//===----------------------------------------------------------------------===//
#include "llvm/CodeGen/SelectionDAG.h"
-#include "SDNodeOrdering.h"
#include "SDNodeDbgValue.h"
-#include "llvm/Constants.h"
-#include "llvm/Analysis/DebugInfo.h"
+#include "SDNodeOrdering.h"
+#include "llvm/ADT/SetVector.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/SmallSet.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/StringExtras.h"
#include "llvm/Analysis/ValueTracking.h"
-#include "llvm/Function.h"
-#include "llvm/GlobalAlias.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/DerivedTypes.h"
#include "llvm/Assembly/Writer.h"
-#include "llvm/CallingConv.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
-#include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/Target/TargetData.h"
-#include "llvm/Target/TargetLowering.h"
-#include "llvm/Target/TargetSelectionDAGInfo.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/Target/TargetIntrinsicInfo.h"
-#include "llvm/Target/TargetMachine.h"
+#include "llvm/DebugInfo.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalAlias.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Intrinsics.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/ManagedStatic.h"
#include "llvm/Support/MathExtras.h"
-#include "llvm/Support/raw_ostream.h"
#include "llvm/Support/Mutex.h"
-#include "llvm/ADT/SetVector.h"
-#include "llvm/ADT/SmallPtrSet.h"
-#include "llvm/ADT/SmallSet.h"
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/StringExtras.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetIntrinsicInfo.h"
+#include "llvm/Target/TargetLowering.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetOptions.h"
+#include "llvm/Target/TargetRegisterInfo.h"
+#include "llvm/Target/TargetSelectionDAGInfo.h"
+#include "llvm/TargetTransformInfo.h"
#include <algorithm>
#include <cmath>
using namespace llvm;
static const fltSemantics *EVTToAPFloatSemantics(EVT VT) {
switch (VT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("Unknown FP format");
+ case MVT::f16: return &APFloat::IEEEhalf;
case MVT::f32: return &APFloat::IEEEsingle;
case MVT::f64: return &APFloat::IEEEdouble;
case MVT::f80: return &APFloat::x87DoubleExtended;
}
}
-SelectionDAG::DAGUpdateListener::anchor() {}
+// Default null implementations of the callbacks.
+void SelectionDAG::DAGUpdateListener::NodeDeleted(SDNode*, SDNode*) {}
+void SelectionDAG::DAGUpdateListener::NodeUpdated(SDNode*) {}
//===----------------------------------------------------------------------===//
// ConstantFPSDNode Class
const APFloat& Val) {
assert(VT.isFloatingPoint() && "Can only convert between FP types");
- // PPC long double cannot be converted to any other type.
- if (VT == MVT::ppcf128 ||
- &Val.getSemantics() == &APFloat::PPCDoubleDouble)
- return false;
-
// convert modifies in place, so make a copy.
APFloat Val2 = APFloat(Val);
bool losesInfo;
if (i == e) return false;
// Do not accept build_vectors that aren't all constants or which have non-~0
- // elements.
+ // elements. We have to be a bit careful here, as the type of the constant
+ // may not be the same as the type of the vector elements due to type
+ // legalization (the elements are promoted to a legal type for the target and
+ // a vector of a type may be legal when the base element type is not).
+ // We only want to check enough bits to cover the vector elements, because
+ // we care if the resultant vector is all ones, not whether the individual
+ // constants are.
SDValue NotZero = N->getOperand(i);
- if (isa<ConstantSDNode>(NotZero)) {
- if (!cast<ConstantSDNode>(NotZero)->isAllOnesValue())
+ unsigned EltSize = N->getValueType(0).getVectorElementType().getSizeInBits();
+ if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(NotZero)) {
+ if (CN->getAPIntValue().countTrailingOnes() < EltSize)
return false;
- } else if (isa<ConstantFPSDNode>(NotZero)) {
- if (!cast<ConstantFPSDNode>(NotZero)->getValueAPF().
- bitcastToAPInt().isAllOnesValue())
+ } else if (ConstantFPSDNode *CFPN = dyn_cast<ConstantFPSDNode>(NotZero)) {
+ if (CFPN->getValueAPF().bitcastToAPInt().countTrailingOnes() < EltSize)
return false;
} else
return false;
// Okay, we have at least one ~0 value, check to see if the rest match or are
- // undefs.
+ // undefs. Even with the above element type twiddling, this should be OK, as
+ // the same type legalization should have applied to all the elements.
for (++i; i != e; ++i)
if (N->getOperand(i) != NotZero &&
N->getOperand(i).getOpcode() != ISD::UNDEF)
// Do not accept build_vectors that aren't all constants or which have non-0
// elements.
SDValue Zero = N->getOperand(i);
- if (isa<ConstantSDNode>(Zero)) {
- if (!cast<ConstantSDNode>(Zero)->isNullValue())
+ if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Zero)) {
+ if (!CN->isNullValue())
return false;
- } else if (isa<ConstantFPSDNode>(Zero)) {
- if (!cast<ConstantFPSDNode>(Zero)->getValueAPF().isPosZero())
+ } else if (ConstantFPSDNode *CFPN = dyn_cast<ConstantFPSDNode>(Zero)) {
+ if (!CFPN->getValueAPF().isPosZero())
return false;
} else
return false;
return true;
}
+/// allOperandsUndef - Return true if the node has at least one operand
+/// and all operands of the specified node are ISD::UNDEF.
+bool ISD::allOperandsUndef(const SDNode *N) {
+ // Return false if the node has no operands.
+ // This is "logically inconsistent" with the definition of "all" but
+ // is probably the desired behavior.
+ if (N->getNumOperands() == 0)
+ return false;
+
+ for (unsigned i = 0, e = N->getNumOperands(); i != e ; ++i)
+ if (N->getOperand(i).getOpcode() != ISD::UNDEF)
+ return false;
+
+ return true;
+}
+
/// getSetCCSwappedOperands - Return the operation corresponding to (Y op X)
/// when given the operation for (X op Y).
ISD::CondCode ISD::getSetCCSwappedOperands(ISD::CondCode Operation) {
ID.AddPointer(GA->getGlobal());
ID.AddInteger(GA->getOffset());
ID.AddInteger(GA->getTargetFlags());
+ ID.AddInteger(GA->getAddressSpace());
break;
}
case ISD::BasicBlock:
ID.AddInteger(CP->getTargetFlags());
break;
}
+ case ISD::TargetIndex: {
+ const TargetIndexSDNode *TI = cast<TargetIndexSDNode>(N);
+ ID.AddInteger(TI->getIndex());
+ ID.AddInteger(TI->getOffset());
+ ID.AddInteger(TI->getTargetFlags());
+ break;
+ }
case ISD::LOAD: {
const LoadSDNode *LD = cast<LoadSDNode>(N);
ID.AddInteger(LD->getMemoryVT().getRawBits());
ID.AddInteger(LD->getRawSubclassData());
+ ID.AddInteger(LD->getPointerInfo().getAddrSpace());
break;
}
case ISD::STORE: {
const StoreSDNode *ST = cast<StoreSDNode>(N);
ID.AddInteger(ST->getMemoryVT().getRawBits());
ID.AddInteger(ST->getRawSubclassData());
+ ID.AddInteger(ST->getPointerInfo().getAddrSpace());
break;
}
case ISD::ATOMIC_CMP_SWAP:
const AtomicSDNode *AT = cast<AtomicSDNode>(N);
ID.AddInteger(AT->getMemoryVT().getRawBits());
ID.AddInteger(AT->getRawSubclassData());
+ ID.AddInteger(AT->getPointerInfo().getAddrSpace());
+ break;
+ }
+ case ISD::PREFETCH: {
+ const MemSDNode *PF = cast<MemSDNode>(N);
+ ID.AddInteger(PF->getPointerInfo().getAddrSpace());
break;
}
case ISD::VECTOR_SHUFFLE: {
}
case ISD::TargetBlockAddress:
case ISD::BlockAddress: {
- ID.AddPointer(cast<BlockAddressSDNode>(N)->getBlockAddress());
- ID.AddInteger(cast<BlockAddressSDNode>(N)->getTargetFlags());
+ const BlockAddressSDNode *BA = cast<BlockAddressSDNode>(N);
+ ID.AddPointer(BA->getBlockAddress());
+ ID.AddInteger(BA->getOffset());
+ ID.AddInteger(BA->getTargetFlags());
break;
}
} // end switch (N->getOpcode())
+
+ // Target specific memory nodes could also have address spaces to check.
+ if (N->isTargetMemoryOpcode())
+ ID.AddInteger(cast<MemSDNode>(N)->getPointerInfo().getAddrSpace());
}
/// AddNodeIDNode - Generic routine for adding a nodes info to the NodeID
/// RemoveDeadNodes - This method deletes the unreachable nodes in the
/// given list, and any nodes that become unreachable as a result.
-void SelectionDAG::RemoveDeadNodes(SmallVectorImpl<SDNode *> &DeadNodes,
- DAGUpdateListener *UpdateListener) {
+void SelectionDAG::RemoveDeadNodes(SmallVectorImpl<SDNode *> &DeadNodes) {
// Process the worklist, deleting the nodes and adding their uses to the
// worklist.
while (!DeadNodes.empty()) {
SDNode *N = DeadNodes.pop_back_val();
- if (UpdateListener)
- UpdateListener->NodeDeleted(N, 0);
+ for (DAGUpdateListener *DUL = UpdateListeners; DUL; DUL = DUL->Next)
+ DUL->NodeDeleted(N, 0);
// Take the node out of the appropriate CSE map.
