; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=amdgcn -mcpu=bonaire < %s | FileCheck -enable-var-scope -check-prefixes=GCN,CI %s
; RUN: llc -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI %s

; Test that doing a shift of a pointer with a constant add will be
; folded into the constant offset addressing mode even if the add has
; multiple uses. This is relevant to accessing 2 separate, adjacent
; LDS globals.

declare i32 @llvm.amdgcn.workitem.id.x() #1

@lds0 = addrspace(3) global [512 x float] poison, align 4
@lds1 = addrspace(3) global [512 x float] poison, align 4


; Make sure the (add tid, 2) << 2 gets folded into the ds's offset as (tid << 2) + 8
define amdgpu_kernel void @load_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: load_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    ds_read_b32 v1, v1 offset:8
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: load_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_read_b32 v1, v1 offset:8
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x float], ptr addrspace(3) @lds0, i32 0, i32 %idx.0
  %val0 = load float, ptr addrspace(3) %arrayidx0, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  store float %val0, ptr addrspace(1) %out
  ret void
}

; Make sure once the first use is folded into the addressing mode, the
; remaining add use goes through the normal shl + add constant fold.
define amdgpu_kernel void @load_shl_base_lds_1(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: load_shl_base_lds_1:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    ds_read_b32 v1, v0 offset:8
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 8, v0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: load_shl_base_lds_1:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_read_b32 v1, v0 offset:8
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 8, v0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x float], ptr addrspace(3) @lds0, i32 0, i32 %idx.0
  %val0 = load float, ptr addrspace(3) %arrayidx0, align 4
  %shl_add_use = shl i32 %idx.0, 2
  store i32 %shl_add_use, ptr addrspace(1) %add_use, align 4
  store float %val0, ptr addrspace(1) %out
  ret void
}

@maxlds = addrspace(3) global [65536 x i8] poison, align 4

define amdgpu_kernel void @load_shl_base_lds_max_offset(ptr addrspace(1) %out, ptr addrspace(3) %lds, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: load_shl_base_lds_max_offset:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
; CI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0xd
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    ds_read_u8 v1, v0 offset:65535
; CI-NEXT:    s_mov_b32 s3, 0xf000
; CI-NEXT:    s_mov_b32 s2, -1
; CI-NEXT:    s_mov_b32 s6, s2
; CI-NEXT:    s_mov_b32 s7, s3
; CI-NEXT:    v_add_i32_e32 v0, vcc, 0xffff, v0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
; CI-NEXT:    buffer_store_byte v1, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: load_shl_base_lds_max_offset:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x34
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_read_u8 v1, v0 offset:65535
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s6, s2
; VI-NEXT:    s_mov_b32 s7, s3
; VI-NEXT:    v_add_u32_e32 v0, vcc, 0xffff, v0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
; VI-NEXT:    buffer_store_byte v1, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 65535
  %arrayidx0 = getelementptr inbounds [65536 x i8], ptr addrspace(3) @maxlds, i32 0, i32 %idx.0
  %val0 = load i8, ptr addrspace(3) %arrayidx0
  store i32 %idx.0, ptr addrspace(1) %add_use
  store i8 %val0, ptr addrspace(1) %out
  ret void
}

; The two globals are placed adjacent in memory, so the same base
; pointer can be used with an offset into the second one.
define amdgpu_kernel void @load_shl_base_lds_2(ptr addrspace(1) %out) #0 {
; CI-LABEL: load_shl_base_lds_2:
; CI:       ; %bb.0:
; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    ds_read2st64_b32 v[0:1], v0 offset0:1 offset1:9
; CI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
; CI-NEXT:    s_mov_b32 s3, 0xf000
; CI-NEXT:    s_mov_b32 s2, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_add_f32_e32 v0, v0, v1
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: load_shl_base_lds_2:
; VI:       ; %bb.0:
; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_read2st64_b32 v[0:1], v0 offset0:1 offset1:9
; VI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_add_f32_e32 v0, v0, v1
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 64
  %arrayidx0 = getelementptr inbounds [512 x float], ptr addrspace(3) @lds0, i32 0, i32 %idx.0
  %val0 = load float, ptr addrspace(3) %arrayidx0, align 4
  %arrayidx1 = getelementptr inbounds [512 x float], ptr addrspace(3) @lds1, i32 0, i32 %idx.0
  %val1 = load float, ptr addrspace(3) %arrayidx1, align 4
  %sum = fadd float %val0, %val1
  store float %sum, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @store_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: store_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xb
; CI-NEXT:    s_mov_b32 s3, 0xf000
; CI-NEXT:    s_mov_b32 s2, -1
; CI-NEXT:    v_add_i32_e32 v1, vcc, 2, v0
; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; CI-NEXT:    v_mov_b32_e32 v2, 1.0
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    ds_write_b32 v0, v2 offset:8
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: store_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    v_add_u32_e32 v1, vcc, 2, v0
; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; VI-NEXT:    v_mov_b32_e32 v2, 1.0
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_write_b32 v0, v2 offset:8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x float], ptr addrspace(3) @lds0, i32 0, i32 %idx.0
  store float 1.0, ptr addrspace(3) %arrayidx0, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  ret void
}


