//
//===----------------------------------------------------------------------===//
-#include "llvm/Analysis/DebugInfo.h"
+#include "llvm/CodeGen/SelectionDAG.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/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineJumpTableInfo.h"
-#include "llvm/CodeGen/SelectionDAG.h"
-#include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetLowering.h"
-#include "llvm/Target/TargetData.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/CallingConv.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/LLVMContext.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DebugInfo.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/LLVMContext.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/DenseMap.h"
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/Target/TargetFrameLowering.h"
+#include "llvm/Target/TargetLowering.h"
+#include "llvm/Target/TargetMachine.h"
using namespace llvm;
//===----------------------------------------------------------------------===//
/// LegalizedNodes - The set of nodes which have already been legalized.
SmallPtrSet<SDNode *, 16> LegalizedNodes;
+ EVT getSetCCResultType(EVT VT) const {
+ return TLI.getSetCCResultType(*DAG.getContext(), VT);
+ }
+
// Libcall insertion helpers.
public:
SDValue OptimizeFloatStore(StoreSDNode *ST);
+ void LegalizeLoadOps(SDNode *Node);
+ void LegalizeStoreOps(SDNode *Node);
+
/// PerformInsertVectorEltInMemory - 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 PerformInsertVectorEltInMemory(SDValue Vec, SDValue Val,
- SDValue Idx, DebugLoc dl);
+ SDValue Idx, SDLoc dl);
SDValue ExpandINSERT_VECTOR_ELT(SDValue Vec, SDValue Val,
- SDValue Idx, DebugLoc dl);
+ SDValue Idx, SDLoc dl);
/// ShuffleWithNarrowerEltType - 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>
- SDValue ShuffleWithNarrowerEltType(EVT NVT, EVT VT, DebugLoc dl,
+ SDValue ShuffleWithNarrowerEltType(EVT NVT, EVT VT, SDLoc dl,
SDValue N1, SDValue N2,
ArrayRef<int> Mask) const;
- void LegalizeSetCCCondCode(EVT VT, SDValue &LHS, SDValue &RHS, SDValue &CC,
- DebugLoc dl);
+ bool LegalizeSetCCCondCode(EVT VT, SDValue &LHS, SDValue &RHS, SDValue &CC,
+ bool &NeedInvert, SDLoc dl);
SDValue ExpandLibCall(RTLIB::Libcall LC, SDNode *Node, bool isSigned);
SDValue ExpandLibCall(RTLIB::Libcall LC, EVT RetVT, const SDValue *Ops,
- unsigned NumOps, bool isSigned, DebugLoc dl);
+ unsigned NumOps, bool isSigned, SDLoc dl);
std::pair<SDValue, SDValue> ExpandChainLibCall(RTLIB::Libcall LC,
SDNode *Node, bool isSigned);
SDValue ExpandFPLibCall(SDNode *Node, RTLIB::Libcall Call_F32,
RTLIB::Libcall Call_F64, RTLIB::Libcall Call_F80,
+ RTLIB::Libcall Call_F128,
RTLIB::Libcall Call_PPCF128);
SDValue ExpandIntLibCall(SDNode *Node, bool isSigned,
RTLIB::Libcall Call_I8,
RTLIB::Libcall Call_I64,
RTLIB::Libcall Call_I128);
void ExpandDivRemLibCall(SDNode *Node, SmallVectorImpl<SDValue> &Results);
+ void ExpandSinCosLibCall(SDNode *Node, SmallVectorImpl<SDValue> &Results);
- SDValue EmitStackConvert(SDValue SrcOp, EVT SlotVT, EVT DestVT, DebugLoc dl);
+ SDValue EmitStackConvert(SDValue SrcOp, EVT SlotVT, EVT DestVT, SDLoc dl);
SDValue ExpandBUILD_VECTOR(SDNode *Node);
SDValue ExpandSCALAR_TO_VECTOR(SDNode *Node);
void ExpandDYNAMIC_STACKALLOC(SDNode *Node,
SmallVectorImpl<SDValue> &Results);
SDValue ExpandFCOPYSIGN(SDNode *Node);
SDValue ExpandLegalINT_TO_FP(bool isSigned, SDValue LegalOp, EVT DestVT,
- DebugLoc dl);
+ SDLoc dl);
SDValue PromoteLegalINT_TO_FP(SDValue LegalOp, EVT DestVT, bool isSigned,
- DebugLoc dl);
+ SDLoc dl);
SDValue PromoteLegalFP_TO_INT(SDValue LegalOp, EVT DestVT, bool isSigned,
- DebugLoc dl);
+ SDLoc dl);
- SDValue ExpandBSWAP(SDValue Op, DebugLoc dl);
- SDValue ExpandBitCount(unsigned Opc, SDValue Op, DebugLoc dl);
+ SDValue ExpandBSWAP(SDValue Op, SDLoc dl);
+ SDValue ExpandBitCount(unsigned Opc, SDValue Op, SDLoc dl);
SDValue ExpandExtractFromVectorThroughStack(SDValue Op);
SDValue ExpandInsertToVectorThroughStack(SDValue Op);
public:
// DAGUpdateListener implementation.
- virtual void NodeDeleted(SDNode *N, SDNode *E) {
+ void NodeDeleted(SDNode *N, SDNode *E) override {
ForgetNode(N);
}
- virtual void NodeUpdated(SDNode *N) {}
+ void NodeUpdated(SDNode *N) override {}
// Node replacement helpers
void ReplacedNode(SDNode *N) {
/// 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>
SDValue
-SelectionDAGLegalize::ShuffleWithNarrowerEltType(EVT NVT, EVT VT, DebugLoc dl,
+SelectionDAGLegalize::ShuffleWithNarrowerEltType(EVT NVT, EVT VT, SDLoc dl,
SDValue N1, SDValue N2,
ArrayRef<int> Mask) const {
unsigned NumMaskElts = VT.getVectorNumElements();
SDValue
SelectionDAGLegalize::ExpandConstantFP(ConstantFPSDNode *CFP, bool UseCP) {
bool Extend = false;
- DebugLoc dl = CFP->getDebugLoc();
+ SDLoc dl(CFP);
// If a FP immediate is precise when represented as a float and if the
// target can do an extending load from float to double, we put it into
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
SDValue Val = ST->getValue();
EVT VT = Val.getValueType();
int Alignment = ST->getAlignment();
- DebugLoc dl = ST->getDebugLoc();
+ unsigned AS = ST->getAddressSpace();
+
+ SDLoc dl(ST);
if (ST->getMemoryVT().isFloatingPoint() ||
ST->getMemoryVT().isVector()) {
EVT intVT = EVT::getIntegerVT(*DAG.getContext(), VT.getSizeInBits());
// Do a (aligned) store to a stack slot, then copy from the stack slot
// to the final destination using (unaligned) integer loads and stores.
EVT StoredVT = ST->getMemoryVT();
- EVT RegVT =
+ MVT RegVT =
TLI.getRegisterType(*DAG.getContext(),
EVT::getIntegerVT(*DAG.getContext(),
StoredVT.getSizeInBits()));
SDValue Store = DAG.getTruncStore(Chain, dl,
Val, StackPtr, MachinePointerInfo(),
StoredVT, false, false, 0);
- SDValue Increment = DAG.getConstant(RegBytes, TLI.getPointerTy());
+ SDValue Increment = DAG.getConstant(RegBytes, TLI.getPointerTy(AS));
SmallVector<SDValue, 8> Stores;
unsigned Offset = 0;
.getWithOffset(Offset),
MemVT, ST->isVolatile(),
ST->isNonTemporal(),
- MinAlign(ST->getAlignment(), Offset)));
+ MinAlign(ST->getAlignment(), Offset),
+ ST->getTBAAInfo()));
// 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;
}
Store1 = DAG.getTruncStore(Chain, dl, TLI.isLittleEndian()?Lo:Hi, Ptr,
ST->getPointerInfo(), NewStoredVT,
ST->isVolatile(), ST->isNonTemporal(), Alignment);
+
Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr,
- DAG.getConstant(IncrementSize, TLI.getPointerTy()));
+ DAG.getConstant(IncrementSize, TLI.getPointerTy(AS)));
Alignment = MinAlign(Alignment, IncrementSize);
Store2 = DAG.getTruncStore(Chain, dl, TLI.isLittleEndian()?Hi:Lo, Ptr,
ST->getPointerInfo().getWithOffset(IncrementSize),
NewStoredVT, ST->isVolatile(), ST->isNonTemporal(),
- Alignment);
+ Alignment, ST->getTBAAInfo());
SDValue Result =
DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Store1, Store2);
SDValue Ptr = LD->getBasePtr();
EVT VT = LD->getValueType(0);
EVT LoadedVT = LD->getMemoryVT();
- DebugLoc dl = LD->getDebugLoc();
+ SDLoc dl(LD);
if (VT.isFloatingPoint() || VT.isVector()) {
EVT intVT = EVT::getIntegerVT(*DAG.getContext(), LoadedVT.getSizeInBits());
- if (TLI.isTypeLegal(intVT)) {
+ if (TLI.isTypeLegal(intVT) && TLI.isTypeLegal(LoadedVT)) {
// Expand to a (misaligned) integer load of the same size,
// then bitconvert to floating point or vector.
- SDValue newLoad = DAG.getLoad(intVT, dl, Chain, Ptr, LD->getPointerInfo(),
- LD->isVolatile(),
- LD->isNonTemporal(),
- LD->isInvariant(), LD->getAlignment());
+ SDValue newLoad = DAG.getLoad(intVT, dl, Chain, Ptr,
+ LD->getMemOperand());
SDValue Result = DAG.getNode(ISD::BITCAST, dl, LoadedVT, newLoad);
- if (VT.isFloatingPoint() && LoadedVT != VT)
- Result = DAG.getNode(ISD::FP_EXTEND, dl, VT, Result);
+ if (LoadedVT != VT)
+ Result = DAG.getNode(VT.isFloatingPoint() ? ISD::FP_EXTEND :
+ ISD::ANY_EXTEND, dl, VT, Result);
ValResult = Result;
ChainResult = Chain;
// Copy the value to a (aligned) stack slot using (unaligned) integer
// loads and stores, then do a (aligned) load from the stack slot.
- EVT RegVT = TLI.getRegisterType(*DAG.getContext(), intVT);
+ MVT RegVT = TLI.getRegisterType(*DAG.getContext(), intVT);
unsigned LoadedBytes = LoadedVT.getSizeInBits() / 8;
unsigned RegBytes = RegVT.getSizeInBits() / 8;
unsigned NumRegs = (LoadedBytes + RegBytes - 1) / RegBytes;
LD->getPointerInfo().getWithOffset(Offset),
LD->isVolatile(), LD->isNonTemporal(),
LD->isInvariant(),
- MinAlign(LD->getAlignment(), Offset));
+ MinAlign(LD->getAlignment(), Offset),
+ LD->getTBAAInfo());
// 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(),
- MinAlign(LD->getAlignment(), Offset));
+ MinAlign(LD->getAlignment(), Offset),
+ LD->getTBAAInfo());
// 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,
if (TLI.isLittleEndian()) {
Lo = DAG.getExtLoad(ISD::ZEXTLOAD, dl, VT, Chain, Ptr, LD->getPointerInfo(),
NewLoadedVT, LD->isVolatile(),
- LD->isNonTemporal(), Alignment);
+ LD->isNonTemporal(), Alignment, LD->getTBAAInfo());
Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr,
- DAG.getConstant(IncrementSize, TLI.getPointerTy()));
+ 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->isNonTemporal(), MinAlign(Alignment, IncrementSize),
+ LD->getTBAAInfo());
} else {
Hi = DAG.getExtLoad(HiExtType, dl, VT, Chain, Ptr, LD->getPointerInfo(),
NewLoadedVT, LD->isVolatile(),
- LD->isNonTemporal(), Alignment);
+ LD->isNonTemporal(), Alignment, LD->getTBAAInfo());
Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr,
- DAG.getConstant(IncrementSize, TLI.getPointerTy()));
+ 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->isNonTemporal(), MinAlign(Alignment, IncrementSize),
+ LD->getTBAAInfo());
}
// aggregate the two parts
/// the insert there, and then read the result back.
SDValue SelectionDAGLegalize::
PerformInsertVectorEltInMemory(SDValue Vec, SDValue Val, SDValue Idx,
- DebugLoc dl) {
+ SDLoc dl) {
SDValue Tmp1 = Vec;
SDValue Tmp2 = Val;
SDValue Tmp3 = Idx;
false, false, 0);
// Load the updated vector.
return DAG.getLoad(VT, dl, Ch, StackPtr,
- MachinePointerInfo::getFixedStack(SPFI), false, false,
+ MachinePointerInfo::getFixedStack(SPFI), false, false,
false, 0);
}
SDValue SelectionDAGLegalize::
-ExpandINSERT_VECTOR_ELT(SDValue Vec, SDValue Val, SDValue Idx, DebugLoc dl) {
+ExpandINSERT_VECTOR_ELT(SDValue Vec, SDValue Val, SDValue Idx, SDLoc dl) {
if (ConstantSDNode *InsertPos = dyn_cast<ConstantSDNode>(Idx)) {
// SCALAR_TO_VECTOR requires that the type of the value being inserted
// match the element type of the vector being created, except for
// probably means that we need to integrate dag combiner and legalizer
// together.
// We generally can't do this one for long doubles.
- SDValue Tmp1 = ST->getChain();
- SDValue Tmp2 = ST->getBasePtr();
- SDValue Tmp3;
+ SDValue Chain = ST->getChain();
+ SDValue Ptr = ST->getBasePtr();
unsigned Alignment = ST->getAlignment();
bool isVolatile = ST->isVolatile();
bool isNonTemporal = ST->isNonTemporal();
- DebugLoc dl = ST->getDebugLoc();
+ const MDNode *TBAAInfo = ST->getTBAAInfo();
+ SDLoc dl(ST);
if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(ST->getValue())) {
if (CFP->getValueType(0) == MVT::f32 &&
TLI.isTypeLegal(MVT::i32)) {
- Tmp3 = DAG.getConstant(CFP->getValueAPF().
+ SDValue Con = DAG.getConstant(CFP->getValueAPF().
bitcastToAPInt().zextOrTrunc(32),
MVT::i32);
- return DAG.getStore(Tmp1, dl, Tmp3, Tmp2, ST->getPointerInfo(),
- isVolatile, isNonTemporal, Alignment);
+ return DAG.getStore(Chain, dl, Con, Ptr, ST->getPointerInfo(),
+ isVolatile, isNonTemporal, Alignment, TBAAInfo);
}
if (CFP->getValueType(0) == MVT::f64) {
// If this target supports 64-bit registers, do a single 64-bit store.
if (TLI.isTypeLegal(MVT::i64)) {
- Tmp3 = DAG.getConstant(CFP->getValueAPF().bitcastToAPInt().
