static cl::opt<bool>
EmitJumpTables("hexagon-emit-jump-tables", cl::init(true), cl::Hidden,
- cl::desc("Control jump table emission on Hexagon target"));
+ cl::desc("Control jump table emission on Hexagon target"));
+
+static cl::opt<bool> EnableHexSDNodeSched("enable-hexagon-sdnode-sched",
+ cl::Hidden, cl::ZeroOrMore, cl::init(false),
+ cl::desc("Enable Hexagon SDNode scheduling"));
+
+static cl::opt<bool> EnableFastMath("ffast-math",
+ cl::Hidden, cl::ZeroOrMore, cl::init(false),
+ cl::desc("Enable Fast Math processing"));
+
+static cl::opt<int> MinimumJumpTables("minimum-jump-tables",
+ cl::Hidden, cl::ZeroOrMore, cl::init(5),
+ cl::desc("Set minimum jump tables"));
+
+static cl::opt<int> MaxStoresPerMemcpyCL("max-store-memcpy",
+ cl::Hidden, cl::ZeroOrMore, cl::init(6),
+ cl::desc("Max #stores to inline memcpy"));
+
+static cl::opt<int> MaxStoresPerMemcpyOptSizeCL("max-store-memcpy-Os",
+ cl::Hidden, cl::ZeroOrMore, cl::init(4),
+ cl::desc("Max #stores to inline memcpy"));
+
+static cl::opt<int> MaxStoresPerMemmoveCL("max-store-memmove",
+ cl::Hidden, cl::ZeroOrMore, cl::init(6),
+ cl::desc("Max #stores to inline memmove"));
+
+static cl::opt<int> MaxStoresPerMemmoveOptSizeCL("max-store-memmove-Os",
+ cl::Hidden, cl::ZeroOrMore, cl::init(4),
+ cl::desc("Max #stores to inline memmove"));
+
+static cl::opt<int> MaxStoresPerMemsetCL("max-store-memset",
+ cl::Hidden, cl::ZeroOrMore, cl::init(8),
+ cl::desc("Max #stores to inline memset"));
+
+static cl::opt<int> MaxStoresPerMemsetOptSizeCL("max-store-memset-Os",
+ cl::Hidden, cl::ZeroOrMore, cl::init(4),
+ cl::desc("Max #stores to inline memset"));
+
namespace {
class HexagonCCState : public CCState {
- int NumNamedVarArgParams;
+ unsigned NumNamedVarArgParams;
public:
HexagonCCState(CallingConv::ID CC, bool isVarArg, MachineFunction &MF,
: CCState(CC, isVarArg, MF, locs, C),
NumNamedVarArgParams(NumNamedVarArgParams) {}
- int getNumNamedVarArgParams() const { return NumNamedVarArgParams; }
+ unsigned getNumNamedVarArgParams() const { return NumNamedVarArgParams; }
};
}
ISD::ArgFlagsTy ArgFlags, CCState &State) {
HexagonCCState &HState = static_cast<HexagonCCState &>(State);
- // NumNamedVarArgParams can not be zero for a VarArg function.
- assert((HState.getNumNamedVarArgParams() > 0) &&
- "NumNamedVarArgParams is not bigger than zero.");
-
- if ((int)ValNo < HState.getNumNamedVarArgParams()) {
+ if (ValNo < HState.getNumNamedVarArgParams()) {
// Deal with named arguments.
return CC_Hexagon(ValNo, ValVT, LocVT, LocInfo, ArgFlags, State);
}
if (ArgFlags.isByVal()) {
// If pass-by-value, the size allocated on stack is decided
// by ArgFlags.getByValSize(), not by the size of LocVT.
- assert ((ArgFlags.getByValSize() > 8) &&
- "ByValSize must be bigger than 8 bytes");
- ofst = State.AllocateStack(ArgFlags.getByValSize(), 4);
+ ofst = State.AllocateStack(ArgFlags.getByValSize(),
+ ArgFlags.getByValAlign());
State.addLoc(CCValAssign::getMem(ValNo, ValVT, ofst, LocVT, LocInfo));
return false;
}
if (ArgFlags.isByVal()) {
// Passed on stack.
- assert ((ArgFlags.getByValSize() > 8) &&
- "ByValSize must be bigger than 8 bytes");
- unsigned Offset = State.AllocateStack(ArgFlags.getByValSize(), 4);
+ unsigned Offset = State.AllocateStack(ArgFlags.getByValSize(),
+ ArgFlags.getByValAlign());
State.addLoc(CCValAssign::getMem(ValNo, ValVT, Offset, LocVT, LocInfo));
return false;
}
ISD::ArgFlagsTy Flags, SelectionDAG &DAG,
SDLoc dl) {
- SDValue SizeNode = DAG.getConstant(Flags.getByValSize(), MVT::i32);
+ SDValue SizeNode = DAG.getConstant(Flags.getByValSize(), dl, MVT::i32);
return DAG.getMemcpy(Chain, dl, Dst, Src, SizeNode, Flags.getByValAlign(),
/*isVolatile=*/false, /*AlwaysInline=*/false,
/*isTailCall=*/false,
return DAG.getNode(HexagonISD::RET_FLAG, dl, MVT::Other, RetOps);
}
+bool HexagonTargetLowering::mayBeEmittedAsTailCall(CallInst *CI) const {
+ // If either no tail call or told not to tail call at all, don't.
+ auto Attr =
+ CI->getParent()->getParent()->getFnAttribute("disable-tail-calls");
+ if (!CI->isTailCall() || Attr.getValueAsString() == "true")
+ return false;
-
+ return true;
+}
/// LowerCallResult - Lower the result values of an ISD::CALL into the
/// appropriate copies out of appropriate physical registers. This assumes that
bool doesNotReturn = CLI.DoesNotReturn;
bool IsStructRet = (Outs.empty()) ? false : Outs[0].Flags.isSRet();
+ MachineFunction &MF = DAG.getMachineFunction();
// Check for varargs.
int NumNamedVarArgParams = -1;
HexagonCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
*DAG.getContext(), NumNamedVarArgParams);
- if (NumNamedVarArgParams > 0)
+ if (isVarArg)
CCInfo.AnalyzeCallOperands(Outs, CC_Hexagon_VarArg);
else
CCInfo.AnalyzeCallOperands(Outs, CC_Hexagon);
+ auto Attr = MF.getFunction()->getFnAttribute("disable-tail-calls");
+ if (Attr.getValueAsString() == "true")
+ isTailCall = false;
- if(isTailCall) {
- bool StructAttrFlag =
- DAG.getMachineFunction().getFunction()->hasStructRetAttr();
+ if (isTailCall) {
+ bool StructAttrFlag = MF.getFunction()->hasStructRetAttr();
isTailCall = IsEligibleForTailCallOptimization(Callee, CallConv,
isVarArg, IsStructRet,
StructAttrFlag,
Outs, OutVals, Ins, DAG);
- for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i){
+ for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
if (VA.isMemLoc()) {
isTailCall = false;
break;
}
}
- if (isTailCall) {
- DEBUG(dbgs () << "Eligible for Tail Call\n");
- } else {
- DEBUG(dbgs () <<
- "Argument must be passed on stack. Not eligible for Tail Call\n");
- }
+ DEBUG(dbgs() << (isTailCall ? "Eligible for Tail Call\n"
+ : "Argument must be passed on stack. "
+ "Not eligible for Tail Call\n"));
}
// Get a count of how many bytes are to be pushed on the stack.
unsigned NumBytes = CCInfo.getNextStackOffset();
SmallVector<std::pair<unsigned, SDValue>, 16> RegsToPass;
SmallVector<SDValue, 8> MemOpChains;
- const HexagonRegisterInfo *QRI = Subtarget->getRegisterInfo();
- SDValue StackPtr =
- DAG.getCopyFromReg(Chain, dl, QRI->getStackRegister(), getPointerTy());
+ auto &HRI = *Subtarget.getRegisterInfo();
+ SDValue StackPtr = DAG.getCopyFromReg(Chain, dl, HRI.getStackRegister(),
+ getPointerTy());
// Walk the register/memloc assignments, inserting copies/loads.
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
default:
// Loc info must be one of Full, SExt, ZExt, or AExt.
llvm_unreachable("Unknown loc info!");
+ case CCValAssign::BCvt:
case CCValAssign::Full:
break;
case CCValAssign::SExt:
if (VA.isMemLoc()) {
unsigned LocMemOffset = VA.getLocMemOffset();
- SDValue PtrOff = DAG.getConstant(LocMemOffset, StackPtr.getValueType());
- PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
-
+ SDValue MemAddr = DAG.getConstant(LocMemOffset, dl,
+ StackPtr.getValueType());
+ MemAddr = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, MemAddr);
if (Flags.isByVal()) {
// The argument is a struct passed by value. According to LLVM, "Arg"
// is is pointer.
- MemOpChains.push_back(CreateCopyOfByValArgument(Arg, PtrOff, Chain,
+ MemOpChains.push_back(CreateCopyOfByValArgument(Arg, MemAddr, Chain,
Flags, DAG, dl));
} else {
- // The argument is not passed by value. "Arg" is a buildin type. It is
- // not a pointer.
- MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
- MachinePointerInfo(),false, false,
- 0));
+ MachinePointerInfo LocPI = MachinePointerInfo::getStack(LocMemOffset);
+ SDValue S = DAG.getStore(Chain, dl, Arg, MemAddr, LocPI, false,
+ false, 0);
+ MemOpChains.push_back(S);
}
continue;
}
// Arguments that can be passed on register must be kept at RegsToPass
// vector.
- if (VA.isRegLoc()) {
+ if (VA.isRegLoc())
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
- }
}
// Transform all store nodes into one single node because all store
// nodes are independent of each other.
- if (!MemOpChains.empty()) {
+ if (!MemOpChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains);
- }
- if (!isTailCall)
- Chain = DAG.getCALLSEQ_START(Chain, DAG.getConstant(NumBytes,
- getPointerTy(), true),
- dl);
+ if (!isTailCall) {
+ SDValue C = DAG.getConstant(NumBytes, dl, getPointerTy(), true);
+ Chain = DAG.getCALLSEQ_START(Chain, C, dl);
+ }
// Build a sequence of copy-to-reg nodes chained together with token
// chain and flag operands which copy the outgoing args into registers.
RegsToPass[i].second, InFlag);
InFlag = Chain.getValue(1);
}
- }
-
- // For tail calls lower the arguments to the 'real' stack slot.
- if (isTailCall) {
+ } else {
+ // For tail calls lower the arguments to the 'real' stack slot.
+ //
// Force all the incoming stack arguments to be loaded from the stack
// before any new outgoing arguments are stored to the stack, because the
// outgoing stack slots may alias the incoming argument stack slots, and
RegsToPass[i].second, InFlag);
InFlag = Chain.getValue(1);
}
- InFlag =SDValue();
+ InFlag = SDValue();
}
// If the callee is a GlobalAddress/ExternalSymbol node (quite common, every
if (flag_aligned_memcpy) {
const char *MemcpyName =
"__hexagon_memcpy_likely_aligned_min32bytes_mult8bytes";
- Callee =
- DAG.getTargetExternalSymbol(MemcpyName, getPointerTy());
+ Callee = DAG.getTargetExternalSymbol(MemcpyName, getPointerTy());
flag_aligned_memcpy = false;
} else if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl, getPointerTy());
RegsToPass[i].second.getValueType()));
}
- if (InFlag.getNode()) {
+ if (InFlag.getNode())
Ops.push_back(InFlag);
- }
- if (isTailCall)
+ if (isTailCall) {
+ MF.getFrameInfo()->setHasTailCall();
return DAG.getNode(HexagonISD::TC_RETURN, dl, NodeTys, Ops);
+ }
int OpCode = doesNotReturn ? HexagonISD::CALLv3nr : HexagonISD::CALLv3;
Chain = DAG.getNode(OpCode, dl, NodeTys, Ops);
InFlag = Chain.getValue(1);
// Create the CALLSEQ_END node.
- Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
- DAG.getIntPtrConstant(0, true), InFlag, dl);
+ Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, dl, true),
+ DAG.getIntPtrConstant(0, dl, true), InFlag, dl);
InFlag = Chain.getValue(1);
// Handle result values, copying them out of physregs into vregs that we
SDValue &Offset, bool &isInc,
SelectionDAG &DAG) {
if (Ptr->getOpcode() != ISD::ADD)
- return false;
+ return false;
if (VT == MVT::i64 || VT == MVT::i32 || VT == MVT::i16 || VT == MVT::i8) {
isInc = (Ptr->getOpcode() == ISD::ADD);
SelectionDAG &DAG) const {
SDNode *Node = Op.getNode();
MachineFunction &MF = DAG.getMachineFunction();
- HexagonMachineFunctionInfo *FuncInfo =
- MF.getInfo<HexagonMachineFunctionInfo>();
+ auto &FuncInfo = *MF.getInfo<HexagonMachineFunctionInfo>();
switch (Node->getOpcode()) {
case ISD::INLINEASM: {
unsigned NumOps = Node->getNumOperands();
--NumOps; // Ignore the flag operand.
for (unsigned i = InlineAsm::Op_FirstOperand; i != NumOps;) {
- if (FuncInfo->hasClobberLR())
+ if (FuncInfo.hasClobberLR())
break;
unsigned Flags =
cast<ConstantSDNode>(Node->getOperand(i))->getZExtValue();
cast<RegisterSDNode>(Node->getOperand(i))->getReg();
// Check it to be lr
- const HexagonRegisterInfo *QRI = Subtarget->getRegisterInfo();
+ const HexagonRegisterInfo *QRI = Subtarget.getRegisterInfo();
if (Reg == QRI->getRARegister()) {
- FuncInfo->setHasClobberLR(true);
+ FuncInfo.setHasClobberLR(true);
break;
}
}
SDValue JumpTableBase = DAG.getNode(HexagonISD::JT, dl,
getPointerTy(), TargetJT);
SDValue ShiftIndex = DAG.getNode(ISD::SHL, dl, MVT::i32, Index,
- DAG.getConstant(2, MVT::i32));
+ DAG.getConstant(2, dl, MVT::i32));
SDValue JTAddress = DAG.getNode(ISD::ADD, dl, MVT::i32, JumpTableBase,
ShiftIndex);
SDValue LoadTarget = DAG.getLoad(MVT::i32, dl, Chain, JTAddress,
SelectionDAG &DAG) const {
SDValue Chain = Op.getOperand(0);
SDValue Size = Op.getOperand(1);
+ SDValue Align = Op.getOperand(2);
SDLoc dl(Op);
- unsigned SPReg = getStackPointerRegisterToSaveRestore();
+ ConstantSDNode *AlignConst = dyn_cast<ConstantSDNode>(Align);
+ assert(AlignConst && "Non-constant Align in LowerDYNAMIC_STACKALLOC");
- // Get a reference to the stack pointer.
