llvm/test/CodeGen/AMDGPU/sdwa-peephole-cndmask-wave32.mir

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
# RUN: llc %s -mtriple=amdgcn -mcpu=gfx1030 -run-pass=si-peephole-sdwa -o - | FileCheck %s

---
name: cndmask_b32 # can be directly converted to SDWA without a copy to VCC
tracksRegLiveness: true
body: |
  bb.0:
    liveins: $vgpr0, $vgpr1, $vcc

    ; CHECK-LABEL: name: cndmask_b32
    ; CHECK: liveins: $vgpr0, $vgpr1, $vcc
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
    ; CHECK-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 16, [[COPY]], implicit $exec
    ; CHECK-NEXT: [[V_LSHRREV_B32_e64_1:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 16, [[COPY1]], implicit $exec
    ; CHECK-NEXT: [[V_CNDMASK_B32_sdwa:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_sdwa 0, [[COPY]], 0, [[COPY1]], 0, 6, 0, 5, 5, implicit $vcc_lo, implicit $exec
    ; CHECK-NEXT: $vgpr0 = COPY [[V_CNDMASK_B32_sdwa]]
    ; CHECK-NEXT: SI_RETURN implicit $vgpr0
    %0:vgpr_32 = COPY $vgpr1
    %1:vgpr_32 = COPY $vgpr0
    %2:vgpr_32 = V_LSHRREV_B32_e64 16, %0, implicit $exec
    %3:vgpr_32 = V_LSHRREV_B32_e64 16, %1, implicit $exec
    %4:vgpr_32 = V_CNDMASK_B32_e32 killed %2, killed %3, implicit $exec, implicit $vcc
    $vgpr0 = COPY %4
    SI_RETURN implicit $vgpr0
...

# For SDWA conversion of V_CNDMASK, the carry-in operand must be
# available in VCC_LO.  This is achieved by introducing a COPY
# instruction. Comparison instructions could be changed to VOP2 form
# intead, but we prefer to use a COPY.

---
name: carry-compare
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $vgpr0

    ; CHECK-LABEL: name: carry-compare
    ; CHECK: liveins: $vgpr0
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
    ; CHECK-NEXT: [[DEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
    ; CHECK-NEXT: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 undef [[DEF]], 1, implicit $exec
    ; CHECK-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 16, undef [[DEF1]], implicit $exec
    ; CHECK-NEXT: $vcc_lo = COPY killed [[V_CMP_EQ_U32_e64_]]
    ; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
    ; CHECK-NEXT: [[V_CNDMASK_B32_sdwa:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_sdwa 0, [[V_MOV_B32_e32_]], 0, undef [[DEF1]], 0, 6, 0, 6, 5, implicit $vcc_lo, implicit $exec
    ; CHECK-NEXT: $vgpr0 = COPY [[V_CNDMASK_B32_sdwa]]
    ; CHECK-NEXT: SI_RETURN implicit $vgpr0
    %0:sreg_32_xm0_xexec = IMPLICIT_DEF
    %1:vgpr_32 = IMPLICIT_DEF
    %2:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 undef %0, 1, implicit $exec
    %3:vgpr_32 = V_LSHRREV_B32_e64 16, undef %1, implicit $exec
    %4:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, %3, killed %2, implicit $exec
    $vgpr0 = COPY %4
    SI_RETURN implicit $vgpr0
...

---
name: carry-compare-class
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $vgpr0

