//===----------------------------------------------------------------------===//
#include "PPCISelLowering.h"
+#include "MCTargetDesc/PPCPredicates.h"
#include "PPCMachineFunctionInfo.h"
#include "PPCPerfectShuffle.h"
#include "PPCTargetMachine.h"
-#include "MCTargetDesc/PPCPredicates.h"
-#include "llvm/CallingConv.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/Intrinsics.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/CodeGen/CallingConvLower.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Intrinsics.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Target/TargetOptions.h"
using namespace llvm;
-static bool CC_PPC_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
- CCValAssign::LocInfo &LocInfo,
- ISD::ArgFlagsTy &ArgFlags,
- CCState &State);
-static bool CC_PPC_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
- MVT &LocVT,
- CCValAssign::LocInfo &LocInfo,
- ISD::ArgFlagsTy &ArgFlags,
- CCState &State);
-static bool CC_PPC_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
+static bool CC_PPC32_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+ CCValAssign::LocInfo &LocInfo,
+ ISD::ArgFlagsTy &ArgFlags,
+ CCState &State);
+static bool CC_PPC32_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
MVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State);
+static bool CC_PPC32_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
+ MVT &LocVT,
+ CCValAssign::LocInfo &LocInfo,
+ ISD::ArgFlagsTy &ArgFlags,
+ CCState &State);
static cl::opt<bool> DisablePPCPreinc("disable-ppc-preinc",
cl::desc("disable preincrement load/store generation on PPC"), cl::Hidden);
// We don't support sin/cos/sqrt/fmod/pow
setOperationAction(ISD::FSIN , MVT::f64, Expand);
setOperationAction(ISD::FCOS , MVT::f64, Expand);
+ setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
setOperationAction(ISD::FREM , MVT::f64, Expand);
setOperationAction(ISD::FPOW , MVT::f64, Expand);
setOperationAction(ISD::FMA , MVT::f64, Legal);
setOperationAction(ISD::FSIN , MVT::f32, Expand);
setOperationAction(ISD::FCOS , MVT::f32, Expand);
+ setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
setOperationAction(ISD::FREM , MVT::f32, Expand);
setOperationAction(ISD::FPOW , MVT::f32, Expand);
setOperationAction(ISD::FMA , MVT::f32, Legal);
setOperationAction(ISD::UREM, VT, Expand);
setOperationAction(ISD::FDIV, VT, Expand);
setOperationAction(ISD::FNEG, VT, Expand);
+ setOperationAction(ISD::FSQRT, VT, Expand);
+ setOperationAction(ISD::FLOG, VT, Expand);
+ setOperationAction(ISD::FLOG10, VT, Expand);
+ setOperationAction(ISD::FLOG2, VT, Expand);
+ setOperationAction(ISD::FEXP, VT, Expand);
+ setOperationAction(ISD::FEXP2, VT, Expand);
+ setOperationAction(ISD::FSIN, VT, Expand);
+ setOperationAction(ISD::FCOS, VT, Expand);
+ setOperationAction(ISD::FABS, VT, Expand);
+ setOperationAction(ISD::FPOWI, VT, Expand);
+ setOperationAction(ISD::FFLOOR, VT, Expand);
+ setOperationAction(ISD::FCEIL, VT, Expand);
+ setOperationAction(ISD::FTRUNC, VT, Expand);
+ setOperationAction(ISD::FRINT, VT, Expand);
+ setOperationAction(ISD::FNEARBYINT, VT, Expand);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Expand);
setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Expand);
