[AArch64] Remove redundant ORRWrs which is generated by zero-extend

[jaykang10]

Summary

There are 2 patterns for zero-extend as below.

//In the case of a 32-bit def that is known to implicitly zero-extend,
//we can use a SUBREG_TO_REG.
def : Pat<(i64 (zext def32:$src)),
          (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;

//When we need to explicitly zero-extend, we use a 32-bit MOV instruction
//and then assert the extension has happened.
def : Pat<(i64 (zext GPR32:$src)),
          (SUBREG_TO_REG (i32 0), (ORRWrs WZR, GPR32:$src, 0), sub_32)>;

The def32 checks the $src needs explicitly zero-extend. However, it can not check the $src in other block and it adds ORRWrs conservatively in this case. This peephole optimization checks ORRWrs is for redundant zero-extend and try to remove it.

[jaykang10]

Fixed a bug

• replaceRegWith changes MI's defintion register. Keep it for SSA form until deleting MI.

[dmgreen]

Can you explain in more details what makes this valid? Does it depend on the top bits already being zero? What verifies that?

It might be useful to add mir tests too, to test specific cases that should/shouldn't be removed.

llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp

216

Do we need to check for WZR?

222

Do we need to avoid when they are in the same block? It would be easier for testing if we didn't.

235

Why do we rule these out? Why don't we rule out anything else?

241

-> definition

[jaykang10]

In D110841#3039405, @dmgreen wrote:

Can you explain in more details what makes this valid? Does it depend on the top bits already being zero? What verifies that?

It might be useful to add mir tests too, to test specific cases that should/shouldn't be removed.

At instruction selection level, we check the operations which do not zero-out the high half of the 64-bit register using isDef32 function as below.

// Any instruction that defines a 32-bit result zeros out the high half of the
// register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
// be copying from a truncate. But any other 32-bit operation will zero-extend
// up to 64 bits. AssertSext/AssertZext aren't saying anything about the upper
// 32 bits, they're probably just qualifying a CopyFromReg.
static inline bool isDef32(const SDNode &N) {
  unsigned Opc = N.getOpcode();
  return Opc != ISD::TRUNCATE && Opc != TargetOpcode::EXTRACT_SUBREG &&
         Opc != ISD::CopyFromReg && Opc != ISD::AssertSext &&
         Opc != ISD::AssertZext && Opc != ISD::AssertAlign &&
         Opc != ISD::FREEZE;
}

As you can see, the isDef32 checks fundamentally ISD::TRUNCATE within a basic block because the below pattern is matched with the ISD::TRUNCATE and EXTRACT_SUBREG does not guarantee the high 32 bits are zero.

// To truncate, we can simply extract from a subregister.
def : Pat<(i32 (trunc GPR64sp:$src)),
          (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;

This patch checks the ORRWrs has EXTRACT_SUBREG in different basic block as operand because the existing pattern with isDef32 resolves case in which the operand is in same basic block.

Let me try to add MIR tests.

llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp

216

You are right! We need to check it because there is below assembly pattern. Let me update the code.

def : InstAlias<"mov $dst, $src", (ORRWrs GPR32:$dst, WZR, GPR32:$src, 0), 2>;

222

If the pattern is in same block, I thought it has already been handled by existing pattern with isDef32 and we do not need to check it.

def def32 : PatLeaf<(i32 GPR32:$src), [{
  return isDef32(*N);
}]>;

// In the case of a 32-bit def that is known to implicitly zero-extend,
// we can use a SUBREG_TO_REG.
def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;

235

I was not sure it is good to keep track of the operands of PHI and COPY in this patch... because it could make code complicated... for example, checking cycled phi. If possible, I would like to solve it in separate patch...

241

Sorry... let me update it.

[jaykang10]

Following comments of @dmgreen, updated patch.

• Checked WZR
• Added MIR tests

[dmgreen]

At instruction selection level, we check the operations which do not zero-out the high half of the 64-bit register using isDef32 function as below.

