#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/ADT/VectorExtras.h"
#include "llvm/CodeGen/CallingConvLower.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
-#include "llvm/CallingConv.h"
-#include "llvm/Constants.h"
-#include "llvm/Function.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/Support/MathExtras.h"
-#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
-#include "llvm/DerivedTypes.h"
+#include "llvm/Target/TargetOptions.h"
using namespace llvm;
static bool CC_PPC_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
// from FP_ROUND: that rounds to nearest, this rounds to zero.
setOperationAction(ISD::FP_ROUND_INREG, MVT::ppcf128, Custom);
+ // We do not currently implment this libm ops for PowerPC.
+ setOperationAction(ISD::FFLOOR, MVT::ppcf128, Expand);
+ setOperationAction(ISD::FCEIL, MVT::ppcf128, Expand);
+ setOperationAction(ISD::FTRUNC, MVT::ppcf128, Expand);
+ setOperationAction(ISD::FRINT, MVT::ppcf128, Expand);
+ setOperationAction(ISD::FNEARBYINT, MVT::ppcf128, Expand);
+
// PowerPC has no SREM/UREM instructions
setOperationAction(ISD::SREM, MVT::i32, Expand);
setOperationAction(ISD::UREM, MVT::i32, Expand);
setOperationAction(ISD::BSWAP, MVT::i32 , Expand);
setOperationAction(ISD::CTPOP, MVT::i32 , Expand);
setOperationAction(ISD::CTTZ , MVT::i32 , Expand);
+ setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i32, Expand);
+ setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i32, Expand);
setOperationAction(ISD::BSWAP, MVT::i64 , Expand);
setOperationAction(ISD::CTPOP, MVT::i64 , Expand);
setOperationAction(ISD::CTTZ , MVT::i64 , Expand);
+ setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i64, Expand);
+ setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i64, Expand);
// PowerPC does not have ROTR
setOperationAction(ISD::ROTR, MVT::i32 , Expand);
// 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()) {
- setOperationAction(ISD::VAARG, MVT::Other, Custom);
- setOperationAction(ISD::VAARG, MVT::i64, Custom);
+ if (TM.getSubtarget<PPCSubtarget>().isSVR4ABI()) {
+ if (TM.getSubtarget<PPCSubtarget>().isPPC64()) {
+ // VAARG always uses double-word chunks, so promote anything smaller.
+ setOperationAction(ISD::VAARG, MVT::i1, Promote);
+ AddPromotedToType (ISD::VAARG, MVT::i1, MVT::i64);
+ setOperationAction(ISD::VAARG, MVT::i8, Promote);
+ AddPromotedToType (ISD::VAARG, MVT::i8, MVT::i64);
+ setOperationAction(ISD::VAARG, MVT::i16, Promote);
+ AddPromotedToType (ISD::VAARG, MVT::i16, MVT::i64);
+ setOperationAction(ISD::VAARG, MVT::i32, Promote);
+ AddPromotedToType (ISD::VAARG, MVT::i32, MVT::i64);
+ setOperationAction(ISD::VAARG, MVT::Other, Expand);
+ } else {
+ // VAARG is custom lowered with the 32-bit SVR4 ABI.
+ setOperationAction(ISD::VAARG, MVT::Other, Custom);
+ setOperationAction(ISD::VAARG, MVT::i64, Custom);
+ }
} else
setOperationAction(ISD::VAARG, MVT::Other, Expand);
setOperationAction(ISD::FPOW, VT, Expand);
setOperationAction(ISD::CTPOP, VT, Expand);
setOperationAction(ISD::CTLZ, VT, Expand);
+ setOperationAction(ISD::CTLZ_ZERO_UNDEF, VT, Expand);
setOperationAction(ISD::CTTZ, VT, Expand);
+ setOperationAction(ISD::CTTZ_ZERO_UNDEF, VT, Expand);
}
// We can custom expand all VECTOR_SHUFFLEs to VPERM, others we can handle
setOperationAction(ISD::BUILD_VECTOR, MVT::v4f32, Custom);
}
+ if (TM.getSubtarget<PPCSubtarget>().has64BitSupport())
+ setOperationAction(ISD::PREFETCH, MVT::Other, Legal);
+
setOperationAction(ISD::ATOMIC_LOAD, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_STORE, MVT::i32, Expand);
setInsertFencesForAtomic(true);
+ setSchedulingPreference(Sched::Hybrid);
+
computeRegisterProperties();
}
// Darwin passes everything on 4 byte boundary.
if (TM.getSubtarget<PPCSubtarget>().isDarwin())
return 4;
- // FIXME SVR4 TBD
+
+ // 16byte and wider vectors are passed on 16byte boundary.
