# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=SI  %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=VI %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=GFX9  %s

---
name: test_fsqrt_s32
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_fsqrt_s32
    ; SI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
    ; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[COPY]]
    ; SI: $vgpr0 = COPY [[FSQRT]](s32)
    ; VI-LABEL: name: test_fsqrt_s32
    ; VI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
    ; VI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[COPY]]
    ; VI: $vgpr0 = COPY [[FSQRT]](s32)
    ; GFX9-LABEL: name: test_fsqrt_s32
    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
    ; GFX9: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[COPY]]
    ; GFX9: $vgpr0 = COPY [[FSQRT]](s32)
    %0:_(s32) = COPY $vgpr0
    %1:_(s32) = G_FSQRT %0
    $vgpr0 = COPY %1
...

---
name: test_fsqrt_s64
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_fsqrt_s64
    ; SI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
    ; SI: [[FSQRT:%[0-9]+]]:_(s64) = G_FSQRT [[COPY]]
    ; SI: $vgpr0_vgpr1 = COPY [[FSQRT]](s64)
    ; VI-LABEL: name: test_fsqrt_s64
    ; VI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
    ; VI: [[FSQRT:%[0-9]+]]:_(s64) = G_FSQRT [[COPY]]
    ; VI: $vgpr0_vgpr1 = COPY [[FSQRT]](s64)
    ; GFX9-LABEL: name: test_fsqrt_s64
    ; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
    ; GFX9: [[FSQRT:%[0-9]+]]:_(s64) = G_FSQRT [[COPY]]
    ; GFX9: $vgpr0_vgpr1 = COPY [[FSQRT]](s64)
    %0:_(s64) = COPY $vgpr0_vgpr1
    %1:_(s64) = G_FSQRT %0
    $vgpr0_vgpr1 = COPY %1

...
---
name: test_fsqrt_s16
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_fsqrt_s16
    ; SI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
    ; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT]]
    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT]](s32)
    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16)
    ; SI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; VI-LABEL: name: test_fsqrt_s16
    ; VI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
    ; VI: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT]](s16)
    ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; GFX9-LABEL: name: test_fsqrt_s16
    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
    ; GFX9: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT]](s16)
    ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
    %0:_(s32) = COPY $vgpr0
    %1:_(s16) = G_TRUNC %0
    %2:_(s16) = G_FSQRT %1
    %3:_(s32) = G_ANYEXT %2
    $vgpr0 = COPY %3
...

---
name: test_fsqrt_v2s32
body: |
  bb.0:
    liveins: $vgpr0_vgpr1

    ; SI-LABEL: name: test_fsqrt_v2s32
    ; SI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
    ; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[UV]]
    ; SI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[UV1]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FSQRT]](s32), [[FSQRT1]](s32)
    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; VI-LABEL: name: test_fsqrt_v2s32
    ; VI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
    ; VI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[UV]]
    ; VI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[UV1]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FSQRT]](s32), [[FSQRT1]](s32)
    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; GFX9-LABEL: name: test_fsqrt_v2s32
    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
    ; GFX9: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[UV]]
    ; GFX9: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[UV1]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FSQRT]](s32), [[FSQRT1]](s32)
    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
    %1:_(<2 x s32>) = G_FSQRT %0
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_fsqrt_v3s32
body: |
  bb.0:
    liveins: $vgpr0_vgpr1_vgpr2

    ; SI-LABEL: name: test_fsqrt_v3s32
    ; SI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
    ; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[UV]]
    ; SI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[UV1]]
    ; SI: [[FSQRT2:%[0-9]+]]:_(s32) = G_FSQRT [[UV2]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FSQRT]](s32), [[FSQRT1]](s32), [[FSQRT2]](s32)
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
    ; VI-LABEL: name: test_fsqrt_v3s32
    ; VI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
    ; VI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[UV]]
    ; VI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[UV1]]
    ; VI: [[FSQRT2:%[0-9]+]]:_(s32) = G_FSQRT [[UV2]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FSQRT]](s32), [[FSQRT1]](s32), [[FSQRT2]](s32)
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
    ; GFX9-LABEL: name: test_fsqrt_v3s32
    ; GFX9: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
    ; GFX9: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[UV]]
    ; GFX9: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[UV1]]
    ; GFX9: [[FSQRT2:%[0-9]+]]:_(s32) = G_FSQRT [[UV2]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FSQRT]](s32), [[FSQRT1]](s32), [[FSQRT2]](s32)
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
    %0:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
    %1:_(<3 x  s32>) = G_FSQRT %0
    $vgpr0_vgpr1_vgpr2 = COPY %1
...