// Any instruction that defines a 32-bit result zeros out the high half of the
// register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
// be copying from a truncate. But any other 32-bit operation will zero-extend
// up to 64 bits. AssertSext/AssertZext aren't saying anything about the upper
// 32 bits, they're probably just qualifying a CopyFromReg.
static inline bool isDef32(const SDNode &N) {
  unsigned Opc = N.getOpcode();
  return Opc != ISD::TRUNCATE && Opc != TargetOpcode::EXTRACT_SUBREG &&
         Opc != ISD::CopyFromReg && Opc != ISD::AssertSext &&
         Opc != ISD::AssertZext && Opc != ISD::AssertAlign &&
         Opc != ISD::FREEZE;
}

As you can see, the isDef32 checks fundamentally ISD::TRUNCATE within a basic block because the below pattern is matched with the ISD::TRUNCATE and EXTRACT_SUBREG does not guarantee the high 32 bits are zero.

// To truncate, we can simply extract from a subregister.
def : Pat<(i32 (trunc GPR64sp:$src)),
          (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;

This patch checks the ORRWrs has EXTRACT_SUBREG in different basic block as operand because the existing pattern with isDef32 resolves case in which the operand is in same basic block.

That explains things in terms of DAG-ISel, but there are other instruction selectors and different optimization between then and here. (Plus the isDef32 has had so many bugs it's difficult to trust!)

We know that all (?) instructions that generate a W register under AArch64 will zero the upper bits of the X register. We seems to say in this patch that certain EXTRACT_SUBREG are not valid, COPY and PHI are currently excluded. Is that really all we have to worry about? Do we know that the top bits are always 0 for all other grp32 sources?

llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp

222

But, is that needed for this pass? I can see it is how DAG ISel works, but if there are this pattern of code in the same block, it should still work fine, shouldn't it?

[jaykang10]

In D110841#3042058, @dmgreen wrote:

At instruction selection level, we check the operations which do not zero-out the high half of the 64-bit register using isDef32 function as below.

// Any instruction that defines a 32-bit result zeros out the high half of the
// register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
// be copying from a truncate. But any other 32-bit operation will zero-extend
// up to 64 bits. AssertSext/AssertZext aren't saying anything about the upper
// 32 bits, they're probably just qualifying a CopyFromReg.
static inline bool isDef32(const SDNode &N) {
  unsigned Opc = N.getOpcode();
  return Opc != ISD::TRUNCATE && Opc != TargetOpcode::EXTRACT_SUBREG &&
         Opc != ISD::CopyFromReg && Opc != ISD::AssertSext &&
         Opc != ISD::AssertZext && Opc != ISD::AssertAlign &&
         Opc != ISD::FREEZE;
}

As you can see, the isDef32 checks fundamentally ISD::TRUNCATE within a basic block because the below pattern is matched with the ISD::TRUNCATE and EXTRACT_SUBREG does not guarantee the high 32 bits are zero.

// To truncate, we can simply extract from a subregister.
def : Pat<(i32 (trunc GPR64sp:$src)),
          (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;

This patch checks the ORRWrs has EXTRACT_SUBREG in different basic block as operand because the existing pattern with isDef32 resolves case in which the operand is in same basic block.

That explains things in terms of DAG-ISel, but there are other instruction selectors and different optimization between then and here. (Plus the isDef32 has had so many bugs it's difficult to trust!)

We know that all (?) instructions that generate a W register under AArch64 will zero the upper bits of the X register. We seems to say in this patch that certain EXTRACT_SUBREG are not valid, COPY and PHI are currently excluded. Is that really all we have to worry about? Do we know that the top bits are always 0 for all other grp32 sources?

I agree with you. I am also not sure about it...

As you mentioned, If AArch64's 32-bit form of instruction defines the source operand of zero-extend, we do not need the zero-extend.

From https://developer.arm.com/documentation/dui0801/b/BABBGCAC
When you use the 32-bit form of an instruction, the upper 32 bits of the source registers are ignored and the upper 32 bits of the destination register are set to zero.

We need to check the zero-extend's source operands which do not come from the destination register of AArch64's 32-bit form instructions. I thought we are not sure the upper 32-bits set to zero in below two cases.