setOperationAction(ISD::BUILD_VECTOR, VT, Expand);
setOperationAction(ISD::CTLZ_ZERO_UNDEF, VT, Expand);
setOperationAction(ISD::CTTZ, VT, Expand);
setOperationAction(ISD::CTTZ_ZERO_UNDEF, VT, Expand);
+ setOperationAction(ISD::VSELECT, VT, Expand);
+ setOperationAction(ISD::SIGN_EXTEND_INREG, VT, Expand);
+
+ for (unsigned j = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
+ j <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++j) {
+ MVT::SimpleValueType InnerVT = (MVT::SimpleValueType)j;
+ setTruncStoreAction(VT, InnerVT, Expand);
+ }
+ setLoadExtAction(ISD::SEXTLOAD, VT, Expand);
+ setLoadExtAction(ISD::ZEXTLOAD, VT, Expand);
+ setLoadExtAction(ISD::EXTLOAD, VT, Expand);
}
// We can custom expand all VECTOR_SHUFFLEs to VPERM, others we can handle
setOperationAction(ISD::FP_TO_UINT, MVT::v4i32, Legal);
setOperationAction(ISD::SINT_TO_FP, MVT::v4i32, Legal);
setOperationAction(ISD::UINT_TO_FP, MVT::v4i32, Legal);
+ setOperationAction(ISD::FFLOOR, MVT::v4f32, Legal);
+ setOperationAction(ISD::FCEIL, MVT::v4f32, Legal);
+ setOperationAction(ISD::FTRUNC, MVT::v4f32, Legal);
+ setOperationAction(ISD::FNEARBYINT, MVT::v4f32, Legal);
addRegisterClass(MVT::v4f32, &PPC::VRRCRegClass);
addRegisterClass(MVT::v4i32, &PPC::VRRCRegClass);
setOperationAction(ISD::BUILD_VECTOR, MVT::v8i16, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v4i32, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v4f32, Custom);
+
+ // Altivec does not contain unordered floating-point compare instructions
+ setCondCodeAction(ISD::SETUO, MVT::v4f32, Expand);
+ setCondCodeAction(ISD::SETUEQ, MVT::v4f32, Expand);
+ setCondCodeAction(ISD::SETUGT, MVT::v4f32, Expand);
+ setCondCodeAction(ISD::SETUGE, MVT::v4f32, Expand);
+ setCondCodeAction(ISD::SETULT, MVT::v4f32, Expand);
+ setCondCodeAction(ISD::SETULE, MVT::v4f32, Expand);
}
if (Subtarget->has64BitSupport()) {
setOperationAction(ISD::ATOMIC_LOAD, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_STORE, MVT::i32, Expand);
+ setOperationAction(ISD::ATOMIC_LOAD, MVT::i64, Expand);
+ setOperationAction(ISD::ATOMIC_STORE, MVT::i64, Expand);
setBooleanContents(ZeroOrOneBooleanContent);
setBooleanVectorContents(ZeroOrOneBooleanContent); // FIXME: Is this correct?
// friends. Gcc uses same threshold of 128 bytes (= 32 word stores).
if (Subtarget->getDarwinDirective() == PPC::DIR_E500mc ||
Subtarget->getDarwinDirective() == PPC::DIR_E5500) {
- maxStoresPerMemset = 32;
- maxStoresPerMemsetOptSize = 16;
- maxStoresPerMemcpy = 32;
- maxStoresPerMemcpyOptSize = 8;
- maxStoresPerMemmove = 32;
- maxStoresPerMemmoveOptSize = 8;
+ MaxStoresPerMemset = 32;
+ MaxStoresPerMemsetOptSize = 16;
+ MaxStoresPerMemcpy = 32;
+ MaxStoresPerMemcpyOptSize = 8;
+ MaxStoresPerMemmove = 32;
+ MaxStoresPerMemmoveOptSize = 8;
setPrefFunctionAlignment(4);
- benefitFromCodePlacementOpt = true;
+ BenefitFromCodePlacementOpt = true;
}
}
case PPCISD::TC_RETURN: return "PPCISD::TC_RETURN";
case PPCISD::CR6SET: return "PPCISD::CR6SET";
case PPCISD::CR6UNSET: return "PPCISD::CR6UNSET";
+ case PPCISD::ADDIS_TOC_HA: return "PPCISD::ADDIS_TOC_HA";
+ case PPCISD::LD_TOC_L: return "PPCISD::LD_TOC_L";
+ case PPCISD::ADDI_TOC_L: return "PPCISD::ADDI_TOC_L";
+ case PPCISD::ADDIS_GOT_TPREL_HA: return "PPCISD::ADDIS_GOT_TPREL_HA";
