#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+#include "llvm/IR/Intrinsics.h"
#include <cctype>
using namespace llvm;
return Op;
}
-SystemZTargetLowering::SystemZTargetLowering(SystemZTargetMachine &tm)
- : TargetLowering(tm, new TargetLoweringObjectFileELF()),
- Subtarget(*tm.getSubtargetImpl()), TM(tm) {
+SystemZTargetLowering::SystemZTargetLowering(const TargetMachine &tm,
+ const SystemZSubtarget &STI)
+ : TargetLowering(tm), Subtarget(STI) {
MVT PtrVT = getPointerTy();
// Set up the register classes.
addRegisterClass(MVT::f128, &SystemZ::FP128BitRegClass);
// Compute derived properties from the register classes
- computeRegisterProperties();
+ computeRegisterProperties(Subtarget.getRegisterInfo());
// Set up special registers.
setExceptionPointerRegister(SystemZ::R6D);
// available, or if the operand is constant.
setOperationAction(ISD::ATOMIC_LOAD_SUB, VT, Custom);
+ // Use POPCNT on z196 and above.
+ if (Subtarget.hasPopulationCount())
+ setOperationAction(ISD::CTPOP, VT, Custom);
+ else
+ setOperationAction(ISD::CTPOP, VT, Expand);
+
// No special instructions for these.
- setOperationAction(ISD::CTPOP, VT, Expand);
setOperationAction(ISD::CTTZ, VT, Expand);
setOperationAction(ISD::CTTZ_ZERO_UNDEF, VT, Expand);
setOperationAction(ISD::CTLZ_ZERO_UNDEF, VT, Expand);
setOperationAction(ISD::SRA_PARTS, MVT::i64, Expand);
// We have native instructions for i8, i16 and i32 extensions, but not i1.
- setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
- setLoadExtAction(ISD::ZEXTLOAD, MVT::i1, Promote);
- setLoadExtAction(ISD::EXTLOAD, MVT::i1, Promote);
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
+ for (MVT VT : MVT::integer_valuetypes()) {
+ setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
+ setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i1, Promote);
+ setLoadExtAction(ISD::EXTLOAD, VT, MVT::i1, Promote);
+ }
// Handle the various types of symbolic address.
setOperationAction(ISD::ConstantPool, PtrVT, Custom);
// Needed so that we don't try to implement f128 constant loads using
// a load-and-extend of a f80 constant (in cases where the constant
// would fit in an f80).
- setLoadExtAction(ISD::EXTLOAD, MVT::f80, Expand);
+ for (MVT VT : MVT::fp_valuetypes())
+ setLoadExtAction(ISD::EXTLOAD, VT, MVT::f80, Expand);
// Floating-point truncation and stores need to be done separately.
setTruncStoreAction(MVT::f64, MVT::f32, Expand);
// Codes for which we want to perform some z-specific combinations.
setTargetDAGCombine(ISD::SIGN_EXTEND);
+ // Handle intrinsics.
+ setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::Other, Custom);
+
// We want to use MVC in preference to even a single load/store pair.
MaxStoresPerMemcpy = 0;
MaxStoresPerMemcpyOptSize = 0;
return Imm.isZero() || Imm.isNegZero();
}
-bool SystemZTargetLowering::allowsUnalignedMemoryAccesses(EVT VT,
- unsigned,
- bool *Fast) const {
+bool SystemZTargetLowering::isLegalICmpImmediate(int64_t Imm) const {
+ // We can use CGFI or CLGFI.
+ return isInt<32>(Imm) || isUInt<32>(Imm);
+}
+
+bool SystemZTargetLowering::isLegalAddImmediate(int64_t Imm) const {
+ // We can use ALGFI or SLGFI.
+ return isUInt<32>(Imm) || isUInt<32>(-Imm);
+}
+
+bool SystemZTargetLowering::allowsMisalignedMemoryAccesses(EVT VT,
+ unsigned,
+ unsigned,
+ bool *Fast) const {
// Unaligned accesses should never be slower than the expanded version.
// We check specifically for aligned accesses in the few cases where
// they are required.
return std::make_pair(0U, nullptr);
}
-std::pair<unsigned, const TargetRegisterClass *> SystemZTargetLowering::
-getRegForInlineAsmConstraint(const std::string &Constraint, MVT VT) const {
+std::pair<unsigned, const TargetRegisterClass *>
+SystemZTargetLowering::getRegForInlineAsmConstraint(
+ const TargetRegisterInfo *TRI, const std::string &Constraint,
+ MVT VT) const {
if (Constraint.size() == 1) {
// GCC Constraint Letters
switch (Constraint[0]) {
SystemZMC::FP64Regs);
}
}
- return TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
+ return TargetLowering::getRegForInlineAsmConstraint(TRI, Constraint, VT);
}
void SystemZTargetLowering::
MachineFrameInfo *MFI = MF.getFrameInfo();
MachineRegisterInfo &MRI = MF.getRegInfo();
SystemZMachineFunctionInfo *FuncInfo =
- MF.getInfo<SystemZMachineFunctionInfo>();
- auto *TFL = static_cast<const SystemZFrameLowering *>(TM.getFrameLowering());
+ MF.getInfo<SystemZMachineFunctionInfo>();
+ auto *TFL =
+ static_cast<const SystemZFrameLowering *>(Subtarget.getFrameLowering());
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
- CCState CCInfo(CallConv, IsVarArg, MF, TM, ArgLocs, *DAG.getContext());
+ CCState CCInfo(CallConv, IsVarArg, MF, ArgLocs, *DAG.getContext());
CCInfo.AnalyzeFormalArguments(Ins, CC_SystemZ);
unsigned NumFixedGPRs = 0;
return Chain;
}
-static bool canUseSiblingCall(CCState ArgCCInfo,
+static bool canUseSiblingCall(const CCState &ArgCCInfo,
SmallVectorImpl<CCValAssign> &ArgLocs) {
// Punt if there are any indirect or stack arguments, or if the call
// needs the call-saved argument register R6.
// Analyze the operands of the call, assigning locations to each operand.
SmallVector<CCValAssign, 16> ArgLocs;
- CCState ArgCCInfo(CallConv, IsVarArg, MF, TM, ArgLocs, *DAG.getContext());
+ CCState ArgCCInfo(CallConv, IsVarArg, MF, ArgLocs, *DAG.getContext());
ArgCCInfo.AnalyzeCallOperands(Outs, CC_SystemZ);
// We don't support GuaranteedTailCallOpt, only automatically-detected
Ops.push_back(DAG.getRegister(RegsToPass[I].first,
RegsToPass[I].second.getValueType()));
+ // Add a register mask operand representing the call-preserved registers.
+ const TargetRegisterInfo *TRI = Subtarget.getRegisterInfo();
+ const uint32_t *Mask = TRI->getCallPreservedMask(MF, CallConv);
+ assert(Mask && "Missing call preserved mask for calling convention");
+ Ops.push_back(DAG.getRegisterMask(Mask));
+
// Glue the call to the argument copies, if any.
if (Glue.getNode())
Ops.push_back(Glue);
// Assign locations to each value returned by this call.
SmallVector<CCValAssign, 16> RetLocs;
- CCState RetCCInfo(CallConv, IsVarArg, MF, TM, RetLocs, *DAG.getContext());
+ CCState RetCCInfo(CallConv, IsVarArg, MF, RetLocs, *DAG.getContext());
RetCCInfo.AnalyzeCallResult(Ins, RetCC_SystemZ);
// Copy all of the result registers out of their specified physreg.
// Assign locations to each returned value.
SmallVector<CCValAssign, 16> RetLocs;
- CCState RetCCInfo(CallConv, IsVarArg, MF, TM, RetLocs, *DAG.getContext());
+ CCState RetCCInfo(CallConv, IsVarArg, MF, RetLocs, *DAG.getContext());
RetCCInfo.AnalyzeReturn(Outs, RetCC_SystemZ);
// Quick exit for void returns
return DAG.getNode(SystemZISD::SERIALIZE, DL, MVT::Other, Chain);
}
+// Return true if Op is an intrinsic node with chain that returns the CC value
+// as its only (other) argument. Provide the associated SystemZISD opcode and
+// the mask of valid CC values if so.
+static bool isIntrinsicWithCCAndChain(SDValue Op, unsigned &Opcode,
+ unsigned &CCValid) {
+ unsigned Id = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
+ switch (Id) {
+ case Intrinsic::s390_tbegin:
+ Opcode = SystemZISD::TBEGIN;
+ CCValid = SystemZ::CCMASK_TBEGIN;
+ return true;
+
+ case Intrinsic::s390_tbegin_nofloat:
+ Opcode = SystemZISD::TBEGIN_NOFLOAT;
+ CCValid = SystemZ::CCMASK_TBEGIN;
+ return true;
+
+ case Intrinsic::s390_tend:
+ Opcode = SystemZISD::TEND;
+ CCValid = SystemZ::CCMASK_TEND;
+ return true;
+
+ default:
+ return false;
+ }
+}
+
+// Emit an intrinsic with chain with a glued value instead of its CC result.
+static SDValue emitIntrinsicWithChainAndGlue(SelectionDAG &DAG, SDValue Op,
+ unsigned Opcode) {
+ // Copy all operands except the intrinsic ID.
+ unsigned NumOps = Op.getNumOperands();
+ SmallVector<SDValue, 6> Ops;
+ Ops.reserve(NumOps - 1);
+ Ops.push_back(Op.getOperand(0));
+ for (unsigned I = 2; I < NumOps; ++I)
+ Ops.push_back(Op.getOperand(I));
+
+ assert(Op->getNumValues() == 2 && "Expected only CC result and chain");
+ SDVTList RawVTs = DAG.getVTList(MVT::Other, MVT::Glue);
+ SDValue Intr = DAG.getNode(Opcode, SDLoc(Op), RawVTs, Ops);
+ SDValue OldChain = SDValue(Op.getNode(), 1);
+ SDValue NewChain = SDValue(Intr.getNode(), 0);
+ DAG.ReplaceAllUsesOfValueWith(OldChain, NewChain);
+ return Intr;
+}
+
// CC is a comparison that will be implemented using an integer or
// floating-point comparison. Return the condition code mask for
// a branch on true. In the integer case, CCMASK_CMP_UO is set for
Load->getChain(), Load->getBasePtr(),
Load->getPointerInfo(), Load->getMemoryVT(),
Load->isVolatile(), Load->isNonTemporal(),
- Load->getAlignment());
+ Load->isInvariant(), Load->getAlignment());
// Make sure that the second operand is an i32 with the right value.
if (C.Op1.getValueType() != MVT::i32 ||
MaskVal = -(CmpVal & -CmpVal);
NewC.ICmpType = SystemZICMP::UnsignedOnly;
}
+ if (!MaskVal)
+ return;
// Check whether the combination of mask, comparison value and comparison
// type are suitable.
C.CCMask = NewCCMask;
}
+// Return a Comparison that tests the condition-code result of intrinsic
+// node Call against constant integer CC using comparison code Cond.
+// Opcode is the opcode of the SystemZISD operation for the intrinsic
+// and CCValid is the set of possible condition-code results.
+static Comparison getIntrinsicCmp(SelectionDAG &DAG, unsigned Opcode,
+ SDValue Call, unsigned CCValid, uint64_t CC,
+ ISD::CondCode Cond) {
+ Comparison C(Call, SDValue());
+ C.Opcode = Opcode;
+ C.CCValid = CCValid;
+ if (Cond == ISD::SETEQ)
+ // bit 3 for CC==0, bit 0 for CC==3, always false for CC>3.
+ C.CCMask = CC < 4 ? 1 << (3 - CC) : 0;
+ else if (Cond == ISD::SETNE)
+ // ...and the inverse of that.
+ C.CCMask = CC < 4 ? ~(1 << (3 - CC)) : -1;
+ else if (Cond == ISD::SETLT || Cond == ISD::SETULT)
+ // bits above bit 3 for CC==0 (always false), bits above bit 0 for CC==3,
+ // always true for CC>3.
+ C.CCMask = CC < 4 ? -1 << (4 - CC) : -1;
+ else if (Cond == ISD::SETGE || Cond == ISD::SETUGE)
+ // ...and the inverse of that.
+ C.CCMask = CC < 4 ? ~(-1 << (4 - CC)) : 0;
+ else if (Cond == ISD::SETLE || Cond == ISD::SETULE)
+ // bit 3 and above for CC==0, bit 0 and above for CC==3 (always true),
+ // always true for CC>3.
+ C.CCMask = CC < 4 ? -1 << (3 - CC) : -1;
+ else if (Cond == ISD::SETGT || Cond == ISD::SETUGT)
+ // ...and the inverse of that.
+ C.CCMask = CC < 4 ? ~(-1 << (3 - CC)) : 0;
+ else
+ llvm_unreachable("Unexpected integer comparison type");
+ C.CCMask &= CCValid;
+ return C;
+}
+
// Decide how to implement a comparison of type Cond between CmpOp0 with CmpOp1.
static Comparison getCmp(SelectionDAG &DAG, SDValue CmpOp0, SDValue CmpOp1,
ISD::CondCode Cond) {
+ if (CmpOp1.getOpcode() == ISD::Constant) {
+ uint64_t Constant = cast<ConstantSDNode>(CmpOp1)->getZExtValue();
+ unsigned Opcode, CCValid;
+ if (CmpOp0.getOpcode() == ISD::INTRINSIC_W_CHAIN &&
+ CmpOp0.getResNo() == 0 && CmpOp0->hasNUsesOfValue(1, 0) &&
+ isIntrinsicWithCCAndChain(CmpOp0, Opcode, CCValid))
+ return getIntrinsicCmp(DAG, Opcode, CmpOp0, CCValid, Constant, Cond);
+ }
Comparison C(CmpOp0, CmpOp1);
C.CCMask = CCMaskForCondCode(Cond);
if (C.Op0.getValueType().isFloatingPoint()) {
// Emit the comparison instruction described by C.
static SDValue emitCmp(SelectionDAG &DAG, SDLoc DL, Comparison &C) {
+ if (!C.Op1.getNode()) {
+ SDValue Op;
+ switch (C.Op0.getOpcode()) {
+ case ISD::INTRINSIC_W_CHAIN:
+ Op = emitIntrinsicWithChainAndGlue(DAG, C.Op0, C.Opcode);
+ break;
+ default:
+ llvm_unreachable("Invalid comparison operands");
+ }
+ return SDValue(Op.getNode(), Op->getNumValues() - 1);
+ }
if (C.Opcode == SystemZISD::ICMP)
return DAG.getNode(SystemZISD::ICMP, DL, MVT::Glue, C.Op0, C.Op1,
DAG.getConstant(C.ICmpType, MVT::i32));
}
SDValue SystemZTargetLowering::lowerBR_CC(SDValue Op, SelectionDAG &DAG) const {
- SDValue Chain = Op.getOperand(0);
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(1))->get();
SDValue CmpOp0 = Op.getOperand(2);
SDValue CmpOp1 = Op.getOperand(3);
Comparison C(getCmp(DAG, CmpOp0, CmpOp1, CC));
SDValue Glue = emitCmp(DAG, DL, C);
return DAG.getNode(SystemZISD::BR_CCMASK, DL, Op.getValueType(),
- Chain, DAG.getConstant(C.CCValid, MVT::i32),
+ Op.getOperand(0), DAG.getConstant(C.CCValid, MVT::i32),
DAG.getConstant(C.CCMask, MVT::i32), Dest, Glue);
}
}
}
- SmallVector<SDValue, 5> Ops;
- Ops.push_back(TrueOp);
- Ops.push_back(FalseOp);
- Ops.push_back(DAG.getConstant(C.CCValid, MVT::i32));
- Ops.push_back(DAG.getConstant(C.CCMask, MVT::i32));
- Ops.push_back(Glue);
+ SDValue Ops[] = {TrueOp, FalseOp, DAG.getConstant(C.CCValid, MVT::i32),
+ DAG.getConstant(C.CCMask, MVT::i32), Glue};
SDVTList VTs = DAG.getVTList(Op.getValueType(), MVT::Glue);
return DAG.getNode(SystemZISD::SELECT_CCMASK, DL, VTs, Ops);
const GlobalValue *GV = Node->getGlobal();
int64_t Offset = Node->getOffset();
EVT PtrVT = getPointerTy();
- Reloc::Model RM = TM.getRelocationModel();
- CodeModel::Model CM = TM.getCodeModel();
+ Reloc::Model RM = DAG.getTarget().getRelocationModel();
+ CodeModel::Model CM = DAG.getTarget().getCodeModel();
SDValue Result;
if (Subtarget.isPC32DBLSymbol(GV, RM, CM)) {
return Result;
}
+SDValue SystemZTargetLowering::lowerTLSGetOffset(GlobalAddressSDNode *Node,
+ SelectionDAG &DAG,
+ unsigned Opcode,
+ SDValue GOTOffset) const {
+ SDLoc DL(Node);
+ EVT PtrVT = getPointerTy();
+ SDValue Chain = DAG.getEntryNode();
+ SDValue Glue;
+
+ // __tls_get_offset takes the GOT offset in %r2 and the GOT in %r12.
+ SDValue GOT = DAG.getGLOBAL_OFFSET_TABLE(PtrVT);
+ Chain = DAG.getCopyToReg(Chain, DL, SystemZ::R12D, GOT, Glue);
+ Glue = Chain.getValue(1);
+ Chain = DAG.getCopyToReg(Chain, DL, SystemZ::R2D, GOTOffset, Glue);
+ Glue = Chain.getValue(1);
+
+ // The first call operand is the chain and the second is the TLS symbol.
+ SmallVector<SDValue, 8> Ops;
+ Ops.push_back(Chain);
+ Ops.push_back(DAG.getTargetGlobalAddress(Node->getGlobal(), DL,
+ Node->getValueType(0),
+ 0, 0));
+
+ // Add argument registers to the end of the list so that they are
+ // known live into the call.
+ Ops.push_back(DAG.getRegister(SystemZ::R2D, PtrVT));
+ Ops.push_back(DAG.getRegister(SystemZ::R12D, PtrVT));
+
+ // Add a register mask operand representing the call-preserved registers.
+ const TargetRegisterInfo *TRI = Subtarget.getRegisterInfo();
+ const uint32_t *Mask =
+ TRI->getCallPreservedMask(DAG.getMachineFunction(), CallingConv::C);
+ assert(Mask && "Missing call preserved mask for calling convention");
+ Ops.push_back(DAG.getRegisterMask(Mask));
+
+ // Glue the call to the argument copies.
+ Ops.push_back(Glue);
+
+ // Emit the call.
+ SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
+ Chain = DAG.getNode(Opcode, DL, NodeTys, Ops);
+ Glue = Chain.getValue(1);
+
+ // Copy the return value from %r2.
+ return DAG.getCopyFromReg(Chain, DL, SystemZ::R2D, PtrVT, Glue);
+}
+
SDValue SystemZTargetLowering::lowerGlobalTLSAddress(GlobalAddressSDNode *Node,
SelectionDAG &DAG) const {
SDLoc DL(Node);
const GlobalValue *GV = Node->getGlobal();
EVT PtrVT = getPointerTy();
- TLSModel::Model model = TM.getTLSModel(GV);
-
- if (model != TLSModel::LocalExec)
- llvm_unreachable("only local-exec TLS mode supported");
+ TLSModel::Model model = DAG.getTarget().getTLSModel(GV);
// The high part of the thread pointer is in access register 0.
SDValue TPHi = DAG.getNode(SystemZISD::EXTRACT_ACCESS, DL, MVT::i32,
DAG.getConstant(32, PtrVT));
SDValue TP = DAG.getNode(ISD::OR, DL, PtrVT, TPHiShifted, TPLo);
- // Get the offset of GA from the thread pointer.
- SystemZConstantPoolValue *CPV =
- SystemZConstantPoolValue::Create(GV, SystemZCP::NTPOFF);
+ // Get the offset of GA from the thread pointer, based on the TLS model.
+ SDValue Offset;
+ switch (model) {
+ case TLSModel::GeneralDynamic: {
+ // Load the GOT offset of the tls_index (module ID / per-symbol offset).
+ SystemZConstantPoolValue *CPV =
+ SystemZConstantPoolValue::Create(GV, SystemZCP::TLSGD);
+
+ Offset = DAG.getConstantPool(CPV, PtrVT, 8);
+ Offset = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(),
+ Offset, MachinePointerInfo::getConstantPool(),
+ false, false, false, 0);
+
+ // Call __tls_get_offset to retrieve the offset.
+ Offset = lowerTLSGetOffset(Node, DAG, SystemZISD::TLS_GDCALL, Offset);
+ break;
+ }
+
+ case TLSModel::LocalDynamic: {
+ // Load the GOT offset of the module ID.
+ SystemZConstantPoolValue *CPV =
+ SystemZConstantPoolValue::Create(GV, SystemZCP::TLSLDM);
+
+ Offset = DAG.getConstantPool(CPV, PtrVT, 8);
+ Offset = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(),
+ Offset, MachinePointerInfo::getConstantPool(),
+ false, false, false, 0);
- // Force the offset into the constant pool and load it from there.
- SDValue CPAddr = DAG.getConstantPool(CPV, PtrVT, 8);
- SDValue Offset = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(),
- CPAddr, MachinePointerInfo::getConstantPool(),
- false, false, false, 0);
+ // Call __tls_get_offset to retrieve the module base offset.
+ Offset = lowerTLSGetOffset(Node, DAG, SystemZISD::TLS_LDCALL, Offset);
+
+ // Note: The SystemZLDCleanupPass will remove redundant computations
+ // of the module base offset. Count total number of local-dynamic
+ // accesses to trigger execution of that pass.
+ SystemZMachineFunctionInfo* MFI =
+ DAG.getMachineFunction().getInfo<SystemZMachineFunctionInfo>();
+ MFI->incNumLocalDynamicTLSAccesses();
+
+ // Add the per-symbol offset.
+ CPV = SystemZConstantPoolValue::Create(GV, SystemZCP::DTPOFF);
+
+ SDValue DTPOffset = DAG.getConstantPool(CPV, PtrVT, 8);
+ DTPOffset = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(),
+ DTPOffset, MachinePointerInfo::getConstantPool(),
+ false, false, false, 0);
+
+ Offset = DAG.getNode(ISD::ADD, DL, PtrVT, Offset, DTPOffset);
+ break;
+ }
+
+ case TLSModel::InitialExec: {
+ // Load the offset from the GOT.
+ Offset = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
+ SystemZII::MO_INDNTPOFF);
+ Offset = DAG.getNode(SystemZISD::PCREL_WRAPPER, DL, PtrVT, Offset);
+ Offset = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(),
+ Offset, MachinePointerInfo::getGOT(),
+ false, false, false, 0);
+ break;
+ }
+
+ case TLSModel::LocalExec: {
+ // Force the offset into the constant pool and load it from there.
+ SystemZConstantPoolValue *CPV =
+ SystemZConstantPoolValue::Create(GV, SystemZCP::NTPOFF);
+
+ Offset = DAG.getConstantPool(CPV, PtrVT, 8);
+ Offset = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(),
+ Offset, MachinePointerInfo::getConstantPool(),
+ false, false, false, 0);
+ break;
+ }
+ }
// Add the base and offset together.
return DAG.getNode(ISD::ADD, DL, PtrVT, TP, Offset);
MVT::i64, HighOp, Low32);
}
+SDValue SystemZTargetLowering::lowerCTPOP(SDValue Op,
+ SelectionDAG &DAG) const {
+ EVT VT = Op.getValueType();
+ int64_t OrigBitSize = VT.getSizeInBits();
+ SDLoc DL(Op);
+
+ // Get the known-zero mask for the operand.
+ Op = Op.getOperand(0);
+ APInt KnownZero, KnownOne;
+ DAG.computeKnownBits(Op, KnownZero, KnownOne);
+ unsigned NumSignificantBits = (~KnownZero).getActiveBits();
+ if (NumSignificantBits == 0)
+ return DAG.getConstant(0, VT);
+
+ // Skip known-zero high parts of the operand.
+ int64_t BitSize = (int64_t)1 << Log2_32_Ceil(NumSignificantBits);
+ BitSize = std::min(BitSize, OrigBitSize);
+
+ // The POPCNT instruction counts the number of bits in each byte.
+ Op = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, Op);
+ Op = DAG.getNode(SystemZISD::POPCNT, DL, MVT::i64, Op);
+ Op = DAG.getNode(ISD::TRUNCATE, DL, VT, Op);
+
+ // Add up per-byte counts in a binary tree. All bits of Op at
+ // position larger than BitSize remain zero throughout.
+ for (int64_t I = BitSize / 2; I >= 8; I = I / 2) {
+ SDValue Tmp = DAG.getNode(ISD::SHL, DL, VT, Op, DAG.getConstant(I, VT));
+ if (BitSize != OrigBitSize)
+ Tmp = DAG.getNode(ISD::AND, DL, VT, Tmp,
+ DAG.getConstant(((uint64_t)1 << BitSize) - 1, VT));
+ Op = DAG.getNode(ISD::ADD, DL, VT, Op, Tmp);
+ }
+
+ // Extract overall result from high byte.
+ if (BitSize > 8)
+ Op = DAG.getNode(ISD::SRL, DL, VT, Op, DAG.getConstant(BitSize - 8, VT));
+
+ return Op;
+}
+
// Op is an atomic load. Lower it into a normal volatile load.
SDValue SystemZTargetLowering::lowerATOMIC_LOAD(SDValue Op,
SelectionDAG &DAG) const {
// Use an addition if the operand is constant and either LAA(G) is
// available or the negative value is in the range of A(G)FHI.
int64_t Value = (-Op2->getAPIntValue()).getSExtValue();
- if (isInt<32>(Value) || TM.getSubtargetImpl()->hasInterlockedAccess1())
+ if (isInt<32>(Value) || Subtarget.hasInterlockedAccess1())
NegSrc2 = DAG.getConstant(Value, MemVT);
- } else if (TM.getSubtargetImpl()->hasInterlockedAccess1())
+ } else if (Subtarget.hasInterlockedAccess1())
// Use LAA(G) if available.
NegSrc2 = DAG.getNode(ISD::SUB, DL, MemVT, DAG.getConstant(0, MemVT),
Src2);
Node->getMemoryVT(), Node->getMemOperand());
}
+// Return an i32 that contains the value of CC immediately after After,
+// whose final operand must be MVT::Glue.
+static SDValue getCCResult(SelectionDAG &DAG, SDNode *After) {
+ SDValue Glue = SDValue(After, After->getNumValues() - 1);
+ SDValue IPM = DAG.getNode(SystemZISD::IPM, SDLoc(After), MVT::i32, Glue);
+ return DAG.getNode(ISD::SRL, SDLoc(After), MVT::i32, IPM,
+ DAG.getConstant(SystemZ::IPM_CC, MVT::i32));
+}
+
+SDValue
+SystemZTargetLowering::lowerINTRINSIC_W_CHAIN(SDValue Op,
+ SelectionDAG &DAG) const {
+ unsigned Opcode, CCValid;
+ if (isIntrinsicWithCCAndChain(Op, Opcode, CCValid)) {
+ assert(Op->getNumValues() == 2 && "Expected only CC result and chain");
+ SDValue Glued = emitIntrinsicWithChainAndGlue(DAG, Op, Opcode);
+ SDValue CC = getCCResult(DAG, Glued.getNode());
+ DAG.ReplaceAllUsesOfValueWith(SDValue(Op.getNode(), 0), CC);
+ return SDValue();
+ }
+
+ return SDValue();
+}
+
SDValue SystemZTargetLowering::LowerOperation(SDValue Op,
SelectionDAG &DAG) const {
switch (Op.getOpcode()) {
return lowerUDIVREM(Op, DAG);
case ISD::OR:
return lowerOR(Op, DAG);
+ case ISD::CTPOP:
+ return lowerCTPOP(Op, DAG);
case ISD::ATOMIC_SWAP:
return lowerATOMIC_LOAD_OP(Op, DAG, SystemZISD::ATOMIC_SWAPW);
case ISD::ATOMIC_STORE:
return lowerSTACKRESTORE(Op, DAG);
case ISD::PREFETCH:
return lowerPREFETCH(Op, DAG);
+ case ISD::INTRINSIC_W_CHAIN:
+ return lowerINTRINSIC_W_CHAIN(Op, DAG);
default:
llvm_unreachable("Unexpected node to lower");
}
OPCODE(SEARCH_STRING);
OPCODE(IPM);
OPCODE(SERIALIZE);
+ OPCODE(TBEGIN);
+ OPCODE(TBEGIN_NOFLOAT);
+ OPCODE(TEND);
OPCODE(ATOMIC_SWAPW);
OPCODE(ATOMIC_LOADW_ADD);
OPCODE(ATOMIC_LOADW_SUB);
MachineBasicBlock *
SystemZTargetLowering::emitSelect(MachineInstr *MI,
MachineBasicBlock *MBB) const {
- const SystemZInstrInfo *TII = TM.getInstrInfo();
+ const SystemZInstrInfo *TII =
+ static_cast<const SystemZInstrInfo *>(Subtarget.getInstrInfo());
unsigned DestReg = MI->getOperand(0).getReg();
unsigned TrueReg = MI->getOperand(1).getReg();
MachineBasicBlock *MBB,
unsigned StoreOpcode, unsigned STOCOpcode,
bool Invert) const {
- const SystemZInstrInfo *TII = TM.getInstrInfo();
+ const SystemZInstrInfo *TII =
+ static_cast<const SystemZInstrInfo *>(Subtarget.getInstrInfo());
unsigned SrcReg = MI->getOperand(0).getReg();
MachineOperand Base = MI->getOperand(1);
// Use STOCOpcode if possible. We could use different store patterns in
// order to avoid matching the index register, but the performance trade-offs
// might be more complicated in that case.
- if (STOCOpcode && !IndexReg && TM.getSubtargetImpl()->hasLoadStoreOnCond()) {
+ if (STOCOpcode && !IndexReg && Subtarget.hasLoadStoreOnCond()) {
if (Invert)
CCMask ^= CCValid;
BuildMI(*MBB, MI, DL, TII->get(STOCOpcode))
unsigned BinOpcode,
unsigned BitSize,
bool Invert) const {
- const SystemZInstrInfo *TII = TM.getInstrInfo();
MachineFunction &MF = *MBB->getParent();
+ const SystemZInstrInfo *TII =
+ static_cast<const SystemZInstrInfo *>(Subtarget.getInstrInfo());
MachineRegisterInfo &MRI = MF.getRegInfo();
bool IsSubWord = (BitSize < 32);
unsigned Tmp = MRI.createVirtualRegister(RC);
BuildMI(MBB, DL, TII->get(BinOpcode), Tmp)
.addReg(RotatedOldVal).addOperand(Src2);
- if (BitSize < 32)
+ if (BitSize <= 32)
// XILF with the upper BitSize bits set.
BuildMI(MBB, DL, TII->get(SystemZ::XILF), RotatedNewVal)
- .addReg(Tmp).addImm(uint32_t(~0 << (32 - BitSize)));
- else if (BitSize == 32)
- // XILF with every bit set.
- BuildMI(MBB, DL, TII->get(SystemZ::XILF), RotatedNewVal)
- .addReg(Tmp).addImm(~uint32_t(0));
+ .addReg(Tmp).addImm(-1U << (32 - BitSize));
else {
// Use LCGR and add -1 to the result, which is more compact than
// an XILF, XILH pair.
unsigned CompareOpcode,
unsigned KeepOldMask,
unsigned BitSize) const {
- const SystemZInstrInfo *TII = TM.getInstrInfo();
MachineFunction &MF = *MBB->getParent();
+ const SystemZInstrInfo *TII =
+ static_cast<const SystemZInstrInfo *>(Subtarget.getInstrInfo());
MachineRegisterInfo &MRI = MF.getRegInfo();
bool IsSubWord = (BitSize < 32);
MachineBasicBlock *
SystemZTargetLowering::emitAtomicCmpSwapW(MachineInstr *MI,
MachineBasicBlock *MBB) const {
- const SystemZInstrInfo *TII = TM.getInstrInfo();
MachineFunction &MF = *MBB->getParent();
+ const SystemZInstrInfo *TII =
+ static_cast<const SystemZInstrInfo *>(Subtarget.getInstrInfo());
MachineRegisterInfo &MRI = MF.getRegInfo();
// Extract the operands. Base can be a register or a frame index.