1. Function argument - There would be copy instructions from physical register to virtual register.
2. EXTRACT_SUBREG

If you feel there are more cases, please let me know. Let me update code with the cases.

For keeping track of PHI and COPY's operands, if possible, I would like to handle them after we clearly understanding the cases which does not guarantee the upper 32-bit set to zero. I am sorry for that...

llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp

222

Yep, I agree with you. Let me remove the check for different block.

[jaykang10]

Rebased

[jaykang10]

Sorry for Ping.

[efriedma]

I'm generally worried about allowing unknown instructions here.

Any AArch64 instruction that produces a 32-bit result zeros the high bits, yes. But some MachineInstr opcodes aren't really instructions in this sense. You've noted EXTRACT_SUBREG, PHI, and COPY specifically. Probably missing IMPLICIT_DEF. Not sure if there are other relevant instructions; any pseudo-instruction is potentially an issue, but auditing them for AArch64, the target-specific ones mostly don't produce GPR32.

In any case, I'd be happier if we had a bit to check for a "real" instruction, in TSFlags or something like that. I don't want to worry about modifying this in the future if, for example, we end up with a FREEZE MachineInstr.

llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp

222

Missed update?

[jaykang10]

In D110841#3067558, @efriedma wrote:

I'm generally worried about allowing unknown instructions here.

Any AArch64 instruction that produces a 32-bit result zeros the high bits, yes. But some MachineInstr opcodes aren't really instructions in this sense. You've noted EXTRACT_SUBREG, PHI, and COPY specifically. Probably missing IMPLICIT_DEF. Not sure if there are other relevant instructions; any pseudo-instruction is potentially an issue, but auditing them for AArch64, the target-specific ones mostly don't produce GPR32.

In any case, I'd be happier if we had a bit to check for a "real" instruction, in TSFlags or something like that. I don't want to worry about modifying this in the future if, for example, we end up with a FREEZE MachineInstr.

I agree with you. Let me try to check the real instructions.

llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp

222

Sorry... let me remove it.

[dmgreen]

In D110841#3064537, @jaykang10 wrote:

Sorry for Ping.

I was looking a bit last week, but didn't get very far into proving this is correct. Eli's suggestions sound good to me, and having a nice big comment explaining why it is valid would be good too.

[jaykang10]

In D110841#3068234, @dmgreen wrote:

In D110841#3064537, @jaykang10 wrote:

Sorry for Ping.

I was looking a bit last week, but didn't get very far into proving this is correct. Eli's suggestions sound good to me, and having a nice big comment explaining why it is valid would be good too.

Yep, once I finish to check the real instructions, let me try to add a big comment.

[jaykang10]

Following the comment of @efriedma, updated patch

• Added a bit to TSFlags for checking real AArch64 instruction

[dmgreen]

Do we need to add an extra TS bit, or can we just use GENERIC_OP_END?

As far as I understand the opcodes are always in the order: [TargetOpcodes, G_ opcodes, A64 Pseudos, A64 instructions]. Do we need to rule out A64 pseudos? If so can we check isPseudo().

[jaykang10]

In D110841#3071072, @dmgreen wrote:

Do we need to add an extra TS bit, or can we just use GENERIC_OP_END?

As far as I understand the opcodes are always in the order: [TargetOpcodes, G_ opcodes, A64 Pseudos, A64 instructions]. Do we need to rule out A64 pseudos? If so can we check isPseudo().

Ah, you are right!!! Let me update code. Thanks @dmgreen

[jaykang10]

Following comment of @dmgreen, updated patch.

• Check TargetOpcode::GENERIC_OP_END to distinguish real AArch64 instruction instead of adding a additional bit to TSFlags. It makes same effect.

[jaykang10]

I have checked bootstrap build and run check-all with the build. It looks OK.

[dmgreen]

Thanks.

I assume you have checked through the AArch64 pseudo instructions (the ones before ABS_ZPmZ_B in build/lib/Target/AArch64/AArch64GenInstrInfo.inc) and they look OK? They will follow the same rules of producing zeroed upper bits for W register definitions.

llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp

21

I don't think that representing this in terms of ISel patterns is useful. It should preferably be described in terms of the Machine Instructions that will be present, no matter where they come from.

llvm/test/CodeGen/AArch64/arm64-assert-zext-sext.ll

40

This looks like the kind of test we could use update_llc_test_checks on.