    ; CHECK-LABEL: name: carry-compare-class
    ; CHECK: liveins: $vgpr0
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
    ; CHECK-NEXT: [[DEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
    ; CHECK-NEXT: [[V_CMP_CLASS_F32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_CLASS_F32_e64 2, undef [[DEF]], 1, implicit $exec
    ; CHECK-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 16, undef [[DEF1]], implicit $exec
    ; CHECK-NEXT: $vcc_lo = COPY killed [[V_CMP_CLASS_F32_e64_]]
    ; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
    ; CHECK-NEXT: [[V_CNDMASK_B32_sdwa:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_sdwa 0, [[V_MOV_B32_e32_]], 0, undef [[DEF1]], 0, 6, 0, 6, 5, implicit $vcc_lo, implicit $exec
    ; CHECK-NEXT: $vgpr0 = COPY [[V_CNDMASK_B32_sdwa]]
    ; CHECK-NEXT: SI_RETURN implicit $vgpr0
    %0:sreg_32_xm0_xexec = IMPLICIT_DEF
    %1:vgpr_32 = IMPLICIT_DEF
    %2:sreg_32_xm0_xexec = V_CMP_CLASS_F32_e64 2, undef %0, 1, implicit $exec
    %3:vgpr_32 = V_LSHRREV_B32_e64 16, undef %1, implicit $exec
    %4:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, %3, killed %2, implicit $exec
    $vgpr0 = COPY %4
    SI_RETURN implicit $vgpr0
...

...
---
name: carry-non-compare
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $vgpr0

    ; CHECK-LABEL: name: carry-non-compare
    ; CHECK: liveins: $vgpr0
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
    ; CHECK-NEXT: [[DEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
    ; CHECK-NEXT: [[V_PK_MAX_F16_:%[0-9]+]]:vgpr_32 = V_PK_MAX_F16 8, [[DEF1]], 8, [[DEF1]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
    ; CHECK-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 16, [[V_PK_MAX_F16_]], implicit $exec
    ; CHECK-NEXT: $vcc_lo = COPY killed undef [[DEF]]
    ; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
    ; CHECK-NEXT: [[V_CNDMASK_B32_sdwa:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_sdwa 0, [[V_MOV_B32_e32_]], 0, [[V_PK_MAX_F16_]], 0, 6, 0, 6, 5, implicit $vcc_lo, implicit $exec
    ; CHECK-NEXT: $vgpr0 = COPY [[V_CNDMASK_B32_sdwa]]
    ; CHECK-NEXT: SI_RETURN implicit $vgpr0
    %0:sreg_32_xm0_xexec = IMPLICIT_DEF
    %1:vgpr_32 = IMPLICIT_DEF
    %2:vgpr_32 = V_PK_MAX_F16 8, %1, 8, %1, 0, 0, 0, 0, 0, implicit $mode, implicit $exec
    %3:vgpr_32 = V_LSHRREV_B32_e64 16, %2, implicit $exec
    %4:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, %3, killed undef %0, implicit $exec
    $vgpr0 = COPY %4
    SI_RETURN implicit $vgpr0
...

...
---
name: carry-multiuse
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $vgpr0, $vgpr1

    ; CHECK-LABEL: name: carry-multiuse
    ; CHECK: liveins: $vgpr0, $vgpr1
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
    ; CHECK-NEXT: [[DEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
    ; CHECK-NEXT: [[V_PK_MAX_F16_:%[0-9]+]]:vgpr_32 = V_PK_MAX_F16 8, [[DEF1]], 8, [[DEF1]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
    ; CHECK-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 16, [[V_PK_MAX_F16_]], implicit $exec
    ; CHECK-NEXT: $vcc_lo = COPY undef [[DEF]]
    ; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
    ; CHECK-NEXT: [[V_CNDMASK_B32_sdwa:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_sdwa 0, [[V_MOV_B32_e32_]], 0, [[V_PK_MAX_F16_]], 0, 6, 0, 6, 5, implicit $vcc_lo, implicit $exec
    ; CHECK-NEXT: $vgpr0 = COPY [[V_CNDMASK_B32_sdwa]]
    ; CHECK-NEXT: $vgpr1 = COPY [[DEF]]
    ; CHECK-NEXT: SI_RETURN implicit $vgpr0
    %0:sreg_32_xm0_xexec = IMPLICIT_DEF
    %1:vgpr_32 = IMPLICIT_DEF
    %2:vgpr_32 = V_PK_MAX_F16 8, %1, 8, %1, 0, 0, 0, 0, 0, implicit $mode, implicit $exec
    %3:vgpr_32 = V_LSHRREV_B32_e64 16, %2, implicit $exec
    %4:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, %3, undef %0, implicit $exec
    $vgpr0 = COPY %4
    $vgpr1 = COPY %0
    SI_RETURN implicit $vgpr0
...

