[AArch64] Redundant copy elimination - remove more zero copies.

[mcrosier]

Summary

This patch removes unnecessary zero copies in BBs that are targets of b.eq/b.ne and we know the result of the compare instruction is zero. For example,

BB#0:
  subs w0, w1, w2
  str w0, [x1]
  b.ne .LBB0_2
BB#1:
  mov w0, wzr  ; <-- redundant
  str w0, [x2]
.LBB0_2

See test cases for additional examples.

Chad

[apinski-cavium]

Why did it not produce:

str wzr, [x2]

Instead? this saves the one instruction like before and also seems better for some better in general as it is not dependent on the subs.

[mcrosier]

In D35075#801162, @apinski-cavium wrote:

Why did it not produce:

str wzr, [x2]

Instead? this saves the one instruction like before and also seems better for some better in general as it is not dependent on the subs.

Yes, with a bit more work that seem very possible. However, I'd prefer to address that in a subsequent patch. I can add a FIXME, if you prefer.

As a concrete example of how this might improve codegen, having the str wzr, [x2] could expose additional opportunities for the AArch64 load/store optimization pass to pair/merge more zero stores.

[mcrosier]

ping.

[mcrosier]

ping. x2

[gberry]

Can you re-upload the diff with full context?

[mcrosier]

-Upload with full diff context, per Geoff's request.

[mcrosier]

lib/Target/AArch64/AArch64RedundantCopyElimination.cpp

231 ↗

(On Diff #107155)

I just caught this little hiccup here.. This should be:

if (DstReg == AArch64::WZR || DstReg == AArch64::XZR)
  return false;

The current code can't possibly evaluate to true. The result is that we push {AArch64::WZR, 0} and {AArch64::XZR, 0} pairs onto the KnownRegs vector. While this is actually very true, this has no effect other then to increase compile-time.

I'll fix this now and rerun correctness just to be safe.

[gberry]

First pass comments below:

lib/Target/AArch64/AArch64RedundantCopyElimination.cpp

116 ↗

(On Diff #107155)

This comment needs updating.

229 ↗

(On Diff #107155)

I think a few opcodes are missing here (it also might be a good idea to sort these so you can visually search through them more easily):
ADCSWr
ADCSXr
SBCSWr
SBCSXr
ADDSWrx
ADDSXrx
ADDSXrx64
SUBSWrx
SUBSXrx
SUBSXrx64

[mcrosier]

lib/Target/AArch64/AArch64RedundantCopyElimination.cpp

116 ↗

(On Diff #107155)

Will do.

229 ↗

(On Diff #107155)

Sure, I can add those (along with test cases).

[gberry]

A few more comments (still looking)...

lib/Target/AArch64/AArch64RedundantCopyElimination.cpp

8 ↗

(On Diff #107155)

This comment needs some updating as well: not just zero copies are removed and not just cbz/cbnz branches are handled.

16 ↗

(On Diff #107155)

Change this to 'if the flag setting opcode defined a register'

37 ↗

(On Diff #107155)

Not really related, but can you add a FIXME to look into handling CCMP opcodes as well?

182 ↗

(On Diff #107155)

Can you move this check down into the if(!ClobberedRegs[SrcReg] ...) check so that we don't miss cases where the destination is not WZR/XZR but the immediate is symbolic?

test/CodeGen/AArch64/machine-zero-copy-remove.mir

1 ↗

(On Diff #107155)

It seems like overkill to test every opcode here. I think more negative test coverage would be better.

[gberry]

Final comments below:

lib/Target/AArch64/AArch64RedundantCopyElimination.cpp

155 ↗

(On Diff #107155)

I find reviewing this part of the change a bit confusing since we're using ClobberedRegs in both knownRegValInBlock() and optimizeCopy(), but the uses are distinct as far as I can tell. I wouldn't object to a refactoring change that split these uses into two variables.

[mcrosier]

Address most of Geoff's feedback.

The one remaining ask is to add additional negative tests, which I'm working on this very moment.

[mcrosier]

-Add one additional FIXME from an earlier feedback.

[mcrosier]

-Add a few more negative tests, per Geoff's request.