[jaykang10]

In D110841#3074725, @dmgreen wrote:

Thanks.

I assume you have checked through the AArch64 pseudo instructions (the ones before ABS_ZPmZ_B in build/lib/Target/AArch64/AArch64GenInstrInfo.inc) and they look OK? They will follow the same rules of producing zeroed upper bits for W register definitions.

um... I have checked the pseudo MIs which inherits Pseudo class in the AArch64InstrInfo.td and AArch64InstrFormats.td files and expandMI function in AArch64ExpandPseudoInsts.cpp. I was able to see the pseudo MIs are expanded to the AArch64 MIs... and I thought they follows the rule.

If you feel some pseudo MIs do not follow the rule, please let me know. I could miss some MIs...

llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp

21

Yep, let me update the comment.

llvm/test/CodeGen/AArch64/arm64-assert-zext-sext.ll

40

It looked it is pre-commit test with NFC tag.

Let me update the expected output with update_llc_test_checks.

[jaykang10]

Following comment of @dmgreen, updated patch.

[dmgreen]

Do we need to remove Kill flags from the uses of the register we are replacing? From something like this:

target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
target triple = "aarch64-none-none-eabi"

@e = dso_local local_unnamed_addr global i16 0, align 4
@b = dso_local local_unnamed_addr global i32 0, align 4
@d = dso_local local_unnamed_addr global i32 0, align 4
@c = dso_local local_unnamed_addr global i32* null, align 8
@a = dso_local local_unnamed_addr global i32 0, align 4

define i32 @i() {
entry:
  %0 = load i32, i32* @b, align 4
  %1 = trunc i32 %0 to i16
  %conv1 = and i16 %1, 255
  %2 = load i32, i32* @d, align 4
  %tobool.not = icmp eq i32 %2, 0
  br i1 %tobool.not, label %if.end, label %if.then

  if.then:                                          ; preds = %entry
  %conv2 = zext i16 %conv1 to i64
  %3 = inttoptr i64 %conv2 to i32*
  store i32* %3, i32** @c, align 8
  br label %if.end

  if.end:                                           ; preds = %if.then, %entry
  %4 = load i32, i32* @a, align 4
  %5 = trunc i32 %4 to i16
  %conv4 = or i16 %conv1, %5
  store i16 %conv4, i16* @e, align 4
  ret i32 0
}

llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp

240

Can we add a dbgs() output explaining what was removed?

[jaykang10]

In D110841#3074973, @dmgreen wrote:

Do we need to remove Kill flags from the uses of the register we are replacing? From something like this:

target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
target triple = "aarch64-none-none-eabi"

@e = dso_local local_unnamed_addr global i16 0, align 4
@b = dso_local local_unnamed_addr global i32 0, align 4
@d = dso_local local_unnamed_addr global i32 0, align 4
@c = dso_local local_unnamed_addr global i32* null, align 8
@a = dso_local local_unnamed_addr global i32 0, align 4

define i32 @i() {
entry:
  %0 = load i32, i32* @b, align 4
  %1 = trunc i32 %0 to i16
  %conv1 = and i16 %1, 255
  %2 = load i32, i32* @d, align 4
  %tobool.not = icmp eq i32 %2, 0
  br i1 %tobool.not, label %if.end, label %if.then

  if.then:                                          ; preds = %entry
  %conv2 = zext i16 %conv1 to i64
  %3 = inttoptr i64 %conv2 to i32*
  store i32* %3, i32** @c, align 8
  br label %if.end

  if.end:                                           ; preds = %if.then, %entry
  %4 = load i32, i32* @a, align 4
  %5 = trunc i32 %4 to i16
  %conv4 = or i16 %conv1, %5
  store i16 %conv4, i16* @e, align 4
  ret i32 0
}

You are right! We need to clear the kill flag of the source register of the ORRWrs. I was confused with the bitmask peephole opt case.

