# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -O0 -run-pass=legalizer -global-isel-abort=0  %s -o - | FileCheck -check-prefix=SI %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=bonaire -mattr=-enable-ds128 -O0 -run-pass=legalizer -global-isel-abort=0  %s -o - | FileCheck -check-prefix=CI %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=bonaire -mattr=+enable-ds128 -O0 -run-pass=legalizer -global-isel-abort=0  %s -o - | FileCheck -check-prefix=CI-DS128 %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -O0 -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=VI %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -O0 -run-pass=legalizer -mattr=-unaligned-access-mode -global-isel-abort=0 %s -o - | FileCheck -check-prefix=GFX9 %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -O0 -run-pass=legalizer -mattr=+unaligned-access-mode -global-isel-abort=0 %s -o - | FileCheck -check-prefix=GFX9-UNALIGNED %s

---
name: test_load_local_s1_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s1_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
    ; SI: $vgpr0 = COPY [[AND]](s32)
    ; CI-LABEL: name: test_load_local_s1_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
    ; CI: $vgpr0 = COPY [[AND]](s32)
    ; CI-DS128-LABEL: name: test_load_local_s1_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
    ; CI-DS128: $vgpr0 = COPY [[AND]](s32)
    ; VI-LABEL: name: test_load_local_s1_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
    ; VI: $vgpr0 = COPY [[AND]](s32)
    ; GFX9-LABEL: name: test_load_local_s1_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
    ; GFX9: $vgpr0 = COPY [[AND]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s1_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[AND]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s1) = G_LOAD %0 :: (load 1, align 1, addrspace 3)
    %2:_(s32) = G_ZEXT %1
    $vgpr0 = COPY %2
...

---
name: test_load_local_s2_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s2_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
    ; SI: $vgpr0 = COPY [[AND]](s32)
    ; CI-LABEL: name: test_load_local_s2_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
    ; CI: $vgpr0 = COPY [[AND]](s32)
    ; CI-DS128-LABEL: name: test_load_local_s2_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
    ; CI-DS128: $vgpr0 = COPY [[AND]](s32)
    ; VI-LABEL: name: test_load_local_s2_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
    ; VI: $vgpr0 = COPY [[AND]](s32)
    ; GFX9-LABEL: name: test_load_local_s2_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
    ; GFX9: $vgpr0 = COPY [[AND]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s2_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[AND]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s2) = G_LOAD %0 :: (load 1, align 1, addrspace 3)
    %2:_(s32) = G_ZEXT %1
    $vgpr0 = COPY %2
...

---
name: test_load_local_s8_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s8_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: $vgpr0 = COPY [[COPY1]](s32)
    ; CI-LABEL: name: test_load_local_s8_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: $vgpr0 = COPY [[COPY1]](s32)
    ; CI-DS128-LABEL: name: test_load_local_s8_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: $vgpr0 = COPY [[COPY1]](s32)
    ; VI-LABEL: name: test_load_local_s8_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: $vgpr0 = COPY [[COPY1]](s32)
    ; GFX9-LABEL: name: test_load_local_s8_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: $vgpr0 = COPY [[COPY1]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s8_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[COPY1]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s8) = G_LOAD %0 :: (load 1, align 4, addrspace 3)
    %2:_(s32) = G_ANYEXT %1
    $vgpr0 = COPY %2
...

---
name: test_load_local_s8_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s8_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: $vgpr0 = COPY [[COPY1]](s32)
    ; CI-LABEL: name: test_load_local_s8_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: $vgpr0 = COPY [[COPY1]](s32)
    ; CI-DS128-LABEL: name: test_load_local_s8_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: $vgpr0 = COPY [[COPY1]](s32)
    ; VI-LABEL: name: test_load_local_s8_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: $vgpr0 = COPY [[COPY1]](s32)
    ; GFX9-LABEL: name: test_load_local_s8_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: $vgpr0 = COPY [[COPY1]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s8_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[COPY1]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s8) = G_LOAD %0 :: (load 1, align 1, addrspace 3)
    %2:_(s32) = G_ANYEXT %1
    $vgpr0 = COPY %2
...

---
name: test_load_local_s16_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s16_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: $vgpr0 = COPY [[COPY1]](s32)
    ; CI-LABEL: name: test_load_local_s16_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: $vgpr0 = COPY [[COPY1]](s32)
    ; CI-DS128-LABEL: name: test_load_local_s16_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: $vgpr0 = COPY [[COPY1]](s32)
    ; VI-LABEL: name: test_load_local_s16_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: $vgpr0 = COPY [[COPY1]](s32)
    ; GFX9-LABEL: name: test_load_local_s16_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: $vgpr0 = COPY [[COPY1]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s16_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[COPY1]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s16) = G_LOAD %0 :: (load 2, align 4, addrspace 3)
    %2:_(s32) = G_ANYEXT %1
    $vgpr0 = COPY %2
...

---
name: test_load_local_s16_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s16_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: $vgpr0 = COPY [[COPY1]](s32)
    ; CI-LABEL: name: test_load_local_s16_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: $vgpr0 = COPY [[COPY1]](s32)
    ; CI-DS128-LABEL: name: test_load_local_s16_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: $vgpr0 = COPY [[COPY1]](s32)
    ; VI-LABEL: name: test_load_local_s16_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: $vgpr0 = COPY [[COPY1]](s32)
    ; GFX9-LABEL: name: test_load_local_s16_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: $vgpr0 = COPY [[COPY1]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s16_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[COPY1]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s16) = G_LOAD %0 :: (load 2, align 2, addrspace 3)
    %2:_(s32) = G_ANYEXT %1
    $vgpr0 = COPY %2
...

---
name: test_load_local_s16_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s16_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
    ; SI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; CI-LABEL: name: test_load_local_s16_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
    ; CI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; CI-DS128-LABEL: name: test_load_local_s16_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI-DS128: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI-DS128: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI-DS128: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
    ; CI-DS128: $vgpr0 = COPY [[ANYEXT]](s32)
    ; VI-LABEL: name: test_load_local_s16_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C1]]
    ; VI: [[C2:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C2]](s16)
    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
    ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; GFX9-LABEL: name: test_load_local_s16_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C1]]
    ; GFX9: [[C2:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C2]](s16)
    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
    ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s16_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 1, addrspace 3)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[COPY1]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s16) = G_LOAD %0 :: (load 2, align 1, addrspace 3)
    %2:_(s32) = G_ANYEXT %1
    $vgpr0 = COPY %2
...

---
name: test_load_local_s32_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s32_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; SI: $vgpr0 = COPY [[LOAD]](s32)
    ; CI-LABEL: name: test_load_local_s32_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI: $vgpr0 = COPY [[LOAD]](s32)
    ; CI-DS128-LABEL: name: test_load_local_s32_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI-DS128: $vgpr0 = COPY [[LOAD]](s32)
    ; VI-LABEL: name: test_load_local_s32_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; VI: $vgpr0 = COPY [[LOAD]](s32)
    ; GFX9-LABEL: name: test_load_local_s32_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9: $vgpr0 = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s32_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[LOAD]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s32) = G_LOAD %0 :: (load 4, align 4, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_load_local_s32_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s32_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: $vgpr0 = COPY [[OR]](s32)
    ; CI-LABEL: name: test_load_local_s32_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: $vgpr0 = COPY [[OR]](s32)
    ; CI-DS128-LABEL: name: test_load_local_s32_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: $vgpr0 = COPY [[OR]](s32)
    ; VI-LABEL: name: test_load_local_s32_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: $vgpr0 = COPY [[OR]](s32)
    ; GFX9-LABEL: name: test_load_local_s32_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: $vgpr0 = COPY [[OR]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s32_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, align 2, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[LOAD]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s32) = G_LOAD %0 :: (load 4, align 2, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_load_local_s32_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s32_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; SI: $vgpr0 = COPY [[OR2]](s32)
    ; CI-LABEL: name: test_load_local_s32_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI: $vgpr0 = COPY [[OR2]](s32)
    ; CI-DS128-LABEL: name: test_load_local_s32_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI-DS128: $vgpr0 = COPY [[OR2]](s32)
    ; VI-LABEL: name: test_load_local_s32_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; VI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; VI: $vgpr0 = COPY [[OR2]](s32)
    ; GFX9-LABEL: name: test_load_local_s32_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; GFX9: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; GFX9: $vgpr0 = COPY [[OR2]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s32_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, align 1, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[LOAD]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s32) = G_LOAD %0 :: (load 4, align 1, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_load_local_s24_align8
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s24_align8
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, align 8, addrspace 3)
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: $vgpr0 = COPY [[COPY1]](s32)
    ; CI-LABEL: name: test_load_local_s24_align8
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, align 8, addrspace 3)
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: $vgpr0 = COPY [[COPY1]](s32)
    ; CI-DS128-LABEL: name: test_load_local_s24_align8
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, align 8, addrspace 3)
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: $vgpr0 = COPY [[COPY1]](s32)
    ; VI-LABEL: name: test_load_local_s24_align8
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, align 8, addrspace 3)
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: $vgpr0 = COPY [[COPY1]](s32)
    ; GFX9-LABEL: name: test_load_local_s24_align8
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, align 8, addrspace 3)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: $vgpr0 = COPY [[COPY1]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s24_align8
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, align 8, addrspace 3)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[COPY1]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s24) = G_LOAD %0 :: (load 3, align 8, addrspace 3)
    %2:_(s32) = G_ANYEXT %1
    $vgpr0 = COPY %2
...

---
name: test_load_local_s24_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s24_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: $vgpr0 = COPY [[COPY1]](s32)
    ; CI-LABEL: name: test_load_local_s24_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: $vgpr0 = COPY [[COPY1]](s32)
    ; CI-DS128-LABEL: name: test_load_local_s24_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: $vgpr0 = COPY [[COPY1]](s32)
    ; VI-LABEL: name: test_load_local_s24_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: $vgpr0 = COPY [[COPY1]](s32)
    ; GFX9-LABEL: name: test_load_local_s24_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: $vgpr0 = COPY [[COPY1]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s24_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[COPY1]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s24) = G_LOAD %0 :: (load 3, align 4, addrspace 3)
    %2:_(s32) = G_ANYEXT %1
    $vgpr0 = COPY %2
...

---
name: test_load_local_s24_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s24_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s16) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 2, align 2, addrspace 3)
    ; SI: [[DEF:%[0-9]+]]:_(s24) = G_IMPLICIT_DEF
    ; SI: [[INSERT:%[0-9]+]]:_(s24) = G_INSERT [[DEF]], [[LOAD]](s16), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(s24) = G_INSERT [[INSERT]], [[LOAD1]](s8), 16
    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[INSERT1]](s24)
    ; SI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; CI-LABEL: name: test_load_local_s24_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s16) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 2, align 2, addrspace 3)
    ; CI: [[DEF:%[0-9]+]]:_(s24) = G_IMPLICIT_DEF
    ; CI: [[INSERT:%[0-9]+]]:_(s24) = G_INSERT [[DEF]], [[LOAD]](s16), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(s24) = G_INSERT [[INSERT]], [[LOAD1]](s8), 16
    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[INSERT1]](s24)
    ; CI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; CI-DS128-LABEL: name: test_load_local_s24_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s16) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 2, align 2, addrspace 3)
    ; CI-DS128: [[DEF:%[0-9]+]]:_(s24) = G_IMPLICIT_DEF
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(s24) = G_INSERT [[DEF]], [[LOAD]](s16), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(s24) = G_INSERT [[INSERT]], [[LOAD1]](s8), 16
    ; CI-DS128: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[INSERT1]](s24)
    ; CI-DS128: $vgpr0 = COPY [[ANYEXT]](s32)
    ; VI-LABEL: name: test_load_local_s24_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s16) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 2, align 2, addrspace 3)
    ; VI: [[DEF:%[0-9]+]]:_(s24) = G_IMPLICIT_DEF
    ; VI: [[INSERT:%[0-9]+]]:_(s24) = G_INSERT [[DEF]], [[LOAD]](s16), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(s24) = G_INSERT [[INSERT]], [[LOAD1]](s8), 16
    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[INSERT1]](s24)
    ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; GFX9-LABEL: name: test_load_local_s24_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s16) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 2, align 2, addrspace 3)
    ; GFX9: [[DEF:%[0-9]+]]:_(s24) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT:%[0-9]+]]:_(s24) = G_INSERT [[DEF]], [[LOAD]](s16), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(s24) = G_INSERT [[INSERT]], [[LOAD1]](s8), 16
    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[INSERT1]](s24)
    ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s24_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[ZEXTLOAD:%[0-9]+]]:_(s32) = G_ZEXTLOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9-UNALIGNED: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 2, align 2, addrspace 3)
    ; GFX9-UNALIGNED: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9-UNALIGNED: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[LOAD]], [[C1]](s32)
    ; GFX9-UNALIGNED: [[OR:%[0-9]+]]:_(s32) = G_OR [[SHL]], [[ZEXTLOAD]]
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[OR]](s32)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[COPY1]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s24) = G_LOAD %0 :: (load 3, align 2, addrspace 3)
    %2:_(s32) = G_ANYEXT %1
    $vgpr0 = COPY %2
...

---
name: test_load_local_s24_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s24_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s16) = G_LOAD [[COPY]](p3) :: (load 2, align 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[DEF:%[0-9]+]]:_(s24) = G_IMPLICIT_DEF
    ; SI: [[INSERT:%[0-9]+]]:_(s24) = G_INSERT [[DEF]], [[LOAD]](s16), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(s24) = G_INSERT [[INSERT]], [[LOAD1]](s8), 16
    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[INSERT1]](s24)
    ; SI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; CI-LABEL: name: test_load_local_s24_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s16) = G_LOAD [[COPY]](p3) :: (load 2, align 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[DEF:%[0-9]+]]:_(s24) = G_IMPLICIT_DEF
    ; CI: [[INSERT:%[0-9]+]]:_(s24) = G_INSERT [[DEF]], [[LOAD]](s16), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(s24) = G_INSERT [[INSERT]], [[LOAD1]](s8), 16
    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[INSERT1]](s24)
    ; CI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; CI-DS128-LABEL: name: test_load_local_s24_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s16) = G_LOAD [[COPY]](p3) :: (load 2, align 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[DEF:%[0-9]+]]:_(s24) = G_IMPLICIT_DEF
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(s24) = G_INSERT [[DEF]], [[LOAD]](s16), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(s24) = G_INSERT [[INSERT]], [[LOAD1]](s8), 16
    ; CI-DS128: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[INSERT1]](s24)
    ; CI-DS128: $vgpr0 = COPY [[ANYEXT]](s32)
    ; VI-LABEL: name: test_load_local_s24_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s16) = G_LOAD [[COPY]](p3) :: (load 2, align 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[DEF:%[0-9]+]]:_(s24) = G_IMPLICIT_DEF
    ; VI: [[INSERT:%[0-9]+]]:_(s24) = G_INSERT [[DEF]], [[LOAD]](s16), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(s24) = G_INSERT [[INSERT]], [[LOAD1]](s8), 16
    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[INSERT1]](s24)
    ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; GFX9-LABEL: name: test_load_local_s24_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s16) = G_LOAD [[COPY]](p3) :: (load 2, align 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[DEF:%[0-9]+]]:_(s24) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT:%[0-9]+]]:_(s24) = G_INSERT [[DEF]], [[LOAD]](s16), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(s24) = G_INSERT [[INSERT]], [[LOAD1]](s8), 16
    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[INSERT1]](s24)
    ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s24_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 1, addrspace 3)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
    ; GFX9-UNALIGNED: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9-UNALIGNED: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9-UNALIGNED: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9-UNALIGNED: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9-UNALIGNED: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[LOAD1]], [[C2]](s32)
    ; GFX9-UNALIGNED: [[OR:%[0-9]+]]:_(s32) = G_OR [[SHL]], [[AND]]
    ; GFX9-UNALIGNED: [[COPY2:%[0-9]+]]:_(s32) = COPY [[OR]](s32)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[COPY2]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s24) = G_LOAD %0 :: (load 3, align 1, addrspace 3)
    %2:_(s32) = G_ANYEXT %1
    $vgpr0 = COPY %2
...

---
name: test_load_local_s48_align8
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s48_align8
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; SI: [[COPY1:%[0-9]+]]:_(s64) = COPY [[LOAD]](s64)
    ; SI: $vgpr0_vgpr1 = COPY [[COPY1]](s64)
    ; CI-LABEL: name: test_load_local_s48_align8
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI: [[COPY1:%[0-9]+]]:_(s64) = COPY [[LOAD]](s64)
    ; CI: $vgpr0_vgpr1 = COPY [[COPY1]](s64)
    ; CI-DS128-LABEL: name: test_load_local_s48_align8
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s64) = COPY [[LOAD]](s64)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[COPY1]](s64)
    ; VI-LABEL: name: test_load_local_s48_align8
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; VI: [[COPY1:%[0-9]+]]:_(s64) = COPY [[LOAD]](s64)
    ; VI: $vgpr0_vgpr1 = COPY [[COPY1]](s64)
    ; GFX9-LABEL: name: test_load_local_s48_align8
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s64) = COPY [[LOAD]](s64)
    ; GFX9: $vgpr0_vgpr1 = COPY [[COPY1]](s64)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s48_align8
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s64) = COPY [[LOAD]](s64)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[COPY1]](s64)
    %0:_(p3) = COPY $vgpr0
    %1:_(s48) = G_LOAD %0 :: (load 6, align 8, addrspace 3)
    %2:_(s64) = G_ANYEXT %1
    $vgpr0_vgpr1 = COPY %2
...

---
name: test_load_local_s64_align8
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s64_align8
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; SI: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
    ; CI-LABEL: name: test_load_local_s64_align8
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
    ; CI-DS128-LABEL: name: test_load_local_s64_align8
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
    ; VI-LABEL: name: test_load_local_s64_align8
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; VI: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
    ; GFX9-LABEL: name: test_load_local_s64_align8
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s64_align8
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
    %0:_(p3) = COPY $vgpr0
    %1:_(s64) = G_LOAD %0 :: (load 8, align 8, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_s64_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s64_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; SI: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
    ; CI-LABEL: name: test_load_local_s64_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
    ; CI-DS128-LABEL: name: test_load_local_s64_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
    ; VI-LABEL: name: test_load_local_s64_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; VI: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
    ; GFX9-LABEL: name: test_load_local_s64_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s64_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
    %0:_(p3) = COPY $vgpr0
    %1:_(s64) = G_LOAD %0 :: (load 8, align 4, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_s64_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s64_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C4]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; SI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
    ; SI: $vgpr0_vgpr1 = COPY [[MV]](s64)
    ; CI-LABEL: name: test_load_local_s64_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C4]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
    ; CI: $vgpr0_vgpr1 = COPY [[MV]](s64)
    ; CI-DS128-LABEL: name: test_load_local_s64_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C4]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI-DS128: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[MV]](s64)
    ; VI-LABEL: name: test_load_local_s64_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C4]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; VI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
    ; VI: $vgpr0_vgpr1 = COPY [[MV]](s64)
    ; GFX9-LABEL: name: test_load_local_s64_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C4]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
    ; GFX9: $vgpr0_vgpr1 = COPY [[MV]](s64)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s64_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 2, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
    %0:_(p3) = COPY $vgpr0
    %1:_(s64) = G_LOAD %0 :: (load 8, align 2, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_s64_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s64_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; SI: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; SI: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; SI: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C9]]
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C9]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY3]](s32)
    ; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; SI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C9]]
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY5]](s32)
    ; SI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL2]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; SI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C9]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C8]](s32)
    ; SI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[TRUNC7]]
    ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; SI: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C10]](s32)
    ; SI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C10]](s32)
    ; SI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; SI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; SI: $vgpr0_vgpr1 = COPY [[MV]](s64)
    ; CI-LABEL: name: test_load_local_s64_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; CI: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C9]]
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
    ; CI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C9]]
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY3]](s32)
    ; CI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; CI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C9]]
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY5]](s32)
    ; CI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL2]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; CI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C9]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C8]](s32)
    ; CI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[TRUNC7]]
    ; CI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; CI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; CI: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C10]](s32)
    ; CI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; CI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; CI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; CI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C10]](s32)
    ; CI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; CI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; CI: $vgpr0_vgpr1 = COPY [[MV]](s64)
    ; CI-DS128-LABEL: name: test_load_local_s64_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI-DS128: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI-DS128: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; CI-DS128: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI-DS128: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C9]]
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
    ; CI-DS128: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI-DS128: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C9]]
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY3]](s32)
    ; CI-DS128: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; CI-DS128: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C9]]
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY5]](s32)
    ; CI-DS128: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL2]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; CI-DS128: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C9]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C8]](s32)
    ; CI-DS128: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[TRUNC7]]
    ; CI-DS128: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; CI-DS128: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; CI-DS128: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C10]](s32)
    ; CI-DS128: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; CI-DS128: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; CI-DS128: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; CI-DS128: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C10]](s32)
    ; CI-DS128: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; CI-DS128: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[MV]](s64)
    ; VI-LABEL: name: test_load_local_s64_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; VI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; VI: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C7]]
    ; VI: [[C8:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C8]](s16)
    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s16) = G_AND [[TRUNC3]], [[C7]]
    ; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[AND3]], [[C8]](s16)
    ; VI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[SHL1]]
    ; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s16) = G_AND [[TRUNC5]], [[C7]]
    ; VI: [[SHL2:%[0-9]+]]:_(s16) = G_SHL [[AND5]], [[C8]](s16)
    ; VI: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[SHL2]]
    ; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s16) = G_AND [[TRUNC7]], [[C7]]
    ; VI: [[SHL3:%[0-9]+]]:_(s16) = G_SHL [[AND7]], [[C8]](s16)
    ; VI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[SHL3]]
    ; VI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; VI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; VI: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C9]](s32)
    ; VI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; VI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; VI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; VI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C9]](s32)
    ; VI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; VI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; VI: $vgpr0_vgpr1 = COPY [[MV]](s64)
    ; GFX9-LABEL: name: test_load_local_s64_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; GFX9: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; GFX9: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C7]]
    ; GFX9: [[C8:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C8]](s16)
    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s16) = G_AND [[TRUNC3]], [[C7]]
    ; GFX9: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[AND3]], [[C8]](s16)
    ; GFX9: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[SHL1]]
    ; GFX9: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; GFX9: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s16) = G_AND [[TRUNC5]], [[C7]]
    ; GFX9: [[SHL2:%[0-9]+]]:_(s16) = G_SHL [[AND5]], [[C8]](s16)
    ; GFX9: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[SHL2]]
    ; GFX9: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; GFX9: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s16) = G_AND [[TRUNC7]], [[C7]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s16) = G_SHL [[AND7]], [[C8]](s16)
    ; GFX9: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[SHL3]]
    ; GFX9: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; GFX9: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; GFX9: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C9]](s32)
    ; GFX9: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; GFX9: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; GFX9: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; GFX9: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C9]](s32)
    ; GFX9: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; GFX9: $vgpr0_vgpr1 = COPY [[MV]](s64)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s64_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 1, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
    %0:_(p3) = COPY $vgpr0
    %1:_(s64) = G_LOAD %0 :: (load 8, align 1, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_s96_align16
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s96_align16
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; SI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; SI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; SI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; SI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; SI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; SI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; SI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; SI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; SI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; SI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; SI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; SI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; SI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; SI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; SI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; SI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; SI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; SI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; SI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; SI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; SI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; SI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; SI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; SI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; SI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; SI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; CI-LABEL: name: test_load_local_s96_align16
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; CI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; CI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; CI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; CI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; CI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; CI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; CI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; CI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; CI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; CI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; CI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; CI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; CI-DS128-LABEL: name: test_load_local_s96_align16
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI-DS128: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI-DS128: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI-DS128: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI-DS128: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI-DS128: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI-DS128: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; CI-DS128: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI-DS128: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI-DS128: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI-DS128: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI-DS128: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI-DS128: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI-DS128: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI-DS128: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI-DS128: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; CI-DS128: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI-DS128: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; CI-DS128: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; CI-DS128: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; CI-DS128: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI-DS128: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; CI-DS128: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; CI-DS128: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; CI-DS128: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI-DS128: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; CI-DS128: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; CI-DS128: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; VI-LABEL: name: test_load_local_s96_align16
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; VI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; VI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; VI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; VI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; VI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; VI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; VI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; VI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; VI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; VI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; VI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; VI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; VI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; VI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; VI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; VI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; VI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; VI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; VI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; VI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; VI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; VI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; VI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; VI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; VI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; VI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; VI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; VI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; VI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; VI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; VI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; VI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; VI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; VI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; VI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; GFX9-LABEL: name: test_load_local_s96_align16
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; GFX9: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; GFX9: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; GFX9: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; GFX9: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; GFX9: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; GFX9: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; GFX9: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; GFX9: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; GFX9: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; GFX9: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; GFX9: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; GFX9: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; GFX9: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; GFX9: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; GFX9: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; GFX9: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; GFX9: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; GFX9: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; GFX9: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; GFX9: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; GFX9: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; GFX9: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; GFX9: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; GFX9: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s96_align16
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 12, align 1, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD]](<3 x s32>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    %0:_(p3) = COPY $vgpr0
    %1:_(s96) = G_LOAD %0 :: (load 12, align 1, addrspace 3)
    $vgpr0_vgpr1_vgpr2 = COPY %1
...

---
name: test_load_local_s96_align8
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s96_align8
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, align 8, addrspace 3)
    ; SI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; SI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; SI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; CI-LABEL: name: test_load_local_s96_align8
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, align 8, addrspace 3)
    ; CI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; CI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; CI-DS128-LABEL: name: test_load_local_s96_align8
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, align 8, addrspace 3)
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; VI-LABEL: name: test_load_local_s96_align8
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, align 8, addrspace 3)
    ; VI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; VI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; VI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; GFX9-LABEL: name: test_load_local_s96_align8
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, align 8, addrspace 3)
    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s96_align8
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 12, align 8, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD]](<3 x s32>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    %0:_(p3) = COPY $vgpr0
    %1:_(s96) = G_LOAD %0 :: (load 12, align 8, addrspace 3)
    $vgpr0_vgpr1_vgpr2 = COPY %1
...

---
name: test_load_local_s96_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s96_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; SI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; SI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; SI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; CI-LABEL: name: test_load_local_s96_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; CI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; CI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; CI-DS128-LABEL: name: test_load_local_s96_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; VI-LABEL: name: test_load_local_s96_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; VI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; VI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; VI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; GFX9-LABEL: name: test_load_local_s96_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s96_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 12, align 4, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD]](<3 x s32>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    %0:_(p3) = COPY $vgpr0
    %1:_(s96) = G_LOAD %0 :: (load 12, align 4, addrspace 3)
    $vgpr0_vgpr1_vgpr2 = COPY %1
...

---
name: test_load_local_s96_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s96_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; SI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; SI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR2]](s32), 64
    ; SI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; CI-LABEL: name: test_load_local_s96_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; CI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR2]](s32), 64
    ; CI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; CI-DS128-LABEL: name: test_load_local_s96_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR2]](s32), 64
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; VI-LABEL: name: test_load_local_s96_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; VI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; VI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; VI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR2]](s32), 64
    ; VI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; GFX9-LABEL: name: test_load_local_s96_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR2]](s32), 64
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s96_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 12, align 2, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD]](<3 x s32>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    %0:_(p3) = COPY $vgpr0
    %1:_(s96) = G_LOAD %0 :: (load 12, align 2, addrspace 3)
    $vgpr0_vgpr1_vgpr2 = COPY %1
...

