1; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=FUNC -check-prefix=SI %s
2
3@ptr_load = addrspace(3) global i32 addrspace(4)* undef, align 8
4
5; Make sure when the load from %ptr2 is folded the chain isn't lost,
6; resulting in losing the store to gptr
7
8; FUNC-LABEL: {{^}}missing_store_reduced:
9; SI: s_load_dwordx4
10; SI: ds_read_b64
11; SI-DAG: buffer_store_dword
12; SI-DAG: v_readfirstlane_b32 s[[PTR_LO:[0-9]+]], v{{[0-9]+}}
13; SI: v_readfirstlane_b32 s[[PTR_HI:[0-9]+]], v{{[0-9]+}}
14; SI: s_nop 3
15; SI: s_load_dword s{{[0-9]+}}, s{{\[}}[[PTR_LO]]:[[PTR_HI]]{{\]}}
16; SI: buffer_store_dword
17; SI: s_endpgm
18define amdgpu_kernel void @missing_store_reduced(i32 addrspace(1)* %out, i32 addrspace(1)* %gptr) #0 {
19  %ptr0 = load i32 addrspace(4)*, i32 addrspace(4)* addrspace(3)* @ptr_load, align 8
20  %ptr2 = getelementptr inbounds i32, i32 addrspace(4)* %ptr0, i64 2
21
22  store i32 99, i32 addrspace(1)* %gptr, align 4
23  %tmp2 = load i32, i32 addrspace(4)* %ptr2, align 4
24
25  store i32 %tmp2, i32 addrspace(1)* %out, align 4
26  ret void
27}
28
29attributes #0 = { nounwind }
30
31