; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn -amdgpu-atomic-optimizer-strategy=Iterative -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX6 %s
; RUN: llc -mtriple=amdgcn -mcpu=tonga -amdgpu-atomic-optimizer-strategy=Iterative -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX8 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -amdgpu-atomic-optimizer-strategy=Iterative -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX9 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx1010 -mattr=-wavefrontsize32,+wavefrontsize64 -amdgpu-atomic-optimizer-strategy=Iterative -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX10,GFX10W64 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx1010 -mattr=+wavefrontsize32,-wavefrontsize64 -amdgpu-atomic-optimizer-strategy=Iterative -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX10,GFX10W32 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -amdgpu-atomic-optimizer-strategy=Iterative -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11W64 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32,-wavefrontsize64 -amdgpu-atomic-optimizer-strategy=Iterative -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11W32 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=-wavefrontsize32,+wavefrontsize64 -amdgpu-atomic-optimizer-strategy=Iterative -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12W64 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=+wavefrontsize32,-wavefrontsize64 -amdgpu-atomic-optimizer-strategy=Iterative -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12W32 %s

declare i32 @llvm.amdgcn.workitem.id.x()
declare i32 @llvm.amdgcn.raw.ptr.buffer.atomic.add(i32, ptr addrspace(8), i32, i32, i32)
declare i32 @llvm.amdgcn.raw.ptr.buffer.atomic.sub(i32, ptr addrspace(8), i32, i32, i32)

; Show what the atomic optimization pass will do for raw buffers.

define amdgpu_kernel void @add_i32_constant(ptr addrspace(1) %out, ptr addrspace(8) %inout) {
; GFX6-LABEL: add_i32_constant:
; GFX6:       ; %bb.0: ; %entry
; GFX6-NEXT:    s_mov_b64 s[4:5], exec
; GFX6-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, s4, 0
; GFX6-NEXT:    v_mbcnt_hi_u32_b32_e32 v0, s5, v0
; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX6-NEXT:    ; implicit-def: $vgpr1
; GFX6-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX6-NEXT:    s_cbranch_execz .LBB0_2
; GFX6-NEXT:  ; %bb.1:
; GFX6-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0xd
; GFX6-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX6-NEXT:    s_mul_i32 s4, s4, 5
; GFX6-NEXT:    v_mov_b32_e32 v1, s4
; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NEXT:    buffer_atomic_add v1, off, s[8:11], 0 glc
; GFX6-NEXT:  .LBB0_2:
; GFX6-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT:    s_mov_b32 s3, 0xf000
; GFX6-NEXT:    s_mov_b32 s2, -1
; GFX6-NEXT:    s_waitcnt vmcnt(0)
; GFX6-NEXT:    v_readfirstlane_b32 s4, v1
; GFX6-NEXT:    v_mad_u32_u24 v0, v0, 5, s4
; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT:    s_endpgm
;
; GFX8-LABEL: add_i32_constant:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_mov_b64 s[4:5], exec
; GFX8-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX8-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX8-NEXT:    ; implicit-def: $vgpr1
; GFX8-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX8-NEXT:    s_cbranch_execz .LBB0_2
; GFX8-NEXT:  ; %bb.1:
; GFX8-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX8-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX8-NEXT:    s_mul_i32 s4, s4, 5
; GFX8-NEXT:    v_mov_b32_e32 v1, s4
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    buffer_atomic_add v1, off, s[8:11], 0 glc
; GFX8-NEXT:  .LBB0_2:
; GFX8-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    v_readfirstlane_b32 s2, v1
; GFX8-NEXT:    v_mad_u32_u24 v2, v0, 5, s2
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_store_dword v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-LABEL: add_i32_constant:
; GFX9:       ; %bb.0: ; %entry
; GFX9-NEXT:    s_mov_b64 s[4:5], exec
; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX9-NEXT:    ; implicit-def: $vgpr1
; GFX9-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX9-NEXT:    s_cbranch_execz .LBB0_2
; GFX9-NEXT:  ; %bb.1:
; GFX9-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX9-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX9-NEXT:    s_mul_i32 s4, s4, 5
; GFX9-NEXT:    v_mov_b32_e32 v1, s4
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    buffer_atomic_add v1, off, s[8:11], 0 glc
; GFX9-NEXT:  .LBB0_2:
; GFX9-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_readfirstlane_b32 s2, v1
; GFX9-NEXT:    v_mov_b32_e32 v2, 0
; GFX9-NEXT:    v_mad_u32_u24 v0, v0, 5, s2
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_store_dword v2, v0, s[0:1]
; GFX9-NEXT:    s_endpgm
;
; GFX10W64-LABEL: add_i32_constant:
; GFX10W64:       ; %bb.0: ; %entry
; GFX10W64-NEXT:    s_mov_b64 s[4:5], exec
; GFX10W64-NEXT:    ; implicit-def: $vgpr1
; GFX10W64-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX10W64-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX10W64-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX10W64-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX10W64-NEXT:    s_cbranch_execz .LBB0_2
; GFX10W64-NEXT:  ; %bb.1:
; GFX10W64-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W64-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX10W64-NEXT:    s_mul_i32 s4, s4, 5
; GFX10W64-NEXT:    v_mov_b32_e32 v1, s4
; GFX10W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT:    buffer_atomic_add v1, off, s[8:11], 0 glc
; GFX10W64-NEXT:  .LBB0_2:
; GFX10W64-NEXT:    s_waitcnt_depctr 0xffe3
; GFX10W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX10W64-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W64-NEXT:    s_waitcnt vmcnt(0)
; GFX10W64-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10W64-NEXT:    v_mov_b32_e32 v1, 0
; GFX10W64-NEXT:    v_mad_u32_u24 v0, v0, 5, s2
; GFX10W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10W64-NEXT:    s_endpgm
;
; GFX10W32-LABEL: add_i32_constant:
; GFX10W32:       ; %bb.0: ; %entry
; GFX10W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX10W32-NEXT:    ; implicit-def: $vgpr1
; GFX10W32-NEXT:    v_mbcnt_lo_u32_b32 v0, s3, 0
; GFX10W32-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX10W32-NEXT:    s_and_saveexec_b32 s2, vcc_lo
; GFX10W32-NEXT:    s_cbranch_execz .LBB0_2
; GFX10W32-NEXT:  ; %bb.1:
; GFX10W32-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX10W32-NEXT:    s_bcnt1_i32_b32 s3, s3
; GFX10W32-NEXT:    s_mul_i32 s3, s3, 5
; GFX10W32-NEXT:    v_mov_b32_e32 v1, s3
; GFX10W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT:    buffer_atomic_add v1, off, s[4:7], 0 glc
; GFX10W32-NEXT:  .LBB0_2:
; GFX10W32-NEXT:    s_waitcnt_depctr 0xffe3
; GFX10W32-NEXT:    s_or_b32 exec_lo, exec_lo, s2
; GFX10W32-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W32-NEXT:    s_waitcnt vmcnt(0)
; GFX10W32-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10W32-NEXT:    v_mov_b32_e32 v1, 0
; GFX10W32-NEXT:    v_mad_u32_u24 v0, v0, 5, s2
; GFX10W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10W32-NEXT:    s_endpgm
;
; GFX11W64-LABEL: add_i32_constant:
; GFX11W64:       ; %bb.0: ; %entry
; GFX11W64-NEXT:    s_mov_b64 s[4:5], exec
; GFX11W64-NEXT:    s_mov_b64 s[2:3], exec
; GFX11W64-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX11W64-NEXT:    ; implicit-def: $vgpr1
; GFX11W64-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX11W64-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX11W64-NEXT:    s_cbranch_execz .LBB0_2
; GFX11W64-NEXT:  ; %bb.1:
; GFX11W64-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W64-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX11W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11W64-NEXT:    s_mul_i32 s4, s4, 5
; GFX11W64-NEXT:    v_mov_b32_e32 v1, s4
; GFX11W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT:    buffer_atomic_add_u32 v1, off, s[8:11], 0 glc
; GFX11W64-NEXT:  .LBB0_2:
; GFX11W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX11W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W64-NEXT:    s_waitcnt vmcnt(0)
; GFX11W64-NEXT:    v_readfirstlane_b32 s2, v1
; GFX11W64-NEXT:    v_mov_b32_e32 v1, 0
; GFX11W64-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11W64-NEXT:    v_mad_u32_u24 v0, v0, 5, s2
; GFX11W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11W64-NEXT:    s_nop 0
; GFX11W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W64-NEXT:    s_endpgm
;
; GFX11W32-LABEL: add_i32_constant:
; GFX11W32:       ; %bb.0: ; %entry
; GFX11W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX11W32-NEXT:    s_mov_b32 s2, exec_lo
; GFX11W32-NEXT:    v_mbcnt_lo_u32_b32 v0, s3, 0
; GFX11W32-NEXT:    ; implicit-def: $vgpr1
; GFX11W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11W32-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX11W32-NEXT:    s_cbranch_execz .LBB0_2
; GFX11W32-NEXT:  ; %bb.1:
; GFX11W32-NEXT:    s_load_b128 s[4:7], s[0:1], 0x34
; GFX11W32-NEXT:    s_bcnt1_i32_b32 s3, s3
; GFX11W32-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11W32-NEXT:    s_mul_i32 s3, s3, 5
; GFX11W32-NEXT:    v_mov_b32_e32 v1, s3
; GFX11W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT:    buffer_atomic_add_u32 v1, off, s[4:7], 0 glc
; GFX11W32-NEXT:  .LBB0_2:
; GFX11W32-NEXT:    s_or_b32 exec_lo, exec_lo, s2
; GFX11W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W32-NEXT:    s_waitcnt vmcnt(0)
; GFX11W32-NEXT:    v_readfirstlane_b32 s2, v1
; GFX11W32-NEXT:    v_mov_b32_e32 v1, 0
; GFX11W32-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11W32-NEXT:    v_mad_u32_u24 v0, v0, 5, s2
; GFX11W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11W32-NEXT:    s_nop 0
; GFX11W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W32-NEXT:    s_endpgm
;
; GFX12W64-LABEL: add_i32_constant:
; GFX12W64:       ; %bb.0: ; %entry
; GFX12W64-NEXT:    s_mov_b64 s[4:5], exec
; GFX12W64-NEXT:    s_mov_b64 s[2:3], exec
; GFX12W64-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX12W64-NEXT:    ; implicit-def: $vgpr1
; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12W64-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX12W64-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX12W64-NEXT:    s_cbranch_execz .LBB0_2