RemoveNodeFromCSEMaps(N);
}
}
-void SelectionDAG::RemoveDeadNode(SDNode *N, DAGUpdateListener *UpdateListener){
+void SelectionDAG::RemoveDeadNode(SDNode *N){
SmallVector<SDNode*, 16> DeadNodes(1, N);
// Create a dummy node that adds a reference to the root node, preventing
// dead node.)
HandleSDNode Dummy(getRoot());
- RemoveDeadNodes(DeadNodes, UpdateListener);
+ RemoveDeadNodes(DeadNodes);
}
void SelectionDAG::DeleteNode(SDNode *N) {
/// node. This transfer can potentially trigger recursive merging.
///
void
-SelectionDAG::AddModifiedNodeToCSEMaps(SDNode *N,
- DAGUpdateListener *UpdateListener) {
+SelectionDAG::AddModifiedNodeToCSEMaps(SDNode *N) {
// For node types that aren't CSE'd, just act as if no identical node
// already exists.
if (!doNotCSE(N)) {
// If there was already an existing matching node, use ReplaceAllUsesWith
// to replace the dead one with the existing one. This can cause
// recursive merging of other unrelated nodes down the line.
- ReplaceAllUsesWith(N, Existing, UpdateListener);
+ ReplaceAllUsesWith(N, Existing);
- // N is now dead. Inform the listener if it exists and delete it.
- if (UpdateListener)
- UpdateListener->NodeDeleted(N, Existing);
+ // N is now dead. Inform the listeners and delete it.
+ for (DAGUpdateListener *DUL = UpdateListeners; DUL; DUL = DUL->Next)
+ DUL->NodeDeleted(N, Existing);
DeleteNodeNotInCSEMaps(N);
return;
}
}
- // If the node doesn't already exist, we updated it. Inform a listener if
- // it exists.
- if (UpdateListener)
- UpdateListener->NodeUpdated(N);
+ // If the node doesn't already exist, we updated it. Inform listeners.
+ for (DAGUpdateListener *DUL = UpdateListeners; DUL; DUL = DUL->Next)
+ DUL->NodeUpdated(N);
}
/// FindModifiedNodeSlot - Find a slot for the specified node if its operands
PointerType::get(Type::getInt8Ty(*getContext()), 0) :
VT.getTypeForEVT(*getContext());
- return TLI.getTargetData()->getABITypeAlignment(Ty);
+ return TLI.getDataLayout()->getABITypeAlignment(Ty);
}
// EntryNode could meaningfully have debug info if we can find it...
SelectionDAG::SelectionDAG(const TargetMachine &tm, CodeGenOpt::Level OL)
: TM(tm), TLI(*tm.getTargetLowering()), TSI(*tm.getSelectionDAGInfo()),
OptLevel(OL), EntryNode(ISD::EntryToken, DebugLoc(), getVTList(MVT::Other)),
- Root(getEntryNode()), Ordering(0) {
+ Root(getEntryNode()), Ordering(0), UpdateListeners(0) {
AllNodes.push_back(&EntryNode);
Ordering = new SDNodeOrdering();
DbgInfo = new SDDbgInfo();
}
SelectionDAG::~SelectionDAG() {
+ assert(!UpdateListeners && "Dangling registered DAGUpdateListeners");
allnodes_clear();
delete Ordering;
delete DbgInfo;
return getConstantFP(APFloat((float)Val), VT, isTarget);
else if (EltVT==MVT::f64)
return getConstantFP(APFloat(Val), VT, isTarget);
- else if (EltVT==MVT::f80 || EltVT==MVT::f128) {
+ else if (EltVT==MVT::f80 || EltVT==MVT::f128 || EltVT==MVT::ppcf128 ||
+ EltVT==MVT::f16) {
bool ignored;
APFloat apf = APFloat(Val);
apf.convert(*EVTToAPFloatSemantics(EltVT), APFloat::rmNearestTiesToEven,
"Cannot set target flags on target-independent globals");
// Truncate (with sign-extension) the offset value to the pointer size.
- EVT PTy = TLI.getPointerTy();
- unsigned BitWidth = PTy.getSizeInBits();
+ unsigned BitWidth = TLI.getPointerTy().getSizeInBits();
if (BitWidth < 64)
- Offset = (Offset << (64 - BitWidth) >> (64 - BitWidth));
+ Offset = SignExtend64(Offset, BitWidth);
const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GV);
if (!GVar) {
ID.AddPointer(GV);
ID.AddInteger(Offset);
ID.AddInteger(TargetFlags);
+ ID.AddInteger(GV->getType()->getAddressSpace());
void *IP = 0;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
return SDValue(E, 0);
assert((TargetFlags == 0 || isTarget) &&
"Cannot set target flags on target-independent globals");
if (Alignment == 0)
- Alignment = TLI.getTargetData()->getPrefTypeAlignment(C->getType());
+ Alignment = TLI.getDataLayout()->getPrefTypeAlignment(C->getType());
unsigned Opc = isTarget ? ISD::TargetConstantPool : ISD::ConstantPool;
FoldingSetNodeID ID;
AddNodeIDNode(ID, Opc, getVTList(VT), 0, 0);
assert((TargetFlags == 0 || isTarget) &&
"Cannot set target flags on target-independent globals");
if (Alignment == 0)
- Alignment = TLI.getTargetData()->getPrefTypeAlignment(C->getType());
+ Alignment = TLI.getDataLayout()->getPrefTypeAlignment(C->getType());
unsigned Opc = isTarget ? ISD::TargetConstantPool : ISD::ConstantPool;
FoldingSetNodeID ID;
AddNodeIDNode(ID, Opc, getVTList(VT), 0, 0);
return SDValue(N, 0);
}
+SDValue SelectionDAG::getTargetIndex(int Index, EVT VT, int64_t Offset,
+ unsigned char TargetFlags) {
+ FoldingSetNodeID ID;
+ AddNodeIDNode(ID, ISD::TargetIndex, getVTList(VT), 0, 0);
+ ID.AddInteger(Index);
+ ID.AddInteger(Offset);
+ ID.AddInteger(TargetFlags);
+ void *IP = 0;
+ if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
+ return SDValue(E, 0);
+
+ SDNode *N = new (NodeAllocator) TargetIndexSDNode(Index, VT, Offset,
+ TargetFlags);
+ CSEMap.InsertNode(N, IP);
+ AllNodes.push_back(N);
+ return SDValue(N, 0);
+}
+
SDValue SelectionDAG::getBasicBlock(MachineBasicBlock *MBB) {
FoldingSetNodeID ID;
AddNodeIDNode(ID, ISD::BasicBlock, getVTList(MVT::Other), 0, 0);
SDValue SelectionDAG::getBlockAddress(const BlockAddress *BA, EVT VT,
+ int64_t Offset,
bool isTarget,
unsigned char TargetFlags) {
unsigned Opc = isTarget ? ISD::TargetBlockAddress : ISD::BlockAddress;
FoldingSetNodeID ID;
AddNodeIDNode(ID, Opc, getVTList(VT), 0, 0);
ID.AddPointer(BA);
+ ID.AddInteger(Offset);
ID.AddInteger(TargetFlags);
void *IP = 0;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
return SDValue(E, 0);
- SDNode *N = new (NodeAllocator) BlockAddressSDNode(Opc, VT, BA, TargetFlags);
+ SDNode *N = new (NodeAllocator) BlockAddressSDNode(Opc, VT, BA, Offset,
+ TargetFlags);
CSEMap.InsertNode(N, IP);
AllNodes.push_back(N);
return SDValue(N, 0);
unsigned ByteSize = VT.getStoreSize();
Type *Ty = VT.getTypeForEVT(*getContext());
unsigned StackAlign =
- std::max((unsigned)TLI.getTargetData()->getPrefTypeAlignment(Ty), minAlign);
+ std::max((unsigned)TLI.getDataLayout()->getPrefTypeAlignment(Ty), minAlign);
int FrameIdx = FrameInfo->CreateStackObject(ByteSize, StackAlign, false);
return getFrameIndex(FrameIdx, TLI.getPointerTy());
VT2.getStoreSizeInBits())/8;
Type *Ty1 = VT1.getTypeForEVT(*getContext());
Type *Ty2 = VT2.getTypeForEVT(*getContext());
- const TargetData *TD = TLI.getTargetData();
+ const DataLayout *TD = TLI.getDataLayout();
unsigned Align = std::max(TD->getPrefTypeAlignment(Ty1),
TD->getPrefTypeAlignment(Ty2));
}
if (ConstantFPSDNode *N1C = dyn_cast<ConstantFPSDNode>(N1.getNode())) {
if (ConstantFPSDNode *N2C = dyn_cast<ConstantFPSDNode>(N2.getNode())) {
- // No compile time operations on this type yet.
