# 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_fma_s32
body: |
  bb.0:
    liveins: $vgpr0, $vgpr1, $vgpr2

    ; SI-LABEL: name: test_fma_s32
    ; SI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
    ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
    ; SI: $vgpr0 = COPY [[FMA]](s32)
    ; VI-LABEL: name: test_fma_s32
    ; VI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
    ; VI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
    ; VI: $vgpr0 = COPY [[FMA]](s32)
    ; GFX9-LABEL: name: test_fma_s32
    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
    ; GFX9: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
    ; GFX9: $vgpr0 = COPY [[FMA]](s32)
    %0:_(s32) = COPY $vgpr0
    %1:_(s32) = COPY $vgpr1
    %2:_(s32) = COPY $vgpr2
    %3:_(s32) = G_FMA %0, %1, %2
    $vgpr0 = COPY %3
...
---
name: test_fma_s64
body: |
  bb.0:
    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3,  $vgpr4_vgpr5

    ; SI-LABEL: name: test_fma_s64
    ; SI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
    ; SI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
    ; SI: [[COPY2:%[0-9]+]]:_(s64) = COPY $vgpr4_vgpr5
    ; SI: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
    ; SI: $vgpr0_vgpr1 = COPY [[FMA]](s64)
    ; VI-LABEL: name: test_fma_s64
    ; VI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
    ; VI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
    ; VI: [[COPY2:%[0-9]+]]:_(s64) = COPY $vgpr4_vgpr5
    ; VI: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
    ; VI: $vgpr0_vgpr1 = COPY [[FMA]](s64)
    ; GFX9-LABEL: name: test_fma_s64
    ; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
    ; GFX9: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
    ; GFX9: [[COPY2:%[0-9]+]]:_(s64) = COPY $vgpr4_vgpr5
    ; GFX9: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
    ; GFX9: $vgpr0_vgpr1 = COPY [[FMA]](s64)
    %0:_(s64) = COPY $vgpr0_vgpr1
    %1:_(s64) = COPY $vgpr2_vgpr3
    %2:_(s64) = COPY $vgpr4_vgpr5
    %3:_(s64) = G_FMA %0, %1, %2
    $vgpr0_vgpr1 = COPY %3
...

---
name: test_fma_s16
body: |
  bb.0:
    liveins: $vgpr0, $vgpr1, $vgpr2

    ; SI-LABEL: name: test_fma_s16
    ; SI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr1
    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
    ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
    ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16)
    ; SI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; VI-LABEL: name: test_fma_s16
    ; VI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr1
    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
    ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
    ; VI: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
    ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; GFX9-LABEL: name: test_fma_s16
    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr1
    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
    ; GFX9: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
    ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
    %0:_(s32) = COPY $vgpr0
    %1:_(s32) = COPY $vgpr1
    %2:_(s32) = COPY $vgpr1
    %3:_(s16) = G_TRUNC %0
    %4:_(s16) = G_TRUNC %1
    %5:_(s16) = G_TRUNC %2

    %6:_(s16) = G_FMA %3, %4, %5
    %7:_(s32) = G_ANYEXT %6
    $vgpr0 = COPY %7
...

---
name: test_fma_v2s32
body: |
  bb.0:
    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5

    ; SI-LABEL: name: test_fma_v2s32
    ; SI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
    ; SI: [[COPY2:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr4_vgpr5
    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
    ; SI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
    ; SI: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<2 x s32>)
    ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV2]], [[UV4]]
    ; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV3]], [[UV5]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32)
    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; VI-LABEL: name: test_fma_v2s32
    ; VI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
    ; VI: [[COPY2:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr4_vgpr5
    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
    ; VI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
    ; VI: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<2 x s32>)
    ; VI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV2]], [[UV4]]
    ; VI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV3]], [[UV5]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32)
    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; GFX9-LABEL: name: test_fma_v2s32
    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
    ; GFX9: [[COPY2:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr4_vgpr5
    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
    ; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
    ; GFX9: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<2 x s32>)
    ; GFX9: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV2]], [[UV4]]
    ; GFX9: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV3]], [[UV5]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32)
    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
    %2:_(<2 x s32>) = COPY $vgpr4_vgpr5
    %3:_(<2 x s32>) = G_FMA %0, %1, %2
    $vgpr0_vgpr1 = COPY %3
...

