# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -march=amdgcn -mcpu=tahiti -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX6 %s
# RUN: llc -march=amdgcn -mcpu=hawaii -mattr=-flat-for-global -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX7 %s
# RUN: llc -march=amdgcn -mcpu=hawaii -mattr=+flat-for-global -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX7-FLAT %s
# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX8 %s
# RUN: llc -march=amdgcn -mcpu=gfx900 -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX9 %s
# RUN: llc -march=amdgcn -mcpu=gfx1010 -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX10 %s
# RUN: llc -march=amdgcn -mcpu=gfx1100 -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX9 %s

---
name:            amdgpu_atomic_cmpxchg_s32_global
legalized:       true
regBankSelected: true
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3

    ; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s32_global
    ; GFX6: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX6-NEXT: {{  $}}
    ; GFX6-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX6-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX6-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX6-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX6-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX6-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
    ; GFX6-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX6-NEXT: [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX6-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN]].sub0
    ; GFX6-NEXT: $vgpr0 = COPY [[COPY3]]
    ; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s32_global
    ; GFX7: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX7-NEXT: {{  $}}
    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX7-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX7-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX7-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX7-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX7-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
    ; GFX7-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX7-NEXT: [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX7-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN]].sub0
    ; GFX7-NEXT: $vgpr0 = COPY [[COPY3]]
    ; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s32_global
    ; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX7-FLAT-NEXT: {{  $}}
    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX7-FLAT-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX7-FLAT-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX7-FLAT-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32), addrspace 1)
    ; GFX7-FLAT-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
    ; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s32_global
    ; GFX8: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX8-NEXT: {{  $}}
    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX8-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX8-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32), addrspace 1)
    ; GFX8-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
    ; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s32_global
    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX9-NEXT: {{  $}}
    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX9-NEXT: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 0, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX9-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
    ; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s32_global
    ; GFX10: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX10-NEXT: {{  $}}
    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX10-NEXT: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 0, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX10-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
    %0:vgpr(p1) = COPY $vgpr0_vgpr1
    %1:vgpr(s32) = COPY $vgpr2
    %2:vgpr(s32) = COPY $vgpr3
    %3:vgpr(<2 x s32>) = G_BUILD_VECTOR %1, %2
    %4:vgpr(s32) = G_AMDGPU_ATOMIC_CMPXCHG %0, %3 :: (load store seq_cst (s32), addrspace 1)
    $vgpr0 = COPY %4

...

---
name:            amdgpu_atomic_cmpxchg_s32_global_gep4
legalized:       true
regBankSelected: true
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3

    ; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gep4
    ; GFX6: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX6-NEXT: {{  $}}
    ; GFX6-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX6-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX6-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX6-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX6-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX6-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
    ; GFX6-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX6-NEXT: [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 4, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX6-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN]].sub0
    ; GFX6-NEXT: $vgpr0 = COPY [[COPY3]]
    ; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gep4
    ; GFX7: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX7-NEXT: {{  $}}
    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX7-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX7-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX7-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX7-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX7-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
    ; GFX7-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX7-NEXT: [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 4, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX7-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN]].sub0
    ; GFX7-NEXT: $vgpr0 = COPY [[COPY3]]
    ; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gep4
    ; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX7-FLAT-NEXT: {{  $}}
    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX7-FLAT-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX7-FLAT-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX7-FLAT-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4, implicit $exec
    ; GFX7-FLAT-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
    ; GFX7-FLAT-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
    ; GFX7-FLAT-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
    ; GFX7-FLAT-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
    ; GFX7-FLAT-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
    ; GFX7-FLAT-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
    ; GFX7-FLAT-NEXT: %12:vgpr_32, dead %14:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %12, %subreg.sub1
    ; GFX7-FLAT-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32), addrspace 1)
    ; GFX7-FLAT-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
    ; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gep4
    ; GFX8: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX8-NEXT: {{  $}}
    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX8-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX8-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4, implicit $exec
    ; GFX8-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
    ; GFX8-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
    ; GFX8-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
    ; GFX8-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
    ; GFX8-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
    ; GFX8-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
    ; GFX8-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
    ; GFX8-NEXT: %12:vgpr_32, dead %14:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
    ; GFX8-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %12, %subreg.sub1
    ; GFX8-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32), addrspace 1)
    ; GFX8-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
    ; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gep4
    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX9-NEXT: {{  $}}
    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX9-NEXT: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 4, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX9-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
    ; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gep4
    ; GFX10: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX10-NEXT: {{  $}}
    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX10-NEXT: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 4, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX10-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
    %0:vgpr(p1) = COPY $vgpr0_vgpr1
    %1:vgpr(s32) = COPY $vgpr2
    %2:vgpr(s32) = COPY $vgpr3
    %3:vgpr(<2 x s32>) = G_BUILD_VECTOR %1, %2
    %4:vgpr(s64) = G_CONSTANT i64 4
    %5:vgpr(p1) = G_PTR_ADD %0, %4
    %6:vgpr(s32) = G_AMDGPU_ATOMIC_CMPXCHG %5, %3 :: (load store seq_cst (s32), addrspace 1)
    $vgpr0 = COPY %6