+ SDValue Con = DAG.getConstant(CFP->getValueAPF().bitcastToAPInt().
zextOrTrunc(64), MVT::i64);
- return DAG.getStore(Tmp1, dl, Tmp3, Tmp2, ST->getPointerInfo(),
- isVolatile, isNonTemporal, Alignment);
+ return DAG.getStore(Chain, dl, Con, Ptr, ST->getPointerInfo(),
+ isVolatile, isNonTemporal, Alignment, TBAAInfo);
}
if (TLI.isTypeLegal(MVT::i32) && !ST->isVolatile()) {
SDValue Hi = DAG.getConstant(IntVal.lshr(32).trunc(32), MVT::i32);
if (TLI.isBigEndian()) std::swap(Lo, Hi);
- Lo = DAG.getStore(Tmp1, dl, Lo, Tmp2, ST->getPointerInfo(), isVolatile,
- isNonTemporal, Alignment);
- Tmp2 = DAG.getNode(ISD::ADD, dl, Tmp2.getValueType(), Tmp2,
- DAG.getIntPtrConstant(4));
- Hi = DAG.getStore(Tmp1, dl, Hi, Tmp2,
+ Lo = DAG.getStore(Chain, dl, Lo, Ptr, ST->getPointerInfo(), isVolatile,
+ isNonTemporal, Alignment, TBAAInfo);
+ 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));
+ isVolatile, isNonTemporal, MinAlign(Alignment, 4U),
+ TBAAInfo);
return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Lo, Hi);
}
}
}
- return SDValue(0, 0);
+ return SDValue(nullptr, 0);
+}
+
+void SelectionDAGLegalize::LegalizeStoreOps(SDNode *Node) {
+ StoreSDNode *ST = cast<StoreSDNode>(Node);
+ SDValue Chain = ST->getChain();
+ SDValue Ptr = ST->getBasePtr();
+ SDLoc dl(Node);
+
+ unsigned Alignment = ST->getAlignment();
+ bool isVolatile = ST->isVolatile();
+ bool isNonTemporal = ST->isNonTemporal();
+ const MDNode *TBAAInfo = ST->getTBAAInfo();
+
+ if (!ST->isTruncatingStore()) {
+ if (SDNode *OptStore = OptimizeFloatStore(ST).getNode()) {
+ ReplaceNode(ST, OptStore);
+ return;
+ }
+
+ {
+ SDValue Value = ST->getValue();
+ MVT VT = Value.getSimpleValueType();
+ switch (TLI.getOperationAction(ISD::STORE, VT)) {
+ default: llvm_unreachable("This action is not supported yet!");
+ case TargetLowering::Legal: {
+ // 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)) {
+ Type *Ty = ST->getMemoryVT().getTypeForEVT(*DAG.getContext());
+ unsigned ABIAlignment= TLI.getDataLayout()->getABITypeAlignment(Ty);
+ if (ST->getAlignment() < ABIAlignment)
+ ExpandUnalignedStore(cast<StoreSDNode>(Node),
+ DAG, TLI, this);
+ }
+ break;
+ }
+ case TargetLowering::Custom: {
+ SDValue Res = TLI.LowerOperation(SDValue(Node, 0), DAG);
+ if (Res.getNode())
+ ReplaceNode(SDValue(Node, 0), Res);
+ return;
+ }
+ case TargetLowering::Promote: {
+ MVT NVT = TLI.getTypeToPromoteTo(ISD::STORE, VT);
+ assert(NVT.getSizeInBits() == VT.getSizeInBits() &&
+ "Can only promote stores to same size type");
+ Value = DAG.getNode(ISD::BITCAST, dl, NVT, Value);
+ SDValue Result =
+ DAG.getStore(Chain, dl, Value, Ptr,
+ ST->getPointerInfo(), isVolatile,
+ isNonTemporal, Alignment, TBAAInfo);
+ ReplaceNode(SDValue(Node, 0), Result);
+ break;
+ }
+ }
+ return;
+ }
+ } else {
+ SDValue Value = ST->getValue();
+
+ EVT StVT = ST->getMemoryVT();
+ unsigned StWidth = StVT.getSizeInBits();
+
+ if (StWidth != StVT.getStoreSizeInBits()) {
+ // Promote to a byte-sized store with upper bits zero if not
+ // storing an integral number of bytes. For example, promote
+ // TRUNCSTORE:i1 X -> TRUNCSTORE:i8 (and X, 1)
+ EVT NVT = EVT::getIntegerVT(*DAG.getContext(),
+ StVT.getStoreSizeInBits());
+ Value = DAG.getZeroExtendInReg(Value, dl, StVT);
+ SDValue Result =
+ DAG.getTruncStore(Chain, dl, Value, Ptr, ST->getPointerInfo(),
+ NVT, isVolatile, isNonTemporal, Alignment,
+ TBAAInfo);
+ ReplaceNode(SDValue(Node, 0), Result);
+ } else if (StWidth & (StWidth - 1)) {
+ // If not storing a power-of-2 number of bits, expand as two stores.
+ assert(!StVT.isVector() && "Unsupported truncstore!");
+ unsigned RoundWidth = 1 << Log2_32(StWidth);
+ assert(RoundWidth < StWidth);
+ unsigned ExtraWidth = StWidth - RoundWidth;
+ assert(ExtraWidth < RoundWidth);
+ assert(!(RoundWidth % 8) && !(ExtraWidth % 8) &&
+ "Store size not an integral number of bytes!");
+ EVT RoundVT = EVT::getIntegerVT(*DAG.getContext(), RoundWidth);
+ EVT ExtraVT = EVT::getIntegerVT(*DAG.getContext(), ExtraWidth);
+ SDValue Lo, Hi;
+ unsigned IncrementSize;
+
+ if (TLI.isLittleEndian()) {
+ // TRUNCSTORE:i24 X -> TRUNCSTORE:i16 X, TRUNCSTORE@+2:i8 (srl X, 16)
+ // Store the bottom RoundWidth bits.
+ Lo = DAG.getTruncStore(Chain, dl, Value, Ptr, ST->getPointerInfo(),
+ RoundVT,
+ isVolatile, isNonTemporal, Alignment,
+ TBAAInfo);
+
+ // Store the remaining ExtraWidth bits.
+ IncrementSize = RoundWidth / 8;
+ Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr,
+ DAG.getConstant(IncrementSize, Ptr.getValueType()));
+ Hi = DAG.getNode(ISD::SRL, dl, Value.getValueType(), Value,
+ DAG.getConstant(RoundWidth,
+ TLI.getShiftAmountTy(Value.getValueType())));
+ Hi = DAG.getTruncStore(Chain, dl, Hi, Ptr,
+ ST->getPointerInfo().getWithOffset(IncrementSize),
+ ExtraVT, isVolatile, isNonTemporal,
+ MinAlign(Alignment, IncrementSize), TBAAInfo);
+ } else {
+ // Big endian - avoid unaligned stores.
+ // TRUNCSTORE:i24 X -> TRUNCSTORE:i16 (srl X, 8), TRUNCSTORE@+2:i8 X
+ // Store the top RoundWidth bits.
+ Hi = DAG.getNode(ISD::SRL, dl, Value.getValueType(), Value,
+ DAG.getConstant(ExtraWidth,
+ TLI.getShiftAmountTy(Value.getValueType())));
+ Hi = DAG.getTruncStore(Chain, dl, Hi, Ptr, ST->getPointerInfo(),
+ RoundVT, isVolatile, isNonTemporal, Alignment,
+ TBAAInfo);
+
+ // Store the remaining ExtraWidth bits.
+ IncrementSize = RoundWidth / 8;
+ Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr,
+ DAG.getConstant(IncrementSize, Ptr.getValueType()));
+ Lo = DAG.getTruncStore(Chain, dl, Value, Ptr,
+ ST->getPointerInfo().getWithOffset(IncrementSize),
+ ExtraVT, isVolatile, isNonTemporal,
+ MinAlign(Alignment, IncrementSize), TBAAInfo);
+ }
+
+ // The order of the stores doesn't matter.
+ SDValue Result = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Lo, Hi);
+ ReplaceNode(SDValue(Node, 0), Result);
+ } else {
+ switch (TLI.getTruncStoreAction(ST->getValue().getSimpleValueType(),
+ StVT.getSimpleVT())) {
+ default: llvm_unreachable("This action is not supported yet!");
+ case TargetLowering::Legal: {
+ unsigned AS = ST->getAddressSpace();
+ // If this is an unaligned store and the target doesn't support it,
+ // expand it.
+ if (!TLI.allowsUnalignedMemoryAccesses(ST->getMemoryVT(), AS)) {
+ Type *Ty = ST->getMemoryVT().getTypeForEVT(*DAG.getContext());
+ unsigned ABIAlignment= TLI.getDataLayout()->getABITypeAlignment(Ty);
+ if (ST->getAlignment() < ABIAlignment)
+ ExpandUnalignedStore(cast<StoreSDNode>(Node), DAG, TLI, this);
+ }
+ break;
+ }
+ case TargetLowering::Custom: {
+ SDValue Res = TLI.LowerOperation(SDValue(Node, 0), DAG);
+ if (Res.getNode())
+ ReplaceNode(SDValue(Node, 0), Res);
+ return;
+ }
+ case TargetLowering::Expand:
+ assert(!StVT.isVector() &&
+ "Vector Stores are handled in LegalizeVectorOps");
+
+ // TRUNCSTORE:i16 i32 -> STORE i16
+ assert(TLI.isTypeLegal(StVT) &&
+ "Do not know how to expand this store!");
+ Value = DAG.getNode(ISD::TRUNCATE, dl, StVT, Value);
+ SDValue Result =
+ DAG.getStore(Chain, dl, Value, Ptr, ST->getPointerInfo(),
+ isVolatile, isNonTemporal, Alignment, TBAAInfo);
+ ReplaceNode(SDValue(Node, 0), Result);
+ break;
+ }
+ }
+ }
+}
+
+void SelectionDAGLegalize::LegalizeLoadOps(SDNode *Node) {
+ LoadSDNode *LD = cast<LoadSDNode>(Node);
+ SDValue Chain = LD->getChain(); // The chain.
+ SDValue Ptr = LD->getBasePtr(); // The base pointer.
+ SDValue Value; // The value returned by the load op.
+ SDLoc dl(Node);
+
+ ISD::LoadExtType ExtType = LD->getExtensionType();
+ if (ExtType == ISD::NON_EXTLOAD) {
+ MVT VT = Node->getSimpleValueType(0);
+ SDValue RVal = SDValue(Node, 0);
+ SDValue RChain = SDValue(Node, 1);
+
+ switch (TLI.getOperationAction(Node->getOpcode(), VT)) {
+ default: llvm_unreachable("This action is not supported yet!");
+ case TargetLowering::Legal: {
+ unsigned AS = LD->getAddressSpace();
+ // If this is an unaligned load and the target doesn't support it,
+ // expand it.
+ if (!TLI.allowsUnalignedMemoryAccesses(LD->getMemoryVT(), AS)) {
+ Type *Ty = LD->getMemoryVT().getTypeForEVT(*DAG.getContext());
+ unsigned ABIAlignment =
+ TLI.getDataLayout()->getABITypeAlignment(Ty);
+ if (LD->getAlignment() < ABIAlignment){
+ ExpandUnalignedLoad(cast<LoadSDNode>(Node), DAG, TLI, RVal, RChain);
+ }
+ }
+ break;
+ }
+ case TargetLowering::Custom: {
+ SDValue Res = TLI.LowerOperation(RVal, DAG);
+ if (Res.getNode()) {
+ RVal = Res;
+ RChain = Res.getValue(1);
+ }
+ break;
+ }
+ case TargetLowering::Promote: {
+ MVT NVT = TLI.getTypeToPromoteTo(Node->getOpcode(), VT);
+ assert(NVT.getSizeInBits() == VT.getSizeInBits() &&
+ "Can only promote loads to same size type");
+
+ SDValue Res = DAG.getLoad(NVT, dl, Chain, Ptr, LD->getMemOperand());
+ RVal = DAG.getNode(ISD::BITCAST, dl, VT, Res);
+ RChain = Res.getValue(1);
+ break;
+ }
+ }
+ if (RChain.getNode() != Node) {
+ assert(RVal.getNode() != Node && "Load must be completely replaced");
+ DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 0), RVal);
+ DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 1), RChain);
+ ReplacedNode(Node);
+ }
+ return;
+ }
+
+ EVT SrcVT = LD->getMemoryVT();
+ unsigned SrcWidth = SrcVT.getSizeInBits();
+ unsigned Alignment = LD->getAlignment();
+ bool isVolatile = LD->isVolatile();
+ bool isNonTemporal = LD->isNonTemporal();
+ const MDNode *TBAAInfo = LD->getTBAAInfo();
+
+ if (SrcWidth != SrcVT.getStoreSizeInBits() &&
+ // Some targets pretend to have an i1 loading operation, and actually
+ // load an i8. This trick is correct for ZEXTLOAD because the top 7
+ // bits are guaranteed to be zero; it helps the optimizers understand
+ // that these bits are zero. It is also useful for EXTLOAD, since it
+ // tells the optimizers that those bits are undefined. It would be
+ // 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)) {
+ // 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();
+ EVT NVT = EVT::getIntegerVT(*DAG.getContext(), NewWidth);
+ SDValue Ch;
+
+ // The extra bits are guaranteed to be zero, since we stored them that
+ // way. A zext load from NVT thus automatically gives zext from SrcVT.
+
+ ISD::LoadExtType NewExtType =
+ ExtType == ISD::ZEXTLOAD ? ISD::ZEXTLOAD : ISD::EXTLOAD;
+
+ SDValue Result =
+ DAG.getExtLoad(NewExtType, dl, Node->getValueType(0),
+ Chain, Ptr, LD->getPointerInfo(),
+ NVT, isVolatile, isNonTemporal, Alignment, TBAAInfo);
+
+ Ch = Result.getValue(1); // The chain.
+
+ if (ExtType == ISD::SEXTLOAD)
+ // Having the top bits zero doesn't help when sign extending.
+ Result = DAG.getNode(ISD::SIGN_EXTEND_INREG, dl,
+ Result.getValueType(),
+ Result, DAG.getValueType(SrcVT));
+ else if (ExtType == ISD::ZEXTLOAD || NVT == Result.getValueType())
+ // All the top bits are guaranteed to be zero - inform the optimizers.
+ Result = DAG.getNode(ISD::AssertZext, dl,
+ Result.getValueType(), Result,
+ DAG.getValueType(SrcVT));
+
+ Value = Result;
+ Chain = Ch;
+ } else if (SrcWidth & (SrcWidth - 1)) {
+ // If not loading a power-of-2 number of bits, expand as two loads.
+ assert(!SrcVT.isVector() && "Unsupported extload!");
+ unsigned RoundWidth = 1 << Log2_32(SrcWidth);
+ assert(RoundWidth < SrcWidth);
+ unsigned ExtraWidth = SrcWidth - RoundWidth;
+ assert(ExtraWidth < RoundWidth);
+ assert(!(RoundWidth % 8) && !(ExtraWidth % 8) &&
+ "Load size not an integral number of bytes!");
+ EVT RoundVT = EVT::getIntegerVT(*DAG.getContext(), RoundWidth);
+ EVT ExtraVT = EVT::getIntegerVT(*DAG.getContext(), ExtraWidth);
+ SDValue Lo, Hi, Ch;
+ unsigned IncrementSize;
+
+ if (TLI.isLittleEndian()) {
+ // EXTLOAD:i24 -> ZEXTLOAD:i16 | (shl EXTLOAD@+2:i8, 16)
+ // Load the bottom RoundWidth bits.