---
name: test_fma_v3s32
body: |
  bb.0:
    liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5, $vgpr6_vgpr7_vgpr8

    ; SI-LABEL: name: test_fma_v3s32
    ; SI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
    ; SI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
    ; SI: [[COPY2:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr6_vgpr7_vgpr8
    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
    ; SI: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
    ; SI: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32), [[UV8:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<3 x s32>)
    ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV3]], [[UV6]]
    ; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV4]], [[UV7]]
    ; SI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV5]], [[UV8]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32)
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
    ; VI-LABEL: name: test_fma_v3s32
    ; VI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
    ; VI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
    ; VI: [[COPY2:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr6_vgpr7_vgpr8
    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
    ; VI: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
    ; VI: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32), [[UV8:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<3 x s32>)
    ; VI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV3]], [[UV6]]
    ; VI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV4]], [[UV7]]
    ; VI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV5]], [[UV8]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32)
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
    ; GFX9-LABEL: name: test_fma_v3s32
    ; GFX9: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
    ; GFX9: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
    ; GFX9: [[COPY2:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr6_vgpr7_vgpr8
    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
    ; GFX9: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
    ; GFX9: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32), [[UV8:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<3 x s32>)
    ; GFX9: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV3]], [[UV6]]
    ; GFX9: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV4]], [[UV7]]
    ; GFX9: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV5]], [[UV8]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32)
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
    %0:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
    %1:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
    %2:_(<3 x s32>) = COPY $vgpr6_vgpr7_vgpr8
    %3:_(<3 x s32>) = G_FMA %0, %1, %2
    $vgpr0_vgpr1_vgpr2 = COPY %3
...

---
name: test_fma_v4s32
body: |
  bb.0:
    liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11

    ; SI-LABEL: name: test_fma_v4s32
    ; SI: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
    ; SI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
    ; SI: [[COPY2:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<4 x s32>)
    ; SI: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
    ; SI: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32), [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<4 x s32>)
    ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV4]], [[UV8]]
    ; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV5]], [[UV9]]
    ; SI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV6]], [[UV10]]
    ; SI: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[UV3]], [[UV7]], [[UV11]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32), [[FMA3]](s32)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
    ; VI-LABEL: name: test_fma_v4s32
    ; VI: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
    ; VI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
    ; VI: [[COPY2:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<4 x s32>)
    ; VI: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
    ; VI: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32), [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<4 x s32>)
    ; VI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV4]], [[UV8]]
    ; VI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV5]], [[UV9]]
    ; VI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV6]], [[UV10]]
    ; VI: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[UV3]], [[UV7]], [[UV11]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32), [[FMA3]](s32)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
    ; GFX9-LABEL: name: test_fma_v4s32
    ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
    ; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
    ; GFX9: [[COPY2:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<4 x s32>)
    ; GFX9: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
    ; GFX9: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32), [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<4 x s32>)
    ; GFX9: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV4]], [[UV8]]
    ; GFX9: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV5]], [[UV9]]
    ; GFX9: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV6]], [[UV10]]
    ; GFX9: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[UV3]], [[UV7]], [[UV11]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32), [[FMA3]](s32)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
    %0:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
    %1:_(<4 x s32>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
    %2:_(<4 x s32>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
    %3:_(<4 x s32>) = G_FMA %0, %1, %2
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
...