+ if (VectorType *VTy = dyn_cast<VectorType>(Ty))
+ if (VTy->getBitWidth() >= 128)
+ return 16;
+
+ // The rest is 8 on PPC64 and 4 on PPC32 boundary.
+ if (PPCSubTarget.isPPC64())
+ return 8;
+
return 4;
}
case PPCISD::EXTSW_32: return "PPCISD::EXTSW_32";
case PPCISD::STD_32: return "PPCISD::STD_32";
case PPCISD::CALL_SVR4: return "PPCISD::CALL_SVR4";
+ case PPCISD::CALL_NOP_SVR4: return "PPCISD::CALL_NOP_SVR4";
case PPCISD::CALL_Darwin: return "PPCISD::CALL_Darwin";
case PPCISD::NOP: return "PPCISD::NOP";
case PPCISD::MTCTR: return "PPCISD::MTCTR";
APInt LHSKnownZero, LHSKnownOne;
APInt RHSKnownZero, RHSKnownOne;
DAG.ComputeMaskedBits(N.getOperand(0),
- APInt::getAllOnesValue(N.getOperand(0)
- .getValueSizeInBits()),
LHSKnownZero, LHSKnownOne);
if (LHSKnownZero.getBoolValue()) {
DAG.ComputeMaskedBits(N.getOperand(1),
- APInt::getAllOnesValue(N.getOperand(1)
- .getValueSizeInBits()),
RHSKnownZero, RHSKnownOne);
// If all of the bits are known zero on the LHS or RHS, the add won't
// carry.
// (for better address arithmetic) if the LHS and RHS of the OR are
// provably disjoint.
APInt LHSKnownZero, LHSKnownOne;
- DAG.ComputeMaskedBits(N.getOperand(0),
- APInt::getAllOnesValue(N.getOperand(0)
- .getValueSizeInBits()),
- LHSKnownZero, LHSKnownOne);
+ DAG.ComputeMaskedBits(N.getOperand(0), LHSKnownZero, LHSKnownOne);
if ((LHSKnownZero.getZExtValue()|~(uint64_t)imm) == ~0ULL) {
// If all of the bits are known zero on the LHS or RHS, the add won't
// (for better address arithmetic) if the LHS and RHS of the OR are
// provably disjoint.
APInt LHSKnownZero, LHSKnownOne;
- DAG.ComputeMaskedBits(N.getOperand(0),
- APInt::getAllOnesValue(N.getOperand(0)
- .getValueSizeInBits()),
- LHSKnownZero, LHSKnownOne);
+ DAG.ComputeMaskedBits(N.getOperand(0), LHSKnownZero, LHSKnownOne);
if ((LHSKnownZero.getZExtValue()|~(uint64_t)imm) == ~0ULL) {
// If all of the bits are known zero on the LHS or RHS, the add won't
// carry.
// Lower to a call to __trampoline_setup(Trmp, TrampSize, FPtr, ctx_reg)
std::pair<SDValue, SDValue> CallResult =
LowerCallTo(Chain, Type::getVoidTy(*DAG.getContext()),
- false, false, false, false, 0, CallingConv::C, false,
- /*isReturnValueUsed=*/true,
+ false, false, false, false, 0, CallingConv::C,
+ /*isTailCall=*/false,
+ /*doesNotRet=*/false, /*isReturnValueUsed=*/true,
DAG.getExternalSymbol("__trampoline_setup", PtrVT),
Args, DAG, dl);
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State) {
- static const unsigned ArgRegs[] = {
+ static const uint16_t ArgRegs[] = {
PPC::R3, PPC::R4, PPC::R5, PPC::R6,
PPC::R7, PPC::R8, PPC::R9, PPC::R10,
};
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
CCState &State) {
- static const unsigned ArgRegs[] = {
+ static const uint16_t ArgRegs[] = {
PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5, PPC::F6, PPC::F7,
PPC::F8
};
/// GetFPR - Get the set of FP registers that should be allocated for arguments,
/// on Darwin.
-static const unsigned *GetFPR() {
- static const unsigned FPR[] = {
+static const uint16_t *GetFPR() {
+ static const uint16_t FPR[] = {
PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5, PPC::F6, PPC::F7,
PPC::F8, PPC::F9, PPC::F10, PPC::F11, PPC::F12, PPC::F13
};
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// Potential tail calls could cause overwriting of argument stack slots.
