blob: 46b1f35a6fc7f5d13ece654f19c6f9607aa93e19 [file] [log] [blame] [edit]
//===- MipsISelLowering.cpp - Mips DAG Lowering Implementation ------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file defines the interfaces that Mips uses to lower LLVM code into a
// selection DAG.
//
//===----------------------------------------------------------------------===//
#include "MipsISelLowering.h"
#include "MCTargetDesc/MipsBaseInfo.h"
#include "MCTargetDesc/MipsInstPrinter.h"
#include "MCTargetDesc/MipsMCTargetDesc.h"
#include "MipsCCState.h"
#include "MipsInstrInfo.h"
#include "MipsMachineFunction.h"
#include "MipsRegisterInfo.h"
#include "MipsSubtarget.h"
#include "MipsTargetMachine.h"
#include "MipsTargetObjectFile.h"
#include "llvm/ADT/APFloat.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/ADT/StringSwitch.h"
#include "llvm/CodeGen/CallingConvLower.h"
#include "llvm/CodeGen/FunctionLoweringInfo.h"
#include "llvm/CodeGen/ISDOpcodes.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineMemOperand.h"
#include "llvm/CodeGen/MachineOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/RuntimeLibcalls.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/SelectionDAGNodes.h"
#include "llvm/CodeGen/TargetFrameLowering.h"
#include "llvm/CodeGen/TargetInstrInfo.h"
#include "llvm/CodeGen/TargetRegisterInfo.h"
#include "llvm/CodeGen/ValueTypes.h"
#include "llvm/IR/CallingConv.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/DebugLoc.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/Type.h"
#include "llvm/IR/Value.h"
#include "llvm/MC/MCContext.h"
#include "llvm/MC/MCRegisterInfo.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CodeGen.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MachineValueType.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
#include <algorithm>
#include <cassert>
#include <cctype>
#include <cstdint>
#include <deque>
#include <iterator>
#include <utility>
#include <vector>
using namespace llvm;
#define DEBUG_TYPE "mips-lower"
STATISTIC(NumTailCalls, "Number of tail calls");
static cl::opt<bool>
NoZeroDivCheck("mno-check-zero-division", cl::Hidden,
cl::desc("MIPS: Don't trap on integer division by zero."),
cl::init(false));
extern cl::opt<bool> EmitJalrReloc;
static const MCPhysReg Mips64DPRegs[8] = {
Mips::D12_64, Mips::D13_64, Mips::D14_64, Mips::D15_64,
Mips::D16_64, Mips::D17_64, Mips::D18_64, Mips::D19_64
};
// If I is a shifted mask, set the size (Size) and the first bit of the
// mask (Pos), and return true.
// For example, if I is 0x003ff800, (Pos, Size) = (11, 11).
static bool isShiftedMask(uint64_t I, uint64_t &Pos, uint64_t &Size) {
if (!isShiftedMask_64(I))
return false;
Size = countPopulation(I);
Pos = countTrailingZeros(I);
return true;
}
// The MIPS MSA ABI passes vector arguments in the integer register set.
// The number of integer registers used is dependant on the ABI used.
MVT MipsTargetLowering::getRegisterTypeForCallingConv(LLVMContext &Context,
CallingConv::ID CC,
EVT VT) const {
if (!VT.isVector())
return getRegisterType(Context, VT);
return Subtarget.isABI_O32() || VT.getSizeInBits() == 32 ? MVT::i32
: MVT::i64;
}
unsigned MipsTargetLowering::getNumRegistersForCallingConv(LLVMContext &Context,
CallingConv::ID CC,
EVT VT) const {
if (VT.isVector())
return std::max(((unsigned)VT.getSizeInBits() /
(Subtarget.isABI_O32() ? 32 : 64)),
1U);
return MipsTargetLowering::getNumRegisters(Context, VT);
}
unsigned MipsTargetLowering::getVectorTypeBreakdownForCallingConv(
LLVMContext &Context, CallingConv::ID CC, EVT VT, EVT &IntermediateVT,
unsigned &NumIntermediates, MVT &RegisterVT) const {
// Break down vector types to either 2 i64s or 4 i32s.
RegisterVT = getRegisterTypeForCallingConv(Context, CC, VT);
IntermediateVT = RegisterVT;
NumIntermediates = VT.getSizeInBits() < RegisterVT.getSizeInBits()
? VT.getVectorNumElements()
: VT.getSizeInBits() / RegisterVT.getSizeInBits();
return NumIntermediates;
}
SDValue MipsTargetLowering::getGlobalReg(SelectionDAG &DAG, EVT Ty) const {
MipsFunctionInfo *FI = DAG.getMachineFunction().getInfo<MipsFunctionInfo>();
return DAG.getRegister(FI->getGlobalBaseReg(), Ty);
}
SDValue MipsTargetLowering::getTargetNode(GlobalAddressSDNode *N, EVT Ty,
SelectionDAG &DAG,
unsigned Flag) const {
return DAG.getTargetGlobalAddress(N->getGlobal(), SDLoc(N), Ty, 0, Flag);
}
SDValue MipsTargetLowering::getTargetNode(ExternalSymbolSDNode *N, EVT Ty,
SelectionDAG &DAG,
unsigned Flag) const {
return DAG.getTargetExternalSymbol(N->getSymbol(), Ty, Flag);
}
SDValue MipsTargetLowering::getTargetNode(BlockAddressSDNode *N, EVT Ty,
SelectionDAG &DAG,
unsigned Flag) const {
return DAG.getTargetBlockAddress(N->getBlockAddress(), Ty, 0, Flag);
}
SDValue MipsTargetLowering::getTargetNode(JumpTableSDNode *N, EVT Ty,
SelectionDAG &DAG,
unsigned Flag) const {
return DAG.getTargetJumpTable(N->getIndex(), Ty, Flag);
}
SDValue MipsTargetLowering::getTargetNode(ConstantPoolSDNode *N, EVT Ty,
SelectionDAG &DAG,
unsigned Flag) const {
return DAG.getTargetConstantPool(N->getConstVal(), Ty, N->getAlignment(),
N->getOffset(), Flag);
}
const char *MipsTargetLowering::getTargetNodeName(unsigned Opcode) const {
switch ((MipsISD::NodeType)Opcode) {
case MipsISD::FIRST_NUMBER: break;
case MipsISD::JmpLink: return "MipsISD::JmpLink";
case MipsISD::TailCall: return "MipsISD::TailCall";
case MipsISD::Highest: return "MipsISD::Highest";
case MipsISD::Higher: return "MipsISD::Higher";
case MipsISD::Hi: return "MipsISD::Hi";
case MipsISD::Lo: return "MipsISD::Lo";
case MipsISD::GotHi: return "MipsISD::GotHi";
case MipsISD::TlsHi: return "MipsISD::TlsHi";
case MipsISD::GPRel: return "MipsISD::GPRel";
case MipsISD::ThreadPointer: return "MipsISD::ThreadPointer";
case MipsISD::Ret: return "MipsISD::Ret";
case MipsISD::ERet: return "MipsISD::ERet";
case MipsISD::EH_RETURN: return "MipsISD::EH_RETURN";
case MipsISD::FMS: return "MipsISD::FMS";
case MipsISD::FPBrcond: return "MipsISD::FPBrcond";
case MipsISD::FPCmp: return "MipsISD::FPCmp";
case MipsISD::FSELECT: return "MipsISD::FSELECT";
case MipsISD::MTC1_D64: return "MipsISD::MTC1_D64";
case MipsISD::CMovFP_T: return "MipsISD::CMovFP_T";
case MipsISD::CMovFP_F: return "MipsISD::CMovFP_F";
case MipsISD::TruncIntFP: return "MipsISD::TruncIntFP";
case MipsISD::MFHI: return "MipsISD::MFHI";
case MipsISD::MFLO: return "MipsISD::MFLO";
case MipsISD::MTLOHI: return "MipsISD::MTLOHI";
case MipsISD::Mult: return "MipsISD::Mult";
case MipsISD::Multu: return "MipsISD::Multu";
case MipsISD::MAdd: return "MipsISD::MAdd";
case MipsISD::MAddu: return "MipsISD::MAddu";
case MipsISD::MSub: return "MipsISD::MSub";
case MipsISD::MSubu: return "MipsISD::MSubu";
case MipsISD::DivRem: return "MipsISD::DivRem";
case MipsISD::DivRemU: return "MipsISD::DivRemU";
case MipsISD::DivRem16: return "MipsISD::DivRem16";
case MipsISD::DivRemU16: return "MipsISD::DivRemU16";
case MipsISD::BuildPairF64: return "MipsISD::BuildPairF64";
case MipsISD::ExtractElementF64: return "MipsISD::ExtractElementF64";
case MipsISD::Wrapper: return "MipsISD::Wrapper";
case MipsISD::DynAlloc: return "MipsISD::DynAlloc";
case MipsISD::Sync: return "MipsISD::Sync";
case MipsISD::Ext: return "MipsISD::Ext";
case MipsISD::Ins: return "MipsISD::Ins";
case MipsISD::CIns: return "MipsISD::CIns";
case MipsISD::LWL: return "MipsISD::LWL";
case MipsISD::LWR: return "MipsISD::LWR";
case MipsISD::SWL: return "MipsISD::SWL";
case MipsISD::SWR: return "MipsISD::SWR";
case MipsISD::LDL: return "MipsISD::LDL";
case MipsISD::LDR: return "MipsISD::LDR";
case MipsISD::SDL: return "MipsISD::SDL";
case MipsISD::SDR: return "MipsISD::SDR";
case MipsISD::EXTP: return "MipsISD::EXTP";
case MipsISD::EXTPDP: return "MipsISD::EXTPDP";
case MipsISD::EXTR_S_H: return "MipsISD::EXTR_S_H";
case MipsISD::EXTR_W: return "MipsISD::EXTR_W";
case MipsISD::EXTR_R_W: return "MipsISD::EXTR_R_W";
case MipsISD::EXTR_RS_W: return "MipsISD::EXTR_RS_W";
case MipsISD::SHILO: return "MipsISD::SHILO";
case MipsISD::MTHLIP: return "MipsISD::MTHLIP";
case MipsISD::MULSAQ_S_W_PH: return "MipsISD::MULSAQ_S_W_PH";
case MipsISD::MAQ_S_W_PHL: return "MipsISD::MAQ_S_W_PHL";
case MipsISD::MAQ_S_W_PHR: return "MipsISD::MAQ_S_W_PHR";
case MipsISD::MAQ_SA_W_PHL: return "MipsISD::MAQ_SA_W_PHL";
case MipsISD::MAQ_SA_W_PHR: return "MipsISD::MAQ_SA_W_PHR";
case MipsISD::DPAU_H_QBL: return "MipsISD::DPAU_H_QBL";
case MipsISD::DPAU_H_QBR: return "MipsISD::DPAU_H_QBR";
case MipsISD::DPSU_H_QBL: return "MipsISD::DPSU_H_QBL";
case MipsISD::DPSU_H_QBR: return "MipsISD::DPSU_H_QBR";
case MipsISD::DPAQ_S_W_PH: return "MipsISD::DPAQ_S_W_PH";
case MipsISD::DPSQ_S_W_PH: return "MipsISD::DPSQ_S_W_PH";
case MipsISD::DPAQ_SA_L_W: return "MipsISD::DPAQ_SA_L_W";
case MipsISD::DPSQ_SA_L_W: return "MipsISD::DPSQ_SA_L_W";
case MipsISD::DPA_W_PH: return "MipsISD::DPA_W_PH";
case MipsISD::DPS_W_PH: return "MipsISD::DPS_W_PH";
case MipsISD::DPAQX_S_W_PH: return "MipsISD::DPAQX_S_W_PH";
case MipsISD::DPAQX_SA_W_PH: return "MipsISD::DPAQX_SA_W_PH";
case MipsISD::DPAX_W_PH: return "MipsISD::DPAX_W_PH";
case MipsISD::DPSX_W_PH: return "MipsISD::DPSX_W_PH";
case MipsISD::DPSQX_S_W_PH: return "MipsISD::DPSQX_S_W_PH";
case MipsISD::DPSQX_SA_W_PH: return "MipsISD::DPSQX_SA_W_PH";
case MipsISD::MULSA_W_PH: return "MipsISD::MULSA_W_PH";
case MipsISD::MULT: return "MipsISD::MULT";
case MipsISD::MULTU: return "MipsISD::MULTU";
case MipsISD::MADD_DSP: return "MipsISD::MADD_DSP";
case MipsISD::MADDU_DSP: return "MipsISD::MADDU_DSP";
case MipsISD::MSUB_DSP: return "MipsISD::MSUB_DSP";
case MipsISD::MSUBU_DSP: return "MipsISD::MSUBU_DSP";
case MipsISD::SHLL_DSP: return "MipsISD::SHLL_DSP";
case MipsISD::SHRA_DSP: return "MipsISD::SHRA_DSP";
case MipsISD::SHRL_DSP: return "MipsISD::SHRL_DSP";
case MipsISD::SETCC_DSP: return "MipsISD::SETCC_DSP";
case MipsISD::SELECT_CC_DSP: return "MipsISD::SELECT_CC_DSP";
case MipsISD::VALL_ZERO: return "MipsISD::VALL_ZERO";
case MipsISD::VANY_ZERO: return "MipsISD::VANY_ZERO";
case MipsISD::VALL_NONZERO: return "MipsISD::VALL_NONZERO";
case MipsISD::VANY_NONZERO: return "MipsISD::VANY_NONZERO";
case MipsISD::VCEQ: return "MipsISD::VCEQ";
case MipsISD::VCLE_S: return "MipsISD::VCLE_S";
case MipsISD::VCLE_U: return "MipsISD::VCLE_U";
case MipsISD::VCLT_S: return "MipsISD::VCLT_S";
case MipsISD::VCLT_U: return "MipsISD::VCLT_U";
case MipsISD::VEXTRACT_SEXT_ELT: return "MipsISD::VEXTRACT_SEXT_ELT";
case MipsISD::VEXTRACT_ZEXT_ELT: return "MipsISD::VEXTRACT_ZEXT_ELT";
case MipsISD::VNOR: return "MipsISD::VNOR";
case MipsISD::VSHF: return "MipsISD::VSHF";
case MipsISD::SHF: return "MipsISD::SHF";
case MipsISD::ILVEV: return "MipsISD::ILVEV";
case MipsISD::ILVOD: return "MipsISD::ILVOD";
case MipsISD::ILVL: return "MipsISD::ILVL";
case MipsISD::ILVR: return "MipsISD::ILVR";
case MipsISD::PCKEV: return "MipsISD::PCKEV";
case MipsISD::PCKOD: return "MipsISD::PCKOD";
case MipsISD::INSVE: return "MipsISD::INSVE";
}
return nullptr;
}
MipsTargetLowering::MipsTargetLowering(const MipsTargetMachine &TM,
const MipsSubtarget &STI)
: TargetLowering(TM), Subtarget(STI), ABI(TM.getABI()) {
// Mips does not have i1 type, so use i32 for
// setcc operations results (slt, sgt, ...).
