#include "llvm/CodeGen/MachineDominators.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/StackMaps.h"
#include "llvm/IR/DerivedTypes.h"
{ X86::MOVSX64rr8, X86::MOVSX64rm8, 0 },
{ X86::MOVUPDrr, X86::MOVUPDrm, TB_ALIGN_16 },
{ X86::MOVUPSrr, X86::MOVUPSrm, 0 },
- { X86::MOVZQI2PQIrr, X86::MOVZQI2PQIrm, 0 },
{ X86::MOVZPQILo2PQIrr, X86::MOVZPQILo2PQIrm, TB_ALIGN_16 },
{ X86::MOVZX16rr8, X86::MOVZX16rm8, 0 },
{ X86::MOVZX32rr16, X86::MOVZX32rm16, 0 },
{ X86::VMOVSHDUPrr, X86::VMOVSHDUPrm, 0 },
{ X86::VMOVUPDrr, X86::VMOVUPDrm, 0 },
{ X86::VMOVUPSrr, X86::VMOVUPSrm, 0 },
- { X86::VMOVZQI2PQIrr, X86::VMOVZQI2PQIrm, 0 },
{ X86::VMOVZPQILo2PQIrr,X86::VMOVZPQILo2PQIrm, TB_ALIGN_16 },
{ X86::VPABSBrr128, X86::VPABSBrm128, 0 },
{ X86::VPABSDrr128, X86::VPABSDrm128, 0 },
{ X86::PEXT32rr, X86::PEXT32rm, 0 },
{ X86::PEXT64rr, X86::PEXT64rm, 0 },
+ // ADX foldable instructions
+ { X86::ADCX32rr, X86::ADCX32rm, 0 },
+ { X86::ADCX64rr, X86::ADCX64rm, 0 },
+ { X86::ADOX32rr, X86::ADOX32rm, 0 },
+ { X86::ADOX64rr, X86::ADOX64rm, 0 },
+
// AVX-512 foldable instructions
{ X86::VADDPSZrr, X86::VADDPSZrm, 0 },
{ X86::VADDPDZrr, X86::VADDPDZrm, 0 },
// It is safe to clobber EFLAGS at the end of a block of no successor has it
// live in.
if (Iter == E) {
- for (MachineBasicBlock::succ_iterator SI = MBB.succ_begin(),
- SE = MBB.succ_end(); SI != SE; ++SI)
- if ((*SI)->isLiveIn(X86::EFLAGS))
+ for (MachineBasicBlock *S : MBB.successors())
+ if (S->isLiveIn(X86::EFLAGS))
return false;
return true;
}
unsigned DestReg, unsigned SubIdx,
const MachineInstr *Orig,
const TargetRegisterInfo &TRI) const {
- // MOV32r0 is implemented with a xor which clobbers condition code.
- // Re-materialize it as movri instructions to avoid side effects.
- unsigned Opc = Orig->getOpcode();
- if (Opc == X86::MOV32r0 && !isSafeToClobberEFLAGS(MBB, I)) {
+ bool ClobbersEFLAGS = false;
+ for (const MachineOperand &MO : Orig->operands()) {
+ if (MO.isReg() && MO.isDef() && MO.getReg() == X86::EFLAGS) {
+ ClobbersEFLAGS = true;
+ break;
+ }
+ }
+
+ if (ClobbersEFLAGS && !isSafeToClobberEFLAGS(MBB, I)) {
+ // The instruction clobbers EFLAGS. Re-materialize as MOV32ri to avoid side
+ // effects.
