[ARM][MachineOutliner] Add Machine Outliner support for ARM

[yroux]

Summary

Enables Machine Outlining support on ARM for ARM and Thumb2 modes. Only the
simplest outlining modes (tailcalls and thunks) are handled here.

This is a follow-up of ARM Machine Outliner support RFC D57054

[samparker]

Hi,

I like reading your code! But I'm concerned around the lack of tests here, specifically:

• CPSR liveness.
• LR liveness.
• All things PIC related.
• Linkage legality.
• A negative test for Thumb-1.
• Inline asm generally concerns me too.

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

5710

When you say 'available', do you mean 'dead'? I'm struggling to follow the logic here...

5720

Do you need to clear LRU first?

5762

I'm surprised to see IT here?

5818

Isn't this already handled in the loop above? I prefer this notation though.

5838

Looks like this MI building could be refactored.

[yroux]

Thanks for the review Sam,

I'll update the patch with more tests (for CPSR/LR it was part of D76068 but you are right it should be done here) and take your comments into account.

In D76066#1921331, @samparker wrote:

Hi,

I like reading your code! But I'm concerned around the lack of tests here, specifically:

• CPSR liveness.
• LR liveness.
• All things PIC related.
• Linkage legality.
• A negative test for Thumb-1.
• Inline asm generally concerns me too.

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

5710

Yes that's it, 'available' in llvm::LiveRegUnit means no part of the register is live so it's 'dead'.

The logic is to identify the unsafe cases where R12 or CPSR are live across the MBB without being defined in that MBB

5720

hmm, I don't think it is necessary to clear it, given that if the register was in the set the XXavailableInBlock is false so it doesn't matter if it is among liveouts or not the candidate will not have the flag UnsafeRegsDead and thus will not be outlined, if it wasn't and is part of the liveouts, we will reture false and the generic part of the MachineOutliner will not map the MBB and we will not outliune it.

5762

Yes, sorry I need to update the comment, this is a conservative way of handling IT blocks, to avoid having it splitted between an outlined region and the call sites

5818

Yes you are right, I'm cleaning this part

5838

yes, I'm refactoring it

[yroux]

Here is an update of the patch

[samparker]

Hi Yvan,

Thanks for adding the tests, I've added a few concerns and questions.

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

5615

Does this code work with the DSP instructions that read/write the Q and GE flags? I have a nasty feeling that we don't model their register usage.

5774

There's the getPredicate helper is ARMBaseInstrInfo which looks like it would be useful here.

5810

Probably worth doing these simple checks earlier. Should we also be concerned about the PC too?

llvm/lib/Target/ARM/ARMTargetMachine.cpp

556

We'll need the LowOverheadLoops pass to run for correctness, so we should instead only add the MachineOutliner if the subtarget doesn't support LOB.

[yroux]

Hi Sam,

Thanks for your comments.

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

5615

Good question, I'll test that

5774

Ah yes, and there is even isPredicated which might simplify the logic here.

5810

Well, the phasing is important here, if we move the checking of LR or SP usage earlier it will make the outlining of calls illegal since they implicitly modifies these registers or we can go back to my previous version which was checking the explicit usage but it introduces redundancy.

hmm.. maybe we can be conservative with PC, it is not verified in AArch64 code @paquette do you think it can be an issue?

llvm/lib/Target/ARM/ARMTargetMachine.cpp

556

What do you mean by "for correctness" ?

I think that it makes more sense that until MachineOutliner and LowOverheadLoops can work together, we have loloops enabled on targets which have LOB support unless it is explicitly disabled by -disable-arm-loloops flag or if the user wants machine outlining with the flag -moutline. If we do that in the opposite way it means that passing the flag -moutline will have no impact on such targets unless the -disable-arm-loloops flag is used

[samparker]

llvm/lib/Target/ARM/ARMTargetMachine.cpp

556

Ok. Well, the HardwareLoops pass inserts intrinsics, which are lowered to pseudos and then finalised by ARMLowOverheadLoops so the compiler could crash with this logic. TTI controls whether we generate a low-overhead loop and it also controls whether the vectorizer tries to optimise for those loops too... So, I think TTI will need to understand when we're trying to use the outliner too.

[yroux]

llvm/lib/Target/ARM/ARMTargetMachine.cpp

556

Ah Ok thanks, I'll try to find how to add that info into TTI

[yroux]

Here is a new update of the patch.

I remove the logic to disable LowOverheadLoops pass since Eli has added the live-ins infos inside outlined functions in D78605.

[yroux]

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

5615

From what I saw in my testing of DSP instructions, these flags are handled by CPSR usage, so this code seems good to me, do you have something else in mind ?

[samparker]

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

5615

Ah, good, then could you please add a couple of tests for those instructions? I think my last concern is back with low-overhead loops and tail-predication. I think we should explicitly avoid outlining any MVE instructions and, if this pass can, then also don't move pseudo instructions too.

[yroux]

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

5615

OK, do you know a pretty way to check if an instruction is a MVE one or a pseudo ? If not I'll add a new helper isMVEOpcode...

[samparker]

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

5615

Sure, for pseudos there's the MachineInstr isPseudo(). For MVE, it's encoded in the target flags:

const MCInstrDesc &MCID = MI.getDesc();
uint64_t Flags = MCID.TSFlags;
if ((Flags & ARMII::DomainMask) == ARMII::DomainMVE)

[yroux]

This update avoids outlining MVE instructions.

[samparker]

Thanks for adding the MVE changes, but I also still don't see any DSP tests, i.e QADD, SADD16.

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

5753

Should IMPLICIT_DEF instructions be considered invisible too?

5817

Cheers. AArch64 is nice and doesn't allow arbitrary writes to the PC, but we won't have that luxury here.

[efriedma]

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

5607

If you control all the instructions that execute, you don't need to worry about what the procedure call standard says. You do need to worry about linker veneers if the outlined function is in a different section, though.

So you need to worry about R12/CPSR on entry to the outlined function, but not on exit from the outlined function.

5806

Maybe also look for TAILJMP opcodes?

5813

Why do you need to forbid outlining code that touches LR or SP? None of the new instructions you're generating read or clobber them. (It might start mattering if you add support for additional forms of outlining, or Thumb1 support, but this patch has neither.)

5817

You might also want to support POP_RET.

[efriedma]

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

5806

Oh, nevermind, TAILJMP is a terminator so it gets handled earlier.

5817

Oh, nevermind, POP_RET is a terminator so it gets handled earlier.

[yroux]

In D76066#2014697, @samparker wrote:

Thanks for adding the MVE changes, but I also still don't see any DSP tests, i.e QADD, SADD16.

hmm... I was wrong w/r to DSP instructions Q and GE flags handling, I don't get what I did in my tests, but there is no info about CPSR in the MIR attached to these instructions. So, you had a good feeling and I can disable them, but as Eli said in his comments we don't really need to worry about APCS but only to linker veneers and I don't think a linker would clear the sticky Q bit or touch the GE ones.

