#include "llvm/Target/TargetFrameLowering.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
using namespace llvm;
#define DEBUG_TYPE "legalizedag"
//===----------------------------------------------------------------------===//
-/// SelectionDAGLegalize - This takes an arbitrary SelectionDAG as input and
+/// This takes an arbitrary SelectionDAG as input and
/// hacks on it until the target machine can handle it. This involves
/// eliminating value sizes the machine cannot handle (promoting small sizes to
/// large sizes or splitting up large values into small values) as well as
/// will attempt merge setcc and brc instructions into brcc's.
///
namespace {
-class SelectionDAGLegalize : public SelectionDAG::DAGUpdateListener {
+class SelectionDAGLegalize {
const TargetMachine &TM;
const TargetLowering &TLI;
SelectionDAG &DAG;
- /// \brief The iterator being used to walk the DAG. We hold a reference to it
- /// in order to update it as necessary on node deletion.
- SelectionDAG::allnodes_iterator &LegalizePosition;
-
/// \brief The set of nodes which have already been legalized. We hold a
/// reference to it in order to update as necessary on node deletion.
SmallPtrSetImpl<SDNode *> &LegalizedNodes;
public:
SelectionDAGLegalize(SelectionDAG &DAG,
- SelectionDAG::allnodes_iterator &LegalizePosition,
SmallPtrSetImpl<SDNode *> &LegalizedNodes,
SmallSetVector<SDNode *, 16> *UpdatedNodes = nullptr)
- : SelectionDAG::DAGUpdateListener(DAG), TM(DAG.getTarget()),
- TLI(DAG.getTargetLoweringInfo()), DAG(DAG),
- LegalizePosition(LegalizePosition), LegalizedNodes(LegalizedNodes),
- UpdatedNodes(UpdatedNodes) {}
+ : TM(DAG.getTarget()), TLI(DAG.getTargetLoweringInfo()), DAG(DAG),
+ LegalizedNodes(LegalizedNodes), UpdatedNodes(UpdatedNodes) {}
/// \brief Legalizes the given operation.
void LegalizeOp(SDNode *Node);
void LegalizeLoadOps(SDNode *Node);
void LegalizeStoreOps(SDNode *Node);
- /// PerformInsertVectorEltInMemory - Some target cannot handle a variable
+ /// Some targets cannot handle a variable
/// insertion index for the INSERT_VECTOR_ELT instruction. In this case, it
/// is necessary to spill the vector being inserted into to memory, perform
/// the insert there, and then read the result back.
SDValue ExpandINSERT_VECTOR_ELT(SDValue Vec, SDValue Val,
SDValue Idx, SDLoc dl);
- /// ShuffleWithNarrowerEltType - Return a vector shuffle operation which
+ /// Return a vector shuffle operation which
/// performs the same shuffe in terms of order or result bytes, but on a type
/// whose vector element type is narrower than the original shuffle type.
/// e.g. <v4i32> <0, 1, 0, 1> -> v8i16 <0, 1, 2, 3, 0, 1, 2, 3>
void ExpandNode(SDNode *Node);
void PromoteNode(SDNode *Node);
- void ForgetNode(SDNode *N) {
- LegalizedNodes.erase(N);
- if (LegalizePosition == SelectionDAG::allnodes_iterator(N))
- ++LegalizePosition;
- if (UpdatedNodes)
- UpdatedNodes->remove(N);
- }
-
public:
- // DAGUpdateListener implementation.
- void NodeDeleted(SDNode *N, SDNode *E) override {
- ForgetNode(N);
- }
- void NodeUpdated(SDNode *N) override {}
-
// Node replacement helpers
void ReplacedNode(SDNode *N) {
- if (N->use_empty()) {
- DAG.RemoveDeadNode(N);
- } else {
- ForgetNode(N);
- }
+ LegalizedNodes.erase(N);
+ if (UpdatedNodes)
+ UpdatedNodes->insert(N);
}
void ReplaceNode(SDNode *Old, SDNode *New) {
DEBUG(dbgs() << " ... replacing: "; Old->dump(&DAG);
};
}
-/// ShuffleWithNarrowerEltType - Return a vector shuffle operation which
+/// Return a vector shuffle operation which
/// performs the same shuffe in terms of order or result bytes, but on a type
/// whose vector element type is narrower than the original shuffle type.
/// e.g. <v4i32> <0, 1, 0, 1> -> v8i16 <0, 1, 2, 3, 0, 1, 2, 3>
return DAG.getVectorShuffle(NVT, dl, N1, N2, &NewMask[0]);
}
-/// ExpandConstantFP - Expands the ConstantFP node to an integer constant or
+/// Expands the ConstantFP node to an integer constant or
/// a load from the constant pool.
