildVectorAllOnes(N1.getNode()))
// do not return N1, because undef node may exist in N1
return DAG.getConstant(
APInt::getAllOnesValue(
N1.getValueType().getScalarType().getSizeInBits()),
- N1.getValueType());
+ SDLoc(N), N1.getValueType());
// fold (or (shuf A, V_0, MA), (shuf B, V_0, MB)) -> (shuf A, B, Mask1)
// fold (or (shuf A, V_0, MA), (shuf B, V_0, MB)) -> (shuf B, A, Mask2)
}
// 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)
- return DAG.FoldConstantArithmetic(ISD::OR, VT, N0C, N1C);
+ if (N0C && N1C && !N1C->isOpaque())
+ return DAG.FoldConstantArithmetic(ISD::OR, SDLoc(N), VT, N0C, N1C);
// canonicalize constant to RHS
if (isConstantIntBuildVectorOrConstantInt(N0) &&
!isConstantIntBuildVectorOrConstantInt(N1))
return DAG.getNode(ISD::OR, SDLoc(N), VT, N1, N0);
// fold (or x, 0) -> x
- if (N1C && N1C->isNullValue())
+ if (isNullConstant(N1))
return N0;
// fold (or x, -1) -> -1
- if (N1C && N1C->isAllOnesValue())
+ if (isAllOnesConstant(N1))
return N1;
// fold (or x, c) -> c iff (x & ~c) == 0
if (N1C && DAG.MaskedValueIsZero(N0, ~N1C->getAPIntValue()))
return Combined;
// Recognize halfword bswaps as (bswap + rotl 16) or (bswap + shl 16)
- SDValue BSwap = MatchBSwapHWord(N, N0, N1);
- if (BSwap.getNode())
+ if (SDValue BSwap = MatchBSwapHWord(N, N0, N1))
return BSwap;
- BSwap = MatchBSwapHWordLow(N, N0, N1);
- if (BSwap.getNode())
+ if (SDValue BSwap = MatchBSwapHWordLow(N, N0, N1))
return BSwap;
// reassociate or
isa<ConstantSDNode>(N0.getOperand(1))) {
ConstantSDNode *C1 = cast<ConstantSDNode>(N0.getOperand(1));
if ((C1->getAPIntValue() & N1C->getAPIntValue()) != 0) {
- if (SDValue COR = DAG.FoldConstantArithmetic(ISD::OR, VT, N1C, C1))
+ if (SDValue COR = DAG.FoldConstantArithmetic(ISD::OR, SDLoc(N1), VT,
+ N1C, C1))
return DAG.getNode(
ISD::AND, SDLoc(N), VT,
DAG.getNode(ISD::OR, SDLoc(N0), VT, N0.getOperand(0), N1), COR);
}
}
// Simplify: (or (op x...), (op y...)) -> (op (or x, y))
- if (N0.getOpcode() == N1.getOpcode()) {
- SDValue Tmp = SimplifyBinOpWithSameOpcodeHands(N);
- if (Tmp.getNode()) return Tmp;
- }
+ if (N0.getOpcode() == N1.getOpcode())
+ if (SDValue Tmp = SimplifyBinOpWithSameOpcodeHands(N))
+ return Tmp;
// See if this is some rotate idiom.
if (SDNode *Rot = MatchRotate(N0, N1, SDLoc(N)))
/// Match "(X shl/srl V1) & V2" where V2 may not be present.
static bool MatchRotateHalf(SDValue Op, SDValue &Shift, SDValue &Mask) {
if (Op.getOpcode() == ISD::AND) {
- if (isa<ConstantSDNode>(Op.getOperand(1))) {
+ if (isConstantIntBuildVectorOrConstantInt(Op.getOperand(1))) {
Mask = Op.getOperand(1);
Op = Op.getOperand(0);
} else {
}
// Return true if we can prove that, whenever Neg and Pos are both in the
-// range [0, OpSize), Neg == (Pos == 0 ? 0 : OpSize - Pos). This means that
+// range [0, EltSize), Neg == (Pos == 0 ? 0 : EltSize - Pos). This means that
// for two opposing shifts shift1 and shift2 and a value X with OpBits bits:
//
// (or (shift1 X, Neg), (shift2 X, Pos))
//
// reduces to a rotate in direction shift2 by Pos or (equivalently) a rotate
-// in direction shift1 by Neg. The range [0, OpSize) means that we only need
+// in direction shift1 by Neg. The range [0, EltSize) means that we only need
// to consider shift amounts with defined behavior.
-static bool matchRotateSub(SDValue Pos, SDValue Neg, unsigned OpSize) {
- // If OpSize is a power of 2 then:
+static bool matchRotateSub(SDValue Pos, SDValue Neg, unsigned EltSize) {
+ // If EltSize is a power of 2 then:
//
- // (a) (Pos == 0 ? 0 : OpSize - Pos) == (OpSize - Pos) & (OpSize - 1)
- // (b) Neg == Neg & (OpSize - 1) whenever Neg is in [0, OpSize).
+ // (a) (Pos == 0 ? 0 : EltSize - Pos) == (EltSize - Pos) & (EltSize - 1)
+ // (b) Neg == Neg & (EltSize - 1) whenever Neg is in [0, EltSize).
//
- // So if OpSize is a power of 2 and Neg is (and Neg', OpSize-1), we check
+ // So if EltSize is a power of 2 and Neg is (and Neg', EltSize-1), we check
// for the stronger condition:
//
- // Neg & (OpSize - 1) == (OpSize - Pos) & (OpSize - 1) [A]
+ // Neg & (EltSize - 1) == (EltSize - Pos) & (EltSize - 1) [A]
//
- // for all Neg and Pos. Since Neg & (OpSize - 1) == Neg' & (OpSize - 1)
+ // for all Neg and Pos. Since Neg & (EltSize - 1) == Neg' & (EltSize - 1)
// we can just replace Neg with Neg' for the rest of the function.
//
// In other cases we check for the even stronger condition:
//
- // Neg == OpSize - Pos [B]
+ // Neg == EltSize - Pos [B]
//
// for all Neg and Pos. Note that the (or ...) then invokes undefined
- // behavior if Pos == 0 (and consequently Neg == OpSize).
+ // behavior if Pos == 0 (and consequently Neg == EltSize).
//
- // We could actually use [A] whenever OpSize is a power of 2, but the
+ // We could actually use [A] whenever EltSize is a power of 2, but the
// only extra cases that it would match are those uninteresting ones
// where Neg and Pos are never in range at the same time. E.g. for
- // OpSize == 32, using [A] would allow a Neg of the form (sub 64, Pos)
+ // EltSize == 32, using [A] would allow a Neg of the form (sub 64, Pos)
// as well as (sub 32, Pos), but:
//
// (or (shift1 X, (sub 64, Pos)), (shift2 X, Pos))
//
// always invokes undefined behavior for 32-bit X.
//
- // Below, Mask == OpSize - 1 when using [A] and is all-ones otherwise.
+ // Below, Mask == EltSize - 1 when using [A] and is all-ones otherwise.
unsigned MaskLoBits = 0;
- if (Neg.getOpcode() == ISD::AND &&
- isPowerOf2_64(OpSize) &&
- Neg.getOperand(1).getOpcode() == ISD::Constant &&
- cast<ConstantSDNode>(Neg.getOperand(1))->getAPIntValue() == OpSize - 1) {
- Neg = Neg.getOperand(0);
- MaskLoBits = Log2_64(OpSize);
+ if (Neg.getOpcode() == ISD::AND && isPowerOf2_64(EltSize)) {
+ if (ConstantSDNode *NegC = isConstOrConstSplat(Neg.getOperand(1))) {
+ if (NegC->getAPIntValue() == EltSize - 1) {
+ Neg = Neg.getOperand(0);
+ MaskLoBits = Log2_64(EltSize);
+ }
+ }
}
// Check whether Neg has the form (sub NegC, NegOp1) for some NegC and NegOp1.
if (Neg.getOpcode() != ISD::SUB)
- return 0;
- ConstantSDNode *NegC = dyn_cast<ConstantSDNode>(Neg.getOperand(0));
+ return false;
+ ConstantSDNode *NegC = isConstOrConstSplat(Neg.getOperand(0));
if (!NegC)
- return 0;
+ return false;
SDValue NegOp1 = Neg.getOperand(1);
- // On the RHS of [A], if Pos is Pos' & (OpSize - 1), just replace Pos with
+ // On the RHS of [A], if Pos is Pos' & (EltSize - 1), just replace Pos with
// Pos'. The truncation is redundant for the purpose of the equality.
- if (MaskLoBits &&
- Pos.getOpcode() == ISD::AND &&
- Pos.getOperand(1).getOpcode() == ISD::Constant &&
- cast<ConstantSDNode>(Pos.getOperand(1))->getAPIntValue() == OpSize - 1)
- Pos = Pos.getOperand(0);
+ if (MaskLoBits && Pos.getOpcode() == ISD::AND)
+ if (ConstantSDNode *PosC = isConstOrConstSplat(Pos.getOperand(1)))
+ if (PosC->getAPIntValue() == EltSize - 1)
+ Pos = Pos.getOperand(0);
// The condition we need is now:
//
- // (NegC - NegOp1) & Mask == (OpSize - Pos) & Mask
+ // (NegC - NegOp1) & Mask == (EltSize - Pos) & Mask
//
// If NegOp1 == Pos then we need:
//
- // OpSize & Mask == NegC & Mask
+ // EltSize & Mask == NegC & Mask
//
// (because "x & Mask" is a truncation and distributes through subtraction).
APInt Width;
if (Pos == NegOp1)
Width = NegC->getAPIntValue();
+
// Check for cases where Pos has the form (add NegOp1, PosC) for some PosC.
// Then the condition we want to prove becomes:
//
- // (NegC - NegOp1) & Mask == (OpSize - (NegOp1 + PosC)) & Mask
+ // (NegC - NegOp1) & Mask == (EltSize - (NegOp1 + PosC)) & Mask
//
// which, again because "x & Mask" is a truncation, becomes:
//
- // NegC & Mask == (OpSize - PosC) & Mask
- // OpSize & Mask == (NegC + PosC) & Mask
- else if (Pos.getOpcode() == ISD::ADD &&
- Pos.getOperand(0) == NegOp1 &&
- Pos.getOperand(1).getOpcode() == ISD::Constant)
- Width = (cast<ConstantSDNode>(Pos.getOperand(1))->getAPIntValue() +
- NegC->getAPIntValue());
- else
+ // NegC & Mask == (EltSize - PosC) & Mask
+ // EltSize & Mask == (NegC + PosC) & Mask
+ else if (Pos.getOpcode() == ISD::ADD && Pos.getOperand(0) == NegOp1) {
+ if (ConstantSDNode *PosC = isConstOrConstSplat(Pos.getOperand(1)))
+ Width = PosC->getAPIntValue() + NegC->getAPIntValue();
+ else
+ return false;
+ } else
return false;
- // Now we just need to check that OpSize & Mask == Width & Mask.
+ // Now we just need to check that EltSize & Mask == Width & Mask.
if (MaskLoBits)
- // Opsize & Mask is 0 since Mask is Opsize - 1.
+ // EltSize & Mask is 0 since Mask is EltSize - 1.
return Width.getLoBits(MaskLoBits) == 0;
- return Width == OpSize;
+ return Width == EltSize;
}
// A subroutine of MatchRotate used once we have found an OR of two opposite
// (srl x, (*ext y))) ->
// (rotr x, y) or (rotl x, (sub 32, y))
EVT VT = Shifted.getValueType();
- if (matchRotateSub(InnerPos, InnerNeg, VT.getSizeInBits())) {
+ if (matchRotateSub(InnerPos, InnerNeg, VT.getScalarSizeInBits())) {
bool HasPos = TLI.isOperationLegalOrCustom(PosOpcode, VT);
return DAG.getNode(HasPos ? PosOpcode : NegOpcode, DL, VT, Shifted,
HasPos ? Pos : Neg).getNode();
if (RHSShift.getOpcode() == ISD::SHL) {
std::swap(LHS, RHS);
std::swap(LHSShift, RHSShift);
- std::swap(LHSMask , RHSMask );
+ std::swap(LHSMask, RHSMask);
}
- unsigned OpSizeInBits = VT.getSizeInBits();
+ unsigned EltSizeInBits = VT.getScalarSizeInBits();
SDValue LHSShiftArg = LHSShift.getOperand(0);
SDValue LHSShiftAmt = LHSShift.getOperand(1);
SDValue RHSShiftArg = RHSShift.getOperand(0);
// fold (or (shl x, C1), (srl x, C2)) -> (rotl x, C1)
// fold (or (shl x, C1), (srl x, C2)) -> (rotr x, C2)
- if (LHSShiftAmt.getOpcode() == ISD::Constant &&
- RHSShiftAmt.getOpcode() == ISD::Constant) {
- uint64_t LShVal = cast<ConstantSDNode>(LHSShiftAmt)->getZExtValue();
- uint64_t RShVal = cast<ConstantSDNode>(RHSShiftAmt)->getZExtValue();
- if ((LShVal + RShVal) != OpSizeInBits)
+ if (isConstOrConstSplat(LHSShiftAmt) && isConstOrConstSplat(RHSShiftAmt)) {
+ uint64_t LShVal = isConstOrConstSplat(LHSShiftAmt)->getZExtValue();
+ uint64_t RShVal = isConstOrConstSplat(RHSShiftAmt)->getZExtValue();
+ if ((LShVal + RShVal) != EltSizeInBits)
return nullptr;
SDValue Rot = DAG.getNode(HasROTL ? ISD::ROTL : ISD::ROTR, DL, VT,
// If there is an AND of either shifted operand, apply it to the result.
if (LHSMask.getNode() || RHSMask.getNode()) {
- APInt Mask = APInt::getAllOnesValue(OpSizeInBits);
+ APInt AllBits = APInt::getAllOnesValue(EltSizeInBits);
+ SDValue Mask = DAG.getConstant(AllBits, DL, VT);
if (LHSMask.getNode()) {
- APInt RHSBits = APInt::getLowBitsSet(OpSizeInBits, LShVal);
- Mask &= cast<ConstantSDNode>(LHSMask)->getAPIntValue() | RHSBits;
+ APInt RHSBits = APInt::getLowBitsSet(EltSizeInBits, LShVal);
+ Mask = DAG.getNode(ISD::AND, DL, VT, Mask,
+ DAG.getNode(ISD::OR, DL, VT, LHSMask,
+ DAG.getConstant(RHSBits, DL, VT)));
}
if (RHSMask.getNode()) {
- APInt LHSBits = APInt::getHighBitsSet(OpSizeInBits, RShVal);
- Mask &= cast<ConstantSDNode>(RHSMask)->getAPIntValue() | LHSBits;
+ APInt LHSBits = APInt::getHighBitsSet(EltSizeInBits, RShVal);
+ Mask = DAG.getNode(ISD::AND, DL, VT, Mask,
+ DAG.getNode(ISD::OR, DL, VT, RHSMask,
+ DAG.getConstant(LHSBits, DL, VT)));
}
- Rot = DAG.getNode(ISD::AND, DL, VT, Rot, DAG.getConstant(Mask, VT));
+ Rot = DAG.getNode(ISD::AND, DL, VT, Rot, Mask);
}
return Rot.getNode();
// fold (xor undef, undef) -> 0. This is a common idiom (misuse).
if (N0.getOpcode() == ISD::UNDEF && N1.getOpcode() == ISD::UNDEF)
- return DAG.getConstant(0, VT);
+ return DAG.getConstant(0, SDLoc(N), VT);
// fold (xor x, undef) -> undef
if (N0.getOpcode() == ISD::UNDEF)
return 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, VT, N0C, N1C);
+ return DAG.FoldConstantArithmetic(ISD::XOR, SDLoc(N), VT, N0C, N1C);
// canonicalize constant to RHS
if (isConstantIntBuildVectorOrConstantInt(N0) &&
!isConstantIntBuildVectorOrConstantInt(N1))
return DAG.getNode(ISD::XOR, SDLoc(N), VT, N1, N0);
// fold (xor x, 0) -> x
- if (N1C && N1C->isNullValue())
+ if (isNullConstant(N1))
return N0;
// reassociate xor
if (SDValue RXOR = ReassociateOps(ISD::XOR, SDLoc(N), N0, N1))
}
// fold (not (zext (setcc x, y))) -> (zext (not (setcc x, y)))
- if (N1C && N1C->getAPIntValue() == 1 && N0.getOpcode() == ISD::ZERO_EXTEND &&
+ if (isOneConstant(N1) && N0.getOpcode() == ISD::ZERO_EXTEND &&
N0.getNode()->hasOneUse() &&
isSetCCEquivalent(N0.getOperand(0), LHS, RHS, CC)){
SDValue V = N0.getOperand(0);
- V = DAG.getNode(ISD::XOR, SDLoc(N0), V.getValueType(), V,
- DAG.getConstant(1, V.getValueType()));
+ SDLoc DL(N0);
+ V = DAG.getNode(ISD::XOR, DL, V.getValueType(), V,
+ DAG.getConstant(1, DL, V.getValueType()));
AddToWorklist(V.getNode());
return DAG.getNode(ISD::ZERO_EXTEND, SDLoc(N), VT, V);
}
// fold (not (or x, y)) -> (and (not x), (not y)) iff x or y are setcc
- if (N1C && N1C->getAPIntValue() == 1 && VT == MVT::i1 &&
+ if (isOneConstant(N1) && VT == MVT::i1 &&
(N0.getOpcode() == ISD::OR || N0.getOpcode() == ISD::AND)) {
SDValue LHS = N0.getOperand(0), RHS = N0.getOperand(1);
if (isOneUseSetCC(RHS) || isOneUseSetCC(LHS)) {
}
}
// fold (not (or x, y)) -> (and (not x), (not y)) iff x or y are constants
- if (N1C && N1C->isAllOnesValue() &&
+ if (isAllOnesConstant(N1) &&
(N0.getOpcode() == ISD::OR || N0.getOpcode() == ISD::AND)) {
SDValue LHS = N0.getOperand(0), RHS = N0.getOperand(1);
if (isa<ConstantSDNode>(RHS) || isa<ConstantSDNode>(LHS)) {
}
// 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)
- return DAG.getNode(ISD::XOR, SDLoc(N), VT, N0.getOperand(1),
+ 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(), VT));
- if (N01C)
- return DAG.getNode(ISD::XOR, SDLoc(N), VT, N0.getOperand(0),
+ N00C->getAPIntValue(), DL, VT));
+ }
+ 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() ^
- N01C->getAPIntValue(), VT));
+ N01C->getAPIntValue(), DL, VT));
+ }
}
// fold (xor x, x) -> 0
if (N0 == N1)
// consistent result.
// - Pushing the zero left requires shifting one bits in from the right.
// A rotate left of ~1 is a nice way of achieving the desired result.
- if (TLI.isOperationLegalOrCustom(ISD::ROTL, VT))
- if (auto *N1C = dyn_cast<ConstantSDNode>(N1.getNode()))
- if (N0.getOpcode() == ISD::SHL)
- if (auto *ShlLHS = dyn_cast<ConstantSDNode>(N0.getOperand(0)))
- if (N1C->isAllOnesValue() && ShlLHS->isOne())
- return DAG.getNode(ISD::ROTL, SDLoc(N), VT, DAG.getConstant(~1, VT),
- N0.getOperand(1));
+ if (TLI.isOperationLegalOrCustom(ISD::ROTL, VT) && N0.getOpcode() == ISD::SHL
+ && isAllOnesConstant(N1) && isOneConstant(N0.getOperand(0))) {
+ SDLoc DL(N);
+ return DAG.getNode(ISD::ROTL, DL, VT, DAG.getConstant(~1, DL, VT),
+ N0.getOperand(1));
+ }
// Simplify: xor (op x...), (op y...) -> (op (xor x, y))
- if (N0.getOpcode() == N1.getOpcode()) {
- SDValue Tmp = SimplifyBinOpWithSameOpcodeHands(N);
- if (Tmp.getNode()) return Tmp;
- }
+ if (N0.getOpcode() == N1.getOpcode())
+ if (SDValue Tmp = SimplifyBinOpWithSameOpcodeHands(N))
+ return Tmp;
// Simplify the expression using non-local knowledge.
if (!VT.isVector() &&
/// 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());
+ 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, SDLoc(N), TruncVT,
- DAG.getNode(ISD::TRUNCATE, SDLoc(N), TruncVT, N00),
- DAG.getConstant(TruncC, TruncVT));
+ return DAG.getNode(ISD::AND, DL, TruncVT,
+ DAG.getNode(ISD::TRUNCATE, DL, TruncVT, N00),
+ DAG.getConstant(TruncC, DL, TruncVT));
+ }
}
}
if (N01CV && N01CV->isConstant() && N00.getOpcode() == ISD::SETCC &&
TLI.getBooleanContents(N00.getOperand(0).getValueType()) ==
TargetLowering::ZeroOrNegativeOneBooleanContent) {
- if (SDValue C = DAG.FoldConstantArithmetic(ISD::SHL, VT, N01CV, N1CV))
+ if (SDValue C = DAG.FoldConstantArithmetic(ISD::SHL, SDLoc(N), VT,
+ N01CV, N1CV))
return DAG.getNode(ISD::AND, SDLoc(N), VT, N00, C);
}
} else {
}
// fold (shl c1, c2) -> c1<<c2
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- if (N0C && N1C)
- return DAG.FoldConstantArithmetic(ISD::SHL, VT, 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 (N0C && N0C->isNullValue())
+ if (isNullConstant(N0))
return N0;
// fold (shl x, c >= size(x)) -> undef
- if (N1C && N1C->getZExtValue() >= OpSizeInBits)
+ if (N1C && N1C->getAPIntValue().uge(OpSizeInBits))
return DAG.getUNDEF(VT);
// fold (shl x, 0) -> x
if (N1C && N1C->isNullValue())
return N0;
// fold (shl undef, x) -> 0
if (N0.getOpcode() == ISD::UNDEF)
- return DAG.getConstant(0, VT);
+ return DAG.getConstant(0, SDLoc(N), VT);
// if (shl x, c) is known to be zero, return 0
if (DAG.MaskedValueIsZero(SDValue(N, 0),
APInt::getAllOnesValue(OpSizeInBits)))
- return DAG.getConstant(0, VT);
+ return DAG.getConstant(0, SDLoc(N), VT);
// fold (shl x, (trunc (and y, c))) -> (shl x, (and (trunc y), (trunc c))).
if (N1.getOpcode() == ISD::TRUNCATE &&
N1.getOperand(0).getOpcode() == ISD::AND) {
if (ConstantSDNode *N0C1 = isConstOrConstSplat(N0.getOperand(1))) {
uint64_t c1 = N0C1->getZExtValue();
uint64_t c2 = N1C->getZExtValue();
+ SDLoc DL(N);
if (c1 + c2 >= OpSizeInBits)
- return DAG.getConstant(0, VT);
- return DAG.getNode(ISD::SHL, SDLoc(N), VT, N0.getOperand(0),
- DAG.getConstant(c1 + c2, N1.getValueType()));
+ return DAG.getConstant(0, DL, VT);
+ return DAG.getNode(ISD::SHL, DL, VT, N0.getOperand(0),
+ DAG.getConstant(c1 + c2, DL, N1.getValueType()));
}
}
EVT InnerShiftVT = N0Op0.getValueType();
uint64_t InnerShiftSize = InnerShiftVT.getScalarSizeInBits();
if (c2 >= OpSizeInBits - InnerShiftSize) {
+ SDLoc DL(N0);
if (c1 + c2 >= OpSizeInBits)
- return DAG.getConstant(0, VT);
- return DAG.getNode(ISD::SHL, SDLoc(N0), VT,
- DAG.getNode(N0.getOpcode(), SDLoc(N0), VT,
+ return DAG.getConstant(0, DL, VT);
+ return DAG.getNode(ISD::SHL, DL, VT,
+ DAG.getNode(N0.getOpcode(), DL, VT,
N0Op0->getOperand(0)),
- DAG.getConstant(c1 + c2, N1.getValueType()));
+ DAG.getConstant(c1 + c2, DL, N1.getValueType()));
}
}
}
if (c1 == c2) {
SDValue NewOp0 = N0.getOperand(0);
EVT CountVT = NewOp0.getOperand(1).getValueType();
- SDValue NewSHL = DAG.getNode(ISD::SHL, SDLoc(N), NewOp0.getValueType(),
- NewOp0, DAG.getConstant(c2, CountVT));
+ SDLoc DL(N);
+ SDValue NewSHL = DAG.getNode(ISD::SHL, DL, NewOp0.getValueType(),
+ NewOp0,
+ DAG.getConstant(c2, DL, CountVT));
AddToWorklist(NewSHL.getNode());
return DAG.getNode(ISD::ZERO_EXTEND, SDLoc(N0), VT, NewSHL);
}
}
}
+ // fold (shl (sr[la] exact X, C1), C2) -> (shl X, (C2-C1)) if C1 <= C2
+ // fold (shl (sr[la] exact X, C1), C2) -> (sr[la] X, (C2-C1)) if C1 > C2
+ if (N1C && (N0.getOpcode() == ISD::SRL || N0.getOpcode() == ISD::SRA) &&
+ cast<BinaryWithFlagsSDNode>(N0)->Flags.hasExact()) {
+ if (ConstantSDNode *N0C1 = isConstOrConstSplat(N0.getOperand(1))) {
+ uint64_t C1 = N0C1->getZExtValue();
+ uint64_t C2 = N1C->getZExtValue();
+ SDLoc DL(N);
+ if (C1 <= C2)
+ return DAG.getNode(ISD::SHL, DL, VT, N0.getOperand(0),
+ DAG.getConstant(C2 - C1, DL, N1.getValueType()));
+ return DAG.getNode(N0.getOpcode(), DL, VT, N0.getOperand(0),
+ DAG.getConstant(C1 - C2, DL, N1.getValueType()));
+ }
+ }
+
// fold (shl (srl x, c1), c2) -> (and (shl x, (sub c2, c1), MASK) or
// (and (srl x, (sub c1, c2), MASK)
// Only fold this if the inner shift has no other uses -- if it does, folding
SDValue Shift;
if (c2 > c1) {
Mask = Mask.shl(c2 - c1);
- Shift = DAG.getNode(ISD::SHL, SDLoc(N), VT, N0.getOperand(0),
- DAG.getConstant(c2 - c1, N1.getValueType()));
+ SDLoc DL(N);
+ Shift = DAG.getNode(ISD::SHL, DL, VT, N0.getOperand(0),
+ DAG.getConstant(c2 - c1, DL, N1.getValueType()));
} else {
Mask = Mask.lshr(c1 - c2);
- Shift = DAG.getNode(ISD::SRL, SDLoc(N), VT, N0.getOperand(0),
- DAG.getConstant(c1 - c2, N1.getValueType()));
+ SDLoc DL(N);
+ Shift = DAG.getNode(ISD::SRL, DL, VT, N0.getOperand(0),
+ DAG.getConstant(c1 - c2, DL, N1.getValueType()));
}
- return DAG.getNode(ISD::AND, SDLoc(N0), VT, Shift,
- DAG.getConstant(Mask, VT));
+ SDLoc DL(N0);
+ return DAG.getNode(ISD::AND, DL, VT, Shift,
+ DAG.getConstant(Mask, DL, VT));
}
}
}
// fold (shl (sra x, c1), c1) -> (and x, (shl -1, c1))
if (N1C && N0.getOpcode() == ISD::SRA && N1 == N0.getOperand(1)) {
unsigned BitSize = VT.getScalarSizeInBits();
+ SDLoc DL(N);
SDValue HiBitsMask =
DAG.getConstant(APInt::getHighBitsSet(BitSize,
- BitSize - N1C->getZExtValue()), VT);
- return DAG.getNode(ISD::AND, SDLoc(N), VT, N0.getOperand(0),
+ BitSize - N1C->getZExtValue()),
+ DL, VT);
+ return DAG.getNode(ISD::AND, DL, VT, N0.getOperand(0),
HiBitsMask);
}
return DAG.getNode(ISD::ADD, SDLoc(N), VT, Shl0, Shl1);
}
- if (N1C) {
- SDValue NewSHL = visitShiftByConstant(N, N1C);
- if (NewSHL.getNode())
- return NewSHL;
+ // fold (shl (mul x, c1), c2) -> (mul x, c1 << c2)
+ if (N1C && N0.getOpcode() == ISD::MUL && N0.getNode()->hasOneUse()) {
+ if (ConstantSDNode *N0C1 = isConstOrConstSplat(N0.getOperand(1))) {
+ if (SDValue Folded =
+ DAG.FoldConstantArithmetic(ISD::SHL, SDLoc(N1), VT, N0C1, N1C))
+ return DAG.getNode(ISD::MUL, SDLoc(N), VT, N0.getOperand(0), Folded);
+ }
}
+ if (N1C && !N1C->isOpaque())
+ if (SDValue NewSHL = visitShiftByConstant(N, N1C))
+ return NewSHL;
+
return SDValue();
}
}
// fold (sra c1, c2) -> (sra c1, c2)
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- if (N0C && N1C)
- return DAG.FoldConstantArithmetic(ISD::SRA, VT, 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 (N0C && N0C->isNullValue())
+ if (isNullConstant(N0))
return N0;
// fold (sra -1, x) -> -1
- if (N0C && N0C->isAllOnesValue())
+ if (isAllOnesConstant(N0))
return N0;
// fold (sra x, (setge c, size(x))) -> undef
if (N1C && N1C->getZExtValue() >= OpSizeInBits)
unsigned Sum = N1C->getZExtValue() + C1->getZExtValue();
if (Sum >= OpSizeInBits)
Sum = OpSizeInBits - 1;
- return DAG.getNode(ISD::SRA, SDLoc(N), VT, N0.getOperand(0),
- DAG.getConstant(Sum, N1.getValueType()));
+ SDLoc DL(N);
+ return DAG.getNode(ISD::SRA, DL, VT, N0.getOperand(0),
+ DAG.getConstant(Sum, DL, N1.getValueType()));
}
}
TLI.isOperationLegalOrCustom(ISD::TRUNCATE, VT) &&
TLI.isTruncateFree(VT, TruncVT)) {
- SDValue Amt = DAG.getConstant(ShiftAmt,
- getShiftAmountTy(N0.getOperand(0).getValueType()));
- SDValue Shift = DAG.getNode(ISD::SRL, SDLoc(N0), VT,
- N0.getOperand(0), Amt);
- SDValue Trunc = DAG.getNode(ISD::TRUNCATE, SDLoc(N0), TruncVT,
- Shift);
- return DAG.getNode(ISD::SIGN_EXTEND, SDLoc(N),
- N->getValueType(0), Trunc);
+ SDLoc DL(N);
+ SDValue Amt = DAG.getConstant(ShiftAmt, DL,
+ getShiftAmountTy(N0.getOperand(0).getValueType()));
+ SDValue Shift = DAG.getNode(ISD::SRL, DL, VT,
+ N0.getOperand(0), Amt);
+ SDValue Trunc = DAG.getNode(ISD::TRUNCATE, DL, TruncVT,
+ Shift);
+ return DAG.getNode(ISD::SIGN_EXTEND, DL,
+ N->getValueType(0), Trunc);
}
}
}
EVT LargeVT = N0Op0.getValueType();
if (LargeVT.getScalarSizeInBits() - OpSizeInBits == LargeShiftVal) {
+ SDLoc DL(N);
SDValue Amt =
- DAG.getConstant(LargeShiftVal + N1C->getZExtValue(),
+ DAG.getConstant(LargeShiftVal + N1C->getZExtValue(), DL,
getShiftAmountTy(N0Op0.getOperand(0).getValueType()));
- SDValue SRA = DAG.getNode(ISD::SRA, SDLoc(N), LargeVT,
+ SDValue SRA = DAG.getNode(ISD::SRA, DL, LargeVT,
N0Op0.getOperand(0), Amt);
- return DAG.getNode(ISD::TRUNCATE, SDLoc(N), VT, SRA);
+ return DAG.getNode(ISD::TRUNCATE, DL, VT, SRA);
}
}
}
if (DAG.SignBitIsZero(N0))
return DAG.getNode(ISD::SRL, SDLoc(N), VT, N0, N1);
- if (N1C) {
- SDValue NewSRA = visitShiftByConstant(N, N1C);
- if (NewSRA.getNode())
+ if (N1C && !N1C->isOpaque())
+ if (SDValue NewSRA = visitShiftByConstant(N, N1C))
return NewSRA;
- }
return SDValue();
}
}
// fold (srl c1, c2) -> c1 >>u c2
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- if (N0C && N1C)
- return DAG.FoldConstantArithmetic(ISD::SRL, VT, 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 (N0C && N0C->isNullValue())
+ if (isNullConstant(N0))
return N0;
// fold (srl x, c >= size(x)) -> undef
if (N1C && N1C->getZExtValue() >= OpSizeInBits)
// if (srl x, c) is known to be zero, return 0
if (N1C && DAG.MaskedValueIsZero(SDValue(N, 0),
APInt::getAllOnesValue(OpSizeInBits)))
- return DAG.getConstant(0, VT);
+ return DAG.getConstant(0, SDLoc(N), VT);
// fold (srl (srl x, c1), c2) -> 0 or (srl x, (add c1, c2))
if (N1C && N0.getOpcode() == ISD::SRL) {
if (ConstantSDNode *N01C = isConstOrConstSplat(N0.getOperand(1))) {
uint64_t c1 = N01C->getZExtValue();
uint64_t c2 = N1C->getZExtValue();
+ SDLoc DL(N);
if (c1 + c2 >= OpSizeInBits)
- return DAG.getConstant(0, VT);
- return DAG.getNode(ISD::SRL, SDLoc(N), VT, N0.getOperand(0),
- DAG.getConstant(c1 + c2, N1.getValueType()));
+ return DAG.getConstant(0, DL, VT);
+ return DAG.getNode(ISD::SRL, DL, VT, N0.getOperand(0),
+ DAG.getConstant(c1 + c2, DL, N1.getValueType()));
}
}
uint64_t InnerShiftSize = InnerShiftVT.getScalarType().getSizeInBits();
// This is only valid if the OpSizeInBits + c1 = size of inner shift.
if (c1 + OpSizeInBits == InnerShiftSize) {
+ SDLoc DL(N0);
if (c1 + c2 >= InnerShiftSize)
- return DAG.getConstant(0, VT);
- return DAG.getNode(ISD::TRUNCATE, SDLoc(N0), VT,
- DAG.getNode(ISD::SRL, SDLoc(N0), InnerShiftVT,
+ return DAG.getConstant(0, DL, VT);
+ return DAG.getNode(ISD::TRUNCATE, DL, VT,
+ DAG.getNode(ISD::SRL, DL, InnerShiftVT,
N0.getOperand(0)->getOperand(0),
- DAG.getConstant(c1 + c2, ShiftCountVT)));
+ DAG.getConstant(c1 + c2, DL,
+ ShiftCountVT)));
}
}
unsigned BitSize = N0.getScalarValueSizeInBits();
if (BitSize <= 64) {
uint64_t ShAmt = N1C->getZExtValue() + 64 - BitSize;
- return DAG.getNode(ISD::AND, SDLoc(N), VT, N0.getOperand(0),
- DAG.getConstant(~0ULL >> ShAmt, VT));
+ SDLoc DL(N);
+ return DAG.getNode(ISD::AND, DL, VT, N0.getOperand(0),
+ DAG.getConstant(~0ULL >> ShAmt, DL, VT));
}
}
if (!LegalTypes || TLI.isTypeDesirableForOp(ISD::SRL, SmallVT)) {
uint64_t ShiftAmt = N1C->getZExtValue();
- SDValue SmallShift = DAG.getNode(ISD::SRL, SDLoc(N0), SmallVT,
+ SDLoc DL0(N0);
+ SDValue SmallShift = DAG.getNode(ISD::SRL, DL0, SmallVT,
N0.getOperand(0),
- DAG.getConstant(ShiftAmt, getShiftAmountTy(SmallVT)));
+ DAG.getConstant(ShiftAmt, DL0,
+ getShiftAmountTy(SmallVT)));
AddToWorklist(SmallShift.getNode());
APInt Mask = APInt::getAllOnesValue(OpSizeInBits).lshr(ShiftAmt);
- return DAG.getNode(ISD::AND, SDLoc(N), VT,
- DAG.getNode(ISD::ANY_EXTEND, SDLoc(N), VT, SmallShift),
- DAG.getConstant(Mask, VT));
+ SDLoc DL(N);
+ return DAG.getNode(ISD::AND, DL, VT,
+ DAG.getNode(ISD::ANY_EXTEND, DL, VT, SmallShift),
+ DAG.getConstant(Mask, DL, VT));
}
}
// If any of the input bits are KnownOne, then the input couldn't be all
// zeros, thus the result of the srl will always be zero.
