#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/CodeGen/SelectionDAGISel.h"
#include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/Target/TargetOptions.h"
+#include "llvm/Target/TargetLoweringObjectFile.h"
#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
#include "llvm/ADT/VectorExtras.h"
using namespace llvm;
SystemZTargetLowering::SystemZTargetLowering(SystemZTargetMachine &tm) :
- TargetLowering(tm), Subtarget(*tm.getSubtargetImpl()), TM(tm) {
+ TargetLowering(tm, new TargetLoweringObjectFileELF()),
+ Subtarget(*tm.getSubtargetImpl()), TM(tm) {
RegInfo = TM.getRegisterInfo();
// Set up the register classes.
- addRegisterClass(MVT::i32, SystemZ::GR32RegisterClass);
- addRegisterClass(MVT::i64, SystemZ::GR64RegisterClass);
+ addRegisterClass(MVT::i32, SystemZ::GR32RegisterClass);
+ addRegisterClass(MVT::i64, SystemZ::GR64RegisterClass);
+ addRegisterClass(MVT::v2i32,SystemZ::GR64PRegisterClass);
+ addRegisterClass(MVT::v2i64,SystemZ::GR128RegisterClass);
+
+ if (!UseSoftFloat) {
+ addRegisterClass(MVT::f32, SystemZ::FP32RegisterClass);
+ addRegisterClass(MVT::f64, SystemZ::FP64RegisterClass);
+ }
// Compute derived properties from the register classes
computeRegisterProperties();
// Set shifts properties
- setShiftAmountFlavor(Extend);
- setShiftAmountType(MVT::i32);
+ setShiftAmountType(MVT::i64);
// Provide all sorts of operation actions
+ setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
+ setLoadExtAction(ISD::ZEXTLOAD, MVT::i1, Promote);
+ setLoadExtAction(ISD::EXTLOAD, MVT::i1, Promote);
+
+ setLoadExtAction(ISD::SEXTLOAD, MVT::f32, Expand);
+ setLoadExtAction(ISD::ZEXTLOAD, MVT::f32, Expand);
+ setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
+
+ setLoadExtAction(ISD::SEXTLOAD, MVT::f64, Expand);
+ setLoadExtAction(ISD::ZEXTLOAD, MVT::f64, Expand);
+ setLoadExtAction(ISD::EXTLOAD, MVT::f64, Expand);
setStackPointerRegisterToSaveRestore(SystemZ::R15D);
- setSchedulingPreference(SchedulingForLatency);
- setOperationAction(ISD::RET, MVT::Other, Custom);
+ // TODO: It may be better to default to latency-oriented scheduling, however
+ // LLVM's current latency-oriented scheduler can't handle physreg definitions
+ // such as SystemZ has with PSW, so set this to the register-pressure
+ // scheduler, because it can.
+ setSchedulingPreference(SchedulingForRegPressure);
+
+ setBooleanContents(ZeroOrOneBooleanContent);
+
+ setOperationAction(ISD::BR_JT, MVT::Other, Expand);
+ setOperationAction(ISD::BRCOND, MVT::Other, Expand);
+ setOperationAction(ISD::BR_CC, MVT::i32, Custom);
+ setOperationAction(ISD::BR_CC, MVT::i64, Custom);
+ setOperationAction(ISD::BR_CC, MVT::f32, Custom);
+ setOperationAction(ISD::BR_CC, MVT::f64, Custom);
+ setOperationAction(ISD::ConstantPool, MVT::i32, Custom);
+ setOperationAction(ISD::ConstantPool, MVT::i64, Custom);
+ setOperationAction(ISD::GlobalAddress, MVT::i64, Custom);
+ setOperationAction(ISD::JumpTable, MVT::i64, Custom);
+ setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64, Expand);
+
+ setOperationAction(ISD::SDIV, MVT::i32, Expand);
+ setOperationAction(ISD::UDIV, MVT::i32, Expand);
+ setOperationAction(ISD::SDIV, MVT::i64, Expand);
+ setOperationAction(ISD::UDIV, MVT::i64, Expand);
+ setOperationAction(ISD::SREM, MVT::i32, Expand);
+ setOperationAction(ISD::UREM, MVT::i32, Expand);
+ setOperationAction(ISD::SREM, MVT::i64, Expand);
+ setOperationAction(ISD::UREM, MVT::i64, Expand);
+
+ setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
+
+ setOperationAction(ISD::CTPOP, MVT::i32, Expand);
+ setOperationAction(ISD::CTPOP, MVT::i64, Expand);
+ setOperationAction(ISD::CTTZ, MVT::i32, Expand);
+ setOperationAction(ISD::CTTZ, MVT::i64, Expand);
+ setOperationAction(ISD::CTLZ, MVT::i32, Promote);
+ setOperationAction(ISD::CTLZ, MVT::i64, Legal);
+
+ // FIXME: Can we lower these 2 efficiently?
