1; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=kaveri -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,CI %s
2; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX9 %s
3
4; GCN-LABEL: {{^}}atomic_load_monotonic_i32:
5; GCN: s_waitcnt
6; GFX9-NOT: s_mov_b32 m0
7; CI-NEXT: s_mov_b32 m0
8; GCN-NEXT: ds_read_b32 v0, v0{{$}}
9; GCN-NEXT: s_waitcnt lgkmcnt(0)
10; GCN-NEXT: s_setpc_b64
11define i32 @atomic_load_monotonic_i32(i32 addrspace(3)* %ptr) {
12  %load = load atomic i32, i32 addrspace(3)* %ptr monotonic, align 4
13  ret i32 %load
14}
15
16; GCN-LABEL: {{^}}atomic_load_monotonic_i32_offset:
17; GCN: s_waitcnt
18; GFX9-NOT: s_mov_b32 m0
19; CI-NEXT: s_mov_b32 m0
20; GCN-NEXT: ds_read_b32 v0, v0 offset:64{{$}}
21; GCN-NEXT: s_waitcnt lgkmcnt(0)
22; GCN-NEXT: s_setpc_b64
23define i32 @atomic_load_monotonic_i32_offset(i32 addrspace(3)* %ptr) {
24  %gep = getelementptr inbounds i32, i32 addrspace(3)* %ptr, i32 16
25  %load = load atomic i32, i32 addrspace(3)* %gep monotonic, align 4
26  ret i32 %load
27}
28
29; GCN-LABEL: {{^}}atomic_load_monotonic_i64:
30; GCN: s_waitcnt
31; GFX9-NOT: s_mov_b32 m0
32; CI-NEXT: s_mov_b32 m0
33; GCN-NEXT: ds_read_b64 v[0:1], v0{{$}}
34; GCN-NEXT: s_waitcnt lgkmcnt(0)
35; GCN-NEXT: s_setpc_b64
36define i64 @atomic_load_monotonic_i64(i64 addrspace(3)* %ptr) {
37  %load = load atomic i64, i64 addrspace(3)* %ptr monotonic, align 8
38  ret i64 %load
39}
40
41; GCN-LABEL: {{^}}atomic_load_monotonic_i64_offset:
42; GCN: s_waitcnt
43; GFX9-NOT: s_mov_b32 m0
44; CI-NEXT: s_mov_b32 m0
45; GCN-NEXT: ds_read_b64 v[0:1], v0 offset:128{{$}}
46; GCN-NEXT: s_waitcnt lgkmcnt(0)
47; GCN-NEXT: s_setpc_b64
48define i64 @atomic_load_monotonic_i64_offset(i64 addrspace(3)* %ptr) {
49  %gep = getelementptr inbounds i64, i64 addrspace(3)* %ptr, i64 16
50  %load = load atomic i64, i64 addrspace(3)* %gep monotonic, align 8
51  ret i64 %load
52}
53