SystemZTargetLowering::emitExt128(MachineInstr *MI,
MachineBasicBlock *MBB,
bool ClearEven, unsigned SubReg) const {
- const SystemZInstrInfo *TII = TM.getInstrInfo();
MachineFunction &MF = *MBB->getParent();
+ const SystemZInstrInfo *TII =
+ static_cast<const SystemZInstrInfo *>(Subtarget.getInstrInfo());
MachineRegisterInfo &MRI = MF.getRegInfo();
DebugLoc DL = MI->getDebugLoc();
SystemZTargetLowering::emitMemMemWrapper(MachineInstr *MI,
MachineBasicBlock *MBB,
unsigned Opcode) const {
- const SystemZInstrInfo *TII = TM.getInstrInfo();
MachineFunction &MF = *MBB->getParent();
+ const SystemZInstrInfo *TII =
+ static_cast<const SystemZInstrInfo *>(Subtarget.getInstrInfo());
MachineRegisterInfo &MRI = MF.getRegInfo();
DebugLoc DL = MI->getDebugLoc();
SystemZTargetLowering::emitStringWrapper(MachineInstr *MI,
MachineBasicBlock *MBB,
unsigned Opcode) const {
- const SystemZInstrInfo *TII = TM.getInstrInfo();
MachineFunction &MF = *MBB->getParent();
+ const SystemZInstrInfo *TII =
+ static_cast<const SystemZInstrInfo *>(Subtarget.getInstrInfo());
MachineRegisterInfo &MRI = MF.getRegInfo();
DebugLoc DL = MI->getDebugLoc();
return DoneMBB;
}
+// Update TBEGIN instruction with final opcode and register clobbers.
+MachineBasicBlock *
+SystemZTargetLowering::emitTransactionBegin(MachineInstr *MI,
+ MachineBasicBlock *MBB,
+ unsigned Opcode,
+ bool NoFloat) const {
+ MachineFunction &MF = *MBB->getParent();
+ const TargetFrameLowering *TFI = Subtarget.getFrameLowering();
+ const SystemZInstrInfo *TII = Subtarget.getInstrInfo();
+
+ // Update opcode.
+ MI->setDesc(TII->get(Opcode));
+
+ // We cannot handle a TBEGIN that clobbers the stack or frame pointer.
+ // Make sure to add the corresponding GRSM bits if they are missing.
+ uint64_t Control = MI->getOperand(2).getImm();
+ static const unsigned GPRControlBit[16] = {
+ 0x8000, 0x8000, 0x4000, 0x4000, 0x2000, 0x2000, 0x1000, 0x1000,
+ 0x0800, 0x0800, 0x0400, 0x0400, 0x0200, 0x0200, 0x0100, 0x0100
+ };
+ Control |= GPRControlBit[15];
+ if (TFI->hasFP(MF))
+ Control |= GPRControlBit[11];
+ MI->getOperand(2).setImm(Control);
+
+ // Add GPR clobbers.
+ for (int I = 0; I < 16; I++) {
+ if ((Control & GPRControlBit[I]) == 0) {
+ unsigned Reg = SystemZMC::GR64Regs[I];
+ MI->addOperand(MachineOperand::CreateReg(Reg, true, true));
+ }
+ }
+
+ // Add FPR clobbers.
+ if (!NoFloat && (Control & 4) != 0) {
+ for (int I = 0; I < 16; I++) {
+ unsigned Reg = SystemZMC::FP64Regs[I];
+ MI->addOperand(MachineOperand::CreateReg(Reg, true, true));
+ }
+ }
+
+ return MBB;
+}
+
MachineBasicBlock *SystemZTargetLowering::
EmitInstrWithCustomInserter(MachineInstr *MI, MachineBasicBlock *MBB) const {
switch (MI->getOpcode()) {
return emitStringWrapper(MI, MBB, SystemZ::MVST);
case SystemZ::SRSTLoop:
return emitStringWrapper(MI, MBB, SystemZ::SRST);
+ case SystemZ::TBEGIN:
+ return emitTransactionBegin(MI, MBB, SystemZ::TBEGIN, false);
+ case SystemZ::TBEGIN_nofloat:
+ return emitTransactionBegin(MI, MBB, SystemZ::TBEGIN, true);
+ case SystemZ::TBEGINC:
+ return emitTransactionBegin(MI, MBB, SystemZ::TBEGINC, true);
default:
llvm_unreachable("Unexpected instr type to insert");
}