# RUN: llc -mtriple=thumbv8.1m.main -mattr=+mve -run-pass=arm-low-overhead-loops %s -o - | FileCheck %s

--- |
  target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
  target triple = "thumbv8.1m.main-arm-none-eabi"

  %struct.arm_2d_size_t = type { i16, i16 }

  define void @__arm_2d_impl_rgb16_cl_key_1x1_paving_x_mirror_xx(ptr noalias %pSource, i16 signext %iSourceStride, ptr noalias %pTarget, i16 signext %iTargetStride, ptr noalias nocapture readonly %ptSrcCopySize, ptr noalias nocapture readonly %ptDstCopySize, i16 zeroext %Colour) local_unnamed_addr #0 !dbg !12 {
  entry:
    call void @llvm.dbg.value(metadata ptr %pSource, metadata !33, metadata !DIExpression()), !dbg !62
    call void @llvm.dbg.value(metadata i16 %iSourceStride, metadata !34, metadata !DIExpression()), !dbg !62
    call void @llvm.dbg.value(metadata ptr %pTarget, metadata !35, metadata !DIExpression()), !dbg !62
    call void @llvm.dbg.value(metadata i16 %iTargetStride, metadata !36, metadata !DIExpression()), !dbg !62
    call void @llvm.dbg.value(metadata ptr %ptSrcCopySize, metadata !37, metadata !DIExpression()), !dbg !62
    call void @llvm.dbg.value(metadata ptr %ptDstCopySize, metadata !38, metadata !DIExpression()), !dbg !62
    call void @llvm.dbg.value(metadata i16 %Colour, metadata !39, metadata !DIExpression()), !dbg !62
    call void @llvm.dbg.value(metadata i16 undef, metadata !40, metadata !DIExpression(DW_OP_LLVM_convert, 16, DW_ATE_signed, DW_OP_LLVM_convert, 32, DW_ATE_signed, DW_OP_stack_value)), !dbg !62
    call void @llvm.dbg.value(metadata i32 0, metadata !43, metadata !DIExpression()), !dbg !63
    %iHeight = getelementptr inbounds %struct.arm_2d_size_t, ptr %ptDstCopySize, i32 0, i32 1
    %0 = load i16, ptr %iHeight, align 2, !tbaa !64
    %conv1 = sext i16 %0 to i32
    %conv8 = sext i16 %iSourceStride to i32
    %conv10 = sext i16 %iTargetStride to i32
    call void @llvm.dbg.value(metadata ptr %pSource, metadata !33, metadata !DIExpression()), !dbg !62
    call void @llvm.dbg.value(metadata ptr %pTarget, metadata !35, metadata !DIExpression()), !dbg !62
    call void @llvm.dbg.value(metadata i32 0, metadata !43, metadata !DIExpression()), !dbg !63
    %cmp34 = icmp sgt i16 %0, 0, !dbg !69
    br i1 %cmp34, label %for.body.lr.ph, label %for.cond.cleanup, !dbg !70

  for.body.lr.ph:                                   ; preds = %entry
    %iWidth2 = bitcast ptr %ptSrcCopySize to ptr, !dbg !71
    %1 = load i16, ptr %iWidth2, align 2, !dbg !71, !tbaa !72
    call void @llvm.dbg.value(metadata i16 %1, metadata !40, metadata !DIExpression(DW_OP_LLVM_convert, 16, DW_ATE_signed, DW_OP_LLVM_convert, 32, DW_ATE_signed, DW_OP_stack_value)), !dbg !62
    %conv = sext i16 %1 to i32, !dbg !73
    call void @llvm.dbg.value(metadata i32 %conv, metadata !40, metadata !DIExpression()), !dbg !62
    %sub = add nsw i32 %conv, -1
    %iWidth33 = bitcast ptr %ptDstCopySize to ptr
    %2 = load i16, ptr %iWidth33, align 2, !tbaa !72
    %conv4 = sext i16 %2 to i32
    %3 = tail call { <8 x i16>, i32 } @llvm.arm.mve.vddup.v8i16(i32 %sub, i32 1), !dbg !62
    %4 = add i32 %conv4, 7, !dbg !70
    %smin = call i32 @llvm.smin.i32(i32 %conv4, i32 8), !dbg !70
    %5 = sub i32 %4, %smin, !dbg !70
    %6 = lshr i32 %5, 3, !dbg !70
    %7 = add nuw nsw i32 %6, 1, !dbg !70
    br label %for.body, !dbg !70

  for.cond.cleanup:                                 ; preds = %do.end, %entry
    ret void, !dbg !74