Let me clear the kill flag. Thanks for catching that. @dmgreen

llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp

240

Yep, let me add debug output.

[jaykang10]

Following comment of @dmgreen, updated patch.

• Cleared kill flag of source operand of ORRWrs.

[jaykang10]

In D110841#3075110, @jaykang10 wrote:

Following comment of @dmgreen, updated patch.

• Cleared kill flag of source operand of ORRWrs.

No error from bootstrap build and check-all on AArch64 machine.

[dmgreen]

No error from bootstrap build and check-all on AArch64 machine.

Thanks. Knowing you were away for a few days I took the time to include this in some csmith testing which I happened to already be running for another reason. It didn't come up with any other problems related to this, which is a good sign (but doesn't mean something else might be wrong, just that csmith like code does OK). So LGTM.

[jaykang10]

In D110841#3083244, @dmgreen wrote:

No error from bootstrap build and check-all on AArch64 machine.

Thanks. Knowing you were away for a few days I took the time to include this in some csmith testing which I happened to already be running for another reason. It didn't come up with any other problems related to this, which is a good sign (but doesn't mean something else might be wrong, just that csmith like code does OK). So LGTM.

Thanks for checking csmith @dmgreen! Let me push this patch after rebase.

llvm/lib/Target/AArch64/AArch64MIPeepholeOpt.cpp

 //
 // 1. MOVi32imm + ANDWrr ==> ANDWri + ANDWri
 //    MOVi64imm + ANDXrr ==> ANDXri + ANDXri
 //
 //    The mov pseudo instruction could be expanded to multiple mov instructions
 //    later. In this case, we could try to split the constant  operand of mov
 //    instruction into two bitmask immediates. It makes two AND instructions
 //    intead of multiple `mov` + `and` instructions.
+//
+// 2. Remove redundant ORRWrs which is generated by zero-extend.
+//
+//    %3:gpr32 = ORRWrs $wzr, %2, 0
+//    %4:gpr64 = SUBREG_TO_REG 0, %3, %subreg.sub_32
+//
+//    If AArch64's 32-bit form of instruction defines the source operand of
+//    ORRWrs, we can remove the ORRWrs because the upper 32 bits of the source
+//    operand are set to zero.
+//
 //===----------------------------------------------------------------------===//
 
 #include "AArch64ExpandImm.h"
 #include "AArch64InstrInfo.h"
 #include "MCTargetDesc/AArch64AddressingModes.h"
 #include "llvm/ADT/SetVector.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/MachineLoopInfo.h"
struct AArch64MIPeepholeOpt : public MachineFunctionPass {
 
   const AArch64InstrInfo *TII;
   MachineLoopInfo *MLI;
   MachineRegisterInfo *MRI;
 
   template <typename T>
   bool visitAND(MachineInstr &MI,
                 SmallSetVector<MachineInstr *, 8> &ToBeRemoved);
+  bool visitORR(MachineInstr &MI,
+                SmallSetVector<MachineInstr *, 8> &ToBeRemoved);
   bool runOnMachineFunction(MachineFunction &MF) override;
 
   StringRef getPassName() const override {
     return "AArch64 MI Peephole Optimization pass";
   }
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.setPreservesCFG();
bool AArch64MIPeepholeOpt::visitAND(
   ToBeRemoved.insert(&MI);
   if (SubregToRegMI)
     ToBeRemoved.insert(SubregToRegMI);
   ToBeRemoved.insert(MovMI);
 
