1; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SICIVI,FUNC %s
2; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-enable-ds128 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SICIVI,FUNC %s
3; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefixes=EG,FUNC %s
4
5; Testing for ds_read/write_128
6; RUN: llc -march=amdgcn -mcpu=tahiti -mattr=+enable-ds128 < %s | FileCheck -check-prefixes=SI,FUNC %s
7; RUN: llc -march=amdgcn -mcpu=tonga -mattr=+enable-ds128 < %s | FileCheck -check-prefixes=CIVI,FUNC %s
8; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=+enable-ds128 < %s | FileCheck -check-prefixes=CIVI,FUNC %s
9
10; FUNC-LABEL: {{^}}load_f32_local:
11; SICIVI: s_mov_b32 m0
12; GFX9-NOT: m0
13; GCN: ds_read_b32
14
15; EG: LDS_READ_RET
16define amdgpu_kernel void @load_f32_local(float addrspace(1)* %out, float addrspace(3)* %in) #0 {
17entry:
18  %tmp0 = load float, float addrspace(3)* %in
19  store float %tmp0, float addrspace(1)* %out
20  ret void
21}
22
23; FUNC-LABEL: {{^}}load_v2f32_local:
24; SICIVI: s_mov_b32 m0
25; GFX9-NOT: m0
26
27; GCN: ds_read_b64
28
29; EG: LDS_READ_RET
30; EG: LDS_READ_RET
31define amdgpu_kernel void @load_v2f32_local(<2 x float> addrspace(1)* %out, <2 x float> addrspace(3)* %in) #0 {
32entry:
33  %tmp0 = load <2 x float>, <2 x float> addrspace(3)* %in
34  store <2 x float> %tmp0, <2 x float> addrspace(1)* %out
35  ret void
36}
37
38; FIXME: should this do a read2_b64?
39; FUNC-LABEL: {{^}}local_load_v3f32:
40; SICIVI: s_mov_b32 m0
41; GFX9-NOT: m0
42
43; SI-DAG: ds_read_b32 v{{[0-9]+}}, v{{[0-9]+}} offset:8
44; SI-DAG: ds_read_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+$}}
45; CIVI-DAG: ds_read_b96 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+$}}
46; GCN: s_waitcnt
47; SI-DAG: ds_write_b64
48; SI-DAG: ds_write_b32 v{{[0-9]+}}, v{{[0-9]+}} offset:8{{$}}
49; CIVI-DAG: ds_write_b96 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}
50
51; EG: LDS_READ_RET
52; EG: LDS_READ_RET
53; EG: LDS_READ_RET
54define amdgpu_kernel void @local_load_v3f32(<3 x float> addrspace(3)* %out, <3 x float> addrspace(3)* %in) #0 {
55entry:
56  %tmp0 = load <3 x float>, <3 x float> addrspace(3)* %in
57  store <3 x float> %tmp0, <3 x float> addrspace(3)* %out
58  ret void
59}
60
61; FUNC-LABEL: {{^}}local_load_v4f32:
62; SICIVI: s_mov_b32 m0
63; GFX9-NOT: m0
64
65; GCN: ds_read2_b64
66
67; EG: LDS_READ_RET
68; EG: LDS_READ_RET
69; EG: LDS_READ_RET
70; EG: LDS_READ_RET
71define amdgpu_kernel void @local_load_v4f32(<4 x float> addrspace(3)* %out, <4 x float> addrspace(3)* %in) #0 {
72entry:
73  %tmp0 = load <4 x float>, <4 x float> addrspace(3)* %in
74  store <4 x float> %tmp0, <4 x float> addrspace(3)* %out
75  ret void
76}
77
78; FUNC-LABEL: {{^}}local_load_v8f32:
79; SICIVI: s_mov_b32 m0
80; GFX9-NOT: m0
81
82; GCN: ds_read2_b64
83; GCN: ds_read2_b64
84
85; EG: LDS_READ_RET
86; EG: LDS_READ_RET
87; EG: LDS_READ_RET
88; EG: LDS_READ_RET
89; EG: LDS_READ_RET
90; EG: LDS_READ_RET
91; EG: LDS_READ_RET
92; EG: LDS_READ_RET
93define amdgpu_kernel void @local_load_v8f32(<8 x float> addrspace(3)* %out, <8 x float> addrspace(3)* %in) #0 {
94entry:
95  %tmp0 = load <8 x float>, <8 x float> addrspace(3)* %in
96  store <8 x float> %tmp0, <8 x float> addrspace(3)* %out
97  ret void
98}
99
100; FUNC-LABEL: {{^}}local_load_v16f32:
101; SICIVI: s_mov_b32 m0
102; GFX9-NOT: m0
103
104; GCN: ds_read2_b64
105; GCN: ds_read2_b64
106; GCN: ds_read2_b64
107; GCN: ds_read2_b64
108
109; EG: LDS_READ_RET
110; EG: LDS_READ_RET
111; EG: LDS_READ_RET
112; EG: LDS_READ_RET
113; EG: LDS_READ_RET
114; EG: LDS_READ_RET
115; EG: LDS_READ_RET
116; EG: LDS_READ_RET
117; EG: LDS_READ_RET
118; EG: LDS_READ_RET
119; EG: LDS_READ_RET
120; EG: LDS_READ_RET
121; EG: LDS_READ_RET
122; EG: LDS_READ_RET
123; EG: LDS_READ_RET
124; EG: LDS_READ_RET
125define amdgpu_kernel void @local_load_v16f32(<16 x float> addrspace(3)* %out, <16 x float> addrspace(3)* %in) #0 {
126entry:
127  %tmp0 = load <16 x float>, <16 x float> addrspace(3)* %in
128  store <16 x float> %tmp0, <16 x float> addrspace(3)* %out
129  ret void
130}
131
132; Tests if ds_read/write_b128 gets generated for the 16 byte aligned load.
133; FUNC-LABEL: {{^}}local_v4f32_to_128:
134
135; SI-NOT: ds_read_b128
136; SI-NOT: ds_write_b128
137
138; CIVI: ds_read_b128
139; CIVI: ds_write_b128
140
141; EG: LDS_READ_RET
142; EG: LDS_READ_RET
143; EG: LDS_READ_RET
144; EG: LDS_READ_RET
145define amdgpu_kernel void @local_v4f32_to_128(<4 x float> addrspace(3)* %out, <4 x float> addrspace(3)* %in) {
146  %ld = load <4 x float>, <4 x float> addrspace(3)* %in, align 16
147  store <4 x float> %ld, <4 x float> addrspace(3)* %out, align 16
148  ret void
149}
150
151attributes #0 = { nounwind }
152