setSchedulingPreference(Sched::RegPressure);
setStackPointerRegisterToSaveRestore(X86StackPtr);
+ if (Subtarget->isTargetWindows() && !Subtarget->isTargetCygMing()) {
+ // Setup Windows compiler runtime calls.
+ setLibcallName(RTLIB::SDIV_I64, "_alldiv");
+ setLibcallName(RTLIB::UDIV_I64, "_aulldiv");
+ setLibcallName(RTLIB::FPTOUINT_F64_I64, "_ftol2");
+ setLibcallName(RTLIB::FPTOUINT_F32_I64, "_ftol2");
+ setLibcallCallingConv(RTLIB::SDIV_I64, CallingConv::X86_StdCall);
+ setLibcallCallingConv(RTLIB::UDIV_I64, CallingConv::X86_StdCall);
+ setLibcallCallingConv(RTLIB::FPTOUINT_F64_I64, CallingConv::C);
+ setLibcallCallingConv(RTLIB::FPTOUINT_F32_I64, CallingConv::C);
+ }
+
if (Subtarget->isTargetDarwin()) {
// Darwin should use _setjmp/_longjmp instead of setjmp/longjmp.
setUseUnderscoreSetJmp(false);
setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
if (Subtarget->is64Bit())
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64, Expand);
- if (Subtarget->isTargetCygMing())
+ if (Subtarget->isTargetCygMing() || Subtarget->isTargetWindows())
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Custom);
else
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand);
// FIXME: In order to prevent SSE instructions being expanded to MMX ones
// with -msoft-float, disable use of MMX as well.
if (!UseSoftFloat && !DisableMMX && Subtarget->hasMMX()) {
- addRegisterClass(MVT::x86mmx, X86::VR64RegisterClass, false);
+ addRegisterClass(MVT::x86mmx, X86::VR64RegisterClass);
// No operations on x86mmx supported, everything uses intrinsics.
}
return Ins[0].Flags.isSRet();
}
-/// CCAssignFnForNode - Selects the correct CCAssignFn for a the
-/// given CallingConvention value.
-CCAssignFn *X86TargetLowering::CCAssignFnForNode(CallingConv::ID CC) const {
- if (Subtarget->is64Bit()) {
- if (CC == CallingConv::GHC)
- return CC_X86_64_GHC;
- else if (Subtarget->isTargetWin64())
- return CC_X86_Win64_C;
- else
- return CC_X86_64_C;
- }
-
- if (CC == CallingConv::X86_FastCall)
- return CC_X86_32_FastCall;
- else if (CC == CallingConv::X86_ThisCall)
- return CC_X86_32_ThisCall;
- else if (CC == CallingConv::Fast)
- return CC_X86_32_FastCC;
- else if (CC == CallingConv::GHC)
- return CC_X86_32_GHC;
- else
- return CC_X86_32_C;
-}
-
/// 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
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
ArgLocs, *DAG.getContext());
- CCInfo.AnalyzeFormalArguments(Ins, CCAssignFnForNode(CallConv));
+ CCInfo.AnalyzeFormalArguments(Ins, CC_X86);
unsigned LastVal = ~0U;
SDValue ArgValue;
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
ArgLocs, *DAG.getContext());
- CCInfo.AnalyzeCallOperands(Outs, CCAssignFnForNode(CallConv));
+ CCInfo.AnalyzeCallOperands(Outs, CC_X86);
// Get a count of how many bytes are to be pushed on the stack.
unsigned NumBytes = CCInfo.getNextStackOffset();
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CalleeCC, isVarArg, getTargetMachine(),
ArgLocs, *DAG.getContext());
- CCInfo.AnalyzeCallOperands(Outs, CCAssignFnForNode(CalleeCC));
+ CCInfo.AnalyzeCallOperands(Outs, CC_X86);
if (CCInfo.getNextStackOffset()) {
MachineFunction &MF = DAG.getMachineFunction();
if (MF.getInfo<X86MachineFunctionInfo>()->getBytesToPopOnReturn())
}
}
+ // An stdcall caller is expected to clean up its arguments; the callee
+ // isn't going to do that. PR 8461.
