#include "llvm/Target/TargetOptions.h"
using namespace llvm;
-static bool CC_PPC32_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
- CCValAssign::LocInfo &LocInfo,
- ISD::ArgFlagsTy &ArgFlags,
- CCState &State);
-static bool CC_PPC32_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
- MVT &LocVT,
- CCValAssign::LocInfo &LocInfo,
- ISD::ArgFlagsTy &ArgFlags,
- CCState &State);
-static bool CC_PPC32_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
- MVT &LocVT,
- CCValAssign::LocInfo &LocInfo,
- ISD::ArgFlagsTy &ArgFlags,
- CCState &State);
-
static cl::opt<bool> DisablePPCPreinc("disable-ppc-preinc",
cl::desc("disable preincrement load/store generation on PPC"), cl::Hidden);
PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
: TargetLowering(TM, CreateTLOF(TM)), PPCSubTarget(*TM.getSubtargetImpl()) {
const PPCSubtarget *Subtarget = &TM.getSubtarget<PPCSubtarget>();
- PPCRegInfo = TM.getRegisterInfo();
- PPCII = TM.getInstrInfo();
setPow2DivIsCheap();
setTargetDAGCombine(ISD::STORE);
setTargetDAGCombine(ISD::BR_CC);
setTargetDAGCombine(ISD::BSWAP);
+ setTargetDAGCombine(ISD::INTRINSIC_WO_CHAIN);
// Use reciprocal estimates.
if (TM.Options.UnsafeFPMath) {
case PPCISD::RET_FLAG: return "PPCISD::RET_FLAG";
case PPCISD::EH_SJLJ_SETJMP: return "PPCISD::EH_SJLJ_SETJMP";
case PPCISD::EH_SJLJ_LONGJMP: return "PPCISD::EH_SJLJ_LONGJMP";
- case PPCISD::MFCR: return "PPCISD::MFCR";
+ case PPCISD::MFOCRF: return "PPCISD::MFOCRF";
case PPCISD::VCMP: return "PPCISD::VCMP";
case PPCISD::VCMPo: return "PPCISD::VCMPo";
case PPCISD::LBRX: return "PPCISD::LBRX";
SelectionDAG &DAG,
bool Aligned) const {
// FIXME dl should come from parent load or store, not from address
- DebugLoc dl = N.getDebugLoc();
+ SDLoc dl(N);
// If this can be more profitably realized as r+r, fail.
if (SelectAddressRegReg(N, Disp, Base, DAG))
return false;
return true;
}
-
/// getPreIndexedAddressParts - returns true by value, base pointer and
/// offset pointer and addressing mode by reference if the node's address
/// can be legally represented as pre-indexed load / store address.
/// PICBase, set the HiOpFlags and LoOpFlags to the target MO flags.
static bool GetLabelAccessInfo(const TargetMachine &TM, unsigned &HiOpFlags,
unsigned &LoOpFlags, const GlobalValue *GV = 0) {
- HiOpFlags = PPCII::MO_HA16;
- LoOpFlags = PPCII::MO_LO16;
+ HiOpFlags = PPCII::MO_HA;
+ LoOpFlags = PPCII::MO_LO;
// Don't use the pic base if not in PIC relocation model. Or if we are on a
// non-darwin platform. We don't support PIC on other platforms yet.
SelectionDAG &DAG) {
EVT PtrVT = HiPart.getValueType();
SDValue Zero = DAG.getConstant(0, PtrVT);
- DebugLoc DL = HiPart.getDebugLoc();
+ SDLoc DL(HiPart);
SDValue Hi = DAG.getNode(PPCISD::Hi, DL, PtrVT, HiPart, Zero);
SDValue Lo = DAG.getNode(PPCISD::Lo, DL, PtrVT, LoPart, Zero);
// The actual address of the GlobalValue is stored in the TOC.
if (PPCSubTarget.isSVR4ABI() && PPCSubTarget.isPPC64()) {
SDValue GA = DAG.getTargetConstantPool(C, PtrVT, CP->getAlignment(), 0);
- return DAG.getNode(PPCISD::TOC_ENTRY, CP->getDebugLoc(), MVT::i64, GA,
+ return DAG.getNode(PPCISD::TOC_ENTRY, SDLoc(CP), MVT::i64, GA,
DAG.getRegister(PPC::X2, MVT::i64));
}
// The actual address of the GlobalValue is stored in the TOC.
