#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
-#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Target/TargetData.h"
-#include "llvm/Target/TargetFrameInfo.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetSelectionDAGInfo.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/ManagedStatic.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
-#include "llvm/System/Mutex.h"
+#include "llvm/Support/Mutex.h"
#include "llvm/ADT/SetVector.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallSet.h"
/// BUILD_VECTOR where all of the elements are ~0 or undef.
bool ISD::isBuildVectorAllOnes(const SDNode *N) {
// Look through a bit convert.
- if (N->getOpcode() == ISD::BIT_CONVERT)
+ if (N->getOpcode() == ISD::BITCAST)
N = N->getOperand(0).getNode();
if (N->getOpcode() != ISD::BUILD_VECTOR) return false;
/// BUILD_VECTOR where all of the elements are 0 or undef.
bool ISD::isBuildVectorAllZeros(const SDNode *N) {
// Look through a bit convert.
- if (N->getOpcode() == ISD::BIT_CONVERT)
+ if (N->getOpcode() == ISD::BITCAST)
N = N->getOperand(0).getNode();
if (N->getOpcode() != ISD::BUILD_VECTOR) return false;
ID.AddInteger(CP->getAlignment());
ID.AddInteger(CP->getOffset());
if (CP->isMachineConstantPoolEntry())
- CP->getMachineCPVal()->AddSelectionDAGCSEId(ID);
+ CP->getMachineCPVal()->addSelectionDAGCSEId(ID);
else
ID.AddPointer(CP->getConstVal());
ID.AddInteger(CP->getTargetFlags());
case ISD::ATOMIC_LOAD_MIN:
case ISD::ATOMIC_LOAD_MAX:
case ISD::ATOMIC_LOAD_UMIN:
- case ISD::ATOMIC_LOAD_UMAX: {
+ case ISD::ATOMIC_LOAD_UMAX:
+ case ISD::ATOMIC_LOAD:
+ case ISD::ATOMIC_STORE: {
const AtomicSDNode *AT = cast<AtomicSDNode>(N);
ID.AddInteger(AT->getMemoryVT().getRawBits());
ID.AddInteger(AT->getRawSubclassData());
///
static inline unsigned
encodeMemSDNodeFlags(int ConvType, ISD::MemIndexedMode AM, bool isVolatile,
- bool isNonTemporal) {
+ bool isNonTemporal, bool isInvariant) {
assert((ConvType & 3) == ConvType &&
"ConvType may not require more than 2 bits!");
assert((AM & 7) == AM &&
return ConvType |
(AM << 2) |
(isVolatile << 5) |
- (isNonTemporal << 6);
+ (isNonTemporal << 6) |
+ (isInvariant << 7);
}
//===----------------------------------------------------------------------===//
/// doNotCSE - Return true if CSE should not be performed for this node.
static bool doNotCSE(SDNode *N) {
- if (N->getValueType(0) == MVT::Flag)
+ if (N->getValueType(0) == MVT::Glue)
return true; // Never CSE anything that produces a flag.
switch (N->getOpcode()) {
// Check that remaining values produced are not flags.
for (unsigned i = 1, e = N->getNumValues(); i != e; ++i)
- if (N->getValueType(i) == MVT::Flag)
+ if (N->getValueType(i) == MVT::Glue)
return true; // Never CSE anything that produces a flag.
return false;
void SelectionDAG::RemoveDeadNode(SDNode *N, DAGUpdateListener *UpdateListener){
SmallVector<SDNode*, 16> DeadNodes(1, N);
+
+ // Create a dummy node that adds a reference to the root node, preventing
+ // it from being deleted. (This matters if the root is an operand of the
+ // dead node.)
+ HandleSDNode Dummy(getRoot());
+
RemoveDeadNodes(DeadNodes, UpdateListener);
}
Ordering->remove(N);
// If any of the SDDbgValue nodes refer to this SDNode, invalidate them.
- SmallVector<SDDbgValue*, 2> &DbgVals = DbgInfo->getSDDbgValues(N);
+ ArrayRef<SDDbgValue*> DbgVals = DbgInfo->getSDDbgValues(N);
for (unsigned i = 0, e = DbgVals.size(); i != e; ++i)
DbgVals[i]->setIsInvalidated();
}
bool SelectionDAG::RemoveNodeFromCSEMaps(SDNode *N) {
bool Erased = false;
switch (N->getOpcode()) {
- case ISD::EntryToken:
- llvm_unreachable("EntryToken should not be in CSEMaps!");
- return false;
case ISD::HANDLENODE: return false; // noop.
case ISD::CONDCODE:
assert(CondCodeNodes[cast<CondCodeSDNode>(N)->get()] &&
}
default:
// Remove it from the CSE Map.
+ assert(N->getOpcode() != ISD::DELETED_NODE && "DELETED_NODE in CSEMap!");
+ assert(N->getOpcode() != ISD::EntryToken && "EntryToken in CSEMap!");
Erased = CSEMap.RemoveNode(N);
break;
}
// Verify that the node was actually in one of the CSE maps, unless it has a
// flag result (which cannot be CSE'd) or is one of the special cases that are
// not subject to CSE.
- if (!Erased && N->getValueType(N->getNumValues()-1) != MVT::Flag &&
+ if (!Erased && N->getValueType(N->getNumValues()-1) != MVT::Glue &&
!N->isMachineOpcode() && !doNotCSE(N)) {
N->dump(this);
dbgs() << "\n";
return Node;
}
-/// VerifyNode - Sanity check the given node. Aborts if it is invalid.
-void SelectionDAG::VerifyNode(SDNode *N) {
+#ifndef NDEBUG
+/// VerifyNodeCommon - Sanity check the given node. Aborts if it is invalid.
+static void VerifyNodeCommon(SDNode *N) {
switch (N->getOpcode()) {
default:
break;
assert(N->getNumOperands() == N->getValueType(0).getVectorNumElements() &&
"Wrong number of operands!");
EVT EltVT = N->getValueType(0).getVectorElementType();
- for (SDNode::op_iterator I = N->op_begin(), E = N->op_end(); I != E; ++I)
+ for (SDNode::op_iterator I = N->op_begin(), E = N->op_end(); I != E; ++I) {
assert((I->getValueType() == EltVT ||
(EltVT.isInteger() && I->getValueType().isInteger() &&
EltVT.bitsLE(I->getValueType()))) &&
"Wrong operand type!");
+ assert(I->getValueType() == N->getOperand(0).getValueType() &&
+ "Operands must all have the same type");
+ }
break;
}
}
}
+/// VerifySDNode - Sanity check the given SDNode. Aborts if it is invalid.
+static void VerifySDNode(SDNode *N) {
+ // The SDNode allocators cannot be used to allocate nodes with fields that are
+ // not present in an SDNode!
+ assert(!isa<MemSDNode>(N) && "Bad MemSDNode!");
+ assert(!isa<ShuffleVectorSDNode>(N) && "Bad ShuffleVectorSDNode!");
+ assert(!isa<ConstantSDNode>(N) && "Bad ConstantSDNode!");
+ assert(!isa<ConstantFPSDNode>(N) && "Bad ConstantFPSDNode!");
+ assert(!isa<GlobalAddressSDNode>(N) && "Bad GlobalAddressSDNode!");
+ assert(!isa<FrameIndexSDNode>(N) && "Bad FrameIndexSDNode!");
+ assert(!isa<JumpTableSDNode>(N) && "Bad JumpTableSDNode!");
+ assert(!isa<ConstantPoolSDNode>(N) && "Bad ConstantPoolSDNode!");
+ assert(!isa<BasicBlockSDNode>(N) && "Bad BasicBlockSDNode!");
+ assert(!isa<SrcValueSDNode>(N) && "Bad SrcValueSDNode!");
+ assert(!isa<MDNodeSDNode>(N) && "Bad MDNodeSDNode!");
+ assert(!isa<RegisterSDNode>(N) && "Bad RegisterSDNode!");
+ assert(!isa<BlockAddressSDNode>(N) && "Bad BlockAddressSDNode!");
+ assert(!isa<EHLabelSDNode>(N) && "Bad EHLabelSDNode!");
+ assert(!isa<ExternalSymbolSDNode>(N) && "Bad ExternalSymbolSDNode!");
+ assert(!isa<CondCodeSDNode>(N) && "Bad CondCodeSDNode!");
+ assert(!isa<CvtRndSatSDNode>(N) && "Bad CvtRndSatSDNode!");
+ assert(!isa<VTSDNode>(N) && "Bad VTSDNode!");
+ assert(!isa<MachineSDNode>(N) && "Bad MachineSDNode!");
+
+ VerifyNodeCommon(N);
+}
+
+/// VerifyMachineNode - Sanity check the given MachineNode. Aborts if it is
+/// invalid.
+static void VerifyMachineNode(SDNode *N) {
+ // The MachineNode allocators cannot be used to allocate nodes with fields
+ // that are not present in a MachineNode!
+ // Currently there are no such nodes.
+
+ VerifyNodeCommon(N);
+}
+#endif // NDEBUG
+
/// getEVTAlignment - Compute the default alignment value for the
/// given type.
///
unsigned SelectionDAG::getEVTAlignment(EVT VT) const {
- const Type *Ty = VT == MVT::iPTR ?
+ Type *Ty = VT == MVT::iPTR ?
PointerType::get(Type::getInt8Ty(*getContext()), 0) :
VT.getTypeForEVT(*getContext());
DbgInfo->clear();
}
+SDValue SelectionDAG::getAnyExtOrTrunc(SDValue Op, DebugLoc DL, EVT VT) {
+ return VT.bitsGT(Op.getValueType()) ?
+ getNode(ISD::ANY_EXTEND, DL, VT, Op) :
+ getNode(ISD::TRUNCATE, DL, VT, Op);
+}
+
SDValue SelectionDAG::getSExtOrTrunc(SDValue Op, DebugLoc DL, EVT VT) {
return VT.bitsGT(Op.getValueType()) ?
getNode(ISD::SIGN_EXTEND, DL, VT, Op) :
assert(VT.isInteger() && "Cannot create FP integer constant!");
EVT EltVT = VT.getScalarType();
- assert(Val.getBitWidth() == EltVT.getSizeInBits() &&
- "APInt size does not match type size!");
+ const ConstantInt *Elt = &Val;
+ // In some cases the vector type is legal but the element type is illegal and
+ // needs to be promoted, for example v8i8 on ARM. In this case, promote the
+ // inserted value (the type does not need to match the vector element type).
+ // Any extra bits introduced will be truncated away.