- if (KnownOne.getBoolValue()) return DAG.getConstant(0, VT);
+ if (KnownOne.getBoolValue()) return DAG.getConstant(0, SDLoc(N0), VT);
// If all of the bits input the to ctlz node are known to be zero, then
// the result of the ctlz is "32" and the result of the shift is one.
APInt UnknownBits = ~KnownZero;
- if (UnknownBits == 0) return DAG.getConstant(1, VT);
+ if (UnknownBits == 0) return DAG.getConstant(1, SDLoc(N0), VT);
// Otherwise, check to see if there is exactly one bit input to the ctlz.
if ((UnknownBits & (UnknownBits - 1)) == 0) {
SDValue Op = N0.getOperand(0);
if (ShAmt) {
- Op = DAG.getNode(ISD::SRL, SDLoc(N0), VT, Op,
- DAG.getConstant(ShAmt, getShiftAmountTy(Op.getValueType())));
+ SDLoc DL(N0);
+ Op = DAG.getNode(ISD::SRL, DL, VT, Op,
+ DAG.getConstant(ShAmt, DL,
+ getShiftAmountTy(Op.getValueType())));
AddToWorklist(Op.getNode());
}
- return DAG.getNode(ISD::XOR, SDLoc(N), VT,
- Op, DAG.getConstant(1, VT));
+ SDLoc DL(N);
+ return DAG.getNode(ISD::XOR, DL, VT,
+ Op, DAG.getConstant(1, DL, VT));
}
}
// fold (srl x, (trunc (and y, c))) -> (srl x, (and (trunc y), (trunc c))).
if (N1.getOpcode() == ISD::TRUNCATE &&
N1.getOperand(0).getOpcode() == ISD::AND) {
- SDValue NewOp1 = distributeTruncateThroughAnd(N1.getNode());
- if (NewOp1.getNode())
+ if (SDValue NewOp1 = distributeTruncateThroughAnd(N1.getNode()))
return DAG.getNode(ISD::SRL, SDLoc(N), VT, N0, NewOp1);
}
if (N1C && SimplifyDemandedBits(SDValue(N, 0)))
return SDValue(N, 0);
- if (N1C) {
- SDValue NewSRL = visitShiftByConstant(N, N1C);
- if (NewSRL.getNode())
+ if (N1C && !N1C->isOpaque())
+ if (SDValue NewSRL = visitShiftByConstant(N, N1C))
return NewSRL;
- }
// Attempt to convert a srl of a load into a narrower zero-extending load.
- SDValue NarrowLoad = ReduceLoadWidth(N);
- if (NarrowLoad.getNode())
+ if (SDValue NarrowLoad = ReduceLoadWidth(N))
return NarrowLoad;
// Here is a common situation. We want to optimize:
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();
}
// fold (select C, X, X) -> X
if (N1 == N2)
return N1;
- // fold (select true, X, Y) -> X
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- if (N0C && !N0C->isNullValue())
- return N1;
- // fold (select false, X, Y) -> Y
- if (N0C && N0C->isNullValue())
- return N2;
+ if (const ConstantSDNode *N0C = dyn_cast<const ConstantSDNode>(N0)) {
+ // fold (select true, X, Y) -> X
+ // fold (select false, X, Y) -> Y
+ return !N0C->isNullValue() ? N1 : N2;
+ }
// fold (select C, 1, X) -> (or C, X)
- ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
- if (VT == MVT::i1 && N1C && N1C->getAPIntValue() == 1)
+ if (VT == MVT::i1 && isOneConstant(N1))
return DAG.getNode(ISD::OR, SDLoc(N), VT, N0, N2);
// fold (select C, 0, 1) -> (xor C, 1)
// We can't do this reliably if integer based booleans have different contents
// undiscoverable (or not reasonably discoverable). For example, it could be
// in another basic block or it could require searching a complicated
// expression.
- ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2);
if (VT.isInteger() &&
(VT0 == MVT::i1 || (VT0.isInteger() &&
TLI.getBooleanContents(false, false) ==
TLI.getBooleanContents(false, true) &&
TLI.getBooleanContents(false, false) ==
TargetLowering::ZeroOrOneBooleanContent)) &&
- N1C && N2C && N1C->isNullValue() && N2C->getAPIntValue() == 1) {
+ isNullConstant(N1) && isOneConstant(N2)) {
SDValue XORNode;
- if (VT == VT0)
- return DAG.getNode(ISD::XOR, SDLoc(N), VT0,
- N0, DAG.getConstant(1, VT0));
- XORNode = DAG.getNode(ISD::XOR, SDLoc(N0), VT0,
- N0, DAG.getConstant(1, VT0));
+ if (VT == VT0) {
+ SDLoc DL(N);
+ return DAG.getNode(ISD::XOR, DL, VT0,
+ N0, DAG.getConstant(1, DL, VT0));
+ }
+ SDLoc DL0(N0);
+ XORNode = DAG.getNode(ISD::XOR, DL0, VT0,
+ N0, DAG.getConstant(1, DL0, VT0));
AddToWorklist(XORNode.getNode());
if (VT.bitsGT(VT0))
return DAG.getNode(ISD::ZERO_EXTEND, SDLoc(N), VT, XORNode);
return DAG.getNode(ISD::TRUNCATE, SDLoc(N), VT, XORNode);
}
// fold (select C, 0, X) -> (and (not C), X)
- if (VT == VT0 && VT == MVT::i1 && N1C && N1C->isNullValue()) {
+ if (VT == VT0 && VT == MVT::i1 && isNullConstant(N1)) {
SDValue NOTNode = DAG.getNOT(SDLoc(N0), N0, VT);
AddToWorklist(NOTNode.getNode());
return DAG.getNode(ISD::AND, SDLoc(N), VT, NOTNode, N2);
}
// fold (select C, X, 1) -> (or (not C), X)
- if (VT == VT0 && VT == MVT::i1 && N2C && N2C->getAPIntValue() == 1) {
+ if (VT == VT0 && VT == MVT::i1 && isOneConstant(N2)) {
SDValue NOTNode = DAG.getNOT(SDLoc(N0), N0, VT);
AddToWorklist(NOTNode.getNode());
return DAG.getNode(ISD::OR, SDLoc(N), VT, NOTNode, N1);
}
// fold (select C, X, 0) -> (and C, X)
- if (VT == MVT::i1 && N2C && N2C->isNullValue())
+ if (VT == MVT::i1 && isNullConstant(N2))
return DAG.getNode(ISD::AND, SDLoc(N), VT, N0, N1);
// fold (select X, X, Y) -> (or X, Y)
// fold (select X, 1, Y) -> (or X, Y)
- if (VT == MVT::i1 && (N0 == N1 || (N1C && N1C->getAPIntValue() == 1)))
+ if (VT == MVT::i1 && (N0 == N1 || isOneConstant(N1)))
return DAG.getNode(ISD::OR, SDLoc(N), VT, N0, N2);
// fold (select X, Y, X) -> (and X, Y)
// fold (select X, Y, 0) -> (and X, Y)
- if (VT == MVT::i1 && (N0 == N2 || (N2C && N2C->getAPIntValue() == 0)))
+ if (VT == MVT::i1 && (N0 == N2 || isNullConstant(N2)))
return DAG.getNode(ISD::AND, SDLoc(N), VT, N0, N1);
// If we can fold this based on the true/false value, do so.
if (SimplifySelectOps(N, N1, N2))
return SDValue(N, 0); // Don't revisit N.
- // fold selects based on a setcc into other things, such as min/max/abs
- if (N0.getOpcode() == ISD::SETCC) {
- // select x, y (fcmp lt x, y) -> fminnum x, y
- // select x, y (fcmp gt x, y) -> fmaxnum x, y
- //
- // This is OK if we don't care about what happens if either operand is a
- // NaN.
- //
-
- // FIXME: Instead of testing for UnsafeFPMath, this should be checking for
- // no signed zeros as well as no nans.
- const TargetOptions &Options = DAG.getTarget().Options;
- if (Options.UnsafeFPMath &&
- VT.isFloatingPoint() && N0.hasOneUse() &&
- DAG.isKnownNeverNaN(N1) && DAG.isKnownNeverNaN(N2)) {
- ISD::CondCode CC = cast<CondCodeSDNode>(N0.getOperand(2))->get();
-
- SDValue FMinMax =
- combineMinNumMaxNum(SDLoc(N), VT, N0.getOperand(0), N0.getOperand(1),
- N1, N2, CC, TLI, DAG);
- if (FMinMax)
- return FMinMax;
- }
-
- if ((!LegalOperations &&
- TLI.isOperationLegalOrCustom(ISD::SELECT_CC, VT)) ||
- TLI.isOperationLegal(ISD::SELECT_CC, VT))
- return DAG.getNode(ISD::SELECT_CC, SDLoc(N), VT,
- N0.getOperand(0), N0.getOperand(1),
- N1, N2, N0.getOperand(2));
- return SimplifySelect(SDLoc(N), N0, N1, N2);
- }
-
if (VT0 == MVT::i1) {
- if (TLI.shouldNormalizeToSelectSequence(*DAG.getContext(), VT)) {
- // select (and Cond0, Cond1), X, Y
- // -> select Cond0, (select Cond1, X, Y), Y
- if (N0->getOpcode() == ISD::AND && N0->hasOneUse()) {
- SDValue Cond0 = N0->getOperand(0);
- SDValue Cond1 = N0->getOperand(1);
- SDValue InnerSelect = DAG.getNode(ISD::SELECT, SDLoc(N),
- N1.getValueType(), Cond1, N1, N2);
+ // The code in this block deals with the following 2 equivalences:
+ // select(C0|C1, x, y) <=> select(C0, x, select(C1, x, y))
+ // select(C0&C1, x, y) <=> select(C0, select(C1, x, y), y)
+ // The target can specify its prefered form with the
+ // shouldNormalizeToSelectSequence() callback. However we always transform
+ // to the right anyway if we find the inner select exists in the DAG anyway
+ // and we always transform to the left side if we know that we can further
+ // optimize the combination of the conditions.
+ bool normalizeToSequence
+ = TLI.shouldNormalizeToSelectSequence(*DAG.getContext(), VT);
+ // select (and Cond0, Cond1), X, Y
+ // -> select Cond0, (select Cond1, X, Y), Y
+ if (N0->getOpcode() == ISD::AND && N0->hasOneUse()) {
+ SDValue Cond0 = N0->getOperand(0);
+ SDValue Cond1 = N0->getOperand(1);
+ SDValue InnerSelect = DAG.getNode(ISD::SELECT, SDLoc(N),
+ N1.getValueType(), Cond1, N1, N2);
+ if (normalizeToSequence || !InnerSelect.use_empty())
return DAG.getNode(ISD::SELECT, SDLoc(N), N1.getValueType(), Cond0,
InnerSelect, N2);
- }
- // select (or Cond0, Cond1), X, Y -> select Cond0, X, (select Cond1, X, Y)
- if (N0->getOpcode() == ISD::OR && N0->hasOneUse()) {
- SDValue Cond0 = N0->getOperand(0);
- SDValue Cond1 = N0->getOperand(1);
- SDValue InnerSelect = DAG.getNode(ISD::SELECT, SDLoc(N),
- N1.getValueType(), Cond1, N1, N2);
+ }
+ // select (or Cond0, Cond1), X, Y -> select Cond0, X, (select Cond1, X, Y)
+ if (N0->getOpcode() == ISD::OR && N0->hasOneUse()) {
+ SDValue Cond0 = N0->getOperand(0);
+ SDValue Cond1 = N0->getOperand(1);
+ SDValue InnerSelect = DAG.getNode(ISD::SELECT, SDLoc(N),
+ N1.getValueType(), Cond1, N1, N2);
+ if (normalizeToSequence || !InnerSelect.use_empty())
return DAG.getNode(ISD::SELECT, SDLoc(N), N1.getValueType(), Cond0, N1,
InnerSelect);
- }
}
// select Cond0, (select Cond1, X, Y), Y -> select (and Cond0, Cond1), X, Y
- if (N1->getOpcode() == ISD::SELECT) {
+ if (N1->getOpcode() == ISD::SELECT && N1->hasOneUse()) {
SDValue N1_0 = N1->getOperand(0);
SDValue N1_1 = N1->getOperand(1);
SDValue N1_2 = N1->getOperand(2);
if (N1_2 == N2 && N0.getValueType() == N1_0.getValueType()) {
// Create the actual and node if we can generate good code for it.
- if (!TLI.shouldNormalizeToSelectSequence(*DAG.getContext(), VT)) {
+ if (!normalizeToSequence) {
SDValue And = DAG.getNode(ISD::AND, SDLoc(N), N0.getValueType(),
N0, N1_0);
return DAG.getNode(ISD::SELECT, SDLoc(N), N1.getValueType(), And,
}
}
// select Cond0, X, (select Cond1, X, Y) -> select (or Cond0, Cond1), X, Y
- if (N2->getOpcode() == ISD::SELECT) {
+ if (N2->getOpcode() == ISD::SELECT && N2->hasOneUse()) {
SDValue N2_0 = N2->getOperand(0);
SDValue N2_1 = N2->getOperand(1);
SDValue N2_2 = N2->getOperand(2);
if (N2_1 == N1 && N0.getValueType() == N2_0.getValueType()) {
// Create the actual or node if we can generate good code for it.
- if (!TLI.shouldNormalizeToSelectSequence(*DAG.getContext(), VT)) {
+ if (!normalizeToSequence) {
SDValue Or = DAG.getNode(ISD::OR, SDLoc(N), N0.getValueType(),
N0, N2_0);
return DAG.getNode(ISD::SELECT, SDLoc(N), N1.getValueType(), Or,
}
}
+ // fold selects based on a setcc into other things, such as min/max/abs
+ if (N0.getOpcode() == ISD::SETCC) {
+ // select x, y (fcmp lt x, y) -> fminnum x, y
+ // select x, y (fcmp gt x, y) -> fmaxnum x, y
+ //
+ // This is OK if we don't care about what happens if either operand is a
+ // NaN.
+ //
+
+ // FIXME: Instead of testing for UnsafeFPMath, this should be checking for
+ // no signed zeros as well as no nans.
+ const TargetOptions &Options = DAG.getTarget().Options;
+ if (Options.UnsafeFPMath &&
+ VT.isFloatingPoint() && N0.hasOneUse() &&
+ DAG.isKnownNeverNaN(N1) && DAG.isKnownNeverNaN(N2)) {
+ ISD::CondCode CC = cast<CondCodeSDNode>(N0.getOperand(2))->get();
+
+ if (SDValue FMinMax = combineMinNumMaxNum(SDLoc(N), VT, N0.getOperand(0),
+ N0.getOperand(1), N1, N2, CC,
+ TLI, DAG))
+ return FMinMax;
+ }
+
+ if ((!LegalOperations &&
+ TLI.isOperationLegalOrCustom(ISD::SELECT_CC, VT)) ||
+ TLI.isOperationLegal(ISD::SELECT_CC, VT))
+ return DAG.getNode(ISD::SELECT_CC, SDLoc(N), VT,
+ N0.getOperand(0), N0.getOperand(1),
+ N1, N2, N0.getOperand(2));
+ return SimplifySelect(SDLoc(N), N0, N1, N2);
+ }
+
return SDValue();
}
TopHalf->isNullValue() ? RHS->getOperand(1) : LHS->getOperand(1));
}
+SDValue DAGCombiner::visitMSCATTER(SDNode *N) {
+
+ if (Level >= AfterLegalizeTypes)
+ return SDValue();
+
+ MaskedScatterSDNode *MSC = cast<MaskedScatterSDNode>(N);
+ SDValue Mask = MSC->getMask();
+ SDValue Data = MSC->getValue();
+ SDLoc DL(N);
+
+ // If the MSCATTER data type requires splitting and the mask is provided by a
+ // SETCC, then split both nodes and its operands before legalization. This
+ // prevents the type legalizer from unrolling SETCC into scalar comparisons
+ // and enables future optimizations (e.g. min/max pattern matching on X86).
+ if (Mask.getOpcode() != ISD::SETCC)
+ return SDValue();
+
+ // Check if any splitting is required.
+ if (TLI.getTypeAction(*DAG.getContext(), Data.getValueType()) !=
+ TargetLowering::TypeSplitVector)
+ return SDValue();
+ SDValue MaskLo, MaskHi, Lo, Hi;
+ std::tie(MaskLo, MaskHi) = SplitVSETCC(Mask.getNode(), DAG);
+
+ EVT LoVT, HiVT;
+ std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(MSC->getValueType(0));
+
+ SDValue Chain = MSC->getChain();
+
+ EVT MemoryVT = MSC->getMemoryVT();
+ unsigned Alignment = MSC->getOriginalAlignment();
+
+ EVT LoMemVT, HiMemVT;
+ std::tie(LoMemVT, HiMemVT) = DAG.GetSplitDestVTs(MemoryVT);
+
+ SDValue DataLo, DataHi;
+ std::tie(DataLo, DataHi) = DAG.SplitVector(Data, DL);
+
+ SDValue BasePtr = MSC->getBasePtr();
+ SDValue IndexLo, IndexHi;
+ std::tie(IndexLo, IndexHi) = DAG.SplitVector(MSC->getIndex(), DL);
+
+ MachineMemOperand *MMO = DAG.getMachineFunction().
+ getMachineMemOperand(MSC->getPointerInfo(),
+ MachineMemOperand::MOStore, LoMemVT.getStoreSize(),
+ Alignment, MSC->getAAInfo(), MSC->getRanges());
+
+ SDValue OpsLo[] = { Chain, DataLo, MaskLo, BasePtr, IndexLo };
+ Lo = DAG.getMaskedScatter(DAG.getVTList(MVT::Other), DataLo.getValueType(),
+ DL, OpsLo, MMO);
+
+ SDValue OpsHi[] = {Chain, DataHi, MaskHi, BasePtr, IndexHi};
+ Hi = DAG.getMaskedScatter(DAG.getVTList(MVT::Other), DataHi.getValueType(),
+ DL, OpsHi, MMO);
+
+ AddToWorklist(Lo.getNode());
+ AddToWorklist(Hi.getNode());
+
+ return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Lo, Hi);
+}
+
SDValue DAGCombiner::visitMSTORE(SDNode *N) {
if (Level >= AfterLegalizeTypes)
unsigned IncrementSize = LoMemVT.getSizeInBits()/8;
Ptr = DAG.getNode(ISD::ADD, DL, Ptr.getValueType(), Ptr,
- DAG.getConstant(IncrementSize, Ptr.getValueType()));
+ DAG.getConstant(IncrementSize, DL, Ptr.getValueType()));
MMO = DAG.getMachineFunction().
getMachineMemOperand(MST->getPointerInfo(),
return SDValue();
}
+SDValue DAGCombiner::visitMGATHER(SDNode *N) {
+
+ if (Level >= AfterLegalizeTypes)
+ return SDValue();
+
+ MaskedGatherSDNode *MGT = dyn_cast<MaskedGatherSDNode>(N);
+ SDValue Mask = MGT->getMask();
+ SDLoc DL(N);
+
+ // If the MGATHER result requires splitting and the mask is provided by a
+ // SETCC, then split both nodes and its operands before legalization. This
+ // prevents the type legalizer from unrolling SETCC into scalar comparisons
+ // and enables future optimizations (e.g. min/max pattern matching on X86).
+
+ if (Mask.getOpcode() != ISD::SETCC)
+ return SDValue();
+
+ EVT VT = N->getValueType(0);
+
+ // Check if any splitting is required.
+ if (TLI.getTypeAction(*DAG.getContext(), VT) !=
+ TargetLowering::TypeSplitVector)
+ return SDValue();
+
+ SDValue MaskLo, MaskHi, Lo, Hi;
+ std::tie(MaskLo, MaskHi) = SplitVSETCC(Mask.getNode(), DAG);
+
+ SDValue Src0 = MGT->getValue();
+ SDValue Src0Lo, Src0Hi;
+ std::tie(Src0Lo, Src0Hi) = DAG.SplitVector(Src0, DL);
+
+ EVT LoVT, HiVT;
+ std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(VT);
+
+ SDValue Chain = MGT->getChain();
+ EVT MemoryVT = MGT->getMemoryVT();
+ unsigned Alignment = MGT->getOriginalAlignment();
+
+ EVT LoMemVT, HiMemVT;
+ std::tie(LoMemVT, HiMemVT) = DAG.GetSplitDestVTs(MemoryVT);
+
+ SDValue BasePtr = MGT->getBasePtr();
+ SDValue Index = MGT->getIndex();
+ SDValue IndexLo, IndexHi;
+ std::tie(IndexLo, IndexHi) = DAG.SplitVector(Index, DL);
+
+ MachineMemOperand *MMO = DAG.getMachineFunction().
+ getMachineMemOperand(MGT->getPointerInfo(),
+ MachineMemOperand::MOLoad, LoMemVT.getStoreSize(),
+ Alignment, MGT->getAAInfo(), MGT->getRanges());
+
+ SDValue OpsLo[] = { Chain, Src0Lo, MaskLo, BasePtr, IndexLo };
+ Lo = DAG.getMaskedGather(DAG.getVTList(LoVT, MVT::Other), LoVT, DL, OpsLo,
+ MMO);
+
+ SDValue OpsHi[] = {Chain, Src0Hi, MaskHi, BasePtr, IndexHi};
+ Hi = DAG.getMaskedGather(DAG.getVTList(HiVT, MVT::Other), HiVT, DL, OpsHi,
+ MMO);
+
+ AddToWorklist(Lo.getNode());
+ AddToWorklist(Hi.getNode());
+
+ // Build a factor node to remember that this load is independent of the
+ // other one.
+ Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Lo.getValue(1),
+ Hi.getValue(1));
+
+ // Legalized the chain result - switch anything that used the old chain to
+ // use the new one.
+ DAG.ReplaceAllUsesOfValueWith(SDValue(MGT, 1), Chain);
+
+ SDValue GatherRes = DAG.getNode(ISD::CONCAT_VECTORS, DL, VT, Lo, Hi);
+
+ SDValue RetOps[] = { GatherRes, Chain };
+ return DAG.getMergeValues(RetOps, DL);
+}
+
SDValue DAGCombiner::visitMLOAD(SDNode *N) {
if (Level >= AfterLegalizeTypes)
unsigned IncrementSize = LoMemVT.getSizeInBits()/8;
Ptr = DAG.getNode(ISD::ADD, DL, Ptr.getValueType(), Ptr,
- DAG.getConstant(IncrementSize, Ptr.getValueType()));
+ DAG.getConstant(IncrementSize, DL, Ptr.getValueType()));
MMO = DAG.getMachineFunction().
getMachineMemOperand(MLD->getPointerInfo(),
EVT VT = LHS.getValueType();
SDValue Shift = DAG.getNode(
ISD::SRA, DL, VT, LHS,
- DAG.getConstant(VT.getScalarType().getSizeInBits() - 1, VT));
+ DAG.getConstant(VT.getScalarType().getSizeInBits() - 1, DL, VT));
SDValue Add = DAG.getNode(ISD::ADD, DL, VT, LHS, Shift);
AddToWorklist(Shift.getNode());
AddToWorklist(Add.getNode());
if (N1.getOpcode() == ISD::CONCAT_VECTORS &&
N2.getOpcode() == ISD::CONCAT_VECTORS &&
ISD::isBuildVectorOfConstantSDNodes(N0.getNode())) {
- SDValue CV = ConvertSelectToConcatVector(N, DAG);
- if (CV.getNode())
+ if (SDValue CV = ConvertSelectToConcatVector(N, DAG))
return CV;
}
SDLoc(N));
}
-// tryToFoldExtendOfConstant - Try to fold a sext/zext/aext
-// dag node into a ConstantSDNode or a build_vector of constants.
-// This function is called by the DAGCombiner when visiting sext/zext/aext
-// dag nodes (see for example method DAGCombiner::visitSIGN_EXTEND).
-// Vector extends are not folded if operations are legal; this is to
-// avoid introducing illegal build_vector dag nodes.
+SDValue DAGCombiner::visitSETCCE(SDNode *N) {
+ SDValue LHS = N->getOperand(0);
+ SDValue RHS = N->getOperand(1);
+ SDValue Carry = N->getOperand(2);
+ SDValue Cond = N->getOperand(3);
+
+ // If Carry is false, fold to a regular SETCC.
+ if (Carry.getOpcode() == ISD::CARRY_FALSE)
+ return DAG.getNode(ISD::SETCC, SDLoc(N), N->getVTList(), LHS, RHS, Cond);
+
+ return SDValue();
+}
+
+/// Try to fold a sext/zext/aext dag node into a ConstantSDNode or
+/// a build_vector of constants.
+/// This function is called by the DAGCombiner when visiting sext/zext/aext
+/// dag nodes (see for example method DAGCombiner::visitSIGN_EXTEND).
+/// Vector extends are not folded if operations are legal; this is to
+/// avoid introducing illegal build_vector dag nodes.
static SDNode *tryToFoldExtendOfConstant(SDNode *N, const TargetLowering &TLI,
SelectionDAG &DAG, bool LegalTypes,
bool LegalOperations) {
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
// We can fold this node into a build_vector.
unsigned VTBits = SVT.getSizeInBits();
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) {
continue;
}
- ConstantSDNode *CurrentND = cast<ConstantSDNode>(Op);
- const APInt &C = APInt(VTBits, CurrentND->getAPIntValue().getZExtValue());
- if (Opcode == ISD::SIGN_EXTEND)
- Elts.push_back(DAG.getConstant(C.shl(ShAmt).ashr(ShAmt).getZExtValue(),
- SVT));
+ SDLoc DL(Op);
+ // Get the constant value and if needed trunc it to the size of the type.
+ // Nodes like build_vector might have constants wider than the scalar type.
+ APInt C = cast<ConstantSDNode>(Op)->getAPIntValue().zextOrTrunc(EVTBits);
+ if (Opcode == ISD::SIGN_EXTEND || Opcode == ISD::SIGN_EXTEND_VECTOR_INREG)
+ Elts.push_back(DAG.getConstant(C.sext(VTBits), DL, SVT));
else
- Elts.push_back(DAG.getConstant(C.shl(ShAmt).lshr(ShAmt).getZExtValue(),
- SVT));
+ Elts.push_back(DAG.getConstant(C.zext(VTBits), DL, SVT));
}
return DAG.getNode(ISD::BUILD_VECTOR, DL, VT, Elts).getNode();
Align, LN0->getAAInfo());
BasePtr = DAG.getNode(ISD::ADD, DL, BasePtr.getValueType(), BasePtr,
- DAG.getConstant(Stride, BasePtr.getValueType()));
+ DAG.getConstant(Stride, DL, BasePtr.getValueType()));
Loads.push_back(SplitLoad.getValue(0));
Chains.push_back(SplitLoad.getValue(1));
if (N0.getOpcode() == ISD::TRUNCATE) {
// fold (sext (truncate (load x))) -> (sext (smaller load x))
// fold (sext (truncate (srl (load x), c))) -> (sext (smaller load (x+c/n)))
- SDValue NarrowLoad = ReduceLoadWidth(N0.getNode());
- if (NarrowLoad.getNode()) {
+ if (SDValue NarrowLoad = ReduceLoadWidth(N0.getNode())) {
SDNode* oye = N0.getNode()->getOperand(0).getNode();
if (NarrowLoad.getNode() != N0.getNode()) {
CombineTo(N0.getNode(), NarrowLoad);
LN0->getMemOperand());
APInt Mask = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
Mask = Mask.sext(VT.getSizeInBits());
- SDValue And = DAG.getNode(N0.getOpcode(), SDLoc(N), VT,
- ExtLoad, DAG.getConstant(Mask, VT));
+ SDLoc DL(N);
+ SDValue And = DAG.getNode(N0.getOpcode(), DL, VT,
+ ExtLoad, DAG.getConstant(Mask, DL, VT));
SDValue Trunc = DAG.getNode(ISD::TRUNCATE,
SDLoc(N0.getOperand(0)),
N0.getOperand(0).getValueType(), ExtLoad);
CombineTo(N, And);
CombineTo(N0.getOperand(0).getNode(), Trunc, ExtLoad.getValue(1));
- ExtendSetCCUses(SetCCs, Trunc, ExtLoad, SDLoc(N),
+ ExtendSetCCUses(SetCCs, Trunc, ExtLoad, DL,
ISD::SIGN_EXTEND);
return SDValue(N, 0); // Return N so it doesn't get rechecked!