[mcrosier]

Just for fun add one more test where we can remove a redundant copy and a redundant move.

llvm/trunk/lib/Target/AArch64/AArch64RedundantCopyElimination.cpp

 //=- AArch64RedundantCopyElimination.cpp - Remove useless copy for AArch64 -=//
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
-// This pass removes unnecessary zero copies in BBs that are targets of
-// cbz/cbnz instructions. For instance, the copy instruction in the code below
-// can be removed because the CBZW jumps to BB#2 when W0 is zero.
+// This pass removes unnecessary copies/moves in BBs based on a dominating
+// condition.
+//
+// We handle three cases:
+// 1. For BBs that are targets of CBZ/CBNZ instructions, we know the value of
+//    the CBZ/CBNZ source register is zero on the taken/not-taken path. For
+//    instance, the copy instruction in the code below can be removed because
+//    the CBZW jumps to BB#2 when w0 is zero.
+//
+//  BB#1:
+//    cbz w0, .LBB0_2
+//  .LBB0_2:
+//    mov w0, wzr  ; <-- redundant
+//
+// 2. If the flag setting instruction defines a register other than WZR/XZR, we
+//    can remove a zero copy in some cases.
+//
+//  BB#0:
+//    subs w0, w1, w2
+//    str w0, [x1]
+//    b.ne .LBB0_2
 //  BB#1:
-//    CBZW %W0, <BB#2>
-//  BB#2:
-//    %W0 = COPY %WZR
-// Similarly, this pass also handles non-zero copies.
+//    mov w0, wzr  ; <-- redundant
+//    str w0, [x2]
+//  .LBB0_2
+//
+// 3. Finally, if the flag setting instruction is a comparison against a
+//    constant (i.e., ADDS[W|X]ri, SUBS[W|X]ri), we can remove a mov immediate
+//    in some cases.
+//
 //  BB#0:
-//    cmp x0, #1
+//    subs xzr, x0, #1
 //    b.eq .LBB0_1
 //  .LBB0_1:
-//    orr x0, xzr, #0x1
+//    orr x0, xzr, #0x1  ; <-- redundant
 //
 // This pass should be run after register allocation.
 //
 // FIXME: This could also be extended to check the whole dominance subtree below
 // the comparison if the compile time regression is acceptable.
 //
+// FIXME: Add support for handling CCMP instructions.
+// FIXME: If the known register value is zero, we should be able to rewrite uses
+//        to use WZR/XZR directly in some cases.
 //===----------------------------------------------------------------------===//
-
 #include "AArch64.h"
 #include "llvm/ADT/Optional.h"
 #include "llvm/ADT/SetVector.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/ADT/iterator_range.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/Support/Debug.h"
 
 using namespace llvm;
 
 #define DEBUG_TYPE "aarch64-copyelim"
 
 STATISTIC(NumCopiesRemoved, "Number of copies removed.");
 
 namespace {
 class AArch64RedundantCopyElimination : public MachineFunctionPass {
   const MachineRegisterInfo *MRI;
   const TargetRegisterInfo *TRI;
-  BitVector ClobberedRegs;
+
+  // DomBBClobberedRegs is used when computing known values in the dominating
+  // BB.
+  BitVector DomBBClobberedRegs;
+
+  // OptBBClobberedRegs is used when optimizing away redundant copies/moves.
+  BitVector OptBBClobberedRegs;
 
 public:
   static char ID;
   AArch64RedundantCopyElimination() : MachineFunctionPass(ID) {
     initializeAArch64RedundantCopyEliminationPass(
         *PassRegistry::getPassRegistry());
   }
 
   struct RegImm {
     MCPhysReg Reg;
     int32_t Imm;
     RegImm(MCPhysReg Reg, int32_t Imm) : Reg(Reg), Imm(Imm) {}
   };
 