---
name: test_load_local_s96_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s96_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; SI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; SI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; SI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; SI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; SI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; SI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; SI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; SI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; SI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; SI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; SI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; SI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; SI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; SI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; SI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; SI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; SI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; SI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; SI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; SI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; SI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; SI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; SI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; SI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; SI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; SI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; CI-LABEL: name: test_load_local_s96_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; CI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; CI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; CI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; CI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; CI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; CI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; CI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; CI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; CI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; CI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; CI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; CI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; CI-DS128-LABEL: name: test_load_local_s96_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI-DS128: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI-DS128: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI-DS128: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI-DS128: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI-DS128: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI-DS128: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; CI-DS128: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI-DS128: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI-DS128: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI-DS128: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI-DS128: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI-DS128: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI-DS128: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI-DS128: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI-DS128: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; CI-DS128: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI-DS128: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; CI-DS128: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; CI-DS128: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; CI-DS128: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI-DS128: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; CI-DS128: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; CI-DS128: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; CI-DS128: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI-DS128: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; CI-DS128: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; CI-DS128: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; VI-LABEL: name: test_load_local_s96_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; VI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; VI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; VI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; VI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; VI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; VI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; VI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; VI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; VI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; VI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; VI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; VI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; VI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; VI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; VI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; VI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; VI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; VI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; VI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; VI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; VI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; VI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; VI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; VI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; VI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; VI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; VI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; VI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; VI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; VI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; VI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; VI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; VI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; VI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; VI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; GFX9-LABEL: name: test_load_local_s96_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; GFX9: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; GFX9: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; GFX9: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; GFX9: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; GFX9: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; GFX9: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; GFX9: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; GFX9: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; GFX9: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; GFX9: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; GFX9: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; GFX9: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; GFX9: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; GFX9: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; GFX9: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; GFX9: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; GFX9: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; GFX9: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; GFX9: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; GFX9: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; GFX9: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; GFX9: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; GFX9: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; GFX9: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s96_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 12, align 1, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD]](<3 x s32>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2 = COPY [[BITCAST]](s96)
    %0:_(p3) = COPY $vgpr0
    %1:_(s96) = G_LOAD %0 :: (load 12, align 1, addrspace 3)
    $vgpr0_vgpr1_vgpr2 = COPY %1
...

---
name: test_load_local_s128_align16
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s128_align16
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; SI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; SI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; SI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; SI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; SI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; SI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; SI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; SI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; SI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; SI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; SI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; SI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; SI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; SI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; SI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; SI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; SI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; SI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; SI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; SI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; SI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; SI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; SI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; SI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C7]](s32)
    ; SI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; SI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; SI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; SI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; SI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; SI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; SI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; SI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; SI: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; SI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; SI: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; SI: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; SI: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; SI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; SI: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; SI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; SI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; SI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; SI: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; SI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; SI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; SI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR8]](s32), [[OR11]](s32)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; SI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; CI-LABEL: name: test_load_local_s128_align16
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; CI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; CI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; CI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; CI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; CI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; CI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; CI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; CI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; CI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; CI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; CI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C7]](s32)
    ; CI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; CI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; CI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; CI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; CI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; CI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; CI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; CI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; CI: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; CI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; CI: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; CI: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; CI: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; CI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; CI: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; CI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; CI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; CI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; CI: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; CI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; CI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; CI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR8]](s32), [[OR11]](s32)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; CI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; CI-DS128-LABEL: name: test_load_local_s128_align16
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI-DS128: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI-DS128: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI-DS128: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI-DS128: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI-DS128: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI-DS128: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI-DS128: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI-DS128: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI-DS128: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI-DS128: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI-DS128: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI-DS128: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI-DS128: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI-DS128: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI-DS128: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; CI-DS128: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI-DS128: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; CI-DS128: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; CI-DS128: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; CI-DS128: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI-DS128: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; CI-DS128: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; CI-DS128: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; CI-DS128: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI-DS128: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; CI-DS128: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; CI-DS128: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; CI-DS128: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI-DS128: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; CI-DS128: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; CI-DS128: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; CI-DS128: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; CI-DS128: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; CI-DS128: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; CI-DS128: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; CI-DS128: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; CI-DS128: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; CI-DS128: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; CI-DS128: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; CI-DS128: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; CI-DS128: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; CI-DS128: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; CI-DS128: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; CI-DS128: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; CI-DS128: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; CI-DS128: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; CI-DS128: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; CI-DS128: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; CI-DS128: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; CI-DS128: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32), [[OR8]](s32), [[OR11]](s32)
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[BUILD_VECTOR]](<4 x s32>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; VI-LABEL: name: test_load_local_s128_align16
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; VI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; VI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; VI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; VI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; VI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; VI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; VI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; VI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; VI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; VI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; VI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; VI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; VI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; VI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; VI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; VI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; VI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; VI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; VI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; VI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; VI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; VI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; VI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; VI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; VI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; VI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; VI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; VI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; VI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; VI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; VI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; VI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; VI: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; VI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; VI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; VI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; VI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; VI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; VI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; VI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; VI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; VI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; VI: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; VI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; VI: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; VI: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; VI: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; VI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; VI: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; VI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; VI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; VI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; VI: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; VI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; VI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32), [[OR8]](s32), [[OR11]](s32)
    ; VI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[BUILD_VECTOR]](<4 x s32>)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; GFX9-LABEL: name: test_load_local_s128_align16
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; GFX9: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; GFX9: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; GFX9: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; GFX9: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; GFX9: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; GFX9: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; GFX9: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; GFX9: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; GFX9: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; GFX9: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; GFX9: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; GFX9: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; GFX9: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; GFX9: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; GFX9: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; GFX9: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; GFX9: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; GFX9: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; GFX9: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; GFX9: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; GFX9: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; GFX9: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; GFX9: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; GFX9: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; GFX9: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; GFX9: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; GFX9: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; GFX9: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; GFX9: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; GFX9: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; GFX9: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; GFX9: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; GFX9: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; GFX9: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; GFX9: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; GFX9: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; GFX9: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; GFX9: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; GFX9: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; GFX9: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; GFX9: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; GFX9: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; GFX9: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; GFX9: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; GFX9: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32), [[OR8]](s32), [[OR11]](s32)
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[BUILD_VECTOR]](<4 x s32>)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s128_align16
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 1, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[LOAD]](<4 x s32>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    %0:_(p3) = COPY $vgpr0
    %1:_(s128) = G_LOAD %0 :: (load 16, align 1, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_load_local_s128_align8
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s128_align8
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, addrspace 3)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[LOAD]](<2 x s32>), [[LOAD1]](<2 x s32>)
    ; SI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; CI-LABEL: name: test_load_local_s128_align8
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, addrspace 3)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[LOAD]](<2 x s32>), [[LOAD1]](<2 x s32>)
    ; CI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; CI-DS128-LABEL: name: test_load_local_s128_align8
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 8, addrspace 3)
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[LOAD]](<4 x s32>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; VI-LABEL: name: test_load_local_s128_align8
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 8, addrspace 3)
    ; VI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[LOAD]](<4 x s32>)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; GFX9-LABEL: name: test_load_local_s128_align8
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 8, addrspace 3)
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[LOAD]](<4 x s32>)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s128_align8
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 8, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[LOAD]](<4 x s32>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    %0:_(p3) = COPY $vgpr0
    %1:_(s128) = G_LOAD %0 :: (load 16, align 8, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_load_local_s128_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s128_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, align 4, addrspace 3)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[LOAD]](<2 x s32>), [[LOAD1]](<2 x s32>)
    ; SI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; CI-LABEL: name: test_load_local_s128_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, align 4, addrspace 3)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[LOAD]](<2 x s32>), [[LOAD1]](<2 x s32>)
    ; CI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; CI-DS128-LABEL: name: test_load_local_s128_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 4, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 4 + 8, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 12, addrspace 3)
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[LOAD]](s32), [[LOAD1]](s32), [[LOAD2]](s32), [[LOAD3]](s32)
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[BUILD_VECTOR]](<4 x s32>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; VI-LABEL: name: test_load_local_s128_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 4, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 4 + 8, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 12, addrspace 3)
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[LOAD]](s32), [[LOAD1]](s32), [[LOAD2]](s32), [[LOAD3]](s32)
    ; VI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[BUILD_VECTOR]](<4 x s32>)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; GFX9-LABEL: name: test_load_local_s128_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 4, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 4 + 8, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 12, addrspace 3)
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[LOAD]](s32), [[LOAD1]](s32), [[LOAD2]](s32), [[LOAD3]](s32)
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[BUILD_VECTOR]](<4 x s32>)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s128_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 4, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[LOAD]](<4 x s32>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    %0:_(p3) = COPY $vgpr0
    %1:_(s128) = G_LOAD %0 :: (load 16, align 4, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_load_local_s128_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s128_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C3]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C1]]
    ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; SI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR3]](s32)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; SI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; CI-LABEL: name: test_load_local_s128_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C3]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C1]]
    ; CI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; CI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR3]](s32)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; CI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; CI-DS128-LABEL: name: test_load_local_s128_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C3]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C1]]
    ; CI-DS128: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; CI-DS128: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR3]](s32)
    ; CI-DS128: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; VI-LABEL: name: test_load_local_s128_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; VI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C3]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; VI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C1]]
    ; VI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; VI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; VI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; VI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR3]](s32)
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; VI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; GFX9-LABEL: name: test_load_local_s128_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C3]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C1]]
    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; GFX9: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR3]](s32)
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s128_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 2, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[LOAD]](<4 x s32>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    %0:_(p3) = COPY $vgpr0
    %1:_(s128) = G_LOAD %0 :: (load 16, align 2, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_load_local_s128_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_s128_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; SI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; SI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; SI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; SI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; SI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; SI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; SI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; SI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; SI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; SI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; SI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; SI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; SI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; SI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; SI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; SI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; SI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; SI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; SI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; SI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; SI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; SI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; SI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; SI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C7]](s32)
    ; SI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; SI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; SI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; SI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; SI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; SI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; SI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; SI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; SI: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; SI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; SI: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; SI: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; SI: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; SI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; SI: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; SI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; SI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; SI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; SI: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; SI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; SI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; SI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR8]](s32), [[OR11]](s32)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; SI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; CI-LABEL: name: test_load_local_s128_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; CI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; CI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; CI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; CI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; CI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; CI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; CI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; CI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; CI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; CI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; CI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C7]](s32)
    ; CI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; CI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; CI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; CI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; CI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; CI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; CI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; CI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; CI: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; CI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; CI: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; CI: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; CI: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; CI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; CI: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; CI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; CI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; CI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; CI: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; CI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; CI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; CI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR8]](s32), [[OR11]](s32)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; CI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; CI-DS128-LABEL: name: test_load_local_s128_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI-DS128: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI-DS128: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI-DS128: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI-DS128: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI-DS128: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI-DS128: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI-DS128: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI-DS128: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI-DS128: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI-DS128: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI-DS128: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI-DS128: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI-DS128: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI-DS128: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI-DS128: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; CI-DS128: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI-DS128: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; CI-DS128: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; CI-DS128: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; CI-DS128: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI-DS128: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; CI-DS128: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; CI-DS128: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; CI-DS128: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI-DS128: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; CI-DS128: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; CI-DS128: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; CI-DS128: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI-DS128: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; CI-DS128: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; CI-DS128: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; CI-DS128: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; CI-DS128: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; CI-DS128: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; CI-DS128: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; CI-DS128: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; CI-DS128: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; CI-DS128: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; CI-DS128: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; CI-DS128: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; CI-DS128: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; CI-DS128: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; CI-DS128: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; CI-DS128: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; CI-DS128: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; CI-DS128: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; CI-DS128: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; CI-DS128: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; CI-DS128: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; CI-DS128: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32), [[OR8]](s32), [[OR11]](s32)
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[BUILD_VECTOR]](<4 x s32>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; VI-LABEL: name: test_load_local_s128_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; VI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; VI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; VI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; VI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; VI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; VI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; VI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; VI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; VI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; VI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; VI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; VI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; VI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; VI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; VI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; VI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; VI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; VI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; VI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; VI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; VI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; VI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; VI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; VI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; VI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; VI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; VI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; VI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; VI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; VI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; VI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; VI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; VI: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; VI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; VI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; VI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; VI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; VI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; VI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; VI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; VI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; VI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; VI: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; VI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; VI: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; VI: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; VI: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; VI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; VI: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; VI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; VI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; VI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; VI: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; VI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; VI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32), [[OR8]](s32), [[OR11]](s32)
    ; VI: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[BUILD_VECTOR]](<4 x s32>)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; GFX9-LABEL: name: test_load_local_s128_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; GFX9: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; GFX9: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; GFX9: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; GFX9: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; GFX9: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; GFX9: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; GFX9: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; GFX9: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; GFX9: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; GFX9: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; GFX9: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; GFX9: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; GFX9: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; GFX9: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; GFX9: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; GFX9: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; GFX9: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; GFX9: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; GFX9: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; GFX9: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; GFX9: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; GFX9: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; GFX9: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; GFX9: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; GFX9: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; GFX9: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; GFX9: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; GFX9: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; GFX9: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; GFX9: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; GFX9: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; GFX9: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; GFX9: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; GFX9: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; GFX9: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; GFX9: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; GFX9: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; GFX9: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; GFX9: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; GFX9: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; GFX9: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; GFX9: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; GFX9: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; GFX9: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; GFX9: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32), [[OR8]](s32), [[OR11]](s32)
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[BUILD_VECTOR]](<4 x s32>)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_s128_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 1, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(s128) = G_BITCAST [[LOAD]](<4 x s32>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](s128)
    %0:_(p3) = COPY $vgpr0
    %1:_(s128) = G_LOAD %0 :: (load 16, align 1, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_load_local_p1_align8
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_p1_align8
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; SI: $vgpr0_vgpr1 = COPY [[LOAD]](p1)
    ; CI-LABEL: name: test_load_local_p1_align8
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI: $vgpr0_vgpr1 = COPY [[LOAD]](p1)
    ; CI-DS128-LABEL: name: test_load_local_p1_align8
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[LOAD]](p1)
    ; VI-LABEL: name: test_load_local_p1_align8
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; VI: $vgpr0_vgpr1 = COPY [[LOAD]](p1)
    ; GFX9-LABEL: name: test_load_local_p1_align8
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](p1)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_p1_align8
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](p1)
    %0:_(p3) = COPY $vgpr0
    %1:_(p1) = G_LOAD %0 :: (load 8, align 8, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_p1_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_p1_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; SI: $vgpr0_vgpr1 = COPY [[LOAD]](p1)
    ; CI-LABEL: name: test_load_local_p1_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI: $vgpr0_vgpr1 = COPY [[LOAD]](p1)
    ; CI-DS128-LABEL: name: test_load_local_p1_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[LOAD]](p1)
    ; VI-LABEL: name: test_load_local_p1_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; VI: $vgpr0_vgpr1 = COPY [[LOAD]](p1)
    ; GFX9-LABEL: name: test_load_local_p1_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](p1)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_p1_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](p1)
    %0:_(p3) = COPY $vgpr0
    %1:_(p1) = G_LOAD %0 :: (load 8, align 4, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_p1_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_p1_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C4]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; SI: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
    ; SI: $vgpr0_vgpr1 = COPY [[MV]](p1)
    ; CI-LABEL: name: test_load_local_p1_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C4]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
    ; CI: $vgpr0_vgpr1 = COPY [[MV]](p1)
    ; CI-DS128-LABEL: name: test_load_local_p1_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C4]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI-DS128: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[MV]](p1)
    ; VI-LABEL: name: test_load_local_p1_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C4]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; VI: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
    ; VI: $vgpr0_vgpr1 = COPY [[MV]](p1)
    ; GFX9-LABEL: name: test_load_local_p1_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C4]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; GFX9: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
    ; GFX9: $vgpr0_vgpr1 = COPY [[MV]](p1)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_p1_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[COPY]](p3) :: (load 8, align 2, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](p1)
    %0:_(p3) = COPY $vgpr0
    %1:_(p1) = G_LOAD %0 :: (load 8, align 2, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_p1_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_p1_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; SI: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; SI: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; SI: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C9]]
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C9]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY3]](s32)
    ; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; SI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C9]]
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY5]](s32)
    ; SI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL2]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; SI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C9]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C8]](s32)
    ; SI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[TRUNC7]]
    ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; SI: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C10]](s32)
    ; SI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C10]](s32)
    ; SI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; SI: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; SI: $vgpr0_vgpr1 = COPY [[MV]](p1)
    ; CI-LABEL: name: test_load_local_p1_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; CI: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C9]]
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
    ; CI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C9]]
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY3]](s32)
    ; CI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; CI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C9]]
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY5]](s32)
    ; CI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL2]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; CI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C9]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C8]](s32)
    ; CI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[TRUNC7]]
    ; CI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; CI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; CI: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C10]](s32)
    ; CI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; CI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; CI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; CI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C10]](s32)
    ; CI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; CI: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; CI: $vgpr0_vgpr1 = COPY [[MV]](p1)
    ; CI-DS128-LABEL: name: test_load_local_p1_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI-DS128: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI-DS128: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; CI-DS128: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI-DS128: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C9]]
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
    ; CI-DS128: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI-DS128: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C9]]
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY3]](s32)
    ; CI-DS128: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; CI-DS128: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C9]]
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY5]](s32)
    ; CI-DS128: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL2]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; CI-DS128: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C9]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C8]](s32)
    ; CI-DS128: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[TRUNC7]]
    ; CI-DS128: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; CI-DS128: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; CI-DS128: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C10]](s32)
    ; CI-DS128: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; CI-DS128: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; CI-DS128: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; CI-DS128: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C10]](s32)
    ; CI-DS128: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; CI-DS128: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[MV]](p1)
    ; VI-LABEL: name: test_load_local_p1_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; VI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; VI: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C7]]
    ; VI: [[C8:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C8]](s16)
    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s16) = G_AND [[TRUNC3]], [[C7]]
    ; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[AND3]], [[C8]](s16)
    ; VI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[SHL1]]
    ; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s16) = G_AND [[TRUNC5]], [[C7]]
    ; VI: [[SHL2:%[0-9]+]]:_(s16) = G_SHL [[AND5]], [[C8]](s16)
    ; VI: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[SHL2]]
    ; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s16) = G_AND [[TRUNC7]], [[C7]]
    ; VI: [[SHL3:%[0-9]+]]:_(s16) = G_SHL [[AND7]], [[C8]](s16)
    ; VI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[SHL3]]
    ; VI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; VI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; VI: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C9]](s32)
    ; VI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; VI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; VI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; VI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C9]](s32)
    ; VI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; VI: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; VI: $vgpr0_vgpr1 = COPY [[MV]](p1)
    ; GFX9-LABEL: name: test_load_local_p1_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; GFX9: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; GFX9: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C7]]
    ; GFX9: [[C8:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C8]](s16)
    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s16) = G_AND [[TRUNC3]], [[C7]]
    ; GFX9: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[AND3]], [[C8]](s16)
    ; GFX9: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[SHL1]]
    ; GFX9: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; GFX9: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s16) = G_AND [[TRUNC5]], [[C7]]
    ; GFX9: [[SHL2:%[0-9]+]]:_(s16) = G_SHL [[AND5]], [[C8]](s16)
    ; GFX9: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[SHL2]]
    ; GFX9: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; GFX9: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s16) = G_AND [[TRUNC7]], [[C7]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s16) = G_SHL [[AND7]], [[C8]](s16)
    ; GFX9: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[SHL3]]
    ; GFX9: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; GFX9: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; GFX9: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C9]](s32)
    ; GFX9: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; GFX9: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; GFX9: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; GFX9: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C9]](s32)
    ; GFX9: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; GFX9: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; GFX9: $vgpr0_vgpr1 = COPY [[MV]](p1)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_p1_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[COPY]](p3) :: (load 8, align 1, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](p1)
    %0:_(p3) = COPY $vgpr0
    %1:_(p1) = G_LOAD %0 :: (load 8, align 1, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_p3_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_p3_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(p3) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; SI: $vgpr0 = COPY [[LOAD]](p3)
    ; CI-LABEL: name: test_load_local_p3_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(p3) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI: $vgpr0 = COPY [[LOAD]](p3)
    ; CI-DS128-LABEL: name: test_load_local_p3_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(p3) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI-DS128: $vgpr0 = COPY [[LOAD]](p3)
    ; VI-LABEL: name: test_load_local_p3_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(p3) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; VI: $vgpr0 = COPY [[LOAD]](p3)
    ; GFX9-LABEL: name: test_load_local_p3_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(p3) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9: $vgpr0 = COPY [[LOAD]](p3)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_p3_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(p3) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[LOAD]](p3)
    %0:_(p3) = COPY $vgpr0
    %1:_(p3) = G_LOAD %0 :: (load 4, align 4, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_load_local_p3_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_p3_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[INTTOPTR:%[0-9]+]]:_(p3) = G_INTTOPTR [[OR]](s32)
    ; SI: $vgpr0 = COPY [[INTTOPTR]](p3)
    ; CI-LABEL: name: test_load_local_p3_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[INTTOPTR:%[0-9]+]]:_(p3) = G_INTTOPTR [[OR]](s32)
    ; CI: $vgpr0 = COPY [[INTTOPTR]](p3)
    ; CI-DS128-LABEL: name: test_load_local_p3_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[INTTOPTR:%[0-9]+]]:_(p3) = G_INTTOPTR [[OR]](s32)
    ; CI-DS128: $vgpr0 = COPY [[INTTOPTR]](p3)
    ; VI-LABEL: name: test_load_local_p3_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[INTTOPTR:%[0-9]+]]:_(p3) = G_INTTOPTR [[OR]](s32)
    ; VI: $vgpr0 = COPY [[INTTOPTR]](p3)
    ; GFX9-LABEL: name: test_load_local_p3_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[INTTOPTR:%[0-9]+]]:_(p3) = G_INTTOPTR [[OR]](s32)
    ; GFX9: $vgpr0 = COPY [[INTTOPTR]](p3)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_p3_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(p3) = G_LOAD [[COPY]](p3) :: (load 4, align 2, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[LOAD]](p3)
    %0:_(p3) = COPY $vgpr0
    %1:_(p3) = G_LOAD %0 :: (load 4, align 2, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_load_local_p3_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_p3_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; SI: [[INTTOPTR:%[0-9]+]]:_(p3) = G_INTTOPTR [[OR2]](s32)
    ; SI: $vgpr0 = COPY [[INTTOPTR]](p3)
    ; CI-LABEL: name: test_load_local_p3_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI: [[INTTOPTR:%[0-9]+]]:_(p3) = G_INTTOPTR [[OR2]](s32)
    ; CI: $vgpr0 = COPY [[INTTOPTR]](p3)
    ; CI-DS128-LABEL: name: test_load_local_p3_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI-DS128: [[INTTOPTR:%[0-9]+]]:_(p3) = G_INTTOPTR [[OR2]](s32)
    ; CI-DS128: $vgpr0 = COPY [[INTTOPTR]](p3)
    ; VI-LABEL: name: test_load_local_p3_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; VI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; VI: [[INTTOPTR:%[0-9]+]]:_(p3) = G_INTTOPTR [[OR2]](s32)
    ; VI: $vgpr0 = COPY [[INTTOPTR]](p3)
    ; GFX9-LABEL: name: test_load_local_p3_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; GFX9: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; GFX9: [[INTTOPTR:%[0-9]+]]:_(p3) = G_INTTOPTR [[OR2]](s32)
    ; GFX9: $vgpr0 = COPY [[INTTOPTR]](p3)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_p3_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(p3) = G_LOAD [[COPY]](p3) :: (load 4, align 1, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[LOAD]](p3)
    %0:_(p3) = COPY $vgpr0
    %1:_(p3) = G_LOAD %0 :: (load 4, align 1, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_load_local_p5_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_p5_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(p5) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; SI: $vgpr0 = COPY [[LOAD]](p5)
    ; CI-LABEL: name: test_load_local_p5_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(p5) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI: $vgpr0 = COPY [[LOAD]](p5)
    ; CI-DS128-LABEL: name: test_load_local_p5_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(p5) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI-DS128: $vgpr0 = COPY [[LOAD]](p5)
    ; VI-LABEL: name: test_load_local_p5_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(p5) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; VI: $vgpr0 = COPY [[LOAD]](p5)
    ; GFX9-LABEL: name: test_load_local_p5_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(p5) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9: $vgpr0 = COPY [[LOAD]](p5)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_p5_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(p5) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[LOAD]](p5)
    %0:_(p3) = COPY $vgpr0
    %1:_(p5) = G_LOAD %0 :: (load 4, align 4, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_load_local_p5_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_p5_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[INTTOPTR:%[0-9]+]]:_(p5) = G_INTTOPTR [[OR]](s32)
    ; SI: $vgpr0 = COPY [[INTTOPTR]](p5)
    ; CI-LABEL: name: test_load_local_p5_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[INTTOPTR:%[0-9]+]]:_(p5) = G_INTTOPTR [[OR]](s32)
    ; CI: $vgpr0 = COPY [[INTTOPTR]](p5)
    ; CI-DS128-LABEL: name: test_load_local_p5_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[INTTOPTR:%[0-9]+]]:_(p5) = G_INTTOPTR [[OR]](s32)
    ; CI-DS128: $vgpr0 = COPY [[INTTOPTR]](p5)
    ; VI-LABEL: name: test_load_local_p5_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[INTTOPTR:%[0-9]+]]:_(p5) = G_INTTOPTR [[OR]](s32)
    ; VI: $vgpr0 = COPY [[INTTOPTR]](p5)
    ; GFX9-LABEL: name: test_load_local_p5_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[INTTOPTR:%[0-9]+]]:_(p5) = G_INTTOPTR [[OR]](s32)
    ; GFX9: $vgpr0 = COPY [[INTTOPTR]](p5)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_p5_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(p5) = G_LOAD [[COPY]](p3) :: (load 4, align 2, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[LOAD]](p5)
    %0:_(p3) = COPY $vgpr0
    %1:_(p5) = G_LOAD %0 :: (load 4, align 2, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_load_local_p5_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_p5_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; SI: [[INTTOPTR:%[0-9]+]]:_(p5) = G_INTTOPTR [[OR2]](s32)
    ; SI: $vgpr0 = COPY [[INTTOPTR]](p5)
    ; CI-LABEL: name: test_load_local_p5_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI: [[INTTOPTR:%[0-9]+]]:_(p5) = G_INTTOPTR [[OR2]](s32)
    ; CI: $vgpr0 = COPY [[INTTOPTR]](p5)
    ; CI-DS128-LABEL: name: test_load_local_p5_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI-DS128: [[INTTOPTR:%[0-9]+]]:_(p5) = G_INTTOPTR [[OR2]](s32)
    ; CI-DS128: $vgpr0 = COPY [[INTTOPTR]](p5)
    ; VI-LABEL: name: test_load_local_p5_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; VI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; VI: [[INTTOPTR:%[0-9]+]]:_(p5) = G_INTTOPTR [[OR2]](s32)
    ; VI: $vgpr0 = COPY [[INTTOPTR]](p5)
    ; GFX9-LABEL: name: test_load_local_p5_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; GFX9: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; GFX9: [[INTTOPTR:%[0-9]+]]:_(p5) = G_INTTOPTR [[OR2]](s32)
    ; GFX9: $vgpr0 = COPY [[INTTOPTR]](p5)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_p5_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(p5) = G_LOAD [[COPY]](p3) :: (load 4, align 1, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[LOAD]](p5)
    %0:_(p3) = COPY $vgpr0
    %1:_(p5) = G_LOAD %0 :: (load 4, align 1, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_load_local_v2s8_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v2s8_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
    ; SI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
    ; SI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; CI-LABEL: name: test_load_local_v2s8_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
    ; CI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
    ; CI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
    ; CI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; CI-DS128-LABEL: name: test_load_local_v2s8_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI-DS128: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI-DS128: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
    ; CI-DS128: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI-DS128: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
    ; CI-DS128: $vgpr0 = COPY [[ANYEXT]](s32)
    ; VI-LABEL: name: test_load_local_v2s8_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
    ; VI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
    ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
    ; GFX9-LABEL: name: test_load_local_v2s8_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
    ; GFX9: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
    ; GFX9: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
    ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v2s8_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9-UNALIGNED: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
    ; GFX9-UNALIGNED: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9-UNALIGNED: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
    ; GFX9-UNALIGNED: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9-UNALIGNED: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
    ; GFX9-UNALIGNED: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; GFX9-UNALIGNED: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; GFX9-UNALIGNED: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
    ; GFX9-UNALIGNED: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
    ; GFX9-UNALIGNED: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
    ; GFX9-UNALIGNED: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; GFX9-UNALIGNED: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
    ; GFX9-UNALIGNED: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; GFX9-UNALIGNED: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[ANYEXT]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s8>) = G_LOAD %0 :: (load 2, align 2, addrspace 3)
    %2:_(s16) = G_BITCAST %1
    %3:_(s32) = G_ANYEXT %2
    $vgpr0 = COPY %3
...