---
name: test_fma_v2s64
body: |
  bb.0:
    liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11

    ; SI-LABEL: name: test_fma_v2s64
    ; SI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
    ; SI: [[COPY2:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
    ; SI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
    ; SI: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
    ; SI: [[UV4:%[0-9]+]]:_(s64), [[UV5:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY2]](<2 x s64>)
    ; SI: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[UV]], [[UV2]], [[UV4]]
    ; SI: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[UV1]], [[UV3]], [[UV5]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FMA]](s64), [[FMA1]](s64)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    ; VI-LABEL: name: test_fma_v2s64
    ; VI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
    ; VI: [[COPY2:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
    ; VI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
    ; VI: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
    ; VI: [[UV4:%[0-9]+]]:_(s64), [[UV5:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY2]](<2 x s64>)
    ; VI: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[UV]], [[UV2]], [[UV4]]
    ; VI: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[UV1]], [[UV3]], [[UV5]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FMA]](s64), [[FMA1]](s64)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    ; GFX9-LABEL: name: test_fma_v2s64
    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
    ; GFX9: [[COPY2:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
    ; GFX9: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
    ; GFX9: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
    ; GFX9: [[UV4:%[0-9]+]]:_(s64), [[UV5:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY2]](<2 x s64>)
    ; GFX9: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[UV]], [[UV2]], [[UV4]]
    ; GFX9: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[UV1]], [[UV3]], [[UV5]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FMA]](s64), [[FMA1]](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>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
    %2:_(<2 x s64>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
    %3:_(<2 x s64>) = G_FMA %0, %1, %2
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
...

---
name: test_fma_v2s16
body: |
  bb.0:
    liveins: $vgpr0, $vgpr1, $vgpr2

    ; SI-LABEL: name: test_fma_v2s16
    ; SI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
    ; SI: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
    ; SI: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
    ; 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: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<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: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY2]](<2 x s16>)
    ; SI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
    ; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
    ; SI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC4]](s16)
    ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
    ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
    ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC5]](s16)
    ; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
    ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC]](s16)
    ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC1]](s16)
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
    ; SI: [[BITCAST3:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; SI: $vgpr0 = COPY [[BITCAST3]](<2 x s16>)
    ; VI-LABEL: name: test_fma_v2s16
    ; VI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
    ; VI: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
    ; VI: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
    ; 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: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<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: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY2]](<2 x s16>)
    ; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
    ; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
    ; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
    ; VI: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC2]], [[TRUNC4]]
    ; VI: [[FMA1:%[0-9]+]]:_(s16) = G_FMA [[TRUNC1]], [[TRUNC3]], [[TRUNC5]]
    ; VI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FMA]](s16)
    ; VI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FMA1]](s16)
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
    ; VI: [[BITCAST3:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; VI: $vgpr0 = COPY [[BITCAST3]](<2 x s16>)
    ; GFX9-LABEL: name: test_fma_v2s16
    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
    ; GFX9: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
    ; GFX9: [[FMA:%[0-9]+]]:_(<2 x s16>) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
    ; GFX9: $vgpr0 = COPY [[FMA]](<2 x s16>)
    %0:_(<2 x s16>) = COPY $vgpr0
    %1:_(<2 x s16>) = COPY $vgpr1
    %2:_(<2 x s16>) = COPY $vgpr2
    %3:_(<2 x s16>) = G_FMA %0, %1, %2
    $vgpr0 = COPY %3
...

---
name: test_fma_v3s16
body: |
  bb.0:
    liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5, $vgpr6_vgpr7_vgpr8