SDValue
SelectionDAGLegalize::ExpandConstantFP(ConstantFPSDNode *CFP, bool UseCP) {
if (ConstantFPSDNode::isValueValidForType(SVT, CFP->getValueAPF()) &&
// Only do this if the target has a native EXTLOAD instruction from
// smaller type.
- TLI.isLoadExtLegal(ISD::EXTLOAD, SVT) &&
+ TLI.isLoadExtLegal(ISD::EXTLOAD, OrigVT, SVT) &&
TLI.ShouldShrinkFPConstant(OrigVT)) {
Type *SType = SVT.getTypeForEVT(*DAG.getContext());
LLVMC = cast<ConstantFP>(ConstantExpr::getFPTrunc(LLVMC, SType));
return Result;
}
-/// ExpandUnalignedStore - Expands an unaligned store to 2 half-size stores.
+/// Expands an unaligned store to 2 half-size stores.
static void ExpandUnalignedStore(StoreSDNode *ST, SelectionDAG &DAG,
const TargetLowering &TLI,
SelectionDAGLegalize *DAGLegalize) {
DAGLegalize->ReplaceNode(SDValue(ST, 0), Result);
}
-/// ExpandUnalignedLoad - Expands an unaligned load to 2 half-size loads.
+/// Expands an unaligned load to 2 half-size loads.
static void
ExpandUnalignedLoad(LoadSDNode *LD, SelectionDAG &DAG,
const TargetLowering &TLI,
ChainResult = TF;
}
-/// PerformInsertVectorEltInMemory - Some target cannot handle a variable
-/// insertion index for the INSERT_VECTOR_ELT instruction. In this case, it
+/// Some target cannot handle a variable insertion index for the
+/// INSERT_VECTOR_ELT instruction. In this case, it
/// is necessary to spill the vector being inserted into to memory, perform
/// the insert there, and then read the result back.
SDValue SelectionDAGLegalize::
Type *Ty = ST->getMemoryVT().getTypeForEVT(*DAG.getContext());
unsigned ABIAlignment= TLI.getDataLayout()->getABITypeAlignment(Ty);
if (Align < ABIAlignment)
- ExpandUnalignedStore(cast<StoreSDNode>(Node),
- DAG, TLI, this);
+ ExpandUnalignedStore(cast<StoreSDNode>(Node), DAG, TLI, this);
}
break;
}
case TargetLowering::Custom: {
SDValue Res = TLI.LowerOperation(SDValue(Node, 0), DAG);
- if (Res.getNode())
+ if (Res && Res != SDValue(Node, 0))
ReplaceNode(SDValue(Node, 0), Res);
return;
}
Value = DAG.getZeroExtendInReg(Value, dl, StVT);
SDValue Result =
DAG.getTruncStore(Chain, dl, Value, Ptr, ST->getPointerInfo(),
- NVT, isVolatile, isNonTemporal, Alignment,
- AAInfo);
+ NVT, isVolatile, isNonTemporal, Alignment, AAInfo);
ReplaceNode(SDValue(Node, 0), Result);
} else if (StWidth & (StWidth - 1)) {
// If not storing a power-of-2 number of bits, expand as two stores.
}
case TargetLowering::Custom: {
SDValue Res = TLI.LowerOperation(SDValue(Node, 0), DAG);
- if (Res.getNode())
+ if (Res && Res != SDValue(Node, 0))
ReplaceNode(SDValue(Node, 0), Res);
return;
}
// nice to have an effective generic way of getting these benefits...
// Until such a way is found, don't insist on promoting i1 here.
(SrcVT != MVT::i1 ||
- TLI.getLoadExtAction(ExtType, MVT::i1) == TargetLowering::Promote)) {
+ TLI.getLoadExtAction(ExtType, Node->getValueType(0), MVT::i1) ==
+ TargetLowering::Promote)) {
// Promote to a byte-sized load if not loading an integral number of
// bytes. For example, promote EXTLOAD:i20 -> EXTLOAD:i24.
