test/CodeGen/AMDGPU/flat_atomics_min_max_system.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN1 %s
; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN2 %s
; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN3 %s

define amdgpu_kernel void @atomic_max_i32_offset(ptr %out, i32 %in) {
; GCN1-LABEL: atomic_max_i32_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_add_u32 s0, s2, 16
; GCN1-NEXT:    s_addc_u32 s1, s3, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB0_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_max_i32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB0_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_max_i32_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_add_u32 s0, s2, 16
; GCN2-NEXT:    s_addc_u32 s1, s3, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB0_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_max_i32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB0_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_max_i32_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s2
; GCN3-NEXT:    v_mov_b32_e32 v1, s3
; GCN3-NEXT:    flat_load_dword v1, v[0:1] offset:16
; GCN3-NEXT:  .LBB0_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v2, s2
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_max_i32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v3, s3
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB0_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %gep = getelementptr i32, ptr %out, i32 4
  %val = atomicrmw volatile max ptr %gep, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_max_i32_ret_offset(ptr %out, ptr %out2, i32 %in) {
; GCN1-LABEL: atomic_max_i32_ret_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s8, s[0:1], 0xd
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_add_u32 s0, s4, 16
; GCN1-NEXT:    s_addc_u32 s1, s5, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:  .LBB1_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    v_max_i32_e32 v0, s8, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB1_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_max_i32_ret_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_add_u32 s0, s4, 16
; GCN2-NEXT:    s_addc_u32 s1, s5, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:  .LBB1_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    v_max_i32_e32 v0, s8, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB1_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_max_i32_ret_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s2, s[0:1], 0x34
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s4
; GCN3-NEXT:    v_mov_b32_e32 v1, s5
; GCN3-NEXT:    flat_load_dword v0, v[0:1] offset:16
; GCN3-NEXT:  .LBB1_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v2, s4
; GCN3-NEXT:    v_mov_b32_e32 v3, s5
; GCN3-NEXT:    v_max_i32_e32 v0, s2, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB1_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %gep = getelementptr i32, ptr %out, i32 4
  %val = atomicrmw volatile max ptr %gep, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_max_i32_addr64_offset(ptr %out, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_max_i32_addr64_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xd
; GCN1-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s6, s0
; GCN1-NEXT:    s_addc_u32 s1, s7, s1
; GCN1-NEXT:    s_add_u32 s0, s0, 16
; GCN1-NEXT:    s_addc_u32 s1, s1, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB2_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_max_i32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB2_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_max_i32_addr64_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN2-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s6, s0
; GCN2-NEXT:    s_addc_u32 s1, s7, s1
; GCN2-NEXT:    s_add_u32 s0, s0, 16
; GCN2-NEXT:    s_addc_u32 s1, s1, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB2_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_max_i32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB2_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_max_i32_addr64_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN3-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s6, s0
; GCN3-NEXT:    s_addc_u32 s1, s7, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v1, v[0:1] offset:16
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB2_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_max_i32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB2_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %gep = getelementptr i32, ptr %ptr, i32 4
  %val = atomicrmw volatile max ptr %gep, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_max_i32_ret_addr64_offset(ptr %out, ptr %out2, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_max_i32_ret_addr64_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xf
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s8, s[0:1], 0xd
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s4, s0
; GCN1-NEXT:    s_addc_u32 s1, s5, s1
; GCN1-NEXT:    s_add_u32 s0, s0, 16
; GCN1-NEXT:    s_addc_u32 s1, s1, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB3_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    v_max_i32_e32 v0, s8, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB3_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_max_i32_ret_addr64_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s4, s0
; GCN2-NEXT:    s_addc_u32 s1, s5, s1
; GCN2-NEXT:    s_add_u32 s0, s0, 16
; GCN2-NEXT:    s_addc_u32 s1, s1, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB3_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    v_max_i32_e32 v0, s8, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB3_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_max_i32_ret_addr64_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s4, s0
; GCN3-NEXT:    s_addc_u32 s1, s5, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v0, v[0:1] offset:16
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB3_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    v_max_i32_e32 v0, s8, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB3_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %gep = getelementptr i32, ptr %ptr, i32 4
  %val = atomicrmw volatile max ptr %gep, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_max_i32(ptr %out, i32 %in) {
; GCN1-LABEL: atomic_max_i32:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_mov_b64 s[0:1], 0
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v0, s2
; GCN1-NEXT:    v_mov_b32_e32 v1, s3
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:  .LBB4_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v2, s2