  for.body:                                         ; preds = %do.end, %for.body.lr.ph
    %pSource.addr.037 = phi ptr [ %pSource, %for.body.lr.ph ], [ %add.ptr9, %do.end ]
    %pTarget.addr.036 = phi ptr [ %pTarget, %for.body.lr.ph ], [ %add.ptr11, %do.end ]
    %y.035 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %do.end ]
    call void @llvm.dbg.value(metadata ptr %pSource.addr.037, metadata !33, metadata !DIExpression()), !dbg !62
    call void @llvm.dbg.value(metadata ptr %pTarget.addr.036, metadata !35, metadata !DIExpression()), !dbg !62
    call void @llvm.dbg.value(metadata i32 %y.035, metadata !43, metadata !DIExpression()), !dbg !63
    call void @llvm.dbg.value(metadata ptr %pTarget.addr.036, metadata !46, metadata !DIExpression()), !dbg !75
    call void @llvm.dbg.value(metadata ptr %pSource.addr.037, metadata !49, metadata !DIExpression()), !dbg !75
    call void @llvm.dbg.value(metadata i32 %conv4, metadata !50, metadata !DIExpression()), !dbg !75
    call void @llvm.dbg.value(metadata i32 undef, metadata !51, metadata !DIExpression()), !dbg !75
    call void @llvm.dbg.value(metadata <8 x i16> undef, metadata !52, metadata !DIExpression()), !dbg !75
    %8 = call i32 @llvm.start.loop.iterations.i32(i32 %7), !dbg !76
    br label %do.body, !dbg !76

  do.body:                                          ; preds = %do.body, %for.body
    %pDst.0 = phi ptr [ %pTarget.addr.036, %for.body ], [ %add.ptr, %do.body ], !dbg !75
    %dstWidth.0 = phi i32 [ %conv4, %for.body ], [ %sub5, %do.body ], !dbg !75
    %.pn = phi { <8 x i16>, i32 } [ %3, %for.body ], [ %12, %do.body ]
    %9 = phi i32 [ %8, %for.body ], [ %17, %do.body ], !dbg !75
    %pDst.01 = bitcast ptr %pDst.0 to ptr, !dbg !75
    %offset.0 = extractvalue { <8 x i16>, i32 } %.pn, 0, !dbg !75
    %curOffsetIdx.0 = extractvalue { <8 x i16>, i32 } %.pn, 1, !dbg !75
    call void @llvm.dbg.value(metadata <8 x i16> %offset.0, metadata !52, metadata !DIExpression()), !dbg !75
    call void @llvm.dbg.value(metadata i32 %curOffsetIdx.0, metadata !51, metadata !DIExpression()), !dbg !75
    call void @llvm.dbg.value(metadata i32 %dstWidth.0, metadata !50, metadata !DIExpression()), !dbg !75
    call void @llvm.dbg.value(metadata ptr %pDst.0, metadata !46, metadata !DIExpression()), !dbg !75
    %10 = tail call <8 x i1> @llvm.arm.mve.vctp16(i32 %dstWidth.0), !dbg !77
    call void @llvm.dbg.value(metadata i32 undef, metadata !58, metadata !DIExpression()), !dbg !78
    %11 = tail call <8 x i16> @llvm.arm.mve.vldr.gather.offset.predicated.v8i16.p0.v8i16.v8i1(ptr %pSource.addr.037, <8 x i16> %offset.0, i32 16, i32 1, i32 1, <8 x i1> %10), !dbg !79
    call void @llvm.dbg.value(metadata <8 x i16> %11, metadata !61, metadata !DIExpression()), !dbg !78
    %12 = tail call { <8 x i16>, i32 } @llvm.arm.mve.vddup.predicated.v8i16.v8i1(<8 x i16> undef, i32 %curOffsetIdx.0, i32 1, <8 x i1> %10), !dbg !80
    call void @llvm.dbg.value(metadata i32 undef, metadata !51, metadata !DIExpression()), !dbg !75
    call void @llvm.dbg.value(metadata <8 x i16> undef, metadata !52, metadata !DIExpression()), !dbg !75
    %13 = insertelement <8 x i16> poison, i16 %Colour, i32 0
    %14 = shufflevector <8 x i16> %13, <8 x i16> poison, <8 x i32> zeroinitializer
    %15 = icmp ne <8 x i16> %11, %14, !dbg !81
    %16 = and <8 x i1> %15, %10, !dbg !81
    call void @llvm.dbg.value(metadata i32 undef, metadata !58, metadata !DIExpression(DW_OP_LLVM_convert, 32, DW_ATE_unsigned, DW_OP_LLVM_convert, 16, DW_ATE_unsigned, DW_OP_stack_value)), !dbg !78
    tail call void @llvm.masked.store.v8i16.p0(<8 x i16> %11, ptr %pDst.01, i32 2, <8 x i1> %16), !dbg !82
    %add.ptr = getelementptr inbounds i16, ptr %pDst.0, i32 8, !dbg !83
    call void @llvm.dbg.value(metadata ptr %add.ptr, metadata !46, metadata !DIExpression()), !dbg !75
    %sub5 = add nsw i32 %dstWidth.0, -8, !dbg !84
    call void @llvm.dbg.value(metadata i32 %sub5, metadata !50, metadata !DIExpression()), !dbg !75
    %17 = call i32 @llvm.loop.decrement.reg.i32(i32 %9, i32 1), !dbg !85
    %18 = icmp ne i32 %17, 0, !dbg !85
    br i1 %18, label %do.body, label %do.end, !dbg !85, !llvm.loop !86