+ int Value;
+ switch (Orig->getOpcode()) {
+ case X86::MOV32r0: Value = 0; break;
+ case X86::MOV32r1: Value = 1; break;
+ case X86::MOV32r_1: Value = -1; break;
+ default:
+ llvm_unreachable("Unexpected instruction!");
+ }
+
DebugLoc DL = Orig->getDebugLoc();
BuildMI(MBB, I, DL, get(X86::MOV32ri)).addOperand(Orig->getOperand(0))
- .addImm(0);
+ .addImm(Value);
} else {
MachineInstr *MI = MBB.getParent()->CloneMachineInstr(Orig);
MBB.insert(I, MI);
bool IsIntrinOpcode;
isFMA3(Opc, &IsIntrinOpcode);
- unsigned GroupsNum;
+ size_t GroupsNum;
const unsigned (*OpcodeGroups)[3];
if (IsIntrinOpcode) {
- GroupsNum = sizeof(IntrinOpcodeGroups) / sizeof(IntrinOpcodeGroups[0]);
+ GroupsNum = array_lengthof(IntrinOpcodeGroups);
OpcodeGroups = IntrinOpcodeGroups;
} else {
- GroupsNum = sizeof(RegularOpcodeGroups) / sizeof(RegularOpcodeGroups[0]);
+ GroupsNum = array_lengthof(RegularOpcodeGroups);
OpcodeGroups = RegularOpcodeGroups;
}
const unsigned *FoundOpcodesGroup = nullptr;
- unsigned FormIndex;
+ size_t FormIndex;
// Look for the input opcode in the corresponding opcodes table.
- unsigned GroupIndex = 0;
- for (; GroupIndex < GroupsNum && !FoundOpcodesGroup; GroupIndex++) {
- for (FormIndex = 0; FormIndex < FormsNum; FormIndex++) {
+ for (size_t GroupIndex = 0; GroupIndex < GroupsNum && !FoundOpcodesGroup;
+ ++GroupIndex) {
+ for (FormIndex = 0; FormIndex < FormsNum; ++FormIndex) {
if (OpcodeGroups[GroupIndex][FormIndex] == Opc) {
FoundOpcodesGroup = OpcodeGroups[GroupIndex];
break;
return 0;
}
-inline static bool MaskRegClassContains(unsigned Reg) {
+static bool MaskRegClassContains(unsigned Reg) {
return X86::VK8RegClass.contains(Reg) ||
X86::VK16RegClass.contains(Reg) ||
X86::VK32RegClass.contains(Reg) ||
X86::VK64RegClass.contains(Reg) ||
X86::VK1RegClass.contains(Reg);
}
+
+static bool GRRegClassContains(unsigned Reg) {
+ return X86::GR64RegClass.contains(Reg) ||
+ X86::GR32RegClass.contains(Reg) ||
+ X86::GR16RegClass.contains(Reg) ||
+ X86::GR8RegClass.contains(Reg);
+}
+static
+unsigned copyPhysRegOpcode_AVX512_DQ(unsigned& DestReg, unsigned& SrcReg) {
+ if (MaskRegClassContains(SrcReg) && X86::GR8RegClass.contains(DestReg)) {
+ DestReg = getX86SubSuperRegister(DestReg, MVT::i32);
+ return X86::KMOVBrk;
+ }
+ if (MaskRegClassContains(DestReg) && X86::GR8RegClass.contains(SrcReg)) {
+ SrcReg = getX86SubSuperRegister(SrcReg, MVT::i32);
+ return X86::KMOVBkr;
+ }
+ return 0;
+}
+
+static
+unsigned copyPhysRegOpcode_AVX512_BW(unsigned& DestReg, unsigned& SrcReg) {
+ if (MaskRegClassContains(SrcReg) && MaskRegClassContains(DestReg))
+ return X86::KMOVQkk;
+ if (MaskRegClassContains(SrcReg) && X86::GR32RegClass.contains(DestReg))
+ return X86::KMOVDrk;
+ if (MaskRegClassContains(SrcReg) && X86::GR64RegClass.contains(DestReg))
+ return X86::KMOVQrk;
+ if (MaskRegClassContains(DestReg) && X86::GR32RegClass.contains(SrcReg))
+ return X86::KMOVDkr;
+ if (MaskRegClassContains(DestReg) && X86::GR64RegClass.contains(SrcReg))
+ return X86::KMOVQkr;
+ return 0;
+}
+
static
-unsigned copyPhysRegOpcode_AVX512(unsigned& DestReg, unsigned& SrcReg) {