---
name: test_load_local_v2s8_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v2s8_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32)
    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; CI-LABEL: name: test_load_local_v2s8_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32)
    ; CI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; CI-DS128-LABEL: name: test_load_local_v2s8_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; VI-LABEL: name: test_load_local_v2s8_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY3]](s32), [[COPY4]](s32)
    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; GFX9-LABEL: name: test_load_local_v2s8_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32)
    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v2s8_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 1, addrspace 3)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9-UNALIGNED: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
    ; GFX9-UNALIGNED: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9-UNALIGNED: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
    ; GFX9-UNALIGNED: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9-UNALIGNED: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s8>) = G_LOAD %0 :: (load 2, align 1, addrspace 3)
    %2:_(<2 x s32>) = G_ANYEXT %1
    $vgpr0_vgpr1 = COPY %2
...

---
name: test_load_local_v3s8_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v3s8_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<3 x s8>) = G_LOAD [[COPY]](p3) :: (load 3, align 4, addrspace 1)
    ; SI: [[DEF:%[0-9]+]]:_(<4 x s8>) = G_IMPLICIT_DEF
    ; SI: [[INSERT:%[0-9]+]]:_(<4 x s8>) = G_INSERT [[DEF]], [[LOAD]](<3 x s8>), 0
    ; SI: $vgpr0 = COPY [[INSERT]](<4 x s8>)
    ; CI-LABEL: name: test_load_local_v3s8_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<3 x s8>) = G_LOAD [[COPY]](p3) :: (load 3, align 4, addrspace 1)
    ; CI: [[DEF:%[0-9]+]]:_(<4 x s8>) = G_IMPLICIT_DEF
    ; CI: [[INSERT:%[0-9]+]]:_(<4 x s8>) = G_INSERT [[DEF]], [[LOAD]](<3 x s8>), 0
    ; CI: $vgpr0 = COPY [[INSERT]](<4 x s8>)
    ; CI-DS128-LABEL: name: test_load_local_v3s8_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<3 x s8>) = G_LOAD [[COPY]](p3) :: (load 3, align 4, addrspace 1)
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<4 x s8>) = G_IMPLICIT_DEF
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<4 x s8>) = G_INSERT [[DEF]], [[LOAD]](<3 x s8>), 0
    ; CI-DS128: $vgpr0 = COPY [[INSERT]](<4 x s8>)
    ; VI-LABEL: name: test_load_local_v3s8_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<3 x s8>) = G_LOAD [[COPY]](p3) :: (load 3, align 4, addrspace 1)
    ; VI: [[DEF:%[0-9]+]]:_(<4 x s8>) = G_IMPLICIT_DEF
    ; VI: [[INSERT:%[0-9]+]]:_(<4 x s8>) = G_INSERT [[DEF]], [[LOAD]](<3 x s8>), 0
    ; VI: $vgpr0 = COPY [[INSERT]](<4 x s8>)
    ; GFX9-LABEL: name: test_load_local_v3s8_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<3 x s8>) = G_LOAD [[COPY]](p3) :: (load 3, align 4, addrspace 1)
    ; GFX9: [[DEF:%[0-9]+]]:_(<4 x s8>) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT:%[0-9]+]]:_(<4 x s8>) = G_INSERT [[DEF]], [[LOAD]](<3 x s8>), 0
    ; GFX9: $vgpr0 = COPY [[INSERT]](<4 x s8>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v3s8_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s8>) = G_LOAD [[COPY]](p3) :: (load 3, align 4, addrspace 1)
    ; GFX9-UNALIGNED: [[DEF:%[0-9]+]]:_(<4 x s8>) = G_IMPLICIT_DEF
    ; GFX9-UNALIGNED: [[INSERT:%[0-9]+]]:_(<4 x s8>) = G_INSERT [[DEF]], [[LOAD]](<3 x s8>), 0
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[INSERT]](<4 x s8>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<3 x s8>) = G_LOAD %0 :: (load 3, addrspace 1, align 4)
    %2:_(<4 x s8>) = G_IMPLICIT_DEF
    %3:_(<4 x s8>) = G_INSERT %2, %1, 0
    $vgpr0 = COPY %3
...

---
name: test_load_local_v3s8_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v3s8_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<3 x s8>) = G_LOAD [[COPY]](p3) :: (load 3, align 1, addrspace 3)
    ; SI: [[DEF:%[0-9]+]]:_(<4 x s8>) = G_IMPLICIT_DEF
    ; SI: [[INSERT:%[0-9]+]]:_(<4 x s8>) = G_INSERT [[DEF]], [[LOAD]](<3 x s8>), 0
    ; SI: $vgpr0 = COPY [[INSERT]](<4 x s8>)
    ; CI-LABEL: name: test_load_local_v3s8_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<3 x s8>) = G_LOAD [[COPY]](p3) :: (load 3, align 1, addrspace 3)
    ; CI: [[DEF:%[0-9]+]]:_(<4 x s8>) = G_IMPLICIT_DEF
    ; CI: [[INSERT:%[0-9]+]]:_(<4 x s8>) = G_INSERT [[DEF]], [[LOAD]](<3 x s8>), 0
    ; CI: $vgpr0 = COPY [[INSERT]](<4 x s8>)
    ; CI-DS128-LABEL: name: test_load_local_v3s8_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<3 x s8>) = G_LOAD [[COPY]](p3) :: (load 3, align 1, addrspace 3)
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<4 x s8>) = G_IMPLICIT_DEF
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<4 x s8>) = G_INSERT [[DEF]], [[LOAD]](<3 x s8>), 0
    ; CI-DS128: $vgpr0 = COPY [[INSERT]](<4 x s8>)
    ; VI-LABEL: name: test_load_local_v3s8_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<3 x s8>) = G_LOAD [[COPY]](p3) :: (load 3, align 1, addrspace 3)
    ; VI: [[DEF:%[0-9]+]]:_(<4 x s8>) = G_IMPLICIT_DEF
    ; VI: [[INSERT:%[0-9]+]]:_(<4 x s8>) = G_INSERT [[DEF]], [[LOAD]](<3 x s8>), 0
    ; VI: $vgpr0 = COPY [[INSERT]](<4 x s8>)
    ; GFX9-LABEL: name: test_load_local_v3s8_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<3 x s8>) = G_LOAD [[COPY]](p3) :: (load 3, align 1, addrspace 3)
    ; GFX9: [[DEF:%[0-9]+]]:_(<4 x s8>) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT:%[0-9]+]]:_(<4 x s8>) = G_INSERT [[DEF]], [[LOAD]](<3 x s8>), 0
    ; GFX9: $vgpr0 = COPY [[INSERT]](<4 x s8>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v3s8_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s8>) = G_LOAD [[COPY]](p3) :: (load 3, align 1, addrspace 3)
    ; GFX9-UNALIGNED: [[DEF:%[0-9]+]]:_(<4 x s8>) = G_IMPLICIT_DEF
    ; GFX9-UNALIGNED: [[INSERT:%[0-9]+]]:_(<4 x s8>) = G_INSERT [[DEF]], [[LOAD]](<3 x s8>), 0
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[INSERT]](<4 x s8>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<3 x s8>) = G_LOAD %0 :: (load 3, align 1, addrspace 3)
    %2:_(<4 x s8>) = G_IMPLICIT_DEF
    %3:_(<4 x s8>) = G_INSERT %2, %1, 0
    $vgpr0 = COPY %3
...

---
name: test_load_local_v4s8_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v4s8_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32)
    ; SI: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR]](<4 x s32>)
    ; SI: $vgpr0 = COPY [[TRUNC]](<4 x s8>)
    ; CI-LABEL: name: test_load_local_v4s8_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32)
    ; CI: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR]](<4 x s32>)
    ; CI: $vgpr0 = COPY [[TRUNC]](<4 x s8>)
    ; CI-DS128-LABEL: name: test_load_local_v4s8_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI-DS128: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32)
    ; CI-DS128: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR]](<4 x s32>)
    ; CI-DS128: $vgpr0 = COPY [[TRUNC]](<4 x s8>)
    ; VI-LABEL: name: test_load_local_v4s8_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32)
    ; VI: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR]](<4 x s32>)
    ; VI: $vgpr0 = COPY [[TRUNC]](<4 x s8>)
    ; GFX9-LABEL: name: test_load_local_v4s8_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY1]](s32), [[COPY2]](s32)
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY3]](s32), [[COPY4]](s32)
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
    ; GFX9: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[CONCAT_VECTORS]](<4 x s16>)
    ; GFX9: $vgpr0 = COPY [[TRUNC]](<4 x s8>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v4s8_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9-UNALIGNED: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
    ; GFX9-UNALIGNED: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9-UNALIGNED: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
    ; GFX9-UNALIGNED: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9-UNALIGNED: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY1]](s32), [[COPY2]](s32)
    ; GFX9-UNALIGNED: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
    ; GFX9-UNALIGNED: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY3]](s32), [[COPY4]](s32)
    ; GFX9-UNALIGNED: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
    ; GFX9-UNALIGNED: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[CONCAT_VECTORS]](<4 x s16>)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[TRUNC]](<4 x s8>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<4 x s8>) = G_LOAD %0 :: (load 4, align 4, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_load_local_v8s8_align8
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v8s8_align8
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD]](<2 x s32>)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C]](s32)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C1]](s32)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C2]](s32)
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32)
    ; SI: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR]](<4 x s32>)
    ; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C]](s32)
    ; SI: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C1]](s32)
    ; SI: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C2]](s32)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
    ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
    ; SI: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
    ; SI: [[TRUNC1:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR1]](<4 x s32>)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s8>) = G_CONCAT_VECTORS [[TRUNC]](<4 x s8>), [[TRUNC1]](<4 x s8>)
    ; SI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<8 x s8>)
    ; CI-LABEL: name: test_load_local_v8s8_align8
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD]](<2 x s32>)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C]](s32)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C1]](s32)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C2]](s32)
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32)
    ; CI: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR]](<4 x s32>)
    ; CI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C]](s32)
    ; CI: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C1]](s32)
    ; CI: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C2]](s32)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
    ; CI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
    ; CI: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
    ; CI: [[TRUNC1:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR1]](<4 x s32>)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s8>) = G_CONCAT_VECTORS [[TRUNC]](<4 x s8>), [[TRUNC1]](<4 x s8>)
    ; CI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<8 x s8>)
    ; CI-DS128-LABEL: name: test_load_local_v8s8_align8
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI-DS128: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD]](<2 x s32>)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C]](s32)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C1]](s32)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI-DS128: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C2]](s32)
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32)
    ; CI-DS128: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR]](<4 x s32>)
    ; CI-DS128: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C]](s32)
    ; CI-DS128: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C1]](s32)
    ; CI-DS128: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C2]](s32)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
    ; CI-DS128: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
    ; CI-DS128: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
    ; CI-DS128: [[TRUNC1:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR1]](<4 x s32>)
    ; CI-DS128: [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s8>) = G_CONCAT_VECTORS [[TRUNC]](<4 x s8>), [[TRUNC1]](<4 x s8>)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<8 x s8>)
    ; VI-LABEL: name: test_load_local_v8s8_align8
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD]](<2 x s32>)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C]](s32)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C1]](s32)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C2]](s32)
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32)
    ; VI: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR]](<4 x s32>)
    ; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C]](s32)
    ; VI: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C1]](s32)
    ; VI: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C2]](s32)
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
    ; VI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
    ; VI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
    ; VI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
    ; VI: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
    ; VI: [[TRUNC1:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR1]](<4 x s32>)
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s8>) = G_CONCAT_VECTORS [[TRUNC]](<4 x s8>), [[TRUNC1]](<4 x s8>)
    ; VI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<8 x s8>)
    ; GFX9-LABEL: name: test_load_local_v8s8_align8
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD]](<2 x s32>)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C]](s32)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C1]](s32)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C2]](s32)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY1]](s32), [[COPY2]](s32)
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY3]](s32), [[COPY4]](s32)
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
    ; GFX9: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[CONCAT_VECTORS]](<4 x s16>)
    ; GFX9: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C]](s32)
    ; GFX9: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C1]](s32)
    ; GFX9: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C2]](s32)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY5]](s32), [[COPY6]](s32)
    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY7]](s32), [[COPY8]](s32)
    ; GFX9: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>)
    ; GFX9: [[TRUNC1:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[CONCAT_VECTORS1]](<4 x s16>)
    ; GFX9: [[CONCAT_VECTORS2:%[0-9]+]]:_(<8 x s8>) = G_CONCAT_VECTORS [[TRUNC]](<4 x s8>), [[TRUNC1]](<4 x s8>)
    ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS2]](<8 x s8>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v8s8_align8
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9-UNALIGNED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD]](<2 x s32>)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9-UNALIGNED: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C]](s32)
    ; GFX9-UNALIGNED: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9-UNALIGNED: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C1]](s32)
    ; GFX9-UNALIGNED: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9-UNALIGNED: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C2]](s32)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
    ; GFX9-UNALIGNED: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY1]](s32), [[COPY2]](s32)
    ; GFX9-UNALIGNED: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
    ; GFX9-UNALIGNED: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY3]](s32), [[COPY4]](s32)
    ; GFX9-UNALIGNED: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
    ; GFX9-UNALIGNED: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[CONCAT_VECTORS]](<4 x s16>)
    ; GFX9-UNALIGNED: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C]](s32)
    ; GFX9-UNALIGNED: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C1]](s32)
    ; GFX9-UNALIGNED: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C2]](s32)
    ; GFX9-UNALIGNED: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
    ; GFX9-UNALIGNED: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY5]](s32), [[COPY6]](s32)
    ; GFX9-UNALIGNED: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
    ; GFX9-UNALIGNED: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY7]](s32), [[COPY8]](s32)
    ; GFX9-UNALIGNED: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>)
    ; GFX9-UNALIGNED: [[TRUNC1:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[CONCAT_VECTORS1]](<4 x s16>)
    ; GFX9-UNALIGNED: [[CONCAT_VECTORS2:%[0-9]+]]:_(<8 x s8>) = G_CONCAT_VECTORS [[TRUNC]](<4 x s8>), [[TRUNC1]](<4 x s8>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS2]](<8 x s8>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<8 x s8>) = G_LOAD %0 :: (load 8, align 8, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_v16s8_align16
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v16s8_align16
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; SI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; SI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; SI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; SI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; SI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; SI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; SI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C3]](s32)
    ; SI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; SI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; SI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; SI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; SI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; SI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; SI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32)
    ; SI: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR]](<4 x s32>)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
    ; SI: [[TRUNC1:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR1]](<4 x s32>)
    ; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; SI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; SI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; SI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; SI: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32), [[COPY12]](s32)
    ; SI: [[TRUNC2:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR2]](<4 x s32>)
    ; SI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; SI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; SI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; SI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; SI: [[BUILD_VECTOR3:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY13]](s32), [[COPY14]](s32), [[COPY15]](s32), [[COPY16]](s32)
    ; SI: [[TRUNC3:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR3]](<4 x s32>)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<16 x s8>) = G_CONCAT_VECTORS [[TRUNC]](<4 x s8>), [[TRUNC1]](<4 x s8>), [[TRUNC2]](<4 x s8>), [[TRUNC3]](<4 x s8>)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<16 x s8>)
    ; CI-LABEL: name: test_load_local_v16s8_align16
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C3]](s32)
    ; CI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; CI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; CI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; CI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; CI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; CI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; CI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32)
    ; CI: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR]](<4 x s32>)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
    ; CI: [[TRUNC1:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR1]](<4 x s32>)
    ; CI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32), [[COPY12]](s32)
    ; CI: [[TRUNC2:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR2]](<4 x s32>)
    ; CI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; CI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; CI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; CI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; CI: [[BUILD_VECTOR3:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY13]](s32), [[COPY14]](s32), [[COPY15]](s32), [[COPY16]](s32)
    ; CI: [[TRUNC3:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR3]](<4 x s32>)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<16 x s8>) = G_CONCAT_VECTORS [[TRUNC]](<4 x s8>), [[TRUNC1]](<4 x s8>), [[TRUNC2]](<4 x s8>), [[TRUNC3]](<4 x s8>)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<16 x s8>)
    ; CI-DS128-LABEL: name: test_load_local_v16s8_align16
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI-DS128: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI-DS128: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI-DS128: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI-DS128: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI-DS128: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI-DS128: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI-DS128: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; CI-DS128: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; CI-DS128: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; CI-DS128: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; CI-DS128: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; CI-DS128: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; CI-DS128: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; CI-DS128: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32)
    ; CI-DS128: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR]](<4 x s32>)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI-DS128: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
    ; CI-DS128: [[TRUNC1:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR1]](<4 x s32>)
    ; CI-DS128: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI-DS128: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI-DS128: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI-DS128: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI-DS128: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32), [[COPY12]](s32)
    ; CI-DS128: [[TRUNC2:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR2]](<4 x s32>)
    ; CI-DS128: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; CI-DS128: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; CI-DS128: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; CI-DS128: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; CI-DS128: [[BUILD_VECTOR3:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY13]](s32), [[COPY14]](s32), [[COPY15]](s32), [[COPY16]](s32)
    ; CI-DS128: [[TRUNC3:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR3]](<4 x s32>)
    ; CI-DS128: [[CONCAT_VECTORS:%[0-9]+]]:_(<16 x s8>) = G_CONCAT_VECTORS [[TRUNC]](<4 x s8>), [[TRUNC1]](<4 x s8>), [[TRUNC2]](<4 x s8>), [[TRUNC3]](<4 x s8>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<16 x s8>)
    ; VI-LABEL: name: test_load_local_v16s8_align16
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; VI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; VI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; VI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; VI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; VI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; VI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; VI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; VI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; VI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; VI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; VI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; VI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; VI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; VI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32)
    ; VI: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR]](<4 x s32>)
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; VI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; VI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; VI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; VI: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
    ; VI: [[TRUNC1:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR1]](<4 x s32>)
    ; VI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; VI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; VI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; VI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; VI: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32), [[COPY12]](s32)
    ; VI: [[TRUNC2:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR2]](<4 x s32>)
    ; VI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; VI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; VI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; VI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; VI: [[BUILD_VECTOR3:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY13]](s32), [[COPY14]](s32), [[COPY15]](s32), [[COPY16]](s32)
    ; VI: [[TRUNC3:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[BUILD_VECTOR3]](<4 x s32>)
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<16 x s8>) = G_CONCAT_VECTORS [[TRUNC]](<4 x s8>), [[TRUNC1]](<4 x s8>), [[TRUNC2]](<4 x s8>), [[TRUNC3]](<4 x s8>)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<16 x s8>)
    ; GFX9-LABEL: name: test_load_local_v16s8_align16
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; GFX9: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; GFX9: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; GFX9: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; GFX9: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; GFX9: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; GFX9: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; GFX9: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; GFX9: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; GFX9: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; GFX9: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; GFX9: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; GFX9: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; GFX9: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY1]](s32), [[COPY2]](s32)
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY3]](s32), [[COPY4]](s32)
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
    ; GFX9: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[CONCAT_VECTORS]](<4 x s16>)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY5]](s32), [[COPY6]](s32)
    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY7]](s32), [[COPY8]](s32)
    ; GFX9: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>)
    ; GFX9: [[TRUNC1:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[CONCAT_VECTORS1]](<4 x s16>)
    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC4:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32)
    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC5:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32)
    ; GFX9: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC4]](<2 x s16>), [[BUILD_VECTOR_TRUNC5]](<2 x s16>)
    ; GFX9: [[TRUNC2:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[CONCAT_VECTORS2]](<4 x s16>)
    ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; GFX9: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC6:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
    ; GFX9: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; GFX9: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC7:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
    ; GFX9: [[CONCAT_VECTORS3:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC6]](<2 x s16>), [[BUILD_VECTOR_TRUNC7]](<2 x s16>)
    ; GFX9: [[TRUNC3:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[CONCAT_VECTORS3]](<4 x s16>)
    ; GFX9: [[CONCAT_VECTORS4:%[0-9]+]]:_(<16 x s8>) = G_CONCAT_VECTORS [[TRUNC]](<4 x s8>), [[TRUNC1]](<4 x s8>), [[TRUNC2]](<4 x s8>), [[TRUNC3]](<4 x s8>)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS4]](<16 x s8>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v16s8_align16
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 1, addrspace 3)
    ; GFX9-UNALIGNED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD]](<4 x s32>)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9-UNALIGNED: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C]](s32)
    ; GFX9-UNALIGNED: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9-UNALIGNED: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C1]](s32)
    ; GFX9-UNALIGNED: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9-UNALIGNED: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C2]](s32)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
    ; GFX9-UNALIGNED: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY1]](s32), [[COPY2]](s32)
    ; GFX9-UNALIGNED: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
    ; GFX9-UNALIGNED: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY3]](s32), [[COPY4]](s32)
    ; GFX9-UNALIGNED: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
    ; GFX9-UNALIGNED: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[CONCAT_VECTORS]](<4 x s16>)
    ; GFX9-UNALIGNED: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C]](s32)
    ; GFX9-UNALIGNED: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C1]](s32)
    ; GFX9-UNALIGNED: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C2]](s32)
    ; GFX9-UNALIGNED: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
    ; GFX9-UNALIGNED: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY5]](s32), [[COPY6]](s32)
    ; GFX9-UNALIGNED: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
    ; GFX9-UNALIGNED: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY7]](s32), [[COPY8]](s32)
    ; GFX9-UNALIGNED: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>)
    ; GFX9-UNALIGNED: [[TRUNC1:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[CONCAT_VECTORS1]](<4 x s16>)
    ; GFX9-UNALIGNED: [[LSHR6:%[0-9]+]]:_(s32) = G_LSHR [[UV2]], [[C]](s32)
    ; GFX9-UNALIGNED: [[LSHR7:%[0-9]+]]:_(s32) = G_LSHR [[UV2]], [[C1]](s32)
    ; GFX9-UNALIGNED: [[LSHR8:%[0-9]+]]:_(s32) = G_LSHR [[UV2]], [[C2]](s32)
    ; GFX9-UNALIGNED: [[COPY9:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
    ; GFX9-UNALIGNED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR6]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR_TRUNC4:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32)
    ; GFX9-UNALIGNED: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR7]](s32)
    ; GFX9-UNALIGNED: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LSHR8]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR_TRUNC5:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32)
    ; GFX9-UNALIGNED: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC4]](<2 x s16>), [[BUILD_VECTOR_TRUNC5]](<2 x s16>)
    ; GFX9-UNALIGNED: [[TRUNC2:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[CONCAT_VECTORS2]](<4 x s16>)
    ; GFX9-UNALIGNED: [[LSHR9:%[0-9]+]]:_(s32) = G_LSHR [[UV3]], [[C]](s32)
    ; GFX9-UNALIGNED: [[LSHR10:%[0-9]+]]:_(s32) = G_LSHR [[UV3]], [[C1]](s32)
    ; GFX9-UNALIGNED: [[LSHR11:%[0-9]+]]:_(s32) = G_LSHR [[UV3]], [[C2]](s32)
    ; GFX9-UNALIGNED: [[COPY13:%[0-9]+]]:_(s32) = COPY [[UV3]](s32)
    ; GFX9-UNALIGNED: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR9]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR_TRUNC6:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
    ; GFX9-UNALIGNED: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR10]](s32)
    ; GFX9-UNALIGNED: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LSHR11]](s32)
    ; GFX9-UNALIGNED: [[BUILD_VECTOR_TRUNC7:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
    ; GFX9-UNALIGNED: [[CONCAT_VECTORS3:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC6]](<2 x s16>), [[BUILD_VECTOR_TRUNC7]](<2 x s16>)
    ; GFX9-UNALIGNED: [[TRUNC3:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[CONCAT_VECTORS3]](<4 x s16>)
    ; GFX9-UNALIGNED: [[CONCAT_VECTORS4:%[0-9]+]]:_(<16 x s8>) = G_CONCAT_VECTORS [[TRUNC]](<4 x s8>), [[TRUNC1]](<4 x s8>), [[TRUNC2]](<4 x s8>), [[TRUNC3]](<4 x s8>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS4]](<16 x s8>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<16 x s8>) = G_LOAD %0 :: (load 16, align 1, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_load_local_v2s16_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v2s16_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s16>) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; SI: $vgpr0 = COPY [[LOAD]](<2 x s16>)
    ; CI-LABEL: name: test_load_local_v2s16_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s16>) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI: $vgpr0 = COPY [[LOAD]](<2 x s16>)
    ; CI-DS128-LABEL: name: test_load_local_v2s16_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<2 x s16>) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI-DS128: $vgpr0 = COPY [[LOAD]](<2 x s16>)
    ; VI-LABEL: name: test_load_local_v2s16_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<2 x s16>) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; VI: $vgpr0 = COPY [[LOAD]](<2 x s16>)
    ; GFX9-LABEL: name: test_load_local_v2s16_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<2 x s16>) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9: $vgpr0 = COPY [[LOAD]](<2 x s16>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v2s16_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s16>) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[LOAD]](<2 x s16>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s16>) = G_LOAD %0 :: (load 4, align 4, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_load_local_v2s16_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v2s16_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; SI: $vgpr0 = COPY [[BITCAST]](<2 x s16>)
    ; CI-LABEL: name: test_load_local_v2s16_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; CI: $vgpr0 = COPY [[BITCAST]](<2 x s16>)
    ; CI-DS128-LABEL: name: test_load_local_v2s16_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; CI-DS128: $vgpr0 = COPY [[BITCAST]](<2 x s16>)
    ; VI-LABEL: name: test_load_local_v2s16_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; VI: $vgpr0 = COPY [[BITCAST]](<2 x s16>)
    ; GFX9-LABEL: name: test_load_local_v2s16_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY1]](s32), [[COPY2]](s32)
    ; GFX9: $vgpr0 = COPY [[BUILD_VECTOR_TRUNC]](<2 x s16>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v2s16_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s16>) = G_LOAD [[COPY]](p3) :: (load 4, align 2, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[LOAD]](<2 x s16>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s16>) = G_LOAD %0 :: (load 4, align 2, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_load_local_v2s16_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v2s16_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY2]](s32)
    ; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C5]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL2]]
    ; SI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
    ; SI: $vgpr0 = COPY [[BITCAST]](<2 x s16>)
    ; CI-LABEL: name: test_load_local_v2s16_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY2]](s32)
    ; CI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; CI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; CI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C5]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL2]]
    ; CI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
    ; CI: $vgpr0 = COPY [[BITCAST]](<2 x s16>)
    ; CI-DS128-LABEL: name: test_load_local_v2s16_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI-DS128: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI-DS128: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY2]](s32)
    ; CI-DS128: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; CI-DS128: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; CI-DS128: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C5]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL2]]
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
    ; CI-DS128: $vgpr0 = COPY [[BITCAST]](<2 x s16>)
    ; VI-LABEL: name: test_load_local_v2s16_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C1]]
    ; VI: [[C2:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C2]](s16)
    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s16) = G_AND [[TRUNC3]], [[C1]]
    ; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[AND3]], [[C2]](s16)
    ; VI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[SHL1]]
    ; VI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; VI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C4]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL2]]
    ; VI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
    ; VI: $vgpr0 = COPY [[BITCAST]](<2 x s16>)
    ; GFX9-LABEL: name: test_load_local_v2s16_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C1]]
    ; GFX9: [[C2:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C2]](s16)
    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s16) = G_AND [[TRUNC3]], [[C1]]
    ; GFX9: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[AND3]], [[C2]](s16)
    ; GFX9: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[SHL1]]
    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
    ; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[OR1]](s16)
    ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[ANYEXT]](s32), [[ANYEXT1]](s32)
    ; GFX9: $vgpr0 = COPY [[BUILD_VECTOR_TRUNC]](<2 x s16>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v2s16_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s16>) = G_LOAD [[COPY]](p3) :: (load 4, align 1, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[LOAD]](<2 x s16>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s16>) = G_LOAD %0 :: (load 4, align 1, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_load_local_v3s16_align8
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v3s16_align8
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[LOAD]](<4 x s16>), [[DEF]](<4 x s16>), [[DEF]](<4 x s16>)
    ; SI: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[UV]](<3 x s16>), 0
    ; SI: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>)
    ; CI-LABEL: name: test_load_local_v3s16_align8
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[LOAD]](<4 x s16>), [[DEF]](<4 x s16>), [[DEF]](<4 x s16>)
    ; CI: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; CI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[UV]](<3 x s16>), 0
    ; CI: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>)
    ; CI-DS128-LABEL: name: test_load_local_v3s16_align8
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; CI-DS128: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[LOAD]](<4 x s16>), [[DEF]](<4 x s16>), [[DEF]](<4 x s16>)
    ; CI-DS128: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[UV]](<3 x s16>), 0
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>)
    ; VI-LABEL: name: test_load_local_v3s16_align8
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[LOAD]](<4 x s16>), [[DEF]](<4 x s16>), [[DEF]](<4 x s16>)
    ; VI: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[UV]](<3 x s16>), 0
    ; VI: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>)
    ; GFX9-LABEL: name: test_load_local_v3s16_align8
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[LOAD]](<4 x s16>), [[DEF]](<4 x s16>), [[DEF]](<4 x s16>)
    ; GFX9: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; GFX9: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[UV]](<3 x s16>), 0
    ; GFX9: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v3s16_align8
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9-UNALIGNED: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; GFX9-UNALIGNED: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[LOAD]](<4 x s16>), [[DEF]](<4 x s16>), [[DEF]](<4 x s16>)
    ; GFX9-UNALIGNED: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; GFX9-UNALIGNED: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[UV]](<3 x s16>), 0
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<3 x s16>) = G_LOAD %0 :: (load 6, align 8, addrspace 3)
    %2:_(<4 x s16>) = G_IMPLICIT_DEF
    %3:_(<4 x s16>) = G_INSERT %2, %1, 0
    $vgpr0_vgpr1 = COPY %3
...