- SDValue StackPointer = DAG.getCopyFromReg(Chain, dl, SPReg, MVT::i32);
+ unsigned A = AlignConst->getSExtValue();
+ auto &HFI = *Subtarget.getFrameLowering();
+ // "Zero" means natural stack alignment.
+ if (A == 0)
+ A = HFI.getStackAlignment();
- // Subtract the dynamic size from the actual stack size to
- // obtain the new stack size.
- SDValue Sub = DAG.getNode(ISD::SUB, dl, MVT::i32, StackPointer, Size);
+ DEBUG({
+ dbgs () << LLVM_FUNCTION_NAME << " Align: " << A << " Size: ";
+ Size.getNode()->dump(&DAG);
+ dbgs() << "\n";
+ });
- //
- // For Hexagon, the outgoing memory arguments area should be on top of the
- // alloca area on the stack i.e., the outgoing memory arguments should be
- // at a lower address than the alloca area. Move the alloca area down the
- // stack by adding back the space reserved for outgoing arguments to SP
- // here.
- //
- // We do not know what the size of the outgoing args is at this point.
- // So, we add a pseudo instruction ADJDYNALLOC that will adjust the
- // stack pointer. We patch this instruction with the correct, known
- // offset in emitPrologue().
- //
- // Use a placeholder immediate (zero) for now. This will be patched up
- // by emitPrologue().
- SDValue ArgAdjust = DAG.getNode(HexagonISD::ADJDYNALLOC, dl,
- MVT::i32,
- Sub,
- DAG.getConstant(0, MVT::i32));
-
- // The Sub result contains the new stack start address, so it
- // must be placed in the stack pointer register.
- const HexagonRegisterInfo *QRI = Subtarget->getRegisterInfo();
- SDValue CopyChain = DAG.getCopyToReg(Chain, dl, QRI->getStackRegister(), Sub);
-
- SDValue Ops[2] = { ArgAdjust, CopyChain };
- return DAG.getMergeValues(Ops, dl);
+ SDValue AC = DAG.getConstant(A, dl, MVT::i32);
+ SDVTList VTs = DAG.getVTList(MVT::i32, MVT::Other);
+ return DAG.getNode(HexagonISD::ALLOCA, dl, VTs, Chain, Size, AC);
}
SDValue
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
- HexagonMachineFunctionInfo *FuncInfo =
- MF.getInfo<HexagonMachineFunctionInfo>();
-
+ auto &FuncInfo = *MF.getInfo<HexagonMachineFunctionInfo>();
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
HEXAGON_LRFP_SIZE +
CCInfo.getNextStackOffset(),
true);
- FuncInfo->setVarArgsFrameIndex(FrameIndex);
+ FuncInfo.setVarArgsFrameIndex(FrameIndex);
}
return Chain;
SDValue InpVal = Op.getOperand(0);
if (isa<ConstantSDNode>(InpVal)) {
uint64_t V = cast<ConstantSDNode>(InpVal)->getZExtValue();
- return DAG.getTargetConstant(countPopulation(V), MVT::i64);
+ return DAG.getTargetConstant(countPopulation(V), dl, MVT::i64);
}
SDValue PopOut = DAG.getNode(HexagonISD::POPCOUNT, dl, MVT::i32, InpVal);
return DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i64, PopOut);
LoadNode->isInvariant(),
Alignment);
// Base+2 load.
- SDValue Increment = DAG.getConstant(2, MVT::i32);
+ SDValue Increment = DAG.getConstant(2, DL, MVT::i32);
Ptr = DAG.getNode(ISD::ADD, DL, Base.getValueType(), Base, Increment);
Loads[1] = DAG.getExtLoad(Ext, DL, MVT::i32, Chain, Ptr,
LoadNode->getPointerInfo(), MVT::i16,
LoadNode->isInvariant(),
Alignment);
// SHL 16, then OR base and base+2.
- SDValue ShiftAmount = DAG.getConstant(16, MVT::i32);
+ SDValue ShiftAmount = DAG.getConstant(16, DL, MVT::i32);
SDValue Tmp1 = DAG.getNode(ISD::SHL, DL, MVT::i32, Loads[1], ShiftAmount);
SDValue Tmp2 = DAG.getNode(ISD::OR, DL, MVT::i32, Tmp1, Loads[0]);
// Base + 4.
- Increment = DAG.getConstant(4, MVT::i32);
+ Increment = DAG.getConstant(4, DL, MVT::i32);
Ptr = DAG.getNode(ISD::ADD, DL, Base.getValueType(), Base, Increment);
Loads[2] = DAG.getExtLoad(Ext, DL, MVT::i32, Chain, Ptr,
LoadNode->getPointerInfo(), MVT::i16,
LoadNode->isInvariant(),
Alignment);
// Base + 6.
- Increment = DAG.getConstant(6, MVT::i32);
+ Increment = DAG.getConstant(6, DL, MVT::i32);
Ptr = DAG.getNode(ISD::ADD, DL, Base.getValueType(), Base, Increment);
Loads[3] = DAG.getExtLoad(Ext, DL, MVT::i32, Chain, Ptr,
LoadNode->getPointerInfo(), MVT::i16,
else
Res = DAG.getTargetConstantPool(CP->getConstVal(), ValTy,
CP->getAlignment());
- return DAG.getNode(HexagonISD::CONST32, dl, ValTy, Res);
+ return DAG.getNode(HexagonISD::CP, dl, ValTy, Res);
}
SDValue
HexagonTargetLowering::LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const {
- const TargetRegisterInfo *TRI = Subtarget->getRegisterInfo();
+ const HexagonRegisterInfo &HRI = *Subtarget.getRegisterInfo();
MachineFunction &MF = DAG.getMachineFunction();
- MachineFrameInfo *MFI = MF.getFrameInfo();
- MFI->setReturnAddressIsTaken(true);
+ MachineFrameInfo &MFI = *MF.getFrameInfo();
+ MFI.setReturnAddressIsTaken(true);
if (verifyReturnAddressArgumentIsConstant(Op, DAG))
return SDValue();
unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
if (Depth) {
SDValue FrameAddr = LowerFRAMEADDR(Op, DAG);
- SDValue Offset = DAG.getConstant(4, MVT::i32);
+ SDValue Offset = DAG.getConstant(4, dl, MVT::i32);
return DAG.getLoad(VT, dl, DAG.getEntryNode(),
DAG.getNode(ISD::ADD, dl, VT, FrameAddr, Offset),
MachinePointerInfo(), false, false, false, 0);
}
// Return LR, which contains the return address. Mark it an implicit live-in.
- unsigned Reg = MF.addLiveIn(TRI->getRARegister(), getRegClassFor(MVT::i32));
+ unsigned Reg = MF.addLiveIn(HRI.getRARegister(), getRegClassFor(MVT::i32));
return DAG.getCopyFromReg(DAG.getEntryNode(), dl, Reg, VT);
}
SDValue
HexagonTargetLowering::LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const {
- const HexagonRegisterInfo *TRI = Subtarget->getRegisterInfo();
- MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
- MFI->setFrameAddressIsTaken(true);
+ const HexagonRegisterInfo &HRI = *Subtarget.getRegisterInfo();
+ MachineFrameInfo &MFI = *DAG.getMachineFunction().getFrameInfo();
+ MFI.setFrameAddressIsTaken(true);
EVT VT = Op.getValueType();
SDLoc dl(Op);
unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), dl,
- TRI->getFrameRegister(), VT);
+ HRI.getFrameRegister(), VT);
while (Depth--)
FrameAddr = DAG.getLoad(VT, dl, DAG.getEntryNode(), FrameAddr,
MachinePointerInfo(),
HexagonTargetLowering::HexagonTargetLowering(const TargetMachine &TM,
const HexagonSubtarget &STI)
- : TargetLowering(TM), Subtarget(&STI) {
-
- // Set up the register classes.
- addRegisterClass(MVT::v2i1, &Hexagon::PredRegsRegClass); // bbbbaaaa
- addRegisterClass(MVT::v4i1, &Hexagon::PredRegsRegClass); // ddccbbaa
- addRegisterClass(MVT::v8i1, &Hexagon::PredRegsRegClass); // hgfedcba
- addRegisterClass(MVT::i32, &Hexagon::IntRegsRegClass);
- addRegisterClass(MVT::v4i8, &Hexagon::IntRegsRegClass);
+ : TargetLowering(TM), HTM(static_cast<const HexagonTargetMachine&>(TM)),
+ Subtarget(STI) {
+ bool IsV4 = !Subtarget.hasV5TOps();
+ auto &HRI = *Subtarget.getRegisterInfo();
+
+ setPrefLoopAlignment(4);
+ setPrefFunctionAlignment(4);
+ setMinFunctionAlignment(2);
+ setInsertFencesForAtomic(false);
+ setExceptionPointerRegister(Hexagon::R0);
+ setExceptionSelectorRegister(Hexagon::R1);
+ setStackPointerRegisterToSaveRestore(HRI.getStackRegister());
+
+ if (EnableHexSDNodeSched)
+ setSchedulingPreference(Sched::VLIW);
+ else
+ setSchedulingPreference(Sched::Source);
+
+ // Limits for inline expansion of memcpy/memmove
+ MaxStoresPerMemcpy = MaxStoresPerMemcpyCL;
+ MaxStoresPerMemcpyOptSize = MaxStoresPerMemcpyOptSizeCL;
+ MaxStoresPerMemmove = MaxStoresPerMemmoveCL;
+ MaxStoresPerMemmoveOptSize = MaxStoresPerMemmoveOptSizeCL;
+ MaxStoresPerMemset = MaxStoresPerMemsetCL;
+ MaxStoresPerMemsetOptSize = MaxStoresPerMemsetOptSizeCL;
+
+ //
+ // Set up register classes.
+ //
+
+ addRegisterClass(MVT::i1, &Hexagon::PredRegsRegClass);
+ addRegisterClass(MVT::v2i1, &Hexagon::PredRegsRegClass); // bbbbaaaa
+ addRegisterClass(MVT::v4i1, &Hexagon::PredRegsRegClass); // ddccbbaa
+ addRegisterClass(MVT::v8i1, &Hexagon::PredRegsRegClass); // hgfedcba
+ addRegisterClass(MVT::i32, &Hexagon::IntRegsRegClass);
+ addRegisterClass(MVT::v4i8, &Hexagon::IntRegsRegClass);
addRegisterClass(MVT::v2i16, &Hexagon::IntRegsRegClass);
- promoteLdStType(MVT::v4i8, MVT::i32);
- promoteLdStType(MVT::v2i16, MVT::i32);
+ addRegisterClass(MVT::i64, &Hexagon::DoubleRegsRegClass);
+ addRegisterClass(MVT::v8i8, &Hexagon::DoubleRegsRegClass);
+ addRegisterClass(MVT::v4i16, &Hexagon::DoubleRegsRegClass);
+ addRegisterClass(MVT::v2i32, &Hexagon::DoubleRegsRegClass);
- if (Subtarget->hasV5TOps()) {
+ if (Subtarget.hasV5TOps()) {
addRegisterClass(MVT::f32, &Hexagon::IntRegsRegClass);
addRegisterClass(MVT::f64, &Hexagon::DoubleRegsRegClass);
}
- addRegisterClass(MVT::i64, &Hexagon::DoubleRegsRegClass);
- addRegisterClass(MVT::v8i8, &Hexagon::DoubleRegsRegClass);
- addRegisterClass(MVT::v4i16, &Hexagon::DoubleRegsRegClass);
- addRegisterClass(MVT::v2i32, &Hexagon::DoubleRegsRegClass);
- promoteLdStType(MVT::v8i8, MVT::i64);
+ //
+ // Handling of scalar operations.
+ //
+ // All operations default to "legal", except:
+ // - indexed loads and stores (pre-/post-incremented),
+ // - ANY_EXTEND_VECTOR_INREG, ATOMIC_CMP_SWAP_WITH_SUCCESS, CONCAT_VECTORS,
+ // ConstantFP, DEBUGTRAP, FCEIL, FCOPYSIGN, FEXP, FEXP2, FFLOOR, FGETSIGN,
+ // FLOG, FLOG2, FLOG10, FMAXNUM, FMINNUM, FNEARBYINT, FRINT, FROUND, TRAP,
+ // FTRUNC, PREFETCH, SIGN_EXTEND_VECTOR_INREG, ZERO_EXTEND_VECTOR_INREG,
+ // which default to "expand" for at least one type.
+
+ // Misc operations.