- bool isImmutable = !(GuaranteedTailCallOpt && (CallConv==CallingConv::Fast));
+ bool isImmutable = !(getTargetMachine().Options.GuaranteedTailCallOpt &&
+ (CallConv == CallingConv::Fast));
unsigned PtrByteSize = 4;
// Assign locations to all of the incoming arguments.
// Arguments stored in registers.
if (VA.isRegLoc()) {
- TargetRegisterClass *RC;
+ const TargetRegisterClass *RC;
EVT ValVT = VA.getValVT();
switch (ValVT.getSimpleVT().SimpleTy) {
// If the function takes variable number of arguments, make a frame index for
// the start of the first vararg value... for expansion of llvm.va_start.
if (isVarArg) {
- static const unsigned GPArgRegs[] = {
+ static const uint16_t GPArgRegs[] = {
PPC::R3, PPC::R4, PPC::R5, PPC::R6,
PPC::R7, PPC::R8, PPC::R9, PPC::R10,
};
const unsigned NumGPArgRegs = array_lengthof(GPArgRegs);
- static const unsigned FPArgRegs[] = {
+ static const uint16_t FPArgRegs[] = {
PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5, PPC::F6, PPC::F7,
PPC::F8
};
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = PtrVT == MVT::i64;
// Potential tail calls could cause overwriting of argument stack slots.
- bool isImmutable = !(GuaranteedTailCallOpt && (CallConv==CallingConv::Fast));
+ bool isImmutable = !(getTargetMachine().Options.GuaranteedTailCallOpt &&
+ (CallConv == CallingConv::Fast));
unsigned PtrByteSize = isPPC64 ? 8 : 4;
unsigned ArgOffset = PPCFrameLowering::getLinkageSize(isPPC64, true);
// Area that is at least reserved in caller of this function.
unsigned MinReservedArea = ArgOffset;
- static const unsigned GPR_32[] = { // 32-bit registers.
+ static const uint16_t GPR_32[] = { // 32-bit registers.
PPC::R3, PPC::R4, PPC::R5, PPC::R6,
PPC::R7, PPC::R8, PPC::R9, PPC::R10,
};
- static const unsigned GPR_64[] = { // 64-bit registers.
+ static const uint16_t GPR_64[] = { // 64-bit registers.
PPC::X3, PPC::X4, PPC::X5, PPC::X6,
PPC::X7, PPC::X8, PPC::X9, PPC::X10,
};
- static const unsigned *FPR = GetFPR();
+ static const uint16_t *FPR = GetFPR();
- static const unsigned VR[] = {
+ static const uint16_t VR[] = {
PPC::V2, PPC::V3, PPC::V4, PPC::V5, PPC::V6, PPC::V7, PPC::V8,
PPC::V9, PPC::V10, PPC::V11, PPC::V12, PPC::V13
};
unsigned GPR_idx = 0, FPR_idx = 0, VR_idx = 0;
- const unsigned *GPR = isPPC64 ? GPR_64 : GPR_32;
+ const uint16_t *GPR = isPPC64 ? GPR_64 : GPR_32;
// In 32-bit non-varargs functions, the stack space for vectors is after the
// stack space for non-vectors. We do not use this space unless we have
for (unsigned ArgNo = 0, e = Ins.size(); ArgNo != e;
++ArgNo) {
EVT ObjectVT = Ins[ArgNo].VT;
- unsigned ObjSize = ObjectVT.getSizeInBits()/8;
ISD::ArgFlagsTy Flags = Ins[ArgNo].Flags;
if (Flags.isByVal()) {
// ObjSize is the true size, ArgSize rounded up to multiple of regs.
- ObjSize = Flags.getByValSize();
+ unsigned ObjSize = Flags.getByValSize();
unsigned ArgSize =
((ObjSize + PtrByteSize - 1)/PtrByteSize) * PtrByteSize;
VecArgOffset += ArgSize;
PPCFrameLowering::getMinCallFrameSize(isPPC64, true));
// Tail call needs the stack to be aligned.
- if (CC==CallingConv::Fast && GuaranteedTailCallOpt) {
- unsigned TargetAlign = DAG.getMachineFunction().getTarget().getFrameLowering()->
- getStackAlignment();
+ if (CC == CallingConv::Fast && DAG.getTarget().Options.GuaranteedTailCallOpt){
+ unsigned TargetAlign = DAG.getMachineFunction().getTarget().