---
name: test_load_local_v3s16_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v3s16_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; SI: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT]], [[BITCAST]](<2 x s16>), 0
    ; SI: [[CONCAT_VECTORS1:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT1]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; SI: [[UV4:%[0-9]+]]:_(<3 x s16>), [[UV5:%[0-9]+]]:_(<3 x s16>), [[UV6:%[0-9]+]]:_(<3 x s16>), [[UV7:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS1]](<12 x s16>)
    ; SI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV4]](<3 x s16>), 0
    ; SI: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT2]], [[TRUNC]](s16), 32
    ; SI: [[CONCAT_VECTORS2:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT3]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; SI: [[UV8:%[0-9]+]]:_(<3 x s16>), [[UV9:%[0-9]+]]:_(<3 x s16>), [[UV10:%[0-9]+]]:_(<3 x s16>), [[UV11:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS2]](<12 x s16>)
    ; SI: [[INSERT4:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV8]](<3 x s16>), 0
    ; SI: $vgpr0_vgpr1 = COPY [[INSERT4]](<4 x s16>)
    ; CI-LABEL: name: test_load_local_v3s16_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; CI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; CI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; CI: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; CI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT]], [[BITCAST]](<2 x s16>), 0
    ; CI: [[CONCAT_VECTORS1:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT1]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; CI: [[UV4:%[0-9]+]]:_(<3 x s16>), [[UV5:%[0-9]+]]:_(<3 x s16>), [[UV6:%[0-9]+]]:_(<3 x s16>), [[UV7:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS1]](<12 x s16>)
    ; CI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV4]](<3 x s16>), 0
    ; CI: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT2]], [[TRUNC]](s16), 32
    ; CI: [[CONCAT_VECTORS2:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT3]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; CI: [[UV8:%[0-9]+]]:_(<3 x s16>), [[UV9:%[0-9]+]]:_(<3 x s16>), [[UV10:%[0-9]+]]:_(<3 x s16>), [[UV11:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS2]](<12 x s16>)
    ; CI: [[INSERT4:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV8]](<3 x s16>), 0
    ; CI: $vgpr0_vgpr1 = COPY [[INSERT4]](<4 x s16>)
    ; CI-DS128-LABEL: name: test_load_local_v3s16_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI-DS128: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; CI-DS128: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; CI-DS128: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; CI-DS128: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT]], [[BITCAST]](<2 x s16>), 0
    ; CI-DS128: [[CONCAT_VECTORS1:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT1]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; CI-DS128: [[UV4:%[0-9]+]]:_(<3 x s16>), [[UV5:%[0-9]+]]:_(<3 x s16>), [[UV6:%[0-9]+]]:_(<3 x s16>), [[UV7:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS1]](<12 x s16>)
    ; CI-DS128: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV4]](<3 x s16>), 0
    ; CI-DS128: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT2]], [[TRUNC]](s16), 32
    ; CI-DS128: [[CONCAT_VECTORS2:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT3]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; CI-DS128: [[UV8:%[0-9]+]]:_(<3 x s16>), [[UV9:%[0-9]+]]:_(<3 x s16>), [[UV10:%[0-9]+]]:_(<3 x s16>), [[UV11:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS2]](<12 x s16>)
    ; CI-DS128: [[INSERT4:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV8]](<3 x s16>), 0
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[INSERT4]](<4 x s16>)
    ; VI-LABEL: name: test_load_local_v3s16_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; VI: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT]], [[BITCAST]](<2 x s16>), 0
    ; VI: [[CONCAT_VECTORS1:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT1]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; VI: [[UV4:%[0-9]+]]:_(<3 x s16>), [[UV5:%[0-9]+]]:_(<3 x s16>), [[UV6:%[0-9]+]]:_(<3 x s16>), [[UV7:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS1]](<12 x s16>)
    ; VI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV4]](<3 x s16>), 0
    ; VI: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT2]], [[TRUNC]](s16), 32
    ; VI: [[CONCAT_VECTORS2:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT3]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; VI: [[UV8:%[0-9]+]]:_(<3 x s16>), [[UV9:%[0-9]+]]:_(<3 x s16>), [[UV10:%[0-9]+]]:_(<3 x s16>), [[UV11:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS2]](<12 x s16>)
    ; VI: [[INSERT4:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV8]](<3 x s16>), 0
    ; VI: $vgpr0_vgpr1 = COPY [[INSERT4]](<4 x s16>)
    ; GFX9-LABEL: name: test_load_local_v3s16_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY1]](s32), [[COPY2]](s32)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; GFX9: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; GFX9: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; GFX9: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; GFX9: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT]], [[BUILD_VECTOR_TRUNC]](<2 x s16>), 0
    ; GFX9: [[CONCAT_VECTORS1:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT1]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; GFX9: [[UV4:%[0-9]+]]:_(<3 x s16>), [[UV5:%[0-9]+]]:_(<3 x s16>), [[UV6:%[0-9]+]]:_(<3 x s16>), [[UV7:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS1]](<12 x s16>)
    ; GFX9: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV4]](<3 x s16>), 0
    ; GFX9: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT2]], [[TRUNC]](s16), 32
    ; GFX9: [[CONCAT_VECTORS2:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT3]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; GFX9: [[UV8:%[0-9]+]]:_(<3 x s16>), [[UV9:%[0-9]+]]:_(<3 x s16>), [[UV10:%[0-9]+]]:_(<3 x s16>), [[UV11:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS2]](<12 x s16>)
    ; GFX9: [[INSERT4:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV8]](<3 x s16>), 0
    ; GFX9: $vgpr0_vgpr1 = COPY [[INSERT4]](<4 x s16>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v3s16_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s16>) = G_LOAD [[COPY]](p3) :: (load 6, align 2, addrspace 3)
    ; GFX9-UNALIGNED: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; GFX9-UNALIGNED: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[LOAD]](<3 x s16>), 0
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<3 x s16>) = G_LOAD %0 :: (load 6, align 2, addrspace 3)
    %2:_(<4 x s16>) = G_IMPLICIT_DEF
    %3:_(<4 x s16>) = G_INSERT %2, %1, 0
    $vgpr0_vgpr1 = COPY %3
...

---
name: test_load_local_v3s16_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v3s16_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY2]](s32)
    ; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C5]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL2]]
    ; SI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; SI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY4]](s32)
    ; SI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; SI: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT]], [[BITCAST]](<2 x s16>), 0
    ; SI: [[CONCAT_VECTORS1:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT1]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; SI: [[UV4:%[0-9]+]]:_(<3 x s16>), [[UV5:%[0-9]+]]:_(<3 x s16>), [[UV6:%[0-9]+]]:_(<3 x s16>), [[UV7:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS1]](<12 x s16>)
    ; SI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV4]](<3 x s16>), 0
    ; SI: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT2]], [[OR3]](s16), 32
    ; SI: [[CONCAT_VECTORS2:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT3]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; SI: [[UV8:%[0-9]+]]:_(<3 x s16>), [[UV9:%[0-9]+]]:_(<3 x s16>), [[UV10:%[0-9]+]]:_(<3 x s16>), [[UV11:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS2]](<12 x s16>)
    ; SI: [[INSERT4:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV8]](<3 x s16>), 0
    ; SI: $vgpr0_vgpr1 = COPY [[INSERT4]](<4 x s16>)
    ; CI-LABEL: name: test_load_local_v3s16_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY2]](s32)
    ; CI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; CI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; CI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C5]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL2]]
    ; CI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY4]](s32)
    ; CI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; CI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; CI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; CI: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; CI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT]], [[BITCAST]](<2 x s16>), 0
    ; CI: [[CONCAT_VECTORS1:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT1]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; CI: [[UV4:%[0-9]+]]:_(<3 x s16>), [[UV5:%[0-9]+]]:_(<3 x s16>), [[UV6:%[0-9]+]]:_(<3 x s16>), [[UV7:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS1]](<12 x s16>)
    ; CI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV4]](<3 x s16>), 0
    ; CI: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT2]], [[OR3]](s16), 32
    ; CI: [[CONCAT_VECTORS2:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT3]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; CI: [[UV8:%[0-9]+]]:_(<3 x s16>), [[UV9:%[0-9]+]]:_(<3 x s16>), [[UV10:%[0-9]+]]:_(<3 x s16>), [[UV11:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS2]](<12 x s16>)
    ; CI: [[INSERT4:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV8]](<3 x s16>), 0
    ; CI: $vgpr0_vgpr1 = COPY [[INSERT4]](<4 x s16>)
    ; CI-DS128-LABEL: name: test_load_local_v3s16_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI-DS128: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI-DS128: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY2]](s32)
    ; CI-DS128: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; CI-DS128: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; CI-DS128: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C5]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL2]]
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI-DS128: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY4]](s32)
    ; CI-DS128: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; CI-DS128: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; CI-DS128: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; CI-DS128: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT]], [[BITCAST]](<2 x s16>), 0
    ; CI-DS128: [[CONCAT_VECTORS1:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT1]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; CI-DS128: [[UV4:%[0-9]+]]:_(<3 x s16>), [[UV5:%[0-9]+]]:_(<3 x s16>), [[UV6:%[0-9]+]]:_(<3 x s16>), [[UV7:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS1]](<12 x s16>)
    ; CI-DS128: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV4]](<3 x s16>), 0
    ; CI-DS128: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT2]], [[OR3]](s16), 32
    ; CI-DS128: [[CONCAT_VECTORS2:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT3]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; CI-DS128: [[UV8:%[0-9]+]]:_(<3 x s16>), [[UV9:%[0-9]+]]:_(<3 x s16>), [[UV10:%[0-9]+]]:_(<3 x s16>), [[UV11:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS2]](<12 x s16>)
    ; CI-DS128: [[INSERT4:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV8]](<3 x s16>), 0
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[INSERT4]](<4 x s16>)
    ; VI-LABEL: name: test_load_local_v3s16_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C1]]
    ; VI: [[C2:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C2]](s16)
    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s16) = G_AND [[TRUNC3]], [[C1]]
    ; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[AND3]], [[C2]](s16)
    ; VI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[SHL1]]
    ; VI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; VI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C4]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL2]]
    ; VI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C1]]
    ; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s16) = G_AND [[TRUNC5]], [[C1]]
    ; VI: [[SHL3:%[0-9]+]]:_(s16) = G_SHL [[AND5]], [[C2]](s16)
    ; VI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[SHL3]]
    ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; VI: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT]], [[BITCAST]](<2 x s16>), 0
    ; VI: [[CONCAT_VECTORS1:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT1]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; VI: [[UV4:%[0-9]+]]:_(<3 x s16>), [[UV5:%[0-9]+]]:_(<3 x s16>), [[UV6:%[0-9]+]]:_(<3 x s16>), [[UV7:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS1]](<12 x s16>)
    ; VI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV4]](<3 x s16>), 0
    ; VI: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT2]], [[OR3]](s16), 32
    ; VI: [[CONCAT_VECTORS2:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT3]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; VI: [[UV8:%[0-9]+]]:_(<3 x s16>), [[UV9:%[0-9]+]]:_(<3 x s16>), [[UV10:%[0-9]+]]:_(<3 x s16>), [[UV11:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS2]](<12 x s16>)
    ; VI: [[INSERT4:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV8]](<3 x s16>), 0
    ; VI: $vgpr0_vgpr1 = COPY [[INSERT4]](<4 x s16>)
    ; GFX9-LABEL: name: test_load_local_v3s16_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C1]]
    ; GFX9: [[C2:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C2]](s16)
    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s16) = G_AND [[TRUNC3]], [[C1]]
    ; GFX9: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[AND3]], [[C2]](s16)
    ; GFX9: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[SHL1]]
    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
    ; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[OR1]](s16)
    ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[ANYEXT]](s32), [[ANYEXT1]](s32)
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; GFX9: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C1]]
    ; GFX9: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s16) = G_AND [[TRUNC5]], [[C1]]
    ; GFX9: [[SHL2:%[0-9]+]]:_(s16) = G_SHL [[AND5]], [[C2]](s16)
    ; GFX9: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[SHL2]]
    ; GFX9: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; GFX9: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; GFX9: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
    ; GFX9: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT]], [[BUILD_VECTOR_TRUNC]](<2 x s16>), 0
    ; GFX9: [[CONCAT_VECTORS1:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT1]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; GFX9: [[UV4:%[0-9]+]]:_(<3 x s16>), [[UV5:%[0-9]+]]:_(<3 x s16>), [[UV6:%[0-9]+]]:_(<3 x s16>), [[UV7:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS1]](<12 x s16>)
    ; GFX9: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV4]](<3 x s16>), 0
    ; GFX9: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT2]], [[OR2]](s16), 32
    ; GFX9: [[CONCAT_VECTORS2:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[INSERT3]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
    ; GFX9: [[UV8:%[0-9]+]]:_(<3 x s16>), [[UV9:%[0-9]+]]:_(<3 x s16>), [[UV10:%[0-9]+]]:_(<3 x s16>), [[UV11:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS2]](<12 x s16>)
    ; GFX9: [[INSERT4:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV8]](<3 x s16>), 0
    ; GFX9: $vgpr0_vgpr1 = COPY [[INSERT4]](<4 x s16>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v3s16_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s16>) = G_LOAD [[COPY]](p3) :: (load 6, align 1, addrspace 3)
    ; GFX9-UNALIGNED: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
    ; GFX9-UNALIGNED: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[LOAD]](<3 x s16>), 0
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<3 x s16>) = G_LOAD %0 :: (load 6, align 1, addrspace 3)
    %2:_(<4 x s16>) = G_IMPLICIT_DEF
    %3:_(<4 x s16>) = G_INSERT %2, %1, 0
    $vgpr0_vgpr1 = COPY %3
...

---
name: test_load_local_v4s16_align8
body: |
  bb.0:
    liveins: $vgpr0
    ; SI-LABEL: name: test_load_local_v4s16_align8
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; SI: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
    ; CI-LABEL: name: test_load_local_v4s16_align8
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
    ; CI-DS128-LABEL: name: test_load_local_v4s16_align8
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
    ; VI-LABEL: name: test_load_local_v4s16_align8
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; VI: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
    ; GFX9-LABEL: name: test_load_local_v4s16_align8
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v4s16_align8
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<4 x s16>) = G_LOAD %0 :: (load 8, align 8, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_v4s16_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v4s16_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; SI: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
    ; CI-LABEL: name: test_load_local_v4s16_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
    ; CI-DS128-LABEL: name: test_load_local_v4s16_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
    ; VI-LABEL: name: test_load_local_v4s16_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; VI: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
    ; GFX9-LABEL: name: test_load_local_v4s16_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v4s16_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<4 x s16>) = G_LOAD %0 :: (load 8, align 4, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_v4s16_align2
body: |
  bb.0:
    liveins: $vgpr0
    ; SI-LABEL: name: test_load_local_v4s16_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C4]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; SI: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>)
    ; SI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    ; CI-LABEL: name: test_load_local_v4s16_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C4]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>)
    ; CI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    ; CI-DS128-LABEL: name: test_load_local_v4s16_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C4]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI-DS128: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
    ; CI-DS128: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    ; VI-LABEL: name: test_load_local_v4s16_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C4]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; VI: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>)
    ; VI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    ; GFX9-LABEL: name: test_load_local_v4s16_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY1]](s32), [[COPY2]](s32)
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY3]](s32), [[COPY4]](s32)
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
    ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v4s16_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, align 2, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<4 x s16>) = G_LOAD %0 :: (load 8, align 2, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_v4s16_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v4s16_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY2]](s32)
    ; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C5]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL2]]
    ; SI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; SI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY4]](s32)
    ; SI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C4]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; SI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C1]]
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[COPY6]](s32)
    ; SI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[SHL4]](s32)
    ; SI: [[OR4:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[TRUNC7]]
    ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR4]](s16)
    ; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C5]](s32)
    ; SI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; SI: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR5]](s32)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>)
    ; SI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    ; CI-LABEL: name: test_load_local_v4s16_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY2]](s32)
    ; CI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; CI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; CI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C5]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL2]]
    ; CI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY4]](s32)
    ; CI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C4]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C1]]
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[COPY6]](s32)
    ; CI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[SHL4]](s32)
    ; CI: [[OR4:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[TRUNC7]]
    ; CI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; CI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR4]](s16)
    ; CI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C5]](s32)
    ; CI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; CI: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR5]](s32)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>)
    ; CI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    ; CI-DS128-LABEL: name: test_load_local_v4s16_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI-DS128: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI-DS128: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY2]](s32)
    ; CI-DS128: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; CI-DS128: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; CI-DS128: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C5]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL2]]
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI-DS128: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY4]](s32)
    ; CI-DS128: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C4]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI-DS128: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C1]]
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI-DS128: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[COPY6]](s32)
    ; CI-DS128: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[SHL4]](s32)
    ; CI-DS128: [[OR4:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[TRUNC7]]
    ; CI-DS128: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; CI-DS128: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR4]](s16)
    ; CI-DS128: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C5]](s32)
    ; CI-DS128: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; CI-DS128: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR5]](s32)
    ; CI-DS128: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    ; VI-LABEL: name: test_load_local_v4s16_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C1]]
    ; VI: [[C2:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C2]](s16)
    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s16) = G_AND [[TRUNC3]], [[C1]]
    ; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[AND3]], [[C2]](s16)
    ; VI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[SHL1]]
    ; VI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; VI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C4]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL2]]
    ; VI: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C1]]
    ; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s16) = G_AND [[TRUNC5]], [[C1]]
    ; VI: [[SHL3:%[0-9]+]]:_(s16) = G_SHL [[AND5]], [[C2]](s16)
    ; VI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[SHL3]]
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C3]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C1]]
    ; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s16) = G_AND [[TRUNC7]], [[C1]]
    ; VI: [[SHL4:%[0-9]+]]:_(s16) = G_SHL [[AND7]], [[C2]](s16)
    ; VI: [[OR4:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[SHL4]]
    ; VI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; VI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR4]](s16)
    ; VI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C4]](s32)
    ; VI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; VI: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR5]](s32)
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>)
    ; VI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    ; GFX9-LABEL: name: test_load_local_v4s16_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C1]]
    ; GFX9: [[C2:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C2]](s16)
    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C1]]
    ; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s16) = G_AND [[TRUNC3]], [[C1]]
    ; GFX9: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[AND3]], [[C2]](s16)
    ; GFX9: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[SHL1]]
    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
    ; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[OR1]](s16)
    ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[ANYEXT]](s32), [[ANYEXT1]](s32)
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; GFX9: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C1]]
    ; GFX9: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s16) = G_AND [[TRUNC5]], [[C1]]
    ; GFX9: [[SHL2:%[0-9]+]]:_(s16) = G_SHL [[AND5]], [[C2]](s16)
    ; GFX9: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[SHL2]]
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C3]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; GFX9: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C1]]
    ; GFX9: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s16) = G_AND [[TRUNC7]], [[C1]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s16) = G_SHL [[AND7]], [[C2]](s16)
    ; GFX9: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[SHL3]]
    ; GFX9: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[OR2]](s16)
    ; GFX9: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[OR3]](s16)
    ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[ANYEXT2]](s32), [[ANYEXT3]](s32)
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
    ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v4s16_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[COPY]](p3) :: (load 8, align 1, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<4 x s16>) = G_LOAD %0 :: (load 8, align 1, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_v2s32_align8
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v2s32_align8
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; SI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; CI-LABEL: name: test_load_local_v2s32_align8
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; CI-DS128-LABEL: name: test_load_local_v2s32_align8
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; VI-LABEL: name: test_load_local_v2s32_align8
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; VI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; GFX9-LABEL: name: test_load_local_v2s32_align8
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v2s32_align8
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s32>) = G_LOAD %0 :: (load 8, align 8, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_v2s32_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v2s32_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; SI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; CI-LABEL: name: test_load_local_v2s32_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; CI-DS128-LABEL: name: test_load_local_v2s32_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; VI-LABEL: name: test_load_local_v2s32_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; VI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; GFX9-LABEL: name: test_load_local_v2s32_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v2s32_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s32>) = G_LOAD %0 :: (load 8, align 4, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_v2s32_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v2s32_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; CI-LABEL: name: test_load_local_v2s32_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; CI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; CI-DS128-LABEL: name: test_load_local_v2s32_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; VI-LABEL: name: test_load_local_v2s32_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; GFX9-LABEL: name: test_load_local_v2s32_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v2s32_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 2, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s32>) = G_LOAD %0 :: (load 8, align 2, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_v2s32_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v2s32_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; SI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; SI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; SI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; CI-LABEL: name: test_load_local_v2s32_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; CI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; CI-DS128-LABEL: name: test_load_local_v2s32_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI-DS128: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI-DS128: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI-DS128: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI-DS128: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI-DS128: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI-DS128: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; VI-LABEL: name: test_load_local_v2s32_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; VI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; VI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; VI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; VI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; VI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; VI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; VI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; VI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; VI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; VI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; VI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; GFX9-LABEL: name: test_load_local_v2s32_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; GFX9: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; GFX9: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; GFX9: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; GFX9: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; GFX9: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; GFX9: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; GFX9: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v2s32_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 1, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s32>) = G_LOAD %0 :: (load 8, align 1, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_load_local_v3s32_align16
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v3s32_align16
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; SI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; SI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; SI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; SI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; SI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; SI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; SI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; SI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; SI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; SI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; SI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; SI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; SI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; SI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; SI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; SI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; SI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; SI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; SI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; SI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; SI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; SI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; SI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; SI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; SI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[INSERT1]](<3 x s32>)
    ; CI-LABEL: name: test_load_local_v3s32_align16
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; CI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; CI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; CI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; CI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; CI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; CI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; CI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; CI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; CI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; CI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; CI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; CI: $vgpr0_vgpr1_vgpr2 = COPY [[INSERT1]](<3 x s32>)
    ; CI-DS128-LABEL: name: test_load_local_v3s32_align16
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI-DS128: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI-DS128: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI-DS128: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI-DS128: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI-DS128: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI-DS128: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; CI-DS128: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI-DS128: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI-DS128: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI-DS128: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI-DS128: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI-DS128: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI-DS128: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI-DS128: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI-DS128: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; CI-DS128: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI-DS128: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; CI-DS128: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; CI-DS128: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; CI-DS128: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI-DS128: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; CI-DS128: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; CI-DS128: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; CI-DS128: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI-DS128: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; CI-DS128: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; CI-DS128: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; CI-DS128: $vgpr0_vgpr1_vgpr2 = COPY [[INSERT1]](<3 x s32>)
    ; VI-LABEL: name: test_load_local_v3s32_align16
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; VI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; VI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; VI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; VI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; VI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; VI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; VI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; VI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; VI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; VI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; VI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; VI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; VI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; VI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; VI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; VI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; VI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; VI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; VI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; VI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; VI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; VI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; VI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; VI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; VI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; VI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; VI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; VI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; VI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; VI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; VI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; VI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; VI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; VI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[INSERT1]](<3 x s32>)
    ; GFX9-LABEL: name: test_load_local_v3s32_align16
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; GFX9: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; GFX9: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; GFX9: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; GFX9: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; GFX9: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; GFX9: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; GFX9: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; GFX9: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; GFX9: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; GFX9: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; GFX9: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; GFX9: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; GFX9: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; GFX9: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; GFX9: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; GFX9: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; GFX9: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; GFX9: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; GFX9: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; GFX9: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; GFX9: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; GFX9: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; GFX9: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; GFX9: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[INSERT1]](<3 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v3s32_align16
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 12, align 1, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<3 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<3 x s32>) = G_LOAD %0 :: (load 12, align 1, addrspace 3)
    $vgpr0_vgpr1_vgpr2 = COPY %1
...