+ if (VT.isVector() && TLI.getTypeAction(*getContext(), EltVT) ==
+ TargetLowering::TypePromoteInteger) {
+ EltVT = TLI.getTypeToTransformTo(*getContext(), EltVT);
+ APInt NewVal = Elt->getValue().zext(EltVT.getSizeInBits());
+ Elt = ConstantInt::get(*getContext(), NewVal);
+ }
+
+ assert(Elt->getBitWidth() == EltVT.getSizeInBits() &&
+ "APInt size does not match type size!");
unsigned Opc = isT ? ISD::TargetConstant : ISD::Constant;
FoldingSetNodeID ID;
AddNodeIDNode(ID, Opc, getVTList(EltVT), 0, 0);
- ID.AddPointer(&Val);
+ ID.AddPointer(Elt);
void *IP = 0;
SDNode *N = NULL;
if ((N = CSEMap.FindNodeOrInsertPos(ID, IP)))
return SDValue(N, 0);
if (!N) {
- N = new (NodeAllocator) ConstantSDNode(isT, &Val, EltVT);
+ N = new (NodeAllocator) ConstantSDNode(isT, Elt, EltVT);
CSEMap.InsertNode(N, IP);
AllNodes.push_back(N);
}
AddNodeIDNode(ID, Opc, getVTList(VT), 0, 0);
ID.AddInteger(Alignment);
ID.AddInteger(Offset);
- C->AddSelectionDAGCSEId(ID);
+ C->addSelectionDAGCSEId(ID);
ID.AddInteger(TargetFlags);
void *IP = 0;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
void *IP = 0;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
return SDValue(E, 0);
-
+
SDNode *N = new (NodeAllocator) EHLabelSDNode(dl, Root, Label);
CSEMap.InsertNode(N, IP);
AllNodes.push_back(N);
FoldingSetNodeID ID;
AddNodeIDNode(ID, ISD::MDNODE_SDNODE, getVTList(MVT::Other), 0, 0);
ID.AddPointer(MD);
-
+
void *IP = 0;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
return SDValue(E, 0);
-
+
SDNode *N = new (NodeAllocator) MDNodeSDNode(MD);
CSEMap.InsertNode(N, IP);
AllNodes.push_back(N);
/// getShiftAmountOperand - Return the specified value casted to
/// the target's desired shift amount type.
-SDValue SelectionDAG::getShiftAmountOperand(SDValue Op) {
+SDValue SelectionDAG::getShiftAmountOperand(EVT LHSTy, SDValue Op) {
EVT OpTy = Op.getValueType();
- MVT ShTy = TLI.getShiftAmountTy();
+ MVT ShTy = TLI.getShiftAmountTy(LHSTy);
if (OpTy == ShTy || OpTy.isVector()) return Op;
ISD::NodeType Opcode = OpTy.bitsGT(ShTy) ? ISD::TRUNCATE : ISD::ZERO_EXTEND;
SDValue SelectionDAG::CreateStackTemporary(EVT VT, unsigned minAlign) {
MachineFrameInfo *FrameInfo = getMachineFunction().getFrameInfo();
unsigned ByteSize = VT.getStoreSize();
- const Type *Ty = VT.getTypeForEVT(*getContext());
+ Type *Ty = VT.getTypeForEVT(*getContext());
unsigned StackAlign =
std::max((unsigned)TLI.getTargetData()->getPrefTypeAlignment(Ty), minAlign);
SDValue SelectionDAG::CreateStackTemporary(EVT VT1, EVT VT2) {
unsigned Bytes = std::max(VT1.getStoreSizeInBits(),
VT2.getStoreSizeInBits())/8;
- const Type *Ty1 = VT1.getTypeForEVT(*getContext());
- const Type *Ty2 = VT2.getTypeForEVT(*getContext());
+ Type *Ty1 = VT1.getTypeForEVT(*getContext());
+ Type *Ty2 = VT2.getTypeForEVT(*getContext());
const TargetData *TD = TLI.getTargetData();
unsigned Align = std::max(TD->getPrefTypeAlignment(Ty1),
TD->getPrefTypeAlignment(Ty2));
// Also compute a conserative estimate for high known-0 bits.
// More trickiness is possible, but this is sufficient for the
// interesting case of alignment computation.
- KnownOne.clear();
+ KnownOne.clearAllBits();
unsigned TrailZ = KnownZero.countTrailingOnes() +
KnownZero2.countTrailingOnes();
unsigned LeadZ = std::max(KnownZero.countLeadingOnes() +
AllOnes, KnownZero2, KnownOne2, Depth+1);
unsigned LeadZ = KnownZero2.countLeadingOnes();
- KnownOne2.clear();
- KnownZero2.clear();
+ KnownOne2.clearAllBits();
+ KnownZero2.clearAllBits();
ComputeMaskedBits(Op.getOperand(1),
AllOnes, KnownZero2, KnownOne2, Depth+1);
unsigned RHSUnknownLeadingOnes = KnownOne2.countLeadingZeros();
// The boolean result conforms to getBooleanContents. Fall through.
case ISD::SETCC:
// If we know the result of a setcc has the top bits zero, use this info.
- if (TLI.getBooleanContents() == TargetLowering::ZeroOrOneBooleanContent &&
- BitWidth > 1)
+ if (TLI.getBooleanContents(Op.getValueType().isVector()) ==
+ TargetLowering::ZeroOrOneBooleanContent && BitWidth > 1)
KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - 1);
return;
case ISD::SHL:
// If the sign extended bits are demanded, we know that the sign
// bit is demanded.
- InSignBit.zext(BitWidth);
+ InSignBit = InSignBit.zext(BitWidth);
if (NewBits.getBoolValue())
InputDemandedBits |= InSignBit;
case ISD::CTPOP: {
unsigned LowBits = Log2_32(BitWidth)+1;
KnownZero = APInt::getHighBitsSet(BitWidth, BitWidth - LowBits);
- KnownOne.clear();
+ KnownOne.clearAllBits();
return;
}
case ISD::LOAD: {
EVT InVT = Op.getOperand(0).getValueType();
unsigned InBits = InVT.getScalarType().getSizeInBits();
APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - InBits) & Mask;
- APInt InMask = Mask;
- InMask.trunc(InBits);
- KnownZero.trunc(InBits);
- KnownOne.trunc(InBits);
+ APInt InMask = Mask.trunc(InBits);
+ KnownZero = KnownZero.trunc(InBits);
+ KnownOne = KnownOne.trunc(InBits);
ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1);
- KnownZero.zext(BitWidth);
- KnownOne.zext(BitWidth);
+ KnownZero = KnownZero.zext(BitWidth);
+ KnownOne = KnownOne.zext(BitWidth);
KnownZero |= NewBits;
return;
}
unsigned InBits = InVT.getScalarType().getSizeInBits();
APInt InSignBit = APInt::getSignBit(InBits);
APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - InBits) & Mask;
- APInt InMask = Mask;
- InMask.trunc(InBits);
+ APInt InMask = Mask.trunc(InBits);
// If any of the sign extended bits are demanded, we know that the sign
// bit is demanded. Temporarily set this bit in the mask for our callee.
if (NewBits.getBoolValue())
InMask |= InSignBit;
- KnownZero.trunc(InBits);
- KnownOne.trunc(InBits);
+ KnownZero = KnownZero.trunc(InBits);
+ KnownOne = KnownOne.trunc(InBits);
ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1);
// Note if the sign bit is known to be zero or one.
// If the sign bit wasn't actually demanded by our caller, we don't
// want it set in the KnownZero and KnownOne result values. Reset the
// mask and reapply it to the result values.
- InMask = Mask;
- InMask.trunc(InBits);
+ InMask = Mask.trunc(InBits);
KnownZero &= InMask;
KnownOne &= InMask;
- KnownZero.zext(BitWidth);
- KnownOne.zext(BitWidth);
+ KnownZero = KnownZero.zext(BitWidth);
+ KnownOne = KnownOne.zext(BitWidth);
// If the sign bit is known zero or one, the top bits match.
if (SignBitKnownZero)
case ISD::ANY_EXTEND: {
EVT InVT = Op.getOperand(0).getValueType();
unsigned InBits = InVT.getScalarType().getSizeInBits();
- APInt InMask = Mask;
- InMask.trunc(InBits);
- KnownZero.trunc(InBits);
- KnownOne.trunc(InBits);
+ APInt InMask = Mask.trunc(InBits);
+ KnownZero = KnownZero.trunc(InBits);
+ KnownOne = KnownOne.trunc(InBits);
ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1);
- KnownZero.zext(BitWidth);
- KnownOne.zext(BitWidth);
+ KnownZero = KnownZero.zext(BitWidth);
+ KnownOne = KnownOne.zext(BitWidth);
return;
}
case ISD::TRUNCATE: {
EVT InVT = Op.getOperand(0).getValueType();
unsigned InBits = InVT.getScalarType().getSizeInBits();
- APInt InMask = Mask;
- InMask.zext(InBits);
- KnownZero.zext(InBits);
- KnownOne.zext(InBits);
+ APInt InMask = Mask.zext(InBits);
+ KnownZero = KnownZero.zext(InBits);
+ KnownOne = KnownOne.zext(InBits);
ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1);
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
- KnownZero.trunc(BitWidth);
- KnownOne.trunc(BitWidth);
+ KnownZero = KnownZero.trunc(BitWidth);
+ KnownOne = KnownOne.trunc(BitWidth);
break;
}
case ISD::AssertZext: {
}
}
// fall through
- case ISD::ADD: {
+ case ISD::ADD:
+ case ISD::ADDE: {
// Output known-0 bits are known if clear or set in both the low clear bits
// common to both LHS & RHS. For example, 8+(X<<3) is known to have the
// low 3 bits clear.
KnownZeroOut = std::min(KnownZeroOut,
KnownZero2.countTrailingOnes());
- KnownZero |= APInt::getLowBitsSet(BitWidth, KnownZeroOut);
+ if (Op.getOpcode() == ISD::ADD) {
+ KnownZero |= APInt::getLowBitsSet(BitWidth, KnownZeroOut);
+ return;
+ }
+
+ // With ADDE, a carry bit may be added in, so we can only use this
+ // information if we know (at least) that the low two bits are clear. We
+ // then return to the caller that the low bit is unknown but that other bits
+ // are known zero.