+ Lo = DAG.getExtLoad(ISD::ZEXTLOAD, dl, Node->getValueType(0),
+ Chain, Ptr,
+ LD->getPointerInfo(), RoundVT, isVolatile,
+ isNonTemporal, Alignment, TBAAInfo);
+
+ // Load the remaining ExtraWidth bits.
+ IncrementSize = RoundWidth / 8;
+ Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr,
+ 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);
+
+ // Build a factor node to remember that this load is independent of
+ // the other one.
+ Ch = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Lo.getValue(1),
+ Hi.getValue(1));
+
+ // Move the top bits to the right place.
+ Hi = DAG.getNode(ISD::SHL, dl, Hi.getValueType(), Hi,
+ DAG.getConstant(RoundWidth,
+ TLI.getShiftAmountTy(Hi.getValueType())));
+
+ // Join the hi and lo parts.
+ Value = DAG.getNode(ISD::OR, dl, Node->getValueType(0), Lo, Hi);
+ } else {
+ // Big endian - avoid unaligned loads.
+ // EXTLOAD:i24 -> (shl EXTLOAD:i16, 8) | ZEXTLOAD@+2:i8
+ // Load the top RoundWidth bits.
+ Hi = DAG.getExtLoad(ExtType, dl, Node->getValueType(0), Chain, Ptr,
+ LD->getPointerInfo(), RoundVT, isVolatile,
+ isNonTemporal, Alignment, TBAAInfo);
+
+ // Load the remaining ExtraWidth bits.
+ IncrementSize = RoundWidth / 8;
+ Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr,
+ DAG.getConstant(IncrementSize, Ptr.getValueType()));
+ Lo = DAG.getExtLoad(ISD::ZEXTLOAD,
+ dl, Node->getValueType(0), Chain, Ptr,
+ LD->getPointerInfo().getWithOffset(IncrementSize),
+ ExtraVT, isVolatile, isNonTemporal,
+ MinAlign(Alignment, IncrementSize), TBAAInfo);
+
+ // Build a factor node to remember that this load is independent of
+ // the other one.
+ Ch = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Lo.getValue(1),
+ Hi.getValue(1));
+
+ // Move the top bits to the right place.
+ Hi = DAG.getNode(ISD::SHL, dl, Hi.getValueType(), Hi,
+ DAG.getConstant(ExtraWidth,
+ TLI.getShiftAmountTy(Hi.getValueType())));
+
+ // Join the hi and lo parts.
+ Value = DAG.getNode(ISD::OR, dl, Node->getValueType(0), Lo, Hi);
+ }
+
+ Chain = Ch;
+ } else {
+ bool isCustom = false;
+ switch (TLI.getLoadExtAction(ExtType, SrcVT.getSimpleVT())) {
+ default: llvm_unreachable("This action is not supported yet!");
+ case TargetLowering::Custom:
+ isCustom = true;
+ // FALLTHROUGH
+ case TargetLowering::Legal: {
+ Value = SDValue(Node, 0);
+ Chain = SDValue(Node, 1);
+
+ if (isCustom) {
+ SDValue Res = TLI.LowerOperation(SDValue(Node, 0), DAG);
+ if (Res.getNode()) {
+ Value = Res;
+ Chain = Res.getValue(1);
+ }
+ } else {
+ // If this is an unaligned load and the target doesn't support
+ // 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);
+ }
+ }
+ }
+ 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);
+ 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() &&
+ "Vector Loads are handled in LegalizeVectorOps");
+
+ // FIXME: This does not work for vectors on most targets. Sign-
+ // and zero-extend operations are currently folded into extending
+ // loads, whether they are legal or not, and then we end up here
+ // without any support for legalizing them.
+ assert(ExtType != ISD::EXTLOAD &&
+ "EXTLOAD should always be supported!");
+ // Turn the unsupported load into an EXTLOAD followed by an
+ // explicit zero/sign extend inreg.
+ SDValue Result = DAG.getExtLoad(ISD::EXTLOAD, dl,
+ Node->getValueType(0),
+ Chain, Ptr, SrcVT,
+ LD->getMemOperand());
+ SDValue ValRes;
+ if (ExtType == ISD::SEXTLOAD)
+ ValRes = DAG.getNode(ISD::SIGN_EXTEND_INREG, dl,
+ Result.getValueType(),
+ Result, DAG.getValueType(SrcVT));
+ else
+ ValRes = DAG.getZeroExtendInReg(Result, dl,
+ SrcVT.getScalarType());
+ Value = ValRes;
+ Chain = Result.getValue(1);
+ break;
+ }
+ }
+
+ // Since loads produce two values, make sure to remember that we legalized
+ // both of them.
+ if (Chain.getNode() != Node) {
+ assert(Value.getNode() != Node && "Load must be completely replaced");
+ DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 0), Value);
+ DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 1), Chain);
+ ReplacedNode(Node);
+ }
}
/// LegalizeOp - Return a legal replacement for the given operation, with
if (Node->getOpcode() == ISD::TargetConstant) // Allow illegal target nodes.
return;
- DebugLoc dl = Node->getDebugLoc();
-
for (unsigned i = 0, e = Node->getNumValues(); i != e; ++i)
assert(TLI.getTypeAction(*DAG.getContext(), Node->getValueType(i)) ==
TargetLowering::TypeLegal &&
Node->getOperand(i).getOpcode() == ISD::TargetConstant) &&
"Unexpected illegal type!");
- SDValue Tmp1, Tmp2, Tmp3, Tmp4;
- bool isCustom = false;
-
// Figure out the correct action; the way to query this varies by opcode
TargetLowering::LegalizeAction Action = TargetLowering::Legal;
bool SimpleFinishLegalizing = true;
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:
unsigned CCOperand = Node->getOpcode() == ISD::SELECT_CC ? 4 :
Node->getOpcode() == ISD::SETCC ? 2 : 1;
unsigned CompareOperand = Node->getOpcode() == ISD::BR_CC ? 2 : 0;
- EVT OpVT = Node->getOperand(CompareOperand).getValueType();
+ MVT OpVT = Node->getOperand(CompareOperand).getSimpleValueType();
ISD::CondCode CCCode =
cast<CondCodeSDNode>(Node->getOperand(CCOperand))->get();
Action = TLI.getCondCodeAction(CCCode, OpVT);
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) {
+ // replace ISD::DEBUGTRAP with ISD::TRAP
+ SDValue NewVal;
+ NewVal = DAG.getNode(ISD::TRAP, SDLoc(Node), Node->getVTList(),
+ Node->getOperand(0));
+ ReplaceNode(Node, NewVal.getNode());
+ LegalizeOp(NewVal.getNode());
+ return;
+ }
+ break;
+
default:
if (Node->getOpcode() >= ISD::BUILTIN_OP_END) {
Action = TargetLowering::Legal;
Node->getOperand(1),
Handle.getValue());
}
- break;
- }
-
- 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);
- Node = NewNode;
- }
- switch (Action) {
- case TargetLowering::Legal:
- return;
- case TargetLowering::Custom:
- // FIXME: The handling for custom lowering with multiple results is
- // a complete mess.
- Tmp1 = TLI.LowerOperation(SDValue(Node, 0), DAG);
- if (Tmp1.getNode()) {
- SmallVector<SDValue, 8> ResultVals;
- for (unsigned i = 0, e = Node->getNumValues(); i != e; ++i) {
- if (e == 1)
- ResultVals.push_back(Tmp1);
- else
- ResultVals.push_back(Tmp1.getValue(i));
- }
- if (Tmp1.getNode() != Node || Tmp1.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);
- }
- return;
- }
-
- // FALL THROUGH
- case TargetLowering::Expand:
- ExpandNode(Node);
- return;
- case TargetLowering::Promote:
- PromoteNode(Node);
- return;
- }
- }
-
- switch (Node->getOpcode()) {
- default:
-#ifndef NDEBUG
- dbgs() << "NODE: ";
- Node->dump( &DAG);
- dbgs() << "\n";
-#endif
- llvm_unreachable("Do not know how to legalize this operator!");
-
- case ISD::CALLSEQ_START:
- case ISD::CALLSEQ_END:
- break;
- case ISD::LOAD: {
- LoadSDNode *LD = cast<LoadSDNode>(Node);
- Tmp1 = LD->getChain(); // Legalize the chain.
- Tmp2 = LD->getBasePtr(); // Legalize the base pointer.
-
- ISD::LoadExtType ExtType = LD->getExtensionType();
- if (ExtType == ISD::NON_EXTLOAD) {
- EVT VT = Node->getValueType(0);
- Tmp3 = SDValue(Node, 0);
- Tmp4 = SDValue(Node, 1);
-
- switch (TLI.getOperationAction(Node->getOpcode(), VT)) {
- default: llvm_unreachable("This action is not supported yet!");
- case TargetLowering::Legal:
- // If this is an unaligned load and the target doesn't support it,
- // expand it.
- if (!TLI.allowsUnalignedMemoryAccesses(LD->getMemoryVT())) {
- Type *Ty = LD->getMemoryVT().getTypeForEVT(*DAG.getContext());
- unsigned ABIAlignment = TLI.getTargetData()->getABITypeAlignment(Ty);
- if (LD->getAlignment() < ABIAlignment){
- ExpandUnalignedLoad(cast<LoadSDNode>(Node),
- DAG, TLI, Tmp3, Tmp4);
- }
- }
- break;
- case TargetLowering::Custom:
- Tmp1 = TLI.LowerOperation(Tmp3, DAG);
- if (Tmp1.getNode()) {
- Tmp3 = Tmp1;
- Tmp4 = Tmp1.getValue(1);
- }
- break;
- case TargetLowering::Promote: {
- // Only promote a load of vector type to another.
- assert(VT.isVector() && "Cannot promote this load!");
- // Change base type to a different vector type.
- EVT NVT = TLI.getTypeToPromoteTo(Node->getOpcode(), VT);
-
- Tmp1 = DAG.getLoad(NVT, dl, Tmp1, Tmp2, LD->getPointerInfo(),
- LD->isVolatile(), LD->isNonTemporal(),
- LD->isInvariant(), LD->getAlignment());
- Tmp3 = DAG.getNode(ISD::BITCAST, dl, VT, Tmp1);
- Tmp4 = Tmp1.getValue(1);
- break;
- }
- }
- if (Tmp4.getNode() != Node) {
- assert(Tmp3.getNode() != Node && "Load must be completely replaced");
- DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 0), Tmp3);
- DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 1), Tmp4);
- ReplacedNode(Node);
- }
- return;
- }
-
- EVT SrcVT = LD->getMemoryVT();
- unsigned SrcWidth = SrcVT.getSizeInBits();
- unsigned Alignment = LD->getAlignment();
- bool isVolatile = LD->isVolatile();
- bool isNonTemporal = LD->isNonTemporal();
-
- if (SrcWidth != SrcVT.getStoreSizeInBits() &&
- // Some targets pretend to have an i1 loading operation, and actually
- // load an i8. This trick is correct for ZEXTLOAD because the top 7
- // bits are guaranteed to be zero; it helps the optimizers understand
- // that these bits are zero. It is also useful for EXTLOAD, since it
- // tells the optimizers that those bits are undefined. It would be
- // 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)) {
- // 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();
- EVT NVT = EVT::getIntegerVT(*DAG.getContext(), NewWidth);
- SDValue Ch;
-
- // The extra bits are guaranteed to be zero, since we stored them that
- // way. A zext load from NVT thus automatically gives zext from SrcVT.
-
- ISD::LoadExtType NewExtType =
- ExtType == ISD::ZEXTLOAD ? ISD::ZEXTLOAD : ISD::EXTLOAD;
-
- SDValue Result =
- DAG.getExtLoad(NewExtType, dl, Node->getValueType(0),
- Tmp1, Tmp2, LD->getPointerInfo(),
- NVT, isVolatile, isNonTemporal, Alignment);
-
- Ch = Result.getValue(1); // The chain.
-
- if (ExtType == ISD::SEXTLOAD)
- // Having the top bits zero doesn't help when sign extending.
- Result = DAG.getNode(ISD::SIGN_EXTEND_INREG, dl,
- Result.getValueType(),
- Result, DAG.getValueType(SrcVT));
- else if (ExtType == ISD::ZEXTLOAD || NVT == Result.getValueType())
- // All the top bits are guaranteed to be zero - inform the optimizers.
- Result = DAG.getNode(ISD::AssertZext, dl,
- Result.getValueType(), Result,
- DAG.getValueType(SrcVT));
-
- Tmp1 = Result;
- Tmp2 = Ch;
- } else if (SrcWidth & (SrcWidth - 1)) {
- // If not loading a power-of-2 number of bits, expand as two loads.
- assert(!SrcVT.isVector() && "Unsupported extload!");
- unsigned RoundWidth = 1 << Log2_32(SrcWidth);
- assert(RoundWidth < SrcWidth);
- unsigned ExtraWidth = SrcWidth - RoundWidth;
- assert(ExtraWidth < RoundWidth);
- assert(!(RoundWidth % 8) && !(ExtraWidth % 8) &&
- "Load size not an integral number of bytes!");
- EVT RoundVT = EVT::getIntegerVT(*DAG.getContext(), RoundWidth);
- EVT ExtraVT = EVT::getIntegerVT(*DAG.getContext(), ExtraWidth);
- SDValue Lo, Hi, Ch;
- unsigned IncrementSize;
-
- if (TLI.isLittleEndian()) {
- // EXTLOAD:i24 -> ZEXTLOAD:i16 | (shl EXTLOAD@+2:i8, 16)
- // Load the bottom RoundWidth bits.
- Lo = DAG.getExtLoad(ISD::ZEXTLOAD, dl, Node->getValueType(0),
- Tmp1, Tmp2,
- LD->getPointerInfo(), RoundVT, isVolatile,
- isNonTemporal, Alignment);
-
- // Load the remaining ExtraWidth bits.
- IncrementSize = RoundWidth / 8;
- Tmp2 = DAG.getNode(ISD::ADD, dl, Tmp2.getValueType(), Tmp2,
- DAG.getIntPtrConstant(IncrementSize));
- Hi = DAG.getExtLoad(ExtType, dl, Node->getValueType(0), Tmp1, Tmp2,
- LD->getPointerInfo().getWithOffset(IncrementSize),
- ExtraVT, isVolatile, isNonTemporal,
- MinAlign(Alignment, IncrementSize));
-
- // Build a factor node to remember that this load is independent of
- // the other one.
- Ch = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Lo.getValue(1),
- Hi.getValue(1));
-
- // Move the top bits to the right place.