  do.end:                                           ; preds = %do.body
    %add.ptr9 = getelementptr inbounds i16, ptr %pSource.addr.037, i32 %conv8, !dbg !89
    call void @llvm.dbg.value(metadata ptr %add.ptr9, metadata !33, metadata !DIExpression()), !dbg !62
    %add.ptr11 = getelementptr inbounds i16, ptr %pTarget.addr.036, i32 %conv10, !dbg !90
    call void @llvm.dbg.value(metadata ptr %add.ptr11, metadata !35, metadata !DIExpression()), !dbg !62
    %inc = add nuw nsw i32 %y.035, 1, !dbg !91
    call void @llvm.dbg.value(metadata i32 %inc, metadata !43, metadata !DIExpression()), !dbg !63
    %exitcond.not = icmp eq i32 %inc, %conv1, !dbg !69
    br i1 %exitcond.not, label %for.cond.cleanup, label %for.body, !dbg !70, !llvm.loop !92
  }

  declare { <8 x i16>, i32 } @llvm.arm.mve.vddup.v8i16(i32, i32)
  declare <8 x i1> @llvm.arm.mve.vctp16(i32)
  declare <8 x i16> @llvm.arm.mve.vldr.gather.offset.predicated.v8i16.p0.v8i16.v8i1(ptr, <8 x i16>, i32, i32, i32, <8 x i1>)
  declare { <8 x i16>, i32 } @llvm.arm.mve.vddup.predicated.v8i16.v8i1(<8 x i16>, i32, i32, <8 x i1>)
  declare void @llvm.masked.store.v8i16.p0(<8 x i16>, ptr, i32 immarg, <8 x i1>)
  declare void @llvm.dbg.value(metadata, metadata, metadata)
  declare i32 @llvm.smin.i32(i32, i32)
  declare i32 @llvm.start.loop.iterations.i32(i32)
  declare i32 @llvm.loop.decrement.reg.i32(i32, i32)

  attributes #0 = { nofree nounwind "frame-pointer"="all" "min-legal-vector-width"="0" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cortex-m55" "target-features"="+armv8.1-m.main,+cdecp0,+dsp,+fp-armv8d16,+fp-armv8d16sp,+fp16,+fp64,+fullfp16,+hwdiv,+lob,+mve,+mve.fp,+ras,+strict-align,+thumb-mode,+vfp2,+vfp2sp,+vfp3d16,+vfp3d16sp,+vfp4d16,+vfp4d16sp,-aes,-bf16,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-fp-armv8,-fp-armv8sp,-fp16fml,-hwdiv-arm,-i8mm,-neon,-sb,-sha2,-vfp3,-vfp3sp,-vfp4,-vfp4sp" "unsafe-fp-math"="true" }