   return true;
 }
 
+bool AArch64MIPeepholeOpt::visitORR(
+    MachineInstr &MI, SmallSetVector<MachineInstr *, 8> &ToBeRemoved) {
+  // Check this ORR comes from below zero-extend pattern.
+  //
+  // def : Pat<(i64 (zext GPR32:$src)),
+  //           (SUBREG_TO_REG (i32 0), (ORRWrs WZR, GPR32:$src, 0), sub_32)>;
+  if (MI.getOperand(3).getImm() != 0)
+    return false;
+
+  if (MI.getOperand(1).getReg() != AArch64::WZR)
+    return false;
+
+  MachineInstr *SrcMI = MRI->getUniqueVRegDef(MI.getOperand(2).getReg());
+  if (!SrcMI)
+    return false;
+
+  // From https://developer.arm.com/documentation/dui0801/b/BABBGCAC
+  //
+  // When you use the 32-bit form of an instruction, the upper 32 bits of the
+  // source registers are ignored and the upper 32 bits of the destination
+  // register are set to zero.
+  //
+  // If AArch64's 32-bit form of instruction defines the source operand of
+  // zero-extend, we do not need the zero-extend. Let's check the MI's opcode is
+  // real AArch64 instruction and if it is not, do not process the opcode
+  // conservatively.
+  if (SrcMI->getOpcode() <= TargetOpcode::GENERIC_OP_END)
+    return false;
+
+  Register DefReg = MI.getOperand(0).getReg();
+  MRI->replaceRegWith(DefReg, MI.getOperand(2).getReg());
+  // replaceRegWith changes MI's definition register. Keep it for SSA form until
+  // deleting MI.
+  MI.getOperand(0).setReg(DefReg);
+  ToBeRemoved.insert(&MI);
+
+  return true;
+}
+
 bool AArch64MIPeepholeOpt::runOnMachineFunction(MachineFunction &MF) {
   if (skipFunction(MF.getFunction()))
     return false;
 
   TII = static_cast<const AArch64InstrInfo *>(MF.getSubtarget().getInstrInfo());
   MLI = &getAnalysis<MachineLoopInfo>();
   MRI = &MF.getRegInfo();
 
for (MachineInstr &MI : MBB) {
       default:
         break;
       case AArch64::ANDWrr:
         Changed = visitAND<uint32_t>(MI, ToBeRemoved);
         break;
       case AArch64::ANDXrr:
         Changed = visitAND<uint64_t>(MI, ToBeRemoved);
         break;
+      case AArch64::ORRWrs:
+        Changed = visitORR(MI, ToBeRemoved);
       }
     }
   }
 
   for (MachineInstr *MI : ToBeRemoved)
     MI->eraseFromParent();
 
   return Changed;
 }
 
 FunctionPass *llvm::createAArch64MIPeepholeOptPass() {
   return new AArch64MIPeepholeOpt();
 }

llvm/test/CodeGen/AArch64/arm64-assert-zext-sext.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -O2 -mtriple=aarch64-linux-gnu < %s | FileCheck %s
 
 declare i32 @test(i32) local_unnamed_addr
 declare i32 @test1(i64) local_unnamed_addr
 
 define i32 @assertzext(i32 %n, i1 %a, i32* %b) local_unnamed_addr {
+; CHECK-LABEL: assertzext:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    stp x30, x19, [sp, #-16]! // 16-byte Folded Spill
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    .cfi_offset w19, -8
+; CHECK-NEXT:    .cfi_offset w30, -16
+; CHECK-NEXT:    mov w8, #33066
+; CHECK-NEXT:    tst w1, #0x1
+; CHECK-NEXT:    movk w8, #28567, lsl #16
+; CHECK-NEXT:    csel w19, wzr, w8, ne
+; CHECK-NEXT:    cbnz w0, .LBB0_2
+; CHECK-NEXT:  // %bb.1: // %if.then
+; CHECK-NEXT:    mov w19, wzr
+; CHECK-NEXT:    str wzr, [x2]
+; CHECK-NEXT:  .LBB0_2: // %if.end
+; CHECK-NEXT:    mov w0, w19
+; CHECK-NEXT:    bl test
+; CHECK-NEXT:    mov w0, w19
+; CHECK-NEXT:    bl test1
+; CHECK-NEXT:    mov w0, wzr
+; CHECK-NEXT:    ldp x30, x19, [sp], #16 // 16-byte Folded Reload
+; CHECK-NEXT:    ret
 entry:
   %i = select i1 %a, i64 0, i64 66296709418
   %conv.i = trunc i64 %i to i32
   %cmp = icmp eq i32 %n, 0
   br i1 %cmp, label %if.then, label %if.end
 