+ setOperationAction(ISD::ConstantFP, MVT::f32, Legal); // Default: expand
+ setOperationAction(ISD::ConstantFP, MVT::f64, Legal); // Default: expand
- // Custom lower v4i16 load only. Let v4i16 store to be
- // promoted for now.
- setOperationAction(ISD::LOAD, MVT::v4i16, Custom);
- AddPromotedToType(ISD::LOAD, MVT::v4i16, MVT::i64);
- setOperationAction(ISD::STORE, MVT::v4i16, Promote);
- AddPromotedToType(ISD::STORE, MVT::v4i16, MVT::i64);
- promoteLdStType(MVT::v2i32, MVT::i64);
+ setOperationAction(ISD::ConstantPool, MVT::i32, Custom);
+ setOperationAction(ISD::BUILD_PAIR, MVT::i64, Expand);
+ setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
+ setOperationAction(ISD::INLINEASM, MVT::Other, Custom);
+ setOperationAction(ISD::EH_RETURN, MVT::Other, Custom);
+ setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Custom);
- for (unsigned i = (unsigned) MVT::FIRST_VECTOR_VALUETYPE;
- i <= (unsigned) MVT::LAST_VECTOR_VALUETYPE; ++i) {
- MVT::SimpleValueType VT = (MVT::SimpleValueType) i;
-
- // Hexagon does not have support for the following operations,
- // so they need to be expanded.
- setOperationAction(ISD::SELECT, VT, Expand);
- setOperationAction(ISD::SDIV, VT, Expand);
- setOperationAction(ISD::SREM, VT, Expand);
- setOperationAction(ISD::UDIV, VT, Expand);
- setOperationAction(ISD::UREM, VT, Expand);
- setOperationAction(ISD::ROTL, VT, Expand);
- setOperationAction(ISD::ROTR, VT, Expand);
- setOperationAction(ISD::FDIV, VT, Expand);
- setOperationAction(ISD::FNEG, VT, Expand);
- setOperationAction(ISD::UMUL_LOHI, VT, Expand);
- setOperationAction(ISD::SMUL_LOHI, VT, Expand);
- setOperationAction(ISD::UDIVREM, VT, Expand);
- setOperationAction(ISD::SDIVREM, VT, Expand);
- setOperationAction(ISD::FPOW, VT, Expand);
- setOperationAction(ISD::CTPOP, VT, Expand);
- setOperationAction(ISD::CTLZ, VT, Expand);
- setOperationAction(ISD::CTLZ_ZERO_UNDEF, VT, Expand);
- setOperationAction(ISD::CTTZ, VT, Expand);
- setOperationAction(ISD::CTTZ_ZERO_UNDEF, VT, Expand);
-
- // Expand all any extend loads.
- for (unsigned j = (unsigned) MVT::FIRST_VECTOR_VALUETYPE;
- j <= (unsigned) MVT::LAST_VECTOR_VALUETYPE; ++j)
- setLoadExtAction(ISD::EXTLOAD, (MVT::SimpleValueType) j, VT, Expand);
-
- // Expand all trunc stores.
- for (unsigned TargetVT = (unsigned) MVT::FIRST_VECTOR_VALUETYPE;
- TargetVT <= (unsigned) MVT::LAST_VECTOR_VALUETYPE; ++TargetVT)
- setTruncStoreAction(VT, (MVT::SimpleValueType) TargetVT, Expand);
-
- setOperationAction(ISD::VECTOR_SHUFFLE, VT, Expand);
- setOperationAction(ISD::ConstantPool, VT, Expand);
- setOperationAction(ISD::SCALAR_TO_VECTOR, VT, Expand);
- setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Expand);
- setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Expand);
- setOperationAction(ISD::BUILD_VECTOR, VT, Expand);
- setOperationAction(ISD::EXTRACT_SUBVECTOR, VT, Expand);
- setOperationAction(ISD::INSERT_SUBVECTOR, VT, Expand);
- setOperationAction(ISD::CONCAT_VECTORS, VT, Expand);
- setOperationAction(ISD::SRA, VT, Custom);
- setOperationAction(ISD::SHL, VT, Custom);
- setOperationAction(ISD::SRL, VT, Custom);
+ // Custom legalize GlobalAddress nodes into CONST32.
+ setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
+ setOperationAction(ISD::GlobalAddress, MVT::i8, Custom);
+ setOperationAction(ISD::BlockAddress, MVT::i32, Custom);
- if (!isTypeLegal(VT))
- continue;
+ // Hexagon needs to optimize cases with negative constants.
+ setOperationAction(ISD::SETCC, MVT::i8, Custom);
+ setOperationAction(ISD::SETCC, MVT::i16, Custom);
- setOperationAction(ISD::ADD, VT, Legal);
- setOperationAction(ISD::SUB, VT, Legal);
- setOperationAction(ISD::MUL, VT, Legal);
+ // VASTART needs to be custom lowered to use the VarArgsFrameIndex.
+ setOperationAction(ISD::VASTART, MVT::Other, Custom);
+ setOperationAction(ISD::VAEND, MVT::Other, Expand);
+ setOperationAction(ISD::VAARG, MVT::Other, Expand);
- setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
- setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
- setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
- setOperationAction(ISD::EXTRACT_SUBVECTOR, VT, Custom);
- setOperationAction(ISD::INSERT_SUBVECTOR, VT, Custom);
- setOperationAction(ISD::CONCAT_VECTORS, VT, Custom);
- }
+ setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
+ setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
+ setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Custom);
- setOperationAction(ISD::SETCC, MVT::v2i16, Custom);
- setOperationAction(ISD::VSELECT, MVT::v2i16, Custom);
- setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v8i8, Custom);
- setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v4i16, Custom);
+ if (EmitJumpTables)
+ setOperationAction(ISD::BR_JT, MVT::Other, Custom);
+ else
+ setOperationAction(ISD::BR_JT, MVT::Other, Expand);
+ // Increase jump tables cutover to 5, was 4.
+ setMinimumJumpTableEntries(MinimumJumpTables);
+
+ // Hexagon has instructions for add/sub with carry. The problem with
+ // modeling these instructions is that they produce 2 results: Rdd and Px.
+ // To model the update of Px, we will have to use Defs[p0..p3] which will
+ // cause any predicate live range to spill. So, we pretend we dont't have
+ // these instructions.
+ setOperationAction(ISD::ADDE, MVT::i8, Expand);
+ setOperationAction(ISD::ADDE, MVT::i16, Expand);
+ setOperationAction(ISD::ADDE, MVT::i32, Expand);
+ setOperationAction(ISD::ADDE, MVT::i64, Expand);
+ setOperationAction(ISD::SUBE, MVT::i8, Expand);
+ setOperationAction(ISD::SUBE, MVT::i16, Expand);
+ setOperationAction(ISD::SUBE, MVT::i32, Expand);
+ setOperationAction(ISD::SUBE, MVT::i64, Expand);
+ setOperationAction(ISD::ADDC, MVT::i8, Expand);
+ setOperationAction(ISD::ADDC, MVT::i16, Expand);
+ setOperationAction(ISD::ADDC, MVT::i32, Expand);
+ setOperationAction(ISD::ADDC, MVT::i64, Expand);
+ setOperationAction(ISD::SUBC, MVT::i8, Expand);
+ setOperationAction(ISD::SUBC, MVT::i16, Expand);
+ setOperationAction(ISD::SUBC, MVT::i32, Expand);
+ setOperationAction(ISD::SUBC, MVT::i64, Expand);
- setOperationAction(ISD::ConstantPool, MVT::i32, Custom);
+ // Only add and sub that detect overflow are the saturating ones.
+ for (MVT VT : MVT::integer_valuetypes()) {
+ setOperationAction(ISD::UADDO, VT, Expand);
+ setOperationAction(ISD::SADDO, VT, Expand);
+ setOperationAction(ISD::USUBO, VT, Expand);
+ setOperationAction(ISD::SSUBO, VT, Expand);
+ }
- addRegisterClass(MVT::i1, &Hexagon::PredRegsRegClass);
+ setOperationAction(ISD::CTLZ, MVT::i8, Promote);
+ setOperationAction(ISD::CTLZ, MVT::i16, Promote);
+ setOperationAction(ISD::CTTZ, MVT::i8, Promote);
+ setOperationAction(ISD::CTTZ, MVT::i16, Promote);
+ setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i8, Promote);
+ setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i16, Promote);
+ setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i8, Promote);
+ setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i16, Promote);
+
+ // In V5, popcount can count # of 1s in i64 but returns i32.
+ // On V4 it will be expanded (set later).
+ setOperationAction(ISD::CTPOP, MVT::i8, Promote);
+ setOperationAction(ISD::CTPOP, MVT::i16, Promote);
+ setOperationAction(ISD::CTPOP, MVT::i32, Promote);
+ setOperationAction(ISD::CTPOP, MVT::i64, Custom);
+
+ // We custom lower i64 to i64 mul, so that it is not considered as a legal
+ // operation. There is a pattern that will match i64 mul and transform it
+ // to a series of instructions.
+ setOperationAction(ISD::MUL, MVT::i64, Expand);
+ setOperationAction(ISD::MULHS, MVT::i64, Expand);
- computeRegisterProperties(Subtarget->getRegisterInfo());
+ for (unsigned IntExpOp :
+ {ISD::SDIV, ISD::UDIV, ISD::SREM, ISD::UREM, ISD::SDIVREM, ISD::UDIVREM,
+ ISD::ROTL, ISD::ROTR, ISD::BSWAP, ISD::SHL_PARTS, ISD::SRA_PARTS,
+ ISD::SRL_PARTS, ISD::SMUL_LOHI, ISD::UMUL_LOHI}) {
+ setOperationAction(IntExpOp, MVT::i32, Expand);
+ setOperationAction(IntExpOp, MVT::i64, Expand);
+ }
- // Align loop entry
- setPrefLoopAlignment(4);
+ for (unsigned FPExpOp :
+ {ISD::FDIV, ISD::FREM, ISD::FSQRT, ISD::FSIN, ISD::FCOS, ISD::FSINCOS,
+ ISD::FPOW, ISD::FCOPYSIGN}) {
+ setOperationAction(FPExpOp, MVT::f32, Expand);
+ setOperationAction(FPExpOp, MVT::f64, Expand);
+ }
- // Limits for inline expansion of memcpy/memmove
- MaxStoresPerMemcpy = 6;
- MaxStoresPerMemmove = 6;
+ // No extending loads from i32.
+ for (MVT VT : MVT::integer_valuetypes()) {
+ setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i32, Expand);
+ setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i32, Expand);
+ setLoadExtAction(ISD::EXTLOAD, VT, MVT::i32, Expand);
+ }
+ // Turn FP truncstore into trunc + store.
+ setTruncStoreAction(MVT::f64, MVT::f32, Expand);
+ // Turn FP extload into load/fextend.
+ for (MVT VT : MVT::fp_valuetypes())
+ setLoadExtAction(ISD::EXTLOAD, VT, MVT::f32, Expand);
+
+ // Expand BR_CC and SELECT_CC for all integer and fp types.
+ for (MVT VT : MVT::integer_valuetypes()) {
+ setOperationAction(ISD::BR_CC, VT, Expand);
+ setOperationAction(ISD::SELECT_CC, VT, Expand);
+ }
+ for (MVT VT : MVT::fp_valuetypes()) {
+ setOperationAction(ISD::BR_CC, VT, Expand);
+ setOperationAction(ISD::SELECT_CC, VT, Expand);
+ }
+ setOperationAction(ISD::BR_CC, MVT::Other, Expand);
//
- // Library calls for unsupported operations
+ // Handling of vector operations.
//
- setLibcallName(RTLIB::SINTTOFP_I128_F64, "__hexagon_floattidf");
- setLibcallName(RTLIB::SINTTOFP_I128_F32, "__hexagon_floattisf");
-
- setLibcallName(RTLIB::FPTOUINT_F32_I128, "__hexagon_fixunssfti");
- setLibcallName(RTLIB::FPTOUINT_F64_I128, "__hexagon_fixunsdfti");
-
- setLibcallName(RTLIB::FPTOSINT_F32_I128, "__hexagon_fixsfti");
- setLibcallName(RTLIB::FPTOSINT_F64_I128, "__hexagon_fixdfti");
-
- setLibcallName(RTLIB::SDIV_I32, "__hexagon_divsi3");
- setOperationAction(ISD::SDIV, MVT::i32, Expand);
- setLibcallName(RTLIB::SREM_I32, "__hexagon_umodsi3");
- setOperationAction(ISD::SREM, MVT::i32, Expand);
-
- setLibcallName(RTLIB::SDIV_I64, "__hexagon_divdi3");
- setOperationAction(ISD::SDIV, MVT::i64, Expand);
- setLibcallName(RTLIB::SREM_I64, "__hexagon_moddi3");
- setOperationAction(ISD::SREM, MVT::i64, Expand);
+ // Custom lower v4i16 load only. Let v4i16 store to be
+ // promoted for now.
+ promoteLdStType(MVT::v4i8, MVT::i32);
+ promoteLdStType(MVT::v2i16, MVT::i32);
+ promoteLdStType(MVT::v8i8, MVT::i64);
+ promoteLdStType(MVT::v2i32, MVT::i64);
- setLibcallName(RTLIB::UDIV_I32, "__hexagon_udivsi3");
- setOperationAction(ISD::UDIV, MVT::i32, Expand);
+ setOperationAction(ISD::LOAD, MVT::v4i16, Custom);
+ setOperationAction(ISD::STORE, MVT::v4i16, Promote);
+ AddPromotedToType(ISD::LOAD, MVT::v4i16, MVT::i64);
+ AddPromotedToType(ISD::STORE, MVT::v4i16, MVT::i64);
- setLibcallName(RTLIB::UDIV_I64, "__hexagon_udivdi3");
- setOperationAction(ISD::UDIV, MVT::i64, Expand);
+ // Set the action for vector operations to "expand", then override it with
+ // either "custom" or "legal" for specific cases.
