//===----------------------------------------------------------------------===//
#include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/ADT/SetVector.h"
#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Triple.h"
#include "llvm/CodeGen/Analysis.h"
#include "llvm/Target/TargetFrameLowering.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
using namespace llvm;
+#define DEBUG_TYPE "legalizedag"
+
//===----------------------------------------------------------------------===//
-/// SelectionDAGLegalize - This takes an arbitrary SelectionDAG as input and
+/// This takes an arbitrary SelectionDAG as input and
/// hacks on it until the target machine can handle it. This involves
/// eliminating value sizes the machine cannot handle (promoting small sizes to
/// large sizes or splitting up large values into small values) as well as
/// will attempt merge setcc and brc instructions into brcc's.
///
namespace {
-class SelectionDAGLegalize : public SelectionDAG::DAGUpdateListener {
+class SelectionDAGLegalize {
const TargetMachine &TM;
const TargetLowering &TLI;
SelectionDAG &DAG;
- /// LegalizePosition - The iterator for walking through the node list.
- SelectionDAG::allnodes_iterator LegalizePosition;
+ /// \brief The set of nodes which have already been legalized. We hold a
+ /// reference to it in order to update as necessary on node deletion.
+ SmallPtrSetImpl<SDNode *> &LegalizedNodes;
- /// LegalizedNodes - The set of nodes which have already been legalized.
- SmallPtrSet<SDNode *, 16> LegalizedNodes;
+ /// \brief A set of all the nodes updated during legalization.
+ SmallSetVector<SDNode *, 16> *UpdatedNodes;
EVT getSetCCResultType(EVT VT) const {
return TLI.getSetCCResultType(*DAG.getContext(), VT);
// Libcall insertion helpers.
public:
- explicit SelectionDAGLegalize(SelectionDAG &DAG);
+ SelectionDAGLegalize(SelectionDAG &DAG,
+ SmallPtrSetImpl<SDNode *> &LegalizedNodes,
+ SmallSetVector<SDNode *, 16> *UpdatedNodes = nullptr)
+ : TM(DAG.getTarget()), TLI(DAG.getTargetLoweringInfo()), DAG(DAG),
+ LegalizedNodes(LegalizedNodes), UpdatedNodes(UpdatedNodes) {}
- void LegalizeDAG();
-
-private:
- /// LegalizeOp - Legalizes the given operation.
+ /// \brief Legalizes the given operation.
void LegalizeOp(SDNode *Node);
+private:
SDValue OptimizeFloatStore(StoreSDNode *ST);
void LegalizeLoadOps(SDNode *Node);
void LegalizeStoreOps(SDNode *Node);
- /// PerformInsertVectorEltInMemory - Some target cannot handle a variable
+ /// Some targets cannot handle a variable
/// insertion index for the INSERT_VECTOR_ELT instruction. In this case, it
/// is necessary to spill the vector being inserted into to memory, perform
/// the insert there, and then read the result back.
SDValue ExpandINSERT_VECTOR_ELT(SDValue Vec, SDValue Val,
SDValue Idx, SDLoc dl);
- /// ShuffleWithNarrowerEltType - Return a vector shuffle operation which
+ /// Return a vector shuffle operation which
/// performs the same shuffe in terms of order or result bytes, but on a type
/// whose vector element type is narrower than the original shuffle type.
/// e.g. <v4i32> <0, 1, 0, 1> -> v8i16 <0, 1, 2, 3, 0, 1, 2, 3>
void ExpandNode(SDNode *Node);
void PromoteNode(SDNode *Node);
- void ForgetNode(SDNode *N) {
- LegalizedNodes.erase(N);
- if (LegalizePosition == SelectionDAG::allnodes_iterator(N))
- ++LegalizePosition;
- }
-
public:
- // DAGUpdateListener implementation.
- void NodeDeleted(SDNode *N, SDNode *E) override {
- ForgetNode(N);
- }
- void NodeUpdated(SDNode *N) override {}
-
// Node replacement helpers
void ReplacedNode(SDNode *N) {
- if (N->use_empty()) {
- DAG.RemoveDeadNode(N);
- } else {
- ForgetNode(N);
- }
+ LegalizedNodes.erase(N);
+ if (UpdatedNodes)
+ UpdatedNodes->insert(N);
}
void ReplaceNode(SDNode *Old, SDNode *New) {
+ DEBUG(dbgs() << " ... replacing: "; Old->dump(&DAG);
+ dbgs() << " with: "; New->dump(&DAG));
+
+ assert(Old->getNumValues() == New->getNumValues() &&
+ "Replacing one node with another that produces a different number "
+ "of values!");
DAG.ReplaceAllUsesWith(Old, New);
+ for (unsigned i = 0, e = Old->getNumValues(); i != e; ++i)
+ DAG.TransferDbgValues(SDValue(Old, i), SDValue(New, i));
+ if (UpdatedNodes)
+ UpdatedNodes->insert(New);
ReplacedNode(Old);
}
void ReplaceNode(SDValue Old, SDValue New) {
+ DEBUG(dbgs() << " ... replacing: "; Old->dump(&DAG);
+ dbgs() << " with: "; New->dump(&DAG));
+
DAG.ReplaceAllUsesWith(Old, New);
+ DAG.TransferDbgValues(Old, New);
+ if (UpdatedNodes)
+ UpdatedNodes->insert(New.getNode());
ReplacedNode(Old.getNode());
}
void ReplaceNode(SDNode *Old, const SDValue *New) {
+ DEBUG(dbgs() << " ... replacing: "; Old->dump(&DAG));
+
DAG.ReplaceAllUsesWith(Old, New);
+ for (unsigned i = 0, e = Old->getNumValues(); i != e; ++i) {
+ DEBUG(dbgs() << (i == 0 ? " with: "
+ : " and: ");
+ New[i]->dump(&DAG));
+ DAG.TransferDbgValues(SDValue(Old, i), New[i]);
+ if (UpdatedNodes)
+ UpdatedNodes->insert(New[i].getNode());
+ }
ReplacedNode(Old);
}
};
}
-/// ShuffleWithNarrowerEltType - Return a vector shuffle operation which
+/// Return a vector shuffle operation which
/// performs the same shuffe in terms of order or result bytes, but on a type
/// whose vector element type is narrower than the original shuffle type.
/// e.g. <v4i32> <0, 1, 0, 1> -> v8i16 <0, 1, 2, 3, 0, 1, 2, 3>
return DAG.getVectorShuffle(NVT, dl, N1, N2, &NewMask[0]);
}
-SelectionDAGLegalize::SelectionDAGLegalize(SelectionDAG &dag)
- : SelectionDAG::DAGUpdateListener(dag),
- TM(dag.getTarget()), TLI(dag.getTargetLoweringInfo()),
- DAG(dag) {
-}
-
-void SelectionDAGLegalize::LegalizeDAG() {
- DAG.AssignTopologicalOrder();
-
- // Visit all the nodes. We start in topological order, so that we see
- // nodes with their original operands intact. Legalization can produce
- // new nodes which may themselves need to be legalized. Iterate until all
- // nodes have been legalized.
- for (;;) {
- bool AnyLegalized = false;
- for (LegalizePosition = DAG.allnodes_end();
- LegalizePosition != DAG.allnodes_begin(); ) {
- --LegalizePosition;
-
- SDNode *N = LegalizePosition;
- if (LegalizedNodes.insert(N)) {
- AnyLegalized = true;
- LegalizeOp(N);
- }
- }
- if (!AnyLegalized)
- break;
-
- }
-
- // Remove dead nodes now.
- DAG.RemoveDeadNodes();
-}
-
-/// ExpandConstantFP - Expands the ConstantFP node to an integer constant or
+/// Expands the ConstantFP node to an integer constant or
/// a load from the constant pool.
SDValue
SelectionDAGLegalize::ExpandConstantFP(ConstantFPSDNode *CFP, bool UseCP) {
EVT OrigVT = VT;
EVT SVT = VT;
- while (SVT != MVT::f32) {
+ while (SVT != MVT::f32 && SVT != MVT::f16) {
SVT = (MVT::SimpleValueType)(SVT.getSimpleVT().SimpleTy - 1);
if (ConstantFPSDNode::isValueValidForType(SVT, CFP->getValueAPF()) &&
// Only do this if the target has a native EXTLOAD instruction from
// smaller type.
- TLI.isLoadExtLegal(ISD::EXTLOAD, SVT) &&
+ TLI.isLoadExtLegal(ISD::EXTLOAD, OrigVT, SVT) &&
TLI.ShouldShrinkFPConstant(OrigVT)) {
Type *SType = SVT.getTypeForEVT(*DAG.getContext());
LLVMC = cast<ConstantFP>(ConstantExpr::getFPTrunc(LLVMC, SType));
DAG.getExtLoad(ISD::EXTLOAD, dl, OrigVT,
DAG.getEntryNode(),
CPIdx, MachinePointerInfo::getConstantPool(),
- VT, false, false, Alignment);
+ VT, false, false, false, Alignment);
return Result;
}
SDValue Result =
return Result;
}
-/// ExpandUnalignedStore - Expands an unaligned store to 2 half-size stores.
+/// Expands an unaligned store to 2 half-size stores.
static void ExpandUnalignedStore(StoreSDNode *ST, SelectionDAG &DAG,
const TargetLowering &TLI,
SelectionDAGLegalize *DAGLegalize) {
// Load from the stack slot.
SDValue Load = DAG.getExtLoad(ISD::EXTLOAD, dl, RegVT, Store, StackPtr,
MachinePointerInfo(),
- MemVT, false, false, 0);
+ MemVT, false, false, false, 0);
Stores.push_back(DAG.getTruncStore(Load.getValue(1), dl, Load, Ptr,
ST->getPointerInfo()
MemVT, ST->isVolatile(),
ST->isNonTemporal(),
MinAlign(ST->getAlignment(), Offset),
- ST->getTBAAInfo()));
+ ST->getAAInfo()));
// The order of the stores doesn't matter - say it with a TokenFactor.
- SDValue Result =
- DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &Stores[0],
- Stores.size());
+ SDValue Result = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Stores);
DAGLegalize->ReplaceNode(SDValue(ST, 0), Result);
return;
}
Store2 = DAG.getTruncStore(Chain, dl, TLI.isLittleEndian()?Hi:Lo, Ptr,
ST->getPointerInfo().getWithOffset(IncrementSize),
NewStoredVT, ST->isVolatile(), ST->isNonTemporal(),
- Alignment, ST->getTBAAInfo());
+ Alignment, ST->getAAInfo());
SDValue Result =
DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Store1, Store2);
DAGLegalize->ReplaceNode(SDValue(ST, 0), Result);
}
-/// ExpandUnalignedLoad - Expands an unaligned load to 2 half-size loads.