- if (N1C->getValueType(0) == MVT::ppcf128)
- return SDValue();
-
APFloat::cmpResult R = N1C->getValueAPF().compare(N2C->getValueAPF());
switch (Cond) {
default: break;
bool SelectionDAG::MaskedValueIsZero(SDValue Op, const APInt &Mask,
unsigned Depth) const {
APInt KnownZero, KnownOne;
- ComputeMaskedBits(Op, Mask, KnownZero, KnownOne, Depth);
+ ComputeMaskedBits(Op, KnownZero, KnownOne, Depth);
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
return (KnownZero & Mask) == Mask;
}
/// known to be either zero or one and return them in the KnownZero/KnownOne
/// bitsets. This code only analyzes bits in Mask, in order to short-circuit
/// processing.
-void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask,
- APInt &KnownZero, APInt &KnownOne,
- unsigned Depth) const {
- unsigned BitWidth = Mask.getBitWidth();
- assert(BitWidth == Op.getValueType().getScalarType().getSizeInBits() &&
- "Mask size mismatches value type size!");
+void SelectionDAG::ComputeMaskedBits(SDValue Op, APInt &KnownZero,
+ APInt &KnownOne, unsigned Depth) const {
+ unsigned BitWidth = Op.getValueType().getScalarType().getSizeInBits();
KnownZero = KnownOne = APInt(BitWidth, 0); // Don't know anything.
- if (Depth == 6 || Mask == 0)
+ if (Depth == 6)
return; // Limit search depth.
APInt KnownZero2, KnownOne2;
switch (Op.getOpcode()) {
case ISD::Constant:
// We know all of the bits for a constant!
- KnownOne = cast<ConstantSDNode>(Op)->getAPIntValue() & Mask;
- KnownZero = ~KnownOne & Mask;
+ KnownOne = cast<ConstantSDNode>(Op)->getAPIntValue();
+ KnownZero = ~KnownOne;
return;
case ISD::AND:
// If either the LHS or the RHS are Zero, the result is zero.
- ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero, KnownOne, Depth+1);
- ComputeMaskedBits(Op.getOperand(0), Mask & ~KnownZero,
- KnownZero2, KnownOne2, Depth+1);
+ ComputeMaskedBits(Op.getOperand(1), KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1);
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
KnownZero |= KnownZero2;
return;
case ISD::OR:
- ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero, KnownOne, Depth+1);
- ComputeMaskedBits(Op.getOperand(0), Mask & ~KnownOne,
- KnownZero2, KnownOne2, Depth+1);
+ ComputeMaskedBits(Op.getOperand(1), KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1);
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
KnownOne |= KnownOne2;
return;
case ISD::XOR: {
- ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero, KnownOne, Depth+1);
- ComputeMaskedBits(Op.getOperand(0), Mask, KnownZero2, KnownOne2, Depth+1);
+ ComputeMaskedBits(Op.getOperand(1), KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1);
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
return;
}
case ISD::MUL: {
- APInt Mask2 = APInt::getAllOnesValue(BitWidth);
- ComputeMaskedBits(Op.getOperand(1), Mask2, KnownZero, KnownOne, Depth+1);
- ComputeMaskedBits(Op.getOperand(0), Mask2, KnownZero2, KnownOne2, Depth+1);
+ ComputeMaskedBits(Op.getOperand(1), KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1);
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
LeadZ = std::min(LeadZ, BitWidth);
KnownZero = APInt::getLowBitsSet(BitWidth, TrailZ) |
APInt::getHighBitsSet(BitWidth, LeadZ);
- KnownZero &= Mask;
return;
}
case ISD::UDIV: {
// For the purposes of computing leading zeros we can conservatively
// treat a udiv as a logical right shift by the power of 2 known to
// be less than the denominator.
- APInt AllOnes = APInt::getAllOnesValue(BitWidth);
- ComputeMaskedBits(Op.getOperand(0),
- AllOnes, KnownZero2, KnownOne2, Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1);
unsigned LeadZ = KnownZero2.countLeadingOnes();
KnownOne2.clearAllBits();
KnownZero2.clearAllBits();
- ComputeMaskedBits(Op.getOperand(1),
- AllOnes, KnownZero2, KnownOne2, Depth+1);
+ ComputeMaskedBits(Op.getOperand(1), KnownZero2, KnownOne2, Depth+1);
unsigned RHSUnknownLeadingOnes = KnownOne2.countLeadingZeros();
if (RHSUnknownLeadingOnes != BitWidth)
LeadZ = std::min(BitWidth,
LeadZ + BitWidth - RHSUnknownLeadingOnes - 1);
- KnownZero = APInt::getHighBitsSet(BitWidth, LeadZ) & Mask;
+ KnownZero = APInt::getHighBitsSet(BitWidth, LeadZ);
return;
}
case ISD::SELECT:
- ComputeMaskedBits(Op.getOperand(2), Mask, KnownZero, KnownOne, Depth+1);
- ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero2, KnownOne2, Depth+1);
+ ComputeMaskedBits(Op.getOperand(2), KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(1), KnownZero2, KnownOne2, Depth+1);
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
KnownZero &= KnownZero2;
return;
case ISD::SELECT_CC:
- ComputeMaskedBits(Op.getOperand(3), Mask, KnownZero, KnownOne, Depth+1);
- ComputeMaskedBits(Op.getOperand(2), Mask, KnownZero2, KnownOne2, Depth+1);
+ ComputeMaskedBits(Op.getOperand(3), KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(2), KnownZero2, KnownOne2, Depth+1);
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
if (ShAmt >= BitWidth)
return;
- ComputeMaskedBits(Op.getOperand(0), Mask.lshr(ShAmt),
- KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1);
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
KnownZero <<= ShAmt;
KnownOne <<= ShAmt;
if (ShAmt >= BitWidth)
return;
- ComputeMaskedBits(Op.getOperand(0), (Mask << ShAmt),
- KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1);
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
KnownZero = KnownZero.lshr(ShAmt);
KnownOne = KnownOne.lshr(ShAmt);
- APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt) & Mask;
+ APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt);
KnownZero |= HighBits; // High bits known zero.
}
return;
if (ShAmt >= BitWidth)
return;
- APInt InDemandedMask = (Mask << ShAmt);
// If any of the demanded bits are produced by the sign extension, we also
// demand the input sign bit.
- APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt) & Mask;
- if (HighBits.getBoolValue())
- InDemandedMask |= APInt::getSignBit(BitWidth);
+ APInt HighBits = APInt::getHighBitsSet(BitWidth, ShAmt);
- ComputeMaskedBits(Op.getOperand(0), InDemandedMask, KnownZero, KnownOne,
- Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1);
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
KnownZero = KnownZero.lshr(ShAmt);
KnownOne = KnownOne.lshr(ShAmt);
// Sign extension. Compute the demanded bits in the result that are not
// present in the input.
- APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - EBits) & Mask;
+ APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - EBits);
APInt InSignBit = APInt::getSignBit(EBits);
- APInt InputDemandedBits = Mask & APInt::getLowBitsSet(BitWidth, EBits);
+ APInt InputDemandedBits = APInt::getLowBitsSet(BitWidth, EBits);
// If the sign extended bits are demanded, we know that the sign
// bit is demanded.
if (NewBits.getBoolValue())
InputDemandedBits |= InSignBit;
- ComputeMaskedBits(Op.getOperand(0), InputDemandedBits,
- KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1);
+ KnownOne &= InputDemandedBits;
+ KnownZero &= InputDemandedBits;
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
// If the sign bit of the input is known set or clear, then we know the
return;
}
case ISD::LOAD: {
+ LoadSDNode *LD = cast<LoadSDNode>(Op);
if (ISD::isZEXTLoad(Op.getNode())) {
- LoadSDNode *LD = cast<LoadSDNode>(Op);
EVT VT = LD->getMemoryVT();
unsigned MemBits = VT.getScalarType().getSizeInBits();
- KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - MemBits) & Mask;
+ KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - MemBits);
+ } else if (const MDNode *Ranges = LD->getRanges()) {
+ computeMaskedBitsLoad(*Ranges, KnownZero);
}
return;
}
case ISD::ZERO_EXTEND: {
EVT InVT = Op.getOperand(0).getValueType();
unsigned InBits = InVT.getScalarType().getSizeInBits();
- APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - InBits) & Mask;
- APInt InMask = Mask.trunc(InBits);
+ APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - InBits);
KnownZero = KnownZero.trunc(InBits);
KnownOne = KnownOne.trunc(InBits);
- ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1);
KnownZero = KnownZero.zext(BitWidth);
KnownOne = KnownOne.zext(BitWidth);
KnownZero |= NewBits;
EVT InVT = Op.getOperand(0).getValueType();
unsigned InBits = InVT.getScalarType().getSizeInBits();
APInt InSignBit = APInt::getSignBit(InBits);
- APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - InBits) & Mask;
- 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;
+ APInt NewBits = APInt::getHighBitsSet(BitWidth, BitWidth - InBits);
KnownZero = KnownZero.trunc(InBits);
KnownOne = KnownOne.trunc(InBits);
- ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1);
// Note if the sign bit is known to be zero or one.