- Hi = DAG.getNode(ISD::SHL, dl, Hi.getValueType(), Hi,
- DAG.getConstant(RoundWidth,
- TLI.getShiftAmountTy(Hi.getValueType())));
-
- // Join the hi and lo parts.
- Tmp1 = DAG.getNode(ISD::OR, dl, Node->getValueType(0), Lo, Hi);
- } else {
- // Big endian - avoid unaligned loads.
- // EXTLOAD:i24 -> (shl EXTLOAD:i16, 8) | ZEXTLOAD@+2:i8
- // Load the top RoundWidth bits.
- Hi = DAG.getExtLoad(ExtType, dl, Node->getValueType(0), Tmp1, Tmp2,
- LD->getPointerInfo(), RoundVT, isVolatile,
- isNonTemporal, Alignment);
-
- // Load the remaining ExtraWidth bits.
- IncrementSize = RoundWidth / 8;
- Tmp2 = DAG.getNode(ISD::ADD, dl, Tmp2.getValueType(), Tmp2,
- DAG.getIntPtrConstant(IncrementSize));
- Lo = DAG.getExtLoad(ISD::ZEXTLOAD,
- dl, Node->getValueType(0), Tmp1, Tmp2,
- LD->getPointerInfo().getWithOffset(IncrementSize),
- ExtraVT, isVolatile, isNonTemporal,
- MinAlign(Alignment, IncrementSize));
-
- // Build a factor node to remember that this load is independent of
- // the other one.
- Ch = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Lo.getValue(1),
- Hi.getValue(1));
-
- // Move the top bits to the right place.
- Hi = DAG.getNode(ISD::SHL, dl, Hi.getValueType(), Hi,
- DAG.getConstant(ExtraWidth,
- TLI.getShiftAmountTy(Hi.getValueType())));
-
- // Join the hi and lo parts.
- Tmp1 = DAG.getNode(ISD::OR, dl, Node->getValueType(0), Lo, Hi);
- }
-
- Tmp2 = Ch;
- } else {
- switch (TLI.getLoadExtAction(ExtType, SrcVT)) {
- default: llvm_unreachable("This action is not supported yet!");
- case TargetLowering::Custom:
- isCustom = true;
- // FALLTHROUGH
- case TargetLowering::Legal:
- Tmp1 = SDValue(Node, 0);
- Tmp2 = SDValue(Node, 1);
-
- if (isCustom) {
- Tmp3 = TLI.LowerOperation(SDValue(Node, 0), DAG);
- if (Tmp3.getNode()) {
- Tmp1 = Tmp3;
- Tmp2 = Tmp3.getValue(1);
- }
- } else {
- // If this is an unaligned load and the target doesn't support it,
- // expand it.
- if (!TLI.allowsUnalignedMemoryAccesses(LD->getMemoryVT())) {
- Type *Ty =
- LD->getMemoryVT().getTypeForEVT(*DAG.getContext());
- unsigned ABIAlignment =
- TLI.getTargetData()->getABITypeAlignment(Ty);
- if (LD->getAlignment() < ABIAlignment){
- ExpandUnalignedLoad(cast<LoadSDNode>(Node),
- DAG, TLI, Tmp1, Tmp2);
- }
- }
- }
- break;
- case TargetLowering::Expand:
- if (!TLI.isLoadExtLegal(ISD::EXTLOAD, SrcVT) && TLI.isTypeLegal(SrcVT)) {
- SDValue Load = DAG.getLoad(SrcVT, dl, Tmp1, Tmp2,
- LD->getPointerInfo(),
- LD->isVolatile(), LD->isNonTemporal(),
- LD->isInvariant(), LD->getAlignment());
- unsigned ExtendOp;
- switch (ExtType) {
- case ISD::EXTLOAD:
- ExtendOp = (SrcVT.isFloatingPoint() ?
- ISD::FP_EXTEND : ISD::ANY_EXTEND);
- break;
- case ISD::SEXTLOAD: ExtendOp = ISD::SIGN_EXTEND; break;
- case ISD::ZEXTLOAD: ExtendOp = ISD::ZERO_EXTEND; break;
- default: llvm_unreachable("Unexpected extend load type!");
- }
- Tmp1 = DAG.getNode(ExtendOp, dl, Node->getValueType(0), Load);
- Tmp2 = Load.getValue(1);
- break;
- }
-
- assert(!SrcVT.isVector() &&
- "Vector Loads are handled in LegalizeVectorOps");
-
- // FIXME: This does not work for vectors on most targets. Sign- and
- // zero-extend operations are currently folded into extending loads,
- // whether they are legal or not, and then we end up here without any
- // support for legalizing them.
- assert(ExtType != ISD::EXTLOAD &&
- "EXTLOAD should always be supported!");
- // Turn the unsupported load into an EXTLOAD followed by an explicit
- // zero/sign extend inreg.
- SDValue Result = DAG.getExtLoad(ISD::EXTLOAD, dl, Node->getValueType(0),
- Tmp1, Tmp2, LD->getPointerInfo(), SrcVT,
- LD->isVolatile(), LD->isNonTemporal(),
- LD->getAlignment());
- SDValue ValRes;
- if (ExtType == ISD::SEXTLOAD)
- ValRes = DAG.getNode(ISD::SIGN_EXTEND_INREG, dl,
- Result.getValueType(),
- Result, DAG.getValueType(SrcVT));
- else
- ValRes = DAG.getZeroExtendInReg(Result, dl, SrcVT.getScalarType());
- Tmp1 = ValRes;
- Tmp2 = Result.getValue(1);
- break;
- }
- }
-
- // Since loads produce two values, make sure to remember that we legalized
- // both of them.
- if (Tmp2.getNode() != Node) {
- assert(Tmp1.getNode() != Node && "Load must be completely replaced");
- DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 0), Tmp1);
- DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 1), Tmp2);
- ReplacedNode(Node);
- }
- break;
- }
- case ISD::STORE: {
- StoreSDNode *ST = cast<StoreSDNode>(Node);
- Tmp1 = ST->getChain();
- Tmp2 = ST->getBasePtr();
- unsigned Alignment = ST->getAlignment();
- bool isVolatile = ST->isVolatile();
- bool isNonTemporal = ST->isNonTemporal();
-
- if (!ST->isTruncatingStore()) {
- if (SDNode *OptStore = OptimizeFloatStore(ST).getNode()) {
- ReplaceNode(ST, OptStore);
- break;
- }
-
- {
- Tmp3 = ST->getValue();
- EVT VT = Tmp3.getValueType();
- switch (TLI.getOperationAction(ISD::STORE, VT)) {
- default: llvm_unreachable("This action is not supported yet!");
- case TargetLowering::Legal:
- // If this is an unaligned store and the target doesn't support it,
- // expand it.
- if (!TLI.allowsUnalignedMemoryAccesses(ST->getMemoryVT())) {
- Type *Ty = ST->getMemoryVT().getTypeForEVT(*DAG.getContext());
- unsigned ABIAlignment= TLI.getTargetData()->getABITypeAlignment(Ty);
- if (ST->getAlignment() < ABIAlignment)
- ExpandUnalignedStore(cast<StoreSDNode>(Node),
- DAG, TLI, this);
- }
- break;
- case TargetLowering::Custom:
- Tmp1 = TLI.LowerOperation(SDValue(Node, 0), DAG);
- if (Tmp1.getNode())
- ReplaceNode(SDValue(Node, 0), Tmp1);
- break;
- case TargetLowering::Promote: {
- assert(VT.isVector() && "Unknown legal promote case!");
- Tmp3 = DAG.getNode(ISD::BITCAST, dl,
- TLI.getTypeToPromoteTo(ISD::STORE, VT), Tmp3);
- SDValue Result =
- DAG.getStore(Tmp1, dl, Tmp3, Tmp2,
- ST->getPointerInfo(), isVolatile,
- isNonTemporal, Alignment);
- ReplaceNode(SDValue(Node, 0), Result);
- break;
- }
- }
- break;
- }
- } else {
- Tmp3 = ST->getValue();
-
- EVT StVT = ST->getMemoryVT();
- unsigned StWidth = StVT.getSizeInBits();
-
- if (StWidth != StVT.getStoreSizeInBits()) {
- // Promote to a byte-sized store with upper bits zero if not
- // storing an integral number of bytes. For example, promote
- // TRUNCSTORE:i1 X -> TRUNCSTORE:i8 (and X, 1)
- EVT NVT = EVT::getIntegerVT(*DAG.getContext(),
- StVT.getStoreSizeInBits());
- Tmp3 = DAG.getZeroExtendInReg(Tmp3, dl, StVT);
- SDValue Result =
- DAG.getTruncStore(Tmp1, dl, Tmp3, Tmp2, ST->getPointerInfo(),
- NVT, isVolatile, isNonTemporal, Alignment);
- ReplaceNode(SDValue(Node, 0), Result);
- } else if (StWidth & (StWidth - 1)) {
- // If not storing a power-of-2 number of bits, expand as two stores.
- assert(!StVT.isVector() && "Unsupported truncstore!");
- unsigned RoundWidth = 1 << Log2_32(StWidth);
- assert(RoundWidth < StWidth);
- unsigned ExtraWidth = StWidth - RoundWidth;
- assert(ExtraWidth < RoundWidth);
- assert(!(RoundWidth % 8) && !(ExtraWidth % 8) &&
- "Store size not an integral number of bytes!");
- EVT RoundVT = EVT::getIntegerVT(*DAG.getContext(), RoundWidth);
- EVT ExtraVT = EVT::getIntegerVT(*DAG.getContext(), ExtraWidth);
- SDValue Lo, Hi;
- unsigned IncrementSize;
-
- if (TLI.isLittleEndian()) {
- // TRUNCSTORE:i24 X -> TRUNCSTORE:i16 X, TRUNCSTORE@+2:i8 (srl X, 16)
- // Store the bottom RoundWidth bits.
- Lo = DAG.getTruncStore(Tmp1, dl, Tmp3, Tmp2, ST->getPointerInfo(),
- RoundVT,
- isVolatile, isNonTemporal, Alignment);
-
- // Store the remaining ExtraWidth bits.
- IncrementSize = RoundWidth / 8;
- Tmp2 = DAG.getNode(ISD::ADD, dl, Tmp2.getValueType(), Tmp2,
- DAG.getIntPtrConstant(IncrementSize));
- Hi = DAG.getNode(ISD::SRL, dl, Tmp3.getValueType(), Tmp3,
- DAG.getConstant(RoundWidth,
- TLI.getShiftAmountTy(Tmp3.getValueType())));
- Hi = DAG.getTruncStore(Tmp1, dl, Hi, Tmp2,
- ST->getPointerInfo().getWithOffset(IncrementSize),
- ExtraVT, isVolatile, isNonTemporal,
- MinAlign(Alignment, IncrementSize));
- } else {
- // Big endian - avoid unaligned stores.
- // TRUNCSTORE:i24 X -> TRUNCSTORE:i16 (srl X, 8), TRUNCSTORE@+2:i8 X
- // Store the top RoundWidth bits.
- Hi = DAG.getNode(ISD::SRL, dl, Tmp3.getValueType(), Tmp3,
- DAG.getConstant(ExtraWidth,
- TLI.getShiftAmountTy(Tmp3.getValueType())));
- Hi = DAG.getTruncStore(Tmp1, dl, Hi, Tmp2, ST->getPointerInfo(),
- RoundVT, isVolatile, isNonTemporal, Alignment);
-
- // Store the remaining ExtraWidth bits.
- IncrementSize = RoundWidth / 8;
- Tmp2 = DAG.getNode(ISD::ADD, dl, Tmp2.getValueType(), Tmp2,
- DAG.getIntPtrConstant(IncrementSize));
- Lo = DAG.getTruncStore(Tmp1, dl, Tmp3, Tmp2,
- ST->getPointerInfo().getWithOffset(IncrementSize),
- ExtraVT, isVolatile, isNonTemporal,
- MinAlign(Alignment, IncrementSize));
- }
-
- // The order of the stores doesn't matter.
- SDValue Result = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Lo, Hi);
- ReplaceNode(SDValue(Node, 0), Result);
- } else {
- switch (TLI.getTruncStoreAction(ST->getValue().getValueType(), StVT)) {
- default: llvm_unreachable("This action is not supported yet!");
- case TargetLowering::Legal:
- // If this is an unaligned store and the target doesn't support it,
- // expand it.
- if (!TLI.allowsUnalignedMemoryAccesses(ST->getMemoryVT())) {
- Type *Ty = ST->getMemoryVT().getTypeForEVT(*DAG.getContext());
- unsigned ABIAlignment= TLI.getTargetData()->getABITypeAlignment(Ty);
- if (ST->getAlignment() < ABIAlignment)
- ExpandUnalignedStore(cast<StoreSDNode>(Node), DAG, TLI, this);
- }
- break;
- case TargetLowering::Custom:
- Tmp1 = TLI.LowerOperation(SDValue(Node, 0), DAG);
- if (Tmp1.getNode())
- ReplaceNode(SDValue(Node, 0), Tmp1);
- break;
- case TargetLowering::Expand:
- assert(!StVT.isVector() &&
- "Vector Stores are handled in LegalizeVectorOps");
-
- // TRUNCSTORE:i16 i32 -> STORE i16
- assert(TLI.isTypeLegal(StVT) && "Do not know how to expand this store!");
- Tmp3 = DAG.getNode(ISD::TRUNCATE, dl, StVT, Tmp3);
- SDValue Result =
- DAG.getStore(Tmp1, dl, Tmp3, Tmp2, ST->getPointerInfo(),
- isVolatile, isNonTemporal, Alignment);
- ReplaceNode(SDValue(Node, 0), Result);
- break;
+ break;
+ }
+
+ 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);
+ Node = NewNode;
+ }
+ switch (Action) {
+ case TargetLowering::Legal:
+ return;
+ case TargetLowering::Custom: {
+ // FIXME: The handling for custom lowering with multiple results is
+ // 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);
}
+ return;
}
}
+ // FALL THROUGH
+ case TargetLowering::Expand:
+ ExpandNode(Node);
+ return;
+ case TargetLowering::Promote:
+ PromoteNode(Node);
+ return;
+ }
+ }
+
+ switch (Node->getOpcode()) {
+ default:
+#ifndef NDEBUG
+ dbgs() << "NODE: ";
+ Node->dump( &DAG);
+ dbgs() << "\n";
+#endif
+ llvm_unreachable("Do not know how to legalize this operator!");
+
+ case ISD::CALLSEQ_START:
+ case ISD::CALLSEQ_END:
break;
+ case ISD::LOAD: {
+ return LegalizeLoadOps(Node);
+ }
+ case ISD::STORE: {
+ return LegalizeStoreOps(Node);
}
}
}
SDValue SelectionDAGLegalize::ExpandExtractFromVectorThroughStack(SDValue Op) {
SDValue Vec = Op.getOperand(0);
SDValue Idx = Op.getOperand(1);
- DebugLoc dl = Op.getDebugLoc();
- // 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);
+ SDLoc dl(Op);
+
+ // 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.getNode(ISD::MUL, dl, Idx.getValueType(), Idx,
DAG.getConstant(EltSize, Idx.getValueType()));
- if (Idx.getValueType().bitsGT(TLI.getPointerTy()))
- Idx = DAG.getNode(ISD::TRUNCATE, dl, TLI.getPointerTy(), Idx);
- else
- Idx = DAG.getNode(ISD::ZERO_EXTEND, dl, TLI.getPointerTy(), Idx);
-
+ Idx = DAG.getZExtOrTrunc(Idx, dl, TLI.getPointerTy());
StackPtr = DAG.getNode(ISD::ADD, dl, Idx.getValueType(), Idx, StackPtr);
if (Op.getValueType().isVector())
SDValue Vec = Op.getOperand(0);
SDValue Part = Op.getOperand(1);
SDValue Idx = Op.getOperand(2);
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
// Store the value to a temporary stack slot, then LOAD the returned part.
