#include "PPCISelLowering.h"
#include "PPCMachineFunctionInfo.h"
#include "PPCPerfectShuffle.h"
-#include "PPCPredicates.h"
#include "PPCTargetMachine.h"
+#include "MCTargetDesc/PPCPredicates.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/VectorExtras.h"
#include "llvm/CodeGen/CallingConvLower.h"
setOperationAction(ISD::FCOS , MVT::f64, Expand);
setOperationAction(ISD::FREM , MVT::f64, Expand);
setOperationAction(ISD::FPOW , MVT::f64, Expand);
+ setOperationAction(ISD::FMA , MVT::f64, Expand);
setOperationAction(ISD::FSIN , MVT::f32, Expand);
setOperationAction(ISD::FCOS , MVT::f32, Expand);
setOperationAction(ISD::FREM , MVT::f32, Expand);
setOperationAction(ISD::FPOW , MVT::f32, Expand);
+ setOperationAction(ISD::FMA , MVT::f32, Expand);
setOperationAction(ISD::FLT_ROUNDS_, MVT::i32, Custom);
setOperationAction(ISD::TRAP, MVT::Other, Legal);
// TRAMPOLINE is custom lowered.
- setOperationAction(ISD::TRAMPOLINE, MVT::Other, Custom);
+ setOperationAction(ISD::INIT_TRAMPOLINE, MVT::Other, Custom);
+ setOperationAction(ISD::ADJUST_TRAMPOLINE, MVT::Other, Custom);
// VASTART needs to be custom lowered to use the VarArgsFrameIndex
setOperationAction(ISD::VASTART , MVT::Other, Custom);
// VAARG is custom lowered with the 32-bit SVR4 ABI.
- if (
TM.getSubtarget<PPCSubtarget>().isSVR4ABI()
- && !TM.getSubtarget<PPCSubtarget>().isPPC64())
+ if (TM.getSubtarget<PPCSubtarget>().isSVR4ABI()
+ && !TM.getSubtarget<PPCSubtarget>().isPPC64()) {
setOperationAction(ISD::VAARG, MVT::Other, Custom);
- else
+ setOperationAction(ISD::VAARG, MVT::i64, Custom);
+ } else
setOperationAction(ISD::VAARG, MVT::Other, Expand);
// Use the default implementation.
setOperationAction(ISD::BUILD_VECTOR, MVT::v4f32, Custom);
}
+ setOperationAction(ISD::ATOMIC_LOAD, MVT::i32, Expand);
+ setOperationAction(ISD::ATOMIC_STORE, MVT::i32, Expand);
+
setBooleanContents(ZeroOrOneBooleanContent);
+ setBooleanVectorContents(ZeroOrOneBooleanContent); // FIXME: Is this correct?
if (TM.getSubtarget<PPCSubtarget>().isPPC64()) {
setStackPointerRegisterToSaveRestore(PPC::X1);
if (PPCSubTarget.isDarwin())
setPrefFunctionAlignment(4);
+ setInsertFencesForAtomic(true);
+
computeRegisterProperties();
}
/// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
/// function arguments in the caller parameter area.
-unsigned PPCTargetLowering::getByValTypeAlignment(const Type *Ty) const {
+unsigned PPCTargetLowering::getByValTypeAlignment(Type *Ty) const {
const TargetMachine &TM = getTargetMachine();
// Darwin passes everything on 4 byte boundary.