  !0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, retainedTypes: !2, splitDebugInlining: false, nameTableKind: None)
  !1 = !DIFile(filename: "tmp.c", directory: "")
  !2 = !{!3}
  !3 = !DIDerivedType(tag: DW_TAG_typedef, name: "int32_t", file: !4, line: 58, baseType: !5)
  !4 = !DIFile(filename: "stdint.h", directory: "")
  !5 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
  !6 = !{i32 7, !"Dwarf Version", i32 4}
  !7 = !{i32 2, !"Debug Info Version", i32 3}
  !8 = !{i32 1, !"wchar_size", i32 4}
  !9 = !{i32 1, !"min_enum_size", i32 4}
  !10 = !{i32 7, !"frame-pointer", i32 2}
  !11 = !{!"clang"}
  !12 = distinct !DISubprogram(name: "__arm_2d_impl_rgb16_cl_key_1x1_paving_x_mirror_xx", scope: !1, file: !1, line: 61, type: !13, scopeLine: 71, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !32)
  !13 = !DISubroutineType(types: !14)
  !14 = !{null, !15, !20, !22, !20, !24, !24, !18}
  !15 = !DIDerivedType(tag: DW_TAG_restrict_type, baseType: !16)
  !16 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !17, size: 32)
  !17 = !DIDerivedType(tag: DW_TAG_const_type, baseType: !18)
  !18 = !DIDerivedType(tag: DW_TAG_typedef, name: "uint16_t", file: !4, line: 63, baseType: !19)
  !19 = !DIBasicType(name: "unsigned short", size: 16, encoding: DW_ATE_unsigned)
  !20 = !DIDerivedType(tag: DW_TAG_typedef, name: "int16_t", file: !4, line: 57, baseType: !21)
  !21 = !DIBasicType(name: "short", size: 16, encoding: DW_ATE_signed)
  !22 = !DIDerivedType(tag: DW_TAG_restrict_type, baseType: !23)
  !23 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !18, size: 32)
  !24 = !DIDerivedType(tag: DW_TAG_restrict_type, baseType: !25)
  !25 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !26, size: 32)
  !26 = !DIDerivedType(tag: DW_TAG_const_type, baseType: !27)
  !27 = !DIDerivedType(tag: DW_TAG_typedef, name: "arm_2d_size_t", file: !1, line: 8, baseType: !28)
  !28 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "arm_2d_size_t", file: !1, line: 5, size: 32, elements: !29)
  !29 = !{!30, !31}
  !30 = !DIDerivedType(tag: DW_TAG_member, name: "iWidth", scope: !28, file: !1, line: 6, baseType: !20, size: 16)
  !31 = !DIDerivedType(tag: DW_TAG_member, name: "iHeight", scope: !28, file: !1, line: 7, baseType: !20, size: 16, offset: 16)
  !32 = !{!33, !34, !35, !36, !37, !38, !39, !40, !43, !46, !49, !50, !51, !52, !58, !61}
  !33 = !DILocalVariable(name: "pSource", arg: 1, scope: !12, file: !1, line: 62, type: !15)
  !34 = !DILocalVariable(name: "iSourceStride", arg: 2, scope: !12, file: !1, line: 63, type: !20)
  !35 = !DILocalVariable(name: "pTarget", arg: 3, scope: !12, file: !1, line: 64, type: !22)
  !36 = !DILocalVariable(name: "iTargetStride", arg: 4, scope: !12, file: !1, line: 65, type: !20)
  !37 = !DILocalVariable(name: "ptSrcCopySize", arg: 5, scope: !12, file: !1, line: 67, type: !24)
  !38 = !DILocalVariable(name: "ptDstCopySize", arg: 6, scope: !12, file: !1, line: 69, type: !24)
  !39 = !DILocalVariable(name: "Colour", arg: 7, scope: !12, file: !1, line: 70, type: !18)
  !40 = !DILocalVariable(name: "srcWidth", scope: !12, file: !1, line: 72, type: !41)
  !41 = !DIDerivedType(tag: DW_TAG_typedef, name: "uint32_t", file: !4, line: 64, baseType: !42)
  !42 = !DIBasicType(name: "unsigned int", size: 32, encoding: DW_ATE_unsigned)
  !43 = !DILocalVariable(name: "y", scope: !44, file: !1, line: 74, type: !45)
  !44 = distinct !DILexicalBlock(scope: !12, file: !1, line: 74, column: 5)
  !45 = !DIDerivedType(tag: DW_TAG_typedef, name: "int_fast16_t", file: !4, line: 86, baseType: !5)
  !46 = !DILocalVariable(name: "pDst", scope: !47, file: !1, line: 75, type: !23)
  !47 = distinct !DILexicalBlock(scope: !48, file: !1, line: 74, column: 63)
  !48 = distinct !DILexicalBlock(scope: !44, file: !1, line: 74, column: 5)
  !49 = !DILocalVariable(name: "pSrc", scope: !47, file: !1, line: 76, type: !15)
  !50 = !DILocalVariable(name: "dstWidth", scope: !47, file: !1, line: 77, type: !41)
  !51 = !DILocalVariable(name: "curOffsetIdx", scope: !47, file: !1, line: 78, type: !41)
  !52 = !DILocalVariable(name: "offset", scope: !47, file: !1, line: 79, type: !53)
  !53 = !DIDerivedType(tag: DW_TAG_typedef, name: "uint16x8_t", file: !54, line: 37, baseType: !55)
  !54 = !DIFile(filename: "lib/clang/14.0.0/include/arm_mve.h", directory: "")
  !55 = !DICompositeType(tag: DW_TAG_array_type, baseType: !18, size: 128, flags: DIFlagVector, elements: !56)
  !56 = !{!57}
  !57 = !DISubrange(count: 8)
  !58 = !DILocalVariable(name: "p", scope: !59, file: !1, line: 81, type: !60)
  !59 = distinct !DILexicalBlock(scope: !47, file: !1, line: 80, column: 12)
  !60 = !DIDerivedType(tag: DW_TAG_typedef, name: "mve_pred16_t", file: !54, line: 24, baseType: !18)
  !61 = !DILocalVariable(name: "in", scope: !59, file: !1, line: 82, type: !53)
  !62 = !DILocation(line: 0, scope: !12)
  !63 = !DILocation(line: 0, scope: !44)
  !64 = !{!65, !66, i64 2}
  !65 = !{!"arm_2d_size_t", !66, i64 0, !66, i64 2}
  !66 = !{!"short", !67, i64 0}
  !67 = !{!"omnipotent char", !68, i64 0}
  !68 = !{!"Simple C/C++ TBAA"}
  !69 = !DILocation(line: 74, column: 32, scope: !48)
  !70 = !DILocation(line: 74, column: 5, scope: !44)
  !71 = !DILocation(line: 72, column: 47, scope: !12)
  !72 = !{!65, !66, i64 0}
  !73 = !DILocation(line: 72, column: 32, scope: !12)
  !74 = !DILocation(line: 93, column: 1, scope: !12)
  !75 = !DILocation(line: 0, scope: !47)
  !76 = !DILocation(line: 80, column: 9, scope: !47)
  !77 = !DILocation(line: 81, column: 33, scope: !59)
  !78 = !DILocation(line: 0, scope: !59)
  !79 = !DILocation(line: 83, column: 18, scope: !59)
  !80 = !DILocation(line: 84, column: 22, scope: !59)
  !81 = !DILocation(line: 85, column: 17, scope: !59)
  !82 = !DILocation(line: 86, column: 13, scope: !59)
  !83 = !DILocation(line: 87, column: 18, scope: !59)
  !84 = !DILocation(line: 88, column: 22, scope: !59)
  !85 = !DILocation(line: 89, column: 9, scope: !59)
  !86 = distinct !{!86, !76, !87, !88}
  !87 = !DILocation(line: 89, column: 40, scope: !47)
  !88 = !{!"llvm.loop.mustprogress"}
  !89 = !DILocation(line: 90, column: 17, scope: !47)
  !90 = !DILocation(line: 91, column: 17, scope: !47)
  !91 = !DILocation(line: 74, column: 59, scope: !48)
  !92 = distinct !{!92, !70, !93, !88}
  !93 = !DILocation(line: 92, column: 5, scope: !44)