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_max_i32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v3, s3
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN1-NEXT:    s_cbranch_execnz .LBB4_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_max_i32:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_mov_b64 s[0:1], 0
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v0, s2
; GCN2-NEXT:    v_mov_b32_e32 v1, s3
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:  .LBB4_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v2, s2
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_max_i32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v3, s3
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN2-NEXT:    s_cbranch_execnz .LBB4_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_max_i32:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s2
; GCN3-NEXT:    v_mov_b32_e32 v1, s3
; GCN3-NEXT:    flat_load_dword v1, v[0:1]
; GCN3-NEXT:  .LBB4_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v2, s2
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_max_i32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v3, s3
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB4_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %val = atomicrmw volatile max ptr %out, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_max_i32_ret(ptr %out, ptr %out2, i32 %in) {
; GCN1-LABEL: atomic_max_i32_ret:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s2, s[0:1], 0xd
; GCN1-NEXT:    s_mov_b64 s[0:1], 0
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v0, s4
; GCN1-NEXT:    v_mov_b32_e32 v1, s5
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:  .LBB5_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v2, s4
; GCN1-NEXT:    v_mov_b32_e32 v3, s5
; GCN1-NEXT:    v_max_i32_e32 v0, s2, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN1-NEXT:    s_cbranch_execnz .LBB5_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_max_i32_ret:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s2, s[0:1], 0x34
; GCN2-NEXT:    s_mov_b64 s[0:1], 0
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v0, s4
; GCN2-NEXT:    v_mov_b32_e32 v1, s5
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:  .LBB5_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v2, s4
; GCN2-NEXT:    v_mov_b32_e32 v3, s5
; GCN2-NEXT:    v_max_i32_e32 v0, s2, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN2-NEXT:    s_cbranch_execnz .LBB5_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_max_i32_ret:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s2, s[0:1], 0x34
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s4
; GCN3-NEXT:    v_mov_b32_e32 v1, s5
; GCN3-NEXT:    flat_load_dword v0, v[0:1]
; GCN3-NEXT:  .LBB5_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v2, s4
; GCN3-NEXT:    v_mov_b32_e32 v3, s5
; GCN3-NEXT:    v_max_i32_e32 v0, s2, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB5_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %val = atomicrmw volatile max ptr %out, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_max_i32_addr64(ptr %out, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_max_i32_addr64:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xd
; GCN1-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s6, s0
; GCN1-NEXT:    s_addc_u32 s1, s7, s1
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB6_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_max_i32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB6_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_max_i32_addr64:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN2-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s6, s0
; GCN2-NEXT:    s_addc_u32 s1, s7, s1
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB6_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_max_i32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB6_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_max_i32_addr64:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN3-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s6, s0
; GCN3-NEXT:    s_addc_u32 s1, s7, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v1, v[0:1]
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB6_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_max_i32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB6_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %val = atomicrmw volatile max ptr %ptr, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_max_i32_ret_addr64(ptr %out, ptr %out2, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_max_i32_ret_addr64:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xf
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s8, s[0:1], 0xd
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s4, s0
; GCN1-NEXT:    s_addc_u32 s1, s5, s1
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB7_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    v_max_i32_e32 v0, s8, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB7_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_max_i32_ret_addr64:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s4, s0
; GCN2-NEXT:    s_addc_u32 s1, s5, s1
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB7_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    v_max_i32_e32 v0, s8, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB7_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_max_i32_ret_addr64:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s4, s0
; GCN3-NEXT:    s_addc_u32 s1, s5, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v0, v[0:1]
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB7_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    v_max_i32_e32 v0, s8, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB7_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %val = atomicrmw volatile max ptr %ptr, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_umax_i32_offset(ptr %out, i32 %in) {
; GCN1-LABEL: atomic_umax_i32_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_add_u32 s0, s2, 16
; GCN1-NEXT:    s_addc_u32 s1, s3, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB8_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_max_u32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB8_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umax_i32_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_add_u32 s0, s2, 16
; GCN2-NEXT:    s_addc_u32 s1, s3, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB8_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_max_u32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB8_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umax_i32_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s2
