summaryrefslogtreecommitdiffstats
path: root/lib/Target/PowerPC/PPCISelLowering.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Target/PowerPC/PPCISelLowering.cpp')
-rw-r--r--lib/Target/PowerPC/PPCISelLowering.cpp154
1 files changed, 139 insertions, 15 deletions
diff --git a/lib/Target/PowerPC/PPCISelLowering.cpp b/lib/Target/PowerPC/PPCISelLowering.cpp
index dbb184c..9741a39 100644
--- a/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -125,10 +125,12 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
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);
@@ -215,10 +217,11 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
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.
@@ -1262,9 +1265,107 @@ SDValue PPCTargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const {
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));
+
+ // 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));
- llvm_unreachable("VAARG not yet implemented for the SVR4 ABI!");
- return SDValue(); // Not reached
+ // 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,
@@ -1870,7 +1971,11 @@ PPCTargetLowering::LowerFormalArguments_Darwin(
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(),
@@ -1889,7 +1994,11 @@ PPCTargetLowering::LowerFormalArguments_Darwin(
// 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);
@@ -2902,6 +3011,12 @@ PPCTargetLowering::LowerCall_SVR4(SDValue Chain, SDValue Callee,
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&MemOpChains[0], MemOpChains.size());
+ // Set CR6 to true if this is a vararg call.
+ if (isVarArg) {
+ SDValue SetCR(DAG.getMachineNode(PPC::CRSET, dl, MVT::i32), 0);
+ RegsToPass.push_back(std::make_pair(unsigned(PPC::CR1EQ), SetCR));
+ }
+
// Build a sequence of copy-to-reg nodes chained together with token chain
// and flag operands which copy the outgoing args into the appropriate regs.
SDValue InFlag;
@@ -2911,13 +3026,6 @@ PPCTargetLowering::LowerCall_SVR4(SDValue Chain, SDValue Callee,
InFlag = Chain.getValue(1);
}
- // Set CR6 to true if this is a vararg call.
- if (isVarArg) {
- SDValue SetCR(DAG.getMachineNode(PPC::CRSET, dl, MVT::i32), 0);
- Chain = DAG.getCopyToReg(Chain, dl, PPC::CR1EQ, SetCR, InFlag);
- InFlag = Chain.getValue(1);
- }
-
if (isTailCall)
PrepareTailCall(DAG, InFlag, Chain, dl, false, SPDiff, NumBytes, LROp, FPOp,
false, TailCallArguments);
@@ -4422,11 +4530,27 @@ SDValue PPCTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
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);
@@ -4676,7 +4800,7 @@ PPCTargetLowering::EmitPartwordAtomicBinary(MachineInstr *MI,
.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);
OpenPOWER on IntegriCloud