---
name: test_fsqrt_v2s64
body: |
  bb.0:
    liveins: $vgpr0_vgpr1_vgpr2_vgpr3

    ; SI-LABEL: name: test_fsqrt_v2s64
    ; SI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
    ; SI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
    ; SI: [[FSQRT:%[0-9]+]]:_(s64) = G_FSQRT [[UV]]
    ; SI: [[FSQRT1:%[0-9]+]]:_(s64) = G_FSQRT [[UV1]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FSQRT]](s64), [[FSQRT1]](s64)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    ; VI-LABEL: name: test_fsqrt_v2s64
    ; VI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
    ; VI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
    ; VI: [[FSQRT:%[0-9]+]]:_(s64) = G_FSQRT [[UV]]
    ; VI: [[FSQRT1:%[0-9]+]]:_(s64) = G_FSQRT [[UV1]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FSQRT]](s64), [[FSQRT1]](s64)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    ; GFX9-LABEL: name: test_fsqrt_v2s64
    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
    ; GFX9: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
    ; GFX9: [[FSQRT:%[0-9]+]]:_(s64) = G_FSQRT [[UV]]
    ; GFX9: [[FSQRT1:%[0-9]+]]:_(s64) = G_FSQRT [[UV1]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FSQRT]](s64), [[FSQRT1]](s64)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
    %1:_(<2 x s64>) = G_FSQRT %0
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_fsqrt_v2s16
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_fsqrt_v2s16
    ; SI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
    ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
    ; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT]]
    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT]](s32)
    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
    ; SI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT1]]
    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT1]](s32)
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
    ; SI: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
    ; VI-LABEL: name: test_fsqrt_v2s16
    ; VI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
    ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
    ; VI: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
    ; VI: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC1]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FSQRT]](s16), [[FSQRT1]](s16)
    ; VI: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
    ; GFX9-LABEL: name: test_fsqrt_v2s16
    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
    ; GFX9: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
    ; GFX9: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC1]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FSQRT]](s16), [[FSQRT1]](s16)
    ; GFX9: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
    %0:_(<2 x s16>) = COPY $vgpr0
    %1:_(<2 x s16>) = G_FSQRT %0
    $vgpr0 = COPY %1
...

---
name: test_fsqrt_v3s16
body: |
  bb.0:

    ; SI-LABEL: name: test_fsqrt_v3s16
    ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0
    ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0
    ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>)
    ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
    ; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
    ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
    ; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
    ; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT]]
    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT]](s32)
    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
    ; SI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT1]]
    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT1]](s32)
    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
    ; SI: [[FSQRT2:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT2]]
    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT2]](s32)
    ; SI: [[DEF2:%[0-9]+]]:_(s16) = G_IMPLICIT_DEF
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
    ; SI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC2]](s16), [[DEF2]](s16)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s16>), [[BUILD_VECTOR1]](<2 x s16>)
    ; SI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS]](<4 x s16>), 0
    ; SI: S_NOP 0, implicit [[EXTRACT1]](<3 x s16>)
    ; VI-LABEL: name: test_fsqrt_v3s16
    ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0
    ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0
    ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>)
    ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
    ; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
    ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
    ; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
    ; VI: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
    ; VI: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC1]]
    ; VI: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC2]]
    ; VI: [[DEF2:%[0-9]+]]:_(s16) = G_IMPLICIT_DEF
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FSQRT]](s16), [[FSQRT1]](s16)
    ; VI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FSQRT2]](s16), [[DEF2]](s16)
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s16>), [[BUILD_VECTOR1]](<2 x s16>)
    ; VI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS]](<4 x s16>), 0
    ; VI: S_NOP 0, implicit [[EXTRACT1]](<3 x s16>)
    ; GFX9-LABEL: name: test_fsqrt_v3s16
    ; GFX9: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; GFX9: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0
    ; GFX9: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0
    ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>)
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
    ; GFX9: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
    ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
    ; GFX9: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
    ; GFX9: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC1]]
    ; GFX9: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC2]]
    ; GFX9: [[DEF2:%[0-9]+]]:_(s16) = G_IMPLICIT_DEF
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FSQRT]](s16), [[FSQRT1]](s16)
    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FSQRT2]](s16), [[DEF2]](s16)
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s16>), [[BUILD_VECTOR1]](<2 x s16>)
    ; GFX9: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS]](<4 x s16>), 0
    ; GFX9: S_NOP 0, implicit [[EXTRACT1]](<3 x s16>)
    %0:_(<3 x s16>) = G_IMPLICIT_DEF
    %1:_(<3 x s16>) = G_FSQRT %0
    S_NOP 0, implicit %1
...

---
name: test_fsqrt_v4s16
body: |
  bb.0:
    liveins: $vgpr0_vgpr1

    ; SI-LABEL: name: test_fsqrt_v4s16
    ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
    ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
    ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
    ; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
    ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
    ; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
    ; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
    ; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT]]
    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT]](s32)
    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
    ; SI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT1]]
    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT1]](s32)
    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
    ; SI: [[FSQRT2:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT2]]
    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT2]](s32)
    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
    ; SI: [[FSQRT3:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT3]]
    ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT3]](s32)
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16)
    ; SI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s16>), [[BUILD_VECTOR1]](<2 x s16>)
    ; SI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    ; VI-LABEL: name: test_fsqrt_v4s16
    ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
    ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
    ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
    ; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
    ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
    ; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
    ; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
    ; VI: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
    ; VI: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC1]]
    ; VI: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC2]]
    ; VI: [[FSQRT3:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC3]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FSQRT]](s16), [[FSQRT1]](s16)
    ; VI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FSQRT2]](s16), [[FSQRT3]](s16)
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s16>), [[BUILD_VECTOR1]](<2 x s16>)
    ; VI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    ; GFX9-LABEL: name: test_fsqrt_v4s16
    ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
    ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
    ; GFX9: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
    ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
    ; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
    ; GFX9: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
    ; GFX9: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC1]]
    ; GFX9: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC2]]
    ; GFX9: [[FSQRT3:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC3]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FSQRT]](s16), [[FSQRT1]](s16)
    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[FSQRT2]](s16), [[FSQRT3]](s16)
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s16>), [[BUILD_VECTOR1]](<2 x s16>)
    ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
    %1:_(<4 x s16>) = G_FSQRT %0
    $vgpr0_vgpr1 = COPY %1
...