unsigned NewWidth = SrcVT.getStoreSizeInBits();
Chain = Ch;
} else {
bool isCustom = false;
- switch (TLI.getLoadExtAction(ExtType, SrcVT.getSimpleVT())) {
+ switch (TLI.getLoadExtAction(ExtType, Node->getValueType(0),
+ SrcVT.getSimpleVT())) {
default: llvm_unreachable("This action is not supported yet!");
case TargetLowering::Custom:
isCustom = true;
unsigned AS = LD->getAddressSpace();
unsigned Align = LD->getAlignment();
if (!TLI.allowsMisalignedMemoryAccesses(MemVT, AS, Align)) {
- Type *Ty =
- LD->getMemoryVT().getTypeForEVT(*DAG.getContext());
- unsigned ABIAlignment =
- TLI.getDataLayout()->getABITypeAlignment(Ty);
+ Type *Ty = LD->getMemoryVT().getTypeForEVT(*DAG.getContext());
+ unsigned ABIAlignment = TLI.getDataLayout()->getABITypeAlignment(Ty);
if (Align < ABIAlignment){
- ExpandUnalignedLoad(cast<LoadSDNode>(Node),
- DAG, TLI, Value, Chain);
+ ExpandUnalignedLoad(cast<LoadSDNode>(Node), DAG, TLI, Value, Chain);
}
}
}
break;
}
case TargetLowering::Expand:
- if (!TLI.isLoadExtLegal(ISD::EXTLOAD, SrcVT) &&
- TLI.isTypeLegal(SrcVT)) {
- SDValue Load = DAG.getLoad(SrcVT, dl, Chain, Ptr,
- LD->getMemOperand());
- unsigned ExtendOp;
- switch (ExtType) {
- case ISD::EXTLOAD:
- ExtendOp = (SrcVT.isFloatingPoint() ?
- ISD::FP_EXTEND : ISD::ANY_EXTEND);
+ if (!TLI.isLoadExtLegal(ISD::EXTLOAD, Node->getValueType(0), SrcVT)) {
+ // If the source type is not legal, see if there is a legal extload to
+ // an intermediate type that we can then extend further.
+ EVT LoadVT = TLI.getRegisterType(SrcVT.getSimpleVT());
+ if (TLI.isTypeLegal(SrcVT) || // Same as SrcVT == LoadVT?
+ TLI.isLoadExtLegal(ExtType, LoadVT, SrcVT)) {
+ // If we are loading a legal type, this is a non-extload followed by a
+ // full extend.
+ ISD::LoadExtType MidExtType =
+ (LoadVT == SrcVT) ? ISD::NON_EXTLOAD : ExtType;
+
+ SDValue Load = DAG.getExtLoad(MidExtType, dl, LoadVT, Chain, Ptr,
+ SrcVT, LD->getMemOperand());
+ unsigned ExtendOp =
+ ISD::getExtForLoadExtType(SrcVT.isFloatingPoint(), ExtType);
+ Value = DAG.getNode(ExtendOp, dl, Node->getValueType(0), Load);
+ Chain = Load.getValue(1);
break;
- case ISD::SEXTLOAD: ExtendOp = ISD::SIGN_EXTEND; break;
- case ISD::ZEXTLOAD: ExtendOp = ISD::ZERO_EXTEND; break;
- default: llvm_unreachable("Unexpected extend load type!");
}
- Value = DAG.getNode(ExtendOp, dl, Node->getValueType(0), Load);
- Chain = Load.getValue(1);
- break;
}
assert(!SrcVT.isVector() &&
Result.getValueType(),
Result, DAG.getValueType(SrcVT));
else
- ValRes = DAG.getZeroExtendInReg(Result, dl,
- SrcVT.getScalarType());
+ ValRes = DAG.getZeroExtendInReg(Result, dl, SrcVT.getScalarType());
Value = ValRes;
Chain = Result.getValue(1);
break;
}
}
-/// LegalizeOp - Return a legal replacement for the given operation, with
-/// all legal operands.
+/// Return a legal replacement for the given operation, with all legal operands.
void SelectionDAGLegalize::LegalizeOp(SDNode *Node) {
DEBUG(dbgs() << "\nLegalizing: "; Node->dump(&DAG));
Idx = DAG.getZExtOrTrunc(Idx, dl, TLI.getPointerTy());
StackPtr = DAG.getNode(ISD::ADD, dl, Idx.getValueType(), Idx, StackPtr);
+ SDValue NewLoad;
+
if (Op.getValueType().isVector())
- return DAG.getLoad(Op.getValueType(), dl, Ch, StackPtr,MachinePointerInfo(),
- false, false, false, 0);
- return DAG.getExtLoad(ISD::EXTLOAD, dl, Op.getValueType(), Ch, StackPtr,
- MachinePointerInfo(),
- Vec.getValueType().getVectorElementType(),
- false, false, false, 0);
+ NewLoad = DAG.getLoad(Op.getValueType(), dl, Ch, StackPtr,
+ MachinePointerInfo(), false, false, false, 0);
+ else
+ NewLoad = DAG.getExtLoad(
+ ISD::EXTLOAD, dl, Op.getValueType(), Ch, StackPtr, MachinePointerInfo(),
+ Vec.getValueType().getVectorElementType(), false, false, false, 0);
+
+ // Replace the chain going out of the store, by the one out of the load.