Idx = DAG.getNode(ISD::MUL, dl, Idx.getValueType(), Idx,
DAG.getConstant(EltSize, Idx.getValueType()));
-
- if (Idx.getValueType().bitsGT(TLI.getPointerTy()))
- Idx = DAG.getNode(ISD::TRUNCATE, dl, TLI.getPointerTy(), Idx);
- else
- Idx = DAG.getNode(ISD::ZERO_EXTEND, dl, TLI.getPointerTy(), Idx);
+ Idx = DAG.getZExtOrTrunc(Idx, dl, TLI.getPointerTy());
SDValue SubStackPtr = DAG.getNode(ISD::ADD, dl, Idx.getValueType(), Idx,
StackPtr);
// Create the stack frame object.
EVT VT = Node->getValueType(0);
EVT EltVT = VT.getVectorElementType();
- DebugLoc dl = Node->getDebugLoc();
+ SDLoc dl(Node);
SDValue FIPtr = DAG.CreateStackTemporary(VT);
int FI = cast<FrameIndexSDNode>(FIPtr.getNode())->getIndex();
MachinePointerInfo PtrInfo = MachinePointerInfo::getFixedStack(FI);
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();
// Result is a load from the stack slot.
- return DAG.getLoad(VT, dl, StoreChain, FIPtr, PtrInfo,
+ return DAG.getLoad(VT, dl, StoreChain, FIPtr, PtrInfo,
false, false, false, 0);
}
SDValue SelectionDAGLegalize::ExpandFCOPYSIGN(SDNode* Node) {
- DebugLoc dl = Node->getDebugLoc();
+ SDLoc dl(Node);
SDValue Tmp1 = Node->getOperand(0);
SDValue Tmp2 = Node->getOperand(1);
// the pointer so that the loaded integer will contain the sign bit.
unsigned Strides = (FloatVT.getSizeInBits()-1)/LoadTy.getSizeInBits();
unsigned ByteOffset = (Strides * LoadTy.getSizeInBits()) / 8;
- LoadPtr = DAG.getNode(ISD::ADD, dl, LoadPtr.getValueType(),
- LoadPtr, DAG.getIntPtrConstant(ByteOffset));
+ LoadPtr = DAG.getNode(ISD::ADD, dl, LoadPtr.getValueType(), LoadPtr,
+ DAG.getConstant(ByteOffset, LoadPtr.getValueType()));
// Load a legal integer containing the sign bit.
SignBit = DAG.getLoad(LoadTy, dl, Ch, LoadPtr, MachinePointerInfo(),
false, false, false, 0);
}
}
// Now get the sign bit proper, by seeing whether the value is negative.
- SignBit = DAG.getSetCC(dl, TLI.getSetCCResultType(SignBit.getValueType()),
+ SignBit = DAG.getSetCC(dl, getSetCCResultType(SignBit.getValueType()),
SignBit, DAG.getConstant(0, SignBit.getValueType()),
ISD::SETLT);
// Get the absolute value of the result.
SDValue AbsVal = DAG.getNode(ISD::FABS, dl, Tmp1.getValueType(), Tmp1);
// Select between the nabs and abs value based on the sign bit of
// the input.
- return DAG.getNode(ISD::SELECT, dl, AbsVal.getValueType(), SignBit,
- DAG.getNode(ISD::FNEG, dl, AbsVal.getValueType(), AbsVal),
- AbsVal);
+ return DAG.getSelect(dl, AbsVal.getValueType(), SignBit,
+ DAG.getNode(ISD::FNEG, dl, AbsVal.getValueType(), AbsVal),
+ AbsVal);
}
void SelectionDAGLegalize::ExpandDYNAMIC_STACKALLOC(SDNode* Node,
unsigned SPReg = TLI.getStackPointerRegisterToSaveRestore();
assert(SPReg && "Target cannot require DYNAMIC_STACKALLOC expansion and"
" not tell us which reg is the stack pointer!");
- DebugLoc dl = Node->getDebugLoc();
+ SDLoc dl(Node);
EVT VT = Node->getValueType(0);
SDValue Tmp1 = SDValue(Node, 0);
SDValue Tmp2 = SDValue(Node, 1);
// Chain the dynamic stack allocation so that it doesn't modify the stack
// pointer when other instructions are using the stack.
- Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(0, true));
+ Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(0, true),
+ SDLoc(Node));
SDValue Size = Tmp2.getOperand(1);
SDValue SP = DAG.getCopyFromReg(Chain, dl, SPReg, VT);
Chain = SP.getValue(1);
unsigned Align = cast<ConstantSDNode>(Tmp3)->getZExtValue();
unsigned StackAlign = TM.getFrameLowering()->getStackAlignment();
- if (Align > StackAlign)
- SP = DAG.getNode(ISD::AND, dl, VT, SP,
- DAG.getConstant(-(uint64_t)Align, VT));
Tmp1 = DAG.getNode(ISD::SUB, dl, VT, SP, Size); // Value
+ if (Align > StackAlign)
+ Tmp1 = DAG.getNode(ISD::AND, dl, VT, Tmp1,
+ DAG.getConstant(-(uint64_t)Align, VT));
Chain = DAG.getCopyToReg(Chain, dl, SPReg, Tmp1); // Output chain
Tmp2 = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(0, true),
- DAG.getIntPtrConstant(0, true), SDValue());
+ DAG.getIntPtrConstant(0, true), SDValue(),
+ SDLoc(Node));
Results.push_back(Tmp1);
Results.push_back(Tmp2);
}
/// LegalizeSetCCCondCode - Legalize a SETCC with given LHS and RHS and
-/// condition code CC on the current target. This routine expands SETCC with
-/// illegal condition code into AND / OR of multiple SETCC values.
-void SelectionDAGLegalize::LegalizeSetCCCondCode(EVT VT,
+/// 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
+/// will be set to false.
+///
+/// If the SETCC has been legalized by using getSetCCSwappedOperands(),
+/// then the values of LHS and RHS will be swapped, CC will be set to the
+/// new condition, and NeedInvert will be set to false.
+///
+/// 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::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,
SDValue &LHS, SDValue &RHS,
SDValue &CC,
- DebugLoc dl) {
- EVT OpVT = LHS.getValueType();
+ bool &NeedInvert,
+ SDLoc dl) {
+ MVT OpVT = LHS.getSimpleValueType();
ISD::CondCode CCCode = cast<CondCodeSDNode>(CC)->get();
+ NeedInvert = false;
switch (TLI.getCondCodeAction(CCCode, OpVT)) {
default: llvm_unreachable("Unknown condition code action!");
case TargetLowering::Legal:
// Nothing to do.
break;
case TargetLowering::Expand: {
+ ISD::CondCode InvCC = ISD::getSetCCSwappedOperands(CCCode);
+ if (TLI.isCondCodeLegal(InvCC, OpVT)) {
+ std::swap(LHS, RHS);
+ CC = DAG.getCondCode(InvCC);
+ return true;
+ }
ISD::CondCode CC1 = ISD::SETCC_INVALID, CC2 = ISD::SETCC_INVALID;
unsigned Opc = 0;
switch (CCCode) {
default: llvm_unreachable("Don't know how to expand this condition!");
- case ISD::SETOEQ: CC1 = ISD::SETEQ; CC2 = ISD::SETO; Opc = ISD::AND; break;
- case ISD::SETOGT: CC1 = ISD::SETGT; CC2 = ISD::SETO; Opc = ISD::AND; break;
- case ISD::SETOGE: CC1 = ISD::SETGE; CC2 = ISD::SETO; Opc = ISD::AND; break;
- case ISD::SETOLT: CC1 = ISD::SETLT; CC2 = ISD::SETO; Opc = ISD::AND; break;
- case ISD::SETOLE: CC1 = ISD::SETLE; CC2 = ISD::SETO; Opc = ISD::AND; break;
- case ISD::SETONE: CC1 = ISD::SETNE; CC2 = ISD::SETO; Opc = ISD::AND; break;
- case ISD::SETUEQ: CC1 = ISD::SETEQ; CC2 = ISD::SETUO; Opc = ISD::OR; break;
- case ISD::SETUGT: CC1 = ISD::SETGT; CC2 = ISD::SETUO; Opc = ISD::OR; break;
- case ISD::SETUGE: CC1 = ISD::SETGE; CC2 = ISD::SETUO; Opc = ISD::OR; break;
- case ISD::SETULT: CC1 = ISD::SETLT; CC2 = ISD::SETUO; Opc = ISD::OR; break;
- case ISD::SETULE: CC1 = ISD::SETLE; CC2 = ISD::SETUO; Opc = ISD::OR; break;
- case ISD::SETUNE: CC1 = ISD::SETNE; CC2 = ISD::SETUO; Opc = ISD::OR; break;
- // FIXME: Implement more expansions.
- }
-
- SDValue SetCC1 = DAG.getSetCC(dl, VT, LHS, RHS, CC1);
- SDValue SetCC2 = DAG.getSetCC(dl, VT, LHS, RHS, CC2);
+ case ISD::SETO:
+ assert(TLI.getCondCodeAction(ISD::SETOEQ, OpVT)
+ == TargetLowering::Legal
+ && "If SETO is expanded, SETOEQ must be legal!");
+ CC1 = ISD::SETOEQ; CC2 = ISD::SETOEQ; Opc = ISD::AND; break;
+ case ISD::SETUO:
+ assert(TLI.getCondCodeAction(ISD::SETUNE, OpVT)
+ == TargetLowering::Legal
+ && "If SETUO is expanded, SETUNE must be legal!");
+ CC1 = ISD::SETUNE; CC2 = ISD::SETUNE; Opc = ISD::OR; break;
+ case ISD::SETOEQ:
+ case ISD::SETOGT:
+ case ISD::SETOGE:
+ case ISD::SETOLT:
+ case ISD::SETOLE:
+ case ISD::SETONE:
+ case ISD::SETUEQ:
+ case ISD::SETUNE:
+ case ISD::SETUGT:
+ case ISD::SETUGE:
+ case ISD::SETULT:
+ case ISD::SETULE:
+ // If we are floating point, assign and break, otherwise fall through.
+ if (!OpVT.isInteger()) {
+ // We can use the 4th bit to tell if we are the unordered
+ // or ordered version of the opcode.
+ CC2 = ((unsigned)CCCode & 0x8U) ? ISD::SETUO : ISD::SETO;
+ Opc = ((unsigned)CCCode & 0x8U) ? ISD::OR : ISD::AND;
+ CC1 = (ISD::CondCode)(((int)CCCode & 0x7) | 0x10);
+ break;
+ }
+ // Fallthrough if we are unsigned integer.
+ case ISD::SETLE:
+ case ISD::SETGT:
+ case ISD::SETGE:
+ case ISD::SETLT:
+ // We only support using the inverted operation, which is computed above
+ // and not a different manner of supporting expanding these cases.
+ llvm_unreachable("Don't know how to expand this condition!");
+ case ISD::SETNE:
+ case ISD::SETEQ:
+ // Try inverting the result of the inverse condition.
+ InvCC = CCCode == ISD::SETEQ ? ISD::SETNE : ISD::SETEQ;
+ if (TLI.isCondCodeLegal(InvCC, OpVT)) {
+ CC = DAG.getCondCode(InvCC);
+ NeedInvert = true;
+ return true;
+ }
+ // If inverting the condition didn't work then we have no means to expand
+ // the condition.
+ llvm_unreachable("Don't know how to expand this condition!");
+ }
+
+ SDValue SetCC1, SetCC2;
+ if (CCCode != ISD::SETO && CCCode != ISD::SETUO) {
+ // If we aren't the ordered or unorder operation,
+ // then the pattern is (LHS CC1 RHS) Opc (LHS CC2 RHS).
+ SetCC1 = DAG.getSetCC(dl, VT, LHS, RHS, CC1);
+ SetCC2 = DAG.getSetCC(dl, VT, LHS, RHS, CC2);
+ } else {
+ // Otherwise, the pattern is (LHS CC1 LHS) Opc (RHS CC2 RHS)
+ SetCC1 = DAG.getSetCC(dl, VT, LHS, LHS, CC1);
+ SetCC2 = DAG.getSetCC(dl, VT, RHS, RHS, CC2);
+ }
LHS = DAG.getNode(Opc, dl, VT, SetCC1, SetCC2);
RHS = SDValue();
CC = SDValue();
- break;
+ return true;
}
}
+ return false;
}
/// EmitStackConvert - Emit a store/load combination to the stack. This stores
SDValue SelectionDAGLegalize::EmitStackConvert(SDValue SrcOp,
EVT SlotVT,
EVT DestVT,
- DebugLoc dl) {
+ SDLoc dl) {
// Create the stack frame object.
unsigned SrcAlign =
- TLI.getTargetData()->getPrefTypeAlignment(SrcOp.getValueType().
+ TLI.getDataLayout()->getPrefTypeAlignment(SrcOp.getValueType().
getTypeForEVT(*DAG.getContext()));
SDValue FIPtr = DAG.CreateStackTemporary(SlotVT, SrcAlign);
unsigned SlotSize = SlotVT.getSizeInBits();
unsigned DestSize = DestVT.getSizeInBits();
Type *DestType = DestVT.getTypeForEVT(*DAG.getContext());
- unsigned DestAlign = TLI.getTargetData()->getPrefTypeAlignment(DestType);
+ unsigned DestAlign = TLI.getDataLayout()->getPrefTypeAlignment(DestType);
// Emit a store to the stack slot. Use a truncstore if the input value is
// later than DestVT.
}
SDValue SelectionDAGLegalize::ExpandSCALAR_TO_VECTOR(SDNode *Node) {
- DebugLoc dl = Node->getDebugLoc();
+ SDLoc dl(Node);
// Create a vector sized/aligned stack slot, store the value to element #0,
// then load the whole vector back out.
SDValue StackPtr = DAG.CreateStackTemporary(Node->getValueType(0));
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, 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
/// support the operation, but do support the resultant vector type.