+ if (!CCMatch && CallerCC==CallingConv::X86_StdCall)
+ return false;
+
return true;
}
// movssrr and movsdrr do not clear top bits. Try to use movd, movq
// instead.
MVT ExtVT = (OpVT == MVT::v2f64) ? MVT::i64 : MVT::i32;
- if ((ExtVT.SimpleTy != MVT::i64 || Subtarget->is64Bit()) &&
+ if ((ExtVT != MVT::i64 || Subtarget->is64Bit()) &&
SrcOp.getOpcode() == ISD::SCALAR_TO_VECTOR &&
SrcOp.getOperand(0).getOpcode() == ISD::BIT_CONVERT &&
SrcOp.getOperand(0).getOperand(0).getValueType() == ExtVT) {
SDValue
X86TargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op,
SelectionDAG &DAG) const {
- assert(Subtarget->isTargetCygMing() &&
- "This should be used only on Cygwin/Mingw targets");
+ assert((Subtarget->isTargetCygMing() || Subtarget->isTargetWindows()) &&
+ "This should be used only on Windows targets");
DebugLoc dl = Op.getDebugLoc();
// Get the inputs.
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Flag);
- Chain = DAG.getNode(X86ISD::MINGW_ALLOCA, dl, NodeTys, Chain, Flag);
+ Chain = DAG.getNode(X86ISD::WIN_ALLOCA, dl, NodeTys, Chain, Flag);
Flag = Chain.getValue(1);
Chain = DAG.getCopyFromReg(Chain, dl, X86StackPtr, SPTy).getValue(1);
}
SDValue X86TargetLowering::LowerVAARG(SDValue Op, SelectionDAG &DAG) const {
- // X86-64 va_list is a struct { i32, i32, i8*, i8* }.
- assert(Subtarget->is64Bit() && "This code only handles 64-bit va_arg!");
+ assert(Subtarget->is64Bit() &&
+ "LowerVAARG only handles 64-bit va_arg!");
+ assert((Subtarget->isTargetLinux() ||
+ Subtarget->isTargetDarwin()) &&
+ "Unhandled target in LowerVAARG");
+ assert(Op.getNode()->getNumOperands() == 4);
+ SDValue Chain = Op.getOperand(0);
+ SDValue SrcPtr = Op.getOperand(1);
+ const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
+ unsigned Align = Op.getConstantOperandVal(3);
+ DebugLoc dl = Op.getDebugLoc();
- report_fatal_error("VAArgInst is not yet implemented for x86-64!");
- return SDValue();
+ EVT ArgVT = Op.getNode()->getValueType(0);
+ const Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
+ uint32_t ArgSize = getTargetData()->getTypeAllocSize(ArgTy);
+ uint8_t ArgMode;
+
+ // Decide which area this value should be read from.
+ // TODO: Implement the AMD64 ABI in its entirety. This simple
+ // selection mechanism works only for the basic types.
+ if (ArgVT == MVT::f80) {
+ llvm_unreachable("va_arg for f80 not yet implemented");
+ } else if (ArgVT.isFloatingPoint() && ArgSize <= 16 /*bytes*/) {
+ ArgMode = 2; // Argument passed in XMM register. Use fp_offset.
+ } else if (ArgVT.isInteger() && ArgSize <= 32 /*bytes*/) {
+ ArgMode = 1; // Argument passed in GPR64 register(s). Use gp_offset.
+ } else {
+ llvm_unreachable("Unhandled argument type in LowerVAARG");
+ }
+
+ if (ArgMode == 2) {
+ // Sanity Check: Make sure using fp_offset makes sense.