...
---
name:            __arm_2d_impl_rgb16_cl_key_1x1_paving_x_mirror_xx
alignment:       2
tracksRegLiveness: true
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:       40
  offsetAdjustment: -32
  maxAlignment:    4
  adjustsStack:    false
  hasCalls:        false
  stackProtector:  ''
  maxCallFrameSize: 0
  cvBytesOfCalleeSavedRegisters: 0
  hasOpaqueSPAdjustment: false
  hasVAStart:      false
  hasMustTailInVarArgFunc: false
  hasTailCall:     false
  localFrameSize:  0
  savePoint:       ''
  restorePoint:    ''
fixedStack:
  - { id: 0, 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: 1, 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: 2, 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: -40, size: 4, alignment: 4,
      stack-id: default, callee-saved-register: '', callee-saved-restored: true,
      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
  - { id: 1, name: '', type: spill-slot, offset: -4, size: 4, alignment: 4,
      stack-id: default, callee-saved-register: '$lr', callee-saved-restored: false,
      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
  - { id: 2, name: '', type: spill-slot, offset: -8, size: 4, alignment: 4,
      stack-id: default, callee-saved-register: '$r7', callee-saved-restored: true,
      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
  - { id: 3, name: '', type: spill-slot, offset: -12, size: 4, alignment: 4,
      stack-id: default, callee-saved-register: '$r6', callee-saved-restored: true,
      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
  - { id: 4, name: '', type: spill-slot, offset: -16, size: 4, alignment: 4,
      stack-id: default, callee-saved-register: '$r5', callee-saved-restored: true,
      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
  - { id: 5, name: '', type: spill-slot, offset: -20, size: 4, alignment: 4,
      stack-id: default, callee-saved-register: '$r4', callee-saved-restored: true,
      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
  - { id: 6, name: '', type: spill-slot, offset: -24, size: 4, alignment: 4,
      stack-id: default, callee-saved-register: '$r11', callee-saved-restored: true,
      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
  - { id: 7, name: '', type: spill-slot, offset: -28, size: 4, alignment: 4,
      stack-id: default, callee-saved-register: '$r10', callee-saved-restored: true,
      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
  - { id: 8, name: '', type: spill-slot, offset: -32, size: 4, alignment: 4,
      stack-id: default, callee-saved-register: '$r9', callee-saved-restored: true,
      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
  - { id: 9, name: '', type: spill-slot, offset: -36, size: 4, alignment: 4,
      stack-id: default, callee-saved-register: '$r8', callee-saved-restored: true,
      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
body:             |
  ; CHECK-LABEL: name: __arm_2d_impl_rgb16_cl_key_1x1_paving_x_mirror_xx
  ; CHECK:   early-clobber renamable $q2 = MVE_VLDRHU16_rq renamable $r0, killed renamable $q1, 0, $noreg, renamable $lr, debug-location !DILocation(line: 83, column: 18, scope: {{.*}}) :: (load (s128), align 1)
  ; CHECK:   renamable $q1, renamable $r8 = MVE_VDDUPu16 killed renamable $r8, 1, 0, $noreg, renamable $lr, undef renamable $q1, debug-location !DILocation(line: 84, column: 22, scope: {{.*}})
  ; CHECK:   MVE_VPTv8i16r 8, renamable $q2, renamable $r11, 1, implicit-def $vpr, debug-location !DILocation(line: 85, column: 17, scope: {{.*}})
  ; CHECK:   renamable $r6 = MVE_VSTRHU16_post killed renamable $q2, killed renamable $r6, 16, 1, killed renamable $vpr, renamable $lr, debug-location !DILocation(line: 86, column: 13, scope: {{.*}}) :: (store (s128) into %ir.pDst.01, align 2)
  ; CHECK:   $lr = MVE_LETP killed renamable $lr, %bb.3, debug-location !DILocation(line: 89, column: 9, scope: {{.*}})
  bb.0.entry:
    successors: %bb.1(0x50000000), %bb.5(0x30000000)
    liveins: $r0, $r1, $r2, $r3, $r4, $r5, $r6, $lr, $r8, $r9, $r10, $r11