SDValue SelectionDAGLegalize::ExpandBUILD_VECTOR(SDNode *Node) {
unsigned NumElems = Node->getNumOperands();
SDValue Value1, Value2;
- DebugLoc dl = Node->getDebugLoc();
+ SDLoc dl(Node);
EVT VT = Node->getValueType(0);
EVT OpVT = Node->getOperand(0).getValueType();
EVT EltVT = VT.getVectorElementType();
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;
}
}
// isTailCall may be true since the callee does not reference caller stack
// frame. Check if it's in the right position.
SDValue TCChain = InChain;
- bool isTailCall = isInTailCallPosition(DAG, Node, TCChain, TLI);
+ bool isTailCall = TLI.isInTailCallPosition(DAG, Node, TCChain);
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, Node->getDebugLoc());
- 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).
/// and returning a result of type RetVT.
SDValue SelectionDAGLegalize::ExpandLibCall(RTLIB::Libcall LC, EVT RetVT,
const SDValue *Ops, unsigned NumOps,
- bool isSigned, DebugLoc dl) {
+ bool isSigned, SDLoc dl) {
TargetLowering::ArgListTy Args;
Args.reserve(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;
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, Node->getDebugLoc());
+
+ 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;
RTLIB::Libcall Call_F32,
RTLIB::Libcall Call_F64,
RTLIB::Libcall Call_F80,
+ RTLIB::Libcall Call_F128,
RTLIB::Libcall Call_PPCF128) {
RTLIB::Libcall LC;
- switch (Node->getValueType(0).getSimpleVT().SimpleTy) {
+ switch (Node->getSimpleValueType(0).SimpleTy) {
default: llvm_unreachable("Unexpected request for libcall!");
case MVT::f32: LC = Call_F32; break;
case MVT::f64: LC = Call_F64; break;
case MVT::f80: LC = Call_F80; break;
+ case MVT::f128: LC = Call_F128; break;
case MVT::ppcf128: LC = Call_PPCF128; break;
}
return ExpandLibCall(LC, Node, false);
RTLIB::Libcall Call_I64,
RTLIB::Libcall Call_I128) {
RTLIB::Libcall LC;
- switch (Node->getValueType(0).getSimpleVT().SimpleTy) {
+ switch (Node->getSimpleValueType(0).SimpleTy) {
default: llvm_unreachable("Unexpected request for libcall!");
case MVT::i8: LC = Call_I8; break;
case MVT::i16: LC = Call_I16; break;
static bool isDivRemLibcallAvailable(SDNode *Node, bool isSigned,
const TargetLowering &TLI) {
RTLIB::Libcall LC;
- switch (Node->getValueType(0).getSimpleVT().SimpleTy) {
+ switch (Node->getSimpleValueType(0).SimpleTy) {
default: llvm_unreachable("Unexpected request for libcall!");
case MVT::i8: LC= isSigned ? RTLIB::SDIVREM_I8 : RTLIB::UDIVREM_I8; break;
case MVT::i16: LC= isSigned ? RTLIB::SDIVREM_I16 : RTLIB::UDIVREM_I16; break;
case MVT::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
+/// useDivRem - Only issue divrem libcall if both quotient and remainder are
/// needed.
-static bool UseDivRem(SDNode *Node, bool isSigned, bool isDIV) {
+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;
unsigned OtherOpcode = 0;
if (isSigned)
OtherOpcode = isDIV ? ISD::SREM : ISD::SDIV;
SDNode *User = *UI;
if (User == Node)
continue;
- if (User->getOpcode() == OtherOpcode &&
+ if ((User->getOpcode() == OtherOpcode || User->getOpcode() == DivRemOpc) &&
User->getOperand(0) == Op0 &&
User->getOperand(1) == Op1)
return true;
bool isSigned = Opcode == ISD::SDIVREM;
RTLIB::Libcall LC;
- switch (Node->getValueType(0).getSimpleVT().SimpleTy) {
+ switch (Node->getSimpleValueType(0).SimpleTy) {
default: llvm_unreachable("Unexpected request for libcall!");
case MVT::i8: LC= isSigned ? RTLIB::SDIVREM_I8 : RTLIB::UDIVREM_I8; break;
case MVT::i16: LC= isSigned ? RTLIB::SDIVREM_I16 : RTLIB::UDIVREM_I16; break;
SDValue Callee = DAG.getExternalSymbol(TLI.getLibcallName(LC),
TLI.getPointerTy());
- DebugLoc dl = Node->getDebugLoc();
- TargetLowering::
- CallLoweringInfo CLI(InChain, RetTy, isSigned, !isSigned, false, false,
- 0, TLI.getLibcallCallingConv(LC), /*isTailCall=*/false,
- /*doesNotReturn=*/false, /*isReturnValueUsed=*/true,
- Callee, Args, DAG, dl);
+ SDLoc dl(Node);
+ 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.
+static bool isSinCosLibcallAvailable(SDNode *Node, const TargetLowering &TLI) {
+ RTLIB::Libcall LC;
+ switch (Node->getSimpleValueType(0).SimpleTy) {
+ default: llvm_unreachable("Unexpected request for libcall!");
+ case MVT::f32: LC = RTLIB::SINCOS_F32; break;
+ case MVT::f64: LC = RTLIB::SINCOS_F64; break;
+ case MVT::f80: LC = RTLIB::SINCOS_F80; break;
+ case MVT::f128: LC = RTLIB::SINCOS_F128; break;
+ case MVT::ppcf128: LC = RTLIB::SINCOS_PPCF128; break;
+ }
+ return TLI.getLibcallName(LC) != nullptr;
+}
+
+/// canCombineSinCosLibcall - 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 false;
+ // GNU sin/cos functions set errno while sincos does not. Therefore
+ // combining sin and cos is only safe if unsafe-fpmath is enabled.
+ bool isGNU = Triple(TM.getTargetTriple()).getEnvironment() == Triple::GNU;
+ if (isGNU && !TM.Options.UnsafeFPMath)
+ return false;
+ return true;
+}
+
+/// useSinCos - 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;
+
+ SDValue Op0 = Node->getOperand(0);
+ for (SDNode::use_iterator UI = Op0.getNode()->use_begin(),
+ UE = Op0.getNode()->use_end(); UI != UE; ++UI) {
+ SDNode *User = *UI;
+ if (User == Node)
+ continue;
+ // The other user might have been turned into sincos already.
+ if (User->getOpcode() == OtherOpcode || User->getOpcode() == ISD::FSINCOS)
+ return true;
+ }
+ return false;
+}
+
+/// ExpandSinCosLibCall - Issue libcalls to sincos to compute sin / cos
+/// pairs.
+void
+SelectionDAGLegalize::ExpandSinCosLibCall(SDNode *Node,
+ SmallVectorImpl<SDValue> &Results) {
+ RTLIB::Libcall LC;
+ switch (Node->getSimpleValueType(0).SimpleTy) {
+ default: llvm_unreachable("Unexpected request for libcall!");
+ case MVT::f32: LC = RTLIB::SINCOS_F32; break;
+ case MVT::f64: LC = RTLIB::SINCOS_F64; break;
+ case MVT::f80: LC = RTLIB::SINCOS_F80; break;
+ case MVT::f128: LC = RTLIB::SINCOS_F128; break;
+ case MVT::ppcf128: LC = RTLIB::SINCOS_PPCF128; break;
+ }
+
+ // The input chain to this libcall is the entry node of the function.
+ // Legalizing the call will automatically add the previous call to the
+ // dependence.
+ SDValue InChain = DAG.getEntryNode();
+
+ EVT RetVT = Node->getValueType(0);
+ Type *RetTy = RetVT.getTypeForEVT(*DAG.getContext());
+
+ TargetLowering::ArgListTy Args;
+ TargetLowering::ArgListEntry Entry;
+
+ // Pass the argument.
+ Entry.Node = Node->getOperand(0);
+ Entry.Ty = RetTy;
+ Entry.isSExt = false;
+ Entry.isZExt = false;
+ Args.push_back(Entry);
+
+ // Pass the return address of sin.
+ SDValue SinPtr = DAG.CreateStackTemporary(RetVT);
+ Entry.Node = SinPtr;
+ Entry.Ty = RetTy->getPointerTo();
+ Entry.isSExt = false;
+ Entry.isZExt = false;
+ Args.push_back(Entry);
+
+ // Also pass the return address of the cos.
+ SDValue CosPtr = DAG.CreateStackTemporary(RetVT);
+ Entry.Node = CosPtr;
+ Entry.Ty = RetTy->getPointerTo();
+ Entry.isSExt = false;
+ Entry.isZExt = false;
+ Args.push_back(Entry);
+
+ SDValue Callee = DAG.getExternalSymbol(TLI.getLibcallName(LC),
+ TLI.getPointerTy());
+
+ SDLoc dl(Node);
+ 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));
+ Results.push_back(DAG.getLoad(RetVT, dl, CallInfo.second, CosPtr,
+ MachinePointerInfo(), false, false, false, 0));
+}
+
/// ExpandLegalINT_TO_FP - 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
SDValue SelectionDAGLegalize::ExpandLegalINT_TO_FP(bool isSigned,
SDValue Op0,
EVT DestVT,
- DebugLoc dl) {
- if (Op0.getValueType() == MVT::i32) {
+ SDLoc dl) {
+ if (Op0.getValueType() == MVT::i32 && TLI.isTypeLegal(MVT::f64)) {
// simple 32-bit [signed|unsigned] integer to float/double expansion
// Get the stack frame index of a 8 byte buffer.
SDValue StackSlot = DAG.CreateStackTemporary(MVT::f64);
// word offset constant for Hi/Lo address computation
- SDValue WordOff = DAG.getConstant(sizeof(int), TLI.getPointerTy());
+ SDValue WordOff = DAG.getConstant(sizeof(int), StackSlot.getValueType());
// set up Hi and Lo (into buffer) address based on endian
SDValue Hi = StackSlot;
- SDValue Lo = DAG.getNode(ISD::ADD, dl,
- TLI.getPointerTy(), StackSlot, WordOff);
+ SDValue Lo = DAG.getNode(ISD::ADD, dl, StackSlot.getValueType(),
+ StackSlot, WordOff);
if (TLI.isLittleEndian())
std::swap(Hi, Lo);
// select. We happen to get lucky and machinesink does the right
// thing most of the time. This would be a good candidate for a
//pseudo-op, or, even better, for whole-function isel.
- SDValue SignBitTest = DAG.getSetCC(dl, TLI.getSetCCResultType(MVT::i64),
+ SDValue SignBitTest = DAG.getSetCC(dl, getSetCCResultType(MVT::i64),
Op0, DAG.getConstant(0, MVT::i64), ISD::SETLT);
- return DAG.getNode(ISD::SELECT, dl, MVT::f32, SignBitTest, Slow, Fast);
+ return DAG.getSelect(dl, MVT::f32, SignBitTest, Slow, Fast);
}
// Otherwise, implement the fully general conversion.
DAG.getConstant(UINT64_C(0x800), MVT::i64));
SDValue And2 = DAG.getNode(ISD::AND, dl, MVT::i64, Op0,
DAG.getConstant(UINT64_C(0x7ff), MVT::i64));
- SDValue Ne = DAG.getSetCC(dl, TLI.getSetCCResultType(MVT::i64),
+ SDValue Ne = DAG.getSetCC(dl, getSetCCResultType(MVT::i64),
And2, DAG.getConstant(UINT64_C(0), MVT::i64), ISD::SETNE);
- SDValue Sel = DAG.getNode(ISD::SELECT, dl, MVT::i64, Ne, Or, Op0);
- SDValue Ge = DAG.getSetCC(dl, TLI.getSetCCResultType(MVT::i64),
+ SDValue Sel = DAG.getSelect(dl, MVT::i64, Ne, Or, Op0);
+ SDValue Ge = DAG.getSetCC(dl, getSetCCResultType(MVT::i64),
Op0, DAG.getConstant(UINT64_C(0x0020000000000000), MVT::i64),
ISD::SETUGE);
- SDValue Sel2 = DAG.getNode(ISD::SELECT, dl, MVT::i64, Ge, Sel, Op0);
+ SDValue Sel2 = DAG.getSelect(dl, MVT::i64, Ge, Sel, Op0);
EVT SHVT = TLI.getShiftAmountTy(Sel2.getValueType());
SDValue Sh = DAG.getNode(ISD::SRL, dl, MVT::i64, Sel2,
SDValue Tmp1 = DAG.getNode(ISD::SINT_TO_FP, dl, DestVT, Op0);
- SDValue SignSet = DAG.getSetCC(dl, TLI.getSetCCResultType(Op0.getValueType()),
+ SDValue SignSet = DAG.getSetCC(dl, getSetCCResultType(Op0.getValueType()),
Op0, DAG.getConstant(0, Op0.getValueType()),
ISD::SETLT);
SDValue Zero = DAG.getIntPtrConstant(0), Four = DAG.getIntPtrConstant(4);
- SDValue CstOffset = DAG.getNode(ISD::SELECT, dl, Zero.getValueType(),
+ SDValue CstOffset = DAG.getSelect(dl, Zero.getValueType(),
SignSet, Four, Zero);
// If the sign bit of the integer is set, the large number will be treated
// as a negative number. To counteract this, the dynamic code adds an
// offset depending on the data type.
uint64_t FF;
- switch (Op0.getValueType().getSimpleVT().SimpleTy) {
+ switch (Op0.getSimpleValueType().SimpleTy) {
default: llvm_unreachable("Unsupported integer type!");
case MVT::i8 : FF = 0x43800000ULL; break; // 2^8 (as a float)
case MVT::i16: FF = 0x47800000ULL; break; // 2^16 (as a float)
SDValue CPIdx = DAG.getConstantPool(FudgeFactor, TLI.getPointerTy());
unsigned Alignment = cast<ConstantPoolSDNode>(CPIdx)->getAlignment();
- CPIdx = DAG.getNode(ISD::ADD, dl, TLI.getPointerTy(), CPIdx, CstOffset);
+ CPIdx = DAG.getNode(ISD::ADD, dl, CPIdx.getValueType(), CPIdx, CstOffset);
Alignment = std::min(Alignment, 4u);
SDValue FudgeInReg;
if (DestVT == MVT::f32)
SDValue SelectionDAGLegalize::PromoteLegalINT_TO_FP(SDValue LegalOp,
EVT DestVT,
bool isSigned,
- DebugLoc dl) {
+ SDLoc dl) {
// First step, figure out the appropriate *INT_TO_FP operation to use.
EVT NewInTy = LegalOp.getValueType();
SDValue SelectionDAGLegalize::PromoteLegalFP_TO_INT(SDValue LegalOp,
EVT DestVT,
bool isSigned,
- DebugLoc dl) {
+ SDLoc dl) {
// First step, figure out the appropriate FP_TO*INT operation to use.
EVT NewOutTy = DestVT;
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;
}
/// ExpandBSWAP - Open code the operations for BSWAP of the specified operation.
///
-SDValue SelectionDAGLegalize::ExpandBSWAP(SDValue Op, DebugLoc dl) {
+SDValue SelectionDAGLegalize::ExpandBSWAP(SDValue Op, SDLoc dl) {
EVT VT = Op.getValueType();
EVT SHVT = TLI.getShiftAmountTy(VT);
SDValue Tmp1, Tmp2, Tmp3, Tmp4, Tmp5, Tmp6, Tmp7, Tmp8;
}
}
-/// SplatByte - Distribute ByteVal over NumBits bits.