+ case PPCISD::LD_GOT_TPREL_L: return "PPCISD::LD_GOT_TPREL_L";
+ case PPCISD::ADD_TLS: return "PPCISD::ADD_TLS";
+ case PPCISD::ADDIS_TLSGD_HA: return "PPCISD::ADDIS_TLSGD_HA";
+ case PPCISD::ADDI_TLSGD_L: return "PPCISD::ADDI_TLSGD_L";
+ case PPCISD::GET_TLS_ADDR: return "PPCISD::GET_TLS_ADDR";
+ case PPCISD::ADDIS_TLSLD_HA: return "PPCISD::ADDIS_TLSLD_HA";
+ case PPCISD::ADDI_TLSLD_L: return "PPCISD::ADDI_TLSLD_L";
+ case PPCISD::GET_TLSLD_ADDR: return "PPCISD::GET_TLSLD_ADDR";
+ case PPCISD::ADDIS_DTPREL_HA: return "PPCISD::ADDIS_DTPREL_HA";
+ case PPCISD::ADDI_DTPREL_L: return "PPCISD::ADDI_DTPREL_L";
+ case PPCISD::VADD_SPLAT: return "PPCISD::VADD_SPLAT";
}
}
EVT PtrVT = getPointerTy();
bool is64bit = PPCSubTarget.isPPC64();
- TLSModel::Model model = getTargetMachine().getTLSModel(GV);
+ TLSModel::Model Model = getTargetMachine().getTLSModel(GV);
+
+ if (Model == TLSModel::LocalExec) {
+ SDValue TGAHi = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
+ PPCII::MO_TPREL16_HA);
+ SDValue TGALo = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
+ PPCII::MO_TPREL16_LO);
+ SDValue TLSReg = DAG.getRegister(is64bit ? PPC::X13 : PPC::R2,
+ is64bit ? MVT::i64 : MVT::i32);
+ SDValue Hi = DAG.getNode(PPCISD::Hi, dl, PtrVT, TGAHi, TLSReg);
+ return DAG.getNode(PPCISD::Lo, dl, PtrVT, TGALo, Hi);
+ }
- SDValue TGAHi = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
- PPCII::MO_TPREL16_HA);
- SDValue TGALo = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
- PPCII::MO_TPREL16_LO);
+ if (!is64bit)
+ llvm_unreachable("only local-exec is currently supported for ppc32");
+
+ if (Model == TLSModel::InitialExec) {
+ SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, 0);
+ SDValue GOTReg = DAG.getRegister(PPC::X2, MVT::i64);
+ SDValue TPOffsetHi = DAG.getNode(PPCISD::ADDIS_GOT_TPREL_HA, dl,
+ PtrVT, GOTReg, TGA);
+ SDValue TPOffset = DAG.getNode(PPCISD::LD_GOT_TPREL_L, dl,
+ PtrVT, TGA, TPOffsetHi);
+ return DAG.getNode(PPCISD::ADD_TLS, dl, PtrVT, TPOffset, TGA);
+ }
- if (model != TLSModel::LocalExec)
- llvm_unreachable("only local-exec TLS mode supported");
- SDValue TLSReg = DAG.getRegister(is64bit ? PPC::X13 : PPC::R2,
- is64bit ? MVT::i64 : MVT::i32);
- SDValue Hi = DAG.getNode(PPCISD::Hi, dl, PtrVT, TGAHi, TLSReg);
- return DAG.getNode(PPCISD::Lo, dl, PtrVT, TGALo, Hi);
+ if (Model == TLSModel::GeneralDynamic) {
+ SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, 0);
+ SDValue GOTReg = DAG.getRegister(PPC::X2, MVT::i64);
+ SDValue GOTEntryHi = DAG.getNode(PPCISD::ADDIS_TLSGD_HA, dl, PtrVT,
+ GOTReg, TGA);
+ SDValue GOTEntry = DAG.getNode(PPCISD::ADDI_TLSGD_L, dl, PtrVT,
+ GOTEntryHi, TGA);
+
+ // We need a chain node, and don't have one handy. The underlying
+ // call has no side effects, so using the function entry node
+ // suffices.
+ SDValue Chain = DAG.getEntryNode();
+ Chain = DAG.getCopyToReg(Chain, dl, PPC::X3, GOTEntry);
+ SDValue ParmReg = DAG.getRegister(PPC::X3, MVT::i64);
+ SDValue TLSAddr = DAG.getNode(PPCISD::GET_TLS_ADDR, dl,
+ PtrVT, ParmReg, TGA);
+ // The return value from GET_TLS_ADDR really is in X3 already, but
+ // some hacks are needed here to tie everything together. The extra
+ // copies dissolve during subsequent transforms.