; GCN3-NEXT:    v_mov_b32_e32 v1, s3
; GCN3-NEXT:    flat_load_dword v1, v[0:1] offset:16
; GCN3-NEXT:  .LBB8_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v2, s2
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_max_u32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v3, s3
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB8_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %gep = getelementptr i32, ptr %out, i32 4
  %val = atomicrmw volatile umax ptr %gep, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_umax_i32_ret_offset(ptr %out, ptr %out2, i32 %in) {
; GCN1-LABEL: atomic_umax_i32_ret_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s8, s[0:1], 0xd
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_add_u32 s0, s4, 16
; GCN1-NEXT:    s_addc_u32 s1, s5, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:  .LBB9_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    v_max_u32_e32 v0, s8, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB9_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umax_i32_ret_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_add_u32 s0, s4, 16
; GCN2-NEXT:    s_addc_u32 s1, s5, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:  .LBB9_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    v_max_u32_e32 v0, s8, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB9_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umax_i32_ret_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s2, s[0:1], 0x34
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s4
; GCN3-NEXT:    v_mov_b32_e32 v1, s5
; GCN3-NEXT:    flat_load_dword v0, v[0:1] offset:16
; GCN3-NEXT:  .LBB9_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v2, s4
; GCN3-NEXT:    v_mov_b32_e32 v3, s5
; GCN3-NEXT:    v_max_u32_e32 v0, s2, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB9_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %gep = getelementptr i32, ptr %out, i32 4
  %val = atomicrmw volatile umax ptr %gep, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_umax_i32_addr64_offset(ptr %out, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_umax_i32_addr64_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xd
; GCN1-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s6, s0
; GCN1-NEXT:    s_addc_u32 s1, s7, s1
; GCN1-NEXT:    s_add_u32 s0, s0, 16
; GCN1-NEXT:    s_addc_u32 s1, s1, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB10_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_max_u32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB10_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umax_i32_addr64_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN2-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s6, s0
; GCN2-NEXT:    s_addc_u32 s1, s7, s1
; GCN2-NEXT:    s_add_u32 s0, s0, 16
; GCN2-NEXT:    s_addc_u32 s1, s1, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB10_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_max_u32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB10_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umax_i32_addr64_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN3-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s6, s0
; GCN3-NEXT:    s_addc_u32 s1, s7, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v1, v[0:1] offset:16
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB10_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_max_u32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB10_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %gep = getelementptr i32, ptr %ptr, i32 4
  %val = atomicrmw volatile umax ptr %gep, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_umax_i32_ret_addr64_offset(ptr %out, ptr %out2, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_umax_i32_ret_addr64_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xf
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s8, s[0:1], 0xd
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s4, s0
; GCN1-NEXT:    s_addc_u32 s1, s5, s1
; GCN1-NEXT:    s_add_u32 s0, s0, 16
; GCN1-NEXT:    s_addc_u32 s1, s1, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB11_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    v_max_u32_e32 v0, s8, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB11_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umax_i32_ret_addr64_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s4, s0
; GCN2-NEXT:    s_addc_u32 s1, s5, s1
; GCN2-NEXT:    s_add_u32 s0, s0, 16
; GCN2-NEXT:    s_addc_u32 s1, s1, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB11_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    v_max_u32_e32 v0, s8, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB11_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umax_i32_ret_addr64_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s4, s0
; GCN3-NEXT:    s_addc_u32 s1, s5, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v0, v[0:1] offset:16
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB11_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    v_max_u32_e32 v0, s8, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB11_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %gep = getelementptr i32, ptr %ptr, i32 4
  %val = atomicrmw volatile umax ptr %gep, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_umax_i32(ptr %out, i32 %in) {
; GCN1-LABEL: atomic_umax_i32:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_mov_b64 s[0:1], 0
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v0, s2
; GCN1-NEXT:    v_mov_b32_e32 v1, s3
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:  .LBB12_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v2, s2
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_max_u32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v3, s3
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN1-NEXT:    s_cbranch_execnz .LBB12_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umax_i32:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_mov_b64 s[0:1], 0
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v0, s2
; GCN2-NEXT:    v_mov_b32_e32 v1, s3
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:  .LBB12_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v2, s2
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_max_u32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v3, s3
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN2-NEXT:    s_cbranch_execnz .LBB12_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umax_i32:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s2
; GCN3-NEXT:    v_mov_b32_e32 v1, s3
; GCN3-NEXT:    flat_load_dword v1, v[0:1]
