1; RUN: llc -march=amdgcn -mcpu=SI < %s | FileCheck -check-prefix=SI %s
2; RUN: llc -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=SI %s
3
4; SI-LABEL: {{^}}no_reorder_v2f64_global_load_store:
5; SI: buffer_load_dwordx2
6; SI: buffer_load_dwordx2
7; SI: buffer_load_dwordx2
8; SI: buffer_load_dwordx2
9; SI: buffer_store_dwordx2
10; SI: buffer_store_dwordx2
11; SI: buffer_store_dwordx2
12; SI: buffer_store_dwordx2
13; SI: s_endpgm
14define void @no_reorder_v2f64_global_load_store(<2 x double> addrspace(1)* nocapture %x, <2 x double> addrspace(1)* nocapture %y) nounwind {
15  %tmp1 = load <2 x double> addrspace(1)* %x, align 16
16  %tmp4 = load <2 x double> addrspace(1)* %y, align 16
17  store <2 x double> %tmp4, <2 x double> addrspace(1)* %x, align 16
18  store <2 x double> %tmp1, <2 x double> addrspace(1)* %y, align 16
19  ret void
20}
21
22; SI-LABEL: {{^}}no_reorder_scalarized_v2f64_local_load_store:
23; SI: ds_read_b64
24; SI: ds_read_b64
25; SI: ds_write_b64
26; SI: ds_write_b64
27; SI: s_endpgm
28define void @no_reorder_scalarized_v2f64_local_load_store(<2 x double> addrspace(3)* nocapture %x, <2 x double> addrspace(3)* nocapture %y) nounwind {
29  %tmp1 = load <2 x double> addrspace(3)* %x, align 16
30  %tmp4 = load <2 x double> addrspace(3)* %y, align 16
31  store <2 x double> %tmp4, <2 x double> addrspace(3)* %x, align 16
32  store <2 x double> %tmp1, <2 x double> addrspace(3)* %y, align 16
33  ret void
34}
35
36; SI-LABEL: {{^}}no_reorder_split_v8i32_global_load_store:
37; SI: buffer_load_dword
38; SI: buffer_load_dword
39; SI: buffer_load_dword
40; SI: buffer_load_dword
41
42; SI: buffer_load_dword
43; SI: buffer_load_dword
44; SI: buffer_load_dword
45; SI: buffer_load_dword
46
47; SI: buffer_load_dword
48; SI: buffer_load_dword
49; SI: buffer_load_dword
50; SI: buffer_load_dword
51
52; SI: buffer_load_dword
53; SI: buffer_load_dword
54; SI: buffer_load_dword
55; SI: buffer_load_dword
56
57
58; SI: buffer_store_dword
59; SI: buffer_store_dword
60; SI: buffer_store_dword
61; SI: buffer_store_dword
62
63; SI: buffer_store_dword
64; SI: buffer_store_dword
65; SI: buffer_store_dword
66; SI: buffer_store_dword
67
68; SI: buffer_store_dword
69; SI: buffer_store_dword
70; SI: buffer_store_dword
71; SI: buffer_store_dword
72
73; SI: buffer_store_dword
74; SI: buffer_store_dword
75; SI: buffer_store_dword
76; SI: buffer_store_dword
77; SI: s_endpgm
78define void @no_reorder_split_v8i32_global_load_store(<8 x i32> addrspace(1)* nocapture %x, <8 x i32> addrspace(1)* nocapture %y) nounwind {
79  %tmp1 = load <8 x i32> addrspace(1)* %x, align 32
80  %tmp4 = load <8 x i32> addrspace(1)* %y, align 32
81  store <8 x i32> %tmp4, <8 x i32> addrspace(1)* %x, align 32
82  store <8 x i32> %tmp1, <8 x i32> addrspace(1)* %y, align 32
83  ret void
84}
85
86; SI-LABEL: {{^}}no_reorder_extload_64:
87; SI: ds_read_b64
88; SI: ds_read_b64
89; SI: ds_write_b64
90; SI-NOT: ds_read
91; SI: ds_write_b64
92; SI: s_endpgm
93define void @no_reorder_extload_64(<2 x i32> addrspace(3)* nocapture %x, <2 x i32> addrspace(3)* nocapture %y) nounwind {
94  %tmp1 = load <2 x i32> addrspace(3)* %x, align 8
95  %tmp4 = load <2 x i32> addrspace(3)* %y, align 8
96  %tmp1ext = zext <2 x i32> %tmp1 to <2 x i64>
97  %tmp4ext = zext <2 x i32> %tmp4 to <2 x i64>
98  %tmp7 = add <2 x i64> %tmp1ext, <i64 1, i64 1>
99  %tmp9 = add <2 x i64> %tmp4ext, <i64 1, i64 1>
100  %trunctmp9 = trunc <2 x i64> %tmp9 to <2 x i32>
101  %trunctmp7 = trunc <2 x i64> %tmp7 to <2 x i32>
102  store <2 x i32> %trunctmp9, <2 x i32> addrspace(3)* %x, align 8
103  store <2 x i32> %trunctmp7, <2 x i32> addrspace(3)* %y, align 8
104  ret void
105}
106