Do you think we can let it as it is ?

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

5607

Yes I agree and it is what is done here, I put the same comment as what was done on AArch64 to explain where it comes from, but as it is mentioned below we remove the candidate when these registers are live into or across, which means that we can outline cases where def and use are inside the outlined region or the def inside and the use after the exit as you said.

I've modify the machine-outliner-unsafe-registers.mir test in the update version I'll submit to exhibit this.

5753

Yes, I'll add it

5813

I agree that there is no issue with SP until we modify it, I'll move that to the right patch, for LR it is needed here, when doing thunk outlining the call to the outlined function is a BL and if an instruction uses LR in the outlined region we will have a problem.

[efriedma]

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

5813

Oh, right, we're effectively hoisting the definition of LR in thunk outlining. Makes sense.

For tail-call outlining, we don't care about LR, but I guess it's trickier to separate out those two cases.

[samparker]

Eli said in his comments we don't really need to worry about APCS but only to linker veneers and I don't think a linker would clear the sticky Q bit or touch the GE ones.

Ah, yes, good point. Then I have nothing else, but definitely worth waiting to see if Eli has further comments.

[yroux]

In D76066#2022025, @samparker wrote:

Ah, yes, good point. Then I have nothing else, but definitely worth waiting to see if Eli has further comments.

Yes I'll wait for Eli's feedback and update the other patches. Thanks for the review Sam

[efriedma]

LGTM with one minor comment

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

5607

Pleas clarify the comment to explain why the procedure call standard is relevant.

[yroux]

I found a last issue when doing a full bootstrap with -moutline used to build clang, there was a case of thunk outlining where the original was call a BLX LR which was broken by the outlining call BL OUTLINE_FUNCTION_X I fixed this issue by checking the use of link register before handling calls and I've added a test case.

Rebased and full bootstrap in ARM and Thumb2 mode are now OK, I'll commit it tomorrow if that's fine for you.

clang/lib/Driver/ToolChains/Clang.cpp

 
   if (!Args.hasFlag(options::OPT_fcxx_static_destructors,
                     options::OPT_fno_cxx_static_destructors, true))
     CmdArgs.push_back("-fno-c++-static-destructors");
 
   if (Arg *A = Args.getLastArg(options::OPT_moutline,
                                options::OPT_mno_outline)) {
     if (A->getOption().matches(options::OPT_moutline)) {
-      // We only support -moutline in AArch64 right now. If we're not compiling
-      // for AArch64, emit a warning and ignore the flag. Otherwise, add the
-      // proper mllvm flags.
-      if (Triple.getArch() != llvm::Triple::aarch64 &&
-          Triple.getArch() != llvm::Triple::aarch64_32) {
+      // We only support -moutline in AArch64 and ARM targets right now. If
+      // we're not compiling for these, emit a warning and ignore the flag.
+      // Otherwise, add the proper mllvm flags.
+      if (!(Triple.isARM() || Triple.isThumb() ||
+            Triple.getArch() == llvm::Triple::aarch64 ||
+            Triple.getArch() == llvm::Triple::aarch64_32)) {
         D.Diag(diag::warn_drv_moutline_unsupported_opt) << Triple.getArchName();
       } else {
         CmdArgs.push_back("-mllvm");
         CmdArgs.push_back("-enable-machine-outliner");
       }
     } else {
       // Disable all outlining behaviour.
       CmdArgs.push_back("-mllvm");

llvm/lib/Target/ARM/ARMBaseInstrInfo.h

public:
 
   std::pair<unsigned, unsigned>
   decomposeMachineOperandsTargetFlags(unsigned TF) const override;
   ArrayRef<std::pair<unsigned, const char *>>
   getSerializableDirectMachineOperandTargetFlags() const override;
   ArrayRef<std::pair<unsigned, const char *>>
   getSerializableBitmaskMachineOperandTargetFlags() const override;
 
+  /// ARM supports the MachineOutliner.
+  bool isFunctionSafeToOutlineFrom(MachineFunction &MF,
+                                   bool OutlineFromLinkOnceODRs) const override;
+  outliner::OutlinedFunction getOutliningCandidateInfo(
+      std::vector<outliner::Candidate> &RepeatedSequenceLocs) const override;
+  outliner::InstrType getOutliningType(MachineBasicBlock::iterator &MIT,
+                                       unsigned Flags) const override;
+  bool isMBBSafeToOutlineFrom(MachineBasicBlock &MBB,
+                              unsigned &Flags) const override;
+  void buildOutlinedFrame(MachineBasicBlock &MBB, MachineFunction &MF,
+                          const outliner::OutlinedFunction &OF) const override;
+  MachineBasicBlock::iterator
+  insertOutlinedCall(Module &M, MachineBasicBlock &MBB,
+                     MachineBasicBlock::iterator &It, MachineFunction &MF,
+                     const outliner::Candidate &C) const override;
+
 private:
   unsigned getInstBundleLength(const MachineInstr &MI) const;
 
   int getVLDMDefCycle(const InstrItineraryData *ItinData,
                       const MCInstrDesc &DefMCID,
                       unsigned DefClass,
                       unsigned DefIdx, unsigned DefAlign) const;
   int getLDMDefCycle(const InstrItineraryData *ItinData,

llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp

 #include "llvm/CodeGen/LiveVariables.h"
 #include "llvm/CodeGen/MachineBasicBlock.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineMemOperand.h"
+#include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineOperand.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/ScoreboardHazardRecognizer.h"
 #include "llvm/CodeGen/SelectionDAGNodes.h"
 #include "llvm/CodeGen/TargetInstrInfo.h"
 #include "llvm/CodeGen/TargetRegisterInfo.h"
 #include "llvm/CodeGen/TargetSchedule.h"
 #include "llvm/IR/Attributes.h"
if (Cost1 < Cost2)
     return true;
   if (Cost1 > Cost2)
     return false;
 