; GFX12W64-NEXT:  ; %bb.1:
; GFX12W64-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX12W64-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX12W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX12W64-NEXT:    s_mul_i32 s4, s4, 5
; GFX12W64-NEXT:    v_mov_b32_e32 v1, s4
; GFX12W64-NEXT:    s_wait_kmcnt 0x0
; GFX12W64-NEXT:    buffer_atomic_add_u32 v1, off, s[8:11], null th:TH_ATOMIC_RETURN
; GFX12W64-NEXT:  .LBB0_2:
; GFX12W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX12W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX12W64-NEXT:    s_wait_loadcnt 0x0
; GFX12W64-NEXT:    v_readfirstlane_b32 s2, v1
; GFX12W64-NEXT:    v_mov_b32_e32 v1, 0
; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12W64-NEXT:    v_mad_u32_u24 v0, v0, 5, s2
; GFX12W64-NEXT:    s_wait_kmcnt 0x0
; GFX12W64-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12W64-NEXT:    s_nop 0
; GFX12W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX12W64-NEXT:    s_endpgm
;
; GFX12W32-LABEL: add_i32_constant:
; GFX12W32:       ; %bb.0: ; %entry
; GFX12W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX12W32-NEXT:    s_mov_b32 s2, exec_lo
; GFX12W32-NEXT:    v_mbcnt_lo_u32_b32 v0, s3, 0
; GFX12W32-NEXT:    ; implicit-def: $vgpr1
; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12W32-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX12W32-NEXT:    s_cbranch_execz .LBB0_2
; GFX12W32-NEXT:  ; %bb.1:
; GFX12W32-NEXT:    s_load_b128 s[4:7], s[0:1], 0x34
; GFX12W32-NEXT:    s_bcnt1_i32_b32 s3, s3
; GFX12W32-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX12W32-NEXT:    s_mul_i32 s3, s3, 5
; GFX12W32-NEXT:    v_mov_b32_e32 v1, s3
; GFX12W32-NEXT:    s_wait_kmcnt 0x0
; GFX12W32-NEXT:    buffer_atomic_add_u32 v1, off, s[4:7], null th:TH_ATOMIC_RETURN
; GFX12W32-NEXT:  .LBB0_2:
; GFX12W32-NEXT:    s_or_b32 exec_lo, exec_lo, s2
; GFX12W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX12W32-NEXT:    s_wait_loadcnt 0x0
; GFX12W32-NEXT:    v_readfirstlane_b32 s2, v1
; GFX12W32-NEXT:    v_mov_b32_e32 v1, 0
; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12W32-NEXT:    v_mad_u32_u24 v0, v0, 5, s2
; GFX12W32-NEXT:    s_wait_kmcnt 0x0
; GFX12W32-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12W32-NEXT:    s_nop 0
; GFX12W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX12W32-NEXT:    s_endpgm
entry:
  %old = call i32 @llvm.amdgcn.raw.ptr.buffer.atomic.add(i32 5, ptr addrspace(8) %inout, i32 0, i32 0, i32 0)
  store i32 %old, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @add_i32_uniform(ptr addrspace(1) %out, ptr addrspace(8) %inout, i32 %additive) {
; GFX6-LABEL: add_i32_uniform:
; GFX6:       ; %bb.0: ; %entry
; GFX6-NEXT:    s_mov_b64 s[4:5], exec
; GFX6-NEXT:    s_load_dword s6, s[0:1], 0x11
; GFX6-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, s4, 0
; GFX6-NEXT:    v_mbcnt_hi_u32_b32_e32 v0, s5, v0
; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX6-NEXT:    ; implicit-def: $vgpr1
; GFX6-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX6-NEXT:    s_cbranch_execz .LBB1_2
; GFX6-NEXT:  ; %bb.1:
; GFX6-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0xd
; GFX6-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NEXT:    s_mul_i32 s4, s6, s4
; GFX6-NEXT:    v_mov_b32_e32 v1, s4
; GFX6-NEXT:    buffer_atomic_add v1, off, s[8:11], 0 glc
; GFX6-NEXT:  .LBB1_2:
; GFX6-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT:    s_mov_b32 s3, 0xf000
; GFX6-NEXT:    s_mov_b32 s2, -1
; GFX6-NEXT:    s_waitcnt vmcnt(0)
; GFX6-NEXT:    v_readfirstlane_b32 s4, v1
; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NEXT:    v_mul_lo_u32 v0, s6, v0
; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s4, v0
; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT:    s_endpgm
;
; GFX8-LABEL: add_i32_uniform:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dword s6, s[0:1], 0x44
; GFX8-NEXT:    s_mov_b64 s[4:5], exec
; GFX8-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX8-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX8-NEXT:    ; implicit-def: $vgpr1
; GFX8-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX8-NEXT:    s_cbranch_execz .LBB1_2
; GFX8-NEXT:  ; %bb.1:
; GFX8-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX8-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_mul_i32 s4, s6, s4
; GFX8-NEXT:    v_mov_b32_e32 v1, s4
; GFX8-NEXT:    buffer_atomic_add v1, off, s[8:11], 0 glc
; GFX8-NEXT:  .LBB1_2:
; GFX8-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mul_lo_u32 v0, s6, v0
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    v_readfirstlane_b32 s2, v1
; GFX8-NEXT:    v_add_u32_e32 v2, vcc, s2, v0
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_store_dword v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-LABEL: add_i32_uniform:
; GFX9:       ; %bb.0: ; %entry
; GFX9-NEXT:    s_load_dword s6, s[0:1], 0x44
; GFX9-NEXT:    s_mov_b64 s[4:5], exec
; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX9-NEXT:    ; implicit-def: $vgpr1
; GFX9-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX9-NEXT:    s_cbranch_execz .LBB1_2
; GFX9-NEXT:  ; %bb.1:
; GFX9-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX9-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_mul_i32 s4, s6, s4
; GFX9-NEXT:    v_mov_b32_e32 v1, s4
; GFX9-NEXT:    buffer_atomic_add v1, off, s[8:11], 0 glc
; GFX9-NEXT:  .LBB1_2:
; GFX9-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_mul_lo_u32 v0, s6, v0
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_readfirstlane_b32 s2, v1
; GFX9-NEXT:    v_mov_b32_e32 v2, 0
; GFX9-NEXT:    v_add_u32_e32 v0, s2, v0
; GFX9-NEXT:    global_store_dword v2, v0, s[0:1]
; GFX9-NEXT:    s_endpgm
;
; GFX10W64-LABEL: add_i32_uniform:
; GFX10W64:       ; %bb.0: ; %entry
; GFX10W64-NEXT:    s_load_dword s6, s[0:1], 0x44
; GFX10W64-NEXT:    s_mov_b64 s[4:5], exec
; GFX10W64-NEXT:    ; implicit-def: $vgpr1
; GFX10W64-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX10W64-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX10W64-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX10W64-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX10W64-NEXT:    s_cbranch_execz .LBB1_2
; GFX10W64-NEXT:  ; %bb.1:
; GFX10W64-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W64-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX10W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT:    s_mul_i32 s4, s6, s4
; GFX10W64-NEXT:    v_mov_b32_e32 v1, s4
; GFX10W64-NEXT:    buffer_atomic_add v1, off, s[8:11], 0 glc
; GFX10W64-NEXT:  .LBB1_2:
; GFX10W64-NEXT:    s_waitcnt_depctr 0xffe3
; GFX10W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX10W64-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W64-NEXT:    s_waitcnt vmcnt(0)
; GFX10W64-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT:    v_mad_u64_u32 v[0:1], s[2:3], s6, v0, s[2:3]
; GFX10W64-NEXT:    v_mov_b32_e32 v1, 0
; GFX10W64-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10W64-NEXT:    s_endpgm
;
; GFX10W32-LABEL: add_i32_uniform:
; GFX10W32:       ; %bb.0: ; %entry
; GFX10W32-NEXT:    s_load_dword s2, s[0:1], 0x44
; GFX10W32-NEXT:    s_mov_b32 s4, exec_lo
; GFX10W32-NEXT:    ; implicit-def: $vgpr1
; GFX10W32-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX10W32-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX10W32-NEXT:    s_and_saveexec_b32 s3, vcc_lo
; GFX10W32-NEXT:    s_cbranch_execz .LBB1_2
; GFX10W32-NEXT:  ; %bb.1:
; GFX10W32-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W32-NEXT:    s_bcnt1_i32_b32 s4, s4
; GFX10W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT:    s_mul_i32 s4, s2, s4
; GFX10W32-NEXT:    v_mov_b32_e32 v1, s4
; GFX10W32-NEXT:    buffer_atomic_add v1, off, s[8:11], 0 glc
; GFX10W32-NEXT:  .LBB1_2:
; GFX10W32-NEXT:    s_waitcnt_depctr 0xffe3
; GFX10W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
; GFX10W32-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W32-NEXT:    s_waitcnt vmcnt(0)
; GFX10W32-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT:    v_mad_u64_u32 v[0:1], s2, s2, v0, s[4:5]
; GFX10W32-NEXT:    v_mov_b32_e32 v1, 0
; GFX10W32-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10W32-NEXT:    s_endpgm
;
; GFX11W64-LABEL: add_i32_uniform:
; GFX11W64:       ; %bb.0: ; %entry
; GFX11W64-NEXT:    s_load_b32 s6, s[0:1], 0x44
; GFX11W64-NEXT:    s_mov_b64 s[4:5], exec
; GFX11W64-NEXT:    s_mov_b64 s[2:3], exec
; GFX11W64-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX11W64-NEXT:    ; implicit-def: $vgpr1
; GFX11W64-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX11W64-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX11W64-NEXT:    s_cbranch_execz .LBB1_2
; GFX11W64-NEXT:  ; %bb.1:
; GFX11W64-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W64-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX11W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT:    s_mul_i32 s4, s6, s4
; GFX11W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W64-NEXT:    v_mov_b32_e32 v1, s4
; GFX11W64-NEXT:    buffer_atomic_add_u32 v1, off, s[8:11], 0 glc
; GFX11W64-NEXT:  .LBB1_2:
; GFX11W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX11W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W64-NEXT:    s_waitcnt vmcnt(0)
; GFX11W64-NEXT:    v_readfirstlane_b32 s2, v1
; GFX11W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11W64-NEXT:    v_mad_u64_u32 v[1:2], null, s6, v0, s[2:3]
; GFX11W64-NEXT:    v_mov_b32_e32 v0, 0
; GFX11W64-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11W64-NEXT:    s_nop 0
; GFX11W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W64-NEXT:    s_endpgm
;
; GFX11W32-LABEL: add_i32_uniform:
; GFX11W32:       ; %bb.0: ; %entry
; GFX11W32-NEXT:    s_load_b32 s2, s[0:1], 0x44
; GFX11W32-NEXT:    s_mov_b32 s4, exec_lo
; GFX11W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX11W32-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX11W32-NEXT:    ; implicit-def: $vgpr1
; GFX11W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11W32-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX11W32-NEXT:    s_cbranch_execz .LBB1_2
; GFX11W32-NEXT:  ; %bb.1:
; GFX11W32-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W32-NEXT:    s_bcnt1_i32_b32 s4, s4