+ getFrameLowering()->getStackAlignment();
unsigned AlignMask = TargetAlign-1;
NumBytes = (NumBytes + AlignMask) & ~AlignMask;
}
bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
SelectionDAG& DAG) const {
- if (!GuaranteedTailCallOpt)
+ if (!getTargetMachine().Options.GuaranteedTailCallOpt)
return false;
// Variable argument functions are not supported.
// the stack. Account for this here so these bytes can be pushed back on in
// PPCRegisterInfo::eliminateCallFramePseudoInstr.
int BytesCalleePops =
- (CallConv==CallingConv::Fast && GuaranteedTailCallOpt) ? NumBytes : 0;
+ (CallConv == CallingConv::Fast &&
+ getTargetMachine().Options.GuaranteedTailCallOpt) ? NumBytes : 0;
+
+ // Add a register mask operand representing the call-preserved registers.
+ const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo();
+ const uint32_t *Mask = TRI->getCallPreservedMask(CallConv);
+ assert(Mask && "Missing call preserved mask for calling convention");
+ Ops.push_back(DAG.getRegisterMask(Mask));
if (InFlag.getNode())
Ops.push_back(InFlag);
return DAG.getNode(PPCISD::TC_RETURN, dl, MVT::Other, &Ops[0], Ops.size());
}
- Chain = DAG.getNode(CallOpc, dl, NodeTys, &Ops[0], Ops.size());
- InFlag = Chain.getValue(1);
-
// Add a NOP immediately after the branch instruction when using the 64-bit
// SVR4 ABI. At link time, if caller and callee are in a different module and
// thus have a different TOC, the call will be replaced with a call to a stub
// which restores the TOC of the caller from the TOC save slot of the current
// stack frame. If caller and callee belong to the same module (and have the
// same TOC), the NOP will remain unchanged.
+
+ bool needsTOCRestore = false;
if (!isTailCall && PPCSubTarget.isSVR4ABI()&& PPCSubTarget.isPPC64()) {
- SDVTList VTs = DAG.getVTList(MVT::Other, MVT::Glue);
if (CallOpc == PPCISD::BCTRL_SVR4) {
// This is a call through a function pointer.
// Restore the caller TOC from the save area into R2.
// since r2 is a reserved register (which prevents the register allocator
// from allocating it), resulting in an additional register being
// allocated and an unnecessary move instruction being generated.
- Chain = DAG.getNode(PPCISD::TOC_RESTORE, dl, VTs, Chain, InFlag);
- InFlag = Chain.getValue(1);
- } else {
+ needsTOCRestore = true;
+ } else if (CallOpc == PPCISD::CALL_SVR4) {
// Otherwise insert NOP.
- InFlag = DAG.getNode(PPCISD::NOP, dl, MVT::Glue, InFlag);
+ CallOpc = PPCISD::CALL_NOP_SVR4;
}
}
+ Chain = DAG.getNode(CallOpc, dl, NodeTys, &Ops[0], Ops.size());
+ InFlag = Chain.getValue(1);
+
+ if (needsTOCRestore) {
+ SDVTList VTs = DAG.getVTList(MVT::Other, MVT::Glue);
+ Chain = DAG.getNode(PPCISD::TOC_RESTORE, dl, VTs, Chain, InFlag);
+ InFlag = Chain.getValue(1);
+ }
+
Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
DAG.getIntPtrConstant(BytesCalleePops, true),
InFlag);
SDValue
PPCTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
CallingConv::ID CallConv, bool isVarArg,
- bool &isTailCall,
+ bool doesNotRet, bool &isTailCall,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SmallVectorImpl<ISD::InputArg> &Ins,
// and restoring the callers stack pointer in this functions epilog. This is
// done because by tail calling the called function might overwrite the value
// in this function's (MF) stack pointer stack slot 0(SP).
- if (GuaranteedTailCallOpt && CallConv==CallingConv::Fast)
+ if (getTargetMachine().Options.GuaranteedTailCallOpt &&
+ CallConv == CallingConv::Fast)
MF.getInfo<PPCFunctionInfo>()->setHasFastCall();
// Count how many bytes are to be pushed on the stack, including the linkage
// and restoring the callers stack pointer in this functions epilog. This is
// done because by tail calling the called function might overwrite the value
// in this function's (MF) stack pointer stack slot 0(SP).