-  Optional<RegImm> knownRegValInBlock(MachineInstr &CondBr,
-                                      MachineBasicBlock *MBB,
-                                      MachineBasicBlock::iterator &FirstUse);
-  bool optimizeCopy(MachineBasicBlock *MBB);
+  bool knownRegValInBlock(MachineInstr &CondBr, MachineBasicBlock *MBB,
+                          SmallVectorImpl<RegImm> &KnownRegs,
+  bool optimizeBlock(MachineBasicBlock *MBB);
   bool runOnMachineFunction(MachineFunction &MF) override;
   MachineFunctionProperties getRequiredProperties() const override {
     return MachineFunctionProperties().set(
         MachineFunctionProperties::Property::NoVRegs);
   }
   StringRef getPassName() const override {
     return "AArch64 Redundant Copy Elimination";
   }
for (const MachineOperand &MO : MI.operands()) {
 
     for (MCRegAliasIterator AI(Reg, TRI, true); AI.isValid(); ++AI)
       ClobberedRegs.set(*AI);
   }
 }
 
 /// It's possible to determine the value of a register based on a dominating
 /// condition.  To do so, this function checks to see if the basic block \p MBB
-/// is the target to which a conditional branch \p CondBr jumps and whose
-/// equality comparison is against a constant.  If so, return a known physical
-/// register and constant value pair.  Otherwise, return None.
-Optional<AArch64RedundantCopyElimination::RegImm>
-AArch64RedundantCopyElimination::knownRegValInBlock(
+/// is the target of a conditional branch \p CondBr with an equality comparison.
+/// If the branch is a CBZ/CBNZ, we know the value of its source operand is zero
+/// in \p MBB for some cases.  Otherwise, we find and inspect the NZCV setting
+/// instruction (e.g., SUBS, ADDS).  If this instruction defines a register
+/// other than WZR/XZR, we know the value of the destination register is zero in
+/// \p MMB for some cases.  In addition, if the NZCV setting instruction is
+/// comparing against a constant we know the other source register is equal to
+/// the constant in \p MBB for some cases.  If we find any constant values, push
+/// a physical register and constant value pair onto the KnownRegs vector and
+/// return true.  Otherwise, return false if no known values were found.
+bool AArch64RedundantCopyElimination::knownRegValInBlock(
     MachineInstr &CondBr, MachineBasicBlock *MBB,
-    MachineBasicBlock::iterator &FirstUse) {
+    SmallVectorImpl<RegImm> &KnownRegs, MachineBasicBlock::iterator &FirstUse) {
   unsigned Opc = CondBr.getOpcode();
 
   // Check if the current basic block is the target block to which the
   // CBZ/CBNZ instruction jumps when its Wt/Xt is zero.
   if (((Opc == AArch64::CBZW || Opc == AArch64::CBZX) &&
        MBB == CondBr.getOperand(1).getMBB()) ||
       ((Opc == AArch64::CBNZW || Opc == AArch64::CBNZX) &&
        MBB != CondBr.getOperand(1).getMBB())) {
     FirstUse = CondBr;
-    return RegImm(CondBr.getOperand(0).getReg(), 0);
+    KnownRegs.push_back(RegImm(CondBr.getOperand(0).getReg(), 0));
+    return true;
   }
 
   // Otherwise, must be a conditional branch.
   if (Opc != AArch64::Bcc)
-    return None;
+    return false;
 
   // Must be an equality check (i.e., == or !=).
   AArch64CC::CondCode CC = (AArch64CC::CondCode)CondBr.getOperand(0).getImm();
   if (CC != AArch64CC::EQ && CC != AArch64CC::NE)
-    return None;
+    return false;
 
   MachineBasicBlock *BrTarget = CondBr.getOperand(1).getMBB();
   if ((CC == AArch64CC::EQ && BrTarget != MBB) ||
       (CC == AArch64CC::NE && BrTarget == MBB))
-    return None;
+    return false;
 
   // Stop if we get to the beginning of PredMBB.
   MachineBasicBlock *PredMBB = *MBB->pred_begin();
   assert(PredMBB == CondBr.getParent() &&
          "Conditional branch not in predecessor block!");
   if (CondBr == PredMBB->begin())
-    return None;
+    return false;
 
   // Registers clobbered in PredMBB between CondBr instruction and current
   // instruction being checked in loop.
-  ClobberedRegs.reset();
+  DomBBClobberedRegs.reset();
 