+ static unsigned VectExpOps[] = {
+ // Integer arithmetic:
+ ISD::ADD, ISD::SUB, ISD::MUL, ISD::SDIV, ISD::UDIV,
+ ISD::SREM, ISD::UREM, ISD::SDIVREM, ISD::UDIVREM, ISD::ADDC,
+ ISD::SUBC, ISD::SADDO, ISD::UADDO, ISD::SSUBO, ISD::USUBO,
+ ISD::SMUL_LOHI, ISD::UMUL_LOHI,
+ // Logical/bit:
+ ISD::AND, ISD::OR, ISD::XOR, ISD::ROTL, ISD::ROTR,
+ ISD::CTPOP, ISD::CTLZ, ISD::CTTZ, ISD::CTLZ_ZERO_UNDEF,
+ ISD::CTTZ_ZERO_UNDEF,
+ // Floating point arithmetic/math functions:
+ ISD::FADD, ISD::FSUB, ISD::FMUL, ISD::FMA, ISD::FDIV,
+ ISD::FREM, ISD::FNEG, ISD::FABS, ISD::FSQRT, ISD::FSIN,
+ ISD::FCOS, ISD::FPOWI, ISD::FPOW, ISD::FLOG, ISD::FLOG2,
+ ISD::FLOG10, ISD::FEXP, ISD::FEXP2, ISD::FCEIL, ISD::FTRUNC,
+ ISD::FRINT, ISD::FNEARBYINT, ISD::FROUND, ISD::FFLOOR,
+ ISD::FMINNUM, ISD::FMAXNUM, ISD::FSINCOS,
+ // Misc:
+ ISD::SELECT, ISD::ConstantPool,
+ // Vector:
+ ISD::BUILD_VECTOR, ISD::SCALAR_TO_VECTOR,
+ ISD::EXTRACT_VECTOR_ELT, ISD::INSERT_VECTOR_ELT,
+ ISD::EXTRACT_SUBVECTOR, ISD::INSERT_SUBVECTOR,
+ ISD::CONCAT_VECTORS, ISD::VECTOR_SHUFFLE
+ };
- setLibcallName(RTLIB::UREM_I32, "__hexagon_umodsi3");
- setOperationAction(ISD::UREM, MVT::i32, Expand);
+ for (MVT VT : MVT::vector_valuetypes()) {
+ for (unsigned VectExpOp : VectExpOps)
+ setOperationAction(VectExpOp, VT, Expand);
- setLibcallName(RTLIB::UREM_I64, "__hexagon_umoddi3");
- setOperationAction(ISD::UREM, MVT::i64, Expand);
+ // Expand all extended loads and truncating stores:
+ for (MVT TargetVT : MVT::vector_valuetypes()) {
+ setLoadExtAction(ISD::EXTLOAD, TargetVT, VT, Expand);
+ setTruncStoreAction(VT, TargetVT, Expand);
+ }
- setLibcallName(RTLIB::DIV_F32, "__hexagon_divsf3");
- setOperationAction(ISD::FDIV, MVT::f32, Expand);
+ setOperationAction(ISD::SRA, VT, Custom);
+ setOperationAction(ISD::SHL, VT, Custom);
+ setOperationAction(ISD::SRL, VT, Custom);
+ }
- setLibcallName(RTLIB::DIV_F64, "__hexagon_divdf3");
- setOperationAction(ISD::FDIV, MVT::f64, Expand);
+ // Types natively supported:
+ for (MVT NativeVT : {MVT::v2i1, MVT::v4i1, MVT::v8i1, MVT::v32i1, MVT::v64i1,
+ MVT::v4i8, MVT::v8i8, MVT::v2i16, MVT::v4i16, MVT::v1i32,
+ MVT::v2i32, MVT::v1i64}) {
+ setOperationAction(ISD::BUILD_VECTOR, NativeVT, Custom);
+ setOperationAction(ISD::EXTRACT_VECTOR_ELT, NativeVT, Custom);
+ setOperationAction(ISD::INSERT_VECTOR_ELT, NativeVT, Custom);
+ setOperationAction(ISD::EXTRACT_SUBVECTOR, NativeVT, Custom);
+ setOperationAction(ISD::INSERT_SUBVECTOR, NativeVT, Custom);
+ setOperationAction(ISD::CONCAT_VECTORS, NativeVT, Custom);
- setLibcallName(RTLIB::ADD_F64, "__hexagon_adddf3");
- setLibcallName(RTLIB::SUB_F64, "__hexagon_subdf3");
- setLibcallName(RTLIB::MUL_F64, "__hexagon_muldf3");
+ setOperationAction(ISD::ADD, NativeVT, Legal);
+ setOperationAction(ISD::SUB, NativeVT, Legal);
+ setOperationAction(ISD::MUL, NativeVT, Legal);
+ setOperationAction(ISD::AND, NativeVT, Legal);
+ setOperationAction(ISD::OR, NativeVT, Legal);
+ setOperationAction(ISD::XOR, NativeVT, Legal);
+ }
- setOperationAction(ISD::FSQRT, MVT::f32, Expand);
- setOperationAction(ISD::FSQRT, MVT::f64, Expand);
- setOperationAction(ISD::FSIN, MVT::f32, Expand);
- setOperationAction(ISD::FSIN, MVT::f64, Expand);
+ setOperationAction(ISD::SETCC, MVT::v2i16, Custom);
+ setOperationAction(ISD::VSELECT, MVT::v2i16, Custom);
+ setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v4i16, Custom);
+ setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v8i8, Custom);
- if (Subtarget->hasV5TOps()) {
- // Hexagon V5 Support.
- setOperationAction(ISD::FADD, MVT::f32, Legal);
+ // Subtarget-specific operation actions.
+ //
+ if (Subtarget.hasV5TOps()) {
+ setOperationAction(ISD::FMA, MVT::f64, Expand);
setOperationAction(ISD::FADD, MVT::f64, Expand);
- setOperationAction(ISD::FSUB, MVT::f32, Legal);
setOperationAction(ISD::FSUB, MVT::f64, Expand);
setOperationAction(ISD::FMUL, MVT::f64, Expand);
- setOperationAction(ISD::FP_EXTEND, MVT::f32, Legal);
- setCondCodeAction(ISD::SETOEQ, MVT::f32, Legal);
- setCondCodeAction(ISD::SETOEQ, MVT::f64, Legal);
- setCondCodeAction(ISD::SETUEQ, MVT::f32, Legal);
- setCondCodeAction(ISD::SETUEQ, MVT::f64, Legal);
-
- setCondCodeAction(ISD::SETOGE, MVT::f32, Legal);
- setCondCodeAction(ISD::SETOGE, MVT::f64, Legal);
- setCondCodeAction(ISD::SETUGE, MVT::f32, Legal);
- setCondCodeAction(ISD::SETUGE, MVT::f64, Legal);
-
- setCondCodeAction(ISD::SETOGT, MVT::f32, Legal);
- setCondCodeAction(ISD::SETOGT, MVT::f64, Legal);
- setCondCodeAction(ISD::SETUGT, MVT::f32, Legal);
- setCondCodeAction(ISD::SETUGT, MVT::f64, Legal);
-
- setCondCodeAction(ISD::SETOLE, MVT::f32, Legal);
- setCondCodeAction(ISD::SETOLE, MVT::f64, Legal);
- setCondCodeAction(ISD::SETOLT, MVT::f32, Legal);
- setCondCodeAction(ISD::SETOLT, MVT::f64, Legal);
-
- setOperationAction(ISD::ConstantFP, MVT::f32, Legal);
- setOperationAction(ISD::ConstantFP, MVT::f64, Legal);
-
- setOperationAction(ISD::FP_TO_UINT, MVT::i1, Promote);
- setOperationAction(ISD::FP_TO_SINT, MVT::i1, Promote);
- setOperationAction(ISD::UINT_TO_FP, MVT::i1, Promote);
- setOperationAction(ISD::SINT_TO_FP, MVT::i1, Promote);
-
- setOperationAction(ISD::FP_TO_UINT, MVT::i8, Promote);
- setOperationAction(ISD::FP_TO_SINT, MVT::i8, Promote);
- setOperationAction(ISD::UINT_TO_FP, MVT::i8, Promote);
- setOperationAction(ISD::SINT_TO_FP, MVT::i8, Promote);
+ setOperationAction(ISD::FP_TO_UINT, MVT::i1, Promote);
+ setOperationAction(ISD::FP_TO_UINT, MVT::i8, Promote);
setOperationAction(ISD::FP_TO_UINT, MVT::i16, Promote);
+ setOperationAction(ISD::FP_TO_SINT, MVT::i1, Promote);
+ setOperationAction(ISD::FP_TO_SINT, MVT::i8, Promote);
setOperationAction(ISD::FP_TO_SINT, MVT::i16, Promote);
+ setOperationAction(ISD::UINT_TO_FP, MVT::i1, Promote);
+ setOperationAction(ISD::UINT_TO_FP, MVT::i8, Promote);
setOperationAction(ISD::UINT_TO_FP, MVT::i16, Promote);
+ setOperationAction(ISD::SINT_TO_FP, MVT::i1, Promote);
+ setOperationAction(ISD::SINT_TO_FP, MVT::i8, Promote);
setOperationAction(ISD::SINT_TO_FP, MVT::i16, Promote);
- setOperationAction(ISD::FP_TO_UINT, MVT::i32, Legal);
- setOperationAction(ISD::FP_TO_SINT, MVT::i32, Legal);
- setOperationAction(ISD::UINT_TO_FP, MVT::i32, Legal);
- setOperationAction(ISD::SINT_TO_FP, MVT::i32, Legal);
-
- setOperationAction(ISD::FP_TO_UINT, MVT::i64, Legal);
- setOperationAction(ISD::FP_TO_SINT, MVT::i64, Legal);
- setOperationAction(ISD::UINT_TO_FP, MVT::i64, Legal);
- setOperationAction(ISD::SINT_TO_FP, MVT::i64, Legal);
-
- setOperationAction(ISD::FABS, MVT::f32, Legal);
- setOperationAction(ISD::FABS, MVT::f64, Expand);
-
- setOperationAction(ISD::FNEG, MVT::f32, Legal);
- setOperationAction(ISD::FNEG, MVT::f64, Expand);
- } else {
-
- // Expand fp<->uint.