...
---
name: live-vcc # cannot convert because of live VCC
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $vgpr0

    ; CHECK-LABEL: name: live-vcc
    ; CHECK: liveins: $vgpr0
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
    ; CHECK-NEXT: [[DEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
    ; CHECK-NEXT: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 [[DEF]], 1, implicit $exec
    ; CHECK-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 16, [[DEF1]], implicit $exec
    ; CHECK-NEXT: V_CMP_EQ_U32_e32 1, undef [[DEF1]], implicit-def $vcc_lo, implicit $exec
    ; CHECK-NEXT: [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, [[V_LSHRREV_B32_e64_]], [[V_CMP_EQ_U32_e64_]], implicit $exec
    ; CHECK-NEXT: [[V_CNDMASK_B32_e32_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e32 killed [[V_LSHRREV_B32_e64_]], killed [[V_LSHRREV_B32_e64_]], implicit $vcc_lo, implicit $exec
    ; CHECK-NEXT: $vgpr0 = COPY [[V_CNDMASK_B32_e64_]]
    ; CHECK-NEXT: SI_RETURN implicit $vgpr0
    %0:sreg_32_xm0_xexec = IMPLICIT_DEF
    %2:vgpr_32 = IMPLICIT_DEF
    %3:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 %0, 1, implicit $exec
    %5:vgpr_32 = V_LSHRREV_B32_e64 16, %2, implicit $exec
    V_CMP_EQ_U32_e32 1, undef %2, implicit-def $vcc, implicit $exec
    %6:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, %5, %3, implicit $exec
    %7:vgpr_32 = V_CNDMASK_B32_e32 killed %5, killed %5, implicit $vcc, implicit $exec
    $vgpr0 = COPY %6
    SI_RETURN implicit $vgpr0
...

...
---
name: cannot-shrink-with-source-mods
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $vgpr0

    ; CHECK-LABEL: name: cannot-shrink-with-source-mods
    ; CHECK: liveins: $vgpr0
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
    ; CHECK-NEXT: [[DEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
    ; CHECK-NEXT: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 [[DEF]], 0, implicit $exec
    ; CHECK-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 16, [[DEF1]], implicit $exec
    ; CHECK-NEXT: [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 1, 0, 0, [[V_LSHRREV_B32_e64_]], killed [[V_CMP_EQ_U32_e64_]], implicit $exec
    ; CHECK-NEXT: $vgpr0 = COPY [[V_CNDMASK_B32_e64_]]
    ; CHECK-NEXT: SI_RETURN implicit $vgpr0
    %0:sreg_32_xm0_xexec = IMPLICIT_DEF
    %1:vgpr_32 = IMPLICIT_DEF
    %2:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 %0, 0, implicit $exec
    %3:vgpr_32 = V_LSHRREV_B32_e64 16, %1, implicit $exec
    %4:vgpr_32 = V_CNDMASK_B32_e64 1, 0, 0, %3, killed %2, implicit $exec
    $vgpr0 = COPY %4
    SI_RETURN implicit $vgpr0
...

...
---
name: missing-carry-def
tracksRegLiveness: true
registers:
  - { id: 0, class: sreg_32_xm0_xexec }
body:             |
  bb.0:
    liveins: $vgpr0

    ; CHECK-LABEL: name: missing-carry-def
    ; CHECK: liveins: $vgpr0
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
    ; CHECK-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 16, [[DEF]], implicit $exec
    ; CHECK-NEXT: [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, [[V_LSHRREV_B32_e64_]], undef %0:sreg_32_xm0_xexec, implicit $exec
    ; CHECK-NEXT: $vgpr0 = COPY [[V_CNDMASK_B32_e64_]]
    ; CHECK-NEXT: SI_RETURN implicit $vgpr0
    %1:sreg_32_xm0_xexec = IMPLICIT_DEF
    %2:vgpr_32 = V_LSHRREV_B32_e64 16, %1:sreg_32_xm0_xexec, implicit $exec
    %3:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, %2, undef %0:sreg_32_xm0_xexec, implicit $exec
    $vgpr0 = COPY %3
    SI_RETURN implicit $vgpr0
...

