#include "llvm/GlobalAlias.h"
#include "llvm/GlobalVariable.h"
#include "llvm/Function.h"
+#include "llvm/Instructions.h"
#include "llvm/Intrinsics.h"
+#include "llvm/LLVMContext.h"
#include "llvm/ADT/BitVector.h"
#include "llvm/ADT/VectorExtras.h"
#include "llvm/CodeGen/MachineFrameInfo.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 SDValue getMOVL(SelectionDAG &DAG, DebugLoc dl, MVT VT, SDValue V1,
SDValue V2);
+static TargetLoweringObjectFile *createTLOF(X86TargetMachine &TM) {
+ switch (TM.getSubtarget<X86Subtarget>().TargetType) {
+ default: llvm_unreachable("unknown subtarget type");
+ case X86Subtarget::isDarwin:
+ return new TargetLoweringObjectFileMachO();
+ case X86Subtarget::isELF:
+ return new TargetLoweringObjectFileELF();
+ case X86Subtarget::isMingw:
+ case X86Subtarget::isCygwin:
+ case X86Subtarget::isWindows:
+ return new TargetLoweringObjectFileCOFF();
+ }
+
+}
+
X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
- : TargetLowering(TM) {
+ : TargetLowering(TM, createTLOF(TM)) {
Subtarget = &TM.getSubtarget<X86Subtarget>();
X86ScalarSSEf64 = Subtarget->hasSSE2();
X86ScalarSSEf32 = Subtarget->hasSSE1();
setShiftAmountType(MVT::i8);
setBooleanContents(ZeroOrOneBooleanContent);
setSchedulingPreference(SchedulingForRegPressure);
- setShiftAmountFlavor(Mask); // shl X, 32 == shl X, 0
setStackPointerRegisterToSaveRestore(X86StackPtr);
if (Subtarget->isTargetDarwin()) {
setOperationAction(ISD::SELECT, MVT::v4i16, Promote);
setOperationAction(ISD::SELECT, MVT::v2i32, Promote);
setOperationAction(ISD::SELECT, MVT::v1i64, Custom);
+ setOperationAction(ISD::VSETCC, MVT::v8i8, Custom);
+ setOperationAction(ISD::VSETCC, MVT::v4i16, Custom);
+ setOperationAction(ISD::VSETCC, MVT::v2i32, Custom);
}
if (!UseSoftFloat && Subtarget->hasSSE1()) {
SelectionDAG &DAG) const {
if (usesGlobalOffsetTable())
return DAG.getGLOBAL_OFFSET_TABLE(getPointerTy());
- if (!Subtarget->isPICStyleRIPRel())
+ if (!Subtarget->is64Bit())
// This doesn't have DebugLoc associated with it, but is not really the
// same as a Register.
return DAG.getNode(X86ISD::GlobalBaseReg, DebugLoc::getUnknownLoc(),
SmallVector<CCValAssign, 16> RVLocs;
unsigned CC = DAG.getMachineFunction().getFunction()->getCallingConv();
bool isVarArg = DAG.getMachineFunction().getFunction()->isVarArg();
- CCState CCInfo(CC, isVarArg, getTargetMachine(), RVLocs);
+ CCState CCInfo(CC, isVarArg, getTargetMachine(), RVLocs, *DAG.getContext());
CCInfo.AnalyzeReturn(Op.getNode(), RetCC_X86);
// If this is the first return lowered for this function, add the regs to the
SmallVector<CCValAssign, 16> RVLocs;
bool isVarArg = TheCall->isVarArg();
bool Is64Bit = Subtarget->is64Bit();
- CCState CCInfo(CallingConv, isVarArg, getTargetMachine(), RVLocs);
+ CCState CCInfo(CallingConv, isVarArg, getTargetMachine(),
+ RVLocs, *DAG.getContext());
CCInfo.AnalyzeCallResult(TheCall, RetCC_X86);
SmallVector<SDValue, 8> ResultVals;
MVT::v2i64, InFlag).getValue(1);
Val = Chain.getValue(0);
Val = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i64,
- Val, DAG.getConstant(0, MVT::i64));
+ Val, DAG.getConstant(0, MVT::i64));
} else {
Chain = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(),
MVT::i64, InFlag).getValue(1);
return TheCall->getArgFlags(0).isSRet();
}
-/// ArgsAreStructReturn - Determines whether a FORMAL_ARGUMENTS node uses struct
+/// ArgsAreStructReturn - Determines whether a function uses struct
/// return semantics.
static bool ArgsAreStructReturn(SDValue Op) {
unsigned NumArgs = Op.getNode()->getNumValues() - 1;
return cast<ARG_FLAGSSDNode>(Op.getOperand(3))->getArgFlags().isSRet();
}
-/// IsCalleePop - Determines whether a CALL or FORMAL_ARGUMENTS node requires
-/// the callee to pop its own arguments. Callee pop is necessary to support tail
-/// calls.
+/// 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) {
if (IsVarArg)
return false;
}
-/// CallRequiresGOTInRegister - Check whether the call requires the GOT pointer
-/// in a register before calling.
-bool X86TargetLowering::CallRequiresGOTPtrInReg(bool Is64Bit, bool IsTailCall) {
- return !IsTailCall && !Is64Bit &&
- getTargetMachine().getRelocationModel() == Reloc::PIC_ &&
- Subtarget->isPICStyleGOT();
-}
-
-/// CallRequiresFnAddressInReg - Check whether the call requires the function
-/// address to be loaded in a register.
-bool
-X86TargetLowering::CallRequiresFnAddressInReg(bool Is64Bit, bool IsTailCall) {
- return !Is64Bit && IsTailCall &&
- getTargetMachine().getRelocationModel() == Reloc::PIC_ &&
- Subtarget->isPICStyleGOT();
-}
-
/// CreateCopyOfByValArgument - Make a copy of an aggregate at address specified
/// by "Src" to address "Dst" with size and alignment information specified by
/// the specific parameter attribute. The copy will be passed as a byval
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
- CCState CCInfo(CC, isVarArg, getTargetMachine(), ArgLocs);
+ CCState CCInfo(CC, isVarArg, getTargetMachine(), ArgLocs, *DAG.getContext());
CCInfo.AnalyzeFormalArguments(Op.getNode(), CCAssignFnForNode(CC));
SmallVector<SDValue, 8> ArgValues;
unsigned LastVal = ~0U;
+ SDValue ArgValue;
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
// TODO: If an arg is passed in two places (e.g. reg and stack), skip later
RC = X86::FR64RegisterClass;
else if (RegVT.isVector() && RegVT.getSizeInBits() == 128)
RC = X86::VR128RegisterClass;
- else if (RegVT.isVector()) {
- assert(RegVT.getSizeInBits() == 64);
- if (!Is64Bit)
- RC = X86::VR64RegisterClass; // MMX values are passed in MMXs.