- if (GuaranteedTailCallOpt && CallConv==CallingConv::Fast)
+ if (getTargetMachine().Options.GuaranteedTailCallOpt &&
+ CallConv == CallingConv::Fast)
MF.getInfo<PPCFunctionInfo>()->setHasFastCall();
unsigned nAltivecParamsAtEnd = 0;
unsigned ArgOffset = PPCFrameLowering::getLinkageSize(isPPC64, true);
unsigned GPR_idx = 0, FPR_idx = 0, VR_idx = 0;
- static const unsigned GPR_32[] = { // 32-bit registers.
+ static const uint16_t GPR_32[] = { // 32-bit registers.
PPC::R3, PPC::R4, PPC::R5, PPC::R6,
PPC::R7, PPC::R8, PPC::R9, PPC::R10,
};
- static const unsigned GPR_64[] = { // 64-bit registers.
+ static const uint16_t GPR_64[] = { // 64-bit registers.
PPC::X3, PPC::X4, PPC::X5, PPC::X6,
PPC::X7, PPC::X8, PPC::X9, PPC::X10,
};
- static const unsigned *FPR = GetFPR();
+ static const uint16_t *FPR = GetFPR();
- static const unsigned VR[] = {
+ static const uint16_t VR[] = {
PPC::V2, PPC::V3, PPC::V4, PPC::V5, PPC::V6, PPC::V7, PPC::V8,
PPC::V9, PPC::V10, PPC::V11, PPC::V12, PPC::V13
};
const unsigned NumFPRs = 13;
const unsigned NumVRs = array_lengthof(VR);
- const unsigned *GPR = isPPC64 ? GPR_64 : GPR_32;
+ const uint16_t *GPR = isPPC64 ? GPR_64 : GPR_32;
SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPass;
SmallVector<TailCallArgumentInfo, 8> TailCallArguments;
// Check to see if this is a shuffle of 4-byte values. If so, we can use our
// perfect shuffle table to emit an optimal matching sequence.
- SmallVector<int, 16> PermMask;
- SVOp->getMask(PermMask);
+ ArrayRef<int> PermMask = SVOp->getMask();
unsigned PFIndexes[4];
bool isFourElementShuffle = true;
case ISD::RETURNADDR: return LowerRETURNADDR(Op, DAG);
case ISD::FRAMEADDR: return LowerFRAMEADDR(Op, DAG);
}
- return SDValue();
}
void PPCTargetLowering::ReplaceNodeResults(SDNode *N,
DebugLoc dl = N->getDebugLoc();
switch (N->getOpcode()) {
default:
- assert(false && "Do not know how to custom type legalize this operation!");
- return;
+ llvm_unreachable("Do not know how to custom type legalize this operation!");
case ISD::VAARG: {
if (!TM.getSubtarget<PPCSubtarget>().isSVR4ABI()
|| TM.getSubtarget<PPCSubtarget>().isPPC64())
//===----------------------------------------------------------------------===//
void PPCTargetLowering::computeMaskedBitsForTargetNode(const SDValue Op,
- const APInt &Mask,
APInt &KnownZero,
APInt &KnownOne,
const SelectionDAG &DAG,
unsigned Depth) const {
- KnownZero = KnownOne = APInt(Mask.getBitWidth(), 0);
+ KnownZero = KnownOne = APInt(KnownZero.getBitWidth(), 0);
switch (Op.getOpcode()) {
default: break;
case PPCISD::LBRX: {
return (V > -(1 << 16) && V < (1 << 16)-1);
}
-bool PPCTargetLowering::isLegalAddressImmediate(llvm::GlobalValue* GV) const {
+bool PPCTargetLowering::isLegalAddressImmediate(GlobalValue* GV) const {
return false;
}
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
MFI->setFrameAddressIsTaken(true);
- bool is31 = (DisableFramePointerElim(MF) || MFI->hasVarSizedObjects()) &&
+ bool is31 = (getTargetMachine().Options.DisableFramePointerElim(MF) ||
+ MFI->hasVarSizedObjects()) &&
MFI->getStackSize() &&
!MF.getFunction()->hasFnAttr(Attribute::Naked);
unsigned FrameReg = isPPC64 ? (is31 ? PPC::X31 : PPC::X1) :
return MVT::i32;
}
}
+
+Sched::Preference PPCTargetLowering::getSchedulingPreference(SDNode *N) const {
+ unsigned Directive = PPCSubTarget.getDarwinDirective();
+ if (Directive == PPC::DIR_440 || Directive == PPC::DIR_A2)
+ return Sched::ILP;
+
+ return TargetLowering::getSchedulingPreference(N);
+}
+