+ Chain = DAG.getCopyToReg(Chain, dl, PPC::X3, TLSAddr);
+ return DAG.getCopyFromReg(Chain, dl, PPC::X3, PtrVT);
+ }
+
+ if (Model == TLSModel::LocalDynamic) {
+ SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, 0);
+ SDValue GOTReg = DAG.getRegister(PPC::X2, MVT::i64);
+ SDValue GOTEntryHi = DAG.getNode(PPCISD::ADDIS_TLSLD_HA, dl, PtrVT,
+ GOTReg, TGA);
+ SDValue GOTEntry = DAG.getNode(PPCISD::ADDI_TLSLD_L, dl, PtrVT,
+ GOTEntryHi, TGA);
+
+ // We need a chain node, and don't have one handy. The underlying
+ // call has no side effects, so using the function entry node
+ // suffices.
+ SDValue Chain = DAG.getEntryNode();
+ Chain = DAG.getCopyToReg(Chain, dl, PPC::X3, GOTEntry);
+ SDValue ParmReg = DAG.getRegister(PPC::X3, MVT::i64);
+ SDValue TLSAddr = DAG.getNode(PPCISD::GET_TLSLD_ADDR, dl,
+ PtrVT, ParmReg, TGA);
+ // The return value from GET_TLSLD_ADDR really is in X3 already, but
+ // some hacks are needed here to tie everything together. The extra
+ // copies dissolve during subsequent transforms.
+ Chain = DAG.getCopyToReg(Chain, dl, PPC::X3, TLSAddr);
+ SDValue DtvOffsetHi = DAG.getNode(PPCISD::ADDIS_DTPREL_HA, dl, PtrVT,
+ Chain, ParmReg, TGA);
+ return DAG.getNode(PPCISD::ADDI_DTPREL_L, dl, PtrVT, DtvOffsetHi, TGA);
+ }
+
+ llvm_unreachable("Unknown TLS model!");
}
SDValue PPCTargetLowering::LowerGlobalAddress(SDValue Op,
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = (PtrVT == MVT::i64);
- unsigned AS = 0;
Type *IntPtrTy =
DAG.getTargetLoweringInfo().getDataLayout()->getIntPtrType(
- *DAG.getContext(), AS);
+ *DAG.getContext());
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
#include "PPCGenCallingConv.inc"
-static bool CC_PPC_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
- CCValAssign::LocInfo &LocInfo,
- ISD::ArgFlagsTy &ArgFlags,
- CCState &State) {
+static bool CC_PPC32_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+ CCValAssign::LocInfo &LocInfo,
+ ISD::ArgFlagsTy &ArgFlags,
+ CCState &State) {
return true;
}
-static bool CC_PPC_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
- MVT &LocVT,
- CCValAssign::LocInfo &LocInfo,
- ISD::ArgFlagsTy &ArgFlags,
- CCState &State) {
+static bool CC_PPC32_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
+ MVT &LocVT,
+ CCValAssign::LocInfo &LocInfo,
+ ISD::ArgFlagsTy &ArgFlags,
+ CCState &State) {
static const uint16_t ArgRegs[] = {
PPC::R3, PPC::R4, PPC::R5, PPC::R6,
PPC::R7, PPC::R8, PPC::R9, PPC::R10,
return false;
}
-static bool CC_PPC_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
- MVT &LocVT,
- CCValAssign::LocInfo &LocInfo,
- ISD::ArgFlagsTy &ArgFlags,
- CCState &State) {
+static bool CC_PPC32_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
+ MVT &LocVT,
+ CCValAssign::LocInfo &LocInfo,
+ ISD::ArgFlagsTy &ArgFlags,
+ CCState &State) {
static const uint16_t ArgRegs[] = {
PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5, PPC::F6, PPC::F7,
PPC::F8
// Reserve space for the linkage area on the stack.
CCInfo.AllocateStack(PPCFrameLowering::getLinkageSize(false, false), PtrByteSize);
- CCInfo.AnalyzeFormalArguments(Ins, CC_PPC_SVR4);
+ CCInfo.AnalyzeFormalArguments(Ins, CC_PPC32_SVR4);
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
// Reserve stack space for the allocations in CCInfo.