; --------------------------------------------------------------------------------
; Atomics.

@lds2 = addrspace(3) global [512 x i32] poison, align 4

; define amdgpu_kernel void @atomic_load_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
;   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
;   %idx.0 = add nsw i32 %tid.x, 2
;   %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
;   %val = load atomic i32, ptr addrspace(3) %arrayidx0 seq_cst, align 4
;   store i32 %val, ptr addrspace(1) %out, align 4
;   store i32 %idx.0, ptr addrspace(1) %add_use, align 4
;   ret void
; }

define amdgpu_kernel void @atomic_cmpxchg_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use, i32 %swap) #0 {
; CI-LABEL: atomic_cmpxchg_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    s_load_dword s8, s[4:5], 0xd
; CI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; CI-NEXT:    v_mov_b32_e32 v2, 7
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v3, s8
; CI-NEXT:    ds_cmpst_rtn_b32 v1, v1, v2, v3 offset:8
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: atomic_cmpxchg_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    s_load_dword s8, s[4:5], 0x34
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    v_mov_b32_e32 v2, 7
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v3, s8
; VI-NEXT:    ds_cmpst_rtn_b32 v1, v1, v2, v3 offset:8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
  %pair = cmpxchg ptr addrspace(3) %arrayidx0, i32 7, i32 %swap seq_cst monotonic
  %result = extractvalue { i32, i1 } %pair, 0
  store i32 %result, ptr addrspace(1) %out, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  ret void
}

define amdgpu_kernel void @atomic_swap_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: atomic_swap_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; CI-NEXT:    v_mov_b32_e32 v2, 3
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    ds_wrxchg_rtn_b32 v1, v1, v2 offset:8
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: atomic_swap_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    v_mov_b32_e32 v2, 3
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    ds_wrxchg_rtn_b32 v1, v1, v2 offset:8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
  %val = atomicrmw xchg ptr addrspace(3) %arrayidx0, i32 3 seq_cst
  store i32 %val, ptr addrspace(1) %out, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  ret void
}

define amdgpu_kernel void @atomic_add_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: atomic_add_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; CI-NEXT:    v_mov_b32_e32 v2, 3
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    ds_add_rtn_u32 v1, v1, v2 offset:8
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: atomic_add_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    v_mov_b32_e32 v2, 3
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    ds_add_rtn_u32 v1, v1, v2 offset:8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
  %val = atomicrmw add ptr addrspace(3) %arrayidx0, i32 3 seq_cst
  store i32 %val, ptr addrspace(1) %out, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  ret void
}

define amdgpu_kernel void @atomic_sub_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: atomic_sub_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; CI-NEXT:    v_mov_b32_e32 v2, 3
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    ds_sub_rtn_u32 v1, v1, v2 offset:8
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: atomic_sub_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    v_mov_b32_e32 v2, 3
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    ds_sub_rtn_u32 v1, v1, v2 offset:8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
  %val = atomicrmw sub ptr addrspace(3) %arrayidx0, i32 3 seq_cst
  store i32 %val, ptr addrspace(1) %out, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  ret void
}

define amdgpu_kernel void @atomic_and_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: atomic_and_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; CI-NEXT:    v_mov_b32_e32 v2, 3
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    ds_and_rtn_b32 v1, v1, v2 offset:8
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: atomic_and_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    v_mov_b32_e32 v2, 3
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    ds_and_rtn_b32 v1, v1, v2 offset:8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
  %val = atomicrmw and ptr addrspace(3) %arrayidx0, i32 3 seq_cst
  store i32 %val, ptr addrspace(1) %out, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  ret void
}