   // If they are equal, try with !ForCodesize
   return ConstantMaterializationCost(Val1, Subtarget, !ForCodesize) <
          ConstantMaterializationCost(Val2, Subtarget, !ForCodesize);
 }
+
+/// Constants defining how certain sequences should be outlined.
+/// This encompasses how an outlined function should be called, and what kind of
+/// frame should be emitted for that outlined function.
+///
+/// \p MachineOutlinerTailCall implies that the function is being created from
+/// a sequence of instructions ending in a return.
+///
+/// That is,
+///
+/// I1                                OUTLINED_FUNCTION:
+/// I2    --> B OUTLINED_FUNCTION     I1
+/// BX LR                             I2
+///                                   BX LR
+///
+/// +-------------------------+--------+-----+
+/// |                         | Thumb2 | ARM |
+/// +-------------------------+--------+-----+
+/// | Call overhead in Bytes  |      4 |   4 |
+/// | Frame overhead in Bytes |      0 |   0 |
+/// | Stack fixup required    |     No |  No |
+/// +-------------------------+--------+-----+
+///
+/// \p MachineOutlinerThunk implies that the function is being created from
+/// a sequence of instructions ending in a call. The outlined function is
+/// called with a BL instruction, and the outlined function tail-calls the
+/// original call destination.
+///
+/// That is,
+///
+/// I1                                OUTLINED_FUNCTION:
+/// I2   --> BL OUTLINED_FUNCTION     I1
+/// BL f                              I2
+///                                   B f
+///
+/// +-------------------------+--------+-----+
+/// |                         | Thumb2 | ARM |
+/// +-------------------------+--------+-----+
+/// | Call overhead in Bytes  |      4 |   4 |
+/// | Frame overhead in Bytes |      0 |   0 |
+/// | Stack fixup required    |     No |  No |
+/// +-------------------------+--------+-----+
+
+enum MachineOutlinerClass {MachineOutlinerTailCall, MachineOutlinerThunk};
+
+enum MachineOutlinerMBBFlags {
+  LRUnavailableSomewhere = 0x2,
+  HasCalls = 0x4,
+  UnsafeRegsDead = 0x8
+};
+
+struct OutlinerCosts {
+  const int CallTailCall;
+  const int FrameTailCall;
+  const int CallThunk;
+  const int FrameThunk;
+
+  OutlinerCosts(const ARMSubtarget &target)
+      : CallTailCall(target.isThumb() ? 4 : 4),
+        FrameTailCall(target.isThumb() ? 0 : 0),
+        CallThunk(target.isThumb() ? 4 : 4),
+        FrameThunk(target.isThumb() ? 0 : 0) {}
+};
+
+outliner::OutlinedFunction ARMBaseInstrInfo::getOutliningCandidateInfo(
+    std::vector<outliner::Candidate> &RepeatedSequenceLocs) const {
+  outliner::Candidate &FirstCand = RepeatedSequenceLocs[0];
+  unsigned SequenceSize =
+      std::accumulate(FirstCand.front(), std::next(FirstCand.back()), 0,
+                      [this](unsigned Sum, const MachineInstr &MI) {
+                        return Sum + getInstSizeInBytes(MI);
+                      });
+
+  // Properties about candidate MBBs that hold for all of them.
+  unsigned FlagsSetInAll = 0xF;
+
+  // Compute liveness information for each candidate, and set FlagsSetInAll.
+  const TargetRegisterInfo &TRI = getRegisterInfo();
+  std::for_each(
+      RepeatedSequenceLocs.begin(), RepeatedSequenceLocs.end(),
+      [&FlagsSetInAll](outliner::Candidate &C) { FlagsSetInAll &= C.Flags; });
+
+  // According to the ARM Procedure Call Standard, the following are
+  // undefined on entry/exit from a function call:
+  //
+  // * Register R12(IP),
+  // * Condition codes (and thus the CPSR register)
+  //
+  // Because of this, we can't outline any sequence of instructions where one
+  // of these registers is live into/across it. Thus, we need to delete those
+  // candidates.
+  auto CantGuaranteeValueAcrossCall = [&TRI](outliner::Candidate &C) {
+    // If the unsafe registers in this block are all dead, then we don't need
+    // to compute liveness here.
+    if (C.Flags & UnsafeRegsDead)
+      return false;
+    C.initLRU(TRI);
+    LiveRegUnits LRU = C.LRU;
+    return (!LRU.available(ARM::R12) || !LRU.available(ARM::CPSR));
+  };
+
+  // Are there any candidates where those registers are live?
+  if (!(FlagsSetInAll & UnsafeRegsDead)) {
+    // Erase every candidate that violates the restrictions above. (It could be
+    // true that we have viable candidates, so it's not worth bailing out in
+    // the case that, say, 1 out of 20 candidates violate the restructions.)
+    RepeatedSequenceLocs.erase(std::remove_if(RepeatedSequenceLocs.begin(),
+                                              RepeatedSequenceLocs.end(),
+                                              CantGuaranteeValueAcrossCall),
+                               RepeatedSequenceLocs.end());
+
+    // If the sequence doesn't have enough candidates left, then we're done.
+    if (RepeatedSequenceLocs.size() < 2)
+      return outliner::OutlinedFunction();
+  }
+
+  // At this point, we have only "safe" candidates to outline. Figure out
+  // frame + call instruction information.
+
+  unsigned LastInstrOpcode = RepeatedSequenceLocs[0].back()->getOpcode();
+
+  // Helper lambda which sets call information for every candidate.
+  auto SetCandidateCallInfo =
+      [&RepeatedSequenceLocs](unsigned CallID, unsigned NumBytesForCall) {
+        for (outliner::Candidate &C : RepeatedSequenceLocs)
+          C.setCallInfo(CallID, NumBytesForCall);
+      };
+
+  OutlinerCosts *Costs = new OutlinerCosts(Subtarget);
+  unsigned FrameID = 0;
+  unsigned NumBytesToCreateFrame = 0;
+
+  // If the last instruction in any candidate is a terminator, then we should
+  // tail call all of the candidates.
+  if (RepeatedSequenceLocs[0].back()->isTerminator()) {
+    FrameID = MachineOutlinerTailCall;
+    NumBytesToCreateFrame = Costs->FrameTailCall;
+    SetCandidateCallInfo(MachineOutlinerTailCall, Costs->CallTailCall);
+  } else if (LastInstrOpcode == ARM::BL || LastInstrOpcode == ARM::BLX ||
+             LastInstrOpcode == ARM::tBL || LastInstrOpcode == ARM::tBLXr ||
+             LastInstrOpcode == ARM::tBLXi) {
+    FrameID = MachineOutlinerThunk;
+    NumBytesToCreateFrame = Costs->FrameThunk;
+    SetCandidateCallInfo(MachineOutlinerThunk, Costs->CallThunk);
+  } else
+    return outliner::OutlinedFunction();
+
+  return outliner::OutlinedFunction(RepeatedSequenceLocs, SequenceSize,
+                                    NumBytesToCreateFrame, FrameID);
+}
+
+bool ARMBaseInstrInfo::isFunctionSafeToOutlineFrom(
+    MachineFunction &MF, bool OutlineFromLinkOnceODRs) const {
+  const Function &F = MF.getFunction();
+
+  // Can F be deduplicated by the linker? If it can, don't outline from it.
+  if (!OutlineFromLinkOnceODRs && F.hasLinkOnceODRLinkage())
+    return false;
+
+  // Don't outline from functions with section markings; the program could
+  // expect that all the code is in the named section.
+  // FIXME: Allow outlining from multiple functions with the same section
+  // marking.
+  if (F.hasSection())
+    return false;
+
+  // FIXME: Thumb1 outlining is not handled
+  if (MF.getInfo<ARMFunctionInfo>()->isThumb1OnlyFunction())
+    return false;
+
+  // It's safe to outline from MF.
+  return true;
+}
+
+bool ARMBaseInstrInfo::isMBBSafeToOutlineFrom(MachineBasicBlock &MBB,
+                                              unsigned &Flags) const {
+  // Check if LR is available through all of the MBB. If it's not, then set