+/// Expands an unaligned load to 2 half-size loads.
static void
ExpandUnalignedLoad(LoadSDNode *LD, SelectionDAG &DAG,
const TargetLowering &TLI,
LD->isVolatile(), LD->isNonTemporal(),
LD->isInvariant(),
MinAlign(LD->getAlignment(), Offset),
- LD->getTBAAInfo());
+ LD->getAAInfo());
// Follow the load with a store to the stack slot. Remember the store.
Stores.push_back(DAG.getStore(Load.getValue(1), dl, Load, StackPtr,
MachinePointerInfo(), false, false, 0));
LD->getPointerInfo().getWithOffset(Offset),
MemVT, LD->isVolatile(),
LD->isNonTemporal(),
+ LD->isInvariant(),
MinAlign(LD->getAlignment(), Offset),
- LD->getTBAAInfo());
+ LD->getAAInfo());
// Follow the load with a store to the stack slot. Remember the store.
// On big-endian machines this requires a truncating store to ensure
// that the bits end up in the right place.
false, false, 0));
// The order of the stores doesn't matter - say it with a TokenFactor.
- SDValue TF = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &Stores[0],
- Stores.size());
+ SDValue TF = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Stores);
// Finally, perform the original load only redirected to the stack slot.
Load = DAG.getExtLoad(LD->getExtensionType(), dl, VT, TF, StackBase,
- MachinePointerInfo(), LoadedVT, false, false, 0);
+ MachinePointerInfo(), LoadedVT, false,false, false,
+ 0);
// Callers expect a MERGE_VALUES node.
ValResult = Load;
if (TLI.isLittleEndian()) {
Lo = DAG.getExtLoad(ISD::ZEXTLOAD, dl, VT, Chain, Ptr, LD->getPointerInfo(),
NewLoadedVT, LD->isVolatile(),
- LD->isNonTemporal(), Alignment, LD->getTBAAInfo());
+ LD->isNonTemporal(), LD->isInvariant(), Alignment,
+ LD->getAAInfo());
Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr,
DAG.getConstant(IncrementSize, Ptr.getValueType()));
Hi = DAG.getExtLoad(HiExtType, dl, VT, Chain, Ptr,
LD->getPointerInfo().getWithOffset(IncrementSize),
NewLoadedVT, LD->isVolatile(),
- LD->isNonTemporal(), MinAlign(Alignment, IncrementSize),
- LD->getTBAAInfo());
+ LD->isNonTemporal(),LD->isInvariant(),
+ MinAlign(Alignment, IncrementSize), LD->getAAInfo());
} else {
Hi = DAG.getExtLoad(HiExtType, dl, VT, Chain, Ptr, LD->getPointerInfo(),
NewLoadedVT, LD->isVolatile(),
- LD->isNonTemporal(), Alignment, LD->getTBAAInfo());
+ LD->isNonTemporal(), LD->isInvariant(), Alignment,
+ LD->getAAInfo());
Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr,
DAG.getConstant(IncrementSize, Ptr.getValueType()));
Lo = DAG.getExtLoad(ISD::ZEXTLOAD, dl, VT, Chain, Ptr,
LD->getPointerInfo().getWithOffset(IncrementSize),
NewLoadedVT, LD->isVolatile(),
- LD->isNonTemporal(), MinAlign(Alignment, IncrementSize),
- LD->getTBAAInfo());
+ LD->isNonTemporal(), LD->isInvariant(),
+ MinAlign(Alignment, IncrementSize), LD->getAAInfo());
}
// aggregate the two parts
ChainResult = TF;
}
-/// PerformInsertVectorEltInMemory - Some target cannot handle a variable
-/// insertion index for the INSERT_VECTOR_ELT instruction. In this case, it
+/// Some target cannot handle a variable insertion index for the
+/// INSERT_VECTOR_ELT instruction. In this case, it
/// is necessary to spill the vector being inserted into to memory, perform
/// the insert there, and then read the result back.
SDValue SelectionDAGLegalize::
unsigned Alignment = ST->getAlignment();
bool isVolatile = ST->isVolatile();
bool isNonTemporal = ST->isNonTemporal();
- const MDNode *TBAAInfo = ST->getTBAAInfo();
+ AAMDNodes AAInfo = ST->getAAInfo();
SDLoc dl(ST);
if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(ST->getValue())) {
if (CFP->getValueType(0) == MVT::f32 &&
bitcastToAPInt().zextOrTrunc(32),
MVT::i32);
return DAG.getStore(Chain, dl, Con, Ptr, ST->getPointerInfo(),
- isVolatile, isNonTemporal, Alignment, TBAAInfo);
+ isVolatile, isNonTemporal, Alignment, AAInfo);
}
if (CFP->getValueType(0) == MVT::f64) {
SDValue Con = DAG.getConstant(CFP->getValueAPF().bitcastToAPInt().
zextOrTrunc(64), MVT::i64);
return DAG.getStore(Chain, dl, Con, Ptr, ST->getPointerInfo(),
- isVolatile, isNonTemporal, Alignment, TBAAInfo);
+ isVolatile, isNonTemporal, Alignment, AAInfo);
}
if (TLI.isTypeLegal(MVT::i32) && !ST->isVolatile()) {
if (TLI.isBigEndian()) std::swap(Lo, Hi);
Lo = DAG.getStore(Chain, dl, Lo, Ptr, ST->getPointerInfo(), isVolatile,
- isNonTemporal, Alignment, TBAAInfo);
+ isNonTemporal, Alignment, AAInfo);
Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr,
DAG.getConstant(4, Ptr.getValueType()));
Hi = DAG.getStore(Chain, dl, Hi, Ptr,
ST->getPointerInfo().getWithOffset(4),
isVolatile, isNonTemporal, MinAlign(Alignment, 4U),
- TBAAInfo);
+ AAInfo);
return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Lo, Hi);
}
}
}
- return SDValue(0, 0);
+ return SDValue(nullptr, 0);
}
void SelectionDAGLegalize::LegalizeStoreOps(SDNode *Node) {
unsigned Alignment = ST->getAlignment();
bool isVolatile = ST->isVolatile();
bool isNonTemporal = ST->isNonTemporal();
- const MDNode *TBAAInfo = ST->getTBAAInfo();
+ AAMDNodes AAInfo = ST->getAAInfo();
if (!ST->isTruncatingStore()) {
if (SDNode *OptStore = OptimizeFloatStore(ST).getNode()) {
// If this is an unaligned store and the target doesn't support it,
// expand it.
unsigned AS = ST->getAddressSpace();
- if (!TLI.allowsUnalignedMemoryAccesses(ST->getMemoryVT(), AS)) {
+ unsigned Align = ST->getAlignment();
+ if (!TLI.allowsMisalignedMemoryAccesses(ST->getMemoryVT(), AS, Align)) {
Type *Ty = ST->getMemoryVT().getTypeForEVT(*DAG.getContext());
unsigned ABIAlignment= TLI.getDataLayout()->getABITypeAlignment(Ty);
- if (ST->getAlignment() < ABIAlignment)
- ExpandUnalignedStore(cast<StoreSDNode>(Node),
- DAG, TLI, this);
+ if (Align < ABIAlignment)
+ ExpandUnalignedStore(cast<StoreSDNode>(Node), DAG, TLI, this);
}
break;
}
case TargetLowering::Custom: {
SDValue Res = TLI.LowerOperation(SDValue(Node, 0), DAG);
- if (Res.getNode())
+ if (Res && Res != SDValue(Node, 0))
ReplaceNode(SDValue(Node, 0), Res);
return;
}
SDValue Result =
DAG.getStore(Chain, dl, Value, Ptr,
ST->getPointerInfo(), isVolatile,
- isNonTemporal, Alignment, TBAAInfo);
+ isNonTemporal, Alignment, AAInfo);
ReplaceNode(SDValue(Node, 0), Result);
break;
}
Value = DAG.getZeroExtendInReg(Value, dl, StVT);
SDValue Result =
DAG.getTruncStore(Chain, dl, Value, Ptr, ST->getPointerInfo(),
- NVT, isVolatile, isNonTemporal, Alignment,
- TBAAInfo);
+ NVT, isVolatile, isNonTemporal, Alignment, AAInfo);
ReplaceNode(SDValue(Node, 0), Result);
} else if (StWidth & (StWidth - 1)) {
// If not storing a power-of-2 number of bits, expand as two stores.
Lo = DAG.getTruncStore(Chain, dl, Value, Ptr, ST->getPointerInfo(),
RoundVT,
isVolatile, isNonTemporal, Alignment,
- TBAAInfo);
+ AAInfo);
// Store the remaining ExtraWidth bits.
IncrementSize = RoundWidth / 8;
Hi = DAG.getTruncStore(Chain, dl, Hi, Ptr,
ST->getPointerInfo().getWithOffset(IncrementSize),
ExtraVT, isVolatile, isNonTemporal,
- MinAlign(Alignment, IncrementSize), TBAAInfo);
+ MinAlign(Alignment, IncrementSize), AAInfo);
} else {
// Big endian - avoid unaligned stores.
// TRUNCSTORE:i24 X -> TRUNCSTORE:i16 (srl X, 8), TRUNCSTORE@+2:i8 X
TLI.getShiftAmountTy(Value.getValueType())));
Hi = DAG.getTruncStore(Chain, dl, Hi, Ptr, ST->getPointerInfo(),
RoundVT, isVolatile, isNonTemporal, Alignment,
- TBAAInfo);
+ AAInfo);
// Store the remaining ExtraWidth bits.
IncrementSize = RoundWidth / 8;
Lo = DAG.getTruncStore(Chain, dl, Value, Ptr,
ST->getPointerInfo().getWithOffset(IncrementSize),
ExtraVT, isVolatile, isNonTemporal,
- MinAlign(Alignment, IncrementSize), TBAAInfo);
+ MinAlign(Alignment, IncrementSize), AAInfo);
}
// The order of the stores doesn't matter.
default: llvm_unreachable("This action is not supported yet!");
case TargetLowering::Legal: {
unsigned AS = ST->getAddressSpace();
+ unsigned Align = ST->getAlignment();
// If this is an unaligned store and the target doesn't support it,
// expand it.
