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