 if.then:                     ; preds = %entry
   store i32 0, i32* %b, align 4
   br label %if.end
 
 if.end:                      ; preds = %if.then, %entry
   %i1 = phi i32 [ 0, %if.then ], [ %conv.i, %entry ]
   %call.i = tail call i32 @test(i32 %i1)
   %i2 = sext i32 %i1 to i64
   %call1.i = tail call i32 @test1(i64 %i2)
   ret i32 0
-; CHECK:       // %if.end
-; CHECK:       mov     w{{[0-9]+}}, w{{[0-9]+}}
-; CHECK:       bl      test
-; CHECK:       mov     w{{[0-9]+}}, w{{[0-9]+}}
-; CHECK:       bl      test1
 }
 
 define i32 @assertsext(i32 %n, i8 %a) local_unnamed_addr {
+; CHECK-LABEL: assertsext:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    cbz w0, .LBB1_2
+; CHECK-NEXT:  // %bb.1:
+; CHECK-NEXT:    mov x0, xzr
+; CHECK-NEXT:    b .LBB1_3
+; CHECK-NEXT:  .LBB1_2: // %if.then
+; CHECK-NEXT:    mov x9, #24575
+; CHECK-NEXT:    sxtb w8, w1
+; CHECK-NEXT:    movk x9, #15873, lsl #16
+; CHECK-NEXT:    movk x9, #474, lsl #32
+; CHECK-NEXT:    udiv x0, x9, x8
+; CHECK-NEXT:  .LBB1_3: // %if.end
+; CHECK-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    .cfi_offset w30, -16
+; CHECK-NEXT:    bl test1
+; CHECK-NEXT:    mov w0, wzr
+; CHECK-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
+; CHECK-NEXT:    ret
 entry:
   %conv.i = sext i8 %a to i32
   %cmp = icmp eq i32 %n, 0
   br i1 %cmp, label %if.then, label %if.end
 
 if.then:                     ; preds = %entry
   %conv1 = zext i32 %conv.i to i64
   %div = udiv i64 2036854775807, %conv1
   br label %if.end
-; CHECK:       // %if.then
-; CHECK:       mov     w{{[0-9]+}}, w{{[0-9]+}}
-; CHECK:       udiv    x{{[0-9]+}}, x{{[0-9]+}}, x{{[0-9]+}}
 
 if.end:                      ; preds = %if.then, %entry
   %i1 = phi i64 [ %div, %if.then ], [ 0, %entry ]
   %call1.i = tail call i32 @test1(i64 %i1)
   ret i32 0
 }

