1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX1
3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX2
4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx | FileCheck %s --check-prefix=ALL --check-prefix=XOP --check-prefix=XOPAVX1
5; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=XOP --check-prefix=XOPAVX2
6; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512BW
7;
8; Variable Shifts
9;
10
11define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind {
12; AVX1-LABEL: var_shift_v4i64:
13; AVX1:       # BB#0:
14; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
15; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
16; AVX1-NEXT:    vpsrlq %xmm2, %xmm3, %xmm4
17; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
18; AVX1-NEXT:    vpsrlq %xmm2, %xmm3, %xmm2
19; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm4[0,1,2,3],xmm2[4,5,6,7]
20; AVX1-NEXT:    vpsrlq %xmm1, %xmm0, %xmm3
21; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
22; AVX1-NEXT:    vpsrlq %xmm1, %xmm0, %xmm0
23; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7]
24; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
25; AVX1-NEXT:    retq
26;
27; AVX2-LABEL: var_shift_v4i64:
28; AVX2:       # BB#0:
29; AVX2-NEXT:    vpsrlvq %ymm1, %ymm0, %ymm0
30; AVX2-NEXT:    retq
31;
32; XOPAVX1-LABEL: var_shift_v4i64:
33; XOPAVX1:       # BB#0:
34; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
35; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
36; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm3, %xmm2
37; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
38; XOPAVX1-NEXT:    vpshlq %xmm2, %xmm4, %xmm2
39; XOPAVX1-NEXT:    vpsubq %xmm1, %xmm3, %xmm1
40; XOPAVX1-NEXT:    vpshlq %xmm1, %xmm0, %xmm0
41; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
42; XOPAVX1-NEXT:    retq
43;
44; XOPAVX2-LABEL: var_shift_v4i64:
45; XOPAVX2:       # BB#0:
46; XOPAVX2-NEXT:    vpsrlvq %ymm1, %ymm0, %ymm0
47; XOPAVX2-NEXT:    retq
48;
49; AVX512-LABEL: var_shift_v4i64:
50; AVX512:       ## BB#0:
51; AVX512-NEXT:    vpsrlvq %ymm1, %ymm0, %ymm0
52; AVX512-NEXT:    retq
53  %shift = lshr <4 x i64> %a, %b
54  ret <4 x i64> %shift
55}
56
57define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
58; AVX1-LABEL: var_shift_v8i32:
59; AVX1:       # BB#0:
60; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
61; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
62; AVX1-NEXT:    vpsrldq {{.*#+}} xmm4 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
63; AVX1-NEXT:    vpsrld %xmm4, %xmm2, %xmm4
64; AVX1-NEXT:    vpsrlq $32, %xmm3, %xmm5
65; AVX1-NEXT:    vpsrld %xmm5, %xmm2, %xmm5
66; AVX1-NEXT:    vpblendw {{.*#+}} xmm4 = xmm5[0,1,2,3],xmm4[4,5,6,7]
67; AVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
68; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm6 = xmm3[2],xmm5[2],xmm3[3],xmm5[3]
69; AVX1-NEXT:    vpsrld %xmm6, %xmm2, %xmm6
70; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero
71; AVX1-NEXT:    vpsrld %xmm3, %xmm2, %xmm2
72; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm6[4,5,6,7]
73; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm4[2,3],xmm2[4,5],xmm4[6,7]
74; AVX1-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
75; AVX1-NEXT:    vpsrld %xmm3, %xmm0, %xmm3
76; AVX1-NEXT:    vpsrlq $32, %xmm1, %xmm4
77; AVX1-NEXT:    vpsrld %xmm4, %xmm0, %xmm4
78; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7]
79; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm4 = xmm1[2],xmm5[2],xmm1[3],xmm5[3]
80; AVX1-NEXT:    vpsrld %xmm4, %xmm0, %xmm4
81; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
82; AVX1-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
83; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm4[4,5,6,7]
84; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3],xmm0[4,5],xmm3[6,7]
85; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
86; AVX1-NEXT:    retq
87;
88; AVX2-LABEL: var_shift_v8i32:
89; AVX2:       # BB#0:
90; AVX2-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
91; AVX2-NEXT:    retq
92;
93; XOPAVX1-LABEL: var_shift_v8i32:
94; XOPAVX1:       # BB#0:
95; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
96; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
97; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm3, %xmm2
98; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
99; XOPAVX1-NEXT:    vpshld %xmm2, %xmm4, %xmm2
100; XOPAVX1-NEXT:    vpsubd %xmm1, %xmm3, %xmm1
101; XOPAVX1-NEXT:    vpshld %xmm1, %xmm0, %xmm0
102; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
103; XOPAVX1-NEXT:    retq
104;
105; XOPAVX2-LABEL: var_shift_v8i32:
106; XOPAVX2:       # BB#0:
107; XOPAVX2-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
108; XOPAVX2-NEXT:    retq
109;
110; AVX512-LABEL: var_shift_v8i32:
111; AVX512:       ## BB#0:
112; AVX512-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
113; AVX512-NEXT:    retq
114  %shift = lshr <8 x i32> %a, %b
115  ret <8 x i32> %shift
116}
117
118define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind {
119; AVX1-LABEL: var_shift_v16i16:
120; AVX1:       # BB#0:
121; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
122; AVX1-NEXT:    vpsllw $12, %xmm2, %xmm3
123; AVX1-NEXT:    vpsllw $4, %xmm2, %xmm2
124; AVX1-NEXT:    vpor %xmm3, %xmm2, %xmm2
125; AVX1-NEXT:    vpaddw %xmm2, %xmm2, %xmm3
126; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
127; AVX1-NEXT:    vpsrlw $8, %xmm4, %xmm5
128; AVX1-NEXT:    vpblendvb %xmm2, %xmm5, %xmm4, %xmm2
129; AVX1-NEXT:    vpsrlw $4, %xmm2, %xmm4
130; AVX1-NEXT:    vpblendvb %xmm3, %xmm4, %xmm2, %xmm2
131; AVX1-NEXT:    vpsrlw $2, %xmm2, %xmm4
132; AVX1-NEXT:    vpaddw %xmm3, %xmm3, %xmm3
133; AVX1-NEXT:    vpblendvb %xmm3, %xmm4, %xmm2, %xmm2
134; AVX1-NEXT:    vpsrlw $1, %xmm2, %xmm4
135; AVX1-NEXT:    vpaddw %xmm3, %xmm3, %xmm3
136; AVX1-NEXT:    vpblendvb %xmm3, %xmm4, %xmm2, %xmm2
137; AVX1-NEXT:    vpsllw $12, %xmm1, %xmm3
138; AVX1-NEXT:    vpsllw $4, %xmm1, %xmm1
139; AVX1-NEXT:    vpor %xmm3, %xmm1, %xmm1
140; AVX1-NEXT:    vpaddw %xmm1, %xmm1, %xmm3
141; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm4
142; AVX1-NEXT:    vpblendvb %xmm1, %xmm4, %xmm0, %xmm0
143; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm1
144; AVX1-NEXT:    vpblendvb %xmm3, %xmm1, %xmm0, %xmm0
145; AVX1-NEXT:    vpsrlw $2, %xmm0, %xmm1
146; AVX1-NEXT:    vpaddw %xmm3, %xmm3, %xmm3
147; AVX1-NEXT:    vpblendvb %xmm3, %xmm1, %xmm0, %xmm0
148; AVX1-NEXT:    vpsrlw $1, %xmm0, %xmm1
149; AVX1-NEXT:    vpaddw %xmm3, %xmm3, %xmm3
150; AVX1-NEXT:    vpblendvb %xmm3, %xmm1, %xmm0, %xmm0
151; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
152; AVX1-NEXT:    retq
153;
154; AVX2-LABEL: var_shift_v16i16:
155; AVX2:       # BB#0:
156; AVX2-NEXT:    vpxor %ymm2, %ymm2, %ymm2
157; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15]
158; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm0[4,4,5,5,6,6,7,7,12,12,13,13,14,14,15,15]
159; AVX2-NEXT:    vpsrlvd %ymm3, %ymm4, %ymm3
160; AVX2-NEXT:    vpsrld $16, %ymm3, %ymm3
161; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11]
162; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
163; AVX2-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
164; AVX2-NEXT:    vpsrld $16, %ymm0, %ymm0
165; AVX2-NEXT:    vpackusdw %ymm3, %ymm0, %ymm0
166; AVX2-NEXT:    retq
167;
168; XOPAVX1-LABEL: var_shift_v16i16:
169; XOPAVX1:       # BB#0:
170; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
171; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
172; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm3, %xmm2
173; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
174; XOPAVX1-NEXT:    vpshlw %xmm2, %xmm4, %xmm2
175; XOPAVX1-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
176; XOPAVX1-NEXT:    vpshlw %xmm1, %xmm0, %xmm0
177; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
178; XOPAVX1-NEXT:    retq
179;
180; XOPAVX2-LABEL: var_shift_v16i16:
181; XOPAVX2:       # BB#0:
182; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
183; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
184; XOPAVX2-NEXT:    vpsubw %xmm2, %xmm3, %xmm2
185; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm4
186; XOPAVX2-NEXT:    vpshlw %xmm2, %xmm4, %xmm2
187; XOPAVX2-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
188; XOPAVX2-NEXT:    vpshlw %xmm1, %xmm0, %xmm0
189; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
190; XOPAVX2-NEXT:    retq
191;
192; AVX512-LABEL: var_shift_v16i16:
193; AVX512:       ## BB#0:
194; AVX512-NEXT:    vpsrlvw %zmm1, %zmm0, %zmm0
195; AVX512-NEXT:    retq
196  %shift = lshr <16 x i16> %a, %b
197  ret <16 x i16> %shift
198}
199
200define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
201; AVX1-LABEL: var_shift_v32i8:
202; AVX1:       # BB#0:
203; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
204; AVX1-NEXT:    vpsrlw $4, %xmm2, %xmm3
205; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
206; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
207; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
208; AVX1-NEXT:    vpsllw $5, %xmm5, %xmm5
209; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm2, %xmm2
210; AVX1-NEXT:    vpsrlw $2, %xmm2, %xmm3
211; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
212; AVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
213; AVX1-NEXT:    vpaddb %xmm5, %xmm5, %xmm5
214; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm2, %xmm2
215; AVX1-NEXT:    vpsrlw $1, %xmm2, %xmm3
216; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
217; AVX1-NEXT:    vpand %xmm7, %xmm3, %xmm3
218; AVX1-NEXT:    vpaddb %xmm5, %xmm5, %xmm5
219; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm2, %xmm2
220; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm3
221; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
222; AVX1-NEXT:    vpsllw $5, %xmm1, %xmm1
223; AVX1-NEXT:    vpblendvb %xmm1, %xmm3, %xmm0, %xmm0
224; AVX1-NEXT:    vpsrlw $2, %xmm0, %xmm3
225; AVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
226; AVX1-NEXT:    vpaddb %xmm1, %xmm1, %xmm1
227; AVX1-NEXT:    vpblendvb %xmm1, %xmm3, %xmm0, %xmm0
228; AVX1-NEXT:    vpsrlw $1, %xmm0, %xmm3