define amdgpu_kernel void @atomic_or_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: atomic_or_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; CI-NEXT:    v_mov_b32_e32 v2, 3
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    ds_or_rtn_b32 v1, v1, v2 offset:8
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: atomic_or_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    v_mov_b32_e32 v2, 3
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    ds_or_rtn_b32 v1, v1, v2 offset:8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
  %val = atomicrmw or ptr addrspace(3) %arrayidx0, i32 3 seq_cst
  store i32 %val, ptr addrspace(1) %out, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  ret void
}

define amdgpu_kernel void @atomic_xor_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: atomic_xor_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; CI-NEXT:    v_mov_b32_e32 v2, 3
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    ds_xor_rtn_b32 v1, v1, v2 offset:8
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: atomic_xor_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    v_mov_b32_e32 v2, 3
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    ds_xor_rtn_b32 v1, v1, v2 offset:8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
  %val = atomicrmw xor ptr addrspace(3) %arrayidx0, i32 3 seq_cst
  store i32 %val, ptr addrspace(1) %out, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  ret void
}

; define amdgpu_kernel void @atomic_nand_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
;   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
;   %idx.0 = add nsw i32 %tid.x, 2
;   %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
;   %val = atomicrmw nand ptr addrspace(3) %arrayidx0, i32 3 seq_cst
;   store i32 %val, ptr addrspace(1) %out, align 4
;   store i32 %idx.0, ptr addrspace(1) %add_use, align 4
;   ret void
; }

define amdgpu_kernel void @atomic_min_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: atomic_min_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; CI-NEXT:    v_mov_b32_e32 v2, 3
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    ds_min_rtn_i32 v1, v1, v2 offset:8
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: atomic_min_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    v_mov_b32_e32 v2, 3
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    ds_min_rtn_i32 v1, v1, v2 offset:8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
  %val = atomicrmw min ptr addrspace(3) %arrayidx0, i32 3 seq_cst
  store i32 %val, ptr addrspace(1) %out, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  ret void
}

define amdgpu_kernel void @atomic_max_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: atomic_max_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; CI-NEXT:    v_mov_b32_e32 v2, 3
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    ds_max_rtn_i32 v1, v1, v2 offset:8
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: atomic_max_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    v_mov_b32_e32 v2, 3
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    ds_max_rtn_i32 v1, v1, v2 offset:8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
  %val = atomicrmw max ptr addrspace(3) %arrayidx0, i32 3 seq_cst
  store i32 %val, ptr addrspace(1) %out, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  ret void
}

define amdgpu_kernel void @atomic_umin_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: atomic_umin_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; CI-NEXT:    v_mov_b32_e32 v2, 3
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    ds_min_rtn_u32 v1, v1, v2 offset:8
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: atomic_umin_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    v_mov_b32_e32 v2, 3
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    ds_min_rtn_u32 v1, v1, v2 offset:8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
  %val = atomicrmw umin ptr addrspace(3) %arrayidx0, i32 3 seq_cst
  store i32 %val, ptr addrspace(1) %out, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  ret void
}

define amdgpu_kernel void @atomic_umax_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: atomic_umax_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; CI-NEXT:    v_mov_b32_e32 v2, 3
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    ds_max_rtn_u32 v1, v1, v2 offset:8
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: atomic_umax_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    v_mov_b32_e32 v2, 3
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    ds_max_rtn_u32 v1, v1, v2 offset:8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
  %val = atomicrmw umax ptr addrspace(3) %arrayidx0, i32 3 seq_cst
  store i32 %val, ptr addrspace(1) %out, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  ret void
}

define amdgpu_kernel void @atomic_inc_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: atomic_inc_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; CI-NEXT:    v_mov_b32_e32 v2, 31
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    ds_inc_rtn_u32 v1, v1, v2 offset:8
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: atomic_inc_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    v_mov_b32_e32 v2, 31
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    ds_inc_rtn_u32 v1, v1, v2 offset:8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
  %val = atomicrmw uinc_wrap ptr addrspace(3) %arrayidx0, i32 31 seq_cst
  store i32 %val, ptr addrspace(1) %out, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  ret void
}