    DBG_VALUE $r0, $noreg, !33, !DIExpression(), debug-location !62
    DBG_VALUE $r1, $noreg, !34, !DIExpression(), debug-location !62
    DBG_VALUE $r2, $noreg, !35, !DIExpression(), debug-location !62
    DBG_VALUE $r3, $noreg, !36, !DIExpression(), debug-location !62
    frame-setup tPUSH 14 /* CC::al */, $noreg, killed $r4, killed $r5, killed $r6, $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
    $r7 = frame-setup tADDrSPi $sp, 3, 14 /* CC::al */, $noreg
    frame-setup CFI_INSTRUCTION def_cfa $r7, 8
    $sp = frame-setup t2STMDB_UPD $sp, 14 /* CC::al */, $noreg, killed $r8, killed $r9, killed $r10, killed $r11
    frame-setup CFI_INSTRUCTION offset $r11, -24
    frame-setup CFI_INSTRUCTION offset $r10, -28
    frame-setup CFI_INSTRUCTION offset $r9, -32
    frame-setup CFI_INSTRUCTION offset $r8, -36
    DBG_VALUE $r7, 0, !37, !DIExpression(DW_OP_plus_uconst, 8), debug-location !62
    DBG_VALUE $r7, 0, !38, !DIExpression(DW_OP_plus_uconst, 12), debug-location !62
    early-clobber $sp = t2STR_PRE killed $r3, $sp, -4, 14 /* CC::al */, $noreg :: (store (s32) into %stack.0)
    DBG_VALUE $sp, 0, !36, !DIExpression(), debug-location !62
    DBG_VALUE $sp, 0, !36, !DIExpression(), debug-location !62
    renamable $r6 = tLDRi $r7, 3, 14 /* CC::al */, $noreg :: (load (s32) from %fixed-stack.1)
    DBG_VALUE 0, $noreg, !43, !DIExpression(), debug-location !63
    DBG_VALUE $noreg, $noreg, !40, !DIExpression(DW_OP_LLVM_convert, 16, DW_ATE_signed, DW_OP_LLVM_convert, 32, DW_ATE_signed, DW_OP_stack_value), debug-location !62
    renamable $r3 = t2LDRSHi12 renamable $r6, 2, 14 /* CC::al */, $noreg :: (load (s16) from %ir.iHeight, !tbaa !64)
    DBG_VALUE $noreg, $noreg, !35, !DIExpression(), debug-location !62
    DBG_VALUE $noreg, $noreg, !34, !DIExpression(), debug-location !62
    DBG_VALUE $noreg, $noreg, !33, !DIExpression(), debug-location !62
    tCMPi8 renamable $r3, 1, 14 /* CC::al */, $noreg, implicit-def $cpsr, debug-location !70
    DBG_VALUE $noreg, $noreg, !39, !DIExpression(), debug-location !62
    DBG_VALUE $r0, $noreg, !33, !DIExpression(), debug-location !62
    tBcc %bb.5, 11 /* CC::lt */, killed $cpsr, debug-location !70

  bb.1.for.body.lr.ph:
    successors: %bb.2(0x80000000)
    liveins: $r0, $r1, $r2, $r3, $r6

    DBG_VALUE 0, $noreg, !43, !DIExpression(), debug-location !63
    DBG_VALUE $sp, 0, !36, !DIExpression(), debug-location !62
    DBG_VALUE $r7, 0, !38, !DIExpression(DW_OP_plus_uconst, 12), debug-location !62
    DBG_VALUE $r7, 0, !37, !DIExpression(DW_OP_plus_uconst, 8), debug-location !62
    DBG_VALUE $r0, $noreg, !33, !DIExpression(), debug-location !62
    renamable $r10 = t2LDRSHi12 killed renamable $r6, 0, 14 /* CC::al */, $noreg :: (load (s16) from %ir.iWidth33, !tbaa !72)
    renamable $r5 = tLDRi $r7, 2, 14 /* CC::al */, $noreg :: (load (s32) from %fixed-stack.2, align 8)
    $r6 = tMOVr $r10, 14 /* CC::al */, $noreg, debug-location !70
    renamable $r5 = t2LDRSHi12 killed renamable $r5, 0, 14 /* CC::al */, $noreg, debug-location !71 :: (load (s16) from %ir.iWidth2, !tbaa !72)
    DBG_VALUE $r5, $noreg, !40, !DIExpression(), debug-location !62
    t2CMPri renamable $r10, 8, 14 /* CC::al */, $noreg, implicit-def $cpsr, debug-location !70
    t2IT 10, 8, implicit-def $itstate, debug-location !70
    renamable $r6 = tMOVi8 $noreg, 8, 10 /* CC::ge */, killed $cpsr, implicit killed renamable $r6, implicit killed $itstate, debug-location !70
    renamable $r6 = t2SUBrr renamable $r10, killed renamable $r6, 14 /* CC::al */, $noreg, $noreg, debug-location !70
    renamable $r11 = t2LDRi12 $r7, 16, 14 /* CC::al */, $noreg :: (load (s32) from %fixed-stack.0, align 8)
    DBG_VALUE $r11, $noreg, !39, !DIExpression(), debug-location !62
    renamable $r12 = nsw t2SUBri killed renamable $r5, 1, 14 /* CC::al */, $noreg, $noreg
    renamable $r6, dead $cpsr = tADDi8 killed renamable $r6, 7, 14 /* CC::al */, $noreg, debug-location !70
    renamable $r5, dead $cpsr = tMOVi8 1, 14 /* CC::al */, $noreg
    renamable $q0, renamable $r12 = MVE_VDDUPu16 killed renamable $r12, 1, 0, $noreg, $noreg, undef renamable $q0, debug-location !62
    renamable $r9 = nuw nsw t2ADDrs killed renamable $r5, killed renamable $r6, 27, 14 /* CC::al */, $noreg, $noreg, debug-location !70
    renamable $r4, dead $cpsr = tMOVi8 0, 14 /* CC::al */, $noreg

