[ARM][LowOverheadLoops] Change predicate inspection

[samparker]

Summary

Use the already provided helper function to get the operand type so that we can detect whether the vpr is being used as a predicate or not. Also use existing helpers to get the predicate indices when we converting the vpt blocks.

[SjoerdMeijer]

LGTM, with just one nit:

At first sight, this looked like a NFC, but it is not. So perhaps it's good to comment somewhere in the description, code, and/or the test case what the motivation is for this change.

llvm/lib/Target/ARM/ARMLowOverheadLoops.cpp

 ///
 //===----------------------------------------------------------------------===//
 
 #include "ARM.h"
 #include "ARMBaseInstrInfo.h"
 #include "ARMBaseRegisterInfo.h"
 #include "ARMBasicBlockInfo.h"
 #include "ARMSubtarget.h"
+#include "Thumb2InstrInfo.h"
 #include "llvm/ADT/SetOperations.h"
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineLoopInfo.h"
 #include "llvm/CodeGen/MachineLoopUtils.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/CodeGen/ReachingDefAnalysis.h"
bool LowOverheadLoop::ValidateTailPredicate(MachineInstr *StartInsertPt,
     ReachingDefAnalysis *RDA, MachineLoopInfo *MLI) {
   assert(VCTP && "VCTP instruction expected but is not set");
   // All predication within the loop should be based on vctp. If the block
   // isn't predicated on entry, check whether the vctp is within the block
   // and that all other instructions are then predicated on it.
   for (auto &Block : VPTBlocks) {
     if (Block.IsPredicatedOn(VCTP))
       continue;
-    if (!Block.HasNonUniformPredicate() || !isVCTP(Block.getDivergent()->MI))
+    if (!Block.HasNonUniformPredicate() || !isVCTP(Block.getDivergent()->MI)) {
+      LLVM_DEBUG(dbgs() << "ARM Loops: Found unsupported diverging predicate: "
+                 << *Block.getDivergent()->MI);
       return false;
+    }
     SmallVectorImpl<PredicatedMI> &Insts = Block.getInsts();
     for (auto &PredMI : Insts) {
       if (PredMI.Predicates.count(VCTP) || isVCTP(PredMI.MI))
         continue;
       LLVM_DEBUG(dbgs() << "ARM Loops: Can't convert: " << *PredMI.MI
                         << " - which is predicated on:\n";
                         for (auto *MI : PredMI.Predicates)
                           dbgs() << "   - " << *MI;
if (auto *ElemDef = RDA->getLocalLiveOutMIDef(InsertBB, NumElements)) {
         ElemDef->removeFromParent();
         InsertBB->insert(MachineBasicBlock::iterator(InsertPt), ElemDef);
         LLVM_DEBUG(dbgs() << "ARM Loops: Moved element count def: "
                    << *ElemDef);
       } else if (IsSafeToMove<MachineBasicBlock::iterator>(InsertPt, ElemDef, RDA)) {
         InsertPt->removeFromParent();
         InsertBB->insertAfter(MachineBasicBlock::iterator(ElemDef), InsertPt);
         LLVM_DEBUG(dbgs() << "ARM Loops: Moved start past: " << *ElemDef);
-      } else
+      } else {
+        LLVM_DEBUG(dbgs() << "ARM Loops: Unable to move element count to loop "
+                   << "start instruction.\n");
         return false;
       }
     }
+  }
 
   // Especially in the case of while loops, InsertBB may not be the
   // preheader, so we need to check that the register isn't redefined
   // before entering the loop.
   auto CannotProvideElements = [&RDA](MachineBasicBlock *MBB,
                                       Register NumElements) {
     // NumElements is redefined in this block.
     if (RDA->getReachingDef(&MBB->back(), NumElements) >= 0)
       return true;
 
     // Don't continue searching up through multiple predecessors.
     if (MBB->pred_size() > 1)
       return true;
 
     return false;
   };
 
   // First, find the block that looks like the preheader.
   MachineBasicBlock *MBB = MLI->findLoopPreheader(ML, true);
-  if (!MBB)
+  if (!MBB) {
+    LLVM_DEBUG(dbgs() << "ARM Loops: Didn't find preheader.\n");
     return false;
+  }
 
   // Then search backwards for a def, until we get to InsertBB.
   while (MBB != InsertBB) {
-    if (CannotProvideElements(MBB, NumElements))
+    if (CannotProvideElements(MBB, NumElements)) {
+      LLVM_DEBUG(dbgs() << "ARM Loops: Unable to provide element count.\n");
       return false;
+    }
     MBB = *MBB->pred_begin();
   }
 