define amdgpu_kernel void @atomic_dec_shl_base_lds_0(ptr addrspace(1) %out, ptr addrspace(1) %add_use) #0 {
; CI-LABEL: atomic_dec_shl_base_lds_0:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; CI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; CI-NEXT:    v_mov_b32_e32 v2, 31
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    ds_dec_rtn_u32 v1, v1, v2 offset:8
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_mov_b32 s6, -1
; CI-NEXT:    s_mov_b32 s4, s0
; CI-NEXT:    s_mov_b32 s5, s1
; CI-NEXT:    s_mov_b32 s0, s2
; CI-NEXT:    s_mov_b32 s1, s3
; CI-NEXT:    s_mov_b32 s2, s6
; CI-NEXT:    s_mov_b32 s3, s7
; CI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
; CI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT:    s_endpgm
;
; VI-LABEL: atomic_dec_shl_base_lds_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    v_mov_b32_e32 v2, 31
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    ds_dec_rtn_u32 v1, v1, v2 offset:8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_mov_b32 s0, s2
; VI-NEXT:    s_mov_b32 s1, s3
; VI-NEXT:    s_mov_b32 s2, s6
; VI-NEXT:    s_mov_b32 s3, s7
; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
  %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
  %idx.0 = add nsw i32 %tid.x, 2
  %arrayidx0 = getelementptr inbounds [512 x i32], ptr addrspace(3) @lds2, i32 0, i32 %idx.0
  %val = atomicrmw udec_wrap ptr addrspace(3) %arrayidx0, i32 31 seq_cst
  store i32 %val, ptr addrspace(1) %out, align 4
  store i32 %idx.0, ptr addrspace(1) %add_use, align 4
  ret void
}

define void @shl_add_ptr_combine_2use_lds(i32 %idx) #0 {
; GCN-LABEL: shl_add_ptr_combine_2use_lds:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
; GCN-NEXT:    v_mov_b32_e32 v2, 9
; GCN-NEXT:    s_mov_b32 m0, -1
; GCN-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
; GCN-NEXT:    ds_write_b32 v1, v2 offset:32
; GCN-NEXT:    v_mov_b32_e32 v1, 10
; GCN-NEXT:    ds_write_b32 v0, v1 offset:64
; GCN-NEXT:    s_waitcnt lgkmcnt(0)
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %idx.add = add nuw i32 %idx, 4
  %shl0 = shl i32 %idx.add, 3
  %shl1 = shl i32 %idx.add, 4
  %ptr0 = inttoptr i32 %shl0 to ptr addrspace(3)
  %ptr1 = inttoptr i32 %shl1 to ptr addrspace(3)
  store volatile i32 9, ptr addrspace(3) %ptr0
  store volatile i32 10, ptr addrspace(3) %ptr1
  ret void
}

define void @shl_add_ptr_combine_2use_max_lds_offset(i32 %idx) #0 {
; CI-LABEL: shl_add_ptr_combine_2use_max_lds_offset:
; CI:       ; %bb.0:
; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT:    v_lshlrev_b32_e32 v1, 4, v0
; CI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; CI-NEXT:    v_mov_b32_e32 v2, 9
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    v_add_i32_e32 v1, vcc, 0x1fff0, v1
; CI-NEXT:    ds_write_b32 v0, v2 offset:65528
; CI-NEXT:    v_mov_b32_e32 v0, 10
; CI-NEXT:    ds_write_b32 v1, v0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_setpc_b64 s[30:31]
;
; VI-LABEL: shl_add_ptr_combine_2use_max_lds_offset:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_lshlrev_b32_e32 v1, 4, v0
; VI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; VI-NEXT:    v_mov_b32_e32 v2, 9
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    v_add_u32_e32 v1, vcc, 0x1fff0, v1
; VI-NEXT:    ds_write_b32 v0, v2 offset:65528
; VI-NEXT:    v_mov_b32_e32 v0, 10
; VI-NEXT:    ds_write_b32 v1, v0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_setpc_b64 s[30:31]
  %idx.add = add nuw i32 %idx, 8191
  %shl0 = shl i32 %idx.add, 3
  %shl1 = shl i32 %idx.add, 4
  %ptr0 = inttoptr i32 %shl0 to ptr addrspace(3)
  %ptr1 = inttoptr i32 %shl1 to ptr addrspace(3)
  store volatile i32 9, ptr addrspace(3) %ptr0
  store volatile i32 10, ptr addrspace(3) %ptr1
  ret void
}