if (TM.getSubtarget<PPCSubtarget>().isDarwin())
}
}
-MVT::SimpleValueType PPCTargetLowering::getSetCCResultType(EVT VT) const {
+EVT PPCTargetLowering::getSetCCResultType(EVT VT) const {
return MVT::i32;
}
SDValue PPCTargetLowering::LowerVAARG(SDValue Op, SelectionDAG &DAG,
const PPCSubtarget &Subtarget) const {
+ SDNode *Node = Op.getNode();
+ EVT VT = Node->getValueType(0);
+ EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ SDValue InChain = Node->getOperand(0);
+ SDValue VAListPtr = Node->getOperand(1);
+ const Value *SV = cast<SrcValueSDNode>(Node->getOperand(2))->getValue();
+ DebugLoc dl = Node->getDebugLoc();
+
+ assert(!Subtarget.isPPC64() && "LowerVAARG is PPC32 only");
+
+ // gpr_index
+ SDValue GprIndex = DAG.getExtLoad(ISD::ZEXTLOAD, dl, MVT::i32, InChain,
+ VAListPtr, MachinePointerInfo(SV), MVT::i8,
+ false, false, 0);
+ InChain = GprIndex.getValue(1);
+
+ if (VT == MVT::i64) {
+ // Check if GprIndex is even
+ SDValue GprAnd = DAG.getNode(ISD::AND, dl, MVT::i32, GprIndex,
+ DAG.getConstant(1, MVT::i32));
+ SDValue CC64 = DAG.getSetCC(dl, MVT::i32, GprAnd,
+ DAG.getConstant(0, MVT::i32), ISD::SETNE);
+ SDValue GprIndexPlusOne = DAG.getNode(ISD::ADD, dl, MVT::i32, GprIndex,
+ DAG.getConstant(1, MVT::i32));
+ // Align GprIndex to be even if it isn't
+ GprIndex = DAG.getNode(ISD::SELECT, dl, MVT::i32, CC64, GprIndexPlusOne,
+ GprIndex);
+ }
+
+ // fpr index is 1 byte after gpr
+ SDValue FprPtr = DAG.getNode(ISD::ADD, dl, PtrVT, VAListPtr,
+ DAG.getConstant(1, MVT::i32));
- llvm_unreachable("VAARG not yet implemented for the SVR4 ABI!");
- return SDValue(); // Not reached
+ // fpr
+ SDValue FprIndex = DAG.getExtLoad(ISD::ZEXTLOAD, dl, MVT::i32, InChain,
+ FprPtr, MachinePointerInfo(SV), MVT::i8,
+ false, false, 0);
+ InChain = FprIndex.getValue(1);
+
+ SDValue RegSaveAreaPtr = DAG.getNode(ISD::ADD, dl, PtrVT, VAListPtr,
+ DAG.getConstant(8, MVT::i32));
+
+ SDValue OverflowAreaPtr = DAG.getNode(ISD::ADD, dl, PtrVT, VAListPtr,
+ DAG.getConstant(4, MVT::i32));
+
+ // areas
+ SDValue OverflowArea = DAG.getLoad(MVT::i32, dl, InChain, OverflowAreaPtr,
+ MachinePointerInfo(), false, false, 0);
+ InChain = OverflowArea.getValue(1);
+
+ SDValue RegSaveArea = DAG.getLoad(MVT::i32, dl, InChain, RegSaveAreaPtr,
+ MachinePointerInfo(), false, false, 0);
+ InChain = RegSaveArea.getValue(1);
+
+ // select overflow_area if index > 8
+ SDValue CC = DAG.getSetCC(dl, MVT::i32, VT.isInteger() ? GprIndex : FprIndex,
+ DAG.getConstant(8, MVT::i32), ISD::SETLT);
+
+ // adjustment constant gpr_index * 4/8
+ SDValue RegConstant = DAG.getNode(ISD::MUL, dl, MVT::i32,
+ VT.isInteger() ? GprIndex : FprIndex,
+ DAG.getConstant(VT.isInteger() ? 4 : 8,
+ MVT::i32));
+
+ // OurReg = RegSaveArea + RegConstant
+ SDValue OurReg = DAG.getNode(ISD::ADD, dl, PtrVT, RegSaveArea,
+ RegConstant);
+
+ // Floating types are 32 bytes into RegSaveArea
+ if (VT.isFloatingPoint())
+ OurReg = DAG.getNode(ISD::ADD, dl, PtrVT, OurReg,
+ DAG.getConstant(32, MVT::i32));
+
+ // increase {f,g}pr_index by 1 (or 2 if VT is i64)
+ SDValue IndexPlus1 = DAG.getNode(ISD::ADD, dl, MVT::i32,
+ VT.isInteger() ? GprIndex : FprIndex,
+ DAG.getConstant(VT == MVT::i64 ? 2 : 1,
+ MVT::i32));
+
+ InChain = DAG.getTruncStore(InChain, dl, IndexPlus1,
+ VT.isInteger() ? VAListPtr : FprPtr,
+ MachinePointerInfo(SV),
+ MVT::i8, false, false, 0);
+
+ // determine if we should load from reg_save_area or overflow_area
+ SDValue Result = DAG.getNode(ISD::SELECT, dl, PtrVT, CC, OurReg, OverflowArea);
+
+ // increase overflow_area by 4/8 if gpr/fpr > 8
+ SDValue OverflowAreaPlusN = DAG.getNode(ISD::ADD, dl, PtrVT, OverflowArea,