CCByValInfo.AllocateStack(CCInfo.getNextStackOffset(), PtrByteSize);
- CCByValInfo.AnalyzeFormalArguments(Ins, CC_PPC_SVR4_ByVal);
+ CCByValInfo.AnalyzeFormalArguments(Ins, CC_PPC32_SVR4_ByVal);
// Area that is at least reserved in the caller of this function.
unsigned MinReservedArea = CCByValInfo.getNextStackOffset();
SmallVector<SDValue, 8> MemOps;
unsigned nAltivecParamsAtEnd = 0;
Function::const_arg_iterator FuncArg = MF.getFunction()->arg_begin();
- for (unsigned ArgNo = 0, e = Ins.size(); ArgNo != e; ++ArgNo, ++FuncArg) {
+ unsigned CurArgIdx = 0;
+ for (unsigned ArgNo = 0, e = Ins.size(); ArgNo != e; ++ArgNo) {
SDValue ArgVal;
bool needsLoad = false;
EVT ObjectVT = Ins[ArgNo].VT;
unsigned ObjSize = ObjectVT.getSizeInBits()/8;
unsigned ArgSize = ObjSize;
ISD::ArgFlagsTy Flags = Ins[ArgNo].Flags;
+ std::advance(FuncArg, Ins[ArgNo].OrigArgIndex - CurArgIdx);
+ CurArgIdx = Ins[ArgNo].OrigArgIndex;
unsigned CurArgOffset = ArgOffset;
// ObjSize is the true size, ArgSize rounded up to multiple of registers.
ObjSize = Flags.getByValSize();
ArgSize = ((ObjSize + PtrByteSize - 1)/PtrByteSize) * PtrByteSize;
+ // Empty aggregate parameters do not take up registers. Examples:
+ // struct { } a;
+ // union { } b;
+ // int c[0];
+ // etc. However, we have to provide a place-holder in InVals, so
+ // pretend we have an 8-byte item at the current address for that
+ // purpose.
+ if (!ObjSize) {
+ int FI = MFI->CreateFixedObject(PtrByteSize, ArgOffset, true);
+ SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
+ InVals.push_back(FIN);
+ continue;
+ }
// All aggregates smaller than 8 bytes must be passed right-justified.
if (ObjSize < PtrByteSize)
CurArgOffset = CurArgOffset + (PtrByteSize - ObjSize);
SmallVector<SDValue, 8> MemOps;
unsigned nAltivecParamsAtEnd = 0;
+ // FIXME: FuncArg and Ins[ArgNo] must reference the same argument.
+ // When passing anonymous aggregates, this is currently not true.
+ // See LowerFormalArguments_64SVR4 for a fix.
Function::const_arg_iterator FuncArg = MF.getFunction()->arg_begin();
for (unsigned ArgNo = 0, e = Ins.size(); ArgNo != e; ++ArgNo, ++FuncArg) {
SDValue ArgVal;
// Copy all of the result registers out of their specified physreg.
for (unsigned i = 0, e = RVLocs.size(); i != e; ++i) {
CCValAssign &VA = RVLocs[i];
- EVT VT = VA.getValVT();
assert(VA.isRegLoc() && "Can only return in registers!");
- Chain = DAG.getCopyFromReg(Chain, dl,
- VA.getLocReg(), VT, InFlag).getValue(1);
- InVals.push_back(Chain.getValue(0));
- InFlag = Chain.getValue(2);
+
+ SDValue Val = DAG.getCopyFromReg(Chain, dl,
+ VA.getLocReg(), VA.getLocVT(), InFlag);
+ Chain = Val.getValue(1);
+ InFlag = Val.getValue(2);
+
+ switch (VA.getLocInfo()) {
+ default: llvm_unreachable("Unknown loc info!");
+ case CCValAssign::Full: break;
+ case CCValAssign::AExt:
+ Val = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), Val);
+ break;
+ case CCValAssign::ZExt:
+ Val = DAG.getNode(ISD::AssertZext, dl, VA.getLocVT(), Val,
+ DAG.getValueType(VA.getValVT()));
+ Val = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), Val);
+ break;
+ case CCValAssign::SExt:
+ Val = DAG.getNode(ISD::AssertSext, dl, VA.getLocVT(), Val,
+ DAG.getValueType(VA.getValVT()));
+ Val = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), Val);
+ break;
+ }
+
+ InVals.push_back(Val);
}
return Chain;
// When performing tail call optimization the callee pops its arguments off
// the stack. Account for this here so these bytes can be pushed back on in
- // PPCRegisterInfo::eliminateCallFramePseudoInstr.