- else {
- // Darwin calling convention passes MMX values in either GPRs or
- // XMMs in x86-64. Other targets pass them in memory.
- if (RegVT != MVT::v1i64 && Subtarget->hasSSE2()) {
- RC = X86::VR128RegisterClass; // MMX values are passed in XMMs.
- RegVT = MVT::v2i64;
- } else {
- RC = X86::GR64RegisterClass; // v1i64 values are passed in GPRs.
- RegVT = MVT::i64;
- }
- }
- } else {
- assert(0 && "Unknown argument type!");
- }
+ else if (RegVT.isVector() && RegVT.getSizeInBits() == 64)
+ RC = X86::VR64RegisterClass;
+ else
+ llvm_unreachable("Unknown argument type!");
- unsigned Reg = DAG.getMachineFunction().addLiveIn(VA.getLocReg(), RC);
- SDValue ArgValue = DAG.getCopyFromReg(Root, dl, Reg, RegVT);
+ unsigned Reg = MF.addLiveIn(VA.getLocReg(), RC);
+ ArgValue = DAG.getCopyFromReg(Root, dl, Reg, RegVT);
// If this is an 8 or 16-bit value, it is really passed promoted to 32
// bits. Insert an assert[sz]ext to capture this, then truncate to the
else if (VA.getLocInfo() == CCValAssign::ZExt)
ArgValue = DAG.getNode(ISD::AssertZext, dl, RegVT, ArgValue,
DAG.getValueType(VA.getValVT()));
+ else if (VA.getLocInfo() == CCValAssign::BCvt)
+ ArgValue = DAG.getNode(ISD::BIT_CONVERT, dl, VA.getValVT(), ArgValue);
- if (VA.getLocInfo() != CCValAssign::Full)
- ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
-
- // Handle MMX values passed in GPRs.
- if (Is64Bit && RegVT != VA.getLocVT()) {
- if (RegVT.getSizeInBits() == 64 && RC == X86::GR64RegisterClass)
- ArgValue = DAG.getNode(ISD::BIT_CONVERT, dl, VA.getLocVT(), ArgValue);
- else if (RC == X86::VR128RegisterClass) {
+ if (VA.isExtInLoc()) {
+ // Handle MMX values passed in XMM regs.
+ if (RegVT.isVector()) {
ArgValue = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i64,
ArgValue, DAG.getConstant(0, MVT::i64));
- ArgValue = DAG.getNode(ISD::BIT_CONVERT, dl, VA.getLocVT(), ArgValue);
- }
+ ArgValue = DAG.getNode(ISD::BIT_CONVERT, dl, VA.getValVT(), ArgValue);
+ } else
+ ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
}
-
- ArgValues.push_back(ArgValue);
} else {
assert(VA.isMemLoc());
- ArgValues.push_back(LowerMemArgument(Op, DAG, VA, MFI, CC, Root, i));
+ ArgValue = LowerMemArgument(Op, DAG, VA, MFI, CC, Root, i);
}
+
+ // If value is passed via pointer - do a load.
+ if (VA.getLocInfo() == CCValAssign::Indirect)
+ ArgValue = DAG.getLoad(VA.getValVT(), dl, Root, ArgValue, NULL, 0);
+
+ ArgValues.push_back(ArgValue);
}
// The x86-64 ABI for returning structs by value requires that we copy
// the sret argument into %rax for the return. Save the argument into
// a virtual register so that we can access it from the return points.
- if (Is64Bit && DAG.getMachineFunction().getFunction()->hasStructRetAttr()) {
- MachineFunction &MF = DAG.getMachineFunction();
+ if (Is64Bit && MF.getFunction()->hasStructRetAttr()) {
X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();
unsigned Reg = FuncInfo->getSRetReturnReg();
if (!Reg) {
const CCValAssign &VA,
SDValue Chain,
SDValue Arg, ISD::ArgFlagsTy Flags) {
+ const unsigned FirstStackArgOffset = (Subtarget->isTargetWin64() ? 32 : 0);
DebugLoc dl = TheCall->getDebugLoc();
- unsigned LocMemOffset = VA.getLocMemOffset();
+ unsigned LocMemOffset = FirstStackArgOffset + VA.getLocMemOffset();
SDValue PtrOff = DAG.getIntPtrConstant(LocMemOffset);
PtrOff = DAG.getNode(ISD::ADD, dl, getPointerTy(), StackPtr, PtrOff);
if (Flags.isByVal()) {
// Analyze operands of the call, assigning locations to each operand.
SmallVector<CCValAssign, 16> ArgLocs;
- CCState CCInfo(CC, isVarArg, getTargetMachine(), ArgLocs);
+ CCState CCInfo(CC, isVarArg, getTargetMachine(), ArgLocs, *DAG.getContext());
CCInfo.AnalyzeCallOperands(TheCall, CCAssignFnForNode(CC));
// Get a count of how many bytes are to be pushed on the stack.
// of tail call optimization arguments are handle later.
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
+ MVT RegVT = VA.getLocVT();
SDValue Arg = TheCall->getArg(i);
ISD::ArgFlagsTy Flags = TheCall->getArgFlags(i);
bool isByVal = Flags.isByVal();
// Promote the value if needed.
switch (VA.getLocInfo()) {
- default: assert(0 && "Unknown loc info!");
+ default: llvm_unreachable("Unknown loc info!");
case CCValAssign::Full: break;
case CCValAssign::SExt:
- Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, VA.getLocVT(), Arg);
+ Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, RegVT, Arg);
break;
case CCValAssign::ZExt:
- Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Arg);
+ Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, RegVT, Arg);
break;
case CCValAssign::AExt:
- Arg = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), Arg);
+ if (RegVT.isVector() && RegVT.getSizeInBits() == 128) {
+ // Special case: passing MMX values in XMM registers.
+ Arg = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i64, Arg);
+ Arg = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2i64, Arg);
+ Arg = getMOVL(DAG, dl, MVT::v2i64, DAG.getUNDEF(MVT::v2i64), Arg);
+ } else
+ Arg = DAG.getNode(ISD::ANY_EXTEND, dl, RegVT, Arg);
+ break;
+ case CCValAssign::BCvt:
+ Arg = DAG.getNode(ISD::BIT_CONVERT, dl, RegVT, Arg);
+ break;
+ case CCValAssign::Indirect: {
+ // Store the argument.
+ SDValue SpillSlot = DAG.CreateStackTemporary(VA.getValVT());
+ int FI = cast<FrameIndexSDNode>(SpillSlot)->getIndex();
+ Chain = DAG.getStore(Chain, dl, Arg, SpillSlot,
+ PseudoSourceValue::getFixedStack(FI), 0);
+ Arg = SpillSlot;
break;
}
+ }
if (VA.isRegLoc()) {
- if (Is64Bit) {
- MVT RegVT = VA.getLocVT();
- if (RegVT.isVector() && RegVT.getSizeInBits() == 64)
- switch (VA.getLocReg()) {
- default:
- break;
- case X86::RDI: case X86::RSI: case X86::RDX: case X86::RCX:
- case X86::R8: {
- // Special case: passing MMX values in GPR registers.