+ if (KnownZeroOut >= 2) // ADDE
+ KnownZero |= APInt::getBitsSet(BitWidth, 1, KnownZeroOut);
return;
}
case ISD::SREM:
uint32_t Leaders = std::max(KnownZero.countLeadingOnes(),
KnownZero2.countLeadingOnes());
- KnownOne.clear();
+ KnownOne.clearAllBits();
KnownZero = APInt::getHighBitsSet(BitWidth, Leaders) & Mask;
return;
}
+ case ISD::FrameIndex:
+ case ISD::TargetFrameIndex:
+ if (unsigned Align = InferPtrAlignment(Op)) {
+ // The low bits are known zero if the pointer is aligned.
+ KnownZero = APInt::getLowBitsSet(BitWidth, Log2_32(Align));
+ return;
+ }
+ break;
+
default:
- // Allow the target to implement this method for its nodes.
- if (Op.getOpcode() >= ISD::BUILTIN_OP_END) {
+ if (Op.getOpcode() < ISD::BUILTIN_OP_END)
+ break;
+ // Fallthrough
case ISD::INTRINSIC_WO_CHAIN:
case ISD::INTRINSIC_W_CHAIN:
case ISD::INTRINSIC_VOID:
- TLI.computeMaskedBitsForTargetNode(Op, Mask, KnownZero, KnownOne, *this,
- Depth);
- }
+ // Allow the target to implement this method for its nodes.
+ TLI.computeMaskedBitsForTargetNode(Op, Mask, KnownZero, KnownOne, *this,
+ Depth);
return;
}
}
case ISD::Constant: {
const APInt &Val = cast<ConstantSDNode>(Op)->getAPIntValue();
- // If negative, return # leading ones.
- if (Val.isNegative())
- return Val.countLeadingOnes();
-
- // Return # leading zeros.
- return Val.countLeadingZeros();
+ return Val.getNumSignBits();
}
case ISD::SIGN_EXTEND:
// The boolean result conforms to getBooleanContents. Fall through.
case ISD::SETCC:
// If setcc returns 0/-1, all bits are sign bits.
- if (TLI.getBooleanContents() ==
+ if (TLI.getBooleanContents(Op.getValueType().isVector()) ==
TargetLowering::ZeroOrNegativeOneBooleanContent)
return VTBits;
break;
return std::max(FirstAnswer, std::min(VTBits, Mask.countLeadingZeros()));
}
+/// isBaseWithConstantOffset - Return true if the specified operand is an
+/// ISD::ADD with a ConstantSDNode on the right-hand side, or if it is an
+/// ISD::OR with a ConstantSDNode that is guaranteed to have the same
+/// semantics as an ADD. This handles the equivalence:
+/// X|Cst == X+Cst iff X&Cst = 0.
+bool SelectionDAG::isBaseWithConstantOffset(SDValue Op) const {
+ if ((Op.getOpcode() != ISD::ADD && Op.getOpcode() != ISD::OR) ||
+ !isa<ConstantSDNode>(Op.getOperand(1)))
+ return false;
+
+ if (Op.getOpcode() == ISD::OR &&
+ !MaskedValueIsZero(Op.getOperand(0),
+ cast<ConstantSDNode>(Op.getOperand(1))->getAPIntValue()))
+ return false;
+
+ return true;
+}
+
+
bool SelectionDAG::isKnownNeverNaN(SDValue Op) const {
// If we're told that NaNs won't happen, assume they won't.
if (NoNaNsFPMath)
return !C->isZero();
// TODO: Recognize more cases here.
+ switch (Op.getOpcode()) {
+ default: break;
+ case ISD::OR:
+ if (const ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op.getOperand(1)))
+ return !C->isNullValue();
+ break;
+ }
return false;
}
return false;
}
-bool SelectionDAG::isVerifiedDebugInfoDesc(SDValue Op) const {
- GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(Op);
- if (!GA) return false;
- if (GA->getOffset() != 0) return false;
- const GlobalVariable *GV = dyn_cast<GlobalVariable>(GA->getGlobal());
- if (!GV) return false;
- return MF->getMMI().hasDebugInfo();
-}
-
-
/// getNode - Gets or creates the specified node.
///
SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, EVT VT) {
AllNodes.push_back(N);
#ifndef NDEBUG
- VerifyNode(N);
+ VerifySDNode(N);
#endif
return SDValue(N, 0);
}
switch (Opcode) {
default: break;
case ISD::SIGN_EXTEND:
- return getConstant(APInt(Val).sextOrTrunc(VT.getSizeInBits()), VT);
+ return getConstant(Val.sextOrTrunc(VT.getSizeInBits()), VT);
case ISD::ANY_EXTEND:
case ISD::ZERO_EXTEND:
case ISD::TRUNCATE:
- return getConstant(APInt(Val).zextOrTrunc(VT.getSizeInBits()), VT);
+ return getConstant(Val.zextOrTrunc(VT.getSizeInBits()), VT);
case ISD::UINT_TO_FP:
case ISD::SINT_TO_FP: {
- const uint64_t zero[] = {0, 0};
// No compile time operations on ppcf128.
if (VT == MVT::ppcf128) break;
- APFloat apf = APFloat(APInt(VT.getSizeInBits(), 2, zero));
+ APFloat apf(APInt::getNullValue(VT.getSizeInBits()));
(void)apf.convertFromAPInt(Val,
Opcode==ISD::SINT_TO_FP,
APFloat::rmNearestTiesToEven);
return getConstantFP(apf, VT);
}
- case ISD::BIT_CONVERT:
+ case ISD::BITCAST:
if (VT == MVT::f32 && C->getValueType(0) == MVT::i32)
return getConstantFP(Val.bitsToFloat(), VT);
else if (VT == MVT::f64 && C->getValueType(0) == MVT::i64)
APFloat::rmTowardZero, &ignored);
if (s==APFloat::opInvalidOp) // inexact is OK, in fact usual
break;
- APInt api(VT.getSizeInBits(), 2, x);
+ APInt api(VT.getSizeInBits(), x);
return getConstant(api, VT);
}
- case ISD::BIT_CONVERT:
+ case ISD::BITCAST:
if (VT == MVT::i32 && C->getValueType(0) == MVT::f32)
return getConstant((uint32_t)V.bitcastToAPInt().getZExtValue(), VT);
else if (VT == MVT::i64 && C->getValueType(0) == MVT::f64)
"Vector element count mismatch!");
if (OpOpcode == ISD::SIGN_EXTEND || OpOpcode == ISD::ZERO_EXTEND)
return getNode(OpOpcode, DL, VT, Operand.getNode()->getOperand(0));
+ else if (OpOpcode == ISD::UNDEF)
+ // sext(undef) = 0, because the top bits will all be the same.
+ return getConstant(0, VT);
break;
case ISD::ZERO_EXTEND:
assert(VT.isInteger() && Operand.getValueType().isInteger() &&
if (OpOpcode == ISD::ZERO_EXTEND) // (zext (zext x)) -> (zext x)
return getNode(ISD::ZERO_EXTEND, DL, VT,
Operand.getNode()->getOperand(0));
+ else if (OpOpcode == ISD::UNDEF)
+ // zext(undef) = 0, because the top bits will be zero.
+ return getConstant(0, VT);
break;
case ISD::ANY_EXTEND:
assert(VT.isInteger() && Operand.getValueType().isInteger() &&
OpOpcode == ISD::ANY_EXTEND)
// (ext (zext x)) -> (zext x) and (ext (sext x)) -> (sext x)
return getNode(OpOpcode, DL, VT, Operand.getNode()->getOperand(0));
+ else if (OpOpcode == ISD::UNDEF)
+ return getUNDEF(VT);
// (ext (trunx x)) -> x
if (OpOpcode == ISD::TRUNCATE) {
return Operand.getNode()->getOperand(0);
}
break;
- case ISD::BIT_CONVERT:
+ case ISD::BITCAST:
// Basic sanity checking.
assert(VT.getSizeInBits() == Operand.getValueType().getSizeInBits()
- && "Cannot BIT_CONVERT between types of different sizes!");
+ && "Cannot BITCAST between types of different sizes!");
if (VT == Operand.getValueType()) return Operand; // noop conversion.
- if (OpOpcode == ISD::BIT_CONVERT) // bitconv(bitconv(x)) -> bitconv(x)
- return getNode(ISD::BIT_CONVERT, DL, VT, Operand.getOperand(0));
+ if (OpOpcode == ISD::BITCAST) // bitconv(bitconv(x)) -> bitconv(x)
+ return getNode(ISD::BITCAST, DL, VT, Operand.getOperand(0));
if (OpOpcode == ISD::UNDEF)
return getUNDEF(VT);
break;
SDNode *N;
SDVTList VTs = getVTList(VT);
- if (VT != MVT::Flag) { // Don't CSE flag producing nodes
+ if (VT != MVT::Glue) { // Don't CSE flag producing nodes
FoldingSetNodeID ID;
SDValue Ops[1] = { Operand };
AddNodeIDNode(ID, Opcode, VTs, Ops, 1);
AllNodes.push_back(N);
#ifndef NDEBUG
- VerifyNode(N);
+ VerifySDNode(N);
#endif
return SDValue(N, 0);
}
"Shift operators return type must be the same as their first arg");
assert(VT.isInteger() && N2.getValueType().isInteger() &&
"Shifts only work on integers");
+ // Verify that the shift amount VT is bit enough to hold valid shift
+ // amounts. This catches things like trying to shift an i1024 value by an
+ // i8, which is easy to fall into in generic code that uses
+ // TLI.getShiftAmount().
+ assert(N2.getValueType().getSizeInBits() >=
+ Log2_32_Ceil(N1.getValueType().getSizeInBits()) &&
+ "Invalid use of small shift amount with oversized value!");
// Always fold shifts of i1 values so the code generator doesn't need to
// handle them. Since we know the size of the shift has to be less than the
EVT.getVectorNumElements() == VT.getVectorNumElements()) &&
"Vector element counts must match in FP_ROUND_INREG");
assert(EVT.bitsLE(VT) && "Not rounding down!");
+ (void)EVT;
if (cast<VTSDNode>(N2)->getVT() == VT) return N1; // Not actually rounding.
break;
}
assert(N2C && (unsigned)N2C->getZExtValue() < 2 && "Bad EXTRACT_ELEMENT!");
assert(!N1.getValueType().isVector() && !VT.isVector() &&
(N1.getValueType().isInteger() == VT.isInteger()) &&
+ N1.getValueType() != VT &&
"Wrong types for EXTRACT_ELEMENT!");
// EXTRACT_ELEMENT of BUILD_PAIR is often formed while legalize is expanding
return getConstant(ShiftedVal.trunc(ElementSize), VT);
}
break;
- case ISD::EXTRACT_SUBVECTOR:
- if (N1.getValueType() == VT) // Trivial extraction.