- if (!TLI.allowsUnalignedMemoryAccesses(ST->getMemoryVT(), AS)) {
+ if (!TLI.allowsMisalignedMemoryAccesses(ST->getMemoryVT(), AS, Align)) {
Type *Ty = ST->getMemoryVT().getTypeForEVT(*DAG.getContext());
unsigned ABIAlignment= TLI.getDataLayout()->getABITypeAlignment(Ty);
- if (ST->getAlignment() < ABIAlignment)
+ if (Align < ABIAlignment)
ExpandUnalignedStore(cast<StoreSDNode>(Node), DAG, TLI, this);
}
break;
}
case TargetLowering::Custom: {
SDValue Res = TLI.LowerOperation(SDValue(Node, 0), DAG);
- if (Res.getNode())
+ if (Res && Res != SDValue(Node, 0))
ReplaceNode(SDValue(Node, 0), Res);
return;
}
Value = DAG.getNode(ISD::TRUNCATE, dl, StVT, Value);
SDValue Result =
DAG.getStore(Chain, dl, Value, Ptr, ST->getPointerInfo(),
- isVolatile, isNonTemporal, Alignment, TBAAInfo);
+ isVolatile, isNonTemporal, Alignment, AAInfo);
ReplaceNode(SDValue(Node, 0), Result);
break;
}
default: llvm_unreachable("This action is not supported yet!");
case TargetLowering::Legal: {
unsigned AS = LD->getAddressSpace();
+ unsigned Align = LD->getAlignment();
// If this is an unaligned load and the target doesn't support it,
// expand it.
- if (!TLI.allowsUnalignedMemoryAccesses(LD->getMemoryVT(), AS)) {
+ if (!TLI.allowsMisalignedMemoryAccesses(LD->getMemoryVT(), AS, Align)) {
Type *Ty = LD->getMemoryVT().getTypeForEVT(*DAG.getContext());
unsigned ABIAlignment =
TLI.getDataLayout()->getABITypeAlignment(Ty);
- if (LD->getAlignment() < ABIAlignment){
+ if (Align < ABIAlignment){
ExpandUnalignedLoad(cast<LoadSDNode>(Node), DAG, TLI, RVal, RChain);
}
}
assert(RVal.getNode() != Node && "Load must be completely replaced");
DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 0), RVal);
DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 1), RChain);
+ if (UpdatedNodes) {
+ UpdatedNodes->insert(RVal.getNode());
+ UpdatedNodes->insert(RChain.getNode());
+ }
ReplacedNode(Node);
}
return;
unsigned Alignment = LD->getAlignment();
bool isVolatile = LD->isVolatile();
bool isNonTemporal = LD->isNonTemporal();
- const MDNode *TBAAInfo = LD->getTBAAInfo();
+ bool isInvariant = LD->isInvariant();
+ AAMDNodes AAInfo = LD->getAAInfo();
if (SrcWidth != SrcVT.getStoreSizeInBits() &&
// Some targets pretend to have an i1 loading operation, and actually
// nice to have an effective generic way of getting these benefits...
// Until such a way is found, don't insist on promoting i1 here.
(SrcVT != MVT::i1 ||
- TLI.getLoadExtAction(ExtType, MVT::i1) == TargetLowering::Promote)) {
+ TLI.getLoadExtAction(ExtType, Node->getValueType(0), MVT::i1) ==
+ TargetLowering::Promote)) {
// Promote to a byte-sized load if not loading an integral number of
// bytes. For example, promote EXTLOAD:i20 -> EXTLOAD:i24.
unsigned NewWidth = SrcVT.getStoreSizeInBits();
SDValue Result =
DAG.getExtLoad(NewExtType, dl, Node->getValueType(0),
Chain, Ptr, LD->getPointerInfo(),
- NVT, isVolatile, isNonTemporal, Alignment, TBAAInfo);
+ NVT, isVolatile, isNonTemporal, isInvariant, Alignment,
+ AAInfo);
Ch = Result.getValue(1); // The chain.
Lo = DAG.getExtLoad(ISD::ZEXTLOAD, dl, Node->getValueType(0),
Chain, Ptr,
LD->getPointerInfo(), RoundVT, isVolatile,
- isNonTemporal, Alignment, TBAAInfo);
+ isNonTemporal, isInvariant, Alignment, AAInfo);
// Load the remaining ExtraWidth bits.
IncrementSize = RoundWidth / 8;
DAG.getConstant(IncrementSize, Ptr.getValueType()));
Hi = DAG.getExtLoad(ExtType, dl, Node->getValueType(0), Chain, Ptr,
LD->getPointerInfo().getWithOffset(IncrementSize),
- ExtraVT, isVolatile, isNonTemporal,
- MinAlign(Alignment, IncrementSize), TBAAInfo);
+ ExtraVT, isVolatile, isNonTemporal, isInvariant,
+ MinAlign(Alignment, IncrementSize), AAInfo);
// Build a factor node to remember that this load is independent of
// the other one.
// Load the top RoundWidth bits.
Hi = DAG.getExtLoad(ExtType, dl, Node->getValueType(0), Chain, Ptr,
LD->getPointerInfo(), RoundVT, isVolatile,
- isNonTemporal, Alignment, TBAAInfo);
+ isNonTemporal, isInvariant, Alignment, AAInfo);
// Load the remaining ExtraWidth bits.
IncrementSize = RoundWidth / 8;
Lo = DAG.getExtLoad(ISD::ZEXTLOAD,
dl, Node->getValueType(0), Chain, Ptr,
LD->getPointerInfo().getWithOffset(IncrementSize),
- ExtraVT, isVolatile, isNonTemporal,
- MinAlign(Alignment, IncrementSize), TBAAInfo);
+ ExtraVT, isVolatile, isNonTemporal, isInvariant,
+ MinAlign(Alignment, IncrementSize), AAInfo);
// Build a factor node to remember that this load is independent of
// the other one.
Chain = Ch;
} else {
bool isCustom = false;
- switch (TLI.getLoadExtAction(ExtType, SrcVT.getSimpleVT())) {
+ switch (TLI.getLoadExtAction(ExtType, Node->getValueType(0),
+ SrcVT.getSimpleVT())) {
default: llvm_unreachable("This action is not supported yet!");
case TargetLowering::Custom:
isCustom = true;
// it, expand it.
EVT MemVT = LD->getMemoryVT();
unsigned AS = LD->getAddressSpace();
- if (!TLI.allowsUnalignedMemoryAccesses(MemVT, AS)) {
- Type *Ty =
- LD->getMemoryVT().getTypeForEVT(*DAG.getContext());
- unsigned ABIAlignment =
- TLI.getDataLayout()->getABITypeAlignment(Ty);
- if (LD->getAlignment() < ABIAlignment){
- ExpandUnalignedLoad(cast<LoadSDNode>(Node),
- DAG, TLI, Value, Chain);
+ unsigned Align = LD->getAlignment();
+ if (!TLI.allowsMisalignedMemoryAccesses(MemVT, AS, Align)) {
+ Type *Ty = LD->getMemoryVT().getTypeForEVT(*DAG.getContext());
+ unsigned ABIAlignment = TLI.getDataLayout()->getABITypeAlignment(Ty);
+ if (Align < ABIAlignment){
+ ExpandUnalignedLoad(cast<LoadSDNode>(Node), DAG, TLI, Value, Chain);
}
}
}
break;
}
case TargetLowering::Expand:
- if (!TLI.isLoadExtLegal(ISD::EXTLOAD, SrcVT) &&
- TLI.isTypeLegal(SrcVT)) {
- SDValue Load = DAG.getLoad(SrcVT, dl, Chain, Ptr,
- LD->getMemOperand());
- unsigned ExtendOp;
- switch (ExtType) {
- case ISD::EXTLOAD:
- ExtendOp = (SrcVT.isFloatingPoint() ?
- ISD::FP_EXTEND : ISD::ANY_EXTEND);
+ if (!TLI.isLoadExtLegal(ISD::EXTLOAD, Node->getValueType(0), SrcVT)) {
+ // If the source type is not legal, see if there is a legal extload to
+ // an intermediate type that we can then extend further.
+ EVT LoadVT = TLI.getRegisterType(SrcVT.getSimpleVT());
+ if (TLI.isTypeLegal(SrcVT) || // Same as SrcVT == LoadVT?
+ TLI.isLoadExtLegal(ExtType, LoadVT, SrcVT)) {
+ // If we are loading a legal type, this is a non-extload followed by a
+ // full extend.
+ ISD::LoadExtType MidExtType =
+ (LoadVT == SrcVT) ? ISD::NON_EXTLOAD : ExtType;
+
+ SDValue Load = DAG.getExtLoad(MidExtType, dl, LoadVT, Chain, Ptr,
+ SrcVT, LD->getMemOperand());
+ unsigned ExtendOp =
+ ISD::getExtForLoadExtType(SrcVT.isFloatingPoint(), ExtType);
+ Value = DAG.getNode(ExtendOp, dl, Node->getValueType(0), Load);
+ Chain = Load.getValue(1);
break;
- case ISD::SEXTLOAD: ExtendOp = ISD::SIGN_EXTEND; break;
- case ISD::ZEXTLOAD: ExtendOp = ISD::ZERO_EXTEND; break;
- default: llvm_unreachable("Unexpected extend load type!");
}
- Value = DAG.getNode(ExtendOp, dl, Node->getValueType(0), Load);
- Chain = Load.getValue(1);
- break;
}
assert(!SrcVT.isVector() &&
Result.getValueType(),
Result, DAG.getValueType(SrcVT));
else
- ValRes = DAG.getZeroExtendInReg(Result, dl,
- SrcVT.getScalarType());
+ ValRes = DAG.getZeroExtendInReg(Result, dl, SrcVT.getScalarType());
Value = ValRes;
Chain = Result.getValue(1);
break;
assert(Value.getNode() != Node && "Load must be completely replaced");
DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 0), Value);
DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 1), Chain);
+ if (UpdatedNodes) {
+ UpdatedNodes->insert(Value.getNode());
+ UpdatedNodes->insert(Chain.getNode());
+ }
ReplacedNode(Node);
}
}
-/// LegalizeOp - Return a legal replacement for the given operation, with
-/// all legal operands.