- setOperationAction(ISD::FP_TO_SINT, MVT::i32, Expand);
- setOperationAction(ISD::FP_TO_UINT, MVT::i32, Expand);
-
+ } else { // V4
setOperationAction(ISD::SINT_TO_FP, MVT::i32, Expand);
+ setOperationAction(ISD::SINT_TO_FP, MVT::i64, Expand);
setOperationAction(ISD::UINT_TO_FP, MVT::i32, Expand);
-
- setLibcallName(RTLIB::SINTTOFP_I64_F32, "__hexagon_floatdisf");
- setLibcallName(RTLIB::UINTTOFP_I64_F32, "__hexagon_floatundisf");
-
- setLibcallName(RTLIB::UINTTOFP_I32_F32, "__hexagon_floatunsisf");
- setLibcallName(RTLIB::SINTTOFP_I32_F32, "__hexagon_floatsisf");
-
- setLibcallName(RTLIB::SINTTOFP_I64_F64, "__hexagon_floatdidf");
- setLibcallName(RTLIB::UINTTOFP_I64_F64, "__hexagon_floatundidf");
-
- setLibcallName(RTLIB::UINTTOFP_I32_F64, "__hexagon_floatunsidf");
- setLibcallName(RTLIB::SINTTOFP_I32_F64, "__hexagon_floatsidf");
-
- setLibcallName(RTLIB::FPTOUINT_F32_I32, "__hexagon_fixunssfsi");
- setLibcallName(RTLIB::FPTOUINT_F32_I64, "__hexagon_fixunssfdi");
-
- setLibcallName(RTLIB::FPTOSINT_F64_I64, "__hexagon_fixdfdi");
- setLibcallName(RTLIB::FPTOSINT_F32_I64, "__hexagon_fixsfdi");
-
- setLibcallName(RTLIB::FPTOUINT_F64_I32, "__hexagon_fixunsdfsi");
- setLibcallName(RTLIB::FPTOUINT_F64_I64, "__hexagon_fixunsdfdi");
-
-
- setLibcallName(RTLIB::ADD_F32, "__hexagon_addsf3");
- setOperationAction(ISD::FADD, MVT::f32, Expand);
- setOperationAction(ISD::FADD, MVT::f64, Expand);
-
- setLibcallName(RTLIB::SUB_F32, "__hexagon_subsf3");
- setOperationAction(ISD::FSUB, MVT::f32, Expand);
- setOperationAction(ISD::FSUB, MVT::f64, Expand);
-
- setLibcallName(RTLIB::FPEXT_F32_F64, "__hexagon_extendsfdf2");
- setOperationAction(ISD::FP_EXTEND, MVT::f32, Expand);
-
- setLibcallName(RTLIB::OEQ_F32, "__hexagon_eqsf2");
- setCondCodeAction(ISD::SETOEQ, MVT::f32, Expand);
-
- setLibcallName(RTLIB::OEQ_F64, "__hexagon_eqdf2");
- setCondCodeAction(ISD::SETOEQ, MVT::f64, Expand);
-
- setLibcallName(RTLIB::OGE_F32, "__hexagon_gesf2");
- setCondCodeAction(ISD::SETOGE, MVT::f32, Expand);
-
- setLibcallName(RTLIB::OGE_F64, "__hexagon_gedf2");
- setCondCodeAction(ISD::SETOGE, MVT::f64, Expand);
-
- setLibcallName(RTLIB::OGT_F32, "__hexagon_gtsf2");
- setCondCodeAction(ISD::SETOGT, MVT::f32, Expand);
-
- setLibcallName(RTLIB::OGT_F64, "__hexagon_gtdf2");
- setCondCodeAction(ISD::SETOGT, MVT::f64, Expand);
-
- setLibcallName(RTLIB::FPTOSINT_F64_I32, "__hexagon_fixdfsi");
+ setOperationAction(ISD::UINT_TO_FP, MVT::i64, Expand);
setOperationAction(ISD::FP_TO_SINT, MVT::f64, Expand);
-
- setLibcallName(RTLIB::FPTOSINT_F32_I32, "__hexagon_fixsfsi");
setOperationAction(ISD::FP_TO_SINT, MVT::f32, Expand);
-
- setLibcallName(RTLIB::OLE_F64, "__hexagon_ledf2");
- setCondCodeAction(ISD::SETOLE, MVT::f64, Expand);
-
- setLibcallName(RTLIB::OLE_F32, "__hexagon_lesf2");
- setCondCodeAction(ISD::SETOLE, MVT::f32, Expand);
-
- setLibcallName(RTLIB::OLT_F64, "__hexagon_ltdf2");
- setCondCodeAction(ISD::SETOLT, MVT::f64, Expand);
-
- setLibcallName(RTLIB::OLT_F32, "__hexagon_ltsf2");
- setCondCodeAction(ISD::SETOLT, MVT::f32, Expand);
-
- setOperationAction(ISD::FMUL, MVT::f64, Expand);
-
- setLibcallName(RTLIB::MUL_F32, "__hexagon_mulsf3");
- setOperationAction(ISD::MUL, MVT::f32, Expand);
-
- setLibcallName(RTLIB::UNE_F64, "__hexagon_nedf2");
+ setOperationAction(ISD::FP_EXTEND, MVT::f32, Expand);
+ setOperationAction(ISD::FP_ROUND, MVT::f64, Expand);
setCondCodeAction(ISD::SETUNE, MVT::f64, Expand);
- setLibcallName(RTLIB::UNE_F32, "__hexagon_nesf2");
+ setOperationAction(ISD::CTPOP, MVT::i8, Expand);
+ setOperationAction(ISD::CTPOP, MVT::i16, Expand);
+ setOperationAction(ISD::CTPOP, MVT::i32, Expand);
+ setOperationAction(ISD::CTPOP, MVT::i64, Expand);
- setLibcallName(RTLIB::SUB_F64, "__hexagon_subdf3");
- setOperationAction(ISD::SUB, MVT::f64, Expand);
-
- setLibcallName(RTLIB::SUB_F32, "__hexagon_subsf3");
- setOperationAction(ISD::SUB, MVT::f32, Expand);
-
- setLibcallName(RTLIB::FPROUND_F64_F32, "__hexagon_truncdfsf2");
- setOperationAction(ISD::FP_ROUND, MVT::f64, Expand);
-
- setLibcallName(RTLIB::UO_F64, "__hexagon_unorddf2");
- setCondCodeAction(ISD::SETUO, MVT::f64, Expand);
-
- setLibcallName(RTLIB::O_F64, "__hexagon_unorddf2");
- setCondCodeAction(ISD::SETO, MVT::f64, Expand);
-
- setLibcallName(RTLIB::O_F32, "__hexagon_unordsf2");
- setCondCodeAction(ISD::SETO, MVT::f32, Expand);
-
- setLibcallName(RTLIB::UO_F32, "__hexagon_unordsf2");
- setCondCodeAction(ISD::SETUO, MVT::f32, Expand);
+ // Expand these operations for both f32 and f64:
+ for (unsigned FPExpOpV4 :
+ {ISD::FADD, ISD::FSUB, ISD::FMUL, ISD::FABS, ISD::FNEG, ISD::FMA}) {
+ setOperationAction(FPExpOpV4, MVT::f32, Expand);
+ setOperationAction(FPExpOpV4, MVT::f64, Expand);
+ }
- setOperationAction(ISD::FABS, MVT::f32, Expand);
- setOperationAction(ISD::FABS, MVT::f64, Expand);
- setOperationAction(ISD::FNEG, MVT::f32, Expand);
- setOperationAction(ISD::FNEG, MVT::f64, Expand);
+ for (ISD::CondCode FPExpCCV4 :
+ {ISD::SETOEQ, ISD::SETOGT, ISD::SETOLT, ISD::SETOGE, ISD::SETOLE,
+ ISD::SETUO, ISD::SETO}) {
+ setCondCodeAction(FPExpCCV4, MVT::f32, Expand);
+ setCondCodeAction(FPExpCCV4, MVT::f64, Expand);
+ }
}
- setLibcallName(RTLIB::SREM_I32, "__hexagon_modsi3");
- setOperationAction(ISD::SREM, MVT::i32, Expand);
-
- setIndexedLoadAction(ISD::POST_INC, MVT::i8, Legal);
- setIndexedLoadAction(ISD::POST_INC, MVT::i16, Legal);
- setIndexedLoadAction(ISD::POST_INC, MVT::i32, Legal);
- setIndexedLoadAction(ISD::POST_INC, MVT::i64, Legal);
-
- setIndexedStoreAction(ISD::POST_INC, MVT::i8, Legal);
- setIndexedStoreAction(ISD::POST_INC, MVT::i16, Legal);
- setIndexedStoreAction(ISD::POST_INC, MVT::i32, Legal);
- setIndexedStoreAction(ISD::POST_INC, MVT::i64, Legal);
-
- setOperationAction(ISD::BUILD_PAIR, MVT::i64, Expand);
-
- // Turn FP extload into load/fextend.
- for (MVT VT : MVT::fp_valuetypes())
- setLoadExtAction(ISD::EXTLOAD, VT, MVT::f32, Expand);
-
- // No extending loads from i32.
- for (MVT VT : MVT::integer_valuetypes()) {
- setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i32, Expand);
- setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i32, Expand);
- setLoadExtAction(ISD::EXTLOAD, VT, MVT::i32, Expand);
+ // Handling of indexed loads/stores: default is "expand".
+ //
+ for (MVT LSXTy : {MVT::i8, MVT::i16, MVT::i32, MVT::i64}) {
+ setIndexedLoadAction(ISD::POST_INC, LSXTy, Legal);
+ setIndexedStoreAction(ISD::POST_INC, LSXTy, Legal);
}
- // Turn FP truncstore into trunc + store.
- setTruncStoreAction(MVT::f64, MVT::f32, Expand);
-
- // Custom legalize GlobalAddress nodes into CONST32.
- setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
- setOperationAction(ISD::GlobalAddress, MVT::i8, Custom);
- setOperationAction(ISD::BlockAddress, MVT::i32, Custom);
- // Truncate action?
- setOperationAction(ISD::TRUNCATE, MVT::i64, Expand);
-
- // Hexagon doesn't have sext_inreg, replace them with shl/sra.
- setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
-
- // Hexagon has no REM or DIVREM operations.
- setOperationAction(ISD::UREM, MVT::i32, Expand);
- setOperationAction(ISD::SREM, MVT::i32, Expand);
- setOperationAction(ISD::SDIVREM, MVT::i32, Expand);
- setOperationAction(ISD::UDIVREM, MVT::i32, Expand);
- setOperationAction(ISD::SREM, MVT::i64, Expand);
- setOperationAction(ISD::SDIVREM, MVT::i64, Expand);
- setOperationAction(ISD::UDIVREM, MVT::i64, Expand);
-
- setOperationAction(ISD::BSWAP, MVT::i64, Expand);
-
- // Lower SELECT_CC to SETCC and SELECT.
- setOperationAction(ISD::SELECT_CC, MVT::i1, Expand);
- setOperationAction(ISD::SELECT_CC, MVT::i32, Expand);
- setOperationAction(ISD::SELECT_CC, MVT::i64, Expand);
-
- if (Subtarget->hasV5TOps()) {
+ computeRegisterProperties(&HRI);
- // We need to make the operation type of SELECT node to be Custom,
- // such that we don't go into the infinite loop of
- // select -> setcc -> select_cc -> select loop.
- setOperationAction(ISD::SELECT, MVT::f32, Custom);
- setOperationAction(ISD::SELECT, MVT::f64, Custom);
+ //
+ // Library calls for unsupported operations
+ //
+ bool FastMath = EnableFastMath;
- setOperationAction(ISD::SELECT_CC, MVT::f32, Expand);
- setOperationAction(ISD::SELECT_CC, MVT::f64, Expand);
+ setLibcallName(RTLIB::SDIV_I32, "__hexagon_divsi3");
+ setLibcallName(RTLIB::SDIV_I64, "__hexagon_divdi3");
+ setLibcallName(RTLIB::UDIV_I32, "__hexagon_udivsi3");
+ setLibcallName(RTLIB::UDIV_I64, "__hexagon_udivdi3");
+ setLibcallName(RTLIB::SREM_I32, "__hexagon_modsi3");
+ setLibcallName(RTLIB::SREM_I64, "__hexagon_moddi3");
+ setLibcallName(RTLIB::UREM_I32, "__hexagon_umodsi3");
+ setLibcallName(RTLIB::UREM_I64, "__hexagon_umoddi3");
- } else {
+ setLibcallName(RTLIB::SINTTOFP_I128_F64, "__hexagon_floattidf");
+ setLibcallName(RTLIB::SINTTOFP_I128_F32, "__hexagon_floattisf");
+ setLibcallName(RTLIB::FPTOUINT_F32_I128, "__hexagon_fixunssfti");
+ setLibcallName(RTLIB::FPTOUINT_F64_I128, "__hexagon_fixunsdfti");
+ setLibcallName(RTLIB::FPTOSINT_F32_I128, "__hexagon_fixsfti");
+ setLibcallName(RTLIB::FPTOSINT_F64_I128, "__hexagon_fixdfti");
- // Hexagon has no select or setcc: expand to SELECT_CC.
- setOperationAction(ISD::SELECT, MVT::f32, Expand);
- setOperationAction(ISD::SELECT, MVT::f64, Expand);
+ if (IsV4) {
+ // Handle single-precision floating point operations on V4.
+ if (FastMath) {
+ setLibcallName(RTLIB::ADD_F32, "__hexagon_fast_addsf3");
+ setLibcallName(RTLIB::SUB_F32, "__hexagon_fast_subsf3");
+ setLibcallName(RTLIB::MUL_F32, "__hexagon_fast_mulsf3");
+ setLibcallName(RTLIB::OGT_F32, "__hexagon_fast_gtsf2");
+ setLibcallName(RTLIB::OLT_F32, "__hexagon_fast_ltsf2");
+ // Double-precision compares.
+ setLibcallName(RTLIB::OGT_F64, "__hexagon_fast_gtdf2");
+ setLibcallName(RTLIB::OLT_F64, "__hexagon_fast_ltdf2");
+ } else {
+ setLibcallName(RTLIB::ADD_F32, "__hexagon_addsf3");
+ setLibcallName(RTLIB::SUB_F32, "__hexagon_subsf3");
+ setLibcallName(RTLIB::MUL_F32, "__hexagon_mulsf3");
+ setLibcallName(RTLIB::OGT_F32, "__hexagon_gtsf2");
+ setLibcallName(RTLIB::OLT_F32, "__hexagon_ltsf2");
+ // Double-precision compares.
+ setLibcallName(RTLIB::OGT_F64, "__hexagon_gtdf2");
+ setLibcallName(RTLIB::OLT_F64, "__hexagon_ltdf2");
+ }
}
- // Hexagon needs to optimize cases with negative constants.
- setOperationAction(ISD::SETCC, MVT::i16, Custom);
- setOperationAction(ISD::SETCC, MVT::i8, Custom);
-
- if (EmitJumpTables) {
- setOperationAction(ISD::BR_JT, MVT::Other, Custom);
+ // This is the only fast library function for sqrtd.
+ if (FastMath)
+ setLibcallName(RTLIB::SQRT_F64, "__hexagon_fast2_sqrtdf2");
+
+ // Prefix is: nothing for "slow-math",
+ // "fast2_" for V4 fast-math and V5+ fast-math double-precision
+ // (actually, keep fast-math and fast-math2 separate for now)
+ if (FastMath) {
+ setLibcallName(RTLIB::ADD_F64, "__hexagon_fast_adddf3");
+ setLibcallName(RTLIB::SUB_F64, "__hexagon_fast_subdf3");
+ setLibcallName(RTLIB::MUL_F64, "__hexagon_fast_muldf3");
+ setLibcallName(RTLIB::DIV_F64, "__hexagon_fast_divdf3");
+ // Calling __hexagon_fast2_divsf3 with fast-math on V5 (ok).
+ setLibcallName(RTLIB::DIV_F32, "__hexagon_fast_divsf3");
} else {
- setOperationAction(ISD::BR_JT, MVT::Other, Expand);
+ setLibcallName(RTLIB::ADD_F64, "__hexagon_adddf3");
+ setLibcallName(RTLIB::SUB_F64, "__hexagon_subdf3");
+ setLibcallName(RTLIB::MUL_F64, "__hexagon_muldf3");
+ setLibcallName(RTLIB::DIV_F64, "__hexagon_divdf3");
+ setLibcallName(RTLIB::DIV_F32, "__hexagon_divsf3");
}
- // Increase jump tables cutover to 5, was 4.