define void @shl_add_ptr_combine_2use_both_max_lds_offset(i32 %idx) #0 {
; CI-LABEL: shl_add_ptr_combine_2use_both_max_lds_offset:
; CI:       ; %bb.0:
; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT:    v_add_i32_e32 v0, vcc, 0x1000, v0
; CI-NEXT:    v_lshlrev_b32_e32 v1, 4, v0
; CI-NEXT:    v_mov_b32_e32 v2, 9
; CI-NEXT:    s_mov_b32 m0, -1
; CI-NEXT:    v_lshlrev_b32_e32 v0, 5, v0
; CI-NEXT:    ds_write_b32 v1, v2
; CI-NEXT:    v_mov_b32_e32 v1, 10
; CI-NEXT:    ds_write_b32 v0, v1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_setpc_b64 s[30:31]
;
; VI-LABEL: shl_add_ptr_combine_2use_both_max_lds_offset:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_add_u32_e32 v0, vcc, 0x1000, v0
; VI-NEXT:    v_lshlrev_b32_e32 v1, 4, v0
; VI-NEXT:    v_mov_b32_e32 v2, 9
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    v_lshlrev_b32_e32 v0, 5, v0
; VI-NEXT:    ds_write_b32 v1, v2
; VI-NEXT:    v_mov_b32_e32 v1, 10
; VI-NEXT:    ds_write_b32 v0, v1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_setpc_b64 s[30:31]
  %idx.add = add nuw i32 %idx, 4096
  %shl0 = shl i32 %idx.add, 4
  %shl1 = shl i32 %idx.add, 5
  %ptr0 = inttoptr i32 %shl0 to ptr addrspace(3)
  %ptr1 = inttoptr i32 %shl1 to ptr addrspace(3)
  store volatile i32 9, ptr addrspace(3) %ptr0
  store volatile i32 10, ptr addrspace(3) %ptr1
  ret void
}

define void @shl_add_ptr_combine_2use_private(i16 zeroext %idx.arg) #0 {
; GCN-LABEL: shl_add_ptr_combine_2use_private:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; GCN-NEXT:    v_mov_b32_e32 v2, 9
; GCN-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; GCN-NEXT:    buffer_store_dword v2, v1, s[0:3], 0 offen offset:16
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v1, 10
; GCN-NEXT:    buffer_store_dword v1, v0, s[0:3], 0 offen offset:32
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %idx = zext i16 %idx.arg to i32
  %idx.add = add nuw i32 %idx, 4
  %shl0 = shl i32 %idx.add, 2
  %shl1 = shl i32 %idx.add, 3
  %ptr0 = inttoptr i32 %shl0 to ptr addrspace(5)
  %ptr1 = inttoptr i32 %shl1 to ptr addrspace(5)
  store volatile i32 9, ptr addrspace(5) %ptr0
  store volatile i32 10, ptr addrspace(5) %ptr1
  ret void
}

define void @shl_add_ptr_combine_2use_max_private_offset(i16 zeroext %idx.arg) #0 {
; CI-LABEL: shl_add_ptr_combine_2use_max_private_offset:
; CI:       ; %bb.0:
; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT:    v_lshlrev_b32_e32 v1, 4, v0
; CI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; CI-NEXT:    v_mov_b32_e32 v2, 9
; CI-NEXT:    v_add_i32_e32 v1, vcc, 0x1ff0, v1
; CI-NEXT:    buffer_store_dword v2, v0, s[0:3], 0 offen offset:4088
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, 10
; CI-NEXT:    buffer_store_dword v0, v1, s[0:3], 0 offen
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    s_setpc_b64 s[30:31]
;
; VI-LABEL: shl_add_ptr_combine_2use_max_private_offset:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_lshlrev_b32_e32 v1, 4, v0
; VI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; VI-NEXT:    v_mov_b32_e32 v2, 9
; VI-NEXT:    v_add_u32_e32 v1, vcc, 0x1ff0, v1
; VI-NEXT:    buffer_store_dword v2, v0, s[0:3], 0 offen offset:4088
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, 10
; VI-NEXT:    buffer_store_dword v0, v1, s[0:3], 0 offen
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_setpc_b64 s[30:31]
  %idx = zext i16 %idx.arg to i32
  %idx.add = add nuw i32 %idx, 511
  %shl0 = shl i32 %idx.add, 3
  %shl1 = shl i32 %idx.add, 4
  %ptr0 = inttoptr i32 %shl0 to ptr addrspace(5)
  %ptr1 = inttoptr i32 %shl1 to ptr addrspace(5)
  store volatile i32 9, ptr addrspace(5) %ptr0
  store volatile i32 10, ptr addrspace(5) %ptr1
  ret void
}