; GCN3-NEXT:  .LBB12_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v2, s2
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_max_u32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v3, s3
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB12_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %val = atomicrmw volatile umax ptr %out, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_umax_i32_ret(ptr %out, ptr %out2, i32 %in) {
; GCN1-LABEL: atomic_umax_i32_ret:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s2, s[0:1], 0xd
; GCN1-NEXT:    s_mov_b64 s[0:1], 0
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v0, s4
; GCN1-NEXT:    v_mov_b32_e32 v1, s5
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:  .LBB13_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v2, s4
; GCN1-NEXT:    v_mov_b32_e32 v3, s5
; GCN1-NEXT:    v_max_u32_e32 v0, s2, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN1-NEXT:    s_cbranch_execnz .LBB13_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umax_i32_ret:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s2, s[0:1], 0x34
; GCN2-NEXT:    s_mov_b64 s[0:1], 0
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v0, s4
; GCN2-NEXT:    v_mov_b32_e32 v1, s5
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:  .LBB13_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v2, s4
; GCN2-NEXT:    v_mov_b32_e32 v3, s5
; GCN2-NEXT:    v_max_u32_e32 v0, s2, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN2-NEXT:    s_cbranch_execnz .LBB13_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umax_i32_ret:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s2, s[0:1], 0x34
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s4
; GCN3-NEXT:    v_mov_b32_e32 v1, s5
; GCN3-NEXT:    flat_load_dword v0, v[0:1]
; GCN3-NEXT:  .LBB13_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v2, s4
; GCN3-NEXT:    v_mov_b32_e32 v3, s5
; GCN3-NEXT:    v_max_u32_e32 v0, s2, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB13_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %val = atomicrmw volatile umax ptr %out, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_umax_i32_addr64(ptr %out, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_umax_i32_addr64:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xd
; GCN1-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s6, s0
; GCN1-NEXT:    s_addc_u32 s1, s7, s1
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB14_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_max_u32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB14_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umax_i32_addr64:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN2-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s6, s0
; GCN2-NEXT:    s_addc_u32 s1, s7, s1
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB14_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_max_u32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB14_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umax_i32_addr64:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN3-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s6, s0
; GCN3-NEXT:    s_addc_u32 s1, s7, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v1, v[0:1]
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB14_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_max_u32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB14_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %val = atomicrmw volatile umax ptr %ptr, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_umax_i32_ret_addr64(ptr %out, ptr %out2, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_umax_i32_ret_addr64:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xf
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s8, s[0:1], 0xd
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s4, s0
; GCN1-NEXT:    s_addc_u32 s1, s5, s1
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB15_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    v_max_u32_e32 v0, s8, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB15_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umax_i32_ret_addr64:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s4, s0
; GCN2-NEXT:    s_addc_u32 s1, s5, s1
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB15_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    v_max_u32_e32 v0, s8, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB15_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umax_i32_ret_addr64:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s4, s0
; GCN3-NEXT:    s_addc_u32 s1, s5, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v0, v[0:1]
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB15_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    v_max_u32_e32 v0, s8, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB15_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %val = atomicrmw volatile umax ptr %ptr, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_min_i32_offset(ptr %out, i32 %in) {
; GCN1-LABEL: atomic_min_i32_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_add_u32 s0, s2, 16
; GCN1-NEXT:    s_addc_u32 s1, s3, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB16_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_min_i32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB16_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_min_i32_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_add_u32 s0, s2, 16
; GCN2-NEXT:    s_addc_u32 s1, s3, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB16_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_min_i32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB16_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_min_i32_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s2
; GCN3-NEXT:    v_mov_b32_e32 v1, s3
; GCN3-NEXT:    flat_load_dword v1, v[0:1] offset:16
; GCN3-NEXT:  .LBB16_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v2, s2
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_min_i32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v3, s3
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB16_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %gep = getelementptr i32, ptr %out, i32 4
  %val = atomicrmw volatile min ptr %gep, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_min_i32_ret_offset(ptr %out, ptr %out2, i32 %in) {
; GCN1-LABEL: atomic_min_i32_ret_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s8, s[0:1], 0xd
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_add_u32 s0, s4, 16