    ; SI-LABEL: name: test_fma_v3s16
    ; SI: [[COPY:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr0_vgpr1_vgpr2
    ; SI: [[COPY1:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr3_vgpr4_vgpr5
    ; SI: [[COPY2:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr6_vgpr7_vgpr8
    ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<6 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: [[UV3:%[0-9]+]]:_(<2 x s16>), [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<6 x s16>)
    ; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
    ; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
    ; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
    ; SI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
    ; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<2 x s16>)
    ; SI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
    ; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
    ; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>), [[UV8:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY2]](<6 x s16>)
    ; SI: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[UV6]](<2 x s16>)
    ; SI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST4]](s32)
    ; SI: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST4]], [[C]](s32)
    ; SI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR4]](s32)
    ; SI: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV7]](<2 x s16>)
    ; SI: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST5]](s32)
    ; SI: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST5]], [[C]](s32)
    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC6]](s16)
    ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
    ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC4]](s16)
    ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC7]](s16)
    ; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
    ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
    ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC5]](s16)
    ; SI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC8]](s16)
    ; SI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[FPEXT6]], [[FPEXT7]], [[FPEXT8]]
    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA2]](s32)
    ; SI: [[DEF:%[0-9]+]]:_(s16) = G_IMPLICIT_DEF
    ; SI: [[DEF1:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
    ; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; SI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; SI: [[UV9:%[0-9]+]]:_(<2 x s16>), [[UV10:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[DEF2]](<4 x s16>)
    ; SI: [[BITCAST6:%[0-9]+]]:_(s32) = G_BITCAST [[UV9]](<2 x s16>)
    ; SI: [[LSHR6:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST6]], [[C]](s32)
    ; SI: [[BITCAST7:%[0-9]+]]:_(s32) = G_BITCAST [[UV10]](<2 x s16>)
    ; SI: [[LSHR7:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST7]], [[C]](s32)
    ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC]](s16)
    ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC1]](s16)
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
    ; SI: [[BITCAST8:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC2]](s16)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST6]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND]], [[C]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
    ; SI: [[BITCAST9:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR6]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[BITCAST7]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND1]], [[SHL2]]
    ; SI: [[BITCAST10:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST8]](<2 x s16>), [[BITCAST9]](<2 x s16>), [[BITCAST10]](<2 x s16>)
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[CONCAT_VECTORS]](<6 x s16>)
    ; VI-LABEL: name: test_fma_v3s16
    ; VI: [[COPY:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr0_vgpr1_vgpr2
    ; VI: [[COPY1:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr3_vgpr4_vgpr5
    ; VI: [[COPY2:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr6_vgpr7_vgpr8
    ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<6 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: [[UV3:%[0-9]+]]:_(<2 x s16>), [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<6 x s16>)
    ; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
    ; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
    ; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
    ; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
    ; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<2 x s16>)
    ; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
    ; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
    ; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>), [[UV8:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY2]](<6 x s16>)
    ; VI: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[UV6]](<2 x s16>)
    ; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST4]](s32)
    ; VI: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST4]], [[C]](s32)
    ; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR4]](s32)
    ; VI: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV7]](<2 x s16>)
    ; VI: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST5]](s32)
    ; VI: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST5]], [[C]](s32)
    ; VI: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC3]], [[TRUNC6]]
    ; VI: [[FMA1:%[0-9]+]]:_(s16) = G_FMA [[TRUNC1]], [[TRUNC4]], [[TRUNC7]]
    ; VI: [[FMA2:%[0-9]+]]:_(s16) = G_FMA [[TRUNC2]], [[TRUNC5]], [[TRUNC8]]
    ; VI: [[DEF:%[0-9]+]]:_(s16) = G_IMPLICIT_DEF
    ; VI: [[DEF1:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
    ; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; VI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; VI: [[UV9:%[0-9]+]]:_(<2 x s16>), [[UV10:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[DEF2]](<4 x s16>)