+ setOperationAction(ISD::SETCC, MVT::i32, Expand);
+ setOperationAction(ISD::SETCC, MVT::i64, Expand);
+ setOperationAction(ISD::SETCC, MVT::f32, Expand);
+ setOperationAction(ISD::SETCC, MVT::f64, Expand);
+ setOperationAction(ISD::SELECT, MVT::i32, Expand);
+ setOperationAction(ISD::SELECT, MVT::i64, Expand);
+ setOperationAction(ISD::SELECT, MVT::f32, Expand);
+ setOperationAction(ISD::SELECT, MVT::f64, Expand);
+ setOperationAction(ISD::SELECT_CC, MVT::i32, Custom);
+ setOperationAction(ISD::SELECT_CC, MVT::i64, Custom);
+ setOperationAction(ISD::SELECT_CC, MVT::f32, Custom);
+ setOperationAction(ISD::SELECT_CC, MVT::f64, Custom);
+
+ setOperationAction(ISD::MULHS, MVT::i64, Expand);
+ setOperationAction(ISD::SMUL_LOHI, MVT::i64, Expand);
+
+ // FIXME: Can we support these natively?
+ setOperationAction(ISD::UMUL_LOHI, MVT::i64, Expand);
+ setOperationAction(ISD::SRL_PARTS, MVT::i64, Expand);
+ setOperationAction(ISD::SHL_PARTS, MVT::i64, Expand);
+ setOperationAction(ISD::SRA_PARTS, MVT::i64, Expand);
+
+ // Lower some FP stuff
+ setOperationAction(ISD::FSIN, MVT::f32, Expand);
+ setOperationAction(ISD::FSIN, MVT::f64, Expand);
+ setOperationAction(ISD::FCOS, MVT::f32, Expand);
+ setOperationAction(ISD::FCOS, MVT::f64, Expand);
+ setOperationAction(ISD::FREM, MVT::f32, Expand);
+ setOperationAction(ISD::FREM, MVT::f64, Expand);
+
+ // We have only 64-bit bitconverts
+ setOperationAction(ISD::BIT_CONVERT, MVT::f32, Expand);
+ setOperationAction(ISD::BIT_CONVERT, MVT::i32, Expand);
+
+ setOperationAction(ISD::UINT_TO_FP, MVT::i32, Expand);
+ setOperationAction(ISD::UINT_TO_FP, MVT::i64, Expand);
+ setOperationAction(ISD::FP_TO_UINT, MVT::i32, Expand);
+ setOperationAction(ISD::FP_TO_UINT, MVT::i64, Expand);
+
+ setTruncStoreAction(MVT::f64, MVT::f32, Expand);
}
SDValue SystemZTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) {
switch (Op.getOpcode()) {
- case ISD::FORMAL_ARGUMENTS: return LowerFORMAL_ARGUMENTS(Op, DAG);
- case ISD::RET: return LowerRET(Op, DAG);
- case ISD::CALL: return LowerCALL(Op, DAG);
+ case ISD::BR_CC: return LowerBR_CC(Op, DAG);
+ case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
+ case ISD::GlobalAddress: return LowerGlobalAddress(Op, DAG);
+ case ISD::JumpTable: return LowerJumpTable(Op, DAG);
+ case ISD::ConstantPool: return LowerConstantPool(Op, DAG);
default:
- assert(0 && "unimplemented operand");
+ llvm_unreachable("Should not custom lower this!");
return SDValue();
}
}
+bool SystemZTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const {
+ if (UseSoftFloat || (VT != MVT::f32 && VT != MVT::f64))
+ return false;
+
+ // +0.0 lzer
+ // +0.0f lzdr
+ // -0.0 lzer + lner
+ // -0.0f lzdr + lndr
+ return Imm.isZero() || Imm.isNegZero();
+}
+
+//===----------------------------------------------------------------------===//
+// SystemZ Inline Assembly Support
+//===----------------------------------------------------------------------===//
+
+/// getConstraintType - Given a constraint letter, return the type of
+/// constraint it is for this target.