+/// Return a legal replacement for the given operation, with all legal operands.
void SelectionDAGLegalize::LegalizeOp(SDNode *Node) {
+ DEBUG(dbgs() << "\nLegalizing: "; Node->dump(&DAG));
+
if (Node->getOpcode() == ISD::TargetConstant) // Allow illegal target nodes.
return;
if (Action != TargetLowering::Promote)
Action = TLI.getOperationAction(Node->getOpcode(), MVT::Other);
break;
+ case ISD::FP_TO_FP16:
case ISD::SINT_TO_FP:
case ISD::UINT_TO_FP:
case ISD::EXTRACT_VECTOR_ELT:
if (Action == TargetLowering::Legal)
Action = TargetLowering::Custom;
break;
+ case ISD::READ_REGISTER:
+ case ISD::WRITE_REGISTER:
+ // Named register is legal in the DAG, but blocked by register name
+ // selection if not implemented by target (to chose the correct register)
+ // They'll be converted to Copy(To/From)Reg.
+ Action = TargetLowering::Legal;
+ break;
case ISD::DEBUGTRAP:
Action = TLI.getOperationAction(Node->getOpcode(), Node->getValueType(0));
if (Action == TargetLowering::Expand) {
}
if (NewNode != Node) {
- DAG.ReplaceAllUsesWith(Node, NewNode);
- for (unsigned i = 0, e = Node->getNumValues(); i != e; ++i)
- DAG.TransferDbgValues(SDValue(Node, i), SDValue(NewNode, i));
- ReplacedNode(Node);
+ ReplaceNode(Node, NewNode);
Node = NewNode;
}
switch (Action) {
// a complete mess.
SDValue Res = TLI.LowerOperation(SDValue(Node, 0), DAG);
if (Res.getNode()) {
- SmallVector<SDValue, 8> ResultVals;
- for (unsigned i = 0, e = Node->getNumValues(); i != e; ++i) {
- if (e == 1)
- ResultVals.push_back(Res);
- else
- ResultVals.push_back(Res.getValue(i));
- }
- if (Res.getNode() != Node || Res.getResNo() != 0) {
- DAG.ReplaceAllUsesWith(Node, ResultVals.data());
- for (unsigned i = 0, e = Node->getNumValues(); i != e; ++i)
- DAG.TransferDbgValues(SDValue(Node, i), ResultVals[i]);
- ReplacedNode(Node);
+ if (!(Res.getNode() != Node || Res.getResNo() != 0))
+ return;
+
+ if (Node->getNumValues() == 1) {
+ // We can just directly replace this node with the lowered value.
+ ReplaceNode(SDValue(Node, 0), Res);
+ return;
}
+
+ SmallVector<SDValue, 8> ResultVals;
+ for (unsigned i = 0, e = Node->getNumValues(); i != e; ++i)
+ ResultVals.push_back(Res.getValue(i));
+ ReplaceNode(Node, ResultVals.data());
return;
}
}
SDValue Vec = Op.getOperand(0);
SDValue Idx = Op.getOperand(1);
SDLoc dl(Op);
- // Store the value to a temporary stack slot, then LOAD the returned part.
- SDValue StackPtr = DAG.CreateStackTemporary(Vec.getValueType());
- SDValue Ch = DAG.getStore(DAG.getEntryNode(), dl, Vec, StackPtr,
- MachinePointerInfo(), false, false, 0);
+
+ // Before we generate a new store to a temporary stack slot, see if there is
+ // already one that we can use. There often is because when we scalarize
+ // vector operations (using SelectionDAG::UnrollVectorOp for example) a whole
+ // series of EXTRACT_VECTOR_ELT nodes are generated, one for each element in
+ // the vector. If all are expanded here, we don't want one store per vector
+ // element.
+ SDValue StackPtr, Ch;
+ for (SDNode::use_iterator UI = Vec.getNode()->use_begin(),
+ UE = Vec.getNode()->use_end(); UI != UE; ++UI) {
+ SDNode *User = *UI;
+ if (StoreSDNode *ST = dyn_cast<StoreSDNode>(User)) {
+ if (ST->isIndexed() || ST->isTruncatingStore() ||
+ ST->getValue() != Vec)
+ continue;
+
+ // Make sure that nothing else could have stored into the destination of
+ // this store.
+ if (!ST->getChain().reachesChainWithoutSideEffects(DAG.getEntryNode()))
+ continue;
+
+ StackPtr = ST->getBasePtr();
+ Ch = SDValue(ST, 0);
+ break;
+ }
+ }
+
+ if (!Ch.getNode()) {
+ // Store the value to a temporary stack slot, then LOAD the returned part.
+ StackPtr = DAG.CreateStackTemporary(Vec.getValueType());
+ Ch = DAG.getStore(DAG.getEntryNode(), dl, Vec, StackPtr,
+ MachinePointerInfo(), false, false, 0);
+ }
// Add the offset to the index.
unsigned EltSize =
Idx = DAG.getZExtOrTrunc(Idx, dl, TLI.getPointerTy());
StackPtr = DAG.getNode(ISD::ADD, dl, Idx.getValueType(), Idx, StackPtr);
+ SDValue NewLoad;
+
if (Op.getValueType().isVector())
- return DAG.getLoad(Op.getValueType(), dl, Ch, StackPtr,MachinePointerInfo(),
- false, false, false, 0);
- return DAG.getExtLoad(ISD::EXTLOAD, dl, Op.getValueType(), Ch, StackPtr,
- MachinePointerInfo(),
- Vec.getValueType().getVectorElementType(),
- false, false, 0);
+ NewLoad = DAG.getLoad(Op.getValueType(), dl, Ch, StackPtr,
+ MachinePointerInfo(), false, false, false, 0);
+ else
+ NewLoad = DAG.getExtLoad(
+ ISD::EXTLOAD, dl, Op.getValueType(), Ch, StackPtr, MachinePointerInfo(),
+ Vec.getValueType().getVectorElementType(), false, false, false, 0);
+
+ // Replace the chain going out of the store, by the one out of the load.
+ DAG.ReplaceAllUsesOfValueWith(Ch, SDValue(NewLoad.getNode(), 1));
+
+ // We introduced a cycle though, so update the loads operands, making sure
+ // to use the original store's chain as an incoming chain.
+ SmallVector<SDValue, 6> NewLoadOperands(NewLoad->op_begin(),
+ NewLoad->op_end());
+ NewLoadOperands[0] = Ch;
+ NewLoad =
+ SDValue(DAG.UpdateNodeOperands(NewLoad.getNode(), NewLoadOperands), 0);
+ return NewLoad;
}
SDValue SelectionDAGLegalize::ExpandInsertToVectorThroughStack(SDValue Op) {
StackPtr);
// Store the subvector.
- Ch = DAG.getStore(DAG.getEntryNode(), dl, Part, SubStackPtr,
+ Ch = DAG.getStore(Ch, dl, Part, SubStackPtr,
MachinePointerInfo(), false, false, 0);
// Finally, load the updated vector.
SDValue StoreChain;
if (!Stores.empty()) // Not all undef elements?
- StoreChain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
- &Stores[0], Stores.size());
+ StoreChain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Stores);
else
StoreChain = DAG.getEntryNode();
SDValue SP = DAG.getCopyFromReg(Chain, dl, SPReg, VT);
Chain = SP.getValue(1);
unsigned Align = cast<ConstantSDNode>(Tmp3)->getZExtValue();
- unsigned StackAlign = TM.getFrameLowering()->getStackAlignment();
+ unsigned StackAlign =
+ DAG.getSubtarget().getFrameLowering()->getStackAlignment();
Tmp1 = DAG.getNode(ISD::SUB, dl, VT, SP, Size); // Value
if (Align > StackAlign)
Tmp1 = DAG.getNode(ISD::AND, dl, VT, Tmp1,
Results.push_back(Tmp2);
}
-/// LegalizeSetCCCondCode - Legalize a SETCC with given LHS and RHS and
-/// condition code CC on the current target.
+/// Legalize a SETCC with given LHS and RHS and condition code CC on the current
+/// target.
///
/// If the SETCC has been legalized using AND / OR, then the legalized node
/// will be stored in LHS. RHS and CC will be set to SDValue(). NeedInvert
/// If the SETCC has been legalized using the inverse condcode, then LHS and
/// RHS will be unchanged, CC will set to the inverted condcode, and NeedInvert
/// will be set to true. The caller must invert the result of the SETCC with
-/// SelectionDAG::getNOT() or take equivalent action to swap the effect of a
-/// true/false result.
+/// SelectionDAG::getLogicalNOT() or take equivalent action to swap the effect
+/// of a true/false result.
///
/// \returns true if the SetCC has been legalized, false if it hasn't.
bool SelectionDAGLegalize::LegalizeSetCCCondCode(EVT VT,
return false;
}
-/// EmitStackConvert - Emit a store/load combination to the stack. This stores
+/// Emit a store/load combination to the stack. This stores
/// SrcOp to a stack slot of type SlotVT, truncating it if needed. It then does
/// a load from the stack slot to DestVT, extending it if needed.
/// The resultant code need not be legal.
assert(SlotSize < DestSize && "Unknown extension!");
return DAG.getExtLoad(ISD::EXTLOAD, dl, DestVT, Store, FIPtr,
- PtrInfo, SlotVT, false, false, DestAlign);
+ PtrInfo, SlotVT, false, false, false, DestAlign);
}
SDValue SelectionDAGLegalize::ExpandSCALAR_TO_VECTOR(SDNode *Node) {
false, false, false, 0);
}
+static bool
+ExpandBVWithShuffles(SDNode *Node, SelectionDAG &DAG,
+ const TargetLowering &TLI, SDValue &Res) {
+ unsigned NumElems = Node->getNumOperands();
+ SDLoc dl(Node);
+ EVT VT = Node->getValueType(0);
+
+ // Try to group the scalars into pairs, shuffle the pairs together, then
+ // shuffle the pairs of pairs together, etc. until the vector has
+ // been built. This will work only if all of the necessary shuffle masks
+ // are legal.
+
+ // We do this in two phases; first to check the legality of the shuffles,
+ // and next, assuming that all shuffles are legal, to create the new nodes.
+ for (int Phase = 0; Phase < 2; ++Phase) {
+ SmallVector<std::pair<SDValue, SmallVector<int, 16> >, 16> IntermedVals,
+ NewIntermedVals;
+ for (unsigned i = 0; i < NumElems; ++i) {
+ SDValue V = Node->getOperand(i);
+ if (V.getOpcode() == ISD::UNDEF)
+ continue;
+
+ SDValue Vec;
+ if (Phase)
+ Vec = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, V);
+ IntermedVals.push_back(std::make_pair(Vec, SmallVector<int, 16>(1, i)));
+ }
+
+ while (IntermedVals.size() > 2) {
+ NewIntermedVals.clear();
+ for (unsigned i = 0, e = (IntermedVals.size() & ~1u); i < e; i += 2) {
+ // This vector and the next vector are shuffled together (simply to
+ // append the one to the other).