    ; VI: [[BITCAST6:%[0-9]+]]:_(s32) = G_BITCAST [[UV9]](<2 x s16>)
    ; VI: [[LSHR6:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST6]], [[C]](s32)
    ; VI: [[BITCAST7:%[0-9]+]]:_(s32) = G_BITCAST [[UV10]](<2 x s16>)
    ; VI: [[LSHR7:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST7]], [[C]](s32)
    ; VI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FMA]](s16)
    ; VI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FMA1]](s16)
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
    ; VI: [[BITCAST8:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; VI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[FMA2]](s16)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST6]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND]], [[C]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
    ; VI: [[BITCAST9:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR6]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[BITCAST7]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND1]], [[SHL2]]
    ; VI: [[BITCAST10:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST8]](<2 x s16>), [[BITCAST9]](<2 x s16>), [[BITCAST10]](<2 x s16>)
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[CONCAT_VECTORS]](<6 x s16>)
    ; GFX9-LABEL: name: test_fma_v3s16
    ; GFX9: [[COPY:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr0_vgpr1_vgpr2
    ; GFX9: [[COPY1:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr3_vgpr4_vgpr5
    ; GFX9: [[COPY2:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr6_vgpr7_vgpr8
    ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<6 x s16>)
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
    ; GFX9: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
    ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY3]](s32), [[COPY4]](s32)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
    ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
    ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY5]](s32), [[DEF]](s32)
    ; GFX9: [[DEF1:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
    ; GFX9: [[UV3:%[0-9]+]]:_(<2 x s16>), [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<6 x s16>)
    ; GFX9: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
    ; GFX9: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
    ; GFX9: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<2 x s16>)
    ; GFX9: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY6]](s32), [[COPY7]](s32)
    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY8]](s32), [[DEF]](s32)
    ; GFX9: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>), [[UV8:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY2]](<6 x s16>)
    ; GFX9: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[UV6]](<2 x s16>)
    ; GFX9: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST4]], [[C]](s32)
    ; GFX9: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV7]](<2 x s16>)
    ; GFX9: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST5]], [[C]](s32)
    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST4]](s32)
    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC4:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32)
    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[BITCAST5]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC5:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[DEF]](s32)
    ; GFX9: [[FMA:%[0-9]+]]:_(<2 x s16>) = G_FMA [[BUILD_VECTOR_TRUNC]], [[BUILD_VECTOR_TRUNC2]], [[BUILD_VECTOR_TRUNC4]]
    ; GFX9: [[FMA1:%[0-9]+]]:_(<2 x s16>) = G_FMA [[BUILD_VECTOR_TRUNC1]], [[BUILD_VECTOR_TRUNC3]], [[BUILD_VECTOR_TRUNC5]]
    ; GFX9: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; GFX9: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; GFX9: [[BITCAST6:%[0-9]+]]:_(s32) = G_BITCAST [[FMA]](<2 x s16>)
    ; GFX9: [[LSHR6:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST6]], [[C]](s32)
    ; GFX9: [[BITCAST7:%[0-9]+]]:_(s32) = G_BITCAST [[FMA1]](<2 x s16>)
    ; GFX9: [[LSHR7:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST7]], [[C]](s32)
    ; GFX9: [[UV9:%[0-9]+]]:_(<2 x s16>), [[UV10:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[DEF2]](<4 x s16>)
    ; GFX9: [[BITCAST8:%[0-9]+]]:_(s32) = G_BITCAST [[UV9]](<2 x s16>)
    ; GFX9: [[LSHR8:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST8]], [[C]](s32)
    ; GFX9: [[BITCAST9:%[0-9]+]]:_(s32) = G_BITCAST [[UV10]](<2 x s16>)
    ; GFX9: [[LSHR9:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST9]], [[C]](s32)
    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST6]](s32)
    ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR6]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC6:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
    ; GFX9: [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST7]](s32)
    ; GFX9: [[COPY15:%[0-9]+]]:_(s32) = COPY [[BITCAST8]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC7:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
    ; GFX9: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LSHR8]](s32)
    ; GFX9: [[COPY17:%[0-9]+]]:_(s32) = COPY [[BITCAST9]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC8:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC6]](<2 x s16>), [[BUILD_VECTOR_TRUNC7]](<2 x s16>), [[BUILD_VECTOR_TRUNC8]](<2 x s16>)
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[CONCAT_VECTORS]](<6 x s16>)
    %0:_(<6 x s16>) = COPY $vgpr0_vgpr1_vgpr2
    %1:_(<6 x s16>) = COPY $vgpr3_vgpr4_vgpr5
    %2:_(<6 x s16>) = COPY $vgpr6_vgpr7_vgpr8
    %3:_(<3 x s16>), %4:_(<3 x s16>) = G_UNMERGE_VALUES %0
    %5:_(<3 x s16>), %6:_(<3 x s16>) = G_UNMERGE_VALUES %1
    %7:_(<3 x s16>), %8:_(<3 x s16>) = G_UNMERGE_VALUES %2
    %9:_(<3 x s16>) = G_FMA %3, %5, %7
    %10:_(<3 x s16>) = G_IMPLICIT_DEF
    %11:_(<6 x s16>) = G_CONCAT_VECTORS %9, %10
    $vgpr0_vgpr1_vgpr2 = COPY %11