   // Find compare instruction that sets NZCV used by CondBr.
   MachineBasicBlock::reverse_iterator RIt = CondBr.getReverseIterator();
   for (MachineInstr &PredI : make_range(std::next(RIt), PredMBB->rend())) {
 
-    // Track clobbered registers.
-    trackRegDefs(PredI, ClobberedRegs, TRI);
-
     bool IsCMN = false;
     switch (PredI.getOpcode()) {
     default:
       break;
 
     // CMN is an alias for ADDS with a dead destination register.
     case AArch64::ADDSWri:
     case AArch64::ADDSXri:
       IsCMN = true;
       LLVM_FALLTHROUGH;
     // CMP is an alias for SUBS with a dead destination register.
     case AArch64::SUBSWri:
     case AArch64::SUBSXri: {
+      MCPhysReg DstReg = PredI.getOperand(0).getReg();
       MCPhysReg SrcReg = PredI.getOperand(1).getReg();
 
-      // Must not be a symbolic immediate.
-      if (!PredI.getOperand(2).isImm())
-        return None;
-
-      // The src register must not be modified between the cmp and conditional
-      // branch.  This includes a self-clobbering compare.
-      if (ClobberedRegs[SrcReg])
-        return None;
-
-      // We've found the Cmp that sets NZCV.
+      bool Res = false;
+      // If we're comparing against a non-symbolic immediate and the source
+      // register of the compare is not modified (including a self-clobbering
+      // compare) between the compare and conditional branch we known the value
+      // of the 1st source operand.
+      if (PredI.getOperand(2).isImm() && !DomBBClobberedRegs[SrcReg] &&
+          SrcReg != DstReg) {
+        // We've found the instruction that sets NZCV.
         int32_t KnownImm = PredI.getOperand(2).getImm();
         int32_t Shift = PredI.getOperand(3).getImm();
         KnownImm <<= Shift;
         if (IsCMN)
           KnownImm = -KnownImm;
         FirstUse = PredI;
-      return RegImm(SrcReg, KnownImm);
+        KnownRegs.push_back(RegImm(SrcReg, KnownImm));
+        Res = true;
+      }
+
+      // If this instructions defines something other than WZR/XZR, we know it's
+      // result is zero in some cases.
+      if (DstReg == AArch64::WZR || DstReg == AArch64::XZR)
+        return Res;
+
+      // The destination register must not be modified between the NZCV setting
+      // instruction and the conditional branch.
+      if (DomBBClobberedRegs[DstReg])
+        return Res;
+
+      FirstUse = PredI;
+      KnownRegs.push_back(RegImm(DstReg, 0));
+      return true;
+    }
+
+    // Look for NZCV setting instructions that define something other than
+    // WZR/XZR.
+    case AArch64::ADCSWr:
+    case AArch64::ADCSXr:
+    case AArch64::ADDSWrr:
+    case AArch64::ADDSWrs:
+    case AArch64::ADDSWrx:
+    case AArch64::ADDSXrr:
+    case AArch64::ADDSXrs:
+    case AArch64::ADDSXrx:
+    case AArch64::ADDSXrx64:
+    case AArch64::ANDSWri:
+    case AArch64::ANDSWrr:
+    case AArch64::ANDSWrs:
+    case AArch64::ANDSXri:
+    case AArch64::ANDSXrr:
+    case AArch64::ANDSXrs:
+    case AArch64::BICSWrr:
+    case AArch64::BICSWrs:
+    case AArch64::BICSXrs:
+    case AArch64::BICSXrr:
+    case AArch64::SBCSWr:
+    case AArch64::SBCSXr:
+    case AArch64::SUBSWrr:
+    case AArch64::SUBSWrs:
+    case AArch64::SUBSWrx:
+    case AArch64::SUBSXrr:
+    case AArch64::SUBSXrs:
+    case AArch64::SUBSXrx:
+    case AArch64::SUBSXrx64: {
+      MCPhysReg DstReg = PredI.getOperand(0).getReg();
+      if (DstReg == AArch64::WZR || DstReg == AArch64::XZR)
+        return false;
+
+      // The destination register of the NZCV setting instruction must not be
+      // modified before the conditional branch.
+      if (DomBBClobberedRegs[DstReg])
+        return false;
+
+      // We've found the instruction that sets NZCV whose DstReg == 0.
+      FirstUse = PredI;
+      KnownRegs.push_back(RegImm(DstReg, 0));
+      return true;
     }
     }
 
