1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -march=amdgcn -mcpu=gfx902 -verify-machineinstrs -amdgpu-enable-global-sgpr-addr < %s | FileCheck -check-prefix=GCN %s
3
4define amdgpu_kernel void @vector_clause(<4 x i32> addrspace(1)* noalias nocapture readonly %arg, <4 x i32> addrspace(1)* noalias nocapture %arg1) {
5; GCN-LABEL: vector_clause:
6; GCN:       ; %bb.0: ; %bb
7; GCN-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
8; GCN-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
9; GCN-NEXT:    v_mov_b32_e32 v17, 0
10; GCN-NEXT:    v_lshlrev_b32_e32 v16, 4, v0
11; GCN-NEXT:    s_nop 0
12; GCN-NEXT:    s_waitcnt lgkmcnt(0)
13; GCN-NEXT:    s_nop 0
14; GCN-NEXT:    global_load_dwordx4 v[0:3], v[16:17], s[2:3]
15; GCN-NEXT:    global_load_dwordx4 v[4:7], v[16:17], s[2:3] offset:16
16; GCN-NEXT:    global_load_dwordx4 v[8:11], v[16:17], s[2:3] offset:32
17; GCN-NEXT:    global_load_dwordx4 v[12:15], v[16:17], s[2:3] offset:48
18; GCN-NEXT:    s_nop 0
19; GCN-NEXT:    s_waitcnt vmcnt(3)
20; GCN-NEXT:    s_nop 0
21; GCN-NEXT:    global_store_dwordx4 v[16:17], v[0:3], s[4:5]
22; GCN-NEXT:    s_waitcnt vmcnt(3)
23; GCN-NEXT:    global_store_dwordx4 v[16:17], v[4:7], s[4:5] offset:16
24; GCN-NEXT:    s_waitcnt vmcnt(3)
25; GCN-NEXT:    global_store_dwordx4 v[16:17], v[8:11], s[4:5] offset:32
26; GCN-NEXT:    s_waitcnt vmcnt(3)
27; GCN-NEXT:    global_store_dwordx4 v[16:17], v[12:15], s[4:5] offset:48
28; GCN-NEXT:    s_endpgm
29bb:
30  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
31  %tmp2 = zext i32 %tmp to i64
32  %tmp3 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg, i64 %tmp2
33  %tmp4 = load <4 x i32>, <4 x i32> addrspace(1)* %tmp3, align 16
34  %tmp5 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg1, i64 %tmp2
35  %tmp6 = add nuw nsw i64 %tmp2, 1
36  %tmp7 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg, i64 %tmp6
37  %tmp8 = load <4 x i32>, <4 x i32> addrspace(1)* %tmp7, align 16
38  %tmp9 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg1, i64 %tmp6
39  %tmp10 = add nuw nsw i64 %tmp2, 2
40  %tmp11 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg, i64 %tmp10
41  %tmp12 = load <4 x i32>, <4 x i32> addrspace(1)* %tmp11, align 16
42  %tmp13 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg1, i64 %tmp10
43  %tmp14 = add nuw nsw i64 %tmp2, 3
44  %tmp15 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg, i64 %tmp14
45  %tmp16 = load <4 x i32>, <4 x i32> addrspace(1)* %tmp15, align 16
46  %tmp17 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg1, i64 %tmp14
47  store <4 x i32> %tmp4, <4 x i32> addrspace(1)* %tmp5, align 16
48  store <4 x i32> %tmp8, <4 x i32> addrspace(1)* %tmp9, align 16
49  store <4 x i32> %tmp12, <4 x i32> addrspace(1)* %tmp13, align 16
50  store <4 x i32> %tmp16, <4 x i32> addrspace(1)* %tmp17, align 16
51  ret void
52}
53
54define amdgpu_kernel void @scalar_clause(<4 x i32> addrspace(1)* noalias nocapture readonly %arg, <4 x i32> addrspace(1)* noalias nocapture %arg1) {