---
name: cndmask-not-converted
tracksRegLiveness: true
body: |
  ; CHECK-LABEL: name: cndmask-not-converted
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.2(0x40000000)
  ; CHECK-NEXT:   liveins: $vgpr0, $sgpr8_sgpr9
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sgpr_64 = COPY $sgpr8_sgpr9
  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr0
  ; CHECK-NEXT:   [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[COPY]], 0, 0
  ; CHECK-NEXT:   S_BITCMP1_B32 [[S_LOAD_DWORDX2_IMM]].sub1, 0, implicit-def $scc
  ; CHECK-NEXT:   [[S_CSELECT_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_CSELECT_B32 -1, 0, implicit $scc
  ; CHECK-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
  ; CHECK-NEXT:   [[S_AND_B32_:%[0-9]+]]:sreg_32 = S_AND_B32 $exec_lo, [[S_CSELECT_B32_]], implicit-def dead $scc
  ; CHECK-NEXT:   $vcc_lo = COPY [[S_AND_B32_]]
  ; CHECK-NEXT:   S_CBRANCH_VCCNZ %bb.2, implicit $vcc_lo
  ; CHECK-NEXT:   S_BRANCH %bb.1
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_64 = COPY [[S_LOAD_DWORDX2_IMM]]
  ; CHECK-NEXT:   [[V_MUL_U32_U24_e64_:%[0-9]+]]:vgpr_32 = V_MUL_U32_U24_e64 [[COPY1]](s32), 5, 0, implicit $exec
  ; CHECK-NEXT:   [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MUL_U32_U24_e64_]], %subreg.sub0, killed [[V_MOV_B32_e32_1]], %subreg.sub1
  ; CHECK-NEXT:   [[GLOBAL_LOAD_USHORT:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_USHORT [[REG_SEQUENCE]], 3, 0, implicit $exec
  ; CHECK-NEXT:   [[V_AND_B32_e64_:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[GLOBAL_LOAD_USHORT]], 255, implicit $exec
  ; CHECK-NEXT:   [[V_MOV_B32_e32_2:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 65535, implicit $exec
  ; CHECK-NEXT:   [[V_AND_B32_sdwa:%[0-9]+]]:vgpr_32 = V_AND_B32_sdwa 0, [[V_MOV_B32_e32_2]], 0, [[GLOBAL_LOAD_USHORT]], 0, 6, 0, 6, 0, implicit $exec
  ; CHECK-NEXT:   S_CMP_EQ_U32 [[COPY2]].sub0, 0, implicit-def $scc
  ; CHECK-NEXT:   [[S_CSELECT_B32_1:%[0-9]+]]:sreg_32_xm0_xexec = S_CSELECT_B32 -1, 0, implicit $scc
  ; CHECK-NEXT:   $vcc_lo = COPY [[S_CSELECT_B32_1]]
  ; CHECK-NEXT:   [[V_CNDMASK_B32_e32_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e32 0, killed [[V_AND_B32_sdwa]], implicit $vcc_lo, implicit $exec
  ; CHECK-NEXT:   [[V_MOV_B32_e32_3:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 24, implicit $exec
  ; CHECK-NEXT:   [[V_LSHRREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_sdwa 0, [[V_MOV_B32_e32_3]], 0, [[V_CNDMASK_B32_e32_]], 0, 1, 0, 6, 6, implicit $exec
  ; CHECK-NEXT:   [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 16, [[V_CNDMASK_B32_e32_]], implicit $exec
  ; CHECK-NEXT:   [[V_MOV_B32_e32_4:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 255, implicit $exec
  ; CHECK-NEXT:   [[V_AND_B32_sdwa1:%[0-9]+]]:vgpr_32 = V_AND_B32_sdwa 0, [[V_CNDMASK_B32_e32_]], 0, [[V_MOV_B32_e32_4]], 0, 6, 0, 5, 6, implicit $exec
  ; CHECK-NEXT:   [[V_OR_B32_sdwa:%[0-9]+]]:vgpr_32 = V_OR_B32_sdwa 0, [[V_AND_B32_sdwa1]], 0, [[V_LSHRREV_B32_sdwa]], 0, 5, 0, 6, 6, implicit $exec
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   [[PHI:%[0-9]+]]:vgpr_32 = PHI [[V_MOV_B32_e32_]], %bb.0, [[V_OR_B32_sdwa]], %bb.1
  ; CHECK-NEXT:   [[V_MOV_B:%[0-9]+]]:vreg_64 = V_MOV_B64_PSEUDO 0, implicit $exec
  ; CHECK-NEXT:   GLOBAL_STORE_BYTE killed [[V_MOV_B]], [[PHI]], 0, 0, implicit $exec
  ; CHECK-NEXT:   S_ENDPGM 0
  bb.0:
    successors: %bb.1(0x40000000), %bb.2(0x40000000)
    liveins: $vgpr0, $sgpr8_sgpr9