- return N1;
+ case ISD::EXTRACT_SUBVECTOR: {
+ SDValue Index = N2;
+ if (VT.isSimple() && N1.getValueType().isSimple()) {
+ assert(VT.isVector() && N1.getValueType().isVector() &&
+ "Extract subvector VTs must be a vectors!");
+ assert(VT.getVectorElementType() == N1.getValueType().getVectorElementType() &&
+ "Extract subvector VTs must have the same element type!");
+ assert(VT.getSimpleVT() <= N1.getValueType().getSimpleVT() &&
+ "Extract subvector must be from larger vector to smaller vector!");
+
+ if (isa<ConstantSDNode>(Index.getNode())) {
+ assert((VT.getVectorNumElements() +
+ cast<ConstantSDNode>(Index.getNode())->getZExtValue()
+ <= N1.getValueType().getVectorNumElements())
+ && "Extract subvector overflow!");
+ }
+
+ // Trivial extraction.
+ if (VT.getSimpleVT() == N1.getValueType().getSimpleVT())
+ return N1;
+ }
break;
}
+ }
if (N1C) {
if (N2C) {
// Memoize this node if possible.
SDNode *N;
SDVTList VTs = getVTList(VT);
- if (VT != MVT::Flag) {
+ if (VT != MVT::Glue) {
SDValue Ops[] = { N1, N2 };
FoldingSetNodeID ID;
AddNodeIDNode(ID, Opcode, VTs, Ops, 2);
AllNodes.push_back(N);
#ifndef NDEBUG
- VerifyNode(N);
+ VerifySDNode(N);
#endif
return SDValue(N, 0);
}
case ISD::VECTOR_SHUFFLE:
llvm_unreachable("should use getVectorShuffle constructor!");
break;
- case ISD::BIT_CONVERT:
+ case ISD::INSERT_SUBVECTOR: {
+ SDValue Index = N3;
+ if (VT.isSimple() && N1.getValueType().isSimple()
+ && N2.getValueType().isSimple()) {
+ assert(VT.isVector() && N1.getValueType().isVector() &&
+ N2.getValueType().isVector() &&
+ "Insert subvector VTs must be a vectors");
+ assert(VT == N1.getValueType() &&
+ "Dest and insert subvector source types must match!");
+ assert(N2.getValueType().getSimpleVT() <= N1.getValueType().getSimpleVT() &&
+ "Insert subvector must be from smaller vector to larger vector!");
+ if (isa<ConstantSDNode>(Index.getNode())) {
+ assert((N2.getValueType().getVectorNumElements() +
+ cast<ConstantSDNode>(Index.getNode())->getZExtValue()
+ <= VT.getVectorNumElements())
+ && "Insert subvector overflow!");
+ }
+
+ // Trivial insertion.
+ if (VT.getSimpleVT() == N2.getValueType().getSimpleVT())
+ return N2;
+ }
+ break;
+ }
+ case ISD::BITCAST:
// Fold bit_convert nodes from a type to themselves.
if (N1.getValueType() == VT)
return N1;
// Memoize node if it doesn't produce a flag.
SDNode *N;
SDVTList VTs = getVTList(VT);
- if (VT != MVT::Flag) {
+ if (VT != MVT::Glue) {
SDValue Ops[] = { N1, N2, N3 };
FoldingSetNodeID ID;
AddNodeIDNode(ID, Opcode, VTs, Ops, 3);
AllNodes.push_back(N);
#ifndef NDEBUG
- VerifyNode(N);
+ VerifySDNode(N);
#endif
return SDValue(N, 0);
}
&ArgChains[0], ArgChains.size());
}
+/// SplatByte - Distribute ByteVal over NumBits bits.
+static APInt SplatByte(unsigned NumBits, uint8_t ByteVal) {
+ APInt Val = APInt(NumBits, ByteVal);
+ unsigned Shift = 8;
+ for (unsigned i = NumBits; i > 8; i >>= 1) {
+ Val = (Val << Shift) | Val;
+ Shift <<= 1;
+ }
+ return Val;
+}
+
/// getMemsetValue - Vectorized representation of the memset value
/// operand.
static SDValue getMemsetValue(SDValue Value, EVT VT, SelectionDAG &DAG,
unsigned NumBits = VT.getScalarType().getSizeInBits();
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Value)) {
- APInt Val = APInt(NumBits, C->getZExtValue() & 255);
- unsigned Shift = 8;
- for (unsigned i = NumBits; i > 8; i >>= 1) {
- Val = (Val << Shift) | Val;
- Shift <<= 1;
- }
+ APInt Val = SplatByte(NumBits, C->getZExtValue() & 255);
if (VT.isInteger())
return DAG.getConstant(Val, VT);
return DAG.getConstantFP(APFloat(Val), VT);
}
- const TargetLowering &TLI = DAG.getTargetLoweringInfo();
Value = DAG.getNode(ISD::ZERO_EXTEND, dl, VT, Value);
- unsigned Shift = 8;
- for (unsigned i = NumBits; i > 8; i >>= 1) {
- Value = DAG.getNode(ISD::OR, dl, VT,
- DAG.getNode(ISD::SHL, dl, VT, Value,
- DAG.getConstant(Shift,
- TLI.getShiftAmountTy())),
- Value);
- Shift <<= 1;
+ if (NumBits > 8) {
+ // Use a multiplication with 0x010101... to extend the input to the
+ // required length.
+ APInt Magic = SplatByte(NumBits, 0x01);
+ Value = DAG.getNode(ISD::MUL, dl, VT, Value, DAG.getConstant(Magic, VT));
}
return Value;
else if (VT.isVector()) {
unsigned NumElts = VT.getVectorNumElements();
MVT EltVT = (VT.getVectorElementType() == MVT::f32) ? MVT::i32 : MVT::i64;
- return DAG.getNode(ISD::BIT_CONVERT, dl, VT,
+ return DAG.getNode(ISD::BITCAST, dl, VT,
DAG.getConstant(0, EVT::getVectorVT(*DAG.getContext(),
EltVT, NumElts)));
} else
static bool FindOptimalMemOpLowering(std::vector<EVT> &MemOps,
unsigned Limit, uint64_t Size,
unsigned DstAlign, unsigned SrcAlign,
- bool NonScalarIntSafe,
+ bool IsZeroVal,
bool MemcpyStrSrc,
SelectionDAG &DAG,
const TargetLowering &TLI) {
assert((SrcAlign == 0 || SrcAlign >= DstAlign) &&
"Expecting memcpy / memset source to meet alignment requirement!");
- // If 'SrcAlign' is zero, that means the memory operation does not need load
- // the value, i.e. memset or memcpy from constant string. Otherwise, it's
- // the inferred alignment of the source. 'DstAlign', on the other hand, is the
- // specified alignment of the memory operation. If it is zero, that means
- // it's possible to change the alignment of the destination. 'MemcpyStrSrc'
- // indicates whether the memcpy source is constant so it does not need to be
- // loaded.
+ // If 'SrcAlign' is zero, that means the memory operation does not need to
+ // load the value, i.e. memset or memcpy from constant string. Otherwise,
+ // it's the inferred alignment of the source. 'DstAlign', on the other hand,
+ // is the specified alignment of the memory operation. If it is zero, that
+ // means it's possible to change the alignment of the destination.
+ // 'MemcpyStrSrc' indicates whether the memcpy source is constant so it does
+ // not need to be loaded.
EVT VT = TLI.getOptimalMemOpType(Size, DstAlign, SrcAlign,
- NonScalarIntSafe, MemcpyStrSrc,
+ IsZeroVal, MemcpyStrSrc,
DAG.getMachineFunction());
if (VT == MVT::Other) {
if (VT.bitsGT(LVT))
VT = LVT;
}
-
- // If we're optimizing for size, and there is a limit, bump the maximum number
- // of operations inserted down to 4. This is a wild guess that approximates
- // the size of a call to memcpy or memset (3 arguments + call).
- if (Limit != ~0U) {
- const Function *F = DAG.getMachineFunction().getFunction();
- if (F->hasFnAttr(Attribute::OptimizeForSize))
- Limit = 4;
- }
unsigned NumMemOps = 0;
while (Size != 0) {
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
std::vector<EVT> MemOps;
bool DstAlignCanChange = false;
- MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
+ MachineFunction &MF = DAG.getMachineFunction();
+ MachineFrameInfo *MFI = MF.getFrameInfo();
+ bool OptSize = MF.getFunction()->hasFnAttr(Attribute::OptimizeForSize);
FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(Dst);
if (FI && !MFI->isFixedObjectIndex(FI->getIndex()))
DstAlignCanChange = true;
std::string Str;
bool CopyFromStr = isMemSrcFromString(Src, Str);
bool isZeroStr = CopyFromStr && Str.empty();
- unsigned Limit = AlwaysInline ? ~0U : TLI.getMaxStoresPerMemcpy();
-
+ unsigned Limit = AlwaysInline ? ~0U : TLI.getMaxStoresPerMemcpy(OptSize);
+
if (!FindOptimalMemOpLowering(MemOps, Limit, Size,
(DstAlignCanChange ? 0 : Align),
(isZeroStr ? 0 : SrcAlign),
return SDValue();
if (DstAlignCanChange) {
- const Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext());
+ Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext());
unsigned NewAlign = (unsigned) TLI.getTargetData()->getABITypeAlignment(Ty);
if (NewAlign > Align) {
// Give the stack frame object a larger alignment if needed.
// FIXME does the case above also need this?
EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), VT);
assert(NVT.bitsGE(VT));
- Value = DAG.getExtLoad(ISD::EXTLOAD, NVT, dl, Chain,
+ Value = DAG.getExtLoad(ISD::EXTLOAD, dl, NVT, Chain,
getMemBasePlusOffset(Src, SrcOff, DAG),
SrcPtrInfo.getWithOffset(SrcOff), VT, isVol, false,
MinAlign(SrcAlign, SrcOff));
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
std::vector<EVT> MemOps;
bool DstAlignCanChange = false;
- MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
+ MachineFunction &MF = DAG.getMachineFunction();
+ MachineFrameInfo *MFI = MF.getFrameInfo();
+ bool OptSize = MF.getFunction()->hasFnAttr(Attribute::OptimizeForSize);
FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(Dst);
if (FI && !MFI->isFixedObjectIndex(FI->getIndex()))
DstAlignCanChange = true;
unsigned SrcAlign = DAG.InferPtrAlignment(Src);
if (Align > SrcAlign)
SrcAlign = Align;
- unsigned Limit = AlwaysInline ? ~0U : TLI.getMaxStoresPerMemmove();
+ unsigned Limit = AlwaysInline ? ~0U : TLI.getMaxStoresPerMemmove(OptSize);
if (!FindOptimalMemOpLowering(MemOps, Limit, Size,
(DstAlignCanChange ? 0 : Align),
return SDValue();
if (DstAlignCanChange) {
- const Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext());
+ Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext());
unsigned NewAlign = (unsigned) TLI.getTargetData()->getABITypeAlignment(Ty);
if (NewAlign > Align) {
// Give the stack frame object a larger alignment if needed.