...

---
name:            amdgpu_atomic_cmpxchg_s64_global
legalized:       true
regBankSelected: true
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5

    ; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s64_global
    ; GFX6: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX6-NEXT: {{  $}}
    ; GFX6-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX6-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX6-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX6-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX6-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX6-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
    ; GFX6-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX6-NEXT: [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN:%[0-9]+]]:vreg_128 = BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 1, implicit $exec :: (load store seq_cst (s64), addrspace 1)
    ; GFX6-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN]].sub0_sub1
    ; GFX6-NEXT: $vgpr0_vgpr1 = COPY [[COPY3]]
    ; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s64_global
    ; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX7-NEXT: {{  $}}
    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX7-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX7-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX7-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX7-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX7-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
    ; GFX7-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX7-NEXT: [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN:%[0-9]+]]:vreg_128 = BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 1, implicit $exec :: (load store seq_cst (s64), addrspace 1)
    ; GFX7-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN]].sub0_sub1
    ; GFX7-NEXT: $vgpr0_vgpr1 = COPY [[COPY3]]
    ; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s64_global
    ; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX7-FLAT-NEXT: {{  $}}
    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX7-FLAT-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX7-FLAT-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX7-FLAT-NEXT: [[FLAT_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s64), addrspace 1)
    ; GFX7-FLAT-NEXT: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_CMPSWAP_X2_RTN]]
    ; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s64_global
    ; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX8-NEXT: {{  $}}
    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX8-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX8-NEXT: [[FLAT_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s64), addrspace 1)
    ; GFX8-NEXT: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_CMPSWAP_X2_RTN]]
    ; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s64_global
    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX9-NEXT: {{  $}}
    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX9-NEXT: [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = GLOBAL_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 0, 1, implicit $exec :: (load store seq_cst (s64), addrspace 1)
    ; GFX9-NEXT: $vgpr0_vgpr1 = COPY [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN]]
    ; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s64_global
    ; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX10-NEXT: {{  $}}
    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX10-NEXT: [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = GLOBAL_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 0, 1, implicit $exec :: (load store seq_cst (s64), addrspace 1)
    ; GFX10-NEXT: $vgpr0_vgpr1 = COPY [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN]]
    %0:vgpr(p1) = COPY $vgpr0_vgpr1
    %1:vgpr(s64) = COPY $vgpr2_vgpr3
    %2:vgpr(s64) = COPY $vgpr4_vgpr5
    %3:vgpr(<2 x s64>) = G_BUILD_VECTOR %1, %2
    %4:vgpr(s64) = G_AMDGPU_ATOMIC_CMPXCHG %0, %3 :: (load store seq_cst (s64), addrspace 1)
    $vgpr0_vgpr1 = COPY %4

...

---
name:            amdgpu_atomic_cmpxchg_s64_global_gep4
legalized:       true
regBankSelected: true
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5

    ; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_gep4
    ; GFX6: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX6-NEXT: {{  $}}
    ; GFX6-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX6-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX6-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX6-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX6-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX6-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
    ; GFX6-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX6-NEXT: [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN:%[0-9]+]]:vreg_128 = BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 4, 1, implicit $exec :: (load store seq_cst (s64), addrspace 1)
    ; GFX6-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN]].sub0_sub1
    ; GFX6-NEXT: $vgpr0_vgpr1 = COPY [[COPY3]]
    ; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_gep4
    ; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX7-NEXT: {{  $}}
    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX7-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX7-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX7-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX7-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX7-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
    ; GFX7-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX7-NEXT: [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN:%[0-9]+]]:vreg_128 = BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 4, 1, implicit $exec :: (load store seq_cst (s64), addrspace 1)
    ; GFX7-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN]].sub0_sub1
    ; GFX7-NEXT: $vgpr0_vgpr1 = COPY [[COPY3]]
    ; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_gep4
    ; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX7-FLAT-NEXT: {{  $}}
    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX7-FLAT-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX7-FLAT-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX7-FLAT-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4, implicit $exec
    ; GFX7-FLAT-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
    ; GFX7-FLAT-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
    ; GFX7-FLAT-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
    ; GFX7-FLAT-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
    ; GFX7-FLAT-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
    ; GFX7-FLAT-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
    ; GFX7-FLAT-NEXT: %12:vgpr_32, dead %14:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %12, %subreg.sub1
    ; GFX7-FLAT-NEXT: [[FLAT_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_CMPSWAP_X2_RTN [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s64), addrspace 1)
    ; GFX7-FLAT-NEXT: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_CMPSWAP_X2_RTN]]
    ; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_gep4
    ; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX8-NEXT: {{  $}}
    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX8-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX8-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4, implicit $exec
    ; GFX8-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
    ; GFX8-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
    ; GFX8-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
    ; GFX8-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
    ; GFX8-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
    ; GFX8-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
    ; GFX8-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
    ; GFX8-NEXT: %12:vgpr_32, dead %14:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
    ; GFX8-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %12, %subreg.sub1
    ; GFX8-NEXT: [[FLAT_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_CMPSWAP_X2_RTN [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s64), addrspace 1)
    ; GFX8-NEXT: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_CMPSWAP_X2_RTN]]
    ; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_gep4
    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX9-NEXT: {{  $}}
    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX9-NEXT: [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = GLOBAL_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 4, 1, implicit $exec :: (load store seq_cst (s64), addrspace 1)
    ; GFX9-NEXT: $vgpr0_vgpr1 = COPY [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN]]
    ; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_gep4
    ; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX10-NEXT: {{  $}}
    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX10-NEXT: [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = GLOBAL_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 4, 1, implicit $exec :: (load store seq_cst (s64), addrspace 1)
    ; GFX10-NEXT: $vgpr0_vgpr1 = COPY [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN]]
    %0:vgpr(p1) = COPY $vgpr0_vgpr1
    %1:vgpr(s64) = COPY $vgpr2_vgpr3
    %2:vgpr(s64) = COPY $vgpr4_vgpr5
    %3:vgpr(<2 x s64>) = G_BUILD_VECTOR %1, %2
    %4:vgpr(s64) = G_CONSTANT i64 4
    %5:vgpr(p1) = G_PTR_ADD %0, %4
    %6:vgpr(s64) = G_AMDGPU_ATOMIC_CMPXCHG %5, %3 :: (load store seq_cst (s64), addrspace 1)
    $vgpr0_vgpr1 = COPY %6

...

---
name:            amdgpu_atomic_cmpxchg_s32_global_gepm4
legalized:       true
regBankSelected: true
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3

    ; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gepm4
    ; GFX6: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX6-NEXT: {{  $}}
    ; GFX6-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX6-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4294967292, implicit $exec
    ; GFX6-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -1, implicit $exec
    ; GFX6-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
    ; GFX6-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
    ; GFX6-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
    ; GFX6-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
    ; GFX6-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
    ; GFX6-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
    ; GFX6-NEXT: %19:vgpr_32, dead %21:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
    ; GFX6-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %19, %subreg.sub1
    ; GFX6-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX6-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX6-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX6-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
    ; GFX6-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE3]], %subreg.sub2_sub3
    ; GFX6-NEXT: [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE2]], [[REG_SEQUENCE4]], 0, 0, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX6-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN]].sub0
    ; GFX6-NEXT: $vgpr0 = COPY [[COPY7]]
    ; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gepm4
    ; GFX7: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX7-NEXT: {{  $}}
    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX7-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX7-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX7-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4294967292, implicit $exec
    ; GFX7-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -1, implicit $exec
    ; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
    ; GFX7-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
    ; GFX7-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
    ; GFX7-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
    ; GFX7-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
    ; GFX7-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
    ; GFX7-NEXT: %19:vgpr_32, dead %21:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
    ; GFX7-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %19, %subreg.sub1
    ; GFX7-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX7-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX7-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX7-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
    ; GFX7-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE3]], %subreg.sub2_sub3
    ; GFX7-NEXT: [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE2]], [[REG_SEQUENCE4]], 0, 0, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX7-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN]].sub0
    ; GFX7-NEXT: $vgpr0 = COPY [[COPY7]]
    ; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gepm4
    ; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX7-FLAT-NEXT: {{  $}}
    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX7-FLAT-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX7-FLAT-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX7-FLAT-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4294967292, implicit $exec
    ; GFX7-FLAT-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -1, implicit $exec
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
    ; GFX7-FLAT-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
    ; GFX7-FLAT-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
    ; GFX7-FLAT-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
    ; GFX7-FLAT-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
    ; GFX7-FLAT-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
    ; GFX7-FLAT-NEXT: %12:vgpr_32, dead %14:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %12, %subreg.sub1
    ; GFX7-FLAT-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32), addrspace 1)
    ; GFX7-FLAT-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
    ; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gepm4
    ; GFX8: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX8-NEXT: {{  $}}
    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX8-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX8-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4294967292, implicit $exec
    ; GFX8-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -1, implicit $exec
    ; GFX8-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
    ; GFX8-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
    ; GFX8-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
    ; GFX8-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
    ; GFX8-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
    ; GFX8-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
    ; GFX8-NEXT: %12:vgpr_32, dead %14:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
    ; GFX8-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %12, %subreg.sub1
    ; GFX8-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32), addrspace 1)
    ; GFX8-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
    ; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gepm4
    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX9-NEXT: {{  $}}
    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX9-NEXT: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], -4, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX9-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
    ; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gepm4
    ; GFX10: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX10-NEXT: {{  $}}
    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX10-NEXT: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], -4, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX10-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
    %0:vgpr(p1) = COPY $vgpr0_vgpr1
    %1:vgpr(s32) = COPY $vgpr2
    %2:vgpr(s32) = COPY $vgpr3
    %3:vgpr(<2 x s32>) = G_BUILD_VECTOR %1, %2
    %4:vgpr(s64) = G_CONSTANT i64 -4
    %5:vgpr(p1) = G_PTR_ADD %0, %4
    %6:vgpr(s32) = G_AMDGPU_ATOMIC_CMPXCHG %5, %3 :: (load store seq_cst (s32), addrspace 1)
    $vgpr0 = COPY %6

...

---
name:            amdgpu_atomic_cmpxchg_s32_global_nortn
legalized:       true
regBankSelected: true
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3

    ; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_nortn
    ; GFX6: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX6-NEXT: {{  $}}
    ; GFX6-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX6-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX6-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX6-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX6-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX6-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
    ; GFX6-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX6-NEXT: BUFFER_ATOMIC_CMPSWAP_ADDR64 [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 0, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_nortn
    ; GFX7: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX7-NEXT: {{  $}}
    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX7-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX7-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX7-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX7-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX7-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
    ; GFX7-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX7-NEXT: BUFFER_ATOMIC_CMPSWAP_ADDR64 [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 0, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_nortn
    ; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX7-FLAT-NEXT: {{  $}}
    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX7-FLAT-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX7-FLAT-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX7-FLAT-NEXT: FLAT_ATOMIC_CMPSWAP [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32), addrspace 1)
    ; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_nortn
    ; GFX8: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX8-NEXT: {{  $}}
    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX8-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX8-NEXT: FLAT_ATOMIC_CMPSWAP [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32), addrspace 1)
    ; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_nortn
    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX9-NEXT: {{  $}}
    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX9-NEXT: GLOBAL_ATOMIC_CMPSWAP [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_nortn
    ; GFX10: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
    ; GFX10-NEXT: {{  $}}
    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX10-NEXT: GLOBAL_ATOMIC_CMPSWAP [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    %0:vgpr(p1) = COPY $vgpr0_vgpr1
    %1:vgpr(s32) = COPY $vgpr2
    %2:vgpr(s32) = COPY $vgpr3
    %3:vgpr(<2 x s32>) = G_BUILD_VECTOR %1, %2
    %4:vgpr(s32) = G_AMDGPU_ATOMIC_CMPXCHG %0, %3 :: (load store seq_cst (s32), addrspace 1)

...