+TargetLowering::ConstraintType
+SystemZTargetLowering::getConstraintType(const std::string &Constraint) const {
+ if (Constraint.size() == 1) {
+ switch (Constraint[0]) {
+ case 'r':
+ return C_RegisterClass;
+ default:
+ break;
+ }
+ }
+ return TargetLowering::getConstraintType(Constraint);
+}
+
+std::pair<unsigned, const TargetRegisterClass*>
+SystemZTargetLowering::
+getRegForInlineAsmConstraint(const std::string &Constraint,
+ EVT VT) const {
+ if (Constraint.size() == 1) {
+ // GCC Constraint Letters
+ switch (Constraint[0]) {
+ default: break;
+ case 'r': // GENERAL_REGS
+ if (VT == MVT::i32)
+ return std::make_pair(0U, SystemZ::GR32RegisterClass);
+ else if (VT == MVT::i128)
+ return std::make_pair(0U, SystemZ::GR128RegisterClass);
+
+ return std::make_pair(0U, SystemZ::GR64RegisterClass);
+ }
+ }
+
+ return TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
+}
+
//===----------------------------------------------------------------------===//
// Calling Convention Implementation
//===----------------------------------------------------------------------===//
#include "SystemZGenCallingConv.inc"
-SDValue SystemZTargetLowering::LowerFORMAL_ARGUMENTS(SDValue Op,
- SelectionDAG &DAG) {
- unsigned CC = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
- switch (CC) {
+SDValue
+SystemZTargetLowering::LowerFormalArguments(SDValue Chain,
+ CallingConv::ID CallConv,
+ bool isVarArg,
+ const SmallVectorImpl<ISD::InputArg>
+ &Ins,
+ DebugLoc dl,
+ SelectionDAG &DAG,
+ SmallVectorImpl<SDValue> &InVals) {
+
+ switch (CallConv) {
default:
- assert(0 && "Unsupported calling convention");
+ llvm_unreachable("Unsupported calling convention");
case CallingConv::C:
case CallingConv::Fast:
- return LowerCCCArguments(Op, DAG);
+ return LowerCCCArguments(Chain, CallConv, isVarArg, Ins, dl, DAG, InVals);
}
}
-SDValue SystemZTargetLowering::LowerCALL(SDValue Op, SelectionDAG &DAG) {
- CallSDNode *TheCall = cast<CallSDNode>(Op.getNode());
- unsigned CallingConv = TheCall->getCallingConv();
- switch (CallingConv) {
+SDValue
+SystemZTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
+ const Type *RetTy,
+ CallingConv::ID CallConv, bool isVarArg,
+ bool &isTailCall,
+ const SmallVectorImpl<ISD::OutputArg> &Outs,
+ const SmallVectorImpl<ISD::InputArg> &Ins,
+ DebugLoc dl, SelectionDAG &DAG,
+ SmallVectorImpl<SDValue> &InVals) {
+ // SystemZ target does not yet support tail call optimization.
+ isTailCall = false;
+
+ switch (CallConv) {
default:
- assert(0 && "Unsupported calling convention");
+ llvm_unreachable("Unsupported calling convention");
case CallingConv::Fast:
case CallingConv::C:
- return LowerCCCCallTo(Op, DAG, CallingConv);
+ return LowerCCCCallTo(Chain, Callee, CallConv, isVarArg, isTailCall,
+ Outs, Ins, dl, DAG, InVals);
}
}
/// generate load operations for arguments places on the stack.
// FIXME: struct return stuff
// FIXME: varargs
-SDValue SystemZTargetLowering::LowerCCCArguments(SDValue Op,
- SelectionDAG &DAG) {
+SDValue
+SystemZTargetLowering::LowerCCCArguments(SDValue Chain,
+ CallingConv::ID CallConv,
+ bool isVarArg,
+ const SmallVectorImpl<ISD::InputArg>
+ &Ins,
+ DebugLoc dl,
+ SelectionDAG &DAG,
+ SmallVectorImpl<SDValue> &InVals) {
+
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
- SDValue Root = Op.getOperand(0);
- bool isVarArg = cast<ConstantSDNode>(Op.getOperand(2))->getZExtValue() != 0;
- unsigned CC = MF.getFunction()->getCallingConv();
- DebugLoc dl = Op.getDebugLoc();
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
- CCState CCInfo(CC, isVarArg, getTargetMachine(), ArgLocs);
- CCInfo.AnalyzeFormalArguments(Op.getNode(), CC_SystemZ);
+ CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
+ ArgLocs, *DAG.getContext());
+ CCInfo.AnalyzeFormalArguments(Ins, CC_SystemZ);
- assert(!isVarArg && "Varargs not supported yet");
+ if (isVarArg)
+ llvm_report_error("Varargs not supported yet");
- SmallVector<SDValue, 16> ArgValues;
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
+ SDValue ArgValue;
CCValAssign &VA = ArgLocs[i];
+ EVT LocVT = VA.getLocVT();
if (VA.isRegLoc()) {
// Arguments passed in registers
- MVT RegVT = VA.getLocVT();
- switch (RegVT.getSimpleVT()) {
+ TargetRegisterClass *RC;
+ switch (LocVT.getSimpleVT().SimpleTy) {
default:
- cerr << "LowerFORMAL_ARGUMENTS Unhandled argument type: "
- << RegVT.getSimpleVT()
+#ifndef NDEBUG
+ errs() << "LowerFormalArguments Unhandled argument type: "
+ << LocVT.getSimpleVT().SimpleTy
<< "\n";
- abort();
+#endif
+ llvm_unreachable(0);
case MVT::i64:
- unsigned VReg =
- RegInfo.createVirtualRegister(SystemZ::GR64RegisterClass);
- RegInfo.addLiveIn(VA.getLocReg(), VReg);
- SDValue ArgValue = DAG.getCopyFromReg(Root, dl, VReg, RegVT);
-
- // If this is an 8/16/32-bit value, it is really passed promoted to 64
- // bits. Insert an assert[sz]ext to capture this, then truncate to the
- // right size.