bool SignBitKnownZero = KnownZero.isNegative();
assert(!(SignBitKnownZero && SignBitKnownOne) &&
"Sign bit can't be known to be both zero and 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.trunc(InBits);
- KnownZero &= InMask;
- KnownOne &= InMask;
-
KnownZero = KnownZero.zext(BitWidth);
KnownOne = KnownOne.zext(BitWidth);
case ISD::ANY_EXTEND: {
EVT InVT = Op.getOperand(0).getValueType();
unsigned InBits = InVT.getScalarType().getSizeInBits();
- APInt InMask = Mask.trunc(InBits);
KnownZero = KnownZero.trunc(InBits);
KnownOne = KnownOne.trunc(InBits);
- ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1);
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.zext(InBits);
KnownZero = KnownZero.zext(InBits);
KnownOne = KnownOne.zext(InBits);
- ComputeMaskedBits(Op.getOperand(0), InMask, KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1);
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
KnownZero = KnownZero.trunc(BitWidth);
KnownOne = KnownOne.trunc(BitWidth);
case ISD::AssertZext: {
EVT VT = cast<VTSDNode>(Op.getOperand(1))->getVT();
APInt InMask = APInt::getLowBitsSet(BitWidth, VT.getSizeInBits());
- ComputeMaskedBits(Op.getOperand(0), Mask & InMask, KnownZero,
- KnownOne, Depth+1);
- KnownZero |= (~InMask) & Mask;
+ ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1);
+ KnownZero |= (~InMask);
+ KnownOne &= (~KnownZero);
return;
}
case ISD::FGETSIGN:
unsigned NLZ = (CLHS->getAPIntValue()+1).countLeadingZeros();
// NLZ can't be BitWidth with no sign bit
APInt MaskV = APInt::getHighBitsSet(BitWidth, NLZ+1);
- ComputeMaskedBits(Op.getOperand(1), MaskV, KnownZero2, KnownOne2,
- Depth+1);
+ ComputeMaskedBits(Op.getOperand(1), KnownZero2, KnownOne2, Depth+1);
// If all of the MaskV bits are known to be zero, then we know the
// output top bits are zero, because we now know that the output is
if ((KnownZero2 & MaskV) == MaskV) {
unsigned NLZ2 = CLHS->getAPIntValue().countLeadingZeros();
// Top bits known zero.
- KnownZero = APInt::getHighBitsSet(BitWidth, NLZ2) & Mask;
+ KnownZero = APInt::getHighBitsSet(BitWidth, NLZ2);
}
}
}
// 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.
- APInt Mask2 = APInt::getLowBitsSet(BitWidth,
- BitWidth - Mask.countLeadingZeros());
- ComputeMaskedBits(Op.getOperand(0), Mask2, KnownZero2, KnownOne2, Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1);
assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
unsigned KnownZeroOut = KnownZero2.countTrailingOnes();
- ComputeMaskedBits(Op.getOperand(1), Mask2, KnownZero2, KnownOne2, Depth+1);
+ ComputeMaskedBits(Op.getOperand(1), KnownZero2, KnownOne2, Depth+1);
assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
KnownZeroOut = std::min(KnownZeroOut,
KnownZero2.countTrailingOnes());
if (RA.isPowerOf2()) {
APInt LowBits = RA - 1;
APInt Mask2 = LowBits | APInt::getSignBit(BitWidth);
- ComputeMaskedBits(Op.getOperand(0), Mask2,KnownZero2,KnownOne2,Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero2,KnownOne2,Depth+1);
// The low bits of the first operand are unchanged by the srem.
KnownZero = KnownZero2 & LowBits;
// the upper bits are all one.
if (KnownOne2[BitWidth-1] && ((KnownOne2 & LowBits) != 0))
KnownOne |= ~LowBits;
-
- KnownZero &= Mask;
- KnownOne &= Mask;
-
assert((KnownZero & KnownOne) == 0&&"Bits known to be one AND zero?");
}
}
const APInt &RA = Rem->getAPIntValue();
if (RA.isPowerOf2()) {
APInt LowBits = (RA - 1);
- APInt Mask2 = LowBits & Mask;
- KnownZero |= ~LowBits & Mask;
- ComputeMaskedBits(Op.getOperand(0), Mask2, KnownZero, KnownOne,Depth+1);
+ KnownZero |= ~LowBits;
+ ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne,Depth+1);
assert((KnownZero & KnownOne) == 0&&"Bits known to be one AND zero?");
break;
}
// Since the result is less than or equal to either operand, any leading
// zero bits in either operand must also exist in the result.
- APInt AllOnes = APInt::getAllOnesValue(BitWidth);
- ComputeMaskedBits(Op.getOperand(0), AllOnes, KnownZero, KnownOne,
- Depth+1);
- ComputeMaskedBits(Op.getOperand(1), AllOnes, KnownZero2, KnownOne2,
- Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(1), KnownZero2, KnownOne2, Depth+1);
uint32_t Leaders = std::max(KnownZero.countLeadingOnes(),
KnownZero2.countLeadingOnes());
KnownOne.clearAllBits();
- KnownZero = APInt::getHighBitsSet(BitWidth, Leaders) & Mask;
+ KnownZero = APInt::getHighBitsSet(BitWidth, Leaders);
return;
}
case ISD::FrameIndex:
case ISD::INTRINSIC_W_CHAIN:
case ISD::INTRINSIC_VOID:
// Allow the target to implement this method for its nodes.
- TLI.computeMaskedBitsForTargetNode(Op, Mask, KnownZero, KnownOne, *this,
- Depth);
+ TLI.computeMaskedBitsForTargetNode(Op, KnownZero, KnownOne, *this, Depth);
return;
}
}
if (ConstantSDNode *CRHS = dyn_cast<ConstantSDNode>(Op.getOperand(1)))
if (CRHS->isAllOnesValue()) {
APInt KnownZero, KnownOne;
- APInt Mask = APInt::getAllOnesValue(VTBits);
- ComputeMaskedBits(Op.getOperand(0), Mask, KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1);
// If the input is known to be 0 or 1, the output is 0/-1, which is all
// sign bits set.
- if ((KnownZero | APInt(VTBits, 1)) == Mask)
+ if ((KnownZero | APInt(VTBits, 1)).isAllOnesValue())
return VTBits;
// If we are subtracting one from a positive number, there is no carry
if (ConstantSDNode *CLHS = dyn_cast<ConstantSDNode>(Op.getOperand(0)))
if (CLHS->isNullValue()) {
APInt KnownZero, KnownOne;
- APInt Mask = APInt::getAllOnesValue(VTBits);
- ComputeMaskedBits(Op.getOperand(1), Mask, KnownZero, KnownOne, Depth+1);
+ ComputeMaskedBits(Op.getOperand(1), KnownZero, KnownOne, Depth+1);
// If the input is known to be 0 or 1, the output is 0/-1, which is all
// sign bits set.
- if ((KnownZero | APInt(VTBits, 1)) == Mask)
+ if ((KnownZero | APInt(VTBits, 1)).isAllOnesValue())
return VTBits;
// If the input is known to be positive (the sign bit is known clear),
}
// Handle LOADX separately here. EXTLOAD case will fallthrough.
- if (Op.getOpcode() == ISD::LOAD) {
- LoadSDNode *LD = cast<LoadSDNode>(Op);
+ if (LoadSDNode *LD = dyn_cast<LoadSDNode>(Op)) {
unsigned ExtType = LD->getExtensionType();
switch (ExtType) {
default: break;
// Finally, if we can prove that the top bits of the result are 0's or 1's,
// use this information.
APInt KnownZero, KnownOne;
- APInt Mask = APInt::getAllOnesValue(VTBits);
- ComputeMaskedBits(Op, Mask, KnownZero, KnownOne, Depth);
+ ComputeMaskedBits(Op, KnownZero, KnownOne, Depth);
+ APInt Mask;
if (KnownZero.isNegative()) { // sign bit is 0
Mask = KnownZero;
} else if (KnownOne.isNegative()) { // sign bit is 1;
return getConstant(Val.zextOrTrunc(VT.getSizeInBits()), VT);
case ISD::UINT_TO_FP:
case ISD::SINT_TO_FP: {
- // No compile time operations on ppcf128.
- if (VT == MVT::ppcf128) break;
APFloat apf(APInt::getNullValue(VT.getSizeInBits()));
(void)apf.convertFromAPInt(Val,
Opcode==ISD::SINT_TO_FP,
}
case ISD::BITCAST:
if (VT == MVT::f32 && C->getValueType(0) == MVT::i32)
- return getConstantFP(Val.bitsToFloat(), VT);
+ return getConstantFP(APFloat(Val), VT);
else if (VT == MVT::f64 && C->getValueType(0) == MVT::i64)
- return getConstantFP(Val.bitsToDouble(), VT);
+ return getConstantFP(APFloat(Val), VT);
break;
case ISD::BSWAP:
return getConstant(Val.byteSwap(), VT);
// Constant fold unary operations with a floating point constant operand.
if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(Operand.getNode())) {
APFloat V = C->getValueAPF(); // make copy
- if (VT != MVT::ppcf128 && Operand.getValueType() != MVT::ppcf128) {
- switch (Opcode) {
- case ISD::FNEG:
- V.changeSign();
+ switch (Opcode) {
+ case ISD::FNEG:
+ V.changeSign();
+ return getConstantFP(V, VT);
+ case ISD::FABS:
+ V.clearSign();
+ return getConstantFP(V, VT);
+ case ISD::FCEIL: {
+ APFloat::opStatus fs = V.roundToIntegral(APFloat::rmTowardPositive);
+ if (fs == APFloat::opOK || fs == APFloat::opInexact)
return getConstantFP(V, VT);
- case ISD::FABS:
- V.clearSign();
+ break;
+ }
+ case ISD::FTRUNC: {
+ APFloat::opStatus fs = V.roundToIntegral(APFloat::rmTowardZero);
+ if (fs == APFloat::opOK || fs == APFloat::opInexact)
return getConstantFP(V, VT);
- case ISD::FP_ROUND:
- case ISD::FP_EXTEND: {
- bool ignored;
- // This can return overflow, underflow, or inexact; we don't care.