- setMinimumJumpTableEntries(5);
-
- setOperationAction(ISD::BR_CC, MVT::f32, Expand);
- setOperationAction(ISD::BR_CC, MVT::f64, Expand);
- setOperationAction(ISD::BR_CC, MVT::i1, Expand);
- setOperationAction(ISD::BR_CC, MVT::i32, Expand);
- setOperationAction(ISD::BR_CC, MVT::i64, Expand);
-
- setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Custom);
-
- setOperationAction(ISD::FSIN, MVT::f64, Expand);
- setOperationAction(ISD::FCOS, MVT::f64, Expand);
- setOperationAction(ISD::FREM, MVT::f64, Expand);
- setOperationAction(ISD::FSIN, MVT::f32, Expand);
- setOperationAction(ISD::FCOS, MVT::f32, Expand);
- setOperationAction(ISD::FREM, MVT::f32, Expand);
- setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
- setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
-
- // In V4, we have double word add/sub with carry. The problem with
- // modelling this instruction is that it produces 2 results - Rdd and Px.
- // To model update of Px, we will have to use Defs[p0..p3] which will
- // cause any predicate live range to spill. So, we pretend we dont't
- // have these instructions.
- setOperationAction(ISD::ADDE, MVT::i8, Expand);
- setOperationAction(ISD::ADDE, MVT::i16, Expand);
- setOperationAction(ISD::ADDE, MVT::i32, Expand);
- setOperationAction(ISD::ADDE, MVT::i64, Expand);
- setOperationAction(ISD::SUBE, MVT::i8, Expand);
- setOperationAction(ISD::SUBE, MVT::i16, Expand);
- setOperationAction(ISD::SUBE, MVT::i32, Expand);
- setOperationAction(ISD::SUBE, MVT::i64, Expand);
- setOperationAction(ISD::ADDC, MVT::i8, Expand);
- setOperationAction(ISD::ADDC, MVT::i16, Expand);
- setOperationAction(ISD::ADDC, MVT::i32, Expand);
- setOperationAction(ISD::ADDC, MVT::i64, Expand);
- setOperationAction(ISD::SUBC, MVT::i8, Expand);
- setOperationAction(ISD::SUBC, MVT::i16, Expand);
- setOperationAction(ISD::SUBC, MVT::i32, Expand);
- setOperationAction(ISD::SUBC, MVT::i64, Expand);
- // Only add and sub that detect overflow are the saturating ones.
- for (MVT VT : MVT::integer_valuetypes()) {
- setOperationAction(ISD::UADDO, VT, Expand);
- setOperationAction(ISD::SADDO, VT, Expand);
- setOperationAction(ISD::USUBO, VT, Expand);
- setOperationAction(ISD::SSUBO, VT, Expand);
+ if (Subtarget.hasV5TOps()) {
+ if (FastMath)
+ setLibcallName(RTLIB::SQRT_F32, "__hexagon_fast2_sqrtf");
+ else
+ setLibcallName(RTLIB::SQRT_F32, "__hexagon_sqrtf");
+ } else {
+ // V4
+ setLibcallName(RTLIB::SINTTOFP_I32_F32, "__hexagon_floatsisf");
+ setLibcallName(RTLIB::SINTTOFP_I32_F64, "__hexagon_floatsidf");
+ setLibcallName(RTLIB::SINTTOFP_I64_F32, "__hexagon_floatdisf");
+ setLibcallName(RTLIB::SINTTOFP_I64_F64, "__hexagon_floatdidf");
+ setLibcallName(RTLIB::UINTTOFP_I32_F32, "__hexagon_floatunsisf");
+ setLibcallName(RTLIB::UINTTOFP_I32_F64, "__hexagon_floatunsidf");
+ setLibcallName(RTLIB::UINTTOFP_I64_F32, "__hexagon_floatundisf");
+ setLibcallName(RTLIB::UINTTOFP_I64_F64, "__hexagon_floatundidf");
+ setLibcallName(RTLIB::FPTOUINT_F32_I32, "__hexagon_fixunssfsi");
+ setLibcallName(RTLIB::FPTOUINT_F32_I64, "__hexagon_fixunssfdi");
+ setLibcallName(RTLIB::FPTOUINT_F64_I32, "__hexagon_fixunsdfsi");
+ setLibcallName(RTLIB::FPTOUINT_F64_I64, "__hexagon_fixunsdfdi");
+ setLibcallName(RTLIB::FPTOSINT_F32_I32, "__hexagon_fixsfsi");
+ setLibcallName(RTLIB::FPTOSINT_F32_I64, "__hexagon_fixsfdi");
+ setLibcallName(RTLIB::FPTOSINT_F64_I32, "__hexagon_fixdfsi");
+ setLibcallName(RTLIB::FPTOSINT_F64_I64, "__hexagon_fixdfdi");
+ setLibcallName(RTLIB::FPEXT_F32_F64, "__hexagon_extendsfdf2");
+ setLibcallName(RTLIB::FPROUND_F64_F32, "__hexagon_truncdfsf2");
+ setLibcallName(RTLIB::OEQ_F32, "__hexagon_eqsf2");
+ setLibcallName(RTLIB::OEQ_F64, "__hexagon_eqdf2");
+ setLibcallName(RTLIB::OGE_F32, "__hexagon_gesf2");
+ setLibcallName(RTLIB::OGE_F64, "__hexagon_gedf2");
+ setLibcallName(RTLIB::OLE_F32, "__hexagon_lesf2");
+ setLibcallName(RTLIB::OLE_F64, "__hexagon_ledf2");
+ setLibcallName(RTLIB::UNE_F32, "__hexagon_nesf2");
+ setLibcallName(RTLIB::UNE_F64, "__hexagon_nedf2");
+ setLibcallName(RTLIB::UO_F32, "__hexagon_unordsf2");
+ setLibcallName(RTLIB::UO_F64, "__hexagon_unorddf2");
+ setLibcallName(RTLIB::O_F32, "__hexagon_unordsf2");
+ setLibcallName(RTLIB::O_F64, "__hexagon_unorddf2");
}
- setOperationAction(ISD::CTPOP, MVT::i32, Expand);
- setOperationAction(ISD::CTPOP, MVT::i64, Expand);
- setOperationAction(ISD::CTTZ, MVT::i32, Expand);
- setOperationAction(ISD::CTTZ, MVT::i64, Expand);
- setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i32, Expand);
- setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i64, Expand);
- setOperationAction(ISD::CTLZ, MVT::i32, Expand);
- setOperationAction(ISD::CTLZ, MVT::i64, Expand);
- setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i32, Expand);
- setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i64, Expand);
-
- setOperationAction(ISD::ROTL, MVT::i32, Expand);
- setOperationAction(ISD::ROTR, MVT::i32, Expand);
- setOperationAction(ISD::BSWAP, MVT::i32, Expand);
- setOperationAction(ISD::ROTL, MVT::i64, Expand);
- setOperationAction(ISD::ROTR, MVT::i64, Expand);
- setOperationAction(ISD::SHL_PARTS, MVT::i64, Expand);
- setOperationAction(ISD::SRA_PARTS, MVT::i64, Expand);
- setOperationAction(ISD::SRL_PARTS, MVT::i64, Expand);
- setOperationAction(ISD::BR_CC, MVT::i64, Expand);
-
- setOperationAction(ISD::FCOPYSIGN, MVT::f64, Expand);
- setOperationAction(ISD::FCOPYSIGN, MVT::f32, Expand);
- setOperationAction(ISD::FPOW, MVT::f64, Expand);
- setOperationAction(ISD::FPOW, MVT::f32, Expand);
-
- setOperationAction(ISD::SHL_PARTS, MVT::i32, Expand);
- setOperationAction(ISD::SRA_PARTS, MVT::i32, Expand);
- setOperationAction(ISD::SRL_PARTS, MVT::i32, Expand);
-
- setOperationAction(ISD::UMUL_LOHI, MVT::i32, Expand);
- setOperationAction(ISD::SMUL_LOHI, MVT::i32, Expand);
-
- setOperationAction(ISD::MULHS, MVT::i64, Expand);
- setOperationAction(ISD::SMUL_LOHI, MVT::i64, Expand);
- setOperationAction(ISD::UMUL_LOHI, MVT::i64, Expand);
-
- setOperationAction(ISD::EH_RETURN, MVT::Other, Custom);
-
- setExceptionPointerRegister(Hexagon::R0);
- setExceptionSelectorRegister(Hexagon::R1);
-
- // VASTART needs to be custom lowered to use the VarArgsFrameIndex.
- setOperationAction(ISD::VASTART, MVT::Other, Custom);
-
- // Use the default implementation.
- setOperationAction(ISD::VAARG, MVT::Other, Expand);
- setOperationAction(ISD::VACOPY, MVT::Other, Expand);
- setOperationAction(ISD::VAEND, MVT::Other, Expand);
- setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
- setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
-
- setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Custom);
- setOperationAction(ISD::INLINEASM, MVT::Other, Custom);
-
- setMinFunctionAlignment(2);
-
- // Needed for DYNAMIC_STACKALLOC expansion.
- const HexagonRegisterInfo *QRI = Subtarget->getRegisterInfo();
- setStackPointerRegisterToSaveRestore(QRI->getStackRegister());
- setSchedulingPreference(Sched::VLIW);
+ // These cause problems when the shift amount is non-constant.
+ setLibcallName(RTLIB::SHL_I128, nullptr);
+ setLibcallName(RTLIB::SRL_I128, nullptr);
+ setLibcallName(RTLIB::SRA_I128, nullptr);
}
-const char*
-HexagonTargetLowering::getTargetNodeName(unsigned Opcode) const {
- switch (Opcode) {
- default: return nullptr;
- case HexagonISD::CONST32: return "HexagonISD::CONST32";
- case HexagonISD::CONST32_GP: return "HexagonISD::CONST32_GP";
- case HexagonISD::CONST32_Int_Real: return "HexagonISD::CONST32_Int_Real";
- case HexagonISD::ADJDYNALLOC: return "HexagonISD::ADJDYNALLOC";
- case HexagonISD::CMPICC: return "HexagonISD::CMPICC";
- case HexagonISD::CMPFCC: return "HexagonISD::CMPFCC";
- case HexagonISD::BRICC: return "HexagonISD::BRICC";
- case HexagonISD::BRFCC: return "HexagonISD::BRFCC";
- case HexagonISD::SELECT_ICC: return "HexagonISD::SELECT_ICC";
- case HexagonISD::SELECT_FCC: return "HexagonISD::SELECT_FCC";
- case HexagonISD::Hi: return "HexagonISD::Hi";
- case HexagonISD::Lo: return "HexagonISD::Lo";
- case HexagonISD::JT: return "HexagonISD::JT";
- case HexagonISD::CP: return "HexagonISD::CP";
- case HexagonISD::POPCOUNT: return "HexagonISD::POPCOUNT";
- case HexagonISD::COMBINE: return "HexagonISD::COMBINE";
- case HexagonISD::PACKHL: return "HexagonISD::PACKHL";
- case HexagonISD::VSPLATB: return "HexagonISD::VSPLTB";
- case HexagonISD::VSPLATH: return "HexagonISD::VSPLATH";
- case HexagonISD::SHUFFEB: return "HexagonISD::SHUFFEB";
- case HexagonISD::SHUFFEH: return "HexagonISD::SHUFFEH";
- case HexagonISD::SHUFFOB: return "HexagonISD::SHUFFOB";
- case HexagonISD::SHUFFOH: return "HexagonISD::SHUFFOH";
- case HexagonISD::VSXTBH: return "HexagonISD::VSXTBH";
- case HexagonISD::VSXTBW: return "HexagonISD::VSXTBW";
- case HexagonISD::VSRAW: return "HexagonISD::VSRAW";
- case HexagonISD::VSRAH: return "HexagonISD::VSRAH";
- case HexagonISD::VSRLW: return "HexagonISD::VSRLW";
- case HexagonISD::VSRLH: return "HexagonISD::VSRLH";
- case HexagonISD::VSHLW: return "HexagonISD::VSHLW";
- case HexagonISD::VSHLH: return "HexagonISD::VSHLH";
- case HexagonISD::VCMPBEQ: return "HexagonISD::VCMPBEQ";
- case HexagonISD::VCMPBGT: return "HexagonISD::VCMPBGT";
- case HexagonISD::VCMPBGTU: return "HexagonISD::VCMPBGTU";
- case HexagonISD::VCMPHEQ: return "HexagonISD::VCMPHEQ";
- case HexagonISD::VCMPHGT: return "HexagonISD::VCMPHGT";
- case HexagonISD::VCMPHGTU: return "HexagonISD::VCMPHGTU";
- case HexagonISD::VCMPWEQ: return "HexagonISD::VCMPWEQ";
- case HexagonISD::VCMPWGT: return "HexagonISD::VCMPWGT";
- case HexagonISD::VCMPWGTU: return "HexagonISD::VCMPWGTU";
- case HexagonISD::INSERT_ri: return "HexagonISD::INSERT_ri";
- case HexagonISD::INSERT_rd: return "HexagonISD::INSERT_rd";
- case HexagonISD::INSERT_riv: return "HexagonISD::INSERT_riv";
- case HexagonISD::INSERT_rdv: return "HexagonISD::INSERT_rdv";
- case HexagonISD::EXTRACTU_ri: return "HexagonISD::EXTRACTU_ri";
- case HexagonISD::EXTRACTU_rd: return "HexagonISD::EXTRACTU_rd";