- if (VA.getLocInfo() == CCValAssign::SExt)
- ArgValue = DAG.getNode(ISD::AssertSext, dl, RegVT, ArgValue,
- DAG.getValueType(VA.getValVT()));
- else if (VA.getLocInfo() == CCValAssign::ZExt)
- ArgValue = DAG.getNode(ISD::AssertZext, dl, RegVT, ArgValue,
- DAG.getValueType(VA.getValVT()));
-
- if (VA.getLocInfo() != CCValAssign::Full)
- ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
-
- ArgValues.push_back(ArgValue);
+ RC = SystemZ::GR64RegisterClass;
+ break;
+ case MVT::f32:
+ RC = SystemZ::FP32RegisterClass;
+ break;
+ case MVT::f64:
+ RC = SystemZ::FP64RegisterClass;
+ break;
}
+
+ unsigned VReg = RegInfo.createVirtualRegister(RC);
+ RegInfo.addLiveIn(VA.getLocReg(), VReg);
+ ArgValue = DAG.getCopyFromReg(Chain, dl, VReg, LocVT);
} else {
// Sanity check
assert(VA.isMemLoc());
- // Load the argument to a virtual register
- unsigned ObjSize = VA.getLocVT().getSizeInBits()/8;
- if (ObjSize > 8) {
- cerr << "LowerFORMAL_ARGUMENTS Unhandled argument type: "
- << VA.getLocVT().getSimpleVT()
- << "\n";
- }
+
+ // Create the nodes corresponding to a load from this parameter slot.
// Create the frame index object for this incoming parameter...
- int FI = MFI->CreateFixedObject(ObjSize, VA.getLocMemOffset());
+ int FI = MFI->CreateFixedObject(LocVT.getSizeInBits()/8,
+ VA.getLocMemOffset(), true, false);
// Create the SelectionDAG nodes corresponding to a load
- //from this parameter
- SDValue FIN = DAG.getFrameIndex(FI, MVT::i64);
- ArgValues.push_back(DAG.getLoad(VA.getLocVT(), dl, Root, FIN,
- PseudoSourceValue::getFixedStack(FI), 0));
+ // from this parameter
+ SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
+ ArgValue = DAG.getLoad(LocVT, dl, Chain, FIN,
+ PseudoSourceValue::getFixedStack(FI), 0);
}
- }
- ArgValues.push_back(Root);
+ // If this is an 8/16/32-bit value, it is really passed promoted to 64
+ // bits. Insert an assert[sz]ext to capture this, then truncate to the
+ // right size.
+ if (VA.getLocInfo() == CCValAssign::SExt)
+ ArgValue = DAG.getNode(ISD::AssertSext, dl, LocVT, ArgValue,
+ DAG.getValueType(VA.getValVT()));
+ else if (VA.getLocInfo() == CCValAssign::ZExt)
+ ArgValue = DAG.getNode(ISD::AssertZext, dl, LocVT, ArgValue,
+ DAG.getValueType(VA.getValVT()));
+
+ if (VA.getLocInfo() != CCValAssign::Full)
+ ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
+
+ InVals.push_back(ArgValue);
+ }
- // Return the new list of results.
- return DAG.getNode(ISD::MERGE_VALUES, dl, Op.getNode()->getVTList(),
- &ArgValues[0], ArgValues.size()).getValue(Op.getResNo());
+ return Chain;
}
/// LowerCCCCallTo - functions arguments are copied from virtual regs to
/// (physical regs)/(stack frame), CALLSEQ_START and CALLSEQ_END are emitted.
/// TODO: sret.