---
name: test_load_local_v3s32_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v3s32_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; SI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; SI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[INSERT1]](<3 x s32>)
    ; CI-LABEL: name: test_load_local_v3s32_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; CI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; CI: $vgpr0_vgpr1_vgpr2 = COPY [[INSERT1]](<3 x s32>)
    ; CI-DS128-LABEL: name: test_load_local_v3s32_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; CI-DS128: $vgpr0_vgpr1_vgpr2 = COPY [[INSERT1]](<3 x s32>)
    ; VI-LABEL: name: test_load_local_v3s32_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; VI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; VI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[INSERT1]](<3 x s32>)
    ; GFX9-LABEL: name: test_load_local_v3s32_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD]](<2 x s32>), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[INSERT1]](<3 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v3s32_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 12, align 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<3 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<3 x s32>) = G_LOAD %0 :: (load 12, align 4, addrspace 3)
    $vgpr0_vgpr1_vgpr2 = COPY %1
...

---
name: test_load_local_v4s32_align16
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v4s32_align16
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 16, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, addrspace 3)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[LOAD]](<2 x s32>), [[LOAD1]](<2 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<4 x s32>)
    ; CI-LABEL: name: test_load_local_v4s32_align16
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 16, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, addrspace 3)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[LOAD]](<2 x s32>), [[LOAD1]](<2 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<4 x s32>)
    ; CI-DS128-LABEL: name: test_load_local_v4s32_align16
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    ; VI-LABEL: name: test_load_local_v4s32_align16
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, addrspace 3)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    ; GFX9-LABEL: name: test_load_local_v4s32_align16
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, addrspace 3)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v4s32_align16
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<4 x s32>) = G_LOAD %0 :: (load 16, align 16, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_load_local_v4s32_align8
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v4s32_align8
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, addrspace 3)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[LOAD]](<2 x s32>), [[LOAD1]](<2 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<4 x s32>)
    ; CI-LABEL: name: test_load_local_v4s32_align8
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, addrspace 3)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[LOAD]](<2 x s32>), [[LOAD1]](<2 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<4 x s32>)
    ; CI-DS128-LABEL: name: test_load_local_v4s32_align8
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 8, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    ; VI-LABEL: name: test_load_local_v4s32_align8
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 8, addrspace 3)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    ; GFX9-LABEL: name: test_load_local_v4s32_align8
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 8, addrspace 3)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v4s32_align8
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 8, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<4 x s32>) = G_LOAD %0 :: (load 16, align 8, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_load_local_v4s32_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v4s32_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, align 4, addrspace 3)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[LOAD]](<2 x s32>), [[LOAD1]](<2 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<4 x s32>)
    ; CI-LABEL: name: test_load_local_v4s32_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, align 4, addrspace 3)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[LOAD]](<2 x s32>), [[LOAD1]](<2 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<4 x s32>)
    ; CI-DS128-LABEL: name: test_load_local_v4s32_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 4, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 4 + 8, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 12, addrspace 3)
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[LOAD]](s32), [[LOAD1]](s32), [[LOAD2]](s32), [[LOAD3]](s32)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
    ; VI-LABEL: name: test_load_local_v4s32_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 4, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 4 + 8, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 12, addrspace 3)
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[LOAD]](s32), [[LOAD1]](s32), [[LOAD2]](s32), [[LOAD3]](s32)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
    ; GFX9-LABEL: name: test_load_local_v4s32_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 4, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 4 + 8, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 12, addrspace 3)
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[LOAD]](s32), [[LOAD1]](s32), [[LOAD2]](s32), [[LOAD3]](s32)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v4s32_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<4 x s32>) = G_LOAD %0 :: (load 16, align 4, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_load_local_v4s32_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v4s32_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C3]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C1]]
    ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; SI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR3]](s32)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<4 x s32>)
    ; CI-LABEL: name: test_load_local_v4s32_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C3]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C1]]
    ; CI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; CI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR3]](s32)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<4 x s32>)
    ; CI-DS128-LABEL: name: test_load_local_v4s32_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C3]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C1]]
    ; CI-DS128: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; CI-DS128: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR3]](s32)
    ; CI-DS128: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<4 x s32>)
    ; VI-LABEL: name: test_load_local_v4s32_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; VI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C3]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; VI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C1]]
    ; VI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; VI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; VI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; VI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR3]](s32)
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<4 x s32>)
    ; GFX9-LABEL: name: test_load_local_v4s32_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C3]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C1]]
    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; GFX9: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR3]](s32)
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<4 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v4s32_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 2, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<4 x s32>) = G_LOAD %0 :: (load 16, align 2, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_load_local_v4s32_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v4s32_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; SI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; SI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; SI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; SI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; SI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; SI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; SI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; SI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; SI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; SI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; SI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; SI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; SI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; SI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; SI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; SI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; SI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; SI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; SI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; SI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; SI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; SI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; SI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; SI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C7]](s32)
    ; SI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; SI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; SI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; SI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; SI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; SI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; SI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; SI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; SI: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; SI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; SI: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; SI: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; SI: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; SI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; SI: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; SI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; SI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; SI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; SI: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; SI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; SI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; SI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR8]](s32), [[OR11]](s32)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<4 x s32>)
    ; CI-LABEL: name: test_load_local_v4s32_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; CI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; CI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; CI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; CI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; CI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; CI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; CI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; CI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; CI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; CI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; CI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C7]](s32)
    ; CI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; CI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; CI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; CI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; CI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; CI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; CI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; CI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; CI: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; CI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; CI: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; CI: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; CI: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; CI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; CI: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; CI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; CI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; CI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; CI: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; CI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; CI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; CI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR8]](s32), [[OR11]](s32)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[BUILD_VECTOR]](<2 x s32>), [[BUILD_VECTOR1]](<2 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[CONCAT_VECTORS]](<4 x s32>)
    ; CI-DS128-LABEL: name: test_load_local_v4s32_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI-DS128: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI-DS128: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI-DS128: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI-DS128: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI-DS128: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI-DS128: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI-DS128: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI-DS128: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI-DS128: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI-DS128: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI-DS128: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI-DS128: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI-DS128: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI-DS128: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI-DS128: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; CI-DS128: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI-DS128: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; CI-DS128: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; CI-DS128: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; CI-DS128: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI-DS128: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; CI-DS128: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; CI-DS128: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; CI-DS128: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI-DS128: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; CI-DS128: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; CI-DS128: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; CI-DS128: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI-DS128: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; CI-DS128: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; CI-DS128: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; CI-DS128: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; CI-DS128: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; CI-DS128: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; CI-DS128: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; CI-DS128: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; CI-DS128: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; CI-DS128: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; CI-DS128: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; CI-DS128: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; CI-DS128: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; CI-DS128: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; CI-DS128: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; CI-DS128: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; CI-DS128: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; CI-DS128: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; CI-DS128: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; CI-DS128: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; CI-DS128: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; CI-DS128: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32), [[OR8]](s32), [[OR11]](s32)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
    ; VI-LABEL: name: test_load_local_v4s32_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; VI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; VI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; VI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; VI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; VI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; VI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; VI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; VI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; VI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; VI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; VI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; VI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; VI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; VI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; VI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; VI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; VI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; VI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; VI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; VI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; VI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; VI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; VI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; VI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; VI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; VI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; VI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; VI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; VI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; VI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; VI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; VI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; VI: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; VI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; VI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; VI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; VI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; VI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; VI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; VI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; VI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; VI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; VI: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; VI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; VI: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; VI: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; VI: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; VI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; VI: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; VI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; VI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; VI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; VI: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; VI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; VI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32), [[OR8]](s32), [[OR11]](s32)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
    ; GFX9-LABEL: name: test_load_local_v4s32_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; GFX9: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; GFX9: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; GFX9: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; GFX9: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; GFX9: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; GFX9: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; GFX9: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; GFX9: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; GFX9: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; GFX9: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; GFX9: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; GFX9: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; GFX9: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; GFX9: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; GFX9: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; GFX9: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; GFX9: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; GFX9: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; GFX9: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; GFX9: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; GFX9: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; GFX9: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; GFX9: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; GFX9: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; GFX9: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; GFX9: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; GFX9: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; GFX9: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; GFX9: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; GFX9: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; GFX9: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; GFX9: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; GFX9: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
    ; GFX9: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; GFX9: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; GFX9: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; GFX9: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; GFX9: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; GFX9: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; GFX9: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; GFX9: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; GFX9: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; GFX9: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; GFX9: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; GFX9: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32), [[OR8]](s32), [[OR11]](s32)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v4s32_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 1, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<4 x s32>) = G_LOAD %0 :: (load 16, align 1, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_load_local_v8s32_align32
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v8s32_align32
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 32, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD1]](p3) :: (load 8 + 16, align 16, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD2]](p3) :: (load 8 + 24, addrspace 3)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s32>) = G_CONCAT_VECTORS [[LOAD]](<2 x s32>), [[LOAD1]](<2 x s32>), [[LOAD2]](<2 x s32>), [[LOAD3]](<2 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[CONCAT_VECTORS]](<8 x s32>)
    ; CI-LABEL: name: test_load_local_v8s32_align32
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 32, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD1]](p3) :: (load 8 + 16, align 16, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD2]](p3) :: (load 8 + 24, addrspace 3)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s32>) = G_CONCAT_VECTORS [[LOAD]](<2 x s32>), [[LOAD1]](<2 x s32>), [[LOAD2]](<2 x s32>), [[LOAD3]](<2 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[CONCAT_VECTORS]](<8 x s32>)
    ; CI-DS128-LABEL: name: test_load_local_v8s32_align32
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 16 + 16, addrspace 3)
    ; CI-DS128: [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s32>) = G_CONCAT_VECTORS [[LOAD]](<4 x s32>), [[LOAD1]](<4 x s32>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[CONCAT_VECTORS]](<8 x s32>)
    ; VI-LABEL: name: test_load_local_v8s32_align32
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 16 + 16, addrspace 3)
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s32>) = G_CONCAT_VECTORS [[LOAD]](<4 x s32>), [[LOAD1]](<4 x s32>)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[CONCAT_VECTORS]](<8 x s32>)
    ; GFX9-LABEL: name: test_load_local_v8s32_align32
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 16 + 16, addrspace 3)
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s32>) = G_CONCAT_VECTORS [[LOAD]](<4 x s32>), [[LOAD1]](<4 x s32>)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[CONCAT_VECTORS]](<8 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v8s32_align32
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9-UNALIGNED: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9-UNALIGNED: [[LOAD1:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 16 + 16, addrspace 3)
    ; GFX9-UNALIGNED: [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s32>) = G_CONCAT_VECTORS [[LOAD]](<4 x s32>), [[LOAD1]](<4 x s32>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[CONCAT_VECTORS]](<8 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<8 x s32>) = G_LOAD %0 :: (load 32, align 32, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY %1
...

---
name: test_load_local_v16s32_align32
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v16s32_align32
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[LOAD]](<16 x s32>)
    ; CI-LABEL: name: test_load_local_v16s32_align32
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[LOAD]](<16 x s32>)
    ; CI-DS128-LABEL: name: test_load_local_v16s32_align32
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[LOAD]](<16 x s32>)
    ; VI-LABEL: name: test_load_local_v16s32_align32
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[LOAD]](<16 x s32>)
    ; GFX9-LABEL: name: test_load_local_v16s32_align32
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[LOAD]](<16 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v16s32_align32
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[LOAD]](<16 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<16 x s32>) = G_LOAD %0 :: (load 16, align 32, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %1
...

---
name: test_load_local_v2s64_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v2s64_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, align 4, addrspace 3)
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[LOAD]](s64), [[LOAD1]](s64)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    ; CI-LABEL: name: test_load_local_v2s64_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, align 4, addrspace 3)
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[LOAD]](s64), [[LOAD1]](s64)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    ; CI-DS128-LABEL: name: test_load_local_v2s64_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, align 4, addrspace 3)
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[LOAD]](s64), [[LOAD1]](s64)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    ; VI-LABEL: name: test_load_local_v2s64_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, align 4, addrspace 3)
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[LOAD]](s64), [[LOAD1]](s64)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    ; GFX9-LABEL: name: test_load_local_v2s64_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, align 4, addrspace 3)
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[LOAD]](s64), [[LOAD1]](s64)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v2s64_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[COPY]](p3) :: (load 16, align 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<2 x s64>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s64>) = G_LOAD %0 :: (load 16, align 4, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_load_local_v2s64_align16
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v2s64_align16
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; SI: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; SI: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; SI: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C9]]
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C9]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY3]](s32)
    ; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; SI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C9]]
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY5]](s32)
    ; SI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL2]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; SI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C9]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[COPY7]](s32)
    ; SI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[TRUNC7]]
    ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; SI: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C10]](s32)
    ; SI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C10]](s32)
    ; SI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; SI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; SI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; SI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; SI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; SI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; SI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; SI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; SI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; SI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; SI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C3]](s32)
    ; SI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; SI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C4]](s32)
    ; SI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; SI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C5]](s32)
    ; SI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; SI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C6]](s32)
    ; SI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; SI: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD8]](s32)
    ; SI: [[AND8:%[0-9]+]]:_(s16) = G_AND [[TRUNC8]], [[C7]]
    ; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; SI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; SI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C9]]
    ; SI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[COPY9]](s32)
    ; SI: [[TRUNC9:%[0-9]+]]:_(s16) = G_TRUNC [[SHL6]](s32)
    ; SI: [[OR6:%[0-9]+]]:_(s16) = G_OR [[AND8]], [[TRUNC9]]
    ; SI: [[TRUNC10:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD10]](s32)
    ; SI: [[AND10:%[0-9]+]]:_(s16) = G_AND [[TRUNC10]], [[C7]]
    ; SI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; SI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; SI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C9]]
    ; SI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[COPY11]](s32)
    ; SI: [[TRUNC11:%[0-9]+]]:_(s16) = G_TRUNC [[SHL7]](s32)
    ; SI: [[OR7:%[0-9]+]]:_(s16) = G_OR [[AND10]], [[TRUNC11]]
    ; SI: [[TRUNC12:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD12]](s32)
    ; SI: [[AND12:%[0-9]+]]:_(s16) = G_AND [[TRUNC12]], [[C7]]
    ; SI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; SI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; SI: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C9]]
    ; SI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[COPY13]](s32)
    ; SI: [[TRUNC13:%[0-9]+]]:_(s16) = G_TRUNC [[SHL8]](s32)
    ; SI: [[OR8:%[0-9]+]]:_(s16) = G_OR [[AND12]], [[TRUNC13]]
    ; SI: [[TRUNC14:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD14]](s32)
    ; SI: [[AND14:%[0-9]+]]:_(s16) = G_AND [[TRUNC14]], [[C7]]
    ; SI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; SI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; SI: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C9]]
    ; SI: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[COPY15]](s32)
    ; SI: [[TRUNC15:%[0-9]+]]:_(s16) = G_TRUNC [[SHL9]](s32)
    ; SI: [[OR9:%[0-9]+]]:_(s16) = G_OR [[AND14]], [[TRUNC15]]
    ; SI: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[OR6]](s16)
    ; SI: [[ZEXT5:%[0-9]+]]:_(s32) = G_ZEXT [[OR7]](s16)
    ; SI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[ZEXT5]], [[C10]](s32)
    ; SI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[ZEXT4]], [[SHL10]]
    ; SI: [[ZEXT6:%[0-9]+]]:_(s32) = G_ZEXT [[OR8]](s16)
    ; SI: [[ZEXT7:%[0-9]+]]:_(s32) = G_ZEXT [[OR9]](s16)
    ; SI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[ZEXT7]], [[C10]](s32)
    ; SI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[ZEXT6]], [[SHL11]]
    ; SI: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR10]](s32), [[OR11]](s32)
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    ; CI-LABEL: name: test_load_local_v2s64_align16
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; CI: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C9]]
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
    ; CI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C9]]
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY3]](s32)
    ; CI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; CI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C9]]
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY5]](s32)
    ; CI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL2]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; CI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C9]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[COPY7]](s32)
    ; CI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[TRUNC7]]
    ; CI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; CI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; CI: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C10]](s32)
    ; CI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; CI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; CI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; CI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C10]](s32)
    ; CI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; CI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; CI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; CI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C3]](s32)
    ; CI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; CI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C4]](s32)
    ; CI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; CI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C5]](s32)
    ; CI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; CI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C6]](s32)
    ; CI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; CI: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD8]](s32)
    ; CI: [[AND8:%[0-9]+]]:_(s16) = G_AND [[TRUNC8]], [[C7]]
    ; CI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C9]]
    ; CI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[COPY9]](s32)
    ; CI: [[TRUNC9:%[0-9]+]]:_(s16) = G_TRUNC [[SHL6]](s32)
    ; CI: [[OR6:%[0-9]+]]:_(s16) = G_OR [[AND8]], [[TRUNC9]]
    ; CI: [[TRUNC10:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD10]](s32)
    ; CI: [[AND10:%[0-9]+]]:_(s16) = G_AND [[TRUNC10]], [[C7]]
    ; CI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C9]]
    ; CI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[COPY11]](s32)
    ; CI: [[TRUNC11:%[0-9]+]]:_(s16) = G_TRUNC [[SHL7]](s32)
    ; CI: [[OR7:%[0-9]+]]:_(s16) = G_OR [[AND10]], [[TRUNC11]]
    ; CI: [[TRUNC12:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD12]](s32)
    ; CI: [[AND12:%[0-9]+]]:_(s16) = G_AND [[TRUNC12]], [[C7]]
    ; CI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; CI: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C9]]
    ; CI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[COPY13]](s32)
    ; CI: [[TRUNC13:%[0-9]+]]:_(s16) = G_TRUNC [[SHL8]](s32)
    ; CI: [[OR8:%[0-9]+]]:_(s16) = G_OR [[AND12]], [[TRUNC13]]
    ; CI: [[TRUNC14:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD14]](s32)
    ; CI: [[AND14:%[0-9]+]]:_(s16) = G_AND [[TRUNC14]], [[C7]]
    ; CI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; CI: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C9]]
    ; CI: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[COPY15]](s32)
    ; CI: [[TRUNC15:%[0-9]+]]:_(s16) = G_TRUNC [[SHL9]](s32)
    ; CI: [[OR9:%[0-9]+]]:_(s16) = G_OR [[AND14]], [[TRUNC15]]
    ; CI: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[OR6]](s16)
    ; CI: [[ZEXT5:%[0-9]+]]:_(s32) = G_ZEXT [[OR7]](s16)
    ; CI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[ZEXT5]], [[C10]](s32)
    ; CI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[ZEXT4]], [[SHL10]]
    ; CI: [[ZEXT6:%[0-9]+]]:_(s32) = G_ZEXT [[OR8]](s16)
    ; CI: [[ZEXT7:%[0-9]+]]:_(s32) = G_ZEXT [[OR9]](s16)
    ; CI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[ZEXT7]], [[C10]](s32)
    ; CI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[ZEXT6]], [[SHL11]]
    ; CI: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR10]](s32), [[OR11]](s32)
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    ; CI-DS128-LABEL: name: test_load_local_v2s64_align16
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI-DS128: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; CI-DS128: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; CI-DS128: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI-DS128: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C9]]
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
    ; CI-DS128: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
    ; CI-DS128: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C9]]
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[COPY3]](s32)
    ; CI-DS128: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
    ; CI-DS128: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C9]]
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[COPY5]](s32)
    ; CI-DS128: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SHL2]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[TRUNC5]]
    ; CI-DS128: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI-DS128: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C9]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[COPY7]](s32)
    ; CI-DS128: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[TRUNC7]]
    ; CI-DS128: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; CI-DS128: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; CI-DS128: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C10]](s32)
    ; CI-DS128: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; CI-DS128: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; CI-DS128: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; CI-DS128: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C10]](s32)
    ; CI-DS128: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; CI-DS128: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; CI-DS128: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; CI-DS128: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI-DS128: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI-DS128: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI-DS128: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI-DS128: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI-DS128: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI-DS128: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI-DS128: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C3]](s32)
    ; CI-DS128: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; CI-DS128: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C4]](s32)
    ; CI-DS128: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; CI-DS128: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C5]](s32)
    ; CI-DS128: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; CI-DS128: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C6]](s32)
    ; CI-DS128: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; CI-DS128: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD8]](s32)
    ; CI-DS128: [[AND8:%[0-9]+]]:_(s16) = G_AND [[TRUNC8]], [[C7]]
    ; CI-DS128: [[COPY9:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI-DS128: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI-DS128: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C9]]
    ; CI-DS128: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[COPY9]](s32)
    ; CI-DS128: [[TRUNC9:%[0-9]+]]:_(s16) = G_TRUNC [[SHL6]](s32)
    ; CI-DS128: [[OR6:%[0-9]+]]:_(s16) = G_OR [[AND8]], [[TRUNC9]]
    ; CI-DS128: [[TRUNC10:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD10]](s32)
    ; CI-DS128: [[AND10:%[0-9]+]]:_(s16) = G_AND [[TRUNC10]], [[C7]]
    ; CI-DS128: [[COPY11:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI-DS128: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI-DS128: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C9]]
    ; CI-DS128: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[COPY11]](s32)
    ; CI-DS128: [[TRUNC11:%[0-9]+]]:_(s16) = G_TRUNC [[SHL7]](s32)
    ; CI-DS128: [[OR7:%[0-9]+]]:_(s16) = G_OR [[AND10]], [[TRUNC11]]
    ; CI-DS128: [[TRUNC12:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD12]](s32)
    ; CI-DS128: [[AND12:%[0-9]+]]:_(s16) = G_AND [[TRUNC12]], [[C7]]
    ; CI-DS128: [[COPY13:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI-DS128: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; CI-DS128: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C9]]
    ; CI-DS128: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[COPY13]](s32)
    ; CI-DS128: [[TRUNC13:%[0-9]+]]:_(s16) = G_TRUNC [[SHL8]](s32)
    ; CI-DS128: [[OR8:%[0-9]+]]:_(s16) = G_OR [[AND12]], [[TRUNC13]]
    ; CI-DS128: [[TRUNC14:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD14]](s32)
    ; CI-DS128: [[AND14:%[0-9]+]]:_(s16) = G_AND [[TRUNC14]], [[C7]]
    ; CI-DS128: [[COPY15:%[0-9]+]]:_(s32) = COPY [[C8]](s32)
    ; CI-DS128: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; CI-DS128: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C9]]
    ; CI-DS128: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[COPY15]](s32)
    ; CI-DS128: [[TRUNC15:%[0-9]+]]:_(s16) = G_TRUNC [[SHL9]](s32)
    ; CI-DS128: [[OR9:%[0-9]+]]:_(s16) = G_OR [[AND14]], [[TRUNC15]]
    ; CI-DS128: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[OR6]](s16)
    ; CI-DS128: [[ZEXT5:%[0-9]+]]:_(s32) = G_ZEXT [[OR7]](s16)
    ; CI-DS128: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[ZEXT5]], [[C10]](s32)
    ; CI-DS128: [[OR10:%[0-9]+]]:_(s32) = G_OR [[ZEXT4]], [[SHL10]]
    ; CI-DS128: [[ZEXT6:%[0-9]+]]:_(s32) = G_ZEXT [[OR8]](s16)
    ; CI-DS128: [[ZEXT7:%[0-9]+]]:_(s32) = G_ZEXT [[OR9]](s16)
    ; CI-DS128: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[ZEXT7]], [[C10]](s32)
    ; CI-DS128: [[OR11:%[0-9]+]]:_(s32) = G_OR [[ZEXT6]], [[SHL11]]
    ; CI-DS128: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR10]](s32), [[OR11]](s32)
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    ; VI-LABEL: name: test_load_local_v2s64_align16
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; VI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; VI: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C7]]
    ; VI: [[C8:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C8]](s16)
    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s16) = G_AND [[TRUNC3]], [[C7]]
    ; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[AND3]], [[C8]](s16)
    ; VI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[SHL1]]
    ; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s16) = G_AND [[TRUNC5]], [[C7]]
    ; VI: [[SHL2:%[0-9]+]]:_(s16) = G_SHL [[AND5]], [[C8]](s16)
    ; VI: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[SHL2]]
    ; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s16) = G_AND [[TRUNC7]], [[C7]]
    ; VI: [[SHL3:%[0-9]+]]:_(s16) = G_SHL [[AND7]], [[C8]](s16)
    ; VI: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[SHL3]]
    ; VI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; VI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; VI: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C9]](s32)
    ; VI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; VI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; VI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; VI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C9]](s32)
    ; VI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; VI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; VI: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C10]](s32)
    ; VI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; VI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; VI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; VI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; VI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; VI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; VI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; VI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C3]](s32)
    ; VI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; VI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C4]](s32)
    ; VI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; VI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C5]](s32)
    ; VI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; VI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C6]](s32)
    ; VI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; VI: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD8]](s32)
    ; VI: [[AND8:%[0-9]+]]:_(s16) = G_AND [[TRUNC8]], [[C7]]
    ; VI: [[TRUNC9:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD9]](s32)
    ; VI: [[AND9:%[0-9]+]]:_(s16) = G_AND [[TRUNC9]], [[C7]]
    ; VI: [[SHL6:%[0-9]+]]:_(s16) = G_SHL [[AND9]], [[C8]](s16)
    ; VI: [[OR6:%[0-9]+]]:_(s16) = G_OR [[AND8]], [[SHL6]]
    ; VI: [[TRUNC10:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD10]](s32)
    ; VI: [[AND10:%[0-9]+]]:_(s16) = G_AND [[TRUNC10]], [[C7]]
    ; VI: [[TRUNC11:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD11]](s32)
    ; VI: [[AND11:%[0-9]+]]:_(s16) = G_AND [[TRUNC11]], [[C7]]
    ; VI: [[SHL7:%[0-9]+]]:_(s16) = G_SHL [[AND11]], [[C8]](s16)
    ; VI: [[OR7:%[0-9]+]]:_(s16) = G_OR [[AND10]], [[SHL7]]
    ; VI: [[TRUNC12:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD12]](s32)
    ; VI: [[AND12:%[0-9]+]]:_(s16) = G_AND [[TRUNC12]], [[C7]]
    ; VI: [[TRUNC13:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD13]](s32)
    ; VI: [[AND13:%[0-9]+]]:_(s16) = G_AND [[TRUNC13]], [[C7]]
    ; VI: [[SHL8:%[0-9]+]]:_(s16) = G_SHL [[AND13]], [[C8]](s16)
    ; VI: [[OR8:%[0-9]+]]:_(s16) = G_OR [[AND12]], [[SHL8]]
    ; VI: [[TRUNC14:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD14]](s32)
    ; VI: [[AND14:%[0-9]+]]:_(s16) = G_AND [[TRUNC14]], [[C7]]
    ; VI: [[TRUNC15:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD15]](s32)
    ; VI: [[AND15:%[0-9]+]]:_(s16) = G_AND [[TRUNC15]], [[C7]]
    ; VI: [[SHL9:%[0-9]+]]:_(s16) = G_SHL [[AND15]], [[C8]](s16)
    ; VI: [[OR9:%[0-9]+]]:_(s16) = G_OR [[AND14]], [[SHL9]]
    ; VI: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[OR6]](s16)
    ; VI: [[ZEXT5:%[0-9]+]]:_(s32) = G_ZEXT [[OR7]](s16)
    ; VI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[ZEXT5]], [[C9]](s32)
    ; VI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[ZEXT4]], [[SHL10]]
    ; VI: [[ZEXT6:%[0-9]+]]:_(s32) = G_ZEXT [[OR8]](s16)
    ; VI: [[ZEXT7:%[0-9]+]]:_(s32) = G_ZEXT [[OR9]](s16)
    ; VI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[ZEXT7]], [[C9]](s32)
    ; VI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[ZEXT6]], [[SHL11]]
    ; VI: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR10]](s32), [[OR11]](s32)
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    ; GFX9-LABEL: name: test_load_local_v2s64_align16
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; GFX9: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C6]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; GFX9: [[C7:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C7]]
    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C7]]
    ; GFX9: [[C8:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C8]](s16)
    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C7]]
    ; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s16) = G_AND [[TRUNC3]], [[C7]]
    ; GFX9: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[AND3]], [[C8]](s16)
    ; GFX9: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[SHL1]]
    ; GFX9: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C7]]
    ; GFX9: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s16) = G_AND [[TRUNC5]], [[C7]]
    ; GFX9: [[SHL2:%[0-9]+]]:_(s16) = G_SHL [[AND5]], [[C8]](s16)
    ; GFX9: [[OR2:%[0-9]+]]:_(s16) = G_OR [[AND4]], [[SHL2]]
    ; GFX9: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC6]], [[C7]]
    ; GFX9: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s16) = G_AND [[TRUNC7]], [[C7]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s16) = G_SHL [[AND7]], [[C8]](s16)
    ; GFX9: [[OR3:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[SHL3]]
    ; GFX9: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[OR]](s16)
    ; GFX9: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[OR1]](s16)
    ; GFX9: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C9]](s32)
    ; GFX9: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL4]]
    ; GFX9: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[OR2]](s16)
    ; GFX9: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[OR3]](s16)
    ; GFX9: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C9]](s32)
    ; GFX9: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL5]]
    ; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR4]](s32), [[OR5]](s32)
    ; GFX9: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C10]](s32)
    ; GFX9: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; GFX9: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; GFX9: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; GFX9: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; GFX9: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; GFX9: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; GFX9: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; GFX9: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C3]](s32)
    ; GFX9: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; GFX9: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C4]](s32)
    ; GFX9: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; GFX9: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C5]](s32)
    ; GFX9: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; GFX9: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C6]](s32)
    ; GFX9: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; GFX9: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD8]](s32)
    ; GFX9: [[AND8:%[0-9]+]]:_(s16) = G_AND [[TRUNC8]], [[C7]]
    ; GFX9: [[TRUNC9:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD9]](s32)
    ; GFX9: [[AND9:%[0-9]+]]:_(s16) = G_AND [[TRUNC9]], [[C7]]
    ; GFX9: [[SHL6:%[0-9]+]]:_(s16) = G_SHL [[AND9]], [[C8]](s16)
    ; GFX9: [[OR6:%[0-9]+]]:_(s16) = G_OR [[AND8]], [[SHL6]]
    ; GFX9: [[TRUNC10:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD10]](s32)
    ; GFX9: [[AND10:%[0-9]+]]:_(s16) = G_AND [[TRUNC10]], [[C7]]
    ; GFX9: [[TRUNC11:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD11]](s32)
    ; GFX9: [[AND11:%[0-9]+]]:_(s16) = G_AND [[TRUNC11]], [[C7]]
    ; GFX9: [[SHL7:%[0-9]+]]:_(s16) = G_SHL [[AND11]], [[C8]](s16)
    ; GFX9: [[OR7:%[0-9]+]]:_(s16) = G_OR [[AND10]], [[SHL7]]
    ; GFX9: [[TRUNC12:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD12]](s32)
    ; GFX9: [[AND12:%[0-9]+]]:_(s16) = G_AND [[TRUNC12]], [[C7]]
    ; GFX9: [[TRUNC13:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD13]](s32)
    ; GFX9: [[AND13:%[0-9]+]]:_(s16) = G_AND [[TRUNC13]], [[C7]]
    ; GFX9: [[SHL8:%[0-9]+]]:_(s16) = G_SHL [[AND13]], [[C8]](s16)
    ; GFX9: [[OR8:%[0-9]+]]:_(s16) = G_OR [[AND12]], [[SHL8]]
    ; GFX9: [[TRUNC14:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD14]](s32)
    ; GFX9: [[AND14:%[0-9]+]]:_(s16) = G_AND [[TRUNC14]], [[C7]]
    ; GFX9: [[TRUNC15:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD15]](s32)
    ; GFX9: [[AND15:%[0-9]+]]:_(s16) = G_AND [[TRUNC15]], [[C7]]
    ; GFX9: [[SHL9:%[0-9]+]]:_(s16) = G_SHL [[AND15]], [[C8]](s16)
    ; GFX9: [[OR9:%[0-9]+]]:_(s16) = G_OR [[AND14]], [[SHL9]]
    ; GFX9: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[OR6]](s16)
    ; GFX9: [[ZEXT5:%[0-9]+]]:_(s32) = G_ZEXT [[OR7]](s16)
    ; GFX9: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[ZEXT5]], [[C9]](s32)
    ; GFX9: [[OR10:%[0-9]+]]:_(s32) = G_OR [[ZEXT4]], [[SHL10]]
    ; GFX9: [[ZEXT6:%[0-9]+]]:_(s32) = G_ZEXT [[OR8]](s16)
    ; GFX9: [[ZEXT7:%[0-9]+]]:_(s32) = G_ZEXT [[OR9]](s16)
    ; GFX9: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[ZEXT7]], [[C9]](s32)
    ; GFX9: [[OR11:%[0-9]+]]:_(s32) = G_OR [[ZEXT6]], [[SHL11]]
    ; GFX9: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR10]](s32), [[OR11]](s32)
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[MV]](s64), [[MV1]](s64)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v2s64_align16
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[COPY]](p3) :: (load 16, align 1, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<2 x s64>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s64>) = G_LOAD %0 :: (load 16, align 1, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_load_local_v3s64_align32
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v3s64_align32
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 32, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD1]](p3) :: (load 8 + 16, align 16, addrspace 3)
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[LOAD]](s64), [[LOAD1]](s64), [[LOAD2]](s64)
    ; SI: [[DEF:%[0-9]+]]:_(<4 x s64>) = G_IMPLICIT_DEF
    ; SI: [[INSERT:%[0-9]+]]:_(<4 x s64>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<3 x s64>), 0
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[INSERT]](<4 x s64>)
    ; CI-LABEL: name: test_load_local_v3s64_align32
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 32, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD1]](p3) :: (load 8 + 16, align 16, addrspace 3)
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[LOAD]](s64), [[LOAD1]](s64), [[LOAD2]](s64)
    ; CI: [[DEF:%[0-9]+]]:_(<4 x s64>) = G_IMPLICIT_DEF
    ; CI: [[INSERT:%[0-9]+]]:_(<4 x s64>) = G_INSERT [[DEF]], [[BUILD_VECTOR]](<3 x s64>), 0
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[INSERT]](<4 x s64>)
    ; CI-DS128-LABEL: name: test_load_local_v3s64_align32
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 16, align 16, addrspace 3)
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<3 x s64>) = G_IMPLICIT_DEF
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<3 x s64>) = G_INSERT [[DEF]], [[LOAD]](<2 x s64>), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(<3 x s64>) = G_INSERT [[INSERT]], [[LOAD1]](s64), 128
    ; CI-DS128: [[DEF1:%[0-9]+]]:_(<4 x s64>) = G_IMPLICIT_DEF
    ; CI-DS128: [[INSERT2:%[0-9]+]]:_(<4 x s64>) = G_INSERT [[DEF1]], [[INSERT1]](<3 x s64>), 0
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[INSERT2]](<4 x s64>)
    ; VI-LABEL: name: test_load_local_v3s64_align32
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 16, align 16, addrspace 3)
    ; VI: [[DEF:%[0-9]+]]:_(<3 x s64>) = G_IMPLICIT_DEF
    ; VI: [[INSERT:%[0-9]+]]:_(<3 x s64>) = G_INSERT [[DEF]], [[LOAD]](<2 x s64>), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(<3 x s64>) = G_INSERT [[INSERT]], [[LOAD1]](s64), 128
    ; VI: [[DEF1:%[0-9]+]]:_(<4 x s64>) = G_IMPLICIT_DEF
    ; VI: [[INSERT2:%[0-9]+]]:_(<4 x s64>) = G_INSERT [[DEF1]], [[INSERT1]](<3 x s64>), 0
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[INSERT2]](<4 x s64>)
    ; GFX9-LABEL: name: test_load_local_v3s64_align32
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 16, align 16, addrspace 3)
    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s64>) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT:%[0-9]+]]:_(<3 x s64>) = G_INSERT [[DEF]], [[LOAD]](<2 x s64>), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(<3 x s64>) = G_INSERT [[INSERT]], [[LOAD1]](s64), 128
    ; GFX9: [[DEF1:%[0-9]+]]:_(<4 x s64>) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT2:%[0-9]+]]:_(<4 x s64>) = G_INSERT [[DEF1]], [[INSERT1]](<3 x s64>), 0
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[INSERT2]](<4 x s64>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v3s64_align32
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9-UNALIGNED: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9-UNALIGNED: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 16, align 16, addrspace 3)
    ; GFX9-UNALIGNED: [[DEF:%[0-9]+]]:_(<3 x s64>) = G_IMPLICIT_DEF
    ; GFX9-UNALIGNED: [[INSERT:%[0-9]+]]:_(<3 x s64>) = G_INSERT [[DEF]], [[LOAD]](<2 x s64>), 0
    ; GFX9-UNALIGNED: [[INSERT1:%[0-9]+]]:_(<3 x s64>) = G_INSERT [[INSERT]], [[LOAD1]](s64), 128
    ; GFX9-UNALIGNED: [[DEF1:%[0-9]+]]:_(<4 x s64>) = G_IMPLICIT_DEF
    ; GFX9-UNALIGNED: [[INSERT2:%[0-9]+]]:_(<4 x s64>) = G_INSERT [[DEF1]], [[INSERT1]](<3 x s64>), 0
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[INSERT2]](<4 x s64>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<3 x s64>) = G_LOAD %0 :: (load 24, align 32, addrspace 3)
    %2:_(<4 x s64>) = G_IMPLICIT_DEF
    %3:_(<4 x s64>) = G_INSERT %2, %1, 0
    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY %3
...