}
// sext(setcc x, y, cc) -> (select (setcc x, y, cc), -1, 0)
unsigned ElementWidth = VT.getScalarType().getSizeInBits();
+ SDLoc DL(N);
SDValue NegOne =
- DAG.getConstant(APInt::getAllOnesValue(ElementWidth), VT);
+ DAG.getConstant(APInt::getAllOnesValue(ElementWidth), DL, VT);
SDValue SCC =
- SimplifySelectCC(SDLoc(N), N0.getOperand(0), N0.getOperand(1),
- NegOne, DAG.getConstant(0, VT),
+ SimplifySelectCC(DL, N0.getOperand(0), N0.getOperand(1),
+ NegOne, DAG.getConstant(0, DL, VT),
cast<CondCodeSDNode>(N0.getOperand(2))->get(), true);
if (SCC.getNode()) return SCC;
if (!VT.isVector()) {
EVT SetCCVT = getSetCCResultType(N0.getOperand(0).getValueType());
- if (!LegalOperations || TLI.isOperationLegal(ISD::SETCC, SetCCVT)) {
+ if (!LegalOperations ||
+ TLI.isOperationLegal(ISD::SETCC, N0.getOperand(0).getValueType())) {
SDLoc DL(N);
ISD::CondCode CC = cast<CondCodeSDNode>(N0.getOperand(2))->get();
SDValue SetCC = DAG.getSetCC(DL, SetCCVT,
N0.getOperand(0), N0.getOperand(1), CC);
return DAG.getSelect(DL, VT, SetCC,
- NegOne, DAG.getConstant(0, VT));
+ NegOne, DAG.getConstant(0, DL, VT));
}
}
}
SDValue Op1 = N->getOperand(1);
assert(Op0.getValueType() == Op1.getValueType());
- ConstantSDNode *COp0 = dyn_cast<ConstantSDNode>(Op0);
- ConstantSDNode *COp1 = dyn_cast<ConstantSDNode>(Op1);
- if (COp0 && COp0->isNullValue())
+ if (isNullConstant(Op0))
Op = Op1;
- else if (COp1 && COp1->isNullValue())
+ else if (isNullConstant(Op1))
Op = Op0;
else
return false;
// fold (zext (truncate (load x))) -> (zext (smaller load x))
// fold (zext (truncate (srl (load x), c))) -> (zext (small load (x+c/n)))
if (N0.getOpcode() == ISD::TRUNCATE) {
- SDValue NarrowLoad = ReduceLoadWidth(N0.getNode());
- if (NarrowLoad.getNode()) {
+ if (SDValue NarrowLoad = ReduceLoadWidth(N0.getNode())) {
SDNode* oye = N0.getNode()->getOperand(0).getNode();
if (NarrowLoad.getNode() != N0.getNode()) {
CombineTo(N0.getNode(), NarrowLoad);
}
// fold (zext (truncate x)) -> (and x, mask)
- if (N0.getOpcode() == ISD::TRUNCATE &&
- (!LegalOperations || TLI.isOperationLegal(ISD::AND, VT))) {
-
+ if (N0.getOpcode() == ISD::TRUNCATE) {
// fold (zext (truncate (load x))) -> (zext (smaller load x))
// fold (zext (truncate (srl (load x), c))) -> (zext (smaller load (x+c/n)))
- SDValue NarrowLoad = ReduceLoadWidth(N0.getNode());
- if (NarrowLoad.getNode()) {
- SDNode* oye = N0.getNode()->getOperand(0).getNode();
+ if (SDValue NarrowLoad = ReduceLoadWidth(N0.getNode())) {
+ SDNode *oye = N0.getNode()->getOperand(0).getNode();
if (NarrowLoad.getNode() != N0.getNode()) {
CombineTo(N0.getNode(), NarrowLoad);
// CombineTo deleted the truncate, if needed, but not what's under it.
AddToWorklist(oye);
}
- return SDValue(N, 0); // Return N so it doesn't get rechecked!
+ return SDValue(N, 0); // Return N so it doesn't get rechecked!
}
- SDValue Op = N0.getOperand(0);
- if (Op.getValueType().bitsLT(VT)) {
- Op = DAG.getNode(ISD::ANY_EXTEND, SDLoc(N), VT, Op);
- AddToWorklist(Op.getNode());
- } else if (Op.getValueType().bitsGT(VT)) {
- Op = DAG.getNode(ISD::TRUNCATE, SDLoc(N), VT, Op);
- AddToWorklist(Op.getNode());
+ EVT SrcVT = N0.getOperand(0).getValueType();
+ EVT MinVT = N0.getValueType();
+
+ // Try to mask before the extension to avoid having to generate a larger mask,
+ // possibly over several sub-vectors.
+ if (SrcVT.bitsLT(VT)) {
+ if (!LegalOperations || (TLI.isOperationLegal(ISD::AND, SrcVT) &&
+ TLI.isOperationLegal(ISD::ZERO_EXTEND, VT))) {
+ SDValue Op = N0.getOperand(0);
+ Op = DAG.getZeroExtendInReg(Op, SDLoc(N), MinVT.getScalarType());
+ AddToWorklist(Op.getNode());
+ return DAG.getZExtOrTrunc(Op, SDLoc(N), VT);
+ }
+ }
+
+ if (!LegalOperations || TLI.isOperationLegal(ISD::AND, VT)) {
+ SDValue Op = N0.getOperand(0);
+ if (SrcVT.bitsLT(VT)) {
+ Op = DAG.getNode(ISD::ANY_EXTEND, SDLoc(N), VT, Op);
+ AddToWorklist(Op.getNode());
+ } else if (SrcVT.bitsGT(VT)) {
+ Op = DAG.getNode(ISD::TRUNCATE, SDLoc(N), VT, Op);
+ AddToWorklist(Op.getNode());
+ }
+ return DAG.getZeroExtendInReg(Op, SDLoc(N), MinVT.getScalarType());
}
- return DAG.getZeroExtendInReg(Op, SDLoc(N),
- N0.getValueType().getScalarType());
}
// Fold (zext (and (trunc x), cst)) -> (and x, cst),
}
APInt Mask = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
Mask = Mask.zext(VT.getSizeInBits());
- return DAG.getNode(ISD::AND, SDLoc(N), VT,
- X, DAG.getConstant(Mask, VT));
+ SDLoc DL(N);
+ return DAG.getNode(ISD::AND, DL, VT,
+ X, DAG.getConstant(Mask, DL, VT));
}
// fold (zext (load x)) -> (zext (truncate (zextload x)))
// fold (zext (and/or/xor (load x), cst)) ->
// (and/or/xor (zextload x), (zext cst))
+ // Unless (and (load x) cst) will match as a zextload already and has
+ // additional users.
if ((N0.getOpcode() == ISD::AND || N0.getOpcode() == ISD::OR ||
N0.getOpcode() == ISD::XOR) &&
isa<LoadSDNode>(N0.getOperand(0)) &&
if (LN0->getExtensionType() != ISD::SEXTLOAD && LN0->isUnindexed()) {
bool DoXform = true;
SmallVector<SDNode*, 4> SetCCs;
- if (!N0.hasOneUse())
- DoXform = ExtendUsesToFormExtLoad(N, N0.getOperand(0), ISD::ZERO_EXTEND,
- SetCCs, TLI);
+ if (!N0.hasOneUse()) {
+ if (N0.getOpcode() == ISD::AND) {
+ auto *AndC = cast<ConstantSDNode>(N0.getOperand(1));
+ auto NarrowLoad = false;
+ EVT LoadResultTy = AndC->getValueType(0);
+ EVT ExtVT, LoadedVT;
+ if (isAndLoadExtLoad(AndC, LN0, LoadResultTy, ExtVT, LoadedVT,
+ NarrowLoad))
+ DoXform = false;
+ }
+ if (DoXform)
+ DoXform = ExtendUsesToFormExtLoad(N, N0.getOperand(0),
+ ISD::ZERO_EXTEND, SetCCs, TLI);
+ }
if (DoXform) {
SDValue ExtLoad = DAG.getExtLoad(ISD::ZEXTLOAD, SDLoc(LN0), VT,
LN0->getChain(), LN0->getBasePtr(),
LN0->getMemOperand());
APInt Mask = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
Mask = Mask.zext(VT.getSizeInBits());
- SDValue And = DAG.getNode(N0.getOpcode(), SDLoc(N), VT,
- ExtLoad, DAG.getConstant(Mask, VT));
+ SDLoc DL(N);
+ SDValue And = DAG.getNode(N0.getOpcode(), DL, VT,
+ ExtLoad, DAG.getConstant(Mask, DL, VT));
SDValue Trunc = DAG.getNode(ISD::TRUNCATE,
SDLoc(N0.getOperand(0)),
N0.getOperand(0).getValueType(), ExtLoad);
CombineTo(N, And);
CombineTo(N0.getOperand(0).getNode(), Trunc, ExtLoad.getValue(1));
- ExtendSetCCUses(SetCCs, Trunc, ExtLoad, SDLoc(N),
+ ExtendSetCCUses(SetCCs, Trunc, ExtLoad, DL,
ISD::ZERO_EXTEND);
return SDValue(N, 0); // Return N so it doesn't get rechecked!
}
// zext(setcc) -> (and (vsetcc), (1, 1, ...) for vectors.
// Only do this before legalize for now.
EVT EltVT = VT.getVectorElementType();
+ SDLoc DL(N);
SmallVector<SDValue,8> OneOps(VT.getVectorNumElements(),
- DAG.getConstant(1, EltVT));
+ DAG.getConstant(1, DL, EltVT));
if (VT.getSizeInBits() == N0VT.getSizeInBits())
// We know that the # elements of the results is the same as the
// # elements of the compare (and the # elements of the compare result
// for that matter). Check to see that they are the same size. If so,
// we know that the element size of the sext'd result matches the
// element size of the compare operands.
- return DAG.getNode(ISD::AND, SDLoc(N), VT,
- DAG.getSetCC(SDLoc(N), VT, N0.getOperand(0),
+ return DAG.getNode(ISD::AND, DL, VT,
+ DAG.getSetCC(DL, VT, N0.getOperand(0),
N0.getOperand(1),
cast<CondCodeSDNode>(N0.getOperand(2))->get()),
- DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), VT,
+ DAG.getNode(ISD::BUILD_VECTOR, DL, VT,
OneOps));
// If the desired elements are smaller or larger than the source
EVT::getVectorVT(*DAG.getContext(), MatchingElementType,
N0VT.getVectorNumElements());
SDValue VsetCC =
- DAG.getSetCC(SDLoc(N), MatchingVectorType, N0.getOperand(0),
+ DAG.getSetCC(DL, MatchingVectorType, N0.getOperand(0),
N0.getOperand(1),
cast<CondCodeSDNode>(N0.getOperand(2))->get());
- return DAG.getNode(ISD::AND, SDLoc(N), VT,
- DAG.getSExtOrTrunc(VsetCC, SDLoc(N), VT),
- DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), VT, OneOps));
+ return DAG.getNode(ISD::AND, DL, VT,
+ DAG.getSExtOrTrunc(VsetCC, DL, VT),
+ DAG.getNode(ISD::BUILD_VECTOR, DL, VT, OneOps));
}
// zext(setcc x,y,cc) -> select_cc x, y, 1, 0, cc
+ SDLoc DL(N);
SDValue SCC =
- SimplifySelectCC(SDLoc(N), N0.getOperand(0), N0.getOperand(1),
- DAG.getConstant(1, VT), DAG.getConstant(0, VT),
+ SimplifySelectCC(DL, N0.getOperand(0), N0.getOperand(1),
+ DAG.getConstant(1, DL, VT), DAG.getConstant(0, DL, VT),
cast<CondCodeSDNode>(N0.getOperand(2))->get(), true);
if (SCC.getNode()) return SCC;
}
// fold (aext (truncate (load x))) -> (aext (smaller load x))
// fold (aext (truncate (srl (load x), c))) -> (aext (small load (x+c/n)))
if (N0.getOpcode() == ISD::TRUNCATE) {
- SDValue NarrowLoad = ReduceLoadWidth(N0.getNode());
- if (NarrowLoad.getNode()) {
+ if (SDValue NarrowLoad = ReduceLoadWidth(N0.getNode())) {
SDNode* oye = N0.getNode()->getOperand(0).getNode();
if (NarrowLoad.getNode() != N0.getNode()) {
CombineTo(N0.getNode(), NarrowLoad);
}
APInt Mask = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
Mask = Mask.zext(VT.getSizeInBits());
- return DAG.getNode(ISD::AND, SDLoc(N), VT,
- X, DAG.getConstant(Mask, VT));
+ SDLoc DL(N);
+ return DAG.getNode(ISD::AND, DL, VT,
+ X, DAG.getConstant(Mask, DL, VT));
}
// fold (aext (load x)) -> (aext (truncate (extload x)))
}
// aext(setcc x,y,cc) -> select_cc x, y, 1, 0, cc
+ SDLoc DL(N);
SDValue SCC =
- SimplifySelectCC(SDLoc(N), N0.getOperand(0), N0.getOperand(1),
- DAG.getConstant(1, VT), DAG.getConstant(0, VT),
+ SimplifySelectCC(DL, N0.getOperand(0), N0.getOperand(1),
+ DAG.getConstant(1, DL, VT), DAG.getConstant(0, DL, VT),
cast<CondCodeSDNode>(N0.getOperand(2))->get(), true);
if (SCC.getNode())
return SCC;
const APInt &CVal = CV->getAPIntValue();
APInt NewVal = CVal & Mask;
if (NewVal != CVal)
- return DAG.getConstant(NewVal, V.getValueType());
+ return DAG.getConstant(NewVal, SDLoc(V), V.getValueType());
break;
}
case ISD::OR:
// 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();
// Watch out for shift count overflow though.
if (Amt >= Mask.getBitWidth()) break;
APInt NewMask = Mask << Amt;
- SDValue SimplifyLHS = GetDemandedBits(V.getOperand(0), NewMask);
- if (SimplifyLHS.getNode())
+ if (SDValue SimplifyLHS = GetDemandedBits(V.getOperand(0), NewMask))
return DAG.getNode(ISD::SRL, SDLoc(V), V.getValueType(),
SimplifyLHS, V.getOperand(1));
}
// For big endian targets, we need to adjust the offset to the pointer to
// load the correct bytes.
- if (TLI.isBigEndian()) {
+ if (DAG.getDataLayout().isBigEndian()) {
unsigned LVTStoreBits = LN0->getMemoryVT().getStoreSizeInBits();
unsigned EVTStoreBits = ExtVT.getStoreSizeInBits();
ShAmt = LVTStoreBits - EVTStoreBits - ShAmt;
uint64_t PtrOff = ShAmt / 8;
unsigned NewAlign = MinAlign(LN0->getAlignment(), PtrOff);
- SDValue NewPtr = DAG.getNode(ISD::ADD, SDLoc(LN0),
+ SDLoc DL(LN0);
+ SDValue NewPtr = DAG.getNode(ISD::ADD, DL,
PtrType, LN0->getBasePtr(),
- DAG.getConstant(PtrOff, PtrType));
+ DAG.getConstant(PtrOff, DL, PtrType));
AddToWorklist(NewPtr.getNode());
SDValue Load;
// no larger than the source) then the useful bits of the result are
// zero; we can't simply return the shortened shift, because the result
// of that operation is undefined.
+ SDLoc DL(N0);
if (ShLeftAmt >= VT.getSizeInBits())
- Result = DAG.getConstant(0, VT);
+ Result = DAG.getConstant(0, DL, VT);
else
- Result = DAG.getNode(ISD::SHL, SDLoc(N0), VT,
- Result, DAG.getConstant(ShLeftAmt, ShImmTy));
+ Result = DAG.getNode(ISD::SHL, DL, VT,
+ Result, DAG.getConstant(ShLeftAmt, DL, ShImmTy));
}
// Return the new loaded value.
unsigned VTBits = VT.getScalarType().getSizeInBits();
unsigned EVTBits = EVT.getScalarType().getSizeInBits();
+ if (N0.isUndef())
+ return DAG.getUNDEF(VT);
+
// fold (sext_in_reg c1) -> c1
- if (isa<ConstantSDNode>(N0) || N0.getOpcode() == ISD::UNDEF)
+ if (isConstantIntBuildVectorOrConstantInt(N0))
return DAG.getNode(ISD::SIGN_EXTEND_INREG, SDLoc(N), VT, N0, N1);
// If the input is already sign extended, just drop the extension.
// fold (sext_in_reg (load x)) -> (smaller sextload x)
// fold (sext_in_reg (srl (load x), c)) -> (smaller sextload (x+c/evtbits))
- SDValue NarrowLoad = ReduceLoadWidth(N);
- if (NarrowLoad.getNode())
+ if (SDValue NarrowLoad = ReduceLoadWidth(N))
return NarrowLoad;
// fold (sext_in_reg (srl X, 24), i8) -> (sra X, 24)
BSwap, N1);
}
- // Fold a sext_inreg of a build_vector of ConstantSDNodes or undefs
- // into a build_vector.
- if (ISD::isBuildVectorOfConstantSDNodes(N0.getNode())) {
- SmallVector<SDValue, 8> Elts;
- unsigned NumElts = N0->getNumOperands();
- unsigned ShAmt = VTBits - EVTBits;
+ return SDValue();
+}
- for (unsigned i = 0; i != NumElts; ++i) {
- SDValue Op = N0->getOperand(i);
- if (Op->getOpcode() == ISD::UNDEF) {
- Elts.push_back(Op);
- continue;
- }
+SDValue DAGCombiner::visitSIGN_EXTEND_VECTOR_INREG(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+ EVT VT = N->getValueType(0);
- ConstantSDNode *CurrentND = cast<ConstantSDNode>(Op);
- const APInt &C = APInt(VTBits, CurrentND->getAPIntValue().getZExtValue());
- Elts.push_back(DAG.getConstant(C.shl(ShAmt).ashr(ShAmt).getZExtValue(),
- Op.getValueType()));
- }
+ if (N0.getOpcode() == ISD::UNDEF)
+ return DAG.getUNDEF(VT);
- return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), VT, Elts);
- }
+ 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 isLE = TLI.isLittleEndian();
+ bool isLE = DAG.getDataLayout().isLittleEndian();
// noop truncate
if (N0.getValueType() == N->getValueType(0))
SDValue EltNo = N0->getOperand(1);
if (isa<ConstantSDNode>(EltNo) && isTypeLegal(NVT)) {
int Elt = cast<ConstantSDNode>(EltNo)->getZExtValue();
- EVT IndexTy = TLI.getVectorIdxTy();
+ EVT IndexTy = TLI.getVectorIdxTy(DAG.getDataLayout());
int Index = isLE ? (Elt*SizeRatio) : (Elt*SizeRatio + (SizeRatio-1));
SDValue V = DAG.getNode(ISD::BITCAST, SDLoc(N),
NVT, N0.getOperand(0));
+ SDLoc DL(N);
return DAG.getNode(ISD::EXTRACT_VECTOR_ELT,
- SDLoc(N), TrTy, V,
- DAG.getConstant(Index, IndexTy));
+ DL, TrTy, V,
+ DAG.getConstant(Index, DL, IndexTy));
}
}
// fold (truncate (load x)) -> (smaller load x)
// fold (truncate (srl (load x), c)) -> (smaller load (x+c/evtbits))
if (!LegalTypes || TLI.isTypeDesirableForOp(N0.getOpcode(), VT)) {
- SDValue Reduced = ReduceLoadWidth(N);
- if (Reduced.getNode())
+ if (SDValue Reduced = ReduceLoadWidth(N))
return Reduced;
+
// Handle the case where the load remains an extending load even
// after truncation.
if (N0.hasOneUse() && ISD::isUNINDEXEDLoad(N0.getNode())) {
!LD2->isVolatile() &&
DAG.isConsecutiveLoad(LD2, LD1, LD1VT.getSizeInBits()/8, 1)) {
unsigned Align = LD1->getAlignment();
- unsigned NewAlign = TLI.getDataLayout()->
- getABITypeAlignment(VT.getTypeForEVT(*DAG.getContext()));
+ unsigned NewAlign = DAG.getDataLayout().getABITypeAlignment(
+ VT.getTypeForEVT(*DAG.getContext()));
if (NewAlign <= Align &&
(!LegalOperations || TLI.isOperationLegal(ISD::LOAD, VT)))
// Do not change the width of a volatile load.
!cast<LoadSDNode>(N0)->isVolatile() &&
// Do not remove the cast if the types differ in endian layout.
- TLI.hasBigEndianPartOrdering(N0.getValueType()) ==
- TLI.hasBigEndianPartOrdering(VT) &&
+ TLI.hasBigEndianPartOrdering(N0.getValueType(), DAG.getDataLayout()) ==
+ TLI.hasBigEndianPartOrdering(VT, DAG.getDataLayout()) &&
(!LegalOperations || TLI.isOperationLegal(ISD::LOAD, VT)) &&
TLI.isLoadBitCastBeneficial(N0.getValueType(), VT)) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
- unsigned Align = TLI.getDataLayout()->
- getABITypeAlignment(VT.getTypeForEVT(*DAG.getContext()));
+ unsigned Align = DAG.getDataLayout().getABITypeAlignment(
+ VT.getTypeForEVT(*DAG.getContext()));
unsigned OrigAlign = LN0->getAlignment();
if (Align <= OrigAlign) {
N0.getOperand(0));
AddToWorklist(NewConv.getNode());
+ SDLoc DL(N);
APInt SignBit = APInt::getSignBit(VT.getSizeInBits());
if (N0.getOpcode() == ISD::FNEG)
- return DAG.getNode(ISD::XOR, SDLoc(N), VT,
- NewConv, DAG.getConstant(SignBit, VT));
+ return DAG.getNode(ISD::XOR, DL, VT,
+ NewConv, DAG.getConstant(SignBit, DL, VT));
assert(N0.getOpcode() == ISD::FABS);
- return DAG.getNode(ISD::AND, SDLoc(N), VT,
- NewConv, DAG.getConstant(~SignBit, VT));
+ return DAG.getNode(ISD::AND, DL, VT,
+ NewConv, DAG.getConstant(~SignBit, DL, VT));
}
// fold (bitconvert (fcopysign cst, x)) ->
} else if (OrigXWidth > VTWidth) {
// To get the sign bit in the right place, we have to shift it right
// before truncating.
- X = DAG.getNode(ISD::SRL, SDLoc(X),
+ SDLoc DL(X);
+ X = DAG.getNode(ISD::SRL, DL,
X.getValueType(), X,
- DAG.getConstant(OrigXWidth-VTWidth, X.getValueType()));
+ DAG.getConstant(OrigXWidth-VTWidth, DL,
+ X.getValueType()));
AddToWorklist(X.getNode());
X = DAG.getNode(ISD::TRUNCATE, SDLoc(X), VT, X);
AddToWorklist(X.getNode());
APInt SignBit = APInt::getSignBit(VT.getSizeInBits());
X = DAG.getNode(ISD::AND, SDLoc(X), VT,
- X, DAG.getConstant(SignBit, VT));
+ X, DAG.getConstant(SignBit, SDLoc(X), VT));
AddToWorklist(X.getNode());
SDValue Cst = DAG.getNode(ISD::BITCAST, SDLoc(N0),
VT, N0.getOperand(0));
Cst = DAG.getNode(ISD::AND, SDLoc(Cst), VT,
- Cst, DAG.getConstant(~SignBit, VT));
+ Cst, DAG.getConstant(~SignBit, SDLoc(Cst), VT));
AddToWorklist(Cst.getNode());
return DAG.getNode(ISD::OR, SDLoc(N), VT, X, Cst);
}
// bitconvert(build_pair(ld, ld)) -> ld iff load locations are consecutive.
- if (N0.getOpcode() == ISD::BUILD_PAIR) {
- SDValue CombineLD = CombineConsecutiveLoads(N0.getNode(), VT);
- if (CombineLD.getNode())
+ if (N0.getOpcode() == ISD::BUILD_PAIR)
+ if (SDValue CombineLD = CombineConsecutiveLoads(N0.getNode(), VT))
return CombineLD;
+
+ // Remove double bitcasts from shuffles - this is often a legacy of
+ // XformToShuffleWithZero being used to combine bitmaskings (of
+ // float vectors bitcast to integer vectors) into shuffles.
+ // bitcast(shuffle(bitcast(s0),bitcast(s1))) -> shuffle(s0,s1)
+ if (Level < AfterLegalizeDAG && TLI.isTypeLegal(VT) && VT.isVector() &&
+ N0->getOpcode() == ISD::VECTOR_SHUFFLE &&
+ VT.getVectorNumElements() >= N0.getValueType().getVectorNumElements() &&
+ !(VT.getVectorNumElements() % N0.getValueType().getVectorNumElements())) {
+ ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(N0);
+
+ // If operands are a bitcast, peek through if it casts the original VT.
+ // If operands are a constant, just bitcast back to original VT.
+ auto PeekThroughBitcast = [&](SDValue Op) {
+ if (Op.getOpcode() == ISD::BITCAST &&
+ Op.getOperand(0).getValueType() == VT)
+ return SDValue(Op.getOperand(0));
+ if (ISD::isBuildVectorOfConstantSDNodes(Op.getNode()) ||
+ ISD::isBuildVectorOfConstantFPSDNodes(Op.getNode()))
+ return DAG.getNode(ISD::BITCAST, SDLoc(N), VT, Op);
+ return SDValue();
+ };
+
+ SDValue SV0 = PeekThroughBitcast(N0->getOperand(0));
+ SDValue SV1 = PeekThroughBitcast(N0->getOperand(1));
+ if (!(SV0 && SV1))
+ return SDValue();
+
+ int MaskScale =
+ VT.getVectorNumElements() / N0.getValueType().getVectorNumElements();
+ SmallVector<int, 8> NewMask;
+ for (int M : SVN->getMask())
+ for (int i = 0; i != MaskScale; ++i)
+ NewMask.push_back(M < 0 ? -1 : M * MaskScale + i);
+
+ bool LegalMask = TLI.isShuffleMaskLegal(NewMask, VT);
+ if (!LegalMask) {
+ std::swap(SV0, SV1);
+ ShuffleVectorSDNode::commuteMask(NewMask);
+ LegalMask = TLI.isShuffleMaskLegal(NewMask, VT);
+ }
+
+ if (LegalMask)
+ return DAG.getVectorShuffle(VT, SDLoc(N), SV0, SV1, NewMask);
}
return SDValue();
DstEltVT, BV->getOperand(0)));
SmallVector<SDValue, 8> Ops;
- for (unsigned i = 0, e = BV->getNumOperands(); i != e; ++i) {
- SDValue Op = BV->getOperand(i);
+ for (SDValue Op : BV->op_values()) {
// If the vector element type is not legal, the BUILD_VECTOR operands
// are promoted and implicitly truncated. Make that explicit here.
if (Op.getValueType() != SrcEltVT)
return ConstantFoldBITCASTofBUILD_VECTOR(Tmp, DstEltVT);
}
+ SDLoc DL(BV);
+
// Okay, we know the src/dst types are both integers of differing types.
// Handling growing first.
assert(SrcEltVT.isInteger() && DstEltVT.isInteger());
SmallVector<SDValue, 8> Ops;
for (unsigned i = 0, e = BV->getNumOperands(); i != e;
i += NumInputsPerOutput) {
- bool isLE = TLI.isLittleEndian();
+ bool isLE = DAG.getDataLayout().isLittleEndian();
APInt NewBits = APInt(DstBitSize, 0);
bool EltIsUndef = true;
for (unsigned j = 0; j != NumInputsPerOutput; ++j) {
if (EltIsUndef)
Ops.push_back(DAG.getUNDEF(DstEltVT));
else
- Ops.push_back(DAG.getConstant(NewBits, DstEltVT));
+ Ops.push_back(DAG.getConstant(NewBits, DL, DstEltVT));
}
EVT VT = EVT::getVectorVT(*DAG.getContext(), DstEltVT, Ops.size());
- return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(BV), VT, Ops);
+ return DAG.getNode(ISD::BUILD_VECTOR, DL, VT, Ops);
}
// Finally, this must be the case where we are shrinking elements: each input
NumOutputsPerInput*BV->getNumOperands());
SmallVector<SDValue, 8> Ops;
- for (unsigned i = 0, e = BV->getNumOperands(); i != e; ++i) {
- if (BV->getOperand(i).getOpcode() == ISD::UNDEF) {
+ for (const SDValue &Op : BV->op_values()) {
+ if (Op.getOpcode() == ISD::UNDEF) {
Ops.append(NumOutputsPerInput, DAG.getUNDEF(DstEltVT));
continue;
}
- APInt OpVal = cast<ConstantSDNode>(BV->getOperand(i))->
+ APInt OpVal = cast<ConstantSDNode>(Op)->
getAPIntValue().zextOrTrunc(SrcBitSize);
for (unsigned j = 0; j != NumOutputsPerInput; ++j) {
APInt ThisVal = OpVal.trunc(DstBitSize);
- Ops.push_back(DAG.getConstant(ThisVal, DstEltVT));
+ Ops.push_back(DAG.getConstant(ThisVal, DL, DstEltVT));
OpVal = OpVal.lshr(DstBitSize);
}
// For big endian targets, swap the order of the pieces of each element.
- if (TLI.isBigEndian())
+ if (DAG.getDataLayout().isBigEndian())
std::reverse(Ops.end()-NumOutputsPerInput, Ops.end());
}
- return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(BV), VT, Ops);
+ return DAG.getNode(ISD::BUILD_VECTOR, DL, VT, Ops);
}
-// Attempt different variants of (fadd (fmul a, b), c) -> fma or fmad
-static SDValue performFaddFmulCombines(unsigned FusedOpcode,
- bool Aggressive,
- SDNode *N,
- const TargetLowering &TLI,
- SelectionDAG &DAG) {
+/// Try to perform FMA combining on a given FADD node.
+SDValue DAGCombiner::visitFADDForFMACombine(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
EVT VT = N->getValueType(0);
+ SDLoc SL(N);
+
+ const TargetOptions &Options = DAG.getTarget().Options;
+ bool AllowFusion =
+ (Options.AllowFPOpFusion == FPOpFusion::Fast || Options.UnsafeFPMath);
+
+ // Floating-point multiply-add with intermediate rounding.
+ bool HasFMAD = (LegalOperations && TLI.isOperationLegal(ISD::FMAD, VT));
+
+ // Floating-point multiply-add without intermediate rounding.
+ bool HasFMA =
+ AllowFusion && TLI.isFMAFasterThanFMulAndFAdd(VT) &&
+ (!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FMA, VT));
+
+ // No valid opcode, do not combine.
+ if (!HasFMAD && !HasFMA)
+ return SDValue();
+
+ // Always prefer FMAD to FMA for precision.
+ unsigned PreferredFusedOpcode = HasFMAD ? ISD::FMAD : ISD::FMA;
+ bool Aggressive = TLI.enableAggressiveFMAFusion(VT);
+ bool LookThroughFPExt = TLI.isFPExtFree(VT);
+
+ // If we have two choices trying to fold (fadd (fmul u, v), (fmul x, y)),
+ // prefer to fold the multiply with fewer uses.
+ if (Aggressive && N0.getOpcode() == ISD::FMUL &&
+ N1.getOpcode() == ISD::FMUL) {
+ if (N0.getNode()->use_size() > N1.getNode()->use_size())
+ std::swap(N0, N1);
+ }
// fold (fadd (fmul x, y), z) -> (fma x, y, z)
if (N0.getOpcode() == ISD::FMUL &&
(Aggressive || N0->hasOneUse())) {
- return DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
N0.getOperand(0), N0.getOperand(1), N1);
}
// Note: Commutes FADD operands.
if (N1.getOpcode() == ISD::FMUL &&
(Aggressive || N1->hasOneUse())) {
- return DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
N1.getOperand(0), N1.getOperand(1), N0);
}
+ // Look through FP_EXTEND nodes to do more combining.
+ if (AllowFusion && LookThroughFPExt) {
+ // fold (fadd (fpext (fmul x, y)), z) -> (fma (fpext x), (fpext y), z)
+ if (N0.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == ISD::FMUL)
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N00.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N00.getOperand(1)), N1);
+ }
+
+ // fold (fadd x, (fpext (fmul y, z))) -> (fma (fpext y), (fpext z), x)
+ // Note: Commutes FADD operands.
+ if (N1.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N10 = N1.getOperand(0);
+ if (N10.getOpcode() == ISD::FMUL)
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N10.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N10.getOperand(1)), N0);
+ }
+ }
+
// More folding opportunities when target permits.