setBooleanContents(ZeroOrOneBooleanContent);
setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
// The cmp.cond.fmt instruction in MIPS32r6/MIPS64r6 uses 0 and -1 like MSA
// does. Integer booleans still use 0 and 1.
if (Subtarget.hasMips32r6())
setBooleanContents(ZeroOrOneBooleanContent,
ZeroOrNegativeOneBooleanContent);
// Load extented operations for i1 types must be promoted
for (MVT VT : MVT::integer_valuetypes()) {
setLoadExtAction(ISD::EXTLOAD, VT, MVT::i1, Promote);
setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i1, Promote);
setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
}
// MIPS doesn't have extending float->double load/store. Set LoadExtAction
// for f32, f16
for (MVT VT : MVT::fp_valuetypes()) {
setLoadExtAction(ISD::EXTLOAD, VT, MVT::f32, Expand);
setLoadExtAction(ISD::EXTLOAD, VT, MVT::f16, Expand);
}
// Set LoadExtAction for f16 vectors to Expand
for (MVT VT : MVT::fp_fixedlen_vector_valuetypes()) {
MVT F16VT = MVT::getVectorVT(MVT::f16, VT.getVectorNumElements());
if (F16VT.isValid())
setLoadExtAction(ISD::EXTLOAD, VT, F16VT, Expand);
}
setTruncStoreAction(MVT::f32, MVT::f16, Expand);
setTruncStoreAction(MVT::f64, MVT::f16, Expand);
setTruncStoreAction(MVT::f64, MVT::f32, Expand);
// Used by legalize types to correctly generate the setcc result.
// Without this, every float setcc comes with a AND/OR with the result,
// we don't want this, since the fpcmp result goes to a flag register,
// which is used implicitly by brcond and select operations.
AddPromotedToType(ISD::SETCC, MVT::i1, MVT::i32);
// Mips Custom Operations
setOperationAction(ISD::BR_JT, MVT::Other, Expand);
setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
setOperationAction(ISD::BlockAddress, MVT::i32, Custom);
setOperationAction(ISD::GlobalTLSAddress, MVT::i32, Custom);
setOperationAction(ISD::JumpTable, MVT::i32, Custom);
setOperationAction(ISD::ConstantPool, MVT::i32, Custom);
setOperationAction(ISD::SELECT, MVT::f32, Custom);
setOperationAction(ISD::SELECT, MVT::f64, Custom);
setOperationAction(ISD::SELECT, MVT::i32, Custom);
setOperationAction(ISD::SETCC, MVT::f32, Custom);
setOperationAction(ISD::SETCC, MVT::f64, Custom);
setOperationAction(ISD::BRCOND, MVT::Other, Custom);
setOperationAction(ISD::FCOPYSIGN, MVT::f32, Custom);
setOperationAction(ISD::FCOPYSIGN, MVT::f64, Custom);
setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
if (!(TM.Options.NoNaNsFPMath || Subtarget.inAbs2008Mode())) {
setOperationAction(ISD::FABS, MVT::f32, Custom);
setOperationAction(ISD::FABS, MVT::f64, Custom);
}
if (Subtarget.isGP64bit()) {
setOperationAction(ISD::GlobalAddress, MVT::i64, Custom);
setOperationAction(ISD::BlockAddress, MVT::i64, Custom);
setOperationAction(ISD::GlobalTLSAddress, MVT::i64, Custom);
setOperationAction(ISD::JumpTable, MVT::i64, Custom);
setOperationAction(ISD::ConstantPool, MVT::i64, Custom);
setOperationAction(ISD::SELECT, MVT::i64, Custom);
setOperationAction(ISD::LOAD, MVT::i64, Custom);
setOperationAction(ISD::STORE, MVT::i64, Custom);
setOperationAction(ISD::FP_TO_SINT, MVT::i64, Custom);
setOperationAction(ISD::SHL_PARTS, MVT::i64, Custom);
setOperationAction(ISD::SRA_PARTS, MVT::i64, Custom);
setOperationAction(ISD::SRL_PARTS, MVT::i64, Custom);
}
if (!Subtarget.isGP64bit()) {
setOperationAction(ISD::SHL_PARTS, MVT::i32, Custom);
setOperationAction(ISD::SRA_PARTS, MVT::i32, Custom);
setOperationAction(ISD::SRL_PARTS, MVT::i32, Custom);
}
setOperationAction(ISD::EH_DWARF_CFA, MVT::i32, Custom);
if (Subtarget.isGP64bit())
setOperationAction(ISD::EH_DWARF_CFA, MVT::i64, Custom);
setOperationAction(ISD::SDIV, MVT::i32, Expand);
setOperationAction(ISD::SREM, MVT::i32, Expand);
setOperationAction(ISD::UDIV, MVT::i32, Expand);
setOperationAction(ISD::UREM, MVT::i32, Expand);
setOperationAction(ISD::SDIV, MVT::i64, Expand);
setOperationAction(ISD::SREM, MVT::i64, Expand);
setOperationAction(ISD::UDIV, MVT::i64, Expand);
setOperationAction(ISD::UREM, MVT::i64, Expand);
// Operations not directly supported by Mips.
setOperationAction(ISD::BR_CC, MVT::f32, Expand);
setOperationAction(ISD::BR_CC, MVT::f64, Expand);
setOperationAction(ISD::BR_CC, MVT::i32, Expand);
setOperationAction(ISD::BR_CC, MVT::i64, Expand);
setOperationAction(ISD::SELECT_CC, MVT::i32, Expand);
setOperationAction(ISD::SELECT_CC, MVT::i64, Expand);
setOperationAction(ISD::SELECT_CC, MVT::f32, Expand);
setOperationAction(ISD::SELECT_CC, MVT::f64, Expand);
setOperationAction(ISD::UINT_TO_FP, MVT::i32, Expand);
setOperationAction(ISD::UINT_TO_FP, MVT::i64, Expand);
setOperationAction(ISD::FP_TO_UINT, MVT::i32, Expand);
setOperationAction(ISD::FP_TO_UINT, MVT::i64, Expand);
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
if (Subtarget.hasCnMips()) {
setOperationAction(ISD::CTPOP, MVT::i32, Legal);
setOperationAction(ISD::CTPOP, MVT::i64, Legal);
} else {
setOperationAction(ISD::CTPOP, MVT::i32, Expand);
setOperationAction(ISD::CTPOP, MVT::i64, Expand);
}
setOperationAction(ISD::CTTZ, MVT::i32, Expand);
setOperationAction(ISD::CTTZ, MVT::i64, Expand);
setOperationAction(ISD::ROTL, MVT::i32, Expand);
setOperationAction(ISD::ROTL, MVT::i64, Expand);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64, Expand);
if (!Subtarget.hasMips32r2())
setOperationAction(ISD::ROTR, MVT::i32, Expand);
if (!Subtarget.hasMips64r2())
setOperationAction(ISD::ROTR, MVT::i64, Expand);
setOperationAction(ISD::FSIN, MVT::f32, Expand);
setOperationAction(ISD::FSIN, MVT::f64, Expand);
setOperationAction(ISD::FCOS, MVT::f32, Expand);
setOperationAction(ISD::FCOS, MVT::f64, Expand);
setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
setOperationAction(ISD::FPOW, MVT::f32, Expand);
setOperationAction(ISD::FPOW, MVT::f64, Expand);
setOperationAction(ISD::FLOG, MVT::f32, Expand);
setOperationAction(ISD::FLOG2, MVT::f32, Expand);
setOperationAction(ISD::FLOG10, MVT::f32, Expand);
setOperationAction(ISD::FEXP, MVT::f32, Expand);
setOperationAction(ISD::FMA, MVT::f32, Expand);
setOperationAction(ISD::FMA, MVT::f64, Expand);
setOperationAction(ISD::FREM, MVT::f32, Expand);
setOperationAction(ISD::FREM, MVT::f64, Expand);
// Lower f16 conversion operations into library calls
setOperationAction(ISD::FP16_TO_FP, MVT::f32, Expand);
setOperationAction(ISD::FP_TO_FP16, MVT::f32, Expand);
setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
setOperationAction(ISD::FP_TO_FP16, MVT::f64, Expand);
setOperationAction(ISD::EH_RETURN, MVT::Other, Custom);
setOperationAction(ISD::VASTART, MVT::Other, Custom);
setOperationAction(ISD::VAARG, MVT::Other, Custom);
setOperationAction(ISD::VACOPY, MVT::Other, Expand);
setOperationAction(ISD::VAEND, MVT::Other, Expand);
// Use the default for now
setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
if (!Subtarget.isGP64bit()) {
setOperationAction(ISD::ATOMIC_LOAD, MVT::i64, Expand);
setOperationAction(ISD::ATOMIC_STORE, MVT::i64, Expand);
}
if (!Subtarget.hasMips32r2()) {
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i8, Expand);
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i16, Expand);
}
// MIPS16 lacks MIPS32's clz and clo instructions.
if (!Subtarget.hasMips32() || Subtarget.inMips16Mode())
setOperationAction(ISD::CTLZ, MVT::i32, Expand);
if (!Subtarget.hasMips64())
setOperationAction(ISD::CTLZ, MVT::i64, Expand);
if (!Subtarget.hasMips32r2())
setOperationAction(ISD::BSWAP, MVT::i32, Expand);
if (!Subtarget.hasMips64r2())
setOperationAction(ISD::BSWAP, MVT::i64, Expand);
if (Subtarget.isGP64bit()) {
setLoadExtAction(ISD::SEXTLOAD, MVT::i64, MVT::i32, Custom);
setLoadExtAction(ISD::ZEXTLOAD, MVT::i64, MVT::i32, Custom);
setLoadExtAction(ISD::EXTLOAD, MVT::i64, MVT::i32, Custom);
setTruncStoreAction(MVT::i64, MVT::i32, Custom);
}
setOperationAction(ISD::TRAP, MVT::Other, Legal);
setTargetDAGCombine(ISD::SDIVREM);
setTargetDAGCombine(ISD::UDIVREM);
setTargetDAGCombine(ISD::SELECT);
setTargetDAGCombine(ISD::AND);
setTargetDAGCombine(ISD::OR);
setTargetDAGCombine(ISD::ADD);
setTargetDAGCombine(ISD::SUB);
setTargetDAGCombine(ISD::AssertZext);
setTargetDAGCombine(ISD::SHL);
if (ABI.IsO32()) {
// These libcalls are not available in 32-bit.
setLibcallName(RTLIB::SHL_I128, nullptr);
setLibcallName(RTLIB::SRL_I128, nullptr);
setLibcallName(RTLIB::SRA_I128, nullptr);
}
setMinFunctionAlignment(Subtarget.isGP64bit() ? Align(8) : Align(4));
// The arguments on the stack are defined in terms of 4-byte slots on O32
// and 8-byte slots on N32/N64.
setMinStackArgumentAlignment((ABI.IsN32() || ABI.IsN64()) ? Align(8)
: Align(4));
setStackPointerRegisterToSaveRestore(ABI.IsN64() ? Mips::SP_64 : Mips::SP);
MaxStoresPerMemcpy = 16;
isMicroMips = Subtarget.inMicroMipsMode();
}
const MipsTargetLowering *
MipsTargetLowering::create(const MipsTargetMachine &TM,
const MipsSubtarget &STI) {
if (STI.inMips16Mode())
return createMips16TargetLowering(TM, STI);
return createMipsSETargetLowering(TM, STI);
}
// Create a fast isel object.