+ SmallVector<int, 16> ShuffleVec(NumElems, -1);
+
+ SmallVector<int, 16> FinalIndices;
+ FinalIndices.reserve(IntermedVals[i].second.size() +
+ IntermedVals[i+1].second.size());
+
+ int k = 0;
+ for (unsigned j = 0, f = IntermedVals[i].second.size(); j != f;
+ ++j, ++k) {
+ ShuffleVec[k] = j;
+ FinalIndices.push_back(IntermedVals[i].second[j]);
+ }
+ for (unsigned j = 0, f = IntermedVals[i+1].second.size(); j != f;
+ ++j, ++k) {
+ ShuffleVec[k] = NumElems + j;
+ FinalIndices.push_back(IntermedVals[i+1].second[j]);
+ }
+
+ SDValue Shuffle;
+ if (Phase)
+ Shuffle = DAG.getVectorShuffle(VT, dl, IntermedVals[i].first,
+ IntermedVals[i+1].first,
+ ShuffleVec.data());
+ else if (!TLI.isShuffleMaskLegal(ShuffleVec, VT))
+ return false;
+ NewIntermedVals.push_back(
+ std::make_pair(Shuffle, std::move(FinalIndices)));
+ }
+
+ // If we had an odd number of defined values, then append the last
+ // element to the array of new vectors.
+ if ((IntermedVals.size() & 1) != 0)
+ NewIntermedVals.push_back(IntermedVals.back());
+
+ IntermedVals.swap(NewIntermedVals);
+ }
+
+ assert(IntermedVals.size() <= 2 && IntermedVals.size() > 0 &&
+ "Invalid number of intermediate vectors");
+ SDValue Vec1 = IntermedVals[0].first;
+ SDValue Vec2;
+ if (IntermedVals.size() > 1)
+ Vec2 = IntermedVals[1].first;
+ else if (Phase)
+ Vec2 = DAG.getUNDEF(VT);
+
+ SmallVector<int, 16> ShuffleVec(NumElems, -1);
+ for (unsigned i = 0, e = IntermedVals[0].second.size(); i != e; ++i)
+ ShuffleVec[IntermedVals[0].second[i]] = i;
+ for (unsigned i = 0, e = IntermedVals[1].second.size(); i != e; ++i)
+ ShuffleVec[IntermedVals[1].second[i]] = NumElems + i;
+
+ if (Phase)
+ Res = DAG.getVectorShuffle(VT, dl, Vec1, Vec2, ShuffleVec.data());
+ else if (!TLI.isShuffleMaskLegal(ShuffleVec, VT))
+ return false;
+ }
+
+ return true;
+}
-/// ExpandBUILD_VECTOR - Expand a BUILD_VECTOR node on targets that don't
+/// Expand a BUILD_VECTOR node on targets that don't
/// support the operation, but do support the resultant vector type.
SDValue SelectionDAGLegalize::ExpandBUILD_VECTOR(SDNode *Node) {
unsigned NumElems = Node->getNumOperands();
false, false, false, Alignment);
}
- if (!MoreThanTwoValues) {
- SmallVector<int, 8> ShuffleVec(NumElems, -1);
- for (unsigned i = 0; i < NumElems; ++i) {
- SDValue V = Node->getOperand(i);
- if (V.getOpcode() == ISD::UNDEF)
- continue;
- ShuffleVec[i] = V == Value1 ? 0 : NumElems;
- }
- if (TLI.isShuffleMaskLegal(ShuffleVec, Node->getValueType(0))) {
- // Get the splatted value into the low element of a vector register.
- SDValue Vec1 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Value1);
- SDValue Vec2;
- if (Value2.getNode())
- Vec2 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Value2);
- else
- Vec2 = DAG.getUNDEF(VT);
+ SmallSet<SDValue, 16> DefinedValues;
+ for (unsigned i = 0; i < NumElems; ++i) {
+ if (Node->getOperand(i).getOpcode() == ISD::UNDEF)
+ continue;
+ DefinedValues.insert(Node->getOperand(i));
+ }
- // Return shuffle(LowValVec, undef, <0,0,0,0>)
- return DAG.getVectorShuffle(VT, dl, Vec1, Vec2, ShuffleVec.data());
+ if (TLI.shouldExpandBuildVectorWithShuffles(VT, DefinedValues.size())) {
+ if (!MoreThanTwoValues) {
+ SmallVector<int, 8> ShuffleVec(NumElems, -1);
+ for (unsigned i = 0; i < NumElems; ++i) {
+ SDValue V = Node->getOperand(i);
+ if (V.getOpcode() == ISD::UNDEF)
+ continue;
+ ShuffleVec[i] = V == Value1 ? 0 : NumElems;
+ }
+ if (TLI.isShuffleMaskLegal(ShuffleVec, Node->getValueType(0))) {
+ // Get the splatted value into the low element of a vector register.
+ SDValue Vec1 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Value1);
+ SDValue Vec2;
+ if (Value2.getNode())
+ Vec2 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Value2);
+ else
+ Vec2 = DAG.getUNDEF(VT);
+
+ // Return shuffle(LowValVec, undef, <0,0,0,0>)
+ return DAG.getVectorShuffle(VT, dl, Vec1, Vec2, ShuffleVec.data());
+ }
+ } else {
+ SDValue Res;
+ if (ExpandBVWithShuffles(Node, DAG, TLI, Res))
+ return Res;
}
}
return ExpandVectorBuildThroughStack(Node);
}
-// ExpandLibCall - Expand a node into a call to a libcall. If the result value
+// Expand a node into a call to a libcall. If the result value
// does not fit into a register, return the lo part and set the hi part to the
// by-reg argument. If it does fit into a single register, return the result
// and leave the Hi part unset.
if (isTailCall)
InChain = TCChain;
- TargetLowering::
- CallLoweringInfo CLI(InChain, RetTy, isSigned, !isSigned, false, false,
- 0, TLI.getLibcallCallingConv(LC), isTailCall,
- /*doesNotReturn=*/false, /*isReturnValueUsed=*/true,
- Callee, Args, DAG, SDLoc(Node));
- std::pair<SDValue, SDValue> CallInfo = TLI.LowerCallTo(CLI);
+ TargetLowering::CallLoweringInfo CLI(DAG);
+ CLI.setDebugLoc(SDLoc(Node)).setChain(InChain)
+ .setCallee(TLI.getLibcallCallingConv(LC), RetTy, Callee, std::move(Args), 0)
+ .setTailCall(isTailCall).setSExtResult(isSigned).setZExtResult(!isSigned);
+ std::pair<SDValue, SDValue> CallInfo = TLI.LowerCallTo(CLI);
if (!CallInfo.second.getNode())
// It's a tailcall, return the chain (which is the DAG root).
return CallInfo.first;
}
-/// ExpandLibCall - Generate a libcall taking the given operands as arguments
+/// Generate a libcall taking the given operands as arguments
/// and returning a result of type RetVT.
SDValue SelectionDAGLegalize::ExpandLibCall(RTLIB::Libcall LC, EVT RetVT,
const SDValue *Ops, unsigned NumOps,
TLI.getPointerTy());
Type *RetTy = RetVT.getTypeForEVT(*DAG.getContext());
- TargetLowering::
- CallLoweringInfo CLI(DAG.getEntryNode(), RetTy, isSigned, !isSigned, false,
- false, 0, TLI.getLibcallCallingConv(LC),
- /*isTailCall=*/false,
- /*doesNotReturn=*/false, /*isReturnValueUsed=*/true,
- Callee, Args, DAG, dl);
+
+ TargetLowering::CallLoweringInfo CLI(DAG);
+ CLI.setDebugLoc(dl).setChain(DAG.getEntryNode())
+ .setCallee(TLI.getLibcallCallingConv(LC), RetTy, Callee, std::move(Args), 0)
+ .setSExtResult(isSigned).setZExtResult(!isSigned);
+
std::pair<SDValue,SDValue> CallInfo = TLI.LowerCallTo(CLI);
return CallInfo.first;
}
-// ExpandChainLibCall - Expand a node into a call to a libcall. Similar to
+// Expand a node into a call to a libcall. Similar to
// ExpandLibCall except that the first operand is the in-chain.
std::pair<SDValue, SDValue>
SelectionDAGLegalize::ExpandChainLibCall(RTLIB::Libcall LC,
TLI.getPointerTy());
Type *RetTy = Node->getValueType(0).getTypeForEVT(*DAG.getContext());
- TargetLowering::
- CallLoweringInfo CLI(InChain, RetTy, isSigned, !isSigned, false, false,
- 0, TLI.getLibcallCallingConv(LC), /*isTailCall=*/false,
- /*doesNotReturn=*/false, /*isReturnValueUsed=*/true,
- Callee, Args, DAG, SDLoc(Node));
+
+ TargetLowering::CallLoweringInfo CLI(DAG);
+ CLI.setDebugLoc(SDLoc(Node)).setChain(InChain)
+ .setCallee(TLI.getLibcallCallingConv(LC), RetTy, Callee, std::move(Args), 0)
+ .setSExtResult(isSigned).setZExtResult(!isSigned);
+
std::pair<SDValue, SDValue> CallInfo = TLI.LowerCallTo(CLI);
return CallInfo;
return ExpandLibCall(LC, Node, isSigned);
}
-/// isDivRemLibcallAvailable - Return true if divmod libcall is available.
+/// Return true if divmod libcall is available.
static bool isDivRemLibcallAvailable(SDNode *Node, bool isSigned,
const TargetLowering &TLI) {
RTLIB::Libcall LC;
case MVT::i128: LC= isSigned ? RTLIB::SDIVREM_I128:RTLIB::UDIVREM_I128; break;
}
- return TLI.getLibcallName(LC) != 0;
+ return TLI.getLibcallName(LC) != nullptr;
}
-/// useDivRem - Only issue divrem libcall if both quotient and remainder are
-/// needed.
+/// Only issue divrem libcall if both quotient and remainder are needed.
static bool useDivRem(SDNode *Node, bool isSigned, bool isDIV) {
// The other use might have been replaced with a divrem already.
unsigned DivRemOpc = isSigned ? ISD::SDIVREM : ISD::UDIVREM;
return false;
}
-/// ExpandDivRemLibCall - Issue libcalls to __{u}divmod to compute div / rem
-/// pairs.
