xref: /llvm-project/llvm/test/CodeGen/AMDGPU/build_vector.ll (revision 6548b6354d1d990e1c98736f5e7c3de876bedc8e) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
2; RUN: llc < %s -mtriple=amdgcn -mcpu=tahiti | FileCheck %s --check-prefixes=GFX6,GCN
3; RUN: llc < %s -mtriple=amdgcn -mcpu=tonga | FileCheck %s --check-prefixes=GFX8,GCN
4; RUN: llc < %s -mtriple=amdgcn-amd-amdpal -mcpu=gfx1030 | FileCheck %s --check-prefixes=GFX10,GCN
5; RUN: llc < %s -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 -amdgpu-enable-vopd=0 | FileCheck %s --check-prefixes=GFX11,GCN
6; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx940 | FileCheck %s --check-prefixes=GFX940,GCN
7
8define amdgpu_kernel void @build_vector2 (ptr addrspace(1) %out) {
9; GFX6-LABEL: build_vector2:
10; GFX6:       ; %bb.0: ; %entry
11; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
12; GFX6-NEXT:    s_mov_b32 s3, 0xf000
13; GFX6-NEXT:    s_mov_b32 s2, -1
14; GFX6-NEXT:    v_mov_b32_e32 v0, 5
15; GFX6-NEXT:    v_mov_b32_e32 v1, 6
16; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
17; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
18; GFX6-NEXT:    s_endpgm
19;
20; GFX8-LABEL: build_vector2:
21; GFX8:       ; %bb.0: ; %entry
22; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
23; GFX8-NEXT:    v_mov_b32_e32 v0, 5
24; GFX8-NEXT:    v_mov_b32_e32 v1, 6
25; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
26; GFX8-NEXT:    v_mov_b32_e32 v3, s1
27; GFX8-NEXT:    v_mov_b32_e32 v2, s0
28; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
29; GFX8-NEXT:    s_endpgm
30;
31; GFX10-LABEL: build_vector2:
32; GFX10:       ; %bb.0: ; %entry
33; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
34; GFX10-NEXT:    v_mov_b32_e32 v2, 0
35; GFX10-NEXT:    v_mov_b32_e32 v0, 5
36; GFX10-NEXT:    v_mov_b32_e32 v1, 6
37; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
38; GFX10-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
39; GFX10-NEXT:    s_endpgm
40;
41; GFX11-LABEL: build_vector2:
42; GFX11:       ; %bb.0: ; %entry
43; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
44; GFX11-NEXT:    v_mov_b32_e32 v2, 0
45; GFX11-NEXT:    v_mov_b32_e32 v0, 5
46; GFX11-NEXT:    v_mov_b32_e32 v1, 6
47; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
48; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
49; GFX11-NEXT:    s_endpgm
50;
51; GFX940-LABEL: build_vector2:
52; GFX940:       ; %bb.0: ; %entry
53; GFX940-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
54; GFX940-NEXT:    v_mov_b32_e32 v2, 0
55; GFX940-NEXT:    v_mov_b32_e32 v0, 5
56; GFX940-NEXT:    v_mov_b32_e32 v1, 6
57; GFX940-NEXT:    s_waitcnt lgkmcnt(0)
58; GFX940-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1] sc0 sc1
59; GFX940-NEXT:    s_endpgm
60entry:
61  store <2 x i32> <i32 5, i32 6>, ptr addrspace(1) %out
62  ret void
63}
64
65define amdgpu_kernel void @build_vector4 (ptr addrspace(1) %out) {
66; GFX6-LABEL: build_vector4:
67; GFX6:       ; %bb.0: ; %entry
68; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
69; GFX6-NEXT:    s_mov_b32 s3, 0xf000
70; GFX6-NEXT:    s_mov_b32 s2, -1
71; GFX6-NEXT:    v_mov_b32_e32 v0, 5
72; GFX6-NEXT:    v_mov_b32_e32 v1, 6
73; GFX6-NEXT:    v_mov_b32_e32 v2, 7
74; GFX6-NEXT:    v_mov_b32_e32 v3, 8
75; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
76; GFX6-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
77; GFX6-NEXT:    s_endpgm
78;
79; GFX8-LABEL: build_vector4:
80; GFX8:       ; %bb.0: ; %entry
81; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
82; GFX8-NEXT:    v_mov_b32_e32 v0, 5
83; GFX8-NEXT:    v_mov_b32_e32 v1, 6
84; GFX8-NEXT:    v_mov_b32_e32 v2, 7
85; GFX8-NEXT:    v_mov_b32_e32 v3, 8
86; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
87; GFX8-NEXT:    v_mov_b32_e32 v5, s1
88; GFX8-NEXT:    v_mov_b32_e32 v4, s0
89; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
90; GFX8-NEXT:    s_endpgm
91;
92; GFX10-LABEL: build_vector4:
93; GFX10:       ; %bb.0: ; %entry
94; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
95; GFX10-NEXT:    v_mov_b32_e32 v4, 0
96; GFX10-NEXT:    v_mov_b32_e32 v0, 5
97; GFX10-NEXT:    v_mov_b32_e32 v1, 6
98; GFX10-NEXT:    v_mov_b32_e32 v2, 7
99; GFX10-NEXT:    v_mov_b32_e32 v3, 8
100; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
101; GFX10-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
102; GFX10-NEXT:    s_endpgm
103;
104; GFX11-LABEL: build_vector4:
105; GFX11:       ; %bb.0: ; %entry
106; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
107; GFX11-NEXT:    v_mov_b32_e32 v4, 0
108; GFX11-NEXT:    v_mov_b32_e32 v0, 5
109; GFX11-NEXT:    v_mov_b32_e32 v1, 6
110; GFX11-NEXT:    v_mov_b32_e32 v2, 7
111; GFX11-NEXT:    v_mov_b32_e32 v3, 8
112; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
113; GFX11-NEXT:    global_store_b128 v4, v[0:3], s[0:1]
114; GFX11-NEXT:    s_endpgm
115;
116; GFX940-LABEL: build_vector4:
117; GFX940:       ; %bb.0: ; %entry
118; GFX940-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
119; GFX940-NEXT:    v_mov_b32_e32 v4, 0
120; GFX940-NEXT:    v_mov_b32_e32 v0, 5
121; GFX940-NEXT:    v_mov_b32_e32 v1, 6
122; GFX940-NEXT:    v_mov_b32_e32 v2, 7
123; GFX940-NEXT:    v_mov_b32_e32 v3, 8
124; GFX940-NEXT:    s_waitcnt lgkmcnt(0)
125; GFX940-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1] sc0 sc1
126; GFX940-NEXT:    s_endpgm
127entry:
128  store <4 x i32> <i32 5, i32 6, i32 7, i32 8>, ptr addrspace(1) %out
129  ret void
130}
131
132define amdgpu_kernel void @build_vector_v2i16 (ptr addrspace(1) %out) {
133; GFX6-LABEL: build_vector_v2i16:
134; GFX6:       ; %bb.0: ; %entry
135; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
136; GFX6-NEXT:    s_mov_b32 s3, 0xf000
137; GFX6-NEXT:    s_mov_b32 s2, -1
138; GFX6-NEXT:    v_mov_b32_e32 v0, 0x60005
139; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
140; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
141; GFX6-NEXT:    s_endpgm
142;
143; GFX8-LABEL: build_vector_v2i16:
144; GFX8:       ; %bb.0: ; %entry
145; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
146; GFX8-NEXT:    v_mov_b32_e32 v2, 0x60005
147; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
148; GFX8-NEXT:    v_mov_b32_e32 v0, s0
149; GFX8-NEXT:    v_mov_b32_e32 v1, s1
150; GFX8-NEXT:    flat_store_dword v[0:1], v2
151; GFX8-NEXT:    s_endpgm
152;
153; GFX10-LABEL: build_vector_v2i16:
154; GFX10:       ; %bb.0: ; %entry
155; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
156; GFX10-NEXT:    v_mov_b32_e32 v0, 0
157; GFX10-NEXT:    v_mov_b32_e32 v1, 0x60005
158; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
159; GFX10-NEXT:    global_store_dword v0, v1, s[0:1]
160; GFX10-NEXT:    s_endpgm
161;
162; GFX11-LABEL: build_vector_v2i16:
163; GFX11:       ; %bb.0: ; %entry
164; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
165; GFX11-NEXT:    v_mov_b32_e32 v0, 0
166; GFX11-NEXT:    v_mov_b32_e32 v1, 0x60005
167; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
168; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
169; GFX11-NEXT:    s_endpgm
170;
171; GFX940-LABEL: build_vector_v2i16:
172; GFX940:       ; %bb.0: ; %entry
173; GFX940-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
174; GFX940-NEXT:    v_mov_b32_e32 v0, 0
175; GFX940-NEXT:    v_mov_b32_e32 v1, 0x60005
176; GFX940-NEXT:    s_waitcnt lgkmcnt(0)
177; GFX940-NEXT:    global_store_dword v0, v1, s[0:1] sc0 sc1
178; GFX940-NEXT:    s_endpgm
179entry:
180  store <2 x i16> <i16 5, i16 6>, ptr addrspace(1) %out
181  ret void
182}
183
184define amdgpu_kernel void @build_vector_v2i16_trunc (ptr addrspace(1) %out, i32 %a) {
185; GFX6-LABEL: build_vector_v2i16_trunc:
186; GFX6:       ; %bb.0:
187; GFX6-NEXT:    s_load_dword s6, s[4:5], 0xb
188; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
189; GFX6-NEXT:    s_mov_b32 s3, 0xf000
190; GFX6-NEXT:    s_mov_b32 s2, -1
191; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
192; GFX6-NEXT:    v_alignbit_b32 v0, 5, s6, 16
193; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
194; GFX6-NEXT:    s_endpgm
195;
196; GFX8-LABEL: build_vector_v2i16_trunc:
197; GFX8:       ; %bb.0:
198; GFX8-NEXT:    s_load_dword s2, s[4:5], 0x2c
199; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
200; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
201; GFX8-NEXT:    s_lshr_b32 s2, s2, 16
202; GFX8-NEXT:    s_or_b32 s2, s2, 0x50000
203; GFX8-NEXT:    v_mov_b32_e32 v0, s0
204; GFX8-NEXT:    v_mov_b32_e32 v1, s1
205; GFX8-NEXT:    v_mov_b32_e32 v2, s2
206; GFX8-NEXT:    flat_store_dword v[0:1], v2
207; GFX8-NEXT:    s_endpgm
208;
209; GFX10-LABEL: build_vector_v2i16_trunc:
210; GFX10:       ; %bb.0:
211; GFX10-NEXT:    s_clause 0x1
212; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x8
213; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
214; GFX10-NEXT:    v_mov_b32_e32 v0, 0
215; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
216; GFX10-NEXT:    s_lshr_b32 s2, s2, 16
217; GFX10-NEXT:    s_pack_ll_b32_b16 s2, s2, 5
218; GFX10-NEXT:    v_mov_b32_e32 v1, s2
219; GFX10-NEXT:    global_store_dword v0, v1, s[0:1]
220; GFX10-NEXT:    s_endpgm
221;
222; GFX11-LABEL: build_vector_v2i16_trunc:
223; GFX11:       ; %bb.0:
224; GFX11-NEXT:    s_clause 0x1
225; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x8
226; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
227; GFX11-NEXT:    v_mov_b32_e32 v0, 0
228; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
229; GFX11-NEXT:    s_pack_hl_b32_b16 s2, s2, 5
230; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
231; GFX11-NEXT:    v_mov_b32_e32 v1, s2
232; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
233; GFX11-NEXT:    s_endpgm
234;
235; GFX940-LABEL: build_vector_v2i16_trunc:
236; GFX940:       ; %bb.0:
237; GFX940-NEXT:    s_load_dword s2, s[4:5], 0x2c
238; GFX940-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
239; GFX940-NEXT:    v_mov_b32_e32 v0, 0
240; GFX940-NEXT:    s_waitcnt lgkmcnt(0)
241; GFX940-NEXT:    s_lshr_b32 s2, s2, 16
242; GFX940-NEXT:    s_pack_ll_b32_b16 s2, s2, 5
243; GFX940-NEXT:    v_mov_b32_e32 v1, s2
244; GFX940-NEXT:    global_store_dword v0, v1, s[0:1] sc0 sc1
245; GFX940-NEXT:    s_endpgm
246  %srl = lshr i32 %a, 16
247  %trunc = trunc i32 %srl to i16
248  %ins.0 = insertelement <2 x i16> undef, i16 %trunc, i32 0
249  %ins.1 = insertelement <2 x i16> %ins.0, i16 5, i32 1
250  store <2 x i16> %ins.1, ptr addrspace(1) %out
251  ret void
252}
253
254define amdgpu_kernel void @build_v2i32_from_v4i16_shuffle(ptr addrspace(1) %out, <4 x i16> %in) {
255; GFX6-LABEL: build_v2i32_from_v4i16_shuffle:
256; GFX6:       ; %bb.0: ; %entry
257; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
258; GFX6-NEXT:    s_mov_b32 s7, 0xf000
259; GFX6-NEXT:    s_mov_b32 s6, -1
260; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
261; GFX6-NEXT:    s_mov_b32 s4, s0
262; GFX6-NEXT:    s_mov_b32 s5, s1
263; GFX6-NEXT:    s_lshl_b32 s0, s3, 16
264; GFX6-NEXT:    s_lshl_b32 s1, s2, 16
265; GFX6-NEXT:    v_mov_b32_e32 v0, s1
266; GFX6-NEXT:    v_mov_b32_e32 v1, s0
267; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
268; GFX6-NEXT:    s_endpgm
269;
270; GFX8-LABEL: build_v2i32_from_v4i16_shuffle:
271; GFX8:       ; %bb.0: ; %entry
272; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
273; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
274; GFX8-NEXT:    s_lshl_b32 s3, s3, 16
275; GFX8-NEXT:    s_lshl_b32 s2, s2, 16
276; GFX8-NEXT:    v_mov_b32_e32 v3, s1
277; GFX8-NEXT:    v_mov_b32_e32 v0, s2
278; GFX8-NEXT:    v_mov_b32_e32 v1, s3
279; GFX8-NEXT:    v_mov_b32_e32 v2, s0
280; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
281; GFX8-NEXT:    s_endpgm
282;
283; GFX10-LABEL: build_v2i32_from_v4i16_shuffle:
284; GFX10:       ; %bb.0: ; %entry
285; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
286; GFX10-NEXT:    v_mov_b32_e32 v2, 0
287; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
288; GFX10-NEXT:    s_lshl_b32 s2, s2, 16
289; GFX10-NEXT:    s_lshl_b32 s3, s3, 16
290; GFX10-NEXT:    v_mov_b32_e32 v0, s2
291; GFX10-NEXT:    v_mov_b32_e32 v1, s3
292; GFX10-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
293; GFX10-NEXT:    s_endpgm
294;
295; GFX11-LABEL: build_v2i32_from_v4i16_shuffle:
296; GFX11:       ; %bb.0: ; %entry
297; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
298; GFX11-NEXT:    v_mov_b32_e32 v2, 0
299; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
300; GFX11-NEXT:    s_lshl_b32 s2, s2, 16
301; GFX11-NEXT:    s_lshl_b32 s3, s3, 16
302; GFX11-NEXT:    v_mov_b32_e32 v0, s2
303; GFX11-NEXT:    v_mov_b32_e32 v1, s3
304; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
305; GFX11-NEXT:    s_endpgm
306;
307; GFX940-LABEL: build_v2i32_from_v4i16_shuffle:
308; GFX940:       ; %bb.0: ; %entry
309; GFX940-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
310; GFX940-NEXT:    v_mov_b32_e32 v2, 0
311; GFX940-NEXT:    s_waitcnt lgkmcnt(0)
312; GFX940-NEXT:    s_lshl_b32 s3, s3, 16
313; GFX940-NEXT:    s_lshl_b32 s2, s2, 16
314; GFX940-NEXT:    v_mov_b32_e32 v0, s2
315; GFX940-NEXT:    v_mov_b32_e32 v1, s3
316; GFX940-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1] sc0 sc1
317; GFX940-NEXT:    s_endpgm
318entry:
319  %shuf = shufflevector <4 x i16> %in, <4 x i16> zeroinitializer, <2 x i32> <i32 0, i32 2>
320  %zextended = zext <2 x i16> %shuf to <2 x i32>
321  %shifted = shl <2 x i32> %zextended, <i32 16, i32 16>
322  store <2 x i32> %shifted, ptr addrspace(1) %out
323  ret void
324}
325;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
326; GCN: {{.*}}
327