-// FIXME: Move this helper to a common place.
-static APInt SplatByte(unsigned NumBits, uint8_t ByteVal) {
- APInt Val = APInt(NumBits, ByteVal);
- unsigned Shift = 8;
- for (unsigned i = NumBits; i > 8; i >>= 1) {
- Val = (Val << Shift) | Val;
- Shift <<= 1;
- }
- return Val;
-}
-
/// ExpandBitCount - Expand the specified bitcount instruction into operations.
///
SDValue SelectionDAGLegalize::ExpandBitCount(unsigned Opc, SDValue Op,
- DebugLoc dl) {
+ SDLoc dl) {
switch (Opc) {
default: llvm_unreachable("Cannot expand this yet!");
case ISD::CTPOP: {
// This is the "best" algorithm from
// http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel
- SDValue Mask55 = DAG.getConstant(SplatByte(Len, 0x55), VT);
- SDValue Mask33 = DAG.getConstant(SplatByte(Len, 0x33), VT);
- SDValue Mask0F = DAG.getConstant(SplatByte(Len, 0x0F), VT);
- SDValue Mask01 = DAG.getConstant(SplatByte(Len, 0x01), VT);
+ SDValue Mask55 = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x55)), VT);
+ SDValue Mask33 = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x33)), VT);
+ SDValue Mask0F = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x0F)), VT);
+ SDValue Mask01 = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x01)), VT);
// v = v - ((v >> 1) & 0x55555555...)
Op = DAG.getNode(ISD::SUB, dl, VT, Op,
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:
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:
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:
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:
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:
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:
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:
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;
}
void SelectionDAGLegalize::ExpandNode(SDNode *Node) {
SmallVector<SDValue, 8> Results;
- DebugLoc dl = Node->getDebugLoc();
+ SDLoc dl(Node);
SDValue Tmp1, Tmp2, Tmp3, Tmp4;
+ bool NeedInvert;
switch (Node->getOpcode()) {
case ISD::CTPOP:
case ISD::CTLZ:
Results.push_back(DAG.getConstant(0, MVT::i32));
Results.push_back(Node->getOperand(0));
break;
- case ISD::ATOMIC_FENCE:
- case ISD::MEMBARRIER: {
+ case ISD::ATOMIC_FENCE: {
// 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();
EVT NVT = Node->getValueType(0);
- APFloat apf(APInt::getNullValue(VT.getSizeInBits()));
+ APFloat apf(DAG.EVTToAPFloatSemantics(VT),
+ APInt::getNullValue(VT.getSizeInBits()));
APInt x = APInt::getSignBit(NVT.getSizeInBits());
(void)apf.convertFromAPInt(x, false, APFloat::rmNearestTiesToEven);
Tmp1 = DAG.getConstantFP(apf, VT);
- Tmp2 = DAG.getSetCC(dl, TLI.getSetCCResultType(VT),
+ Tmp2 = DAG.getSetCC(dl, getSetCCResultType(VT),
Node->getOperand(0),
Tmp1, ISD::SETLT);
True = DAG.getNode(ISD::FP_TO_SINT, dl, NVT, Node->getOperand(0));
Node->getOperand(0), Tmp1));
False = DAG.getNode(ISD::XOR, dl, NVT, False,
DAG.getConstant(x, NVT));
- Tmp1 = DAG.getNode(ISD::SELECT, dl, NVT, Tmp2, True, False);
+ Tmp1 = DAG.getSelect(dl, NVT, Tmp2, True, False);
Results.push_back(Tmp1);
break;
}
unsigned Align = Node->getConstantOperandVal(3);
SDValue VAListLoad = DAG.getLoad(TLI.getPointerTy(), dl, Tmp1, Tmp2,
- MachinePointerInfo(V),
+ MachinePointerInfo(V),
false, false, false, 0);
SDValue VAList = VAListLoad;
if (Align > TLI.getMinStackArgumentAlignment()) {
assert(((Align & (Align-1)) == 0) && "Expected Align to be a power of 2");
- VAList = DAG.getNode(ISD::ADD, dl, TLI.getPointerTy(), VAList,
+ VAList = DAG.getNode(ISD::ADD, dl, VAList.getValueType(), VAList,
DAG.getConstant(Align - 1,
- TLI.getPointerTy()));
+ VAList.getValueType()));
- VAList = DAG.getNode(ISD::AND, dl, TLI.getPointerTy(), VAList,
+ VAList = DAG.getNode(ISD::AND, dl, VAList.getValueType(), VAList,
DAG.getConstant(-(int64_t)Align,
- TLI.getPointerTy()));
+ VAList.getValueType()));
}
// Increment the pointer, VAList, to the next vaarg
- Tmp3 = DAG.getNode(ISD::ADD, dl, TLI.getPointerTy(), VAList,
- DAG.getConstant(TLI.getTargetData()->
+ Tmp3 = DAG.getNode(ISD::ADD, dl, VAList.getValueType(), VAList,
+ DAG.getConstant(TLI.getDataLayout()->
getTypeAllocSize(VT.getTypeForEVT(*DAG.getContext())),
- TLI.getPointerTy()));
+ VAList.getValueType()));
// Store the incremented VAList to the legalized pointer
Tmp3 = DAG.getStore(VAListLoad.getValue(1), dl, Tmp3, Tmp2,
MachinePointerInfo(V), false, false, 0);
EVT NewEltVT = TLI.getTypeToTransformTo(*DAG.getContext(), EltVT);
// BUILD_VECTOR operands are allowed to be wider than the element type.
- // But if NewEltVT is smaller that EltVT the BUILD_VECTOR does not accept it
+ // But if NewEltVT is smaller that EltVT the BUILD_VECTOR does not accept
+ // it.
if (NewEltVT.bitsLT(EltVT)) {
// Convert shuffle node.
// cast operands to v8i32 and re-build the mask.
// Calculate new VT, the size of the new VT should be equal to original.
- EVT NewVT = EVT::getVectorVT(*DAG.getContext(), NewEltVT,
- VT.getSizeInBits()/NewEltVT.getSizeInBits());
+ EVT NewVT =
+ EVT::getVectorVT(*DAG.getContext(), NewEltVT,
+ VT.getSizeInBits() / NewEltVT.getSizeInBits());
assert(NewVT.bitsEq(VT));
// cast operands to new VT
Op1 = DAG.getNode(ISD::BITCAST, dl, NewVT, Op1);
// Convert the shuffle mask
- unsigned int factor = NewVT.getVectorNumElements()/VT.getVectorNumElements();
+ unsigned int factor =
+ NewVT.getVectorNumElements()/VT.getVectorNumElements();
// EltVT gets smaller
assert(factor > 0);
if (Idx < NumElems)
Ops.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT,
Op0,
- DAG.getIntPtrConstant(Idx)));
+ DAG.getConstant(Idx, TLI.getVectorIdxTy())));
else
Ops.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT,
Op1,
- DAG.getIntPtrConstant(Idx - NumElems)));
+ DAG.getConstant(Idx - NumElems,
+ 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);
EVT VT = Node->getValueType(0);
Tmp1 = Node->getOperand(0);
Tmp2 = DAG.getConstantFP(0.0, VT);
- Tmp2 = DAG.getSetCC(dl, TLI.getSetCCResultType(Tmp1.getValueType()),
+ Tmp2 = DAG.getSetCC(dl, getSetCCResultType(Tmp1.getValueType()),
Tmp1, Tmp2, ISD::SETUGT);
Tmp3 = DAG.getNode(ISD::FNEG, dl, VT, Tmp1);
- Tmp1 = DAG.getNode(ISD::SELECT, dl, VT, Tmp2, Tmp1, Tmp3);
+ Tmp1 = DAG.getSelect(dl, VT, Tmp2, Tmp1, Tmp3);
Results.push_back(Tmp1);
break;
}
case ISD::FSQRT:
Results.push_back(ExpandFPLibCall(Node, RTLIB::SQRT_F32, RTLIB::SQRT_F64,
- RTLIB::SQRT_F80, RTLIB::SQRT_PPCF128));
+ RTLIB::SQRT_F80, RTLIB::SQRT_F128,
+ RTLIB::SQRT_PPCF128));
break;
case ISD::FSIN:
- Results.push_back(ExpandFPLibCall(Node, RTLIB::SIN_F32, RTLIB::SIN_F64,
- RTLIB::SIN_F80, RTLIB::SIN_PPCF128));
+ case ISD::FCOS: {
+ EVT VT = Node->getValueType(0);
+ bool isSIN = Node->getOpcode() == ISD::FSIN;
+ // Turn fsin / fcos into ISD::FSINCOS node if there are a pair of fsin /
+ // fcos which share the same operand and both are used.
+ if ((TLI.isOperationLegalOrCustom(ISD::FSINCOS, VT) ||
+ canCombineSinCosLibcall(Node, TLI, TM))
+ && useSinCos(Node)) {
+ SDVTList VTs = DAG.getVTList(VT, VT);
+ Tmp1 = DAG.getNode(ISD::FSINCOS, dl, VTs, Node->getOperand(0));
+ if (!isSIN)
+ Tmp1 = Tmp1.getValue(1);
+ Results.push_back(Tmp1);
+ } else if (isSIN) {
+ Results.push_back(ExpandFPLibCall(Node, RTLIB::SIN_F32, RTLIB::SIN_F64,
+ RTLIB::SIN_F80, RTLIB::SIN_F128,
+ RTLIB::SIN_PPCF128));
+ } else {
+ Results.push_back(ExpandFPLibCall(Node, RTLIB::COS_F32, RTLIB::COS_F64,
+ RTLIB::COS_F80, RTLIB::COS_F128,
+ RTLIB::COS_PPCF128));
+ }
break;
- case ISD::FCOS:
- Results.push_back(ExpandFPLibCall(Node, RTLIB::COS_F32, RTLIB::COS_F64,
- RTLIB::COS_F80, RTLIB::COS_PPCF128));
+ }
+ case ISD::FSINCOS:
+ // Expand into sincos libcall.
+ ExpandSinCosLibCall(Node, Results);
break;
case ISD::FLOG:
Results.push_back(ExpandFPLibCall(Node, RTLIB::LOG_F32, RTLIB::LOG_F64,
- RTLIB::LOG_F80, RTLIB::LOG_PPCF128));
+ RTLIB::LOG_F80, RTLIB::LOG_F128,
+ RTLIB::LOG_PPCF128));
break;
case ISD::FLOG2:
Results.push_back(ExpandFPLibCall(Node, RTLIB::LOG2_F32, RTLIB::LOG2_F64,
- RTLIB::LOG2_F80, RTLIB::LOG2_PPCF128));
+ RTLIB::LOG2_F80, RTLIB::LOG2_F128,
+ RTLIB::LOG2_PPCF128));
break;
case ISD::FLOG10:
Results.push_back(ExpandFPLibCall(Node, RTLIB::LOG10_F32, RTLIB::LOG10_F64,
- RTLIB::LOG10_F80, RTLIB::LOG10_PPCF128));
+ RTLIB::LOG10_F80, RTLIB::LOG10_F128,
+ RTLIB::LOG10_PPCF128));
break;
case ISD::FEXP:
Results.push_back(ExpandFPLibCall(Node, RTLIB::EXP_F32, RTLIB::EXP_F64,
- RTLIB::EXP_F80, RTLIB::EXP_PPCF128));
+ RTLIB::EXP_F80, RTLIB::EXP_F128,
+ RTLIB::EXP_PPCF128));
break;
case ISD::FEXP2:
Results.push_back(ExpandFPLibCall(Node, RTLIB::EXP2_F32, RTLIB::EXP2_F64,
- RTLIB::EXP2_F80, RTLIB::EXP2_PPCF128));
+ RTLIB::EXP2_F80, RTLIB::EXP2_F128,
+ RTLIB::EXP2_PPCF128));
break;
case ISD::FTRUNC:
Results.push_back(ExpandFPLibCall(Node, RTLIB::TRUNC_F32, RTLIB::TRUNC_F64,
- RTLIB::TRUNC_F80, RTLIB::TRUNC_PPCF128));
+ RTLIB::TRUNC_F80, RTLIB::TRUNC_F128,
+ RTLIB::TRUNC_PPCF128));
break;
case ISD::FFLOOR:
Results.push_back(ExpandFPLibCall(Node, RTLIB::FLOOR_F32, RTLIB::FLOOR_F64,
- RTLIB::FLOOR_F80, RTLIB::FLOOR_PPCF128));
+ RTLIB::FLOOR_F80, RTLIB::FLOOR_F128,
+ RTLIB::FLOOR_PPCF128));
break;
case ISD::FCEIL:
Results.push_back(ExpandFPLibCall(Node, RTLIB::CEIL_F32, RTLIB::CEIL_F64,
- RTLIB::CEIL_F80, RTLIB::CEIL_PPCF128));
+ RTLIB::CEIL_F80, RTLIB::CEIL_F128,
+ RTLIB::CEIL_PPCF128));
break;
case ISD::FRINT:
Results.push_back(ExpandFPLibCall(Node, RTLIB::RINT_F32, RTLIB::RINT_F64,
- RTLIB::RINT_F80, RTLIB::RINT_PPCF128));
+ RTLIB::RINT_F80, RTLIB::RINT_F128,
+ RTLIB::RINT_PPCF128));
break;
case ISD::FNEARBYINT:
Results.push_back(ExpandFPLibCall(Node, RTLIB::NEARBYINT_F32,
RTLIB::NEARBYINT_F64,
RTLIB::NEARBYINT_F80,
+ RTLIB::NEARBYINT_F128,
RTLIB::NEARBYINT_PPCF128));
break;
+ case ISD::FROUND:
+ Results.push_back(ExpandFPLibCall(Node, RTLIB::ROUND_F32,
+ RTLIB::ROUND_F64,
+ RTLIB::ROUND_F80,
+ RTLIB::ROUND_F128,
+ RTLIB::ROUND_PPCF128));
+ break;
case ISD::FPOWI:
Results.push_back(ExpandFPLibCall(Node, RTLIB::POWI_F32, RTLIB::POWI_F64,
- RTLIB::POWI_F80, RTLIB::POWI_PPCF128));
+ RTLIB::POWI_F80, RTLIB::POWI_F128,
+ RTLIB::POWI_PPCF128));
break;
case ISD::FPOW:
Results.push_back(ExpandFPLibCall(Node, RTLIB::POW_F32, RTLIB::POW_F64,
- RTLIB::POW_F80, RTLIB::POW_PPCF128));
+ RTLIB::POW_F80, RTLIB::POW_F128,
+ RTLIB::POW_PPCF128));
break;
case ISD::FDIV:
Results.push_back(ExpandFPLibCall(Node, RTLIB::DIV_F32, RTLIB::DIV_F64,
- RTLIB::DIV_F80, RTLIB::DIV_PPCF128));
+ RTLIB::DIV_F80, RTLIB::DIV_F128,
+ RTLIB::DIV_PPCF128));
break;
case ISD::FREM:
Results.push_back(ExpandFPLibCall(Node, RTLIB::REM_F32, RTLIB::REM_F64,
- RTLIB::REM_F80, RTLIB::REM_PPCF128));
+ RTLIB::REM_F80, RTLIB::REM_F128,
+ RTLIB::REM_PPCF128));
break;
case ISD::FMA:
Results.push_back(ExpandFPLibCall(Node, RTLIB::FMA_F32, RTLIB::FMA_F64,
- RTLIB::FMA_F80, RTLIB::FMA_PPCF128));
+ 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::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::FP32_TO_FP16:
- Results.push_back(ExpandLibCall(RTLIB::FPROUND_F32_F16, Node, false));
+ }
+ case ISD::FP_TO_FP16: {
+ 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.