+ // PPCFrameLowering::eliminateCallFramePseudoInstr.
int BytesCalleePops =
(CallConv == CallingConv::Fast &&
getTargetMachine().Options.GuaranteedTailCallOpt) ? NumBytes : 0;
// Emit tail call.
if (isTailCall) {
- // If this is the first return lowered for this function, add the regs
- // to the liveout set for the function.
- if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
- SmallVector<CCValAssign, 16> RVLocs;
- CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
- getTargetMachine(), RVLocs, *DAG.getContext());
- CCInfo.AnalyzeCallResult(Ins, RetCC_PPC);
- for (unsigned i = 0; i != RVLocs.size(); ++i)
- DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
- }
-
assert(((Callee.getOpcode() == ISD::Register &&
cast<RegisterSDNode>(Callee)->getReg() == PPC::CTR) ||
Callee.getOpcode() == ISD::TargetExternalSymbol ||
bool Result;
if (Outs[i].IsFixed) {
- Result = CC_PPC_SVR4(i, ArgVT, ArgVT, CCValAssign::Full, ArgFlags,
- CCInfo);
+ Result = CC_PPC32_SVR4(i, ArgVT, ArgVT, CCValAssign::Full, ArgFlags,
+ CCInfo);
} else {
- Result = CC_PPC_SVR4_VarArg(i, ArgVT, ArgVT, CCValAssign::Full,
- ArgFlags, CCInfo);
+ Result = CC_PPC32_SVR4_VarArg(i, ArgVT, ArgVT, CCValAssign::Full,
+ ArgFlags, CCInfo);
}
if (Result) {
}
} else {
// All arguments are treated the same.
- CCInfo.AnalyzeCallOperands(Outs, CC_PPC_SVR4);
+ CCInfo.AnalyzeCallOperands(Outs, CC_PPC32_SVR4);
}
// Assign locations to all of the outgoing aggregate by value arguments.
// Reserve stack space for the allocations in CCInfo.
CCByValInfo.AllocateStack(CCInfo.getNextStackOffset(), PtrByteSize);
- CCByValInfo.AnalyzeCallOperands(Outs, CC_PPC_SVR4_ByVal);
+ CCByValInfo.AnalyzeCallOperands(Outs, CC_PPC32_SVR4_ByVal);
// Size of the linkage area, parameter list area and the part of the local
// space variable where copies of aggregates which are passed by value are
// These are the proper values we need for right-justifying the
// aggregate in a parameter register.
unsigned Size = Flags.getByValSize();
+
+ // An empty aggregate parameter takes up no storage and no
+ // registers.
+ if (Size == 0)
+ continue;
+
// All aggregates smaller than 8 bytes must be passed right-justified.
if (Size==1 || Size==2 || Size==4) {
EVT VT = (Size==1) ? MVT::i8 : ((Size==2) ? MVT::i16 : MVT::i32);
RegsToPass.push_back(std::make_pair(FPR[FPR_idx++], Arg));
if (isVarArg) {
- SDValue Store = DAG.getStore(Chain, dl, Arg, PtrOff,
+ // A single float or an aggregate containing only a single float
+ // must be passed right-justified in the stack doubleword, and
+ // in the GPR, if one is available.
+ SDValue StoreOff;
+ if (Arg.getValueType().getSimpleVT().SimpleTy == MVT::f32) {
+ SDValue ConstFour = DAG.getConstant(4, PtrOff.getValueType());
+ StoreOff = DAG.getNode(ISD::ADD, dl, PtrVT, PtrOff, ConstFour);
+ } else
+ StoreOff = PtrOff;
+
+ SDValue Store = DAG.getStore(Chain, dl, Arg, StoreOff,
MachinePointerInfo(), false, false, 0);
MemOpChains.push_back(Store);
getTargetMachine(), RVLocs, *DAG.getContext());
CCInfo.AnalyzeReturn(Outs, RetCC_PPC);
- // If this is the first return lowered for this function, add the regs to the
- // liveout set for the function.