if (PPCSubTarget.isSVR4ABI() && PPCSubTarget.isPPC64()) {
SDValue GA = DAG.getTargetJumpTable(JT->getIndex(), PtrVT);
- return DAG.getNode(PPCISD::TOC_ENTRY, JT->getDebugLoc(), MVT::i64, GA,
+ return DAG.getNode(PPCISD::TOC_ENTRY, SDLoc(JT), MVT::i64, GA,
DAG.getRegister(PPC::X2, MVT::i64));
}
SelectionDAG &DAG) const {
GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(Op);
- DebugLoc dl = GA->getDebugLoc();
+ SDLoc dl(GA);
const GlobalValue *GV = GA->getGlobal();
EVT PtrVT = getPointerTy();
bool is64bit = PPCSubTarget.isPPC64();
if (Model == TLSModel::LocalExec) {
SDValue TGAHi = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
- PPCII::MO_TPREL16_HA);
+ PPCII::MO_TPREL_HA);
SDValue TGALo = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
- PPCII::MO_TPREL16_LO);
+ PPCII::MO_TPREL_LO);
SDValue TLSReg = DAG.getRegister(is64bit ? PPC::X13 : PPC::R2,
is64bit ? MVT::i64 : MVT::i32);
SDValue Hi = DAG.getNode(PPCISD::Hi, dl, PtrVT, TGAHi, TLSReg);
SelectionDAG &DAG) const {
EVT PtrVT = Op.getValueType();
GlobalAddressSDNode *GSDN = cast<GlobalAddressSDNode>(Op);
- DebugLoc DL = GSDN->getDebugLoc();
+ SDLoc DL(GSDN);
const GlobalValue *GV = GSDN->getGlobal();
// 64-bit SVR4 ABI code is always position-independent.
SDValue PPCTargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const {
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
// If we're comparing for equality to zero, expose the fact that this is
// implented as a ctlz/srl pair on ppc, so that the dag combiner can
SDValue InChain = Node->getOperand(0);
SDValue VAListPtr = Node->getOperand(1);
const Value *SV = cast<SrcValueSDNode>(Node->getOperand(2))->getValue();
- DebugLoc dl = Node->getDebugLoc();
+ SDLoc dl(Node);
assert(!Subtarget.isPPC64() && "LowerVAARG is PPC32 only");
SDValue Trmp = Op.getOperand(1); // trampoline
SDValue FPtr = Op.getOperand(2); // nested function
SDValue Nest = Op.getOperand(3); // 'nest' parameter value
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = (PtrVT == MVT::i64);
MachineFunction &MF = DAG.getMachineFunction();
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
if (Subtarget.isDarwinABI() || Subtarget.isPPC64()) {
// vastart just stores the address of the VarArgsFrameIndex slot into the
#include "PPCGenCallingConv.inc"
-static bool CC_PPC32_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
- CCValAssign::LocInfo &LocInfo,
- ISD::ArgFlagsTy &ArgFlags,
- CCState &State) {
+bool llvm::CC_PPC32_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+ CCValAssign::LocInfo &LocInfo,
+ ISD::ArgFlagsTy &ArgFlags,
+ CCState &State) {
return true;
}
-static bool CC_PPC32_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
- MVT &LocVT,
- CCValAssign::LocInfo &LocInfo,
- ISD::ArgFlagsTy &ArgFlags,
- CCState &State) {
+bool llvm::CC_PPC32_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
+ MVT &LocVT,
+ CCValAssign::LocInfo &LocInfo,
+ ISD::ArgFlagsTy &ArgFlags,
+ CCState &State) {
static const uint16_t ArgRegs[] = {
PPC::R3, PPC::R4, PPC::R5, PPC::R6,
PPC::R7, PPC::R8, PPC::R9, PPC::R10,
return false;
}
-static bool CC_PPC32_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
- MVT &LocVT,
- CCValAssign::LocInfo &LocInfo,
- ISD::ArgFlagsTy &ArgFlags,
- CCState &State) {
+bool llvm::CC_PPC32_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
+ MVT &LocVT,
+ CCValAssign::LocInfo &LocInfo,
+ ISD::ArgFlagsTy &ArgFlags,
+ CCState &State) {
static const uint16_t ArgRegs[] = {
PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5, PPC::F6, PPC::F7,
PPC::F8
CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg>
&Ins,
- DebugLoc dl, SelectionDAG &DAG,
+ SDLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals)
const {
if (PPCSubTarget.isSVR4ABI()) {
CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg>
&Ins,
- DebugLoc dl, SelectionDAG &DAG,
+ SDLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
// 32-bit SVR4 ABI Stack Frame Layout:
SDValue
PPCTargetLowering::extendArgForPPC64(ISD::ArgFlagsTy Flags, EVT ObjectVT,
SelectionDAG &DAG, SDValue ArgVal,
- DebugLoc dl) const {
+ SDLoc dl) const {
if (Flags.isSExt())
ArgVal = DAG.getNode(ISD::AssertSext, dl, MVT::i64, ArgVal,
DAG.getValueType(ObjectVT));
CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg>
&Ins,
- DebugLoc dl, SelectionDAG &DAG,
+ SDLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
// TODO: add description of PPC stack frame format, or at least some docs.