+ assert(!UseSoftFloat &&
+ !(DAG.getMachineFunction()
+ .getFunction()->hasFnAttr(Attribute::NoImplicitFloat)) &&
+ Subtarget->hasSSE1());
+ }
+
+ // Insert VAARG_64 node into the DAG
+ // VAARG_64 returns two values: Variable Argument Address, Chain
+ SmallVector<SDValue, 11> InstOps;
+ InstOps.push_back(Chain);
+ InstOps.push_back(SrcPtr);
+ InstOps.push_back(DAG.getConstant(ArgSize, MVT::i32));
+ InstOps.push_back(DAG.getConstant(ArgMode, MVT::i8));
+ InstOps.push_back(DAG.getConstant(Align, MVT::i32));
+ SDVTList VTs = DAG.getVTList(getPointerTy(), MVT::Other);
+ SDValue VAARG = DAG.getMemIntrinsicNode(X86ISD::VAARG_64, dl,
+ VTs, &InstOps[0], InstOps.size(),
+ MVT::i64,
+ MachinePointerInfo(SV),
+ /*Align=*/0,
+ /*Volatile=*/false,
+ /*ReadMem=*/true,
+ /*WriteMem=*/true);
+ Chain = VAARG.getValue(1);
+
+ // Load the next argument and return it
+ return DAG.getLoad(ArgVT, dl,
+ Chain,
+ VAARG,
+ MachinePointerInfo(),
+ false, false, 0);
}
SDValue X86TargetLowering::LowerVACOPY(SDValue Op, SelectionDAG &DAG) const {
N->getOperand(1),
swapInH.getValue(1) };
SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Flag);
- SDValue Result = DAG.getNode(X86ISD::LCMPXCHG8_DAG, dl, Tys, Ops, 3);
+ MachineMemOperand *MMO = cast<AtomicSDNode>(N)->getMemOperand();
+ SDValue Result = DAG.getMemIntrinsicNode(X86ISD::LCMPXCHG8_DAG, dl, Tys,
+ Ops, 3, T, MMO);
SDValue cpOutL = DAG.getCopyFromReg(Result.getValue(0), dl, X86::EAX,
MVT::i32, Result.getValue(1));
SDValue cpOutH = DAG.getCopyFromReg(cpOutL.getValue(1), dl, X86::EDX,
case X86ISD::PUNPCKHDQ: return "X86ISD::PUNPCKHDQ";
case X86ISD::PUNPCKHQDQ: return "X86ISD::PUNPCKHQDQ";
case X86ISD::VASTART_SAVE_XMM_REGS: return "X86ISD::VASTART_SAVE_XMM_REGS";
- case X86ISD::MINGW_ALLOCA: return "X86ISD::MINGW_ALLOCA";
+ case X86ISD::VAARG_64: return "X86ISD::VAARG_64";
+ case X86ISD::WIN_ALLOCA: return "X86ISD::WIN_ALLOCA";
}
}
return BB;
}
+MachineBasicBlock *
+X86TargetLowering::EmitVAARG64WithCustomInserter(
+ MachineInstr *MI,
+ MachineBasicBlock *MBB) const {
+ // Emit va_arg instruction on X86-64.
+
+ // Operands to this pseudo-instruction:
+ // 0 ) Output : destination address (reg)
+ // 1-5) Input : va_list address (addr, i64mem)
+ // 6 ) ArgSize : Size (in bytes) of vararg type
+ // 7 ) ArgMode : 0=overflow only, 1=use gp_offset, 2=use fp_offset
+ // 8 ) Align : Alignment of type
+ // 9 ) EFLAGS (implicit-def)
+
+ assert(MI->getNumOperands() == 10 && "VAARG_64 should have 10 operands!");
+ assert(X86::AddrNumOperands == 5 && "VAARG_64 assumes 5 address operands");
+
+ unsigned DestReg = MI->getOperand(0).getReg();
+ MachineOperand &Base = MI->getOperand(1);
+ MachineOperand &Scale = MI->getOperand(2);
+ MachineOperand &Index = MI->getOperand(3);
+ MachineOperand &Disp = MI->getOperand(4);
+ MachineOperand &Segment = MI->getOperand(5);
+ unsigned ArgSize = MI->getOperand(6).getImm();
+ unsigned ArgMode = MI->getOperand(7).getImm();
+ unsigned Align = MI->getOperand(8).getImm();
+
+ // Memory Reference
+ assert(MI->hasOneMemOperand() && "Expected VAARG_64 to have one memoperand");
+ MachineInstr::mmo_iterator MMOBegin = MI->memoperands_begin();
+ MachineInstr::mmo_iterator MMOEnd = MI->memoperands_end();
+
+ // Machine Information