   LLVM_DEBUG(dbgs() << "ARM Loops: Will use tail predication.\n");
   return true;
 }
 
 void LowOverheadLoop::CheckLegality(ARMBasicBlockUtils *BBUtils,
bool LowOverheadLoop::RecordVPTBlocks(MachineInstr* MI) {
   // 1) It will use the VPR as a predicate operand, but doesn't have to be
   //    instead a VPT block, which means we can assert while building up
   //    the VPT block because we don't find another VPST to being a new
   //    one.
   // 2) VPSEL still requires a VPR operand even after tail predicating,
   //    which means we can't remove it unless there is another
   //    instruction, such as vcmp, that can provide the VPR def.
 
-  unsigned VPROpNum = MI->getNumOperands() - 1;
   bool IsUse = false;
-  if (MI->getOperand(VPROpNum).isReg() &&
-      MI->getOperand(VPROpNum).getReg() == ARM::VPR &&
-      MI->getOperand(VPROpNum).isUse()) {
-    // If this instruction is predicated by VPR, it will be its last
-    // operand.  Also check that it's only 'Then' predicated.
-    if (!MI->getOperand(VPROpNum-1).isImm() ||
-        MI->getOperand(VPROpNum-1).getImm() != ARMVCC::Then) {
-      LLVM_DEBUG(dbgs() << "ARM Loops: Found unhandled predicate on: "
-                 << *MI);
-      return false;
-    }
-    CurrentBlock->addInst(MI, CurrentPredicate);
-    IsUse = true;
-  }
-
   bool IsDef = false;
-  for (unsigned i = 0; i < MI->getNumOperands() - 1; ++i) {
+  const MCInstrDesc &MCID = MI->getDesc();
+  for (int i = MI->getNumOperands() - 1; i >= 0; --i) {
     const MachineOperand &MO = MI->getOperand(i);
     if (!MO.isReg() || MO.getReg() != ARM::VPR)
       continue;
 
     if (MO.isDef()) {
       CurrentPredicate.insert(MI);
       IsDef = true;
+    } else if (ARM::isVpred(MCID.OpInfo[i].OperandType)) {
+      CurrentBlock->addInst(MI, CurrentPredicate);
+      IsUse = true;
     } else {
       LLVM_DEBUG(dbgs() << "ARM Loops: Found instruction using vpr: " << *MI);
       return false;
     }
   }
 
   // If we find a vpr def that is not already predicated on the vctp, we've
   // got disjoint predicates that may not be equivalent when we do the
void ARMLowOverheadLoops::RemoveLoopUpdate(LowOverheadLoop &LoLoop) {
 
   LLVM_DEBUG(dbgs() << "ARM Loops: Can't remove loop update, it's used by:\n";
              for (auto U : Uses) U->dump());
 }
 
 void ARMLowOverheadLoops::ConvertVPTBlocks(LowOverheadLoop &LoLoop) {
   auto RemovePredicate = [](MachineInstr *MI) {
     LLVM_DEBUG(dbgs() << "ARM Loops: Removing predicate from: " << *MI);
-    unsigned OpNum = MI->getNumOperands() - 1;
-    assert(MI->getOperand(OpNum-1).getImm() == ARMVCC::Then &&
+    if (int PIdx = llvm::findFirstVPTPredOperandIdx(*MI)) {
+      assert(MI->getOperand(PIdx).getImm() == ARMVCC::Then &&
              "Expected Then predicate!");
-    MI->getOperand(OpNum-1).setImm(ARMVCC::None);
-    MI->getOperand(OpNum).setReg(0);
+      MI->getOperand(PIdx).setImm(ARMVCC::None);
+      MI->getOperand(PIdx+1).setReg(0);
+    } else
+      llvm_unreachable("trying to unpredicate a non-predicated instruction");
   };
 