+  // a flag.
+  assert(MBB.getParent()->getRegInfo().tracksLiveness() &&
+         "Suitable Machine Function for outlining must track liveness");
+
+  LiveRegUnits LRU(getRegisterInfo());
+
+  std::for_each(MBB.rbegin(), MBB.rend(),
+                [&LRU](MachineInstr &MI) { LRU.accumulate(MI); });
+
+  // Check if each of the unsafe registers are available...
+  bool R12AvailableInBlock = LRU.available(ARM::R12);
+  bool CPSRAvailableInBlock = LRU.available(ARM::CPSR);
+
+  // If all of these are dead (and not live out), we know we don't have to check
+  // them later.
+  if (R12AvailableInBlock && CPSRAvailableInBlock)
+    Flags |= MachineOutlinerMBBFlags::UnsafeRegsDead;
+
+  // Now, add the live outs to the set.
+  LRU.addLiveOuts(MBB);
+
+  // If any of these registers is available in the MBB, but also a live out of
+  // the block, then we know outlining is unsafe.
+  if (R12AvailableInBlock && !LRU.available(ARM::R12))
+    return false;
+  if (CPSRAvailableInBlock && !LRU.available(ARM::CPSR))
+    return false;
+
+  // Check if there's a call inside this MachineBasicBlock.  If there is, then
+  // set a flag.
+  if (any_of(MBB, [](MachineInstr &MI) { return MI.isCall(); }))
+    Flags |= MachineOutlinerMBBFlags::HasCalls;
+
+  if (!LRU.available(ARM::LR))
+    Flags |= MachineOutlinerMBBFlags::LRUnavailableSomewhere;
+
+  return true;
+}
+
+outliner::InstrType
+ARMBaseInstrInfo::getOutliningType(MachineBasicBlock::iterator &MIT,
+                                   unsigned Flags) const {
+  MachineInstr &MI = *MIT;
+  const TargetRegisterInfo *TRI = &getRegisterInfo();
+
+  // Be conservative with inline ASM
+  if (MI.isInlineAsm())
+    return outliner::InstrType::Illegal;
+
+  // Don't allow debug values to impact outlining type.
+  if (MI.isDebugInstr() || MI.isIndirectDebugValue())
+    return outliner::InstrType::Invisible;
+
+  // At this point, KILL instructions don't really tell us much so we can go
+  // ahead and skip over them.
+  if (MI.isKill())
+    return outliner::InstrType::Invisible;
+
+  // PIC instructions contain labels, outlining them would break offset
+  // computing.  unsigned Opc = MI.getOpcode();
+  unsigned Opc = MI.getOpcode();
+  if (Opc == ARM::tPICADD || Opc == ARM::PICADD || Opc == ARM::PICSTR ||
+      Opc == ARM::PICSTRB || Opc == ARM::PICSTRH || Opc == ARM::PICLDR ||
+      Opc == ARM::PICLDRB || Opc == ARM::PICLDRH || Opc == ARM::PICLDRSB ||
+      Opc == ARM::PICLDRSH || Opc == ARM::t2LDRpci_pic ||
+      Opc == ARM::t2MOVi16_ga_pcrel || Opc == ARM::t2MOVTi16_ga_pcrel ||
+      Opc == ARM::t2MOV_ga_pcrel)
+    return outliner::InstrType::Illegal;
+
+  // Be conservative with ARMv8.1 MVE instructions.
+  if (Opc == ARM::t2BF_LabelPseudo || Opc == ARM::t2DoLoopStart ||
+      Opc == ARM::t2WhileLoopStart || Opc == ARM::t2LoopDec ||
+      Opc == ARM::t2LoopEnd)
+     return outliner::InstrType::Illegal;
+
+  const MCInstrDesc &MCID = MI.getDesc();
+  uint64_t MIFlags = MCID.TSFlags;
+  if ((MIFlags & ARMII::DomainMask) == ARMII::DomainMVE)
+     return outliner::InstrType::Illegal;
+
+  // Is this a terminator for a basic block?
+  if (MI.isTerminator()) {
+    // Don't outline if the branch is not unconditional.
+    if (isPredicated(MI))
+      return outliner::InstrType::Illegal;
+
+    // Is this the end of a function?
+    if (MI.getParent()->succ_empty())
+      return outliner::InstrType::Legal;
+
+    // It's not, so don't outline it.
+    return outliner::InstrType::Illegal;
+  }
+
+  // Make sure none of the operands are un-outlinable.
+  for (const MachineOperand &MOP : MI.operands()) {
+    if (MOP.isCPI() || MOP.isJTI() || MOP.isCFIIndex() || MOP.isFI() ||
+        MOP.isTargetIndex())
+      return outliner::InstrType::Illegal;
+  }
+
+  if (MI.isCall()) {
+    // If we don't know anything about the callee, assume it depends on the
+    // stack layout of the caller. In that case, it's only legal to outline
+    // as a tail-call.  Whitelist the call instructions we know about so we
+    // don't get unexpected results with call pseudo-instructions.
+    auto UnknownCallOutlineType = outliner::InstrType::Illegal;
+    if (Opc == ARM::BL || Opc == ARM::tBL || Opc == ARM::BLX ||
+        Opc == ARM::tBLXr || Opc == ARM::tBLXi)
+      UnknownCallOutlineType = outliner::InstrType::LegalTerminator;
+
+    return UnknownCallOutlineType;
+  }
+
+  // Don't touch the link register
+  if (MI.readsRegister(ARM::LR, TRI) || MI.modifiesRegister(ARM::LR, TRI))
+    return outliner::InstrType::Illegal;
+
+  // Be conservative with program counter.
+  if (MI.readsRegister(ARM::PC, TRI) || MI.modifiesRegister(ARM::PC, TRI))
+    return outliner::InstrType::Illegal;
+
+  // Does this use the stack?
+  if (MI.modifiesRegister(ARM::SP, TRI) || MI.readsRegister(ARM::SP, TRI))
+    return outliner::InstrType::Illegal;
+
+  // Be conservative with IT blocks.
+  if (MI.readsRegister(ARM::ITSTATE, TRI) ||
+      MI.modifiesRegister(ARM::ITSTATE, TRI))
+    return outliner::InstrType::Illegal;
+
+  // Don't outline positions.
+  if (MI.isPosition())
+    return outliner::InstrType::Illegal;
+
+  return outliner::InstrType::Legal;
+}
+
+void ARMBaseInstrInfo::buildOutlinedFrame(
+    MachineBasicBlock &MBB, MachineFunction &MF,
+    const outliner::OutlinedFunction &OF) const {
+  // For thunk outlining, rewrite the last instruction from a call to a
+  // tail-call.
+  if (OF.FrameConstructionID == MachineOutlinerThunk) {
+    MachineInstr *Call = &*--MBB.instr_end();
+    bool isThumb = Subtarget.isThumb();
+    unsigned FuncOp = isThumb ? 2 : 0;
+    unsigned Opc = Call->getOperand(FuncOp).isReg()
+                       ? isThumb ? ARM::tTAILJMPr : ARM::TAILJMPr
+                       : isThumb ? Subtarget.isTargetMachO() ? ARM::tTAILJMPd
+                                                             : ARM::tTAILJMPdND
+                                 : ARM::TAILJMPd;
+    MachineInstrBuilder MIB = BuildMI(MBB, MBB.end(), DebugLoc(), get(Opc))
+                                  .add(Call->getOperand(FuncOp));
+    if (isThumb && !Call->getOperand(FuncOp).isReg())
+      MIB.add(predOps(ARMCC::AL));
+    Call->eraseFromParent();
+  }
+}
+
+MachineBasicBlock::iterator ARMBaseInstrInfo::insertOutlinedCall(
+    Module &M, MachineBasicBlock &MBB, MachineBasicBlock::iterator &It,
+    MachineFunction &MF, const outliner::Candidate &C) const {
+  MachineInstrBuilder MIB;
+  MachineBasicBlock::iterator CallPt;
+  unsigned Opc;
+  bool isThumb = Subtarget.isThumb();
+
+  // Are we tail calling?
+  if (C.CallConstructionID == MachineOutlinerTailCall) {
+    // If yes, then we can just branch to the label.
+    Opc = isThumb
+              ? Subtarget.isTargetMachO() ? ARM::tTAILJMPd : ARM::tTAILJMPdND
+              : ARM::TAILJMPd;
+    MIB = BuildMI(MF, DebugLoc(), get(Opc))
+              .addGlobalAddress(M.getNamedValue(MF.getName()));
+    if (isThumb)
+      MIB.add(predOps(ARMCC::AL));
+    It = MBB.insert(It, MIB);
+    return It;
+  }
+
+  // Create the call instruction.
+  Opc = isThumb ? ARM::tBL : ARM::BL;
+  MachineInstrBuilder CallMIB = BuildMI(MF, DebugLoc(), get(Opc));
+  if (isThumb)
+    CallMIB.add(predOps(ARMCC::AL));
+  CallMIB.addGlobalAddress(M.getNamedValue(MF.getName()));
+
+  // Insert the call.
+  It = MBB.insert(It, CallMIB);
+  return It;
+}