- Arg = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i64, Arg);
- break;
- }
- case X86::XMM0: case X86::XMM1: case X86::XMM2: case X86::XMM3:
- case X86::XMM4: case X86::XMM5: case X86::XMM6: case X86::XMM7: {
- // Special case: passing MMX values in XMM registers.
- Arg = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i64, Arg);
- Arg = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2i64, Arg);
- Arg = getMOVL(DAG, dl, MVT::v2i64, DAG.getUNDEF(MVT::v2i64), Arg);
- break;
- }
- }
- }
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
} else {
if (!IsTailCall || (IsTailCall && isByVal)) {
InFlag = Chain.getValue(1);
}
- // ELF / PIC requires GOT in the EBX register before function calls via PLT
- // GOT pointer.
- if (CallRequiresGOTPtrInReg(Is64Bit, IsTailCall)) {
- Chain = DAG.getCopyToReg(Chain, dl, X86::EBX,
- DAG.getNode(X86ISD::GlobalBaseReg,
- DebugLoc::getUnknownLoc(),
- getPointerTy()),
- InFlag);
- InFlag = Chain.getValue(1);
- }
- // If we are tail calling and generating PIC/GOT style code load the address
- // of the callee into ecx. The value in ecx is used as target of the tail
- // jump. This is done to circumvent the ebx/callee-saved problem for tail
- // calls on PIC/GOT architectures. Normally we would just put the address of
- // GOT into ebx and then call target@PLT. But for tail callss ebx would be
- // restored (since ebx is callee saved) before jumping to the target@PLT.
- if (CallRequiresFnAddressInReg(Is64Bit, IsTailCall)) {
- // Note: The actual moving to ecx is done further down.
- GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee);
- if (G && !G->getGlobal()->hasHiddenVisibility() &&
- !G->getGlobal()->hasProtectedVisibility())
- Callee = LowerGlobalAddress(Callee, DAG);
- else if (isa<ExternalSymbolSDNode>(Callee))
- Callee = LowerExternalSymbol(Callee,DAG);
+
+ if (Subtarget->isPICStyleGOT()) {
+ // ELF / PIC requires GOT in the EBX register before function calls via PLT
+ // GOT pointer.
+ if (!IsTailCall) {
+ Chain = DAG.getCopyToReg(Chain, dl, X86::EBX,
+ DAG.getNode(X86ISD::GlobalBaseReg,
+ DebugLoc::getUnknownLoc(),
+ getPointerTy()),
+ InFlag);
+ InFlag = Chain.getValue(1);
+ } else {
+ // If we are tail calling and generating PIC/GOT style code load the
+ // address of the callee into ECX. The value in ecx is used as target of
+ // the tail jump. This is done to circumvent the ebx/callee-saved problem
+ // for tail calls on PIC/GOT architectures. Normally we would just put the
+ // address of GOT into ebx and then call target@PLT. But for tail calls
+ // ebx would be restored (since ebx is callee saved) before jumping to the
+ // target@PLT.
+
+ // Note: The actual moving to ECX is done further down.
+ GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee);
+ if (G && !G->getGlobal()->hasHiddenVisibility() &&
+ !G->getGlobal()->hasProtectedVisibility())
+ Callee = LowerGlobalAddress(Callee, DAG);
+ else if (isa<ExternalSymbolSDNode>(Callee))
+ Callee = LowerExternalSymbol(Callee, DAG);
+ }
}
if (Is64Bit && isVarArg) {
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
// We should use extra load for direct calls to dllimported functions in
// non-JIT mode.
- if (!Subtarget->GVRequiresExtraLoad(G->getGlobal(),
- getTargetMachine(), true))
- Callee = DAG.getTargetGlobalAddress(G->getGlobal(), getPointerTy(),
- G->getOffset());
+ GlobalValue *GV = G->getGlobal();
+ if (!GV->hasDLLImportLinkage()) {
+ unsigned char OpFlags = 0;
+
+ // On ELF targets, in both X86-64 and X86-32 mode, direct calls to
+ // external symbols most go through the PLT in PIC mode. If the symbol
+ // has hidden or protected visibility, or if it is static or local, then
+ // we don't need to use the PLT - we can directly call it.
+ if (Subtarget->isTargetELF() &&
+ getTargetMachine().getRelocationModel() == Reloc::PIC_ &&
+ GV->hasDefaultVisibility() && !GV->hasLocalLinkage()) {
+ OpFlags = X86II::MO_PLT;
+ } else if (Subtarget->isPICStyleStubAny() &&
+ (GV->isDeclaration() || GV->isWeakForLinker()) &&
+ Subtarget->getDarwinVers() < 9) {
+ // PC-relative references to external symbols should go through $stub,
+ // unless we're building with the leopard linker or later, which
+ // automatically synthesizes these stubs.
+ OpFlags = X86II::MO_DARWIN_STUB;
+ }
+
+ Callee = DAG.getTargetGlobalAddress(GV, getPointerTy(),
+ G->getOffset(), OpFlags);
+ }
} else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
- Callee = DAG.getTargetExternalSymbol(S->getSymbol(), getPointerTy());
+ unsigned char OpFlags = 0;
+
+ // On ELF targets, in either X86-64 or X86-32 mode, direct calls to external
+ // symbols should go through the PLT.
+ if (Subtarget->isTargetELF() &&
+ getTargetMachine().getRelocationModel() == Reloc::PIC_) {
+ OpFlags = X86II::MO_PLT;
+ } else if (Subtarget->isPICStyleStubAny() &&
+ Subtarget->getDarwinVers() < 9) {
+ // PC-relative references to external symbols should go through $stub,
+ // unless we're building with the leopard linker or later, which
+ // automatically synthesizes these stubs.
+ OpFlags = X86II::MO_DARWIN_STUB;
+ }
+
+ Callee = DAG.getTargetExternalSymbol(S->getSymbol(), getPointerTy(),
+ OpFlags);
} else if (IsTailCall) {
unsigned Opc = Is64Bit ? X86::R11 : X86::EAX;
Callee,InFlag);
Callee = DAG.getRegister(Opc, getPointerTy());
// Add register as live out.
- DAG.getMachineFunction().getRegInfo().addLiveOut(Opc);
+ MF.getRegInfo().addLiveOut(Opc);
}
// Returns a chain & a flag for retval copy to use.
RegsToPass[i].second.getValueType()));
// Add an implicit use GOT pointer in EBX.