-SDValue SystemZTargetLowering::LowerCCCCallTo(SDValue Op, SelectionDAG &DAG,
- unsigned CC) {
- CallSDNode *TheCall = cast<CallSDNode>(Op.getNode());
- SDValue Chain = TheCall->getChain();
- SDValue Callee = TheCall->getCallee();
- bool isVarArg = TheCall->isVarArg();
- DebugLoc dl = Op.getDebugLoc();
+SDValue
+SystemZTargetLowering::LowerCCCCallTo(SDValue Chain, SDValue Callee,
+ CallingConv::ID CallConv, bool isVarArg,
+ bool isTailCall,
+ const SmallVectorImpl<ISD::OutputArg>
+ &Outs,
+ const SmallVectorImpl<ISD::InputArg> &Ins,
+ DebugLoc dl, SelectionDAG &DAG,
+ SmallVectorImpl<SDValue> &InVals) {
+
MachineFunction &MF = DAG.getMachineFunction();
+ // Offset to first argument stack slot.
+ const unsigned FirstArgOffset = 160;
+
// Analyze operands of the call, assigning locations to each operand.
SmallVector<CCValAssign, 16> ArgLocs;
- CCState CCInfo(CC, isVarArg, getTargetMachine(), ArgLocs);
+ CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
+ ArgLocs, *DAG.getContext());
- CCInfo.AnalyzeCallOperands(TheCall, CC_SystemZ);
+ CCInfo.AnalyzeCallOperands(Outs, CC_SystemZ);
// Get a count of how many bytes are to be pushed on the stack.
unsigned NumBytes = CCInfo.getNextStackOffset();
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
- // Arguments start after the 5 first operands of ISD::CALL
- SDValue Arg = TheCall->getArg(i);
+ SDValue Arg = Outs[i].Val;
// Promote the value if needed.
switch (VA.getLocInfo()) {
SystemZ::R11D : SystemZ::R15D),
getPointerTy());
- SDValue PtrOff =
- DAG.getNode(ISD::ADD, dl, getPointerTy(),
- StackPtr,
- DAG.getIntPtrConstant(160+VA.getLocMemOffset()));
+ unsigned Offset = FirstArgOffset + VA.getLocMemOffset();
+ SDValue PtrOff = DAG.getNode(ISD::ADD, dl, getPointerTy(),
+ StackPtr,
+ DAG.getIntPtrConstant(Offset));
MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
- PseudoSourceValue::getStack(),
- VA.getLocMemOffset()));
+ PseudoSourceValue::getStack(), Offset));
}
}
// Handle result values, copying them out of physregs into vregs that we
// return.
- return SDValue(LowerCallResult(Chain, InFlag, TheCall, CC, DAG),
- Op.getResNo());
+ return LowerCallResult(Chain, InFlag, CallConv, isVarArg, Ins, dl,
+ DAG, InVals);
}
-/// LowerCallResult - Lower the result values of an ISD::CALL into the
-/// appropriate copies out of appropriate physical registers. This assumes that
-/// Chain/InFlag are the input chain/flag to use, and that TheCall is the call
-/// being lowered. Returns a SDNode with the same number of values as the
-/// ISD::CALL.
-SDNode*
+/// LowerCallResult - Lower the result values of a call into the
+/// appropriate copies out of appropriate physical registers.
+///
+SDValue
SystemZTargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
- CallSDNode *TheCall,
- unsigned CallingConv,
- SelectionDAG &DAG) {
- bool isVarArg = TheCall->isVarArg();
- DebugLoc dl = TheCall->getDebugLoc();
+ CallingConv::ID CallConv, bool isVarArg,
+ const SmallVectorImpl<ISD::InputArg>
+ &Ins,
+ DebugLoc dl, SelectionDAG &DAG,
+ SmallVectorImpl<SDValue> &InVals) {
// Assign locations to each value returned by this call.
SmallVector<CCValAssign, 16> RVLocs;
- CCState CCInfo(CallingConv, isVarArg, getTargetMachine(), RVLocs);
+ CCState CCInfo(CallConv, isVarArg, getTargetMachine(), RVLocs,
+ *DAG.getContext());
- CCInfo.AnalyzeCallResult(TheCall, RetCC_SystemZ);
- SmallVector<SDValue, 8> ResultVals;
+ CCInfo.AnalyzeCallResult(Ins, RetCC_SystemZ);
// Copy all of the result registers out of their specified physreg.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
- Chain = DAG.getCopyFromReg(Chain, dl, RVLocs[i].getLocReg(),
- RVLocs[i].getValVT(), InFlag).getValue(1);
+ CCValAssign &VA = RVLocs[i];
+
+ Chain = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(),
+ VA.getLocVT(), InFlag).getValue(1);
+ SDValue RetValue = Chain.getValue(0);
InFlag = Chain.getValue(2);
- ResultVals.push_back(Chain.getValue(0));
- }
- ResultVals.push_back(Chain);
+ // If this is an 8/16/32-bit value, it is really passed promoted to 64
+ // bits. Insert an assert[sz]ext to capture this, then truncate to the
+ // right size.