+ DAG.ReplaceAllUsesOfValueWith(Ch, SDValue(NewLoad.getNode(), 1));
+
+ // We introduced a cycle though, so update the loads operands, making sure
+ // to use the original store's chain as an incoming chain.
+ SmallVector<SDValue, 6> NewLoadOperands(NewLoad->op_begin(),
+ NewLoad->op_end());
+ NewLoadOperands[0] = Ch;
+ NewLoad =
+ SDValue(DAG.UpdateNodeOperands(NewLoad.getNode(), NewLoadOperands), 0);
+ return NewLoad;
}
SDValue SelectionDAGLegalize::ExpandInsertToVectorThroughStack(SDValue Op) {
StackPtr);
// Store the subvector.
- Ch = DAG.getStore(DAG.getEntryNode(), dl, Part, SubStackPtr,
+ Ch = DAG.getStore(Ch, dl, Part, SubStackPtr,
MachinePointerInfo(), false, false, 0);
// Finally, load the updated vector.
SDValue SP = DAG.getCopyFromReg(Chain, dl, SPReg, VT);
Chain = SP.getValue(1);
unsigned Align = cast<ConstantSDNode>(Tmp3)->getZExtValue();
- unsigned StackAlign = TM.getFrameLowering()->getStackAlignment();
+ unsigned StackAlign =
+ DAG.getSubtarget().getFrameLowering()->getStackAlignment();
Tmp1 = DAG.getNode(ISD::SUB, dl, VT, SP, Size); // Value
if (Align > StackAlign)
Tmp1 = DAG.getNode(ISD::AND, dl, VT, Tmp1,
Results.push_back(Tmp2);
}
-/// LegalizeSetCCCondCode - Legalize a SETCC with given LHS and RHS and
-/// condition code CC on the current target.
+/// Legalize a SETCC with given LHS and RHS and condition code CC on the current
+/// target.
///
/// If the SETCC has been legalized using AND / OR, then the legalized node
/// will be stored in LHS. RHS and CC will be set to SDValue(). NeedInvert
return false;
}
-/// EmitStackConvert - Emit a store/load combination to the stack. This stores
+/// Emit a store/load combination to the stack. This stores
/// SrcOp to a stack slot of type SlotVT, truncating it if needed. It then does
/// a load from the stack slot to DestVT, extending it if needed.
/// The resultant code need not be legal.
ShuffleVec.data());
else if (!TLI.isShuffleMaskLegal(ShuffleVec, VT))
return false;
- NewIntermedVals.push_back(std::make_pair(Shuffle, FinalIndices));
+ NewIntermedVals.push_back(
+ std::make_pair(Shuffle, std::move(FinalIndices)));
}
// If we had an odd number of defined values, then append the last
return true;
}
-/// ExpandBUILD_VECTOR - Expand a BUILD_VECTOR node on targets that don't
+/// Expand a BUILD_VECTOR node on targets that don't
/// support the operation, but do support the resultant vector type.
SDValue SelectionDAGLegalize::ExpandBUILD_VECTOR(SDNode *Node) {
unsigned NumElems = Node->getNumOperands();
return ExpandVectorBuildThroughStack(Node);
}
-// ExpandLibCall - Expand a node into a call to a libcall. If the result value
+// Expand a node into a call to a libcall. If the result value
// does not fit into a register, return the lo part and set the hi part to the
// by-reg argument. If it does fit into a single register, return the result
// and leave the Hi part unset.
return CallInfo.first;
}
-/// ExpandLibCall - Generate a libcall taking the given operands as arguments
+/// Generate a libcall taking the given operands as arguments
/// and returning a result of type RetVT.
SDValue SelectionDAGLegalize::ExpandLibCall(RTLIB::Libcall LC, EVT RetVT,
const SDValue *Ops, unsigned NumOps,
return CallInfo.first;
}
-// ExpandChainLibCall - Expand a node into a call to a libcall. Similar to
+// Expand a node into a call to a libcall. Similar to
// ExpandLibCall except that the first operand is the in-chain.
std::pair<SDValue, SDValue>
SelectionDAGLegalize::ExpandChainLibCall(RTLIB::Libcall LC,
return ExpandLibCall(LC, Node, isSigned);
}
-/// isDivRemLibcallAvailable - Return true if divmod libcall is available.
+/// Return true if divmod libcall is available.
static bool isDivRemLibcallAvailable(SDNode *Node, bool isSigned,
const TargetLowering &TLI) {
RTLIB::Libcall LC;
return TLI.getLibcallName(LC) != nullptr;
}
-/// useDivRem - Only issue divrem libcall if both quotient and remainder are
-/// needed.