    %0:sgpr_64 = COPY $sgpr8_sgpr9
    %1:vgpr_32 = COPY $vgpr0
    %2:sreg_64_xexec = S_LOAD_DWORDX2_IMM %0, 0, 0
    S_BITCMP1_B32 %2.sub1, 0, implicit-def $scc
    %3:sreg_32_xm0_xexec = S_CSELECT_B32 -1, 0, implicit $scc
    %4:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
    %5:sreg_32 = S_AND_B32 $exec_lo, %3, implicit-def dead $scc
    $vcc_lo = COPY %5
    S_CBRANCH_VCCNZ %bb.2, implicit $vcc
    S_BRANCH %bb.1

  bb.1:
    successors: %bb.2(0x80000000)

    %6:sreg_64 = COPY %2
    %7:vgpr_32 = V_MUL_U32_U24_e64 %1(s32), 5, 0, implicit $exec
    %8:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
    %9:vreg_64 = REG_SEQUENCE %7, %subreg.sub0, killed %8, %subreg.sub1
    %10:vgpr_32 = GLOBAL_LOAD_USHORT %9, 3, 0, implicit $exec
    %11:vgpr_32 = V_AND_B32_e64 %10, 255, implicit $exec
    %12:vgpr_32 = V_AND_B32_e64 65535, killed %11, implicit $exec
    S_CMP_EQ_U32 %6.sub0, 0, implicit-def $scc
    %13:sreg_32_xm0_xexec = S_CSELECT_B32 -1, 0, implicit $scc
    %14:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, killed %12, %13, implicit $exec
    %15:vgpr_32 = V_LSHRREV_B32_e64 24, %14, implicit $exec
    %16:vgpr_32 = V_LSHLREV_B16_e64 8, %15, implicit $exec
    %17:vgpr_32 = V_LSHRREV_B32_e64 16, %14, implicit $exec
    %18:vgpr_32 = V_AND_B32_e64 %17, 255, implicit $exec
    %19:vgpr_32 = V_OR_B32_e64 killed %18, killed %16, implicit $exec
    %20:vgpr_32 = V_LSHLREV_B32_e64 16, killed %19, implicit $exec

  bb.2:
    %21:vgpr_32 = PHI %4, %bb.0, %20, %bb.1
    %22:vreg_64 = V_MOV_B64_PSEUDO 0, implicit $exec
    GLOBAL_STORE_BYTE killed %22, %21, 0, 0, implicit $exec
    S_ENDPGM 0
...