+unsigned copyPhysRegOpcode_AVX512(unsigned& DestReg, unsigned& SrcReg,
+ const X86Subtarget &Subtarget)
+{
+ if (Subtarget.hasDQI())
+ if (auto Opc = copyPhysRegOpcode_AVX512_DQ(DestReg, SrcReg))
+ return Opc;
+ if (Subtarget.hasBWI())
+ if (auto Opc = copyPhysRegOpcode_AVX512_BW(DestReg, SrcReg))
+ return Opc;
if (X86::VR128XRegClass.contains(DestReg, SrcReg) ||
X86::VR256XRegClass.contains(DestReg, SrcReg) ||
X86::VR512RegClass.contains(DestReg, SrcReg)) {
SrcReg = get512BitSuperRegister(SrcReg);
return X86::VMOVAPSZrr;
}
- if (MaskRegClassContains(DestReg) &&
- MaskRegClassContains(SrcReg))
+ if (MaskRegClassContains(DestReg) && MaskRegClassContains(SrcReg))
return X86::KMOVWkk;
- if (MaskRegClassContains(DestReg) &&
- (X86::GR32RegClass.contains(SrcReg) ||
- X86::GR16RegClass.contains(SrcReg) ||
- X86::GR8RegClass.contains(SrcReg))) {
+ if (MaskRegClassContains(DestReg) && GRRegClassContains(SrcReg)) {
SrcReg = getX86SubSuperRegister(SrcReg, MVT::i32);
return X86::KMOVWkr;
}
- if ((X86::GR32RegClass.contains(DestReg) ||
- X86::GR16RegClass.contains(DestReg) ||
- X86::GR8RegClass.contains(DestReg)) &&
- MaskRegClassContains(SrcReg)) {
+ if (GRRegClassContains(DestReg) && MaskRegClassContains(SrcReg)) {
DestReg = getX86SubSuperRegister(DestReg, MVT::i32);
return X86::KMOVWrk;
}
else if (X86::VR64RegClass.contains(DestReg, SrcReg))
Opc = X86::MMX_MOVQ64rr;
else if (HasAVX512)
- Opc = copyPhysRegOpcode_AVX512(DestReg, SrcReg);
+ Opc = copyPhysRegOpcode_AVX512(DestReg, SrcReg, Subtarget);
else if (X86::VR128RegClass.contains(DestReg, SrcReg))
Opc = HasAVX ? X86::VMOVAPSrr : X86::MOVAPSrr;
else if (X86::VR256RegClass.contains(DestReg, SrcReg))
int Reg = FromEFLAGS ? DestReg : SrcReg;
bool is32 = X86::GR32RegClass.contains(Reg);
bool is64 = X86::GR64RegClass.contains(Reg);
+
if ((FromEFLAGS || ToEFLAGS) && (is32 || is64)) {
+ int Mov = is64 ? X86::MOV64rr : X86::MOV32rr;
+ int Push = is64 ? X86::PUSH64r : X86::PUSH32r;
+ int PushF = is64 ? X86::PUSHF64 : X86::PUSHF32;
+ int Pop = is64 ? X86::POP64r : X86::POP32r;
+ int PopF = is64 ? X86::POPF64 : X86::POPF32;
+ int AX = is64 ? X86::RAX : X86::EAX;
+
+ if (!Subtarget.hasLAHFSAHF()) {
+ assert(Subtarget.is64Bit() &&
+ "Not having LAHF/SAHF only happens on 64-bit.");
+ // Moving EFLAGS to / from another register requires a push and a pop.
+ // Notice that we have to adjust the stack if we don't want to clobber the
+ // first frame index. See X86FrameLowering.cpp - clobbersTheStack.
+ if (FromEFLAGS) {
+ BuildMI(MBB, MI, DL, get(PushF));
+ BuildMI(MBB, MI, DL, get(Pop), DestReg);
+ }
+ if (ToEFLAGS) {
+ BuildMI(MBB, MI, DL, get(Push))
+ .addReg(SrcReg, getKillRegState(KillSrc));
+ BuildMI(MBB, MI, DL, get(PopF));
+ }
+ return;
+ }
+
// The flags need to be saved, but saving EFLAGS with PUSHF/POPF is
// inefficient. Instead:
// - Save the overflow flag OF into AL using SETO, and restore it using a
// Notice that we have to adjust the stack if we don't want to clobber the
// first frame index. See X86FrameLowering.cpp - clobbersTheStack.