  bb.2.for.body (align 4):
    successors: %bb.3(0x80000000)
    liveins: $q0, $r0, $r1, $r2, $r3, $r4, $r9, $r10, $r11, $r12

    DBG_VALUE $r11, $noreg, !39, !DIExpression(), debug-location !62
    DBG_VALUE $sp, 0, !36, !DIExpression(), debug-location !62
    DBG_VALUE $r7, 0, !38, !DIExpression(DW_OP_plus_uconst, 12), debug-location !62
    DBG_VALUE $r7, 0, !37, !DIExpression(DW_OP_plus_uconst, 8), debug-location !62
    DBG_VALUE $r0, $noreg, !33, !DIExpression(), debug-location !62
    DBG_VALUE $r0, $noreg, !33, !DIExpression(), debug-location !62
    DBG_VALUE $r2, $noreg, !35, !DIExpression(), debug-location !62
    DBG_VALUE $r4, $noreg, !43, !DIExpression(), debug-location !63
    DBG_VALUE $r2, $noreg, !46, !DIExpression(), debug-location !75
    DBG_VALUE $r0, $noreg, !49, !DIExpression(), debug-location !75
    DBG_VALUE $r10, $noreg, !50, !DIExpression(), debug-location !75
    DBG_VALUE $noreg, $noreg, !51, !DIExpression(), debug-location !75
    DBG_VALUE $noreg, $noreg, !52, !DIExpression(), debug-location !75
    $r6 = tMOVr $r2, 14 /* CC::al */, $noreg
    $r5 = tMOVr $r10, 14 /* CC::al */, $noreg
    $q1 = MVE_VORR $q0, $q0, 0, $noreg, $noreg, undef $q1
    $r8 = tMOVr $r12, 14 /* CC::al */, $noreg
    renamable $lr = t2DoLoopStartTP renamable $r9, renamable $r10, debug-location !76

  bb.3.do.body (align 4):
    successors: %bb.3(0x7c000000), %bb.4(0x04000000)
    liveins: $lr, $q0, $q1, $r0, $r1, $r2, $r3, $r4, $r5, $r6, $r8, $r9, $r10, $r11, $r12

    DBG_VALUE $r0, $noreg, !49, !DIExpression(), debug-location !75
    DBG_VALUE $r4, $noreg, !43, !DIExpression(), debug-location !63
    DBG_VALUE $r11, $noreg, !39, !DIExpression(), debug-location !62
    DBG_VALUE $sp, 0, !36, !DIExpression(), debug-location !62
    DBG_VALUE $r7, 0, !38, !DIExpression(DW_OP_plus_uconst, 12), debug-location !62
    DBG_VALUE $r7, 0, !37, !DIExpression(DW_OP_plus_uconst, 8), debug-location !62
    DBG_VALUE $r2, $noreg, !35, !DIExpression(), debug-location !62
    DBG_VALUE $r0, $noreg, !33, !DIExpression(), debug-location !62
    DBG_VALUE $r6, $noreg, !46, !DIExpression(), debug-location !75
    DBG_VALUE $r5, $noreg, !50, !DIExpression(), debug-location !75
    DBG_VALUE $r8, $noreg, !51, !DIExpression(), debug-location !75
    DBG_VALUE $q1, $noreg, !52, !DIExpression(), debug-location !75
    renamable $vpr = MVE_VCTP16 renamable $r5, 0, $noreg, $noreg, debug-location !77
    DBG_VALUE $noreg, $noreg, !58, !DIExpression(), debug-location !78
    MVE_VPST 1, implicit $vpr, debug-location !79
    early-clobber renamable $q2 = MVE_VLDRHU16_rq renamable $r0, killed renamable $q1, 1, renamable $vpr, renamable $lr, debug-location !79 :: (load (s128), align 1)
    DBG_VALUE $q2, $noreg, !61, !DIExpression(), debug-location !78
    renamable $q1, renamable $r8 = MVE_VDDUPu16 killed renamable $r8, 1, 1, renamable $vpr, renamable $lr, undef renamable $q1, debug-location !80
    DBG_VALUE $noreg, $noreg, !52, !DIExpression(), debug-location !75
    DBG_VALUE $noreg, $noreg, !51, !DIExpression(), debug-location !75
    renamable $vpr = MVE_VCMPi16r renamable $q2, renamable $r11, 1, 1, killed renamable $vpr, renamable $lr, debug-location !81
    DBG_VALUE $noreg, $noreg, !58, !DIExpression(DW_OP_LLVM_convert, 32, DW_ATE_unsigned, DW_OP_LLVM_convert, 16, DW_ATE_unsigned, DW_OP_stack_value), debug-location !78
    renamable $r6 = MVE_VSTRHU16_post killed renamable $q2, killed renamable $r6, 16, 1, killed renamable $vpr, renamable $lr, debug-location !82 :: (store (s128) into %ir.pDst.01, align 2)
    DBG_VALUE $r6, $noreg, !46, !DIExpression(), debug-location !75
    renamable $r5, dead $cpsr = nsw tSUBi8 killed renamable $r5, 8, 14 /* CC::al */, $noreg, debug-location !84
    DBG_VALUE $r5, $noreg, !50, !DIExpression(), debug-location !75
    renamable $lr = t2LoopEndDec killed renamable $lr, %bb.3, implicit-def dead $cpsr, debug-location !85
    tB %bb.4, 14 /* CC::al */, $noreg, debug-location !85

  bb.4.do.end:
    successors: %bb.5(0x04000000), %bb.2(0x7c000000)
    liveins: $q0, $r0, $r1, $r2, $r3, $r4, $r9, $r10, $r11, $r12

    DBG_VALUE $r5, $noreg, !50, !DIExpression(), debug-location !75
    DBG_VALUE $r6, $noreg, !46, !DIExpression(), debug-location !75
    DBG_VALUE $r0, $noreg, !49, !DIExpression(), debug-location !75
    DBG_VALUE $r4, $noreg, !43, !DIExpression(), debug-location !63
    DBG_VALUE $r11, $noreg, !39, !DIExpression(), debug-location !62
    DBG_VALUE $sp, 0, !36, !DIExpression(), debug-location !62
    DBG_VALUE $r7, 0, !38, !DIExpression(DW_OP_plus_uconst, 12), debug-location !62
    DBG_VALUE $r7, 0, !37, !DIExpression(DW_OP_plus_uconst, 8), debug-location !62
    DBG_VALUE $r2, $noreg, !35, !DIExpression(), debug-location !62
    DBG_VALUE $r0, $noreg, !33, !DIExpression(), debug-location !62
    renamable $r6 = tLDRspi $sp, 0, 14 /* CC::al */, $noreg, debug-location !90 :: (load (s32) from %stack.0)
    renamable $r4, dead $cpsr = nuw nsw tADDi8 killed renamable $r4, 1, 14 /* CC::al */, $noreg, debug-location !91
    DBG_VALUE $r4, $noreg, !43, !DIExpression(), debug-location !63
    renamable $r0 = t2ADDrs killed renamable $r0, renamable $r1, 10, 14 /* CC::al */, $noreg, $noreg, debug-location !89
    DBG_VALUE $r0, $noreg, !33, !DIExpression(), debug-location !62
    tCMPr renamable $r4, renamable $r3, 14 /* CC::al */, $noreg, implicit-def $cpsr, debug-location !70
    renamable $r2 = t2ADDrs killed renamable $r2, killed renamable $r6, 10, 14 /* CC::al */, $noreg, $noreg, debug-location !90
    DBG_VALUE $r2, $noreg, !35, !DIExpression(), debug-location !62
    tBcc %bb.2, 1 /* CC::ne */, killed $cpsr, debug-location !70

  bb.5.for.cond.cleanup:
    DBG_VALUE $sp, 0, !36, !DIExpression(), debug-location !62
    DBG_VALUE $r7, 0, !38, !DIExpression(DW_OP_plus_uconst, 12), debug-location !62
    DBG_VALUE $r7, 0, !37, !DIExpression(DW_OP_plus_uconst, 8), debug-location !62
    DBG_VALUE $r0, $noreg, !33, !DIExpression(), debug-location !62
    $sp = frame-destroy tADDspi $sp, 1, 14 /* CC::al */, $noreg, debug-location !74
    $sp = frame-destroy t2LDMIA_UPD $sp, 14 /* CC::al */, $noreg, def $r8, def $r9, def $r10, def $r11, debug-location !74
    frame-destroy tPOP_RET 14 /* CC::al */, $noreg, def $r4, def $r5, def $r6, def $r7, def $pc, debug-location !74

...