+ DAG.getConstant(VT.isInteger() ? 4 : 8,
+ MVT::i32));
+
+ OverflowArea = DAG.getNode(ISD::SELECT, dl, MVT::i32, CC, OverflowArea,
+ OverflowAreaPlusN);
+
+ InChain = DAG.getTruncStore(InChain, dl, OverflowArea,
+ OverflowAreaPtr,
+ MachinePointerInfo(),
+ MVT::i32, false, false, 0);
+
+ return DAG.getLoad(VT, dl, InChain, Result, MachinePointerInfo(), false, false, 0);
}
-SDValue PPCTargetLowering::LowerTRAMPOLINE(SDValue Op,
- SelectionDAG &DAG) const {
+SDValue PPCTargetLowering::LowerADJUST_TRAMPOLINE(SDValue Op,
+ SelectionDAG &DAG) const {
+ return Op.getOperand(0);
+}
+
+SDValue PPCTargetLowering::LowerINIT_TRAMPOLINE(SDValue Op,
+ SelectionDAG &DAG) const {
SDValue Chain = Op.getOperand(0);
SDValue Trmp = Op.getOperand(1); // trampoline
SDValue FPtr = Op.getOperand(2); // nested function
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = (PtrVT == MVT::i64);
- const Type *IntPtrTy =
+ Type *IntPtrTy =
DAG.getTargetLoweringInfo().getTargetData()->getIntPtrType(
*DAG.getContext());
// Lower to a call to __trampoline_setup(Trmp, TrampSize, FPtr, ctx_reg)
std::pair<SDValue, SDValue> CallResult =
- LowerCallTo(Chain, Op.getValueType().getTypeForEVT(*DAG.getContext()),
+ LowerCallTo(Chain, Type::getVoidTy(*DAG.getContext()),
false, false, false, false, 0, CallingConv::C, false,
/*isReturnValueUsed=*/true,
DAG.getExternalSymbol("__trampoline_setup", PtrVT),
Args, DAG, dl);
- SDValue Ops[] =
- { CallResult.first, CallResult.second };
-
- return DAG.getMergeValues(Ops, 2, dl);
+ return CallResult.second;
}
SDValue PPCTargetLowering::LowerVASTART(SDValue Op, SelectionDAG &DAG,
InVals.push_back(FIN);
if (ObjSize==1 || ObjSize==2) {
if (GPR_idx != Num_GPR_Regs) {
- unsigned VReg = MF.addLiveIn(GPR[GPR_idx], &PPC::GPRCRegClass);
+ unsigned VReg;
+ if (isPPC64)
+ VReg = MF.addLiveIn(GPR[GPR_idx], &PPC::G8RCRegClass);
+ else
+ VReg = MF.addLiveIn(GPR[GPR_idx], &PPC::GPRCRegClass);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, PtrVT);
SDValue Store = DAG.getTruncStore(Val.getValue(1), dl, Val, FIN,
MachinePointerInfo(),
// to memory. ArgVal will be address of the beginning of
// the object.
if (GPR_idx != Num_GPR_Regs) {
- unsigned VReg = MF.addLiveIn(GPR[GPR_idx], &PPC::GPRCRegClass);
+ unsigned VReg;
+ if (isPPC64)
+ VReg = MF.addLiveIn(GPR[GPR_idx], &PPC::G8RCRegClass);
+ else
+ VReg = MF.addLiveIn(GPR[GPR_idx], &PPC::GPRCRegClass);
int FI = MFI->CreateFixedObject(PtrByteSize, ArgOffset, true);
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, PtrVT);
if (!DAG.getTarget().getSubtarget<PPCSubtarget>().isJITCodeModel()) {
unsigned OpFlags = 0;
if (DAG.getTarget().getRelocationModel() != Reloc::Static &&
- (!PPCSubTarget.getTargetTriple().isMacOSX() ||
+ (PPCSubTarget.getTargetTriple().isMacOSX() &&
PPCSubTarget.getTargetTriple().isMacOSXVersionLT(10, 5)) &&
(G->getGlobal()->isDeclaration() ||
G->getGlobal()->isWeakForLinker())) {
unsigned char OpFlags = 0;
if (DAG.getTarget().getRelocationModel() != Reloc::Static &&
- (!PPCSubTarget.getTargetTriple().isMacOSX() ||
+ (PPCSubTarget.getTargetTriple().isMacOSX() &&
PPCSubTarget.getTargetTriple().isMacOSXVersionLT(10, 5))) {
// PC-relative references to external symbols should go through $stub,
// unless we're building with the leopard linker or later, which
SmallVector<TailCallArgumentInfo, 8> TailCallArguments;
SmallVector<SDValue, 8> MemOpChains;
+ bool seenFloatArg = false;
// Walk the register/memloc assignments, inserting copies/loads.
for (unsigned i = 0, j = 0, e = ArgLocs.size();
i != e;
}
if (VA.isRegLoc()) {
+ seenFloatArg |= VA.getLocVT().isFloatingPoint();
// Put argument in a physical register.