+ const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+ MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
+ const TargetRegisterClass *AddrRegClass = getRegClassFor(MVT::i64);
+ const TargetRegisterClass *OffsetRegClass = getRegClassFor(MVT::i32);
+ DebugLoc DL = MI->getDebugLoc();
+
+ // struct va_list {
+ // i32 gp_offset
+ // i32 fp_offset
+ // i64 overflow_area (address)
+ // i64 reg_save_area (address)
+ // }
+ // sizeof(va_list) = 24
+ // alignment(va_list) = 8
+
+ unsigned TotalNumIntRegs = 6;
+ unsigned TotalNumXMMRegs = 8;
+ bool UseGPOffset = (ArgMode == 1);
+ bool UseFPOffset = (ArgMode == 2);
+ unsigned MaxOffset = TotalNumIntRegs * 8 +
+ (UseFPOffset ? TotalNumXMMRegs * 16 : 0);
+
+ /* Align ArgSize to a multiple of 8 */
+ unsigned ArgSizeA8 = (ArgSize + 7) & ~7;
+ bool NeedsAlign = (Align > 8);
+
+ MachineBasicBlock *thisMBB = MBB;
+ MachineBasicBlock *overflowMBB;
+ MachineBasicBlock *offsetMBB;
+ MachineBasicBlock *endMBB;
+
+ unsigned OffsetDestReg = 0; // Argument address computed by offsetMBB
+ unsigned OverflowDestReg = 0; // Argument address computed by overflowMBB
+ unsigned OffsetReg = 0;
+
+ if (!UseGPOffset && !UseFPOffset) {
+ // If we only pull from the overflow region, we don't create a branch.
+ // We don't need to alter control flow.
+ OffsetDestReg = 0; // unused
+ OverflowDestReg = DestReg;
+
+ offsetMBB = NULL;
+ overflowMBB = thisMBB;
+ endMBB = thisMBB;
+ } else {
+ // First emit code to check if gp_offset (or fp_offset) is below the bound.
+ // If so, pull the argument from reg_save_area. (branch to offsetMBB)
+ // If not, pull from overflow_area. (branch to overflowMBB)
+ //
+ // thisMBB
+ // | .
+ // | .
+ // offsetMBB overflowMBB
+ // | .
+ // | .
+ // endMBB
+
+ // Registers for the PHI in endMBB
+ OffsetDestReg = MRI.createVirtualRegister(AddrRegClass);
+ OverflowDestReg = MRI.createVirtualRegister(AddrRegClass);
+
+ const BasicBlock *LLVM_BB = MBB->getBasicBlock();
+ MachineFunction *MF = MBB->getParent();
+ overflowMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+ offsetMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+ endMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+
+ MachineFunction::iterator MBBIter = MBB;
+ ++MBBIter;
+
+ // Insert the new basic blocks
+ MF->insert(MBBIter, offsetMBB);
+ MF->insert(MBBIter, overflowMBB);
+ MF->insert(MBBIter, endMBB);
+
+ // Transfer the remainder of MBB and its successor edges to endMBB.
+ endMBB->splice(endMBB->begin(), thisMBB,
+ llvm::next(MachineBasicBlock::iterator(MI)),
+ thisMBB->end());
+ endMBB->transferSuccessorsAndUpdatePHIs(thisMBB);
+
+ // Make offsetMBB and overflowMBB successors of thisMBB
+ thisMBB->addSuccessor(offsetMBB);
+ thisMBB->addSuccessor(overflowMBB);
+
+ // endMBB is a successor of both offsetMBB and overflowMBB
+ offsetMBB->addSuccessor(endMBB);
+ overflowMBB->addSuccessor(endMBB);
+
+ // Load the offset value into a register
+ OffsetReg = MRI.createVirtualRegister(OffsetRegClass);
+ BuildMI(thisMBB, DL, TII->get(X86::MOV32rm), OffsetReg)
+ .addOperand(Base)
+ .addOperand(Scale)
+ .addOperand(Index)
+ .addDisp(Disp, UseFPOffset ? 4 : 0)
+ .addOperand(Segment)
+ .setMemRefs(MMOBegin, MMOEnd);
+
+ // Check if there is enough room left to pull this argument.