llvm/lib/Target/ARM/ARMTargetMachine.cpp

if (TT.isOSBinFormatMachO()) {
     this->Options.TrapUnreachable = true;
     this->Options.NoTrapAfterNoreturn = true;
   }
 
   // ARM supports the debug entry values.
   setSupportsDebugEntryValues(true);
 
   initAsmInfo();
+
+  // ARM supports the MachineOutliner.
+  setMachineOutliner(true);
+  setSupportsDefaultOutlining(false);
 }
 
 ARMBaseTargetMachine::~ARMBaseTargetMachine() = default;
 
 const ARMSubtarget *
 ARMBaseTargetMachine::getSubtargetImpl(const Function &F) const {
   Attribute CPUAttr = F.getFnAttribute("target-cpu");
   Attribute FSAttr = F.getFnAttribute("target-features");
void ARMPassConfig::addPreEmitPass() {
 
   // Don't optimize barriers at -O0.
   if (getOptLevel() != CodeGenOpt::None)
     addPass(createARMOptimizeBarriersPass());
 }
 
 void ARMPassConfig::addPreEmitPass2() {
   addPass(createARMConstantIslandPass());
   addPass(createARMLowOverheadLoopsPass());
 
   // Identify valid longjmp targets for Windows Control Flow Guard.
   if (TM->getTargetTriple().isOSWindows())
     addPass(createCFGuardLongjmpPass());
 }

llvm/test/CodeGen/ARM/machine-outliner-tail.ll

+; RUN: llc -enable-machine-outliner -verify-machineinstrs -mtriple=arm-- \
+; RUN: --stop-after=machine-outliner < %s | FileCheck %s --check-prefix=ARM
+; RUN: llc -enable-machine-outliner -verify-machineinstrs -mtriple=thumbv7-- \
+; RUN: --stop-after=machine-outliner < %s | FileCheck %s --check-prefix=THUMB
+; RUN: llc -enable-machine-outliner -verify-machineinstrs \
+; RUN: -mtriple=thumbv7-apple-darwin --stop-after=machine-outliner < %s \
+; RUN: | FileCheck %s --check-prefix=MACHO
+; RUN: llc -enable-machine-outliner -verify-machineinstrs -mtriple=thumbv5-- \
+; RUN: --stop-after=machine-outliner < %s | FileCheck %s --check-prefix=THUMB1
+
+; ARM-LABEL: name:            OUTLINED_FUNCTION_0
+; ARM: $r0 = MOVi 1, 14 /* CC::al */, $noreg, $noreg
+; ARM-NEXT: $r1 = MOVi 2, 14 /* CC::al */, $noreg, $noreg
+; ARM-NEXT: $r2 = MOVi 3, 14 /* CC::al */, $noreg, $noreg
+; ARM-NEXT: $r3 = MOVi 4, 14 /* CC::al */, $noreg, $noreg
+; ARM-NEXT: TAILJMPd @z
+
+; THUMB-LABEL: name:            OUTLINED_FUNCTION_0
+; THUMB: $r0, dead $cpsr = tMOVi8 1, 14 /* CC::al */, $noreg
+; THUMB-NEXT: $r1, dead $cpsr = tMOVi8 2, 14 /* CC::al */, $noreg
+; THUMB-NEXT: $r2, dead $cpsr = tMOVi8 3, 14 /* CC::al */, $noreg
+; THUMB-NEXT: $r3, dead $cpsr = tMOVi8 4, 14 /* CC::al */, $noreg
+; THUMB-NEXT: tTAILJMPdND @z, 14 /* CC::al */, $noreg
+
+; MACHO-LABEL: name:            OUTLINED_FUNCTION_0
+; MACHO: $r0, dead $cpsr = tMOVi8 1, 14 /* CC::al */, $noreg
+; MACHO-NEXT: $r1, dead $cpsr = tMOVi8 2, 14 /* CC::al */, $noreg
+; MACHO-NEXT: $r2, dead $cpsr = tMOVi8 3, 14 /* CC::al */, $noreg
+; MACHO-NEXT: $r3, dead $cpsr = tMOVi8 4, 14 /* CC::al */, $noreg
+; MACHO-NEXT: tTAILJMPd @z, 14 /* CC::al */, $noreg
+
+; THUMB1-NOT: OUTLINED_FUNCTION_0
+
+define void @a() {
+entry:
+  tail call void @z(i32 1, i32 2, i32 3, i32 4)
+  ret void
+}
+
+declare void @z(i32, i32, i32, i32)
+
+define dso_local void @b(i32* nocapture readnone %p) {
+entry:
+  tail call void @z(i32 1, i32 2, i32 3, i32 4)
+  ret void
+}