+ if (VA.getLocInfo() == CCValAssign::SExt)
+ RetValue = DAG.getNode(ISD::AssertSext, dl, VA.getLocVT(), RetValue,
+ DAG.getValueType(VA.getValVT()));
+ else if (VA.getLocInfo() == CCValAssign::ZExt)
+ RetValue = DAG.getNode(ISD::AssertZext, dl, VA.getLocVT(), RetValue,
+ DAG.getValueType(VA.getValVT()));
+
+ if (VA.getLocInfo() != CCValAssign::Full)
+ RetValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), RetValue);
+
+ InVals.push_back(RetValue);
+ }
- // Merge everything together with a MERGE_VALUES node.
- return DAG.getNode(ISD::MERGE_VALUES, dl, TheCall->getVTList(),
- &ResultVals[0], ResultVals.size()).getNode();
+ return Chain;
}
-SDValue SystemZTargetLowering::LowerRET(SDValue Op, SelectionDAG &DAG) {
+SDValue
+SystemZTargetLowering::LowerReturn(SDValue Chain,
+ CallingConv::ID CallConv, bool isVarArg,
+ const SmallVectorImpl<ISD::OutputArg> &Outs,
+ DebugLoc dl, SelectionDAG &DAG) {
+
// CCValAssign - represent the assignment of the return value to a location
SmallVector<CCValAssign, 16> RVLocs;
- unsigned CC = DAG.getMachineFunction().getFunction()->getCallingConv();
- bool isVarArg = DAG.getMachineFunction().getFunction()->isVarArg();
- DebugLoc dl = Op.getDebugLoc();
// CCState - Info about the registers and stack slot.
- CCState CCInfo(CC, isVarArg, getTargetMachine(), RVLocs);
+ CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
+ RVLocs, *DAG.getContext());
- // Analize return values of ISD::RET
- CCInfo.AnalyzeReturn(Op.getNode(), RetCC_SystemZ);
+ // Analize return values.
+ CCInfo.AnalyzeReturn(Outs, RetCC_SystemZ);
// If this is the first return lowered for this function, add the regs to the
// liveout set for the function.
DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
}
- // The chain is always operand #0
- SDValue Chain = Op.getOperand(0);
SDValue Flag;
// Copy the result values into the output registers.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
CCValAssign &VA = RVLocs[i];
- SDValue ResValue = Op.getOperand(i*2+1);
+ SDValue ResValue = Outs[i].Val;
assert(VA.isRegLoc() && "Can only return in registers!");
// If this is an 8/16/32-bit value, it is really should be passed promoted
else if (VA.getLocInfo() == CCValAssign::AExt)
ResValue = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), ResValue);
- // ISD::RET => ret chain, (regnum1,val1), ...
- // So i*2+1 index only the regnums
Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), ResValue, Flag);
// Guarantee that all emitted copies are stuck together,
return DAG.getNode(SystemZISD::RET_FLAG, dl, MVT::Other, Chain);
}
+SDValue SystemZTargetLowering::EmitCmp(SDValue LHS, SDValue RHS,
+ ISD::CondCode CC, SDValue &SystemZCC,
+ SelectionDAG &DAG) {
+ // FIXME: Emit a test if RHS is zero
+
+ bool isUnsigned = false;
+ SystemZCC::CondCodes TCC;
+ switch (CC) {
+ default:
+ llvm_unreachable("Invalid integer condition!");
+ case ISD::SETEQ:
+ case ISD::SETOEQ:
+ TCC = SystemZCC::E;
+ break;
+ case ISD::SETUEQ:
+ TCC = SystemZCC::NLH;
+ break;
+ case ISD::SETNE:
+ case ISD::SETONE:
+ TCC = SystemZCC::NE;
+ break;
+ case ISD::SETUNE:
+ TCC = SystemZCC::LH;
+ break;
+ case ISD::SETO:
+ TCC = SystemZCC::O;
+ break;
+ case ISD::SETUO:
+ TCC = SystemZCC::NO;
+ break;