; GCN1-NEXT:    s_addc_u32 s1, s5, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:  .LBB17_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    v_min_i32_e32 v0, s8, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB17_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_min_i32_ret_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_add_u32 s0, s4, 16
; GCN2-NEXT:    s_addc_u32 s1, s5, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:  .LBB17_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    v_min_i32_e32 v0, s8, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB17_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_min_i32_ret_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s2, s[0:1], 0x34
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s4
; GCN3-NEXT:    v_mov_b32_e32 v1, s5
; GCN3-NEXT:    flat_load_dword v0, v[0:1] offset:16
; GCN3-NEXT:  .LBB17_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v2, s4
; GCN3-NEXT:    v_mov_b32_e32 v3, s5
; GCN3-NEXT:    v_min_i32_e32 v0, s2, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB17_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %gep = getelementptr i32, ptr %out, i32 4
  %val = atomicrmw volatile min ptr %gep, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_min_i32_addr64_offset(ptr %out, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_min_i32_addr64_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xd
; GCN1-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s6, s0
; GCN1-NEXT:    s_addc_u32 s1, s7, s1
; GCN1-NEXT:    s_add_u32 s0, s0, 16
; GCN1-NEXT:    s_addc_u32 s1, s1, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB18_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_min_i32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB18_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_min_i32_addr64_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN2-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s6, s0
; GCN2-NEXT:    s_addc_u32 s1, s7, s1
; GCN2-NEXT:    s_add_u32 s0, s0, 16
; GCN2-NEXT:    s_addc_u32 s1, s1, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB18_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_min_i32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB18_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_min_i32_addr64_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN3-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s6, s0
; GCN3-NEXT:    s_addc_u32 s1, s7, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v1, v[0:1] offset:16
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB18_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_min_i32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB18_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %gep = getelementptr i32, ptr %ptr, i32 4
  %val = atomicrmw volatile min ptr %gep, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_min_i32_ret_addr64_offset(ptr %out, ptr %out2, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_min_i32_ret_addr64_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xf
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s8, s[0:1], 0xd
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s4, s0
; GCN1-NEXT:    s_addc_u32 s1, s5, s1
; GCN1-NEXT:    s_add_u32 s0, s0, 16
; GCN1-NEXT:    s_addc_u32 s1, s1, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB19_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    v_min_i32_e32 v0, s8, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB19_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_min_i32_ret_addr64_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s4, s0
; GCN2-NEXT:    s_addc_u32 s1, s5, s1
; GCN2-NEXT:    s_add_u32 s0, s0, 16
; GCN2-NEXT:    s_addc_u32 s1, s1, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB19_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    v_min_i32_e32 v0, s8, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB19_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_min_i32_ret_addr64_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s4, s0
; GCN3-NEXT:    s_addc_u32 s1, s5, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v0, v[0:1] offset:16
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB19_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    v_min_i32_e32 v0, s8, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB19_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %gep = getelementptr i32, ptr %ptr, i32 4
  %val = atomicrmw volatile min ptr %gep, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_min_i32(ptr %out, i32 %in) {
; GCN1-LABEL: atomic_min_i32:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_mov_b64 s[0:1], 0
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v0, s2
; GCN1-NEXT:    v_mov_b32_e32 v1, s3
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:  .LBB20_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v2, s2
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_min_i32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v3, s3
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN1-NEXT:    s_cbranch_execnz .LBB20_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_min_i32:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_mov_b64 s[0:1], 0
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v0, s2
; GCN2-NEXT:    v_mov_b32_e32 v1, s3
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:  .LBB20_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v2, s2
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_min_i32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v3, s3
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN2-NEXT:    s_cbranch_execnz .LBB20_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_min_i32:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s2
; GCN3-NEXT:    v_mov_b32_e32 v1, s3
; GCN3-NEXT:    flat_load_dword v1, v[0:1]
; GCN3-NEXT:  .LBB20_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v2, s2
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_min_i32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v3, s3
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB20_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %val = atomicrmw volatile min ptr %out, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_min_i32_ret(ptr %out, ptr %out2, i32 %in) {
; GCN1-LABEL: atomic_min_i32_ret:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s2, s[0:1], 0xd
; GCN1-NEXT:    s_mov_b64 s[0:1], 0
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v0, s4
; GCN1-NEXT:    v_mov_b32_e32 v1, s5
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:  .LBB21_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v2, s4
; GCN1-NEXT:    v_mov_b32_e32 v3, s5
; GCN1-NEXT:    v_min_i32_e32 v0, s2, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN1-NEXT:    s_cbranch_execnz .LBB21_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_min_i32_ret:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s2, s[0:1], 0x34