Value = DAG.getLoad(VT, dl, Chain,
getMemBasePlusOffset(Src, SrcOff, DAG),
SrcPtrInfo.getWithOffset(SrcOff), isVol,
- false, SrcAlign);
+ false, false, SrcAlign);
LoadValues.push_back(Value);
LoadChains.push_back(Value.getValue(1));
SrcOff += VTSize;
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
std::vector<EVT> MemOps;
bool DstAlignCanChange = false;
- MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
+ MachineFunction &MF = DAG.getMachineFunction();
+ MachineFrameInfo *MFI = MF.getFrameInfo();
+ bool OptSize = MF.getFunction()->hasFnAttr(Attribute::OptimizeForSize);
FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(Dst);
if (FI && !MFI->isFixedObjectIndex(FI->getIndex()))
DstAlignCanChange = true;
- bool NonScalarIntSafe =
+ bool IsZeroVal =
isa<ConstantSDNode>(Src) && cast<ConstantSDNode>(Src)->isNullValue();
- if (!FindOptimalMemOpLowering(MemOps, TLI.getMaxStoresPerMemset(),
+ if (!FindOptimalMemOpLowering(MemOps, TLI.getMaxStoresPerMemset(OptSize),
Size, (DstAlignCanChange ? 0 : Align), 0,
- NonScalarIntSafe, false, DAG, TLI))
+ IsZeroVal, false, DAG, TLI))
return SDValue();
if (DstAlignCanChange) {
- const Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext());
+ Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext());
unsigned NewAlign = (unsigned) TLI.getTargetData()->getABITypeAlignment(Ty);
if (NewAlign > Align) {
// Give the stack frame object a larger alignment if needed.
SmallVector<SDValue, 8> OutChains;
uint64_t DstOff = 0;
unsigned NumMemOps = MemOps.size();
+
+ // Find the largest store and generate the bit pattern for it.
+ EVT LargestVT = MemOps[0];
+ for (unsigned i = 1; i < NumMemOps; i++)
+ if (MemOps[i].bitsGT(LargestVT))
+ LargestVT = MemOps[i];
+ SDValue MemSetValue = getMemsetValue(Src, LargestVT, DAG, dl);
+
for (unsigned i = 0; i < NumMemOps; i++) {
EVT VT = MemOps[i];
- unsigned VTSize = VT.getSizeInBits() / 8;
- SDValue Value = getMemsetValue(Src, VT, DAG, dl);
+
+ // If this store is smaller than the largest store see whether we can get
+ // the smaller value for free with a truncate.
+ SDValue Value = MemSetValue;
+ if (VT.bitsLT(LargestVT)) {
+ if (!LargestVT.isVector() && !VT.isVector() &&
+ TLI.isTruncateFree(LargestVT, VT))
+ Value = DAG.getNode(ISD::TRUNCATE, dl, VT, MemSetValue);
+ else
+ Value = getMemsetValue(Src, VT, DAG, dl);
+ }
+ assert(Value.getValueType() == VT && "Value with wrong type.");
SDValue Store = DAG.getStore(Chain, dl, Value,
getMemBasePlusOffset(Dst, DstOff, DAG),
DstPtrInfo.getWithOffset(DstOff),
isVol, false, Align);
OutChains.push_back(Store);
- DstOff += VTSize;
+ DstOff += VT.getSizeInBits() / 8;
}
return DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
if (Result.getNode())
return Result;
- // Emit a library call.
- const Type *IntPtrTy = TLI.getTargetData()->getIntPtrType(*getContext());
+ // Emit a library call.
+ Type *IntPtrTy = TLI.getTargetData()->getIntPtrType(*getContext());
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Node = Dst; Entry.Ty = IntPtrTy;
SDValue SelectionDAG::getAtomic(unsigned Opcode, DebugLoc dl, EVT MemVT,
SDValue Chain, SDValue Ptr, SDValue Cmp,
SDValue Swp, MachinePointerInfo PtrInfo,
- unsigned Alignment) {
+ unsigned Alignment,
+ AtomicOrdering Ordering,
+ SynchronizationScope SynchScope) {
if (Alignment == 0) // Ensure that codegen never sees alignment 0
Alignment = getEVTAlignment(MemVT);
unsigned Flags = MachineMemOperand::MOLoad | MachineMemOperand::MOStore;
// For now, atomics are considered to be volatile always.
+ // FIXME: Volatile isn't really correct; we should keep track of atomic
+ // orderings in the memoperand.
Flags |= MachineMemOperand::MOVolatile;
MachineMemOperand *MMO =
MF.getMachineMemOperand(PtrInfo, Flags, MemVT.getStoreSize(), Alignment);
- return getAtomic(Opcode, dl, MemVT, Chain, Ptr, Cmp, Swp, MMO);
+ return getAtomic(Opcode, dl, MemVT, Chain, Ptr, Cmp, Swp, MMO,
+ Ordering, SynchScope);
}
SDValue SelectionDAG::getAtomic(unsigned Opcode, DebugLoc dl, EVT MemVT,
SDValue Chain,
SDValue Ptr, SDValue Cmp,
- SDValue Swp, MachineMemOperand *MMO) {
+ SDValue Swp, MachineMemOperand *MMO,
+ AtomicOrdering Ordering,
+ SynchronizationScope SynchScope) {
assert(Opcode == ISD::ATOMIC_CMP_SWAP && "Invalid Atomic Op");
assert(Cmp.getValueType() == Swp.getValueType() && "Invalid Atomic Op Types");
return SDValue(E, 0);
}
SDNode *N = new (NodeAllocator) AtomicSDNode(Opcode, dl, VTs, MemVT, Chain,
- Ptr, Cmp, Swp, MMO);
+ Ptr, Cmp, Swp, MMO, Ordering,
+ SynchScope);
CSEMap.InsertNode(N, IP);
AllNodes.push_back(N);
return SDValue(N, 0);
SDValue Chain,
SDValue Ptr, SDValue Val,
const Value* PtrVal,
- unsigned Alignment) {
+ unsigned Alignment,
+ AtomicOrdering Ordering,
+ SynchronizationScope SynchScope) {
if (Alignment == 0) // Ensure that codegen never sees alignment 0
Alignment = getEVTAlignment(MemVT);
MachineFunction &MF = getMachineFunction();
- unsigned Flags = MachineMemOperand::MOLoad | MachineMemOperand::MOStore;
+ // A monotonic store does not load; a release store "loads" in the sense
+ // that other stores cannot be sunk past it.
+ // (An atomicrmw obviously both loads and stores.)
+ unsigned Flags = MachineMemOperand::MOStore;
+ if (Opcode != ISD::ATOMIC_STORE || Ordering > Monotonic)
+ Flags |= MachineMemOperand::MOLoad;
// For now, atomics are considered to be volatile always.
+ // FIXME: Volatile isn't really correct; we should keep track of atomic
+ // orderings in the memoperand.
Flags |= MachineMemOperand::MOVolatile;
MachineMemOperand *MMO =
MF.getMachineMemOperand(MachinePointerInfo(PtrVal), Flags,
MemVT.getStoreSize(), Alignment);
- return getAtomic(Opcode, dl, MemVT, Chain, Ptr, Val, MMO);
+ return getAtomic(Opcode, dl, MemVT, Chain, Ptr, Val, MMO,
+ Ordering, SynchScope);
}
SDValue SelectionDAG::getAtomic(unsigned Opcode, DebugLoc dl, EVT MemVT,
SDValue Chain,
SDValue Ptr, SDValue Val,
- MachineMemOperand *MMO) {
+ MachineMemOperand *MMO,
+ AtomicOrdering Ordering,
+ SynchronizationScope SynchScope) {
assert((Opcode == ISD::ATOMIC_LOAD_ADD ||
Opcode == ISD::ATOMIC_LOAD_SUB ||
Opcode == ISD::ATOMIC_LOAD_AND ||
Opcode == ISD::ATOMIC_LOAD_MAX ||
Opcode == ISD::ATOMIC_LOAD_UMIN ||
Opcode == ISD::ATOMIC_LOAD_UMAX ||
- Opcode == ISD::ATOMIC_SWAP) &&
+ Opcode == ISD::ATOMIC_SWAP ||
+ Opcode == ISD::ATOMIC_STORE) &&
"Invalid Atomic Op");
EVT VT = Val.getValueType();
- SDVTList VTs = getVTList(VT, MVT::Other);
+ SDVTList VTs = Opcode == ISD::ATOMIC_STORE ? getVTList(MVT::Other) :
+ getVTList(VT, MVT::Other);
FoldingSetNodeID ID;
ID.AddInteger(MemVT.getRawBits());
SDValue Ops[] = {Chain, Ptr, Val};
return SDValue(E, 0);
}
SDNode *N = new (NodeAllocator) AtomicSDNode(Opcode, dl, VTs, MemVT, Chain,
- Ptr, Val, MMO);
+ Ptr, Val, MMO,
+ Ordering, SynchScope);
+ CSEMap.InsertNode(N, IP);
+ AllNodes.push_back(N);
+ return SDValue(N, 0);
+}
+
+SDValue SelectionDAG::getAtomic(unsigned Opcode, DebugLoc dl, EVT MemVT,
+ EVT VT, SDValue Chain,
+ SDValue Ptr,
+ const Value* PtrVal,
+ unsigned Alignment,
+ AtomicOrdering Ordering,
+ SynchronizationScope SynchScope) {
+ if (Alignment == 0) // Ensure that codegen never sees alignment 0
+ Alignment = getEVTAlignment(MemVT);
+
+ MachineFunction &MF = getMachineFunction();
+ // A monotonic load does not store; an acquire load "stores" in the sense
+ // that other loads cannot be hoisted past it.
+ unsigned Flags = MachineMemOperand::MOLoad;
+ if (Ordering > Monotonic)
+ Flags |= MachineMemOperand::MOStore;
+
+ // For now, atomics are considered to be volatile always.
+ // FIXME: Volatile isn't really correct; we should keep track of atomic
+ // orderings in the memoperand.