+/// Only issue divrem libcall if both quotient and remainder are needed.
static bool useDivRem(SDNode *Node, bool isSigned, bool isDIV) {
// The other use might have been replaced with a divrem already.
unsigned DivRemOpc = isSigned ? ISD::SDIVREM : ISD::UDIVREM;
return false;
}
-/// ExpandDivRemLibCall - Issue libcalls to __{u}divmod to compute div / rem
-/// pairs.
+/// Issue libcalls to __{u}divmod to compute div / rem pairs.
void
SelectionDAGLegalize::ExpandDivRemLibCall(SDNode *Node,
SmallVectorImpl<SDValue> &Results) {
Results.push_back(Rem);
}
-/// isSinCosLibcallAvailable - Return true if sincos libcall is available.
+/// Return true if sincos libcall is available.
static bool isSinCosLibcallAvailable(SDNode *Node, const TargetLowering &TLI) {
RTLIB::Libcall LC;
switch (Node->getSimpleValueType(0).SimpleTy) {
return TLI.getLibcallName(LC) != nullptr;
}
-/// canCombineSinCosLibcall - Return true if sincos libcall is available and
-/// can be used to combine sin and cos.
+/// Return true if sincos libcall is available and can be used to combine sin
+/// and cos.
static bool canCombineSinCosLibcall(SDNode *Node, const TargetLowering &TLI,
const TargetMachine &TM) {
if (!isSinCosLibcallAvailable(Node, TLI))
return true;
}
-/// useSinCos - Only issue sincos libcall if both sin and cos are
-/// needed.
+/// Only issue sincos libcall if both sin and cos are needed.
static bool useSinCos(SDNode *Node) {
unsigned OtherOpcode = Node->getOpcode() == ISD::FSIN
? ISD::FCOS : ISD::FSIN;
return false;
}
-/// ExpandSinCosLibCall - Issue libcalls to sincos to compute sin / cos
-/// pairs.
+/// Issue libcalls to sincos to compute sin / cos pairs.
void
SelectionDAGLegalize::ExpandSinCosLibCall(SDNode *Node,
SmallVectorImpl<SDValue> &Results) {
MachinePointerInfo(), false, false, false, 0));
}
-/// ExpandLegalINT_TO_FP - This function is responsible for legalizing a
+/// This function is responsible for legalizing a
/// INT_TO_FP operation of the specified operand when the target requests that
/// we expand it. At this point, we know that the result and operand types are
/// legal for the target.
return DAG.getNode(ISD::FADD, dl, DestVT, Tmp1, FudgeInReg);
}
-/// PromoteLegalINT_TO_FP - This function is responsible for legalizing a
+/// This function is responsible for legalizing a
/// *INT_TO_FP operation of the specified operand when the target requests that
/// we promote it. At this point, we know that the result and operand types are
/// legal for the target, and that there is a legal UINT_TO_FP or SINT_TO_FP
dl, NewInTy, LegalOp));
}
-/// PromoteLegalFP_TO_INT - This function is responsible for legalizing a
+/// This function is responsible for legalizing a
/// FP_TO_*INT operation of the specified operand when the target requests that
/// we promote it. At this point, we know that the result and operand types are
/// legal for the target, and that there is a legal FP_TO_UINT or FP_TO_SINT
return DAG.getNode(ISD::TRUNCATE, dl, DestVT, Operation);
}
-/// ExpandBSWAP - Open code the operations for BSWAP of the specified operation.
-///
+/// Open code the operations for BSWAP of the specified operation.
SDValue SelectionDAGLegalize::ExpandBSWAP(SDValue Op, SDLoc dl) {
EVT VT = Op.getValueType();
EVT SHVT = TLI.getShiftAmountTy(VT);
}
}
-/// ExpandBitCount - Expand the specified bitcount instruction into operations.
-///
+/// Expand the specified bitcount instruction into operations.