//
CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg>
&Ins,
- DebugLoc dl, SelectionDAG &DAG,
+ SDLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
// TODO: add description of PPC stack frame format, or at least some docs.
//
SDValue Chain,
const SmallVector<TailCallArgumentInfo, 8> &TailCallArgs,
SmallVector<SDValue, 8> &MemOpChains,
- DebugLoc dl) {
+ SDLoc dl) {
for (unsigned i = 0, e = TailCallArgs.size(); i != e; ++i) {
SDValue Arg = TailCallArgs[i].Arg;
SDValue FIN = TailCallArgs[i].FrameIdxOp;
int SPDiff,
bool isPPC64,
bool isDarwinABI,
- DebugLoc dl) {
+ SDLoc dl) {
if (SPDiff) {
// Calculate the new stack slot for the return address.
int SlotSize = isPPC64 ? 8 : 4;
SDValue &LROpOut,
SDValue &FPOpOut,
bool isDarwinABI,
- DebugLoc dl) const {
+ SDLoc dl) const {
if (SPDiff) {
// Load the LR and FP stack slot for later adjusting.
EVT VT = PPCSubTarget.isPPC64() ? MVT::i64 : MVT::i32;
static SDValue
CreateCopyOfByValArgument(SDValue Src, SDValue Dst, SDValue Chain,
ISD::ArgFlagsTy Flags, SelectionDAG &DAG,
- DebugLoc dl) {
+ SDLoc dl) {
SDValue SizeNode = DAG.getConstant(Flags.getByValSize(), MVT::i32);
return DAG.getMemcpy(Chain, dl, Dst, Src, SizeNode, Flags.getByValAlign(),
false, false, MachinePointerInfo(0),
unsigned ArgOffset, bool isPPC64, bool isTailCall,
bool isVector, SmallVector<SDValue, 8> &MemOpChains,
SmallVector<TailCallArgumentInfo, 8> &TailCallArguments,
- DebugLoc dl) {
+ SDLoc dl) {
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
if (!isTailCall) {
if (isVector) {
static
void PrepareTailCall(SelectionDAG &DAG, SDValue &InFlag, SDValue &Chain,
- DebugLoc dl, bool isPPC64, int SPDiff, unsigned NumBytes,
+ SDLoc dl, bool isPPC64, int SPDiff, unsigned NumBytes,
SDValue LROp, SDValue FPOp, bool isDarwinABI,
SmallVector<TailCallArgumentInfo, 8> &TailCallArguments) {
MachineFunction &MF = DAG.getMachineFunction();
// Emit callseq_end just before tailcall node.
Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
- DAG.getIntPtrConstant(0, true), InFlag);
+ DAG.getIntPtrConstant(0, true), InFlag, dl);
InFlag = Chain.getValue(1);
}
static
unsigned PrepareCall(SelectionDAG &DAG, SDValue &Callee, SDValue &InFlag,
- SDValue &Chain, DebugLoc dl, int SPDiff, bool isTailCall,
+ SDValue &Chain, SDLoc dl, int SPDiff, bool isTailCall,
SmallVector<std::pair<unsigned, SDValue>, 8> &RegsToPass,
SmallVector<SDValue, 8> &Ops, std::vector<EVT> &NodeTys,
const PPCSubtarget &PPCSubTarget) {
PPCTargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
- DebugLoc dl, SelectionDAG &DAG,
+ SDLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
SmallVector<CCValAssign, 16> RVLocs;
}
SDValue
-PPCTargetLowering::FinishCall(CallingConv::ID CallConv, DebugLoc dl,
+PPCTargetLowering::FinishCall(CallingConv::ID CallConv, SDLoc dl,
bool isTailCall, bool isVarArg,
SelectionDAG &DAG,
SmallVector<std::pair<unsigned, SDValue>, 8>
Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
DAG.getIntPtrConstant(BytesCalleePops, true),
- InFlag);
+ InFlag, dl);
if (!Ins.empty())
InFlag = Chain.getValue(1);
PPCTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
SmallVectorImpl<SDValue> &InVals) const {
SelectionDAG &DAG = CLI.DAG;
- DebugLoc &dl = CLI.DL;
+ SDLoc &dl = CLI.DL;
SmallVector<ISD::OutputArg, 32> &Outs = CLI.Outs;
SmallVector<SDValue, 32> &OutVals = CLI.OutVals;
SmallVector<ISD::InputArg, 32> &Ins = CLI.Ins;
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SmallVectorImpl<ISD::InputArg> &Ins,
- DebugLoc dl, SelectionDAG &DAG,
+ SDLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
// See PPCTargetLowering::LowerFormalArguments_32SVR4() for a description
// of the 32-bit SVR4 ABI stack frame layout.