---
name:            amdgpu_atomic_cmpxchg_s64_global_nortn
legalized:       true
regBankSelected: true
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5

    ; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_nortn
    ; GFX6: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX6-NEXT: {{  $}}
    ; GFX6-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX6-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX6-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX6-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX6-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX6-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
    ; GFX6-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX6-NEXT: BUFFER_ATOMIC_CMPSWAP_X2_ADDR64 [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 0, implicit $exec :: (load store seq_cst (s64), addrspace 1)
    ; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_nortn
    ; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX7-NEXT: {{  $}}
    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX7-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX7-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX7-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX7-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX7-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
    ; GFX7-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX7-NEXT: BUFFER_ATOMIC_CMPSWAP_X2_ADDR64 [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 0, implicit $exec :: (load store seq_cst (s64), addrspace 1)
    ; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_nortn
    ; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX7-FLAT-NEXT: {{  $}}
    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX7-FLAT-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX7-FLAT-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX7-FLAT-NEXT: FLAT_ATOMIC_CMPSWAP_X2 [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst (s64), addrspace 1)
    ; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_nortn
    ; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX8-NEXT: {{  $}}
    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX8-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX8-NEXT: FLAT_ATOMIC_CMPSWAP_X2 [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst (s64), addrspace 1)
    ; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_nortn
    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX9-NEXT: {{  $}}
    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX9-NEXT: GLOBAL_ATOMIC_CMPSWAP_X2 [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst (s64), addrspace 1)
    ; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_nortn
    ; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; GFX10-NEXT: {{  $}}
    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
    ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
    ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
    ; GFX10-NEXT: GLOBAL_ATOMIC_CMPSWAP_X2 [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst (s64), addrspace 1)
    %0:vgpr(p1) = COPY $vgpr0_vgpr1
    %1:vgpr(s64) = COPY $vgpr2_vgpr3
    %2:vgpr(s64) = COPY $vgpr4_vgpr5
    %3:vgpr(<2 x s64>) = G_BUILD_VECTOR %1, %2
    %4:vgpr(s64) = G_AMDGPU_ATOMIC_CMPXCHG %0, %3 :: (load store seq_cst (s64), addrspace 1)

...

---
name:            amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr
legalized:       true
regBankSelected: true
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3

    ; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr
    ; GFX6: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
    ; GFX6-NEXT: {{  $}}
    ; GFX6-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX6-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX6-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4294967295
    ; GFX6-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX6-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX6-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX6-NEXT: [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE2]], 0, 0, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX6-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN]].sub0
    ; GFX6-NEXT: $vgpr0 = COPY [[COPY3]]
    ; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr
    ; GFX7: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
    ; GFX7-NEXT: {{  $}}
    ; GFX7-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; GFX7-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX7-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX7-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4294967295
    ; GFX7-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX7-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX7-NEXT: [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE2]], 0, 0, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX7-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN]].sub0
    ; GFX7-NEXT: $vgpr0 = COPY [[COPY3]]
    ; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr
    ; GFX7-FLAT: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
    ; GFX7-FLAT-NEXT: {{  $}}
    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; GFX7-FLAT-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX7-FLAT-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX7-FLAT-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[COPY]]
    ; GFX7-FLAT-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY3]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32), addrspace 1)
    ; GFX7-FLAT-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
    ; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr
    ; GFX8: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
    ; GFX8-NEXT: {{  $}}
    ; GFX8-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX8-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX8-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[COPY]]
    ; GFX8-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY3]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32), addrspace 1)
    ; GFX8-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
    ; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr
    ; GFX9: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
    ; GFX9-NEXT: {{  $}}
    ; GFX9-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
    ; GFX9-NEXT: [[GLOBAL_ATOMIC_CMPSWAP_SADDR_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_SADDR_RTN [[V_MOV_B32_e32_]], [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX9-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_SADDR_RTN]]
    ; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr
    ; GFX10: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
    ; GFX10-NEXT: {{  $}}
    ; GFX10-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX10-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
    ; GFX10-NEXT: [[GLOBAL_ATOMIC_CMPSWAP_SADDR_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_SADDR_RTN [[V_MOV_B32_e32_]], [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX10-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_SADDR_RTN]]
    %0:sgpr(p1) = COPY $sgpr0_sgpr1
    %1:vgpr(s32) = COPY $vgpr2
    %2:vgpr(s32) = COPY $vgpr3
    %3:vgpr(<2 x s32>) = G_BUILD_VECTOR %1, %2
    %4:vgpr(s32) = G_AMDGPU_ATOMIC_CMPXCHG %0, %3 :: (load store seq_cst (s32), addrspace 1)
    $vgpr0 = COPY %4

...