- if (Aggressive) {
+ if ((AllowFusion || HasFMAD) && Aggressive) {
// fold (fadd (fma x, y, (fmul u, v)), z) -> (fma x, y (fma u, v, z))
- if (N0.getOpcode() == ISD::FMA &&
+ if (N0.getOpcode() == PreferredFusedOpcode &&
N0.getOperand(2).getOpcode() == ISD::FMUL) {
- return DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
N0.getOperand(0), N0.getOperand(1),
- DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
N0.getOperand(2).getOperand(0),
N0.getOperand(2).getOperand(1),
N1));
}
// fold (fadd x, (fma y, z, (fmul u, v)) -> (fma y, z (fma u, v, x))
- if (N1->getOpcode() == ISD::FMA &&
+ if (N1->getOpcode() == PreferredFusedOpcode &&
N1.getOperand(2).getOpcode() == ISD::FMUL) {
- return DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
N1.getOperand(0), N1.getOperand(1),
- DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
N1.getOperand(2).getOperand(0),
N1.getOperand(2).getOperand(1),
N0));
}
+
+ if (AllowFusion && LookThroughFPExt) {
+ // fold (fadd (fma x, y, (fpext (fmul u, v))), z)
+ // -> (fma x, y, (fma (fpext u), (fpext v), z))
+ auto FoldFAddFMAFPExtFMul = [&] (
+ SDValue X, SDValue Y, SDValue U, SDValue V, SDValue Z) {
+ return DAG.getNode(PreferredFusedOpcode, SL, VT, X, Y,
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT, U),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT, V),
+ Z));
+ };
+ if (N0.getOpcode() == PreferredFusedOpcode) {
+ SDValue N02 = N0.getOperand(2);
+ if (N02.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N020 = N02.getOperand(0);
+ if (N020.getOpcode() == ISD::FMUL)
+ return FoldFAddFMAFPExtFMul(N0.getOperand(0), N0.getOperand(1),
+ N020.getOperand(0), N020.getOperand(1),
+ N1);
+ }
+ }
+
+ // fold (fadd (fpext (fma x, y, (fmul u, v))), z)
+ // -> (fma (fpext x), (fpext y), (fma (fpext u), (fpext v), z))
+ // FIXME: This turns two single-precision and one double-precision
+ // operation into two double-precision operations, which might not be
+ // interesting for all targets, especially GPUs.
+ auto FoldFAddFPExtFMAFMul = [&] (
+ SDValue X, SDValue Y, SDValue U, SDValue V, SDValue Z) {
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT, X),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT, Y),
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT, U),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT, V),
+ Z));
+ };
+ if (N0.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == PreferredFusedOpcode) {
+ SDValue N002 = N00.getOperand(2);
+ if (N002.getOpcode() == ISD::FMUL)
+ return FoldFAddFPExtFMAFMul(N00.getOperand(0), N00.getOperand(1),
+ N002.getOperand(0), N002.getOperand(1),
+ N1);
+ }
+ }
+
+ // fold (fadd x, (fma y, z, (fpext (fmul u, v)))
+ // -> (fma y, z, (fma (fpext u), (fpext v), x))
+ if (N1.getOpcode() == PreferredFusedOpcode) {
+ SDValue N12 = N1.getOperand(2);
+ if (N12.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N120 = N12.getOperand(0);
+ if (N120.getOpcode() == ISD::FMUL)
+ return FoldFAddFMAFPExtFMul(N1.getOperand(0), N1.getOperand(1),
+ N120.getOperand(0), N120.getOperand(1),
+ N0);
+ }
+ }
+
+ // fold (fadd x, (fpext (fma y, z, (fmul u, v)))
+ // -> (fma (fpext y), (fpext z), (fma (fpext u), (fpext v), x))
+ // FIXME: This turns two single-precision and one double-precision
+ // operation into two double-precision operations, which might not be
+ // interesting for all targets, especially GPUs.
+ if (N1.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N10 = N1.getOperand(0);
+ if (N10.getOpcode() == PreferredFusedOpcode) {
+ SDValue N102 = N10.getOperand(2);
+ if (N102.getOpcode() == ISD::FMUL)
+ return FoldFAddFPExtFMAFMul(N10.getOperand(0), N10.getOperand(1),
+ N102.getOperand(0), N102.getOperand(1),
+ N0);
+ }
+ }
+ }
}
return SDValue();
}
-static SDValue performFsubFmulCombines(unsigned FusedOpcode,
- bool Aggressive,
- SDNode *N,
- const TargetLowering &TLI,
- SelectionDAG &DAG) {
+/// Try to perform FMA combining on a given FSUB node.
+SDValue DAGCombiner::visitFSUBForFMACombine(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
EVT VT = N->getValueType(0);
-
SDLoc SL(N);
+ const TargetOptions &Options = DAG.getTarget().Options;
+ bool AllowFusion =
+ (Options.AllowFPOpFusion == FPOpFusion::Fast || Options.UnsafeFPMath);
+
+ // Floating-point multiply-add with intermediate rounding.
+ bool HasFMAD = (LegalOperations && TLI.isOperationLegal(ISD::FMAD, VT));
+
+ // Floating-point multiply-add without intermediate rounding.
+ bool HasFMA =
+ AllowFusion && TLI.isFMAFasterThanFMulAndFAdd(VT) &&
+ (!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FMA, VT));
+
+ // No valid opcode, do not combine.
+ if (!HasFMAD && !HasFMA)
+ return SDValue();
+
+ // Always prefer FMAD to FMA for precision.
+ unsigned PreferredFusedOpcode = HasFMAD ? ISD::FMAD : ISD::FMA;
+ bool Aggressive = TLI.enableAggressiveFMAFusion(VT);
+ bool LookThroughFPExt = TLI.isFPExtFree(VT);
+
// fold (fsub (fmul x, y), z) -> (fma x, y, (fneg z))
if (N0.getOpcode() == ISD::FMUL &&
(Aggressive || N0->hasOneUse())) {
- return DAG.getNode(FusedOpcode, SL, VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
N0.getOperand(0), N0.getOperand(1),
DAG.getNode(ISD::FNEG, SL, VT, N1));
}
// Note: Commutes FSUB operands.
if (N1.getOpcode() == ISD::FMUL &&
(Aggressive || N1->hasOneUse()))
- return DAG.getNode(FusedOpcode, SL, VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
DAG.getNode(ISD::FNEG, SL, VT,
N1.getOperand(0)),
N1.getOperand(1), N0);
(Aggressive || (N0->hasOneUse() && N0.getOperand(0).hasOneUse()))) {
SDValue N00 = N0.getOperand(0).getOperand(0);
SDValue N01 = N0.getOperand(0).getOperand(1);
- return DAG.getNode(FusedOpcode, SL, VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
DAG.getNode(ISD::FNEG, SL, VT, N00), N01,
DAG.getNode(ISD::FNEG, SL, VT, N1));
}
+ // Look through FP_EXTEND nodes to do more combining.
+ if (AllowFusion && LookThroughFPExt) {
+ // fold (fsub (fpext (fmul x, y)), z)
+ // -> (fma (fpext x), (fpext y), (fneg z))
+ if (N0.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == ISD::FMUL)
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N00.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N00.getOperand(1)),
+ DAG.getNode(ISD::FNEG, SL, VT, N1));
+ }
+
+ // fold (fsub x, (fpext (fmul y, z)))
+ // -> (fma (fneg (fpext y)), (fpext z), x)
+ // Note: Commutes FSUB operands.
+ if (N1.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N10 = N1.getOperand(0);
+ if (N10.getOpcode() == ISD::FMUL)
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N10.getOperand(0))),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N10.getOperand(1)),
+ N0);
+ }
+
+ // fold (fsub (fpext (fneg (fmul, x, y))), z)
+ // -> (fneg (fma (fpext x), (fpext y), z))
+ // Note: This could be removed with appropriate canonicalization of the
+ // input expression into (fneg (fadd (fpext (fmul, x, y)), z). However, the
+ // orthogonal flags -fp-contract=fast and -enable-unsafe-fp-math prevent
+ // from implementing the canonicalization in visitFSUB.
+ if (N0.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == ISD::FNEG) {
+ SDValue N000 = N00.getOperand(0);
+ if (N000.getOpcode() == ISD::FMUL) {
+ return DAG.getNode(ISD::FNEG, SL, VT,
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N000.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N000.getOperand(1)),
+ N1));
+ }
+ }
+ }
+
+ // fold (fsub (fneg (fpext (fmul, x, y))), z)
+ // -> (fneg (fma (fpext x)), (fpext y), z)
+ // Note: This could be removed with appropriate canonicalization of the
+ // input expression into (fneg (fadd (fpext (fmul, x, y)), z). However, the
+ // orthogonal flags -fp-contract=fast and -enable-unsafe-fp-math prevent
+ // from implementing the canonicalization in visitFSUB.
+ if (N0.getOpcode() == ISD::FNEG) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N000 = N00.getOperand(0);
+ if (N000.getOpcode() == ISD::FMUL) {
+ return DAG.getNode(ISD::FNEG, SL, VT,
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N000.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N000.getOperand(1)),
+ N1));
+ }
+ }
+ }
+
+ }
+
// More folding opportunities when target permits.
- if (Aggressive) {
+ if ((AllowFusion || HasFMAD) && Aggressive) {
// fold (fsub (fma x, y, (fmul u, v)), z)
// -> (fma x, y (fma u, v, (fneg z)))
- if (N0.getOpcode() == FusedOpcode &&
+ if (N0.getOpcode() == PreferredFusedOpcode &&
N0.getOperand(2).getOpcode() == ISD::FMUL) {
- return DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
N0.getOperand(0), N0.getOperand(1),
- DAG.getNode(FusedOpcode, SDLoc(N), VT,
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
N0.getOperand(2).getOperand(0),
N0.getOperand(2).getOperand(1),
- DAG.getNode(ISD::FNEG, SDLoc(N), VT,
+ DAG.getNode(ISD::FNEG, SL, VT,
N1)));
}
// fold (fsub x, (fma y, z, (fmul u, v)))
// -> (fma (fneg y), z, (fma (fneg u), v, x))
- if (N1.getOpcode() == FusedOpcode &&
+ if (N1.getOpcode() == PreferredFusedOpcode &&
N1.getOperand(2).getOpcode() == ISD::FMUL) {
SDValue N20 = N1.getOperand(2).getOperand(0);
SDValue N21 = N1.getOperand(2).getOperand(1);
- return DAG.getNode(FusedOpcode, SDLoc(N), VT,
- DAG.getNode(ISD::FNEG, SDLoc(N), VT,
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT,
N1.getOperand(0)),
N1.getOperand(1),
- DAG.getNode(FusedOpcode, SDLoc(N), VT,
- DAG.getNode(ISD::FNEG, SDLoc(N), VT,
- N20),
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT, N20),
+
N21, N0));
}
+
+ if (AllowFusion && LookThroughFPExt) {
+ // fold (fsub (fma x, y, (fpext (fmul u, v))), z)
+ // -> (fma x, y (fma (fpext u), (fpext v), (fneg z)))
+ if (N0.getOpcode() == PreferredFusedOpcode) {
+ SDValue N02 = N0.getOperand(2);
+ if (N02.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N020 = N02.getOperand(0);
+ if (N020.getOpcode() == ISD::FMUL)
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ N0.getOperand(0), N0.getOperand(1),
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N020.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N020.getOperand(1)),
+ DAG.getNode(ISD::FNEG, SL, VT,
+ N1)));
+ }
+ }
+
+ // fold (fsub (fpext (fma x, y, (fmul u, v))), z)
+ // -> (fma (fpext x), (fpext y),
+ // (fma (fpext u), (fpext v), (fneg z)))
+ // FIXME: This turns two single-precision and one double-precision
+ // operation into two double-precision operations, which might not be
+ // interesting for all targets, especially GPUs.
+ if (N0.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == PreferredFusedOpcode) {
+ SDValue N002 = N00.getOperand(2);
+ if (N002.getOpcode() == ISD::FMUL)
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N00.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N00.getOperand(1)),
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N002.getOperand(0)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N002.getOperand(1)),
+ DAG.getNode(ISD::FNEG, SL, VT,
+ N1)));
+ }
+ }
+
+ // fold (fsub x, (fma y, z, (fpext (fmul u, v))))
+ // -> (fma (fneg y), z, (fma (fneg (fpext u)), (fpext v), x))
+ if (N1.getOpcode() == PreferredFusedOpcode &&
+ N1.getOperand(2).getOpcode() == ISD::FP_EXTEND) {
+ SDValue N120 = N1.getOperand(2).getOperand(0);
+ if (N120.getOpcode() == ISD::FMUL) {
+ SDValue N1200 = N120.getOperand(0);
+ SDValue N1201 = N120.getOperand(1);
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT, N1.getOperand(0)),
+ N1.getOperand(1),
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL,
+ VT, N1200)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N1201),
+ N0));
+ }
+ }
+
+ // fold (fsub x, (fpext (fma y, z, (fmul u, v))))
+ // -> (fma (fneg (fpext y)), (fpext z),
+ // (fma (fneg (fpext u)), (fpext v), x))
+ // FIXME: This turns two single-precision and one double-precision
+ // operation into two double-precision operations, which might not be
+ // interesting for all targets, especially GPUs.
+ if (N1.getOpcode() == ISD::FP_EXTEND &&
+ N1.getOperand(0).getOpcode() == PreferredFusedOpcode) {
+ SDValue N100 = N1.getOperand(0).getOperand(0);
+ SDValue N101 = N1.getOperand(0).getOperand(1);
+ SDValue N102 = N1.getOperand(0).getOperand(2);
+ if (N102.getOpcode() == ISD::FMUL) {
+ SDValue N1020 = N102.getOperand(0);
+ SDValue N1021 = N102.getOperand(1);
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N100)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT, N101),
+ DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT,
+ DAG.getNode(ISD::FP_EXTEND, SL,
+ VT, N1020)),
+ DAG.getNode(ISD::FP_EXTEND, SL, VT,
+ N1021),
+ N0));
+ }
+ }
+ }
}
return SDValue();
}
+/// Try to perform FMA combining on a given FMUL node.
+SDValue DAGCombiner::visitFMULForFMACombine(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+ SDValue N1 = N->getOperand(1);
+ EVT VT = N->getValueType(0);
+ SDLoc SL(N);
+
+ assert(N->getOpcode() == ISD::FMUL && "Expected FMUL Operation");
+
+ const TargetOptions &Options = DAG.getTarget().Options;
+ bool AllowFusion =
+ (Options.AllowFPOpFusion == FPOpFusion::Fast || Options.UnsafeFPMath);
+
+ // Floating-point multiply-add with intermediate rounding.
+ bool HasFMAD = (LegalOperations && TLI.isOperationLegal(ISD::FMAD, VT));
+
+ // Floating-point multiply-add without intermediate rounding.
+ bool HasFMA =
+ AllowFusion && TLI.isFMAFasterThanFMulAndFAdd(VT) &&
+ (!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FMA, VT));
+
+ // No valid opcode, do not combine.
+ if (!HasFMAD && !HasFMA)
+ return SDValue();
+
+ // Always prefer FMAD to FMA for precision.
+ unsigned PreferredFusedOpcode = HasFMAD ? ISD::FMAD : ISD::FMA;
+ bool Aggressive = TLI.enableAggressiveFMAFusion(VT);
+
+ // fold (fmul (fadd x, +1.0), y) -> (fma x, y, y)
+ // fold (fmul (fadd x, -1.0), y) -> (fma x, y, (fneg y))
+ auto FuseFADD = [&](SDValue X, SDValue Y) {
+ if (X.getOpcode() == ISD::FADD && (Aggressive || X->hasOneUse())) {
+ auto XC1 = isConstOrConstSplatFP(X.getOperand(1));
+ if (XC1 && XC1->isExactlyValue(+1.0))
+ return DAG.getNode(PreferredFusedOpcode, SL, VT, X.getOperand(0), Y, Y);
+ if (XC1 && XC1->isExactlyValue(-1.0))
+ return DAG.getNode(PreferredFusedOpcode, SL, VT, X.getOperand(0), Y,
+ DAG.getNode(ISD::FNEG, SL, VT, Y));
+ }
+ return SDValue();
+ };
+
+ if (SDValue FMA = FuseFADD(N0, N1))
+ return FMA;
+ if (SDValue FMA = FuseFADD(N1, N0))
+ return FMA;
+
+ // fold (fmul (fsub +1.0, x), y) -> (fma (fneg x), y, y)
+ // fold (fmul (fsub -1.0, x), y) -> (fma (fneg x), y, (fneg y))
+ // fold (fmul (fsub x, +1.0), y) -> (fma x, y, (fneg y))
+ // fold (fmul (fsub x, -1.0), y) -> (fma x, y, y)
+ auto FuseFSUB = [&](SDValue X, SDValue Y) {
+ if (X.getOpcode() == ISD::FSUB && (Aggressive || X->hasOneUse())) {
+ auto XC0 = isConstOrConstSplatFP(X.getOperand(0));
+ if (XC0 && XC0->isExactlyValue(+1.0))
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT, X.getOperand(1)), Y,
+ Y);
+ if (XC0 && XC0->isExactlyValue(-1.0))
+ return DAG.getNode(PreferredFusedOpcode, SL, VT,
+ DAG.getNode(ISD::FNEG, SL, VT, X.getOperand(1)), Y,
+ DAG.getNode(ISD::FNEG, SL, VT, Y));
+
+ auto XC1 = isConstOrConstSplatFP(X.getOperand(1));
+ if (XC1 && XC1->isExactlyValue(+1.0))
+ return DAG.getNode(PreferredFusedOpcode, SL, VT, X.getOperand(0), Y,
+ DAG.getNode(ISD::FNEG, SL, VT, Y));
+ if (XC1 && XC1->isExactlyValue(-1.0))
+ return DAG.getNode(PreferredFusedOpcode, SL, VT, X.getOperand(0), Y, Y);
+ }
+ return SDValue();
+ };
+
+ if (SDValue FMA = FuseFSUB(N0, N1))
+ return FMA;
+ if (SDValue FMA = FuseFSUB(N1, N0))
+ return FMA;
+
+ return SDValue();
+}
+
SDValue DAGCombiner::visitFADD(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
- ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
- ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1);
+ bool N0CFP = isConstantFPBuildVectorOrConstantFP(N0);
+ bool N1CFP = isConstantFPBuildVectorOrConstantFP(N1);
EVT VT = N->getValueType(0);
+ SDLoc DL(N);
const TargetOptions &Options = DAG.getTarget().Options;
+ const SDNodeFlags *Flags = &cast<BinaryWithFlagsSDNode>(N)->Flags;
// fold vector ops
if (VT.isVector())
// fold (fadd c1, c2) -> c1 + c2
if (N0CFP && N1CFP)
- return DAG.getNode(ISD::FADD, SDLoc(N), VT, N0, N1);
+ return DAG.getNode(ISD::FADD, DL, VT, N0, N1, Flags);
// canonicalize constant to RHS
if (N0CFP && !N1CFP)
- return DAG.getNode(ISD::FADD, SDLoc(N), VT, N1, N0);
+ return DAG.getNode(ISD::FADD, DL, VT, N1, N0, Flags);
// fold (fadd A, (fneg B)) -> (fsub A, B)
if ((!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FSUB, VT)) &&
isNegatibleForFree(N1, LegalOperations, TLI, &Options) == 2)
- return DAG.getNode(ISD::FSUB, SDLoc(N), VT, N0,
- GetNegatedExpression(N1, DAG, LegalOperations));
+ return DAG.getNode(ISD::FSUB, DL, VT, N0,
+ GetNegatedExpression(N1, DAG, LegalOperations), Flags);
// fold (fadd (fneg A), B) -> (fsub B, A)
if ((!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FSUB, VT)) &&
isNegatibleForFree(N0, LegalOperations, TLI, &Options) == 2)
- return DAG.getNode(ISD::FSUB, SDLoc(N), VT, N1,
- GetNegatedExpression(N0, DAG, LegalOperations));
+ return DAG.getNode(ISD::FSUB, DL, VT, N1,
+ GetNegatedExpression(N0, DAG, LegalOperations), Flags);
// If 'unsafe math' is enabled, fold lots of things.
if (Options.UnsafeFPMath) {
bool AllowNewConst = (Level < AfterLegalizeDAG);
// fold (fadd A, 0) -> A
- if (N1CFP && N1CFP->getValueAPF().isZero())
- return N0;
+ if (ConstantFPSDNode *N1C = isConstOrConstSplatFP(N1))
+ if (N1C->isZero())
+ return N0;
// fold (fadd (fadd x, c1), c2) -> (fadd x, (fadd c1, c2))
if (N1CFP && N0.getOpcode() == ISD::FADD && N0.getNode()->hasOneUse() &&
- isa<ConstantFPSDNode>(N0.getOperand(1)))
- return DAG.getNode(ISD::FADD, SDLoc(N), VT, N0.getOperand(0),
- DAG.getNode(ISD::FADD, SDLoc(N), VT,
- N0.getOperand(1), N1));
+ isConstantFPBuildVectorOrConstantFP(N0.getOperand(1)))
+ return DAG.getNode(ISD::FADD, DL, VT, N0.getOperand(0),
+ DAG.getNode(ISD::FADD, DL, VT, N0.getOperand(1), N1,
+ Flags),
+ Flags);
// If allowed, fold (fadd (fneg x), x) -> 0.0
if (AllowNewConst && N0.getOpcode() == ISD::FNEG && N0.getOperand(0) == N1)
- return DAG.getConstantFP(0.0, VT);
+ return DAG.getConstantFP(0.0, DL, VT);
// If allowed, fold (fadd x, (fneg x)) -> 0.0
if (AllowNewConst && N1.getOpcode() == ISD::FNEG && N1.getOperand(0) == N0)
- return DAG.getConstantFP(0.0, VT);
+ return DAG.getConstantFP(0.0, DL, VT);
// We can fold chains of FADD's of the same value into multiplications.
// This transform is not safe in general because we are reducing the number
// of rounding steps.
if (TLI.isOperationLegalOrCustom(ISD::FMUL, VT) && !N0CFP && !N1CFP) {
if (N0.getOpcode() == ISD::FMUL) {
- ConstantFPSDNode *CFP00 = dyn_cast<ConstantFPSDNode>(N0.getOperand(0));
- ConstantFPSDNode *CFP01 = dyn_cast<ConstantFPSDNode>(N0.getOperand(1));
+ bool CFP00 = isConstantFPBuildVectorOrConstantFP(N0.getOperand(0));
+ bool CFP01 = isConstantFPBuildVectorOrConstantFP(N0.getOperand(1));
// (fadd (fmul x, c), x) -> (fmul x, c+1)
if (CFP01 && !CFP00 && N0.getOperand(0) == N1) {
- SDValue NewCFP = DAG.getNode(ISD::FADD, SDLoc(N), VT,
- SDValue(CFP01, 0),
- DAG.getConstantFP(1.0, VT));
- return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N1, NewCFP);
+ SDValue NewCFP = DAG.getNode(ISD::FADD, DL, VT, N0.getOperand(1),
+ DAG.getConstantFP(1.0, DL, VT), Flags);
+ return DAG.getNode(ISD::FMUL, DL, VT, N1, NewCFP, Flags);
}
// (fadd (fmul x, c), (fadd x, x)) -> (fmul x, c+2)
if (CFP01 && !CFP00 && N1.getOpcode() == ISD::FADD &&
N1.getOperand(0) == N1.getOperand(1) &&
N0.getOperand(0) == N1.getOperand(0)) {
- SDValue NewCFP = DAG.getNode(ISD::FADD, SDLoc(N), VT,
- SDValue(CFP01, 0),
- DAG.getConstantFP(2.0, VT));
- return DAG.getNode(ISD::FMUL, SDLoc(N), VT,
- N0.getOperand(0), NewCFP);
+ SDValue NewCFP = DAG.getNode(ISD::FADD, DL, VT, N0.getOperand(1),
+ DAG.getConstantFP(2.0, DL, VT), Flags);
+ return DAG.getNode(ISD::FMUL, DL, VT, N0.getOperand(0), NewCFP, Flags);
}
}
if (N1.getOpcode() == ISD::FMUL) {
- ConstantFPSDNode *CFP10 = dyn_cast<ConstantFPSDNode>(N1.getOperand(0));
- ConstantFPSDNode *CFP11 = dyn_cast<ConstantFPSDNode>(N1.getOperand(1));
+ bool CFP10 = isConstantFPBuildVectorOrConstantFP(N1.getOperand(0));
+ bool CFP11 = isConstantFPBuildVectorOrConstantFP(N1.getOperand(1));
// (fadd x, (fmul x, c)) -> (fmul x, c+1)
if (CFP11 && !CFP10 && N1.getOperand(0) == N0) {
- SDValue NewCFP = DAG.getNode(ISD::FADD, SDLoc(N), VT,
- SDValue(CFP11, 0),
- DAG.getConstantFP(1.0, VT));
- return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N0, NewCFP);
+ SDValue NewCFP = DAG.getNode(ISD::FADD, DL, VT, N1.getOperand(1),
+ DAG.getConstantFP(1.0, DL, VT), Flags);
+ return DAG.getNode(ISD::FMUL, DL, VT, N0, NewCFP, Flags);
}
// (fadd (fadd x, x), (fmul x, c)) -> (fmul x, c+2)
if (CFP11 && !CFP10 && N0.getOpcode() == ISD::FADD &&
N0.getOperand(0) == N0.getOperand(1) &&
N1.getOperand(0) == N0.getOperand(0)) {
- SDValue NewCFP = DAG.getNode(ISD::FADD, SDLoc(N), VT,
- SDValue(CFP11, 0),
- DAG.getConstantFP(2.0, VT));
- return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N1.getOperand(0), NewCFP);
+ SDValue NewCFP = DAG.getNode(ISD::FADD, DL, VT, N1.getOperand(1),
+ DAG.getConstantFP(2.0, DL, VT), Flags);
+ return DAG.getNode(ISD::FMUL, DL, VT, N1.getOperand(0), NewCFP, Flags);
}
}
if (N0.getOpcode() == ISD::FADD && AllowNewConst) {
- ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(N0.getOperand(0));
+ bool CFP00 = isConstantFPBuildVectorOrConstantFP(N0.getOperand(0));
// (fadd (fadd x, x), x) -> (fmul x, 3.0)
- if (!CFP && N0.getOperand(0) == N0.getOperand(1) &&
- (N0.getOperand(0) == N1))
- return DAG.getNode(ISD::FMUL, SDLoc(N), VT,
- N1, DAG.getConstantFP(3.0, VT));
+ if (!CFP00 && N0.getOperand(0) == N0.getOperand(1) &&
+ (N0.getOperand(0) == N1)) {
+ return DAG.getNode(ISD::FMUL, DL, VT,
+ N1, DAG.getConstantFP(3.0, DL, VT), Flags);
+ }
}
if (N1.getOpcode() == ISD::FADD && AllowNewConst) {
- ConstantFPSDNode *CFP10 = dyn_cast<ConstantFPSDNode>(N1.getOperand(0));
+ bool CFP10 = isConstantFPBuildVectorOrConstantFP(N1.getOperand(0));
// (fadd x, (fadd x, x)) -> (fmul x, 3.0)
if (!CFP10 && N1.getOperand(0) == N1.getOperand(1) &&
- N1.getOperand(0) == N0)
- return DAG.getNode(ISD::FMUL, SDLoc(N), VT,
- N0, DAG.getConstantFP(3.0, VT));
+ N1.getOperand(0) == N0) {
+ return DAG.getNode(ISD::FMUL, DL, VT,
+ N0, DAG.getConstantFP(3.0, DL, VT), Flags);
+ }
}
// (fadd (fadd x, x), (fadd x, x)) -> (fmul x, 4.0)
N0.getOpcode() == ISD::FADD && N1.getOpcode() == ISD::FADD &&
N0.getOperand(0) == N0.getOperand(1) &&
N1.getOperand(0) == N1.getOperand(1) &&
- N0.getOperand(0) == N1.getOperand(0))
- return DAG.getNode(ISD::FMUL, SDLoc(N), VT,
- N0.getOperand(0), DAG.getConstantFP(4.0, VT));
+ N0.getOperand(0) == N1.getOperand(0)) {
+ return DAG.getNode(ISD::FMUL, DL, VT, N0.getOperand(0),
+ DAG.getConstantFP(4.0, DL, VT), Flags);
+ }
}
} // enable-unsafe-fp-math
- if (LegalOperations && TLI.isOperationLegal(ISD::FMAD, VT)) {
- // Assume if there is an fmad instruction that it should be aggressively
- // used.
- if (SDValue Fused = performFaddFmulCombines(ISD::FMAD, true, N, TLI, DAG))
- return Fused;
- }
-
// FADD -> FMA combines:
- if ((Options.AllowFPOpFusion == FPOpFusion::Fast || Options.UnsafeFPMath) &&
- TLI.isFMAFasterThanFMulAndFAdd(VT) &&
- (!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FMA, VT))) {
-
- if (!TLI.isOperationLegal(ISD::FMAD, VT)) {
- // Don't form FMA if we are preferring FMAD.
- if (SDValue Fused
- = performFaddFmulCombines(ISD::FMA,
- TLI.enableAggressiveFMAFusion(VT),
- N, TLI, DAG)) {
- return Fused;
- }
- }
-
- // When FP_EXTEND nodes are free on the target, and there is an opportunity
- // to combine into FMA, arrange such nodes accordingly.
- if (TLI.isFPExtFree(VT)) {
-
- // fold (fadd (fpext (fmul x, y)), z) -> (fma (fpext x), (fpext y), z)
- if (N0.getOpcode() == ISD::FP_EXTEND) {
- SDValue N00 = N0.getOperand(0);
- if (N00.getOpcode() == ISD::FMUL)
- return DAG.getNode(ISD::FMA, SDLoc(N), VT,
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N00.getOperand(0)),
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N00.getOperand(1)), N1);
- }
-
- // fold (fadd x, (fpext (fmul y, z)), z) -> (fma (fpext y), (fpext z), x)
- // Note: Commutes FADD operands.
- if (N1.getOpcode() == ISD::FP_EXTEND) {
- SDValue N10 = N1.getOperand(0);
- if (N10.getOpcode() == ISD::FMUL)
- return DAG.getNode(ISD::FMA, SDLoc(N), VT,
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N10.getOperand(0)),
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N10.getOperand(1)), N0);
- }
- }
+ if (SDValue Fused = visitFADDForFMACombine(N)) {
+ AddToWorklist(Fused.getNode());
+ return Fused;
}
return SDValue();
EVT VT = N->getValueType(0);
SDLoc dl(N);
const TargetOptions &Options = DAG.getTarget().Options;
+ const SDNodeFlags *Flags = &cast<BinaryWithFlagsSDNode>(N)->Flags;
// fold vector ops
if (VT.isVector())
// fold (fsub c1, c2) -> c1-c2
if (N0CFP && N1CFP)
- return DAG.getNode(ISD::FSUB, SDLoc(N), VT, N0, N1);
+ return DAG.getNode(ISD::FSUB, dl, VT, N0, N1, Flags);
// fold (fsub A, (fneg B)) -> (fadd A, B)
if (isNegatibleForFree(N1, LegalOperations, TLI, &Options))
return DAG.getNode(ISD::FADD, dl, VT, N0,
- GetNegatedExpression(N1, DAG, LegalOperations));
+ GetNegatedExpression(N1, DAG, LegalOperations), Flags);
// 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))
// (fsub x, x) -> 0.0
if (N0 == N1)
- return DAG.getConstantFP(0.0f, VT);
+ return DAG.getConstantFP(0.0f, dl, VT);
// (fsub x, (fadd x, y)) -> (fneg y)
// (fsub x, (fadd y, x)) -> (fneg y)
SDValue N10 = N1->getOperand(0);
SDValue N11 = N1->getOperand(1);
- if (N10 == N0 && isNegatibleForFree(N11, LegalOperations, TLI, &Options))
- return GetNegatedExpression(N11, DAG, LegalOperations);
-
- if (N11 == N0 && isNegatibleForFree(N10, LegalOperations, TLI, &Options))
- return GetNegatedExpression(N10, DAG, LegalOperations);
- }
- }
-
- if (LegalOperations && TLI.isOperationLegal(ISD::FMAD, VT)) {
- // Assume if there is an fmad instruction that it should be aggressively
- // used.
- if (SDValue Fused = performFsubFmulCombines(ISD::FMAD, true, N, TLI, DAG))
- return Fused;
- }
-
- // FSUB -> FMA combines:
- if ((Options.AllowFPOpFusion == FPOpFusion::Fast || Options.UnsafeFPMath) &&
- TLI.isFMAFasterThanFMulAndFAdd(VT) &&
- (!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FMA, VT))) {
-
- if (!TLI.isOperationLegal(ISD::FMAD, VT)) {
- // Don't form FMA if we are preferring FMAD.
-
- if (SDValue Fused
- = performFsubFmulCombines(ISD::FMA,
- TLI.enableAggressiveFMAFusion(VT),
- N, TLI, DAG)) {
- return Fused;
- }
- }
-
- // When FP_EXTEND nodes are free on the target, and there is an opportunity
- // to combine into FMA, arrange such nodes accordingly.