SDValue SelectionDAGLegalize::ExpandBitCount(unsigned Opc, SDValue Op,
SDLoc dl) {
switch (Opc) {
// x = x | (x >>32); // for 64-bit input
// return popcount(~x);
//
- // but see also: http://www.hackersdelight.org/HDcode/nlz.cc
+ // Ref: "Hacker's Delight" by Henry Warren
EVT VT = Op.getValueType();
EVT ShVT = TLI.getShiftAmountTy(VT);
unsigned len = VT.getSizeInBits();
// for now, we use: { return popcount(~x & (x - 1)); }
// unless the target has ctlz but not ctpop, in which case we use:
// { return 32 - nlz(~x & (x-1)); }
- // see also http://www.hackersdelight.org/HDcode/ntz.cc
+ // Ref: "Hacker's Delight" by Henry Warren
EVT VT = Op.getValueType();
SDValue Tmp3 = DAG.getNode(ISD::AND, dl, VT,
DAG.getNOT(dl, Op, VT),
std::pair <SDValue, SDValue> SelectionDAGLegalize::ExpandAtomic(SDNode *Node) {
unsigned Opc = Node->getOpcode();
MVT VT = cast<AtomicSDNode>(Node)->getMemoryVT().getSimpleVT();
- RTLIB::Libcall LC;
-
- switch (Opc) {
- default:
- llvm_unreachable("Unhandled atomic intrinsic Expand!");
- case ISD::ATOMIC_SWAP:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_LOCK_TEST_AND_SET_1; break;
- case MVT::i16: LC = RTLIB::SYNC_LOCK_TEST_AND_SET_2; break;
- case MVT::i32: LC = RTLIB::SYNC_LOCK_TEST_AND_SET_4; break;
- case MVT::i64: LC = RTLIB::SYNC_LOCK_TEST_AND_SET_8; break;
- case MVT::i128:LC = RTLIB::SYNC_LOCK_TEST_AND_SET_16;break;
- }
- break;
- case ISD::ATOMIC_CMP_SWAP:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_1; break;
- case MVT::i16: LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_2; break;
- case MVT::i32: LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_4; break;
- case MVT::i64: LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_8; break;
- case MVT::i128:LC = RTLIB::SYNC_VAL_COMPARE_AND_SWAP_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_ADD:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_ADD_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_ADD_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_ADD_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_ADD_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_ADD_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_SUB:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_SUB_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_SUB_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_SUB_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_SUB_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_SUB_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_AND:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_AND_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_AND_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_AND_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_AND_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_AND_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_OR:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_OR_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_OR_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_OR_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_OR_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_OR_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_XOR:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_XOR_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_XOR_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_XOR_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_XOR_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_XOR_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_NAND:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_NAND_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_NAND_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_NAND_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_NAND_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_NAND_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_MAX:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_MAX_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_MAX_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_MAX_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_MAX_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_MAX_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_UMAX:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_UMAX_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_UMAX_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_UMAX_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_UMAX_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_UMAX_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_MIN:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_MIN_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_MIN_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_MIN_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_MIN_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_MIN_16;break;
- }
- break;
- case ISD::ATOMIC_LOAD_UMIN:
- switch (VT.SimpleTy) {
- default: llvm_unreachable("Unexpected value type for atomic!");
- case MVT::i8: LC = RTLIB::SYNC_FETCH_AND_UMIN_1; break;
- case MVT::i16: LC = RTLIB::SYNC_FETCH_AND_UMIN_2; break;
- case MVT::i32: LC = RTLIB::SYNC_FETCH_AND_UMIN_4; break;
- case MVT::i64: LC = RTLIB::SYNC_FETCH_AND_UMIN_8; break;
- case MVT::i128:LC = RTLIB::SYNC_FETCH_AND_UMIN_16;break;
- }
- break;
- }
+ RTLIB::Libcall LC = RTLIB::getATOMIC(Opc, VT);
+ assert(LC != RTLIB::UNKNOWN_LIBCALL && "Unexpected atomic op or value type!");
return ExpandChainLibCall(LC, Node, false);
}
Results.push_back(Tmp1);
break;
}
+ case ISD::FMINNUM:
+ Results.push_back(ExpandFPLibCall(Node, RTLIB::FMIN_F32, RTLIB::FMIN_F64,
+ RTLIB::FMIN_F80, RTLIB::FMIN_F128,
+ RTLIB::FMIN_PPCF128));
+ break;
+ case ISD::FMAXNUM:
+ Results.push_back(ExpandFPLibCall(Node, RTLIB::FMAX_F32, RTLIB::FMAX_F64,
+ RTLIB::FMAX_F80, RTLIB::FMAX_F128,
+ RTLIB::FMAX_PPCF128));
+ break;
case ISD::FSQRT:
Results.push_back(ExpandFPLibCall(Node, RTLIB::SQRT_F32, RTLIB::SQRT_F64,
RTLIB::SQRT_F80, RTLIB::SQRT_F128,
RTLIB::FMA_F80, RTLIB::FMA_F128,
RTLIB::FMA_PPCF128));
break;
+ case ISD::FMAD:
+ llvm_unreachable("Illegal fmad should never be formed");
+
+ case ISD::FADD:
+ Results.push_back(ExpandFPLibCall(Node, RTLIB::ADD_F32, RTLIB::ADD_F64,
+ RTLIB::ADD_F80, RTLIB::ADD_F128,
+ RTLIB::ADD_PPCF128));
+ break;
+ case ISD::FMUL:
+ Results.push_back(ExpandFPLibCall(Node, RTLIB::MUL_F32, RTLIB::MUL_F64,
+ RTLIB::MUL_F80, RTLIB::MUL_F128,
+ RTLIB::MUL_PPCF128));
+ break;
case ISD::FP16_TO_FP: {
if (Node->getValueType(0) == MVT::f32) {
Results.push_back(ExpandLibCall(RTLIB::FPEXT_F16_F32, Node, false));
break;
}
case ISD::FP_TO_FP16: {
+ if (!TM.Options.UseSoftFloat && TM.Options.UnsafeFPMath) {
+ SDValue Op = Node->getOperand(0);
+ MVT SVT = Op.getSimpleValueType();
+ if ((SVT == MVT::f64 || SVT == MVT::f80) &&
+ TLI.isOperationLegalOrCustom(ISD::FP_TO_FP16, MVT::f32)) {
+ // Under fastmath, we can expand this node into a fround followed by
+ // a float-half conversion.