229; AVX1-NEXT:    vpand %xmm7, %xmm3, %xmm3
230; AVX1-NEXT:    vpaddb %xmm1, %xmm1, %xmm1
231; AVX1-NEXT:    vpblendvb %xmm1, %xmm3, %xmm0, %xmm0
232; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
233; AVX1-NEXT:    retq
234;
235; AVX2-LABEL: var_shift_v32i8:
236; AVX2:       # BB#0:
237; AVX2-NEXT:    vpsllw $5, %ymm1, %ymm1
238; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm2
239; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
240; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
241; AVX2-NEXT:    vpsrlw $2, %ymm0, %ymm2
242; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
243; AVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
244; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
245; AVX2-NEXT:    vpsrlw $1, %ymm0, %ymm2
246; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
247; AVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
248; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
249; AVX2-NEXT:    retq
250;
251; XOPAVX1-LABEL: var_shift_v32i8:
252; XOPAVX1:       # BB#0:
253; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
254; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
255; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
256; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
257; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm4, %xmm2
258; XOPAVX1-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
259; XOPAVX1-NEXT:    vpshlb %xmm1, %xmm0, %xmm0
260; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
261; XOPAVX1-NEXT:    retq
262;
263; XOPAVX2-LABEL: var_shift_v32i8:
264; XOPAVX2:       # BB#0:
265; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
266; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
267; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
268; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm4
269; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm4, %xmm2
270; XOPAVX2-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
271; XOPAVX2-NEXT:    vpshlb %xmm1, %xmm0, %xmm0
272; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
273; XOPAVX2-NEXT:    retq
274;
275; AVX512-LABEL: var_shift_v32i8:
276; AVX512:       ## BB#0:
277; AVX512-NEXT:    vpsllw $5, %ymm1, %ymm1
278; AVX512-NEXT:    vpsrlw $4, %ymm0, %ymm2
279; AVX512-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
280; AVX512-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
281; AVX512-NEXT:    vpsrlw $2, %ymm0, %ymm2
282; AVX512-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
283; AVX512-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
284; AVX512-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
285; AVX512-NEXT:    vpsrlw $1, %ymm0, %ymm2
286; AVX512-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
287; AVX512-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
288; AVX512-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
289; AVX512-NEXT:    retq
290  %shift = lshr <32 x i8> %a, %b
291  ret <32 x i8> %shift
292}
293
294;
295; Uniform Variable Shifts
296;
297
298define <4 x i64> @splatvar_shift_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind {
299; AVX1-LABEL: splatvar_shift_v4i64:
300; AVX1:       # BB#0:
301; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
302; AVX1-NEXT:    vpsrlq %xmm1, %xmm2, %xmm2
303; AVX1-NEXT:    vpsrlq %xmm1, %xmm0, %xmm0
304; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
305; AVX1-NEXT:    retq
306;
307; AVX2-LABEL: splatvar_shift_v4i64:
308; AVX2:       # BB#0:
309; AVX2-NEXT:    vpsrlq %xmm1, %ymm0, %ymm0
310; AVX2-NEXT:    retq
311;
312; XOPAVX1-LABEL: splatvar_shift_v4i64:
313; XOPAVX1:       # BB#0:
314; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
315; XOPAVX1-NEXT:    vpsrlq %xmm1, %xmm2, %xmm2
316; XOPAVX1-NEXT:    vpsrlq %xmm1, %xmm0, %xmm0
317; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
318; XOPAVX1-NEXT:    retq
319;
320; XOPAVX2-LABEL: splatvar_shift_v4i64:
321; XOPAVX2:       # BB#0:
322; XOPAVX2-NEXT:    vpsrlq %xmm1, %ymm0, %ymm0
323; XOPAVX2-NEXT:    retq
324;
325; AVX512-LABEL: splatvar_shift_v4i64:
326; AVX512:       ## BB#0:
327; AVX512-NEXT:    vpsrlq %xmm1, %ymm0, %ymm0
328; AVX512-NEXT:    retq
329  %splat = shufflevector <4 x i64> %b, <4 x i64> undef, <4 x i32> zeroinitializer
330  %shift = lshr <4 x i64> %a, %splat
331  ret <4 x i64> %shift
332}
333
334define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
335; AVX1-LABEL: splatvar_shift_v8i32:
336; AVX1:       # BB#0:
337; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
338; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3,4,5,6,7]
339; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
340; AVX1-NEXT:    vpsrld %xmm1, %xmm2, %xmm2
341; AVX1-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
342; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
343; AVX1-NEXT:    retq
344;
345; AVX2-LABEL: splatvar_shift_v8i32:
346; AVX2:       # BB#0:
347; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
348; AVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3,4,5,6,7]