llvm/test/CodeGen/ARM/machine-outliner-thunk.ll

+; RUN: llc -enable-machine-outliner -verify-machineinstrs -mtriple=armv7-- \
+; RUN: -stop-after=machine-outliner < %s | FileCheck %s --check-prefix=ARM
+; RUN: llc -enable-machine-outliner -verify-machineinstrs -mtriple=thumbv7-- \
+; RUN: -stop-after=machine-outliner < %s | FileCheck %s --check-prefix=THUMB
+; RUN: llc -enable-machine-outliner -verify-machineinstrs \
+; RUN: -mtriple=thumbv7-apple-darwin -stop-after=machine-outliner < %s \
+; RUN: | FileCheck %s --check-prefix=MACHO
+; RUN: llc -enable-machine-outliner -verify-machineinstrs -mtriple=thumbv5-- \
+; RUN: --stop-after=machine-outliner < %s | FileCheck %s --check-prefix=THUMB1
+
+declare i32 @thunk_called_fn(i32, i32, i32, i32)
+
+define i32 @a() {
+; ARM-LABEL: name:             a
+; ARM:       bb.0.entry:
+; ARM-NEXT:    liveins: $r11, $lr
+; ARM:         $sp = frame-setup STMDB_UPD $sp, 14 /* CC::al */, $noreg, killed $r11, killed $lr
+; ARM-NEXT:    frame-setup CFI_INSTRUCTION def_cfa_offset 8
+; ARM-NEXT:    frame-setup CFI_INSTRUCTION offset $lr, -4
+; ARM-NEXT:    frame-setup CFI_INSTRUCTION offset $r11, -8
+; ARM-NEXT:    BL @OUTLINED_FUNCTION_0{{.*}}
+; ARM-NEXT:    renamable $r0 = ADDri killed renamable $r0, 8, 14 /* CC::al */, $noreg, $noreg
+; ARM-NEXT:    $sp = frame-destroy LDMIA_RET $sp, 14 /* CC::al */, $noreg, def $r11, def $pc, implicit killed $r0
+
+; THUMB-LABEL: name:             a
+; THUMB:       bb.0.entry:
+; THUMB-NEXT:    liveins: $r7, $lr
+; THUMB:         frame-setup tPUSH 14 /* CC::al */, $noreg, killed $r7, killed $lr
+; THUMB-NEXT:    frame-setup CFI_INSTRUCTION def_cfa_offset 8
+; THUMB-NEXT:    frame-setup CFI_INSTRUCTION offset $lr, -4
+; THUMB-NEXT:    frame-setup CFI_INSTRUCTION offset $r7, -8
+; THUMB-NEXT:    tBL 14 /* CC::al */, $noreg, @OUTLINED_FUNCTION_0{{.*}}
+; THUMB-NEXT:    renamable $r0, dead $cpsr = tADDi8 killed renamable $r0, 8, 14 /* CC::al */, $noreg
+; THUMB-NEXT:    tPOP_RET 14 /* CC::al */, $noreg, def $r7, def $pc
+
+; MACHO-LABEL: name:             a
+; MACHO:       bb.0.entry:
+; MACHO-NEXT:    liveins: $lr
+; MACHO:         early-clobber $sp = frame-setup t2STR_PRE killed $lr, $sp, -4, 14 /* CC::al */, $noreg
+; MACHO-NEXT:    frame-setup CFI_INSTRUCTION def_cfa_offset 4
+; MACHO-NEXT:    frame-setup CFI_INSTRUCTION offset $lr, -4
+; MACHO-NEXT:    tBL 14 /* CC::al */, $noreg, @OUTLINED_FUNCTION_0{{.*}}
+; MACHO-NEXT:    renamable $r0, dead $cpsr = tADDi8 killed renamable $r0, 8, 14 /* CC::al */, $noreg
+; MACHO-NEXT:    $lr, $sp = frame-destroy t2LDR_POST $sp, 4, 14 /* CC::al */, $noreg
+; MACHO-NEXT:    tBX_RET 14 /* CC::al */, $noreg, implicit killed $r0
+
+; THUMB1-NOT: OUTLINED_FUNCTION_0
+
+entry:
+  %call = tail call i32 @thunk_called_fn(i32 1, i32 2, i32 3, i32 4)
+  %cx = add i32 %call, 8
+  ret i32 %cx
+}
+
+define i32 @b() {
+; ARM-LABEL: name:             b
+; ARM:       bb.0.entry:
+; ARM-NEXT:    liveins: $r11, $lr
+; ARM:         $sp = frame-setup STMDB_UPD $sp, 14 /* CC::al */, $noreg, killed $r11, killed $lr
+; ARM-NEXT:    frame-setup CFI_INSTRUCTION def_cfa_offset 8
+; ARM-NEXT:    frame-setup CFI_INSTRUCTION offset $lr, -4
+; ARM-NEXT:    frame-setup CFI_INSTRUCTION offset $r11, -8
+; ARM-NEXT:    BL @OUTLINED_FUNCTION_0{{.*}}
+; ARM-NEXT:    renamable $r0 = ADDri killed renamable $r0, 88, 14 /* CC::al */, $noreg, $noreg
+; ARM-NEXT:    $sp = frame-destroy LDMIA_RET $sp, 14 /* CC::al */, $noreg, def $r11, def $pc, implicit killed $r0
+
+; THUMB-LABEL: name:             b
+; THUMB:       bb.0.entry:
+; THUMB-NEXT:    liveins: $r7, $lr
+; THUMB:         frame-setup tPUSH 14 /* CC::al */, $noreg, killed $r7, killed $lr
+; THUMB-NEXT:    frame-setup CFI_INSTRUCTION def_cfa_offset 8
+; THUMB-NEXT:    frame-setup CFI_INSTRUCTION offset $lr, -4
+; THUMB-NEXT:    frame-setup CFI_INSTRUCTION offset $r7, -8
+; THUMB-NEXT:    tBL 14 /* CC::al */, $noreg, @OUTLINED_FUNCTION_0{{.*}}
+; THUMB-NEXT:    renamable $r0, dead $cpsr = tADDi8 killed renamable $r0, 88, 14 /* CC::al */, $noreg
+; THUMB-NEXT:    tPOP_RET 14 /* CC::al */, $noreg, def $r7, def $pc
+
+; MACHO-LABEL: name:             b
+; MACHO:       bb.0.entry:
+; MACHO-NEXT:    liveins: $lr
+; MACHO:         early-clobber $sp = frame-setup t2STR_PRE killed $lr, $sp, -4, 14 /* CC::al */, $noreg
+; MACHO-NEXT:    frame-setup CFI_INSTRUCTION def_cfa_offset 4
+; MACHO-NEXT:    frame-setup CFI_INSTRUCTION offset $lr, -4
+; MACHO-NEXT:    tBL 14 /* CC::al */, $noreg, @OUTLINED_FUNCTION_0{{.*}}
+; MACHO-NEXT:    renamable $r0, dead $cpsr = tADDi8 killed renamable $r0, 88, 14 /* CC::al */, $noreg
+; MACHO-NEXT:    $lr, $sp = frame-destroy t2LDR_POST $sp, 4, 14 /* CC::al */, $noreg
+; MACHO-NEXT:    tBX_RET 14 /* CC::al */, $noreg, implicit killed $r0
+entry:
+  %call = tail call i32 @thunk_called_fn(i32 1, i32 2, i32 3, i32 4)
+  %cx = add i32 %call, 88
+  ret i32 %cx
+}
+
+; ARM-LABEL: name:            OUTLINED_FUNCTION_0
+; ARM:        bb.0:
+; ARM-NEXT:   liveins: $r10, $r9, $r8, $r7, $r6, $r5, $r4, $d15, $d14, $d13, $d12, $d11, $d10, $d9, $d8
+; ARM:        $r0 = MOVi 1, 14 /* CC::al */, $noreg, $noreg
+; ARM-NEXT:   $r1 = MOVi 2, 14 /* CC::al */, $noreg, $noreg
+; ARM-NEXT:   $r2 = MOVi 3, 14 /* CC::al */, $noreg, $noreg
+; ARM-NEXT:   $r3 = MOVi 4, 14 /* CC::al */, $noreg, $noreg
+; ARM-NEXT:   TAILJMPd @thunk_called_fn, implicit $sp
+
+; THUMB-LABEL: name:            OUTLINED_FUNCTION_0
+; THUMB:        bb.0:
+; THUMB-NEXT:   liveins: $r11, $r10, $r9, $r8, $r6, $r5, $r4, $d15, $d14, $d13, $d12, $d11, $d10, $d9, $d8
+; THUMB:        $r0, dead $cpsr = tMOVi8 1, 14 /* CC::al */, $noreg
+; THUMB-NEXT:   $r1, dead $cpsr = tMOVi8 2, 14 /* CC::al */, $noreg
+; THUMB-NEXT:   $r2, dead $cpsr = tMOVi8 3, 14 /* CC::al */, $noreg
+; THUMB-NEXT:   $r3, dead $cpsr = tMOVi8 4, 14 /* CC::al */, $noreg
+; THUMB-NEXT:   tTAILJMPdND @thunk_called_fn, 14 /* CC::al */, $noreg, implicit $sp
+
+; MACHO-LABEL: name:            OUTLINED_FUNCTION_0
+; MACHO:        bb.0:
+; MACHO-NEXT:   liveins: $r7, $r6, $r5, $r4, $r11, $r10, $r8, $d15, $d14, $d13, $d12, $d11, $d10, $d9, $d8
+; MACHO:        $r0, dead $cpsr = tMOVi8 1, 14 /* CC::al */, $noreg
+; MACHO-NEXT:   $r1, dead $cpsr = tMOVi8 2, 14 /* CC::al */, $noreg
+; MACHO-NEXT:   $r2, dead $cpsr = tMOVi8 3, 14 /* CC::al */, $noreg
+; MACHO-NEXT:   $r3, dead $cpsr = tMOVi8 4, 14 /* CC::al */, $noreg
+; MACHO-NEXT:   tTAILJMPd @thunk_called_fn, 14 /* CC::al */, $noreg, implicit $sp