---
name:            amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr_offset_4095
legalized:       true
regBankSelected: true
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3

    ; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr_offset_4095
    ; GFX6: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
    ; GFX6-NEXT: {{  $}}
    ; GFX6-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX6-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX6-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4294967295
    ; GFX6-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX6-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX6-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX6-NEXT: [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE2]], 0, 4095, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX6-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN]].sub0
    ; GFX6-NEXT: $vgpr0 = COPY [[COPY3]]
    ; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr_offset_4095
    ; GFX7: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
    ; GFX7-NEXT: {{  $}}
    ; GFX7-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; GFX7-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX7-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX7-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4294967295
    ; GFX7-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
    ; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX7-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
    ; GFX7-NEXT: [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE2]], 0, 4095, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX7-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN]].sub0
    ; GFX7-NEXT: $vgpr0 = COPY [[COPY3]]
    ; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr_offset_4095
    ; GFX7-FLAT: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
    ; GFX7-FLAT-NEXT: {{  $}}
    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; GFX7-FLAT-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX7-FLAT-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX7-FLAT-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
    ; GFX7-FLAT-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX7-FLAT-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
    ; GFX7-FLAT-NEXT: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
    ; GFX7-FLAT-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
    ; GFX7-FLAT-NEXT: [[COPY6:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
    ; GFX7-FLAT-NEXT: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY3]], [[COPY4]], implicit-def $scc
    ; GFX7-FLAT-NEXT: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY5]], [[COPY6]], implicit-def $scc, implicit $scc
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
    ; GFX7-FLAT-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX7-FLAT-NEXT: [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
    ; GFX7-FLAT-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY7]], [[REG_SEQUENCE2]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32), addrspace 1)
    ; GFX7-FLAT-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
    ; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr_offset_4095
    ; GFX8: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
    ; GFX8-NEXT: {{  $}}
    ; GFX8-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX8-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
    ; GFX8-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 0
    ; GFX8-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
    ; GFX8-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
    ; GFX8-NEXT: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
    ; GFX8-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
    ; GFX8-NEXT: [[COPY6:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
    ; GFX8-NEXT: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY3]], [[COPY4]], implicit-def $scc
    ; GFX8-NEXT: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY5]], [[COPY6]], implicit-def $scc, implicit $scc
    ; GFX8-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
    ; GFX8-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX8-NEXT: [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
    ; GFX8-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY7]], [[REG_SEQUENCE2]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32), addrspace 1)
    ; GFX8-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
    ; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr_offset_4095
    ; GFX9: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
    ; GFX9-NEXT: {{  $}}
    ; GFX9-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
    ; GFX9-NEXT: [[GLOBAL_ATOMIC_CMPSWAP_SADDR_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_SADDR_RTN [[V_MOV_B32_e32_]], [[REG_SEQUENCE]], [[COPY]], 4095, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX9-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_SADDR_RTN]]
    ; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr_offset_4095
    ; GFX10: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
    ; GFX10-NEXT: {{  $}}
    ; GFX10-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
    ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
    ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
    ; GFX10-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2048, implicit $exec
    ; GFX10-NEXT: [[GLOBAL_ATOMIC_CMPSWAP_SADDR_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_SADDR_RTN [[V_MOV_B32_e32_]], [[REG_SEQUENCE]], [[COPY]], 2047, 1, implicit $exec :: (load store seq_cst (s32), addrspace 1)
    ; GFX10-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_SADDR_RTN]]
    %0:sgpr(p1) = COPY $sgpr0_sgpr1
    %1:vgpr(s32) = COPY $vgpr2
    %2:vgpr(s32) = COPY $vgpr3
    %3:sgpr(s64) = G_CONSTANT i64 4095
    %4:sgpr(p1) = G_PTR_ADD %0, %3
    %5:vgpr(<2 x s32>) = G_BUILD_VECTOR %1, %2
    %6:vgpr(s32) = G_AMDGPU_ATOMIC_CMPXCHG %4, %5 :: (load store seq_cst (s32), addrspace 1)
    $vgpr0 = COPY %6

...