- if (TLI.isFPExtFree(VT)) {
- // fold (fsub (fpext (fmul x, y)), z)
- // -> (fma (fpext x), (fpext y), (fneg z))
- if (N0.getOpcode() == ISD::FP_EXTEND) {
- SDValue N00 = N0.getOperand(0);
- if (N00.getOpcode() == ISD::FMUL)
- return DAG.getNode(ISD::FMA, SDLoc(N), VT,
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N00.getOperand(0)),
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N00.getOperand(1)),
- DAG.getNode(ISD::FNEG, SDLoc(N), VT, N1));
- }
-
- // fold (fsub x, (fpext (fmul y, z)))
- // -> (fma (fneg (fpext y)), (fpext z), x)
- // Note: Commutes FSUB operands.
- if (N1.getOpcode() == ISD::FP_EXTEND) {
- SDValue N10 = N1.getOperand(0);
- if (N10.getOpcode() == ISD::FMUL)
- return DAG.getNode(ISD::FMA, SDLoc(N), VT,
- DAG.getNode(ISD::FNEG, SDLoc(N), VT,
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N),
- VT, N10.getOperand(0))),
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N10.getOperand(1)),
- N0);
- }
-
- // fold (fsub (fpext (fneg (fmul, x, y))), z)
- // -> (fma (fneg (fpext x)), (fpext y), (fneg z))
- if (N0.getOpcode() == ISD::FP_EXTEND) {
- SDValue N00 = N0.getOperand(0);
- if (N00.getOpcode() == ISD::FNEG) {
- SDValue N000 = N00.getOperand(0);
- if (N000.getOpcode() == ISD::FMUL) {
- return DAG.getNode(ISD::FMA, dl, VT,
- DAG.getNode(ISD::FNEG, dl, VT,
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N),
- VT, N000.getOperand(0))),
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N000.getOperand(1)),
- DAG.getNode(ISD::FNEG, dl, VT, N1));
- }
- }
- }
+ if (N10 == N0 && isNegatibleForFree(N11, LegalOperations, TLI, &Options))
+ return GetNegatedExpression(N11, DAG, LegalOperations);
- // fold (fsub (fneg (fpext (fmul, x, y))), z)
- // -> (fma (fneg (fpext x)), (fpext y), (fneg z))
- if (N0.getOpcode() == ISD::FNEG) {
- SDValue N00 = N0.getOperand(0);
- if (N00.getOpcode() == ISD::FP_EXTEND) {
- SDValue N000 = N00.getOperand(0);
- if (N000.getOpcode() == ISD::FMUL) {
- return DAG.getNode(ISD::FMA, dl, VT,
- DAG.getNode(ISD::FNEG, dl, VT,
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N),
- VT, N000.getOperand(0))),
- DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT,
- N000.getOperand(1)),
- DAG.getNode(ISD::FNEG, dl, VT, N1));
- }
- }
- }
+ if (N11 == N0 && isNegatibleForFree(N10, LegalOperations, TLI, &Options))
+ return GetNegatedExpression(N10, DAG, LegalOperations);
}
}
+ // FSUB -> FMA combines:
+ if (SDValue Fused = visitFSUBForFMACombine(N)) {
+ AddToWorklist(Fused.getNode());
+ return Fused;
+ }
+
return SDValue();
}
ConstantFPSDNode *N0CFP = isConstOrConstSplatFP(N0);
ConstantFPSDNode *N1CFP = isConstOrConstSplatFP(N1);
EVT VT = N->getValueType(0);
+ SDLoc DL(N);
const TargetOptions &Options = DAG.getTarget().Options;
+ const SDNodeFlags *Flags = &cast<BinaryWithFlagsSDNode>(N)->Flags;
// fold vector ops
if (VT.isVector()) {
// fold (fmul c1, c2) -> c1*c2
if (N0CFP && N1CFP)
- return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N0, N1);
+ return DAG.getNode(ISD::FMUL, DL, VT, N0, N1, Flags);
// canonicalize constant to RHS
if (isConstantFPBuildVectorOrConstantFP(N0) &&
!isConstantFPBuildVectorOrConstantFP(N1))
- return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N1, N0);
+ return DAG.getNode(ISD::FMUL, DL, VT, N1, N0, Flags);
// fold (fmul A, 1.0) -> A
if (N1CFP && N1CFP->isExactlyValue(1.0))
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()))) {
// the second operand of the outer multiply are constants.
if ((N1CFP && isConstOrConstSplatFP(N01)) ||
(BV1 && BV01 && BV1->isConstant() && BV01->isConstant())) {
- SDLoc SL(N);
- SDValue MulConsts = DAG.getNode(ISD::FMUL, SL, VT, N01, N1);
- return DAG.getNode(ISD::FMUL, SL, VT, N00, MulConsts);
+ SDValue MulConsts = DAG.getNode(ISD::FMUL, DL, VT, N01, N1, Flags);
+ return DAG.getNode(ISD::FMUL, DL, VT, N00, MulConsts, Flags);
}
}
}
// Undo the fmul 2.0, x -> fadd x, x transformation, since if it occurs
// during an early run of DAGCombiner can prevent folding with fmuls
// inserted during lowering.
- if (N0.getOpcode() == ISD::FADD && N0.getOperand(0) == N0.getOperand(1)) {
- SDLoc SL(N);
- const SDValue Two = DAG.getConstantFP(2.0, VT);
- SDValue MulConsts = DAG.getNode(ISD::FMUL, SL, VT, Two, N1);
- return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N0.getOperand(0), MulConsts);
+ if (N0.getOpcode() == ISD::FADD &&
+ (N0.getOperand(0) == N0.getOperand(1)) &&
+ N0.hasOneUse()) {
+ const SDValue Two = DAG.getConstantFP(2.0, DL, VT);
+ SDValue MulConsts = DAG.getNode(ISD::FMUL, DL, VT, Two, N1, Flags);
+ return DAG.getNode(ISD::FMUL, DL, VT, N0.getOperand(0), MulConsts, Flags);
}
}
// fold (fmul X, 2.0) -> (fadd X, X)
if (N1CFP && N1CFP->isExactlyValue(+2.0))
- return DAG.getNode(ISD::FADD, SDLoc(N), VT, N0, N0);
+ return DAG.getNode(ISD::FADD, DL, VT, N0, N0, Flags);
// fold (fmul X, -1.0) -> (fneg X)
if (N1CFP && N1CFP->isExactlyValue(-1.0))
if (!LegalOperations || TLI.isOperationLegal(ISD::FNEG, VT))
- return DAG.getNode(ISD::FNEG, SDLoc(N), VT, N0);
+ return DAG.getNode(ISD::FNEG, DL, VT, N0);
// fold (fmul (fneg X), (fneg Y)) -> (fmul X, Y)
if (char LHSNeg = isNegatibleForFree(N0, LegalOperations, TLI, &Options)) {
// Both can be negated for free, check to see if at least one is cheaper
// negated.
if (LHSNeg == 2 || RHSNeg == 2)
- return DAG.getNode(ISD::FMUL, SDLoc(N), VT,
+ return DAG.getNode(ISD::FMUL, DL, VT,
GetNegatedExpression(N0, DAG, LegalOperations),
- GetNegatedExpression(N1, DAG, LegalOperations));
+ GetNegatedExpression(N1, DAG, LegalOperations),
+ Flags);
}
}
+ // FMUL -> FMA combines:
+ if (SDValue Fused = visitFMULForFMACombine(N)) {
+ AddToWorklist(Fused.getNode());
+ return Fused;
+ }
+
return SDValue();
}
if (N1CFP && N1CFP->isZero())
return N2;
}
+ // TODO: The FMA node should have flags that propagate to these nodes.
if (N0CFP && N0CFP->isExactlyValue(1.0))
return DAG.getNode(ISD::FADD, SDLoc(N), VT, N1, N2);
if (N1CFP && N1CFP->isExactlyValue(1.0))
return DAG.getNode(ISD::FADD, SDLoc(N), VT, N0, N2);
// Canonicalize (fma c, x, y) -> (fma x, c, y)
- if (N0CFP && !N1CFP)
+ if (isConstantFPBuildVectorOrConstantFP(N0) &&
+ !isConstantFPBuildVectorOrConstantFP(N1))
return DAG.getNode(ISD::FMA, SDLoc(N), VT, N1, N0, N2);
- // (fma x, c1, (fmul x, c2)) -> (fmul x, c1+c2)
- if (Options.UnsafeFPMath && N1CFP &&
- N2.getOpcode() == ISD::FMUL &&
- N0 == N2.getOperand(0) &&
- N2.getOperand(1).getOpcode() == ISD::ConstantFP) {
- return DAG.getNode(ISD::FMUL, dl, VT, N0,
- DAG.getNode(ISD::FADD, dl, VT, N1, N2.getOperand(1)));
- }
+ // TODO: FMA nodes should have flags that propagate to the created nodes.
+ // For now, create a Flags object for use with all unsafe math transforms.
+ SDNodeFlags Flags;
+ Flags.setUnsafeAlgebra(true);
+ if (Options.UnsafeFPMath) {
+ // (fma x, c1, (fmul x, c2)) -> (fmul x, c1+c2)
+ if (N2.getOpcode() == ISD::FMUL && N0 == N2.getOperand(0) &&
+ isConstantFPBuildVectorOrConstantFP(N1) &&
+ isConstantFPBuildVectorOrConstantFP(N2.getOperand(1))) {
+ return DAG.getNode(ISD::FMUL, dl, VT, N0,
+ DAG.getNode(ISD::FADD, dl, VT, N1, N2.getOperand(1),
+ &Flags), &Flags);
+ }
- // (fma (fmul x, c1), c2, y) -> (fma x, c1*c2, y)
- if (Options.UnsafeFPMath &&
- N0.getOpcode() == ISD::FMUL && N1CFP &&
- N0.getOperand(1).getOpcode() == ISD::ConstantFP) {
- return DAG.getNode(ISD::FMA, dl, VT,
- N0.getOperand(0),
- DAG.getNode(ISD::FMUL, dl, VT, N1, N0.getOperand(1)),
- N2);
+ // (fma (fmul x, c1), c2, y) -> (fma x, c1*c2, y)
+ if (N0.getOpcode() == ISD::FMUL &&
+ isConstantFPBuildVectorOrConstantFP(N1) &&
+ isConstantFPBuildVectorOrConstantFP(N0.getOperand(1))) {
+ return DAG.getNode(ISD::FMA, dl, VT,
+ N0.getOperand(0),
+ DAG.getNode(ISD::FMUL, dl, VT, N1, N0.getOperand(1),
+ &Flags),
+ N2);
+ }
}
// (fma x, 1, y) -> (fadd x, y)
// (fma x, -1, y) -> (fadd (fneg x), y)
if (N1CFP) {
if (N1CFP->isExactlyValue(1.0))
+ // TODO: The FMA node should have flags that propagate to this node.
return DAG.getNode(ISD::FADD, dl, VT, N0, N2);
if (N1CFP->isExactlyValue(-1.0) &&
(!LegalOperations || TLI.isOperationLegal(ISD::FNEG, VT))) {
SDValue RHSNeg = DAG.getNode(ISD::FNEG, dl, VT, N0);
AddToWorklist(RHSNeg.getNode());
+ // TODO: The FMA node should have flags that propagate to this node.
return DAG.getNode(ISD::FADD, dl, VT, N2, RHSNeg);
}
}
- // (fma x, c, x) -> (fmul x, (c+1))
- if (Options.UnsafeFPMath && N1CFP && N0 == N2)
- return DAG.getNode(ISD::FMUL, dl, VT, N0,
- DAG.getNode(ISD::FADD, dl, VT,
- N1, DAG.getConstantFP(1.0, VT)));
-
- // (fma x, c, (fneg x)) -> (fmul x, (c-1))
- if (Options.UnsafeFPMath && N1CFP &&
- N2.getOpcode() == ISD::FNEG && N2.getOperand(0) == N0)
+ if (Options.UnsafeFPMath) {
+ // (fma x, c, x) -> (fmul x, (c+1))
+ if (N1CFP && N0 == N2) {
return DAG.getNode(ISD::FMUL, dl, VT, N0,
- DAG.getNode(ISD::FADD, dl, VT,
- N1, DAG.getConstantFP(-1.0, VT)));
+ DAG.getNode(ISD::FADD, dl, VT,
+ N1, DAG.getConstantFP(1.0, dl, VT),
+ &Flags), &Flags);
+ }
+ // (fma x, c, (fneg x)) -> (fmul x, (c-1))
+ if (N1CFP && N2.getOpcode() == ISD::FNEG && N2.getOperand(0) == N0) {
+ return DAG.getNode(ISD::FMUL, dl, VT, N0,
+ DAG.getNode(ISD::FADD, dl, VT,
+ N1, DAG.getConstantFP(-1.0, dl, VT),
+ &Flags), &Flags);
+ }
+ }
return SDValue();
}
+// Combine multiple FDIVs with the same divisor into multiple FMULs by the
+// reciprocal.
+// E.g., (a / D; b / D;) -> (recip = 1.0 / D; a * recip; b * recip)
+// Notice that this is not always beneficial. One reason is different target
+// may have different costs for FDIV and FMUL, so sometimes the cost of two
+// FDIVs may be lower than the cost of one FDIV and two FMULs. Another reason
+// is the critical path is increased from "one FDIV" to "one FDIV + one FMUL".
+SDValue DAGCombiner::combineRepeatedFPDivisors(SDNode *N) {
+ bool UnsafeMath = DAG.getTarget().Options.UnsafeFPMath;
+ const SDNodeFlags *Flags = N->getFlags();
+ if (!UnsafeMath && !Flags->hasAllowReciprocal())
+ return SDValue();
+
+ // Skip if current node is a reciprocal.
+ SDValue N0 = N->getOperand(0);
+ ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
+ if (N0CFP && N0CFP->isExactlyValue(1.0))
+ return SDValue();
+
+ // Exit early if the target does not want this transform or if there can't
+ // possibly be enough uses of the divisor to make the transform worthwhile.
+ SDValue N1 = N->getOperand(1);
+ unsigned MinUses = TLI.combineRepeatedFPDivisors();
+ if (!MinUses || N1->use_size() < MinUses)
+ return SDValue();
+
+ // Find all FDIV users of the same divisor.
+ // Use a set because duplicates may be present in the user list.
+ SetVector<SDNode *> Users;
+ for (auto *U : N1->uses()) {
+ if (U->getOpcode() == ISD::FDIV && U->getOperand(1) == N1) {
+ // This division is eligible for optimization only if global unsafe math
+ // is enabled or if this division allows reciprocal formation.
+ if (UnsafeMath || U->getFlags()->hasAllowReciprocal())
+ Users.insert(U);
+ }
+ }
+
+ // Now that we have the actual number of divisor uses, make sure it meets
+ // the minimum threshold specified by the target.
+ if (Users.size() < MinUses)
+ return SDValue();
+
+ EVT VT = N->getValueType(0);
+ SDLoc DL(N);
+ SDValue FPOne = DAG.getConstantFP(1.0, DL, VT);
+ SDValue Reciprocal = DAG.getNode(ISD::FDIV, DL, VT, FPOne, N1, Flags);
+
+ // Dividend / Divisor -> Dividend * Reciprocal
+ for (auto *U : Users) {
+ SDValue Dividend = U->getOperand(0);
+ if (Dividend != FPOne) {
+ SDValue NewNode = DAG.getNode(ISD::FMUL, SDLoc(U), VT, Dividend,
+ Reciprocal, Flags);
+ CombineTo(U, NewNode);
+ } else if (U != Reciprocal.getNode()) {
+ // In the absence of fast-math-flags, this user node is always the
+ // same node as Reciprocal, but with FMF they may be different nodes.
+ CombineTo(U, Reciprocal);
+ }
+ }
+ return SDValue(N, 0); // N was replaced.
+}
+
SDValue DAGCombiner::visitFDIV(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
EVT VT = N->getValueType(0);
SDLoc DL(N);
const TargetOptions &Options = DAG.getTarget().Options;
+ SDNodeFlags *Flags = &cast<BinaryWithFlagsSDNode>(N)->Flags;
// fold vector ops
if (VT.isVector())
// fold (fdiv c1, c2) -> c1/c2
if (N0CFP && N1CFP)
- return DAG.getNode(ISD::FDIV, SDLoc(N), VT, N0, N1);
+ return DAG.getNode(ISD::FDIV, SDLoc(N), VT, N0, N1, Flags);
if (Options.UnsafeFPMath) {
// fold (fdiv X, c2) -> fmul X, 1/c2 if losing precision is acceptable.
// TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT) ||
TLI.isOperationLegal(llvm::ISD::ConstantFP, VT) ||
TLI.isFPImmLegal(Recip, VT)))
- return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N0,
- DAG.getConstantFP(Recip, VT));
+ return DAG.getNode(ISD::FMUL, DL, VT, N0,
+ DAG.getConstantFP(Recip, DL, VT), Flags);
}
// If this FDIV is part of a reciprocal square root, it may be folded
// into a target-specific square root estimate instruction.
if (N1.getOpcode() == ISD::FSQRT) {
- if (SDValue RV = BuildRsqrtEstimate(N1.getOperand(0))) {
- return DAG.getNode(ISD::FMUL, DL, VT, N0, RV);
+ if (SDValue RV = BuildRsqrtEstimate(N1.getOperand(0), Flags)) {
+ return DAG.getNode(ISD::FMUL, DL, VT, N0, RV, Flags);
}
} else if (N1.getOpcode() == ISD::FP_EXTEND &&
N1.getOperand(0).getOpcode() == ISD::FSQRT) {
- if (SDValue RV = BuildRsqrtEstimate(N1.getOperand(0).getOperand(0))) {
+ if (SDValue RV = BuildRsqrtEstimate(N1.getOperand(0).getOperand(0),
+ Flags)) {
RV = DAG.getNode(ISD::FP_EXTEND, SDLoc(N1), VT, RV);
AddToWorklist(RV.getNode());
- return DAG.getNode(ISD::FMUL, DL, VT, N0, RV);
+ return DAG.getNode(ISD::FMUL, DL, VT, N0, RV, Flags);
}
} else if (N1.getOpcode() == ISD::FP_ROUND &&
N1.getOperand(0).getOpcode() == ISD::FSQRT) {
- if (SDValue RV = BuildRsqrtEstimate(N1.getOperand(0).getOperand(0))) {
+ if (SDValue RV = BuildRsqrtEstimate(N1.getOperand(0).getOperand(0),
+ Flags)) {
RV = DAG.getNode(ISD::FP_ROUND, SDLoc(N1), VT, RV, N1.getOperand(1));
AddToWorklist(RV.getNode());
- return DAG.getNode(ISD::FMUL, DL, VT, N0, RV);
+ return DAG.getNode(ISD::FMUL, DL, VT, N0, RV, Flags);
}
} else if (N1.getOpcode() == ISD::FMUL) {
// Look through an FMUL. Even though this won't remove the FDIV directly,
if (SqrtOp.getNode()) {
// We found a FSQRT, so try to make this fold:
// x / (y * sqrt(z)) -> x * (rsqrt(z) / y)
- if (SDValue RV = BuildRsqrtEstimate(SqrtOp.getOperand(0))) {
- RV = DAG.getNode(ISD::FDIV, SDLoc(N1), VT, RV, OtherOp);
+ if (SDValue RV = BuildRsqrtEstimate(SqrtOp.getOperand(0), Flags)) {
+ RV = DAG.getNode(ISD::FDIV, SDLoc(N1), VT, RV, OtherOp, Flags);
AddToWorklist(RV.getNode());
- return DAG.getNode(ISD::FMUL, DL, VT, N0, RV);
+ return DAG.getNode(ISD::FMUL, DL, VT, N0, RV, Flags);
}
}
}
// Fold into a reciprocal estimate and multiply instead of a real divide.
- if (SDValue RV = BuildReciprocalEstimate(N1)) {
+ if (SDValue RV = BuildReciprocalEstimate(N1, Flags)) {
AddToWorklist(RV.getNode());
- return DAG.getNode(ISD::FMUL, DL, VT, N0, RV);
+ return DAG.getNode(ISD::FMUL, DL, VT, N0, RV, Flags);
}
}
if (LHSNeg == 2 || RHSNeg == 2)
return DAG.getNode(ISD::FDIV, SDLoc(N), VT,
GetNegatedExpression(N0, DAG, LegalOperations),
- GetNegatedExpression(N1, DAG, LegalOperations));
+ GetNegatedExpression(N1, DAG, LegalOperations),
+ Flags);
}
}
- // Combine multiple FDIVs with the same divisor into multiple FMULs by the
- // reciprocal.
- // E.g., (a / D; b / D;) -> (recip = 1.0 / D; a * recip; b * recip)
- // Notice that this is not always beneficial. One reason is different target
- // may have different costs for FDIV and FMUL, so sometimes the cost of two
- // FDIVs may be lower than the cost of one FDIV and two FMULs. Another reason
- // is the critical path is increased from "one FDIV" to "one FDIV + one FMUL".
- if (Options.UnsafeFPMath) {
- // Skip if current node is a reciprocal.
- if (N0CFP && N0CFP->isExactlyValue(1.0))
- return SDValue();
-
- SmallVector<SDNode *, 4> Users;
- // Find all FDIV users of the same divisor.
- for (SDNode::use_iterator UI = N1.getNode()->use_begin(),
- UE = N1.getNode()->use_end();
- UI != UE; ++UI) {
- SDNode *User = UI.getUse().getUser();
- if (User->getOpcode() == ISD::FDIV && User->getOperand(1) == N1)
- Users.push_back(User);
- }
-
- if (TLI.combineRepeatedFPDivisors(Users.size())) {
- SDValue FPOne = DAG.getConstantFP(1.0, VT); // floating point 1.0
- SDValue Reciprocal = DAG.getNode(ISD::FDIV, SDLoc(N), VT, FPOne, N1);
-
- // Dividend / Divisor -> Dividend * Reciprocal
- for (auto I = Users.begin(), E = Users.end(); I != E; ++I) {
- if ((*I)->getOperand(0) != FPOne) {
- SDValue NewNode = DAG.getNode(ISD::FMUL, SDLoc(*I), VT,
- (*I)->getOperand(0), Reciprocal);
- DAG.ReplaceAllUsesWith(*I, NewNode.getNode());
- }
- }
- return SDValue();
- }
- }
+ if (SDValue CombineRepeatedDivisors = combineRepeatedFPDivisors(N))
+ return CombineRepeatedDivisors;
return SDValue();
}
// fold (frem c1, c2) -> fmod(c1,c2)
if (N0CFP && N1CFP)
- return DAG.getNode(ISD::FREM, SDLoc(N), VT, N0, N1);
+ return DAG.getNode(ISD::FREM, SDLoc(N), VT, N0, N1,
+ &cast<BinaryWithFlagsSDNode>(N)->Flags);
return SDValue();
}
SDValue DAGCombiner::visitFSQRT(SDNode *N) {
- if (DAG.getTarget().Options.UnsafeFPMath &&
- !TLI.isFsqrtCheap()) {
- // Compute this as X * (1/sqrt(X)) = X * (X ** -0.5)
- if (SDValue RV = BuildRsqrtEstimate(N->getOperand(0))) {
- EVT VT = RV.getValueType();
- RV = DAG.getNode(ISD::FMUL, SDLoc(N), VT, N->getOperand(0), RV);
- AddToWorklist(RV.getNode());
+ if (!DAG.getTarget().Options.UnsafeFPMath || TLI.isFsqrtCheap())
+ return SDValue();
- // Unfortunately, RV is now NaN if the input was exactly 0.
- // Select out this case and force the answer to 0.
- SDValue Zero = DAG.getConstantFP(0.0, VT);
- SDValue ZeroCmp =
- DAG.getSetCC(SDLoc(N), TLI.getSetCCResultType(*DAG.getContext(), VT),
- N->getOperand(0), Zero, ISD::SETEQ);
- AddToWorklist(ZeroCmp.getNode());
- AddToWorklist(RV.getNode());
+ // TODO: FSQRT nodes should have flags that propagate to the created nodes.
+ // For now, create a Flags object for use with all unsafe math transforms.
+ SDNodeFlags Flags;
+ Flags.setUnsafeAlgebra(true);
- RV = DAG.getNode(VT.isVector() ? ISD::VSELECT : ISD::SELECT,
- SDLoc(N), VT, ZeroCmp, Zero, RV);
- return RV;
- }
- }
- return SDValue();
+ // Compute this as X * (1/sqrt(X)) = X * (X ** -0.5)
+ SDValue RV = BuildRsqrtEstimate(N->getOperand(0), &Flags);
+ if (!RV)
+ return SDValue();
+
+ EVT VT = RV.getValueType();
+ SDLoc DL(N);
+ RV = DAG.getNode(ISD::FMUL, DL, VT, N->getOperand(0), RV, &Flags);
+ AddToWorklist(RV.getNode());
+
+ // Unfortunately, RV is now NaN if the input was exactly 0.
+ // Select out this case and force the answer to 0.
+ SDValue Zero = DAG.getConstantFP(0.0, DL, VT);
+ EVT CCVT = getSetCCResultType(VT);
+ SDValue ZeroCmp = DAG.getSetCC(DL, CCVT, N->getOperand(0), Zero, ISD::SETEQ);
+ AddToWorklist(ZeroCmp.getNode());
+ AddToWorklist(RV.getNode());
+
+ return DAG.getNode(VT.isVector() ? ISD::VSELECT : ISD::SELECT, DL, VT,
+ ZeroCmp, Zero, RV);
+}
+
+static inline bool CanCombineFCOPYSIGN_EXTEND_ROUND(SDNode *N) {
+ // copysign(x, fp_extend(y)) -> copysign(x, y)
+ // copysign(x, fp_round(y)) -> copysign(x, y)
+ // Do not optimize out type conversion of f128 type yet.
+ // For some target like x86_64, configuration is changed
+ // to keep one f128 value in one SSE register, but
+ // instruction selection cannot handle FCOPYSIGN on
+ // SSE registers yet.
+ SDValue N1 = N->getOperand(1);
+ EVT N1VT = N1->getValueType(0);
+ EVT N1Op0VT = N1->getOperand(0)->getValueType(0);
+ return (N1.getOpcode() == ISD::FP_EXTEND ||
+ N1.getOpcode() == ISD::FP_ROUND) &&
+ (N1VT == N1Op0VT || N1Op0VT != MVT::f128);
}
SDValue DAGCombiner::visitFCOPYSIGN(SDNode *N) {
// copysign(x, fp_extend(y)) -> copysign(x, y)
// copysign(x, fp_round(y)) -> copysign(x, y)
- if (N1.getOpcode() == ISD::FP_EXTEND || N1.getOpcode() == ISD::FP_ROUND)
+ if (CanCombineFCOPYSIGN_EXTEND_ROUND(N))
return DAG.getNode(ISD::FCOPYSIGN, SDLoc(N), VT,
N0, N1.getOperand(0));
!VT.isVector() &&
(!LegalOperations ||
TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT))) {
+ SDLoc DL(N);
SDValue Ops[] =
{ N0.getOperand(0), N0.getOperand(1),
- DAG.getConstantFP(-1.0, VT) , DAG.getConstantFP(0.0, VT),
+ DAG.getConstantFP(-1.0, DL, VT), DAG.getConstantFP(0.0, DL, VT),
N0.getOperand(2) };
- return DAG.getNode(ISD::SELECT_CC, SDLoc(N), VT, Ops);
+ return DAG.getNode(ISD::SELECT_CC, DL, VT, Ops);
}
// fold (sint_to_fp (zext (setcc x, y, cc))) ->
N0.getOperand(0).getOpcode() == ISD::SETCC &&!VT.isVector() &&
(!LegalOperations ||
TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT))) {
+ SDLoc DL(N);
SDValue Ops[] =
{ N0.getOperand(0).getOperand(0), N0.getOperand(0).getOperand(1),
- DAG.getConstantFP(1.0, VT) , DAG.getConstantFP(0.0, VT),
+ DAG.getConstantFP(1.0, DL, VT), DAG.getConstantFP(0.0, DL, VT),
N0.getOperand(0).getOperand(2) };
- return DAG.getNode(ISD::SELECT_CC, SDLoc(N), VT, Ops);
+ return DAG.getNode(ISD::SELECT_CC, DL, VT, Ops);
}
}
if (N0.getOpcode() == ISD::SETCC && !VT.isVector() &&
(!LegalOperations ||
TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT))) {
+ SDLoc DL(N);
SDValue Ops[] =
{ N0.getOperand(0), N0.getOperand(1),
- DAG.getConstantFP(1.0, VT), DAG.getConstantFP(0.0, VT),
+ DAG.getConstantFP(1.0, DL, VT), DAG.getConstantFP(0.0, DL, VT),
N0.getOperand(2) };
- return DAG.getNode(ISD::SELECT_CC, SDLoc(N), VT, Ops);
+ return DAG.getNode(ISD::SELECT_CC, DL, VT, Ops);
}
}
SDValue DAGCombiner::visitFP_TO_SINT(SDNode *N) {
SDValue N0 = N->getOperand(0);
- ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
EVT VT = N->getValueType(0);
// fold (fp_to_sint c1fp) -> c1
- if (N0CFP)
+ if (isConstantFPBuildVectorOrConstantFP(N0))
return DAG.getNode(ISD::FP_TO_SINT, SDLoc(N), VT, N0);
return FoldIntToFPToInt(N, DAG);
SDValue DAGCombiner::visitFP_TO_UINT(SDNode *N) {
SDValue N0 = N->getOperand(0);
- ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
EVT VT = N->getValueType(0);
// fold (fp_to_uint c1fp) -> c1
- if (N0CFP)
+ if (isConstantFPBuildVectorOrConstantFP(N0))
return DAG.getNode(ISD::FP_TO_UINT, SDLoc(N), VT, N0);
return FoldIntToFPToInt(N, DAG);
// single-step fp_round we want to fold to.
// In other words, double rounding isn't the same as rounding.
// Also, this is a value preserving truncation iff both fp_round's are.
- if (DAG.getTarget().Options.UnsafeFPMath || N0IsTrunc)
- return DAG.getNode(ISD::FP_ROUND, SDLoc(N), VT, N0.getOperand(0),
- DAG.getIntPtrConstant(NIsTrunc && N0IsTrunc));
+ if (DAG.getTarget().Options.UnsafeFPMath || N0IsTrunc) {
+ SDLoc DL(N);
+ return DAG.getNode(ISD::FP_ROUND, DL, VT, N0.getOperand(0),
+ DAG.getIntPtrConstant(NIsTrunc && N0IsTrunc, DL));
+ }
}
// fold (fp_round (copysign X, Y)) -> (copysign (fp_round X), Y)
// fold (fp_round_inreg c1fp) -> c1fp
if (N0CFP && isTypeLegal(EVT)) {
- SDValue Round = DAG.getConstantFP(*N0CFP->getConstantFPValue(), EVT);
- return DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT, Round);
+ SDLoc DL(N);
+ SDValue Round = DAG.getConstantFP(*N0CFP->getConstantFPValue(), DL, EVT);
+ return DAG.getNode(ISD::FP_EXTEND, DL, VT, Round);
}
return SDValue();
CombineTo(N, ExtLoad);
CombineTo(N0.getNode(),
DAG.getNode(ISD::FP_ROUND, SDLoc(N0),
- N0.getValueType(), ExtLoad, DAG.getIntPtrConstant(1)),
+ N0.getValueType(), ExtLoad,
+ DAG.getIntPtrConstant(1, SDLoc(N0))),
ExtLoad.getValue(1));
return SDValue(N, 0); // Return N so it doesn't get rechecked!
}
// For a scalar, just generate 0x80...