- case HexagonISD::EXTRACTU_riv: return "HexagonISD::EXTRACTU_riv";
- case HexagonISD::EXTRACTU_rdv: return "HexagonISD::EXTRACTU_rdv";
- case HexagonISD::FTOI: return "HexagonISD::FTOI";
- case HexagonISD::ITOF: return "HexagonISD::ITOF";
- case HexagonISD::CALLv3: return "HexagonISD::CALLv3";
- case HexagonISD::CALLv3nr: return "HexagonISD::CALLv3nr";
- case HexagonISD::CALLR: return "HexagonISD::CALLR";
- case HexagonISD::RET_FLAG: return "HexagonISD::RET_FLAG";
- case HexagonISD::BR_JT: return "HexagonISD::BR_JT";
- case HexagonISD::TC_RETURN: return "HexagonISD::TC_RETURN";
- case HexagonISD::EH_RETURN: return "HexagonISD::EH_RETURN";
- }
+
+const char* HexagonTargetLowering::getTargetNodeName(unsigned Opcode) const {
+ switch ((HexagonISD::NodeType)Opcode) {
+ case HexagonISD::ALLOCA: return "HexagonISD::ALLOCA";
+ case HexagonISD::ARGEXTEND: return "HexagonISD::ARGEXTEND";
+ case HexagonISD::AT_GOT: return "HexagonISD::AT_GOT";
+ case HexagonISD::AT_PCREL: return "HexagonISD::AT_PCREL";
+ case HexagonISD::BARRIER: return "HexagonISD::BARRIER";
+ case HexagonISD::BR_JT: return "HexagonISD::BR_JT";
+ case HexagonISD::CALLR: return "HexagonISD::CALLR";
+ case HexagonISD::CALLv3nr: return "HexagonISD::CALLv3nr";
+ case HexagonISD::CALLv3: return "HexagonISD::CALLv3";
+ case HexagonISD::COMBINE: return "HexagonISD::COMBINE";
+ case HexagonISD::CONST32_GP: return "HexagonISD::CONST32_GP";
+ case HexagonISD::CONST32: return "HexagonISD::CONST32";
+ case HexagonISD::CP: return "HexagonISD::CP";
+ case HexagonISD::DCFETCH: return "HexagonISD::DCFETCH";
+ case HexagonISD::EH_RETURN: return "HexagonISD::EH_RETURN";
+ case HexagonISD::EXTRACTU: return "HexagonISD::EXTRACTU";
+ case HexagonISD::EXTRACTURP: return "HexagonISD::EXTRACTURP";
+ case HexagonISD::FCONST32: return "HexagonISD::FCONST32";
+ case HexagonISD::INSERT: return "HexagonISD::INSERT";
+ case HexagonISD::INSERTRP: return "HexagonISD::INSERTRP";
+ case HexagonISD::JT: return "HexagonISD::JT";
+ case HexagonISD::PACKHL: return "HexagonISD::PACKHL";
+ case HexagonISD::PIC_ADD: return "HexagonISD::PIC_ADD";
+ case HexagonISD::POPCOUNT: return "HexagonISD::POPCOUNT";
+ case HexagonISD::RET_FLAG: return "HexagonISD::RET_FLAG";
+ case HexagonISD::SHUFFEB: return "HexagonISD::SHUFFEB";
+ case HexagonISD::SHUFFEH: return "HexagonISD::SHUFFEH";
+ case HexagonISD::SHUFFOB: return "HexagonISD::SHUFFOB";
+ case HexagonISD::SHUFFOH: return "HexagonISD::SHUFFOH";
+ case HexagonISD::TC_RETURN: return "HexagonISD::TC_RETURN";
+ case HexagonISD::VCMPBEQ: return "HexagonISD::VCMPBEQ";
+ case HexagonISD::VCMPBGT: return "HexagonISD::VCMPBGT";
+ case HexagonISD::VCMPBGTU: return "HexagonISD::VCMPBGTU";
+ case HexagonISD::VCMPHEQ: return "HexagonISD::VCMPHEQ";
+ case HexagonISD::VCMPHGT: return "HexagonISD::VCMPHGT";
+ case HexagonISD::VCMPHGTU: return "HexagonISD::VCMPHGTU";
+ case HexagonISD::VCMPWEQ: return "HexagonISD::VCMPWEQ";
+ case HexagonISD::VCMPWGT: return "HexagonISD::VCMPWGT";
+ case HexagonISD::VCMPWGTU: return "HexagonISD::VCMPWGTU";
+ case HexagonISD::VSHLH: return "HexagonISD::VSHLH";
+ case HexagonISD::VSHLW: return "HexagonISD::VSHLW";
+ case HexagonISD::VSPLATB: return "HexagonISD::VSPLTB";
+ case HexagonISD::VSPLATH: return "HexagonISD::VSPLATH";
+ case HexagonISD::VSRAH: return "HexagonISD::VSRAH";
+ case HexagonISD::VSRAW: return "HexagonISD::VSRAW";
+ case HexagonISD::VSRLH: return "HexagonISD::VSRLH";
+ case HexagonISD::VSRLW: return "HexagonISD::VSRLW";
+ case HexagonISD::VSXTBH: return "HexagonISD::VSXTBH";
+ case HexagonISD::VSXTBW: return "HexagonISD::VSXTBW";
+ case HexagonISD::OP_END: break;
+ }
+ return nullptr;
}
-bool
-HexagonTargetLowering::isTruncateFree(Type *Ty1, Type *Ty2) const {
+bool HexagonTargetLowering::isTruncateFree(Type *Ty1, Type *Ty2) const {
EVT MTy1 = EVT::getEVT(Ty1);
EVT MTy2 = EVT::getEVT(Ty2);
- if (!MTy1.isSimple() || !MTy2.isSimple()) {
+ if (!MTy1.isSimple() || !MTy2.isSimple())
return false;
- }
- return ((MTy1.getSimpleVT() == MVT::i64) && (MTy2.getSimpleVT() == MVT::i32));
+ return (MTy1.getSimpleVT() == MVT::i64) && (MTy2.getSimpleVT() == MVT::i32);
}
bool HexagonTargetLowering::isTruncateFree(EVT VT1, EVT VT2) const {
- if (!VT1.isSimple() || !VT2.isSimple()) {
+ if (!VT1.isSimple() || !VT2.isSimple())
return false;
- }
- return ((VT1.getSimpleVT() == MVT::i64) && (VT2.getSimpleVT() == MVT::i32));
+ return (VT1.getSimpleVT() == MVT::i64) && (VT2.getSimpleVT() == MVT::i32);
}
// shouldExpandBuildVectorWithShuffles
if (IsScalarToVector)
return createSplat(DAG, dl, VT, V1.getOperand(0));
}
- return createSplat(DAG, dl, VT, DAG.getConstant(Lane, MVT::i32));
+ return createSplat(DAG, dl, VT, DAG.getConstant(Lane, dl, MVT::i32));
}
// FIXME: We need to support more general vector shuffles. See
unsigned SplatBits = APSplatBits.getZExtValue();
int32_t SextVal = ((int32_t) (SplatBits << (32 - SplatBitSize)) >>
(32 - SplatBitSize));
- return createSplat(DAG, dl, VT, DAG.getConstant(SextVal, MVT::i32));
+ return createSplat(DAG, dl, VT, DAG.getConstant(SextVal, dl, MVT::i32));
}
// Try to generate COMBINE to build v2i32 vectors.
SDValue V1 = BVN->getOperand(1);
if (V0.getOpcode() == ISD::UNDEF)
- V0 = DAG.getConstant(0, MVT::i32);
+ V0 = DAG.getConstant(0, dl, MVT::i32);
if (V1.getOpcode() == ISD::UNDEF)
- V1 = DAG.getConstant(0, MVT::i32);
+ V1 = DAG.getConstant(0, dl, MVT::i32);
ConstantSDNode *C0 = dyn_cast<ConstantSDNode>(V0);
ConstantSDNode *C1 = dyn_cast<ConstantSDNode>(V1);
}
if (Size == 64)
- ConstVal = DAG.getConstant(Res, MVT::i64);
+ ConstVal = DAG.getConstant(Res, dl, MVT::i64);
else
- ConstVal = DAG.getConstant(Res, MVT::i32);
+ ConstVal = DAG.getConstant(Res, dl, MVT::i32);
// When there are non constant operands, add them with INSERT_VECTOR_ELT to
// ConstVal, the constant part of the vector.
if (HasNonConstantElements) {
EVT EltVT = VT.getVectorElementType();
- SDValue Width = DAG.getConstant(EltVT.getSizeInBits(), MVT::i64);
+ SDValue Width = DAG.getConstant(EltVT.getSizeInBits(), dl, MVT::i64);
SDValue Shifted = DAG.getNode(ISD::SHL, dl, MVT::i64, Width,
- DAG.getConstant(32, MVT::i64));
+ DAG.getConstant(32, dl, MVT::i64));
for (unsigned i = 0, e = NElts; i != e; ++i) {
// LLVM's BUILD_VECTOR operands are in Little Endian mode, whereas Hexagon
if (VT.getSizeInBits() == 64 &&
Operand.getValueType().getSizeInBits() == 32) {
- SDValue C = DAG.getConstant(0, MVT::i32);
+ SDValue C = DAG.getConstant(0, dl, MVT::i32);
Operand = DAG.getNode(HexagonISD::COMBINE, dl, VT, C, Operand);
}
- SDValue Idx = DAG.getConstant(OpIdx, MVT::i64);
+ SDValue Idx = DAG.getConstant(OpIdx, dl, MVT::i64);
SDValue Offset = DAG.getNode(ISD::MUL, dl, MVT::i64, Idx, Width);
SDValue Combined = DAG.getNode(ISD::OR, dl, MVT::i64, Shifted, Offset);
const SDValue Ops[] = {ConstVal, Operand, Combined};
if (VT.getSizeInBits() == 32)
- ConstVal = DAG.getNode(HexagonISD::INSERT_riv, dl, MVT::i32, Ops);
+ ConstVal = DAG.getNode(HexagonISD::INSERTRP, dl, MVT::i32, Ops);
else
- ConstVal = DAG.getNode(HexagonISD::INSERT_rdv, dl, MVT::i64, Ops);
+ ConstVal = DAG.getNode(HexagonISD::INSERTRP, dl, MVT::i64, Ops);
}
}
unsigned NElts = Op.getNumOperands();
SDValue Vec = Op.getOperand(0);
EVT VecVT = Vec.getValueType();
- SDValue Width = DAG.getConstant(VecVT.getSizeInBits(), MVT::i64);
+ SDValue Width = DAG.getConstant(VecVT.getSizeInBits(), dl, MVT::i64);
SDValue Shifted = DAG.getNode(ISD::SHL, dl, MVT::i64, Width,
- DAG.getConstant(32, MVT::i64));
- SDValue ConstVal = DAG.getConstant(0, MVT::i64);
+ DAG.getConstant(32, dl, MVT::i64));
+ SDValue ConstVal = DAG.getConstant(0, dl, MVT::i64);
ConstantSDNode *W = dyn_cast<ConstantSDNode>(Width);
ConstantSDNode *S = dyn_cast<ConstantSDNode>(Shifted);
if (VT.getSizeInBits() == 64 &&
Operand.getValueType().getSizeInBits() == 32) {
- SDValue C = DAG.getConstant(0, MVT::i32);
+ SDValue C = DAG.getConstant(0, dl, MVT::i32);
Operand = DAG.getNode(HexagonISD::COMBINE, dl, VT, C, Operand);
}
- SDValue Idx = DAG.getConstant(OpIdx, MVT::i64);
+ SDValue Idx = DAG.getConstant(OpIdx, dl, MVT::i64);
SDValue Offset = DAG.getNode(ISD::MUL, dl, MVT::i64, Idx, Width);
SDValue Combined = DAG.getNode(ISD::OR, dl, MVT::i64, Shifted, Offset);
const SDValue Ops[] = {ConstVal, Operand, Combined};
if (VT.getSizeInBits() == 32)
- ConstVal = DAG.getNode(HexagonISD::INSERT_riv, dl, MVT::i32, Ops);
+ ConstVal = DAG.getNode(HexagonISD::INSERTRP, dl, MVT::i32, Ops);
else
- ConstVal = DAG.getNode(HexagonISD::INSERT_rdv, dl, MVT::i64, Ops);
+ ConstVal = DAG.getNode(HexagonISD::INSERTRP, dl, MVT::i64, Ops);
}
return DAG.getNode(ISD::BITCAST, dl, VT, ConstVal);
EVT EltVT = VecVT.getVectorElementType();
int EltSize = EltVT.getSizeInBits();
SDValue Width = DAG.getConstant(Op.getOpcode() == ISD::EXTRACT_VECTOR_ELT ?
- EltSize : VTN * EltSize, MVT::i64);
+ EltSize : VTN * EltSize, dl, MVT::i64);
// Constant element number.
- if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Idx)) {
- SDValue Offset = DAG.getConstant(C->getZExtValue() * EltSize, MVT::i32);
+ if (ConstantSDNode *CI = dyn_cast<ConstantSDNode>(Idx)) {
+ uint64_t X = CI->getZExtValue();
+ SDValue Offset = DAG.getConstant(X * EltSize, dl, MVT::i32);
const SDValue Ops[] = {Vec, Width, Offset};
- ConstantSDNode *W = dyn_cast<ConstantSDNode>(Width);
- assert(W && "Non constant width in LowerEXTRACT_VECTOR");
+ ConstantSDNode *CW = dyn_cast<ConstantSDNode>(Width);
+ assert(CW && "Non constant width in LowerEXTRACT_VECTOR");
SDValue N;
- // For certain extracts, it is a simple _hi/_lo subreg.
- if (VecVT.getSimpleVT() == MVT::v2i32) {
- // v2i32 -> i32 vselect.
- if (C->getZExtValue() == 0)
- N = DAG.getTargetExtractSubreg(Hexagon::subreg_loreg, dl,
- MVT::i32, Vec);
- else if (C->getZExtValue() == 1)
- N = DAG.getTargetExtractSubreg(Hexagon::subreg_hireg, dl,
- MVT::i32, Vec);
- else
- llvm_unreachable("Bad offset");
- } else if ((VecVT.getSimpleVT() == MVT::v4i16) &&
- (W->getZExtValue() == 32)) {
- // v4i16 -> v2i16/i32 vselect.