Results.push_back(ExpandConstantFP(CFP, true));
break;
}
- case ISD::EHSELECTION: {
- unsigned Reg = TLI.getExceptionSelectorRegister();
- assert(Reg && "Can't expand to unknown register!");
- Results.push_back(DAG.getCopyFromReg(Node->getOperand(1), dl, Reg,
- Node->getValueType(0)));
- Results.push_back(Results[0].getValue(1));
- break;
- }
- case ISD::EXCEPTIONADDR: {
- unsigned Reg = TLI.getExceptionPointerRegister();
- assert(Reg && "Can't expand to unknown register!");
- Results.push_back(DAG.getCopyFromReg(Node->getOperand(0), dl, Reg,
- Node->getValueType(0)));
- Results.push_back(Results[0].getValue(1));
- break;
- }
case ISD::FSUB: {
EVT VT = Node->getValueType(0);
assert(TLI.isOperationLegalOrCustom(ISD::FADD, VT) &&
case ISD::UREM:
case ISD::SREM: {
EVT VT = Node->getValueType(0);
- SDVTList VTs = DAG.getVTList(VT, VT);
bool isSigned = Node->getOpcode() == ISD::SREM;
unsigned DivOpc = isSigned ? ISD::SDIV : ISD::UDIV;
unsigned DivRemOpc = isSigned ? ISD::SDIVREM : ISD::UDIVREM;
Tmp3 = Node->getOperand(1);
if (TLI.isOperationLegalOrCustom(DivRemOpc, VT) ||
(isDivRemLibcallAvailable(Node, isSigned, TLI) &&
- UseDivRem(Node, isSigned, false))) {
+ // If div is legal, it's better to do the normal expansion
+ !TLI.isOperationLegalOrCustom(DivOpc, Node->getValueType(0)) &&
+ useDivRem(Node, isSigned, false))) {
+ SDVTList VTs = DAG.getVTList(VT, VT);
Tmp1 = DAG.getNode(DivRemOpc, dl, VTs, Tmp2, Tmp3).getValue(1);
} else if (TLI.isOperationLegalOrCustom(DivOpc, VT)) {
// X % Y -> X-X/Y*Y
SDVTList VTs = DAG.getVTList(VT, VT);
if (TLI.isOperationLegalOrCustom(DivRemOpc, VT) ||
(isDivRemLibcallAvailable(Node, isSigned, TLI) &&
- UseDivRem(Node, isSigned, true)))
+ useDivRem(Node, isSigned, true)))
Tmp1 = DAG.getNode(DivRemOpc, dl, VTs, Node->getOperand(0),
Node->getOperand(1));
else if (isSigned)
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);
Tmp1 = DAG.getConstant(VT.getSizeInBits() - 1,
TLI.getShiftAmountTy(BottomHalf.getValueType()));
Tmp1 = DAG.getNode(ISD::SRA, dl, VT, BottomHalf, Tmp1);
- TopHalf = DAG.getSetCC(dl, TLI.getSetCCResultType(VT), TopHalf, Tmp1,
+ TopHalf = DAG.getSetCC(dl, getSetCCResultType(VT), TopHalf, Tmp1,
ISD::SETNE);
} else {
- TopHalf = DAG.getSetCC(dl, TLI.getSetCCResultType(VT), TopHalf,
+ TopHalf = DAG.getSetCC(dl, getSetCCResultType(VT), TopHalf,
DAG.getConstant(0, VT), ISD::SETNE);
}
Results.push_back(BottomHalf);
EVT PTy = TLI.getPointerTy();
- const TargetData &TD = *TLI.getTargetData();
+ const DataLayout &TD = *TLI.getDataLayout();
unsigned EntrySize =
DAG.getMachineFunction().getJumpTableInfo()->getEntrySize(TD);
- Index = DAG.getNode(ISD::MUL, dl, PTy,
- Index, DAG.getConstant(EntrySize, PTy));
- SDValue Addr = DAG.getNode(ISD::ADD, dl, PTy, Index, Table);
+ Index = DAG.getNode(ISD::MUL, dl, Index.getValueType(),
+ Index, DAG.getConstant(EntrySize, Index.getValueType()));
+ SDValue Addr = DAG.getNode(ISD::ADD, dl, Index.getValueType(),
+ Index, Table);
EVT MemVT = EVT::getIntegerVT(*DAG.getContext(), EntrySize * 8);
SDValue LD = DAG.getExtLoad(ISD::SEXTLOAD, dl, PTy, Chain, Addr,
Tmp1 = Node->getOperand(0);
Tmp2 = Node->getOperand(1);
Tmp3 = Node->getOperand(2);
- LegalizeSetCCCondCode(Node->getValueType(0), Tmp1, Tmp2, Tmp3, dl);
+ bool Legalized = LegalizeSetCCCondCode(Node->getValueType(0), Tmp1, Tmp2,
+ Tmp3, NeedInvert, dl);
+
+ if (Legalized) {
+ // If we expanded the SETCC by swapping LHS and RHS, or by inverting the
+ // condition code, create a new SETCC node.
+ if (Tmp3.getNode())
+ Tmp1 = DAG.getNode(ISD::SETCC, dl, Node->getValueType(0),
+ Tmp1, Tmp2, Tmp3);
+
+ // 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.getLogicalNOT(dl, Tmp1, Tmp1->getValueType(0));
- // If we expanded the SETCC into an AND/OR, return the new node
- if (Tmp2.getNode() == 0) {
Results.push_back(Tmp1);
break;
}
// Otherwise, SETCC for the given comparison type must be completely
// illegal; expand it into a SELECT_CC.
EVT VT = Node->getValueType(0);
+ int TrueValue;
+ switch (TLI.getBooleanContents(Tmp1->getValueType(0))) {
+ case TargetLowering::ZeroOrOneBooleanContent:
+ case TargetLowering::UndefinedBooleanContent:
+ TrueValue = 1;
+ break;
+ case TargetLowering::ZeroOrNegativeOneBooleanContent:
+ TrueValue = -1;
+ break;
+ }
Tmp1 = DAG.getNode(ISD::SELECT_CC, dl, VT, Tmp1, Tmp2,
- DAG.getConstant(1, VT), DAG.getConstant(0, VT), Tmp3);
+ DAG.getConstant(TrueValue, VT), DAG.getConstant(0, VT),
+ Tmp3);
Results.push_back(Tmp1);
break;
}
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(CCOp,
+ Tmp1.getValueType().isInteger());
+ if (TLI.isCondCodeLegal(InvCC, Tmp1.getSimpleValueType())) {
+ // Use the new condition code and swap true and false
+ Legalized = true;
+ Tmp1 = DAG.getSelectCC(dl, Tmp1, Tmp2, Tmp4, Tmp3, InvCC);
+ } else {
+ // If The inverse is not legal, then try to swap the arguments using
+ // the inverse condition code.
+ ISD::CondCode SwapInvCC = ISD::getSetCCSwappedOperands(InvCC);
+ if (TLI.isCondCodeLegal(SwapInvCC, Tmp1.getSimpleValueType())) {
+ // The swapped inverse condition is legal, so swap true and false,
+ // lhs and rhs.
+ Legalized = true;
+ Tmp1 = DAG.getSelectCC(dl, Tmp2, Tmp1, Tmp4, Tmp3, SwapInvCC);
+ }
+ }
- LegalizeSetCCCondCode(TLI.getSetCCResultType(Tmp1.getValueType()),
- Tmp1, Tmp2, CC, dl);
+ if (!Legalized) {
+ Legalized = LegalizeSetCCCondCode(
+ getSetCCResultType(Tmp1.getValueType()), Tmp1, Tmp2, CC, NeedInvert,
+ dl);
- assert(!Tmp2.getNode() && "Can't legalize SELECT_CC with legal condition!");
- Tmp2 = DAG.getConstant(0, Tmp1.getValueType());
- CC = DAG.getCondCode(ISD::SETNE);
- Tmp1 = DAG.getNode(ISD::SELECT_CC, dl, Node->getValueType(0), Tmp1, Tmp2,
- Tmp3, Tmp4, CC);
+ assert(Legalized && "Can't legalize SELECT_CC with legal condition!");
+
+ // If we expanded the SETCC by inverting the condition code, then swap
+ // the True/False operands to match.
+ if (NeedInvert)
+ std::swap(Tmp3, Tmp4);
+
+ // If we expanded the SETCC by swapping LHS and RHS, or by inverting the
+ // condition code, create a new SELECT_CC node.
+ if (CC.getNode()) {
+ Tmp1 = DAG.getNode(ISD::SELECT_CC, dl, Node->getValueType(0),
+ Tmp1, Tmp2, Tmp3, Tmp4, CC);
+ } else {
+ Tmp2 = DAG.getConstant(0, Tmp1.getValueType());
+ CC = DAG.getCondCode(ISD::SETNE);
+ Tmp1 = DAG.getNode(ISD::SELECT_CC, dl, Node->getValueType(0), Tmp1,
+ Tmp2, Tmp3, Tmp4, CC);
+ }
+ }
Results.push_back(Tmp1);
break;
}
Tmp3 = Node->getOperand(3); // RHS
Tmp4 = Node->getOperand(1); // CC
- LegalizeSetCCCondCode(TLI.getSetCCResultType(Tmp2.getValueType()),
- Tmp2, Tmp3, Tmp4, dl);
-
- assert(!Tmp3.getNode() && "Can't legalize BR_CC with legal condition!");
- Tmp3 = DAG.getConstant(0, Tmp2.getValueType());
- Tmp4 = DAG.getCondCode(ISD::SETNE);
- Tmp1 = DAG.getNode(ISD::BR_CC, dl, Node->getValueType(0), Tmp1, Tmp4, Tmp2,
- Tmp3, Node->getOperand(4));
+ bool Legalized = LegalizeSetCCCondCode(getSetCCResultType(
+ Tmp2.getValueType()), Tmp2, Tmp3, Tmp4, NeedInvert, dl);
+ (void)Legalized;
+ assert(Legalized && "Can't legalize BR_CC with legal condition!");
+
+ // 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)
+ Tmp4 = DAG.getNOT(dl, Tmp4, Tmp4->getValueType(0));
+
+ // If we expanded the SETCC by swapping LHS and RHS, create a new BR_CC
+ // node.
+ if (Tmp4.getNode()) {
+ Tmp1 = DAG.getNode(ISD::BR_CC, dl, Node->getValueType(0), Tmp1,
+ Tmp4, Tmp2, Tmp3, Node->getOperand(4));
+ } else {
+ Tmp3 = DAG.getConstant(0, Tmp2.getValueType());
+ Tmp4 = DAG.getCondCode(ISD::SETNE);
+ Tmp1 = DAG.getNode(ISD::BR_CC, dl, Node->getValueType(0), Tmp1, Tmp4,
+ Tmp2, Tmp3, Node->getOperand(4));
+ }
Results.push_back(Tmp1);
break;
}
for (unsigned Idx = 0; Idx < NumElem; Idx++) {
SDValue Ex = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl,
VT.getScalarType(),
- Node->getOperand(0), DAG.getIntPtrConstant(Idx));
+ Node->getOperand(0), DAG.getConstant(Idx,
+ TLI.getVectorIdxTy()));
SDValue Sh = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl,
VT.getScalarType(),
- Node->getOperand(1), DAG.getIntPtrConstant(Idx));
+ Node->getOperand(1), DAG.getConstant(Idx,
+ TLI.getVectorIdxTy()));
Scalars.push_back(DAG.getNode(Node->getOpcode(), dl,
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;
}
void SelectionDAGLegalize::PromoteNode(SDNode *Node) {
SmallVector<SDValue, 8> Results;
- EVT OVT = Node->getValueType(0);
+ MVT OVT = Node->getSimpleValueType(0);
if (Node->getOpcode() == ISD::UINT_TO_FP ||
Node->getOpcode() == ISD::SINT_TO_FP ||
Node->getOpcode() == ISD::SETCC) {
- OVT = Node->getOperand(0).getValueType();
+ OVT = Node->getOperand(0).getSimpleValueType();
}
- EVT NVT = TLI.getTypeToPromoteTo(Node->getOpcode(), OVT);
- DebugLoc dl = Node->getDebugLoc();
+ MVT NVT = TLI.getTypeToPromoteTo(Node->getOpcode(), OVT);
+ SDLoc dl(Node);
SDValue Tmp1, Tmp2, Tmp3;
switch (Node->getOpcode()) {
case ISD::CTTZ:
Tmp1 = DAG.getNode(Node->getOpcode(), dl, NVT, Tmp1);
if (Node->getOpcode() == ISD::CTTZ) {
// FIXME: This should set a bit in the zero extended value instead.
- Tmp2 = DAG.getSetCC(dl, TLI.getSetCCResultType(NVT),
+ Tmp2 = DAG.getSetCC(dl, getSetCCResultType(NVT),
Tmp1, DAG.getConstant(NVT.getSizeInBits(), NVT),
ISD::SETEQ);
- Tmp1 = DAG.getNode(ISD::SELECT, dl, NVT, Tmp2,
- DAG.getConstant(OVT.getSizeInBits(), NVT), Tmp1);
+ Tmp1 = DAG.getSelect(dl, NVT, Tmp2,
+ DAG.getConstant(OVT.getSizeInBits(), NVT), Tmp1);
} else if (Node->getOpcode() == ISD::CTLZ ||
Node->getOpcode() == ISD::CTLZ_ZERO_UNDEF) {
// Tmp1 = Tmp1 - (sizeinbits(NVT) - sizeinbits(Old VT))
}
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()) {
Tmp2 = DAG.getNode(ExtOp, dl, NVT, Node->getOperand(1));
Tmp3 = DAG.getNode(ExtOp, dl, NVT, Node->getOperand(2));
// Perform the larger operation, then round down.
- Tmp1 = DAG.getNode(ISD::SELECT, dl, NVT, Tmp1, Tmp2, Tmp3);
+ Tmp1 = DAG.getSelect(dl, NVT, Tmp1, Tmp2, Tmp3);
if (TruncOp != ISD::FP_ROUND)
Tmp1 = DAG.getNode(TruncOp, dl, Node->getValueType(0), Tmp1);
else
projects / oota-llvm.git / blobdiff