   // There are a few scenarios which we have to fix up:
   // 1) A VPT block with is only predicated by the vctp and has no internal vpr
   //    defs.
   // 2) A VPT block which is only predicated by the vctp but has an internal
   //    vpr def.
   // 3) A VPT block which is predicated upon the vctp as well as another vpr

llvm/test/CodeGen/Thumb2/LowOverheadLoops/vmaxmin_vpred_r.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=thumbv8.1m.main -mattr=+mve -run-pass=arm-low-overhead-loops --verify-machineinstrs %s -o - | FileCheck %s
+
+--- |
+  define hidden i32 @arm_elementwise_mul_s8(i8* %input_1_vect, i8* %input_2_vect, i32 %input_1_offset, i32 %input_2_offset, i8* %output, i32 %out_offset, i32 %out_mult, i32 %out_shift, i32 %out_activation_min, i32 %out_activation_max, i32 %block_size) local_unnamed_addr #0 {
+  entry:
+    %add = add i32 %block_size, 3
+    %div = lshr i32 %add, 2
+    %0 = call i1 @llvm.test.set.loop.iterations.i32(i32 %div)
+    br i1 %0, label %for.body.lr.ph, label %for.cond.cleanup
+
+  for.body.lr.ph:                                   ; preds = %entry
+    %.splatinsert.i41 = insertelement <4 x i32> undef, i32 %out_activation_min, i32 0
+    %.splat.i42 = shufflevector <4 x i32> %.splatinsert.i41, <4 x i32> undef, <4 x i32> zeroinitializer
+    %.splatinsert.i = insertelement <4 x i32> undef, i32 %out_activation_max, i32 0
+    %.splat.i = shufflevector <4 x i32> %.splatinsert.i, <4 x i32> undef, <4 x i32> zeroinitializer
+    br label %for.body
+
+  for.cond.cleanup:                                 ; preds = %for.body, %entry
+    ret i32 0
+
+  for.body:                                         ; preds = %for.body, %for.body.lr.ph
+    %input_1_vect.addr.052 = phi i8* [ %input_1_vect, %for.body.lr.ph ], [ %add.ptr, %for.body ]
+    %input_2_vect.addr.051 = phi i8* [ %input_2_vect, %for.body.lr.ph ], [ %add.ptr14, %for.body ]
+    %output.addr.050 = phi i8* [ %output, %for.body.lr.ph ], [ %add.ptr15, %for.body ]
+    %num_elements.049 = phi i32 [ %block_size, %for.body.lr.ph ], [ %sub, %for.body ]
+    %iv = phi i32 [ %div, %for.body.lr.ph ], [ %iv.next, %for.body ]
+    %output_cast = bitcast i8* %output.addr.050 to <4 x i32>*
+    %input_2_cast = bitcast i8* %input_2_vect.addr.051 to <4 x i32>*
+    %input_1_cast = bitcast i8* %input_1_vect.addr.052 to <4 x i32>*
+    %pred = tail call <4 x i1> @llvm.arm.mve.vctp32(i32 %num_elements.049)
+    %load.1 = tail call <4 x i32> @llvm.masked.load.v4i32.p0v4i32(<4 x i32>* %input_1_cast, i32 4, <4 x i1> %pred, <4 x i32> undef)
+    %insert.input_1_offset = insertelement <4 x i32> undef, i32 %input_1_offset, i32 0
+    %splat.input_1_offset = shufflevector <4 x i32> %insert.input_1_offset, <4 x i32> undef, <4 x i32> zeroinitializer
+    %insert.input_2_offset = insertelement <4 x i32> undef, i32 %input_2_offset, i32 0
+    %splat.input_2_offset = shufflevector <4 x i32> %insert.input_2_offset, <4 x i32> undef, <4 x i32> zeroinitializer
+    %add.1 = add <4 x i32> %load.1, %splat.input_1_offset
+    %load.2 = tail call <4 x i32> @llvm.masked.load.v4i32.p0v4i32(<4 x i32>* %input_2_cast, i32 4, <4 x i1> %pred, <4 x i32> undef)
+    %add.2 = add <4 x i32> %load.2, %splat.input_2_offset
+    %mul = mul <4 x i32> %add.1, %add.2
+    %insert.output = insertelement <4 x i32> undef, i32 %out_offset, i32 0
+    %splat.output = shufflevector <4 x i32> %insert.output, <4 x i32> undef, <4 x i32> zeroinitializer
+    %add7 = add <4 x i32> %mul, %splat.output
+    %max = tail call <4 x i32> @llvm.arm.mve.max.predicated.v4i32.v4i1(<4 x i32> %add7, <4 x i32> %.splat.i42, i32 1, <4 x i1> %pred, <4 x i32> undef)