SignMask = APInt::getSignBit(IntVT.getSizeInBits());
}
- Int = DAG.getNode(ISD::XOR, SDLoc(N0), IntVT, Int,
- DAG.getConstant(SignMask, IntVT));
+ SDLoc DL0(N0);
+ Int = DAG.getNode(ISD::XOR, DL0, IntVT, Int,
+ DAG.getConstant(SignMask, DL0, IntVT));
AddToWorklist(Int.getNode());
return DAG.getNode(ISD::BITCAST, SDLoc(N), VT, Int);
}
}
// (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();
CVal.changeSign();
if (Level >= AfterLegalizeDAG &&
- (TLI.isFPImmLegal(CVal, N->getValueType(0)) ||
- TLI.isOperationLegal(ISD::ConstantFP, N->getValueType(0))))
- return DAG.getNode(
- ISD::FMUL, SDLoc(N), VT, N0.getOperand(0),
- DAG.getNode(ISD::FNEG, SDLoc(N), VT, N0.getOperand(1)));
+ (TLI.isFPImmLegal(CVal, VT) ||
+ TLI.isOperationLegal(ISD::ConstantFP, VT)))
+ return DAG.getNode(ISD::FMUL, SDLoc(N), VT, N0.getOperand(0),
+ DAG.getNode(ISD::FNEG, SDLoc(N), VT,
+ N0.getOperand(1)),
+ &cast<BinaryWithFlagsSDNode>(N0)->Flags);
}
}
SDValue DAGCombiner::visitFMINNUM(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
- const ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
- const ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1);
+ EVT VT = N->getValueType(0);
+ const ConstantFPSDNode *N0CFP = isConstOrConstSplatFP(N0);
+ const ConstantFPSDNode *N1CFP = isConstOrConstSplatFP(N1);
if (N0CFP && N1CFP) {
const APFloat &C0 = N0CFP->getValueAPF();
const APFloat &C1 = N1CFP->getValueAPF();
- return DAG.getConstantFP(minnum(C0, C1), N->getValueType(0));
+ return DAG.getConstantFP(minnum(C0, C1), SDLoc(N), VT);
}
- if (N0CFP) {
- EVT VT = N->getValueType(0);
- // Canonicalize to constant on RHS.
+ // Canonicalize to constant on RHS.
+ if (isConstantFPBuildVectorOrConstantFP(N0) &&
+ !isConstantFPBuildVectorOrConstantFP(N1))
return DAG.getNode(ISD::FMINNUM, SDLoc(N), VT, N1, N0);
- }
return SDValue();
}
SDValue DAGCombiner::visitFMAXNUM(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
- const ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
- const ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1);
+ EVT VT = N->getValueType(0);
+ const ConstantFPSDNode *N0CFP = isConstOrConstSplatFP(N0);
+ const ConstantFPSDNode *N1CFP = isConstOrConstSplatFP(N1);
if (N0CFP && N1CFP) {
const APFloat &C0 = N0CFP->getValueAPF();
const APFloat &C1 = N1CFP->getValueAPF();
- return DAG.getConstantFP(maxnum(C0, C1), N->getValueType(0));
+ return DAG.getConstantFP(maxnum(C0, C1), SDLoc(N), VT);
}
- if (N0CFP) {
- EVT VT = N->getValueType(0);
- // Canonicalize to constant on RHS.
+ // Canonicalize to constant on RHS.
+ if (isConstantFPBuildVectorOrConstantFP(N0) &&
+ !isConstantFPBuildVectorOrConstantFP(N1))
return DAG.getNode(ISD::FMAXNUM, SDLoc(N), VT, N1, N0);
- }
return SDValue();
}
// For a scalar, just generate 0x7f...
SignMask = ~APInt::getSignBit(IntVT.getSizeInBits());
}
- Int = DAG.getNode(ISD::AND, SDLoc(N0), IntVT, Int,
- DAG.getConstant(SignMask, IntVT));
+ SDLoc DL(N0);
+ Int = DAG.getNode(ISD::AND, DL, IntVT, Int,
+ DAG.getConstant(SignMask, DL, IntVT));
AddToWorklist(Int.getNode());
return DAG.getNode(ISD::BITCAST, SDLoc(N), N->getValueType(0), Int);
}
if (AndConst.isPowerOf2() &&
cast<ConstantSDNode>(Op1)->getAPIntValue()==AndConst.logBase2()) {
+ SDLoc DL(N);
SDValue SetCC =
- DAG.getSetCC(SDLoc(N),
+ DAG.getSetCC(DL,
getSetCCResultType(Op0.getValueType()),
- Op0, DAG.getConstant(0, Op0.getValueType()),
+ Op0, DAG.getConstant(0, DL, Op0.getValueType()),
ISD::SETNE);
- SDValue NewBRCond = DAG.getNode(ISD::BRCOND, SDLoc(N),
+ SDValue NewBRCond = DAG.getNode(ISD::BRCOND, DL,
MVT::Other, Chain, SetCC, N2);
// Don't add the new BRCond into the worklist or else SimplifySelectCC
// will convert it back to (X & C1) >> C2.
SDValue Op1 = TheXor->getOperand(1);
if (Op0.getOpcode() == Op1.getOpcode()) {
// Avoid missing important xor optimizations.
- SDValue Tmp = visitXOR(TheXor);
- if (Tmp.getNode()) {
+ if (SDValue Tmp = visitXOR(TheXor)) {
if (Tmp.getNode() != TheXor) {
DEBUG(dbgs() << "\nReplacing.8 ";
TheXor->dump(&DAG);
if (Op0.getOpcode() != ISD::SETCC && Op1.getOpcode() != ISD::SETCC) {
bool Equal = false;
- if (ConstantSDNode *RHSCI = dyn_cast<ConstantSDNode>(Op0))
- if (RHSCI->getAPIntValue() == 1 && Op0.hasOneUse() &&
- Op0.getOpcode() == ISD::XOR) {
- TheXor = Op0.getNode();
- Equal = true;
- }
+ if (isOneConstant(Op0) && Op0.hasOneUse() &&
+ Op0.getOpcode() == ISD::XOR) {
+ TheXor = Op0.getNode();
+ Equal = true;
+ }
EVT SetCCVT = N1.getValueType();
if (LegalTypes)
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 = Use->getValueType(0);
+ 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->getValue().getValueType();
+ VT = ST->getMemoryVT();
+ AS = ST->getAddressSpace();
} else
return false;
} else
return false;
- return TLI.isLegalAddressingMode(AM, VT.getTypeForEVT(*DAG.getContext()));
+ return TLI.isLegalAddressingMode(DAG.getDataLayout(), AM,
+ VT.getTypeForEVT(*DAG.getContext()), AS);
}
/// Try turning a load/store into a pre-indexed load/store when the base
}
// Don't create a indexed load / store with zero offset.
- if (isa<ConstantSDNode>(Offset) &&
- cast<ConstantSDNode>(Offset)->isNullValue())
+ if (isNullConstant(Offset))
return false;
// Try turning it into a pre-indexed load / store except when:
// a copy of the original base pointer.
SmallVector<SDNode *, 16> OtherUses;
if (isa<ConstantSDNode>(Offset))
- for (SDNode *Use : BasePtr.getNode()->uses()) {
- if (Use == Ptr.getNode())
+ for (SDNode::use_iterator UI = BasePtr.getNode()->use_begin(),
+ UE = BasePtr.getNode()->use_end();
+ UI != UE; ++UI) {
+ SDUse &Use = UI.getUse();
+ // Skip the use that is Ptr and uses of other results from BasePtr's
+ // node (important for nodes that return multiple results).
+ if (Use.getUser() == Ptr.getNode() || Use != BasePtr)
continue;
- if (Use->isPredecessorOf(N))
+ if (Use.getUser()->isPredecessorOf(N))
continue;
- if (Use->getOpcode() != ISD::ADD && Use->getOpcode() != ISD::SUB) {
+ if (Use.getUser()->getOpcode() != ISD::ADD &&
+ Use.getUser()->getOpcode() != ISD::SUB) {
OtherUses.clear();
break;
}
- SDValue Op0 = Use->getOperand(0), Op1 = Use->getOperand(1);
- if (Op1.getNode() == BasePtr.getNode())
- std::swap(Op0, Op1);
- assert(Op0.getNode() == BasePtr.getNode() &&
- "Use of ADD/SUB but not an operand");
-
+ SDValue Op1 = Use.getUser()->getOperand((UI.getOperandNo() + 1) & 1);
if (!isa<ConstantSDNode>(Op1)) {
OtherUses.clear();
break;
break;
}
- OtherUses.push_back(Use);
+ OtherUses.push_back(Use.getUser());
}
if (Swapped)
if (X1 * Y0 * Y1 < 0) CNV = CNV + Offset1;
else CNV = CNV - Offset1;
+ SDLoc DL(OtherUses[i]);
+
// We can now generate the new expression.
- SDValue NewOp1 = DAG.getConstant(CNV, CN->getValueType(0));
+ SDValue NewOp1 = DAG.getConstant(CNV, DL, CN->getValueType(0));
SDValue NewOp2 = Result.getValue(isLoad ? 1 : 0);
SDValue NewUse = DAG.getNode(Opcode,
- SDLoc(OtherUses[i]),
+ DL,
OtherUses[i]->getValueType(0), NewOp1, NewOp2);
DAG.ReplaceAllUsesOfValueWith(SDValue(OtherUses[i], 0), NewUse);
deleteAndRecombine(OtherUses[i]);
ISD::MemIndexedMode AM = ISD::UNINDEXED;
if (TLI.getPostIndexedAddressParts(N, Op, BasePtr, Offset, AM, DAG)) {
// Don't create a indexed load / store with zero offset.
- if (isa<ConstantSDNode>(Offset) &&
- cast<ConstantSDNode>(Offset)->isNullValue())
+ if (isNullConstant(Offset))
continue;
// Try turning it into a post-indexed load / store except when
"Cannot split out indexing using opaque target constants");
if (Inc.getOpcode() == ISD::TargetConstant) {
ConstantSDNode *ConstInc = cast<ConstantSDNode>(Inc);
- Inc = DAG.getConstant(*ConstInc->getConstantIntValue(),
+ Inc = DAG.getConstant(*ConstInc->getConstantIntValue(), SDLoc(Inc),
ConstInc->getValueType(0));
}
void addSliceGain(const LoadedSlice &LS) {
// Each slice saves a truncate.
const TargetLowering &TLI = LS.DAG->getTargetLoweringInfo();
- if (!TLI.isTruncateFree(LS.Inst->getValueType(0),
- LS.Inst->getOperand(0).getValueType()))
+ if (!TLI.isTruncateFree(LS.Inst->getOperand(0).getValueType(),
+ LS.Inst->getValueType(0)))
++Truncates;
// If there is a shift amount, this slice gets rid of it.
if (LS.Shift)
/// \pre DAG != nullptr.
uint64_t getOffsetFromBase() const {
assert(DAG && "Missing context.");
- bool IsBigEndian =
- DAG->getTargetLoweringInfo().getDataLayout()->isBigEndian();
+ bool IsBigEndian = DAG->getDataLayout().isBigEndian();
assert(!(Shift & 0x7) && "Shifts not aligned on Bytes are not supported.");
uint64_t Offset = Shift / 8;
unsigned TySizeInBytes = Origin->getValueSizeInBits(0) / 8;
if (Offset) {
// BaseAddr = BaseAddr + Offset.
EVT ArithType = BaseAddr.getValueType();
- BaseAddr = DAG->getNode(ISD::ADD, SDLoc(Origin), ArithType, BaseAddr,
- DAG->getConstant(Offset, ArithType));
+ SDLoc DL(Origin);
+ BaseAddr = DAG->getNode(ISD::ADD, DL, ArithType, BaseAddr,
+ DAG->getConstant(Offset, DL, ArithType));
}
// Create the type of the loaded slice according to its size.
// Check if it will be merged with the load.
// 1. Check the alignment constraint.
- unsigned RequiredAlignment = TLI.getDataLayout()->getABITypeAlignment(
+ unsigned RequiredAlignment = DAG->getDataLayout().getABITypeAlignment(
ResVT.getTypeForEVT(*DAG->getContext()));
if (RequiredAlignment > getAlignment())
return Result; // Fail.
else {
bool isOk = false;
- for (unsigned i = 0, e = Chain->getNumOperands(); i != e; ++i)
- if (Chain->getOperand(i).getNode() == LD) {
+ for (const SDValue &ChainOp : Chain->op_values())
+ if (ChainOp.getNode() == LD) {
isOk = true;
break;
}
// Okay, we can do this! Replace the 'St' store with a store of IVal that is
// shifted by ByteShift and truncated down to NumBytes.
- if (ByteShift)
- IVal = DAG.getNode(ISD::SRL, SDLoc(IVal), IVal.getValueType(), IVal,
- DAG.getConstant(ByteShift*8,
+ if (ByteShift) {
+ SDLoc DL(IVal);
+ IVal = DAG.getNode(ISD::SRL, DL, IVal.getValueType(), IVal,
+ DAG.getConstant(ByteShift*8, DL,
DC->getShiftAmountTy(IVal.getValueType())));
+ }
// Figure out the offset for the store and the alignment of the access.
unsigned StOffset;
unsigned NewAlign = St->getAlignment();
- if (DAG.getTargetLoweringInfo().isLittleEndian())
+ if (DAG.getDataLayout().isLittleEndian())
StOffset = ByteShift;
else
StOffset = IVal.getValueType().getStoreSize() - ByteShift - NumBytes;
SDValue Ptr = St->getBasePtr();
if (StOffset) {
- Ptr = DAG.getNode(ISD::ADD, SDLoc(IVal), Ptr.getValueType(),
- Ptr, DAG.getConstant(StOffset, Ptr.getValueType()));
+ SDLoc DL(IVal);
+ Ptr = DAG.getNode(ISD::ADD, DL, Ptr.getValueType(),
+ Ptr, DAG.getConstant(StOffset, DL, Ptr.getValueType()));
NewAlign = MinAlign(NewAlign, StOffset);
}
uint64_t PtrOff = ShAmt / 8;
// For big endian targets, we need to adjust the offset to the pointer to
// load the correct bytes.
- if (TLI.isBigEndian())
+ if (DAG.getDataLayout().isBigEndian())
PtrOff = (BitWidth + 7 - NewBW) / 8 - PtrOff;
unsigned NewAlign = MinAlign(LD->getAlignment(), PtrOff);
Type *NewVTTy = NewVT.getTypeForEVT(*DAG.getContext());
- if (NewAlign < TLI.getDataLayout()->getABITypeAlignment(NewVTTy))
+ if (NewAlign < DAG.getDataLayout().getABITypeAlignment(NewVTTy))
return SDValue();
SDValue NewPtr = DAG.getNode(ISD::ADD, SDLoc(LD),
Ptr.getValueType(), Ptr,
- DAG.getConstant(PtrOff, Ptr.getValueType()));
+ DAG.getConstant(PtrOff, SDLoc(LD),
+ Ptr.getValueType()));
SDValue NewLD = DAG.getLoad(NewVT, SDLoc(N0),
LD->getChain(), NewPtr,
LD->getPointerInfo().getWithOffset(PtrOff),
LD->isInvariant(), NewAlign,
LD->getAAInfo());
SDValue NewVal = DAG.getNode(Opc, SDLoc(Value), NewVT, NewLD,
- DAG.getConstant(NewImm, NewVT));
+ DAG.getConstant(NewImm, SDLoc(Value),
+ NewVT));
SDValue NewST = DAG.getStore(Chain, SDLoc(N),
NewVal, NewPtr,
ST->getPointerInfo().getWithOffset(PtrOff),
unsigned LDAlign = LD->getAlignment();
unsigned STAlign = ST->getAlignment();
Type *IntVTTy = IntVT.getTypeForEVT(*DAG.getContext());
- unsigned ABIAlign = TLI.getDataLayout()->getABITypeAlignment(IntVTTy);
+ unsigned ABIAlign = DAG.getDataLayout().getABITypeAlignment(IntVTTy);
if (LDAlign < ABIAlign || STAlign < ABIAlign)
return SDValue();
};
} // namespace
+// This is a helper function for visitMUL to check the profitability
+// of folding (mul (add x, c1), c2) -> (add (mul x, c2), c1*c2).
+// MulNode is the original multiply, AddNode is (add x, c1),
+// and ConstNode is c2.
+//
+// If the (add x, c1) has multiple uses, we could increase
+// the number of adds if we make this transformation.
+// It would only be worth doing this if we can remove a
+// multiply in the process. Check for that here.
+// To illustrate:
+// (A + c1) * c3
+// (A + c2) * c3
+// We're checking for cases where we have common "c3 * A" expressions.
+bool DAGCombiner::isMulAddWithConstProfitable(SDNode *MulNode,
+ SDValue &AddNode,
+ SDValue &ConstNode) {
+ APInt Val;
+
+ // If the add only has one use, this would be OK to do.
+ if (AddNode.getNode()->hasOneUse())
+ return true;
+
+ // Walk all the users of the constant with which we're multiplying.
+ for (SDNode *Use : ConstNode->uses()) {
+
+ if (Use == MulNode) // This use is the one we're on right now. Skip it.
+ continue;
+
+ if (Use->getOpcode() == ISD::MUL) { // We have another multiply use.
+ SDNode *OtherOp;
+ SDNode *MulVar = AddNode.getOperand(0).getNode();
+
+ // OtherOp is what we're multiplying against the constant.
+ if (Use->getOperand(0) == ConstNode)
+ OtherOp = Use->getOperand(1).getNode();
+ else
+ OtherOp = Use->getOperand(0).getNode();
+
+ // Check to see if multiply is with the same operand of our "add".
+ //
+ // ConstNode = CONST
+ // Use = ConstNode * A <-- visiting Use. OtherOp is A.
+ // ...
+ // AddNode = (A + c1) <-- MulVar is A.
+ // = AddNode * ConstNode <-- current visiting instruction.
+ //
+ // If we make this transformation, we will have a common
+ // multiply (ConstNode * A) that we can save.
+ if (OtherOp == MulVar)
+ return true;
+
+ // Now check to see if a future expansion will give us a common
+ // multiply.
+ //
+ // ConstNode = CONST
+ // AddNode = (A + c1)
+ // ... = AddNode * ConstNode <-- current visiting instruction.
+ // ...
+ // OtherOp = (A + c2)
+ // Use = OtherOp * ConstNode <-- visiting Use.
+ //
+ // If we make this transformation, we will have a common
+ // multiply (CONST * A) after we also do the same transformation
+ // to the "t2" instruction.
+ if (OtherOp->getOpcode() == ISD::ADD &&
+ isConstantIntBuildVectorOrConstantInt(OtherOp->getOperand(1)) &&
+ OtherOp->getOperand(0).getNode() == MulVar)
+ return true;
+ }
+ }
+
+ // Didn't find a case where this would be profitable.
+ return false;
+}
+
+SDValue DAGCombiner::getMergedConstantVectorStore(SelectionDAG &DAG,
+ SDLoc SL,
+ ArrayRef<MemOpLink> Stores,
+ SmallVectorImpl<SDValue> &Chains,
+ EVT Ty) const {
+ SmallVector<SDValue, 8> BuildVector;
+
+ for (unsigned I = 0, E = Ty.getVectorNumElements(); I != E; ++I) {
+ StoreSDNode *St = cast<StoreSDNode>(Stores[I].MemNode);
+ Chains.push_back(St->getChain());
+ BuildVector.push_back(St->getValue());
+ }
+
+ return DAG.getNode(ISD::BUILD_VECTOR, SL, Ty, BuildVector);
+}
+
bool DAGCombiner::MergeStoresOfConstantsOrVecElts(
SmallVectorImpl<MemOpLink> &StoreNodes, EVT MemVT,
- unsigned NumElem, bool IsConstantSrc, bool UseVector) {
+ unsigned NumStores, bool IsConstantSrc, bool UseVector) {
// Make sure we have something to merge.
- if (NumElem < 2)
+ if (NumStores < 2)
return false;
int64_t ElementSizeBytes = MemVT.getSizeInBits() / 8;
LSBaseSDNode *FirstInChain = StoreNodes[0].MemNode;
unsigned LatestNodeUsed = 0;
- for (unsigned i=0; i < NumElem; ++i) {
+ for (unsigned i=0; i < NumStores; ++i) {
// Find a chain for the new wide-store operand. Notice that some
// of the store nodes that we found may not be selected for inclusion
// in the wide store. The chain we use needs to be the chain of the
LatestNodeUsed = i;
}
+ SmallVector<SDValue, 8> Chains;
+
// The latest Node in the DAG.
LSBaseSDNode *LatestOp = StoreNodes[LatestNodeUsed].MemNode;
SDLoc DL(StoreNodes[0].MemNode);
SDValue StoredVal;
if (UseVector) {
- // Find a legal type for the vector store.
- EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT, NumElem);
+ bool IsVec = MemVT.isVector();
+ unsigned Elts = NumStores;
+ if (IsVec) {
+ // When merging vector stores, get the total number of elements.
+ Elts *= MemVT.getVectorNumElements();
+ }
+ // Get the type for the merged vector store.
+ EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT.getScalarType(), Elts);
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, Ty);
+ StoredVal = getMergedConstantVectorStore(DAG, DL, StoreNodes, Chains, Ty);
} else {
SmallVector<SDValue, 8> Ops;
- for (unsigned i = 0; i < NumElem ; ++i) {
+ for (unsigned i = 0; i < NumStores; ++i) {
StoreSDNode *St = cast<StoreSDNode>(StoreNodes[i].MemNode);
SDValue Val = St->getValue();
- // All of the operands of a BUILD_VECTOR must have the same type.
+ // All operands of BUILD_VECTOR / CONCAT_VECTOR must have the same type.
if (Val.getValueType() != MemVT)
return false;
Ops.push_back(Val);
+ Chains.push_back(St->getChain());
}
// Build the extracted vector elements back into a vector.
- StoredVal = DAG.getNode(ISD::BUILD_VECTOR, DL, Ty, Ops);
- }
+ StoredVal = DAG.getNode(IsVec ? ISD::CONCAT_VECTORS : ISD::BUILD_VECTOR,
+ DL, Ty, Ops); }
} else {
// We should always use a vector store when merging extracted vector
// 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 = NumStores * ElementSizeBytes * 8;
+ APInt StoreInt(SizeInBits, 0);
// Construct a single integer constant which is made of the smaller
// constant inputs.
- bool IsLE = TLI.isLittleEndian();
- for (unsigned i = 0; i < NumElem ; ++i) {
- unsigned Idx = IsLE ? (NumElem - 1 - i) : i;
+ bool IsLE = DAG.getDataLayout().isLittleEndian();
+ for (unsigned i = 0; i < NumStores; ++i) {
+ unsigned Idx = IsLE ? (NumStores - 1 - i) : i;
StoreSDNode *St = cast<StoreSDNode>(StoreNodes[Idx].MemNode);
+ Chains.push_back(St->getChain());
+
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);
- StoredVal = DAG.getConstant(StoreInt, StoreTy);
+ EVT StoreTy = EVT::getIntegerVT(*DAG.getContext(), SizeInBits);
+ StoredVal = DAG.getConstant(StoreInt, DL, StoreTy);
}
- SDValue NewStore = DAG.getStore(LatestOp->getChain(), DL, StoredVal,
+ assert(!Chains.empty());
+
+ SDValue NewChain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Chains);
+ SDValue NewStore = DAG.getStore(NewChain, DL, StoredVal,
FirstInChain->getBasePtr(),
FirstInChain->getPointerInfo(),
false, false,
// Replace the last store with the new store
CombineTo(LatestOp, NewStore);
// Erase all other stores.
- for (unsigned i = 0; i < NumElem ; ++i) {
+ for (unsigned i = 0; i < NumStores; ++i) {
if (StoreNodes[i].MemNode == LatestOp)
continue;
StoreSDNode *St = cast<StoreSDNode>(StoreNodes[i].MemNode);
return true;
}
-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);
-
- // 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;
+
+
+ bool UseAA = CombinerAA.getNumOccurrences() > 0 ? CombinerAA
+ : DAG.getSubtarget().useAA();
+
+ if (UseAA) {
+ // Look at other users of the same chain. Stores on the same chain do not
+ // alias. If combiner-aa is enabled, non-aliasing stores are canonicalized
+ // to be on the same chain, so don't bother looking at adjacent chains.
+
+ SDValue Chain = St->getChain();
+ for (auto I = Chain->use_begin(), E = Chain->use_end(); I != E; ++I) {
+ if (StoreSDNode *OtherST = dyn_cast<StoreSDNode>(*I)) {
+ if (I.getOperandNo() != 0)
+ continue;
+
+ if (OtherST->isVolatile() || OtherST->isIndexed())
+ continue;
+
+ if (OtherST->getMemoryVT() != MemVT)
+ continue;
+
+ BaseIndexOffset Ptr = BaseIndexOffset::match(OtherST->getBasePtr());
+
+ if (Ptr.equalBaseIndex(BasePtr))
+ StoreNodes.push_back(MemOpLink(OtherST, Ptr.Offset, Seq++));
+ }
+ }
+
+ return;
+ }
+
while (Index) {
// If the chain has more than one use, then we can't reorder the mem ops.
if (Index != St && !SDValue(Index, 0)->hasOneUse())
if (!Ptr.equalBaseIndex(BasePtr))
break;
- // Check that the alignment is the same.
- if (Index->getAlignment() != St->getAlignment())
- break;
-
// The memory operands must not be volatile.
if (Index->isVolatile() || Index->isIndexed())
break;
if (Index->getMemoryVT() != MemVT)
break;
- // We do not allow unaligned stores because we want to prevent overriding
- // stores.
- if (Index->getAlignment()*8 != MemVT.getSizeInBits())
+ // We do not allow under-aligned stores in order to prevent
+ // overriding stores. NOTE: this is a bad hack. Alignment SHOULD
+ // be irrelevant here; what MATTERS is that we not move memory
+ // operations that potentially overlap past each-other.
+ if (Index->getAlignment() < MemVT.getStoreSize())
break;
// We found a potential memory operand to merge.
break;
}
- // Save the load node for later. Continue the scan.
- AliasLoadNodes.push_back(Ldn);
- NextInChain = Ldn->getChain().getNode();
- continue;
- } else {
- Index = nullptr;
- break;
- }
- }
- }
+ // Save the load node for later. Continue the scan.
+ AliasLoadNodes.push_back(Ldn);
+ NextInChain = Ldn->getChain().getNode();
+ continue;
+ } else {
+ Index = nullptr;
+ break;
+ }
+ }
+ }
+}
+
+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;
+
+ if (!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 IsExtractVecSrc = (StoredVal.getOpcode() == ISD::EXTRACT_VECTOR_ELT ||
+ StoredVal.getOpcode() == ISD::EXTRACT_SUBVECTOR);
+
+ if (!IsConstantSrc && !IsLoadSrc && !IsExtractVecSrc)
+ return false;
+
+ // Don't merge vectors into wider vectors if the source data comes from loads.
+ // TODO: This restriction can be lifted by using logic similar to the
+ // ExtractVecSrc case.
+ if (MemVT.isVector() && IsLoadSrc)
+ 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;
- // Sort the memory operands according to their distance from the base pointer.
+ // Sort the memory operands according to their distance from the
+ // base pointer. As a secondary criteria: make sure stores coming
+ // later in the code come first in the list. This is important for
+ // the non-UseAA case, because we're merging stores into the FINAL
+ // store along a chain which potentially contains aliasing stores.
+ // Thus, if there are multiple stores to the same address, the last
+ // one can be considered for merging but not the others.
std::sort(StoreNodes.begin(), StoreNodes.end(),
[](MemOpLink LHS, MemOpLink RHS) {
return LHS.OffsetFromBase < RHS.OffsetFromBase ||
(LHS.OffsetFromBase == RHS.OffsetFromBase &&
- LHS.SequenceNum > RHS.SequenceNum);
+ LHS.SequenceNum < RHS.SequenceNum);
});
// Scan the memory operations on the chain and find the first non-consecutive
break;
}
- bool Alias = false;
// Check if this store interferes with any of the loads that we found.
- for (unsigned ld = 0, lde = AliasLoadNodes.size(); ld < lde; ++ld)
- if (isAlias(AliasLoadNodes[ld], StoreNodes[i].MemNode)) {
- Alias = true;
- break;
- }
- // We found a load that alias with this store. Stop the sequence.
- if (Alias)
+ // If we find a load that alias with this store. Stop the sequence.
+ if (std::any_of(AliasLoadNodes.begin(), AliasLoadNodes.end(),
+ [&](LSBaseSDNode* Ldn) {
+ return isAlias(Ldn, StoreNodes[i].MemNode);
+ }))
break;
// Mark this node as useful.
// The node with the lowest store address.
LSBaseSDNode *FirstInChain = StoreNodes[0].MemNode;
+ unsigned FirstStoreAS = FirstInChain->getAddressSpace();
+ unsigned FirstStoreAlign = FirstInChain->getAlignment();
+ LLVMContext &Context = *DAG.getContext();
+ const DataLayout &DL = DAG.getDataLayout();
// Store the constants into memory as one consecutive store.
if (IsConstantSrc) {
}
// Find a legal type for the constant store.
- unsigned StoreBW = (i+1) * ElementSizeBytes * 8;
- EVT StoreTy = EVT::getIntegerVT(*DAG.getContext(), StoreBW);
- if (TLI.isTypeLegal(StoreTy))
+ unsigned SizeInBits = (i+1) * ElementSizeBytes * 8;
+ EVT StoreTy = EVT::getIntegerVT(Context, SizeInBits);
+ bool IsFast;
+ if (TLI.isTypeLegal(StoreTy) &&
+ TLI.allowsMemoryAccess(Context, DL, StoreTy, FirstStoreAS,
+ FirstStoreAlign, &IsFast) && IsFast) {
LastLegalType = i+1;
// Or check whether a truncstore is legal.
- else if (TLI.getTypeAction(*DAG.getContext(), StoreTy) ==
- TargetLowering::TypePromoteInteger) {
+ } else if (TLI.getTypeAction(Context, StoreTy) ==
+ TargetLowering::TypePromoteInteger) {
EVT LegalizedStoredValueTy =
- TLI.getTypeToTransformTo(*DAG.getContext(), StoredVal.getValueType());
- if (TLI.isTruncStoreLegal(LegalizedStoredValueTy, StoreTy))
- LastLegalType = i+1;
+ TLI.getTypeToTransformTo(Context, StoredVal.getValueType());
+ if (TLI.isTruncStoreLegal(LegalizedStoredValueTy, StoreTy) &&
+ TLI.allowsMemoryAccess(Context, DL, LegalizedStoredValueTy,
+ FirstStoreAS, FirstStoreAlign, &IsFast) &&
+ IsFast) {
+ LastLegalType = i + 1;
+ }
}
- // Find a legal type for the vector store.
- EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT, i+1);
- if (TLI.isTypeLegal(Ty))
- LastLegalVectorType = i + 1;
+ // 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 ((!NonZero || TLI.storeOfVectorConstantIsCheap(MemVT, i+1,
+ FirstStoreAS)) &&
+ !NoVectors) {
+ // Find a legal type for the vector store.
+ EVT Ty = EVT::getVectorVT(Context, MemVT, i+1);
+ if (TLI.isTypeLegal(Ty) &&
+ TLI.allowsMemoryAccess(Context, DL, Ty, FirstStoreAS,
+ FirstStoreAlign, &IsFast) && IsFast)
+ LastLegalVectorType = i + 1;
+ }
}
- // 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)
- LastLegalVectorType = 0;
-
// Check if we found a legal integer type to store.
if (LastLegalType == 0 && LastLegalVectorType == 0)
return false;
// When extracting multiple vector elements, try to store them
// in one vector store rather than a sequence of scalar stores.
- if (IsExtractVecEltSrc) {
- unsigned NumElem = 0;
+ if (IsExtractVecSrc) {
+ unsigned NumStoresToMerge = 0;
+ bool IsVec = MemVT.isVector();
for (unsigned i = 0; i < LastConsecutiveStore + 1; ++i) {
StoreSDNode *St = cast<StoreSDNode>(StoreNodes[i].MemNode);
- SDValue StoredVal = St->getValue();
+ unsigned StoreValOpcode = St->getValue().getOpcode();
// This restriction could be loosened.
// Bail out if any stored values are not elements extracted from a vector.
// It should be possible to handle mixed sources, but load sources need
// more careful handling (see the block of code below that handles
// consecutive loads).
- if (StoredVal.getOpcode() != ISD::EXTRACT_VECTOR_ELT)
+ if (StoreValOpcode != ISD::EXTRACT_VECTOR_ELT &&
+ StoreValOpcode != ISD::EXTRACT_SUBVECTOR)
return false;
// Find a legal type for the vector store.
- EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT, i+1);
- if (TLI.isTypeLegal(Ty))
- NumElem = i + 1;
+ unsigned Elts = i + 1;
+ if (IsVec) {
+ // When merging vector stores, get the total number of elements.
+ Elts *= MemVT.getVectorNumElements();
+ }
+ EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT.getScalarType(), Elts);
+ bool IsFast;
+ if (TLI.isTypeLegal(Ty) &&
+ TLI.allowsMemoryAccess(Context, DL, Ty, FirstStoreAS,
+ FirstStoreAlign, &IsFast) && IsFast)
+ NumStoresToMerge = i + 1;
}
- return MergeStoresOfConstantsOrVecElts(StoreNodes, MemVT, NumElem,
+ return MergeStoresOfConstantsOrVecElts(StoreNodes, MemVT, NumStoresToMerge,
false, true);
}
if (!Ld->hasNUsesOfValue(1, 0))
break;
- // Check that the alignment is the same as the stores.