- // FIXME need to be more flexible about rounding mode.
- (void)V.convert(*EVTToAPFloatSemantics(VT),
- APFloat::rmNearestTiesToEven, &ignored);
+ break;
+ }
+ case ISD::FFLOOR: {
+ APFloat::opStatus fs = V.roundToIntegral(APFloat::rmTowardNegative);
+ if (fs == APFloat::opOK || fs == APFloat::opInexact)
return getConstantFP(V, VT);
- }
- case ISD::FP_TO_SINT:
- case ISD::FP_TO_UINT: {
- integerPart x[2];
- bool ignored;
- assert(integerPartWidth >= 64);
- // FIXME need to be more flexible about rounding mode.
- APFloat::opStatus s = V.convertToInteger(x, VT.getSizeInBits(),
- Opcode==ISD::FP_TO_SINT,
- APFloat::rmTowardZero, &ignored);
- if (s==APFloat::opInvalidOp) // inexact is OK, in fact usual
- break;
- APInt api(VT.getSizeInBits(), x);
- return getConstant(api, VT);
- }
- 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)
- return getConstant(V.bitcastToAPInt().getZExtValue(), VT);
+ break;
+ }
+ case ISD::FP_EXTEND: {
+ bool ignored;
+ // This can return overflow, underflow, or inexact; we don't care.
+ // FIXME need to be more flexible about rounding mode.
+ (void)V.convert(*EVTToAPFloatSemantics(VT),
+ APFloat::rmNearestTiesToEven, &ignored);
+ return getConstantFP(V, VT);
+ }
+ case ISD::FP_TO_SINT:
+ case ISD::FP_TO_UINT: {
+ integerPart x[2];
+ bool ignored;
+ assert(integerPartWidth >= 64);
+ // FIXME need to be more flexible about rounding mode.
+ APFloat::opStatus s = V.convertToInteger(x, VT.getSizeInBits(),
+ Opcode==ISD::FP_TO_SINT,
+ APFloat::rmTowardZero, &ignored);
+ if (s==APFloat::opInvalidOp) // inexact is OK, in fact usual
break;
- }
+ APInt api(VT.getSizeInBits(), x);
+ return getConstant(api, VT);
+ }
+ 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)
+ return getConstant(V.bitcastToAPInt().getZExtValue(), VT);
+ break;
}
}
if (N1 == N2) return N1;
break;
case ISD::CONCAT_VECTORS:
+ // Concat of UNDEFs is UNDEF.
+ if (N1.getOpcode() == ISD::UNDEF &&
+ N2.getOpcode() == ISD::UNDEF)
+ return getUNDEF(VT);
+
// A CONCAT_VECTOR with all operands BUILD_VECTOR can be simplified to
// one big BUILD_VECTOR.
if (N1.getOpcode() == ISD::BUILD_VECTOR &&
if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(N2))
if (CFP->getValueAPF().isZero())
return N1;
+ } else if (Opcode == ISD::FMUL) {
+ ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(N1);
+ SDValue V = N2;
+
+ // If the first operand isn't the constant, try the second
+ if (!CFP) {
+ CFP = dyn_cast<ConstantFPSDNode>(N2);
+ V = N1;
+ }
+
+ if (CFP) {
+ // 0*x --> 0
+ if (CFP->isZero())
+ return SDValue(CFP,0);
+ // 1*x --> x
+ if (CFP->isExactlyValue(1.0))
+ return V;
+ }
}
}
assert(VT.isFloatingPoint() && "This operator only applies to FP types!");
// expanding large vector constants.
if (N2C && N1.getOpcode() == ISD::BUILD_VECTOR) {
SDValue Elt = N1.getOperand(N2C->getZExtValue());
- EVT VEltTy = N1.getValueType().getVectorElementType();
- if (Elt.getValueType() != VEltTy) {
+
+ if (VT != Elt.getValueType())
// If the vector element type is not legal, the BUILD_VECTOR operands
- // are promoted and implicitly truncated. Make that explicit here.
- Elt = getNode(ISD::TRUNCATE, DL, VEltTy, Elt);
- }
- if (VT != VEltTy) {
- // If the vector element type is not legal, the EXTRACT_VECTOR_ELT
- // result is implicitly extended.
- Elt = getNode(ISD::ANY_EXTEND, DL, VT, Elt);
- }
+ // are promoted and implicitly truncated, and the result implicitly
+ // extended. Make that explicit here.
+ Elt = getAnyExtOrTrunc(Elt, DL, VT);
+
return Elt;
}
// Cannonicalize constant to RHS if commutative
std::swap(N1CFP, N2CFP);
std::swap(N1, N2);
- } else if (N2CFP && VT != MVT::ppcf128) {
+ } else if (N2CFP) {
APFloat V1 = N1CFP->getValueAPF(), V2 = N2CFP->getValueAPF();
APFloat::opStatus s;
switch (Opcode) {
default: break;
}
}
+
+ if (Opcode == ISD::FP_ROUND) {
+ APFloat V = N1CFP->getValueAPF(); // make copy
+ bool ignored;
+ // This can return overflow, underflow, or inexact; we don't care.
+ // FIXME need to be more flexible about rounding mode.
+ (void)V.convert(*EVTToAPFloatSemantics(VT),
+ APFloat::rmNearestTiesToEven, &ignored);
+ return getConstantFP(V, VT);
+ }
}
// Canonicalize an UNDEF to the RHS, even over a constant.
unsigned NumVTBytes = VT.getSizeInBits() / 8;
unsigned NumBytes = std::min(NumVTBytes, unsigned(Str.size()));
- uint64_t Val = 0;
+ APInt Val(NumBytes*8, 0);
if (TLI.isLittleEndian()) {
for (unsigned i = 0; i != NumBytes; ++i)
Val |= (uint64_t)(unsigned char)Str[i] << i*8;
Val |= (uint64_t)(unsigned char)Str[i] << (NumVTBytes-i-1)*8;
}
- return DAG.getConstant(Val, VT);
+ // If the "cost" of materializing the integer immediate is 1 or free, then
+ // it is cost effective to turn the load into the immediate.
+ if (DAG.getTarget().getScalarTargetTransformInfo()->
+ getIntImmCost(Val, VT.getTypeForEVT(*DAG.getContext())) < 2)
+ return DAG.getConstant(Val, VT);
+ return SDValue(0, 0);
}
/// getMemBasePlusOffset - Returns base and offset node for the
static bool FindOptimalMemOpLowering(std::vector<EVT> &MemOps,
unsigned Limit, uint64_t Size,
unsigned DstAlign, unsigned SrcAlign,
- bool IsZeroVal,
+ bool IsMemset,
+ bool ZeroMemset,
bool MemcpyStrSrc,
+ bool AllowOverlap,
SelectionDAG &DAG,
const TargetLowering &TLI) {
assert((SrcAlign == 0 || SrcAlign >= DstAlign) &&
// 'MemcpyStrSrc' indicates whether the memcpy source is constant so it does
// not need to be loaded.
EVT VT = TLI.getOptimalMemOpType(Size, DstAlign, SrcAlign,
- IsZeroVal, MemcpyStrSrc,
+ IsMemset, ZeroMemset, MemcpyStrSrc,
DAG.getMachineFunction());
if (VT == MVT::Other) {
- if (DstAlign >= TLI.getTargetData()->getPointerPrefAlignment() ||
+ if (DstAlign >= TLI.getDataLayout()->getPointerPrefAlignment() ||
TLI.allowsUnalignedMemoryAccesses(VT)) {
VT = TLI.getPointerTy();
} else {
unsigned VTSize = VT.getSizeInBits() / 8;
while (VTSize > Size) {
// For now, only use non-vector load / store's for the left-over pieces.
+ EVT NewVT = VT;
+ unsigned NewVTSize;
+
+ bool Found = false;
if (VT.isVector() || VT.isFloatingPoint()) {
- VT = MVT::i64;
- while (!TLI.isTypeLegal(VT))
- VT = (MVT::SimpleValueType)(VT.getSimpleVT().SimpleTy - 1);
- VTSize = VT.getSizeInBits() / 8;
- } else {
- // This can result in a type that is not legal on the target, e.g.
- // 1 or 2 bytes on PPC.