// Adjust the stack pointer for the new arguments...
// These operations are automatically eliminated by the prolog/epilog pass
- Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumBytes, true));
+ Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumBytes, true),
+ dl);
SDValue CallSeqStart = Chain;
// Load the return address and frame pointer so it can be moved somewhere else
// This must go outside the CALLSEQ_START..END.
SDValue NewCallSeqStart = DAG.getCALLSEQ_START(MemcpyCall,
- CallSeqStart.getNode()->getOperand(1));
+ CallSeqStart.getNode()->getOperand(1),
+ SDLoc(MemcpyCall));
DAG.ReplaceAllUsesWith(CallSeqStart.getNode(),
NewCallSeqStart.getNode());
Chain = CallSeqStart = NewCallSeqStart;
SDValue CallSeqStart,
ISD::ArgFlagsTy Flags,
SelectionDAG &DAG,
- DebugLoc dl) const {
+ SDLoc dl) const {
SDValue MemcpyCall = CreateCopyOfByValArgument(Arg, PtrOff,
CallSeqStart.getNode()->getOperand(0),
Flags, DAG, dl);
// The MEMCPY must go outside the CALLSEQ_START..END.
SDValue NewCallSeqStart = DAG.getCALLSEQ_START(MemcpyCall,
- CallSeqStart.getNode()->getOperand(1));
+ CallSeqStart.getNode()->getOperand(1),
+ SDLoc(MemcpyCall));
DAG.ReplaceAllUsesWith(CallSeqStart.getNode(),
NewCallSeqStart.getNode());
return NewCallSeqStart;
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SmallVectorImpl<ISD::InputArg> &Ins,
- DebugLoc dl, SelectionDAG &DAG,
+ SDLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
unsigned NumOps = Outs.size();
// Adjust the stack pointer for the new arguments...
// These operations are automatically eliminated by the prolog/epilog pass
- Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumBytes, true));
+ Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumBytes, true),
+ dl);
SDValue CallSeqStart = Chain;
// Load the return address and frame pointer so it can be move somewhere else
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SmallVectorImpl<ISD::InputArg> &Ins,
- DebugLoc dl, SelectionDAG &DAG,
+ SDLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
unsigned NumOps = Outs.size();
// Adjust the stack pointer for the new arguments...
// These operations are automatically eliminated by the prolog/epilog pass
- Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumBytes, true));
+ Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumBytes, true),
+ dl);
SDValue CallSeqStart = Chain;
// Load the return address and frame pointer so it can be move somewhere else
CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
- DebugLoc dl, SelectionDAG &DAG) const {
+ SDLoc dl, SelectionDAG &DAG) const {
SmallVector<CCValAssign, 16> RVLocs;
CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
SDValue PPCTargetLowering::LowerSTACKRESTORE(SDValue Op, SelectionDAG &DAG,
const PPCSubtarget &Subtarget) const {
// When we pop the dynamic allocation we need to restore the SP link.
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
// Get the corect type for pointers.
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// Get the inputs.
SDValue Chain = Op.getOperand(0);
SDValue Size = Op.getOperand(1);
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
// Get the corect type for pointers.
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDValue PPCTargetLowering::lowerEH_SJLJ_SETJMP(SDValue Op,
SelectionDAG &DAG) const {
- DebugLoc DL = Op.getDebugLoc();
+ SDLoc DL(Op);
return DAG.getNode(PPCISD::EH_SJLJ_SETJMP, DL,
DAG.getVTList(MVT::i32, MVT::Other),
Op.getOperand(0), Op.getOperand(1));
SDValue PPCTargetLowering::lowerEH_SJLJ_LONGJMP(SDValue Op,
SelectionDAG &DAG) const {
- DebugLoc DL = Op.getDebugLoc();
+ SDLoc DL(Op);
return DAG.getNode(PPCISD::EH_SJLJ_LONGJMP, DL, MVT::Other,
Op.getOperand(0), Op.getOperand(1));
}
EVT CmpVT = Op.getOperand(0).getValueType();
SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
SDValue TV = Op.getOperand(2), FV = Op.getOperand(3);
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
// If the RHS of the comparison is a 0.0, we don't need to do the
// subtraction at all.
// FIXME: Split this code up when LegalizeDAGTypes lands.
