1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -march=amdgcn -mcpu=tahiti -verify-machineinstrs | FileCheck %s -check-prefixes=GCN,GFX89,SI
3; RUN: llc < %s -march=amdgcn -mcpu=tonga  -verify-machineinstrs | FileCheck %s -check-prefixes=GCN,GFX89,VI
4; RUN: llc < %s -march=amdgcn -mcpu=gfx900 -verify-machineinstrs | FileCheck %s -check-prefixes=GCN,GFX89,GFX9
5; RUN: llc < %s -march=r600 -mcpu=redwood  -verify-machineinstrs | FileCheck %s -check-prefixes=GCN,R600
6
7declare i32 @llvm.fshl.i32(i32, i32, i32) nounwind readnone
8declare <2 x i32> @llvm.fshl.v2i32(<2 x i32>, <2 x i32>, <2 x i32>) nounwind readnone
9declare <4 x i32> @llvm.fshl.v4i32(<4 x i32>, <4 x i32>, <4 x i32>) nounwind readnone
10
11define amdgpu_kernel void @fshl_i32(i32 addrspace(1)* %in, i32 %x, i32 %y, i32 %z) {
12; SI-LABEL: fshl_i32:
13; SI:       ; %bb.0: ; %entry
14; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
15; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xb
16; SI-NEXT:    s_mov_b32 s7, 0xf000
17; SI-NEXT:    s_mov_b32 s6, -1
18; SI-NEXT:    s_waitcnt lgkmcnt(0)
19; SI-NEXT:    v_mov_b32_e32 v0, s1
20; SI-NEXT:    s_lshr_b32 s1, s0, 1
21; SI-NEXT:    v_alignbit_b32 v0, s0, v0, 1
22; SI-NEXT:    s_not_b32 s0, s2
23; SI-NEXT:    v_mov_b32_e32 v1, s0
24; SI-NEXT:    v_alignbit_b32 v0, s1, v0, v1
25; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
26; SI-NEXT:    s_endpgm
27;
28; VI-LABEL: fshl_i32:
29; VI:       ; %bb.0: ; %entry
30; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x24
31; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x2c
32; VI-NEXT:    s_waitcnt lgkmcnt(0)
33; VI-NEXT:    v_mov_b32_e32 v0, s1
34; VI-NEXT:    s_not_b32 s2, s2
35; VI-NEXT:    s_lshr_b32 s1, s0, 1
36; VI-NEXT:    v_alignbit_b32 v0, s0, v0, 1
37; VI-NEXT:    v_mov_b32_e32 v1, s2
38; VI-NEXT:    v_alignbit_b32 v2, s1, v0, v1
39; VI-NEXT:    v_mov_b32_e32 v0, s4
40; VI-NEXT:    v_mov_b32_e32 v1, s5
41; VI-NEXT:    flat_store_dword v[0:1], v2
42; VI-NEXT:    s_endpgm
43;
44; GFX9-LABEL: fshl_i32:
45; GFX9:       ; %bb.0: ; %entry
46; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x24
47; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x2c
48; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
49; GFX9-NEXT:    v_mov_b32_e32 v0, s1
50; GFX9-NEXT:    s_not_b32 s2, s2
51; GFX9-NEXT:    s_lshr_b32 s1, s0, 1
52; GFX9-NEXT:    v_alignbit_b32 v0, s0, v0, 1
53; GFX9-NEXT:    v_mov_b32_e32 v1, s2
54; GFX9-NEXT:    v_alignbit_b32 v2, s1, v0, v1
55; GFX9-NEXT:    v_mov_b32_e32 v0, s4
56; GFX9-NEXT:    v_mov_b32_e32 v1, s5
57; GFX9-NEXT:    global_store_dword v[0:1], v2, off
58; GFX9-NEXT:    s_endpgm
59;
60; R600-LABEL: fshl_i32:
61; R600:       ; %bb.0: ; %entry
62; R600-NEXT:    ALU 5, @4, KC0[CB0:0-32], KC1[]
63; R600-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
64; R600-NEXT:    CF_END
65; R600-NEXT:    PAD
66; R600-NEXT:    ALU clause starting at 4:
67; R600-NEXT:     LSHR T0.Z, KC0[2].Z, 1,
68; R600-NEXT:     BIT_ALIGN_INT T0.W, KC0[2].Z, KC0[2].W, 1,
69; R600-NEXT:     NOT_INT * T1.W, KC0[3].X,
70; R600-NEXT:     BIT_ALIGN_INT T0.X, PV.Z, PV.W, PS,
71; R600-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
72; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
73entry:
74  %0 = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 %z)
75  store i32 %0, i32 addrspace(1)* %in
76  ret void
77}
78
79define amdgpu_kernel void @fshl_i32_imm(i32 addrspace(1)* %in, i32 %x, i32 %y) {
80; SI-LABEL: fshl_i32_imm:
81; SI:       ; %bb.0: ; %entry
82; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
83; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
84; SI-NEXT:    s_mov_b32 s7, 0xf000
85; SI-NEXT:    s_mov_b32 s6, -1
86; SI-NEXT:    s_waitcnt lgkmcnt(0)
87; SI-NEXT:    v_mov_b32_e32 v0, s1
88; SI-NEXT:    v_alignbit_b32 v0, s0, v0, 25
89; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
90; SI-NEXT:    s_endpgm
91;
92; VI-LABEL: fshl_i32_imm:
93; VI:       ; %bb.0: ; %entry
94; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
95; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2c
96; VI-NEXT:    s_waitcnt lgkmcnt(0)
97; VI-NEXT:    v_mov_b32_e32 v0, s1
98; VI-NEXT:    v_alignbit_b32 v2, s0, v0, 25
99; VI-NEXT:    v_mov_b32_e32 v0, s2
100; VI-NEXT:    v_mov_b32_e32 v1, s3
101; VI-NEXT:    flat_store_dword v[0:1], v2
102; VI-NEXT:    s_endpgm
103;
104; GFX9-LABEL: fshl_i32_imm:
105; GFX9:       ; %bb.0: ; %entry
106; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
107; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2c
108; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
109; GFX9-NEXT:    v_mov_b32_e32 v0, s1
110; GFX9-NEXT:    v_alignbit_b32 v2, s0, v0, 25
111; GFX9-NEXT:    v_mov_b32_e32 v0, s2
112; GFX9-NEXT:    v_mov_b32_e32 v1, s3
113; GFX9-NEXT:    global_store_dword v[0:1], v2, off
114; GFX9-NEXT:    s_endpgm
115;
116; R600-LABEL: fshl_i32_imm:
117; R600:       ; %bb.0: ; %entry
118; R600-NEXT:    ALU 3, @4, KC0[CB0:0-32], KC1[]
119; R600-NEXT:    MEM_RAT_CACHELESS STORE_RAW T1.X, T0.X, 1
120; R600-NEXT:    CF_END
121; R600-NEXT:    PAD
122; R600-NEXT:    ALU clause starting at 4:
123; R600-NEXT:     LSHR * T0.X, KC0[2].Y, literal.x,
124; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
125; R600-NEXT:     BIT_ALIGN_INT * T1.X, KC0[2].Z, KC0[2].W, literal.x,
126; R600-NEXT:    25(3.503246e-44), 0(0.000000e+00)
127entry:
128  %0 = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 7)
129  store i32 %0, i32 addrspace(1)* %in
130  ret void
131}
132
133define amdgpu_kernel void @fshl_v2i32(<2 x i32> addrspace(1)* %in, <2 x i32> %x, <2 x i32> %y, <2 x i32> %z) {
134; SI-LABEL: fshl_v2i32:
135; SI:       ; %bb.0: ; %entry
136; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
137; SI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
138; SI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
139; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xf
140; SI-NEXT:    s_mov_b32 s7, 0xf000
141; SI-NEXT:    s_mov_b32 s6, -1
142; SI-NEXT:    s_waitcnt lgkmcnt(0)
143; SI-NEXT:    v_mov_b32_e32 v0, s9
144; SI-NEXT:    s_not_b32 s1, s1
145; SI-NEXT:    v_alignbit_b32 v0, s3, v0, 1
146; SI-NEXT:    v_mov_b32_e32 v1, s1
147; SI-NEXT:    s_lshr_b32 s3, s3, 1
148; SI-NEXT:    v_alignbit_b32 v1, s3, v0, v1
149; SI-NEXT:    v_mov_b32_e32 v0, s8
150; SI-NEXT:    s_not_b32 s0, s0
151; SI-NEXT:    v_alignbit_b32 v0, s2, v0, 1
152; SI-NEXT:    s_lshr_b32 s1, s2, 1
153; SI-NEXT:    v_mov_b32_e32 v2, s0
154; SI-NEXT:    v_alignbit_b32 v0, s1, v0, v2
155; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
156; SI-NEXT:    s_endpgm
157;
158; VI-LABEL: fshl_v2i32:
159; VI:       ; %bb.0: ; %entry
160; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
161; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
162; VI-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x34
163; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x3c
164; VI-NEXT:    s_waitcnt lgkmcnt(0)
165; VI-NEXT:    v_mov_b32_e32 v0, s7
166; VI-NEXT:    s_not_b32 s1, s1
167; VI-NEXT:    v_mov_b32_e32 v1, s1
168; VI-NEXT:    s_lshr_b32 s7, s5, 1
169; VI-NEXT:    v_alignbit_b32 v0, s5, v0, 1
170; VI-NEXT:    v_alignbit_b32 v1, s7, v0, v1
171; VI-NEXT:    v_mov_b32_e32 v0, s6
172; VI-NEXT:    s_not_b32 s0, s0
173; VI-NEXT:    v_alignbit_b32 v0, s4, v0, 1
174; VI-NEXT:    s_lshr_b32 s1, s4, 1
175; VI-NEXT:    v_mov_b32_e32 v2, s0
176; VI-NEXT:    v_alignbit_b32 v0, s1, v0, v2
177; VI-NEXT:    v_mov_b32_e32 v2, s2
178; VI-NEXT:    v_mov_b32_e32 v3, s3
179; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
180; VI-NEXT:    s_endpgm
181;
182; GFX9-LABEL: fshl_v2i32:
183; GFX9:       ; %bb.0: ; %entry
184; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
185; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
186; GFX9-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x34
187; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x3c
188; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
189; GFX9-NEXT:    v_mov_b32_e32 v0, s7
190; GFX9-NEXT:    s_not_b32 s1, s1
191; GFX9-NEXT:    v_mov_b32_e32 v1, s1
192; GFX9-NEXT:    s_lshr_b32 s7, s5, 1
193; GFX9-NEXT:    v_alignbit_b32 v0, s5, v0, 1
194; GFX9-NEXT:    v_alignbit_b32 v1, s7, v0, v1
195; GFX9-NEXT:    v_mov_b32_e32 v0, s6
196; GFX9-NEXT:    s_not_b32 s0, s0
197; GFX9-NEXT:    v_alignbit_b32 v0, s4, v0, 1
198; GFX9-NEXT:    s_lshr_b32 s1, s4, 1
199; GFX9-NEXT:    v_mov_b32_e32 v2, s0
200; GFX9-NEXT:    v_alignbit_b32 v0, s1, v0, v2
201; GFX9-NEXT:    v_mov_b32_e32 v2, s2
202; GFX9-NEXT:    v_mov_b32_e32 v3, s3
203; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
204; GFX9-NEXT:    s_endpgm
205;
206; R600-LABEL: fshl_v2i32:
207; R600:       ; %bb.0: ; %entry
208; R600-NEXT:    ALU 9, @4, KC0[CB0:0-32], KC1[]
209; R600-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
210; R600-NEXT:    CF_END
211; R600-NEXT:    PAD
212; R600-NEXT:    ALU clause starting at 4:
213; R600-NEXT:     LSHR T0.Z, KC0[3].X, 1,
214; R600-NEXT:     BIT_ALIGN_INT * T0.W, KC0[3].X, KC0[3].Z, 1,
215; R600-NEXT:     NOT_INT * T1.W, KC0[4].X,
216; R600-NEXT:     BIT_ALIGN_INT T0.Y, T0.Z, T0.W, PV.W,
217; R600-NEXT:     LSHR T0.Z, KC0[2].W, 1,
218; R600-NEXT:     BIT_ALIGN_INT * T0.W, KC0[2].W, KC0[3].Y, 1,
219; R600-NEXT:     NOT_INT * T1.W, KC0[3].W,
220; R600-NEXT:     BIT_ALIGN_INT T0.X, T0.Z, T0.W, PV.W,
221; R600-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
222; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
223entry:
224  %0 = call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %x, <2 x i32> %y, <2 x i32> %z)
225  store <2 x i32> %0, <2 x i32> addrspace(1)* %in
226  ret void
227}
228
229define amdgpu_kernel void @fshl_v2i32_imm(<2 x i32> addrspace(1)* %in, <2 x i32> %x, <2 x i32> %y) {
230; SI-LABEL: fshl_v2i32_imm:
231; SI:       ; %bb.0: ; %entry
232; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
233; SI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
234; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
235; SI-NEXT:    s_mov_b32 s7, 0xf000
236; SI-NEXT:    s_mov_b32 s6, -1
237; SI-NEXT:    s_waitcnt lgkmcnt(0)
238; SI-NEXT:    v_mov_b32_e32 v0, s1
239; SI-NEXT:    v_alignbit_b32 v1, s3, v0, 23
240; SI-NEXT:    v_mov_b32_e32 v0, s0
241; SI-NEXT:    v_alignbit_b32 v0, s2, v0, 25
242; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
243; SI-NEXT:    s_endpgm
244;
245; VI-LABEL: fshl_v2i32_imm:
246; VI:       ; %bb.0: ; %entry
247; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
248; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
249; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
250; VI-NEXT:    s_waitcnt lgkmcnt(0)
251; VI-NEXT:    v_mov_b32_e32 v0, s1
252; VI-NEXT:    v_mov_b32_e32 v2, s0
253; VI-NEXT:    v_alignbit_b32 v1, s5, v0, 23
254; VI-NEXT:    v_alignbit_b32 v0, s4, v2, 25
255; VI-NEXT:    v_mov_b32_e32 v2, s2
256; VI-NEXT:    v_mov_b32_e32 v3, s3
257; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
258; VI-NEXT:    s_endpgm
259;
260; GFX9-LABEL: fshl_v2i32_imm:
261; GFX9:       ; %bb.0: ; %entry
262; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
263; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
264; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
265; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
266; GFX9-NEXT:    v_mov_b32_e32 v0, s1
267; GFX9-NEXT:    v_mov_b32_e32 v2, s0
268; GFX9-NEXT:    v_alignbit_b32 v1, s5, v0, 23
269; GFX9-NEXT:    v_alignbit_b32 v0, s4, v2, 25
270; GFX9-NEXT:    v_mov_b32_e32 v2, s2
271; GFX9-NEXT:    v_mov_b32_e32 v3, s3
272; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
273; GFX9-NEXT:    s_endpgm
274;
275; R600-LABEL: fshl_v2i32_imm:
276; R600:       ; %bb.0: ; %entry
277; R600-NEXT:    ALU 5, @4, KC0[CB0:0-32], KC1[]
278; R600-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
279; R600-NEXT:    CF_END
280; R600-NEXT:    PAD
281; R600-NEXT:    ALU clause starting at 4:
282; R600-NEXT:     BIT_ALIGN_INT * T0.Y, KC0[3].X, KC0[3].Z, literal.x,
283; R600-NEXT:    23(3.222986e-44), 0(0.000000e+00)
284; R600-NEXT:     BIT_ALIGN_INT * T0.X, KC0[2].W, KC0[3].Y, literal.x,
285; R600-NEXT:    25(3.503246e-44), 0(0.000000e+00)
286; R600-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
287; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
288entry:
289  %0 = call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %x, <2 x i32> %y, <2 x i32> <i32 7, i32 9>)
290  store <2 x i32> %0, <2 x i32> addrspace(1)* %in
291  ret void
292}
293
294define amdgpu_kernel void @fshl_v4i32(<4 x i32> addrspace(1)* %in, <4 x i32> %x, <4 x i32> %y, <4 x i32> %z) {
295; SI-LABEL: fshl_v4i32:
296; SI:       ; %bb.0: ; %entry
297; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
298; SI-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0xd
299; SI-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0x11
300; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x15
301; SI-NEXT:    s_mov_b32 s7, 0xf000
302; SI-NEXT:    s_mov_b32 s6, -1
303; SI-NEXT:    s_waitcnt lgkmcnt(0)
304; SI-NEXT:    v_mov_b32_e32 v0, s15
305; SI-NEXT:    s_not_b32 s3, s3
306; SI-NEXT:    v_alignbit_b32 v0, s11, v0, 1
307; SI-NEXT:    v_mov_b32_e32 v1, s3
308; SI-NEXT:    s_lshr_b32 s11, s11, 1
309; SI-NEXT:    v_alignbit_b32 v3, s11, v0, v1
310; SI-NEXT:    v_mov_b32_e32 v0, s14
311; SI-NEXT:    s_not_b32 s2, s2
312; SI-NEXT:    v_mov_b32_e32 v1, s2
313; SI-NEXT:    v_alignbit_b32 v0, s10, v0, 1
314; SI-NEXT:    s_lshr_b32 s3, s10, 1
315; SI-NEXT:    v_alignbit_b32 v2, s3, v0, v1
316; SI-NEXT:    v_mov_b32_e32 v0, s13
317; SI-NEXT:    s_not_b32 s1, s1
318; SI-NEXT:    v_mov_b32_e32 v1, s1
319; SI-NEXT:    v_alignbit_b32 v0, s9, v0, 1
320; SI-NEXT:    s_lshr_b32 s2, s9, 1
321; SI-NEXT:    v_alignbit_b32 v1, s2, v0, v1
322; SI-NEXT:    v_mov_b32_e32 v0, s12
323; SI-NEXT:    s_not_b32 s0, s0
324; SI-NEXT:    v_alignbit_b32 v0, s8, v0, 1
325; SI-NEXT:    s_lshr_b32 s1, s8, 1
326; SI-NEXT:    v_mov_b32_e32 v4, s0
327; SI-NEXT:    v_alignbit_b32 v0, s1, v0, v4
328; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
329; SI-NEXT:    s_endpgm
330;
331; VI-LABEL: fshl_v4i32:
332; VI:       ; %bb.0: ; %entry
333; VI-NEXT:    s_load_dwordx2 s[12:13], s[0:1], 0x24
334; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
335; VI-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x44
336; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x54
337; VI-NEXT:    s_waitcnt lgkmcnt(0)
338; VI-NEXT:    v_mov_b32_e32 v0, s11
339; VI-NEXT:    s_not_b32 s3, s3
340; VI-NEXT:    v_mov_b32_e32 v1, s3
341; VI-NEXT:    s_lshr_b32 s11, s7, 1
342; VI-NEXT:    v_alignbit_b32 v0, s7, v0, 1
343; VI-NEXT:    v_alignbit_b32 v3, s11, v0, v1
344; VI-NEXT:    v_mov_b32_e32 v0, s10
345; VI-NEXT:    s_not_b32 s2, s2
346; VI-NEXT:    v_mov_b32_e32 v1, s2
347; VI-NEXT:    v_alignbit_b32 v0, s6, v0, 1
348; VI-NEXT:    s_lshr_b32 s3, s6, 1
349; VI-NEXT:    v_alignbit_b32 v2, s3, v0, v1
350; VI-NEXT:    v_mov_b32_e32 v0, s9
351; VI-NEXT:    s_not_b32 s1, s1
352; VI-NEXT:    v_mov_b32_e32 v1, s1
353; VI-NEXT:    v_alignbit_b32 v0, s5, v0, 1
354; VI-NEXT:    s_lshr_b32 s2, s5, 1
355; VI-NEXT:    v_alignbit_b32 v1, s2, v0, v1
356; VI-NEXT:    v_mov_b32_e32 v0, s8
357; VI-NEXT:    s_not_b32 s0, s0
358; VI-NEXT:    v_alignbit_b32 v0, s4, v0, 1
359; VI-NEXT:    s_lshr_b32 s1, s4, 1
360; VI-NEXT:    v_mov_b32_e32 v4, s0
361; VI-NEXT:    v_alignbit_b32 v0, s1, v0, v4
362; VI-NEXT:    v_mov_b32_e32 v4, s12
363; VI-NEXT:    v_mov_b32_e32 v5, s13
364; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
365; VI-NEXT:    s_endpgm
366;
367; GFX9-LABEL: fshl_v4i32:
368; GFX9:       ; %bb.0: ; %entry
369; GFX9-NEXT:    s_load_dwordx2 s[12:13], s[0:1], 0x24
370; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
371; GFX9-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x44
372; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x54
373; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
374; GFX9-NEXT:    v_mov_b32_e32 v0, s11
375; GFX9-NEXT:    s_not_b32 s3, s3
376; GFX9-NEXT:    v_mov_b32_e32 v1, s3
377; GFX9-NEXT:    s_lshr_b32 s11, s7, 1
378; GFX9-NEXT:    v_alignbit_b32 v0, s7, v0, 1
379; GFX9-NEXT:    v_alignbit_b32 v3, s11, v0, v1
380; GFX9-NEXT:    v_mov_b32_e32 v0, s10
381; GFX9-NEXT:    s_not_b32 s2, s2
382; GFX9-NEXT:    v_mov_b32_e32 v1, s2
383; GFX9-NEXT:    v_alignbit_b32 v0, s6, v0, 1
384; GFX9-NEXT:    s_lshr_b32 s3, s6, 1
385; GFX9-NEXT:    v_alignbit_b32 v2, s3, v0, v1
386; GFX9-NEXT:    v_mov_b32_e32 v0, s9
387; GFX9-NEXT:    s_not_b32 s1, s1
388; GFX9-NEXT:    v_mov_b32_e32 v1, s1
389; GFX9-NEXT:    v_alignbit_b32 v0, s5, v0, 1
390; GFX9-NEXT:    s_lshr_b32 s2, s5, 1
391; GFX9-NEXT:    v_alignbit_b32 v1, s2, v0, v1
392; GFX9-NEXT:    v_mov_b32_e32 v0, s8
393; GFX9-NEXT:    s_not_b32 s0, s0
394; GFX9-NEXT:    v_alignbit_b32 v0, s4, v0, 1
395; GFX9-NEXT:    s_lshr_b32 s1, s4, 1
396; GFX9-NEXT:    v_mov_b32_e32 v4, s0
397; GFX9-NEXT:    v_alignbit_b32 v0, s1, v0, v4
398; GFX9-NEXT:    v_mov_b32_e32 v4, s12
399; GFX9-NEXT:    v_mov_b32_e32 v5, s13
400; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
401; GFX9-NEXT:    s_endpgm
402;
403; R600-LABEL: fshl_v4i32:
404; R600:       ; %bb.0: ; %entry
405; R600-NEXT:    ALU 17, @4, KC0[CB0:0-32], KC1[]
406; R600-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.XYZW, T1.X, 1
407; R600-NEXT:    CF_END
408; R600-NEXT:    PAD
409; R600-NEXT:    ALU clause starting at 4:
410; R600-NEXT:     LSHR T0.Z, KC0[4].X, 1,
411; R600-NEXT:     BIT_ALIGN_INT * T0.W, KC0[4].X, KC0[5].X, 1,
412; R600-NEXT:     NOT_INT * T1.W, KC0[6].X,
413; R600-NEXT:     LSHR T0.Y, KC0[3].W, 1,
414; R600-NEXT:     BIT_ALIGN_INT T1.Z, KC0[3].W, KC0[4].W, 1,
415; R600-NEXT:     BIT_ALIGN_INT * T0.W, T0.Z, T0.W, PV.W,
416; R600-NEXT:     NOT_INT * T1.W, KC0[5].W,
417; R600-NEXT:     LSHR T1.Y, KC0[3].Z, 1,
418; R600-NEXT:     BIT_ALIGN_INT T0.Z, T0.Y, T1.Z, PV.W,
419; R600-NEXT:     BIT_ALIGN_INT * T1.W, KC0[3].Z, KC0[4].Z, 1,
420; R600-NEXT:     NOT_INT * T2.W, KC0[5].Z,
421; R600-NEXT:     BIT_ALIGN_INT T0.Y, T1.Y, T1.W, PV.W,
422; R600-NEXT:     LSHR T1.Z, KC0[3].Y, 1,
423; R600-NEXT:     BIT_ALIGN_INT * T1.W, KC0[3].Y, KC0[4].Y, 1,
424; R600-NEXT:     NOT_INT * T2.W, KC0[5].Y,
425; R600-NEXT:     BIT_ALIGN_INT T0.X, T1.Z, T1.W, PV.W,
426; R600-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
427; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
428entry:
429  %0 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z)
430  store <4 x i32> %0, <4 x i32> addrspace(1)* %in
431  ret void
432}
433
434define amdgpu_kernel void @fshl_v4i32_imm(<4 x i32> addrspace(1)* %in, <4 x i32> %x, <4 x i32> %y) {
435; SI-LABEL: fshl_v4i32_imm:
436; SI:       ; %bb.0: ; %entry
437; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
438; SI-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0xd
439; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x11
440; SI-NEXT:    s_mov_b32 s7, 0xf000
441; SI-NEXT:    s_mov_b32 s6, -1
442; SI-NEXT:    s_waitcnt lgkmcnt(0)
443; SI-NEXT:    v_mov_b32_e32 v0, s3
444; SI-NEXT:    v_alignbit_b32 v3, s11, v0, 31
445; SI-NEXT:    v_mov_b32_e32 v0, s2
446; SI-NEXT:    v_alignbit_b32 v2, s10, v0, 23
447; SI-NEXT:    v_mov_b32_e32 v0, s1
448; SI-NEXT:    v_alignbit_b32 v1, s9, v0, 25
449; SI-NEXT:    v_mov_b32_e32 v0, s0
450; SI-NEXT:    v_alignbit_b32 v0, s8, v0, 31
451; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
452; SI-NEXT:    s_endpgm
453;
454; VI-LABEL: fshl_v4i32_imm:
455; VI:       ; %bb.0: ; %entry
456; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x24
457; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
458; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x44
459; VI-NEXT:    s_waitcnt lgkmcnt(0)
460; VI-NEXT:    v_mov_b32_e32 v4, s8
461; VI-NEXT:    v_mov_b32_e32 v5, s9
462; VI-NEXT:    v_mov_b32_e32 v0, s3
463; VI-NEXT:    v_mov_b32_e32 v1, s2
464; VI-NEXT:    v_alignbit_b32 v3, s7, v0, 31
465; VI-NEXT:    v_mov_b32_e32 v0, s1
466; VI-NEXT:    v_alignbit_b32 v2, s6, v1, 23
467; VI-NEXT:    v_alignbit_b32 v1, s5, v0, 25
468; VI-NEXT:    v_mov_b32_e32 v0, s0
469; VI-NEXT:    v_alignbit_b32 v0, s4, v0, 31
470; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
471; VI-NEXT:    s_endpgm
472;
473; GFX9-LABEL: fshl_v4i32_imm:
474; GFX9:       ; %bb.0: ; %entry
475; GFX9-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x24
476; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
477; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x44
478; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
479; GFX9-NEXT:    v_mov_b32_e32 v4, s8
480; GFX9-NEXT:    v_mov_b32_e32 v5, s9
481; GFX9-NEXT:    v_mov_b32_e32 v0, s3
482; GFX9-NEXT:    v_mov_b32_e32 v1, s2
483; GFX9-NEXT:    v_alignbit_b32 v3, s7, v0, 31
484; GFX9-NEXT:    v_mov_b32_e32 v0, s1
485; GFX9-NEXT:    v_alignbit_b32 v2, s6, v1, 23
486; GFX9-NEXT:    v_alignbit_b32 v1, s5, v0, 25
487; GFX9-NEXT:    v_mov_b32_e32 v0, s0
488; GFX9-NEXT:    v_alignbit_b32 v0, s4, v0, 31
489; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
490; GFX9-NEXT:    s_endpgm
491;
492; R600-LABEL: fshl_v4i32_imm:
493; R600:       ; %bb.0: ; %entry
494; R600-NEXT:    ALU 9, @4, KC0[CB0:0-32], KC1[]
495; R600-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.XYZW, T1.X, 1
496; R600-NEXT:    CF_END
497; R600-NEXT:    PAD
498; R600-NEXT:    ALU clause starting at 4:
499; R600-NEXT:     BIT_ALIGN_INT * T0.W, KC0[4].X, KC0[5].X, literal.x,
500; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
501; R600-NEXT:     BIT_ALIGN_INT * T0.Z, KC0[3].W, KC0[4].W, literal.x,
502; R600-NEXT:    23(3.222986e-44), 0(0.000000e+00)
503; R600-NEXT:     BIT_ALIGN_INT * T0.Y, KC0[3].Z, KC0[4].Z, literal.x,
504; R600-NEXT:    25(3.503246e-44), 0(0.000000e+00)
505; R600-NEXT:     BIT_ALIGN_INT * T0.X, KC0[3].Y, KC0[4].Y, literal.x,
506; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
507; R600-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
508; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
509entry:
510  %0 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> <i32 1, i32 7, i32 9, i32 33>)
511  store <4 x i32> %0, <4 x i32> addrspace(1)* %in
512  ret void
513}
514