+ SDValue FloatVal = DAG.getNode(ISD::FP_ROUND, dl, MVT::f32, Op,
+ DAG.getIntPtrConstant(0));
+ Results.push_back(
+ DAG.getNode(ISD::FP_TO_FP16, dl, MVT::i16, FloatVal));
+ break;
+ }
+ }
+
RTLIB::Libcall LC =
RTLIB::getFPROUND(Node->getOperand(0).getValueType(), MVT::f16);
assert(LC != RTLIB::UNKNOWN_LIBCALL && "Unable to expand fp_to_fp16");
}
case ISD::FSUB: {
EVT VT = Node->getValueType(0);
- assert(TLI.isOperationLegalOrCustom(ISD::FADD, VT) &&
- TLI.isOperationLegalOrCustom(ISD::FNEG, VT) &&
- "Don't know how to expand this FP subtraction!");
- Tmp1 = DAG.getNode(ISD::FNEG, dl, VT, Node->getOperand(1));
- Tmp1 = DAG.getNode(ISD::FADD, dl, VT, Node->getOperand(0), Tmp1);
- Results.push_back(Tmp1);
+ if (TLI.isOperationLegalOrCustom(ISD::FADD, VT) &&
+ TLI.isOperationLegalOrCustom(ISD::FNEG, VT)) {
+ Tmp1 = DAG.getNode(ISD::FNEG, dl, VT, Node->getOperand(1));
+ Tmp1 = DAG.getNode(ISD::FADD, dl, VT, Node->getOperand(0), Tmp1);
+ Results.push_back(Tmp1);
+ } else {
+ Results.push_back(ExpandFPLibCall(Node, RTLIB::SUB_F32, RTLIB::SUB_F64,
+ RTLIB::SUB_F80, RTLIB::SUB_F128,
+ RTLIB::SUB_PPCF128));
+ }
break;
}
case ISD::SUB: {
TopHalf = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, VT, Ret,
DAG.getIntPtrConstant(1));
// Ret is a node with an illegal type. Because such things are not
- // generally permitted during this phase of legalization, delete the
- // node. The above EXTRACT_ELEMENT nodes should have been folded.
- DAG.DeleteNode(Ret.getNode());
+ // generally permitted during this phase of legalization, make sure the
+ // node has no more uses. The above EXTRACT_ELEMENT nodes should have been
+ // folded.