- if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
- for (unsigned i = 0; i != RVLocs.size(); ++i)
- DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
- }
-
SDValue Flag;
+ SmallVector<SDValue, 4> RetOps(1, Chain);
// Copy the result values into the output registers.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
CCValAssign &VA = RVLocs[i];
assert(VA.isRegLoc() && "Can only return in registers!");
- Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
- OutVals[i], Flag);
+
+ SDValue Arg = OutVals[i];
+
+ switch (VA.getLocInfo()) {
+ default: llvm_unreachable("Unknown loc info!");
+ case CCValAssign::Full: break;
+ case CCValAssign::AExt:
+ Arg = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), Arg);
+ break;
+ case CCValAssign::ZExt:
+ Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Arg);
+ break;
+ case CCValAssign::SExt:
+ Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, VA.getLocVT(), Arg);
+ break;
+ }
+
+ Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), Arg, Flag);
Flag = Chain.getValue(1);
+ RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
}
+ RetOps[0] = Chain; // Update chain.
+
+ // Add the flag if we have it.
if (Flag.getNode())
- return DAG.getNode(PPCISD::RET_FLAG, dl, MVT::Other, Chain, Flag);
- else
- return DAG.getNode(PPCISD::RET_FLAG, dl, MVT::Other, Chain);
+ RetOps.push_back(Flag);
+
+ return DAG.getNode(PPCISD::RET_FLAG, dl, MVT::Other,
+ &RetOps[0], RetOps.size());
}
SDValue PPCTargetLowering::LowerSTACKRESTORE(SDValue Op, SelectionDAG &DAG,
MachineFunction &MF = DAG.getMachineFunction();
EVT VT = Op.getValueType();
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
- std::vector<EVT> NodeTys;
SDValue MFFSreg, InFlag;
// Save FP Control Word to register
- NodeTys.push_back(MVT::f64); // return register
- NodeTys.push_back(MVT::Glue); // unused in this context
+ EVT NodeTys[] = {
+ MVT::f64, // return register
+ MVT::Glue // unused in this context
+ };
SDValue Chain = DAG.getNode(PPCISD::MFFS, dl, NodeTys, &InFlag, 0);
// Save FP register to stack slot
// Two instruction sequences.
// If this value is in the range [-32,30] and is even, use:
- // tmp = VSPLTI[bhw], result = add tmp, tmp
- if (SextVal >= -32 && SextVal <= 30 && (SextVal & 1) == 0) {
- SDValue Res = BuildSplatI(SextVal >> 1, SplatSize, MVT::Other, DAG, dl);
- Res = DAG.getNode(ISD::ADD, dl, Res.getValueType(), Res, Res);
- return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), Res);
+ // VSPLTI[bhw](val/2) + VSPLTI[bhw](val/2)
+ // If this value is in the range [17,31] and is odd, use:
+ // VSPLTI[bhw](val-16) - VSPLTI[bhw](-16)
+ // If this value is in the range [-31,-17] and is odd, use:
+ // VSPLTI[bhw](val+16) + VSPLTI[bhw](-16)
+ // Note the last two are three-instruction sequences.
+ if (SextVal >= -32 && SextVal <= 31) {
+ // To avoid having these optimizations undone by constant folding,
+ // we convert to a pseudo that will be expanded later into one of
+ // the above forms.
+ SDValue Elt = DAG.getConstant(SextVal, MVT::i32);
+ EVT VT = Op.getValueType();
+ int Size = VT == MVT::v16i8 ? 1 : (VT == MVT::v8i16 ? 2 : 4);
+ SDValue EltSize = DAG.getConstant(Size, MVT::i32);
+ return DAG.getNode(PPCISD::VADD_SPLAT, dl, VT, Elt, EltSize);
}
// If this is 0x8000_0000 x 4, turn into vspltisw + vslw. If it is
}
}
- // Three instruction sequences.
-
- // Odd, in range [17,31]: (vsplti C)-(vsplti -16).
- if (SextVal >= 0 && SextVal <= 31) {
- SDValue LHS = BuildSplatI(SextVal-16, SplatSize, MVT::Other, DAG, dl);
- SDValue RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG, dl);
- LHS = DAG.getNode(ISD::SUB, dl, LHS.getValueType(), LHS, RHS);
- return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), LHS);
- }
- // Odd, in range [-31,-17]: (vsplti C)+(vsplti -16).