349; AVX2-NEXT:    vpsrld %xmm1, %ymm0, %ymm0
350; AVX2-NEXT:    retq
351;
352; XOPAVX1-LABEL: splatvar_shift_v8i32:
353; XOPAVX1:       # BB#0:
354; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
355; XOPAVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3,4,5,6,7]
356; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
357; XOPAVX1-NEXT:    vpsrld %xmm1, %xmm2, %xmm2
358; XOPAVX1-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
359; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
360; XOPAVX1-NEXT:    retq
361;
362; XOPAVX2-LABEL: splatvar_shift_v8i32:
363; XOPAVX2:       # BB#0:
364; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
365; XOPAVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3,4,5,6,7]
366; XOPAVX2-NEXT:    vpsrld %xmm1, %ymm0, %ymm0
367; XOPAVX2-NEXT:    retq
368;
369; AVX512-LABEL: splatvar_shift_v8i32:
370; AVX512:       ## BB#0:
371; AVX512-NEXT:    vxorps %xmm2, %xmm2, %xmm2
372; AVX512-NEXT:    vmovss {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3]
373; AVX512-NEXT:    vpsrld %xmm1, %ymm0, %ymm0
374; AVX512-NEXT:    retq
375  %splat = shufflevector <8 x i32> %b, <8 x i32> undef, <8 x i32> zeroinitializer
376  %shift = lshr <8 x i32> %a, %splat
377  ret <8 x i32> %shift
378}
379
380define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind {
381; AVX1-LABEL: splatvar_shift_v16i16:
382; AVX1:       # BB#0:
383; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
384; AVX1-NEXT:    vmovd %xmm1, %eax
385; AVX1-NEXT:    movzwl %ax, %eax
386; AVX1-NEXT:    vmovd %eax, %xmm1
387; AVX1-NEXT:    vpsrlw %xmm1, %xmm2, %xmm2
388; AVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
389; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
390; AVX1-NEXT:    retq
391;
392; AVX2-LABEL: splatvar_shift_v16i16:
393; AVX2:       # BB#0:
394; AVX2-NEXT:    vmovd %xmm1, %eax
395; AVX2-NEXT:    movzwl %ax, %eax
396; AVX2-NEXT:    vmovd %eax, %xmm1
397; AVX2-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
398; AVX2-NEXT:    retq
399;
400; XOPAVX1-LABEL: splatvar_shift_v16i16:
401; XOPAVX1:       # BB#0:
402; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
403; XOPAVX1-NEXT:    vmovd %xmm1, %eax
404; XOPAVX1-NEXT:    movzwl %ax, %eax
405; XOPAVX1-NEXT:    vmovd %eax, %xmm1
406; XOPAVX1-NEXT:    vpsrlw %xmm1, %xmm2, %xmm2
407; XOPAVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
408; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
409; XOPAVX1-NEXT:    retq
410;
411; XOPAVX2-LABEL: splatvar_shift_v16i16:
412; XOPAVX2:       # BB#0:
413; XOPAVX2-NEXT:    vmovd %xmm1, %eax
414; XOPAVX2-NEXT:    movzwl %ax, %eax
415; XOPAVX2-NEXT:    vmovd %eax, %xmm1
416; XOPAVX2-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
417; XOPAVX2-NEXT:    retq
418;
419; AVX512-LABEL: splatvar_shift_v16i16:
420; AVX512:       ## BB#0:
421; AVX512-NEXT:    vmovd %xmm1, %eax
422; AVX512-NEXT:    movzwl %ax, %eax
423; AVX512-NEXT:    vmovd %eax, %xmm1
424; AVX512-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
425; AVX512-NEXT:    retq
426  %splat = shufflevector <16 x i16> %b, <16 x i16> undef, <16 x i32> zeroinitializer
427  %shift = lshr <16 x i16> %a, %splat
428  ret <16 x i16> %shift
429}
430
431define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
432; AVX1-LABEL: splatvar_shift_v32i8:
433; AVX1:       # BB#0:
434; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
435; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
436; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
437; AVX1-NEXT:    vpsrlw $4, %xmm2, %xmm3
438; AVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
439; AVX1-NEXT:    vpand %xmm8, %xmm3, %xmm3
440; AVX1-NEXT:    vpsllw $5, %xmm1, %xmm1
441; AVX1-NEXT:    vpblendvb %xmm1, %xmm3, %xmm2, %xmm2
442; AVX1-NEXT:    vpsrlw $2, %xmm2, %xmm3
443; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
444; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
445; AVX1-NEXT:    vpaddb %xmm1, %xmm1, %xmm6
446; AVX1-NEXT:    vpblendvb %xmm6, %xmm3, %xmm2, %xmm2
447; AVX1-NEXT:    vpsrlw $1, %xmm2, %xmm3
448; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
449; AVX1-NEXT:    vpand %xmm7, %xmm3, %xmm3
450; AVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm4
451; AVX1-NEXT:    vpblendvb %xmm4, %xmm3, %xmm2, %xmm2
452; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm3
453; AVX1-NEXT:    vpand %xmm8, %xmm3, %xmm3
454; AVX1-NEXT:    vpblendvb %xmm1, %xmm3, %xmm0, %xmm0
455; AVX1-NEXT:    vpsrlw $2, %xmm0, %xmm1
456; AVX1-NEXT:    vpand %xmm5, %xmm1, %xmm1
457; AVX1-NEXT:    vpblendvb %xmm6, %xmm1, %xmm0, %xmm0
458; AVX1-NEXT:    vpsrlw $1, %xmm0, %xmm1
459; AVX1-NEXT:    vpand %xmm7, %xmm1, %xmm1
460; AVX1-NEXT:    vpblendvb %xmm4, %xmm1, %xmm0, %xmm0
461; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
462; AVX1-NEXT:    retq
463;
464; AVX2-LABEL: splatvar_shift_v32i8:
465; AVX2:       # BB#0:
466; AVX2-NEXT:    vpbroadcastb %xmm1, %ymm1
467; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm2
468; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
469; AVX2-NEXT:    vpsllw $5, %ymm1, %ymm1
470; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
471; AVX2-NEXT:    vpsrlw $2, %ymm0, %ymm2
472; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
473; AVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
474; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
475; AVX2-NEXT:    vpsrlw $1, %ymm0, %ymm2
476; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
477; AVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
478; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
479; AVX2-NEXT:    retq
480;
481; XOPAVX1-LABEL: splatvar_shift_v32i8:
482; XOPAVX1:       # BB#0:
483; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
484; XOPAVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
485; XOPAVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
486; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
487; XOPAVX1-NEXT:    vpshlb %xmm1, %xmm2, %xmm2
488; XOPAVX1-NEXT:    vpshlb %xmm1, %xmm0, %xmm0
489; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
490; XOPAVX1-NEXT:    retq
491;
492; XOPAVX2-LABEL: splatvar_shift_v32i8:
493; XOPAVX2:       # BB#0:
494; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %ymm1
495; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
496; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
497; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
498; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm4
499; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm4, %xmm2
500; XOPAVX2-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
501; XOPAVX2-NEXT:    vpshlb %xmm1, %xmm0, %xmm0
502; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
503; XOPAVX2-NEXT:    retq
504;
505; AVX512-LABEL: splatvar_shift_v32i8:
506; AVX512:       ## BB#0:
507; AVX512-NEXT:    vpbroadcastb %xmm1, %ymm1
508; AVX512-NEXT:    vpsrlw $4, %ymm0, %ymm2
509; AVX512-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
510; AVX512-NEXT:    vpsllw $5, %ymm1, %ymm1
511; AVX512-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
512; AVX512-NEXT:    vpsrlw $2, %ymm0, %ymm2
513; AVX512-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
514; AVX512-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
515; AVX512-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
516; AVX512-NEXT:    vpsrlw $1, %ymm0, %ymm2
517; AVX512-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
518; AVX512-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
519; AVX512-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
520; AVX512-NEXT:    retq
521  %splat = shufflevector <32 x i8> %b, <32 x i8> undef, <32 x i32> zeroinitializer
522  %shift = lshr <32 x i8> %a, %splat
523  ret <32 x i8> %shift
524}
525
526;
527; Constant Shifts
528;
529
530define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) nounwind {
531; AVX1-LABEL: constant_shift_v4i64:
532; AVX1:       # BB#0:
533; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
534; AVX1-NEXT:    vpsrlq $62, %xmm1, %xmm2
535; AVX1-NEXT:    vpsrlq $31, %xmm1, %xmm1
536; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5,6,7]
537; AVX1-NEXT:    vpsrlq $7, %xmm0, %xmm2
538; AVX1-NEXT:    vpsrlq $1, %xmm0, %xmm0
539; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
540; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
541; AVX1-NEXT:    retq
542;
543; AVX2-LABEL: constant_shift_v4i64:
544; AVX2:       # BB#0:
545; AVX2-NEXT:    vpsrlvq {{.*}}(%rip), %ymm0, %ymm0
546; AVX2-NEXT:    retq
547;
548; XOPAVX1-LABEL: constant_shift_v4i64:
549; XOPAVX1:       # BB#0:
550; XOPAVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
551; XOPAVX1-NEXT:    vpsubq {{.*}}(%rip), %xmm1, %xmm2
552; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
553; XOPAVX1-NEXT:    vpshlq %xmm2, %xmm3, %xmm2
554; XOPAVX1-NEXT:    vpsubq {{.*}}(%rip), %xmm1, %xmm1
555; XOPAVX1-NEXT:    vpshlq %xmm1, %xmm0, %xmm0
556; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
557; XOPAVX1-NEXT:    retq
558;
559; XOPAVX2-LABEL: constant_shift_v4i64:
560; XOPAVX2:       # BB#0:
561; XOPAVX2-NEXT:    vpsrlvq {{.*}}(%rip), %ymm0, %ymm0
562; XOPAVX2-NEXT:    retq
563;
564; AVX512-LABEL: constant_shift_v4i64:
565; AVX512:       ## BB#0:
566; AVX512-NEXT:    vpsrlvq {{.*}}(%rip), %ymm0, %ymm0
567; AVX512-NEXT:    retq
568  %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 31, i64 62>
569  ret <4 x i64> %shift
570}
571
572define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) nounwind {
573; AVX1-LABEL: constant_shift_v8i32:
574; AVX1:       # BB#0:
575; AVX1-NEXT:    vpsrld $7, %xmm0, %xmm1
576; AVX1-NEXT:    vpsrld $5, %xmm0, %xmm2
577; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
578; AVX1-NEXT:    vpsrld $6, %xmm0, %xmm2
579; AVX1-NEXT:    vpsrld $4, %xmm0, %xmm3
580; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
581; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
582; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
583; AVX1-NEXT:    vpsrld $7, %xmm0, %xmm2
584; AVX1-NEXT:    vpsrld $9, %xmm0, %xmm3
585; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
586; AVX1-NEXT:    vpsrld $8, %xmm0, %xmm0
587; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7]
588; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
589; AVX1-NEXT:    retq
590;
591; AVX2-LABEL: constant_shift_v8i32:
592; AVX2:       # BB#0:
593; AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %ymm0, %ymm0
594; AVX2-NEXT:    retq
595;
596; XOPAVX1-LABEL: constant_shift_v8i32:
597; XOPAVX1:       # BB#0:
598; XOPAVX1-NEXT:    vpshld {{.*}}(%rip), %xmm0, %xmm1
599; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
600; XOPAVX1-NEXT:    vpshld {{.*}}(%rip), %xmm0, %xmm0
601; XOPAVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
602; XOPAVX1-NEXT:    retq
603;
604; XOPAVX2-LABEL: constant_shift_v8i32:
605; XOPAVX2:       # BB#0:
606; XOPAVX2-NEXT:    vpsrlvd {{.*}}(%rip), %ymm0, %ymm0
607; XOPAVX2-NEXT:    retq
608;
609; AVX512-LABEL: constant_shift_v8i32:
610; AVX512:       ## BB#0:
611; AVX512-NEXT:    vpsrlvd {{.*}}(%rip), %ymm0, %ymm0
612; AVX512-NEXT:    retq
613  %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 8, i32 7>
614  ret <8 x i32> %shift
615}
616
617define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) nounwind {
618; AVX1-LABEL: constant_shift_v16i16:
619; AVX1:       # BB#0:
620; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
621; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm2
622; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [32896,37008,41120,45232,49344,53456,57568,61680]
623; AVX1-NEXT:    vpblendvb %xmm3, %xmm2, %xmm1, %xmm1
624; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm2
625; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [256,8480,16704,24928,33152,41376,49600,57824]
626; AVX1-NEXT:    vpblendvb %xmm3, %xmm2, %xmm1, %xmm1
627; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm2
628; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [512,16960,33408,49856,768,17216,33664,50112]
629; AVX1-NEXT:    vpblendvb %xmm3, %xmm2, %xmm1, %xmm1
630; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm2
631; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [1024,33920,1280,34176,1536,34432,1792,34688]
632; AVX1-NEXT:    vpblendvb %xmm3, %xmm2, %xmm1, %xmm1
633; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
634; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,4112,8224,12336,16448,20560,24672,28784]
635; AVX1-NEXT:    vpblendvb %xmm3, %xmm2, %xmm0, %xmm0
636; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm2
637; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,8224,16448,24672,32896,41120,49344,57568]
638; AVX1-NEXT:    vpblendvb %xmm3, %xmm2, %xmm0, %xmm0
639; AVX1-NEXT:    vpsrlw $2, %xmm0, %xmm2
640; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,16448,32896,49344,256,16704,33152,49600]
641; AVX1-NEXT:    vpblendvb %xmm3, %xmm2, %xmm0, %xmm0
642; AVX1-NEXT:    vpsrlw $1, %xmm0, %xmm2
643; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,32896,256,33152,512,33408,768,33664]
644; AVX1-NEXT:    vpblendvb %xmm3, %xmm2, %xmm0, %xmm0
645; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
646; AVX1-NEXT:    retq
647;
648; AVX2-LABEL: constant_shift_v16i16:
649; AVX2:       # BB#0:
650; AVX2-NEXT:    vpxor %ymm1, %ymm1, %ymm1
651; AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
652; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm2[4],ymm1[4],ymm2[5],ymm1[5],ymm2[6],ymm1[6],ymm2[7],ymm1[7],ymm2[12],ymm1[12],ymm2[13],ymm1[13],ymm2[14],ymm1[14],ymm2[15],ymm1[15]
653; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm0[4,4,5,5,6,6,7,7,12,12,13,13,14,14,15,15]
654; AVX2-NEXT:    vpsrlvd %ymm3, %ymm4, %ymm3
655; AVX2-NEXT:    vpsrld $16, %ymm3, %ymm3
656; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm2[0],ymm1[0],ymm2[1],ymm1[1],ymm2[2],ymm1[2],ymm2[3],ymm1[3],ymm2[8],ymm1[8],ymm2[9],ymm1[9],ymm2[10],ymm1[10],ymm2[11],ymm1[11]
657; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
658; AVX2-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
659; AVX2-NEXT:    vpsrld $16, %ymm0, %ymm0
660; AVX2-NEXT:    vpackusdw %ymm3, %ymm0, %ymm0
661; AVX2-NEXT:    retq
662;
663; XOPAVX1-LABEL: constant_shift_v16i16:
664; XOPAVX1:       # BB#0:
665; XOPAVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
666; XOPAVX1-NEXT:    vpsubw {{.*}}(%rip), %xmm1, %xmm2
667; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
668; XOPAVX1-NEXT:    vpshlw %xmm2, %xmm3, %xmm2
669; XOPAVX1-NEXT:    vpsubw {{.*}}(%rip), %xmm1, %xmm1
670; XOPAVX1-NEXT:    vpshlw %xmm1, %xmm0, %xmm0
671; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
672; XOPAVX1-NEXT:    retq
673;
674; XOPAVX2-LABEL: constant_shift_v16i16:
675; XOPAVX2:       # BB#0:
676; XOPAVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
677; XOPAVX2-NEXT:    vpsubw {{.*}}(%rip), %xmm1, %xmm2
678; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
679; XOPAVX2-NEXT:    vpshlw %xmm2, %xmm3, %xmm2
680; XOPAVX2-NEXT:    vpsubw {{.*}}(%rip), %xmm1, %xmm1
681; XOPAVX2-NEXT:    vpshlw %xmm1, %xmm0, %xmm0
682; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
683; XOPAVX2-NEXT:    retq
684;
685; AVX512-LABEL: constant_shift_v16i16:
686; AVX512:       ## BB#0:
687; AVX512-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
688; AVX512-NEXT:    vpsrlvw %zmm1, %zmm0, %zmm0
689; AVX512-NEXT:    retq
690  %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
691  ret <16 x i16> %shift
692}
693
694define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) nounwind {
695; AVX1-LABEL: constant_shift_v32i8:
696; AVX1:       # BB#0:
697; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
698; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm2
699; AVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
700; AVX1-NEXT:    vpand %xmm8, %xmm2, %xmm2
701; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [0,1,2,3,4,5,6,7,7,6,5,4,3,2,1,0]
702; AVX1-NEXT:    vpsllw $5, %xmm4, %xmm4
703; AVX1-NEXT:    vpblendvb %xmm4, %xmm2, %xmm1, %xmm1
704; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm2
705; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
706; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
707; AVX1-NEXT:    vpaddb %xmm4, %xmm4, %xmm6
708; AVX1-NEXT:    vpblendvb %xmm6, %xmm2, %xmm1, %xmm1
709; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm2
710; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
711; AVX1-NEXT:    vpand %xmm7, %xmm2, %xmm2
712; AVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm3
713; AVX1-NEXT:    vpblendvb %xmm3, %xmm2, %xmm1, %xmm1
714; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm2
715; AVX1-NEXT:    vpand %xmm8, %xmm2, %xmm2
716; AVX1-NEXT:    vpblendvb %xmm4, %xmm2, %xmm0, %xmm0
717; AVX1-NEXT:    vpsrlw $2, %xmm0, %xmm2
718; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
719; AVX1-NEXT:    vpblendvb %xmm6, %xmm2, %xmm0, %xmm0
720; AVX1-NEXT:    vpsrlw $1, %xmm0, %xmm2
721; AVX1-NEXT:    vpand %xmm7, %xmm2, %xmm2
722; AVX1-NEXT:    vpblendvb %xmm3, %xmm2, %xmm0, %xmm0
723; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
724; AVX1-NEXT:    retq
725;
726; AVX2-LABEL: constant_shift_v32i8:
727; AVX2:       # BB#0:
728; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,1,2,3,4,5,6,7,7,6,5,4,3,2,1,0,0,1,2,3,4,5,6,7,7,6,5,4,3,2,1,0]
729; AVX2-NEXT:    vpsllw $5, %ymm1, %ymm1
730; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm2
731; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
732; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
733; AVX2-NEXT:    vpsrlw $2, %ymm0, %ymm2
734; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
735; AVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
736; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
737; AVX2-NEXT:    vpsrlw $1, %ymm0, %ymm2
738; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
739; AVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
740; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
741; AVX2-NEXT:    retq
742;
743; XOPAVX1-LABEL: constant_shift_v32i8:
744; XOPAVX1:       # BB#0:
745; XOPAVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
746; XOPAVX1-NEXT:    vpsubb {{.*}}(%rip), %xmm1, %xmm1
747; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
748; XOPAVX1-NEXT:    vpshlb %xmm1, %xmm2, %xmm2
749; XOPAVX1-NEXT:    vpshlb %xmm1, %xmm0, %xmm0
750; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
751; XOPAVX1-NEXT:    retq
752;
753; XOPAVX2-LABEL: constant_shift_v32i8:
754; XOPAVX2:       # BB#0:
755; XOPAVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
756; XOPAVX2-NEXT:    vpsubb {{.*}}(%rip), %xmm1, %xmm1
757; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
758; XOPAVX2-NEXT:    vpshlb %xmm1, %xmm2, %xmm2
759; XOPAVX2-NEXT:    vpshlb %xmm1, %xmm0, %xmm0
760; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
761; XOPAVX2-NEXT:    retq
762;
763; AVX512-LABEL: constant_shift_v32i8:
764; AVX512:       ## BB#0:
765; AVX512-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,1,2,3,4,5,6,7,7,6,5,4,3,2,1,0,0,1,2,3,4,5,6,7,7,6,5,4,3,2,1,0]
766; AVX512-NEXT:    vpsllw $5, %ymm1, %ymm1
767; AVX512-NEXT:    vpsrlw $4, %ymm0, %ymm2
768; AVX512-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
769; AVX512-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
770; AVX512-NEXT:    vpsrlw $2, %ymm0, %ymm2
771; AVX512-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
772; AVX512-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
773; AVX512-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
774; AVX512-NEXT:    vpsrlw $1, %ymm0, %ymm2
775; AVX512-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
776; AVX512-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
777; AVX512-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
778; AVX512-NEXT:    retq
779  %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
780  ret <32 x i8> %shift
781}
782
783;
784; Uniform Constant Shifts
785;
786
787define <4 x i64> @splatconstant_shift_v4i64(<4 x i64> %a) nounwind {
788; AVX1-LABEL: splatconstant_shift_v4i64:
789; AVX1:       # BB#0:
790; AVX1-NEXT:    vpsrlq $7, %xmm0, %xmm1
791; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
792; AVX1-NEXT:    vpsrlq $7, %xmm0, %xmm0
793; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
794; AVX1-NEXT:    retq