55; GCN-LABEL: scalar_clause:
56; GCN:       ; %bb.0: ; %bb
57; GCN-NEXT:    s_load_dwordx2 s[16:17], s[0:1], 0x24
58; GCN-NEXT:    s_load_dwordx2 s[18:19], s[0:1], 0x2c
59; GCN-NEXT:    s_nop 0
60; GCN-NEXT:    s_waitcnt lgkmcnt(0)
61; GCN-NEXT:    s_load_dwordx4 s[0:3], s[16:17], 0x0
62; GCN-NEXT:    s_load_dwordx4 s[4:7], s[16:17], 0x10
63; GCN-NEXT:    s_load_dwordx4 s[8:11], s[16:17], 0x20
64; GCN-NEXT:    s_load_dwordx4 s[12:15], s[16:17], 0x30
65; GCN-NEXT:    v_mov_b32_e32 v12, s18
66; GCN-NEXT:    s_waitcnt lgkmcnt(0)
67; GCN-NEXT:    v_mov_b32_e32 v0, s0
68; GCN-NEXT:    v_mov_b32_e32 v4, s4
69; GCN-NEXT:    v_mov_b32_e32 v8, s8
70; GCN-NEXT:    v_mov_b32_e32 v13, s19
71; GCN-NEXT:    v_mov_b32_e32 v1, s1
72; GCN-NEXT:    v_mov_b32_e32 v2, s2
73; GCN-NEXT:    v_mov_b32_e32 v3, s3
74; GCN-NEXT:    v_mov_b32_e32 v5, s5
75; GCN-NEXT:    v_mov_b32_e32 v6, s6
76; GCN-NEXT:    v_mov_b32_e32 v7, s7
77; GCN-NEXT:    s_nop 0
78; GCN-NEXT:    s_nop 0
79; GCN-NEXT:    global_store_dwordx4 v[12:13], v[0:3], off
80; GCN-NEXT:    global_store_dwordx4 v[12:13], v[4:7], off offset:16
81; GCN-NEXT:    v_mov_b32_e32 v0, s12
82; GCN-NEXT:    v_mov_b32_e32 v9, s9
83; GCN-NEXT:    v_mov_b32_e32 v10, s10
84; GCN-NEXT:    v_mov_b32_e32 v11, s11
85; GCN-NEXT:    v_mov_b32_e32 v1, s13
86; GCN-NEXT:    v_mov_b32_e32 v2, s14
87; GCN-NEXT:    v_mov_b32_e32 v3, s15
88; GCN-NEXT:    s_nop 0
89; GCN-NEXT:    s_nop 0
90; GCN-NEXT:    global_store_dwordx4 v[12:13], v[8:11], off offset:32
91; GCN-NEXT:    global_store_dwordx4 v[12:13], v[0:3], off offset:48
92; GCN-NEXT:    s_endpgm
93bb:
94  %tmp = load <4 x i32>, <4 x i32> addrspace(1)* %arg, align 16
95  %tmp2 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg, i64 1
96  %tmp3 = load <4 x i32>, <4 x i32> addrspace(1)* %tmp2, align 16
97  %tmp4 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg1, i64 1
98  %tmp5 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg, i64 2
99  %tmp6 = load <4 x i32>, <4 x i32> addrspace(1)* %tmp5, align 16
100  %tmp7 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg1, i64 2
101  %tmp8 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg, i64 3
102  %tmp9 = load <4 x i32>, <4 x i32> addrspace(1)* %tmp8, align 16
103  %tmp10 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg1, i64 3
104  store <4 x i32> %tmp, <4 x i32> addrspace(1)* %arg1, align 16
105  store <4 x i32> %tmp3, <4 x i32> addrspace(1)* %tmp4, align 16
106  store <4 x i32> %tmp6, <4 x i32> addrspace(1)* %tmp7, align 16
107  store <4 x i32> %tmp9, <4 x i32> addrspace(1)* %tmp10, align 16
108  ret void
109}
110
111define void @mubuf_clause(<4 x i32> addrspace(5)* noalias nocapture readonly %arg, <4 x i32> addrspace(5)* noalias nocapture %arg1) {
112; GCN-LABEL: mubuf_clause:
113; GCN:       ; %bb.0: ; %bb
114; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
115; GCN-NEXT:    v_and_b32_e32 v2, 0x3ff, v2
116; GCN-NEXT:    v_lshlrev_b32_e32 v2, 4, v2
117; GCN-NEXT:    v_add_u32_e32 v0, v0, v2
118; GCN-NEXT:    v_add_u32_e32 v1, v1, v2