FastISel *
MipsTargetLowering::createFastISel(FunctionLoweringInfo &funcInfo,
const TargetLibraryInfo *libInfo) const {
const MipsTargetMachine &TM =
static_cast<const MipsTargetMachine &>(funcInfo.MF->getTarget());
// We support only the standard encoding [MIPS32,MIPS32R5] ISAs.
bool UseFastISel = TM.Options.EnableFastISel && Subtarget.hasMips32() &&
!Subtarget.hasMips32r6() && !Subtarget.inMips16Mode() &&
!Subtarget.inMicroMipsMode();
// Disable if either of the following is true:
// We do not generate PIC, the ABI is not O32, XGOT is being used.
if (!TM.isPositionIndependent() || !TM.getABI().IsO32() ||
Subtarget.useXGOT())
UseFastISel = false;
return UseFastISel ? Mips::createFastISel(funcInfo, libInfo) : nullptr;
}
EVT MipsTargetLowering::getSetCCResultType(const DataLayout &, LLVMContext &,
EVT VT) const {
if (!VT.isVector())
return MVT::i32;
return VT.changeVectorElementTypeToInteger();
}
static SDValue performDivRemCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const MipsSubtarget &Subtarget) {
if (DCI.isBeforeLegalizeOps())
return SDValue();
EVT Ty = N->getValueType(0);
unsigned LO = (Ty == MVT::i32) ? Mips::LO0 : Mips::LO0_64;
unsigned HI = (Ty == MVT::i32) ? Mips::HI0 : Mips::HI0_64;
unsigned Opc = N->getOpcode() == ISD::SDIVREM ? MipsISD::DivRem16 :
MipsISD::DivRemU16;
SDLoc DL(N);
SDValue DivRem = DAG.getNode(Opc, DL, MVT::Glue,
N->getOperand(0), N->getOperand(1));
SDValue InChain = DAG.getEntryNode();
SDValue InGlue = DivRem;
// insert MFLO
if (N->hasAnyUseOfValue(0)) {
SDValue CopyFromLo = DAG.getCopyFromReg(InChain, DL, LO, Ty,
InGlue);
DAG.ReplaceAllUsesOfValueWith(SDValue(N, 0), CopyFromLo);
InChain = CopyFromLo.getValue(1);
InGlue = CopyFromLo.getValue(2);
}
// insert MFHI
if (N->hasAnyUseOfValue(1)) {
SDValue CopyFromHi = DAG.getCopyFromReg(InChain, DL,
HI, Ty, InGlue);
DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), CopyFromHi);
}
return SDValue();
}
static Mips::CondCode condCodeToFCC(ISD::CondCode CC) {
switch (CC) {
default: llvm_unreachable("Unknown fp condition code!");
case ISD::SETEQ:
case ISD::SETOEQ: return Mips::FCOND_OEQ;
case ISD::SETUNE: return Mips::FCOND_UNE;
case ISD::SETLT:
case ISD::SETOLT: return Mips::FCOND_OLT;
case ISD::SETGT:
case ISD::SETOGT: return Mips::FCOND_OGT;
case ISD::SETLE:
case ISD::SETOLE: return Mips::FCOND_OLE;
case ISD::SETGE:
case ISD::SETOGE: return Mips::FCOND_OGE;
case ISD::SETULT: return Mips::FCOND_ULT;
case ISD::SETULE: return Mips::FCOND_ULE;
case ISD::SETUGT: return Mips::FCOND_UGT;
case ISD::SETUGE: return Mips::FCOND_UGE;
case ISD::SETUO: return Mips::FCOND_UN;
case ISD::SETO: return Mips::FCOND_OR;
case ISD::SETNE:
case ISD::SETONE: return Mips::FCOND_ONE;
case ISD::SETUEQ: return Mips::FCOND_UEQ;
}
}
/// This function returns true if the floating point conditional branches and
/// conditional moves which use condition code CC should be inverted.
static bool invertFPCondCodeUser(Mips::CondCode CC) {
if (CC >= Mips::FCOND_F && CC <= Mips::FCOND_NGT)
return false;
assert((CC >= Mips::FCOND_T && CC <= Mips::FCOND_GT) &&
"Illegal Condition Code");
return true;
}
// Creates and returns an FPCmp node from a setcc node.
// Returns Op if setcc is not a floating point comparison.
static SDValue createFPCmp(SelectionDAG &DAG, const SDValue &Op) {
// must be a SETCC node
if (Op.getOpcode() != ISD::SETCC)
return Op;
SDValue LHS = Op.getOperand(0);
if (!LHS.getValueType().isFloatingPoint())
return Op;
SDValue RHS = Op.getOperand(1);
SDLoc DL(Op);
// Assume the 3rd operand is a CondCodeSDNode. Add code to check the type of
// node if necessary.
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
return DAG.getNode(MipsISD::FPCmp, DL, MVT::Glue, LHS, RHS,
DAG.getConstant(condCodeToFCC(CC), DL, MVT::i32));
}
// Creates and returns a CMovFPT/F node.
static SDValue createCMovFP(SelectionDAG &DAG, SDValue Cond, SDValue True,
SDValue False, const SDLoc &DL) {
ConstantSDNode *CC = cast<ConstantSDNode>(Cond.getOperand(2));
bool invert = invertFPCondCodeUser((Mips::CondCode)CC->getSExtValue());
SDValue FCC0 = DAG.getRegister(Mips::FCC0, MVT::i32);
return DAG.getNode((invert ? MipsISD::CMovFP_F : MipsISD::CMovFP_T), DL,
True.getValueType(), True, FCC0, False, Cond);
}
static SDValue performSELECTCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const MipsSubtarget &Subtarget) {
if (DCI.isBeforeLegalizeOps())
return SDValue();
SDValue SetCC = N->getOperand(0);
if ((SetCC.getOpcode() != ISD::SETCC) ||
!SetCC.getOperand(0).getValueType().isInteger())
return SDValue();
SDValue False = N->getOperand(2);
EVT FalseTy = False.getValueType();
if (!FalseTy.isInteger())
return SDValue();
ConstantSDNode *FalseC = dyn_cast<ConstantSDNode>(False);
// If the RHS (False) is 0, we swap the order of the operands
// of ISD::SELECT (obviously also inverting the condition) so that we can
// take advantage of conditional moves using the $0 register.
// Example:
// return (a != 0) ? x : 0;
// load $reg, x
// movz $reg, $0, a
if (!FalseC)
return SDValue();
const SDLoc DL(N);
if (!FalseC->getZExtValue()) {
ISD::CondCode CC = cast<CondCodeSDNode>(SetCC.getOperand(2))->get();
SDValue True = N->getOperand(1);
SetCC = DAG.getSetCC(DL, SetCC.getValueType(), SetCC.getOperand(0),
SetCC.getOperand(1),
ISD::getSetCCInverse(CC, SetCC.getValueType()));
return DAG.getNode(ISD::SELECT, DL, FalseTy, SetCC, False, True);
}
// If both operands are integer constants there's a possibility that we
// can do some interesting optimizations.
SDValue True = N->getOperand(1);
ConstantSDNode *TrueC = dyn_cast<ConstantSDNode>(True);
if (!TrueC || !True.getValueType().isInteger())
return SDValue();
// We'll also ignore MVT::i64 operands as this optimizations proves
// to be ineffective because of the required sign extensions as the result
// of a SETCC operator is always MVT::i32 for non-vector types.
if (True.getValueType() == MVT::i64)
return SDValue();
int64_t Diff = TrueC->getSExtValue() - FalseC->getSExtValue();
// 1) (a < x) ? y : y-1
// slti $reg1, a, x
// addiu $reg2, $reg1, y-1
if (Diff == 1)
return DAG.getNode(ISD::ADD, DL, SetCC.getValueType(), SetCC, False);
// 2) (a < x) ? y-1 : y
// slti $reg1, a, x
// xor $reg1, $reg1, 1
// addiu $reg2, $reg1, y-1
if (Diff == -1) {
ISD::CondCode CC = cast<CondCodeSDNode>(SetCC.getOperand(2))->get();
SetCC = DAG.getSetCC(DL, SetCC.getValueType(), SetCC.getOperand(0),
SetCC.getOperand(1),
ISD::getSetCCInverse(CC, SetCC.getValueType()));
return DAG.getNode(ISD::ADD, DL, SetCC.getValueType(), SetCC, True);
}
// Could not optimize.
return SDValue();
}
static SDValue performCMovFPCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const MipsSubtarget &Subtarget) {
if (DCI.isBeforeLegalizeOps())
return SDValue();
SDValue ValueIfTrue = N->getOperand(0), ValueIfFalse = N->getOperand(2);
ConstantSDNode *FalseC = dyn_cast<ConstantSDNode>(ValueIfFalse);
if (!FalseC || FalseC->getZExtValue())
return SDValue();
// Since RHS (False) is 0, we swap the order of the True/False operands
// (obviously also inverting the condition) so that we can
// take advantage of conditional moves using the $0 register.
// Example:
// return (a != 0) ? x : 0;
// load $reg, x
// movz $reg, $0, a
unsigned Opc = (N->getOpcode() == MipsISD::CMovFP_T) ? MipsISD::CMovFP_F :
MipsISD::CMovFP_T;
SDValue FCC = N->getOperand(1), Glue = N->getOperand(3);
return DAG.getNode(Opc, SDLoc(N), ValueIfFalse.getValueType(),
ValueIfFalse, FCC, ValueIfTrue, Glue);
}
static SDValue performANDCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const MipsSubtarget &Subtarget) {
if (DCI.isBeforeLegalizeOps() || !Subtarget.hasExtractInsert())
return SDValue();
SDValue FirstOperand = N->getOperand(0);
unsigned FirstOperandOpc = FirstOperand.getOpcode();
SDValue Mask = N->getOperand(1);
EVT ValTy = N->getValueType(0);
SDLoc DL(N);
uint64_t Pos = 0, SMPos, SMSize;
ConstantSDNode *CN;
SDValue NewOperand;
unsigned Opc;
// Op's second operand must be a shifted mask.
if (!(CN = dyn_cast<ConstantSDNode>(Mask)) ||
!isShiftedMask(CN->getZExtValue(), SMPos, SMSize))
return SDValue();
if (FirstOperandOpc == ISD::SRA || FirstOperandOpc == ISD::SRL) {
// Pattern match EXT.
// $dst = and ((sra or srl) $src , pos), (2**size - 1)
// => ext $dst, $src, pos, size
// The second operand of the shift must be an immediate.
if (!(CN = dyn_cast<ConstantSDNode>(FirstOperand.getOperand(1))))
return SDValue();
Pos = CN->getZExtValue();
// Return if the shifted mask does not start at bit 0 or the sum of its size
// and Pos exceeds the word's size.
if (SMPos != 0 || Pos + SMSize > ValTy.getSizeInBits())
return SDValue();
Opc = MipsISD::Ext;
NewOperand = FirstOperand.getOperand(0);
} else if (FirstOperandOpc == ISD::SHL && Subtarget.hasCnMips()) {
// Pattern match CINS.
// $dst = and (shl $src , pos), mask
// => cins $dst, $src, pos, size
// mask is a shifted mask with consecutive 1's, pos = shift amount,
// size = population count.
// The second operand of the shift must be an immediate.
if (!(CN = dyn_cast<ConstantSDNode>(FirstOperand.getOperand(1))))
return SDValue();
Pos = CN->getZExtValue();
if (SMPos != Pos || Pos >= ValTy.getSizeInBits() || SMSize >= 32 ||
Pos + SMSize > ValTy.getSizeInBits())
return SDValue();
NewOperand = FirstOperand.getOperand(0);
// SMSize is 'location' (position) in this case, not size.
SMSize--;
Opc = MipsISD::CIns;
} else {
// Pattern match EXT.
// $dst = and $src, (2**size - 1) , if size > 16
// => ext $dst, $src, pos, size , pos = 0
// If the mask is <= 0xffff, andi can be used instead.
if (CN->getZExtValue() <= 0xffff)
return SDValue();
// Return if the mask doesn't start at position 0.
if (SMPos)
return SDValue();
Opc = MipsISD::Ext;
NewOperand = FirstOperand;
}
return DAG.getNode(Opc, DL, ValTy, NewOperand,
DAG.getConstant(Pos, DL, MVT::i32),
DAG.getConstant(SMSize, DL, MVT::i32));
}
static SDValue performORCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const MipsSubtarget &Subtarget) {
// Pattern match INS.
// $dst = or (and $src1 , mask0), (and (shl $src, pos), mask1),
// where mask1 = (2**size - 1) << pos, mask0 = ~mask1
// => ins $dst, $src, size, pos, $src1
if (DCI.isBeforeLegalizeOps() || !Subtarget.hasExtractInsert())
return SDValue();
SDValue And0 = N->getOperand(0), And1 = N->getOperand(1);
uint64_t SMPos0, SMSize0, SMPos1, SMSize1;
ConstantSDNode *CN, *CN1;
// See if Op's first operand matches (and $src1 , mask0).
if (And0.getOpcode() != ISD::AND)
return SDValue();
if (!(CN = dyn_cast<ConstantSDNode>(And0.getOperand(1))) ||
!isShiftedMask(~CN->getSExtValue(), SMPos0, SMSize0))
return SDValue();
// See if Op's second operand matches (and (shl $src, pos), mask1).
if (And1.getOpcode() == ISD::AND &&
And1.getOperand(0).getOpcode() == ISD::SHL) {
if (!(CN = dyn_cast<ConstantSDNode>(And1.getOperand(1))) ||
!isShiftedMask(CN->getZExtValue(), SMPos1, SMSize1))
return SDValue();
// The shift masks must have the same position and size.
if (SMPos0 != SMPos1 || SMSize0 != SMSize1)
return SDValue();
SDValue Shl = And1.getOperand(0);
if (!(CN = dyn_cast<ConstantSDNode>(Shl.getOperand(1))))
return SDValue();
unsigned Shamt = CN->getZExtValue();
// Return if the shift amount and the first bit position of mask are not the
// same.
EVT ValTy = N->getValueType(0);
if ((Shamt != SMPos0) || (SMPos0 + SMSize0 > ValTy.getSizeInBits()))
return SDValue();
SDLoc DL(N);
return DAG.getNode(MipsISD::Ins, DL, ValTy, Shl.getOperand(0),
DAG.getConstant(SMPos0, DL, MVT::i32),
DAG.getConstant(SMSize0, DL, MVT::i32),
And0.getOperand(0));
} else {
// Pattern match DINS.