795;
796; AVX2-LABEL: splatconstant_shift_v4i64:
797; AVX2:       # BB#0:
798; AVX2-NEXT:    vpsrlq $7, %ymm0, %ymm0
799; AVX2-NEXT:    retq
800;
801; XOPAVX1-LABEL: splatconstant_shift_v4i64:
802; XOPAVX1:       # BB#0:
803; XOPAVX1-NEXT:    vpsrlq $7, %xmm0, %xmm1
804; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
805; XOPAVX1-NEXT:    vpsrlq $7, %xmm0, %xmm0
806; XOPAVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
807; XOPAVX1-NEXT:    retq
808;
809; XOPAVX2-LABEL: splatconstant_shift_v4i64:
810; XOPAVX2:       # BB#0:
811; XOPAVX2-NEXT:    vpsrlq $7, %ymm0, %ymm0
812; XOPAVX2-NEXT:    retq
813;
814; AVX512-LABEL: splatconstant_shift_v4i64:
815; AVX512:       ## BB#0:
816; AVX512-NEXT:    vpsrlq $7, %ymm0, %ymm0
817; AVX512-NEXT:    retq
818  %shift = lshr <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7>
819  ret <4 x i64> %shift
820}
821
822define <8 x i32> @splatconstant_shift_v8i32(<8 x i32> %a) nounwind {
823; AVX1-LABEL: splatconstant_shift_v8i32:
824; AVX1:       # BB#0:
825; AVX1-NEXT:    vpsrld $5, %xmm0, %xmm1
826; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
827; AVX1-NEXT:    vpsrld $5, %xmm0, %xmm0
828; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
829; AVX1-NEXT:    retq
830;
831; AVX2-LABEL: splatconstant_shift_v8i32:
832; AVX2:       # BB#0:
833; AVX2-NEXT:    vpsrld $5, %ymm0, %ymm0
834; AVX2-NEXT:    retq
835;
836; XOPAVX1-LABEL: splatconstant_shift_v8i32:
837; XOPAVX1:       # BB#0:
838; XOPAVX1-NEXT:    vpsrld $5, %xmm0, %xmm1
839; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
840; XOPAVX1-NEXT:    vpsrld $5, %xmm0, %xmm0
841; XOPAVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
842; XOPAVX1-NEXT:    retq
843;
844; XOPAVX2-LABEL: splatconstant_shift_v8i32:
845; XOPAVX2:       # BB#0:
846; XOPAVX2-NEXT:    vpsrld $5, %ymm0, %ymm0
847; XOPAVX2-NEXT:    retq
848;
849; AVX512-LABEL: splatconstant_shift_v8i32:
850; AVX512:       ## BB#0:
851; AVX512-NEXT:    vpsrld $5, %ymm0, %ymm0
852; AVX512-NEXT:    retq
853  %shift = lshr <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
854  ret <8 x i32> %shift
855}
856
857define <16 x i16> @splatconstant_shift_v16i16(<16 x i16> %a) nounwind {
858; AVX1-LABEL: splatconstant_shift_v16i16:
859; AVX1:       # BB#0:
860; AVX1-NEXT:    vpsrlw $3, %xmm0, %xmm1
861; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
862; AVX1-NEXT:    vpsrlw $3, %xmm0, %xmm0
863; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
864; AVX1-NEXT:    retq
865;
866; AVX2-LABEL: splatconstant_shift_v16i16:
867; AVX2:       # BB#0:
868; AVX2-NEXT:    vpsrlw $3, %ymm0, %ymm0
869; AVX2-NEXT:    retq
870;
871; XOPAVX1-LABEL: splatconstant_shift_v16i16:
872; XOPAVX1:       # BB#0:
873; XOPAVX1-NEXT:    vpsrlw $3, %xmm0, %xmm1
874; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
875; XOPAVX1-NEXT:    vpsrlw $3, %xmm0, %xmm0
876; XOPAVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
877; XOPAVX1-NEXT:    retq
878;
879; XOPAVX2-LABEL: splatconstant_shift_v16i16:
880; XOPAVX2:       # BB#0:
881; XOPAVX2-NEXT:    vpsrlw $3, %ymm0, %ymm0
882; XOPAVX2-NEXT:    retq
883;
884; AVX512-LABEL: splatconstant_shift_v16i16:
885; AVX512:       ## BB#0:
886; AVX512-NEXT:    vpsrlw $3, %ymm0, %ymm0
887; AVX512-NEXT:    retq
888  %shift = lshr <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
889  ret <16 x i16> %shift
890}
891
892define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) nounwind {
893; AVX1-LABEL: splatconstant_shift_v32i8:
894; AVX1:       # BB#0:
895; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
896; AVX1-NEXT:    vpsrlw $3, %xmm1, %xmm1
897; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
898; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
899; AVX1-NEXT:    vpsrlw $3, %xmm0, %xmm0
900; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
901; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
902; AVX1-NEXT:    retq
903;
904; AVX2-LABEL: splatconstant_shift_v32i8:
905; AVX2:       # BB#0:
906; AVX2-NEXT:    vpsrlw $3, %ymm0, %ymm0
907; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
908; AVX2-NEXT:    retq
909;
910; XOPAVX1-LABEL: splatconstant_shift_v32i8:
911; XOPAVX1:       # BB#0:
912; XOPAVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
913; XOPAVX1-NEXT:    vpsubb {{.*}}(%rip), %xmm1, %xmm1
914; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
915; XOPAVX1-NEXT:    vpshlb %xmm1, %xmm2, %xmm2
916; XOPAVX1-NEXT:    vpshlb %xmm1, %xmm0, %xmm0
917; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
918; XOPAVX1-NEXT:    retq
919;
920; XOPAVX2-LABEL: splatconstant_shift_v32i8:
921; XOPAVX2:       # BB#0:
922; XOPAVX2-NEXT:    vpsrlw $3, %ymm0, %ymm0
923; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
924; XOPAVX2-NEXT:    retq
925;
926; AVX512-LABEL: splatconstant_shift_v32i8:
927; AVX512:       ## BB#0:
928; AVX512-NEXT:    vpsrlw $3, %ymm0, %ymm0
929; AVX512-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
930; AVX512-NEXT:    retq
931  %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
932  ret <32 x i8> %shift
933}
934