+/// Issue libcalls to __{u}divmod to compute div / rem pairs.
void
SelectionDAGLegalize::ExpandDivRemLibCall(SDNode *Node,
SmallVectorImpl<SDValue> &Results) {
TLI.getPointerTy());
SDLoc dl(Node);
- TargetLowering::
- CallLoweringInfo CLI(InChain, RetTy, isSigned, !isSigned, false, false,
- 0, TLI.getLibcallCallingConv(LC), /*isTailCall=*/false,
- /*doesNotReturn=*/false, /*isReturnValueUsed=*/true,
- Callee, Args, DAG, dl);
+ TargetLowering::CallLoweringInfo CLI(DAG);
+ CLI.setDebugLoc(dl).setChain(InChain)
+ .setCallee(TLI.getLibcallCallingConv(LC), RetTy, Callee, std::move(Args), 0)
+ .setSExtResult(isSigned).setZExtResult(!isSigned);
+
std::pair<SDValue, SDValue> CallInfo = TLI.LowerCallTo(CLI);
// Remainder is loaded back from the stack frame.
Results.push_back(Rem);
}
-/// isSinCosLibcallAvailable - Return true if sincos libcall is available.
+/// Return true if sincos libcall is available.
static bool isSinCosLibcallAvailable(SDNode *Node, const TargetLowering &TLI) {
RTLIB::Libcall LC;
switch (Node->getSimpleValueType(0).SimpleTy) {
case MVT::f128: LC = RTLIB::SINCOS_F128; break;
case MVT::ppcf128: LC = RTLIB::SINCOS_PPCF128; break;
}
- return TLI.getLibcallName(LC) != 0;
+ return TLI.getLibcallName(LC) != nullptr;
}
-/// canCombineSinCosLibcall - Return true if sincos libcall is available and
-/// can be used to combine sin and cos.
+/// Return true if sincos libcall is available and can be used to combine sin
+/// and cos.
static bool canCombineSinCosLibcall(SDNode *Node, const TargetLowering &TLI,
const TargetMachine &TM) {
if (!isSinCosLibcallAvailable(Node, TLI))
return true;
}
-/// useSinCos - Only issue sincos libcall if both sin and cos are
-/// needed.
+/// Only issue sincos libcall if both sin and cos are needed.
static bool useSinCos(SDNode *Node) {
unsigned OtherOpcode = Node->getOpcode() == ISD::FSIN
? ISD::FCOS : ISD::FSIN;
return false;
}
-/// ExpandSinCosLibCall - Issue libcalls to sincos to compute sin / cos
-/// pairs.
+/// Issue libcalls to sincos to compute sin / cos pairs.
void
SelectionDAGLegalize::ExpandSinCosLibCall(SDNode *Node,
SmallVectorImpl<SDValue> &Results) {
TLI.getPointerTy());
SDLoc dl(Node);
- TargetLowering::
- CallLoweringInfo CLI(InChain, Type::getVoidTy(*DAG.getContext()),
- false, false, false, false,
- 0, TLI.getLibcallCallingConv(LC), /*isTailCall=*/false,
- /*doesNotReturn=*/false, /*isReturnValueUsed=*/true,
- Callee, Args, DAG, dl);
+ TargetLowering::CallLoweringInfo CLI(DAG);
+ CLI.setDebugLoc(dl).setChain(InChain)
+ .setCallee(TLI.getLibcallCallingConv(LC),
+ Type::getVoidTy(*DAG.getContext()), Callee, std::move(Args), 0);
+
std::pair<SDValue, SDValue> CallInfo = TLI.LowerCallTo(CLI);
Results.push_back(DAG.getLoad(RetVT, dl, CallInfo.second, SinPtr,
MachinePointerInfo(), false, false, false, 0));
}
-/// ExpandLegalINT_TO_FP - This function is responsible for legalizing a
+/// This function is responsible for legalizing a
/// INT_TO_FP operation of the specified operand when the target requests that
/// we expand it. At this point, we know that the result and operand types are
/// legal for the target.
SDValue Load = DAG.getExtLoad(ISD::EXTLOAD, dl, DestVT,
DAG.getEntryNode(), CPIdx,
MachinePointerInfo::getConstantPool(),
- MVT::f32, false, false, Alignment);
+ MVT::f32, false, false, false, Alignment);
HandleSDNode Handle(Load);
LegalizeOp(Load.getNode());
FudgeInReg = Handle.getValue();
return DAG.getNode(ISD::FADD, dl, DestVT, Tmp1, FudgeInReg);
}
-/// PromoteLegalINT_TO_FP - This function is responsible for legalizing a
+/// This function is responsible for legalizing a
/// *INT_TO_FP operation of the specified operand when the target requests that
/// we promote it. At this point, we know that the result and operand types are
/// legal for the target, and that there is a legal UINT_TO_FP or SINT_TO_FP
dl, NewInTy, LegalOp));
}
-/// PromoteLegalFP_TO_INT - This function is responsible for legalizing a
+/// This function is responsible for legalizing a
/// FP_TO_*INT operation of the specified operand when the target requests that
/// we promote it. At this point, we know that the result and operand types are
/// legal for the target, and that there is a legal FP_TO_UINT or FP_TO_SINT
NewOutTy = (MVT::SimpleValueType)(NewOutTy.getSimpleVT().SimpleTy+1);
assert(NewOutTy.isInteger() && "Ran out of possibilities!");
+ // A larger signed type can hold all unsigned values of the requested type,
+ // so using FP_TO_SINT is valid
if (TLI.isOperationLegalOrCustom(ISD::FP_TO_SINT, NewOutTy)) {
OpToUse = ISD::FP_TO_SINT;
break;
}
- if (TLI.isOperationLegalOrCustom(ISD::FP_TO_UINT, NewOutTy)) {
+ // However, if the value may be < 0.0, we *must* use some FP_TO_SINT.
+ if (!isSigned && TLI.isOperationLegalOrCustom(ISD::FP_TO_UINT, NewOutTy)) {
OpToUse = ISD::FP_TO_UINT;
break;
}
return DAG.getNode(ISD::TRUNCATE, dl, DestVT, Operation);
}
-/// ExpandBSWAP - Open code the operations for BSWAP of the specified operation.
-///
+/// Open code the operations for BSWAP of the specified operation.
SDValue SelectionDAGLegalize::ExpandBSWAP(SDValue Op, SDLoc dl) {
EVT VT = Op.getValueType();
EVT SHVT = TLI.getShiftAmountTy(VT);
}
}
-/// ExpandBitCount - Expand the specified bitcount instruction into operations.
-///
+/// Expand the specified bitcount instruction into operations.
SDValue SelectionDAGLegalize::ExpandBitCount(unsigned Opc, SDValue Op,
SDLoc dl) {
switch (Opc) {
// x = x | (x >>32); // for 64-bit input
// return popcount(~x);
//
- // but see also: http://www.hackersdelight.org/HDcode/nlz.cc
+ // Ref: "Hacker's Delight" by Henry Warren
EVT VT = Op.getValueType();
EVT ShVT = TLI.getShiftAmountTy(VT);
unsigned len = VT.getSizeInBits();
// for now, we use: { return popcount(~x & (x - 1)); }
// unless the target has ctlz but not ctpop, in which case we use:
// { return 32 - nlz(~x & (x-1)); }
- // see also http://www.hackersdelight.org/HDcode/ntz.cc
+ // Ref: "Hacker's Delight" by Henry Warren
EVT VT = Op.getValueType();
SDValue Tmp3 = DAG.getNode(ISD::AND, dl, VT,
DAG.getNOT(dl, Op, VT),
std::pair <SDValue, SDValue> SelectionDAGLegalize::ExpandAtomic(SDNode *Node) {
unsigned Opc = Node->getOpcode();
MVT VT = cast<AtomicSDNode>(Node)->getMemoryVT().getSimpleVT();
- RTLIB::Libcall LC;
-
- switch (Opc) {
- default:
- llvm_unreachable("Unhandled atomic intrinsic Expand!");
- case ISD::ATOMIC_SWAP:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_LOCK_TEST_AND_SET_1; break;
- case MVT::i16: LC = RTLIB::SYNC_LOCK_TEST_AND_SET_2; break;
- case MVT::i32: LC = RTLIB::SYNC_LOCK_TEST_AND_SET_4; break;
- case MVT::i64: LC = RTLIB::SYNC_LOCK_TEST_AND_SET_8; break;
- case MVT::i128:LC = RTLIB::SYNC_LOCK_TEST_AND_SET_16;break;
- }
- break;
- case ISD::ATOMIC_CMP_SWAP:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_1; break;
- case MVT::i16: LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_2; break;
- case MVT::i32: LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_4; break;
- case MVT::i64: LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_8; break;
- case MVT::i128:LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_ADD:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_ADD_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_ADD_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_ADD_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_ADD_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_ADD_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_SUB:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_SUB_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_SUB_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_SUB_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_SUB_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_SUB_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_AND:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_AND_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_AND_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_AND_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_AND_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_AND_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_OR:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_OR_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_OR_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_OR_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_OR_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_OR_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_XOR:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_XOR_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_XOR_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_XOR_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_XOR_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_XOR_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_NAND:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_NAND_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_NAND_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_NAND_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_NAND_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_NAND_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_MAX:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_MAX_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_MAX_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_MAX_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_MAX_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_MAX_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_UMAX:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_UMAX_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_UMAX_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_UMAX_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_UMAX_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_UMAX_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_MIN:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_MIN_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_MIN_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_MIN_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_MIN_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_MIN_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_UMIN:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_UMIN_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_UMIN_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_UMIN_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_UMIN_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_UMIN_16;break;
- }
- break;
- }
+ RTLIB::Libcall LC = RTLIB::getATOMIC(Opc, VT);
+ assert(LC != RTLIB::UNKNOWN_LIBCALL && "Unexpected atomic op or value type!");
return ExpandChainLibCall(LC, Node, false);
}
// If the target didn't lower this, lower it to '__sync_synchronize()' call
// FIXME: handle "fence singlethread" more efficiently.