- if (C->getZExtValue() == 0)
- N = DAG.getTargetExtractSubreg(Hexagon::subreg_loreg, dl,
- MVT::i32, Vec);
- else if (C->getZExtValue() == 2)
- N = DAG.getTargetExtractSubreg(Hexagon::subreg_hireg, dl,
- MVT::i32, Vec);
- else
- llvm_unreachable("Bad offset");
- } else if ((VecVT.getSimpleVT() == MVT::v8i8) &&
- (W->getZExtValue() == 32)) {
- // v8i8 -> v4i8/i32 vselect.
- if (C->getZExtValue() == 0)
- N = DAG.getTargetExtractSubreg(Hexagon::subreg_loreg, dl,
- MVT::i32, Vec);
- else if (C->getZExtValue() == 4)
- N = DAG.getTargetExtractSubreg(Hexagon::subreg_hireg, dl,
- MVT::i32, Vec);
+ MVT SVT = VecVT.getSimpleVT();
+ uint64_t W = CW->getZExtValue();
+
+ if (W == 32) {
+ // Translate this node into EXTRACT_SUBREG.
+ unsigned Subreg = (X == 0) ? Hexagon::subreg_loreg : 0;
+
+ if (X == 0)
+ Subreg = Hexagon::subreg_loreg;
+ else if (SVT == MVT::v2i32 && X == 1)
+ Subreg = Hexagon::subreg_hireg;
+ else if (SVT == MVT::v4i16 && X == 2)
+ Subreg = Hexagon::subreg_hireg;
+ else if (SVT == MVT::v8i8 && X == 4)
+ Subreg = Hexagon::subreg_hireg;
else
llvm_unreachable("Bad offset");
+ N = DAG.getTargetExtractSubreg(Subreg, dl, MVT::i32, Vec);
+
} else if (VecVT.getSizeInBits() == 32) {
- N = DAG.getNode(HexagonISD::EXTRACTU_ri, dl, MVT::i32, Ops);
+ N = DAG.getNode(HexagonISD::EXTRACTU, dl, MVT::i32, Ops);
} else {
- N = DAG.getNode(HexagonISD::EXTRACTU_rd, dl, MVT::i64, Ops);
+ N = DAG.getNode(HexagonISD::EXTRACTU, dl, MVT::i64, Ops);
if (VT.getSizeInBits() == 32)
N = DAG.getTargetExtractSubreg(Hexagon::subreg_loreg, dl, MVT::i32, N);
}
// Variable element number.
SDValue Offset = DAG.getNode(ISD::MUL, dl, MVT::i32, Idx,
- DAG.getConstant(EltSize, MVT::i32));
+ DAG.getConstant(EltSize, dl, MVT::i32));
SDValue Shifted = DAG.getNode(ISD::SHL, dl, MVT::i64, Width,
- DAG.getConstant(32, MVT::i64));
+ DAG.getConstant(32, dl, MVT::i64));
SDValue Combined = DAG.getNode(ISD::OR, dl, MVT::i64, Shifted, Offset);
const SDValue Ops[] = {Vec, Combined};
SDValue N;
if (VecVT.getSizeInBits() == 32) {
- N = DAG.getNode(HexagonISD::EXTRACTU_riv, dl, MVT::i32, Ops);
+ N = DAG.getNode(HexagonISD::EXTRACTURP, dl, MVT::i32, Ops);
} else {
- N = DAG.getNode(HexagonISD::EXTRACTU_rdv, dl, MVT::i64, Ops);
+ N = DAG.getNode(HexagonISD::EXTRACTURP, dl, MVT::i64, Ops);
if (VT.getSizeInBits() == 32)
N = DAG.getTargetExtractSubreg(Hexagon::subreg_loreg, dl, MVT::i32, N);
}
EVT EltVT = VecVT.getVectorElementType();
int EltSize = EltVT.getSizeInBits();
SDValue Width = DAG.getConstant(Op.getOpcode() == ISD::INSERT_VECTOR_ELT ?
- EltSize : VTN * EltSize, MVT::i64);
+ EltSize : VTN * EltSize, dl, MVT::i64);
if (ConstantSDNode *C = cast<ConstantSDNode>(Idx)) {
- SDValue Offset = DAG.getConstant(C->getSExtValue() * EltSize, MVT::i32);
+ SDValue Offset = DAG.getConstant(C->getSExtValue() * EltSize, dl, MVT::i32);
const SDValue Ops[] = {Vec, Val, Width, Offset};
SDValue N;
if (VT.getSizeInBits() == 32)
- N = DAG.getNode(HexagonISD::INSERT_ri, dl, MVT::i32, Ops);
+ N = DAG.getNode(HexagonISD::INSERT, dl, MVT::i32, Ops);
else
- N = DAG.getNode(HexagonISD::INSERT_rd, dl, MVT::i64, Ops);
+ N = DAG.getNode(HexagonISD::INSERT, dl, MVT::i64, Ops);
return DAG.getNode(ISD::BITCAST, dl, VT, N);
}
// Variable element number.
SDValue Offset = DAG.getNode(ISD::MUL, dl, MVT::i32, Idx,
- DAG.getConstant(EltSize, MVT::i32));
+ DAG.getConstant(EltSize, dl, MVT::i32));
SDValue Shifted = DAG.getNode(ISD::SHL, dl, MVT::i64, Width,
- DAG.getConstant(32, MVT::i64));
+ DAG.getConstant(32, dl, MVT::i64));
SDValue Combined = DAG.getNode(ISD::OR, dl, MVT::i64, Shifted, Offset);
if (VT.getSizeInBits() == 64 &&
Val.getValueType().getSizeInBits() == 32) {
- SDValue C = DAG.getConstant(0, MVT::i32);
+ SDValue C = DAG.getConstant(0, dl, MVT::i32);
Val = DAG.getNode(HexagonISD::COMBINE, dl, VT, C, Val);
}
SDValue N;
if (VT.getSizeInBits() == 32)
- N = DAG.getNode(HexagonISD::INSERT_riv, dl, MVT::i32, Ops);
+ N = DAG.getNode(HexagonISD::INSERTRP, dl, MVT::i32, Ops);
else
- N = DAG.getNode(HexagonISD::INSERT_rdv, dl, MVT::i64, Ops);
+ N = DAG.getNode(HexagonISD::INSERTRP, dl, MVT::i64, Ops);
return DAG.getNode(ISD::BITCAST, dl, VT, N);
}
SDValue StoreAddr = DAG.getNode(ISD::ADD, dl, getPointerTy(),
DAG.getRegister(Hexagon::R30, getPointerTy()),
- DAG.getIntPtrConstant(4));
+ DAG.getIntPtrConstant(4, dl));
Chain = DAG.getStore(Chain, dl, Handler, StoreAddr, MachinePointerInfo(),
false, false, 0);
Chain = DAG.getCopyToReg(Chain, dl, OffsetReg, Offset);
SDValue
HexagonTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
- switch (Op.getOpcode()) {
- default: llvm_unreachable("Should not custom lower this!");
- case ISD::CONCAT_VECTORS: return LowerCONCAT_VECTORS(Op, DAG);
- case ISD::INSERT_SUBVECTOR: return LowerINSERT_VECTOR(Op, DAG);
- case ISD::INSERT_VECTOR_ELT: return LowerINSERT_VECTOR(Op, DAG);
- case ISD::EXTRACT_SUBVECTOR: return LowerEXTRACT_VECTOR(Op, DAG);
- case ISD::EXTRACT_VECTOR_ELT: return LowerEXTRACT_VECTOR(Op, DAG);
- case ISD::BUILD_VECTOR: return LowerBUILD_VECTOR(Op, DAG);
- case ISD::VECTOR_SHUFFLE: return LowerVECTOR_SHUFFLE(Op, DAG);
+ unsigned Opc = Op.getOpcode();
+ switch (Opc) {
+ default:
+#ifndef NDEBUG
+ Op.getNode()->dumpr(&DAG);
+ if (Opc > HexagonISD::OP_BEGIN && Opc < HexagonISD::OP_END)
+ errs() << "Check for a non-legal type in this operation\n";
+#endif
+ llvm_unreachable("Should not custom lower this!");
+ case ISD::CONCAT_VECTORS: return LowerCONCAT_VECTORS(Op, DAG);
+ case ISD::INSERT_SUBVECTOR: return LowerINSERT_VECTOR(Op, DAG);
+ case ISD::INSERT_VECTOR_ELT: return LowerINSERT_VECTOR(Op, DAG);
+ case ISD::EXTRACT_SUBVECTOR: return LowerEXTRACT_VECTOR(Op, DAG);
+ case ISD::EXTRACT_VECTOR_ELT: return LowerEXTRACT_VECTOR(Op, DAG);
+ case ISD::BUILD_VECTOR: return LowerBUILD_VECTOR(Op, DAG);
+ case ISD::VECTOR_SHUFFLE: return LowerVECTOR_SHUFFLE(Op, DAG);
case ISD::SRA:
case ISD::SHL:
- case ISD::SRL:
- return LowerVECTOR_SHIFT(Op, DAG);
- case ISD::ConstantPool:
- return LowerConstantPool(Op, DAG);
- case ISD::EH_RETURN: return LowerEH_RETURN(Op, DAG);
- // Frame & Return address. Currently unimplemented.
- case ISD::RETURNADDR: return LowerRETURNADDR(Op, DAG);
- case ISD::FRAMEADDR: return LowerFRAMEADDR(Op, DAG);
- case ISD::GlobalTLSAddress:
- llvm_unreachable("TLS not implemented for Hexagon.");
- case ISD::ATOMIC_FENCE: return LowerATOMIC_FENCE(Op, DAG);
- case ISD::GlobalAddress: return LowerGLOBALADDRESS(Op, DAG);
- case ISD::BlockAddress: return LowerBlockAddress(Op, DAG);
- case ISD::VASTART: return LowerVASTART(Op, DAG);
- case ISD::BR_JT: return LowerBR_JT(Op, DAG);
+ case ISD::SRL: return LowerVECTOR_SHIFT(Op, DAG);
+ case ISD::ConstantPool: return LowerConstantPool(Op, DAG);
+ case ISD::EH_RETURN: return LowerEH_RETURN(Op, DAG);
+ // Frame & Return address. Currently unimplemented.
+ case ISD::RETURNADDR: return LowerRETURNADDR(Op, DAG);
+ case ISD::FRAMEADDR: return LowerFRAMEADDR(Op, DAG);
+ case ISD::ATOMIC_FENCE: return LowerATOMIC_FENCE(Op, DAG);
+ case ISD::GlobalAddress: return LowerGLOBALADDRESS(Op, DAG);
+ case ISD::BlockAddress: return LowerBlockAddress(Op, DAG);
+ case ISD::VASTART: return LowerVASTART(Op, DAG);
+ case ISD::BR_JT: return LowerBR_JT(Op, DAG);
// Custom lower some vector loads.
- case ISD::LOAD: return LowerLOAD(Op, DAG);
-
- case ISD::DYNAMIC_STACKALLOC: return LowerDYNAMIC_STACKALLOC(Op, DAG);
- case ISD::SELECT: return Op;
- case ISD::SETCC: return LowerSETCC(Op, DAG);
- case ISD::VSELECT: return LowerVSELECT(Op, DAG);
- case ISD::CTPOP: return LowerCTPOP(Op, DAG);
- case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG);
- case ISD::INLINEASM: return LowerINLINEASM(Op, DAG);
-
+ case ISD::LOAD: return LowerLOAD(Op, DAG);
+ case ISD::DYNAMIC_STACKALLOC: return LowerDYNAMIC_STACKALLOC(Op, DAG);
+ case ISD::SETCC: return LowerSETCC(Op, DAG);
+ case ISD::VSELECT: return LowerVSELECT(Op, DAG);
+ case ISD::CTPOP: return LowerCTPOP(Op, DAG);
+ case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG);
+ case ISD::INLINEASM: return LowerINLINEASM(Op, DAG);
}
}
-
-
-//===----------------------------------------------------------------------===//
-// Hexagon Scheduler Hooks
-//===----------------------------------------------------------------------===//
MachineBasicBlock *
HexagonTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
MachineBasicBlock *BB)
-const {
+ const {
switch (MI->getOpcode()) {
- case Hexagon::ADJDYNALLOC: {
+ case Hexagon::ALLOCA: {
MachineFunction *MF = BB->getParent();
- HexagonMachineFunctionInfo *FuncInfo =
- MF->getInfo<HexagonMachineFunctionInfo>();
+ auto *FuncInfo = MF->getInfo<HexagonMachineFunctionInfo>();
FuncInfo->addAllocaAdjustInst(MI);
return BB;
}
/// specified FP immediate natively. If false, the legalizer will
/// materialize the FP immediate as a load from a constant pool.
bool HexagonTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const {
- return Subtarget->hasV5TOps();
+ return Subtarget.hasV5TOps();
}
/// isLegalAddressingMode - Return true if the addressing mode represented by
/// AM is legal for this target, for a load/store of the specified type.
bool HexagonTargetLowering::isLegalAddressingMode(const AddrMode &AM,
- Type *Ty) const {
+ Type *Ty,
+ unsigned AS) const {
// Allows a signed-extended 11-bit immediate field.
- if (AM.BaseOffs <= -(1LL << 13) || AM.BaseOffs >= (1LL << 13)-1) {
+ if (AM.BaseOffs <= -(1LL << 13) || AM.BaseOffs >= (1LL << 13)-1)
return false;
- }
// No global is ever allowed as a base.
- if (AM.BaseGV) {
+ if (AM.BaseGV)
return false;
- }
int Scale = AM.Scale;
if (Scale < 0) Scale = -Scale;