+ BuildMI(thisMBB, DL, TII->get(X86::CMP32ri))
+ .addReg(OffsetReg)
+ .addImm(MaxOffset + 8 - ArgSizeA8);
+
+ // Branch to "overflowMBB" if offset >= max
+ // Fall through to "offsetMBB" otherwise
+ BuildMI(thisMBB, DL, TII->get(X86::GetCondBranchFromCond(X86::COND_AE)))
+ .addMBB(overflowMBB);
+ }
+
+ // In offsetMBB, emit code to use the reg_save_area.
+ if (offsetMBB) {
+ assert(OffsetReg != 0);
+
+ // Read the reg_save_area address.
+ unsigned RegSaveReg = MRI.createVirtualRegister(AddrRegClass);
+ BuildMI(offsetMBB, DL, TII->get(X86::MOV64rm), RegSaveReg)
+ .addOperand(Base)
+ .addOperand(Scale)
+ .addOperand(Index)
+ .addDisp(Disp, 16)
+ .addOperand(Segment)
+ .setMemRefs(MMOBegin, MMOEnd);
+
+ // Zero-extend the offset
+ unsigned OffsetReg64 = MRI.createVirtualRegister(AddrRegClass);
+ BuildMI(offsetMBB, DL, TII->get(X86::SUBREG_TO_REG), OffsetReg64)
+ .addImm(0)
+ .addReg(OffsetReg)
+ .addImm(X86::sub_32bit);
+
+ // Add the offset to the reg_save_area to get the final address.
+ BuildMI(offsetMBB, DL, TII->get(X86::ADD64rr), OffsetDestReg)
+ .addReg(OffsetReg64)
+ .addReg(RegSaveReg);
+
+ // Compute the offset for the next argument
+ unsigned NextOffsetReg = MRI.createVirtualRegister(OffsetRegClass);
+ BuildMI(offsetMBB, DL, TII->get(X86::ADD32ri), NextOffsetReg)
+ .addReg(OffsetReg)
+ .addImm(UseFPOffset ? 16 : 8);
+
+ // Store it back into the va_list.
+ BuildMI(offsetMBB, DL, TII->get(X86::MOV32mr))
+ .addOperand(Base)
+ .addOperand(Scale)
+ .addOperand(Index)
+ .addDisp(Disp, UseFPOffset ? 4 : 0)
+ .addOperand(Segment)
+ .addReg(NextOffsetReg)
+ .setMemRefs(MMOBegin, MMOEnd);
+
+ // Jump to endMBB
+ BuildMI(offsetMBB, DL, TII->get(X86::JMP_4))
+ .addMBB(endMBB);
+ }
+
+ //
+ // Emit code to use overflow area
+ //
+
+ // Load the overflow_area address into a register.
+ unsigned OverflowAddrReg = MRI.createVirtualRegister(AddrRegClass);
+ BuildMI(overflowMBB, DL, TII->get(X86::MOV64rm), OverflowAddrReg)
+ .addOperand(Base)
+ .addOperand(Scale)
+ .addOperand(Index)
+ .addDisp(Disp, 8)
+ .addOperand(Segment)
+ .setMemRefs(MMOBegin, MMOEnd);
+
+ // If we need to align it, do so. Otherwise, just copy the address
+ // to OverflowDestReg.
+ if (NeedsAlign) {
+ // Align the overflow address
+ assert((Align & (Align-1)) == 0 && "Alignment must be a power of 2");
+ unsigned TmpReg = MRI.createVirtualRegister(AddrRegClass);
+
+ // aligned_addr = (addr + (align-1)) & ~(align-1)
+ BuildMI(overflowMBB, DL, TII->get(X86::ADD64ri32), TmpReg)
+ .addReg(OverflowAddrReg)
+ .addImm(Align-1);
+
+ BuildMI(overflowMBB, DL, TII->get(X86::AND64ri32), OverflowDestReg)
+ .addReg(TmpReg)
+ .addImm(~(uint64_t)(Align-1));
+ } else {
+ BuildMI(overflowMBB, DL, TII->get(TargetOpcode::COPY), OverflowDestReg)
+ .addReg(OverflowAddrReg);
+ }
+
+ // Compute the next overflow address after this argument.