llvm/test/CodeGen/ARM/machine-outliner-unoutlinable.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=thumbv7-- -run-pass=machine-outliner -verify-machineinstrs \
+# RUN: %s -o - | FileCheck %s
+
+--- |
+  define void @dont_outline_asm() #0 { ret void }
+  define void @dont_outline_lr() #0 { ret void }
+  define void @dont_outline_it() #0 { ret void }
+  define void @dont_outline_pic() #0 { ret void }
+  define void @dont_outline_mve() #0 { ret void }
+  declare void @z(i32, i32, i32, i32)
+
+  attributes #0 = { minsize optsize }
+...
+---
+
+name:           dont_outline_asm
+tracksRegLiveness: true
+body:             |
+  ; CHECK-LABEL: name: dont_outline_asm
+  ; CHECK: bb.0:
+  ; CHECK:   INLINEASM &"movs  r0, #42", 1
+  ; CHECK:   tBL 14 /* CC::al */, $noreg, @OUTLINED_FUNCTION_0
+  ; CHECK: bb.1:
+  ; CHECK:   INLINEASM &"movs  r0, #42", 1
+  ; CHECK:   tBL 14 /* CC::al */, $noreg, @OUTLINED_FUNCTION_0
+  bb.0:
+    INLINEASM &"movs  r0, #42", 1
+    $r0, dead $cpsr = tMOVi8 1, 14, $noreg
+    $r1, dead $cpsr = tMOVi8 1, 14, $noreg
+    $r2, dead $cpsr = tMOVi8 1, 14, $noreg
+    $r3, dead $cpsr = tMOVi8 1, 14, $noreg
+    tBL 14, $noreg, @z
+  bb.1:
+    INLINEASM &"movs  r0, #42", 1
+    $r0, dead $cpsr = tMOVi8 1, 14, $noreg
+    $r1, dead $cpsr = tMOVi8 1, 14, $noreg
+    $r2, dead $cpsr = tMOVi8 1, 14, $noreg
+    $r3, dead $cpsr = tMOVi8 1, 14, $noreg
+    tBL 14, $noreg, @z
+  bb.2:
+    tBX_RET 14, $noreg
+...
+---
+
+name:           dont_outline_lr
+tracksRegLiveness: true
+body:             |
+  ; CHECK-LABEL: name: dont_outline_lr
+  ; CHECK-NOT: tBL 14 /* CC::al */, $noreg, @OUTLINED_FUNCTION
+  bb.0:
+    liveins: $lr
+    $r0 = tMOVr $lr, 14, $noreg
+    $r1 = tMOVr $lr, 14, $noreg
+    $r2 = tMOVr $lr, 14, $noreg
+    $r3 = tMOVr $lr, 14, $noreg
+    tBL 14, $noreg, @z
+  bb.1:
+    liveins: $lr
+    $r0 = tMOVr $lr, 14, $noreg
+    $r1 = tMOVr $lr, 14, $noreg
+    $r2 = tMOVr $lr, 14, $noreg
+    $r3 = tMOVr $lr, 14, $noreg
+    tBL 14, $noreg, @z
+  bb.2:
+    tBX_RET 14, $noreg
+...
+---
+
+name:           dont_outline_it
+tracksRegLiveness: true
+body:             |
+  ; CHECK-LABEL: name: dont_outline_it
+  ; CHECK-NOT: tBL 14 /* CC::al */, $noreg, @OUTLINED_FUNCTION
+  bb.0:
+    t2IT 0, 1, implicit-def $itstate
+    $r0, dead $cpsr = tMOVi8 1, 0, $noreg, implicit $itstate
+    $r1, dead $cpsr = tMOVi8 1, 0, $noreg, implicit $itstate
+    $r2, dead $cpsr = tMOVi8 1, 0, $noreg, implicit $itstate
+    $r3, dead $cpsr = tMOVi8 1, 0, $noreg, implicit $itstate
+    tBL 14, $noreg, @z
+  bb.1:
+    t2IT 0, 1, implicit-def $itstate
+    $r0, dead $cpsr = tMOVi8 1, 0, $noreg, implicit $itstate
+    $r1, dead $cpsr = tMOVi8 1, 0, $noreg, implicit $itstate
+    $r2, dead $cpsr = tMOVi8 1, 0, $noreg, implicit $itstate
+    $r3, dead $cpsr = tMOVi8 1, 0, $noreg, implicit $itstate
+    tBL 14, $noreg, @z
+  bb.2:
+    tBX_RET 14, $noreg
+...
+---
+
+name:           dont_outline_pic
+tracksRegLiveness: true
+body:             |
+  ; CHECK-LABEL: name: dont_outline_pic
+  ; CHECK-NOT: tBL 14 /* CC::al */, $noreg, @OUTLINED_FUNCTION
+  bb.0:
+    $r0 = t2MOVi16_ga_pcrel target-flags(arm-lo16, arm-nonlazy) @z, 0
+    $r0 = t2MOVTi16_ga_pcrel $r0, target-flags(arm-lo16, arm-nonlazy) @z, 0
+    $r0 = PICADD $r0, 1, 14, $noreg
+    $r1 = PICLDR $r0, 2, 14, $noreg
+    PICSTR $r0, $r1, 3, 14, $noreg
+    tBL 14, $noreg, @z
+  bb.1:
+    $r0 = t2MOVi16_ga_pcrel target-flags(arm-lo16, arm-nonlazy) @z, 0
+    $r0 = t2MOVTi16_ga_pcrel $r0, target-flags(arm-lo16, arm-nonlazy) @z, 0
+    $r0 = PICADD $r0, 1, 14, $noreg
+    $r1 = PICLDR $r0, 2, 14, $noreg
+    PICSTR $r0, $r1, 3, 14, $noreg
+    tBL 14, $noreg, @z
+  bb.2:
+    tBX_RET 14, $noreg
+...
+---
+
+name:           dont_outline_mve
+tracksRegLiveness: true
+body:             |
+  ; CHECK-LABEL: name: dont_outline_mve
+  ; CHECK-NOT: tBL 14 /* CC::al */, $noreg, @OUTLINED_FUNCTION
+  bb.0:
+    liveins: $r3, $r4, $q0, $q3, $q4, $q5
+    $q5 = MVE_VDUP32 $r3, 0, $noreg, $q5
+    $q4 = MVE_VDUP32 $r4, 0, $noreg, $q4
+    $q0 = MVE_VADDf32 $q4, $q5, 0, $noreg, $q0
+    t2DoLoopStart $r4
+    $r0 = MVE_VMOV_from_lane_32 renamable $q0, 1, 14, $noreg
+    tBL 14, $noreg, @z
+  bb.1:
+    liveins: $r3, $r4, $q0, $q3, $q4, $q5
+    $q5 = MVE_VDUP32 $r3, 0, $noreg, $q5
+    $q4 = MVE_VDUP32 $r4, 0, $noreg, $q4
+    $q0 = MVE_VADDf32 $q4, $q5, 0, $noreg, $q0
+    t2DoLoopStart $r4
+    $r0 = MVE_VMOV_from_lane_32 renamable $q0, 1, 14, $noreg
+    tBL 14, $noreg, @z
+  bb.2:
+    tBX_RET 14, $noreg