llvm/test/CodeGen/AArch64/redundant-mov-from-zero-extend.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -O3 -mtriple=aarch64-linux-gnu < %s | FileCheck %s
+
+define i32 @test(i32 %input, i32 %n, i32 %a) {
+; CHECK-LABEL: test:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    cbz w1, .LBB0_2
+; CHECK-NEXT:  // %bb.1:
+; CHECK-NEXT:    mov w0, wzr
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB0_2: // %bb.0
+; CHECK-NEXT:    add w8, w0, w1
+; CHECK-NEXT:    mov w0, #100
+; CHECK-NEXT:    cmp w8, #4
+; CHECK-NEXT:    b.hi .LBB0_5
+; CHECK-NEXT:  // %bb.3: // %bb.0
+; CHECK-NEXT:    adrp x9, .LJTI0_0
+; CHECK-NEXT:    add x9, x9, :lo12:.LJTI0_0
+; CHECK-NEXT:    adr x10, .LBB0_4
+; CHECK-NEXT:    ldrb w11, [x9, x8]
+; CHECK-NEXT:    add x10, x10, x11, lsl #2
+; CHECK-NEXT:    br x10
+; CHECK-NEXT:  .LBB0_4: // %sw.bb
+; CHECK-NEXT:    add w0, w2, #1
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB0_5: // %bb.0
+; CHECK-NEXT:    cmp w8, #200
+; CHECK-NEXT:    b.ne .LBB0_10
+; CHECK-NEXT:  // %bb.6: // %sw.bb7
+; CHECK-NEXT:    add w0, w2, #7
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB0_7: // %sw.bb1
+; CHECK-NEXT:    add w0, w2, #3
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB0_8: // %sw.bb3
+; CHECK-NEXT:    add w0, w2, #4
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB0_9: // %sw.bb5
+; CHECK-NEXT:    add w0, w2, #5
+; CHECK-NEXT:  .LBB0_10: // %return
+; CHECK-NEXT:    ret
+entry:
+  %b = add nsw i32 %input, %n
+  %cmp = icmp eq i32 %n, 0
+  br i1 %cmp, label %bb.0, label %return
+
+bb.0:
+  switch i32 %b, label %return [
+    i32 0, label %sw.bb
+    i32 1, label %sw.bb1
+    i32 2, label %sw.bb3
+    i32 4, label %sw.bb5
+    i32 200, label %sw.bb7
+  ]
+
+sw.bb:
+  %add = add nsw i32 %a, 1
+  br label %return
+
+sw.bb1:
+  %add2 = add nsw i32 %a, 3
+  br label %return
+
+sw.bb3:
+  %add4 = add nsw i32 %a, 4
+  br label %return
+
+sw.bb5:
+  %add6 = add nsw i32 %a, 5
+  br label %return
+
+sw.bb7:
+  %add8 = add nsw i32 %a, 7
+  br label %return
+
+return:
+  %retval.0 = phi i32 [ %add8, %sw.bb7 ], [ %add6, %sw.bb5 ], [ %add4, %sw.bb3 ], [ %add2, %sw.bb1 ], [ %add, %sw.bb ], [ 100, %bb.0 ], [ 0, %entry ]
+  ret i32 %retval.0
+}

llvm/test/CodeGen/AArch64/redundant-orrwrs-from-zero-extend.mir

+# RUN: llc -mtriple=aarch64 -run-pass aarch64-mi-peephole-opt -verify-machineinstrs -o - %s | FileCheck %s
+---
+name:            test1
+# CHECK-LABEL: name: test1
+alignment:       4
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: gpr32 }
+  - { id: 1, class: gpr32 }
+  - { id: 2, class: gpr32 }
+  - { id: 3, class: gpr32 }
+  - { id: 4, class: gpr64 }
+body:             |
+  bb.0:
+    liveins: $w0, $w1
+
+    %0:gpr32 = COPY $w0
+    %1:gpr32 = COPY $w1
+    B %bb.1
+
+  bb.1:
+    %2:gpr32 = nsw ADDWrr %0, %1
+    B %bb.2
+
+  bb.2:
+    ; CHECK-LABEL: bb.2:
+    ; CHECK-NOT: %3:gpr32 = ORRWrs $wzr, %2, 0
+    ; The ORRWrs should be removed.
+    %3:gpr32 = ORRWrs $wzr, %2, 0
+    %4:gpr64 = SUBREG_TO_REG 0, %3, %subreg.sub_32
+    B %bb.3
+
+  bb.3:
+    $x0 = COPY %4
+    RET_ReallyLR implicit $x0
+...
+---
+name:            test2
+# CHECK-LABEL: name: test2
+alignment:       4
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: gpr64 }
+  - { id: 1, class: gpr32 }
+  - { id: 2, class: gpr32 }
+  - { id: 3, class: gpr64 }
+body:             |
+  bb.0:
+    liveins: $x0
+
+    %0:gpr64 = COPY $x0
+    B %bb.1
+
+  bb.1:
+    %1:gpr32 = EXTRACT_SUBREG %0, %subreg.sub_32
+    B %bb.2
+
+  bb.2:
+    ; CHECK-LABEL: bb.2:
+    ; CHECK: %2:gpr32 = ORRWrs $wzr, %1, 0
+    ; The ORRWrs should not be removed.
+    %2:gpr32 = ORRWrs $wzr, %1, 0
+    %3:gpr64 = SUBREG_TO_REG 0, %2, %subreg.sub_32
+    B %bb.3
+
+  bb.3:
+    $x0 = COPY %3
+    RET_ReallyLR implicit $x0
+...