+ // (the overflow address should be kept 8-byte aligned)
+ unsigned NextAddrReg = MRI.createVirtualRegister(AddrRegClass);
+ BuildMI(overflowMBB, DL, TII->get(X86::ADD64ri32), NextAddrReg)
+ .addReg(OverflowDestReg)
+ .addImm(ArgSizeA8);
+
+ // Store the new overflow address.
+ BuildMI(overflowMBB, DL, TII->get(X86::MOV64mr))
+ .addOperand(Base)
+ .addOperand(Scale)
+ .addOperand(Index)
+ .addDisp(Disp, 8)
+ .addOperand(Segment)
+ .addReg(NextAddrReg)
+ .setMemRefs(MMOBegin, MMOEnd);
+
+ // If we branched, emit the PHI to the front of endMBB.
+ if (offsetMBB) {
+ BuildMI(*endMBB, endMBB->begin(), DL,
+ TII->get(X86::PHI), DestReg)
+ .addReg(OffsetDestReg).addMBB(offsetMBB)
+ .addReg(OverflowDestReg).addMBB(overflowMBB);
+ }
+
+ // Erase the pseudo instruction
+ MI->eraseFromParent();
+
+ return endMBB;
+}
+
MachineBasicBlock *
X86TargetLowering::EmitVAStartSaveXMMRegsWithCustomInserter(
MachineInstr *MI,
}
MachineBasicBlock *
-X86TargetLowering::EmitLoweredMingwAlloca(MachineInstr *MI,
+X86TargetLowering::EmitLoweredWinAlloca(MachineInstr *MI,
MachineBasicBlock *BB) const {
const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
DebugLoc DL = MI->getDebugLoc();
// FIXME: The code should be tweaked as soon as we'll try to do codegen for
// mingw-w64.
+ const char *StackProbeSymbol =
+ Subtarget->isTargetWindows() ? "_chkstk" : "_alloca";
+
BuildMI(*BB, MI, DL, TII->get(X86::CALLpcrel32))
- .addExternalSymbol("_alloca")
+ .addExternalSymbol(StackProbeSymbol)
.addReg(X86::EAX, RegState::Implicit)
.addReg(X86::ESP, RegState::Implicit)
.addReg(X86::EAX, RegState::Define | RegState::Implicit)
MachineBasicBlock *BB) const {
switch (MI->getOpcode()) {
default: assert(false && "Unexpected instr type to insert");
- case X86::MINGW_ALLOCA:
- return EmitLoweredMingwAlloca(MI, BB);
+ case X86::WIN_ALLOCA:
+ return EmitLoweredWinAlloca(MI, BB);
case X86::TLSCall_32:
case X86::TLSCall_64:
return EmitLoweredTLSCall(MI, BB);
false);
case X86::VASTART_SAVE_XMM_REGS:
return EmitVAStartSaveXMMRegsWithCustomInserter(MI, BB);
+
+ case X86::VAARG_64:
+ return EmitVAARG64WithCustomInserter(MI, BB);
}
}
bool X86TargetLowering::ExpandInlineAsm(CallInst *CI) const {
InlineAsm *IA = cast<InlineAsm>(CI->getCalledValue());
- std::vector<InlineAsm::ConstraintInfo> Constraints = IA->ParseConstraints();
+ InlineAsm::ConstraintInfoVector Constraints = IA->ParseConstraints();
std::string AsmStr = IA->getAsmString();
X86TargetLowering::getConstraintType(const std::string &Constraint) const {
if (Constraint.size() == 1) {
switch (Constraint[0]) {
- case 'A':
- return C_Register;
- case 'f':
- case 'r':
case 'R':
- case 'l':
case 'q':
case 'Q':
- case 'x':
+ case 'f':
+ case 't':
+ case 'u':
case 'y':
+ case 'x':
case 'Y':
return C_RegisterClass;
+ case 'a':
+ case 'b':
+ case 'c':
+ case 'd':
+ case 'S':
+ case 'D':
+ case 'A':
+ return C_Register;
+ case 'I':
+ case 'J':
+ case 'K':
+ case 'L':
+ case 'M':
+ case 'N':
+ case 'G':
+ case 'C':
case 'e':
case 'Z':
return C_Other;
return TargetLowering::getConstraintType(Constraint);
}
-/// Examine constraint type and operand type and determine a weight value,
-/// where: -1 = invalid match, and 0 = so-so match to 3 = good match.