+ Flags |= MachineMemOperand::MOVolatile;
+
+ MachineMemOperand *MMO =
+ MF.getMachineMemOperand(MachinePointerInfo(PtrVal), Flags,
+ MemVT.getStoreSize(), Alignment);
+
+ return getAtomic(Opcode, dl, MemVT, VT, Chain, Ptr, MMO,
+ Ordering, SynchScope);
+}
+
+SDValue SelectionDAG::getAtomic(unsigned Opcode, DebugLoc dl, EVT MemVT,
+ EVT VT, SDValue Chain,
+ SDValue Ptr,
+ MachineMemOperand *MMO,
+ AtomicOrdering Ordering,
+ SynchronizationScope SynchScope) {
+ assert(Opcode == ISD::ATOMIC_LOAD && "Invalid Atomic Op");
+
+ SDVTList VTs = getVTList(VT, MVT::Other);
+ FoldingSetNodeID ID;
+ ID.AddInteger(MemVT.getRawBits());
+ SDValue Ops[] = {Chain, Ptr};
+ AddNodeIDNode(ID, Opcode, VTs, Ops, 2);
+ void* IP = 0;
+ if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) {
+ cast<AtomicSDNode>(E)->refineAlignment(MMO);
+ return SDValue(E, 0);
+ }
+ SDNode *N = new (NodeAllocator) AtomicSDNode(Opcode, dl, VTs, MemVT, Chain,
+ Ptr, MMO, Ordering, SynchScope);
CSEMap.InsertNode(N, IP);
AllNodes.push_back(N);
return SDValue(N, 0);
}
/// getMergeValues - Create a MERGE_VALUES node from the given operands.
-/// Allowed to return something different (and simpler) if Simplify is true.
SDValue SelectionDAG::getMergeValues(const SDValue *Ops, unsigned NumOps,
DebugLoc dl) {
if (NumOps == 1)
// Memoize the node unless it returns a flag.
MemIntrinsicSDNode *N;
- if (VTList.VTs[VTList.NumVTs-1] != MVT::Flag) {
+ if (VTList.VTs[VTList.NumVTs-1] != MVT::Glue) {
FoldingSetNodeID ID;
AddNodeIDNode(ID, Opcode, VTList, Ops, NumOps);
void *IP = 0;
!isa<ConstantSDNode>(Ptr.getOperand(1)) ||
!isa<FrameIndexSDNode>(Ptr.getOperand(0)))
return MachinePointerInfo();
-
+
int FI = cast<FrameIndexSDNode>(Ptr.getOperand(0))->getIndex();
return MachinePointerInfo::getFixedStack(FI, Offset+
cast<ConstantSDNode>(Ptr.getOperand(1))->getSExtValue());
return InferPointerInfo(Ptr);
return MachinePointerInfo();
}
-
+
SDValue
SelectionDAG::getLoad(ISD::MemIndexedMode AM, ISD::LoadExtType ExtType,
EVT VT, DebugLoc dl, SDValue Chain,
SDValue Ptr, SDValue Offset,
MachinePointerInfo PtrInfo, EVT MemVT,
- bool isVolatile, bool isNonTemporal,
+ bool isVolatile, bool isNonTemporal, bool isInvariant,
unsigned Alignment, const MDNode *TBAAInfo) {
+ assert(Chain.getValueType() == MVT::Other &&
+ "Invalid chain type");
if (Alignment == 0) // Ensure that codegen never sees alignment 0
Alignment = getEVTAlignment(VT);
Flags |= MachineMemOperand::MOVolatile;
if (isNonTemporal)
Flags |= MachineMemOperand::MONonTemporal;
-
+ if (isInvariant)
+ Flags |= MachineMemOperand::MOInvariant;
+
// If we don't have a PtrInfo, infer the trivial frame index case to simplify
// clients.
if (PtrInfo.V == 0)
PtrInfo = InferPointerInfo(Ptr, Offset);
-
+
MachineFunction &MF = getMachineFunction();
MachineMemOperand *MMO =
MF.getMachineMemOperand(PtrInfo, Flags, MemVT.getStoreSize(), Alignment,
}
SDValue
-SelectionDAG::getLoad(ISD::MemIndexedMode AM, ISD::LoadExtType ExtType,
+SelectionDAG::getLoad(ISD::MemIndexedMode AM, ISD::LoadExtType ExtType,
EVT VT, DebugLoc dl, SDValue Chain,
SDValue Ptr, SDValue Offset, EVT MemVT,
MachineMemOperand *MMO) {
AddNodeIDNode(ID, ISD::LOAD, VTs, Ops, 3);
ID.AddInteger(MemVT.getRawBits());
ID.AddInteger(encodeMemSDNodeFlags(ExtType, AM, MMO->isVolatile(),
- MMO->isNonTemporal()));
+ MMO->isNonTemporal(),
+ MMO->isInvariant()));
void *IP = 0;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) {
cast<LoadSDNode>(E)->refineAlignment(MMO);
SDValue Chain, SDValue Ptr,
MachinePointerInfo PtrInfo,
bool isVolatile, bool isNonTemporal,
- unsigned Alignment, const MDNode *TBAAInfo) {
+ bool isInvariant, unsigned Alignment,
+ const MDNode *TBAAInfo) {
SDValue Undef = getUNDEF(Ptr.getValueType());
return getLoad(ISD::UNINDEXED, ISD::NON_EXTLOAD, VT, dl, Chain, Ptr, Undef,
- PtrInfo, VT, isVolatile, isNonTemporal, Alignment, TBAAInfo);
+ PtrInfo, VT, isVolatile, isNonTemporal, isInvariant, Alignment,
+ TBAAInfo);
}
-SDValue SelectionDAG::getExtLoad(ISD::LoadExtType ExtType, EVT VT, DebugLoc dl,
+SDValue SelectionDAG::getExtLoad(ISD::LoadExtType ExtType, DebugLoc dl, EVT VT,
SDValue Chain, SDValue Ptr,
MachinePointerInfo PtrInfo, EVT MemVT,
bool isVolatile, bool isNonTemporal,
unsigned Alignment, const MDNode *TBAAInfo) {
SDValue Undef = getUNDEF(Ptr.getValueType());
return getLoad(ISD::UNINDEXED, ExtType, VT, dl, Chain, Ptr, Undef,
- PtrInfo, MemVT, isVolatile, isNonTemporal, Alignment,
+ PtrInfo, MemVT, isVolatile, isNonTemporal, false, Alignment,
TBAAInfo);
}
"Load is already a indexed load!");
return getLoad(AM, LD->getExtensionType(), OrigLoad.getValueType(), dl,
LD->getChain(), Base, Offset, LD->getPointerInfo(),
- LD->getMemoryVT(),
- LD->isVolatile(), LD->isNonTemporal(), LD->getAlignment());
+ LD->getMemoryVT(), LD->isVolatile(), LD->isNonTemporal(),
+ false, LD->getAlignment());
}
SDValue SelectionDAG::getStore(SDValue Chain, DebugLoc dl, SDValue Val,
SDValue Ptr, MachinePointerInfo PtrInfo,
bool isVolatile, bool isNonTemporal,
unsigned Alignment, const MDNode *TBAAInfo) {
+ assert(Chain.getValueType() == MVT::Other &&
+ "Invalid chain type");
if (Alignment == 0) // Ensure that codegen never sees alignment 0
Alignment = getEVTAlignment(Val.getValueType());
Flags |= MachineMemOperand::MOVolatile;
if (isNonTemporal)
Flags |= MachineMemOperand::MONonTemporal;
-
+
if (PtrInfo.V == 0)
PtrInfo = InferPointerInfo(Ptr);
SDValue SelectionDAG::getStore(SDValue Chain, DebugLoc dl, SDValue Val,
SDValue Ptr, MachineMemOperand *MMO) {
+ assert(Chain.getValueType() == MVT::Other &&
+ "Invalid chain type");
EVT VT = Val.getValueType();
SDVTList VTs = getVTList(MVT::Other);
SDValue Undef = getUNDEF(Ptr.getValueType());
AddNodeIDNode(ID, ISD::STORE, VTs, Ops, 4);
ID.AddInteger(VT.getRawBits());
ID.AddInteger(encodeMemSDNodeFlags(false, ISD::UNINDEXED, MMO->isVolatile(),
- MMO->isNonTemporal()));
+ MMO->isNonTemporal(), MMO->isInvariant()));
void *IP = 0;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) {
cast<StoreSDNode>(E)->refineAlignment(MMO);
EVT SVT,bool isVolatile, bool isNonTemporal,
unsigned Alignment,
const MDNode *TBAAInfo) {
+ assert(Chain.getValueType() == MVT::Other &&
+ "Invalid chain type");
if (Alignment == 0) // Ensure that codegen never sees alignment 0
Alignment = getEVTAlignment(SVT);
Flags |= MachineMemOperand::MOVolatile;
if (isNonTemporal)
Flags |= MachineMemOperand::MONonTemporal;
-
+
if (PtrInfo.V == 0)
PtrInfo = InferPointerInfo(Ptr);
MachineMemOperand *MMO) {
EVT VT = Val.getValueType();
+ assert(Chain.getValueType() == MVT::Other &&
+ "Invalid chain type");
if (VT == SVT)
return getStore(Chain, dl, Val, Ptr, MMO);
AddNodeIDNode(ID, ISD::STORE, VTs, Ops, 4);
ID.AddInteger(SVT.getRawBits());
ID.AddInteger(encodeMemSDNodeFlags(true, ISD::UNINDEXED, MMO->isVolatile(),
- MMO->isNonTemporal()));
+ MMO->isNonTemporal(), MMO->isInvariant()));
void *IP = 0;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) {
cast<StoreSDNode>(E)->refineAlignment(MMO);
SDNode *N;
SDVTList VTs = getVTList(VT);
- if (VT != MVT::Flag) {
+ if (VT != MVT::Glue) {
FoldingSetNodeID ID;
AddNodeIDNode(ID, Opcode, VTs, Ops, NumOps);
void *IP = 0;
AllNodes.push_back(N);
#ifndef NDEBUG
- VerifyNode(N);
+ VerifySDNode(N);
#endif
return SDValue(N, 0);
}
// Memoize the node unless it returns a flag.