SDValue PPCTargetLowering::LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG,
- DebugLoc dl) const {
+ SDLoc dl) const {
assert(Op.getOperand(0).getValueType().isFloatingPoint());
SDValue Src = Op.getOperand(0);
if (Src.getValueType() == MVT::f32)
SDValue PPCTargetLowering::LowerINT_TO_FP(SDValue Op,
SelectionDAG &DAG) const {
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
// Don't handle ppc_fp128 here; let it be lowered to a libcall.
if (Op.getValueType() != MVT::f32 && Op.getValueType() != MVT::f64)
return SDValue();
SDValue PPCTargetLowering::LowerFLT_ROUNDS_(SDValue Op,
SelectionDAG &DAG) const {
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
/*
The rounding mode is in bits 30:31 of FPSR, and has the following
settings:
SDValue PPCTargetLowering::LowerSHL_PARTS(SDValue Op, SelectionDAG &DAG) const {
EVT VT = Op.getValueType();
unsigned BitWidth = VT.getSizeInBits();
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
assert(Op.getNumOperands() == 3 &&
VT == Op.getOperand(1).getValueType() &&
"Unexpected SHL!");
SDValue PPCTargetLowering::LowerSRL_PARTS(SDValue Op, SelectionDAG &DAG) const {
EVT VT = Op.getValueType();
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
unsigned BitWidth = VT.getSizeInBits();
assert(Op.getNumOperands() == 3 &&
VT == Op.getOperand(1).getValueType() &&
}
SDValue PPCTargetLowering::LowerSRA_PARTS(SDValue Op, SelectionDAG &DAG) const {
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
EVT VT = Op.getValueType();
unsigned BitWidth = VT.getSizeInBits();
assert(Op.getNumOperands() == 3 &&
/// BuildSplatI - Build a canonical splati of Val with an element size of
/// SplatSize. Cast the result to VT.
static SDValue BuildSplatI(int Val, unsigned SplatSize, EVT VT,
- SelectionDAG &DAG, DebugLoc dl) {
+ SelectionDAG &DAG, SDLoc dl) {
assert(Val >= -16 && Val <= 15 && "vsplti is out of range!");
static const EVT VTys[] = { // canonical VT to use for each size.
/// BuildIntrinsicOp - Return a unary operator intrinsic node with the
/// specified intrinsic ID.
static SDValue BuildIntrinsicOp(unsigned IID, SDValue Op,
- SelectionDAG &DAG, DebugLoc dl,
+ SelectionDAG &DAG, SDLoc dl,
EVT DestVT = MVT::Other) {
if (DestVT == MVT::Other) DestVT = Op.getValueType();
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, DestVT,
/// BuildIntrinsicOp - Return a binary operator intrinsic node with the
/// specified intrinsic ID.
static SDValue BuildIntrinsicOp(unsigned IID, SDValue LHS, SDValue RHS,
- SelectionDAG &DAG, DebugLoc dl,
+ SelectionDAG &DAG, SDLoc dl,
EVT DestVT = MVT::Other) {
if (DestVT == MVT::Other) DestVT = LHS.getValueType();
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, DestVT,
/// specified intrinsic ID.
static SDValue BuildIntrinsicOp(unsigned IID, SDValue Op0, SDValue Op1,
SDValue Op2, SelectionDAG &DAG,
- DebugLoc dl, EVT DestVT = MVT::Other) {
+ SDLoc dl, EVT DestVT = MVT::Other) {
if (DestVT == MVT::Other) DestVT = Op0.getValueType();
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, DestVT,
DAG.getConstant(IID, MVT::i32), Op0, Op1, Op2);
/// BuildVSLDOI - Return a VECTOR_SHUFFLE that is a vsldoi of the specified
/// amount. The result has the specified value type.
static SDValue BuildVSLDOI(SDValue LHS, SDValue RHS, unsigned Amt,
- EVT VT, SelectionDAG &DAG, DebugLoc dl) {
+ EVT VT, SelectionDAG &DAG, SDLoc dl) {
// Force LHS/RHS to be the right type.
LHS = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, LHS);
RHS = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, RHS);
// sequence of ops that should be used.
SDValue PPCTargetLowering::LowerBUILD_VECTOR(SDValue Op,
SelectionDAG &DAG) const {
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(Op.getNode());
assert(BVN != 0 && "Expected a BuildVectorSDNode in LowerBUILD_VECTOR");
/// the specified operations to build the shuffle.
static SDValue GeneratePerfectShuffle(unsigned PFEntry, SDValue LHS,
SDValue RHS, SelectionDAG &DAG,
- DebugLoc dl) {
+ SDLoc dl) {
unsigned OpNum = (PFEntry >> 26) & 0x0F;
unsigned LHSID = (PFEntry >> 13) & ((1 << 13)-1);
unsigned RHSID = (PFEntry >> 0) & ((1 << 13)-1);
/// lowered into a vperm.
SDValue PPCTargetLowering::LowerVECTOR_SHUFFLE(SDValue Op,
SelectionDAG &DAG) const {
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
SDValue V1 = Op.getOperand(0);
SDValue V2 = Op.getOperand(1);
ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(Op);
SelectionDAG &DAG) const {
// If this is a lowered altivec predicate compare, CompareOpc is set to the
// opcode number of the comparison.