+    %min = tail call <4 x i32> @llvm.arm.mve.min.predicated.v4i32.v4i1(<4 x i32> %max, <4 x i32> %.splat.i, i32 1, <4 x i1> %pred, <4 x i32> undef)
+    tail call void @llvm.masked.store.v4i32.p0v4i32(<4 x i32> %min, <4 x i32>* %output_cast, i32 4, <4 x i1> %pred)
+    %add.ptr = getelementptr inbounds i8, i8* %input_1_vect.addr.052, i32 4
+    %add.ptr14 = getelementptr inbounds i8, i8* %input_2_vect.addr.051, i32 4
+    %add.ptr15 = getelementptr inbounds i8, i8* %output.addr.050, i32 4
+    %sub = add i32 %num_elements.049, -4
+    %iv.next = call i32 @llvm.loop.decrement.reg.i32.i32.i32(i32 %iv, i32 1)
+    %cmp = icmp ne i32 %iv.next, 0
+    br i1 %cmp, label %for.body, label %for.cond.cleanup
+  }
+  declare <4 x i1> @llvm.arm.mve.vctp32(i32) #1
+  declare <4 x i32> @llvm.masked.load.v4i32.p0v4i32(<4 x i32>*, i32 immarg, <4 x i1>, <4 x i32>) #2
+  declare void @llvm.masked.store.v4i32.p0v4i32(<4 x i32>, <4 x i32>*, i32 immarg, <4 x i1>) #3
+  declare <4 x i32> @llvm.arm.mve.max.predicated.v4i32.v4i1(<4 x i32>, <4 x i32>, i32, <4 x i1>, <4 x i32>) #1
+  declare <4 x i32> @llvm.arm.mve.min.predicated.v4i32.v4i1(<4 x i32>, <4 x i32>, i32, <4 x i1>, <4 x i32>) #1
+  declare i1 @llvm.test.set.loop.iterations.i32(i32) #4
+  declare i32 @llvm.loop.decrement.reg.i32.i32.i32(i32, i32) #4
+  declare void @llvm.stackprotector(i8*, i8**) #5
+...
+---
+name:            arm_elementwise_mul_s8
+alignment:       2
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+failedISel:      false
+tracksRegLiveness: true
+hasWinCFI:       false
+registers:       []
+liveins:
+  - { reg: '$r0', virtual-reg: '' }
+  - { reg: '$r1', virtual-reg: '' }
+  - { reg: '$r2', virtual-reg: '' }
+  - { reg: '$r3', virtual-reg: '' }
+frameInfo:
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       20
+  offsetAdjustment: 0
+  maxAlignment:    4
+  adjustsStack:    false
+  hasCalls:        false
+  stackProtector:  ''
+  maxCallFrameSize: 0
+  cvBytesOfCalleeSavedRegisters: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+  localFrameSize:  0
+  savePoint:       ''
+  restorePoint:    ''
+fixedStack:
+  - { id: 0, type: default, offset: 24, size: 4, alignment: 8, stack-id: default,
+      isImmutable: true, isAliased: false, callee-saved-register: '', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 1, type: default, offset: 20, size: 4, alignment: 4, stack-id: default,
+      isImmutable: true, isAliased: false, callee-saved-register: '', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 2, type: default, offset: 16, size: 4, alignment: 8, stack-id: default,
+      isImmutable: true, isAliased: false, callee-saved-register: '', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 3, type: default, offset: 12, size: 4, alignment: 4, stack-id: default,
+      isImmutable: true, isAliased: false, callee-saved-register: '', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 4, type: default, offset: 8, size: 4, alignment: 8, stack-id: default,
+      isImmutable: true, isAliased: false, callee-saved-register: '', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 5, type: default, offset: 4, size: 4, alignment: 4, stack-id: default,
+      isImmutable: true, isAliased: false, callee-saved-register: '', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 6, type: default, offset: 0, size: 4, alignment: 8, stack-id: default,
+      isImmutable: true, isAliased: false, callee-saved-register: '', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+stack:
+  - { id: 0, name: '', type: spill-slot, offset: -4, size: 4, alignment: 4,
+      stack-id: default, callee-saved-register: '$lr', callee-saved-restored: false,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 1, name: '', type: spill-slot, offset: -8, size: 4, alignment: 4,
+      stack-id: default, callee-saved-register: '$r7', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 2, name: '', type: spill-slot, offset: -12, size: 4, alignment: 4,
+      stack-id: default, callee-saved-register: '$r6', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 3, name: '', type: spill-slot, offset: -16, size: 4, alignment: 4,
+      stack-id: default, callee-saved-register: '$r5', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 4, name: '', type: spill-slot, offset: -20, size: 4, alignment: 4,
+      stack-id: default, callee-saved-register: '$r4', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+callSites:       []
+constants:       []
+machineFunctionInfo: {}
+body:             |
+  ; CHECK-LABEL: name: arm_elementwise_mul_s8
+  ; CHECK: bb.0.entry:
+  ; CHECK:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK:   liveins: $r0, $r1, $r2, $r3, $r4, $r5, $r6, $r7, $lr
+  ; CHECK:   frame-setup tPUSH 14, $noreg, killed $r4, killed $r5, killed $r6, killed $r7, killed $lr, implicit-def $sp, implicit $sp
+  ; CHECK:   frame-setup CFI_INSTRUCTION def_cfa_offset 20
+  ; CHECK:   frame-setup CFI_INSTRUCTION offset $lr, -4
+  ; CHECK:   frame-setup CFI_INSTRUCTION offset $r7, -8
+  ; CHECK:   frame-setup CFI_INSTRUCTION offset $r6, -12
+  ; CHECK:   frame-setup CFI_INSTRUCTION offset $r5, -16
+  ; CHECK:   frame-setup CFI_INSTRUCTION offset $r4, -20
+  ; CHECK:   renamable $r12 = t2LDRi12 $sp, 44, 14, $noreg :: (load 4 from %fixed-stack.6, align 8)
+  ; CHECK:   $lr = MVE_WLSTP_32 renamable $r12, %bb.3
+  ; CHECK: bb.1.for.body.lr.ph:
+  ; CHECK:   successors: %bb.2(0x80000000)
+  ; CHECK:   liveins: $lr, $r0, $r1, $r2, $r3, $r12
+  ; CHECK:   $r7, $r6 = t2LDRDi8 $sp, 36, 14, $noreg :: (load 4 from %fixed-stack.4, align 8), (load 4 from %fixed-stack.5)
+  ; CHECK:   $r5, $r4 = t2LDRDi8 $sp, 20, 14, $noreg :: (load 4 from %fixed-stack.0, align 8), (load 4 from %fixed-stack.1)
+  ; CHECK:   renamable $q0 = MVE_VDUP32 killed renamable $r6, 0, $noreg, undef renamable $q0
+  ; CHECK:   renamable $q1 = MVE_VDUP32 killed renamable $r7, 0, $noreg, undef renamable $q1
+  ; CHECK: bb.2.for.body:
+  ; CHECK:   successors: %bb.2(0x7c000000), %bb.3(0x04000000)
+  ; CHECK:   liveins: $lr, $q0, $q1, $r0, $r1, $r2, $r3, $r4, $r5, $r12
+  ; CHECK:   renamable $r1, renamable $q2 = MVE_VLDRWU32_post killed renamable $r1, 4, 0, $noreg :: (load 16 from %ir.input_2_cast, align 4)
+  ; CHECK:   renamable $r0, renamable $q3 = MVE_VLDRWU32_post killed renamable $r0, 4, 0, $noreg :: (load 16 from %ir.input_1_cast, align 4)
+  ; CHECK:   renamable $q2 = MVE_VADD_qr_i32 killed renamable $q2, renamable $r3, 0, $noreg, undef renamable $q2
+  ; CHECK:   renamable $q3 = MVE_VADD_qr_i32 killed renamable $q3, renamable $r2, 0, $noreg, undef renamable $q3
+  ; CHECK:   renamable $r12 = t2SUBri killed renamable $r12, 4, 14, $noreg, $noreg
+  ; CHECK:   renamable $q2 = MVE_VMULi32 killed renamable $q3, killed renamable $q2, 0, $noreg, undef renamable $q2
+  ; CHECK:   renamable $q2 = MVE_VADD_qr_i32 killed renamable $q2, renamable $r4, 0, $noreg, undef renamable $q2
+  ; CHECK:   renamable $q2 = MVE_VMAXu32 killed renamable $q2, renamable $q1, 0, $noreg, undef renamable $q2
+  ; CHECK:   renamable $q2 = MVE_VMINu32 killed renamable $q2, renamable $q0, 0, $noreg, undef renamable $q2