119; GCN-NEXT:    s_nop 0
120; GCN-NEXT:    s_nop 0
121; GCN-NEXT:    buffer_load_dword v6, v0, s[0:3], 0 offen offset:20
122; GCN-NEXT:    buffer_load_dword v7, v0, s[0:3], 0 offen offset:24
123; GCN-NEXT:    buffer_load_dword v8, v0, s[0:3], 0 offen offset:28
124; GCN-NEXT:    buffer_load_dword v9, v0, s[0:3], 0 offen offset:32
125; GCN-NEXT:    buffer_load_dword v10, v0, s[0:3], 0 offen offset:36
126; GCN-NEXT:    buffer_load_dword v11, v0, s[0:3], 0 offen offset:40
127; GCN-NEXT:    buffer_load_dword v12, v0, s[0:3], 0 offen offset:44
128; GCN-NEXT:    buffer_load_dword v13, v0, s[0:3], 0 offen offset:48
129; GCN-NEXT:    buffer_load_dword v14, v0, s[0:3], 0 offen offset:52
130; GCN-NEXT:    buffer_load_dword v15, v0, s[0:3], 0 offen offset:56
131; GCN-NEXT:    buffer_load_dword v16, v0, s[0:3], 0 offen offset:60
132; GCN-NEXT:    buffer_load_dword v2, v0, s[0:3], 0 offen
133; GCN-NEXT:    buffer_load_dword v3, v0, s[0:3], 0 offen offset:4
134; GCN-NEXT:    buffer_load_dword v4, v0, s[0:3], 0 offen offset:8
135; GCN-NEXT:    buffer_load_dword v5, v0, s[0:3], 0 offen offset:12
136; GCN-NEXT:    s_nop 0
137; GCN-NEXT:    s_nop 0
138; GCN-NEXT:    buffer_load_dword v0, v0, s[0:3], 0 offen offset:16
139; GCN-NEXT:    s_nop 0
140; GCN-NEXT:    s_waitcnt vmcnt(4)
141; GCN-NEXT:    s_nop 0
142; GCN-NEXT:    buffer_store_dword v2, v1, s[0:3], 0 offen
143; GCN-NEXT:    s_waitcnt vmcnt(4)
144; GCN-NEXT:    buffer_store_dword v3, v1, s[0:3], 0 offen offset:4
145; GCN-NEXT:    s_waitcnt vmcnt(4)
146; GCN-NEXT:    buffer_store_dword v4, v1, s[0:3], 0 offen offset:8
147; GCN-NEXT:    s_waitcnt vmcnt(4)
148; GCN-NEXT:    buffer_store_dword v5, v1, s[0:3], 0 offen offset:12
149; GCN-NEXT:    s_waitcnt vmcnt(4)
150; GCN-NEXT:    buffer_store_dword v0, v1, s[0:3], 0 offen offset:16
151; GCN-NEXT:    buffer_store_dword v6, v1, s[0:3], 0 offen offset:20
152; GCN-NEXT:    buffer_store_dword v7, v1, s[0:3], 0 offen offset:24
153; GCN-NEXT:    buffer_store_dword v8, v1, s[0:3], 0 offen offset:28
154; GCN-NEXT:    buffer_store_dword v9, v1, s[0:3], 0 offen offset:32
155; GCN-NEXT:    buffer_store_dword v10, v1, s[0:3], 0 offen offset:36
156; GCN-NEXT:    buffer_store_dword v11, v1, s[0:3], 0 offen offset:40
157; GCN-NEXT:    buffer_store_dword v12, v1, s[0:3], 0 offen offset:44
158; GCN-NEXT:    buffer_store_dword v13, v1, s[0:3], 0 offen offset:48
159; GCN-NEXT:    buffer_store_dword v14, v1, s[0:3], 0 offen offset:52
160; GCN-NEXT:    buffer_store_dword v15, v1, s[0:3], 0 offen offset:56
161; GCN-NEXT:    buffer_store_dword v16, v1, s[0:3], 0 offen offset:60
162; GCN-NEXT:    s_waitcnt vmcnt(0)
163; GCN-NEXT:    s_setpc_b64 s[30:31]
164bb:
165  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
166  %tmp2 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(5)* %arg, i32 %tmp
167  %tmp3 = load <4 x i32>, <4 x i32> addrspace(5)* %tmp2, align 16
168  %tmp4 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(5)* %arg1, i32 %tmp
169  %tmp5 = add nuw nsw i32 %tmp, 1