- VT = (MVT::SimpleValueType)(VT.getSimpleVT().SimpleTy - 1);
- VTSize >>= 1;
+ NewVT = (VT.getSizeInBits() > 64) ? MVT::i64 : MVT::i32;
+ if (TLI.isOperationLegalOrCustom(ISD::STORE, NewVT) &&
+ TLI.isSafeMemOpType(NewVT.getSimpleVT()))
+ Found = true;
+ else if (NewVT == MVT::i64 &&
+ TLI.isOperationLegalOrCustom(ISD::STORE, MVT::f64) &&
+ TLI.isSafeMemOpType(MVT::f64)) {
+ // i64 is usually not legal on 32-bit targets, but f64 may be.
+ NewVT = MVT::f64;
+ Found = true;
+ }
+ }
+
+ if (!Found) {
+ do {
+ NewVT = (MVT::SimpleValueType)(NewVT.getSimpleVT().SimpleTy - 1);
+ if (NewVT == MVT::i8)
+ break;
+ } while (!TLI.isSafeMemOpType(NewVT.getSimpleVT()));
+ }
+ NewVTSize = NewVT.getSizeInBits() / 8;
+
+ // If the new VT cannot cover all of the remaining bits, then consider
+ // issuing a (or a pair of) unaligned and overlapping load / store.
+ // FIXME: Only does this for 64-bit or more since we don't have proper
+ // cost model for unaligned load / store.
+ bool Fast;
+ if (NumMemOps && AllowOverlap &&
+ VTSize >= 8 && NewVTSize < Size &&
+ TLI.allowsUnalignedMemoryAccesses(VT, &Fast) && Fast)
+ VTSize = Size;
+ else {
+ VT = NewVT;
+ VTSize = NewVTSize;
}
}
if (++NumMemOps > Limit)
return false;
+
MemOps.push_back(VT);
Size -= VTSize;
}
bool DstAlignCanChange = false;
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
- bool OptSize = MF.getFunction()->hasFnAttr(Attribute::OptimizeForSize);
+ bool OptSize =
+ MF.getFunction()->getAttributes().
+ hasAttribute(AttributeSet::FunctionIndex, Attribute::OptimizeForSize);
FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(Dst);
if (FI && !MFI->isFixedObjectIndex(FI->getIndex()))
DstAlignCanChange = true;
if (!FindOptimalMemOpLowering(MemOps, Limit, Size,
(DstAlignCanChange ? 0 : Align),
(isZeroStr ? 0 : SrcAlign),
- true, CopyFromStr, DAG, TLI))
+ false, false, CopyFromStr, true, DAG, TLI))
return SDValue();
if (DstAlignCanChange) {
Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext());
- unsigned NewAlign = (unsigned) TLI.getTargetData()->getABITypeAlignment(Ty);
+ unsigned NewAlign = (unsigned) TLI.getDataLayout()->getABITypeAlignment(Ty);
if (NewAlign > Align) {
// Give the stack frame object a larger alignment if needed.
if (MFI->getObjectAlignment(FI->getIndex()) < NewAlign)
unsigned VTSize = VT.getSizeInBits() / 8;
SDValue Value, Store;
+ if (VTSize > Size) {
+ // Issuing an unaligned load / store pair that overlaps with the previous
+ // pair. Adjust the offset accordingly.
+ assert(i == NumMemOps-1 && i != 0);
+ SrcOff -= VTSize - Size;
+ DstOff -= VTSize - Size;
+ }
+
if (CopyFromStr &&
(isZeroStr || (VT.isInteger() && !VT.isVector()))) {
// It's unlikely a store of a vector immediate can be done in a single
// FIXME: Handle other cases where store of vector immediate is done in
// a single instruction.
Value = getMemsetStringVal(VT, dl, DAG, TLI, Str.substr(SrcOff));
- Store = DAG.getStore(Chain, dl, Value,
- getMemBasePlusOffset(Dst, DstOff, DAG),
- DstPtrInfo.getWithOffset(DstOff), isVol,
- false, Align);
- } else {
+ if (Value.getNode())
+ Store = DAG.getStore(Chain, dl, Value,
+ getMemBasePlusOffset(Dst, DstOff, DAG),
+ DstPtrInfo.getWithOffset(DstOff), isVol,
+ false, Align);
+ }
+
+ if (!Store.getNode()) {
// The type might not be legal for the target. This should only happen
// if the type is smaller than a legal type, as on PPC, so the right
// thing to do is generate a LoadExt/StoreTrunc pair. These simplify
OutChains.push_back(Store);
SrcOff += VTSize;
DstOff += VTSize;
+ Size -= VTSize;
}
return DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
bool DstAlignCanChange = false;
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
- bool OptSize = MF.getFunction()->hasFnAttr(Attribute::OptimizeForSize);
+ bool OptSize = MF.getFunction()->getAttributes().
+ hasAttribute(AttributeSet::FunctionIndex, Attribute::OptimizeForSize);
FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(Dst);
if (FI && !MFI->isFixedObjectIndex(FI->getIndex()))
DstAlignCanChange = true;
unsigned Limit = AlwaysInline ? ~0U : TLI.getMaxStoresPerMemmove(OptSize);
if (!FindOptimalMemOpLowering(MemOps, Limit, Size,
- (DstAlignCanChange ? 0 : Align),
- SrcAlign, true, false, DAG, TLI))
+ (DstAlignCanChange ? 0 : Align), SrcAlign,
+ false, false, false, false, DAG, TLI))
return SDValue();
if (DstAlignCanChange) {
Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext());
- unsigned NewAlign = (unsigned) TLI.getTargetData()->getABITypeAlignment(Ty);
+ unsigned NewAlign = (unsigned) TLI.getDataLayout()->getABITypeAlignment(Ty);
if (NewAlign > Align) {
// Give the stack frame object a larger alignment if needed.
if (MFI->getObjectAlignment(FI->getIndex()) < NewAlign)
bool DstAlignCanChange = false;
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
- bool OptSize = MF.getFunction()->hasFnAttr(Attribute::OptimizeForSize);
+ bool OptSize = MF.getFunction()->getAttributes().
+ hasAttribute(AttributeSet::FunctionIndex, Attribute::OptimizeForSize);
FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(Dst);
if (FI && !MFI->isFixedObjectIndex(FI->getIndex()))
DstAlignCanChange = true;
isa<ConstantSDNode>(Src) && cast<ConstantSDNode>(Src)->isNullValue();
if (!FindOptimalMemOpLowering(MemOps, TLI.getMaxStoresPerMemset(OptSize),
Size, (DstAlignCanChange ? 0 : Align), 0,
- IsZeroVal, false, DAG, TLI))
+ true, IsZeroVal, false, true, DAG, TLI))
return SDValue();
if (DstAlignCanChange) {
Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext());
- unsigned NewAlign = (unsigned) TLI.getTargetData()->getABITypeAlignment(Ty);
+ unsigned NewAlign = (unsigned) TLI.getDataLayout()->getABITypeAlignment(Ty);
if (NewAlign > Align) {
// Give the stack frame object a larger alignment if needed.
if (MFI->getObjectAlignment(FI->getIndex()) < NewAlign)
for (unsigned i = 0; i < NumMemOps; i++) {
EVT VT = MemOps[i];
+ unsigned VTSize = VT.getSizeInBits() / 8;
+ if (VTSize > Size) {
+ // Issuing an unaligned load / store pair that overlaps with the previous
+ // pair. Adjust the offset accordingly.
+ assert(i == NumMemOps-1 && i != 0);
+ DstOff -= VTSize - Size;
+ }
// If this store is smaller than the largest store see whether we can get
// the smaller value for free with a truncate.
isVol, false, Align);
OutChains.push_back(Store);
DstOff += VT.getSizeInBits() / 8;
+ Size -= VTSize;
}
return DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
// Emit a library call.
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
- Entry.Ty = TLI.getTargetData()->getIntPtrType(*getContext());
+ Entry.Ty = TLI.getDataLayout()->getIntPtrType(*getContext());
Entry.Node = Dst; Args.push_back(Entry);
Entry.Node = Src; Args.push_back(Entry);
Entry.Node = Size; Args.push_back(Entry);
// FIXME: pass in DebugLoc
- std::pair<SDValue,SDValue> CallResult =
- TLI.LowerCallTo(Chain, Type::getVoidTy(*getContext()),
+ TargetLowering::
+ CallLoweringInfo CLI(Chain, Type::getVoidTy(*getContext()),
false, false, false, false, 0,
TLI.getLibcallCallingConv(RTLIB::MEMCPY),
/*isTailCall=*/false,
getExternalSymbol(TLI.getLibcallName(RTLIB::MEMCPY),
TLI.getPointerTy()),
Args, *this, dl);
+ std::pair<SDValue,SDValue> CallResult = TLI.LowerCallTo(CLI);
+
return CallResult.second;
}
// Emit a library call.
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
- Entry.Ty = TLI.getTargetData()->getIntPtrType(*getContext());
+ Entry.Ty = TLI.getDataLayout()->getIntPtrType(*getContext());
Entry.Node = Dst; Args.push_back(Entry);
Entry.Node = Src; Args.push_back(Entry);
Entry.Node = Size; Args.push_back(Entry);
// FIXME: pass in DebugLoc
- std::pair<SDValue,SDValue> CallResult =
- TLI.LowerCallTo(Chain, Type::getVoidTy(*getContext()),
+ TargetLowering::
+ CallLoweringInfo CLI(Chain, Type::getVoidTy(*getContext()),
false, false, false, false, 0,
TLI.getLibcallCallingConv(RTLIB::MEMMOVE),
/*isTailCall=*/false,
getExternalSymbol(TLI.getLibcallName(RTLIB::MEMMOVE),
TLI.getPointerTy()),
Args, *this, dl);
+ std::pair<SDValue,SDValue> CallResult = TLI.LowerCallTo(CLI);
+
return CallResult.second;
}
return Result;
// Emit a library call.