+/// Examine constraint type and operand type and determine a weight value.
/// This object must already have been set up with the operand type
/// and the current alternative constraint selected.
-int X86TargetLowering::getSingleConstraintMatchWeight(
+TargetLowering::ConstraintWeight
+ X86TargetLowering::getSingleConstraintMatchWeight(
AsmOperandInfo &info, const char *constraint) const {
- int weight = -1;
+ ConstraintWeight weight = CW_Invalid;
Value *CallOperandVal = info.CallOperandVal;
// If we don't have a value, we can't do a match,
// but allow it at the lowest weight.
if (CallOperandVal == NULL)
- return 0;
+ return CW_Default;
+ const Type *type = CallOperandVal->getType();
// Look at the constraint type.
switch (*constraint) {
default:
- return TargetLowering::getSingleConstraintMatchWeight(info, constraint);
+ weight = TargetLowering::getSingleConstraintMatchWeight(info, constraint);
+ case 'R':
+ case 'q':
+ case 'Q':
+ case 'a':
+ case 'b':
+ case 'c':
+ case 'd':
+ case 'S':
+ case 'D':
+ case 'A':
+ if (CallOperandVal->getType()->isIntegerTy())
+ weight = CW_SpecificReg;
+ break;
+ case 'f':
+ case 't':
+ case 'u':
+ if (type->isFloatingPointTy())
+ weight = CW_SpecificReg;
+ break;
+ case 'y':
+ if (type->isX86_MMXTy() && !DisableMMX && Subtarget->hasMMX())
+ weight = CW_SpecificReg;
+ break;
+ case 'x':
+ case 'Y':
+ if ((type->getPrimitiveSizeInBits() == 128) && Subtarget->hasSSE1())
+ weight = CW_Register;
break;
case 'I':
if (ConstantInt *C = dyn_cast<ConstantInt>(info.CallOperandVal)) {
if (C->getZExtValue() <= 31)
- weight = 3;
+ weight = CW_Constant;
+ }
+ break;
+ case 'J':
+ if (ConstantInt *C = dyn_cast<ConstantInt>(CallOperandVal)) {
+ if (C->getZExtValue() <= 63)
+ weight = CW_Constant;
+ }
+ break;
+ case 'K':
+ if (ConstantInt *C = dyn_cast<ConstantInt>(CallOperandVal)) {
+ if ((C->getSExtValue() >= -0x80) && (C->getSExtValue() <= 0x7f))
+ weight = CW_Constant;
+ }
+ break;
+ case 'L':
+ if (ConstantInt *C = dyn_cast<ConstantInt>(CallOperandVal)) {
+ if ((C->getZExtValue() == 0xff) || (C->getZExtValue() == 0xffff))
+ weight = CW_Constant;
+ }
+ break;
+ case 'M':
+ if (ConstantInt *C = dyn_cast<ConstantInt>(CallOperandVal)) {
+ if (C->getZExtValue() <= 3)
+ weight = CW_Constant;
+ }
+ break;
+ case 'N':
+ if (ConstantInt *C = dyn_cast<ConstantInt>(CallOperandVal)) {
+ if (C->getZExtValue() <= 0xff)
+ weight = CW_Constant;
+ }
+ break;
+ case 'G':
+ case 'C':
+ if (dyn_cast<ConstantFP>(CallOperandVal)) {
+ weight = CW_Constant;
+ }
+ break;
+ case 'e':
+ if (ConstantInt *C = dyn_cast<ConstantInt>(CallOperandVal)) {
+ if ((C->getSExtValue() >= -0x80000000LL) &&
+ (C->getSExtValue() <= 0x7fffffffLL))
+ weight = CW_Constant;
+ }
+ break;
+ case 'Z':
+ if (ConstantInt *C = dyn_cast<ConstantInt>(CallOperandVal)) {
+ if (C->getZExtValue() <= 0xffffffff)
+ weight = CW_Constant;
}
break;
- // etc.
}
return weight;
}
projects / oota-llvm.git / blobdiff