- if (SextVal >= -31 && SextVal <= 0) {
- SDValue LHS = BuildSplatI(SextVal+16, SplatSize, MVT::Other, DAG, dl);
- SDValue RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG, dl);
- LHS = DAG.getNode(ISD::ADD, dl, LHS.getValueType(), LHS, RHS);
- return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), LHS);
- }
-
return SDValue();
}
Op.getOperand(3), // RHS
DAG.getConstant(CompareOpc, MVT::i32)
};
- std::vector<EVT> VTs;
- VTs.push_back(Op.getOperand(2).getValueType());
- VTs.push_back(MVT::Glue);
+ EVT VTs[] = { Op.getOperand(2).getValueType(), MVT::Glue };
SDValue CompNode = DAG.getNode(PPCISD::VCMPo, dl, VTs, Ops, 3);
// Now that we have the comparison, emit a copy from the CR to a GPR.
bool BranchOnWhenPredTrue = (CC == ISD::SETEQ) ^ (Val == 0);
// Create the PPCISD altivec 'dot' comparison node.
- std::vector<EVT> VTs;
SDValue Ops[] = {
LHS.getOperand(2), // LHS of compare
LHS.getOperand(3), // RHS of compare
DAG.getConstant(CompareOpc, MVT::i32)
};
- VTs.push_back(LHS.getOperand(2).getValueType());
- VTs.push_back(MVT::Glue);
+ EVT VTs[] = { LHS.getOperand(2).getValueType(), MVT::Glue };
SDValue CompNode = DAG.getNode(PPCISD::VCMPo, dl, VTs, Ops, 3);
// Unpack the result based on how the target uses it.
case 'v':
case 'y':
return C_RegisterClass;
+ case 'Z':
+ // FIXME: While Z does indicate a memory constraint, it specifically
+ // indicates an r+r address (used in conjunction with the 'y' modifier
+ // in the replacement string). Currently, we're forcing the base
+ // register to be r0 in the asm printer (which is interpreted as zero)
+ // and forming the complete address in the second register. This is
+ // suboptimal.
+ return C_Memory;
}
}
return TargetLowering::getConstraintType(Constraint);
case 'y':
weight = CW_Register;
break;
+ case 'Z':
+ weight = CW_Memory;
+ break;
}
return weight;
}
return std::make_pair(0U, &PPC::G8RCRegClass);
return std::make_pair(0U, &PPC::GPRCRegClass);
case 'f':
- if (VT == MVT::f32)
+ if (VT == MVT::f32 || VT == MVT::i32)
return std::make_pair(0U, &PPC::F4RCRegClass);
- if (VT == MVT::f64)
+ if (VT == MVT::f64 || VT == MVT::i64)
return std::make_pair(0U, &PPC::F8RCRegClass);
break;
case 'v':
bool is31 = (getTargetMachine().Options.DisableFramePointerElim(MF) ||
MFI->hasVarSizedObjects()) &&
MFI->getStackSize() &&
- !MF.getFunction()->getFnAttributes().
- hasAttribute(Attributes::Naked);
+ !MF.getFunction()->getAttributes().
+ hasAttribute(AttributeSet::FunctionIndex, Attribute::Naked);
unsigned FrameReg = isPPC64 ? (is31 ? PPC::X31 : PPC::X1) :
(is31 ? PPC::R31 : PPC::R1);
SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), dl, FrameReg,
/// lowering. If DstAlign is zero that means it's safe to destination
/// alignment can satisfy any constraint. Similarly if SrcAlign is zero it
/// means there isn't a need to check it against alignment requirement,
-/// probably because the source does not need to be loaded. If
-/// 'IsZeroVal' is true, that means it's safe to return a
-/// non-scalar-integer type, e.g. empty string source, constant, or loaded
-/// from memory. 'MemcpyStrSrc' indicates whether the memcpy source is
-/// constant so it does not need to be loaded.
+/// probably because the source does not need to be loaded. If 'IsMemset' is
+/// true, that means it's expanding a memset. If 'ZeroMemset' is true, that
+/// means it's a memset of zero. 'MemcpyStrSrc' indicates whether the memcpy
+/// source is constant so it does not need to be loaded.
/// It returns EVT::Other if the type should be determined using generic
/// target-independent logic.
EVT PPCTargetLowering::getOptimalMemOpType(uint64_t Size,
unsigned DstAlign, unsigned SrcAlign,
- bool IsZeroVal,
+ bool IsMemset, bool ZeroMemset,
bool MemcpyStrSrc,
MachineFunction &MF) const {
if (this->PPCSubTarget.isPPC64()) {