TargetLowering::ArgListTy Args;
- TargetLowering::
- CallLoweringInfo CLI(Node->getOperand(0),
- Type::getVoidTy(*DAG.getContext()),
- false, false, false, false, 0, CallingConv::C,
- /*isTailCall=*/false,
- /*doesNotReturn=*/false, /*isReturnValueUsed=*/true,
- DAG.getExternalSymbol("__sync_synchronize",
- TLI.getPointerTy()),
- Args, DAG, dl);
+
+ TargetLowering::CallLoweringInfo CLI(DAG);
+ CLI.setDebugLoc(dl).setChain(Node->getOperand(0))
+ .setCallee(CallingConv::C, Type::getVoidTy(*DAG.getContext()),
+ DAG.getExternalSymbol("__sync_synchronize",
+ TLI.getPointerTy()), std::move(Args), 0);
+
std::pair<SDValue, SDValue> CallResult = TLI.LowerCallTo(CLI);
Results.push_back(CallResult.second);
case ISD::ATOMIC_LOAD: {
// There is no libcall for atomic load; fake it with ATOMIC_CMP_SWAP.
SDValue Zero = DAG.getConstant(0, Node->getValueType(0));
- SDValue Swap = DAG.getAtomic(ISD::ATOMIC_CMP_SWAP, dl,
- cast<AtomicSDNode>(Node)->getMemoryVT(),
- Node->getOperand(0),
- Node->getOperand(1), Zero, Zero,
- cast<AtomicSDNode>(Node)->getMemOperand(),
- cast<AtomicSDNode>(Node)->getOrdering(),
- cast<AtomicSDNode>(Node)->getSynchScope());
+ SDVTList VTs = DAG.getVTList(Node->getValueType(0), MVT::Other);
+ SDValue Swap = DAG.getAtomicCmpSwap(
+ ISD::ATOMIC_CMP_SWAP, dl, cast<AtomicSDNode>(Node)->getMemoryVT(), VTs,
+ Node->getOperand(0), Node->getOperand(1), Zero, Zero,
+ cast<AtomicSDNode>(Node)->getMemOperand(),
+ cast<AtomicSDNode>(Node)->getOrdering(),
+ cast<AtomicSDNode>(Node)->getOrdering(),
+ cast<AtomicSDNode>(Node)->getSynchScope());
Results.push_back(Swap.getValue(0));
Results.push_back(Swap.getValue(1));
break;
Results.push_back(Tmp.second);
break;
}
+ case ISD::ATOMIC_CMP_SWAP_WITH_SUCCESS: {
+ // Expanding an ATOMIC_CMP_SWAP_WITH_SUCCESS produces an ATOMIC_CMP_SWAP and
+ // splits out the success value as a comparison. Expanding the resulting
+ // ATOMIC_CMP_SWAP will produce a libcall.
+ SDVTList VTs = DAG.getVTList(Node->getValueType(0), MVT::Other);
+ SDValue Res = DAG.getAtomicCmpSwap(
+ ISD::ATOMIC_CMP_SWAP, dl, cast<AtomicSDNode>(Node)->getMemoryVT(), VTs,
+ Node->getOperand(0), Node->getOperand(1), Node->getOperand(2),
+ Node->getOperand(3), cast<MemSDNode>(Node)->getMemOperand(),
+ cast<AtomicSDNode>(Node)->getSuccessOrdering(),
+ cast<AtomicSDNode>(Node)->getFailureOrdering(),
+ cast<AtomicSDNode>(Node)->getSynchScope());
+
+ SDValue Success = DAG.getSetCC(SDLoc(Node), Node->getValueType(1),
+ Res, Node->getOperand(2), ISD::SETEQ);
+
+ Results.push_back(Res.getValue(0));
+ Results.push_back(Success);
+ Results.push_back(Res.getValue(1));
+ break;
+ }
case ISD::DYNAMIC_STACKALLOC:
ExpandDYNAMIC_STACKALLOC(Node, Results);
break;
case ISD::TRAP: {
// If this operation is not supported, lower it to 'abort()' call
TargetLowering::ArgListTy Args;
- TargetLowering::
- CallLoweringInfo CLI(Node->getOperand(0),
- Type::getVoidTy(*DAG.getContext()),
- false, false, false, false, 0, CallingConv::C,
- /*isTailCall=*/false,
- /*doesNotReturn=*/false, /*isReturnValueUsed=*/true,
- DAG.getExternalSymbol("abort", TLI.getPointerTy()),
- Args, DAG, dl);
+ TargetLowering::CallLoweringInfo CLI(DAG);
+ CLI.setDebugLoc(dl).setChain(Node->getOperand(0))
+ .setCallee(CallingConv::C, Type::getVoidTy(*DAG.getContext()),
+ DAG.getExternalSymbol("abort", TLI.getPointerTy()),
+ std::move(Args), 0);
std::pair<SDValue, SDValue> CallResult = TLI.LowerCallTo(CLI);
Results.push_back(CallResult.second);
Node->getOperand(0), Node->getValueType(0), dl);
Results.push_back(Tmp1);
break;
+ case ISD::FP_TO_SINT:
+ if (TLI.expandFP_TO_SINT(Node, Tmp1, DAG))
+ Results.push_back(Tmp1);
+ break;
case ISD::FP_TO_UINT: {
SDValue True, False;
EVT VT = Node->getOperand(0).getValueType();
TLI.getVectorIdxTy())));
}
- Tmp1 = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &Ops[0], Ops.size());
+ Tmp1 = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Ops);
// We may have changed the BUILD_VECTOR type. Cast it back to the Node type.
Tmp1 = DAG.getNode(ISD::BITCAST, dl, Node->getValueType(0), Tmp1);
Results.push_back(Tmp1);
Results.push_back(Tmp1);
break;
}
+ case ISD::FMINNUM:
+ Results.push_back(ExpandFPLibCall(Node, RTLIB::FMIN_F32, RTLIB::FMIN_F64,
+ RTLIB::FMIN_F80, RTLIB::FMIN_F128,
+ RTLIB::FMIN_PPCF128));
+ break;
+ case ISD::FMAXNUM:
+ Results.push_back(ExpandFPLibCall(Node, RTLIB::FMAX_F32, RTLIB::FMAX_F64,
+ RTLIB::FMAX_F80, RTLIB::FMAX_F128,
+ RTLIB::FMAX_PPCF128));
+ break;
case ISD::FSQRT:
Results.push_back(ExpandFPLibCall(Node, RTLIB::SQRT_F32, RTLIB::SQRT_F64,
RTLIB::SQRT_F80, RTLIB::SQRT_F128,
RTLIB::FMA_F80, RTLIB::FMA_F128,
RTLIB::FMA_PPCF128));
break;
- case ISD::FP16_TO_FP32:
- Results.push_back(ExpandLibCall(RTLIB::FPEXT_F16_F32, Node, false));
+ case ISD::FMAD:
+ llvm_unreachable("Illegal fmad should never be formed");
+
+ case ISD::FADD:
+ Results.push_back(ExpandFPLibCall(Node, RTLIB::ADD_F32, RTLIB::ADD_F64,
+ RTLIB::ADD_F80, RTLIB::ADD_F128,
+ RTLIB::ADD_PPCF128));
+ break;
+ case ISD::FMUL:
+ Results.push_back(ExpandFPLibCall(Node, RTLIB::MUL_F32, RTLIB::MUL_F64,
+ RTLIB::MUL_F80, RTLIB::MUL_F128,
+ RTLIB::MUL_PPCF128));
break;
- case ISD::FP32_TO_FP16:
- Results.push_back(ExpandLibCall(RTLIB::FPROUND_F32_F16, Node, false));
+ case ISD::FP16_TO_FP: {
+ if (Node->getValueType(0) == MVT::f32) {
+ Results.push_back(ExpandLibCall(RTLIB::FPEXT_F16_F32, Node, false));
+ break;
+ }
+
+ // We can extend to types bigger than f32 in two steps without changing the
+ // result. Since "f16 -> f32" is much more commonly available, give CodeGen
+ // the option of emitting that before resorting to a libcall.
+ SDValue Res =
+ DAG.getNode(ISD::FP16_TO_FP, dl, MVT::f32, Node->getOperand(0));
+ Results.push_back(
+ DAG.getNode(ISD::FP_EXTEND, dl, Node->getValueType(0), Res));
+ break;
+ }
+ case ISD::FP_TO_FP16: {
+ if (!TM.Options.UseSoftFloat && TM.Options.UnsafeFPMath) {
+ SDValue Op = Node->getOperand(0);
+ MVT SVT = Op.getSimpleValueType();
+ if ((SVT == MVT::f64 || SVT == MVT::f80) &&
+ TLI.isOperationLegalOrCustom(ISD::FP_TO_FP16, MVT::f32)) {
+ // Under fastmath, we can expand this node into a fround followed by
+ // a float-half conversion.
+ SDValue FloatVal = DAG.getNode(ISD::FP_ROUND, dl, MVT::f32, Op,
+ DAG.getIntPtrConstant(0));
+ Results.push_back(
+ DAG.getNode(ISD::FP_TO_FP16, dl, MVT::i16, FloatVal));
+ break;
+ }
+ }
+
+ RTLIB::Libcall LC =
+ RTLIB::getFPROUND(Node->getOperand(0).getValueType(), MVT::f16);
+ assert(LC != RTLIB::UNKNOWN_LIBCALL && "Unable to expand fp_to_fp16");
+ Results.push_back(ExpandLibCall(LC, Node, false));
break;
+ }
case ISD::ConstantFP: {
ConstantFPSDNode *CFP = cast<ConstantFPSDNode>(Node);
// Check to see if this FP immediate is already legal.