// $dst = or (and $src, mask0), mask1
// where mask0 = ((1 << SMSize0) -1) << SMPos0
// => dins $dst, $src, pos, size
if (~CN->getSExtValue() == ((((int64_t)1 << SMSize0) - 1) << SMPos0) &&
((SMSize0 + SMPos0 <= 64 && Subtarget.hasMips64r2()) ||
(SMSize0 + SMPos0 <= 32))) {
// Check if AND instruction has constant as argument
bool isConstCase = And1.getOpcode() != ISD::AND;
if (And1.getOpcode() == ISD::AND) {
if (!(CN1 = dyn_cast<ConstantSDNode>(And1->getOperand(1))))
return SDValue();
} else {
if (!(CN1 = dyn_cast<ConstantSDNode>(N->getOperand(1))))
return SDValue();
}
// Don't generate INS if constant OR operand doesn't fit into bits
// cleared by constant AND operand.
if (CN->getSExtValue() & CN1->getSExtValue())
return SDValue();
SDLoc DL(N);
EVT ValTy = N->getOperand(0)->getValueType(0);
SDValue Const1;
SDValue SrlX;
if (!isConstCase) {
Const1 = DAG.getConstant(SMPos0, DL, MVT::i32);
SrlX = DAG.getNode(ISD::SRL, DL, And1->getValueType(0), And1, Const1);
}
return DAG.getNode(
MipsISD::Ins, DL, N->getValueType(0),
isConstCase
? DAG.getConstant(CN1->getSExtValue() >> SMPos0, DL, ValTy)
: SrlX,
DAG.getConstant(SMPos0, DL, MVT::i32),
DAG.getConstant(ValTy.getSizeInBits() / 8 < 8 ? SMSize0 & 31
: SMSize0,
DL, MVT::i32),
And0->getOperand(0));
}
return SDValue();
}
}
static SDValue performMADD_MSUBCombine(SDNode *ROOTNode, SelectionDAG &CurDAG,
const MipsSubtarget &Subtarget) {
// ROOTNode must have a multiplication as an operand for the match to be
// successful.
if (ROOTNode->getOperand(0).getOpcode() != ISD::MUL &&
ROOTNode->getOperand(1).getOpcode() != ISD::MUL)
return SDValue();
// We don't handle vector types here.
if (ROOTNode->getValueType(0).isVector())
return SDValue();
// For MIPS64, madd / msub instructions are inefficent to use with 64 bit
// arithmetic. E.g.
// (add (mul a b) c) =>
// let res = (madd (mthi (drotr c 32))x(mtlo c) a b) in
// MIPS64: (or (dsll (mfhi res) 32) (dsrl (dsll (mflo res) 32) 32)
// or
// MIPS64R2: (dins (mflo res) (mfhi res) 32 32)
//
// The overhead of setting up the Hi/Lo registers and reassembling the
// result makes this a dubious optimzation for MIPS64. The core of the
// problem is that Hi/Lo contain the upper and lower 32 bits of the
// operand and result.
//
// It requires a chain of 4 add/mul for MIPS64R2 to get better code
// density than doing it naively, 5 for MIPS64. Additionally, using
// madd/msub on MIPS64 requires the operands actually be 32 bit sign
// extended operands, not true 64 bit values.
//
// FIXME: For the moment, disable this completely for MIPS64.
if (Subtarget.hasMips64())
return SDValue();
SDValue Mult = ROOTNode->getOperand(0).getOpcode() == ISD::MUL
? ROOTNode->getOperand(0)
: ROOTNode->getOperand(1);
SDValue AddOperand = ROOTNode->getOperand(0).getOpcode() == ISD::MUL
? ROOTNode->getOperand(1)
: ROOTNode->getOperand(0);
// Transform this to a MADD only if the user of this node is the add.
// If there are other users of the mul, this function returns here.
if (!Mult.hasOneUse())
return SDValue();
// maddu and madd are unusual instructions in that on MIPS64 bits 63..31
// must be in canonical form, i.e. sign extended. For MIPS32, the operands
// of the multiply must have 32 or more sign bits, otherwise we cannot
// perform this optimization. We have to check this here as we're performing
// this optimization pre-legalization.
SDValue MultLHS = Mult->getOperand(0);
SDValue MultRHS = Mult->getOperand(1);
bool IsSigned = MultLHS->getOpcode() == ISD::SIGN_EXTEND &&
MultRHS->getOpcode() == ISD::SIGN_EXTEND;
bool IsUnsigned = MultLHS->getOpcode() == ISD::ZERO_EXTEND &&
MultRHS->getOpcode() == ISD::ZERO_EXTEND;
if (!IsSigned && !IsUnsigned)
return SDValue();
// Initialize accumulator.
SDLoc DL(ROOTNode);
SDValue TopHalf;
SDValue BottomHalf;
BottomHalf = CurDAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, AddOperand,
CurDAG.getIntPtrConstant(0, DL));
TopHalf = CurDAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, AddOperand,
CurDAG.getIntPtrConstant(1, DL));
SDValue ACCIn = CurDAG.getNode(MipsISD::MTLOHI, DL, MVT::Untyped,
BottomHalf,
TopHalf);
// Create MipsMAdd(u) / MipsMSub(u) node.
bool IsAdd = ROOTNode->getOpcode() == ISD::ADD;
unsigned Opcode = IsAdd ? (IsUnsigned ? MipsISD::MAddu : MipsISD::MAdd)
: (IsUnsigned ? MipsISD::MSubu : MipsISD::MSub);
SDValue MAddOps[3] = {
CurDAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Mult->getOperand(0)),
CurDAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Mult->getOperand(1)), ACCIn};
EVT VTs[2] = {MVT::i32, MVT::i32};
SDValue MAdd = CurDAG.getNode(Opcode, DL, VTs, MAddOps);
SDValue ResLo = CurDAG.getNode(MipsISD::MFLO, DL, MVT::i32, MAdd);
SDValue ResHi = CurDAG.getNode(MipsISD::MFHI, DL, MVT::i32, MAdd);
SDValue Combined =
CurDAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, ResLo, ResHi);
return Combined;
}
static SDValue performSUBCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const MipsSubtarget &Subtarget) {
// (sub v0 (mul v1, v2)) => (msub v1, v2, v0)
if (DCI.isBeforeLegalizeOps()) {
if (Subtarget.hasMips32() && !Subtarget.hasMips32r6() &&
!Subtarget.inMips16Mode() && N->getValueType(0) == MVT::i64)
return performMADD_MSUBCombine(N, DAG, Subtarget);
return SDValue();
}
return SDValue();
}
static SDValue performADDCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const MipsSubtarget &Subtarget) {
// (add v0 (mul v1, v2)) => (madd v1, v2, v0)
if (DCI.isBeforeLegalizeOps()) {
if (Subtarget.hasMips32() && !Subtarget.hasMips32r6() &&
!Subtarget.inMips16Mode() && N->getValueType(0) == MVT::i64)
return performMADD_MSUBCombine(N, DAG, Subtarget);
return SDValue();
}
// (add v0, (add v1, abs_lo(tjt))) => (add (add v0, v1), abs_lo(tjt))
SDValue Add = N->getOperand(1);
if (Add.getOpcode() != ISD::ADD)
return SDValue();
SDValue Lo = Add.getOperand(1);
if ((Lo.getOpcode() != MipsISD::Lo) ||
(Lo.getOperand(0).getOpcode() != ISD::TargetJumpTable))
return SDValue();
EVT ValTy = N->getValueType(0);
SDLoc DL(N);
SDValue Add1 = DAG.getNode(ISD::ADD, DL, ValTy, N->getOperand(0),
Add.getOperand(0));
return DAG.getNode(ISD::ADD, DL, ValTy, Add1, Lo);
}
static SDValue performSHLCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const MipsSubtarget &Subtarget) {
// Pattern match CINS.
// $dst = shl (and $src , imm), pos
// => cins $dst, $src, pos, size
if (DCI.isBeforeLegalizeOps() || !Subtarget.hasCnMips())
return SDValue();
SDValue FirstOperand = N->getOperand(0);
unsigned FirstOperandOpc = FirstOperand.getOpcode();
SDValue SecondOperand = N->getOperand(1);
EVT ValTy = N->getValueType(0);
SDLoc DL(N);
uint64_t Pos = 0, SMPos, SMSize;
ConstantSDNode *CN;
SDValue NewOperand;
// The second operand of the shift must be an immediate.
if (!(CN = dyn_cast<ConstantSDNode>(SecondOperand)))
return SDValue();
Pos = CN->getZExtValue();
if (Pos >= ValTy.getSizeInBits())
return SDValue();
if (FirstOperandOpc != ISD::AND)
return SDValue();
// AND's second operand must be a shifted mask.
if (!(CN = dyn_cast<ConstantSDNode>(FirstOperand.getOperand(1))) ||
!isShiftedMask(CN->getZExtValue(), SMPos, SMSize))
return SDValue();
// Return if the shifted mask does not start at bit 0 or the sum of its size
// and Pos exceeds the word's size.
if (SMPos != 0 || SMSize > 32 || Pos + SMSize > ValTy.getSizeInBits())
return SDValue();
NewOperand = FirstOperand.getOperand(0);
// SMSize is 'location' (position) in this case, not size.
SMSize--;
return DAG.getNode(MipsISD::CIns, DL, ValTy, NewOperand,
DAG.getConstant(Pos, DL, MVT::i32),
DAG.getConstant(SMSize, DL, MVT::i32));
}
SDValue MipsTargetLowering::PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI)
const {
SelectionDAG &DAG = DCI.DAG;
unsigned Opc = N->getOpcode();
switch (Opc) {
default: break;
case ISD::SDIVREM:
case ISD::UDIVREM:
return performDivRemCombine(N, DAG, DCI, Subtarget);
case ISD::SELECT:
return performSELECTCombine(N, DAG, DCI, Subtarget);
case MipsISD::CMovFP_F:
case MipsISD::CMovFP_T:
return performCMovFPCombine(N, DAG, DCI, Subtarget);
case ISD::AND:
return performANDCombine(N, DAG, DCI, Subtarget);
case ISD::OR:
return performORCombine(N, DAG, DCI, Subtarget);
case ISD::ADD:
return performADDCombine(N, DAG, DCI, Subtarget);
case ISD::SHL:
return performSHLCombine(N, DAG, DCI, Subtarget);
case ISD::SUB:
return performSUBCombine(N, DAG, DCI, Subtarget);
}
return SDValue();
}
bool MipsTargetLowering::isCheapToSpeculateCttz() const {
return Subtarget.hasMips32();
}
bool MipsTargetLowering::isCheapToSpeculateCtlz() const {
return Subtarget.hasMips32();
}
bool MipsTargetLowering::shouldFoldConstantShiftPairToMask(
const SDNode *N, CombineLevel Level) const {
if (N->getOperand(0).getValueType().isVector())
return false;
return true;
}
void
MipsTargetLowering::LowerOperationWrapper(SDNode *N,
SmallVectorImpl<SDValue> &Results,
SelectionDAG &DAG) const {
SDValue Res = LowerOperation(SDValue(N, 0), DAG);
if (Res)
for (unsigned I = 0, E = Res->getNumValues(); I != E; ++I)
Results.push_back(Res.getValue(I));
}
void
MipsTargetLowering::ReplaceNodeResults(SDNode *N,
SmallVectorImpl<SDValue> &Results,
SelectionDAG &DAG) const {
return LowerOperationWrapper(N, Results, DAG);
}
SDValue MipsTargetLowering::
LowerOperation(SDValue Op, SelectionDAG &DAG) const
{
switch (Op.getOpcode())
{
case ISD::BRCOND: return lowerBRCOND(Op, DAG);
case ISD::ConstantPool: return lowerConstantPool(Op, DAG);
case ISD::GlobalAddress: return lowerGlobalAddress(Op, DAG);
case ISD::BlockAddress: return lowerBlockAddress(Op, DAG);
case ISD::GlobalTLSAddress: return lowerGlobalTLSAddress(Op, DAG);
case ISD::JumpTable: return lowerJumpTable(Op, DAG);
case ISD::SELECT: return lowerSELECT(Op, DAG);
case ISD::SETCC: return lowerSETCC(Op, DAG);
case ISD::VASTART: return lowerVASTART(Op, DAG);
case ISD::VAARG: return lowerVAARG(Op, DAG);
case ISD::FCOPYSIGN: return lowerFCOPYSIGN(Op, DAG);
case ISD::FABS: return lowerFABS(Op, DAG);
case ISD::FRAMEADDR: return lowerFRAMEADDR(Op, DAG);
case ISD::RETURNADDR: return lowerRETURNADDR(Op, DAG);
case ISD::EH_RETURN: return lowerEH_RETURN(Op, DAG);
case ISD::ATOMIC_FENCE: return lowerATOMIC_FENCE(Op, DAG);
case ISD::SHL_PARTS: return lowerShiftLeftParts(Op, DAG);
case ISD::SRA_PARTS: return lowerShiftRightParts(Op, DAG, true);
case ISD::SRL_PARTS: return lowerShiftRightParts(Op, DAG, false);
case ISD::LOAD: return lowerLOAD(Op, DAG);
case ISD::STORE: return lowerSTORE(Op, DAG);
case ISD::EH_DWARF_CFA: return lowerEH_DWARF_CFA(Op, DAG);
case ISD::FP_TO_SINT: return lowerFP_TO_SINT(Op, DAG);
}
return SDValue();
}
//===----------------------------------------------------------------------===//
// Lower helper functions
//===----------------------------------------------------------------------===//
// addLiveIn - This helper function adds the specified physical register to the
// MachineFunction as a live in value. It also creates a corresponding
// virtual register for it.
static unsigned
addLiveIn(MachineFunction &MF, unsigned PReg, const TargetRegisterClass *RC)
{
Register VReg = MF.getRegInfo().createVirtualRegister(RC);
MF.getRegInfo().addLiveIn(PReg, VReg);
return VReg;
}
static MachineBasicBlock *insertDivByZeroTrap(MachineInstr &MI,
MachineBasicBlock &MBB,
const TargetInstrInfo &TII,
bool Is64Bit, bool IsMicroMips) {
if (NoZeroDivCheck)
return &MBB;
// Insert instruction "teq $divisor_reg, $zero, 7".
MachineBasicBlock::iterator I(MI);
MachineInstrBuilder MIB;
MachineOperand &Divisor = MI.getOperand(2);
MIB = BuildMI(MBB, std::next(I), MI.getDebugLoc(),
TII.get(IsMicroMips ? Mips::TEQ_MM : Mips::TEQ))
.addReg(Divisor.getReg(), getKillRegState(Divisor.isKill()))
.addReg(Mips::ZERO)
.addImm(7);
// Use the 32-bit sub-register if this is a 64-bit division.
if (Is64Bit)
MIB->getOperand(0).setSubReg(Mips::sub_32);
// Clear Divisor's kill flag.
Divisor.setIsKill(false);
// We would normally delete the original instruction here but in this case
// we only needed to inject an additional instruction rather than replace it.
return &MBB;
}
MachineBasicBlock *
MipsTargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
MachineBasicBlock *BB) const {
switch (MI.getOpcode()) {
default:
llvm_unreachable("Unexpected instr type to insert");
case Mips::ATOMIC_LOAD_ADD_I8:
return emitAtomicBinaryPartword(MI, BB, 1);
case Mips::ATOMIC_LOAD_ADD_I16:
return emitAtomicBinaryPartword(MI, BB, 2);
case Mips::ATOMIC_LOAD_ADD_I32:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_ADD_I64:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_AND_I8:
return emitAtomicBinaryPartword(MI, BB, 1);
case Mips::ATOMIC_LOAD_AND_I16:
return emitAtomicBinaryPartword(MI, BB, 2);
case Mips::ATOMIC_LOAD_AND_I32:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_AND_I64:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_OR_I8:
return emitAtomicBinaryPartword(MI, BB, 1);
case Mips::ATOMIC_LOAD_OR_I16:
return emitAtomicBinaryPartword(MI, BB, 2);
case Mips::ATOMIC_LOAD_OR_I32:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_OR_I64:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_XOR_I8:
return emitAtomicBinaryPartword(MI, BB, 1);
case Mips::ATOMIC_LOAD_XOR_I16:
return emitAtomicBinaryPartword(MI, BB, 2);
case Mips::ATOMIC_LOAD_XOR_I32:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_XOR_I64:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_NAND_I8:
return emitAtomicBinaryPartword(MI, BB, 1);
case Mips::ATOMIC_LOAD_NAND_I16:
return emitAtomicBinaryPartword(MI, BB, 2);
case Mips::ATOMIC_LOAD_NAND_I32:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_NAND_I64:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_SUB_I8:
return emitAtomicBinaryPartword(MI, BB, 1);
case Mips::ATOMIC_LOAD_SUB_I16:
return emitAtomicBinaryPartword(MI, BB, 2);
case Mips::ATOMIC_LOAD_SUB_I32:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_SUB_I64:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_SWAP_I8:
return emitAtomicBinaryPartword(MI, BB, 1);
case Mips::ATOMIC_SWAP_I16:
return emitAtomicBinaryPartword(MI, BB, 2);
case Mips::ATOMIC_SWAP_I32:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_SWAP_I64:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_CMP_SWAP_I8:
return emitAtomicCmpSwapPartword(MI, BB, 1);
case Mips::ATOMIC_CMP_SWAP_I16:
return emitAtomicCmpSwapPartword(MI, BB, 2);
case Mips::ATOMIC_CMP_SWAP_I32:
return emitAtomicCmpSwap(MI, BB);
case Mips::ATOMIC_CMP_SWAP_I64:
return emitAtomicCmpSwap(MI, BB);
case Mips::ATOMIC_LOAD_MIN_I8:
return emitAtomicBinaryPartword(MI, BB, 1);
case Mips::ATOMIC_LOAD_MIN_I16:
return emitAtomicBinaryPartword(MI, BB, 2);
case Mips::ATOMIC_LOAD_MIN_I32:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_MIN_I64:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_MAX_I8:
return emitAtomicBinaryPartword(MI, BB, 1);
case Mips::ATOMIC_LOAD_MAX_I16:
return emitAtomicBinaryPartword(MI, BB, 2);
case Mips::ATOMIC_LOAD_MAX_I32:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_MAX_I64:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_UMIN_I8:
return emitAtomicBinaryPartword(MI, BB, 1);
case Mips::ATOMIC_LOAD_UMIN_I16:
return emitAtomicBinaryPartword(MI, BB, 2);
case Mips::ATOMIC_LOAD_UMIN_I32:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_UMIN_I64:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_UMAX_I8:
return emitAtomicBinaryPartword(MI, BB, 1);
case Mips::ATOMIC_LOAD_UMAX_I16:
return emitAtomicBinaryPartword(MI, BB, 2);
case Mips::ATOMIC_LOAD_UMAX_I32:
return emitAtomicBinary(MI, BB);
case Mips::ATOMIC_LOAD_UMAX_I64:
return emitAtomicBinary(MI, BB);
case Mips::PseudoSDIV:
case Mips::PseudoUDIV:
case Mips::DIV:
case Mips::DIVU:
case Mips::MOD:
case Mips::MODU:
return insertDivByZeroTrap(MI, *BB, *Subtarget.getInstrInfo(), false,
false);
case Mips::SDIV_MM_Pseudo:
case Mips::UDIV_MM_Pseudo:
case Mips::SDIV_MM:
case Mips::UDIV_MM:
case Mips::DIV_MMR6:
case Mips::DIVU_MMR6:
case Mips::MOD_MMR6:
case Mips::MODU_MMR6:
return insertDivByZeroTrap(MI, *BB, *Subtarget.getInstrInfo(), false, true);
case Mips::PseudoDSDIV:
case Mips::PseudoDUDIV:
case Mips::DDIV:
case Mips::DDIVU:
case Mips::DMOD:
case Mips::DMODU:
return insertDivByZeroTrap(MI, *BB, *Subtarget.getInstrInfo(), true, false);
case Mips::PseudoSELECT_I:
case Mips::PseudoSELECT_I64:
case Mips::PseudoSELECT_S:
case Mips::PseudoSELECT_D32:
case Mips::PseudoSELECT_D64:
return emitPseudoSELECT(MI, BB, false, Mips::BNE);
case Mips::PseudoSELECTFP_F_I:
case Mips::PseudoSELECTFP_F_I64:
case Mips::PseudoSELECTFP_F_S:
case Mips::PseudoSELECTFP_F_D32:
case Mips::PseudoSELECTFP_F_D64:
return emitPseudoSELECT(MI, BB, true, Mips::BC1F);
case Mips::PseudoSELECTFP_T_I:
case Mips::PseudoSELECTFP_T_I64:
case Mips::PseudoSELECTFP_T_S:
case Mips::PseudoSELECTFP_T_D32:
case Mips::PseudoSELECTFP_T_D64:
return emitPseudoSELECT(MI, BB, true, Mips::BC1T);
case Mips::PseudoD_SELECT_I:
case Mips::PseudoD_SELECT_I64:
return emitPseudoD_SELECT(MI, BB);
}
}
// This function also handles Mips::ATOMIC_SWAP_I32 (when BinOpcode == 0), and
// Mips::ATOMIC_LOAD_NAND_I32 (when Nand == true)
MachineBasicBlock *
MipsTargetLowering::emitAtomicBinary(MachineInstr &MI,
MachineBasicBlock *BB) const {
MachineFunction *MF = BB->getParent();
MachineRegisterInfo &RegInfo = MF->getRegInfo();
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
DebugLoc DL = MI.getDebugLoc();
unsigned AtomicOp;
bool NeedsAdditionalReg = false;
switch (MI.getOpcode()) {
case Mips::ATOMIC_LOAD_ADD_I32:
AtomicOp = Mips::ATOMIC_LOAD_ADD_I32_POSTRA;
break;
case Mips::ATOMIC_LOAD_SUB_I32:
AtomicOp = Mips::ATOMIC_LOAD_SUB_I32_POSTRA;
break;
case Mips::ATOMIC_LOAD_AND_I32:
AtomicOp = Mips::ATOMIC_LOAD_AND_I32_POSTRA;
break;
case Mips::ATOMIC_LOAD_OR_I32:
AtomicOp = Mips::ATOMIC_LOAD_OR_I32_POSTRA;
break;
case Mips::ATOMIC_LOAD_XOR_I32:
AtomicOp = Mips::ATOMIC_LOAD_XOR_I32_POSTRA;
break;
case Mips::ATOMIC_LOAD_NAND_I32:
AtomicOp = Mips::ATOMIC_LOAD_NAND_I32_POSTRA;
break;
case Mips::ATOMIC_SWAP_I32:
AtomicOp = Mips::ATOMIC_SWAP_I32_POSTRA;
break;
case Mips::ATOMIC_LOAD_ADD_I64:
AtomicOp = Mips::ATOMIC_LOAD_ADD_I64_POSTRA;
break;
case Mips::ATOMIC_LOAD_SUB_I64:
AtomicOp = Mips::ATOMIC_LOAD_SUB_I64_POSTRA;
break;
case Mips::ATOMIC_LOAD_AND_I64:
AtomicOp = Mips::ATOMIC_LOAD_AND_I64_POSTRA;
break;
case Mips::ATOMIC_LOAD_OR_I64:
AtomicOp = Mips::ATOMIC_LOAD_OR_I64_POSTRA;
break;
case Mips::ATOMIC_LOAD_XOR_I64:
AtomicOp = Mips::ATOMIC_LOAD_XOR_I64_POSTRA;
break;
case Mips::ATOMIC_LOAD_NAND_I64:
AtomicOp = Mips::ATOMIC_LOAD_NAND_I64_POSTRA;
break;
case Mips::ATOMIC_SWAP_I64:
AtomicOp = Mips::ATOMIC_SWAP_I64_POSTRA;
break;
case Mips::ATOMIC_LOAD_MIN_I32:
AtomicOp = Mips::ATOMIC_LOAD_MIN_I32_POSTRA;
NeedsAdditionalReg = true;
break;
case Mips::ATOMIC_LOAD_MAX_I32:
AtomicOp = Mips::ATOMIC_LOAD_MAX_I32_POSTRA;
NeedsAdditionalReg = true;
break;
case Mips::ATOMIC_LOAD_UMIN_I32:
AtomicOp = Mips::ATOMIC_LOAD_UMIN_I32_POSTRA;
NeedsAdditionalReg = true;
break;
case Mips::ATOMIC_LOAD_UMAX_I32:
AtomicOp = Mips::ATOMIC_LOAD_UMAX_I32_POSTRA;
NeedsAdditionalReg = true;
break;
case Mips::ATOMIC_LOAD_MIN_I64:
AtomicOp = Mips::ATOMIC_LOAD_MIN_I64_POSTRA;
NeedsAdditionalReg = true;
break;
case Mips::ATOMIC_LOAD_MAX_I64:
AtomicOp = Mips::ATOMIC_LOAD_MAX_I64_POSTRA;
NeedsAdditionalReg = true;
break;
case Mips::ATOMIC_LOAD_UMIN_I64:
AtomicOp = Mips::ATOMIC_LOAD_UMIN_I64_POSTRA;
NeedsAdditionalReg = true;
break;
case Mips::ATOMIC_LOAD_UMAX_I64:
AtomicOp = Mips::ATOMIC_LOAD_UMAX_I64_POSTRA;
NeedsAdditionalReg = true;
break;
default:
llvm_unreachable("Unknown pseudo atomic for replacement!");
}
Register OldVal = MI.getOperand(0).getReg();
Register Ptr = MI.getOperand(1).getReg();
Register Incr = MI.getOperand(2).getReg();
Register Scratch = RegInfo.createVirtualRegister(RegInfo.getRegClass(OldVal));
MachineBasicBlock::iterator II(MI);
// The scratch registers here with the EarlyClobber | Define | Implicit
// flags is used to persuade the register allocator and the machine
// verifier to accept the usage of this register. This has to be a real
// register which has an UNDEF value but is dead after the instruction which
// is unique among the registers chosen for the instruction.
// The EarlyClobber flag has the semantic properties that the operand it is
// attached to is clobbered before the rest of the inputs are read. Hence it
// must be unique among the operands to the instruction.
// The Define flag is needed to coerce the machine verifier that an Undef
// value isn't a problem.
// The Dead flag is needed as the value in scratch isn't used by any other
// instruction. Kill isn't used as Dead is more precise.
// The implicit flag is here due to the interaction between the other flags
// and the machine verifier.
// For correctness purpose, a new pseudo is introduced here. We need this
// new pseudo, so that FastRegisterAllocator does not see an ll/sc sequence
// that is spread over >1 basic blocks. A register allocator which
// introduces (or any codegen infact) a store, can violate the expectations
// of the hardware.
//
// An atomic read-modify-write sequence starts with a linked load
// instruction and ends with a store conditional instruction. The atomic
// read-modify-write sequence fails if any of the following conditions
// occur between the execution of ll and sc:
// * A coherent store is completed by another process or coherent I/O
// module into the block of synchronizable physical memory containing
// the word. The size and alignment of the block is
// implementation-dependent.
// * A coherent store is executed between an LL and SC sequence on the
// same processor to the block of synchornizable physical memory
// containing the word.
//
Register PtrCopy = RegInfo.createVirtualRegister(RegInfo.getRegClass(Ptr));
Register IncrCopy = RegInfo.createVirtualRegister(RegInfo.getRegClass(Incr));
BuildMI(*BB, II, DL, TII->get(Mips::COPY), IncrCopy).addReg(Incr);
BuildMI(*BB, II, DL, TII->get(Mips::COPY), PtrCopy).addReg(Ptr);
MachineInstrBuilder MIB =
BuildMI(*BB, II, DL, TII->get(AtomicOp))
.addReg(OldVal, RegState::Define | RegState::EarlyClobber)
.addReg(PtrCopy)
.addReg(IncrCopy)
.addReg(Scratch, RegState::Define | RegState::EarlyClobber |
RegState::Implicit | RegState::Dead);
if (NeedsAdditionalReg) {
Register Scratch2 =
RegInfo.createVirtualRegister(RegInfo.getRegClass(OldVal));
MIB.addReg(Scratch2, RegState::Define | RegState::EarlyClobber |
RegState::Implicit | RegState::Dead);
}
MI.eraseFromParent();
return BB;
}
MachineBasicBlock *MipsTargetLowering::emitSignExtendToI32InReg(
MachineInstr &MI, MachineBasicBlock *BB, unsigned Size, unsigned DstReg,
unsigned SrcReg) const {
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
const DebugLoc &DL = MI.getDebugLoc();
if (Subtarget.hasMips32r2() && Size == 1) {
BuildMI(BB, DL, TII->get(Mips::SEB), DstReg).addReg(SrcReg);
return BB;
}
if (Subtarget.hasMips32r2() && Size == 2) {
BuildMI(BB, DL, TII->get(Mips::SEH), DstReg).addReg(SrcReg);
return BB;
}
MachineFunction *MF = BB->getParent();
MachineRegisterInfo &RegInfo = MF->getRegInfo();
const TargetRegisterClass *RC = getRegClassFor(MVT::i32);
Register ScrReg = RegInfo.createVirtualRegister(RC);
assert(Size < 32);
int64_t ShiftImm = 32 - (Size * 8);
BuildMI(BB, DL, TII->get(Mips::SLL), ScrReg).addReg(SrcReg).addImm(ShiftImm);
BuildMI(BB, DL, TII->get(Mips::SRA), DstReg).addReg(ScrReg).addImm(ShiftImm);
return BB;
}
MachineBasicBlock *MipsTargetLowering::emitAtomicBinaryPartword(
MachineInstr &MI, MachineBasicBlock *BB, unsigned Size) const {
assert((Size == 1 || Size == 2) &&
"Unsupported size for EmitAtomicBinaryPartial.");
MachineFunction *MF = BB->getParent();
MachineRegisterInfo &RegInfo = MF->getRegInfo();
const TargetRegisterClass *RC = getRegClassFor(MVT::i32);
const bool ArePtrs64bit = ABI.ArePtrs64bit();
const TargetRegisterClass *RCp =
getRegClassFor(ArePtrs64bit ? MVT::i64 : MVT::i32);
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
DebugLoc DL = MI.getDebugLoc();
Register Dest = MI.getOperand(0).getReg();
Register Ptr = MI.getOperand(1).getReg();
Register Incr = MI.getOperand(2).getReg();
Register AlignedAddr = RegInfo.createVirtualRegister(RCp);
Register ShiftAmt = RegInfo.createVirtualRegister(RC);
Register Mask = RegInfo.createVirtualRegister(RC);
Register Mask2 = RegInfo.createVirtualRegister(RC);
Register Incr2 = RegInfo.createVirtualRegister(RC);
Register MaskLSB2 = RegInfo.createVirtualRegister(RCp);
Register PtrLSB2 = RegInfo.createVirtualRegister(RC);
Register MaskUpper = RegInfo.createVirtualRegister(RC);
Register Scratch = RegInfo.createVirtualRegister(RC);
Register Scratch2 = RegInfo.createVirtualRegister(RC);
Register Scratch3 = RegInfo.createVirtualRegister(RC);
unsigned AtomicOp = 0;
bool NeedsAdditionalReg = false;
switch (MI.getOpcode()) {
case Mips::ATOMIC_LOAD_NAND_I8:
AtomicOp = Mips::ATOMIC_LOAD_NAND_I8_POSTRA;
break;
case Mips::ATOMIC_LOAD_NAND_I16:
AtomicOp = Mips::ATOMIC_LOAD_NAND_I16_POSTRA;
break;
case Mips::ATOMIC_SWAP_I8:
AtomicOp = Mips::ATOMIC_SWAP_I8_POSTRA;
break;
case Mips::ATOMIC_SWAP_I16:
AtomicOp = Mips::ATOMIC_SWAP_I16_POSTRA;
break;
case Mips::ATOMIC_LOAD_ADD_I8:
AtomicOp = Mips::ATOMIC_LOAD_ADD_I8_POSTRA;
break;
case Mips::ATOMIC_LOAD_ADD_I16:
AtomicOp = Mips::ATOMIC_LOAD_ADD_I16_POSTRA;
break;
case Mips::ATOMIC_LOAD_SUB_I8:
AtomicOp = Mips::ATOMIC_LOAD_SUB_I8_POSTRA;
break;
case Mips::ATOMIC_LOAD_SUB_I16:
AtomicOp = Mips::ATOMIC_LOAD_SUB_I16_POSTRA;
break;
case Mips::ATOMIC_LOAD_AND_I8:
AtomicOp = Mips::ATOMIC_LOAD_AND_I8_POSTRA;
break;
case Mips::ATOMIC_LOAD_AND_I16:
AtomicOp = Mips::ATOMIC_LOAD_AND_I16_POSTRA;
break;
case Mips::ATOMIC_LOAD_OR_I8:
AtomicOp = Mips::ATOMIC_LOAD_OR_I8_POSTRA;
break;
case Mips::ATOMIC_LOAD_OR_I16:
AtomicOp = Mips::ATOMIC_LOAD_OR_I16_POSTRA;
break;
case Mips::ATOMIC_LOAD_XOR_I8:
AtomicOp = Mips::ATOMIC_LOAD_XOR_I8_POSTRA;
break;
case Mips::ATOMIC_LOAD_XOR_I16:
AtomicOp = Mips::ATOMIC_LOAD_XOR_I16_POSTRA;
break;
case Mips::ATOMIC_LOAD_MIN_I8:
AtomicOp = Mips::ATOMIC_LOAD_MIN_I8_POSTRA;
NeedsAdditionalReg = true;
break;
case Mips::ATOMIC_LOAD_MIN_I16:
AtomicOp = Mips::ATOMIC_LOAD_MIN_I16_POSTRA;
NeedsAdditionalReg = true;
break;
case Mips::ATOMIC_LOAD_MAX_I8:
AtomicOp = Mips::ATOMIC_LOAD_MAX_I8_POSTRA;
NeedsAdditionalReg = true;
break;
case Mips::ATOMIC_LOAD_MAX_I16:
AtomicOp = Mips::ATOMIC_LOAD_MAX_I16_POSTRA;
NeedsAdditionalReg = true;
break;
case Mips::ATOMIC_LOAD_UMIN_I8:
AtomicOp = Mips::ATOMIC_LOAD_UMIN_I8_POSTRA;
NeedsAdditionalReg = true;
break;
case Mips::ATOMIC_LOAD_UMIN_I16:
AtomicOp = Mips::ATOMIC_LOAD_UMIN_I16_POSTRA;
NeedsAdditionalReg = true;
break;
case Mips::ATOMIC_LOAD_UMAX_I8:
AtomicOp = Mips::ATOMIC_LOAD_UMAX_I8_POSTRA;
NeedsAdditionalReg = true;
break;
case Mips::ATOMIC_LOAD_UMAX_I16:
AtomicOp = Mips::ATOMIC_LOAD_UMAX_I16_POSTRA;
NeedsAdditionalReg = true;
break;
default:
llvm_unreachable("Unknown subword atomic pseudo for expansion!");
}
// insert new blocks after the current block
const BasicBlock *LLVM_BB = BB->getBasicBlock();
MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineFunction::iterator It = ++BB->getIterator();
MF->insert(It, exitMBB);
// Transfer the remainder of BB and its successor edges to exitMBB.
exitMBB->splice(exitMBB->begin(), BB,
std::next(MachineBasicBlock::iterator(MI)), BB->end());
exitMBB->transferSuccessorsAndUpdatePHIs(BB);
BB->addSuccessor(exitMBB, BranchProbability::getOne());
// thisMBB:
// addiu masklsb2,$0,-4 # 0xfffffffc
// and alignedaddr,ptr,masklsb2
// andi ptrlsb2,ptr,3
// sll shiftamt,ptrlsb2,3
// ori maskupper,$0,255 # 0xff
// sll mask,maskupper,shiftamt
// nor mask2,$0,mask
// sll incr2,incr,shiftamt
int64_t MaskImm = (Size == 1) ? 255 : 65535;
BuildMI(BB, DL, TII->get(ABI.GetPtrAddiuOp()), MaskLSB2)
.addReg(ABI.GetNullPtr()).addImm(-4);
BuildMI(BB, DL, TII->get(ABI.GetPtrAndOp()), AlignedAddr)
.addReg(Ptr).addReg(MaskLSB2);
BuildMI(BB, DL, TII->get(Mips::ANDi), PtrLSB2)
.addReg(Ptr, 0, ArePtrs64bit ? Mips::sub_32 : 0).addImm(3);
if (Subtarget.isLittle()) {
BuildMI(BB, DL, TII->get(Mips::SLL), ShiftAmt).addReg(PtrLSB2).addImm(3);
} else {
Register Off = RegInfo.createVirtualRegister(RC);
BuildMI(BB, DL, TII->get(Mips::XORi), Off)
.addReg(PtrLSB2).addImm((Size == 1) ? 3 : 2);
BuildMI(BB, DL, TII->get(Mips::SLL), ShiftAmt).addReg(Off).addImm(3);
}
BuildMI(BB, DL, TII->get(Mips::ORi), MaskUpper)
.addReg(Mips::ZERO).addImm(MaskImm);
BuildMI(BB, DL, TII->get(Mips::SLLV), Mask)
.addReg(MaskUpper).addReg(ShiftAmt);
BuildMI(BB, DL, TII->get(Mips::NOR), Mask2).addReg(Mips::ZERO).addReg(Mask);
BuildMI(BB, DL, TII->get(Mips::SLLV), Incr2).addReg(Incr).addReg(ShiftAmt);
// The purposes of the flags on the scratch registers is explained in
// emitAtomicBinary. In summary, we need a scratch register which is going to
// be undef, that is unique among registers chosen for the instruction.
MachineInstrBuilder MIB =
BuildMI(BB, DL, TII->get(AtomicOp))
.addReg(Dest, RegState::Define | RegState::EarlyClobber)
.addReg(AlignedAddr)
.addReg(Incr2)
.addReg(Mask)
.addReg(Mask2)
.addReg(ShiftAmt)
.addReg(Scratch, RegState::EarlyClobber | RegState::Define |
RegState::Dead | RegState::Implicit)
.addReg(Scratch2, RegState::EarlyClobber | RegState::Define |
RegState::Dead | RegState::Implicit)
.addReg(Scratch3, RegState::EarlyClobber | RegState::Define |
RegState::Dead | RegState::Implicit);
if (NeedsAdditionalReg) {
Register Scratch4 = RegInfo.createVirtualRegister(RC);
MIB.addReg(Scratch4, RegState::EarlyClobber | RegState::Define |
RegState::Dead | RegState::Implicit);
}
MI.eraseFromParent(); // The instruction is gone now.
return exitMBB;
}
// Lower atomic compare and swap to a pseudo instruction, taking care to
// define a scratch register for the pseudo instruction's expansion. The
// instruction is expanded after the register allocator as to prevent
// the insertion of stores between the linked load and the store conditional.
MachineBasicBlock *
MipsTargetLowering::emitAtomicCmpSwap(MachineInstr &MI,
MachineBasicBlock *BB) const {
assert((MI.getOpcode() == Mips::ATOMIC_CMP_SWAP_I32 ||
MI.getOpcode() == Mips::ATOMIC_CMP_SWAP_I64) &&
"Unsupported atomic pseudo for EmitAtomicCmpSwap.");
const unsigned Size = MI.getOpcode() == Mips::ATOMIC_CMP_SWAP_I32 ? 4 : 8;
MachineFunction *MF = BB->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
const TargetRegisterClass *RC = getRegClassFor(MVT::getIntegerVT(Size * 8));
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
DebugLoc DL = MI.getDebugLoc();
unsigned AtomicOp = MI.getOpcode() == Mips::ATOMIC_CMP_SWAP_I32
? Mips::ATOMIC_CMP_SWAP_I32_POSTRA
: Mips::ATOMIC_CMP_SWAP_I64_POSTRA;
Register Dest = MI.getOperand(0).getReg();
Register Ptr = MI.getOperand(1).getReg();
Register OldVal = MI.getOperand(2).getReg();
Register NewVal = MI.getOperand(3).getReg();
Register Scratch = MRI.createVirtualRegister(RC);
MachineBasicBlock::iterator II(MI);
// We need to create copies of the various registers and kill them at the
// atomic pseudo. If the copies are not made, when the atomic is expanded
// after fast register allocation, the spills will end up outside of the
// blocks that their values are defined in, causing livein errors.
Register PtrCopy = MRI.createVirtualRegister(MRI.getRegClass(Ptr));
Register OldValCopy = MRI.createVirtualRegister(MRI.getRegClass(OldVal));
Register NewValCopy = MRI.createVirtualRegister(MRI.getRegClass(NewVal));
BuildMI(*BB, II, DL, TII->get(Mips::COPY), PtrCopy).addReg(Ptr);
BuildMI(*BB, II, DL, TII->get(Mips::COPY), OldValCopy).addReg(OldVal);
BuildMI(*BB, II, DL, TII->get(Mips::COPY), NewValCopy).addReg(NewVal);
// The purposes of the flags on the scratch registers is explained in
// emitAtomicBinary. In summary, we need a scratch register which is going to
// be undef, that is unique among registers chosen for the instruction.
BuildMI(*BB, II, DL, TII->get(AtomicOp))
.addReg(Dest, RegState::Define | RegState::EarlyClobber)
.addReg(PtrCopy, RegState::Kill)
.addReg(OldValCopy, RegState::Kill)
.addReg(NewValCopy, RegState::Kill)
.addReg(Scratch, RegState::EarlyClobber | RegState::Define |
RegState::Dead | RegState::Implicit);
MI.eraseFromParent(); // The instruction is gone now.
return BB;
}
MachineBasicBlock *MipsTargetLowering::emitAtomicCmpSwapPartword(
MachineInstr &MI, MachineBasicBlock *BB, unsigned Size) const {
assert((Size == 1 || Size == 2) &&
"Unsupported size for EmitAtomicCmpSwapPartial.");
MachineFunction *MF = BB->getParent();
MachineRegisterInfo &RegInfo = MF->getRegInfo();
const TargetRegisterClass *RC = getRegClassFor(MVT::i32);
const bool ArePtrs64bit = ABI.ArePtrs64bit();
const TargetRegisterClass *RCp =
getRegClassFor(ArePtrs64bit ? MVT::i64 : MVT::i32);
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
DebugLoc DL = MI.getDebugLoc();
Register Dest = MI.getOperand(0).getReg();
Register Ptr = MI.getOperand(1).getReg();
Register CmpVal = MI.getOperand(2).getReg();
Register NewVal = MI.getOperand(3).getReg();
Register AlignedAddr = RegInfo.createVirtualRegister(RCp);
Register ShiftAmt = RegInfo.createVirtualRegister(RC);
Register Mask = RegInfo.createVirtualRegister(RC);
Register Mask2 = RegInfo.createVirtualRegister(RC);
Register ShiftedCmpVal = RegInfo.createVirtualRegister(RC);
Register ShiftedNewVal = RegInfo.createVirtualRegister(RC);
Register MaskLSB2 = RegInfo.createVirtualRegister(RCp);
Register PtrLSB2 = RegInfo.createVirtualRegister(RC);
Register MaskUpper = RegInfo.createVirtualRegister(RC);
Register MaskedCmpVal = RegInfo.createVirtualRegister(RC);
Register MaskedNewVal = RegInfo.createVirtualRegister(RC);
unsigned AtomicOp = MI.getOpcode() == Mips::ATOMIC_CMP_SWAP_I8
? Mips::ATOMIC_CMP_SWAP_I8_POSTRA
: Mips::ATOMIC_CMP_SWAP_I16_POSTRA;
// The scratch registers here with the EarlyClobber | Define | Dead | Implicit
// flags are used to coerce the register allocator and the machine verifier to
// accept the usage of these registers.
// The EarlyClobber flag has the semantic properties that the operand it is
// attached to is clobbered before the rest of the inputs are read. Hence it
// must be unique among the operands to the instruction.
// The Define flag is needed to coerce the machine verifier that an Undef
// value isn't a problem.
// The Dead flag is needed as the value in scratch isn't used by any other
// instruction. Kill isn't used as Dead is more precise.
Register Scratch = RegInfo.createVirtualRegister(RC);
Register Scratch2 = RegInfo.createVirtualRegister(RC);
// insert new blocks after the current block
const BasicBlock *LLVM_BB = BB->getBasicBlock();
MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineFunction::iterator It = ++BB->getIterator();
MF->insert(It, exitMBB);
// Transfer the remainder of BB and its successor edges to exitMBB.
exitMBB->splice(exitMBB->begin(), BB,
std::next(MachineBasicBlock::iterator(MI)), BB->end());
exitMBB->transferSuccessorsAndUpdatePHIs(BB);
BB->addSuccessor(exitMBB, BranchProbability::getOne());
// thisMBB:
// addiu masklsb2,$0,-4 # 0xfffffffc
// and alignedaddr,ptr,masklsb2
// andi ptrlsb2,ptr,3
// xori ptrlsb2,ptrlsb2,3 # Only for BE
// sll shiftamt,ptrlsb2,3
// ori maskupper,$0,255 # 0xff
// sll mask,maskupper,shiftamt
// nor mask2,$0,mask
// andi maskedcmpval,cmpval,255
// sll shiftedcmpval,maskedcmpval,shiftamt
// andi maskednewval,newval,255
// sll shiftednewval,maskednewval,shiftamt
int64_t MaskImm = (Size == 1) ? 255 : 65535;
BuildMI(BB, DL, TII->get(ArePtrs64bit ? Mips::DADDiu : Mips::ADDiu), MaskLSB2)
.addReg(ABI.GetNullPtr()).addImm(-4);
BuildMI(BB, DL, TII->get(ArePtrs64bit ? Mips::AND64 : Mips::AND), AlignedAddr)
.addReg(Ptr).addReg(MaskLSB2);
BuildMI(BB, DL, TII->get(Mips::ANDi), PtrLSB2)
.addReg(Ptr, 0, ArePtrs64bit ? Mips::sub_32 : 0).addImm(3);
if (Subtarget.isLittle()) {
BuildMI(BB, DL, TII->get(Mips::SLL), ShiftAmt).addReg(PtrLSB2).addImm(3);
} else {
Register Off = RegInfo.createVirtualRegister(RC);
BuildMI(BB, DL, TII->get(Mips::XORi), Off)
.addReg(PtrLSB2).addImm((Size == 1) ? 3 : 2);
BuildMI(BB, DL, TII->get(Mips::SLL), ShiftAmt).addReg(Off).addImm(3);
}
BuildMI(BB, DL, TII->get(Mips::ORi), MaskUpper)
.addReg(Mips::ZERO).addImm(MaskImm);
BuildMI(BB, DL, TII->get(Mips::SLLV), Mask)
.addReg(MaskUpper).addReg(ShiftAmt);
BuildMI(BB, DL, TII->get(Mips::NOR), Mask2).addReg(Mips::ZERO).addReg(Mask);
BuildMI(BB, DL, TII->get(Mips::ANDi), MaskedCmpVal)
.addReg(CmpVal).addImm(MaskImm);
BuildMI(BB, DL, TII->get(Mips::SLLV), ShiftedCmpVal)
.addReg(MaskedCmpVal).addReg(ShiftAmt);
BuildMI(BB, DL, TII->get(Mips::ANDi), MaskedNewVal)
.addReg(NewVal).addImm(MaskImm);
BuildMI(BB, DL, TII->get(Mips::SLLV), ShiftedNewVal)
.addReg(MaskedNewVal).addReg(ShiftAmt);
// The purposes of the flags on the scratch registers are explained in
// emitAtomicBinary. In summary, we need a scratch register which is going to
// be undef, that is unique among the register chosen for the instruction.
BuildMI(BB, DL, TII->get(AtomicOp))
.addReg(Dest, RegState::Define | RegState::EarlyClobber)
.addReg(AlignedAddr)
.addReg(Mask)
.addReg(ShiftedCmpVal)
.addReg(Mask2)
.addReg(ShiftedNewVal)
.addReg(ShiftAmt)
.addReg(Scratch, RegState::EarlyClobber | RegState::Define |
RegState::Dead | RegState::Implicit)
.addReg(Scratch2, RegState::EarlyClobber | RegState::Define |
RegState::Dead | RegState::Implicit);
MI.eraseFromParent(); // The instruction is gone now.
return exitMBB;
}
SDValue MipsTargetLowering::lowerBRCOND(SDValue Op, SelectionDAG &DAG) const {
// The first operand is the chain, the second is the condition, the third is
// the block to branch to if the condition is true.
SDValue Chain = Op.getOperand(0);
SDValue Dest = Op.getOperand(2);
SDLoc DL(Op);
assert(!Subtarget.hasMips32r6() && !Subtarget.hasMips64r6());
SDValue CondRes = createFPCmp(DAG, Op.getOperand(1));
// Return if flag is not set by a floating point comparison.
if (CondRes.getOpcode() != MipsISD::FPCmp)
return Op;
SDValue CCNode = CondRes.getOperand(2);
Mips::CondCode CC =
(Mips::CondCode)cast<ConstantSDNode>(CCNode)->getZExtValue();
unsigned Opc = invertFPCondCodeUser(CC) ? Mips::BRANCH_F : Mips::BRANCH_T;
SDValue BrCode = DAG.getConstant(Opc, DL, MVT::i32);
SDValue FCC0 = DAG.getRegister(Mips::FCC0, MVT::i32);
return DAG.getNode(MipsISD::FPBrcond, DL, Op.getValueType(), Chain, BrCode,
FCC0, Dest, CondRes);
}
SDValue MipsTargetLowering::
lowerSELECT(SDValue Op, SelectionDAG &DAG) const
{
assert(!Subtarget.hasMips32r6() && !Subtarget.hasMips64r6());
SDValue Cond = createFPCmp(DAG, Op.getOperand(0));
// Return if flag is not set by a floating point comparison.
if (Cond.getOpcode() != MipsISD::FPCmp)
return Op;
return createCMovFP(DAG, Cond, Op.getOperand(1), Op.getOperand(2),
SDLoc(Op));
}
SDValue MipsTargetLowering::lowerSETCC(SDValue Op, SelectionDAG &DAG) const {
assert(!Subtarget.hasMips32r6() && !Subtarget.hasMips64r6());
SDValue Cond = createFPCmp(DAG, Op);
assert(Cond.getOpcode() == MipsISD::FPCmp &&
"Floating point operand expected.");
SDLoc DL(Op);
SDValue True = DAG.getConstant(1, DL, MVT::i32);
SDValue False = DAG.getConstant(0, DL, MVT::i32);
return createCMovFP(DAG, Cond, True, False, DL);
}
SDValue MipsTargetLowering::lowerGlobalAddress(SDValue Op,
SelectionDAG &DAG) const {
EVT Ty = Op.getValueType();
GlobalAddressSDNode *N = cast<GlobalAddressSDNode>(Op);
const GlobalValue *GV = N->getGlobal();
if (!isPositionIndependent()) {
const MipsTargetObjectFile *TLOF =
static_cast<const MipsTargetObjectFile *>(
getTargetMachine().getObjFileLowering());
const GlobalObject *GO = GV->getBaseObject();
if (GO && TLOF->IsGlobalInSmallSection(GO, getTargetMachine()))
// %gp_rel relocation
return getAddrGPRel(N, SDLoc(N), Ty, DAG, ABI.IsN64());
// %hi/%lo relocation
return Subtarget.hasSym32() ? getAddrNonPIC(N, SDLoc(N), Ty, DAG)
// %highest/%higher/%hi/%lo relocation
: getAddrNonPICSym64(N, SDLoc(N), Ty, DAG);
}
// Every other architecture would use shouldAssumeDSOLocal in here, but
// mips is special.
// * In PIC code mips requires got loads even for local statics!
// * To save on got entries, for local statics the got entry contains the
// page and an additional add instruction takes care of the low bits.
// * It is legal to access a hidden symbol with a non hidden undefined,
// so one cannot guarantee that all access to a hidden symbol will know
// it is hidden.
// * Mips linkers don't support creating a page and a full got entry for
// the same symbol.
// * Given all that, we have to use a full got entry for hidden symbols :-(
if (GV->hasLocalLinkage())
return getAddrLocal(N, SDLoc(N), Ty, DAG, ABI.IsN32() || ABI.IsN64());
if (Subtarget.useXGOT())
return getAddrGlobalLargeGOT(
N, SDLoc(N), Ty, DAG, MipsII::MO_GOT_HI16, MipsII::MO_GOT_LO16,
DAG.getEntryNode(),
MachinePointerInfo::getGOT(DAG.getMachineFunction()));
return getAddrGlobal(
N, SDLoc(N), Ty, DAG,
(ABI.IsN32() || ABI.IsN64()) ? MipsII::MO_GOT_DISP : MipsII::MO_GOT,
DAG.getEntryNode(), MachinePointerInfo::getGOT(DAG.getMachineFunction()));
}
SDValue MipsTargetLowering::lowerBlockAddress(SDValue Op,
SelectionDAG &DAG) const {
BlockAddressSDNode *N = cast<BlockAddressSDNode>(Op);
EVT Ty = Op.getValueType();
if (!isPositionIndependent())
return Subtarget.hasSym32() ? getAddrNonPIC(N, SDLoc(N), Ty, DAG)
: getAddrNonPICSym64(N, SDLoc(N), Ty, DAG);
return getAddrLocal(N, SDLoc(N), Ty, DAG, ABI.IsN32() || ABI.IsN64());
}
SDValue MipsTargetLowering::
lowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const
{
// If the relocation model is PIC, use the General Dynamic TLS Model or
// Local Dynamic TLS model, otherwise use the Initial Exec or
// Local Exec TLS Model.
GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(Op);
if (DAG.getTarget().useEmulatedTLS())
return LowerToTLSEmulatedModel(GA, DAG);
SDLoc DL(GA);
const GlobalValue *GV = GA->getGlobal();
EVT PtrVT = getPointerTy(DAG.getDataLayout());
TLSModel::Model model = getTargetMachine().getTLSModel(GV);
if (model == TLSModel::GeneralDynamic || model == TLSModel::LocalDynamic) {
// General Dynamic and Local Dynamic TLS Model.
unsigned Flag = (model == TLSModel::LocalDynamic) ? MipsII::MO_TLSLDM
: MipsII::MO_TLSGD;
SDValue TGA = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, Flag);
SDValue Argument = DAG.getNode(MipsISD::Wrapper, DL, PtrVT,
getGlobalReg(DAG, PtrVT), TGA);
unsigned PtrSize = PtrVT.getSizeInBits();
IntegerType *PtrTy = Type::getIntNTy(*DAG.getContext(), PtrSize);
SDValue TlsGetAddr = DAG.getExternalSymbol("__tls_get_addr", PtrVT);
ArgListTy Args;
ArgListEntry Entry;
Entry.Node = Argument;
Entry.Ty = PtrTy;
Args.push_back(Entry);
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(DL)
.setChain(DAG.getEntryNode())
.setLibCallee(CallingConv::C, PtrTy, TlsGetAddr, std::move(Args));
std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);
SDValue Ret = CallResult.first;
if (model != TLSModel::LocalDynamic)
return Ret;
SDValue TGAHi = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
MipsII::MO_DTPREL_HI);
SDValue Hi = DAG.getNode(MipsISD::TlsHi, DL, PtrVT, TGAHi);
SDValue TGALo = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
MipsII::MO_DTPREL_LO);
SDValue Lo = DAG.getNode(MipsISD::Lo, DL, PtrVT, TGALo);
SDValue Add = DAG.getNode(ISD::ADD, DL, PtrVT, Hi, Ret);
return DAG.getNode(ISD::ADD, DL, PtrVT, Add, Lo);
}
SDValue Offset;
if (model == TLSModel::InitialExec) {
// Initial Exec TLS Model
SDValue TGA = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
MipsII::MO_GOTTPREL);
TGA = DAG.getNode(MipsISD::Wrapper, DL, PtrVT, getGlobalReg(DAG, PtrVT),
TGA);
Offset =
DAG.getLoad(PtrVT, DL, DAG.getEntryNode(), TGA, MachinePointerInfo());
} else {
// Local Exec TLS Model
assert(model == TLSModel::LocalExec);
SDValue TGAHi = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
MipsII::MO_TPREL_HI);
SDValue TGALo = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
MipsII::MO_TPREL_LO);
SDValue Hi = DAG.getNode(MipsISD::TlsHi, DL, PtrVT, TGAHi);
SDValue Lo = DAG.getNode(MipsISD::Lo, DL, PtrVT, TGALo);
Offset = DAG.getNode(ISD::ADD, DL, PtrVT, Hi, Lo);
}
SDValue ThreadPointer = DAG.getNode(MipsISD::ThreadPointer, DL, PtrVT);
return DAG.getNode(ISD::ADD, DL, PtrVT, ThreadPointer, Offset);
}
SDValue MipsTargetLowering::
lowerJumpTable(SDValue Op, SelectionDAG &DAG) const
{
JumpTableSDNode *N = cast<JumpTableSDNode>(Op);
EVT Ty = Op.getValueType();
if (!isPositionIndependent())
return Subtarget.hasSym32() ? getAddrNonPIC(N, SDLoc(N), Ty, DAG)
: getAddrNonPICSym64(N, SDLoc(N), Ty, DAG);
return getAddrLocal(N, SDLoc(N), Ty, DAG, ABI.IsN32() || ABI.IsN64());
}
SDValue MipsTargetLowering::
lowerConstantPool(SDValue Op, SelectionDAG &DAG) const
{
ConstantPoolSDNode *N = cast<ConstantPoolSDNode>(Op);
EVT Ty = Op.getValueType();
if (!isPositionIndependent()) {
const MipsTargetObjectFile *TLOF =
static_cast<const MipsTargetObjectFile *>(
getTargetMachine().getObjFileLowering());
if (TLOF->IsConstantInSmallSection(DAG.getDataLayout(), N->getConstVal(),
getTargetMachine()))
// %gp_rel relocation
return getAddrGPRel(N, SDLoc(N), Ty, DAG, ABI.IsN64());
return Subtarget.hasSym32() ? getAddrNonPIC(N, SDLoc(N), Ty, DAG)
: getAddrNonPICSym64(N, SDLoc(N), Ty, DAG);
}
return getAddrLocal(N, SDLoc(N), Ty, DAG, ABI.IsN32() || ABI.IsN64());
}
SDValue MipsTargetLowering::lowerVASTART(SDValue Op, SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
MipsFunctionInfo *FuncInfo = MF.getInfo<MipsFunctionInfo>();
SDLoc DL(Op);
SDValue FI = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(),
getPointerTy(MF.getDataLayout()));
// vastart just stores the address of the VarArgsFrameIndex slot into the
// memory location argument.
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
return DAG.getStore(Op.getOperand(0), DL, FI, Op.getOperand(1),
MachinePointerInfo(SV));
}
SDValue MipsTargetLowering::lowerVAARG(SDValue Op, SelectionDAG &DAG) const {
SDNode *Node = Op.getNode();
EVT VT = Node->getValueType(0);
SDValue Chain = Node->getOperand(0);
SDValue VAListPtr = Node->getOperand(1);
const Align Align =
llvm::MaybeAlign(Node->getConstantOperandVal(3)).valueOrOne();
const Value *SV = cast<SrcValueSDNode>(Node->getOperand(2))->getValue();
SDLoc DL(Node);
unsigned ArgSlotSizeInBytes = (ABI.IsN32() || ABI.IsN64()) ? 8 : 4;
SDValue VAListLoad = DAG.getLoad(getPointerTy(DAG.getDataLayout()), DL, Chain,
VAListPtr, MachinePointerInfo(SV));
SDValue VAList = VAListLoad;
// Re-align the pointer if necessary.
// It should only ever be necessary for 64-bit types on O32 since the minimum
// argument alignment is the same as the maximum type alignment for N32/N64.
//
// FIXME: We currently align too often. The code generator doesn't notice
// when the pointer is still aligned from the last va_arg (or pair of
// va_args for the i64 on O32 case).
if (Align > getMinStackArgumentAlignment()) {
VAList = DAG.getNode(
ISD::ADD, DL, VAList.getValueType(), VAList,
DAG.getConstant(Align.value() - 1, DL, VAList.getValueType()));
VAList = DAG.getNode(
ISD::AND, DL, VAList.getValueType(), VAList,
DAG.getConstant(-(int64_t)Align.value(), DL, VAList.getValueType()));
}
// Increment the pointer, VAList, to the next vaarg.
auto &TD = DAG.getDataLayout();
unsigned ArgSizeInBytes =
TD.getTypeAllocSize(VT.getTypeForEVT(*DAG.getContext()));
SDValue Tmp3 =
DAG.getNode(ISD::ADD, DL, VAList.getValueType(), VAList,
DAG.getConstant(alignTo(ArgSizeInBytes, ArgSlotSizeInBytes),
DL, VAList.getValueType()));
// Store the incremented VAList to the legalized pointer
Chain = DAG.getStore(VAListLoad.getValue(1), DL, Tmp3, VAListPtr,
MachinePointerInfo(SV));
// In big-endian mode we must adjust the pointer when the load size is smaller
// than the argument slot size. We must also reduce the known alignment to
// match. For example in the N64 ABI, we must add 4 bytes to the offset to get
// the correct half of the slot, and reduce the alignment from 8 (slot
// alignment) down to 4 (type alignment).
if (!Subtarget.isLittle() && ArgSizeInBytes < ArgSlotSizeInBytes) {
unsigned Adjustment = ArgSlotSizeInBytes - ArgSizeInBytes;
VAList = DAG.getNode(ISD::ADD, DL, VAListPtr.getValueType(), VAList,
DAG.getIntPtrConstant(Adjustment, DL));
}
// Load the actual argument out of the pointer VAList
return DAG.getLoad(VT, DL, Chain, VAList, MachinePointerInfo());
}
static SDValue lowerFCOPYSIGN32(SDValue Op, SelectionDAG &DAG,
bool HasExtractInsert) {
EVT TyX = Op.getOperand(0).getValueType();
EVT TyY = Op.getOperand(1).getValueType();
SDLoc DL(Op);
SDValue Const1 = DAG.getConstant(1, DL, MVT::i32);
SDValue Const31 = DAG.getConstant(31, DL, MVT::i32);
SDValue Res;
// If operand is of type f64, extract the upper 32-bit. Otherwise, bitcast it
// to i32.
SDValue X = (TyX == MVT::f32) ?
DAG.getNode(ISD::BITCAST, DL, MVT::i32, Op.getOperand(0)) :
DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32, Op.getOperand(0),
Const1);
SDValue Y = (TyY == MVT::f32) ?
DAG.getNode(ISD::BITCAST, DL, MVT::i32, Op.getOperand(1)) :
DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32, Op.getOperand(1),
Const1);
if (HasExtractInsert) {
// ext E, Y, 31, 1 ; extract bit31 of Y
// ins X, E, 31, 1 ; insert extracted bit at bit31 of X
SDValue E = DAG.getNode(MipsISD::Ext, DL, MVT::i32, Y, Const31, Const1);
Res = DAG.getNode(MipsISD::Ins, DL, MVT::i32, E, Const31, Const1, X);
} else {
// sll SllX, X, 1
// srl SrlX, SllX, 1
// srl SrlY, Y, 31
// sll SllY, SrlX, 31
// or Or, SrlX, SllY
SDValue SllX = DAG.getNode(ISD::SHL, DL, MVT::i32, X, Const1);
SDValue SrlX = DAG.getNode(ISD::SRL, DL, MVT::i32, SllX, Const1);
SDValue SrlY = DAG.getNode(ISD::SRL, DL, MVT::i32, Y, Const31);
SDValue SllY = DAG.getNode(ISD::SHL, DL, MVT::i32, SrlY, Const31);
Res = DAG.getNode(ISD::OR, DL, MVT::i32, SrlX, SllY);
}
if (TyX == MVT::f32)
return DAG.getNode(ISD::BITCAST, DL, Op.getOperand(0).getValueType(), Res);
SDValue LowX = DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32,
Op.getOperand(0),
DAG.getConstant(0, DL, MVT::i32));
return DAG.getNode(MipsISD::BuildPairF64, DL, MVT::f64, LowX, Res);
}
static SDValue lowerFCOPYSIGN64(SDValue Op, SelectionDAG &DAG,
bool HasExtractInsert) {
unsigned WidthX = Op.getOperand(0).getValueSizeInBits();
unsigned WidthY = Op.getOperand(1).getValueSizeInBits();
EVT TyX = MVT::getIntegerVT(WidthX), TyY = MVT::getIntegerVT(WidthY);
SDLoc DL(Op);
SDValue Const1 = DAG.getConstant(1, DL, MVT::i32);
// Bitcast to integer nodes.
SDValue X = DAG.getNode(ISD::BITCAST, DL, TyX, Op.getOperand(0));
SDValue Y = DAG.getNode(ISD::BITCAST, DL, TyY, Op.getOperand(1));
if (HasExtractInsert) {
// ext E, Y, width(Y) - 1, 1 ; extract bit width(Y)-1 of Y
// ins X, E, width(X) - 1, 1 ; insert extracted bit at bit width(X)-1 of X
SDValue E = DAG.getNode(MipsISD::Ext, DL, TyY, Y,
DAG.getConstant(WidthY - 1, DL, MVT::i32), Const1);
if (WidthX > WidthY)
E = DAG.getNode(ISD::ZERO_EXTEND, DL, TyX, E);
else if (WidthY > WidthX)
E = DAG.getNode(ISD::TRUNCATE, DL, TyX, E);
SDValue I = DAG.getNode(MipsISD::Ins, DL, TyX, E,
DAG.getConstant(WidthX - 1, DL, MVT::i32), Const1,
X);
return DAG.getNode(ISD::BITCAST, DL, Op.getOperand(0).getValueType(), I);
}
// (d)sll SllX, X, 1
// (d)srl SrlX, SllX, 1
// (d)srl SrlY, Y, width(Y)-1
// (d)sll SllY, SrlX, width(Y)-1
// or Or, SrlX, SllY
SDValue SllX = DAG.getNode(ISD::SHL, DL, TyX, X, Const1);
SDValue SrlX = DAG.getNode(ISD::SRL, DL, TyX, SllX, Const1);
SDValue SrlY = DAG.getNode(ISD::SRL, DL, TyY, Y,
DAG.getConstant(WidthY - 1, DL, MVT::i32));
if (WidthX > WidthY)
SrlY = DAG.getNode(ISD::ZERO_EXTEND, DL, TyX, SrlY);
else if (WidthY > WidthX)
SrlY = DAG.getNode(ISD::TRUNCATE, DL, TyX, SrlY);
SDValue SllY = DAG.getNode(ISD::SHL, DL, TyX, SrlY,
DAG.getConstant(WidthX - 1, DL, MVT::i32));
SDValue Or = DAG.getNode(ISD::OR, DL, TyX, SrlX, SllY);
return DAG.getNode(ISD::BITCAST, DL, Op.getOperand(0).getValueType(), Or);
}
SDValue
MipsTargetLowering::lowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const {
if (Subtarget.isGP64bit())
return lowerFCOPYSIGN64(Op, DAG, Subtarget.hasExtractInsert());
return lowerFCOPYSIGN32(Op, DAG, Subtarget.hasExtractInsert());
}
static SDValue lowerFABS32(SDValue Op, SelectionDAG &DAG,
bool HasExtractInsert) {
SDLoc DL(Op);
SDValue Res, Const1 = DAG.getConstant(1, DL, MVT::i32);
// If operand is of type f64, extract the upper 32-bit. Otherwise, bitcast it
// to i32.
SDValue X = (Op.getValueType() == MVT::f32)
? DAG.getNode(ISD::BITCAST, DL, MVT::i32, Op.getOperand(0))
: DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32,
Op.getOperand(0), Const1);
// Clear MSB.
if (HasExtractInsert)
Res = DAG.getNode(MipsISD::Ins, DL, MVT::i32,
DAG.getRegister(Mips::ZERO, MVT::i32),
DAG.getConstant(31, DL, MVT::i32), Const1, X);
else {
// TODO: Provide DAG patterns which transform (and x, cst)
// back to a (shl (srl x (clz cst)) (clz cst)) sequence.
SDValue SllX = DAG.getNode(ISD::SHL, DL, MVT::i32, X, Const1);
Res = DAG.getNode(ISD::SRL, DL, MVT::i32, SllX, Const1);
}