     // Bail if we see an instruction that defines NZCV that we don't handle.
     if (PredI.definesRegister(AArch64::NZCV))
-      return None;
+      return false;
+
+    // Track clobbered registers.
+    trackRegDefs(PredI, DomBBClobberedRegs, TRI);
   }
-  return None;
+  return false;
 }
 
-bool AArch64RedundantCopyElimination::optimizeCopy(MachineBasicBlock *MBB) {
+bool AArch64RedundantCopyElimination::optimizeBlock(MachineBasicBlock *MBB) {
   // Check if the current basic block has a single predecessor.
   if (MBB->pred_size() != 1)
     return false;
 
   // Check if the predecessor has two successors, implying the block ends in a
   // conditional branch.
   MachineBasicBlock *PredMBB = *MBB->pred_begin();
   if (PredMBB->succ_size() != 2)
bool AArch64RedundantCopyElimination::optimizeBlock(MachineBasicBlock *MBB) {
   bool SeenFirstUse = false;
   // Registers that contain a known value at the start of MBB.
   SmallVector<RegImm, 4> KnownRegs;
 
   MachineBasicBlock::iterator Itr = std::next(CondBr);
   do {
     --Itr;
 
-    Optional<RegImm> KnownRegImm = knownRegValInBlock(*Itr, MBB, FirstUse);
-    if (KnownRegImm == None)
+    if (!knownRegValInBlock(*Itr, MBB, KnownRegs, FirstUse))
       continue;
 
-    KnownRegs.push_back(*KnownRegImm);
-
-    // Reset the clobber list, which is used by knownRegValInBlock.
-    ClobberedRegs.reset();
+    // Reset the clobber list.
+    OptBBClobberedRegs.reset();
 
     // Look backward in PredMBB for COPYs from the known reg to find other
     // registers that are known to be a constant value.
     for (auto PredI = Itr;; --PredI) {
       if (FirstUse == PredI)
         SeenFirstUse = true;
 
       if (PredI->isCopy()) {
         MCPhysReg CopyDstReg = PredI->getOperand(0).getReg();
         MCPhysReg CopySrcReg = PredI->getOperand(1).getReg();
         for (auto &KnownReg : KnownRegs) {
-          if (ClobberedRegs[KnownReg.Reg])
+          if (OptBBClobberedRegs[KnownReg.Reg])
             continue;
           // If we have X = COPY Y, and Y is known to be zero, then now X is
           // known to be zero.
-          if (CopySrcReg == KnownReg.Reg && !ClobberedRegs[CopyDstReg]) {
+          if (CopySrcReg == KnownReg.Reg && !OptBBClobberedRegs[CopyDstReg]) {
             KnownRegs.push_back(RegImm(CopyDstReg, KnownReg.Imm));
             if (SeenFirstUse)
               FirstUse = PredI;
             break;
           }
           // If we have X = COPY Y, and X is known to be zero, then now Y is
           // known to be zero.
-          if (CopyDstReg == KnownReg.Reg && !ClobberedRegs[CopySrcReg]) {
+          if (CopyDstReg == KnownReg.Reg && !OptBBClobberedRegs[CopySrcReg]) {
             KnownRegs.push_back(RegImm(CopySrcReg, KnownReg.Imm));
             if (SeenFirstUse)
               FirstUse = PredI;
             break;
           }
         }
       }
 
       // Stop if we get to the beginning of PredMBB.
       if (PredI == PredMBB->begin())
         break;
 
-      trackRegDefs(*PredI, ClobberedRegs, TRI);
+      trackRegDefs(*PredI, OptBBClobberedRegs, TRI);
       // Stop if all of the known-zero regs have been clobbered.
       if (all_of(KnownRegs, [&](RegImm KnownReg) {
-            return ClobberedRegs[KnownReg.Reg];
+            return OptBBClobberedRegs[KnownReg.Reg];
           }))
         break;
     }
     break;
 