; GCN2-NEXT:    s_mov_b64 s[0:1], 0
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v0, s4
; GCN2-NEXT:    v_mov_b32_e32 v1, s5
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:  .LBB21_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v2, s4
; GCN2-NEXT:    v_mov_b32_e32 v3, s5
; GCN2-NEXT:    v_min_i32_e32 v0, s2, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN2-NEXT:    s_cbranch_execnz .LBB21_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_min_i32_ret:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s2, s[0:1], 0x34
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s4
; GCN3-NEXT:    v_mov_b32_e32 v1, s5
; GCN3-NEXT:    flat_load_dword v0, v[0:1]
; GCN3-NEXT:  .LBB21_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v2, s4
; GCN3-NEXT:    v_mov_b32_e32 v3, s5
; GCN3-NEXT:    v_min_i32_e32 v0, s2, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB21_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %val = atomicrmw volatile min ptr %out, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_min_i32_addr64(ptr %out, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_min_i32_addr64:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xd
; GCN1-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s6, s0
; GCN1-NEXT:    s_addc_u32 s1, s7, s1
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB22_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_min_i32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB22_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_min_i32_addr64:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN2-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s6, s0
; GCN2-NEXT:    s_addc_u32 s1, s7, s1
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB22_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_min_i32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB22_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_min_i32_addr64:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN3-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s6, s0
; GCN3-NEXT:    s_addc_u32 s1, s7, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v1, v[0:1]
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB22_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_min_i32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB22_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %val = atomicrmw volatile min ptr %ptr, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_min_i32_ret_addr64(ptr %out, ptr %out2, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_min_i32_ret_addr64:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xf
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s8, s[0:1], 0xd
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s4, s0
; GCN1-NEXT:    s_addc_u32 s1, s5, s1
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB23_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    v_min_i32_e32 v0, s8, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB23_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_min_i32_ret_addr64:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s4, s0
; GCN2-NEXT:    s_addc_u32 s1, s5, s1
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB23_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    v_min_i32_e32 v0, s8, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB23_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_min_i32_ret_addr64:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s4, s0
; GCN3-NEXT:    s_addc_u32 s1, s5, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v0, v[0:1]
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB23_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    v_min_i32_e32 v0, s8, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB23_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %val = atomicrmw volatile min ptr %ptr, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_umin_i32_offset(ptr %out, i32 %in) {
; GCN1-LABEL: atomic_umin_i32_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_add_u32 s0, s2, 16
; GCN1-NEXT:    s_addc_u32 s1, s3, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB24_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_min_u32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB24_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umin_i32_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_add_u32 s0, s2, 16
; GCN2-NEXT:    s_addc_u32 s1, s3, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB24_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_min_u32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB24_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umin_i32_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s2
; GCN3-NEXT:    v_mov_b32_e32 v1, s3
; GCN3-NEXT:    flat_load_dword v1, v[0:1] offset:16
; GCN3-NEXT:  .LBB24_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v2, s2
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_min_u32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v3, s3
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB24_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %gep = getelementptr i32, ptr %out, i32 4
  %val = atomicrmw volatile umin ptr %gep, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_umin_i32_ret_offset(ptr %out, ptr %out2, i32 %in) {
; GCN1-LABEL: atomic_umin_i32_ret_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s8, s[0:1], 0xd
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_add_u32 s0, s4, 16
; GCN1-NEXT:    s_addc_u32 s1, s5, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:  .LBB25_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    v_min_u32_e32 v0, s8, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB25_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umin_i32_ret_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_add_u32 s0, s4, 16
; GCN2-NEXT:    s_addc_u32 s1, s5, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:  .LBB25_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    v_min_u32_e32 v0, s8, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB25_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umin_i32_ret_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s2, s[0:1], 0x34
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s4
; GCN3-NEXT:    v_mov_b32_e32 v1, s5
; GCN3-NEXT:    flat_load_dword v0, v[0:1] offset:16
; GCN3-NEXT:  .LBB25_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v2, s4
; GCN3-NEXT:    v_mov_b32_e32 v3, s5
; GCN3-NEXT:    v_min_u32_e32 v0, s2, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB25_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %gep = getelementptr i32, ptr %out, i32 4