- Type *IntPtrTy = TLI.getTargetData()->getIntPtrType(*getContext());
+ Type *IntPtrTy = TLI.getDataLayout()->getIntPtrType(*getContext());
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Node = Dst; Entry.Ty = IntPtrTy;
Entry.isSExt = false;
Args.push_back(Entry);
// FIXME: pass in DebugLoc
- std::pair<SDValue,SDValue> CallResult =
- TLI.LowerCallTo(Chain, Type::getVoidTy(*getContext()),
+ TargetLowering::
+ CallLoweringInfo CLI(Chain, Type::getVoidTy(*getContext()),
false, false, false, false, 0,
TLI.getLibcallCallingConv(RTLIB::MEMSET),
/*isTailCall=*/false,
getExternalSymbol(TLI.getLibcallName(RTLIB::MEMSET),
TLI.getPointerTy()),
Args, *this, dl);
+ std::pair<SDValue,SDValue> CallResult = TLI.LowerCallTo(CLI);
+
return CallResult.second;
}
SDValue Swp, MachinePointerInfo PtrInfo,
unsigned Alignment,
AtomicOrdering Ordering,
- SynchronizationScope SynchScope) {
+ SynchronizationScope SynchScope) {
if (Alignment == 0) // Ensure that codegen never sees alignment 0
Alignment = getEVTAlignment(MemVT);
MachineFunction &MF = getMachineFunction();
- unsigned Flags = MachineMemOperand::MOLoad | MachineMemOperand::MOStore;
+ // All atomics are load and store, except for ATMOIC_LOAD and ATOMIC_STORE.
// 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;
+ unsigned Flags = MachineMemOperand::MOVolatile;
+ if (Opcode != ISD::ATOMIC_STORE)
+ Flags |= MachineMemOperand::MOLoad;
+ if (Opcode != ISD::ATOMIC_LOAD)
+ Flags |= MachineMemOperand::MOStore;
MachineMemOperand *MMO =
MF.getMachineMemOperand(PtrInfo, Flags, MemVT.getStoreSize(), Alignment);
ID.AddInteger(MemVT.getRawBits());
SDValue Ops[] = {Chain, Ptr, Cmp, Swp};
AddNodeIDNode(ID, Opcode, VTs, Ops, 4);
+ ID.AddInteger(MMO->getPointerInfo().getAddrSpace());
void* IP = 0;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) {
cast<AtomicSDNode>(E)->refineAlignment(MMO);
Alignment = getEVTAlignment(MemVT);
MachineFunction &MF = getMachineFunction();
- // A monotonic store does not load; a release store "loads" in the sense
- // that other stores cannot be sunk past it.
+ // An atomic store does not load. An atomic load does not store.
// (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.
+ // For now, atomics are considered to be volatile always, and they are
+ // chained as such.
// FIXME: Volatile isn't really correct; we should keep track of atomic
// orderings in the memoperand.
- Flags |= MachineMemOperand::MOVolatile;
+ unsigned Flags = MachineMemOperand::MOVolatile;
+ if (Opcode != ISD::ATOMIC_STORE)
+ Flags |= MachineMemOperand::MOLoad;
+ if (Opcode != ISD::ATOMIC_LOAD)
+ Flags |= MachineMemOperand::MOStore;
MachineMemOperand *MMO =
MF.getMachineMemOperand(MachinePointerInfo(PtrVal), Flags,
ID.AddInteger(MemVT.getRawBits());
SDValue Ops[] = {Chain, Ptr, Val};
AddNodeIDNode(ID, Opcode, VTs, Ops, 3);
+ ID.AddInteger(MMO->getPointerInfo().getAddrSpace());
void* IP = 0;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) {
cast<AtomicSDNode>(E)->refineAlignment(MMO);
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.
+ // An atomic store does not load. An atomic load does not store.
+ // (An atomicrmw obviously both loads and stores.)
+ // For now, atomics are considered to be volatile always, and they are
+ // chained as such.
// FIXME: Volatile isn't really correct; we should keep track of atomic
// orderings in the memoperand.
- Flags |= MachineMemOperand::MOVolatile;
+ unsigned Flags = MachineMemOperand::MOVolatile;
+ if (Opcode != ISD::ATOMIC_STORE)
+ Flags |= MachineMemOperand::MOLoad;
+ if (Opcode != ISD::ATOMIC_LOAD)
+ Flags |= MachineMemOperand::MOStore;
MachineMemOperand *MMO =
MF.getMachineMemOperand(MachinePointerInfo(PtrVal), Flags,
ID.AddInteger(MemVT.getRawBits());
SDValue Ops[] = {Chain, Ptr};
AddNodeIDNode(ID, Opcode, VTs, Ops, 2);
+ ID.AddInteger(MMO->getPointerInfo().getAddrSpace());
void* IP = 0;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) {
cast<AtomicSDNode>(E)->refineAlignment(MMO);
assert((Opcode == ISD::INTRINSIC_VOID ||
Opcode == ISD::INTRINSIC_W_CHAIN ||
Opcode == ISD::PREFETCH ||
+ Opcode == ISD::LIFETIME_START ||
+ Opcode == ISD::LIFETIME_END ||
(Opcode <= INT_MAX &&
(int)Opcode >= ISD::FIRST_TARGET_MEMORY_OPCODE)) &&
"Opcode is not a memory-accessing opcode!");
if (VTList.VTs[VTList.NumVTs-1] != MVT::Glue) {
FoldingSetNodeID ID;
AddNodeIDNode(ID, Opcode, VTList, Ops, NumOps);
+ ID.AddInteger(MMO->getPointerInfo().getAddrSpace());
void *IP = 0;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) {
cast<MemIntrinsicSDNode>(E)->refineAlignment(MMO);
SDValue Ptr, SDValue Offset,
MachinePointerInfo PtrInfo, EVT MemVT,
bool isVolatile, bool isNonTemporal, bool isInvariant,
- unsigned Alignment, const MDNode *TBAAInfo) {
- assert(Chain.getValueType() == MVT::Other &&
+ unsigned Alignment, const MDNode *TBAAInfo,
+ const MDNode *Ranges) {
+ assert(Chain.getValueType() == MVT::Other &&
"Invalid chain type");
if (Alignment == 0) // Ensure that codegen never sees alignment 0
Alignment = getEVTAlignment(VT);
MachineFunction &MF = getMachineFunction();
MachineMemOperand *MMO =
MF.getMachineMemOperand(PtrInfo, Flags, MemVT.getStoreSize(), Alignment,
- TBAAInfo);
+ TBAAInfo, Ranges);
return getLoad(AM, ExtType, VT, dl, Chain, Ptr, Offset, MemVT, 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()));
+ ID.AddInteger(MMO->getPointerInfo().getAddrSpace());
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,
- bool isInvariant, unsigned Alignment,
- const MDNode *TBAAInfo) {
+ bool isInvariant, unsigned Alignment,
+ const MDNode *TBAAInfo,
+ const MDNode *Ranges) {
SDValue Undef = getUNDEF(Ptr.getValueType());
return getLoad(ISD::UNINDEXED, ISD::NON_EXTLOAD, VT, dl, Chain, Ptr, Undef,
- PtrInfo, VT, isVolatile, isNonTemporal, isInvariant, Alignment,
- TBAAInfo);
+ PtrInfo, VT, isVolatile, isNonTemporal, isInvariant, Alignment,
+ TBAAInfo, Ranges);
}
SDValue SelectionDAG::getExtLoad(ISD::LoadExtType ExtType, DebugLoc dl, EVT VT,
"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->getMemoryVT(), LD->isVolatile(), LD->isNonTemporal(),
false, LD->getAlignment());
}
SDValue Ptr, MachinePointerInfo PtrInfo,
bool isVolatile, bool isNonTemporal,
unsigned Alignment, const MDNode *TBAAInfo) {
- assert(Chain.getValueType() == MVT::Other &&
+ assert(Chain.getValueType() == MVT::Other &&
"Invalid chain type");
if (Alignment == 0) // Ensure that codegen never sees alignment 0
Alignment = getEVTAlignment(Val.getValueType());
SDValue SelectionDAG::getStore(SDValue Chain, DebugLoc dl, SDValue Val,
SDValue Ptr, MachineMemOperand *MMO) {
- assert(Chain.getValueType() == MVT::Other &&
+ assert(Chain.getValueType() == MVT::Other &&
"Invalid chain type");
EVT VT = Val.getValueType();
SDVTList VTs = getVTList(MVT::Other);
ID.AddInteger(VT.getRawBits());
ID.AddInteger(encodeMemSDNodeFlags(false, ISD::UNINDEXED, MMO->isVolatile(),
MMO->isNonTemporal(), MMO->isInvariant()));
+ ID.AddInteger(MMO->getPointerInfo().getAddrSpace());
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 &&
+ assert(Chain.getValueType() == MVT::Other &&
"Invalid chain type");
if (Alignment == 0) // Ensure that codegen never sees alignment 0
Alignment = getEVTAlignment(SVT);
MachineMemOperand *MMO) {
EVT VT = Val.getValueType();
- assert(Chain.getValueType() == MVT::Other &&
+ assert(Chain.getValueType() == MVT::Other &&
"Invalid chain type");
if (VT == SVT)
return getStore(Chain, dl, Val, Ptr, MMO);
ID.AddInteger(SVT.getRawBits());
ID.AddInteger(encodeMemSDNodeFlags(true, ISD::UNINDEXED, MMO->isVolatile(),
MMO->isNonTemporal(), MMO->isInvariant()));
+ ID.AddInteger(MMO->getPointerInfo().getAddrSpace());
void *IP = 0;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) {
cast<StoreSDNode>(E)->refineAlignment(MMO);
AddNodeIDNode(ID, ISD::STORE, VTs, Ops, 4);
ID.AddInteger(ST->getMemoryVT().getRawBits());
ID.AddInteger(ST->getRawSubclassData());
+ ID.AddInteger(ST->getPointerInfo().getAddrSpace());
void *IP = 0;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
return SDValue(E, 0);
if (I->NumVTs != NumVTs || VTs[0] != I->VTs[0] || VTs[1] != I->VTs[1])
continue;
- bool NoMatch = false;
- for (unsigned i = 2; i != NumVTs; ++i)
- if (VTs[i] != I->VTs[i]) {
- NoMatch = true;
- break;
- }
- if (!NoMatch)
+ if (std::equal(&VTs[2], &VTs[NumVTs], &I->VTs[2]))
return *I;
}
/// pointed to by a use iterator is deleted, increment the use iterator
/// so that it doesn't dangle.