   } while (Itr != PredMBB->begin() && Itr->isTerminator());
 
   // We've not found a registers with a known value, time to bail out.
   if (KnownRegs.empty())
     return false;
 
   bool Changed = false;
   // UsedKnownRegs is the set of KnownRegs that have had uses added to MBB.
   SmallSetVector<unsigned, 4> UsedKnownRegs;
   MachineBasicBlock::iterator LastChange = MBB->begin();
-  // Remove redundant Copy instructions unless KnownReg is modified.
+  // Remove redundant copy/move instructions unless KnownReg is modified.
   for (MachineBasicBlock::iterator I = MBB->begin(), E = MBB->end(); I != E;) {
     MachineInstr *MI = &*I;
     ++I;
     bool RemovedMI = false;
     bool IsCopy = MI->isCopy();
     bool IsMoveImm = MI->isMoveImmediate();
     if (IsCopy || IsMoveImm) {
       MCPhysReg DefReg = MI->getOperand(0).getReg();

 
 bool AArch64RedundantCopyElimination::runOnMachineFunction(
     MachineFunction &MF) {
   if (skipFunction(*MF.getFunction()))
     return false;
   TRI = MF.getSubtarget().getRegisterInfo();
   MRI = &MF.getRegInfo();
 
-  // Resize the clobber register bitfield tracker.  We do this once per
-  // function and then clear the bitfield each time we optimize a copy.
-  ClobberedRegs.resize(TRI->getNumRegs());
+  // Resize the clobber register bitfield trackers.  We do this once per
+  // function.
+  DomBBClobberedRegs.resize(TRI->getNumRegs());
+  OptBBClobberedRegs.resize(TRI->getNumRegs());
 
   bool Changed = false;
   for (MachineBasicBlock &MBB : MF)
-    Changed |= optimizeCopy(&MBB);
+    Changed |= optimizeBlock(&MBB);
   return Changed;
 }
 
 FunctionPass *llvm::createAArch64RedundantCopyEliminationPass() {
   return new AArch64RedundantCopyElimination();
 }