+ case ISD::SETULE:
+ if (LHS.getValueType().isFloatingPoint()) {
+ TCC = SystemZCC::NH;
+ break;
+ }
+ isUnsigned = true; // FALLTHROUGH
+ case ISD::SETLE:
+ case ISD::SETOLE:
+ TCC = SystemZCC::LE;
+ break;
+ case ISD::SETUGE:
+ if (LHS.getValueType().isFloatingPoint()) {
+ TCC = SystemZCC::NL;
+ break;
+ }
+ isUnsigned = true; // FALLTHROUGH
+ case ISD::SETGE:
+ case ISD::SETOGE:
+ TCC = SystemZCC::HE;
+ break;
+ case ISD::SETUGT:
+ if (LHS.getValueType().isFloatingPoint()) {
+ TCC = SystemZCC::NLE;
+ break;
+ }
+ isUnsigned = true; // FALLTHROUGH
+ case ISD::SETGT:
+ case ISD::SETOGT:
+ TCC = SystemZCC::H;
+ break;
+ case ISD::SETULT:
+ if (LHS.getValueType().isFloatingPoint()) {
+ TCC = SystemZCC::NHE;
+ break;
+ }
+ isUnsigned = true; // FALLTHROUGH
+ case ISD::SETLT:
+ case ISD::SETOLT:
+ TCC = SystemZCC::L;
+ break;
+ }
+
+ SystemZCC = DAG.getConstant(TCC, MVT::i32);
+
+ DebugLoc dl = LHS.getDebugLoc();
+ return DAG.getNode((isUnsigned ? SystemZISD::UCMP : SystemZISD::CMP),
+ dl, MVT::i64, LHS, RHS);
+}
+
+
+SDValue SystemZTargetLowering::LowerBR_CC(SDValue Op, SelectionDAG &DAG) {
+ SDValue Chain = Op.getOperand(0);
+ ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(1))->get();
+ SDValue LHS = Op.getOperand(2);
+ SDValue RHS = Op.getOperand(3);
+ SDValue Dest = Op.getOperand(4);
+ DebugLoc dl = Op.getDebugLoc();
+
+ SDValue SystemZCC;
+ SDValue Flag = EmitCmp(LHS, RHS, CC, SystemZCC, DAG);
+ return DAG.getNode(SystemZISD::BRCOND, dl, Op.getValueType(),
+ Chain, Dest, SystemZCC, Flag);
+}
+
+SDValue SystemZTargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) {
+ SDValue LHS = Op.getOperand(0);
+ SDValue RHS = Op.getOperand(1);
+ SDValue TrueV = Op.getOperand(2);
+ SDValue FalseV = Op.getOperand(3);
+ ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(4))->get();
+ DebugLoc dl = Op.getDebugLoc();
+
+ SDValue SystemZCC;
+ SDValue Flag = EmitCmp(LHS, RHS, CC, SystemZCC, DAG);
+
+ SDVTList VTs = DAG.getVTList(Op.getValueType(), MVT::Flag);
+ SmallVector<SDValue, 4> Ops;
+ Ops.push_back(TrueV);
+ Ops.push_back(FalseV);
+ Ops.push_back(SystemZCC);
+ Ops.push_back(Flag);
+
+ return DAG.getNode(SystemZISD::SELECT, dl, VTs, &Ops[0], Ops.size());
+}
+
+SDValue SystemZTargetLowering::LowerGlobalAddress(SDValue Op,
+ SelectionDAG &DAG) {
+ DebugLoc dl = Op.getDebugLoc();
+ GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
+ int64_t Offset = cast<GlobalAddressSDNode>(Op)->getOffset();
+
+ bool IsPic = getTargetMachine().getRelocationModel() == Reloc::PIC_;
+ bool ExtraLoadRequired =
+ Subtarget.GVRequiresExtraLoad(GV, getTargetMachine(), false);
+
+ SDValue Result;
+ if (!IsPic && !ExtraLoadRequired) {
+ Result = DAG.getTargetGlobalAddress(GV, getPointerTy(), Offset);
+ Offset = 0;
+ } else {
+ unsigned char OpFlags = 0;
+ if (ExtraLoadRequired)
+ OpFlags = SystemZII::MO_GOTENT;
+
+ Result = DAG.getTargetGlobalAddress(GV, getPointerTy(), 0, OpFlags);
+ }
+
+ Result = DAG.getNode(SystemZISD::PCRelativeWrapper, dl,
+ getPointerTy(), Result);
+
+ if (ExtraLoadRequired)
+ Result = DAG.getLoad(getPointerTy(), dl, DAG.getEntryNode(), Result,
+ PseudoSourceValue::getGOT(), 0);
+
+ // If there was a non-zero offset that we didn't fold, create an explicit
+ // addition for it.