---
name: test_load_local_v4s64_align32
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v4s64_align32
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 32, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD1]](p3) :: (load 8 + 16, align 16, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD2]](p3) :: (load 8 + 24, addrspace 3)
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[LOAD]](s64), [[LOAD1]](s64), [[LOAD2]](s64), [[LOAD3]](s64)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[BUILD_VECTOR]](<4 x s64>)
    ; CI-LABEL: name: test_load_local_v4s64_align32
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s64) = G_LOAD [[COPY]](p3) :: (load 8, align 32, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD1]](p3) :: (load 8 + 16, align 16, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD2]](p3) :: (load 8 + 24, addrspace 3)
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[LOAD]](s64), [[LOAD1]](s64), [[LOAD2]](s64), [[LOAD3]](s64)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[BUILD_VECTOR]](<4 x s64>)
    ; CI-DS128-LABEL: name: test_load_local_v4s64_align32
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[PTR_ADD]](p3) :: (load 16 + 16, addrspace 3)
    ; CI-DS128: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s64>) = G_CONCAT_VECTORS [[LOAD]](<2 x s64>), [[LOAD1]](<2 x s64>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[CONCAT_VECTORS]](<4 x s64>)
    ; VI-LABEL: name: test_load_local_v4s64_align32
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[PTR_ADD]](p3) :: (load 16 + 16, addrspace 3)
    ; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s64>) = G_CONCAT_VECTORS [[LOAD]](<2 x s64>), [[LOAD1]](<2 x s64>)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[CONCAT_VECTORS]](<4 x s64>)
    ; GFX9-LABEL: name: test_load_local_v4s64_align32
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[PTR_ADD]](p3) :: (load 16 + 16, addrspace 3)
    ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s64>) = G_CONCAT_VECTORS [[LOAD]](<2 x s64>), [[LOAD1]](<2 x s64>)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[CONCAT_VECTORS]](<4 x s64>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v4s64_align32
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[COPY]](p3) :: (load 16, align 32, addrspace 3)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9-UNALIGNED: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9-UNALIGNED: [[LOAD1:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[PTR_ADD]](p3) :: (load 16 + 16, addrspace 3)
    ; GFX9-UNALIGNED: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s64>) = G_CONCAT_VECTORS [[LOAD]](<2 x s64>), [[LOAD1]](<2 x s64>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[CONCAT_VECTORS]](<4 x s64>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<4 x s64>) = G_LOAD %0 :: (load 32, align 32, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY %1
...

---
name: test_load_local_v2p1_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v2p1_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, align 4, addrspace 3)
    ; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[LOAD]](<2 x s32>), [[LOAD1]](<2 x s32>)
    ; SI: [[BITCAST:%[0-9]+]]:_(<2 x p1>) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](<2 x p1>)
    ; CI-LABEL: name: test_load_local_v2p1_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 8, align 4, addrspace 3)
    ; CI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[LOAD]](<2 x s32>), [[LOAD1]](<2 x s32>)
    ; CI: [[BITCAST:%[0-9]+]]:_(<2 x p1>) = G_BITCAST [[CONCAT_VECTORS]](<4 x s32>)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](<2 x p1>)
    ; CI-DS128-LABEL: name: test_load_local_v2p1_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 4, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 4 + 8, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 12, addrspace 3)
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[LOAD]](s32), [[LOAD1]](s32), [[LOAD2]](s32), [[LOAD3]](s32)
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(<2 x p1>) = G_BITCAST [[BUILD_VECTOR]](<4 x s32>)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](<2 x p1>)
    ; VI-LABEL: name: test_load_local_v2p1_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 4, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 4 + 8, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 12, addrspace 3)
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[LOAD]](s32), [[LOAD1]](s32), [[LOAD2]](s32), [[LOAD3]](s32)
    ; VI: [[BITCAST:%[0-9]+]]:_(<2 x p1>) = G_BITCAST [[BUILD_VECTOR]](<4 x s32>)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](<2 x p1>)
    ; GFX9-LABEL: name: test_load_local_v2p1_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 4, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 4 + 8, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 12, addrspace 3)
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[LOAD]](s32), [[LOAD1]](s32), [[LOAD2]](s32), [[LOAD3]](s32)
    ; GFX9: [[BITCAST:%[0-9]+]]:_(<2 x p1>) = G_BITCAST [[BUILD_VECTOR]](<4 x s32>)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](<2 x p1>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v2p1_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 16, align 4, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(<2 x p1>) = G_BITCAST [[LOAD]](<4 x s32>)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BITCAST]](<2 x p1>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x p1>) = G_LOAD %0 :: (load 16, align 4, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_load_local_v2p3_align8
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_load_local_v2p3_align8
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x p3>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; SI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
    ; CI-LABEL: name: test_load_local_v2p3_align8
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x p3>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
    ; CI-DS128-LABEL: name: test_load_local_v2p3_align8
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<2 x p3>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
    ; VI-LABEL: name: test_load_local_v2p3_align8
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<2 x p3>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; VI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
    ; GFX9-LABEL: name: test_load_local_v2p3_align8
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<2 x p3>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
    ; GFX9-UNALIGNED-LABEL: name: test_load_local_v2p3_align8
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x p3>) = G_LOAD [[COPY]](p3) :: (load 8, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x p3>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x p3>) = G_LOAD %0 :: (load 8, align 8, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_extload_local_s32_from_1_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_s32_from_1_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; SI: $vgpr0 = COPY [[LOAD]](s32)
    ; CI-LABEL: name: test_extload_local_s32_from_1_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; CI: $vgpr0 = COPY [[LOAD]](s32)
    ; CI-DS128-LABEL: name: test_extload_local_s32_from_1_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; CI-DS128: $vgpr0 = COPY [[LOAD]](s32)
    ; VI-LABEL: name: test_extload_local_s32_from_1_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; VI: $vgpr0 = COPY [[LOAD]](s32)
    ; GFX9-LABEL: name: test_extload_local_s32_from_1_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; GFX9: $vgpr0 = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_s32_from_1_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[LOAD]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s32) = G_LOAD %0 :: (load 1, align 4, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_extload_local_s32_from_2_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_s32_from_2_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; SI: $vgpr0 = COPY [[LOAD]](s32)
    ; CI-LABEL: name: test_extload_local_s32_from_2_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; CI: $vgpr0 = COPY [[LOAD]](s32)
    ; CI-DS128-LABEL: name: test_extload_local_s32_from_2_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; CI-DS128: $vgpr0 = COPY [[LOAD]](s32)
    ; VI-LABEL: name: test_extload_local_s32_from_2_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; VI: $vgpr0 = COPY [[LOAD]](s32)
    ; GFX9-LABEL: name: test_extload_local_s32_from_2_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; GFX9: $vgpr0 = COPY [[LOAD]](s32)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_s32_from_2_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0 = COPY [[LOAD]](s32)
    %0:_(p3) = COPY $vgpr0
    %1:_(s32) = G_LOAD %0 :: (load 2, align 4, addrspace 3)
    $vgpr0 = COPY %1
...

---
name: test_extload_local_s64_from_1_align4
body: |
  bb.0:
    liveins: $vgpr0


    ; SI-LABEL: name: test_extload_local_s64_from_1_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; SI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; SI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; CI-LABEL: name: test_extload_local_s64_from_1_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; CI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; CI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; CI-DS128-LABEL: name: test_extload_local_s64_from_1_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; CI-DS128: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; VI-LABEL: name: test_extload_local_s64_from_1_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; VI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; VI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; GFX9-LABEL: name: test_extload_local_s64_from_1_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; GFX9: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_s64_from_1_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; GFX9-UNALIGNED: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    %0:_(p3) = COPY $vgpr0
    %1:_(s64) = G_LOAD %0 :: (load 1, align 4, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_extload_local_s64_from_2_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_s64_from_2_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; SI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; SI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; CI-LABEL: name: test_extload_local_s64_from_2_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; CI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; CI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; CI-DS128-LABEL: name: test_extload_local_s64_from_2_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; CI-DS128: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; VI-LABEL: name: test_extload_local_s64_from_2_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; VI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; VI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; GFX9-LABEL: name: test_extload_local_s64_from_2_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; GFX9: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_s64_from_2_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; GFX9-UNALIGNED: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    %0:_(p3) = COPY $vgpr0
    %1:_(s64) = G_LOAD %0 :: (load 2, align 4, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_extload_local_s64_from_4_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_s64_from_4_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; SI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; SI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; CI-LABEL: name: test_extload_local_s64_from_4_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; CI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; CI-DS128-LABEL: name: test_extload_local_s64_from_4_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI-DS128: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; VI-LABEL: name: test_extload_local_s64_from_4_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; VI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; VI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; GFX9-LABEL: name: test_extload_local_s64_from_4_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; GFX9: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_s64_from_4_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9-UNALIGNED: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    %0:_(p3) = COPY $vgpr0
    %1:_(s64) = G_LOAD %0 :: (load 4, align 4, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_extload_local_s128_from_4_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_s128_from_4_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; SI: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
    ; SI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[LOAD]](s32), [[DEF]](s32)
    ; SI: [[DEF1:%[0-9]+]]:_(s64) = G_IMPLICIT_DEF
    ; SI: [[MV1:%[0-9]+]]:_(s128) = G_MERGE_VALUES [[MV]](s64), [[DEF1]](s64)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[MV1]](s128)
    ; CI-LABEL: name: test_extload_local_s128_from_4_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
    ; CI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[LOAD]](s32), [[DEF]](s32)
    ; CI: [[DEF1:%[0-9]+]]:_(s64) = G_IMPLICIT_DEF
    ; CI: [[MV1:%[0-9]+]]:_(s128) = G_MERGE_VALUES [[MV]](s64), [[DEF1]](s64)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[MV1]](s128)
    ; CI-DS128-LABEL: name: test_extload_local_s128_from_4_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI-DS128: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
    ; CI-DS128: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[LOAD]](s32), [[DEF]](s32)
    ; CI-DS128: [[DEF1:%[0-9]+]]:_(s64) = G_IMPLICIT_DEF
    ; CI-DS128: [[MV1:%[0-9]+]]:_(s128) = G_MERGE_VALUES [[MV]](s64), [[DEF1]](s64)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[MV1]](s128)
    ; VI-LABEL: name: test_extload_local_s128_from_4_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; VI: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
    ; VI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[LOAD]](s32), [[DEF]](s32)
    ; VI: [[DEF1:%[0-9]+]]:_(s64) = G_IMPLICIT_DEF
    ; VI: [[MV1:%[0-9]+]]:_(s128) = G_MERGE_VALUES [[MV]](s64), [[DEF1]](s64)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[MV1]](s128)
    ; GFX9-LABEL: name: test_extload_local_s128_from_4_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
    ; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[LOAD]](s32), [[DEF]](s32)
    ; GFX9: [[DEF1:%[0-9]+]]:_(s64) = G_IMPLICIT_DEF
    ; GFX9: [[MV1:%[0-9]+]]:_(s128) = G_MERGE_VALUES [[MV]](s64), [[DEF1]](s64)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[MV1]](s128)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_s128_from_4_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9-UNALIGNED: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
    ; GFX9-UNALIGNED: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[LOAD]](s32), [[DEF]](s32)
    ; GFX9-UNALIGNED: [[DEF1:%[0-9]+]]:_(s64) = G_IMPLICIT_DEF
    ; GFX9-UNALIGNED: [[MV1:%[0-9]+]]:_(s128) = G_MERGE_VALUES [[MV]](s64), [[DEF1]](s64)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[MV1]](s128)
    %0:_(p3) = COPY $vgpr0
    %1:_(s128) = G_LOAD %0 :: (load 4, align 4, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_extload_local_s64_from_2_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_s64_from_2_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; SI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; SI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; CI-LABEL: name: test_extload_local_s64_from_2_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; CI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; CI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; CI-DS128-LABEL: name: test_extload_local_s64_from_2_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; CI-DS128: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; VI-LABEL: name: test_extload_local_s64_from_2_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; VI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; VI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; GFX9-LABEL: name: test_extload_local_s64_from_2_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; GFX9: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_s64_from_2_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, align 4, addrspace 3)
    ; GFX9-UNALIGNED: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    %0:_(p3) = COPY $vgpr0
    %1:_(s64) = G_LOAD %0 :: (load 2, align 4, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_extload_local_s64_from_1_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_s64_from_1_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; SI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; SI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; CI-LABEL: name: test_extload_local_s64_from_1_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; CI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; CI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; CI-DS128-LABEL: name: test_extload_local_s64_from_1_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; CI-DS128: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; VI-LABEL: name: test_extload_local_s64_from_1_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; VI: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; VI: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; GFX9-LABEL: name: test_extload_local_s64_from_1_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; GFX9: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_s64_from_1_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, align 4, addrspace 3)
    ; GFX9-UNALIGNED: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LOAD]](s32)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[ANYEXT]](s64)
    %0:_(p3) = COPY $vgpr0
    %1:_(s64) = G_LOAD %0 :: (load 1, align 4, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_extload_local_v2s32_from_4_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_v2s32_from_4_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, align 1, addrspace 3)
    ; SI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; CI-LABEL: name: test_extload_local_v2s32_from_4_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, align 1, addrspace 3)
    ; CI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; CI-DS128-LABEL: name: test_extload_local_v2s32_from_4_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, align 1, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; VI-LABEL: name: test_extload_local_v2s32_from_4_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, align 1, addrspace 3)
    ; VI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; GFX9-LABEL: name: test_extload_local_v2s32_from_4_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, align 1, addrspace 3)
    ; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_v2s32_from_4_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, align 1, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s32>) = G_LOAD %0 :: (load 4, align 1, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_extload_local_v2s32_from_4_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_v2s32_from_4_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, align 2, addrspace 3)
    ; SI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; CI-LABEL: name: test_extload_local_v2s32_from_4_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, align 2, addrspace 3)
    ; CI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; CI-DS128-LABEL: name: test_extload_local_v2s32_from_4_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, align 2, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; VI-LABEL: name: test_extload_local_v2s32_from_4_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, align 2, addrspace 3)
    ; VI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; GFX9-LABEL: name: test_extload_local_v2s32_from_4_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, align 2, addrspace 3)
    ; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_v2s32_from_4_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, align 2, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s32>) = G_LOAD %0 :: (load 4, align 2, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_extload_local_v2s32_from_4_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_v2s32_from_4_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; SI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; CI-LABEL: name: test_extload_local_v2s32_from_4_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; CI-DS128-LABEL: name: test_extload_local_v2s32_from_4_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; VI-LABEL: name: test_extload_local_v2s32_from_4_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; VI: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; GFX9-LABEL: name: test_extload_local_v2s32_from_4_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_v2s32_from_4_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s32>) = G_LOAD %0 :: (load 4, align 4, addrspace 3)
    $vgpr0_vgpr1 = COPY %1
...

---
name: test_extload_local_v3s32_from_6_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_v3s32_from_6_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 6, align 4, addrspace 3)
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<3 x s32>)
    ; CI-LABEL: name: test_extload_local_v3s32_from_6_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 6, align 4, addrspace 3)
    ; CI: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<3 x s32>)
    ; CI-DS128-LABEL: name: test_extload_local_v3s32_from_6_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 6, align 4, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<3 x s32>)
    ; VI-LABEL: name: test_extload_local_v3s32_from_6_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 6, align 4, addrspace 3)
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<3 x s32>)
    ; GFX9-LABEL: name: test_extload_local_v3s32_from_6_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 6, align 4, addrspace 3)
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<3 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_v3s32_from_6_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 6, align 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<3 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<3 x s32>) = G_LOAD %0 :: (load 6, align 4, addrspace 3)
    $vgpr0_vgpr1_vgpr2 = COPY %1
...