///
-/// This class also manages a "downlink" DAGUpdateListener, to forward
-/// messages to ReplaceAllUsesWith's callers.
-///
class RAUWUpdateListener : public SelectionDAG::DAGUpdateListener {
- SelectionDAG::DAGUpdateListener *DownLink;
SDNode::use_iterator &UI;
SDNode::use_iterator &UE;
// Increment the iterator as needed.
while (UI != UE && N == *UI)
++UI;
-
- // Then forward the message.
- if (DownLink) DownLink->NodeDeleted(N, E);
- }
-
- virtual void NodeUpdated(SDNode *N) {
- // Just forward the message.
- if (DownLink) DownLink->NodeUpdated(N);
}
public:
- RAUWUpdateListener(SelectionDAG::DAGUpdateListener *dl,
+ RAUWUpdateListener(SelectionDAG &d,
SDNode::use_iterator &ui,
SDNode::use_iterator &ue)
- : DownLink(dl), UI(ui), UE(ue) {}
+ : SelectionDAG::DAGUpdateListener(d), UI(ui), UE(ue) {}
};
}
///
/// This version assumes From has a single result value.
///
-void SelectionDAG::ReplaceAllUsesWith(SDValue FromN, SDValue To,
- DAGUpdateListener *UpdateListener) {
+void SelectionDAG::ReplaceAllUsesWith(SDValue FromN, SDValue To) {
SDNode *From = FromN.getNode();
assert(From->getNumValues() == 1 && FromN.getResNo() == 0 &&
"Cannot replace with this method!");
// is replaced by To, we don't want to replace of all its users with To
// too. See PR3018 for more info.
SDNode::use_iterator UI = From->use_begin(), UE = From->use_end();
- RAUWUpdateListener Listener(UpdateListener, UI, UE);
+ RAUWUpdateListener Listener(*this, UI, UE);
while (UI != UE) {
SDNode *User = *UI;
// Now that we have modified User, add it back to the CSE maps. If it
// already exists there, recursively merge the results together.
- AddModifiedNodeToCSEMaps(User, &Listener);
+ AddModifiedNodeToCSEMaps(User);
}
// If we just RAUW'd the root, take note.
/// This version assumes that for each value of From, there is a
/// corresponding value in To in the same position with the same type.
///
-void SelectionDAG::ReplaceAllUsesWith(SDNode *From, SDNode *To,
- DAGUpdateListener *UpdateListener) {
+void SelectionDAG::ReplaceAllUsesWith(SDNode *From, SDNode *To) {
#ifndef NDEBUG
for (unsigned i = 0, e = From->getNumValues(); i != e; ++i)
assert((!From->hasAnyUseOfValue(i) ||
// Iterate over just the existing users of From. See the comments in
// the ReplaceAllUsesWith above.
SDNode::use_iterator UI = From->use_begin(), UE = From->use_end();
- RAUWUpdateListener Listener(UpdateListener, UI, UE);
+ RAUWUpdateListener Listener(*this, UI, UE);
while (UI != UE) {
SDNode *User = *UI;
// Now that we have modified User, add it back to the CSE maps. If it
// already exists there, recursively merge the results together.
- AddModifiedNodeToCSEMaps(User, &Listener);
+ AddModifiedNodeToCSEMaps(User);
}
// If we just RAUW'd the root, take note.
///
/// This version can replace From with any result values. To must match the
/// number and types of values returned by From.
-void SelectionDAG::ReplaceAllUsesWith(SDNode *From,
- const SDValue *To,
- DAGUpdateListener *UpdateListener) {
+void SelectionDAG::ReplaceAllUsesWith(SDNode *From, const SDValue *To) {
if (From->getNumValues() == 1) // Handle the simple case efficiently.
- return ReplaceAllUsesWith(SDValue(From, 0), To[0], UpdateListener);
+ return ReplaceAllUsesWith(SDValue(From, 0), To[0]);
// Iterate over just the existing users of From. See the comments in
// the ReplaceAllUsesWith above.
SDNode::use_iterator UI = From->use_begin(), UE = From->use_end();
- RAUWUpdateListener Listener(UpdateListener, UI, UE);
+ RAUWUpdateListener Listener(*this, UI, UE);
while (UI != UE) {
SDNode *User = *UI;
// Now that we have modified User, add it back to the CSE maps. If it
// already exists there, recursively merge the results together.
- AddModifiedNodeToCSEMaps(User, &Listener);
+ AddModifiedNodeToCSEMaps(User);
}
// If we just RAUW'd the root, take note.
/// ReplaceAllUsesOfValueWith - Replace any uses of From with To, leaving
/// uses of other values produced by From.getNode() alone. The Deleted
/// vector is handled the same way as for ReplaceAllUsesWith.
-void SelectionDAG::ReplaceAllUsesOfValueWith(SDValue From, SDValue To,
- DAGUpdateListener *UpdateListener){
+void SelectionDAG::ReplaceAllUsesOfValueWith(SDValue From, SDValue To){
// Handle the really simple, really trivial case efficiently.
if (From == To) return;
// Handle the simple, trivial, case efficiently.
if (From.getNode()->getNumValues() == 1) {
- ReplaceAllUsesWith(From, To, UpdateListener);
+ ReplaceAllUsesWith(From, To);
return;
}
// the ReplaceAllUsesWith above.
SDNode::use_iterator UI = From.getNode()->use_begin(),
UE = From.getNode()->use_end();
- RAUWUpdateListener Listener(UpdateListener, UI, UE);
+ RAUWUpdateListener Listener(*this, UI, UE);
while (UI != UE) {
SDNode *User = *UI;
bool UserRemovedFromCSEMaps = false;
// Now that we have modified User, add it back to the CSE maps. If it
// already exists there, recursively merge the results together.
- AddModifiedNodeToCSEMaps(User, &Listener);
+ AddModifiedNodeToCSEMaps(User);
}
// If we just RAUW'd the root, take note.
/// handled the same way as for ReplaceAllUsesWith.
void SelectionDAG::ReplaceAllUsesOfValuesWith(const SDValue *From,
const SDValue *To,
- unsigned Num,
- DAGUpdateListener *UpdateListener){
+ unsigned Num){
// Handle the simple, trivial case efficiently.
if (Num == 1)
- return ReplaceAllUsesOfValueWith(*From, *To, UpdateListener);
+ return ReplaceAllUsesOfValueWith(*From, *To);
// Read up all the uses and make records of them. This helps
// processing new uses that are introduced during the
// Now that we have modified User, add it back to the CSE maps. If it
// already exists there, recursively merge the results together.
- AddModifiedNodeToCSEMaps(User, UpdateListener);
+ AddModifiedNodeToCSEMaps(User);
}
}
}
}
- // Visit all the nodes. As we iterate, moves nodes into sorted order,
+ // Visit all the nodes. As we iterate, move nodes into sorted order,
// such that by the time the end is reached all nodes will be sorted.
for (allnodes_iterator I = allnodes_begin(),E = allnodes_end(); I != E; ++I) {
SDNode *N = I;
int64_t GVOffset = 0;
if (TLI.isGAPlusOffset(Ptr.getNode(), GV, 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());
+ llvm::ComputeMaskedBits(const_cast<GlobalValue*>(GV), KnownZero, KnownOne,
+ TLI.getDataLayout());
unsigned AlignBits = KnownZero.countTrailingOnes();
unsigned Align = AlignBits ? 1 << std::min(31U, AlignBits) : 0;
if (Align)
projects / oota-llvm.git / blobdiff