llvm/test/CodeGen/ARM/machine-outliner-unsafe-registers.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=arm-- -run-pass=machine-outliner -verify-machineinstrs \
+# RUN: %s -o - | FileCheck %s
+
+--- |
+  define void @outline_cpsr_r12_ok() #0 { ret void }
+  define void @dont_outline_cpsr() #0 { ret void }
+  define void @dont_outline_r12() #0 { ret void }
+  declare void @z(i32, i32, i32, i32)
+
+  attributes #0 = { minsize optsize }
+...
+---
+
+name:           outline_cpsr_r12_ok
+tracksRegLiveness: true
+body:             |
+  ; CHECK-LABEL: name: outline_cpsr_r12_ok
+  ; CHECK: bb.0:
+  ; CHECK:   BL @OUTLINED_FUNCTION_0
+  ; CHECK: bb.1:
+  ; CHECK:   $r12 = MOVi 1, 14 /* CC::al */, $noreg, $noreg
+  ; CHECK:   CMPri $r12, 42, 14 /* CC::al */, $noreg, implicit-def $cpsr
+  ; CHECK:   BL @OUTLINED_FUNCTION_0
+  bb.0:
+    liveins: $cpsr, $r12
+    $r0 = MOVi 1, 14, $noreg, $noreg
+    $r1 = MOVi 1, 14, $noreg, $noreg
+    $r2 = MOVi 1, 14, $noreg, $noreg
+    $r3 = MOVi 1, 14, $noreg, $noreg
+    BL @z
+  bb.1:
+    $r12 = MOVi 1, 14, $noreg, $noreg
+    CMPri $r12, 42, 14, $noreg, implicit-def $cpsr
+    $r0 = MOVi 1, 14, $noreg, $noreg
+    $r1 = MOVi 1, 14, $noreg, $noreg
+    $r2 = MOVi 1, 14, $noreg, $noreg
+    $r3 = MOVi 1, 14, $noreg, $noreg
+    BL @z
+  bb.2:
+    BX_RET 14, $noreg
+...
+---
+
+name:           dont_outline_cpsr
+tracksRegLiveness: true
+body:             |
+  ; CHECK-LABEL: name: dont_outline_cpsr
+  ; CHECK-NOT: BL @OUTLINED_FUNCTION_0
+  bb.0:
+    liveins: $cpsr
+    $r0 = MOVi 1, 14, $noreg, $noreg
+    $r1 = MOVi 1, 14, $noreg, $noreg
+    $r2 = MOVi 1, 14, $noreg, $noreg
+    $r3 = MOVi 1, 14, $noreg, $noreg
+    BL @z
+  bb.1:
+    liveins: $cpsr
+    BX_RET 14, $noreg
+...
+---
+
+name:           dont_outline_r12
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $r12
+    ; CHECK-LABEL: name: dont_outline_r12
+    ; CHECK-NOT: BL @OUTLINED_FUNCTION_0
+    $r0 = MOVi 1, 14, $noreg, $noreg
+    $r1 = MOVi 1, 14, $noreg, $noreg
+    $r2 = MOVi 1, 14, $noreg, $noreg
+    $r3 = MOVi 1, 14, $noreg, $noreg
+    BL @z
+    $r2 = MOVr $r12, 14, $noreg, $noreg
+    BX_RET 14, $noreg