; GFX11W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT:    s_mul_i32 s4, s2, s4
; GFX11W32-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W32-NEXT:    v_mov_b32_e32 v1, s4
; GFX11W32-NEXT:    buffer_atomic_add_u32 v1, off, s[8:11], 0 glc
; GFX11W32-NEXT:  .LBB1_2:
; GFX11W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
; GFX11W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W32-NEXT:    s_waitcnt vmcnt(0)
; GFX11W32-NEXT:    v_readfirstlane_b32 s4, v1
; GFX11W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11W32-NEXT:    v_mad_u64_u32 v[1:2], null, s2, v0, s[4:5]
; GFX11W32-NEXT:    v_mov_b32_e32 v0, 0
; GFX11W32-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11W32-NEXT:    s_nop 0
; GFX11W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W32-NEXT:    s_endpgm
;
; GFX12W64-LABEL: add_i32_uniform:
; GFX12W64:       ; %bb.0: ; %entry
; GFX12W64-NEXT:    s_load_b32 s6, s[0:1], 0x44
; GFX12W64-NEXT:    s_mov_b64 s[4:5], exec
; GFX12W64-NEXT:    s_mov_b64 s[2:3], exec
; GFX12W64-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX12W64-NEXT:    ; implicit-def: $vgpr1
; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12W64-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX12W64-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX12W64-NEXT:    s_cbranch_execz .LBB1_2
; GFX12W64-NEXT:  ; %bb.1:
; GFX12W64-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX12W64-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX12W64-NEXT:    s_wait_kmcnt 0x0
; GFX12W64-NEXT:    s_mul_i32 s4, s6, s4
; GFX12W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX12W64-NEXT:    v_mov_b32_e32 v1, s4
; GFX12W64-NEXT:    buffer_atomic_add_u32 v1, off, s[8:11], null th:TH_ATOMIC_RETURN
; GFX12W64-NEXT:  .LBB1_2:
; GFX12W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX12W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX12W64-NEXT:    s_wait_loadcnt 0x0
; GFX12W64-NEXT:    v_readfirstlane_b32 s2, v1
; GFX12W64-NEXT:    s_wait_kmcnt 0x0
; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12W64-NEXT:    v_mad_co_u64_u32 v[0:1], null, s6, v0, s[2:3]
; GFX12W64-NEXT:    v_mov_b32_e32 v1, 0
; GFX12W64-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12W64-NEXT:    s_nop 0
; GFX12W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX12W64-NEXT:    s_endpgm
;
; GFX12W32-LABEL: add_i32_uniform:
; GFX12W32:       ; %bb.0: ; %entry
; GFX12W32-NEXT:    s_load_b32 s2, s[0:1], 0x44
; GFX12W32-NEXT:    s_mov_b32 s4, exec_lo
; GFX12W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX12W32-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX12W32-NEXT:    ; implicit-def: $vgpr1
; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12W32-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX12W32-NEXT:    s_cbranch_execz .LBB1_2
; GFX12W32-NEXT:  ; %bb.1:
; GFX12W32-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX12W32-NEXT:    s_bcnt1_i32_b32 s4, s4
; GFX12W32-NEXT:    s_wait_kmcnt 0x0
; GFX12W32-NEXT:    s_mul_i32 s4, s2, s4
; GFX12W32-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX12W32-NEXT:    v_mov_b32_e32 v1, s4
; GFX12W32-NEXT:    buffer_atomic_add_u32 v1, off, s[8:11], null th:TH_ATOMIC_RETURN
; GFX12W32-NEXT:  .LBB1_2:
; GFX12W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
; GFX12W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX12W32-NEXT:    s_wait_loadcnt 0x0
; GFX12W32-NEXT:    v_readfirstlane_b32 s4, v1
; GFX12W32-NEXT:    s_wait_kmcnt 0x0
; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12W32-NEXT:    v_mad_co_u64_u32 v[0:1], null, s2, v0, s[4:5]
; GFX12W32-NEXT:    v_mov_b32_e32 v1, 0
; GFX12W32-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12W32-NEXT:    s_nop 0
; GFX12W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX12W32-NEXT:    s_endpgm
entry:
  %old = call i32 @llvm.amdgcn.raw.ptr.buffer.atomic.add(i32 %additive, ptr addrspace(8) %inout, i32 0, i32 0, i32 0)
  store i32 %old, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @add_i32_varying_vdata(ptr addrspace(1) %out, ptr addrspace(8) %inout) {
; GFX6-LABEL: add_i32_varying_vdata:
; GFX6:       ; %bb.0: ; %entry
; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NEXT:    buffer_atomic_add v0, off, s[4:7], 0 glc
; GFX6-NEXT:    s_mov_b32 s3, 0xf000
; GFX6-NEXT:    s_mov_b32 s2, -1
; GFX6-NEXT:    s_waitcnt vmcnt(0)
; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT:    s_endpgm
;
; GFX8-LABEL: add_i32_varying_vdata:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_mov_b64 s[2:3], exec
; GFX8-NEXT:    s_mov_b32 s4, 0
; GFX8-NEXT:    ; implicit-def: $vgpr1
; GFX8-NEXT:  .LBB2_1: ; %ComputeLoop
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_ff1_i32_b64 s5, s[2:3]
; GFX8-NEXT:    s_mov_b32 m0, s5
; GFX8-NEXT:    v_readlane_b32 s8, v0, s5
; GFX8-NEXT:    s_lshl_b64 s[6:7], 1, s5
; GFX8-NEXT:    v_writelane_b32 v1, s4, m0
; GFX8-NEXT:    s_add_i32 s4, s4, s8
; GFX8-NEXT:    s_andn2_b64 s[2:3], s[2:3], s[6:7]
; GFX8-NEXT:    s_cmp_lg_u64 s[2:3], 0
; GFX8-NEXT:    s_cbranch_scc1 .LBB2_1
; GFX8-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX8-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX8-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX8-NEXT:    ; implicit-def: $vgpr0
; GFX8-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX8-NEXT:    s_xor_b64 s[2:3], exec, s[2:3]
; GFX8-NEXT:    s_cbranch_execz .LBB2_4
; GFX8-NEXT:  ; %bb.3:
; GFX8-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX8-NEXT:    v_mov_b32_e32 v0, s4
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    buffer_atomic_add v0, off, s[8:11], 0 glc
; GFX8-NEXT:  .LBB2_4:
; GFX8-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    v_readfirstlane_b32 s2, v0
; GFX8-NEXT:    v_add_u32_e32 v2, vcc, s2, v1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_store_dword v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-LABEL: add_i32_varying_vdata:
; GFX9:       ; %bb.0: ; %entry
; GFX9-NEXT:    s_mov_b64 s[2:3], exec
; GFX9-NEXT:    s_mov_b32 s4, 0
; GFX9-NEXT:    ; implicit-def: $vgpr1
; GFX9-NEXT:  .LBB2_1: ; %ComputeLoop
; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX9-NEXT:    s_ff1_i32_b64 s5, s[2:3]
; GFX9-NEXT:    s_mov_b32 m0, s5
; GFX9-NEXT:    v_readlane_b32 s8, v0, s5
; GFX9-NEXT:    s_lshl_b64 s[6:7], 1, s5
; GFX9-NEXT:    v_writelane_b32 v1, s4, m0
; GFX9-NEXT:    s_add_i32 s4, s4, s8
; GFX9-NEXT:    s_andn2_b64 s[2:3], s[2:3], s[6:7]
; GFX9-NEXT:    s_cmp_lg_u64 s[2:3], 0
; GFX9-NEXT:    s_cbranch_scc1 .LBB2_1
; GFX9-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX9-NEXT:    ; implicit-def: $vgpr0
; GFX9-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX9-NEXT:    s_xor_b64 s[2:3], exec, s[2:3]
; GFX9-NEXT:    s_cbranch_execz .LBB2_4
; GFX9-NEXT:  ; %bb.3:
; GFX9-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX9-NEXT:    v_mov_b32_e32 v0, s4
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    buffer_atomic_add v0, off, s[8:11], 0 glc
; GFX9-NEXT:  .LBB2_4:
; GFX9-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_readfirstlane_b32 s2, v0
; GFX9-NEXT:    v_mov_b32_e32 v2, 0
; GFX9-NEXT:    v_add_u32_e32 v0, s2, v1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_store_dword v2, v0, s[0:1]
; GFX9-NEXT:    s_endpgm
;
; GFX10W64-LABEL: add_i32_varying_vdata:
; GFX10W64:       ; %bb.0: ; %entry
; GFX10W64-NEXT:    s_mov_b64 s[2:3], exec
; GFX10W64-NEXT:    s_mov_b32 s4, 0
; GFX10W64-NEXT:    ; implicit-def: $vgpr1
; GFX10W64-NEXT:  .LBB2_1: ; %ComputeLoop
; GFX10W64-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX10W64-NEXT:    s_ff1_i32_b64 s5, s[2:3]
; GFX10W64-NEXT:    v_readlane_b32 s8, v0, s5
; GFX10W64-NEXT:    s_lshl_b64 s[6:7], 1, s5
; GFX10W64-NEXT:    v_writelane_b32 v1, s4, s5
; GFX10W64-NEXT:    s_andn2_b64 s[2:3], s[2:3], s[6:7]
; GFX10W64-NEXT:    s_add_i32 s4, s4, s8
; GFX10W64-NEXT:    s_cmp_lg_u64 s[2:3], 0
; GFX10W64-NEXT:    s_cbranch_scc1 .LBB2_1
; GFX10W64-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX10W64-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX10W64-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
; GFX10W64-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX10W64-NEXT:    ; implicit-def: $vgpr0
; GFX10W64-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX10W64-NEXT:    s_xor_b64 s[2:3], exec, s[2:3]
; GFX10W64-NEXT:    s_cbranch_execz .LBB2_4
; GFX10W64-NEXT:  ; %bb.3:
; GFX10W64-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W64-NEXT:    v_mov_b32_e32 v0, s4
; GFX10W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT:    buffer_atomic_add v0, off, s[8:11], 0 glc
; GFX10W64-NEXT:  .LBB2_4:
; GFX10W64-NEXT:    s_waitcnt_depctr 0xffe3
; GFX10W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX10W64-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W64-NEXT:    s_waitcnt vmcnt(0)
; GFX10W64-NEXT:    v_readfirstlane_b32 s2, v0
; GFX10W64-NEXT:    v_mov_b32_e32 v0, 0
; GFX10W64-NEXT:    v_add_nc_u32_e32 v1, s2, v1
; GFX10W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX10W64-NEXT:    s_endpgm
;
; GFX10W32-LABEL: add_i32_varying_vdata:
; GFX10W32:       ; %bb.0: ; %entry
; GFX10W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX10W32-NEXT:    s_mov_b32 s2, 0
; GFX10W32-NEXT:    ; implicit-def: $vgpr1
; GFX10W32-NEXT:  .LBB2_1: ; %ComputeLoop
; GFX10W32-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX10W32-NEXT:    s_ff1_i32_b32 s4, s3
; GFX10W32-NEXT:    v_readlane_b32 s5, v0, s4
; GFX10W32-NEXT:    s_lshl_b32 s6, 1, s4
; GFX10W32-NEXT:    v_writelane_b32 v1, s2, s4
; GFX10W32-NEXT:    s_andn2_b32 s3, s3, s6
; GFX10W32-NEXT:    s_add_i32 s2, s2, s5
; GFX10W32-NEXT:    s_cmp_lg_u32 s3, 0
; GFX10W32-NEXT:    s_cbranch_scc1 .LBB2_1
; GFX10W32-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX10W32-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX10W32-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX10W32-NEXT:    ; implicit-def: $vgpr0
; GFX10W32-NEXT:    s_and_saveexec_b32 s3, vcc_lo
; GFX10W32-NEXT:    s_xor_b32 s3, exec_lo, s3
; GFX10W32-NEXT:    s_cbranch_execz .LBB2_4
; GFX10W32-NEXT:  ; %bb.3:
; GFX10W32-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX10W32-NEXT:    v_mov_b32_e32 v0, s2
; GFX10W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT:    buffer_atomic_add v0, off, s[4:7], 0 glc
; GFX10W32-NEXT:  .LBB2_4:
; GFX10W32-NEXT:    s_waitcnt_depctr 0xffe3
; GFX10W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
; GFX10W32-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W32-NEXT:    s_waitcnt vmcnt(0)
; GFX10W32-NEXT:    v_readfirstlane_b32 s2, v0
; GFX10W32-NEXT:    v_mov_b32_e32 v0, 0
; GFX10W32-NEXT:    v_add_nc_u32_e32 v1, s2, v1
; GFX10W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX10W32-NEXT:    s_endpgm
;
; GFX11W64-LABEL: add_i32_varying_vdata:
; GFX11W64:       ; %bb.0: ; %entry
; GFX11W64-NEXT:    s_mov_b64 s[2:3], exec
; GFX11W64-NEXT:    s_mov_b32 s4, 0
; GFX11W64-NEXT:    ; implicit-def: $vgpr1
; GFX11W64-NEXT:  .LBB2_1: ; %ComputeLoop
; GFX11W64-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX11W64-NEXT:    s_ctz_i32_b64 s5, s[2:3]
; GFX11W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
; GFX11W64-NEXT:    v_readlane_b32 s8, v0, s5
; GFX11W64-NEXT:    s_lshl_b64 s[6:7], 1, s5
; GFX11W64-NEXT:    v_writelane_b32 v1, s4, s5
; GFX11W64-NEXT:    s_and_not1_b64 s[2:3], s[2:3], s[6:7]
; GFX11W64-NEXT:    s_add_i32 s4, s4, s8
; GFX11W64-NEXT:    s_cmp_lg_u64 s[2:3], 0
; GFX11W64-NEXT:    s_cbranch_scc1 .LBB2_1
; GFX11W64-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX11W64-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX11W64-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
; GFX11W64-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX11W64-NEXT:    ; implicit-def: $vgpr0
; GFX11W64-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX11W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W64-NEXT:    s_xor_b64 s[2:3], exec, s[2:3]
; GFX11W64-NEXT:    s_cbranch_execz .LBB2_4
; GFX11W64-NEXT:  ; %bb.3:
; GFX11W64-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W64-NEXT:    v_mov_b32_e32 v0, s4
; GFX11W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT:    buffer_atomic_add_u32 v0, off, s[8:11], 0 glc
; GFX11W64-NEXT:  .LBB2_4:
; GFX11W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX11W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W64-NEXT:    s_waitcnt vmcnt(0)
; GFX11W64-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11W64-NEXT:    v_mov_b32_e32 v0, 0
; GFX11W64-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11W64-NEXT:    v_add_nc_u32_e32 v1, s2, v1
; GFX11W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11W64-NEXT:    s_nop 0
; GFX11W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W64-NEXT:    s_endpgm
;
; GFX11W32-LABEL: add_i32_varying_vdata:
; GFX11W32:       ; %bb.0: ; %entry
; GFX11W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX11W32-NEXT:    s_mov_b32 s2, 0
; GFX11W32-NEXT:    ; implicit-def: $vgpr1
; GFX11W32-NEXT:  .LBB2_1: ; %ComputeLoop
; GFX11W32-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX11W32-NEXT:    s_ctz_i32_b32 s4, s3
; GFX11W32-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
; GFX11W32-NEXT:    v_readlane_b32 s5, v0, s4
; GFX11W32-NEXT:    s_lshl_b32 s6, 1, s4
; GFX11W32-NEXT:    v_writelane_b32 v1, s2, s4
; GFX11W32-NEXT:    s_and_not1_b32 s3, s3, s6
; GFX11W32-NEXT:    s_add_i32 s2, s2, s5
; GFX11W32-NEXT:    s_cmp_lg_u32 s3, 0
; GFX11W32-NEXT:    s_cbranch_scc1 .LBB2_1
; GFX11W32-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX11W32-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX11W32-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX11W32-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX11W32-NEXT:    ; implicit-def: $vgpr0
; GFX11W32-NEXT:    s_and_saveexec_b32 s3, vcc_lo
; GFX11W32-NEXT:    s_xor_b32 s3, exec_lo, s3
; GFX11W32-NEXT:    s_cbranch_execz .LBB2_4
; GFX11W32-NEXT:  ; %bb.3:
; GFX11W32-NEXT:    s_load_b128 s[4:7], s[0:1], 0x34
; GFX11W32-NEXT:    v_mov_b32_e32 v0, s2
; GFX11W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT:    buffer_atomic_add_u32 v0, off, s[4:7], 0 glc
; GFX11W32-NEXT:  .LBB2_4:
; GFX11W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
; GFX11W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W32-NEXT:    s_waitcnt vmcnt(0)
; GFX11W32-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11W32-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_add_nc_u32 v1, s2, v1
; GFX11W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11W32-NEXT:    s_nop 0
; GFX11W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W32-NEXT:    s_endpgm
;
; GFX12W64-LABEL: add_i32_varying_vdata:
; GFX12W64:       ; %bb.0: ; %entry
; GFX12W64-NEXT:    s_mov_b64 s[2:3], exec
; GFX12W64-NEXT:    s_mov_b32 s4, 0
; GFX12W64-NEXT:    ; implicit-def: $vgpr1
; GFX12W64-NEXT:  .LBB2_1: ; %ComputeLoop
; GFX12W64-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX12W64-NEXT:    s_ctz_i32_b64 s5, s[2:3]
; GFX12W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
; GFX12W64-NEXT:    v_readlane_b32 s8, v0, s5
; GFX12W64-NEXT:    s_lshl_b64 s[6:7], 1, s5
; GFX12W64-NEXT:    v_writelane_b32 v1, s4, s5
; GFX12W64-NEXT:    s_and_not1_b64 s[2:3], s[2:3], s[6:7]
; GFX12W64-NEXT:    s_add_co_i32 s4, s4, s8
; GFX12W64-NEXT:    s_cmp_lg_u64 s[2:3], 0
; GFX12W64-NEXT:    s_cbranch_scc1 .LBB2_1
; GFX12W64-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX12W64-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12W64-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
; GFX12W64-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX12W64-NEXT:    ; implicit-def: $vgpr0
; GFX12W64-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX12W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX12W64-NEXT:    s_xor_b64 s[2:3], exec, s[2:3]
; GFX12W64-NEXT:    s_cbranch_execz .LBB2_4
; GFX12W64-NEXT:  ; %bb.3:
; GFX12W64-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX12W64-NEXT:    v_mov_b32_e32 v0, s4
; GFX12W64-NEXT:    s_wait_kmcnt 0x0
; GFX12W64-NEXT:    buffer_atomic_add_u32 v0, off, s[8:11], null th:TH_ATOMIC_RETURN
; GFX12W64-NEXT:  .LBB2_4:
; GFX12W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX12W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX12W64-NEXT:    s_wait_loadcnt 0x0
; GFX12W64-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12W64-NEXT:    v_mov_b32_e32 v0, 0
; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12W64-NEXT:    v_add_nc_u32_e32 v1, s2, v1
; GFX12W64-NEXT:    s_wait_kmcnt 0x0
; GFX12W64-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12W64-NEXT:    s_nop 0
; GFX12W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX12W64-NEXT:    s_endpgm
;
; GFX12W32-LABEL: add_i32_varying_vdata:
; GFX12W32:       ; %bb.0: ; %entry
; GFX12W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX12W32-NEXT:    s_mov_b32 s2, 0
; GFX12W32-NEXT:    ; implicit-def: $vgpr1
; GFX12W32-NEXT:  .LBB2_1: ; %ComputeLoop
; GFX12W32-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX12W32-NEXT:    s_ctz_i32_b32 s4, s3
; GFX12W32-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
; GFX12W32-NEXT:    v_readlane_b32 s5, v0, s4
; GFX12W32-NEXT:    s_lshl_b32 s6, 1, s4
; GFX12W32-NEXT:    v_writelane_b32 v1, s2, s4
; GFX12W32-NEXT:    s_and_not1_b32 s3, s3, s6
; GFX12W32-NEXT:    s_add_co_i32 s2, s2, s5
; GFX12W32-NEXT:    s_cmp_lg_u32 s3, 0
; GFX12W32-NEXT:    s_cbranch_scc1 .LBB2_1
; GFX12W32-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX12W32-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX12W32-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX12W32-NEXT:    ; implicit-def: $vgpr0
; GFX12W32-NEXT:    s_and_saveexec_b32 s3, vcc_lo
; GFX12W32-NEXT:    s_xor_b32 s3, exec_lo, s3
; GFX12W32-NEXT:    s_cbranch_execz .LBB2_4
; GFX12W32-NEXT:  ; %bb.3:
; GFX12W32-NEXT:    s_load_b128 s[4:7], s[0:1], 0x34
; GFX12W32-NEXT:    v_mov_b32_e32 v0, s2
; GFX12W32-NEXT:    s_wait_kmcnt 0x0
; GFX12W32-NEXT:    buffer_atomic_add_u32 v0, off, s[4:7], null th:TH_ATOMIC_RETURN
; GFX12W32-NEXT:  .LBB2_4:
; GFX12W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
; GFX12W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX12W32-NEXT:    s_wait_loadcnt 0x0
; GFX12W32-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12W32-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_add_nc_u32 v1, s2, v1
; GFX12W32-NEXT:    s_wait_kmcnt 0x0
; GFX12W32-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12W32-NEXT:    s_nop 0
; GFX12W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX12W32-NEXT:    s_endpgm
entry:
  %lane = call i32 @llvm.amdgcn.workitem.id.x()
  %old = call i32 @llvm.amdgcn.raw.ptr.buffer.atomic.add(i32 %lane, ptr addrspace(8) %inout, i32 0, i32 0, i32 0)
  store i32 %old, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @add_i32_varying_offset(ptr addrspace(1) %out, ptr addrspace(8) %inout) {
; GFX6-LABEL: add_i32_varying_offset:
; GFX6:       ; %bb.0: ; %entry
; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT:    v_mov_b32_e32 v1, 1
; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NEXT:    buffer_atomic_add v1, v0, s[4:7], 0 offen glc
; GFX6-NEXT:    s_mov_b32 s3, 0xf000
; GFX6-NEXT:    s_mov_b32 s2, -1
; GFX6-NEXT:    s_waitcnt vmcnt(0)
; GFX6-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; GFX6-NEXT:    s_endpgm
;
; GFX8-LABEL: add_i32_varying_offset:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX8-NEXT:    v_mov_b32_e32 v2, 1
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    buffer_atomic_add v2, v0, s[4:7], 0 offen glc
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    flat_store_dword v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-LABEL: add_i32_varying_offset:
; GFX9:       ; %bb.0: ; %entry
; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX9-NEXT:    v_mov_b32_e32 v1, 1
; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    buffer_atomic_add v1, v0, s[4:7], 0 offen glc
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX9-NEXT:    s_endpgm
;
; GFX10-LABEL: add_i32_varying_offset:
; GFX10:       ; %bb.0: ; %entry
; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    buffer_atomic_add v1, v0, s[4:7], 0 offen glc
; GFX10-NEXT:    v_mov_b32_e32 v0, 0
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: add_i32_varying_offset:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x34
; GFX11-NEXT:    v_mov_b32_e32 v1, 1
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    buffer_atomic_add_u32 v1, v0, s[4:7], 0 offen glc
; GFX11-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: add_i32_varying_offset:
; GFX12:       ; %bb.0: ; %entry
; GFX12-NEXT:    s_load_b128 s[4:7], s[0:1], 0x34
; GFX12-NEXT:    v_mov_b32_e32 v1, 1
; GFX12-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    buffer_atomic_add_u32 v1, v0, s[4:7], null offen th:TH_ATOMIC_RETURN
; GFX12-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-NEXT:    s_wait_loadcnt 0x0
; GFX12-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-NEXT:    s_nop 0
; GFX12-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX12-NEXT:    s_endpgm
entry:
  %lane = call i32 @llvm.amdgcn.workitem.id.x()
  %old = call i32 @llvm.amdgcn.raw.ptr.buffer.atomic.add(i32 1, ptr addrspace(8) %inout, i32 %lane, i32 0, i32 0)
  store i32 %old, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @sub_i32_constant(ptr addrspace(1) %out, ptr addrspace(8) %inout) {
; GFX6-LABEL: sub_i32_constant:
; GFX6:       ; %bb.0: ; %entry
; GFX6-NEXT:    s_mov_b64 s[4:5], exec
; GFX6-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, s4, 0
; GFX6-NEXT:    v_mbcnt_hi_u32_b32_e32 v0, s5, v0
; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX6-NEXT:    ; implicit-def: $vgpr1
; GFX6-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX6-NEXT:    s_cbranch_execz .LBB4_2
; GFX6-NEXT:  ; %bb.1:
; GFX6-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0xd
; GFX6-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX6-NEXT:    s_mul_i32 s4, s4, 5
; GFX6-NEXT:    v_mov_b32_e32 v1, s4
; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NEXT:    buffer_atomic_sub v1, off, s[8:11], 0 glc
; GFX6-NEXT:  .LBB4_2:
; GFX6-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT:    s_mov_b32 s3, 0xf000
; GFX6-NEXT:    s_mov_b32 s2, -1
; GFX6-NEXT:    s_waitcnt vmcnt(0)
; GFX6-NEXT:    v_readfirstlane_b32 s4, v1
; GFX6-NEXT:    v_mul_u32_u24_e32 v0, 5, v0
; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT:    s_endpgm
;
; GFX8-LABEL: sub_i32_constant:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_mov_b64 s[4:5], exec
; GFX8-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX8-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX8-NEXT:    ; implicit-def: $vgpr1
; GFX8-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX8-NEXT:    s_cbranch_execz .LBB4_2
; GFX8-NEXT:  ; %bb.1:
; GFX8-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX8-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX8-NEXT:    s_mul_i32 s4, s4, 5
; GFX8-NEXT:    v_mov_b32_e32 v1, s4
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    buffer_atomic_sub v1, off, s[8:11], 0 glc
; GFX8-NEXT:  .LBB4_2:
; GFX8-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    v_readfirstlane_b32 s2, v1
; GFX8-NEXT:    v_mul_u32_u24_e32 v0, 5, v0
; GFX8-NEXT:    v_sub_u32_e32 v2, vcc, s2, v0
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_store_dword v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-LABEL: sub_i32_constant:
; GFX9:       ; %bb.0: ; %entry
; GFX9-NEXT:    s_mov_b64 s[4:5], exec
; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX9-NEXT:    ; implicit-def: $vgpr1
; GFX9-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX9-NEXT:    s_cbranch_execz .LBB4_2
; GFX9-NEXT:  ; %bb.1:
; GFX9-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX9-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX9-NEXT:    s_mul_i32 s4, s4, 5
; GFX9-NEXT:    v_mov_b32_e32 v1, s4
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    buffer_atomic_sub v1, off, s[8:11], 0 glc
; GFX9-NEXT:  .LBB4_2:
; GFX9-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_readfirstlane_b32 s2, v1
; GFX9-NEXT:    v_mul_u32_u24_e32 v0, 5, v0
; GFX9-NEXT:    v_mov_b32_e32 v2, 0
; GFX9-NEXT:    v_sub_u32_e32 v0, s2, v0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_store_dword v2, v0, s[0:1]
; GFX9-NEXT:    s_endpgm
;
; GFX10W64-LABEL: sub_i32_constant:
; GFX10W64:       ; %bb.0: ; %entry
; GFX10W64-NEXT:    s_mov_b64 s[4:5], exec
; GFX10W64-NEXT:    ; implicit-def: $vgpr1
; GFX10W64-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX10W64-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX10W64-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX10W64-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX10W64-NEXT:    s_cbranch_execz .LBB4_2
; GFX10W64-NEXT:  ; %bb.1:
; GFX10W64-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W64-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX10W64-NEXT:    s_mul_i32 s4, s4, 5
; GFX10W64-NEXT:    v_mov_b32_e32 v1, s4
; GFX10W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT:    buffer_atomic_sub v1, off, s[8:11], 0 glc
; GFX10W64-NEXT:  .LBB4_2:
; GFX10W64-NEXT:    s_waitcnt_depctr 0xffe3
; GFX10W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX10W64-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W64-NEXT:    s_waitcnt vmcnt(0)
; GFX10W64-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10W64-NEXT:    v_mul_u32_u24_e32 v0, 5, v0
; GFX10W64-NEXT:    v_mov_b32_e32 v1, 0
; GFX10W64-NEXT:    v_sub_nc_u32_e32 v0, s2, v0
; GFX10W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10W64-NEXT:    s_endpgm
;
; GFX10W32-LABEL: sub_i32_constant:
; GFX10W32:       ; %bb.0: ; %entry
; GFX10W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX10W32-NEXT:    ; implicit-def: $vgpr1
; GFX10W32-NEXT:    v_mbcnt_lo_u32_b32 v0, s3, 0
; GFX10W32-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX10W32-NEXT:    s_and_saveexec_b32 s2, vcc_lo
; GFX10W32-NEXT:    s_cbranch_execz .LBB4_2
; GFX10W32-NEXT:  ; %bb.1:
; GFX10W32-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX10W32-NEXT:    s_bcnt1_i32_b32 s3, s3
; GFX10W32-NEXT:    s_mul_i32 s3, s3, 5
; GFX10W32-NEXT:    v_mov_b32_e32 v1, s3
; GFX10W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT:    buffer_atomic_sub v1, off, s[4:7], 0 glc
; GFX10W32-NEXT:  .LBB4_2:
; GFX10W32-NEXT:    s_waitcnt_depctr 0xffe3
; GFX10W32-NEXT:    s_or_b32 exec_lo, exec_lo, s2
; GFX10W32-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W32-NEXT:    s_waitcnt vmcnt(0)
; GFX10W32-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10W32-NEXT:    v_mul_u32_u24_e32 v0, 5, v0
; GFX10W32-NEXT:    v_mov_b32_e32 v1, 0
; GFX10W32-NEXT:    v_sub_nc_u32_e32 v0, s2, v0
; GFX10W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10W32-NEXT:    s_endpgm
;
; GFX11W64-LABEL: sub_i32_constant:
; GFX11W64:       ; %bb.0: ; %entry
; GFX11W64-NEXT:    s_mov_b64 s[4:5], exec
; GFX11W64-NEXT:    s_mov_b64 s[2:3], exec
; GFX11W64-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX11W64-NEXT:    ; implicit-def: $vgpr1
; GFX11W64-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX11W64-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX11W64-NEXT:    s_cbranch_execz .LBB4_2
; GFX11W64-NEXT:  ; %bb.1:
; GFX11W64-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W64-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX11W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11W64-NEXT:    s_mul_i32 s4, s4, 5
; GFX11W64-NEXT:    v_mov_b32_e32 v1, s4
; GFX11W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT:    buffer_atomic_sub_u32 v1, off, s[8:11], 0 glc
; GFX11W64-NEXT:  .LBB4_2:
; GFX11W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX11W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W64-NEXT:    s_waitcnt vmcnt(0)
; GFX11W64-NEXT:    v_readfirstlane_b32 s2, v1
; GFX11W64-NEXT:    v_mul_u32_u24_e32 v0, 5, v0
; GFX11W64-NEXT:    v_mov_b32_e32 v1, 0
; GFX11W64-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11W64-NEXT:    v_sub_nc_u32_e32 v0, s2, v0
; GFX11W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11W64-NEXT:    s_nop 0
; GFX11W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W64-NEXT:    s_endpgm
;
; GFX11W32-LABEL: sub_i32_constant:
; GFX11W32:       ; %bb.0: ; %entry
; GFX11W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX11W32-NEXT:    s_mov_b32 s2, exec_lo
; GFX11W32-NEXT:    v_mbcnt_lo_u32_b32 v0, s3, 0
; GFX11W32-NEXT:    ; implicit-def: $vgpr1
; GFX11W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11W32-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX11W32-NEXT:    s_cbranch_execz .LBB4_2
; GFX11W32-NEXT:  ; %bb.1:
; GFX11W32-NEXT:    s_load_b128 s[4:7], s[0:1], 0x34
; GFX11W32-NEXT:    s_bcnt1_i32_b32 s3, s3
; GFX11W32-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11W32-NEXT:    s_mul_i32 s3, s3, 5
; GFX11W32-NEXT:    v_mov_b32_e32 v1, s3
; GFX11W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT:    buffer_atomic_sub_u32 v1, off, s[4:7], 0 glc
; GFX11W32-NEXT:  .LBB4_2:
; GFX11W32-NEXT:    s_or_b32 exec_lo, exec_lo, s2
; GFX11W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W32-NEXT:    s_waitcnt vmcnt(0)
; GFX11W32-NEXT:    v_readfirstlane_b32 s2, v1
; GFX11W32-NEXT:    v_mul_u32_u24_e32 v0, 5, v0
; GFX11W32-NEXT:    v_mov_b32_e32 v1, 0
; GFX11W32-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11W32-NEXT:    v_sub_nc_u32_e32 v0, s2, v0
; GFX11W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11W32-NEXT:    s_nop 0
; GFX11W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W32-NEXT:    s_endpgm
;
; GFX12W64-LABEL: sub_i32_constant:
; GFX12W64:       ; %bb.0: ; %entry
; GFX12W64-NEXT:    s_mov_b64 s[4:5], exec
; GFX12W64-NEXT:    s_mov_b64 s[2:3], exec
; GFX12W64-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX12W64-NEXT:    ; implicit-def: $vgpr1
; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12W64-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX12W64-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX12W64-NEXT:    s_cbranch_execz .LBB4_2
; GFX12W64-NEXT:  ; %bb.1:
; GFX12W64-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX12W64-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX12W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX12W64-NEXT:    s_mul_i32 s4, s4, 5
; GFX12W64-NEXT:    v_mov_b32_e32 v1, s4
; GFX12W64-NEXT:    s_wait_kmcnt 0x0
; GFX12W64-NEXT:    buffer_atomic_sub_u32 v1, off, s[8:11], null th:TH_ATOMIC_RETURN
; GFX12W64-NEXT:  .LBB4_2:
; GFX12W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX12W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX12W64-NEXT:    s_wait_loadcnt 0x0
; GFX12W64-NEXT:    v_readfirstlane_b32 s2, v1
; GFX12W64-NEXT:    v_mul_u32_u24_e32 v0, 5, v0
; GFX12W64-NEXT:    v_mov_b32_e32 v1, 0
; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12W64-NEXT:    v_sub_nc_u32_e32 v0, s2, v0
; GFX12W64-NEXT:    s_wait_kmcnt 0x0