- int Mov = is64 ? X86::MOV64rr : X86::MOV32rr;
- int Push = is64 ? X86::PUSH64r : X86::PUSH32r;
- int Pop = is64 ? X86::POP64r : X86::POP32r;
- int AX = is64 ? X86::RAX : X86::EAX;
bool AXDead = (Reg == AX) ||
(MachineBasicBlock::LQR_Dead ==
MBB.computeRegisterLiveness(&getRegisterInfo(), AX, MI));
-
- if (!AXDead)
+ if (!AXDead) {
+ // FIXME: If computeRegisterLiveness() reported LQR_Unknown then AX may
+ // actually be dead. This is not a problem for correctness as we are just
+ // (unnecessarily) saving+restoring a dead register. However the
+ // MachineVerifier expects operands that read from dead registers
+ // to be marked with the "undef" flag.
BuildMI(MBB, MI, DL, get(Push)).addReg(AX, getKillRegState(true));
+ }
if (FromEFLAGS) {
BuildMI(MBB, MI, DL, get(X86::SETOr), X86::AL);
BuildMI(MBB, MI, DL, get(X86::LAHF));
return true;
}
+static bool expandMOV32r1(MachineInstrBuilder &MIB, const TargetInstrInfo &TII,
+ bool MinusOne) {
+ MachineBasicBlock &MBB = *MIB->getParent();
+ DebugLoc DL = MIB->getDebugLoc();
+ unsigned Reg = MIB->getOperand(0).getReg();
+
+ // Insert the XOR.
+ BuildMI(MBB, MIB.getInstr(), DL, TII.get(X86::XOR32rr), Reg)
+ .addReg(Reg, RegState::Undef)
+ .addReg(Reg, RegState::Undef);
+
+ // Turn the pseudo into an INC or DEC.
+ MIB->setDesc(TII.get(MinusOne ? X86::DEC32r : X86::INC32r));
+ MIB.addReg(Reg);
+
+ return true;
+}
+
+bool X86InstrInfo::ExpandMOVImmSExti8(MachineInstrBuilder &MIB) const {
+ MachineBasicBlock &MBB = *MIB->getParent();
+ DebugLoc DL = MIB->getDebugLoc();
+ int64_t Imm = MIB->getOperand(1).getImm();
+ assert(Imm != 0 && "Using push/pop for 0 is not efficient.");
+ MachineBasicBlock::iterator I = MIB.getInstr();
+
+ int StackAdjustment;
+
+ if (Subtarget.is64Bit()) {
+ assert(MIB->getOpcode() == X86::MOV64ImmSExti8 ||
+ MIB->getOpcode() == X86::MOV32ImmSExti8);
+ // 64-bit mode doesn't have 32-bit push/pop, so use 64-bit operations and
+ // widen the register if necessary.
+ StackAdjustment = 8;
+ BuildMI(MBB, I, DL, get(X86::PUSH64i8)).addImm(Imm);
+ MIB->setDesc(get(X86::POP64r));
+ MIB->getOperand(0)
+ .setReg(getX86SubSuperRegister(MIB->getOperand(0).getReg(), MVT::i64));
+ } else {
+ assert(MIB->getOpcode() == X86::MOV32ImmSExti8);
+ StackAdjustment = 4;
+ BuildMI(MBB, I, DL, get(X86::PUSH32i8)).addImm(Imm);
+ MIB->setDesc(get(X86::POP32r));
+ }
+
+ // Build CFI if necessary.
+ MachineFunction &MF = *MBB.getParent();
+ const X86FrameLowering *TFL = Subtarget.getFrameLowering();
+ bool IsWin64Prologue = MF.getTarget().getMCAsmInfo()->usesWindowsCFI();
+ bool NeedsDwarfCFI =
+ !IsWin64Prologue &&
+ (MF.getMMI().hasDebugInfo() || MF.getFunction()->needsUnwindTableEntry());
+ bool EmitCFI = !TFL->hasFP(MF) && NeedsDwarfCFI;
+ if (EmitCFI) {
+ TFL->BuildCFI(MBB, I, DL,
+ MCCFIInstruction::createAdjustCfaOffset(nullptr, StackAdjustment));
+ TFL->BuildCFI(MBB, std::next(I), DL,
+ MCCFIInstruction::createAdjustCfaOffset(nullptr, -StackAdjustment));
+ }
+
+ return true;
+}
+
// LoadStackGuard has so far only been implemented for 64-bit MachO. Different
// code sequence is needed for other targets.