- if (Ld->getAlignment() != St->getAlignment())
- break;
-
// The memory operands must not be volatile.
if (Ld->isVolatile() || Ld->isIndexed())
break;
St->getAlignment() >= RequiredAlignment)
return false;
+ LoadSDNode *FirstLoad = cast<LoadSDNode>(LoadNodes[0].MemNode);
+ unsigned FirstLoadAS = FirstLoad->getAddressSpace();
+ unsigned FirstLoadAlign = FirstLoad->getAlignment();
+
// Scan the memory operations on the chain and find the first non-consecutive
// load memory address. These variables hold the index in the store node
// array.
unsigned LastLegalVectorType = 0;
unsigned LastLegalIntegerType = 0;
StartAddress = LoadNodes[0].OffsetFromBase;
- SDValue FirstChain = LoadNodes[0].MemNode->getChain();
+ SDValue FirstChain = FirstLoad->getChain();
for (unsigned i = 1; i < LoadNodes.size(); ++i) {
- // All loads much share the same chain.
+ // All loads must share the same chain.
if (LoadNodes[i].MemNode->getChain() != FirstChain)
break;
if (CurrAddress - StartAddress != (ElementSizeBytes * i))
break;
LastConsecutiveLoad = i;
-
// Find a legal type for the vector store.
- EVT StoreTy = EVT::getVectorVT(*DAG.getContext(), MemVT, i+1);
- if (TLI.isTypeLegal(StoreTy))
+ EVT StoreTy = EVT::getVectorVT(Context, MemVT, i+1);
+ bool IsFastSt, IsFastLd;
+ if (TLI.isTypeLegal(StoreTy) &&
+ TLI.allowsMemoryAccess(Context, DL, StoreTy, FirstStoreAS,
+ FirstStoreAlign, &IsFastSt) && IsFastSt &&
+ TLI.allowsMemoryAccess(Context, DL, StoreTy, FirstLoadAS,
+ FirstLoadAlign, &IsFastLd) && IsFastLd) {
LastLegalVectorType = i + 1;
+ }
// Find a legal type for the integer store.
- unsigned StoreBW = (i+1) * ElementSizeBytes * 8;
- StoreTy = EVT::getIntegerVT(*DAG.getContext(), StoreBW);
- if (TLI.isTypeLegal(StoreTy))
+ unsigned SizeInBits = (i+1) * ElementSizeBytes * 8;
+ StoreTy = EVT::getIntegerVT(Context, SizeInBits);
+ if (TLI.isTypeLegal(StoreTy) &&
+ TLI.allowsMemoryAccess(Context, DL, StoreTy, FirstStoreAS,
+ FirstStoreAlign, &IsFastSt) && IsFastSt &&
+ TLI.allowsMemoryAccess(Context, DL, StoreTy, FirstLoadAS,
+ FirstLoadAlign, &IsFastLd) && IsFastLd)
LastLegalIntegerType = i + 1;
// Or check whether a truncstore and extload is legal.
- else if (TLI.getTypeAction(*DAG.getContext(), StoreTy) ==
+ else if (TLI.getTypeAction(Context, StoreTy) ==
TargetLowering::TypePromoteInteger) {
EVT LegalizedStoredValueTy =
- TLI.getTypeToTransformTo(*DAG.getContext(), StoreTy);
+ TLI.getTypeToTransformTo(Context, StoreTy);
if (TLI.isTruncStoreLegal(LegalizedStoredValueTy, StoreTy) &&
TLI.isLoadExtLegal(ISD::ZEXTLOAD, LegalizedStoredValueTy, StoreTy) &&
TLI.isLoadExtLegal(ISD::SEXTLOAD, LegalizedStoredValueTy, StoreTy) &&
- TLI.isLoadExtLegal(ISD::EXTLOAD, LegalizedStoredValueTy, StoreTy))
+ TLI.isLoadExtLegal(ISD::EXTLOAD, LegalizedStoredValueTy, StoreTy) &&
+ TLI.allowsMemoryAccess(Context, DL, LegalizedStoredValueTy,
+ FirstStoreAS, FirstStoreAlign, &IsFastSt) &&
+ IsFastSt &&
+ TLI.allowsMemoryAccess(Context, DL, LegalizedStoredValueTy,
+ FirstLoadAS, FirstLoadAlign, &IsFastLd) &&
+ IsFastLd)
LastLegalIntegerType = i+1;
}
}
if (NumElem < 2)
return false;
+ // Collect the chains from all merged stores.
+ SmallVector<SDValue, 8> MergeStoreChains;
+ MergeStoreChains.push_back(StoreNodes[0].MemNode->getChain());
+
// The latest Node in the DAG.
unsigned LatestNodeUsed = 0;
for (unsigned i=1; i<NumElem; ++i) {
// latest store node which is *used* and replaced by the wide store.
if (StoreNodes[i].SequenceNum < StoreNodes[LatestNodeUsed].SequenceNum)
LatestNodeUsed = i;
+
+ MergeStoreChains.push_back(StoreNodes[i].MemNode->getChain());
}
LSBaseSDNode *LatestOp = StoreNodes[LatestNodeUsed].MemNode;
// to memory.
EVT JointMemOpVT;
if (UseVectorTy) {
- JointMemOpVT = EVT::getVectorVT(*DAG.getContext(), MemVT, NumElem);
+ JointMemOpVT = EVT::getVectorVT(Context, MemVT, NumElem);
} else {
- unsigned StoreBW = NumElem * ElementSizeBytes * 8;
- JointMemOpVT = EVT::getIntegerVT(*DAG.getContext(), StoreBW);
+ unsigned SizeInBits = NumElem * ElementSizeBytes * 8;
+ JointMemOpVT = EVT::getIntegerVT(Context, SizeInBits);
}
SDLoc LoadDL(LoadNodes[0].MemNode);
SDLoc StoreDL(StoreNodes[0].MemNode);
- LoadSDNode *FirstLoad = cast<LoadSDNode>(LoadNodes[0].MemNode);
- SDValue NewLoad = DAG.getLoad(JointMemOpVT, LoadDL,
- FirstLoad->getChain(),
- FirstLoad->getBasePtr(),
- FirstLoad->getPointerInfo(),
- false, false, false,
- FirstLoad->getAlignment());
-
- SDValue NewStore = DAG.getStore(LatestOp->getChain(), StoreDL, NewLoad,
- FirstInChain->getBasePtr(),
- FirstInChain->getPointerInfo(), false, false,
- FirstInChain->getAlignment());
+ // The merged loads are required to have the same incoming chain, so
+ // using the first's chain is acceptable.
+ SDValue NewLoad = DAG.getLoad(
+ JointMemOpVT, LoadDL, FirstLoad->getChain(), FirstLoad->getBasePtr(),
+ FirstLoad->getPointerInfo(), false, false, false, FirstLoadAlign);
+
+ SDValue NewStoreChain =
+ DAG.getNode(ISD::TokenFactor, StoreDL, MVT::Other, MergeStoreChains);
- // Replace one of the loads with the new load.
- LoadSDNode *Ld = cast<LoadSDNode>(LoadNodes[0].MemNode);
- DAG.ReplaceAllUsesOfValueWith(SDValue(Ld, 1),
- SDValue(NewLoad.getNode(), 1));
+ SDValue NewStore = DAG.getStore(
+ NewStoreChain, StoreDL, NewLoad, FirstInChain->getBasePtr(),
+ FirstInChain->getPointerInfo(), false, false, FirstStoreAlign);
- // Remove the rest of the load chains.
- for (unsigned i = 1; i < NumElem ; ++i) {
- // Replace all chain users of the old load nodes with the chain of the new
- // load node.
+ // Transfer chain users from old loads to the new load.
+ for (unsigned i = 0; i < NumElem; ++i) {
LoadSDNode *Ld = cast<LoadSDNode>(LoadNodes[i].MemNode);
- DAG.ReplaceAllUsesOfValueWith(SDValue(Ld, 1), Ld->getChain());
+ DAG.ReplaceAllUsesOfValueWith(SDValue(Ld, 1),
+ SDValue(NewLoad.getNode(), 1));
}
// Replace the last store with the new store.
return true;
}
+SDValue DAGCombiner::replaceStoreChain(StoreSDNode *ST, SDValue BetterChain) {
+ SDLoc SL(ST);
+ SDValue ReplStore;
+
+ // Replace the chain to avoid dependency.
+ if (ST->isTruncatingStore()) {
+ ReplStore = DAG.getTruncStore(BetterChain, SL, ST->getValue(),
+ ST->getBasePtr(), ST->getMemoryVT(),
+ ST->getMemOperand());
+ } else {
+ ReplStore = DAG.getStore(BetterChain, SL, ST->getValue(), ST->getBasePtr(),
+ ST->getMemOperand());
+ }
+
+ // Create token to keep both nodes around.
+ SDValue Token = DAG.getNode(ISD::TokenFactor, SL,
+ MVT::Other, ST->getChain(), ReplStore);
+
+ // Make sure the new and old chains are cleaned up.
+ AddToWorklist(Token.getNode());
+
+ // Don't add users to work list.
+ return CombineTo(ST, Token, false);
+}
+
+SDValue DAGCombiner::replaceStoreOfFPConstant(StoreSDNode *ST) {
+ SDValue Value = ST->getValue();
+ if (Value.getOpcode() == ISD::TargetConstantFP)
+ return SDValue();
+
+ SDLoc DL(ST);
+
+ SDValue Chain = ST->getChain();
+ SDValue Ptr = ST->getBasePtr();
+
+ const ConstantFPSDNode *CFP = cast<ConstantFPSDNode>(Value);
+
+ // NOTE: If the original store is volatile, this transform must not increase
+ // the number of stores. For example, on x86-32 an f64 can be stored in one
+ // processor operation but an i64 (which is not legal) requires two. So the
+ // transform should not be done in this case.
+
+ SDValue Tmp;
+ switch (CFP->getSimpleValueType(0).SimpleTy) {
+ default:
+ llvm_unreachable("Unknown FP type");
+ case MVT::f16: // We don't do this for these yet.
+ case MVT::f80:
+ case MVT::f128:
+ case MVT::ppcf128:
+ return SDValue();
+ case MVT::f32:
+ if ((isTypeLegal(MVT::i32) && !LegalOperations && !ST->isVolatile()) ||
+ TLI.isOperationLegalOrCustom(ISD::STORE, MVT::i32)) {
+ ;
+ Tmp = DAG.getConstant((uint32_t)CFP->getValueAPF().
+ bitcastToAPInt().getZExtValue(), SDLoc(CFP),
+ MVT::i32);
+ return DAG.getStore(Chain, DL, Tmp, Ptr, ST->getMemOperand());
+ }
+
+ return SDValue();
+ case MVT::f64:
+ if ((TLI.isTypeLegal(MVT::i64) && !LegalOperations &&
+ !ST->isVolatile()) ||
+ TLI.isOperationLegalOrCustom(ISD::STORE, MVT::i64)) {
+ ;
+ Tmp = DAG.getConstant(CFP->getValueAPF().bitcastToAPInt().
+ getZExtValue(), SDLoc(CFP), MVT::i64);
+ return DAG.getStore(Chain, DL, Tmp,
+ Ptr, ST->getMemOperand());
+ }
+
+ if (!ST->isVolatile() &&
+ TLI.isOperationLegalOrCustom(ISD::STORE, MVT::i32)) {
+ // Many FP stores are not made apparent until after legalize, e.g. for
+ // argument passing. Since this is so common, custom legalize the
+ // 64-bit integer store into two 32-bit stores.
+ uint64_t Val = CFP->getValueAPF().bitcastToAPInt().getZExtValue();
+ SDValue Lo = DAG.getConstant(Val & 0xFFFFFFFF, SDLoc(CFP), MVT::i32);
+ SDValue Hi = DAG.getConstant(Val >> 32, SDLoc(CFP), MVT::i32);
+ if (DAG.getDataLayout().isBigEndian())
+ std::swap(Lo, Hi);
+
+ unsigned Alignment = ST->getAlignment();
+ bool isVolatile = ST->isVolatile();
+ bool isNonTemporal = ST->isNonTemporal();
+ AAMDNodes AAInfo = ST->getAAInfo();
+
+ SDValue St0 = DAG.getStore(Chain, DL, Lo,
+ Ptr, ST->getPointerInfo(),
+ isVolatile, isNonTemporal,
+ ST->getAlignment(), AAInfo);
+ Ptr = DAG.getNode(ISD::ADD, DL, Ptr.getValueType(), Ptr,
+ DAG.getConstant(4, DL, Ptr.getValueType()));
+ Alignment = MinAlign(Alignment, 4U);
+ SDValue St1 = DAG.getStore(Chain, DL, Hi,
+ Ptr, ST->getPointerInfo().getWithOffset(4),
+ isVolatile, isNonTemporal,
+ Alignment, AAInfo);
+ return DAG.getNode(ISD::TokenFactor, DL, MVT::Other,
+ St0, St1);
+ }
+
+ return SDValue();
+ }
+}
+
SDValue DAGCombiner::visitSTORE(SDNode *N) {
StoreSDNode *ST = cast<StoreSDNode>(N);
SDValue Chain = ST->getChain();
ST->isUnindexed()) {
unsigned OrigAlign = ST->getAlignment();
EVT SVT = Value.getOperand(0).getValueType();
- unsigned Align = TLI.getDataLayout()->
- getABITypeAlignment(SVT.getTypeForEVT(*DAG.getContext()));
+ unsigned Align = DAG.getDataLayout().getABITypeAlignment(
+ SVT.getTypeForEVT(*DAG.getContext()));
if (Align <= OrigAlign &&
((!LegalOperations && !ST->isVolatile()) ||
TLI.isOperationLegalOrCustom(ISD::STORE, SVT)))
if (Value.getOpcode() == ISD::UNDEF && ST->isUnindexed())
return Chain;
- // Turn 'store float 1.0, Ptr' -> 'store int 0x12345678, Ptr'
- if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(Value)) {
- // NOTE: If the original store is volatile, this transform must not increase
- // the number of stores. For example, on x86-32 an f64 can be stored in one
- // processor operation but an i64 (which is not legal) requires two. So the
- // transform should not be done in this case.
- if (Value.getOpcode() != ISD::TargetConstantFP) {
- SDValue Tmp;
- switch (CFP->getSimpleValueType(0).SimpleTy) {
- default: llvm_unreachable("Unknown FP type");
- case MVT::f16: // We don't do this for these yet.
- case MVT::f80:
- case MVT::f128:
- case MVT::ppcf128:
- break;
- case MVT::f32:
- if ((isTypeLegal(MVT::i32) && !LegalOperations && !ST->isVolatile()) ||
- TLI.isOperationLegalOrCustom(ISD::STORE, MVT::i32)) {
- Tmp = DAG.getConstant((uint32_t)CFP->getValueAPF().
- bitcastToAPInt().getZExtValue(), MVT::i32);
- return DAG.getStore(Chain, SDLoc(N), Tmp,
- Ptr, ST->getMemOperand());
- }
- break;
- case MVT::f64:
- if ((TLI.isTypeLegal(MVT::i64) && !LegalOperations &&
- !ST->isVolatile()) ||
- TLI.isOperationLegalOrCustom(ISD::STORE, MVT::i64)) {
- Tmp = DAG.getConstant(CFP->getValueAPF().bitcastToAPInt().
- getZExtValue(), MVT::i64);
- return DAG.getStore(Chain, SDLoc(N), Tmp,
- Ptr, ST->getMemOperand());
- }
-
- if (!ST->isVolatile() &&
- TLI.isOperationLegalOrCustom(ISD::STORE, MVT::i32)) {
- // Many FP stores are not made apparent until after legalize, e.g. for
- // argument passing. Since this is so common, custom legalize the
- // 64-bit integer store into two 32-bit stores.
- uint64_t Val = CFP->getValueAPF().bitcastToAPInt().getZExtValue();
- SDValue Lo = DAG.getConstant(Val & 0xFFFFFFFF, MVT::i32);
- SDValue Hi = DAG.getConstant(Val >> 32, MVT::i32);
- if (TLI.isBigEndian()) std::swap(Lo, Hi);
-
- unsigned Alignment = ST->getAlignment();
- bool isVolatile = ST->isVolatile();
- bool isNonTemporal = ST->isNonTemporal();
- AAMDNodes AAInfo = ST->getAAInfo();
-
- SDValue St0 = DAG.getStore(Chain, SDLoc(ST), Lo,
- Ptr, ST->getPointerInfo(),
- isVolatile, isNonTemporal,
- ST->getAlignment(), AAInfo);
- Ptr = DAG.getNode(ISD::ADD, SDLoc(N), Ptr.getValueType(), Ptr,
- DAG.getConstant(4, Ptr.getValueType()));
- Alignment = MinAlign(Alignment, 4U);
- SDValue St1 = DAG.getStore(Chain, SDLoc(ST), Hi,
- Ptr, ST->getPointerInfo().getWithOffset(4),
- isVolatile, isNonTemporal,
- Alignment, AAInfo);
- return DAG.getNode(ISD::TokenFactor, SDLoc(N), MVT::Other,
- St0, St1);
- }
-
- break;
- }
- }
- }
-
// Try to infer better alignment information than the store already has.
if (OptLevel != CodeGenOpt::None && ST->isUnindexed()) {
if (unsigned Align = DAG.InferPtrAlignment(Ptr)) {
// Try transforming a pair floating point load / store ops to integer
// load / store ops.
- SDValue NewST = TransformFPLoadStorePair(N);
- if (NewST.getNode())
+ if (SDValue NewST = TransformFPLoadStorePair(N))
return NewST;
bool UseAA = CombinerAA.getNumOccurrences() > 0 ? CombinerAA
UseAA = false;
#endif
if (UseAA && ST->isUnindexed()) {
- // Walk up chain skipping non-aliasing memory nodes.
- SDValue BetterChain = FindBetterChain(N, Chain);
+ // FIXME: We should do this even without AA enabled. AA will just allow
+ // FindBetterChain to work in more situations. The problem with this is that
+ // any combine that expects memory operations to be on consecutive chains
+ // first needs to be updated to look for users of the same chain.
- // If there is a better chain.
- if (Chain != BetterChain) {
- SDValue ReplStore;
-
- // Replace the chain to avoid dependency.
- if (ST->isTruncatingStore()) {
- ReplStore = DAG.getTruncStore(BetterChain, SDLoc(N), Value, Ptr,
- ST->getMemoryVT(), ST->getMemOperand());
- } else {
- ReplStore = DAG.getStore(BetterChain, SDLoc(N), Value, Ptr,
- ST->getMemOperand());
- }
-
- // Create token to keep both nodes around.
- SDValue Token = DAG.getNode(ISD::TokenFactor, SDLoc(N),
- MVT::Other, Chain, ReplStore);
-
- // Make sure the new and old chains are cleaned up.
- AddToWorklist(Token.getNode());
-
- // Don't add users to work list.
- return CombineTo(N, Token, false);
+ // Walk up chain skipping non-aliasing memory nodes, on this store and any
+ // adjacent stores.
+ if (findBetterNeighborChains(ST)) {
+ // replaceStoreChain uses CombineTo, which handled all of the worklist
+ // manipulation. Return the original node to not do anything else.
+ return SDValue(ST, 0);
}
}
return SDValue(N, 0);
}
+ // Turn 'store float 1.0, Ptr' -> 'store int 0x12345678, Ptr'
+ //
+ // Make sure to do this only after attempting to merge stores in order to
+ // avoid changing the types of some subset of stores due to visit order,
+ // preventing their merging.
+ if (isa<ConstantFPSDNode>(Value)) {
+ if (SDValue NewSt = replaceStoreOfFPConstant(ST))
+ return NewSt;
+ }
+
return ReduceLoadOpStoreWidth(N);
}
EVT ResultVT = EVE->getValueType(0);
EVT VecEltVT = InVecVT.getVectorElementType();
unsigned Align = OriginalLoad->getAlignment();
- unsigned NewAlign = TLI.getDataLayout()->getABITypeAlignment(
+ unsigned NewAlign = DAG.getDataLayout().getABITypeAlignment(
VecEltVT.getTypeForEVT(*DAG.getContext()));
if (NewAlign > Align || !TLI.isOperationLegalOrCustom(ISD::LOAD, VecEltVT))
SDValue Offset;
EVT PtrType = NewPtr.getValueType();
MachinePointerInfo MPI;
+ SDLoc DL(EVE);
if (auto *ConstEltNo = dyn_cast<ConstantSDNode>(EltNo)) {
int Elt = ConstEltNo->getZExtValue();
unsigned PtrOff = VecEltVT.getSizeInBits() * Elt / 8;
- if (TLI.isBigEndian())
- PtrOff = InVecVT.getSizeInBits() / 8 - PtrOff;
- Offset = DAG.getConstant(PtrOff, PtrType);
+ Offset = DAG.getConstant(PtrOff, DL, PtrType);
MPI = OriginalLoad->getPointerInfo().getWithOffset(PtrOff);
} else {
+ Offset = DAG.getZExtOrTrunc(EltNo, DL, PtrType);
Offset = DAG.getNode(
- ISD::MUL, SDLoc(EVE), EltNo.getValueType(), EltNo,
- DAG.getConstant(VecEltVT.getStoreSize(), EltNo.getValueType()));
- if (TLI.isBigEndian())
- Offset = DAG.getNode(
- ISD::SUB, SDLoc(EVE), EltNo.getValueType(),
- DAG.getConstant(InVecVT.getStoreSize(), EltNo.getValueType()), Offset);
+ ISD::MUL, DL, PtrType, Offset,
+ DAG.getConstant(VecEltVT.getStoreSize(), DL, PtrType));
MPI = OriginalLoad->getPointerInfo();
}
- NewPtr = DAG.getNode(ISD::ADD, SDLoc(EVE), PtrType, NewPtr, Offset);
+ NewPtr = DAG.getNode(ISD::ADD, DL, PtrType, NewPtr, Offset);
// The replacement we need to do here is a little tricky: we need to
// replace an extractelement of a load with a load.
}
SDValue EltNo = N->getOperand(1);
- bool ConstEltNo = isa<ConstantSDNode>(EltNo);
+ ConstantSDNode *ConstEltNo = dyn_cast<ConstantSDNode>(EltNo);
+
+ // extract_vector_elt (build_vector x, y), 1 -> y
+ if (ConstEltNo &&
+ InVec.getOpcode() == ISD::BUILD_VECTOR &&
+ TLI.isTypeLegal(VT) &&
+ (InVec.hasOneUse() ||
+ TLI.aggressivelyPreferBuildVectorSources(VT))) {
+ SDValue Elt = InVec.getOperand(ConstEltNo->getZExtValue());
+ EVT InEltVT = Elt.getValueType();
+
+ // Sometimes build_vector's scalar input types do not match result type.
+ if (NVT == InEltVT)
+ return Elt;
+
+ // TODO: It may be useful to truncate if free if the build_vector implicitly
+ // converts.
+ }
// Transform: (EXTRACT_VECTOR_ELT( VECTOR_SHUFFLE )) -> EXTRACT_VECTOR_ELT.
// We only perform this optimization before the op legalization phase because
// patterns. For example on AVX, extracting elements from a wide vector
// without using extract_subvector. However, if we can find an underlying
// scalar value, then we can always use that.
- if (InVec.getOpcode() == ISD::VECTOR_SHUFFLE
- && ConstEltNo) {
- int Elt = cast<ConstantSDNode>(EltNo)->getZExtValue();
+ if (ConstEltNo && InVec.getOpcode() == ISD::VECTOR_SHUFFLE) {
int NumElem = VT.getVectorNumElements();
ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(InVec);
// Find the new index to extract from.
- int OrigElt = SVOp->getMaskElt(Elt);
+ int OrigElt = SVOp->getMaskElt(ConstEltNo->getZExtValue());
// Extracting an undef index is undef.
if (OrigElt == -1)
// scalar_to_vector here as well.
if (!LegalOperations) {
- EVT IndexTy = TLI.getVectorIdxTy();
- return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(N), NVT,
- SVInVec, DAG.getConstant(OrigElt, IndexTy));
+ EVT IndexTy = TLI.getVectorIdxTy(DAG.getDataLayout());
+ return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(N), NVT, SVInVec,
+ DAG.getConstant(OrigElt, SDLoc(SVOp), IndexTy));
}
}
if (ISD::isNormalLoad(InVec.getNode())) {
LN0 = cast<LoadSDNode>(InVec);
Elt = (Idx < (int)NumElems) ? Idx : Idx - (int)NumElems;
- EltNo = DAG.getConstant(Elt, EltNo.getValueType());
+ EltNo = DAG.getConstant(Elt, SDLoc(EltNo), EltNo.getValueType());
}
}
if (!ValidTypes)
return SDValue();
- bool isLE = TLI.isLittleEndian();
+ bool isLE = DAG.getDataLayout().isLittleEndian();
unsigned ElemRatio = OutScalarTy.getSizeInBits()/SourceType.getSizeInBits();
assert(ElemRatio > 1 && "Invalid element size ratio");
SDValue Filler = AllAnyExt ? DAG.getUNDEF(SourceType):
- DAG.getConstant(0, SourceType);
+ DAG.getConstant(0, SDLoc(N), SourceType);
unsigned NewBVElems = ElemRatio * VT.getVectorNumElements();
SmallVector<SDValue, 8> Ops(NewBVElems, Filler);
if (Op.getOpcode() == ISD::UNDEF) continue;
// See if we can combine this build_vector into a blend with a zero vector.
- if (!VecIn2.getNode() && ((Op.getOpcode() == ISD::Constant &&
- cast<ConstantSDNode>(Op.getNode())->isNullValue()) ||
- (Op.getOpcode() == ISD::ConstantFP &&
- cast<ConstantFPSDNode>(Op.getNode())->getValueAPF().isZero()))) {
+ if (!VecIn2.getNode() && (isNullConstant(Op) || isNullFPConstant(Op))) {
UsesZeroVector = true;
continue;
}
// Try to replace VecIn1 with two extract_subvectors
// No need to update the masks, they should still be correct.
- VecIn2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, VecIn1,
- DAG.getConstant(VT.getVectorNumElements(), TLI.getVectorIdxTy()));
- VecIn1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, VecIn1,
- DAG.getConstant(0, TLI.getVectorIdxTy()));
+ VecIn2 = DAG.getNode(
+ ISD::EXTRACT_SUBVECTOR, dl, VT, VecIn1,
+ DAG.getConstant(VT.getVectorNumElements(), dl,
+ TLI.getVectorIdxTy(DAG.getDataLayout())));
+ VecIn1 = DAG.getNode(
+ ISD::EXTRACT_SUBVECTOR, dl, VT, VecIn1,
+ DAG.getConstant(0, dl, TLI.getVectorIdxTy(DAG.getDataLayout())));
} else
return SDValue();
}
if (UsesZeroVector)
- VecIn2 = VT.isInteger() ? DAG.getConstant(0, VT) :
- DAG.getConstantFP(0.0, VT);
+ VecIn2 = VT.isInteger() ? DAG.getConstant(0, dl, VT) :
+ DAG.getConstantFP(0.0, dl, VT);
else
// If VecIn2 is unused then change it to undef.
VecIn2 = VecIn2.getNode() ? VecIn2 : DAG.getUNDEF(VT);
}
// 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());
DAG.getNode(ISD::BUILD_VECTOR, DL, VecVT, Ops));
}
-SDValue DAGCombiner::visitCONCAT_VECTORS(SDNode *N) {
- // TODO: Check to see if this is a CONCAT_VECTORS of a bunch of
- // EXTRACT_SUBVECTOR operations. If so, and if the EXTRACT_SUBVECTOR vector
- // inputs come from at most two distinct vectors, turn this into a shuffle
- // node.
+// Check to see if this is a CONCAT_VECTORS of a bunch of EXTRACT_SUBVECTOR
+// operations. If so, and if the EXTRACT_SUBVECTOR vector inputs come from at
+// most two distinct vectors the same size as the result, attempt to turn this
+// into a legal shuffle.
+static SDValue combineConcatVectorOfExtracts(SDNode *N, SelectionDAG &DAG) {
+ EVT VT = N->getValueType(0);
+ EVT OpVT = N->getOperand(0).getValueType();
+ int NumElts = VT.getVectorNumElements();
+ int NumOpElts = OpVT.getVectorNumElements();
+
+ SDValue SV0 = DAG.getUNDEF(VT), SV1 = DAG.getUNDEF(VT);
+ SmallVector<int, 8> Mask;
+
+ for (SDValue Op : N->ops()) {
+ // Peek through any bitcast.
+ while (Op.getOpcode() == ISD::BITCAST)
+ Op = Op.getOperand(0);
+
+ // UNDEF nodes convert to UNDEF shuffle mask values.
+ if (Op.getOpcode() == ISD::UNDEF) {
+ Mask.append((unsigned)NumOpElts, -1);
+ continue;
+ }
+
+ if (Op.getOpcode() != ISD::EXTRACT_SUBVECTOR)
+ return SDValue();
+
+ // What vector are we extracting the subvector from and at what index?
+ SDValue ExtVec = Op.getOperand(0);
+
+ // We want the EVT of the original extraction to correctly scale the
+ // extraction index.
+ EVT ExtVT = ExtVec.getValueType();
+
+ // Peek through any bitcast.
+ while (ExtVec.getOpcode() == ISD::BITCAST)
+ ExtVec = ExtVec.getOperand(0);
+
+ // UNDEF nodes convert to UNDEF shuffle mask values.
+ if (ExtVec.getOpcode() == ISD::UNDEF) {
+ Mask.append((unsigned)NumOpElts, -1);
+ continue;
+ }
+
+ if (!isa<ConstantSDNode>(Op.getOperand(1)))
+ return SDValue();
+ int ExtIdx = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
+
+ // Ensure that we are extracting a subvector from a vector the same
+ // size as the result.
+ if (ExtVT.getSizeInBits() != VT.getSizeInBits())
+ return SDValue();
+
+ // Scale the subvector index to account for any bitcast.
+ int NumExtElts = ExtVT.getVectorNumElements();
+ if (0 == (NumExtElts % NumElts))
+ ExtIdx /= (NumExtElts / NumElts);
+ else if (0 == (NumElts % NumExtElts))
+ ExtIdx *= (NumElts / NumExtElts);
+ else
+ return SDValue();
+
+ // At most we can reference 2 inputs in the final shuffle.
+ if (SV0.getOpcode() == ISD::UNDEF || SV0 == ExtVec) {
+ SV0 = ExtVec;
+ for (int i = 0; i != NumOpElts; ++i)
+ Mask.push_back(i + ExtIdx);
+ } else if (SV1.getOpcode() == ISD::UNDEF || SV1 == ExtVec) {
+ SV1 = ExtVec;
+ for (int i = 0; i != NumOpElts; ++i)
+ Mask.push_back(i + ExtIdx + NumElts);
+ } else {
+ return SDValue();
+ }
+ }
+
+ if (!DAG.getTargetLoweringInfo().isShuffleMaskLegal(Mask, VT))
+ return SDValue();
+
+ return DAG.getVectorShuffle(VT, SDLoc(N), DAG.getBitcast(VT, SV0),
+ DAG.getBitcast(VT, SV1), Mask);
+}
+SDValue DAGCombiner::visitCONCAT_VECTORS(SDNode *N) {
// If we only have one input vector, we don't need to do any concatenation.
if (N->getNumOperands() == 1)
return N->getOperand(0);
if (SDValue V = combineConcatVectorOfScalars(N, DAG))
return V;
+ // Fold CONCAT_VECTORS of EXTRACT_SUBVECTOR (or undef) to VECTOR_SHUFFLE.
+ if (Level < AfterLegalizeVectorOps && TLI.isTypeLegal(VT))
+ if (SDValue V = combineConcatVectorOfExtracts(N, DAG))
+ return V;
+
// Type legalization of vectors and DAG canonicalization of SHUFFLE_VECTOR
// nodes often generate nop CONCAT_VECTOR nodes.
// Scan the CONCAT_VECTOR operands and look for a CONCAT operations that
std::all_of(SVN->getMask().begin() + NumElemsPerConcat,
SVN->getMask().end(), [](int i) { return i == -1; })) {
N0 = DAG.getVectorShuffle(ConcatVT, SDLoc(N), N0.getOperand(0), N0.getOperand(1),
- ArrayRef<int>(SVN->getMask().begin(), NumElemsPerConcat));
+ makeArrayRef(SVN->getMask().begin(), NumElemsPerConcat));
N1 = DAG.getUNDEF(ConcatVT);
return DAG.getNode(ISD::CONCAT_VECTORS, SDLoc(N), VT, N0, N1);
}
return SDValue();
}
+SDValue DAGCombiner::visitFP16_TO_FP(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+
+ // fold fp16_to_fp(op & 0xffff) -> fp16_to_fp(op)
+ if (N0->getOpcode() == ISD::AND) {
+ ConstantSDNode *AndConst = getAsNonOpaqueConstant(N0.getOperand(1));
+ if (AndConst && AndConst->getAPIntValue() == 0xffff) {
+ return DAG.getNode(ISD::FP16_TO_FP, SDLoc(N), N->getValueType(0),
+ N0.getOperand(0));
+ }
+ }
+
+ return SDValue();
+}
+
/// Returns a vector_shuffle if it able to transform an AND to a vector_shuffle
/// with the destination vector and a zero vector.