  %val = atomicrmw volatile umin ptr %gep, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_umin_i32_addr64_offset(ptr %out, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_umin_i32_addr64_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xd
; GCN1-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s6, s0
; GCN1-NEXT:    s_addc_u32 s1, s7, s1
; GCN1-NEXT:    s_add_u32 s0, s0, 16
; GCN1-NEXT:    s_addc_u32 s1, s1, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB26_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_min_u32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB26_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umin_i32_addr64_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN2-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s6, s0
; GCN2-NEXT:    s_addc_u32 s1, s7, s1
; GCN2-NEXT:    s_add_u32 s0, s0, 16
; GCN2-NEXT:    s_addc_u32 s1, s1, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB26_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_min_u32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB26_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umin_i32_addr64_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN3-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s6, s0
; GCN3-NEXT:    s_addc_u32 s1, s7, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v1, v[0:1] offset:16
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB26_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_min_u32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB26_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %gep = getelementptr i32, ptr %ptr, i32 4
  %val = atomicrmw volatile umin ptr %gep, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_umin_i32_ret_addr64_offset(ptr %out, ptr %out2, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_umin_i32_ret_addr64_offset:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xf
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s8, s[0:1], 0xd
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s4, s0
; GCN1-NEXT:    s_addc_u32 s1, s5, s1
; GCN1-NEXT:    s_add_u32 s0, s0, 16
; GCN1-NEXT:    s_addc_u32 s1, s1, 0
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB27_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    v_min_u32_e32 v0, s8, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB27_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umin_i32_ret_addr64_offset:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s4, s0
; GCN2-NEXT:    s_addc_u32 s1, s5, s1
; GCN2-NEXT:    s_add_u32 s0, s0, 16
; GCN2-NEXT:    s_addc_u32 s1, s1, 0
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB27_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    v_min_u32_e32 v0, s8, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB27_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umin_i32_ret_addr64_offset:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s4, s0
; GCN3-NEXT:    s_addc_u32 s1, s5, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v0, v[0:1] offset:16
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB27_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    v_min_u32_e32 v0, s8, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] offset:16 glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB27_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %gep = getelementptr i32, ptr %ptr, i32 4
  %val = atomicrmw volatile umin ptr %gep, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_umin_i32(ptr %out, i32 %in) {
; GCN1-LABEL: atomic_umin_i32:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_mov_b64 s[0:1], 0
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v0, s2
; GCN1-NEXT:    v_mov_b32_e32 v1, s3
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:  .LBB28_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v2, s2
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_min_u32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v3, s3
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN1-NEXT:    s_cbranch_execnz .LBB28_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umin_i32:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_mov_b64 s[0:1], 0
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v0, s2
; GCN2-NEXT:    v_mov_b32_e32 v1, s3
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:  .LBB28_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v2, s2
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_min_u32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v3, s3
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN2-NEXT:    s_cbranch_execnz .LBB28_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umin_i32:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s2
; GCN3-NEXT:    v_mov_b32_e32 v1, s3
; GCN3-NEXT:    flat_load_dword v1, v[0:1]
; GCN3-NEXT:  .LBB28_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v2, s2
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_min_u32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v3, s3
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB28_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %val = atomicrmw volatile umin ptr %out, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_umin_i32_ret(ptr %out, ptr %out2, i32 %in) {
; GCN1-LABEL: atomic_umin_i32_ret:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s2, s[0:1], 0xd
; GCN1-NEXT:    s_mov_b64 s[0:1], 0
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v0, s4
; GCN1-NEXT:    v_mov_b32_e32 v1, s5
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:  .LBB29_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v2, s4
; GCN1-NEXT:    v_mov_b32_e32 v3, s5
; GCN1-NEXT:    v_min_u32_e32 v0, s2, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN1-NEXT:    s_cbranch_execnz .LBB29_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umin_i32_ret:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s2, s[0:1], 0x34
; GCN2-NEXT:    s_mov_b64 s[0:1], 0
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v0, s4
; GCN2-NEXT:    v_mov_b32_e32 v1, s5
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:  .LBB29_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v2, s4
; GCN2-NEXT:    v_mov_b32_e32 v3, s5
; GCN2-NEXT:    v_min_u32_e32 v0, s2, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN2-NEXT:    s_cbranch_execnz .LBB29_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umin_i32_ret:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s2, s[0:1], 0x34
; GCN3-NEXT:    s_mov_b64 s[0:1], 0
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v0, s4
; GCN3-NEXT:    v_mov_b32_e32 v1, s5
; GCN3-NEXT:    flat_load_dword v0, v[0:1]