+  ; CHECK:   renamable $r5 = MVE_VSTRWU32_post killed renamable $q2, killed renamable $r5, 4, 0, killed $noreg :: (store 16 into %ir.output_cast, align 4)
+  ; CHECK:   $lr = MVE_LETP renamable $lr, %bb.2
+  ; CHECK: bb.3.for.cond.cleanup:
+  ; CHECK:   $r0, dead $cpsr = tMOVi8 0, 14, $noreg
+  ; CHECK:   tPOP_RET 14, $noreg, def $r4, def $r5, def $r6, def $r7, def $pc, implicit killed $r0
+  bb.0.entry:
+    successors: %bb.1(0x40000000), %bb.3(0x40000000)
+    liveins: $r0, $r1, $r2, $r3, $r4, $r5, $r6, $r7, $lr
+
+    frame-setup tPUSH 14, $noreg, killed $r4, killed $r5, killed $r6, killed $r7, killed $lr, implicit-def $sp, implicit $sp
+    frame-setup CFI_INSTRUCTION def_cfa_offset 20
+    frame-setup CFI_INSTRUCTION offset $lr, -4
+    frame-setup CFI_INSTRUCTION offset $r7, -8
+    frame-setup CFI_INSTRUCTION offset $r6, -12
+    frame-setup CFI_INSTRUCTION offset $r5, -16
+    frame-setup CFI_INSTRUCTION offset $r4, -20
+    renamable $r12 = t2LDRi12 $sp, 44, 14, $noreg :: (load 4 from %fixed-stack.0, align 8)
+    renamable $r5 = t2ADDri renamable $r12, 3, 14, $noreg, $noreg
+    renamable $lr = t2LSRri killed renamable $r5, 2, 14, $noreg, $noreg
+    t2WhileLoopStart renamable $lr, %bb.3, implicit-def dead $cpsr
+    tB %bb.1, 14, $noreg
+
+  bb.1.for.body.lr.ph:
+    successors: %bb.2(0x80000000)
+    liveins: $lr, $r0, $r1, $r2, $r3, $r12
+
+    $r7, $r6 = t2LDRDi8 $sp, 36, 14, $noreg :: (load 4 from %fixed-stack.2, align 8), (load 4 from %fixed-stack.1)
+    $r5, $r4 = t2LDRDi8 $sp, 20, 14, $noreg :: (load 4 from %fixed-stack.6, align 8), (load 4 from %fixed-stack.5)
+    renamable $q0 = MVE_VDUP32 killed renamable $r6, 0, $noreg, undef renamable $q0
+    renamable $q1 = MVE_VDUP32 killed renamable $r7, 0, $noreg, undef renamable $q1
+
+  bb.2.for.body:
+    successors: %bb.2(0x7c000000), %bb.3(0x04000000)
+    liveins: $lr, $q0, $q1, $r0, $r1, $r2, $r3, $r4, $r5, $r12
+
+    renamable $vpr = MVE_VCTP32 renamable $r12, 0, $noreg
+    MVE_VPST 8, implicit $vpr
+    renamable $r1, renamable $q2 = MVE_VLDRWU32_post killed renamable $r1, 4, 1, renamable $vpr :: (load 16 from %ir.input_2_cast, align 4)
+    MVE_VPST 8, implicit $vpr
+    renamable $r0, renamable $q3 = MVE_VLDRWU32_post killed renamable $r0, 4, 1, renamable $vpr :: (load 16 from %ir.input_1_cast, align 4)
+    renamable $q2 = MVE_VADD_qr_i32 killed renamable $q2, renamable $r3, 0, $noreg, undef renamable $q2
+    renamable $q3 = MVE_VADD_qr_i32 killed renamable $q3, renamable $r2, 0, $noreg, undef renamable $q3
+    renamable $r12 = t2SUBri killed renamable $r12, 4, 14, $noreg, $noreg
+    renamable $q2 = MVE_VMULi32 killed renamable $q3, killed renamable $q2, 0, $noreg, undef renamable $q2
+    renamable $q2 = MVE_VADD_qr_i32 killed renamable $q2, renamable $r4, 0, $noreg, undef renamable $q2
+    MVE_VPST 2, implicit $vpr
+    renamable $q2 = MVE_VMAXu32 killed renamable $q2, renamable $q1, 1, renamable $vpr, undef renamable $q2
+    renamable $q2 = MVE_VMINu32 killed renamable $q2, renamable $q0, 1, renamable $vpr, undef renamable $q2
+    renamable $r5 = MVE_VSTRWU32_post killed renamable $q2, killed renamable $r5, 4, 1, killed renamable $vpr :: (store 16 into %ir.output_cast, align 4)
+    renamable $lr = t2LoopDec killed renamable $lr, 1
+    t2LoopEnd renamable $lr, %bb.2, implicit-def dead $cpsr
+    tB %bb.3, 14, $noreg
+
+  bb.3.for.cond.cleanup:
+    $r0, dead $cpsr = tMOVi8 0, 14, $noreg
+    tPOP_RET 14, $noreg, def $r4, def $r5, def $r6, def $r7, def $pc, implicit killed $r0
+
+...