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
int CompareOpc;
bool isDot;
if (!getAltivecCompareInfo(Op, CompareOpc, isDot))
// Now that we have the comparison, emit a copy from the CR to a GPR.
// This is flagged to the above dot comparison.
- SDValue Flags = DAG.getNode(PPCISD::MFCR, dl, MVT::i32,
+ SDValue Flags = DAG.getNode(PPCISD::MFOCRF, dl, MVT::i32,
DAG.getRegister(PPC::CR6, MVT::i32),
CompNode.getValue(1));
SDValue PPCTargetLowering::LowerSCALAR_TO_VECTOR(SDValue Op,
SelectionDAG &DAG) const {
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
// Create a stack slot that is 16-byte aligned.
MachineFrameInfo *FrameInfo = DAG.getMachineFunction().getFrameInfo();
int FrameIdx = FrameInfo->CreateStackObject(16, 16, false);
}
SDValue PPCTargetLowering::LowerMUL(SDValue Op, SelectionDAG &DAG) const {
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
if (Op.getValueType() == MVT::v4i32) {
SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
case ISD::FP_TO_UINT:
case ISD::FP_TO_SINT: return LowerFP_TO_INT(Op, DAG,
- Op.getDebugLoc());
+ SDLoc(Op));
case ISD::UINT_TO_FP:
case ISD::SINT_TO_FP: return LowerINT_TO_FP(Op, DAG);
case ISD::FLT_ROUNDS_: return LowerFLT_ROUNDS_(Op, DAG);
SmallVectorImpl<SDValue>&Results,
SelectionDAG &DAG) const {
const TargetMachine &TM = getTargetMachine();
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
switch (N->getOpcode()) {
default:
llvm_unreachable("Do not know how to custom type legalize this operation!");
// Setup
MIB = BuildMI(*thisMBB, MI, DL, TII->get(PPC::BCLalways)).addMBB(mainMBB);
- MIB.addRegMask(PPCRegInfo->getNoPreservedMask());
+ const PPCRegisterInfo *TRI =
+ static_cast<const PPCRegisterInfo*>(getTargetMachine().getRegisterInfo());
+ MIB.addRegMask(TRI->getNoPreservedMask());
BuildMI(*thisMBB, MI, DL, TII->get(PPC::LI), restoreDstReg).addImm(1);
Cond.push_back(MI->getOperand(1));
DebugLoc dl = MI->getDebugLoc();
- PPCII->insertSelect(*BB, MI, dl, MI->getOperand(0).getReg(), Cond,
- MI->getOperand(2).getReg(), MI->getOperand(3).getReg());
+ const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+ TII->insertSelect(*BB, MI, dl, MI->getOperand(0).getReg(),
+ Cond, MI->getOperand(2).getReg(),
+ MI->getOperand(3).getReg());
} else if (MI->getOpcode() == PPC::SELECT_CC_I4 ||
MI->getOpcode() == PPC::SELECT_CC_I8 ||
MI->getOpcode() == PPC::SELECT_CC_F4 ||
++Iterations;
SelectionDAG &DAG = DCI.DAG;
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
SDValue FPOne =
DAG.getConstantFP(1.0, VT.getScalarType());
++Iterations;
SelectionDAG &DAG = DCI.DAG;
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
SDValue FPThreeHalves =
DAG.getConstantFP(1.5, VT.getScalarType());
return SDValue();
}
+// Like SelectionDAG::isConsecutiveLoad, but also works for stores, and does
+// not enforce equality of the chain operands.
+static bool isConsecutiveLS(LSBaseSDNode *LS, LSBaseSDNode *Base,
+ unsigned Bytes, int Dist,
+ SelectionDAG &DAG) {
+ EVT VT = LS->getMemoryVT();
+ if (VT.getSizeInBits() / 8 != Bytes)
+ return false;
+
+ SDValue Loc = LS->getBasePtr();
+ SDValue BaseLoc = Base->getBasePtr();
+ if (Loc.getOpcode() == ISD::FrameIndex) {
+ if (BaseLoc.getOpcode() != ISD::FrameIndex)
+ return false;
+ const MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
+ int FI = cast<FrameIndexSDNode>(Loc)->getIndex();
+ int BFI = cast<FrameIndexSDNode>(BaseLoc)->getIndex();
+ int FS = MFI->getObjectSize(FI);
+ int BFS = MFI->getObjectSize(BFI);
+ if (FS != BFS || FS != (int)Bytes) return false;
+ return MFI->getObjectOffset(FI) == (MFI->getObjectOffset(BFI) + Dist*Bytes);
+ }
+
+ // Handle X+C
+ if (DAG.isBaseWithConstantOffset(Loc) && Loc.getOperand(0) == BaseLoc &&
+ cast<ConstantSDNode>(Loc.getOperand(1))->getSExtValue() == Dist*Bytes)
+ return true;
+
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+ const GlobalValue *GV1 = NULL;
+ const GlobalValue *GV2 = NULL;
+ int64_t Offset1 = 0;
+ int64_t Offset2 = 0;
+ bool isGA1 = TLI.isGAPlusOffset(Loc.getNode(), GV1, Offset1);
+ bool isGA2 = TLI.isGAPlusOffset(BaseLoc.getNode(), GV2, Offset2);
+ if (isGA1 && isGA2 && GV1 == GV2)
+ return Offset1 == (Offset2 + Dist*Bytes);
+ return false;
+}
+
+// Return true is there is a nearyby consecutive load to the one provided
+// (regardless of alignment). We search up and down the chain, looking though
+// token factors and other loads (but nothing else). As a result, a true
+// results indicates that it is safe to create a new consecutive load adjacent
+// to the load provided.