SDNode *N;
- if (VTList.VTs[VTList.NumVTs-1] != MVT::Flag) {
+ if (VTList.VTs[VTList.NumVTs-1] != MVT::Glue) {
FoldingSetNodeID ID;
AddNodeIDNode(ID, Opcode, VTList, Ops, NumOps);
void *IP = 0;
}
AllNodes.push_back(N);
#ifndef NDEBUG
- VerifyNode(N);
+ VerifySDNode(N);
#endif
return SDValue(N, 0);
}
unsigned NumOps) {
// If an identical node already exists, use it.
void *IP = 0;
- if (VTs.VTs[VTs.NumVTs-1] != MVT::Flag) {
+ if (VTs.VTs[VTs.NumVTs-1] != MVT::Glue) {
FoldingSetNodeID ID;
AddNodeIDNode(ID, Opc, VTs, Ops, NumOps);
if (SDNode *ON = CSEMap.FindNodeOrInsertPos(ID, IP))
MachineSDNode *
SelectionDAG::getMachineNode(unsigned Opcode, DebugLoc DL, SDVTList VTs,
const SDValue *Ops, unsigned NumOps) {
- bool DoCSE = VTs.VTs[VTs.NumVTs-1] != MVT::Flag;
+ bool DoCSE = VTs.VTs[VTs.NumVTs-1] != MVT::Glue;
MachineSDNode *N;
- void *IP;
+ void *IP = 0;
if (DoCSE) {
FoldingSetNodeID ID;
AllNodes.push_back(N);
#ifndef NDEBUG
- VerifyNode(N);
+ VerifyMachineNode(N);
#endif
return N;
}
/// else return NULL.
SDNode *SelectionDAG::getNodeIfExists(unsigned Opcode, SDVTList VTList,
const SDValue *Ops, unsigned NumOps) {
- if (VTList.VTs[VTList.NumVTs-1] != MVT::Flag) {
+ if (VTList.VTs[VTList.NumVTs-1] != MVT::Glue) {
FoldingSetNodeID ID;
AddNodeIDNode(ID, Opcode, VTList, Ops, NumOps);
void *IP = 0;
// already exists there, recursively merge the results together.
AddModifiedNodeToCSEMaps(User, &Listener);
}
+
+ // If we just RAUW'd the root, take note.
+ if (FromN == getRoot())
+ setRoot(To);
}
/// ReplaceAllUsesWith - Modify anything using 'From' to use 'To' instead.
// already exists there, recursively merge the results together.
AddModifiedNodeToCSEMaps(User, &Listener);
}
+
+ // If we just RAUW'd the root, take note.
+ if (From == getRoot().getNode())
+ setRoot(SDValue(To, getRoot().getResNo()));
}
/// ReplaceAllUsesWith - Modify anything using 'From' to use 'To' instead.
// already exists there, recursively merge the results together.
AddModifiedNodeToCSEMaps(User, &Listener);
}
+
+ // If we just RAUW'd the root, take note.
+ if (From == getRoot().getNode())
+ setRoot(SDValue(To[getRoot().getResNo()]));
}
/// ReplaceAllUsesOfValueWith - Replace any uses of From with To, leaving
// already exists there, recursively merge the results together.
AddModifiedNodeToCSEMaps(User, &Listener);
}
+
+ // If we just RAUW'd the root, take note.
+ if (From == getRoot())
+ setRoot(To);
}
namespace {
SD->setHasDebugValue(true);
}
+/// TransferDbgValues - Transfer SDDbgValues.
+void SelectionDAG::TransferDbgValues(SDValue From, SDValue To) {
+ if (From == To || !From.getNode()->getHasDebugValue())
+ return;
+ SDNode *FromNode = From.getNode();
+ SDNode *ToNode = To.getNode();
+ ArrayRef<SDDbgValue *> DVs = GetDbgValues(FromNode);
+ SmallVector<SDDbgValue *, 2> ClonedDVs;
+ for (ArrayRef<SDDbgValue *>::iterator I = DVs.begin(), E = DVs.end();
+ I != E; ++I) {
+ SDDbgValue *Dbg = *I;
+ if (Dbg->getKind() == SDDbgValue::SDNODE) {
+ SDDbgValue *Clone = getDbgValue(Dbg->getMDPtr(), ToNode, To.getResNo(),
+ Dbg->getOffset(), Dbg->getDebugLoc(),
+ Dbg->getOrder());
+ ClonedDVs.push_back(Clone);
+ }
+ }
+ for (SmallVector<SDDbgValue *, 2>::iterator I = ClonedDVs.begin(),
+ E = ClonedDVs.end(); I != E; ++I)
+ AddDbgValue(*I, ToNode, false);
+}
+
//===----------------------------------------------------------------------===//
// SDNode Class
//===----------------------------------------------------------------------===//
MachineMemOperand *mmo)
: SDNode(Opc, dl, VTs), MemoryVT(memvt), MMO(mmo) {
SubclassData = encodeMemSDNodeFlags(0, ISD::UNINDEXED, MMO->isVolatile(),
- MMO->isNonTemporal());
+ MMO->isNonTemporal(), MMO->isInvariant());
assert(isVolatile() == MMO->isVolatile() && "Volatile encoding error!");
assert(isNonTemporal() == MMO->isNonTemporal() &&
"Non-temporal encoding error!");
}
MemSDNode::MemSDNode(unsigned Opc, DebugLoc dl, SDVTList VTs,
- const SDValue *Ops, unsigned NumOps, EVT memvt,
+ const SDValue *Ops, unsigned NumOps, EVT memvt,
MachineMemOperand *mmo)
: SDNode(Opc, dl, VTs, Ops, NumOps),
MemoryVT(memvt), MMO(mmo) {
SubclassData = encodeMemSDNodeFlags(0, ISD::UNINDEXED, MMO->isVolatile(),
- MMO->isNonTemporal());
+ MMO->isNonTemporal(), MMO->isInvariant());
assert(isVolatile() == MMO->isVolatile() && "Volatile encoding error!");
assert(memvt.getStoreSize() == MMO->getSize() && "Size mismatch!");
}
namespace {
struct EVTArray {
std::vector<EVT> VTs;
-
+
EVTArray() {
VTs.reserve(MVT::LAST_VALUETYPE);
for (unsigned i = 0; i < MVT::LAST_VALUETYPE; ++i)
/// reachesChainWithoutSideEffects - Return true if this operand (which must
/// be a chain) reaches the specified operand without crossing any
-/// side-effecting instructions on any chain path. In practice, this looks
-/// through token factors and non-volatile loads. In order to remain efficient,
+/// side-effecting instructions on any chain path. In practice, this looks
+/// through token factors and non-volatile loads. In order to remain efficient,
/// this only looks a couple of nodes in, it does not do an exhaustive search.
bool SDValue::reachesChainWithoutSideEffects(SDValue Dest,
unsigned Depth) const {
return false;
}
-/// isPredecessorOf - Return true if this node is a predecessor of N. This node
-/// is either an operand of N or it can be reached by traversing up the operands.
-/// NOTE: this is an expensive method. Use it carefully.
-bool SDNode::isPredecessorOf(SDNode *N) const {
- SmallPtrSet<SDNode *, 32> Visited;
- SmallVector<SDNode *, 16> Worklist;
- Worklist.push_back(N);
+/// hasPredecessor - Return true if N is a predecessor of this node.
+/// N is either an operand of this node, or can be reached by recursively
+/// traversing up the operands.
+/// NOTE: This is an expensive method. Use it carefully.
+bool SDNode::hasPredecessor(const SDNode *N) const {
+ SmallPtrSet<const SDNode *, 32> Visited;
+ SmallVector<const SDNode *, 16> Worklist;
+ return hasPredecessorHelper(N, Visited, Worklist);
+}
- do {
- N = Worklist.pop_back_val();
- for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
- SDNode *Op = N->getOperand(i).getNode();
- if (Op == this)
- return true;
+bool SDNode::hasPredecessorHelper(const SDNode *N,
+ SmallPtrSet<const SDNode *, 32> &Visited,
+ SmallVector<const SDNode *, 16> &Worklist) const {
+ if (Visited.empty()) {
+ Worklist.push_back(this);
+ } else {
+ // Take a look in the visited set. If we've already encountered this node
+ // we needn't search further.
+ if (Visited.count(N))
+ return true;
+ }
+
+ // Haven't visited N yet. Continue the search.
+ while (!Worklist.empty()) {
+ const SDNode *M = Worklist.pop_back_val();
+ for (unsigned i = 0, e = M->getNumOperands(); i != e; ++i) {
+ SDNode *Op = M->getOperand(i).getNode();
if (Visited.insert(Op))
Worklist.push_back(Op);
+ if (Op == N)
+ return true;
}
- } while (!Worklist.empty());
+ }
return false;
}
#endif
case ISD::PREFETCH: return "Prefetch";
case ISD::MEMBARRIER: return "MemBarrier";
+ case ISD::ATOMIC_FENCE: return "AtomicFence";
case ISD::ATOMIC_CMP_SWAP: return "AtomicCmpSwap";
case ISD::ATOMIC_SWAP: return "AtomicSwap";
case ISD::ATOMIC_LOAD_ADD: return "AtomicLoadAdd";
case ISD::ATOMIC_LOAD_MAX: return "AtomicLoadMax";
case ISD::ATOMIC_LOAD_UMIN: return "AtomicLoadUMin";
case ISD::ATOMIC_LOAD_UMAX: return "AtomicLoadUMax";
+ case ISD::ATOMIC_LOAD: return "AtomicLoad";
+ case ISD::ATOMIC_STORE: return "AtomicStore";
case ISD::PCMARKER: return "PCMarker";
case ISD::READCYCLECOUNTER: return "ReadCycleCounter";
case ISD::SRCVALUE: return "SrcValue";
case ISD::EH_RETURN: return "EH_RETURN";
case ISD::EH_SJLJ_SETJMP: return "EH_SJLJ_SETJMP";
case ISD::EH_SJLJ_LONGJMP: return "EH_SJLJ_LONGJMP";
- case ISD::EH_SJLJ_DISPATCHSETUP: return "EH_SJLJ_DISPATCHSETUP";
case ISD::ConstantPool: return "ConstantPool";
case ISD::ExternalSymbol: return "ExternalSymbol";
case ISD::BlockAddress: return "BlockAddress";
case ISD::FSUB: return "fsub";
case ISD::FMUL: return "fmul";
case ISD::FDIV: return "fdiv";
+ case ISD::FMA: return "fma";
case ISD::FREM: return "frem";
case ISD::FCOPYSIGN: return "fcopysign";
case ISD::FGETSIGN: return "fgetsign";
case ISD::FPOWI: return "fpowi";
case ISD::SETCC: return "setcc";
- case ISD::VSETCC: return "vsetcc";
case ISD::SELECT: return "select";
+ case ISD::VSELECT: return "vselect";
case ISD::SELECT_CC: return "select_cc";
case ISD::INSERT_VECTOR_ELT: return "insert_vector_elt";
case ISD::EXTRACT_VECTOR_ELT: return "extract_vector_elt";
case ISD::CONCAT_VECTORS: return "concat_vectors";
+ case ISD::INSERT_SUBVECTOR: return "insert_subvector";
case ISD::EXTRACT_SUBVECTOR: return "extract_subvector";
case ISD::SCALAR_TO_VECTOR: return "scalar_to_vector";
case ISD::VECTOR_SHUFFLE: return "vector_shuffle";
case ISD::UINT_TO_FP: return "uint_to_fp";
case ISD::FP_TO_SINT: return "fp_to_sint";
case ISD::FP_TO_UINT: return "fp_to_uint";
- case ISD::BIT_CONVERT: return "bit_convert";
+ case ISD::BITCAST: return "bitcast";
case ISD::FP16_TO_FP32: return "fp16_to_fp32";
case ISD::FP32_TO_FP16: return "fp32_to_fp16";
case ISD::CTLZ: return "ctlz";
// Trampolines
- case ISD::TRAMPOLINE: return "trampoline";
+ case ISD::INIT_TRAMPOLINE: return "init_trampoline";
+ case ISD::ADJUST_TRAMPOLINE: return "adjust_trampoline";
case ISD::CONDCODE:
switch (cast<CondCodeSDNode>(this)->get()) {
OS << LBB->getName() << " ";
OS << (const void*)BBDN->getBasicBlock() << ">";
} else if (const RegisterSDNode *R = dyn_cast<RegisterSDNode>(this)) {
- if (G && R->getReg() &&
- TargetRegisterInfo::isPhysicalRegister(R->getReg())) {
- OS << " %" << G->getTarget().getRegisterInfo()->getName(R->getReg());
- } else {
- OS << " %reg" << R->getReg();
- }
+ OS << ' ' << PrintReg(R->getReg(), G ? G->getTarget().getRegisterInfo() :0);
} else if (const ExternalSymbolSDNode *ES =
dyn_cast<ExternalSymbolSDNode>(this)) {
OS << "'" << ES->getSymbol() << "'";
const char *AM = getIndexedModeName(ST->getAddressingMode());
if (*AM)
OS << ", " << AM;
-
+
OS << ">";
} else if (const MemSDNode* M = dyn_cast<MemSDNode>(this)) {
OS << "<" << *M->getMemOperand() << ">";
static void printrWithDepthHelper(raw_ostream &OS, const SDNode *N,
const SelectionDAG *G, unsigned depth,
- unsigned indent)
-{
+ unsigned indent) {
if (depth == 0)
return;
return;
for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
+ // Don't follow chain operands.