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
} else {
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&MemOpChains[0], MemOpChains.size());
- // Set CR6 to true if this is a vararg call.
+ // Set CR6 to true if this is a vararg call with floating args passed in
+ // registers.
if (isVarArg) {
- SDValue SetCR(DAG.getMachineNode(PPC::CRSET, dl, MVT::i32), 0);
+ SDValue SetCR(DAG.getMachineNode(seenFloatArg ? PPC::CRSET : PPC::CRUNSET,
+ dl, MVT::i32), 0);
RegsToPass.push_back(std::make_pair(unsigned(PPC::CR1EQ), SetCR));
}
case ISD::GlobalTLSAddress: llvm_unreachable("TLS not implemented for PPC");
case ISD::JumpTable: return LowerJumpTable(Op, DAG);
case ISD::SETCC: return LowerSETCC(Op, DAG);
- case ISD::TRAMPOLINE: return LowerTRAMPOLINE(Op, DAG);
+ case ISD::INIT_TRAMPOLINE: return LowerINIT_TRAMPOLINE(Op, DAG);
+ case ISD::ADJUST_TRAMPOLINE: return LowerADJUST_TRAMPOLINE(Op, DAG);
case ISD::VASTART:
return LowerVASTART(Op, DAG, PPCSubTarget);
void PPCTargetLowering::ReplaceNodeResults(SDNode *N,
SmallVectorImpl<SDValue>&Results,
SelectionDAG &DAG) const {
+ const TargetMachine &TM = getTargetMachine();
DebugLoc dl = N->getDebugLoc();
switch (N->getOpcode()) {
default:
assert(false && "Do not know how to custom type legalize this operation!");
return;
+ case ISD::VAARG: {
+ if (!TM.getSubtarget<PPCSubtarget>().isSVR4ABI()
+ || TM.getSubtarget<PPCSubtarget>().isPPC64())
+ return;
+
+ EVT VT = N->getValueType(0);
+
+ if (VT == MVT::i64) {
+ SDValue NewNode = LowerVAARG(SDValue(N, 1), DAG, PPCSubTarget);
+
+ Results.push_back(NewNode);
+ Results.push_back(NewNode.getValue(1));
+ }
+ return;
+ }
case ISD::FP_ROUND_INREG: {
assert(N->getValueType(0) == MVT::ppcf128);
assert(N->getOperand(0).getValueType() == MVT::ppcf128);
.addReg(TmpReg).addReg(MaskReg);
BuildMI(BB, dl, TII->get(is64bit ? PPC::OR8 : PPC::OR), Tmp4Reg)
.addReg(Tmp3Reg).addReg(Tmp2Reg);
- BuildMI(BB, dl, TII->get(PPC::STWCX))
+ BuildMI(BB, dl, TII->get(is64bit ? PPC::STDCX : PPC::STWCX))
.addReg(Tmp4Reg).addReg(ZeroReg).addReg(PtrReg);
BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(loopMBB);
// but allow it at the lowest weight.
if (CallOperandVal == NULL)
return CW_Default;
- const Type *type = CallOperandVal->getType();
+ Type *type = CallOperandVal->getType();
// Look at the constraint type.
switch (*constraint) {
default:
// isLegalAddressingMode - Return true if the addressing mode represented
// by AM is legal for this target, for a load/store of the specified type.
bool PPCTargetLowering::isLegalAddressingMode(const AddrMode &AM,
- const Type *Ty) const {
+ Type *Ty) const {
// FIXME: PPC does not allow r+i addressing modes for vectors!
// PPC allows a sign-extended 16-bit immediate field.
/// isLegalAddressImmediate - Return true if the integer value can be used
/// as the offset of the target addressing mode for load / store of the
/// given type.
-bool PPCTargetLowering::isLegalAddressImmediate(int64_t V,const Type *Ty) const{
+bool PPCTargetLowering::isLegalAddressImmediate(int64_t V,Type *Ty) const{
// PPC allows a sign-extended 16-bit immediate field.
return (V > -(1 << 16) && V < (1 << 16)-1);
}
projects / oota-llvm.git / blobdiff