+ if (Offset != 0)
+ Result = DAG.getNode(ISD::ADD, dl, getPointerTy(), Result,
+ DAG.getConstant(Offset, getPointerTy()));
+
+ return Result;
+}
+
+// FIXME: PIC here
+SDValue SystemZTargetLowering::LowerJumpTable(SDValue Op,
+ SelectionDAG &DAG) {
+ DebugLoc dl = Op.getDebugLoc();
+ JumpTableSDNode *JT = cast<JumpTableSDNode>(Op);
+ SDValue Result = DAG.getTargetJumpTable(JT->getIndex(), getPointerTy());
+
+ return DAG.getNode(SystemZISD::PCRelativeWrapper, dl, getPointerTy(), Result);
+}
+
+
+// FIXME: PIC here
+// FIXME: This is just dirty hack. We need to lower cpool properly
+SDValue SystemZTargetLowering::LowerConstantPool(SDValue Op,
+ SelectionDAG &DAG) {
+ DebugLoc dl = Op.getDebugLoc();
+ ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
+
+ SDValue Result = DAG.getTargetConstantPool(CP->getConstVal(), getPointerTy(),
+ CP->getAlignment(),
+ CP->getOffset());
+
+ return DAG.getNode(SystemZISD::PCRelativeWrapper, dl, getPointerTy(), Result);
+}
+
const char *SystemZTargetLowering::getTargetNodeName(unsigned Opcode) const {
switch (Opcode) {
case SystemZISD::RET_FLAG: return "SystemZISD::RET_FLAG";
case SystemZISD::CALL: return "SystemZISD::CALL";
+ case SystemZISD::BRCOND: return "SystemZISD::BRCOND";
+ case SystemZISD::CMP: return "SystemZISD::CMP";
+ case SystemZISD::UCMP: return "SystemZISD::UCMP";
+ case SystemZISD::SELECT: return "SystemZISD::SELECT";
+ case SystemZISD::PCRelativeWrapper: return "SystemZISD::PCRelativeWrapper";
default: return NULL;
}
}
+//===----------------------------------------------------------------------===//
+// Other Lowering Code
+//===----------------------------------------------------------------------===//
+
+MachineBasicBlock*
+SystemZTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
+ MachineBasicBlock *BB,
+ DenseMap<MachineBasicBlock*, MachineBasicBlock*> *EM) const {
+ const SystemZInstrInfo &TII = *TM.getInstrInfo();
+ DebugLoc dl = MI->getDebugLoc();
+ assert((MI->getOpcode() == SystemZ::Select32 ||
+ MI->getOpcode() == SystemZ::SelectF32 ||
+ MI->getOpcode() == SystemZ::Select64 ||
+ MI->getOpcode() == SystemZ::SelectF64) &&
+ "Unexpected instr type to insert");
+
+ // To "insert" a SELECT 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 I = BB;
+ ++I;
+
+ // thisMBB:
+ // ...
+ // TrueVal = ...
+ // cmpTY ccX, r1, r2
+ // jCC copy1MBB
+ // fallthrough --> copy0MBB
+ MachineBasicBlock *thisMBB = BB;
+ MachineFunction *F = BB->getParent();
+ MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB);
+ MachineBasicBlock *copy1MBB = F->CreateMachineBasicBlock(LLVM_BB);
+ SystemZCC::CondCodes CC = (SystemZCC::CondCodes)MI->getOperand(3).getImm();
+ BuildMI(BB, dl, TII.getBrCond(CC)).addMBB(copy1MBB);
+ F->insert(I, copy0MBB);
+ F->insert(I, copy1MBB);
+ // Inform sdisel of the edge changes.
+ for (MachineBasicBlock::succ_iterator SI = BB->succ_begin(),
+ SE = BB->succ_end(); SI != SE; ++SI)
+ EM->insert(std::make_pair(*SI, copy1MBB));
+ // 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.
+ copy1MBB->transferSuccessors(BB);
+ // Next, add the true and fallthrough blocks as its successors.
+ BB->addSuccessor(copy0MBB);
+ BB->addSuccessor(copy1MBB);
+
+ // copy0MBB:
+ // %FalseValue = ...
+ // # fallthrough to copy1MBB
+ BB = copy0MBB;
+
+ // Update machine-CFG edges
+ BB->addSuccessor(copy1MBB);
+
+ // copy1MBB:
+ // %Result = phi [ %FalseValue, copy0MBB ], [ %TrueValue, thisMBB ]
+ // ...
+ BB = copy1MBB;
+ BuildMI(BB, dl, TII.get(SystemZ::PHI),
+ MI->getOperand(0).getReg())
+ .addReg(MI->getOperand(2).getReg()).addMBB(copy0MBB)
+ .addReg(MI->getOperand(1).getReg()).addMBB(thisMBB);
+
+ F->DeleteMachineInstr(MI); // The pseudo instruction is gone now.
+ return BB;
+}
projects / oota-llvm.git / blobdiff