llvm/trunk/test/CodeGen/AArch64/machine-zero-copy-remove.mir

+# RUN: llc -mtriple=aarch64--linux-gnu -run-pass=aarch64-copyelim %s -verify-machineinstrs -o - | FileCheck %s
+---
+# CHECK-LABEL: name: test1
+# CHECK: ANDSWri %w0, 1, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %wzr
+name:            test1
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %w0, %x1, %x2
+
+    %w0 = ANDSWri %w0, 1, implicit-def %nzcv
+    STRWui killed %w0, killed %x1, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x2
+
+    %w0 = COPY %wzr
+    STRWui killed %w0, killed %x2, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test2
+# CHECK: ANDSXri %x0, 1, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %xzr
+name:            test2
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %x0, %x1, %x2
+
+    %x0 = ANDSXri %x0, 1, implicit-def %nzcv
+    STRXui killed %x0, killed %x1, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x2
+
+    %x0 = COPY %xzr
+    STRXui killed %x0, killed %x2, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test3
+# CHECK: ADDSWri %w0, 1, 0, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %wzr
+name:            test3
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %w0, %x1, %x2
+
+    %w0 = ADDSWri %w0, 1, 0, implicit-def %nzcv
+    STRWui killed %w0, killed %x1, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x2
+
+    %w0 = COPY %wzr
+    STRWui killed %w0, killed %x2, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test4
+# CHECK: ADDSXri %x0, 1, 0, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %xzr
+name:            test4
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %x0, %x1, %x2
+
+    %x0 = ADDSXri %x0, 1, 0, implicit-def %nzcv
+    STRXui killed %x0, killed %x1, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x2
+
+    %x0 = COPY %xzr
+    STRXui killed %x0, killed %x2, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test5
+# CHECK: SUBSWri %w0, 1, 0, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %wzr
+name:            test5
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %w0, %x1, %x2
+
+    %w0 = SUBSWri %w0, 1, 0, implicit-def %nzcv
+    STRWui killed %w0, killed %x1, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x2
+
+    %w0 = COPY %wzr
+    STRWui killed %w0, killed %x2, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test6
+# CHECK: SUBSXri %x0, 1, 0, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %xzr
+name:            test6
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %x0, %x1, %x2
+
+    %x0 = SUBSXri %x0, 1, 0, implicit-def %nzcv
+    STRXui killed %x0, killed %x1, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x2
+
+    %x0 = COPY %xzr
+    STRXui killed %x0, killed %x2, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test7
+# CHECK: ADDSWrr %w0, %w1, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %wzr
+name:            test7
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %w0, %w1, %x2, %x3
+
+    %w0 = ADDSWrr %w0, %w1, implicit-def %nzcv
+    STRWui killed %w0, killed %x2, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x3
+
+    %w0 = COPY %wzr
+    STRWui killed %w0, killed %x3, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test8
+# CHECK: ADDSXrr %x0, %x1, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %xzr
+name:            test8
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %x0, %x1, %x2, %x3
+
+    %x0 = ADDSXrr %x0, %x1, implicit-def %nzcv
+    STRXui killed %x0, killed %x2, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x3
+
+    %x0 = COPY %xzr
+    STRXui killed %x0, killed %x3, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test9
+# CHECK: ANDSWrr %w0, %w1, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %wzr
+name:            test9
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %w0, %w1, %x2, %x3
+
+    %w0 = ANDSWrr %w0, %w1, implicit-def %nzcv
+    STRWui killed %w0, killed %x2, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x3
+
+    %w0 = COPY %wzr
+    STRWui killed %w0, killed %x3, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test10
+# CHECK: ANDSXrr %x0, %x1, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %xzr
+name:            test10
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %x0, %x1, %x2, %x3
+
+    %x0 = ANDSXrr %x0, %x1, implicit-def %nzcv
+    STRXui killed %x0, killed %x2, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x3
+
+    %x0 = COPY %xzr
+    STRXui killed %x0, killed %x3, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test11
+# CHECK: BICSWrr %w0, %w1, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %wzr
+name:            test11
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %w0, %w1, %x2, %x3
+
+    %w0 = BICSWrr %w0, %w1, implicit-def %nzcv
+    STRWui killed %w0, killed %x2, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x3
+
+    %w0 = COPY %wzr
+    STRWui killed %w0, killed %x3, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test12
+# CHECK: BICSXrr %x0, %x1, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %xzr
+name:            test12
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %x0, %x1, %x2, %x3
+
+    %x0 = BICSXrr %x0, %x1, implicit-def %nzcv
+    STRXui killed %x0, killed %x2, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x3
+
+    %x0 = COPY %xzr
+    STRXui killed %x0, killed %x3, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test13