; GFX12W64-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12W64-NEXT:    s_nop 0
; GFX12W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX12W64-NEXT:    s_endpgm
;
; GFX12W32-LABEL: sub_i32_constant:
; GFX12W32:       ; %bb.0: ; %entry
; GFX12W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX12W32-NEXT:    s_mov_b32 s2, exec_lo
; GFX12W32-NEXT:    v_mbcnt_lo_u32_b32 v0, s3, 0
; GFX12W32-NEXT:    ; implicit-def: $vgpr1
; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12W32-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX12W32-NEXT:    s_cbranch_execz .LBB4_2
; GFX12W32-NEXT:  ; %bb.1:
; GFX12W32-NEXT:    s_load_b128 s[4:7], s[0:1], 0x34
; GFX12W32-NEXT:    s_bcnt1_i32_b32 s3, s3
; GFX12W32-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX12W32-NEXT:    s_mul_i32 s3, s3, 5
; GFX12W32-NEXT:    v_mov_b32_e32 v1, s3
; GFX12W32-NEXT:    s_wait_kmcnt 0x0
; GFX12W32-NEXT:    buffer_atomic_sub_u32 v1, off, s[4:7], null th:TH_ATOMIC_RETURN
; GFX12W32-NEXT:  .LBB4_2:
; GFX12W32-NEXT:    s_or_b32 exec_lo, exec_lo, s2
; GFX12W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX12W32-NEXT:    s_wait_loadcnt 0x0
; GFX12W32-NEXT:    v_readfirstlane_b32 s2, v1
; GFX12W32-NEXT:    v_mul_u32_u24_e32 v0, 5, v0
; GFX12W32-NEXT:    v_mov_b32_e32 v1, 0
; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12W32-NEXT:    v_sub_nc_u32_e32 v0, s2, v0
; GFX12W32-NEXT:    s_wait_kmcnt 0x0
; GFX12W32-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12W32-NEXT:    s_nop 0
; GFX12W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX12W32-NEXT:    s_endpgm
entry:
  %old = call i32 @llvm.amdgcn.raw.ptr.buffer.atomic.sub(i32 5, ptr addrspace(8) %inout, i32 0, i32 0, i32 0)
  store i32 %old, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @sub_i32_uniform(ptr addrspace(1) %out, ptr addrspace(8) %inout, i32 %subitive) {
; GFX6-LABEL: sub_i32_uniform:
; GFX6:       ; %bb.0: ; %entry
; GFX6-NEXT:    s_mov_b64 s[4:5], exec
; GFX6-NEXT:    s_load_dword s6, s[0:1], 0x11
; GFX6-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, s4, 0
; GFX6-NEXT:    v_mbcnt_hi_u32_b32_e32 v0, s5, v0
; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX6-NEXT:    ; implicit-def: $vgpr1
; GFX6-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX6-NEXT:    s_cbranch_execz .LBB5_2
; GFX6-NEXT:  ; %bb.1:
; GFX6-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0xd
; GFX6-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NEXT:    s_mul_i32 s4, s6, s4
; GFX6-NEXT:    v_mov_b32_e32 v1, s4
; GFX6-NEXT:    buffer_atomic_sub v1, off, s[8:11], 0 glc
; GFX6-NEXT:  .LBB5_2:
; GFX6-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT:    s_mov_b32 s3, 0xf000
; GFX6-NEXT:    s_mov_b32 s2, -1
; GFX6-NEXT:    s_waitcnt vmcnt(0)
; GFX6-NEXT:    v_readfirstlane_b32 s4, v1
; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NEXT:    v_mul_lo_u32 v0, s6, v0
; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT:    s_endpgm
;
; GFX8-LABEL: sub_i32_uniform:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dword s6, s[0:1], 0x44
; GFX8-NEXT:    s_mov_b64 s[4:5], exec
; GFX8-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX8-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX8-NEXT:    ; implicit-def: $vgpr1
; GFX8-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX8-NEXT:    s_cbranch_execz .LBB5_2
; GFX8-NEXT:  ; %bb.1:
; GFX8-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX8-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_mul_i32 s4, s6, s4
; GFX8-NEXT:    v_mov_b32_e32 v1, s4
; GFX8-NEXT:    buffer_atomic_sub v1, off, s[8:11], 0 glc
; GFX8-NEXT:  .LBB5_2:
; GFX8-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mul_lo_u32 v0, s6, v0
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    v_readfirstlane_b32 s2, v1
; GFX8-NEXT:    v_sub_u32_e32 v2, vcc, s2, v0
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_store_dword v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-LABEL: sub_i32_uniform:
; GFX9:       ; %bb.0: ; %entry
; GFX9-NEXT:    s_load_dword s6, s[0:1], 0x44
; GFX9-NEXT:    s_mov_b64 s[4:5], exec
; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX9-NEXT:    ; implicit-def: $vgpr1
; GFX9-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX9-NEXT:    s_cbranch_execz .LBB5_2
; GFX9-NEXT:  ; %bb.1:
; GFX9-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX9-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_mul_i32 s4, s6, s4
; GFX9-NEXT:    v_mov_b32_e32 v1, s4
; GFX9-NEXT:    buffer_atomic_sub v1, off, s[8:11], 0 glc
; GFX9-NEXT:  .LBB5_2:
; GFX9-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_mul_lo_u32 v0, s6, v0
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_readfirstlane_b32 s2, v1
; GFX9-NEXT:    v_mov_b32_e32 v2, 0
; GFX9-NEXT:    v_sub_u32_e32 v0, s2, v0
; GFX9-NEXT:    global_store_dword v2, v0, s[0:1]
; GFX9-NEXT:    s_endpgm
;
; GFX10W64-LABEL: sub_i32_uniform:
; GFX10W64:       ; %bb.0: ; %entry
; GFX10W64-NEXT:    s_load_dword s6, s[0:1], 0x44
; GFX10W64-NEXT:    s_mov_b64 s[4:5], exec
; GFX10W64-NEXT:    ; implicit-def: $vgpr1
; GFX10W64-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX10W64-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX10W64-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX10W64-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX10W64-NEXT:    s_cbranch_execz .LBB5_2
; GFX10W64-NEXT:  ; %bb.1:
; GFX10W64-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W64-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX10W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT:    s_mul_i32 s4, s6, s4
; GFX10W64-NEXT:    v_mov_b32_e32 v1, s4
; GFX10W64-NEXT:    buffer_atomic_sub v1, off, s[8:11], 0 glc
; GFX10W64-NEXT:  .LBB5_2:
; GFX10W64-NEXT:    s_waitcnt_depctr 0xffe3
; GFX10W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX10W64-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT:    v_mul_lo_u32 v0, s6, v0
; GFX10W64-NEXT:    s_waitcnt vmcnt(0)
; GFX10W64-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10W64-NEXT:    v_mov_b32_e32 v1, 0
; GFX10W64-NEXT:    v_sub_nc_u32_e32 v0, s2, v0
; GFX10W64-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10W64-NEXT:    s_endpgm
;
; GFX10W32-LABEL: sub_i32_uniform:
; GFX10W32:       ; %bb.0: ; %entry
; GFX10W32-NEXT:    s_load_dword s2, s[0:1], 0x44
; GFX10W32-NEXT:    s_mov_b32 s4, exec_lo
; GFX10W32-NEXT:    ; implicit-def: $vgpr1
; GFX10W32-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX10W32-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX10W32-NEXT:    s_and_saveexec_b32 s3, vcc_lo
; GFX10W32-NEXT:    s_cbranch_execz .LBB5_2
; GFX10W32-NEXT:  ; %bb.1:
; GFX10W32-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W32-NEXT:    s_bcnt1_i32_b32 s4, s4
; GFX10W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT:    s_mul_i32 s4, s2, s4
; GFX10W32-NEXT:    v_mov_b32_e32 v1, s4
; GFX10W32-NEXT:    buffer_atomic_sub v1, off, s[8:11], 0 glc
; GFX10W32-NEXT:  .LBB5_2:
; GFX10W32-NEXT:    s_waitcnt_depctr 0xffe3
; GFX10W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
; GFX10W32-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT:    v_mul_lo_u32 v0, s2, v0
; GFX10W32-NEXT:    s_waitcnt vmcnt(0)
; GFX10W32-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10W32-NEXT:    v_mov_b32_e32 v1, 0
; GFX10W32-NEXT:    v_sub_nc_u32_e32 v0, s2, v0
; GFX10W32-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10W32-NEXT:    s_endpgm
;
; GFX11W64-LABEL: sub_i32_uniform:
; GFX11W64:       ; %bb.0: ; %entry
; GFX11W64-NEXT:    s_load_b32 s6, s[0:1], 0x44
; GFX11W64-NEXT:    s_mov_b64 s[4:5], exec
; GFX11W64-NEXT:    s_mov_b64 s[2:3], exec
; GFX11W64-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX11W64-NEXT:    ; implicit-def: $vgpr1
; GFX11W64-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX11W64-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX11W64-NEXT:    s_cbranch_execz .LBB5_2
; GFX11W64-NEXT:  ; %bb.1:
; GFX11W64-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W64-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX11W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT:    s_mul_i32 s4, s6, s4
; GFX11W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W64-NEXT:    v_mov_b32_e32 v1, s4
; GFX11W64-NEXT:    buffer_atomic_sub_u32 v1, off, s[8:11], 0 glc
; GFX11W64-NEXT:  .LBB5_2:
; GFX11W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX11W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT:    v_mul_lo_u32 v0, s6, v0
; GFX11W64-NEXT:    s_waitcnt vmcnt(0)
; GFX11W64-NEXT:    v_readfirstlane_b32 s2, v1
; GFX11W64-NEXT:    v_mov_b32_e32 v1, 0
; GFX11W64-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11W64-NEXT:    v_sub_nc_u32_e32 v0, s2, v0
; GFX11W64-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11W64-NEXT:    s_nop 0
; GFX11W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W64-NEXT:    s_endpgm
;
; GFX11W32-LABEL: sub_i32_uniform:
; GFX11W32:       ; %bb.0: ; %entry
; GFX11W32-NEXT:    s_load_b32 s2, s[0:1], 0x44
; GFX11W32-NEXT:    s_mov_b32 s4, exec_lo
; GFX11W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX11W32-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX11W32-NEXT:    ; implicit-def: $vgpr1
; GFX11W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11W32-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX11W32-NEXT:    s_cbranch_execz .LBB5_2
; GFX11W32-NEXT:  ; %bb.1:
; GFX11W32-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W32-NEXT:    s_bcnt1_i32_b32 s4, s4
; GFX11W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT:    s_mul_i32 s4, s2, s4
; GFX11W32-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W32-NEXT:    v_mov_b32_e32 v1, s4
; GFX11W32-NEXT:    buffer_atomic_sub_u32 v1, off, s[8:11], 0 glc
; GFX11W32-NEXT:  .LBB5_2:
; GFX11W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
; GFX11W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT:    v_mul_lo_u32 v0, s2, v0
; GFX11W32-NEXT:    s_waitcnt vmcnt(0)
; GFX11W32-NEXT:    v_readfirstlane_b32 s2, v1
; GFX11W32-NEXT:    v_mov_b32_e32 v1, 0
; GFX11W32-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11W32-NEXT:    v_sub_nc_u32_e32 v0, s2, v0
; GFX11W32-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11W32-NEXT:    s_nop 0
; GFX11W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W32-NEXT:    s_endpgm
;
; GFX12W64-LABEL: sub_i32_uniform:
; GFX12W64:       ; %bb.0: ; %entry
; GFX12W64-NEXT:    s_load_b32 s6, s[0:1], 0x44
; GFX12W64-NEXT:    s_mov_b64 s[4:5], exec
; GFX12W64-NEXT:    s_mov_b64 s[2:3], exec
; GFX12W64-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX12W64-NEXT:    ; implicit-def: $vgpr1
; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12W64-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX12W64-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX12W64-NEXT:    s_cbranch_execz .LBB5_2
; GFX12W64-NEXT:  ; %bb.1:
; GFX12W64-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX12W64-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX12W64-NEXT:    s_wait_kmcnt 0x0
; GFX12W64-NEXT:    s_mul_i32 s4, s6, s4
; GFX12W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX12W64-NEXT:    v_mov_b32_e32 v1, s4
; GFX12W64-NEXT:    buffer_atomic_sub_u32 v1, off, s[8:11], null th:TH_ATOMIC_RETURN
; GFX12W64-NEXT:  .LBB5_2:
; GFX12W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX12W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX12W64-NEXT:    s_wait_kmcnt 0x0
; GFX12W64-NEXT:    v_mul_lo_u32 v0, s6, v0
; GFX12W64-NEXT:    s_wait_loadcnt 0x0
; GFX12W64-NEXT:    v_readfirstlane_b32 s2, v1
; GFX12W64-NEXT:    v_mov_b32_e32 v1, 0
; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12W64-NEXT:    v_sub_nc_u32_e32 v0, s2, v0
; GFX12W64-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12W64-NEXT:    s_nop 0
; GFX12W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX12W64-NEXT:    s_endpgm
;
; GFX12W32-LABEL: sub_i32_uniform:
; GFX12W32:       ; %bb.0: ; %entry
; GFX12W32-NEXT:    s_load_b32 s2, s[0:1], 0x44
; GFX12W32-NEXT:    s_mov_b32 s4, exec_lo
; GFX12W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX12W32-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX12W32-NEXT:    ; implicit-def: $vgpr1
; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12W32-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX12W32-NEXT:    s_cbranch_execz .LBB5_2
; GFX12W32-NEXT:  ; %bb.1:
; GFX12W32-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX12W32-NEXT:    s_bcnt1_i32_b32 s4, s4
; GFX12W32-NEXT:    s_wait_kmcnt 0x0
; GFX12W32-NEXT:    s_mul_i32 s4, s2, s4
; GFX12W32-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX12W32-NEXT:    v_mov_b32_e32 v1, s4
; GFX12W32-NEXT:    buffer_atomic_sub_u32 v1, off, s[8:11], null th:TH_ATOMIC_RETURN
; GFX12W32-NEXT:  .LBB5_2:
; GFX12W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
; GFX12W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX12W32-NEXT:    s_wait_kmcnt 0x0
; GFX12W32-NEXT:    v_mul_lo_u32 v0, s2, v0
; GFX12W32-NEXT:    s_wait_loadcnt 0x0
; GFX12W32-NEXT:    v_readfirstlane_b32 s2, v1
; GFX12W32-NEXT:    v_mov_b32_e32 v1, 0
; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12W32-NEXT:    v_sub_nc_u32_e32 v0, s2, v0
; GFX12W32-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12W32-NEXT:    s_nop 0
; GFX12W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX12W32-NEXT:    s_endpgm
entry:
  %old = call i32 @llvm.amdgcn.raw.ptr.buffer.atomic.sub(i32 %subitive, ptr addrspace(8) %inout, i32 0, i32 0, i32 0)
  store i32 %old, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @sub_i32_varying_vdata(ptr addrspace(1) %out, ptr addrspace(8) %inout) {
; GFX6-LABEL: sub_i32_varying_vdata:
; GFX6:       ; %bb.0: ; %entry
; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NEXT:    buffer_atomic_sub v0, off, s[4:7], 0 glc
; GFX6-NEXT:    s_mov_b32 s3, 0xf000
; GFX6-NEXT:    s_mov_b32 s2, -1
; GFX6-NEXT:    s_waitcnt vmcnt(0)
; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT:    s_endpgm
;
; GFX8-LABEL: sub_i32_varying_vdata:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_mov_b64 s[2:3], exec
; GFX8-NEXT:    s_mov_b32 s4, 0
; GFX8-NEXT:    ; implicit-def: $vgpr1
; GFX8-NEXT:  .LBB6_1: ; %ComputeLoop
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_ff1_i32_b64 s5, s[2:3]
; GFX8-NEXT:    s_mov_b32 m0, s5
; GFX8-NEXT:    v_readlane_b32 s8, v0, s5
; GFX8-NEXT:    s_lshl_b64 s[6:7], 1, s5
; GFX8-NEXT:    v_writelane_b32 v1, s4, m0
; GFX8-NEXT:    s_add_i32 s4, s4, s8
; GFX8-NEXT:    s_andn2_b64 s[2:3], s[2:3], s[6:7]
; GFX8-NEXT:    s_cmp_lg_u64 s[2:3], 0
; GFX8-NEXT:    s_cbranch_scc1 .LBB6_1
; GFX8-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX8-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX8-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX8-NEXT:    ; implicit-def: $vgpr0
; GFX8-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX8-NEXT:    s_xor_b64 s[2:3], exec, s[2:3]
; GFX8-NEXT:    s_cbranch_execz .LBB6_4
; GFX8-NEXT:  ; %bb.3:
; GFX8-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX8-NEXT:    v_mov_b32_e32 v0, s4
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    buffer_atomic_sub v0, off, s[8:11], 0 glc
; GFX8-NEXT:  .LBB6_4:
; GFX8-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    v_readfirstlane_b32 s2, v0
; GFX8-NEXT:    v_sub_u32_e32 v2, vcc, s2, v1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_store_dword v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-LABEL: sub_i32_varying_vdata:
; GFX9:       ; %bb.0: ; %entry
; GFX9-NEXT:    s_mov_b64 s[2:3], exec
; GFX9-NEXT:    s_mov_b32 s4, 0
; GFX9-NEXT:    ; implicit-def: $vgpr1
; GFX9-NEXT:  .LBB6_1: ; %ComputeLoop
; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX9-NEXT:    s_ff1_i32_b64 s5, s[2:3]
; GFX9-NEXT:    s_mov_b32 m0, s5
; GFX9-NEXT:    v_readlane_b32 s8, v0, s5
; GFX9-NEXT:    s_lshl_b64 s[6:7], 1, s5
; GFX9-NEXT:    v_writelane_b32 v1, s4, m0
; GFX9-NEXT:    s_add_i32 s4, s4, s8
; GFX9-NEXT:    s_andn2_b64 s[2:3], s[2:3], s[6:7]
; GFX9-NEXT:    s_cmp_lg_u64 s[2:3], 0
; GFX9-NEXT:    s_cbranch_scc1 .LBB6_1
; GFX9-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX9-NEXT:    ; implicit-def: $vgpr0
; GFX9-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX9-NEXT:    s_xor_b64 s[2:3], exec, s[2:3]
; GFX9-NEXT:    s_cbranch_execz .LBB6_4
; GFX9-NEXT:  ; %bb.3:
; GFX9-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX9-NEXT:    v_mov_b32_e32 v0, s4
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    buffer_atomic_sub v0, off, s[8:11], 0 glc
; GFX9-NEXT:  .LBB6_4:
; GFX9-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_readfirstlane_b32 s2, v0
; GFX9-NEXT:    v_mov_b32_e32 v2, 0
; GFX9-NEXT:    v_sub_u32_e32 v0, s2, v1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_store_dword v2, v0, s[0:1]
; GFX9-NEXT:    s_endpgm
;
; GFX10W64-LABEL: sub_i32_varying_vdata:
; GFX10W64:       ; %bb.0: ; %entry
; GFX10W64-NEXT:    s_mov_b64 s[2:3], exec
; GFX10W64-NEXT:    s_mov_b32 s4, 0
; GFX10W64-NEXT:    ; implicit-def: $vgpr1
; GFX10W64-NEXT:  .LBB6_1: ; %ComputeLoop
; GFX10W64-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX10W64-NEXT:    s_ff1_i32_b64 s5, s[2:3]
; GFX10W64-NEXT:    v_readlane_b32 s8, v0, s5
; GFX10W64-NEXT:    s_lshl_b64 s[6:7], 1, s5
; GFX10W64-NEXT:    v_writelane_b32 v1, s4, s5
; GFX10W64-NEXT:    s_andn2_b64 s[2:3], s[2:3], s[6:7]
; GFX10W64-NEXT:    s_add_i32 s4, s4, s8
; GFX10W64-NEXT:    s_cmp_lg_u64 s[2:3], 0
; GFX10W64-NEXT:    s_cbranch_scc1 .LBB6_1
; GFX10W64-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX10W64-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX10W64-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
; GFX10W64-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX10W64-NEXT:    ; implicit-def: $vgpr0
; GFX10W64-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX10W64-NEXT:    s_xor_b64 s[2:3], exec, s[2:3]
; GFX10W64-NEXT:    s_cbranch_execz .LBB6_4
; GFX10W64-NEXT:  ; %bb.3:
; GFX10W64-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W64-NEXT:    v_mov_b32_e32 v0, s4
; GFX10W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT:    buffer_atomic_sub v0, off, s[8:11], 0 glc
; GFX10W64-NEXT:  .LBB6_4:
; GFX10W64-NEXT:    s_waitcnt_depctr 0xffe3
; GFX10W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX10W64-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W64-NEXT:    s_waitcnt vmcnt(0)
; GFX10W64-NEXT:    v_readfirstlane_b32 s2, v0
; GFX10W64-NEXT:    v_mov_b32_e32 v0, 0
; GFX10W64-NEXT:    v_sub_nc_u32_e32 v1, s2, v1
; GFX10W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX10W64-NEXT:    s_endpgm
;
; GFX10W32-LABEL: sub_i32_varying_vdata:
; GFX10W32:       ; %bb.0: ; %entry
; GFX10W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX10W32-NEXT:    s_mov_b32 s2, 0
; GFX10W32-NEXT:    ; implicit-def: $vgpr1
; GFX10W32-NEXT:  .LBB6_1: ; %ComputeLoop
; GFX10W32-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX10W32-NEXT:    s_ff1_i32_b32 s4, s3
; GFX10W32-NEXT:    v_readlane_b32 s5, v0, s4
; GFX10W32-NEXT:    s_lshl_b32 s6, 1, s4
; GFX10W32-NEXT:    v_writelane_b32 v1, s2, s4
; GFX10W32-NEXT:    s_andn2_b32 s3, s3, s6
; GFX10W32-NEXT:    s_add_i32 s2, s2, s5
; GFX10W32-NEXT:    s_cmp_lg_u32 s3, 0
; GFX10W32-NEXT:    s_cbranch_scc1 .LBB6_1
; GFX10W32-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX10W32-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX10W32-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX10W32-NEXT:    ; implicit-def: $vgpr0
; GFX10W32-NEXT:    s_and_saveexec_b32 s3, vcc_lo
; GFX10W32-NEXT:    s_xor_b32 s3, exec_lo, s3
; GFX10W32-NEXT:    s_cbranch_execz .LBB6_4
; GFX10W32-NEXT:  ; %bb.3:
; GFX10W32-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX10W32-NEXT:    v_mov_b32_e32 v0, s2
; GFX10W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT:    buffer_atomic_sub v0, off, s[4:7], 0 glc
; GFX10W32-NEXT:  .LBB6_4:
; GFX10W32-NEXT:    s_waitcnt_depctr 0xffe3
; GFX10W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
; GFX10W32-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W32-NEXT:    s_waitcnt vmcnt(0)
; GFX10W32-NEXT:    v_readfirstlane_b32 s2, v0
; GFX10W32-NEXT:    v_mov_b32_e32 v0, 0
; GFX10W32-NEXT:    v_sub_nc_u32_e32 v1, s2, v1
; GFX10W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX10W32-NEXT:    s_endpgm
;
; GFX11W64-LABEL: sub_i32_varying_vdata:
; GFX11W64:       ; %bb.0: ; %entry
; GFX11W64-NEXT:    s_mov_b64 s[2:3], exec
; GFX11W64-NEXT:    s_mov_b32 s4, 0
; GFX11W64-NEXT:    ; implicit-def: $vgpr1
; GFX11W64-NEXT:  .LBB6_1: ; %ComputeLoop
; GFX11W64-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX11W64-NEXT:    s_ctz_i32_b64 s5, s[2:3]