; GCN3-NEXT:  .LBB29_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v2, s4
; GCN3-NEXT:    v_mov_b32_e32 v3, s5
; GCN3-NEXT:    v_min_u32_e32 v0, s2, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[0:1], vcc, s[0:1]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[0:1]
; GCN3-NEXT:    s_cbranch_execnz .LBB29_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[0:1]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %val = atomicrmw volatile umin ptr %out, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}

define amdgpu_kernel void @atomic_umin_i32_addr64(ptr %out, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_umin_i32_addr64:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xd
; GCN1-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s4, s[0:1], 0xb
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s6, s0
; GCN1-NEXT:    s_addc_u32 s1, s7, s1
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v1, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB30_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_min_u32_e32 v0, s4, v1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB30_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umin_i32_addr64:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN2-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s6, s0
; GCN2-NEXT:    s_addc_u32 s1, s7, s1
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v1, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB30_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_min_u32_e32 v0, s4, v1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB30_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umin_i32_addr64:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GCN3-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s6, s0
; GCN3-NEXT:    s_addc_u32 s1, s7, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v1, v[0:1]
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB30_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_min_u32_e32 v0, s4, v1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB30_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %val = atomicrmw volatile umin ptr %ptr, i32 %in seq_cst
  ret void
}

define amdgpu_kernel void @atomic_umin_i32_ret_addr64(ptr %out, ptr %out2, i32 %in, i64 %index) {
; GCN1-LABEL: atomic_umin_i32_ret_addr64:
; GCN1:       ; %bb.0: ; %entry
; GCN1-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xf
; GCN1-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GCN1-NEXT:    s_load_dword s8, s[0:1], 0xd
; GCN1-NEXT:    s_waitcnt lgkmcnt(0)
; GCN1-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN1-NEXT:    s_add_u32 s0, s4, s0
; GCN1-NEXT:    s_addc_u32 s1, s5, s1
; GCN1-NEXT:    v_mov_b32_e32 v0, s0
; GCN1-NEXT:    v_mov_b32_e32 v1, s1
; GCN1-NEXT:    flat_load_dword v0, v[0:1]
; GCN1-NEXT:    s_mov_b64 s[2:3], 0
; GCN1-NEXT:  .LBB31_1: ; %atomicrmw.start
; GCN1-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    v_mov_b32_e32 v1, v0
; GCN1-NEXT:    v_mov_b32_e32 v3, s1
; GCN1-NEXT:    v_mov_b32_e32 v2, s0
; GCN1-NEXT:    v_min_u32_e32 v0, s8, v1
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN1-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN1-NEXT:    buffer_wbinvl1_vol
; GCN1-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN1-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN1-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN1-NEXT:    s_cbranch_execnz .LBB31_1
; GCN1-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN1-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN1-NEXT:    v_mov_b32_e32 v1, s6
; GCN1-NEXT:    v_mov_b32_e32 v2, s7
; GCN1-NEXT:    flat_store_dword v[1:2], v0
; GCN1-NEXT:    s_endpgm
;
; GCN2-LABEL: atomic_umin_i32_ret_addr64:
; GCN2:       ; %bb.0: ; %entry
; GCN2-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN2-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN2-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN2-NEXT:    s_waitcnt lgkmcnt(0)
; GCN2-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN2-NEXT:    s_add_u32 s0, s4, s0
; GCN2-NEXT:    s_addc_u32 s1, s5, s1
; GCN2-NEXT:    v_mov_b32_e32 v0, s0
; GCN2-NEXT:    v_mov_b32_e32 v1, s1
; GCN2-NEXT:    flat_load_dword v0, v[0:1]
; GCN2-NEXT:    s_mov_b64 s[2:3], 0
; GCN2-NEXT:  .LBB31_1: ; %atomicrmw.start
; GCN2-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    v_mov_b32_e32 v1, v0
; GCN2-NEXT:    v_mov_b32_e32 v3, s1
; GCN2-NEXT:    v_mov_b32_e32 v2, s0
; GCN2-NEXT:    v_min_u32_e32 v0, s8, v1
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN2-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN2-NEXT:    buffer_wbinvl1_vol
; GCN2-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN2-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN2-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN2-NEXT:    s_cbranch_execnz .LBB31_1
; GCN2-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN2-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN2-NEXT:    v_mov_b32_e32 v1, s6
; GCN2-NEXT:    v_mov_b32_e32 v2, s7
; GCN2-NEXT:    flat_store_dword v[1:2], v0
; GCN2-NEXT:    s_endpgm
;
; GCN3-LABEL: atomic_umin_i32_ret_addr64:
; GCN3:       ; %bb.0: ; %entry
; GCN3-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
; GCN3-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GCN3-NEXT:    s_load_dword s8, s[0:1], 0x34
; GCN3-NEXT:    s_waitcnt lgkmcnt(0)
; GCN3-NEXT:    s_lshl_b64 s[0:1], s[2:3], 2
; GCN3-NEXT:    s_add_u32 s0, s4, s0
; GCN3-NEXT:    s_addc_u32 s1, s5, s1
; GCN3-NEXT:    v_mov_b32_e32 v0, s0
; GCN3-NEXT:    v_mov_b32_e32 v1, s1
; GCN3-NEXT:    flat_load_dword v0, v[0:1]
; GCN3-NEXT:    s_mov_b64 s[2:3], 0
; GCN3-NEXT:  .LBB31_1: ; %atomicrmw.start
; GCN3-NEXT:    ; =>This Inner Loop Header: Depth=1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    v_mov_b32_e32 v1, v0
; GCN3-NEXT:    v_mov_b32_e32 v3, s1
; GCN3-NEXT:    v_mov_b32_e32 v2, s0
; GCN3-NEXT:    v_min_u32_e32 v0, s8, v1
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    flat_atomic_cmpswap v0, v[2:3], v[0:1] glc
; GCN3-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN3-NEXT:    buffer_wbinvl1_vol
; GCN3-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
; GCN3-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GCN3-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GCN3-NEXT:    s_cbranch_execnz .LBB31_1
; GCN3-NEXT:  ; %bb.2: ; %atomicrmw.end
; GCN3-NEXT:    s_or_b64 exec, exec, s[2:3]
; GCN3-NEXT:    v_mov_b32_e32 v1, s6
; GCN3-NEXT:    v_mov_b32_e32 v2, s7
; GCN3-NEXT:    flat_store_dword v[1:2], v0
; GCN3-NEXT:    s_endpgm
entry:
  %ptr = getelementptr i32, ptr %out, i64 %index
  %val = atomicrmw volatile umin ptr %ptr, i32 %in seq_cst
  store i32 %val, ptr %out2
  ret void
}