static void expandLoadStackGuard(MachineInstrBuilder &MIB,
switch (MI->getOpcode()) {
case X86::MOV32r0:
return Expand2AddrUndef(MIB, get(X86::XOR32rr));
+ case X86::MOV32r1:
+ return expandMOV32r1(MIB, *this, /*MinusOne=*/ false);
+ case X86::MOV32r_1:
+ return expandMOV32r1(MIB, *this, /*MinusOne=*/ true);
+ case X86::MOV32ImmSExti8:
+ case X86::MOV64ImmSExti8:
+ return ExpandMOVImmSExti8(MIB);
case X86::SETB_C8r:
return Expand2AddrUndef(MIB, get(X86::SBB8rr));
case X86::SETB_C16r:
for (unsigned i = 0; i != NumAddrOps; ++i) {
const MachineOperand &MO = MOs[i];
if (i == 3 && PtrOffset != 0) {
- assert((MO.isImm() || MO.isGlobal()) &&
- "Unexpected memory operand type");
- if (MO.isImm()) {
- MIB.addImm(MO.getImm() + PtrOffset);
- } else {
- MIB.addGlobalAddress(MO.getGlobal(), MO.getOffset() + PtrOffset,
- MO.getTargetFlags());
- }
+ MIB.addDisp(MO, PtrOffset);
} else {
MIB.addOperand(MO);
}
case X86::DIVSSrr_Int: case X86::VDIVSSrr_Int:
case X86::MULSSrr_Int: case X86::VMULSSrr_Int:
case X86::SUBSSrr_Int: case X86::VSUBSSrr_Int:
+ case X86::VFMADDSSr132r_Int: case X86::VFNMADDSSr132r_Int:
+ case X86::VFMADDSSr213r_Int: case X86::VFNMADDSSr213r_Int:
+ case X86::VFMADDSSr231r_Int: case X86::VFNMADDSSr231r_Int:
+ case X86::VFMSUBSSr132r_Int: case X86::VFNMSUBSSr132r_Int:
+ case X86::VFMSUBSSr213r_Int: case X86::VFNMSUBSSr213r_Int:
+ case X86::VFMSUBSSr231r_Int: case X86::VFNMSUBSSr231r_Int:
return false;
default:
return true;
case X86::DIVSDrr_Int: case X86::VDIVSDrr_Int:
case X86::MULSDrr_Int: case X86::VMULSDrr_Int:
case X86::SUBSDrr_Int: case X86::VSUBSDrr_Int:
+ case X86::VFMADDSDr132r_Int: case X86::VFNMADDSDr132r_Int:
+ case X86::VFMADDSDr213r_Int: case X86::VFNMADDSDr213r_Int:
+ case X86::VFMADDSDr231r_Int: case X86::VFNMADDSDr231r_Int:
+ case X86::VFMSUBSDr132r_Int: case X86::VFNMSUBSDr132r_Int:
+ case X86::VFMSUBSDr213r_Int: case X86::VFNMSUBSDr213r_Int:
+ case X86::VFMSUBSDr231r_Int: case X86::VFNMSUBSDr231r_Int:
return false;
default:
return true;
// domains, but they require a bit more work than just switching opcodes.
static const uint16_t *lookup(unsigned opcode, unsigned domain) {
- for (unsigned i = 0, e = array_lengthof(ReplaceableInstrs); i != e; ++i)
- if (ReplaceableInstrs[i][domain-1] == opcode)
- return ReplaceableInstrs[i];
+ for (const uint16_t (&Row)[3] : ReplaceableInstrs)
+ if (Row[domain-1] == opcode)
+ return Row;
return nullptr;
}
static const uint16_t *lookupAVX2(unsigned opcode, unsigned domain) {
- for (unsigned i = 0, e = array_lengthof(ReplaceableInstrsAVX2); i != e; ++i)
- if (ReplaceableInstrsAVX2[i][domain-1] == opcode)
- return ReplaceableInstrsAVX2[i];
+ for (const uint16_t (&Row)[3] : ReplaceableInstrsAVX2)
+ if (Row[domain-1] == opcode)
+ return Row;
return nullptr;
}
projects / oota-llvm.git / blobdiff