+ if (N->getOperand(i).getValueType() == MVT::Other)
+ continue;
OS << '\n';
printrWithDepthHelper(OS, N->getOperand(i).getNode(), G, depth-1, indent+2);
}
void SDNode::printrWithDepth(raw_ostream &OS, const SelectionDAG *G,
unsigned depth) const {
printrWithDepthHelper(OS, this, G, depth, 0);
-}
+}
void SDNode::printrFull(raw_ostream &OS, const SelectionDAG *G) const {
// Don't print impossibly deep things.
- printrWithDepth(OS, G, 100);
+ printrWithDepth(OS, G, 10);
}
void SDNode::dumprWithDepth(const SelectionDAG *G, unsigned depth) const {
void SDNode::dumprFull(const SelectionDAG *G) const {
// Don't print impossibly deep things.
- dumprWithDepth(G, 100);
-}
+ dumprWithDepth(G, 10);
+}
static void DumpNodes(const SDNode *N, unsigned indent, const SelectionDAG *G) {
for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
Operands[j] = getNode(ISD::EXTRACT_VECTOR_ELT, dl,
OperandEltVT,
Operand,
- getConstant(i, MVT::i32));
+ getConstant(i, TLI.getPointerTy()));
} else {
// A scalar operand; just use it as is.
Operands[j] = Operand;
Scalars.push_back(getNode(N->getOpcode(), dl, EltVT,
&Operands[0], Operands.size()));
break;
+ case ISD::VSELECT:
+ Scalars.push_back(getNode(ISD::SELECT, dl, EltVT,
+ &Operands[0], Operands.size()));
+ break;
case ISD::SHL:
case ISD::SRA:
case ISD::SRL:
case ISD::ROTL:
case ISD::ROTR:
Scalars.push_back(getNode(N->getOpcode(), dl, EltVT, Operands[0],
- getShiftAmountOperand(Operands[1])));
+ getShiftAmountOperand(Operands[0].getValueType(),
+ Operands[1])));
break;
case ISD::SIGN_EXTEND_INREG:
case ISD::FP_ROUND_INREG: {
}
-/// isConsecutiveLoad - Return true if LD is loading 'Bytes' bytes from a
-/// location that is 'Dist' units away from the location that the 'Base' load
+/// isConsecutiveLoad - Return true if LD is loading 'Bytes' bytes from a
+/// location that is 'Dist' units away from the location that the 'Base' load
/// is loading from.
-bool SelectionDAG::isConsecutiveLoad(LoadSDNode *LD, LoadSDNode *Base,
+bool SelectionDAG::isConsecutiveLoad(LoadSDNode *LD, LoadSDNode *Base,
unsigned Bytes, int Dist) const {
if (LD->getChain() != Base->getChain())
return false;
if (FS != BFS || FS != (int)Bytes) return false;
return MFI->getObjectOffset(FI) == (MFI->getObjectOffset(BFI) + Dist*Bytes);
}
- if (Loc.getOpcode() == ISD::ADD && Loc.getOperand(0) == BaseLoc) {
- ConstantSDNode *V = dyn_cast<ConstantSDNode>(Loc.getOperand(1));
- if (V && (V->getSExtValue() == Dist*Bytes))
- return true;
- }
+
+ // Handle X+C
+ if (isBaseWithConstantOffset(Loc) && Loc.getOperand(0) == BaseLoc &&
+ cast<ConstantSDNode>(Loc.getOperand(1))->getSExtValue() == Dist*Bytes)
+ return true;
const GlobalValue *GV1 = NULL;
const GlobalValue *GV2 = NULL;
const GlobalValue *GV;
int64_t GVOffset = 0;
if (TLI.isGAPlusOffset(Ptr.getNode(), GV, GVOffset)) {
- // If GV has specified alignment, then use it. Otherwise, use the preferred
- // alignment.
- unsigned Align = GV->getAlignment();
- if (!Align) {
- if (const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GV)) {
- if (GVar->hasInitializer()) {
- const TargetData *TD = TLI.getTargetData();
- Align = TD->getPreferredAlignment(GVar);
- }
- }
- }
- return MinAlign(Align, GVOffset);
+ unsigned PtrWidth = TLI.getPointerTy().getSizeInBits();
+ APInt AllOnes = APInt::getAllOnesValue(PtrWidth);
+ APInt KnownZero(PtrWidth, 0), KnownOne(PtrWidth, 0);
+ llvm::ComputeMaskedBits(const_cast<GlobalValue*>(GV), AllOnes,
+ KnownZero, KnownOne, TLI.getTargetData());
+ unsigned AlignBits = KnownZero.countTrailingOnes();
+ unsigned Align = AlignBits ? 1 << std::min(31U, AlignBits) : 0;
+ if (Align)
+ return MinAlign(Align, GVOffset);
}
// If this is a direct reference to a stack slot, use information about the
int64_t FrameOffset = 0;
if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(Ptr)) {
FrameIdx = FI->getIndex();
- } else if (Ptr.getOpcode() == ISD::ADD &&
- isa<ConstantSDNode>(Ptr.getOperand(1)) &&
+ } else if (isBaseWithConstantOffset(Ptr) &&
isa<FrameIndexSDNode>(Ptr.getOperand(0))) {
+ // Handle FI+Cst
FrameIdx = cast<FrameIndexSDNode>(Ptr.getOperand(0))->getIndex();
FrameOffset = Ptr.getConstantOperandVal(1);
}
if (FrameIdx != (1 << 31)) {
- // FIXME: Handle FI+CST.
const MachineFrameInfo &MFI = *getMachineFunction().getFrameInfo();
unsigned FIInfoAlign = MinAlign(MFI.getObjectAlignment(FrameIdx),
FrameOffset);
return;
// Dump the current SDNode, but don't end the line yet.
- OS << std::string(indent, ' ');
+ OS.indent(indent);
N->printr(OS, G);
// Having printed this SDNode, walk the children:
}
-const Type *ConstantPoolSDNode::getType() const {
+Type *ConstantPoolSDNode::getType() const {
if (isMachineConstantPoolEntry())
return Val.MachineCPVal->getType();
return Val.ConstVal->getType();
if (OpVal.getOpcode() == ISD::UNDEF)
SplatUndef |= APInt::getBitsSet(sz, BitPos, BitPos + EltBitSize);
else if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(OpVal))
- SplatValue |= APInt(CN->getAPIntValue()).zextOrTrunc(EltBitSize).
+ SplatValue |= CN->getAPIntValue().zextOrTrunc(EltBitSize).
zextOrTrunc(sz) << BitPos;
else if (ConstantFPSDNode *CN = dyn_cast<ConstantFPSDNode>(OpVal))
SplatValue |= CN->getValueAPF().bitcastToAPInt().zextOrTrunc(sz) <<BitPos;
while (sz > 8) {
unsigned HalfSize = sz / 2;
- APInt HighValue = APInt(SplatValue).lshr(HalfSize).trunc(HalfSize);
- APInt LowValue = APInt(SplatValue).trunc(HalfSize);
- APInt HighUndef = APInt(SplatUndef).lshr(HalfSize).trunc(HalfSize);
- APInt LowUndef = APInt(SplatUndef).trunc(HalfSize);
+ APInt HighValue = SplatValue.lshr(HalfSize).trunc(HalfSize);
+ APInt LowValue = SplatValue.trunc(HalfSize);
+ APInt HighUndef = SplatUndef.lshr(HalfSize).trunc(HalfSize);
+ APInt LowUndef = SplatUndef.trunc(HalfSize);
// If the two halves do not match (ignoring undef bits), stop here.
if ((HighValue & ~LowUndef) != (LowValue & ~HighUndef) ||
// If this node has already been checked, don't check it again.
if (Checked.count(N))
return;
-
+
// If a node has already been visited on this depth-first walk, reject it as
// a cycle.
if (!Visited.insert(N)) {
errs() << "Detected cycle in SelectionDAG\n";
abort();
}
-
+
for(unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
checkForCyclesHelper(N->getOperand(i).getNode(), Visited, Checked);
-
+
Checked.insert(N);
Visited.erase(N);
}