---
name: test_extload_local_v4s32_from_8_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_v4s32_from_8_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    ; CI-LABEL: name: test_extload_local_v4s32_from_8_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    ; CI-DS128-LABEL: name: test_extload_local_v4s32_from_8_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    ; VI-LABEL: name: test_extload_local_v4s32_from_8_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    ; GFX9-LABEL: name: test_extload_local_v4s32_from_8_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_v4s32_from_8_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
    %0:_(p3) = COPY $vgpr0
    %1:_(<4 x s32>) = G_LOAD %0 :: (load 8, align 4, addrspace 3)
    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...

---
name: test_extload_local_v2s96_from_24_align1
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_v2s96_from_24_align1
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; SI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; SI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; SI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; SI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; SI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; SI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; SI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; SI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; SI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; SI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; SI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; SI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; SI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; SI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; SI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; SI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; SI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; SI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; SI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; SI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; SI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; SI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; SI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; SI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; SI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; SI: [[COPY13:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; SI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY13]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; SI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; SI: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; SI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; SI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; SI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; SI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; SI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; SI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; SI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; SI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; SI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; SI: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; SI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; SI: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; SI: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; SI: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; SI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; SI: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; SI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; SI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; SI: [[COPY17:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; SI: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY17]], [[C3]]
    ; SI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; SI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; SI: [[PTR_ADD15:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C7]](s32)
    ; SI: [[LOAD16:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD15]](p3) :: (load 1 + 16, addrspace 3)
    ; SI: [[PTR_ADD16:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C]](s32)
    ; SI: [[LOAD17:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD16]](p3) :: (load 1 + 17, addrspace 3)
    ; SI: [[PTR_ADD17:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C1]](s32)
    ; SI: [[LOAD18:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD17]](p3) :: (load 1 + 18, addrspace 3)
    ; SI: [[PTR_ADD18:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C2]](s32)
    ; SI: [[LOAD19:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD18]](p3) :: (load 1 + 19, addrspace 3)
    ; SI: [[COPY18:%[0-9]+]]:_(s32) = COPY [[LOAD16]](s32)
    ; SI: [[AND16:%[0-9]+]]:_(s32) = G_AND [[COPY18]], [[C3]]
    ; SI: [[COPY19:%[0-9]+]]:_(s32) = COPY [[LOAD17]](s32)
    ; SI: [[AND17:%[0-9]+]]:_(s32) = G_AND [[COPY19]], [[C3]]
    ; SI: [[SHL12:%[0-9]+]]:_(s32) = G_SHL [[AND17]], [[C4]](s32)
    ; SI: [[OR12:%[0-9]+]]:_(s32) = G_OR [[AND16]], [[SHL12]]
    ; SI: [[COPY20:%[0-9]+]]:_(s32) = COPY [[LOAD18]](s32)
    ; SI: [[AND18:%[0-9]+]]:_(s32) = G_AND [[COPY20]], [[C3]]
    ; SI: [[SHL13:%[0-9]+]]:_(s32) = G_SHL [[AND18]], [[C5]](s32)
    ; SI: [[OR13:%[0-9]+]]:_(s32) = G_OR [[OR12]], [[SHL13]]
    ; SI: [[COPY21:%[0-9]+]]:_(s32) = COPY [[LOAD19]](s32)
    ; SI: [[AND19:%[0-9]+]]:_(s32) = G_AND [[COPY21]], [[C3]]
    ; SI: [[SHL14:%[0-9]+]]:_(s32) = G_SHL [[AND19]], [[C6]](s32)
    ; SI: [[OR14:%[0-9]+]]:_(s32) = G_OR [[OR13]], [[SHL14]]
    ; SI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR11]](s32), [[OR14]](s32)
    ; SI: [[PTR_ADD19:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C4]](s32)
    ; SI: [[LOAD20:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD19]](p3) :: (load 1 + 20, addrspace 3)
    ; SI: [[PTR_ADD20:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C]](s32)
    ; SI: [[LOAD21:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD20]](p3) :: (load 1 + 21, addrspace 3)
    ; SI: [[PTR_ADD21:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C1]](s32)
    ; SI: [[LOAD22:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD21]](p3) :: (load 1 + 22, addrspace 3)
    ; SI: [[PTR_ADD22:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C2]](s32)
    ; SI: [[LOAD23:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD22]](p3) :: (load 1 + 23, addrspace 3)
    ; SI: [[COPY22:%[0-9]+]]:_(s32) = COPY [[LOAD20]](s32)
    ; SI: [[AND20:%[0-9]+]]:_(s32) = G_AND [[COPY22]], [[C3]]
    ; SI: [[COPY23:%[0-9]+]]:_(s32) = COPY [[LOAD21]](s32)
    ; SI: [[AND21:%[0-9]+]]:_(s32) = G_AND [[COPY23]], [[C3]]
    ; SI: [[SHL15:%[0-9]+]]:_(s32) = G_SHL [[AND21]], [[C4]](s32)
    ; SI: [[OR15:%[0-9]+]]:_(s32) = G_OR [[AND20]], [[SHL15]]
    ; SI: [[COPY24:%[0-9]+]]:_(s32) = COPY [[LOAD22]](s32)
    ; SI: [[AND22:%[0-9]+]]:_(s32) = G_AND [[COPY24]], [[C3]]
    ; SI: [[SHL16:%[0-9]+]]:_(s32) = G_SHL [[AND22]], [[C5]](s32)
    ; SI: [[OR16:%[0-9]+]]:_(s32) = G_OR [[OR15]], [[SHL16]]
    ; SI: [[COPY25:%[0-9]+]]:_(s32) = COPY [[LOAD23]](s32)
    ; SI: [[AND23:%[0-9]+]]:_(s32) = G_AND [[COPY25]], [[C3]]
    ; SI: [[SHL17:%[0-9]+]]:_(s32) = G_SHL [[AND23]], [[C6]](s32)
    ; SI: [[OR17:%[0-9]+]]:_(s32) = G_OR [[OR16]], [[SHL17]]
    ; SI: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR1]](<2 x s32>), 0
    ; SI: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[OR17]](s32), 64
    ; SI: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; SI: [[COPY26:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; SI: [[COPY27:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[COPY26]](s96)
    ; SI: $vgpr3_vgpr4_vgpr5 = COPY [[COPY27]](s96)
    ; CI-LABEL: name: test_extload_local_v2s96_from_24_align1
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; CI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; CI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; CI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; CI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; CI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; CI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; CI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; CI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; CI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; CI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; CI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI: [[COPY13:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; CI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY13]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; CI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; CI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; CI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; CI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; CI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; CI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; CI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; CI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; CI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; CI: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; CI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; CI: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; CI: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; CI: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; CI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; CI: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; CI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; CI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; CI: [[COPY17:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; CI: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY17]], [[C3]]
    ; CI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; CI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; CI: [[PTR_ADD15:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C7]](s32)
    ; CI: [[LOAD16:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD15]](p3) :: (load 1 + 16, addrspace 3)
    ; CI: [[PTR_ADD16:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C]](s32)
    ; CI: [[LOAD17:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD16]](p3) :: (load 1 + 17, addrspace 3)
    ; CI: [[PTR_ADD17:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C1]](s32)
    ; CI: [[LOAD18:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD17]](p3) :: (load 1 + 18, addrspace 3)
    ; CI: [[PTR_ADD18:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C2]](s32)
    ; CI: [[LOAD19:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD18]](p3) :: (load 1 + 19, addrspace 3)
    ; CI: [[COPY18:%[0-9]+]]:_(s32) = COPY [[LOAD16]](s32)
    ; CI: [[AND16:%[0-9]+]]:_(s32) = G_AND [[COPY18]], [[C3]]
    ; CI: [[COPY19:%[0-9]+]]:_(s32) = COPY [[LOAD17]](s32)
    ; CI: [[AND17:%[0-9]+]]:_(s32) = G_AND [[COPY19]], [[C3]]
    ; CI: [[SHL12:%[0-9]+]]:_(s32) = G_SHL [[AND17]], [[C4]](s32)
    ; CI: [[OR12:%[0-9]+]]:_(s32) = G_OR [[AND16]], [[SHL12]]
    ; CI: [[COPY20:%[0-9]+]]:_(s32) = COPY [[LOAD18]](s32)
    ; CI: [[AND18:%[0-9]+]]:_(s32) = G_AND [[COPY20]], [[C3]]
    ; CI: [[SHL13:%[0-9]+]]:_(s32) = G_SHL [[AND18]], [[C5]](s32)
    ; CI: [[OR13:%[0-9]+]]:_(s32) = G_OR [[OR12]], [[SHL13]]
    ; CI: [[COPY21:%[0-9]+]]:_(s32) = COPY [[LOAD19]](s32)
    ; CI: [[AND19:%[0-9]+]]:_(s32) = G_AND [[COPY21]], [[C3]]
    ; CI: [[SHL14:%[0-9]+]]:_(s32) = G_SHL [[AND19]], [[C6]](s32)
    ; CI: [[OR14:%[0-9]+]]:_(s32) = G_OR [[OR13]], [[SHL14]]
    ; CI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR11]](s32), [[OR14]](s32)
    ; CI: [[PTR_ADD19:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C4]](s32)
    ; CI: [[LOAD20:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD19]](p3) :: (load 1 + 20, addrspace 3)
    ; CI: [[PTR_ADD20:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C]](s32)
    ; CI: [[LOAD21:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD20]](p3) :: (load 1 + 21, addrspace 3)
    ; CI: [[PTR_ADD21:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C1]](s32)
    ; CI: [[LOAD22:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD21]](p3) :: (load 1 + 22, addrspace 3)
    ; CI: [[PTR_ADD22:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C2]](s32)
    ; CI: [[LOAD23:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD22]](p3) :: (load 1 + 23, addrspace 3)
    ; CI: [[COPY22:%[0-9]+]]:_(s32) = COPY [[LOAD20]](s32)
    ; CI: [[AND20:%[0-9]+]]:_(s32) = G_AND [[COPY22]], [[C3]]
    ; CI: [[COPY23:%[0-9]+]]:_(s32) = COPY [[LOAD21]](s32)
    ; CI: [[AND21:%[0-9]+]]:_(s32) = G_AND [[COPY23]], [[C3]]
    ; CI: [[SHL15:%[0-9]+]]:_(s32) = G_SHL [[AND21]], [[C4]](s32)
    ; CI: [[OR15:%[0-9]+]]:_(s32) = G_OR [[AND20]], [[SHL15]]
    ; CI: [[COPY24:%[0-9]+]]:_(s32) = COPY [[LOAD22]](s32)
    ; CI: [[AND22:%[0-9]+]]:_(s32) = G_AND [[COPY24]], [[C3]]
    ; CI: [[SHL16:%[0-9]+]]:_(s32) = G_SHL [[AND22]], [[C5]](s32)
    ; CI: [[OR16:%[0-9]+]]:_(s32) = G_OR [[OR15]], [[SHL16]]
    ; CI: [[COPY25:%[0-9]+]]:_(s32) = COPY [[LOAD23]](s32)
    ; CI: [[AND23:%[0-9]+]]:_(s32) = G_AND [[COPY25]], [[C3]]
    ; CI: [[SHL17:%[0-9]+]]:_(s32) = G_SHL [[AND23]], [[C6]](s32)
    ; CI: [[OR17:%[0-9]+]]:_(s32) = G_OR [[OR16]], [[SHL17]]
    ; CI: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR1]](<2 x s32>), 0
    ; CI: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[OR17]](s32), 64
    ; CI: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; CI: [[COPY26:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; CI: [[COPY27:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; CI: $vgpr0_vgpr1_vgpr2 = COPY [[COPY26]](s96)
    ; CI: $vgpr3_vgpr4_vgpr5 = COPY [[COPY27]](s96)
    ; CI-DS128-LABEL: name: test_extload_local_v2s96_from_24_align1
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; CI-DS128: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; CI-DS128: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; CI-DS128: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; CI-DS128: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; CI-DS128: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; CI-DS128: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; CI-DS128: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; CI-DS128: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; CI-DS128: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI-DS128: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; CI-DS128: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; CI-DS128: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; CI-DS128: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; CI-DS128: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; CI-DS128: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI-DS128: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; CI-DS128: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI-DS128: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; CI-DS128: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; CI-DS128: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; CI-DS128: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI-DS128: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; CI-DS128: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; CI-DS128: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; CI-DS128: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI-DS128: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; CI-DS128: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; CI-DS128: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI-DS128: [[COPY13:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY13]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI-DS128: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI-DS128: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; CI-DS128: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; CI-DS128: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; CI-DS128: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; CI-DS128: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; CI-DS128: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; CI-DS128: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; CI-DS128: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; CI-DS128: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; CI-DS128: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; CI-DS128: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; CI-DS128: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; CI-DS128: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; CI-DS128: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; CI-DS128: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; CI-DS128: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; CI-DS128: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; CI-DS128: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; CI-DS128: [[COPY17:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; CI-DS128: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY17]], [[C3]]
    ; CI-DS128: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; CI-DS128: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; CI-DS128: [[PTR_ADD15:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C7]](s32)
    ; CI-DS128: [[LOAD16:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD15]](p3) :: (load 1 + 16, addrspace 3)
    ; CI-DS128: [[PTR_ADD16:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C]](s32)
    ; CI-DS128: [[LOAD17:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD16]](p3) :: (load 1 + 17, addrspace 3)
    ; CI-DS128: [[PTR_ADD17:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C1]](s32)
    ; CI-DS128: [[LOAD18:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD17]](p3) :: (load 1 + 18, addrspace 3)
    ; CI-DS128: [[PTR_ADD18:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C2]](s32)
    ; CI-DS128: [[LOAD19:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD18]](p3) :: (load 1 + 19, addrspace 3)
    ; CI-DS128: [[COPY18:%[0-9]+]]:_(s32) = COPY [[LOAD16]](s32)
    ; CI-DS128: [[AND16:%[0-9]+]]:_(s32) = G_AND [[COPY18]], [[C3]]
    ; CI-DS128: [[COPY19:%[0-9]+]]:_(s32) = COPY [[LOAD17]](s32)
    ; CI-DS128: [[AND17:%[0-9]+]]:_(s32) = G_AND [[COPY19]], [[C3]]
    ; CI-DS128: [[SHL12:%[0-9]+]]:_(s32) = G_SHL [[AND17]], [[C4]](s32)
    ; CI-DS128: [[OR12:%[0-9]+]]:_(s32) = G_OR [[AND16]], [[SHL12]]
    ; CI-DS128: [[COPY20:%[0-9]+]]:_(s32) = COPY [[LOAD18]](s32)
    ; CI-DS128: [[AND18:%[0-9]+]]:_(s32) = G_AND [[COPY20]], [[C3]]
    ; CI-DS128: [[SHL13:%[0-9]+]]:_(s32) = G_SHL [[AND18]], [[C5]](s32)
    ; CI-DS128: [[OR13:%[0-9]+]]:_(s32) = G_OR [[OR12]], [[SHL13]]
    ; CI-DS128: [[COPY21:%[0-9]+]]:_(s32) = COPY [[LOAD19]](s32)
    ; CI-DS128: [[AND19:%[0-9]+]]:_(s32) = G_AND [[COPY21]], [[C3]]
    ; CI-DS128: [[SHL14:%[0-9]+]]:_(s32) = G_SHL [[AND19]], [[C6]](s32)
    ; CI-DS128: [[OR14:%[0-9]+]]:_(s32) = G_OR [[OR13]], [[SHL14]]
    ; CI-DS128: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR11]](s32), [[OR14]](s32)
    ; CI-DS128: [[PTR_ADD19:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C4]](s32)
    ; CI-DS128: [[LOAD20:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD19]](p3) :: (load 1 + 20, addrspace 3)
    ; CI-DS128: [[PTR_ADD20:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C]](s32)
    ; CI-DS128: [[LOAD21:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD20]](p3) :: (load 1 + 21, addrspace 3)
    ; CI-DS128: [[PTR_ADD21:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C1]](s32)
    ; CI-DS128: [[LOAD22:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD21]](p3) :: (load 1 + 22, addrspace 3)
    ; CI-DS128: [[PTR_ADD22:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C2]](s32)
    ; CI-DS128: [[LOAD23:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD22]](p3) :: (load 1 + 23, addrspace 3)
    ; CI-DS128: [[COPY22:%[0-9]+]]:_(s32) = COPY [[LOAD20]](s32)
    ; CI-DS128: [[AND20:%[0-9]+]]:_(s32) = G_AND [[COPY22]], [[C3]]
    ; CI-DS128: [[COPY23:%[0-9]+]]:_(s32) = COPY [[LOAD21]](s32)
    ; CI-DS128: [[AND21:%[0-9]+]]:_(s32) = G_AND [[COPY23]], [[C3]]
    ; CI-DS128: [[SHL15:%[0-9]+]]:_(s32) = G_SHL [[AND21]], [[C4]](s32)
    ; CI-DS128: [[OR15:%[0-9]+]]:_(s32) = G_OR [[AND20]], [[SHL15]]
    ; CI-DS128: [[COPY24:%[0-9]+]]:_(s32) = COPY [[LOAD22]](s32)
    ; CI-DS128: [[AND22:%[0-9]+]]:_(s32) = G_AND [[COPY24]], [[C3]]
    ; CI-DS128: [[SHL16:%[0-9]+]]:_(s32) = G_SHL [[AND22]], [[C5]](s32)
    ; CI-DS128: [[OR16:%[0-9]+]]:_(s32) = G_OR [[OR15]], [[SHL16]]
    ; CI-DS128: [[COPY25:%[0-9]+]]:_(s32) = COPY [[LOAD23]](s32)
    ; CI-DS128: [[AND23:%[0-9]+]]:_(s32) = G_AND [[COPY25]], [[C3]]
    ; CI-DS128: [[SHL17:%[0-9]+]]:_(s32) = G_SHL [[AND23]], [[C6]](s32)
    ; CI-DS128: [[OR17:%[0-9]+]]:_(s32) = G_OR [[OR16]], [[SHL17]]
    ; CI-DS128: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR1]](<2 x s32>), 0
    ; CI-DS128: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[OR17]](s32), 64
    ; CI-DS128: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; CI-DS128: [[COPY26:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; CI-DS128: [[COPY27:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2 = COPY [[COPY26]](s96)
    ; CI-DS128: $vgpr3_vgpr4_vgpr5 = COPY [[COPY27]](s96)
    ; VI-LABEL: name: test_extload_local_v2s96_from_24_align1
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; VI: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; VI: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; VI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; VI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; VI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; VI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; VI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; VI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; VI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; VI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; VI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; VI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; VI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; VI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; VI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; VI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; VI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; VI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; VI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; VI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; VI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; VI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; VI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; VI: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; VI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; VI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; VI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; VI: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; VI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; VI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; VI: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; VI: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; VI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; VI: [[COPY13:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; VI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY13]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; VI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; VI: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; VI: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; VI: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; VI: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; VI: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; VI: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; VI: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; VI: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; VI: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; VI: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; VI: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; VI: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; VI: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; VI: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; VI: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; VI: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; VI: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; VI: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; VI: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; VI: [[COPY17:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; VI: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY17]], [[C3]]
    ; VI: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; VI: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; VI: [[PTR_ADD15:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C7]](s32)
    ; VI: [[LOAD16:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD15]](p3) :: (load 1 + 16, addrspace 3)
    ; VI: [[PTR_ADD16:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C]](s32)
    ; VI: [[LOAD17:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD16]](p3) :: (load 1 + 17, addrspace 3)
    ; VI: [[PTR_ADD17:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C1]](s32)
    ; VI: [[LOAD18:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD17]](p3) :: (load 1 + 18, addrspace 3)
    ; VI: [[PTR_ADD18:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C2]](s32)
    ; VI: [[LOAD19:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD18]](p3) :: (load 1 + 19, addrspace 3)
    ; VI: [[COPY18:%[0-9]+]]:_(s32) = COPY [[LOAD16]](s32)
    ; VI: [[AND16:%[0-9]+]]:_(s32) = G_AND [[COPY18]], [[C3]]
    ; VI: [[COPY19:%[0-9]+]]:_(s32) = COPY [[LOAD17]](s32)
    ; VI: [[AND17:%[0-9]+]]:_(s32) = G_AND [[COPY19]], [[C3]]
    ; VI: [[SHL12:%[0-9]+]]:_(s32) = G_SHL [[AND17]], [[C4]](s32)
    ; VI: [[OR12:%[0-9]+]]:_(s32) = G_OR [[AND16]], [[SHL12]]
    ; VI: [[COPY20:%[0-9]+]]:_(s32) = COPY [[LOAD18]](s32)
    ; VI: [[AND18:%[0-9]+]]:_(s32) = G_AND [[COPY20]], [[C3]]
    ; VI: [[SHL13:%[0-9]+]]:_(s32) = G_SHL [[AND18]], [[C5]](s32)
    ; VI: [[OR13:%[0-9]+]]:_(s32) = G_OR [[OR12]], [[SHL13]]
    ; VI: [[COPY21:%[0-9]+]]:_(s32) = COPY [[LOAD19]](s32)
    ; VI: [[AND19:%[0-9]+]]:_(s32) = G_AND [[COPY21]], [[C3]]
    ; VI: [[SHL14:%[0-9]+]]:_(s32) = G_SHL [[AND19]], [[C6]](s32)
    ; VI: [[OR14:%[0-9]+]]:_(s32) = G_OR [[OR13]], [[SHL14]]
    ; VI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR11]](s32), [[OR14]](s32)
    ; VI: [[PTR_ADD19:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C4]](s32)
    ; VI: [[LOAD20:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD19]](p3) :: (load 1 + 20, addrspace 3)
    ; VI: [[PTR_ADD20:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C]](s32)
    ; VI: [[LOAD21:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD20]](p3) :: (load 1 + 21, addrspace 3)
    ; VI: [[PTR_ADD21:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C1]](s32)
    ; VI: [[LOAD22:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD21]](p3) :: (load 1 + 22, addrspace 3)
    ; VI: [[PTR_ADD22:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C2]](s32)
    ; VI: [[LOAD23:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD22]](p3) :: (load 1 + 23, addrspace 3)
    ; VI: [[COPY22:%[0-9]+]]:_(s32) = COPY [[LOAD20]](s32)
    ; VI: [[AND20:%[0-9]+]]:_(s32) = G_AND [[COPY22]], [[C3]]
    ; VI: [[COPY23:%[0-9]+]]:_(s32) = COPY [[LOAD21]](s32)
    ; VI: [[AND21:%[0-9]+]]:_(s32) = G_AND [[COPY23]], [[C3]]
    ; VI: [[SHL15:%[0-9]+]]:_(s32) = G_SHL [[AND21]], [[C4]](s32)
    ; VI: [[OR15:%[0-9]+]]:_(s32) = G_OR [[AND20]], [[SHL15]]
    ; VI: [[COPY24:%[0-9]+]]:_(s32) = COPY [[LOAD22]](s32)
    ; VI: [[AND22:%[0-9]+]]:_(s32) = G_AND [[COPY24]], [[C3]]
    ; VI: [[SHL16:%[0-9]+]]:_(s32) = G_SHL [[AND22]], [[C5]](s32)
    ; VI: [[OR16:%[0-9]+]]:_(s32) = G_OR [[OR15]], [[SHL16]]
    ; VI: [[COPY25:%[0-9]+]]:_(s32) = COPY [[LOAD23]](s32)
    ; VI: [[AND23:%[0-9]+]]:_(s32) = G_AND [[COPY25]], [[C3]]
    ; VI: [[SHL17:%[0-9]+]]:_(s32) = G_SHL [[AND23]], [[C6]](s32)
    ; VI: [[OR17:%[0-9]+]]:_(s32) = G_OR [[OR16]], [[SHL17]]
    ; VI: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR1]](<2 x s32>), 0
    ; VI: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[OR17]](s32), 64
    ; VI: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; VI: [[COPY26:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; VI: [[COPY27:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[COPY26]](s96)
    ; VI: $vgpr3_vgpr4_vgpr5 = COPY [[COPY27]](s96)
    ; GFX9-LABEL: name: test_extload_local_v2s96_from_24_align1
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 1, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 1 + 1, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 1 + 2, addrspace 3)
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C2]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 1 + 3, addrspace 3)
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C3]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C4]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C5]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
    ; GFX9: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C6]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
    ; GFX9: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C7]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 1 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 1 + 5, addrspace 3)
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C1]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 1 + 6, addrspace 3)
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C2]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 1 + 7, addrspace 3)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C4]](s32)
    ; GFX9: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL3]]
    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
    ; GFX9: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND6]], [[C5]](s32)
    ; GFX9: [[OR4:%[0-9]+]]:_(s32) = G_OR [[OR3]], [[SHL4]]
    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
    ; GFX9: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C6]](s32)
    ; GFX9: [[OR5:%[0-9]+]]:_(s32) = G_OR [[OR4]], [[SHL5]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR2]](s32), [[OR5]](s32)
    ; GFX9: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 1 + 8, addrspace 3)
    ; GFX9: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; GFX9: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 1 + 9, addrspace 3)
    ; GFX9: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C1]](s32)
    ; GFX9: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 1 + 10, addrspace 3)
    ; GFX9: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C2]](s32)
    ; GFX9: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 1 + 11, addrspace 3)
    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; GFX9: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; GFX9: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
    ; GFX9: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C4]](s32)
    ; GFX9: [[OR6:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL6]]
    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; GFX9: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
    ; GFX9: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[AND10]], [[C5]](s32)
    ; GFX9: [[OR7:%[0-9]+]]:_(s32) = G_OR [[OR6]], [[SHL7]]
    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; GFX9: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
    ; GFX9: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C6]](s32)
    ; GFX9: [[OR8:%[0-9]+]]:_(s32) = G_OR [[OR7]], [[SHL8]]
    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; GFX9: [[COPY13:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; GFX9: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY13]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR8]](s32), 64
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; GFX9: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9: [[PTR_ADD11:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C8]](s32)
    ; GFX9: [[LOAD12:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD11]](p3) :: (load 1 + 12, addrspace 3)
    ; GFX9: [[PTR_ADD12:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C]](s32)
    ; GFX9: [[LOAD13:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD12]](p3) :: (load 1 + 13, addrspace 3)
    ; GFX9: [[PTR_ADD13:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C1]](s32)
    ; GFX9: [[LOAD14:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD13]](p3) :: (load 1 + 14, addrspace 3)
    ; GFX9: [[PTR_ADD14:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C2]](s32)
    ; GFX9: [[LOAD15:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD14]](p3) :: (load 1 + 15, addrspace 3)
    ; GFX9: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LOAD12]](s32)
    ; GFX9: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C3]]
    ; GFX9: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LOAD13]](s32)
    ; GFX9: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C3]]
    ; GFX9: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C4]](s32)
    ; GFX9: [[OR9:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL9]]
    ; GFX9: [[COPY16:%[0-9]+]]:_(s32) = COPY [[LOAD14]](s32)
    ; GFX9: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C3]]
    ; GFX9: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C5]](s32)
    ; GFX9: [[OR10:%[0-9]+]]:_(s32) = G_OR [[OR9]], [[SHL10]]
    ; GFX9: [[COPY17:%[0-9]+]]:_(s32) = COPY [[LOAD15]](s32)
    ; GFX9: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY17]], [[C3]]
    ; GFX9: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C6]](s32)
    ; GFX9: [[OR11:%[0-9]+]]:_(s32) = G_OR [[OR10]], [[SHL11]]
    ; GFX9: [[PTR_ADD15:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C7]](s32)
    ; GFX9: [[LOAD16:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD15]](p3) :: (load 1 + 16, addrspace 3)
    ; GFX9: [[PTR_ADD16:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C]](s32)
    ; GFX9: [[LOAD17:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD16]](p3) :: (load 1 + 17, addrspace 3)
    ; GFX9: [[PTR_ADD17:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C1]](s32)
    ; GFX9: [[LOAD18:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD17]](p3) :: (load 1 + 18, addrspace 3)
    ; GFX9: [[PTR_ADD18:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD15]], [[C2]](s32)
    ; GFX9: [[LOAD19:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD18]](p3) :: (load 1 + 19, addrspace 3)
    ; GFX9: [[COPY18:%[0-9]+]]:_(s32) = COPY [[LOAD16]](s32)
    ; GFX9: [[AND16:%[0-9]+]]:_(s32) = G_AND [[COPY18]], [[C3]]
    ; GFX9: [[COPY19:%[0-9]+]]:_(s32) = COPY [[LOAD17]](s32)
    ; GFX9: [[AND17:%[0-9]+]]:_(s32) = G_AND [[COPY19]], [[C3]]
    ; GFX9: [[SHL12:%[0-9]+]]:_(s32) = G_SHL [[AND17]], [[C4]](s32)
    ; GFX9: [[OR12:%[0-9]+]]:_(s32) = G_OR [[AND16]], [[SHL12]]
    ; GFX9: [[COPY20:%[0-9]+]]:_(s32) = COPY [[LOAD18]](s32)
    ; GFX9: [[AND18:%[0-9]+]]:_(s32) = G_AND [[COPY20]], [[C3]]
    ; GFX9: [[SHL13:%[0-9]+]]:_(s32) = G_SHL [[AND18]], [[C5]](s32)
    ; GFX9: [[OR13:%[0-9]+]]:_(s32) = G_OR [[OR12]], [[SHL13]]
    ; GFX9: [[COPY21:%[0-9]+]]:_(s32) = COPY [[LOAD19]](s32)
    ; GFX9: [[AND19:%[0-9]+]]:_(s32) = G_AND [[COPY21]], [[C3]]
    ; GFX9: [[SHL14:%[0-9]+]]:_(s32) = G_SHL [[AND19]], [[C6]](s32)
    ; GFX9: [[OR14:%[0-9]+]]:_(s32) = G_OR [[OR13]], [[SHL14]]
    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR11]](s32), [[OR14]](s32)
    ; GFX9: [[PTR_ADD19:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD11]], [[C4]](s32)
    ; GFX9: [[LOAD20:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD19]](p3) :: (load 1 + 20, addrspace 3)
    ; GFX9: [[PTR_ADD20:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C]](s32)
    ; GFX9: [[LOAD21:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD20]](p3) :: (load 1 + 21, addrspace 3)
    ; GFX9: [[PTR_ADD21:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C1]](s32)
    ; GFX9: [[LOAD22:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD21]](p3) :: (load 1 + 22, addrspace 3)
    ; GFX9: [[PTR_ADD22:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD19]], [[C2]](s32)
    ; GFX9: [[LOAD23:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD22]](p3) :: (load 1 + 23, addrspace 3)
    ; GFX9: [[COPY22:%[0-9]+]]:_(s32) = COPY [[LOAD20]](s32)
    ; GFX9: [[AND20:%[0-9]+]]:_(s32) = G_AND [[COPY22]], [[C3]]
    ; GFX9: [[COPY23:%[0-9]+]]:_(s32) = COPY [[LOAD21]](s32)
    ; GFX9: [[AND21:%[0-9]+]]:_(s32) = G_AND [[COPY23]], [[C3]]
    ; GFX9: [[SHL15:%[0-9]+]]:_(s32) = G_SHL [[AND21]], [[C4]](s32)
    ; GFX9: [[OR15:%[0-9]+]]:_(s32) = G_OR [[AND20]], [[SHL15]]
    ; GFX9: [[COPY24:%[0-9]+]]:_(s32) = COPY [[LOAD22]](s32)
    ; GFX9: [[AND22:%[0-9]+]]:_(s32) = G_AND [[COPY24]], [[C3]]
    ; GFX9: [[SHL16:%[0-9]+]]:_(s32) = G_SHL [[AND22]], [[C5]](s32)
    ; GFX9: [[OR16:%[0-9]+]]:_(s32) = G_OR [[OR15]], [[SHL16]]
    ; GFX9: [[COPY25:%[0-9]+]]:_(s32) = COPY [[LOAD23]](s32)
    ; GFX9: [[AND23:%[0-9]+]]:_(s32) = G_AND [[COPY25]], [[C3]]
    ; GFX9: [[SHL17:%[0-9]+]]:_(s32) = G_SHL [[AND23]], [[C6]](s32)
    ; GFX9: [[OR17:%[0-9]+]]:_(s32) = G_OR [[OR16]], [[SHL17]]
    ; GFX9: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR1]](<2 x s32>), 0
    ; GFX9: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[OR17]](s32), 64
    ; GFX9: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; GFX9: [[COPY26:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; GFX9: [[COPY27:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[COPY26]](s96)
    ; GFX9: $vgpr3_vgpr4_vgpr5 = COPY [[COPY27]](s96)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_v2s96_from_24_align1
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 12, align 1, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD]](<3 x s32>)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9-UNALIGNED: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9-UNALIGNED: [[LOAD1:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 12 + 12, align 1, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD1]](<3 x s32>)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; GFX9-UNALIGNED: [[COPY2:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2 = COPY [[COPY1]](s96)
    ; GFX9-UNALIGNED: $vgpr3_vgpr4_vgpr5 = COPY [[COPY2]](s96)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s96>) = G_LOAD %0 :: (load 24, align 1, addrspace 3)
    %2:_(s96) = G_EXTRACT %1, 0
    %3:_(s96) = G_EXTRACT %1, 96
    $vgpr0_vgpr1_vgpr2 = COPY %2
    $vgpr3_vgpr4_vgpr5 = COPY %3
...