+static bool findConsecutiveLoad(LoadSDNode *LD, SelectionDAG &DAG) {
+ SDValue Chain = LD->getChain();
+ EVT VT = LD->getMemoryVT();
+
+ SmallSet<SDNode *, 16> LoadRoots;
+ SmallVector<SDNode *, 8> Queue(1, Chain.getNode());
+ SmallSet<SDNode *, 16> Visited;
+
+ // First, search up the chain, branching to follow all token-factor operands.
+ // If we find a consecutive load, then we're done, otherwise, record all
+ // nodes just above the top-level loads and token factors.
+ while (!Queue.empty()) {
+ SDNode *ChainNext = Queue.pop_back_val();
+ if (!Visited.insert(ChainNext))
+ continue;
+
+ if (LoadSDNode *ChainLD = dyn_cast<LoadSDNode>(ChainNext)) {
+ if (isConsecutiveLS(ChainLD, LD, VT.getStoreSize(), 1, DAG))
+ return true;
+
+ if (!Visited.count(ChainLD->getChain().getNode()))
+ Queue.push_back(ChainLD->getChain().getNode());
+ } else if (ChainNext->getOpcode() == ISD::TokenFactor) {
+ for (SDNode::op_iterator O = ChainNext->op_begin(),
+ OE = ChainNext->op_end(); O != OE; ++O)
+ if (!Visited.count(O->getNode()))
+ Queue.push_back(O->getNode());
+ } else
+ LoadRoots.insert(ChainNext);
+ }
+
+ // Second, search down the chain, starting from the top-level nodes recorded
+ // in the first phase. These top-level nodes are the nodes just above all
+ // loads and token factors. Starting with their uses, recursively look though
+ // all loads (just the chain uses) and token factors to find a consecutive
+ // load.
+ Visited.clear();
+ Queue.clear();
+
+ for (SmallSet<SDNode *, 16>::iterator I = LoadRoots.begin(),
+ IE = LoadRoots.end(); I != IE; ++I) {
+ Queue.push_back(*I);
+
+ while (!Queue.empty()) {
+ SDNode *LoadRoot = Queue.pop_back_val();
+ if (!Visited.insert(LoadRoot))
+ continue;
+
+ if (LoadSDNode *ChainLD = dyn_cast<LoadSDNode>(LoadRoot))
+ if (isConsecutiveLS(ChainLD, LD, VT.getStoreSize(), 1, DAG))
+ return true;
+
+ for (SDNode::use_iterator UI = LoadRoot->use_begin(),
+ UE = LoadRoot->use_end(); UI != UE; ++UI)
+ if (((isa<LoadSDNode>(*UI) &&
+ cast<LoadSDNode>(*UI)->getChain().getNode() == LoadRoot) ||
+ UI->getOpcode() == ISD::TokenFactor) && !Visited.count(*UI))
+ Queue.push_back(*UI);
+ }
+ }
+
+ return false;
+}
+
SDValue PPCTargetLowering::PerformDAGCombine(SDNode *N,
DAGCombinerInfo &DCI) const {
const TargetMachine &TM = getTargetMachine();
SelectionDAG &DAG = DCI.DAG;
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
switch (N->getOpcode()) {
default: break;
case PPCISD::SHL:
DCI);
if (RV.getNode() != 0) {
DCI.AddToWorklist(RV.getNode());
- RV = DAG.getNode(ISD::FP_EXTEND, N->getOperand(1).getDebugLoc(),
+ RV = DAG.getNode(ISD::FP_EXTEND, SDLoc(N->getOperand(1)),
N->getValueType(0), RV);
DCI.AddToWorklist(RV.getNode());
return DAG.getNode(ISD::FMUL, dl, N->getValueType(0),
DCI);
if (RV.getNode() != 0) {
DCI.AddToWorklist(RV.getNode());
- RV = DAG.getNode(ISD::FP_ROUND, N->getOperand(1).getDebugLoc(),
+ RV = DAG.getNode(ISD::FP_ROUND, SDLoc(N->getOperand(1)),
N->getValueType(0), RV,
N->getOperand(1).getOperand(1));
DCI.AddToWorklist(RV.getNode());
// cause the last vector in the sequence to be (re)loaded. Otherwise,
// the next vector will be fetched as you might suspect was necessary.