+ assert(Ret->use_empty() &&
+ "Unexpected uses of illegally type from expanded lib call.");
}
if (isSigned) {
Node->getOpcode() == ISD::SETCC) {
OVT = Node->getOperand(0).getSimpleValueType();
}
+ if (Node->getOpcode() == ISD::BR_CC)
+ OVT = Node->getOperand(2).getSimpleValueType();
MVT NVT = TLI.getTypeToPromoteTo(Node->getOpcode(), OVT);
SDLoc dl(Node);
SDValue Tmp1, Tmp2, Tmp3;
Tmp1, Tmp2, Node->getOperand(2)));
break;
}
+ case ISD::BR_CC: {
+ unsigned ExtOp = ISD::FP_EXTEND;
+ if (NVT.isInteger()) {
+ ISD::CondCode CCCode =
+ cast<CondCodeSDNode>(Node->getOperand(1))->get();
+ ExtOp = isSignedIntSetCC(CCCode) ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND;
+ }
+ Tmp1 = DAG.getNode(ExtOp, dl, NVT, Node->getOperand(2));
+ Tmp2 = DAG.getNode(ExtOp, dl, NVT, Node->getOperand(3));
+ Results.push_back(DAG.getNode(ISD::BR_CC, dl, Node->getValueType(0),
+ Node->getOperand(0), Node->getOperand(1),
+ Tmp1, Tmp2, Node->getOperand(4)));
+ break;
+ }
+ case ISD::FADD:
+ case ISD::FSUB:
+ case ISD::FMUL:
case ISD::FDIV:
case ISD::FREM:
+ case ISD::FMINNUM:
+ case ISD::FMAXNUM:
+ case ISD::FCOPYSIGN:
case ISD::FPOW: {
Tmp1 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(0));
Tmp2 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(1));
Tmp3, DAG.getIntPtrConstant(0)));
break;
}
- case ISD::FLOG2:
- case ISD::FEXP2:
+ case ISD::FMA: {
+ Tmp1 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(0));
+ Tmp2 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(1));
+ Tmp3 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(2));
+ Results.push_back(
+ DAG.getNode(ISD::FP_ROUND, dl, OVT,
+ DAG.getNode(Node->getOpcode(), dl, NVT, Tmp1, Tmp2, Tmp3),
+ DAG.getIntPtrConstant(0)));
+ break;
+ }
+ case ISD::FPOWI: {
+ Tmp1 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(0));
+ Tmp2 = Node->getOperand(1);
+ Tmp3 = DAG.getNode(Node->getOpcode(), dl, NVT, Tmp1, Tmp2);
+ Results.push_back(DAG.getNode(ISD::FP_ROUND, dl, OVT,
+ Tmp3, DAG.getIntPtrConstant(0)));
+ break;
+ }
+ case ISD::FFLOOR:
+ case ISD::FCEIL:
+ case ISD::FRINT:
+ case ISD::FNEARBYINT:
+ case ISD::FROUND:
+ case ISD::FTRUNC:
+ case ISD::FNEG:
+ case ISD::FSQRT:
+ case ISD::FSIN:
+ case ISD::FCOS:
case ISD::FLOG:
- case ISD::FEXP: {
+ case ISD::FLOG2:
+ case ISD::FLOG10:
+ case ISD::FABS:
+ case ISD::FEXP:
+ case ISD::FEXP2: {
Tmp1 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(0));
Tmp2 = DAG.getNode(Node->getOpcode(), dl, NVT, Tmp1);
Results.push_back(DAG.getNode(ISD::FP_ROUND, dl, OVT,
ReplaceNode(Node, Results.data());
}
-// SelectionDAG::Legalize - This is the entry point for the file.
-//
+/// This is the entry point for the file.
void SelectionDAG::Legalize() {
AssignTopologicalOrder();
- allnodes_iterator LegalizePosition;
SmallPtrSet<SDNode *, 16> LegalizedNodes;
- SelectionDAGLegalize Legalizer(*this, LegalizePosition, LegalizedNodes);
+ SelectionDAGLegalize Legalizer(*this, LegalizedNodes);
// Visit all the nodes. We start in topological order, so that we see
// nodes with their original operands intact. Legalization can produce
// nodes have been legalized.
for (;;) {
bool AnyLegalized = false;
- for (LegalizePosition = allnodes_end();
- LegalizePosition != allnodes_begin(); ) {
- --LegalizePosition;
+ for (auto NI = allnodes_end(); NI != allnodes_begin();) {
+ --NI;
- SDNode *N = LegalizePosition;
- if (LegalizedNodes.insert(N)) {
+ SDNode *N = NI;
+ if (N->use_empty() && N != getRoot().getNode()) {
+ ++NI;
+ DeleteNode(N);
+ continue;
+ }
+
+ if (LegalizedNodes.insert(N).second) {
AnyLegalized = true;
Legalizer.LegalizeOp(N);
+
+ if (N->use_empty() && N != getRoot().getNode()) {
+ ++NI;
+ DeleteNode(N);
+ }
}
}
if (!AnyLegalized)
bool SelectionDAG::LegalizeOp(SDNode *N,
SmallSetVector<SDNode *, 16> &UpdatedNodes) {
- allnodes_iterator LegalizePosition(N);
SmallPtrSet<SDNode *, 16> LegalizedNodes;
- SelectionDAGLegalize Legalizer(*this, LegalizePosition, LegalizedNodes,
- &UpdatedNodes);
+ SelectionDAGLegalize Legalizer(*this, LegalizedNodes, &UpdatedNodes);
// Directly insert the node in question, and legalize it. This will recurse
// as needed through operands.
projects / oota-llvm.git / blobdiff