---
name: test_extload_local_v2s96_from_24_align2
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_v2s96_from_24_align2
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; SI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; SI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; SI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; SI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; SI: [[COPY7:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; SI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY7]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR2]](s32), 64
    ; SI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; SI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; SI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; SI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; SI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; SI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; SI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C1]]
    ; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; SI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; SI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C3]](s32)
    ; SI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 2 + 16, addrspace 3)
    ; SI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; SI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 2 + 18, addrspace 3)
    ; SI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; SI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C1]]
    ; SI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; SI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C1]]
    ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C2]](s32)
    ; SI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL4]]
    ; SI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR3]](s32), [[OR4]](s32)
    ; SI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C4]](s32)
    ; SI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 2 + 20, addrspace 3)
    ; SI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD9]], [[C]](s32)
    ; SI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 2 + 22, addrspace 3)
    ; SI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; SI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C1]]
    ; SI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; SI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C1]]
    ; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C2]](s32)
    ; SI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[AND10]], [[SHL5]]
    ; SI: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR1]](<2 x s32>), 0
    ; SI: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[OR5]](s32), 64
    ; SI: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; SI: [[COPY14:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; SI: [[COPY15:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[COPY14]](s96)
    ; SI: $vgpr3_vgpr4_vgpr5 = COPY [[COPY15]](s96)
    ; CI-LABEL: name: test_extload_local_v2s96_from_24_align2
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; CI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; CI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; CI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; CI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; CI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; CI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; CI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; CI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; CI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI: [[COPY7:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; CI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY7]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR2]](s32), 64
    ; CI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; CI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; CI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; CI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; CI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; CI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C1]]
    ; CI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; CI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; CI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C3]](s32)
    ; CI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 2 + 16, addrspace 3)
    ; CI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 2 + 18, addrspace 3)
    ; CI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C1]]
    ; CI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C1]]
    ; CI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C2]](s32)
    ; CI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL4]]
    ; CI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR3]](s32), [[OR4]](s32)
    ; CI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C4]](s32)
    ; CI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 2 + 20, addrspace 3)
    ; CI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD9]], [[C]](s32)
    ; CI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 2 + 22, addrspace 3)
    ; CI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C1]]
    ; CI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C1]]
    ; CI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C2]](s32)
    ; CI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[AND10]], [[SHL5]]
    ; CI: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR1]](<2 x s32>), 0
    ; CI: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[OR5]](s32), 64
    ; CI: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; CI: [[COPY14:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; CI: [[COPY15:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; CI: $vgpr0_vgpr1_vgpr2 = COPY [[COPY14]](s96)
    ; CI: $vgpr3_vgpr4_vgpr5 = COPY [[COPY15]](s96)
    ; CI-DS128-LABEL: name: test_extload_local_v2s96_from_24_align2
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; CI-DS128: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; CI-DS128: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; CI-DS128: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; CI-DS128: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; CI-DS128: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; CI-DS128: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; CI-DS128: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; CI-DS128: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; CI-DS128: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; CI-DS128: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; CI-DS128: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; CI-DS128: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; CI-DS128: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; CI-DS128: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; CI-DS128: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; CI-DS128: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; CI-DS128: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; CI-DS128: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI-DS128: [[COPY7:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY7]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR2]](s32), 64
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI-DS128: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI-DS128: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; CI-DS128: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; CI-DS128: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; CI-DS128: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; CI-DS128: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; CI-DS128: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; CI-DS128: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; CI-DS128: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C1]]
    ; CI-DS128: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; CI-DS128: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; CI-DS128: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C3]](s32)
    ; CI-DS128: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 2 + 16, addrspace 3)
    ; CI-DS128: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; CI-DS128: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 2 + 18, addrspace 3)
    ; CI-DS128: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; CI-DS128: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C1]]
    ; CI-DS128: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; CI-DS128: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C1]]
    ; CI-DS128: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C2]](s32)
    ; CI-DS128: [[OR4:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL4]]
    ; CI-DS128: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR3]](s32), [[OR4]](s32)
    ; CI-DS128: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C4]](s32)
    ; CI-DS128: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 2 + 20, addrspace 3)
    ; CI-DS128: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD9]], [[C]](s32)
    ; CI-DS128: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 2 + 22, addrspace 3)
    ; CI-DS128: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; CI-DS128: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C1]]
    ; CI-DS128: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; CI-DS128: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C1]]
    ; CI-DS128: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C2]](s32)
    ; CI-DS128: [[OR5:%[0-9]+]]:_(s32) = G_OR [[AND10]], [[SHL5]]
    ; CI-DS128: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR1]](<2 x s32>), 0
    ; CI-DS128: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[OR5]](s32), 64
    ; CI-DS128: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; CI-DS128: [[COPY14:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; CI-DS128: [[COPY15:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2 = COPY [[COPY14]](s96)
    ; CI-DS128: $vgpr3_vgpr4_vgpr5 = COPY [[COPY15]](s96)
    ; VI-LABEL: name: test_extload_local_v2s96_from_24_align2
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; VI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; VI: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; VI: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; VI: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; VI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; VI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; VI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; VI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; VI: [[COPY7:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; VI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY7]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR2]](s32), 64
    ; VI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; VI: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; VI: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; VI: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; VI: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; VI: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; VI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; VI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; VI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; VI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C1]]
    ; VI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; VI: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; VI: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C3]](s32)
    ; VI: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 2 + 16, addrspace 3)
    ; VI: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; VI: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 2 + 18, addrspace 3)
    ; VI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; VI: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C1]]
    ; VI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; VI: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C1]]
    ; VI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C2]](s32)
    ; VI: [[OR4:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL4]]
    ; VI: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR3]](s32), [[OR4]](s32)
    ; VI: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C4]](s32)
    ; VI: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 2 + 20, addrspace 3)
    ; VI: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD9]], [[C]](s32)
    ; VI: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 2 + 22, addrspace 3)
    ; VI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; VI: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C1]]
    ; VI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; VI: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C1]]
    ; VI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C2]](s32)
    ; VI: [[OR5:%[0-9]+]]:_(s32) = G_OR [[AND10]], [[SHL5]]
    ; VI: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR1]](<2 x s32>), 0
    ; VI: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[OR5]](s32), 64
    ; VI: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; VI: [[COPY14:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; VI: [[COPY15:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[COPY14]](s96)
    ; VI: $vgpr3_vgpr4_vgpr5 = COPY [[COPY15]](s96)
    ; GFX9-LABEL: name: test_extload_local_v2s96_from_24_align2
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 2 + 2, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LOAD]](s32)
    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LOAD1]](s32)
    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32)
    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C3]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 2 + 4, addrspace 3)
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 2 + 6, addrspace 3)
    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LOAD2]](s32)
    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LOAD3]](s32)
    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR]](s32), [[OR1]](s32)
    ; GFX9: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD3:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C4]](s32)
    ; GFX9: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p3) :: (load 2 + 8, addrspace 3)
    ; GFX9: [[PTR_ADD4:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD3]], [[C]](s32)
    ; GFX9: [[LOAD5:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD4]](p3) :: (load 2 + 10, addrspace 3)
    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LOAD4]](s32)
    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LOAD5]](s32)
    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND5]], [[C2]](s32)
    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[AND4]], [[SHL2]]
    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; GFX9: [[COPY7:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; GFX9: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY7]], [[BUILD_VECTOR]](<2 x s32>), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[OR2]](s32), 64
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; GFX9: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9: [[PTR_ADD5:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C5]](s32)
    ; GFX9: [[LOAD6:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD5]](p3) :: (load 2 + 12, addrspace 3)
    ; GFX9: [[PTR_ADD6:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C]](s32)
    ; GFX9: [[LOAD7:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD6]](p3) :: (load 2 + 14, addrspace 3)
    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LOAD6]](s32)
    ; GFX9: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LOAD7]](s32)
    ; GFX9: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C1]]
    ; GFX9: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C2]](s32)
    ; GFX9: [[OR3:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL3]]
    ; GFX9: [[PTR_ADD7:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C3]](s32)
    ; GFX9: [[LOAD8:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD7]](p3) :: (load 2 + 16, addrspace 3)
    ; GFX9: [[PTR_ADD8:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD7]], [[C]](s32)
    ; GFX9: [[LOAD9:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD8]](p3) :: (load 2 + 18, addrspace 3)
    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LOAD8]](s32)
    ; GFX9: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C1]]
    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LOAD9]](s32)
    ; GFX9: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C1]]
    ; GFX9: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[AND9]], [[C2]](s32)
    ; GFX9: [[OR4:%[0-9]+]]:_(s32) = G_OR [[AND8]], [[SHL4]]
    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[OR3]](s32), [[OR4]](s32)
    ; GFX9: [[PTR_ADD9:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD5]], [[C4]](s32)
    ; GFX9: [[LOAD10:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p3) :: (load 2 + 20, addrspace 3)
    ; GFX9: [[PTR_ADD10:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD9]], [[C]](s32)
    ; GFX9: [[LOAD11:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD10]](p3) :: (load 2 + 22, addrspace 3)
    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LOAD10]](s32)
    ; GFX9: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C1]]
    ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LOAD11]](s32)
    ; GFX9: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C1]]
    ; GFX9: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[AND11]], [[C2]](s32)
    ; GFX9: [[OR5:%[0-9]+]]:_(s32) = G_OR [[AND10]], [[SHL5]]
    ; GFX9: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[BUILD_VECTOR1]](<2 x s32>), 0
    ; GFX9: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[OR5]](s32), 64
    ; GFX9: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; GFX9: [[COPY14:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; GFX9: [[COPY15:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[COPY14]](s96)
    ; GFX9: $vgpr3_vgpr4_vgpr5 = COPY [[COPY15]](s96)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_v2s96_from_24_align2
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 12, align 2, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD]](<3 x s32>)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9-UNALIGNED: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9-UNALIGNED: [[LOAD1:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 12 + 12, align 2, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD1]](<3 x s32>)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; GFX9-UNALIGNED: [[COPY2:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2 = COPY [[COPY1]](s96)
    ; GFX9-UNALIGNED: $vgpr3_vgpr4_vgpr5 = COPY [[COPY2]](s96)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s96>) = G_LOAD %0 :: (load 24, align 2, addrspace 3)
    %2:_(s96) = G_EXTRACT %1, 0
    %3:_(s96) = G_EXTRACT %1, 96
    $vgpr0_vgpr1_vgpr2 = COPY %2
    $vgpr3_vgpr4_vgpr5 = COPY %3
...

---
name: test_extload_local_v2s96_from_24_align4
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_v2s96_from_24_align4
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; SI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; SI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; SI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY1]], [[LOAD]](<2 x s32>), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; SI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD1]](p3) :: (load 8 + 12, align 4, addrspace 3)
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 20, addrspace 3)
    ; SI: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD2]](<2 x s32>), 0
    ; SI: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[LOAD3]](s32), 64
    ; SI: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; SI: [[COPY2:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; SI: [[COPY3:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[COPY2]](s96)
    ; SI: $vgpr3_vgpr4_vgpr5 = COPY [[COPY3]](s96)
    ; CI-LABEL: name: test_extload_local_v2s96_from_24_align4
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; CI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; CI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY1]], [[LOAD]](<2 x s32>), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; CI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD1]](p3) :: (load 8 + 12, align 4, addrspace 3)
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 20, addrspace 3)
    ; CI: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD2]](<2 x s32>), 0
    ; CI: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[LOAD3]](s32), 64
    ; CI: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; CI: [[COPY2:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; CI: [[COPY3:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; CI: $vgpr0_vgpr1_vgpr2 = COPY [[COPY2]](s96)
    ; CI: $vgpr3_vgpr4_vgpr5 = COPY [[COPY3]](s96)
    ; CI-DS128-LABEL: name: test_extload_local_v2s96_from_24_align4
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY1]], [[LOAD]](<2 x s32>), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD1]](p3) :: (load 8 + 12, align 4, addrspace 3)
    ; CI-DS128: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI-DS128: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 20, addrspace 3)
    ; CI-DS128: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD2]](<2 x s32>), 0
    ; CI-DS128: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[LOAD3]](s32), 64
    ; CI-DS128: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; CI-DS128: [[COPY3:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2 = COPY [[COPY2]](s96)
    ; CI-DS128: $vgpr3_vgpr4_vgpr5 = COPY [[COPY3]](s96)
    ; VI-LABEL: name: test_extload_local_v2s96_from_24_align4
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; VI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; VI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; VI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY1]], [[LOAD]](<2 x s32>), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; VI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD1]](p3) :: (load 8 + 12, align 4, addrspace 3)
    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; VI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 20, addrspace 3)
    ; VI: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD2]](<2 x s32>), 0
    ; VI: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[LOAD3]](s32), 64
    ; VI: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; VI: [[COPY2:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; VI: [[COPY3:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[COPY2]](s96)
    ; VI: $vgpr3_vgpr4_vgpr5 = COPY [[COPY3]](s96)
    ; GFX9-LABEL: name: test_extload_local_v2s96_from_24_align4
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 4, addrspace 3)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, addrspace 3)
    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; GFX9: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; GFX9: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY1]], [[LOAD]](<2 x s32>), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD1]](p3) :: (load 8 + 12, align 4, addrspace 3)
    ; GFX9: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; GFX9: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 20, addrspace 3)
    ; GFX9: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD2]](<2 x s32>), 0
    ; GFX9: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[LOAD3]](s32), 64
    ; GFX9: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; GFX9: [[COPY2:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; GFX9: [[COPY3:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[COPY2]](s96)
    ; GFX9: $vgpr3_vgpr4_vgpr5 = COPY [[COPY3]](s96)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_v2s96_from_24_align4
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 12, align 4, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD]](<3 x s32>)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9-UNALIGNED: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9-UNALIGNED: [[LOAD1:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 12 + 12, align 4, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD1]](<3 x s32>)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; GFX9-UNALIGNED: [[COPY2:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2 = COPY [[COPY1]](s96)
    ; GFX9-UNALIGNED: $vgpr3_vgpr4_vgpr5 = COPY [[COPY2]](s96)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s96>) = G_LOAD %0 :: (load 24, align 4, addrspace 3)
    %2:_(s96) = G_EXTRACT %1, 0
    %3:_(s96) = G_EXTRACT %1, 96
    $vgpr0_vgpr1_vgpr2 = COPY %2
    $vgpr3_vgpr4_vgpr5 = COPY %3
...

---
name: test_extload_local_v2s96_from_24_align16
body: |
  bb.0:
    liveins: $vgpr0

    ; SI-LABEL: name: test_extload_local_v2s96_from_24_align16
    ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; SI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 16, addrspace 3)
    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; SI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; SI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, align 8, addrspace 3)
    ; SI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; SI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; SI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY1]], [[LOAD]](<2 x s32>), 0
    ; SI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; SI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; SI: [[LOAD2:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD1]](p3) :: (load 8 + 12, align 4, addrspace 3)
    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; SI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 20, addrspace 3)
    ; SI: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD2]](<2 x s32>), 0
    ; SI: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[LOAD3]](s32), 64
    ; SI: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; SI: [[COPY2:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; SI: [[COPY3:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; SI: $vgpr0_vgpr1_vgpr2 = COPY [[COPY2]](s96)
    ; SI: $vgpr3_vgpr4_vgpr5 = COPY [[COPY3]](s96)
    ; CI-LABEL: name: test_extload_local_v2s96_from_24_align16
    ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p3) :: (load 8, align 16, addrspace 3)
    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p3) :: (load 4 + 8, align 8, addrspace 3)
    ; CI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY [[DEF]](<3 x s32>)
    ; CI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[COPY1]], [[LOAD]](<2 x s32>), 0
    ; CI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD1]](s32), 64
    ; CI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C1]](s32)
    ; CI: [[LOAD2:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD1]](p3) :: (load 8 + 12, align 4, addrspace 3)
    ; CI: [[PTR_ADD2:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD1]], [[C]](s32)
    ; CI: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p3) :: (load 4 + 20, addrspace 3)
    ; CI: [[INSERT2:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD2]](<2 x s32>), 0
    ; CI: [[INSERT3:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT2]], [[LOAD3]](s32), 64
    ; CI: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT3]](<3 x s32>)
    ; CI: [[COPY2:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; CI: [[COPY3:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; CI: $vgpr0_vgpr1_vgpr2 = COPY [[COPY2]](s96)
    ; CI: $vgpr3_vgpr4_vgpr5 = COPY [[COPY3]](s96)
    ; CI-DS128-LABEL: name: test_extload_local_v2s96_from_24_align16
    ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 12, align 16, addrspace 3)
    ; CI-DS128: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD]](<3 x s32>)
    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; CI-DS128: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; CI-DS128: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 12, align 4, addrspace 3)
    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; CI-DS128: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD]], [[C1]](s32)
    ; CI-DS128: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 4 + 20, addrspace 3)
    ; CI-DS128: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; CI-DS128: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD1]](<2 x s32>), 0
    ; CI-DS128: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD2]](s32), 64
    ; CI-DS128: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; CI-DS128: $vgpr0_vgpr1_vgpr2 = COPY [[COPY1]](s96)
    ; CI-DS128: $vgpr3_vgpr4_vgpr5 = COPY [[COPY2]](s96)
    ; VI-LABEL: name: test_extload_local_v2s96_from_24_align16
    ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; VI: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 12, align 16, addrspace 3)
    ; VI: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD]](<3 x s32>)
    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; VI: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; VI: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 12, align 4, addrspace 3)
    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD]], [[C1]](s32)
    ; VI: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 4 + 20, addrspace 3)
    ; VI: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; VI: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD1]](<2 x s32>), 0
    ; VI: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD2]](s32), 64
    ; VI: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; VI: [[COPY1:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; VI: [[COPY2:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; VI: $vgpr0_vgpr1_vgpr2 = COPY [[COPY1]](s96)
    ; VI: $vgpr3_vgpr4_vgpr5 = COPY [[COPY2]](s96)
    ; GFX9-LABEL: name: test_extload_local_v2s96_from_24_align16
    ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 12, align 16, addrspace 3)
    ; GFX9: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD]](<3 x s32>)
    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 8 + 12, align 4, addrspace 3)
    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
    ; GFX9: [[PTR_ADD1:%[0-9]+]]:_(p3) = G_PTR_ADD [[PTR_ADD]], [[C1]](s32)
    ; GFX9: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p3) :: (load 4 + 20, addrspace 3)
    ; GFX9: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
    ; GFX9: [[INSERT:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[DEF]], [[LOAD1]](<2 x s32>), 0
    ; GFX9: [[INSERT1:%[0-9]+]]:_(<3 x s32>) = G_INSERT [[INSERT]], [[LOAD2]](s32), 64
    ; GFX9: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[INSERT1]](<3 x s32>)
    ; GFX9: [[COPY1:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; GFX9: [[COPY2:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[COPY1]](s96)
    ; GFX9: $vgpr3_vgpr4_vgpr5 = COPY [[COPY2]](s96)
    ; GFX9-UNALIGNED-LABEL: name: test_extload_local_v2s96_from_24_align16
    ; GFX9-UNALIGNED: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
    ; GFX9-UNALIGNED: [[LOAD:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[COPY]](p3) :: (load 12, align 16, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD]](<3 x s32>)
    ; GFX9-UNALIGNED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
    ; GFX9-UNALIGNED: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32)
    ; GFX9-UNALIGNED: [[LOAD1:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[PTR_ADD]](p3) :: (load 12 + 12, align 4, addrspace 3)
    ; GFX9-UNALIGNED: [[BITCAST1:%[0-9]+]]:_(s96) = G_BITCAST [[LOAD1]](<3 x s32>)
    ; GFX9-UNALIGNED: [[COPY1:%[0-9]+]]:_(s96) = COPY [[BITCAST]](s96)
    ; GFX9-UNALIGNED: [[COPY2:%[0-9]+]]:_(s96) = COPY [[BITCAST1]](s96)
    ; GFX9-UNALIGNED: $vgpr0_vgpr1_vgpr2 = COPY [[COPY1]](s96)
    ; GFX9-UNALIGNED: $vgpr3_vgpr4_vgpr5 = COPY [[COPY2]](s96)
    %0:_(p3) = COPY $vgpr0
    %1:_(<2 x s96>) = G_LOAD %0 :: (load 24, align 16, addrspace 3)
    %2:_(s96) = G_EXTRACT %1, 0
    %3:_(s96) = G_EXTRACT %1, 96
    $vgpr0_vgpr1_vgpr2 = COPY %2
    $vgpr3_vgpr4_vgpr5 = COPY %3
...