; GFX11W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
; GFX11W64-NEXT:    v_readlane_b32 s8, v0, s5
; GFX11W64-NEXT:    s_lshl_b64 s[6:7], 1, s5
; GFX11W64-NEXT:    v_writelane_b32 v1, s4, s5
; GFX11W64-NEXT:    s_and_not1_b64 s[2:3], s[2:3], s[6:7]
; GFX11W64-NEXT:    s_add_i32 s4, s4, s8
; GFX11W64-NEXT:    s_cmp_lg_u64 s[2:3], 0
; GFX11W64-NEXT:    s_cbranch_scc1 .LBB6_1
; GFX11W64-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX11W64-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX11W64-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
; GFX11W64-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX11W64-NEXT:    ; implicit-def: $vgpr0
; GFX11W64-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX11W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W64-NEXT:    s_xor_b64 s[2:3], exec, s[2:3]
; GFX11W64-NEXT:    s_cbranch_execz .LBB6_4
; GFX11W64-NEXT:  ; %bb.3:
; GFX11W64-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W64-NEXT:    v_mov_b32_e32 v0, s4
; GFX11W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT:    buffer_atomic_sub_u32 v0, off, s[8:11], 0 glc
; GFX11W64-NEXT:  .LBB6_4:
; GFX11W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX11W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W64-NEXT:    s_waitcnt vmcnt(0)
; GFX11W64-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11W64-NEXT:    v_mov_b32_e32 v0, 0
; GFX11W64-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11W64-NEXT:    v_sub_nc_u32_e32 v1, s2, v1
; GFX11W64-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11W64-NEXT:    s_nop 0
; GFX11W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W64-NEXT:    s_endpgm
;
; GFX11W32-LABEL: sub_i32_varying_vdata:
; GFX11W32:       ; %bb.0: ; %entry
; GFX11W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX11W32-NEXT:    s_mov_b32 s2, 0
; GFX11W32-NEXT:    ; implicit-def: $vgpr1
; GFX11W32-NEXT:  .LBB6_1: ; %ComputeLoop
; GFX11W32-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX11W32-NEXT:    s_ctz_i32_b32 s4, s3
; GFX11W32-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
; GFX11W32-NEXT:    v_readlane_b32 s5, v0, s4
; GFX11W32-NEXT:    s_lshl_b32 s6, 1, s4
; GFX11W32-NEXT:    v_writelane_b32 v1, s2, s4
; GFX11W32-NEXT:    s_and_not1_b32 s3, s3, s6
; GFX11W32-NEXT:    s_add_i32 s2, s2, s5
; GFX11W32-NEXT:    s_cmp_lg_u32 s3, 0
; GFX11W32-NEXT:    s_cbranch_scc1 .LBB6_1
; GFX11W32-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX11W32-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX11W32-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX11W32-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX11W32-NEXT:    ; implicit-def: $vgpr0
; GFX11W32-NEXT:    s_and_saveexec_b32 s3, vcc_lo
; GFX11W32-NEXT:    s_xor_b32 s3, exec_lo, s3
; GFX11W32-NEXT:    s_cbranch_execz .LBB6_4
; GFX11W32-NEXT:  ; %bb.3:
; GFX11W32-NEXT:    s_load_b128 s[4:7], s[0:1], 0x34
; GFX11W32-NEXT:    v_mov_b32_e32 v0, s2
; GFX11W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT:    buffer_atomic_sub_u32 v0, off, s[4:7], 0 glc
; GFX11W32-NEXT:  .LBB6_4:
; GFX11W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
; GFX11W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W32-NEXT:    s_waitcnt vmcnt(0)
; GFX11W32-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11W32-NEXT:    v_mov_b32_e32 v0, 0
; GFX11W32-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11W32-NEXT:    v_sub_nc_u32_e32 v1, s2, v1
; GFX11W32-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11W32-NEXT:    s_nop 0
; GFX11W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W32-NEXT:    s_endpgm
;
; GFX12W64-LABEL: sub_i32_varying_vdata:
; GFX12W64:       ; %bb.0: ; %entry
; GFX12W64-NEXT:    s_mov_b64 s[2:3], exec
; GFX12W64-NEXT:    s_mov_b32 s4, 0
; GFX12W64-NEXT:    ; implicit-def: $vgpr1
; GFX12W64-NEXT:  .LBB6_1: ; %ComputeLoop
; GFX12W64-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX12W64-NEXT:    s_ctz_i32_b64 s5, s[2:3]
; GFX12W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
; GFX12W64-NEXT:    v_readlane_b32 s8, v0, s5
; GFX12W64-NEXT:    s_lshl_b64 s[6:7], 1, s5
; GFX12W64-NEXT:    v_writelane_b32 v1, s4, s5
; GFX12W64-NEXT:    s_and_not1_b64 s[2:3], s[2:3], s[6:7]
; GFX12W64-NEXT:    s_add_co_i32 s4, s4, s8
; GFX12W64-NEXT:    s_cmp_lg_u64 s[2:3], 0
; GFX12W64-NEXT:    s_cbranch_scc1 .LBB6_1
; GFX12W64-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX12W64-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12W64-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
; GFX12W64-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX12W64-NEXT:    ; implicit-def: $vgpr0
; GFX12W64-NEXT:    s_and_saveexec_b64 s[2:3], vcc
; GFX12W64-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX12W64-NEXT:    s_xor_b64 s[2:3], exec, s[2:3]
; GFX12W64-NEXT:    s_cbranch_execz .LBB6_4
; GFX12W64-NEXT:  ; %bb.3:
; GFX12W64-NEXT:    s_load_b128 s[8:11], s[0:1], 0x34
; GFX12W64-NEXT:    v_mov_b32_e32 v0, s4
; GFX12W64-NEXT:    s_wait_kmcnt 0x0
; GFX12W64-NEXT:    buffer_atomic_sub_u32 v0, off, s[8:11], null th:TH_ATOMIC_RETURN
; GFX12W64-NEXT:  .LBB6_4:
; GFX12W64-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX12W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX12W64-NEXT:    s_wait_loadcnt 0x0
; GFX12W64-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12W64-NEXT:    v_mov_b32_e32 v0, 0
; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12W64-NEXT:    v_sub_nc_u32_e32 v1, s2, v1
; GFX12W64-NEXT:    s_wait_kmcnt 0x0
; GFX12W64-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12W64-NEXT:    s_nop 0
; GFX12W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX12W64-NEXT:    s_endpgm
;
; GFX12W32-LABEL: sub_i32_varying_vdata:
; GFX12W32:       ; %bb.0: ; %entry
; GFX12W32-NEXT:    s_mov_b32 s3, exec_lo
; GFX12W32-NEXT:    s_mov_b32 s2, 0
; GFX12W32-NEXT:    ; implicit-def: $vgpr1
; GFX12W32-NEXT:  .LBB6_1: ; %ComputeLoop
; GFX12W32-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX12W32-NEXT:    s_ctz_i32_b32 s4, s3
; GFX12W32-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
; GFX12W32-NEXT:    v_readlane_b32 s5, v0, s4
; GFX12W32-NEXT:    s_lshl_b32 s6, 1, s4
; GFX12W32-NEXT:    v_writelane_b32 v1, s2, s4
; GFX12W32-NEXT:    s_and_not1_b32 s3, s3, s6
; GFX12W32-NEXT:    s_add_co_i32 s2, s2, s5
; GFX12W32-NEXT:    s_cmp_lg_u32 s3, 0
; GFX12W32-NEXT:    s_cbranch_scc1 .LBB6_1
; GFX12W32-NEXT:  ; %bb.2: ; %ComputeEnd
; GFX12W32-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX12W32-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX12W32-NEXT:    ; implicit-def: $vgpr0
; GFX12W32-NEXT:    s_and_saveexec_b32 s3, vcc_lo
; GFX12W32-NEXT:    s_xor_b32 s3, exec_lo, s3
; GFX12W32-NEXT:    s_cbranch_execz .LBB6_4
; GFX12W32-NEXT:  ; %bb.3:
; GFX12W32-NEXT:    s_load_b128 s[4:7], s[0:1], 0x34
; GFX12W32-NEXT:    v_mov_b32_e32 v0, s2
; GFX12W32-NEXT:    s_wait_kmcnt 0x0
; GFX12W32-NEXT:    buffer_atomic_sub_u32 v0, off, s[4:7], null th:TH_ATOMIC_RETURN
; GFX12W32-NEXT:  .LBB6_4:
; GFX12W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
; GFX12W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX12W32-NEXT:    s_wait_loadcnt 0x0
; GFX12W32-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12W32-NEXT:    v_mov_b32_e32 v0, 0
; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12W32-NEXT:    v_sub_nc_u32_e32 v1, s2, v1
; GFX12W32-NEXT:    s_wait_kmcnt 0x0
; GFX12W32-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12W32-NEXT:    s_nop 0
; GFX12W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX12W32-NEXT:    s_endpgm
entry:
  %lane = call i32 @llvm.amdgcn.workitem.id.x()
  %old = call i32 @llvm.amdgcn.raw.ptr.buffer.atomic.sub(i32 %lane, ptr addrspace(8) %inout, i32 0, i32 0, i32 0)
  store i32 %old, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @sub_i32_varying_offset(ptr addrspace(1) %out, ptr addrspace(8) %inout) {
; GFX6-LABEL: sub_i32_varying_offset:
; GFX6:       ; %bb.0: ; %entry
; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT:    v_mov_b32_e32 v1, 1
; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NEXT:    buffer_atomic_sub v1, v0, s[4:7], 0 offen glc
; GFX6-NEXT:    s_mov_b32 s3, 0xf000
; GFX6-NEXT:    s_mov_b32 s2, -1
; GFX6-NEXT:    s_waitcnt vmcnt(0)
; GFX6-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; GFX6-NEXT:    s_endpgm
;
; GFX8-LABEL: sub_i32_varying_offset:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX8-NEXT:    v_mov_b32_e32 v2, 1
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    buffer_atomic_sub v2, v0, s[4:7], 0 offen glc
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    flat_store_dword v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-LABEL: sub_i32_varying_offset:
; GFX9:       ; %bb.0: ; %entry
; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX9-NEXT:    v_mov_b32_e32 v1, 1
; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    buffer_atomic_sub v1, v0, s[4:7], 0 offen glc
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX9-NEXT:    s_endpgm
;
; GFX10-LABEL: sub_i32_varying_offset:
; GFX10:       ; %bb.0: ; %entry
; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    buffer_atomic_sub v1, v0, s[4:7], 0 offen glc
; GFX10-NEXT:    v_mov_b32_e32 v0, 0
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: sub_i32_varying_offset:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x34
; GFX11-NEXT:    v_mov_b32_e32 v1, 1
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    buffer_atomic_sub_u32 v1, v0, s[4:7], 0 offen glc
; GFX11-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: sub_i32_varying_offset:
; GFX12:       ; %bb.0: ; %entry
; GFX12-NEXT:    s_load_b128 s[4:7], s[0:1], 0x34
; GFX12-NEXT:    v_mov_b32_e32 v1, 1
; GFX12-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    buffer_atomic_sub_u32 v1, v0, s[4:7], null offen th:TH_ATOMIC_RETURN
; GFX12-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-NEXT:    s_wait_loadcnt 0x0
; GFX12-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-NEXT:    s_nop 0
; GFX12-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX12-NEXT:    s_endpgm
entry:
  %lane = call i32 @llvm.amdgcn.workitem.id.x()
  %old = call i32 @llvm.amdgcn.raw.ptr.buffer.atomic.sub(i32 1, ptr addrspace(8) %inout, i32 %lane, i32 0, i32 0)
  store i32 %old, ptr addrspace(1) %out
  ret void
}