/// e.g. AND V, <0xffffffff, 0, 0xffffffff, 0>. ==>
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 (RHS.getOpcode() == ISD::BITCAST)
RHS = RHS.getOperand(0);
- if (RHS.getOpcode() == ISD::BUILD_VECTOR) {
+ if (RHS.getOpcode() != ISD::BUILD_VECTOR)
+ return SDValue();
+
+ EVT RVT = RHS.getValueType();
+ unsigned NumElts = RHS.getNumOperands();
+
+ // Attempt to create a valid clear mask, splitting the mask into
+ // sub elements and checking to see if each is
+ // all zeros or all ones - suitable for shuffle masking.
+ auto BuildClearMask = [&](int Split) {
+ int NumSubElts = NumElts * Split;
+ int NumSubBits = RVT.getScalarSizeInBits() / Split;
+
SmallVector<int, 8> Indices;
- unsigned NumElts = RHS.getNumOperands();
+ for (int i = 0; i != NumSubElts; ++i) {
+ int EltIdx = i / Split;
+ int SubIdx = i % Split;
+ SDValue Elt = RHS.getOperand(EltIdx);
+ if (Elt.getOpcode() == ISD::UNDEF) {
+ Indices.push_back(-1);
+ continue;
+ }
- for (unsigned i = 0; i != NumElts; ++i) {
- SDValue Elt = RHS.getOperand(i);
- if (!isa<ConstantSDNode>(Elt))
+ APInt Bits;
+ if (isa<ConstantSDNode>(Elt))
+ Bits = cast<ConstantSDNode>(Elt)->getAPIntValue();
+ else if (isa<ConstantFPSDNode>(Elt))
+ Bits = cast<ConstantFPSDNode>(Elt)->getValueAPF().bitcastToAPInt();
+ else
return SDValue();
- if (cast<ConstantSDNode>(Elt)->isAllOnesValue())
+ // Extract the sub element from the constant bit mask.
+ if (DAG.getDataLayout().isBigEndian()) {
+ Bits = Bits.lshr((Split - SubIdx - 1) * NumSubBits);
+ } else {
+ Bits = Bits.lshr(SubIdx * NumSubBits);
+ }
+
+ if (Split > 1)
+ Bits = Bits.trunc(NumSubBits);
+
+ if (Bits.isAllOnesValue())
Indices.push_back(i);
- else if (cast<ConstantSDNode>(Elt)->isNullValue())
- Indices.push_back(NumElts+i);
+ else if (Bits == 0)
+ Indices.push_back(i + NumSubElts);
else
return SDValue();
}
// Let's see if the target supports this vector_shuffle.
- EVT RVT = RHS.getValueType();
- if (!TLI.isVectorClearMaskLegal(Indices, RVT))
+ EVT ClearSVT = EVT::getIntegerVT(*DAG.getContext(), NumSubBits);
+ EVT ClearVT = EVT::getVectorVT(*DAG.getContext(), ClearSVT, NumSubElts);
+ if (!TLI.isVectorClearMaskLegal(Indices, ClearVT))
return SDValue();
- // Return the new VECTOR_SHUFFLE node.
- EVT EltVT = RVT.getVectorElementType();
- SmallVector<SDValue,8> ZeroOps(RVT.getVectorNumElements(),
- DAG.getConstant(0, EltVT));
- SDValue Zero = DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), RVT, ZeroOps);
- LHS = DAG.getNode(ISD::BITCAST, dl, RVT, LHS);
- SDValue Shuf = DAG.getVectorShuffle(RVT, dl, LHS, Zero, &Indices[0]);
- return DAG.getNode(ISD::BITCAST, dl, VT, Shuf);
- }
+ SDValue Zero = DAG.getConstant(0, dl, ClearVT);
+ return DAG.getBitcast(VT, DAG.getVectorShuffle(ClearVT, dl,
+ DAG.getBitcast(ClearVT, LHS),
+ Zero, &Indices[0]));
+ };
+
+ // Determine maximum split level (byte level masking).
+ int MaxSplit = 1;
+ if (RVT.getScalarSizeInBits() % 8 == 0)
+ MaxSplit = RVT.getScalarSizeInBits() / 8;
+
+ for (int Split = 1; Split <= MaxSplit; ++Split)
+ if (RVT.getScalarSizeInBits() % Split == 0)
+ if (SDValue S = BuildClearMask(Split))
+ return S;
return SDValue();
}
SDValue LHS = N->getOperand(0);
SDValue RHS = N->getOperand(1);
+ SDValue Ops[] = {LHS, RHS};
+ // See if we can constant fold the vector operation.
+ if (SDValue Fold = DAG.FoldConstantVectorArithmetic(
+ N->getOpcode(), SDLoc(LHS), LHS.getValueType(), Ops, N->getFlags()))
+ return Fold;
+
+ // Try to convert a constant mask AND into a shuffle clear mask.
if (SDValue Shuffle = XformToShuffleWithZero(N))
return Shuffle;
- // If the LHS and RHS are BUILD_VECTOR nodes, see if we can constant fold
- // this operation.
- if (LHS.getOpcode() == ISD::BUILD_VECTOR &&
- RHS.getOpcode() == ISD::BUILD_VECTOR) {
- // Check if both vectors are constants. If not bail out.
- if (!(cast<BuildVectorSDNode>(LHS)->isConstant() &&
- cast<BuildVectorSDNode>(RHS)->isConstant()))
- return SDValue();
-
- SmallVector<SDValue, 8> Ops;
- for (unsigned i = 0, e = LHS.getNumOperands(); i != e; ++i) {
- SDValue LHSOp = LHS.getOperand(i);
- SDValue RHSOp = RHS.getOperand(i);
-
- // Can't fold divide by zero.
- if (N->getOpcode() == ISD::SDIV || N->getOpcode() == ISD::UDIV ||
- N->getOpcode() == ISD::FDIV) {
- if ((RHSOp.getOpcode() == ISD::Constant &&
- cast<ConstantSDNode>(RHSOp.getNode())->isNullValue()) ||
- (RHSOp.getOpcode() == ISD::ConstantFP &&
- cast<ConstantFPSDNode>(RHSOp.getNode())->getValueAPF().isZero()))
- break;
- }
-
- EVT VT = LHSOp.getValueType();
- EVT RVT = RHSOp.getValueType();
- if (RVT != VT) {
- // Integer BUILD_VECTOR operands may have types larger than the element
- // size (e.g., when the element type is not legal). Prior to type
- // legalization, the types may not match between the two BUILD_VECTORS.
- // Truncate one of the operands to make them match.
- if (RVT.getSizeInBits() > VT.getSizeInBits()) {
- RHSOp = DAG.getNode(ISD::TRUNCATE, SDLoc(N), VT, RHSOp);
- } else {
- LHSOp = DAG.getNode(ISD::TRUNCATE, SDLoc(N), RVT, LHSOp);
- VT = RVT;
- }
- }
- SDValue FoldOp = DAG.getNode(N->getOpcode(), SDLoc(LHS), VT,
- LHSOp, RHSOp);
- if (FoldOp.getOpcode() != ISD::UNDEF &&
- FoldOp.getOpcode() != ISD::Constant &&
- FoldOp.getOpcode() != ISD::ConstantFP)
- break;
- Ops.push_back(FoldOp);
- AddToWorklist(FoldOp.getNode());
- }
-
- if (Ops.size() == LHS.getNumOperands())
- return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), LHS.getValueType(), Ops);
- }
-
// Type legalization might introduce new shuffles in the DAG.
// Fold (VBinOp (shuffle (A, Undef, Mask)), (shuffle (B, Undef, Mask)))
// -> (shuffle (VBinOp (A, B)), Undef, Mask).
EVT VT = N->getValueType(0);
SDValue UndefVector = LHS.getOperand(1);
SDValue NewBinOp = DAG.getNode(N->getOpcode(), SDLoc(N), VT,
- LHS.getOperand(0), RHS.getOperand(0));
+ LHS.getOperand(0), RHS.getOperand(0),
+ N->getFlags());
AddUsersToWorklist(N);
return DAG.getVectorShuffle(VT, SDLoc(N), NewBinOp, UndefVector,
&SVN0->getMask()[0]);
if (LHS.getOperand(0) != RHS.getOperand(0) ||
// Do not let this transformation reduce the number of volatile loads.
LLD->isVolatile() || RLD->isVolatile() ||
+ // FIXME: If either is a pre/post inc/dec load,
+ // we'd need to split out the address adjustment.
+ LLD->isIndexed() || RLD->isIndexed() ||
// If this is an EXTLOAD, the VT's must match.
LLD->getMemoryVT() != RLD->getMemoryVT() ||
// If this is an EXTLOAD, the kind of extension must match.
EVT VT = N2.getValueType();
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.getNode());
ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2.getNode());
- ConstantSDNode *N3C = dyn_cast<ConstantSDNode>(N3.getNode());
// Determine if the condition we're dealing with is constant
SDValue SCC = SimplifySetCC(getSetCCResultType(N0.getValueType()),
N0, N1, CC, DL, false);
if (SCC.getNode()) AddToWorklist(SCC.getNode());
- ConstantSDNode *SCCC = dyn_cast_or_null<ConstantSDNode>(SCC.getNode());
- // fold select_cc true, x, y -> x
- if (SCCC && !SCCC->isNullValue())
- return N2;
- // fold select_cc false, x, y -> y
- if (SCCC && SCCC->isNullValue())
- return N3;
+ if (ConstantSDNode *SCCC = dyn_cast_or_null<ConstantSDNode>(SCC.getNode())) {
+ // fold select_cc true, x, y -> x
+ // fold select_cc false, x, y -> y
+ return !SCCC->isNullValue() ? N2 : N3;
+ }
// 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 &&
const_cast<ConstantFP*>(TV->getConstantFPValue())
};
Type *FPTy = Elts[0]->getType();
- const DataLayout &TD = *TLI.getDataLayout();
+ const DataLayout &TD = DAG.getDataLayout();
// Create a ConstantArray of the two constants.
Constant *CA = ConstantArray::get(ArrayType::get(FPTy, 2), Elts);
- SDValue CPIdx = DAG.getConstantPool(CA, TLI.getPointerTy(),
- TD.getPrefTypeAlignment(FPTy));
+ SDValue CPIdx =
+ DAG.getConstantPool(CA, TLI.getPointerTy(DAG.getDataLayout()),
+ TD.getPrefTypeAlignment(FPTy));
unsigned Alignment = cast<ConstantPoolSDNode>(CPIdx)->getAlignment();
// Get the offsets to the 0 and 1 element of the array so that we can
// select between them.
- SDValue Zero = DAG.getIntPtrConstant(0);
+ SDValue Zero = DAG.getIntPtrConstant(0, DL);
unsigned EltSize = (unsigned)TD.getTypeAllocSize(Elts[0]->getType());
- SDValue One = DAG.getIntPtrConstant(EltSize);
+ SDValue One = DAG.getIntPtrConstant(EltSize, SDLoc(FV));
SDValue Cond = DAG.getSetCC(DL,
getSetCCResultType(N0.getValueType()),
CPIdx = DAG.getNode(ISD::ADD, DL, CPIdx.getValueType(), CPIdx,
CstOffset);
AddToWorklist(CPIdx.getNode());
- return DAG.getLoad(TV->getValueType(0), DL, DAG.getEntryNode(), CPIdx,
- MachinePointerInfo::getConstantPool(), false,
- false, false, Alignment);
-
+ return DAG.getLoad(
+ TV->getValueType(0), DL, DAG.getEntryNode(), CPIdx,
+ MachinePointerInfo::getConstantPool(DAG.getMachineFunction()),
+ false, false, false, Alignment);
}
}
// Check to see if we can perform the "gzip trick", transforming
// (select_cc setlt X, 0, A, 0) -> (and (sra X, (sub size(X), 1), A)
- if (N1C && N3C && N3C->isNullValue() && CC == ISD::SETLT &&
- (N1C->isNullValue() || // (a < 0) ? b : 0
- (N1C->getAPIntValue() == 1 && N0 == N2))) { // (a < 1) ? a : 0
+ if (isNullConstant(N3) && CC == ISD::SETLT &&
+ (isNullConstant(N1) || // (a < 0) ? b : 0
+ (isOneConstant(N1) && N0 == N2))) { // (a < 1) ? a : 0
EVT XType = N0.getValueType();
EVT AType = N2.getValueType();
if (XType.bitsGE(AType)) {
// and (sra X, size(X)-1, A) -> "and (srl X, C2), A" iff A is a
// single-bit constant.
- if (N2C && ((N2C->getAPIntValue() & (N2C->getAPIntValue()-1)) == 0)) {
+ if (N2C && ((N2C->getAPIntValue() & (N2C->getAPIntValue() - 1)) == 0)) {
unsigned ShCtV = N2C->getAPIntValue().logBase2();
- ShCtV = XType.getSizeInBits()-ShCtV-1;
- SDValue ShCt = DAG.getConstant(ShCtV,
+ ShCtV = XType.getSizeInBits() - ShCtV - 1;
+ SDValue ShCt = DAG.getConstant(ShCtV, SDLoc(N0),
getShiftAmountTy(N0.getValueType()));
SDValue Shift = DAG.getNode(ISD::SRL, SDLoc(N0),
XType, N0, ShCt);
SDValue Shift = DAG.getNode(ISD::SRA, SDLoc(N0),
XType, N0,
- DAG.getConstant(XType.getSizeInBits()-1,
+ DAG.getConstant(XType.getSizeInBits() - 1,
+ SDLoc(N0),
getShiftAmountTy(N0.getValueType())));
AddToWorklist(Shift.getNode());
// single bit-test can be materialized as an all-ones register with
// shift-left and shift-right-arith.
if (CC == ISD::SETEQ && N0->getOpcode() == ISD::AND &&
- N0->getValueType(0) == VT &&
- N1C && N1C->isNullValue() &&
- N2C && N2C->isNullValue()) {
+ N0->getValueType(0) == VT && isNullConstant(N1) && isNullConstant(N2)) {
SDValue AndLHS = N0->getOperand(0);
ConstantSDNode *ConstAndRHS = dyn_cast<ConstantSDNode>(N0->getOperand(1));
if (ConstAndRHS && ConstAndRHS->getAPIntValue().countPopulation() == 1) {
// Shift the tested bit over the sign bit.
APInt AndMask = ConstAndRHS->getAPIntValue();
SDValue ShlAmt =
- DAG.getConstant(AndMask.countLeadingZeros(),
+ DAG.getConstant(AndMask.countLeadingZeros(), SDLoc(AndLHS),
getShiftAmountTy(AndLHS.getValueType()));
SDValue Shl = DAG.getNode(ISD::SHL, SDLoc(N0), VT, AndLHS, ShlAmt);
// Now arithmetic right shift it all the way over, so the result is either
// all-ones, or zero.
SDValue ShrAmt =
- DAG.getConstant(AndMask.getBitWidth()-1,
+ DAG.getConstant(AndMask.getBitWidth() - 1, SDLoc(Shl),
getShiftAmountTy(Shl.getValueType()));
SDValue Shr = DAG.getNode(ISD::SRA, SDLoc(N0), VT, Shl, ShrAmt);
}
// fold select C, 16, 0 -> shl C, 4
- if (N2C && N3C && N3C->isNullValue() && N2C->getAPIntValue().isPowerOf2() &&
+ if (N2C && isNullConstant(N3) && N2C->getAPIntValue().isPowerOf2() &&
TLI.getBooleanContents(N0.getValueType()) ==
TargetLowering::ZeroOrOneBooleanContent) {
// If the caller doesn't want us to simplify this into a zext of a compare,
// don't do it.
- if (NotExtCompare && N2C->getAPIntValue() == 1)
+ if (NotExtCompare && N2C->isOne())
return SDValue();
// Get a SetCC of the condition
// NOTE: Don't create a SETCC if it's not legal on this target.
if (!LegalOperations ||
- TLI.isOperationLegal(ISD::SETCC,
- LegalTypes ? getSetCCResultType(N0.getValueType()) : MVT::i1)) {
+ TLI.isOperationLegal(ISD::SETCC, N0.getValueType())) {
SDValue Temp, SCC;
// cast from setcc result type to select result type
if (LegalTypes) {
AddToWorklist(SCC.getNode());
AddToWorklist(Temp.getNode());
- if (N2C->getAPIntValue() == 1)
+ if (N2C->isOne())
return Temp;
// shl setcc result by log2 n2c
return DAG.getNode(
ISD::SHL, DL, N2.getValueType(), Temp,
- DAG.getConstant(N2C->getAPIntValue().logBase2(),
+ DAG.getConstant(N2C->getAPIntValue().logBase2(), SDLoc(Temp),
getShiftAmountTy(Temp.getValueType())));
}
}
- // Check to see if this is the equivalent of setcc
- // FIXME: Turn all of these into setcc if setcc if setcc is legal
- // otherwise, go ahead with the folds.
- if (0 && N3C && N3C->isNullValue() && N2C && (N2C->getAPIntValue() == 1ULL)) {
- EVT XType = N0.getValueType();
- if (!LegalOperations ||
- TLI.isOperationLegal(ISD::SETCC, getSetCCResultType(XType))) {
- SDValue Res = DAG.getSetCC(DL, getSetCCResultType(XType), N0, N1, CC);
- if (Res.getValueType() != VT)
- Res = DAG.getNode(ISD::ZERO_EXTEND, DL, VT, Res);
- return Res;
- }
-
- // fold (seteq X, 0) -> (srl (ctlz X, log2(size(X))))
- if (N1C && N1C->isNullValue() && CC == ISD::SETEQ &&
- (!LegalOperations ||
- TLI.isOperationLegal(ISD::CTLZ, XType))) {
- SDValue Ctlz = DAG.getNode(ISD::CTLZ, SDLoc(N0), XType, N0);
- return DAG.getNode(ISD::SRL, DL, XType, Ctlz,
- DAG.getConstant(Log2_32(XType.getSizeInBits()),
- getShiftAmountTy(Ctlz.getValueType())));
- }
- // fold (setgt X, 0) -> (srl (and (-X, ~X), size(X)-1))
- if (N1C && N1C->isNullValue() && CC == ISD::SETGT) {
- SDValue NegN0 = DAG.getNode(ISD::SUB, SDLoc(N0),
- XType, DAG.getConstant(0, XType), N0);
- SDValue NotN0 = DAG.getNOT(SDLoc(N0), N0, XType);
- return DAG.getNode(ISD::SRL, DL, XType,
- DAG.getNode(ISD::AND, DL, XType, NegN0, NotN0),
- DAG.getConstant(XType.getSizeInBits()-1,
- getShiftAmountTy(XType)));
- }
- // fold (setgt X, -1) -> (xor (srl (X, size(X)-1), 1))
- if (N1C && N1C->isAllOnesValue() && CC == ISD::SETGT) {
- SDValue Sign = DAG.getNode(ISD::SRL, SDLoc(N0), XType, N0,
- DAG.getConstant(XType.getSizeInBits()-1,
- getShiftAmountTy(N0.getValueType())));
- return DAG.getNode(ISD::XOR, DL, XType, Sign, DAG.getConstant(1, XType));
- }
- }
-
// Check to see if this is an integer abs.
// select_cc setg[te] X, 0, X, -X ->
// select_cc setgt X, -1, X, -X ->
EVT XType = N0.getValueType();
if (SubC && SubC->isNullValue() && XType.isInteger()) {
- SDValue Shift = DAG.getNode(ISD::SRA, SDLoc(N0), XType,
+ SDLoc DL(N0);
+ SDValue Shift = DAG.getNode(ISD::SRA, DL, XType,
N0,
- DAG.getConstant(XType.getSizeInBits()-1,
+ DAG.getConstant(XType.getSizeInBits() - 1, DL,
getShiftAmountTy(N0.getValueType())));
- SDValue Add = DAG.getNode(ISD::ADD, SDLoc(N0),
+ SDValue Add = DAG.getNode(ISD::ADD, DL,
XType, N0, Shift);
AddToWorklist(Shift.getNode());
AddToWorklist(Add.getNode());
return SDValue();
// Avoid division by zero.
- if (!C->getAPIntValue())
+ if (C->isNullValue())
return SDValue();
std::vector<SDNode*> Built;
return SDValue();
// Avoid division by zero.
- if (!C->getAPIntValue())
+ if (C->isNullValue())
return SDValue();
std::vector<SDNode *> Built;
return SDValue();
// Avoid division by zero.
- if (!C->getAPIntValue())
+ if (C->isNullValue())
return SDValue();
std::vector<SDNode*> Built;
return S;
}
-SDValue DAGCombiner::BuildReciprocalEstimate(SDValue Op) {
+SDValue DAGCombiner::BuildReciprocalEstimate(SDValue Op, SDNodeFlags *Flags) {
if (Level >= AfterLegalizeDAG)
return SDValue();
// does not require additional intermediate precision]
EVT VT = Op.getValueType();
SDLoc DL(Op);
- SDValue FPOne = DAG.getConstantFP(1.0, VT);
+ SDValue FPOne = DAG.getConstantFP(1.0, DL, VT);
AddToWorklist(Est.getNode());
// Newton iterations: Est = Est + Est (1 - Arg * Est)
for (unsigned i = 0; i < Iterations; ++i) {
- SDValue NewEst = DAG.getNode(ISD::FMUL, DL, VT, Op, Est);
+ SDValue NewEst = DAG.getNode(ISD::FMUL, DL, VT, Op, Est, Flags);
AddToWorklist(NewEst.getNode());
- NewEst = DAG.getNode(ISD::FSUB, DL, VT, FPOne, NewEst);
+ NewEst = DAG.getNode(ISD::FSUB, DL, VT, FPOne, NewEst, Flags);
AddToWorklist(NewEst.getNode());
- NewEst = DAG.getNode(ISD::FMUL, DL, VT, Est, NewEst);
+ NewEst = DAG.getNode(ISD::FMUL, DL, VT, Est, NewEst, Flags);
AddToWorklist(NewEst.getNode());
- Est = DAG.getNode(ISD::FADD, DL, VT, Est, NewEst);
+ Est = DAG.getNode(ISD::FADD, DL, VT, Est, NewEst, Flags);
AddToWorklist(Est.getNode());
}
}
/// X_{i+1} = X_i (1.5 - A X_i^2 / 2)
/// As a result, we precompute A/2 prior to the iteration loop.
SDValue DAGCombiner::BuildRsqrtNROneConst(SDValue Arg, SDValue Est,
- unsigned Iterations) {
+ unsigned Iterations,
+ SDNodeFlags *Flags) {
EVT VT = Arg.getValueType();
SDLoc DL(Arg);
- SDValue ThreeHalves = DAG.getConstantFP(1.5, VT);
+ SDValue ThreeHalves = DAG.getConstantFP(1.5, DL, VT);
// We now need 0.5 * Arg which we can write as (1.5 * Arg - Arg) so that
// this entire sequence requires only one FP constant.
- SDValue HalfArg = DAG.getNode(ISD::FMUL, DL, VT, ThreeHalves, Arg);
+ SDValue HalfArg = DAG.getNode(ISD::FMUL, DL, VT, ThreeHalves, Arg, Flags);
AddToWorklist(HalfArg.getNode());
- HalfArg = DAG.getNode(ISD::FSUB, DL, VT, HalfArg, Arg);
+ HalfArg = DAG.getNode(ISD::FSUB, DL, VT, HalfArg, Arg, Flags);
AddToWorklist(HalfArg.getNode());
// Newton iterations: Est = Est * (1.5 - HalfArg * Est * Est)
for (unsigned i = 0; i < Iterations; ++i) {
- SDValue NewEst = DAG.getNode(ISD::FMUL, DL, VT, Est, Est);
+ SDValue NewEst = DAG.getNode(ISD::FMUL, DL, VT, Est, Est, Flags);
AddToWorklist(NewEst.getNode());
- NewEst = DAG.getNode(ISD::FMUL, DL, VT, HalfArg, NewEst);
+ NewEst = DAG.getNode(ISD::FMUL, DL, VT, HalfArg, NewEst, Flags);
AddToWorklist(NewEst.getNode());
- NewEst = DAG.getNode(ISD::FSUB, DL, VT, ThreeHalves, NewEst);
+ NewEst = DAG.getNode(ISD::FSUB, DL, VT, ThreeHalves, NewEst, Flags);
AddToWorklist(NewEst.getNode());
- Est = DAG.getNode(ISD::FMUL, DL, VT, Est, NewEst);
+ Est = DAG.getNode(ISD::FMUL, DL, VT, Est, NewEst, Flags);
AddToWorklist(Est.getNode());
}
return Est;
/// =>
/// X_{i+1} = (-0.5 * X_i) * (A * X_i * X_i + (-3.0))
SDValue DAGCombiner::BuildRsqrtNRTwoConst(SDValue Arg, SDValue Est,
- unsigned Iterations) {
+ unsigned Iterations,
+ SDNodeFlags *Flags) {
EVT VT = Arg.getValueType();
SDLoc DL(Arg);
- SDValue MinusThree = DAG.getConstantFP(-3.0, VT);
- SDValue MinusHalf = DAG.getConstantFP(-0.5, VT);
+ SDValue MinusThree = DAG.getConstantFP(-3.0, DL, VT);
+ SDValue MinusHalf = DAG.getConstantFP(-0.5, DL, VT);
// Newton iterations: Est = -0.5 * Est * (-3.0 + Arg * Est * Est)
for (unsigned i = 0; i < Iterations; ++i) {
- SDValue HalfEst = DAG.getNode(ISD::FMUL, DL, VT, Est, MinusHalf);
+ SDValue HalfEst = DAG.getNode(ISD::FMUL, DL, VT, Est, MinusHalf, Flags);
AddToWorklist(HalfEst.getNode());
- Est = DAG.getNode(ISD::FMUL, DL, VT, Est, Est);
+ Est = DAG.getNode(ISD::FMUL, DL, VT, Est, Est, Flags);
AddToWorklist(Est.getNode());
- Est = DAG.getNode(ISD::FMUL, DL, VT, Est, Arg);
+ Est = DAG.getNode(ISD::FMUL, DL, VT, Est, Arg, Flags);
AddToWorklist(Est.getNode());
- Est = DAG.getNode(ISD::FADD, DL, VT, Est, MinusThree);
+ Est = DAG.getNode(ISD::FADD, DL, VT, Est, MinusThree, Flags);
AddToWorklist(Est.getNode());
- Est = DAG.getNode(ISD::FMUL, DL, VT, Est, HalfEst);
+ Est = DAG.getNode(ISD::FMUL, DL, VT, Est, HalfEst, Flags);
AddToWorklist(Est.getNode());
}
return Est;
}
-SDValue DAGCombiner::BuildRsqrtEstimate(SDValue Op) {
+SDValue DAGCombiner::BuildRsqrtEstimate(SDValue Op, SDNodeFlags *Flags) {
if (Level >= AfterLegalizeDAG)
return SDValue();
AddToWorklist(Est.getNode());
if (Iterations) {
Est = UseOneConstNR ?
- BuildRsqrtNROneConst(Op, Est, Iterations) :
- BuildRsqrtNRTwoConst(Op, Est, Iterations);
+ BuildRsqrtNROneConst(Op, Est, Iterations, Flags) :
+ BuildRsqrtNRTwoConst(Op, Est, Iterations, Flags);
}
return Est;
}
// If they are both volatile then they cannot be reordered.
if (Op0->isVolatile() && Op1->isVolatile()) return true;
+ // If one operation reads from invariant memory, and the other may store, they
+ // cannot alias. These should really be checking the equivalent of mayWrite,
+ // but it only matters for memory nodes other than load /store.
+ if (Op0->isInvariant() && Op1->writeMem())
+ return false;
+
+ if (Op1->isInvariant() && Op0->writeMem())
+ return false;
+
// Gather base node and offset information.
SDValue Base1, Base2;
int64_t Offset1, Offset2;
Op0->getSrcValueOffset() - MinOffset;
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()));
- if (AAResult == AliasAnalysis::NoAlias)
+ AliasResult AAResult =
+ AA.alias(MemoryLocation(Op0->getMemOperand()->getValue(), Overlap1,
+ UseTBAA ? Op0->getAAInfo() : AAMDNodes()),
+ MemoryLocation(Op1->getMemOperand()->getValue(), Overlap2,
+ UseTBAA ? Op1->getAAInfo() : AAMDNodes()));
+ if (AAResult == NoAlias)
return false;
}
// aliases list. If not, then continue up the chain looking for the next
// candidate.
while (!Chains.empty()) {
- SDValue Chain = Chains.back();
- Chains.pop_back();
+ SDValue Chain = Chains.pop_back_val();
// For TokenFactor nodes, look at each operand and only continue up the
- // chain until we find two aliases. If we've seen two aliases, assume we'll
- // find more and revert to original chain since the xform is unlikely to be
- // profitable.
+ // chain until we reach the depth limit.
//
// FIXME: The depth check could be made to return the last non-aliasing
// chain we found before we hit a tokenfactor rather than the original
// chain.
- if (Depth > 6 || Aliases.size() == 2) {
+ if (Depth > TLI.getGatherAllAliasesMaxDepth()) {
Aliases.clear();
Aliases.push_back(OriginalChain);
return;
UIE = M->use_end(); UI != UIE; ++UI)
if (UI.getUse().getValueType() == MVT::Other &&
Visited.insert(*UI).second) {
- if (isa<MemIntrinsicSDNode>(*UI) || isa<MemSDNode>(*UI)) {
+ if (isa<MemSDNode>(*UI)) {
// We've not visited this use, and we care about it (it could have an
// ordering dependency with the original node).
Aliases.clear();
return DAG.getNode(ISD::TokenFactor, SDLoc(N), MVT::Other, Aliases);
}
+bool DAGCombiner::findBetterNeighborChains(StoreSDNode* St) {
+ // 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;
+
+ // Do not handle stores to undef base pointers.
+ if (BasePtr.Base.getOpcode() == ISD::UNDEF)
+ return false;
+
+ SmallVector<StoreSDNode *, 8> ChainedStores;
+ ChainedStores.push_back(St);
+
+ // 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.
+ StoreSDNode *Index = St;
+ while (Index) {
+ // If the chain has more than one use, then we can't reorder the mem ops.
+ if (Index != St && !SDValue(Index, 0)->hasOneUse())
+ break;
+
+ if (Index->isVolatile() || Index->isIndexed())
+ break;
+
+ // Find the base pointer and offset for this memory node.
+ BaseIndexOffset Ptr = BaseIndexOffset::match(Index->getBasePtr());
+
+ // Check that the base pointer is the same as the original one.
+ if (!Ptr.equalBaseIndex(BasePtr))
+ break;
+
+ // Find the next memory operand in the chain. If the next operand in the
+ // chain is a store then move up and continue the scan with the next
+ // memory operand. If the next operand is a load save it and use alias
+ // information to check if it interferes with anything.
+ SDNode *NextInChain = Index->getChain().getNode();
+ while (true) {
+ if (StoreSDNode *STn = dyn_cast<StoreSDNode>(NextInChain)) {
+ // We found a store node. Use it for the next iteration.
+ ChainedStores.push_back(STn);
+ Index = STn;
+ break;
+ } else if (LoadSDNode *Ldn = dyn_cast<LoadSDNode>(NextInChain)) {
+ NextInChain = Ldn->getChain().getNode();
+ continue;
+ } else {
+ Index = nullptr;
+ break;
+ }
+ }
+ }
+
+ bool MadeChange = false;
+ SmallVector<std::pair<StoreSDNode *, SDValue>, 8> BetterChains;
+
+ for (StoreSDNode *ChainedStore : ChainedStores) {
+ SDValue Chain = ChainedStore->getChain();
+ SDValue BetterChain = FindBetterChain(ChainedStore, Chain);
+
+ if (Chain != BetterChain) {
+ MadeChange = true;
+ BetterChains.push_back(std::make_pair(ChainedStore, BetterChain));
+ }
+ }
+
+ // Do all replacements after finding the replacements to make to avoid making
+ // the chains more complicated by introducing new TokenFactors.
+ for (auto Replacement : BetterChains)
+ replaceStoreChain(Replacement.first, Replacement.second);
+
+ return MadeChange;
+}
+
/// This is the entry point for the file.
void SelectionDAG::Combine(CombineLevel Level, AliasAnalysis &AA,
CodeGenOpt::Level OptLevel) {