+# CHECK: SUBSWrr %w0, %w1, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %wzr
+name:            test13
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %w0, %w1, %x2, %x3
+
+    %w0 = SUBSWrr %w0, %w1, implicit-def %nzcv
+    STRWui killed %w0, killed %x2, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x3
+
+    %w0 = COPY %wzr
+    STRWui killed %w0, killed %x3, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test14
+# CHECK: SUBSXrr %x0, %x1, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %xzr
+name:            test14
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %x0, %x1, %x2, %x3
+
+    %x0 = SUBSXrr %x0, %x1, implicit-def %nzcv
+    STRXui killed %x0, killed %x2, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x3
+
+    %x0 = COPY %xzr
+    STRXui killed %x0, killed %x3, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test15
+# CHECK: ADDSWrs %w0, %w1, 0, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %wzr
+name:            test15
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %w0, %w1, %x2, %x3
+
+    %w0 = ADDSWrs %w0, %w1, 0, implicit-def %nzcv
+    STRWui killed %w0, killed %x2, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x3
+
+    %w0 = COPY %wzr
+    STRWui killed %w0, killed %x3, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test16
+# CHECK: ADDSXrs %x0, %x1, 0, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %xzr
+name:            test16
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %x0, %x1, %x2, %x3
+
+    %x0 = ADDSXrs %x0, %x1, 0, implicit-def %nzcv
+    STRXui killed %x0, killed %x2, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x3
+
+    %x0 = COPY %xzr
+    STRXui killed %x0, killed %x3, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test17
+# CHECK: ANDSWrs %w0, %w1, 0, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %wzr
+name:            test17
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %w0, %w1, %x2, %x3
+
+    %w0 = ANDSWrs %w0, %w1, 0, implicit-def %nzcv
+    STRWui killed %w0, killed %x2, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x3
+
+    %w0 = COPY %wzr
+    STRWui killed %w0, killed %x3, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test18
+# CHECK: ANDSXrs %x0, %x1, 0, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %xzr
+name:            test18
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %x0, %x1, %x2, %x3
+
+    %x0 = ANDSXrs %x0, %x1, 0, implicit-def %nzcv
+    STRXui killed %x0, killed %x2, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x3
+
+    %x0 = COPY %xzr
+    STRXui killed %x0, killed %x3, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# CHECK-LABEL: name: test19
+# CHECK: BICSWrs %w0, %w1, 0, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: COPY %wzr
+name:            test19
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %w0, %w1, %x2, %x3
+
+    %w0 = BICSWrs %w0, %w1, 0, implicit-def %nzcv
+    STRWui killed %w0, killed %x2, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x3
+
+    %w0 = COPY %wzr
+    STRWui killed %w0, killed %x3, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# Unicorn test - we can remove a redundant copy and a redundant mov
+# CHECK-LABEL: name: test20
+# CHECK: SUBSWri %w1, 1, 0, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK-NOT: %w0 = COPY %wzr
+# CHECK-NOT: %w1 = MOVi32imm 1
+name:            test20
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %w1, %x2
+
+    %w0 = SUBSWri %w1, 1, 0, implicit-def %nzcv
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x2
+
+    %w0 = COPY %wzr
+    %w1 = MOVi32imm 1
+    STRWui killed %w0, %x2, 0
+    STRWui killed %w1, killed %x2, 1
+
+  bb.2:
+    RET_ReallyLR
+
+...
+# Negative test - MOVi32imm clobbers %w0
+# CHECK-LABEL: name: test21
+# CHECK: ANDSWri %w0, 1, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK: %w0 = COPY %wzr
+name:            test21
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %w0, %x1, %x2
+
+    %w0 = ANDSWri %w0, 1, implicit-def %nzcv
+    STRWui killed %w0, %x1, 0
+    %w0 = MOVi32imm -1
+    STRWui killed %w0, killed %x1, 1
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x2
+
+    %w0 = COPY %wzr
+    STRWui killed %w0, killed %x2, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# Negative test - SUBSXri self-clobbers x0, so MOVi64imm can't be removed
+# CHECK-LABEL: name: test22
+# CHECK: SUBSXri %x0, 1, 0, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK: %x0 = MOVi64imm 1
+name:            test22
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %x0, %x1, %x2
+
+    %x0 = SUBSXri %x0, 1, 0, implicit-def %nzcv
+    STRXui killed %x0, killed %x1, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: %x2
+
+    %x0 = MOVi64imm 1
+    STRXui killed %x0, killed %x2, 0
+
+  bb.2:
+    RET_ReallyLR
+...
+# Negative test - bb.1 has multiple preds
+# CHECK-LABEL: name: test23
+# CHECK: ADDSWri %w0, 1, 0, implicit-def %nzcv
+# CHECK: bb.1:
+# CHECK: COPY %wzr
+name:            test23
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: %w0, %x1, %x2
+
+    %w0 = ADDSWri %w0, 1, 0, implicit-def %nzcv
+    STRWui killed %w0, killed %x1, 0
+    Bcc 1, %bb.2, implicit killed %nzcv
+    B %bb.1
+
+  bb.3:
+    B %bb.1
+
+  bb.1:
+    liveins: %x2
+
+    %w0 = COPY %wzr
+    STRWui killed %w0, killed %x2, 0
+
+  bb.2:
+    RET_ReallyLR