170  %tmp6 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(5)* %arg, i32 %tmp5
171  %tmp7 = load <4 x i32>, <4 x i32> addrspace(5)* %tmp6, align 16
172  %tmp8 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(5)* %arg1, i32 %tmp5
173  %tmp9 = add nuw nsw i32 %tmp, 2
174  %tmp10 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(5)* %arg, i32 %tmp9
175  %tmp11 = load <4 x i32>, <4 x i32> addrspace(5)* %tmp10, align 16
176  %tmp12 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(5)* %arg1, i32 %tmp9
177  %tmp13 = add nuw nsw i32 %tmp, 3
178  %tmp14 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(5)* %arg, i32 %tmp13
179  %tmp15 = load <4 x i32>, <4 x i32> addrspace(5)* %tmp14, align 16
180  %tmp16 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(5)* %arg1, i32 %tmp13
181  store <4 x i32> %tmp3, <4 x i32> addrspace(5)* %tmp4, align 16
182  store <4 x i32> %tmp7, <4 x i32> addrspace(5)* %tmp8, align 16
183  store <4 x i32> %tmp11, <4 x i32> addrspace(5)* %tmp12, align 16
184  store <4 x i32> %tmp15, <4 x i32> addrspace(5)* %tmp16, align 16
185  ret void
186}
187
188define amdgpu_kernel void @vector_clause_indirect(i64 addrspace(1)* noalias nocapture readonly %arg, <4 x i32> addrspace(1)* noalias nocapture readnone %arg1, <4 x i32> addrspace(1)* noalias nocapture %arg2) {
189; GCN-LABEL: vector_clause_indirect:
190; GCN:       ; %bb.0: ; %bb
191; GCN-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
192; GCN-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x34
193; GCN-NEXT:    v_mov_b32_e32 v1, 0
194; GCN-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
195; GCN-NEXT:    s_nop 0
196; GCN-NEXT:    s_waitcnt lgkmcnt(0)
197; GCN-NEXT:    s_nop 0
198; GCN-NEXT:    global_load_dwordx2 v[8:9], v[0:1], s[2:3]
199; GCN-NEXT:    s_nop 0
200; GCN-NEXT:    s_waitcnt vmcnt(0)
201; GCN-NEXT:    s_nop 0
202; GCN-NEXT:    global_load_dwordx4 v[0:3], v[8:9], off
203; GCN-NEXT:    global_load_dwordx4 v[4:7], v[8:9], off offset:16
204; GCN-NEXT:    v_mov_b32_e32 v9, s5
205; GCN-NEXT:    v_mov_b32_e32 v8, s4
206; GCN-NEXT:    s_nop 0
207; GCN-NEXT:    s_waitcnt vmcnt(1)
208; GCN-NEXT:    s_nop 0
209; GCN-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off
210; GCN-NEXT:    s_waitcnt vmcnt(1)
211; GCN-NEXT:    global_store_dwordx4 v[8:9], v[4:7], off offset:16
212; GCN-NEXT:    s_endpgm
213bb:
214  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
215  %tmp3 = zext i32 %tmp to i64
216  %tmp4 = getelementptr inbounds i64, i64 addrspace(1)* %arg, i64 %tmp3
217  %tmp5 = bitcast i64 addrspace(1)* %tmp4 to <4 x i32> addrspace(1)* addrspace(1)*
218  %tmp6 = load <4 x i32> addrspace(1)*, <4 x i32> addrspace(1)* addrspace(1)* %tmp5, align 8
219  %tmp7 = load <4 x i32>, <4 x i32> addrspace(1)* %tmp6, align 16
220  %tmp8 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %tmp6, i64 1
221  %tmp9 = load <4 x i32>, <4 x i32> addrspace(1)* %tmp8, align 16
222  store <4 x i32> %tmp7, <4 x i32> addrspace(1)* %arg2, align 16
223  %tmp10 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg2, i64 1