define void @shl_add_ptr_combine_2use_both_max_private_offset(i16 zeroext %idx.arg) #0 {
; CI-LABEL: shl_add_ptr_combine_2use_both_max_private_offset:
; CI:       ; %bb.0:
; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT:    v_add_i32_e32 v0, vcc, 0x100, v0
; CI-NEXT:    v_lshlrev_b32_e32 v1, 4, v0
; CI-NEXT:    v_mov_b32_e32 v2, 9
; CI-NEXT:    v_lshlrev_b32_e32 v0, 5, v0
; CI-NEXT:    buffer_store_dword v2, v1, s[0:3], 0 offen
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v1, 10
; CI-NEXT:    buffer_store_dword v1, v0, s[0:3], 0 offen
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    s_setpc_b64 s[30:31]
;
; VI-LABEL: shl_add_ptr_combine_2use_both_max_private_offset:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_add_u32_e32 v0, vcc, 0x100, v0
; VI-NEXT:    v_lshlrev_b32_e32 v1, 4, v0
; VI-NEXT:    v_mov_b32_e32 v2, 9
; VI-NEXT:    v_lshlrev_b32_e32 v0, 5, v0
; VI-NEXT:    buffer_store_dword v2, v1, s[0:3], 0 offen
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, 10
; VI-NEXT:    buffer_store_dword v1, v0, s[0:3], 0 offen
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_setpc_b64 s[30:31]
  %idx = zext i16 %idx.arg to i32
  %idx.add = add nuw i32 %idx, 256
  %shl0 = shl i32 %idx.add, 4
  %shl1 = shl i32 %idx.add, 5
  %ptr0 = inttoptr i32 %shl0 to ptr addrspace(5)
  %ptr1 = inttoptr i32 %shl1 to ptr addrspace(5)
  store volatile i32 9, ptr addrspace(5) %ptr0
  store volatile i32 10, ptr addrspace(5) %ptr1
  ret void
}

define void @shl_or_ptr_combine_2use_lds(i32 %idx) #0 {
; GCN-LABEL: shl_or_ptr_combine_2use_lds:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_lshlrev_b32_e32 v1, 1, v0
; GCN-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
; GCN-NEXT:    v_mov_b32_e32 v2, 9
; GCN-NEXT:    s_mov_b32 m0, -1
; GCN-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
; GCN-NEXT:    ds_write_b32 v0, v2 offset:8
; GCN-NEXT:    v_mov_b32_e32 v0, 10
; GCN-NEXT:    ds_write_b32 v1, v0 offset:16
; GCN-NEXT:    s_waitcnt lgkmcnt(0)
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %idx.shl = shl i32 %idx, 1
  %idx.add = or i32 %idx.shl, 1
  %shl0 = shl i32 %idx.add, 3
  %shl1 = shl i32 %idx.add, 4
  %ptr0 = inttoptr i32 %shl0 to ptr addrspace(3)
  %ptr1 = inttoptr i32 %shl1 to ptr addrspace(3)
  store volatile i32 9, ptr addrspace(3) %ptr0
  store volatile i32 10, ptr addrspace(3) %ptr1
  ret void
}

define void @shl_or_ptr_not_combine_2use_lds(i32 %idx) #0 {
; GCN-LABEL: shl_or_ptr_not_combine_2use_lds:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_or_b32_e32 v0, 1, v0
; GCN-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
; GCN-NEXT:    v_mov_b32_e32 v2, 9
; GCN-NEXT:    s_mov_b32 m0, -1
; GCN-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
; GCN-NEXT:    ds_write_b32 v1, v2
; GCN-NEXT:    v_mov_b32_e32 v1, 10
; GCN-NEXT:    ds_write_b32 v0, v1
; GCN-NEXT:    s_waitcnt lgkmcnt(0)
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %idx.add = or i32 %idx, 1
  %shl0 = shl i32 %idx.add, 3
  %shl1 = shl i32 %idx.add, 4
  %ptr0 = inttoptr i32 %shl0 to ptr addrspace(3)
  %ptr1 = inttoptr i32 %shl1 to ptr addrspace(3)
  store volatile i32 9, ptr addrspace(3) %ptr0
  store volatile i32 10, ptr addrspace(3) %ptr1
  ret void
}

attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }