#include "X86InstrBuilder.h"
#include "X86ISelLowering.h"
#include "X86TargetMachine.h"
+#include "X86TargetObjectFile.h"
#include "llvm/CallingConv.h"
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Target/TargetLoweringObjectFile.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/StringExtras.h"
static cl::opt<bool>
DisableMMX("disable-mmx", cl::Hidden, cl::desc("Disable use of MMX"));
+// Disable16Bit - 16-bit operations typically have a larger encoding than
+// corresponding 32-bit instructions, and 16-bit code is slow on some
+// processors. This is an experimental flag to disable 16-bit operations
+// (which forces them to be Legalized to 32-bit operations).
+static cl::opt<bool>
+Disable16Bit("disable-16bit", cl::Hidden,
+ cl::desc("Disable use of 16-bit instructions"));
+
// Forward declarations.
static SDValue getMOVL(SelectionDAG &DAG, DebugLoc dl, EVT VT, SDValue V1,
SDValue V2);
switch (TM.getSubtarget<X86Subtarget>().TargetType) {
default: llvm_unreachable("unknown subtarget type");
case X86Subtarget::isDarwin:
+ if (TM.getSubtarget<X86Subtarget>().is64Bit())
+ return new X8664_MachoTargetObjectFile();
return new TargetLoweringObjectFileMachO();
case X86Subtarget::isELF:
return new TargetLoweringObjectFileELF();
// Set up the register classes.
addRegisterClass(MVT::i8, X86::GR8RegisterClass);
- addRegisterClass(MVT::i16, X86::GR16RegisterClass);
+ if (!Disable16Bit)
+ addRegisterClass(MVT::i16, X86::GR16RegisterClass);
addRegisterClass(MVT::i32, X86::GR32RegisterClass);
if (Subtarget->is64Bit())
addRegisterClass(MVT::i64, X86::GR64RegisterClass);
// We don't accept any truncstore of integer registers.
setTruncStoreAction(MVT::i64, MVT::i32, Expand);
- setTruncStoreAction(MVT::i64, MVT::i16, Expand);
+ if (!Disable16Bit)
+ setTruncStoreAction(MVT::i64, MVT::i16, Expand);
setTruncStoreAction(MVT::i64, MVT::i8 , Expand);
- setTruncStoreAction(MVT::i32, MVT::i16, Expand);
+ if (!Disable16Bit)
+ setTruncStoreAction(MVT::i32, MVT::i16, Expand);
setTruncStoreAction(MVT::i32, MVT::i8 , Expand);
setTruncStoreAction(MVT::i16, MVT::i8, Expand);
setOperationAction(ISD::CTTZ , MVT::i8 , Custom);
setOperationAction(ISD::CTLZ , MVT::i8 , Custom);
setOperationAction(ISD::CTPOP , MVT::i16 , Expand);
- setOperationAction(ISD::CTTZ , MVT::i16 , Custom);
- setOperationAction(ISD::CTLZ , MVT::i16 , Custom);
+ if (Disable16Bit) {
+ setOperationAction(ISD::CTTZ , MVT::i16 , Expand);
+ setOperationAction(ISD::CTLZ , MVT::i16 , Expand);
+ } else {
+ setOperationAction(ISD::CTTZ , MVT::i16 , Custom);
+ setOperationAction(ISD::CTLZ , MVT::i16 , Custom);
+ }
setOperationAction(ISD::CTPOP , MVT::i32 , Expand);
setOperationAction(ISD::CTTZ , MVT::i32 , Custom);
setOperationAction(ISD::CTLZ , MVT::i32 , Custom);
setOperationAction(ISD::SELECT , MVT::i1 , Promote);
// X86 wants to expand cmov itself.
setOperationAction(ISD::SELECT , MVT::i8 , Custom);
- setOperationAction(ISD::SELECT , MVT::i16 , Custom);
+ if (Disable16Bit)
+ setOperationAction(ISD::SELECT , MVT::i16 , Expand);
+ else
+ setOperationAction(ISD::SELECT , MVT::i16 , Custom);
setOperationAction(ISD::SELECT , MVT::i32 , Custom);
setOperationAction(ISD::SELECT , MVT::f32 , Custom);
setOperationAction(ISD::SELECT , MVT::f64 , Custom);
setOperationAction(ISD::SELECT , MVT::f80 , Custom);
setOperationAction(ISD::SETCC , MVT::i8 , Custom);
- setOperationAction(ISD::SETCC , MVT::i16 , Custom);
+ if (Disable16Bit)
+ setOperationAction(ISD::SETCC , MVT::i16 , Expand);
+ else
+ setOperationAction(ISD::SETCC , MVT::i16 , Custom);
setOperationAction(ISD::SETCC , MVT::i32 , Custom);
setOperationAction(ISD::SETCC , MVT::f32 , Custom);
setOperationAction(ISD::SETCC , MVT::f64 , Custom);
return F->hasFnAttr(Attribute::OptimizeForSize) ? 0 : 4;
}
-/// getPreferredLSDADataFormat - Return the preferred exception handling data
-/// format for the LSDA.
-unsigned X86TargetLowering::getPreferredLSDADataFormat() const {
- if (Subtarget->isTargetDarwin())
- return dwarf::DW_EH_PE_pcrel;
-
- CodeModel::Model M = getTargetMachine().getCodeModel();
-
- if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
- if (!Subtarget->is64Bit() || M == CodeModel::Small)
- return dwarf::DW_EH_PE_pcrel | dwarf::DW_EH_PE_sdata4;
-
- return dwarf::DW_EH_PE_pcrel | dwarf::DW_EH_PE_sdata8;
- }
-
- if (M == CodeModel::Small)
- return dwarf::DW_EH_PE_sdata4;
-
- return dwarf::DW_EH_PE_absptr;
-}
-
-/// getPreferredFDEDataFormat - Return the preferred exception handling data
-/// format for the FDE.
-unsigned X86TargetLowering::getPreferredFDEDataFormat() const {
- if (Subtarget->isTargetDarwin())
- return dwarf::DW_EH_PE_pcrel;
-
- CodeModel::Model M = getTargetMachine().getCodeModel();
-
- if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
- if (!Subtarget->is64Bit() ||
- M == CodeModel::Small || M == CodeModel::Medium)
- return dwarf::DW_EH_PE_pcrel | dwarf::DW_EH_PE_sdata4;
-
- return dwarf::DW_EH_PE_pcrel | dwarf::DW_EH_PE_sdata8;
- }
-
- if (M == CodeModel::Small || M == CodeModel::Medium)
- return dwarf::DW_EH_PE_sdata4;
-
- return dwarf::DW_EH_PE_absptr;
-}
-
//===----------------------------------------------------------------------===//
// Return Value Calling Convention Implementation
//===----------------------------------------------------------------------===//
SDValue
X86TargetLowering::LowerReturn(SDValue Chain,
- unsigned CallConv, bool isVarArg,
+ CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
DebugLoc dl, SelectionDAG &DAG) {
SmallVector<SDValue, 6> RetOps;
RetOps.push_back(Chain); // Operand #0 = Chain (updated below)
// Operand #1 = Bytes To Pop
- RetOps.push_back(DAG.getConstant(getBytesToPopOnReturn(), MVT::i16));
+ RetOps.push_back(DAG.getTargetConstant(getBytesToPopOnReturn(), MVT::i16));
// Copy the result values into the output registers.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
///
SDValue
X86TargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
- unsigned CallConv, bool isVarArg,
+ CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) {
/// IsCalleePop - Determines whether the callee is required to pop its
/// own arguments. Callee pop is necessary to support tail calls.
-bool X86TargetLowering::IsCalleePop(bool IsVarArg, unsigned CallingConv) {
+bool X86TargetLowering::IsCalleePop(bool IsVarArg, CallingConv::ID CallingConv){
if (IsVarArg)
return false;
/// CCAssignFnForNode - Selects the correct CCAssignFn for a the
/// given CallingConvention value.
-CCAssignFn *X86TargetLowering::CCAssignFnForNode(unsigned CC) const {
+CCAssignFn *X86TargetLowering::CCAssignFnForNode(CallingConv::ID CC) const {
if (Subtarget->is64Bit()) {
if (Subtarget->isTargetWin64())
return CC_X86_Win64_C;
/// NameDecorationForCallConv - Selects the appropriate decoration to
/// apply to a MachineFunction containing a given calling convention.
NameDecorationStyle
-X86TargetLowering::NameDecorationForCallConv(unsigned CallConv) {
+X86TargetLowering::NameDecorationForCallConv(CallingConv::ID CallConv) {
if (CallConv == CallingConv::X86_FastCall)
return FastCall;
else if (CallConv == CallingConv::X86_StdCall)
SDValue
X86TargetLowering::LowerMemArgument(SDValue Chain,
- unsigned CallConv,
+ CallingConv::ID CallConv,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
const CCValAssign &VA,
SDValue
X86TargetLowering::LowerFormalArguments(SDValue Chain,
- unsigned CallConv,
+ CallingConv::ID CallConv,
bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl,
SDValue
X86TargetLowering::LowerCall(SDValue Chain, SDValue Callee,
- unsigned CallConv, bool isVarArg, bool isTailCall,
+ CallingConv::ID CallConv, bool isVarArg,
+ bool isTailCall,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
/// optimization should implement this function.
bool
X86TargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
- unsigned CalleeCC,
+ CallingConv::ID CalleeCC,
bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
SelectionDAG& DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
- unsigned CallerCC = MF.getFunction()->getCallingConv();
+ CallingConv::ID CallerCC = MF.getFunction()->getCallingConv();
return CalleeCC == CallingConv::Fast && CallerCC == CalleeCC;
}
DebugLoc dl = Op.getDebugLoc();
EVT VT = Op.getValueType();
EVT EltVT = VT;
- unsigned EltNum = 1;
- if (VT.isVector()) {
+ if (VT.isVector())
EltVT = VT.getVectorElementType();
- EltNum = VT.getVectorNumElements();
- }
std::vector<Constant*> CV;
if (EltVT == MVT::f64) {
Constant *C = ConstantFP::get(*Context, APFloat(APInt(64, 1ULL << 63)));
break;
}
}
+
+ // The vector shift intrinsics with scalars uses 32b shift amounts but
+ // the sse2/mmx shift instructions reads 64 bits. Set the upper 32 bits
+ // to be zero.
+ SDValue ShOps[4];
+ ShOps[0] = ShAmt;
+ ShOps[1] = DAG.getConstant(0, MVT::i32);
+ if (ShAmtVT == MVT::v4i32) {
+ ShOps[2] = DAG.getUNDEF(MVT::i32);
+ ShOps[3] = DAG.getUNDEF(MVT::i32);
+ ShAmt = DAG.getNode(ISD::BUILD_VECTOR, dl, ShAmtVT, &ShOps[0], 4);
+ } else {
+ ShAmt = DAG.getNode(ISD::BUILD_VECTOR, dl, ShAmtVT, &ShOps[0], 2);
+ }
+
EVT VT = Op.getValueType();
- ShAmt = DAG.getNode(ISD::BIT_CONVERT, dl, VT,
- DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, ShAmtVT, ShAmt));
+ ShAmt = DAG.getNode(ISD::BIT_CONVERT, dl, VT, ShAmt);
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, VT,
DAG.getConstant(NewIntNo, MVT::i32),
Op.getOperand(1), ShAmt);
} else {
const Function *Func =
cast<Function>(cast<SrcValueSDNode>(Op.getOperand(5))->getValue());
- unsigned CC = Func->getCallingConv();
+ CallingConv::ID CC = Func->getCallingConv();
unsigned NestReg;
switch (CC) {
return EndMBB;
}
+MachineBasicBlock *
+X86TargetLowering::EmitLoweredSelect(MachineInstr *MI,
+ MachineBasicBlock *BB) const {
+ const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+ DebugLoc DL = MI->getDebugLoc();
+
+ // To "insert" a SELECT_CC instruction, we actually have to insert the
+ // diamond control-flow pattern. The incoming instruction knows the
+ // destination vreg to set, the condition code register to branch on, the
+ // true/false values to select between, and a branch opcode to use.
+ const BasicBlock *LLVM_BB = BB->getBasicBlock();
+ MachineFunction::iterator It = BB;
+ ++It;
+
+ // thisMBB:
+ // ...
+ // TrueVal = ...
+ // cmpTY ccX, r1, r2
+ // bCC copy1MBB
+ // fallthrough --> copy0MBB
+ MachineBasicBlock *thisMBB = BB;
+ MachineFunction *F = BB->getParent();
+ MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB);
+ MachineBasicBlock *sinkMBB = F->CreateMachineBasicBlock(LLVM_BB);
+ unsigned Opc =
+ X86::GetCondBranchFromCond((X86::CondCode)MI->getOperand(3).getImm());
+ BuildMI(BB, DL, TII->get(Opc)).addMBB(sinkMBB);
+ F->insert(It, copy0MBB);
+ F->insert(It, sinkMBB);
+ // Update machine-CFG edges by transferring all successors of the current
+ // block to the new block which will contain the Phi node for the select.
+ sinkMBB->transferSuccessors(BB);
+
+ // Add the true and fallthrough blocks as its successors.
+ BB->addSuccessor(copy0MBB);
+ BB->addSuccessor(sinkMBB);
+
+ // copy0MBB:
+ // %FalseValue = ...
+ // # fallthrough to sinkMBB
+ BB = copy0MBB;
+
+ // Update machine-CFG edges
+ BB->addSuccessor(sinkMBB);
+
+ // sinkMBB:
+ // %Result = phi [ %FalseValue, copy0MBB ], [ %TrueValue, thisMBB ]
+ // ...
+ BB = sinkMBB;
+ BuildMI(BB, DL, TII->get(X86::PHI), MI->getOperand(0).getReg())
+ .addReg(MI->getOperand(1).getReg()).addMBB(copy0MBB)
+ .addReg(MI->getOperand(2).getReg()).addMBB(thisMBB);
+
+ F->DeleteMachineInstr(MI); // The pseudo instruction is gone now.
+ return BB;
+}
+
+
MachineBasicBlock *
X86TargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
MachineBasicBlock *BB) const {
- DebugLoc dl = MI->getDebugLoc();
- const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
switch (MI->getOpcode()) {
default: assert(false && "Unexpected instr type to insert");
case X86::CMOV_GR8:
case X86::CMOV_FR64:
case X86::CMOV_V4F32:
case X86::CMOV_V2F64:
- case X86::CMOV_V2I64: {
- // To "insert" a SELECT_CC instruction, we actually have to insert the
- // diamond control-flow pattern. The incoming instruction knows the
- // destination vreg to set, the condition code register to branch on, the
- // true/false values to select between, and a branch opcode to use.
- const BasicBlock *LLVM_BB = BB->getBasicBlock();
- MachineFunction::iterator It = BB;
- ++It;
-
- // thisMBB:
- // ...
- // TrueVal = ...
- // cmpTY ccX, r1, r2
- // bCC copy1MBB
- // fallthrough --> copy0MBB
- MachineBasicBlock *thisMBB = BB;
- MachineFunction *F = BB->getParent();
- MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB);
- MachineBasicBlock *sinkMBB = F->CreateMachineBasicBlock(LLVM_BB);
- unsigned Opc =
- X86::GetCondBranchFromCond((X86::CondCode)MI->getOperand(3).getImm());
- BuildMI(BB, dl, TII->get(Opc)).addMBB(sinkMBB);
- F->insert(It, copy0MBB);
- F->insert(It, sinkMBB);
- // Update machine-CFG edges by transferring all successors of the current
- // block to the new block which will contain the Phi node for the select.
- sinkMBB->transferSuccessors(BB);
-
- // Add the true and fallthrough blocks as its successors.
- BB->addSuccessor(copy0MBB);
- BB->addSuccessor(sinkMBB);
-
- // copy0MBB:
- // %FalseValue = ...
- // # fallthrough to sinkMBB
- BB = copy0MBB;
-
- // Update machine-CFG edges
- BB->addSuccessor(sinkMBB);
-
- // sinkMBB:
- // %Result = phi [ %FalseValue, copy0MBB ], [ %TrueValue, thisMBB ]
- // ...
- BB = sinkMBB;
- BuildMI(BB, dl, TII->get(X86::PHI), MI->getOperand(0).getReg())
- .addReg(MI->getOperand(1).getReg()).addMBB(copy0MBB)
- .addReg(MI->getOperand(2).getReg()).addMBB(thisMBB);
-
- F->DeleteMachineInstr(MI); // The pseudo instruction is gone now.
- return BB;
- }
+ case X86::CMOV_V2I64:
+ return EmitLoweredSelect(MI, BB);
case X86::FP32_TO_INT16_IN_MEM:
case X86::FP32_TO_INT32_IN_MEM:
case X86::FP80_TO_INT16_IN_MEM:
case X86::FP80_TO_INT32_IN_MEM:
case X86::FP80_TO_INT64_IN_MEM: {
+ const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+ DebugLoc DL = MI->getDebugLoc();
+
// Change the floating point control register to use "round towards zero"
// mode when truncating to an integer value.
MachineFunction *F = BB->getParent();
int CWFrameIdx = F->getFrameInfo()->CreateStackObject(2, 2);
- addFrameReference(BuildMI(BB, dl, TII->get(X86::FNSTCW16m)), CWFrameIdx);
+ addFrameReference(BuildMI(BB, DL, TII->get(X86::FNSTCW16m)), CWFrameIdx);
// Load the old value of the high byte of the control word...
unsigned OldCW =
F->getRegInfo().createVirtualRegister(X86::GR16RegisterClass);
- addFrameReference(BuildMI(BB, dl, TII->get(X86::MOV16rm), OldCW),
+ addFrameReference(BuildMI(BB, DL, TII->get(X86::MOV16rm), OldCW),
CWFrameIdx);
// Set the high part to be round to zero...
- addFrameReference(BuildMI(BB, dl, TII->get(X86::MOV16mi)), CWFrameIdx)
+ addFrameReference(BuildMI(BB, DL, TII->get(X86::MOV16mi)), CWFrameIdx)
.addImm(0xC7F);
// Reload the modified control word now...
- addFrameReference(BuildMI(BB, dl, TII->get(X86::FLDCW16m)), CWFrameIdx);
+ addFrameReference(BuildMI(BB, DL, TII->get(X86::FLDCW16m)), CWFrameIdx);
// Restore the memory image of control word to original value
- addFrameReference(BuildMI(BB, dl, TII->get(X86::MOV16mr)), CWFrameIdx)
+ addFrameReference(BuildMI(BB, DL, TII->get(X86::MOV16mr)), CWFrameIdx)
.addReg(OldCW);
// Get the X86 opcode to use.
} else {
AM.Disp = Op.getImm();
}
- addFullAddress(BuildMI(BB, dl, TII->get(Opc)), AM)
+ addFullAddress(BuildMI(BB, DL, TII->get(Opc)), AM)
.addReg(MI->getOperand(X86AddrNumOperands).getReg());
// Reload the original control word now.
- addFrameReference(BuildMI(BB, dl, TII->get(X86::FLDCW16m)), CWFrameIdx);
+ addFrameReference(BuildMI(BB, DL, TII->get(X86::FLDCW16m)), CWFrameIdx);
F->DeleteMachineInstr(MI); // The pseudo instruction is gone now.
return BB;
} else if (LHS == Cond.getOperand(1) && RHS == Cond.getOperand(0)) {
switch (CC) {
default: break;
- case ISD::SETOGT: // (X > Y) ? Y : X -> min
- case ISD::SETUGT:
+ case ISD::SETOGT:
+ // This can use a min only if the LHS isn't NaN.
+ if (DAG.isKnownNeverNaN(LHS))
+ Opcode = X86ISD::FMIN;
+ else if (DAG.isKnownNeverNaN(RHS)) {
+ Opcode = X86ISD::FMIN;
+ // Put the potential NaN in the RHS so that SSE will preserve it.
+ std::swap(LHS, RHS);
+ }
+ break;
+
+ case ISD::SETUGT: // (X > Y) ? Y : X -> min
case ISD::SETGT:
if (!UnsafeFPMath) break;
// FALL THROUGH.
Opcode = X86ISD::FMIN;
break;
- case ISD::SETOLE: // (X <= Y) ? Y : X -> max
case ISD::SETULE:
+ // This can use a max only if the LHS isn't NaN.
+ if (DAG.isKnownNeverNaN(LHS))
+ Opcode = X86ISD::FMAX;
+ else if (DAG.isKnownNeverNaN(RHS)) {
+ Opcode = X86ISD::FMAX;
+ // Put the potential NaN in the RHS so that SSE will preserve it.
+ std::swap(LHS, RHS);
+ }
+ break;
+
+ case ISD::SETOLE: // (X <= Y) ? Y : X -> max
case ISD::SETLE:
if (!UnsafeFPMath) break;
// FALL THROUGH.
SDValue ShAmtOp = N->getOperand(1);
EVT EltVT = VT.getVectorElementType();
DebugLoc DL = N->getDebugLoc();
- SDValue BaseShAmt;
+ SDValue BaseShAmt = SDValue();
if (ShAmtOp.getOpcode() == ISD::BUILD_VECTOR) {
unsigned NumElts = VT.getVectorNumElements();
unsigned i = 0;
}
} else if (ShAmtOp.getOpcode() == ISD::VECTOR_SHUFFLE &&
cast<ShuffleVectorSDNode>(ShAmtOp)->isSplat()) {
- BaseShAmt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, EltVT, ShAmtOp,
- DAG.getIntPtrConstant(0));
+ SDValue InVec = ShAmtOp.getOperand(0);
+ if (InVec.getOpcode() == ISD::BUILD_VECTOR) {
+ unsigned NumElts = InVec.getValueType().getVectorNumElements();
+ unsigned i = 0;
+ for (; i != NumElts; ++i) {
+ SDValue Arg = InVec.getOperand(i);
+ if (Arg.getOpcode() == ISD::UNDEF) continue;
+ BaseShAmt = Arg;
+ break;
+ }
+ } else if (InVec.getOpcode() == ISD::INSERT_VECTOR_ELT) {
+ if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(InVec.getOperand(2))) {
+ unsigned SplatIdx = cast<ShuffleVectorSDNode>(ShAmtOp)->getSplatIndex();
+ if (C->getZExtValue() == SplatIdx)
+ BaseShAmt = InVec.getOperand(1);
+ }
+ }
+ if (BaseShAmt.getNode() == 0)
+ BaseShAmt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, EltVT, ShAmtOp,
+ DAG.getIntPtrConstant(0));
} else
return SDValue();
+ // The shift amount is an i32.
if (EltVT.bitsGT(MVT::i32))
BaseShAmt = DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, BaseShAmt);
else if (EltVT.bitsLT(MVT::i32))
- BaseShAmt = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i32, BaseShAmt);
+ BaseShAmt = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i32, BaseShAmt);
// The shift amount is identical so we can do a vector shift.
SDValue ValOp = N->getOperand(0);
// Not found as a standard register?
if (Res.second == 0) {
- // GCC calls "st(0)" just plain "st".
+ // Map st(0) -> st(7) -> ST0
+ if (Constraint.size() == 7 && Constraint[0] == '{' &&
+ tolower(Constraint[1]) == 's' &&
+ tolower(Constraint[2]) == 't' &&
+ Constraint[3] == '(' &&
+ (Constraint[4] >= '0' && Constraint[4] <= '7') &&
+ Constraint[5] == ')' &&
+ Constraint[6] == '}') {
+
+ Res.first = X86::ST0+Constraint[4]-'0';
+ Res.second = X86::RFP80RegisterClass;
+ return Res;
+ }
+
+ // GCC allows "st(0)" to be called just plain "st".
if (StringsEqualNoCase("{st}", Constraint)) {
Res.first = X86::ST0;
Res.second = X86::RFP80RegisterClass;
+ return Res;
}
+
+ // flags -> EFLAGS
+ if (StringsEqualNoCase("{flags}", Constraint)) {
+ Res.first = X86::EFLAGS;
+ Res.second = X86::CCRRegisterClass;
+ return Res;
+ }
+
// 'A' means EAX + EDX.
if (Constraint == "A") {
Res.first = X86::EAX;
Res.second = X86::GR32_ADRegisterClass;
+ return Res;
}
return Res;
}