224  store <4 x i32> %tmp9, <4 x i32> addrspace(1)* %tmp10, align 16
225  ret void
226}
227
228define void @load_global_d16_hi(i16 addrspace(1)* %in, i16 %reg, <2 x i16> addrspace(1)* %out) {
229; GCN-LABEL: load_global_d16_hi:
230; GCN:       ; %bb.0: ; %entry
231; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
232; GCN-NEXT:    v_mov_b32_e32 v5, v2
233; GCN-NEXT:    s_nop 0
234; GCN-NEXT:    s_nop 0
235; GCN-NEXT:    global_load_short_d16_hi v5, v[0:1], off
236; GCN-NEXT:    s_nop 0
237; GCN-NEXT:    s_nop 0
238; GCN-NEXT:    global_load_short_d16_hi v2, v[0:1], off offset:64
239; GCN-NEXT:    s_nop 0
240; GCN-NEXT:    s_waitcnt vmcnt(1)
241; GCN-NEXT:    s_nop 0
242; GCN-NEXT:    global_store_dword v[3:4], v5, off
243; GCN-NEXT:    s_waitcnt vmcnt(1)
244; GCN-NEXT:    global_store_dword v[3:4], v2, off offset:128
245; GCN-NEXT:    s_waitcnt vmcnt(0)
246; GCN-NEXT:    s_setpc_b64 s[30:31]
247entry:
248  %gep = getelementptr inbounds i16, i16 addrspace(1)* %in, i64 32
249  %load1 = load i16, i16 addrspace(1)* %in
250  %load2 = load i16, i16 addrspace(1)* %gep
251  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
252  %build1 = insertelement <2 x i16> %build0, i16 %load1, i32 1
253  store <2 x i16> %build1, <2 x i16> addrspace(1)* %out
254  %build2 = insertelement <2 x i16> undef, i16 %reg, i32 0
255  %build3 = insertelement <2 x i16> %build2, i16 %load2, i32 1
256  %gep2 = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 32
257  store <2 x i16> %build3, <2 x i16> addrspace(1)* %gep2
258  ret void
259}
260
261define void @load_global_d16_lo(i16 addrspace(1)* %in, i32 %reg, <2 x i16> addrspace(1)* %out) {
262; GCN-LABEL: load_global_d16_lo:
263; GCN:       ; %bb.0: ; %entry
264; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
265; GCN-NEXT:    v_mov_b32_e32 v5, v2
266; GCN-NEXT:    s_nop 0
267; GCN-NEXT:    s_nop 0
268; GCN-NEXT:    global_load_short_d16 v5, v[0:1], off
269; GCN-NEXT:    s_nop 0
270; GCN-NEXT:    s_nop 0
271; GCN-NEXT:    global_load_short_d16 v2, v[0:1], off offset:64
272; GCN-NEXT:    s_nop 0
273; GCN-NEXT:    s_waitcnt vmcnt(1)
274; GCN-NEXT:    s_nop 0
275; GCN-NEXT:    global_store_dword v[3:4], v5, off
276; GCN-NEXT:    s_waitcnt vmcnt(1)
277; GCN-NEXT:    global_store_dword v[3:4], v2, off offset:128
278; GCN-NEXT:    s_waitcnt vmcnt(0)
279; GCN-NEXT:    s_setpc_b64 s[30:31]
280entry:
281  %gep = getelementptr inbounds i16, i16 addrspace(1)* %in, i64 32
282  %reg.bc1 = bitcast i32 %reg to <2 x i16>
283  %reg.bc2 = bitcast i32 %reg to <2 x i16>
284  %load1 = load i16, i16 addrspace(1)* %in
285  %load2 = load i16, i16 addrspace(1)* %gep
286  %build1 = insertelement <2 x i16> %reg.bc1, i16 %load1, i32 0
287  %build2 = insertelement <2 x i16> %reg.bc2, i16 %load2, i32 0
288  %gep2 = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 32
289  store <2 x i16> %build1, <2 x i16> addrspace(1)* %out
290  store <2 x i16> %build2, <2 x i16> addrspace(1)* %gep2
291  ret void
292}
293
294declare i32 @llvm.amdgcn.workitem.id.x()
295