- if (!IsTailCall && !Is64Bit &&
- getTargetMachine().getRelocationModel() == Reloc::PIC_ &&
- Subtarget->isPICStyleGOT())
+ if (!IsTailCall && Subtarget->isPICStyleGOT())
Ops.push_back(DAG.getRegister(X86::EBX, getPointerTy()));
// Add an implicit use of AL for x86 vararg functions.
return false;
if (CheckTailCallReturnConstraints(TheCall, Ret)) {
- MachineFunction &MF = DAG.getMachineFunction();
- unsigned CallerCC = MF.getFunction()->getCallingConv();
- unsigned CalleeCC= TheCall->getCallingConv();
- if (CalleeCC == CallingConv::Fast && CallerCC == CalleeCC) {
- SDValue Callee = TheCall->getCallee();
- // On x86/32Bit PIC/GOT tail calls are supported.
- if (getTargetMachine().getRelocationModel() != Reloc::PIC_ ||
- !Subtarget->isPICStyleGOT()|| !Subtarget->is64Bit())
- return true;
-
- // Can only do local tail calls (in same module, hidden or protected) on
- // x86_64 PIC/GOT at the moment.
- if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee))
- return G->getGlobal()->hasHiddenVisibility()
- || G->getGlobal()->hasProtectedVisibility();
- }
+ unsigned CallerCC =
+ DAG.getMachineFunction().getFunction()->getCallingConv();
+ unsigned CalleeCC = TheCall->getCallingConv();
+ if (CalleeCC == CallingConv::Fast && CallerCC == CalleeCC)
+ return true;
}
return false;
}
switch (SetCCOpcode) {
- default: assert(0 && "Invalid integer condition!");
+ default: llvm_unreachable("Invalid integer condition!");
case ISD::SETEQ: return X86::COND_E;
case ISD::SETGT: return X86::COND_G;
case ISD::SETGE: return X86::COND_GE;
// 1 | 0 | 0 | X == Y
// 1 | 1 | 1 | unordered
switch (SetCCOpcode) {
- default: assert(0 && "Condcode should be pre-legalized away");
+ default: llvm_unreachable("Condcode should be pre-legalized away");
case ISD::SETUEQ:
case ISD::SETEQ: return X86::COND_E;
case ISD::SETOLT: // flipped
return Mask;
}
+/// isZeroNode - Returns true if Elt is a constant zero or a floating point
+/// constant +0.0.
+bool X86::isZeroNode(SDValue Elt) {
+ return ((isa<ConstantSDNode>(Elt) &&
+ cast<ConstantSDNode>(Elt)->getZExtValue() == 0) ||
+ (isa<ConstantFPSDNode>(Elt) &&
+ cast<ConstantFPSDNode>(Elt)->getValueAPF().isPosZero()));
+}
+
/// CommuteVectorShuffle - Swap vector_shuffle operands as well as values in
/// their permute mask.
static SDValue CommuteVectorShuffle(ShuffleVectorSDNode *SVOp,
return true;
}
-/// isZeroNode - Returns true if Elt is a constant zero or a floating point
-/// constant +0.0.
-static inline bool isZeroNode(SDValue Elt) {
- return ((isa<ConstantSDNode>(Elt) &&
- cast<ConstantSDNode>(Elt)->getZExtValue() == 0) ||
- (isa<ConstantFPSDNode>(Elt) &&
- cast<ConstantFPSDNode>(Elt)->getValueAPF().isPosZero()));
-}
-
/// isZeroShuffle - Returns true if N is a VECTOR_SHUFFLE that can be resolved
/// to an zero vector.
/// FIXME: move to dag combiner / method on ShuffleVectorSDNode
unsigned Opc = V2.getOpcode();
if (Opc == ISD::UNDEF || ISD::isBuildVectorAllZeros(V2.getNode()))
continue;
- if (Opc != ISD::BUILD_VECTOR || !isZeroNode(V2.getOperand(Idx-NumElems)))
+ if (Opc != ISD::BUILD_VECTOR ||
+ !X86::isZeroNode(V2.getOperand(Idx-NumElems)))
return false;
} else if (Idx >= 0) {
unsigned Opc = V1.getOpcode();
if (Opc == ISD::UNDEF || ISD::isBuildVectorAllZeros(V1.getNode()))
continue;
- if (Opc != ISD::BUILD_VECTOR || !isZeroNode(V1.getOperand(Idx)))
+ if (Opc != ISD::BUILD_VECTOR ||
+ !X86::isZeroNode(V1.getOperand(Idx)))
return false;
}
}
continue;
}
SDValue Elt = DAG.getShuffleScalarElt(SVOp, Index);
- if (Elt.getNode() && isZeroNode(Elt))
+ if (Elt.getNode() && X86::isZeroNode(Elt))
++NumZeros;
else
break;
if (Elt.getOpcode() != ISD::Constant &&
Elt.getOpcode() != ISD::ConstantFP)
IsAllConstants = false;
- if (isZeroNode(Elt))
+ if (X86::isZeroNode(Elt))
NumZero++;
else {
NonZeros |= (1 << i);
// Is it a vector logical left shift?
if (NumElems == 2 && Idx == 1 &&
- isZeroNode(Op.getOperand(0)) && !isZeroNode(Op.getOperand(1))) {
+ X86::isZeroNode(Op.getOperand(0)) &&
+ !X86::isZeroNode(Op.getOperand(1))) {
unsigned NumBits = VT.getSizeInBits();
return getVShift(true, VT,
DAG.getNode(ISD::SCALAR_TO_VECTOR, dl,
// Bits [3:0] of the constant are the zero mask. The DAG Combiner may
// combine either bitwise AND or insert of float 0.0 to set these bits.
N2 = DAG.getIntPtrConstant(cast<ConstantSDNode>(N2)->getZExtValue() << 4);
+ // Create this as a scalar to vector..
+ N1 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v4f32, N1);
return DAG.getNode(X86ISD::INSERTPS, dl, VT, N0, N1, N2);
- } else if (EVT == MVT::i32) {
- // InsertPS works with constant index.
- if (isa<ConstantSDNode>(N2))
- return Op;
+ } else if (EVT == MVT::i32 && isa<ConstantSDNode>(N2)) {
+ // PINSR* works with constant index.
+ return Op;
}
return SDValue();
}
DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32,
Op.getOperand(0))));
+ if (Op.getValueType() == MVT::v1i64 && Op.getOperand(0).getValueType() == MVT::i64)
+ return DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v1i64, Op.getOperand(0));
+
SDValue AnyExt = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i32, Op.getOperand(0));
MVT VT = MVT::v2i32;
switch (Op.getValueType().getSimpleVT()) {
// global base reg.
unsigned char OpFlag = 0;
unsigned WrapperKind = X86ISD::Wrapper;
- if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
- if (Subtarget->isPICStyleStub())
- OpFlag = X86II::MO_PIC_BASE_OFFSET;
- else if (Subtarget->isPICStyleGOT())
- OpFlag = X86II::MO_GOTOFF;
- else if (Subtarget->isPICStyleRIPRel() &&
- getTargetMachine().getCodeModel() == CodeModel::Small)
- WrapperKind = X86ISD::WrapperRIP;
- }
+
+ if (Subtarget->isPICStyleRIPRel() &&
+ getTargetMachine().getCodeModel() == CodeModel::Small)
+ WrapperKind = X86ISD::WrapperRIP;
+ else if (Subtarget->isPICStyleGOT())
+ OpFlag = X86II::MO_GOTOFF;
+ else if (Subtarget->isPICStyleStubPIC())
+ OpFlag = X86II::MO_PIC_BASE_OFFSET;
SDValue Result = DAG.getTargetConstantPool(CP->getConstVal(), getPointerTy(),
CP->getAlignment(),
// global base reg.
unsigned char OpFlag = 0;
unsigned WrapperKind = X86ISD::Wrapper;
- if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
- if (Subtarget->isPICStyleStub())
- OpFlag = X86II::MO_PIC_BASE_OFFSET;
- else if (Subtarget->isPICStyleGOT())
- OpFlag = X86II::MO_GOTOFF;
- else if (Subtarget->isPICStyleRIPRel())
- WrapperKind = X86ISD::WrapperRIP;
- }
+
+ if (Subtarget->isPICStyleRIPRel() &&
+ getTargetMachine().getCodeModel() == CodeModel::Small)
+ WrapperKind = X86ISD::WrapperRIP;
+ else if (Subtarget->isPICStyleGOT())
+ OpFlag = X86II::MO_GOTOFF;
+ else if (Subtarget->isPICStyleStubPIC())
+ OpFlag = X86II::MO_PIC_BASE_OFFSET;
SDValue Result = DAG.getTargetJumpTable(JT->getIndex(), getPointerTy(),
OpFlag);
// global base reg.
unsigned char OpFlag = 0;
unsigned WrapperKind = X86ISD::Wrapper;
- if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
- if (Subtarget->isPICStyleStub())
- OpFlag = X86II::MO_PIC_BASE_OFFSET;
- else if (Subtarget->isPICStyleGOT())
- OpFlag = X86II::MO_GOTOFF;
- else if (Subtarget->isPICStyleRIPRel())
- WrapperKind = X86ISD::WrapperRIP;
- }
+ if (Subtarget->isPICStyleRIPRel() &&
+ getTargetMachine().getCodeModel() == CodeModel::Small)
+ WrapperKind = X86ISD::WrapperRIP;
+ else if (Subtarget->isPICStyleGOT())
+ OpFlag = X86II::MO_GOTOFF;
+ else if (Subtarget->isPICStyleStubPIC())
+ OpFlag = X86II::MO_PIC_BASE_OFFSET;
SDValue Result = DAG.getTargetExternalSymbol(Sym, getPointerTy(), OpFlag);
// With PIC, the address is actually $g + Offset.
if (getTargetMachine().getRelocationModel() == Reloc::PIC_ &&
- !Subtarget->isPICStyleRIPRel()) {
+ !Subtarget->is64Bit()) {
Result = DAG.getNode(ISD::ADD, DL, getPointerTy(),
DAG.getNode(X86ISD::GlobalBaseReg,
DebugLoc::getUnknownLoc(),
X86TargetLowering::LowerGlobalAddress(const GlobalValue *GV, DebugLoc dl,
int64_t Offset,
SelectionDAG &DAG) const {
- bool IsPic = getTargetMachine().getRelocationModel() == Reloc::PIC_;
- bool ExtraLoadRequired =
- Subtarget->GVRequiresExtraLoad(GV, getTargetMachine(), false);
-
// Create the TargetGlobalAddress node, folding in the constant
// offset if it is legal.
+ unsigned char OpFlags =
+ Subtarget->ClassifyGlobalReference(GV, getTargetMachine());
SDValue Result;
- if (!IsPic && !ExtraLoadRequired && isInt32(Offset)) {
+ if (OpFlags == X86II::MO_NO_FLAG && isInt32(Offset)) {
+ // A direct static reference to a global.
Result = DAG.getTargetGlobalAddress(GV, getPointerTy(), Offset);
Offset = 0;
} else {
- unsigned char OpFlags = 0;
-
- if (Subtarget->isPICStyleRIPRel() &&
- getTargetMachine().getRelocationModel() != Reloc::Static) {
- if (ExtraLoadRequired)
- OpFlags = X86II::MO_GOTPCREL;
- } else if (Subtarget->isPICStyleGOT() &&
- getTargetMachine().getRelocationModel() == Reloc::PIC_) {
- if (ExtraLoadRequired)
- OpFlags = X86II::MO_GOT;
- else
- OpFlags = X86II::MO_GOTOFF;
- }
-
Result = DAG.getTargetGlobalAddress(GV, getPointerTy(), 0, OpFlags);
}
Result = DAG.getNode(X86ISD::Wrapper, dl, getPointerTy(), Result);
// With PIC, the address is actually $g + Offset.
- if (IsPic && !Subtarget->isPICStyleRIPRel()) {
+ if (isGlobalRelativeToPICBase(OpFlags)) {
Result = DAG.getNode(ISD::ADD, dl, getPointerTy(),
DAG.getNode(X86ISD::GlobalBaseReg, dl, getPointerTy()),
Result);
}
- // For Darwin & Mingw32, external and weak symbols are indirect, so we want to
- // load the value at address GV, not the value of GV itself. This means that
- // the GlobalAddress must be in the base or index register of the address, not
- // the GV offset field. Platform check is inside GVRequiresExtraLoad() call
- // The same applies for external symbols during PIC codegen
- if (ExtraLoadRequired)
+ // For globals that require a load from a stub to get the address, emit the
+ // load.
+ if (isGlobalStubReference(OpFlags))
Result = DAG.getLoad(getPointerTy(), dl, DAG.getEntryNode(), Result,
PseudoSourceValue::getGOT(), 0);
Subtarget->is64Bit());
}
- assert(0 && "Unreachable");
+ llvm_unreachable("Unreachable");
return SDValue();
}
SDValue ShOpLo = Op.getOperand(0);
SDValue ShOpHi = Op.getOperand(1);
SDValue ShAmt = Op.getOperand(2);
- SDValue Tmp1 = isSRA ?
- DAG.getNode(ISD::SRA, dl, VT, ShOpHi,
- DAG.getConstant(VTBits - 1, MVT::i8)) :
- DAG.getConstant(0, VT);
+ SDValue Tmp1 = isSRA ? DAG.getNode(ISD::SRA, dl, VT, ShOpHi,
+ DAG.getConstant(VTBits - 1, MVT::i8))
+ : DAG.getConstant(0, VT);
SDValue Tmp2, Tmp3;
if (Op.getOpcode() == ISD::SHL_PARTS) {
}
SDValue AndNode = DAG.getNode(ISD::AND, dl, MVT::i8, ShAmt,
- DAG.getConstant(VTBits, MVT::i8));
+ DAG.getConstant(VTBits, MVT::i8));
SDValue Cond = DAG.getNode(X86ISD::CMP, dl, VT,
- AndNode, DAG.getConstant(0, MVT::i8));
+ AndNode, DAG.getConstant(0, MVT::i8));
SDValue Hi, Lo;
SDValue CC = DAG.getConstant(X86::COND_NE, MVT::i8);
*/
DebugLoc dl = Op.getDebugLoc();
+ LLVMContext *Context = DAG.getContext();
// Build some magic constants.
std::vector<Constant*> CV0;
- CV0.push_back(ConstantInt::get(APInt(32, 0x45300000)));
- CV0.push_back(ConstantInt::get(APInt(32, 0x43300000)));
- CV0.push_back(ConstantInt::get(APInt(32, 0)));
- CV0.push_back(ConstantInt::get(APInt(32, 0)));
+ CV0.push_back(ConstantInt::get(*Context, APInt(32, 0x45300000)));
+ CV0.push_back(ConstantInt::get(*Context, APInt(32, 0x43300000)));
+ CV0.push_back(ConstantInt::get(*Context, APInt(32, 0)));
+ CV0.push_back(ConstantInt::get(*Context, APInt(32, 0)));
Constant *C0 = ConstantVector::get(CV0);
SDValue CPIdx0 = DAG.getConstantPool(C0, getPointerTy(), 16);
std::vector<Constant*> CV1;
- CV1.push_back(ConstantFP::get(APFloat(APInt(64, 0x4530000000000000ULL))));
- CV1.push_back(ConstantFP::get(APFloat(APInt(64, 0x4330000000000000ULL))));
+ CV1.push_back(
+ ConstantFP::get(*Context, APFloat(APInt(64, 0x4530000000000000ULL))));
+ CV1.push_back(
+ ConstantFP::get(*Context, APFloat(APInt(64, 0x4330000000000000ULL))));
Constant *C1 = ConstantVector::get(CV1);
SDValue CPIdx1 = DAG.getConstantPool(C1, getPointerTy(), 16);
unsigned Opc;
switch (DstTy.getSimpleVT()) {
- default: assert(0 && "Invalid FP_TO_SINT to lower!");
+ default: llvm_unreachable("Invalid FP_TO_SINT to lower!");
case MVT::i16: Opc = X86ISD::FP_TO_INT16_IN_MEM; break;
case MVT::i32: Opc = X86ISD::FP_TO_INT32_IN_MEM; break;
case MVT::i64: Opc = X86ISD::FP_TO_INT64_IN_MEM; break;
}
SDValue X86TargetLowering::LowerFABS(SDValue Op, SelectionDAG &DAG) {
+ LLVMContext *Context = DAG.getContext();
DebugLoc dl = Op.getDebugLoc();
MVT VT = Op.getValueType();
MVT EltVT = VT;
EltVT = VT.getVectorElementType();
std::vector<Constant*> CV;
if (EltVT == MVT::f64) {
- Constant *C = ConstantFP::get(APFloat(APInt(64, ~(1ULL << 63))));
+ Constant *C = ConstantFP::get(*Context, APFloat(APInt(64, ~(1ULL << 63))));
CV.push_back(C);
CV.push_back(C);
} else {
- Constant *C = ConstantFP::get(APFloat(APInt(32, ~(1U << 31))));
+ Constant *C = ConstantFP::get(*Context, APFloat(APInt(32, ~(1U << 31))));
CV.push_back(C);
CV.push_back(C);
CV.push_back(C);
}
SDValue X86TargetLowering::LowerFNEG(SDValue Op, SelectionDAG &DAG) {
+ LLVMContext *Context = DAG.getContext();
DebugLoc dl = Op.getDebugLoc();
MVT VT = Op.getValueType();
MVT EltVT = VT;
}
std::vector<Constant*> CV;
if (EltVT == MVT::f64) {
- Constant *C = ConstantFP::get(APFloat(APInt(64, 1ULL << 63)));
+ Constant *C = ConstantFP::get(*Context, APFloat(APInt(64, 1ULL << 63)));
CV.push_back(C);
CV.push_back(C);
} else {
- Constant *C = ConstantFP::get(APFloat(APInt(32, 1U << 31)));
+ Constant *C = ConstantFP::get(*Context, APFloat(APInt(32, 1U << 31)));
CV.push_back(C);
CV.push_back(C);
CV.push_back(C);
}
SDValue X86TargetLowering::LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) {
+ LLVMContext *Context = DAG.getContext();
SDValue Op0 = Op.getOperand(0);
SDValue Op1 = Op.getOperand(1);
DebugLoc dl = Op.getDebugLoc();
// First get the sign bit of second operand.
std::vector<Constant*> CV;
if (SrcVT == MVT::f64) {
- CV.push_back(ConstantFP::get(APFloat(APInt(64, 1ULL << 63))));
- CV.push_back(ConstantFP::get(APFloat(APInt(64, 0))));
+ CV.push_back(ConstantFP::get(*Context, APFloat(APInt(64, 1ULL << 63))));
+ CV.push_back(ConstantFP::get(*Context, APFloat(APInt(64, 0))));
} else {
- CV.push_back(ConstantFP::get(APFloat(APInt(32, 1U << 31))));
- CV.push_back(ConstantFP::get(APFloat(APInt(32, 0))));
- CV.push_back(ConstantFP::get(APFloat(APInt(32, 0))));
- CV.push_back(ConstantFP::get(APFloat(APInt(32, 0))));
+ CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, 1U << 31))));
+ CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, 0))));
+ CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, 0))));
+ CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, 0))));
}
Constant *C = ConstantVector::get(CV);
SDValue CPIdx = DAG.getConstantPool(C, getPointerTy(), 16);
// Clear first operand sign bit.
CV.clear();
if (VT == MVT::f64) {
- CV.push_back(ConstantFP::get(APFloat(APInt(64, ~(1ULL << 63)))));
- CV.push_back(ConstantFP::get(APFloat(APInt(64, 0))));
+ CV.push_back(ConstantFP::get(*Context, APFloat(APInt(64, ~(1ULL << 63)))));
+ CV.push_back(ConstantFP::get(*Context, APFloat(APInt(64, 0))));
} else {
- CV.push_back(ConstantFP::get(APFloat(APInt(32, ~(1U << 31)))));
- CV.push_back(ConstantFP::get(APFloat(APInt(32, 0))));
- CV.push_back(ConstantFP::get(APFloat(APInt(32, 0))));
- CV.push_back(ConstantFP::get(APFloat(APInt(32, 0))));
+ CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, ~(1U << 31)))));
+ CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, 0))));
+ CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, 0))));
+ CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, 0))));
}
C = ConstantVector::get(CV);
CPIdx = DAG.getConstantPool(C, getPointerTy(), 16);
NEQ = DAG.getNode(Opc, dl, VT, Op0, Op1, DAG.getConstant(4, MVT::i8));
return DAG.getNode(ISD::AND, dl, VT, ORD, NEQ);
}
- assert(0 && "Illegal FP comparison");
+ llvm_unreachable("Illegal FP comparison");
}
// Handle all other FP comparisons here.
return DAG.getNode(Opc, dl, VT, Op0, Op1, DAG.getConstant(SSECC, MVT::i8));
switch (VT.getSimpleVT()) {
default: break;
+ case MVT::v8i8:
case MVT::v16i8: EQOpc = X86ISD::PCMPEQB; GTOpc = X86ISD::PCMPGTB; break;
+ case MVT::v4i16:
case MVT::v8i16: EQOpc = X86ISD::PCMPEQW; GTOpc = X86ISD::PCMPGTW; break;
+ case MVT::v2i32:
case MVT::v4i32: EQOpc = X86ISD::PCMPEQD; GTOpc = X86ISD::PCMPGTD; break;
case MVT::v2i64: EQOpc = X86ISD::PCMPEQQ; GTOpc = X86ISD::PCMPGTQ; break;
}
DAG.getConstant(X86CC, MVT::i8), Cond);
return DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, SetCC);
}
+ // ptest intrinsics. The intrinsic these come from are designed to return
+ // an integer value, not just an instruction so lower it to the ptest
+ // pattern and a setcc for the result.
+ case Intrinsic::x86_sse41_ptestz:
+ case Intrinsic::x86_sse41_ptestc:
+ case Intrinsic::x86_sse41_ptestnzc:{
+ unsigned X86CC = 0;
+ switch (IntNo) {
+ default: llvm_unreachable("Bad fallthrough in Intrinsic lowering.");
+ case Intrinsic::x86_sse41_ptestz:
+ // ZF = 1
+ X86CC = X86::COND_E;
+ break;
+ case Intrinsic::x86_sse41_ptestc:
+ // CF = 1
+ X86CC = X86::COND_B;
+ break;
+ case Intrinsic::x86_sse41_ptestnzc:
+ // ZF and CF = 0
+ X86CC = X86::COND_A;
+ break;
+ }
+
+ SDValue LHS = Op.getOperand(1);
+ SDValue RHS = Op.getOperand(2);
+ SDValue Test = DAG.getNode(X86ISD::PTEST, dl, MVT::i32, LHS, RHS);
+ SDValue CC = DAG.getConstant(X86CC, MVT::i8);
+ SDValue SetCC = DAG.getNode(X86ISD::SETCC, dl, MVT::i8, CC, Test);
+ return DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, SetCC);
+ }
// Fix vector shift instructions where the last operand is a non-immediate
// i32 value.
case Intrinsic::x86_mmx_psrai_d:
NewIntNo = Intrinsic::x86_mmx_psra_d;
break;
- default: LLVM_UNREACHABLE("Impossible intrinsic"); // Can't reach here.
+ default: llvm_unreachable("Impossible intrinsic"); // Can't reach here.
}
break;
}
switch (CC) {
default:
- assert(0 && "Unsupported calling convention");
+ llvm_unreachable("Unsupported calling convention");
case CallingConv::C:
case CallingConv::X86_StdCall: {
// Pass 'nest' parameter in ECX.
DebugLoc dl = Op.getDebugLoc();
switch (Op.getOpcode()) {
- default: assert(0 && "Unknown ovf instruction!");
+ default: llvm_unreachable("Unknown ovf instruction!");
case ISD::SADDO:
// A subtract of one will be selected as a INC. Note that INC doesn't
// set CF, so we can't do this for UADDO.
///
SDValue X86TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) {
switch (Op.getOpcode()) {
- default: assert(0 && "Should not custom lower this!");
+ default: llvm_unreachable("Should not custom lower this!");
case ISD::ATOMIC_CMP_SWAP: return LowerCMP_SWAP(Op,DAG);
case ISD::ATOMIC_LOAD_SUB: return LowerLOAD_SUB(Op,DAG);
case ISD::BUILD_VECTOR: return LowerBUILD_VECTOR(Op, DAG);
case X86ISD::INC: return "X86ISD::INC";
case X86ISD::DEC: return "X86ISD::DEC";
case X86ISD::MUL_IMM: return "X86ISD::MUL_IMM";
+ case X86ISD::PTEST: return "X86ISD::PTEST";
}
}
return false;
if (AM.BaseGV) {
- // We can only fold this if we don't need an extra load.
- if (Subtarget->GVRequiresExtraLoad(AM.BaseGV, getTargetMachine(), false))
+ unsigned GVFlags =
+ Subtarget->ClassifyGlobalReference(AM.BaseGV, getTargetMachine());
+
+ // If a reference to this global requires an extra load, we can't fold it.
+ if (isGlobalStubReference(GVFlags))
return false;
- // If BaseGV requires a register, we cannot also have a BaseReg.
- if (Subtarget->GVRequiresRegister(AM.BaseGV, getTargetMachine(), false) &&
- AM.HasBaseReg)
+
+ // If BaseGV requires a register for the PIC base, we cannot also have a
+ // BaseReg specified.
+ if (AM.HasBaseReg && isGlobalRelativeToPICBase(GVFlags))
return false;
// X86-64 only supports addr of globals in small code model.
// Get the X86 opcode to use.
unsigned Opc;
switch (MI->getOpcode()) {
- default: assert(0 && "illegal opcode!");
+ default: llvm_unreachable("illegal opcode!");
case X86::FP32_TO_INT16_IN_MEM: Opc = X86::IST_Fp16m32; break;
case X86::FP32_TO_INT32_IN_MEM: Opc = X86::IST_Fp32m32; break;
case X86::FP32_TO_INT64_IN_MEM: Opc = X86::IST_Fp64m32; break;
SDValue ValOp = N->getOperand(0);
switch (N->getOpcode()) {
default:
- assert(0 && "Unknown shift opcode!");
+ llvm_unreachable("Unknown shift opcode!");
break;
case ISD::SHL:
if (VT == MVT::v2i64)
// X86 Inline Assembly Support
//===----------------------------------------------------------------------===//
+static bool LowerToBSwap(CallInst *CI) {
+ // FIXME: this should verify that we are targetting a 486 or better. If not,
+ // we will turn this bswap into something that will be lowered to logical ops
+ // instead of emitting the bswap asm. For now, we don't support 486 or lower
+ // so don't worry about this.
+
+ // Verify this is a simple bswap.
+ if (CI->getNumOperands() != 2 ||
+ CI->getType() != CI->getOperand(1)->getType() ||
+ !CI->getType()->isInteger())
+ return false;
+
+ const IntegerType *Ty = dyn_cast<IntegerType>(CI->getType());
+ if (!Ty || Ty->getBitWidth() % 16 != 0)
+ return false;
+
+ // Okay, we can do this xform, do so now.
+ const Type *Tys[] = { Ty };
+ Module *M = CI->getParent()->getParent()->getParent();
+ Constant *Int = Intrinsic::getDeclaration(M, Intrinsic::bswap, Tys, 1);
+
+ Value *Op = CI->getOperand(1);
+ Op = CallInst::Create(Int, Op, CI->getName(), CI);
+
+ CI->replaceAllUsesWith(Op);
+ CI->eraseFromParent();
+ return true;
+}
+
+bool X86TargetLowering::ExpandInlineAsm(CallInst *CI) const {
+ InlineAsm *IA = cast<InlineAsm>(CI->getCalledValue());
+ std::vector<InlineAsm::ConstraintInfo> Constraints = IA->ParseConstraints();
+
+ std::string AsmStr = IA->getAsmString();
+
+ // TODO: should remove alternatives from the asmstring: "foo {a|b}" -> "foo a"
+ std::vector<std::string> AsmPieces;
+ SplitString(AsmStr, AsmPieces, "\n"); // ; as separator?
+
+ switch (AsmPieces.size()) {
+ default: return false;
+ case 1:
+ AsmStr = AsmPieces[0];
+ AsmPieces.clear();
+ SplitString(AsmStr, AsmPieces, " \t"); // Split with whitespace.
+
+ // bswap $0
+ if (AsmPieces.size() == 2 &&
+ (AsmPieces[0] == "bswap" ||
+ AsmPieces[0] == "bswapq" ||
+ AsmPieces[0] == "bswapl") &&
+ (AsmPieces[1] == "$0" ||
+ AsmPieces[1] == "${0:q}")) {
+ // No need to check constraints, nothing other than the equivalent of
+ // "=r,0" would be valid here.
+ return LowerToBSwap(CI);
+ }
+ // rorw $$8, ${0:w} --> llvm.bswap.i16
+ if (CI->getType() == Type::Int16Ty &&
+ AsmPieces.size() == 3 &&
+ AsmPieces[0] == "rorw" &&
+ AsmPieces[1] == "$$8," &&
+ AsmPieces[2] == "${0:w}" &&
+ IA->getConstraintString() == "=r,0,~{dirflag},~{fpsr},~{flags},~{cc}") {
+ return LowerToBSwap(CI);
+ }
+ break;
+ case 3:
+ if (CI->getType() == Type::Int64Ty && Constraints.size() >= 2 &&
+ Constraints[0].Codes.size() == 1 && Constraints[0].Codes[0] == "A" &&
+ Constraints[1].Codes.size() == 1 && Constraints[1].Codes[0] == "0") {
+ // bswap %eax / bswap %edx / xchgl %eax, %edx -> llvm.bswap.i64
+ std::vector<std::string> Words;
+ SplitString(AsmPieces[0], Words, " \t");
+ if (Words.size() == 2 && Words[0] == "bswap" && Words[1] == "%eax") {
+ Words.clear();
+ SplitString(AsmPieces[1], Words, " \t");
+ if (Words.size() == 2 && Words[0] == "bswap" && Words[1] == "%edx") {
+ Words.clear();
+ SplitString(AsmPieces[2], Words, " \t,");
+ if (Words.size() == 3 && Words[0] == "xchgl" && Words[1] == "%eax" &&
+ Words[2] == "%edx") {
+ return LowerToBSwap(CI);
+ }
+ }
+ }
+ }
+ break;
+ }
+ return false;
+}
+
+
+
/// getConstraintType - Given a constraint letter, return the type of
/// constraint it is for this target.
X86TargetLowering::ConstraintType
// Otherwise, this isn't something we can handle, reject it.
return;
}
+
+ GlobalValue *GV = GA->getGlobal();
// If we require an extra load to get this address, as in PIC mode, we
// can't accept it.
- if (Subtarget->GVRequiresExtraLoad(GA->getGlobal(),
- getTargetMachine(), false))
+ if (isGlobalStubReference(Subtarget->ClassifyGlobalReference(GV,
+ getTargetMachine())))
return;
if (hasMemory)
- Op = LowerGlobalAddress(GA->getGlobal(), Op.getDebugLoc(), Offset, DAG);
+ Op = LowerGlobalAddress(GV, Op.getDebugLoc(), Offset, DAG);
else
- Op = DAG.getTargetGlobalAddress(GA->getGlobal(), GA->getValueType(0),
- Offset);
+ Op = DAG.getTargetGlobalAddress(GV, GA->getValueType(0), Offset);
Result = Op;
break;
}
// FIXME: not handling fp-stack yet!
switch (Constraint[0]) { // GCC X86 Constraint Letters
default: break; // Unknown constraint letter
- case 'q': // Q_REGS (GENERAL_REGS in 64-bit mode)
+ case 'q': // GENERAL_REGS in 64-bit mode, Q_REGS in 32-bit mode.
+ if (Subtarget->is64Bit()) {
+ if (VT == MVT::i32)
+ return make_vector<unsigned>(X86::EAX, X86::EDX, X86::ECX, X86::EBX,
+ X86::ESI, X86::EDI, X86::R8D, X86::R9D,
+ X86::R10D,X86::R11D,X86::R12D,
+ X86::R13D,X86::R14D,X86::R15D,
+ X86::EBP, X86::ESP, 0);
+ else if (VT == MVT::i16)
+ return make_vector<unsigned>(X86::AX, X86::DX, X86::CX, X86::BX,
+ X86::SI, X86::DI, X86::R8W,X86::R9W,
+ X86::R10W,X86::R11W,X86::R12W,
+ X86::R13W,X86::R14W,X86::R15W,
+ X86::BP, X86::SP, 0);
+ else if (VT == MVT::i8)
+ return make_vector<unsigned>(X86::AL, X86::DL, X86::CL, X86::BL,
+ X86::SIL, X86::DIL, X86::R8B,X86::R9B,
+ X86::R10B,X86::R11B,X86::R12B,
+ X86::R13B,X86::R14B,X86::R15B,
+ X86::BPL, X86::SPL, 0);
+
+ else if (VT == MVT::i64)
+ return make_vector<unsigned>(X86::RAX, X86::RDX, X86::RCX, X86::RBX,
+ X86::RSI, X86::RDI, X86::R8, X86::R9,
+ X86::R10, X86::R11, X86::R12,
+ X86::R13, X86::R14, X86::R15,
+ X86::RBP, X86::RSP, 0);
+
+ break;
+ }
+ // 32-bit fallthrough
case 'Q': // Q_REGS
if (VT == MVT::i32)
return make_vector<unsigned>(X86::EAX, X86::EDX, X86::ECX, X86::EBX, 0);
// 'A' means EAX + EDX.
if (Constraint == "A") {
Res.first = X86::EAX;
- Res.second = X86::GRADRegisterClass;
+ Res.second = X86::GR32_ADRegisterClass;
}
return Res;
}