}
case ISD::FSUB: {
EVT VT = Node->getValueType(0);
- assert(TLI.isOperationLegalOrCustom(ISD::FADD, VT) &&
- TLI.isOperationLegalOrCustom(ISD::FNEG, VT) &&
- "Don't know how to expand this FP subtraction!");
- Tmp1 = DAG.getNode(ISD::FNEG, dl, VT, Node->getOperand(1));
- Tmp1 = DAG.getNode(ISD::FADD, dl, VT, Node->getOperand(0), Tmp1);
- Results.push_back(Tmp1);
+ if (TLI.isOperationLegalOrCustom(ISD::FADD, VT) &&
+ TLI.isOperationLegalOrCustom(ISD::FNEG, VT)) {
+ Tmp1 = DAG.getNode(ISD::FNEG, dl, VT, Node->getOperand(1));
+ Tmp1 = DAG.getNode(ISD::FADD, dl, VT, Node->getOperand(0), Tmp1);
+ Results.push_back(Tmp1);
+ } else {
+ Results.push_back(ExpandFPLibCall(Node, RTLIB::SUB_F32, RTLIB::SUB_F64,
+ RTLIB::SUB_F80, RTLIB::SUB_F128,
+ RTLIB::SUB_PPCF128));
+ }
break;
}
case ISD::SUB: {
Node->getOperand(1)));
break;
}
+
+ SDValue Lo, Hi;
+ EVT HalfType = VT.getHalfSizedIntegerVT(*DAG.getContext());
+ if (TLI.isOperationLegalOrCustom(ISD::ZERO_EXTEND, VT) &&
+ TLI.isOperationLegalOrCustom(ISD::ANY_EXTEND, VT) &&
+ TLI.isOperationLegalOrCustom(ISD::SHL, VT) &&
+ TLI.isOperationLegalOrCustom(ISD::OR, VT) &&
+ TLI.expandMUL(Node, Lo, Hi, HalfType, DAG)) {
+ Lo = DAG.getNode(ISD::ZERO_EXTEND, dl, VT, Lo);
+ Hi = DAG.getNode(ISD::ANY_EXTEND, dl, VT, Hi);
+ SDValue Shift = DAG.getConstant(HalfType.getSizeInBits(),
+ TLI.getShiftAmountTy(HalfType));
+ Hi = DAG.getNode(ISD::SHL, dl, VT, Hi, Shift);
+ Results.push_back(DAG.getNode(ISD::OR, dl, VT, Lo, Hi));
+ break;
+ }
+
Tmp1 = ExpandIntLibCall(Node, false,
RTLIB::MUL_I8,
RTLIB::MUL_I16, RTLIB::MUL_I32,
ISD::ADD : ISD::SUB, dl, LHS.getValueType(),
LHS, RHS);
Results.push_back(Sum);
- EVT OType = Node->getValueType(1);
+ EVT ResultType = Node->getValueType(1);
+ EVT OType = getSetCCResultType(Node->getValueType(0));
SDValue Zero = DAG.getConstant(0, LHS.getValueType());
SDValue SumSignNE = DAG.getSetCC(dl, OType, LHSSign, SumSign, ISD::SETNE);
SDValue Cmp = DAG.getNode(ISD::AND, dl, OType, SignsMatch, SumSignNE);
- Results.push_back(Cmp);
+ Results.push_back(DAG.getBoolExtOrTrunc(Cmp, dl, ResultType, ResultType));
break;
}
case ISD::UADDO:
ISD::ADD : ISD::SUB, dl, LHS.getValueType(),
LHS, RHS);
Results.push_back(Sum);
- Results.push_back(DAG.getSetCC(dl, Node->getValueType(1), Sum, LHS,
- Node->getOpcode () == ISD::UADDO ?
- ISD::SETULT : ISD::SETUGT));
+
+ EVT ResultType = Node->getValueType(1);
+ EVT SetCCType = getSetCCResultType(Node->getValueType(0));
+ ISD::CondCode CC
+ = Node->getOpcode() == ISD::UADDO ? ISD::SETULT : ISD::SETUGT;
+ SDValue SetCC = DAG.getSetCC(dl, SetCCType, Sum, LHS, CC);
+
+ Results.push_back(DAG.getBoolExtOrTrunc(SetCC, dl, ResultType, ResultType));
break;
}
case ISD::UMULO:
BottomHalf = DAG.getNode(Ops[isSigned][1], dl, DAG.getVTList(VT, VT), LHS,
RHS);
TopHalf = BottomHalf.getValue(1);
- } else if (TLI.isTypeLegal(EVT::getIntegerVT(*DAG.getContext(),
- VT.getSizeInBits() * 2))) {
+ } else if (TLI.isTypeLegal(WideVT)) {
LHS = DAG.getNode(Ops[isSigned][2], dl, WideVT, LHS);
RHS = DAG.getNode(Ops[isSigned][2], dl, WideVT, RHS);
Tmp1 = DAG.getNode(ISD::MUL, dl, WideVT, LHS, RHS);
TopHalf = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, VT, Ret,
DAG.getIntPtrConstant(1));
// Ret is a node with an illegal type. Because such things are not
- // generally permitted during this phase of legalization, delete the
- // node. The above EXTRACT_ELEMENT nodes should have been folded.
- DAG.DeleteNode(Ret.getNode());
+ // generally permitted during this phase of legalization, make sure the
+ // node has no more uses. The above EXTRACT_ELEMENT nodes should have been
+ // folded.
+ assert(Ret->use_empty() &&
+ "Unexpected uses of illegally type from expanded lib call.");
}
if (isSigned) {
EVT MemVT = EVT::getIntegerVT(*DAG.getContext(), EntrySize * 8);
SDValue LD = DAG.getExtLoad(ISD::SEXTLOAD, dl, PTy, Chain, Addr,
MachinePointerInfo::getJumpTable(), MemVT,
- false, false, 0);
+ false, false, false, 0);
Addr = LD;
if (TM.getRelocationModel() == Reloc::PIC_) {
// For PIC, the sequence is:
// If we expanded the SETCC by inverting the condition code, then wrap
// the existing SETCC in a NOT to restore the intended condition.
if (NeedInvert)
- Tmp1 = DAG.getNOT(dl, Tmp1, Tmp1->getValueType(0));
+ Tmp1 = DAG.getLogicalNOT(dl, Tmp1, Tmp1->getValueType(0));
Results.push_back(Tmp1);
break;
// illegal; expand it into a SELECT_CC.
EVT VT = Node->getValueType(0);
int TrueValue;
- switch (TLI.getBooleanContents(VT.isVector())) {
+ switch (TLI.getBooleanContents(Tmp1->getValueType(0))) {
case TargetLowering::ZeroOrOneBooleanContent:
case TargetLowering::UndefinedBooleanContent:
TrueValue = 1;
Tmp2 = Node->getOperand(1); // RHS
Tmp3 = Node->getOperand(2); // True
Tmp4 = Node->getOperand(3); // False
+ EVT VT = Node->getValueType(0);
SDValue CC = Node->getOperand(4);
+ ISD::CondCode CCOp = cast<CondCodeSDNode>(CC)->get();
+
+ if (TLI.isCondCodeLegal(CCOp, Tmp1.getSimpleValueType())) {
+ // If the condition code is legal, then we need to expand this
+ // node using SETCC and SELECT.
+ EVT CmpVT = Tmp1.getValueType();
+ assert(!TLI.isOperationExpand(ISD::SELECT, VT) &&
+ "Cannot expand ISD::SELECT_CC when ISD::SELECT also needs to be "
+ "expanded.");
+ EVT CCVT = TLI.getSetCCResultType(*DAG.getContext(), CmpVT);
+ SDValue Cond = DAG.getNode(ISD::SETCC, dl, CCVT, Tmp1, Tmp2, CC);
+ Results.push_back(DAG.getSelect(dl, VT, Cond, Tmp3, Tmp4));
+ break;
+ }
+ // SELECT_CC is legal, so the condition code must not be.
bool Legalized = false;
// Try to legalize by inverting the condition. This is for targets that
// might support an ordered version of a condition, but not the unordered
// version (or vice versa).
- ISD::CondCode InvCC = ISD::getSetCCInverse(cast<CondCodeSDNode>(CC)->get(),
+ ISD::CondCode InvCC = ISD::getSetCCInverse(CCOp,
Tmp1.getValueType().isInteger());
if (TLI.isCondCodeLegal(InvCC, Tmp1.getSimpleValueType())) {
// Use the new condition code and swap true and false
VT.getScalarType(), Ex, Sh));
}
SDValue Result =
- DAG.getNode(ISD::BUILD_VECTOR, dl, Node->getValueType(0),
- &Scalars[0], Scalars.size());
+ DAG.getNode(ISD::BUILD_VECTOR, dl, Node->getValueType(0), Scalars);
ReplaceNode(SDValue(Node, 0), Result);
break;
}
// use the new one.
DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 0), Tmp2);
DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 1), Chain);
+ if (UpdatedNodes) {
+ UpdatedNodes->insert(Tmp2.getNode());
+ UpdatedNodes->insert(Chain.getNode());
+ }
ReplacedNode(Node);
break;
}
}
case ISD::SELECT: {
unsigned ExtOp, TruncOp;
- if (Node->getValueType(0).isVector()) {
+ if (Node->getValueType(0).isVector() ||
+ Node->getValueType(0).getSizeInBits() == NVT.getSizeInBits()) {
ExtOp = ISD::BITCAST;
TruncOp = ISD::BITCAST;
} else if (Node->getValueType(0).isInteger()) {
Tmp1, Tmp2, Node->getOperand(2)));
break;
}
+ case ISD::FADD:
+ case ISD::FSUB:
+ case ISD::FMUL:
case ISD::FDIV:
case ISD::FREM:
case ISD::FPOW: {
ReplaceNode(Node, Results.data());
}
-// SelectionDAG::Legalize - This is the entry point for the file.
-//
+/// This is the entry point for the file.
void SelectionDAG::Legalize() {
- /// run - This is the main entry point to this class.
- ///
- SelectionDAGLegalize(*this).LegalizeDAG();
+ AssignTopologicalOrder();
+
+ SmallPtrSet<SDNode *, 16> LegalizedNodes;
+ SelectionDAGLegalize Legalizer(*this, LegalizedNodes);
+
+ // Visit all the nodes. We start in topological order, so that we see
+ // nodes with their original operands intact. Legalization can produce
+ // new nodes which may themselves need to be legalized. Iterate until all
+ // nodes have been legalized.
+ for (;;) {
+ bool AnyLegalized = false;
+ for (auto NI = allnodes_end(); NI != allnodes_begin();) {
+ --NI;
+
+ SDNode *N = NI;
+ if (N->use_empty() && N != getRoot().getNode()) {
+ ++NI;
+ DeleteNode(N);
+ continue;
+ }
+
+ if (LegalizedNodes.insert(N).second) {
+ AnyLegalized = true;
+ Legalizer.LegalizeOp(N);
+
+ if (N->use_empty() && N != getRoot().getNode()) {
+ ++NI;
+ DeleteNode(N);
+ }
+ }
+ }
+ if (!AnyLegalized)
+ break;
+
+ }
+
+ // Remove dead nodes now.
+ RemoveDeadNodes();
+}
+
+bool SelectionDAG::LegalizeOp(SDNode *N,
+ SmallSetVector<SDNode *, 16> &UpdatedNodes) {
+ SmallPtrSet<SDNode *, 16> LegalizedNodes;
+ SelectionDAGLegalize Legalizer(*this, LegalizedNodes, &UpdatedNodes);
+
+ // Directly insert the node in question, and legalize it. This will recurse
+ // as needed through operands.
+ LegalizedNodes.insert(N);
+ Legalizer.LegalizeOp(N);
+
+ return LegalizedNodes.count(N);
}