- // FIXME: We might be able to reuse the permutation generation from
+ // We might be able to reuse the permutation generation from
// a different base address offset from this one by an aligned amount.
+ // The INTRINSIC_WO_CHAIN DAG combine will attempt to perform this
+ // optimization later.
SDValue PermCntl = BuildIntrinsicOp(Intrinsic::ppc_altivec_lvsl, Ptr,
DAG, dl, MVT::v16i8);
// require the next load to appear to be aligned, even though it
// is actually offset from the base pointer by a lesser amount.
int IncOffset = VT.getSizeInBits() / 8;
- int IncValue = IncOffset - 1;
+ int IncValue = IncOffset;
+
+ // Walk (both up and down) the chain looking for another load at the real
+ // (aligned) offset (the alignment of the other load does not matter in
+ // this case). If found, then do not use the offset reduction trick, as
+ // that will prevent the loads from being later combined (as they would
+ // otherwise be duplicates).
+ if (!findConsecutiveLoad(LD, DAG))
+ --IncValue;
+
SDValue Increment = DAG.getConstant(IncValue, getPointerTy());
Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr, Increment);
- // FIXME: We might have another load (with a slightly-different
- // real offset) that we can reuse here.
SDValue ExtraLoad =
DAG.getLoad(VT, dl, Chain, Ptr,
LD->getPointerInfo().getWithOffset(IncOffset),
}
}
break;
+ case ISD::INTRINSIC_WO_CHAIN:
+ if (cast<ConstantSDNode>(N->getOperand(0))->getZExtValue() ==
+ Intrinsic::ppc_altivec_lvsl &&
+ N->getOperand(1)->getOpcode() == ISD::ADD) {
+ SDValue Add = N->getOperand(1);
+
+ if (DAG.MaskedValueIsZero(Add->getOperand(1),
+ APInt::getAllOnesValue(4 /* 16 byte alignment */).zext(
+ Add.getValueType().getScalarType().getSizeInBits()))) {
+ SDNode *BasePtr = Add->getOperand(0).getNode();
+ for (SDNode::use_iterator UI = BasePtr->use_begin(),
+ UE = BasePtr->use_end(); UI != UE; ++UI) {
+ if (UI->getOpcode() == ISD::INTRINSIC_WO_CHAIN &&
+ cast<ConstantSDNode>(UI->getOperand(0))->getZExtValue() ==
+ Intrinsic::ppc_altivec_lvsl) {
+ // We've found another LVSL, and this address if an aligned
+ // multiple of that one. The results will be the same, so use the
+ // one we've just found instead.
+
+ return SDValue(*UI, 0);
+ }
+ }
+ }
+ }
case ISD::BSWAP:
// Turn BSWAP (LOAD) -> lhbrx/lwbrx.
if (ISD::isNON_EXTLoad(N->getOperand(0).getNode()) &&
}
}
- // If the user is a MFCR instruction, we know this is safe. Otherwise we
- // give up for right now.
- if (FlagUser->getOpcode() == PPCISD::MFCR)
+ // If the user is a MFOCRF instruction, we know this is safe.
+ // Otherwise we give up for right now.
+ if (FlagUser->getOpcode() == PPCISD::MFOCRF)
return SDValue(VCMPoNode, 0);
}
break;
}
case ISD::BR_CC: {
// If this is a branch on an altivec predicate comparison, lower this so
- // that we don't have to do a MFCR: instead, branch directly on CR6. This
+ // that we don't have to do a MFOCRF: instead, branch directly on CR6. This
// lowering is done pre-legalize, because the legalizer lowers the predicate
// compare down to code that is difficult to reassemble.
ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(1))->get();
std::pair<unsigned, const TargetRegisterClass*>
PPCTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
- EVT VT) const {
+ MVT VT) const {
if (Constraint.size() == 1) {
// GCC RS6000 Constraint Letters
switch (Constraint[0]) {
MachineFrameInfo *MFI = MF.getFrameInfo();
MFI->setReturnAddressIsTaken(true);
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
// Make sure the function does not optimize away the store of the RA to
SDValue PPCTargetLowering::LowerFRAMEADDR(SDValue Op,
SelectionDAG &DAG) const {
- DebugLoc dl = Op.getDebugLoc();
+ SDLoc dl(Op);
unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();