...

---
name: test_fma_v4s16
body: |
  bb.0:
    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5

    ; SI-LABEL: name: test_fma_v4s16
    ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
    ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
    ; SI: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5
    ; 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: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
    ; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
    ; SI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
    ; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
    ; SI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
    ; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
    ; SI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
    ; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
    ; SI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
    ; SI: [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY2]](<4 x s16>)
    ; SI: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<2 x s16>)
    ; SI: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST4]](s32)
    ; SI: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST4]], [[C]](s32)
    ; SI: [[TRUNC9:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR4]](s32)
    ; SI: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV5]](<2 x s16>)
    ; SI: [[TRUNC10:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST5]](s32)
    ; SI: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST5]], [[C]](s32)
    ; SI: [[TRUNC11:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR5]](s32)
    ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
    ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC4]](s16)
    ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC8]](s16)
    ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
    ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
    ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
    ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC5]](s16)
    ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC9]](s16)
    ; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
    ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
    ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
    ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC6]](s16)
    ; SI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC10]](s16)
    ; SI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[FPEXT6]], [[FPEXT7]], [[FPEXT8]]
    ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA2]](s32)
    ; SI: [[FPEXT9:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
    ; SI: [[FPEXT10:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC7]](s16)
    ; SI: [[FPEXT11:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC11]](s16)
    ; SI: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[FPEXT9]], [[FPEXT10]], [[FPEXT11]]
    ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA3]](s32)
    ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC]](s16)
    ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC1]](s16)
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
    ; SI: [[BITCAST6:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC2]](s16)
    ; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC3]](s16)
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
    ; SI: [[BITCAST7:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST6]](<2 x s16>), [[BITCAST7]](<2 x s16>)
    ; SI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    ; VI-LABEL: name: test_fma_v4s16
    ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
    ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
    ; VI: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5
    ; 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: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
    ; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
    ; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
    ; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
    ; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
    ; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
    ; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
    ; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
    ; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
    ; VI: [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY2]](<4 x s16>)
    ; VI: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<2 x s16>)
    ; VI: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST4]](s32)
    ; VI: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST4]], [[C]](s32)
    ; VI: [[TRUNC9:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR4]](s32)
    ; VI: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV5]](<2 x s16>)
    ; VI: [[TRUNC10:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST5]](s32)
    ; VI: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST5]], [[C]](s32)
    ; VI: [[TRUNC11:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR5]](s32)
    ; VI: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC4]], [[TRUNC8]]
    ; VI: [[FMA1:%[0-9]+]]:_(s16) = G_FMA [[TRUNC1]], [[TRUNC5]], [[TRUNC9]]
    ; VI: [[FMA2:%[0-9]+]]:_(s16) = G_FMA [[TRUNC2]], [[TRUNC6]], [[TRUNC10]]
    ; VI: [[FMA3:%[0-9]+]]:_(s16) = G_FMA [[TRUNC3]], [[TRUNC7]], [[TRUNC11]]
    ; VI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FMA]](s16)
    ; VI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FMA1]](s16)
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
    ; VI: [[BITCAST6:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; VI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[FMA2]](s16)
    ; VI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[FMA3]](s16)
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
    ; VI: [[BITCAST7:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST6]](<2 x s16>), [[BITCAST7]](<2 x s16>)
    ; VI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    ; GFX9-LABEL: name: test_fma_v4s16
    ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
    ; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
    ; GFX9: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5
    ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
    ; GFX9: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
    ; GFX9: [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY2]](<4 x s16>)
    ; GFX9: [[FMA:%[0-9]+]]:_(<2 x s16>) = G_FMA [[UV]], [[UV2]], [[UV4]]
    ; GFX9: [[FMA1:%[0-9]+]]:_(<2 x s16>) = G_FMA [[UV1]], [[UV3]], [[UV5]]
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[FMA]](<2 x s16>), [[FMA1]](<2 x s16>)
    ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
    %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
    %2:_(<4 x s16>) = COPY $vgpr4_vgpr5
    %3:_(<4 x s16>) = G_FMA %0, %1, %2
    $vgpr0_vgpr1 = COPY %3
...