• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+prefer-256-bit | FileCheck %s --check-prefix=CHECK --check-prefix=AVX256 --check-prefix=AVX256VL
3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,-prefer-256-bit | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512NOBW --check-prefix=AVX512VL
4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+prefer-256-bit | FileCheck %s --check-prefix=CHECK --check-prefix=AVX256 --check-prefix=AVX256VLBW
5; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,-prefer-256-bit | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512VLBW
6; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+prefer-256-bit | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512NOBW --check-prefix=AVX512F
7; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,-prefer-256-bit | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512NOBW --check-prefix=AVX512F
8; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+prefer-256-bit | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512BW
9; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,-prefer-256-bit | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512BW
10
11define <16 x i1> @shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0(<8 x i32>* %a, <8 x i32>* %b) {
12; AVX256VL-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
13; AVX256VL:       # %bb.0:
14; AVX256VL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
15; AVX256VL-NEXT:    vpcmpeqd (%rdi), %ymm0, %k1
16; AVX256VL-NEXT:    vpcmpeqd (%rsi), %ymm0, %k2
17; AVX256VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
18; AVX256VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k2} {z}
19; AVX256VL-NEXT:    vpmovdw %ymm1, %xmm1
20; AVX256VL-NEXT:    vmovdqa32 %ymm0, %ymm2 {%k1} {z}
21; AVX256VL-NEXT:    vpmovdw %ymm2, %xmm2
22; AVX256VL-NEXT:    vpblendw {{.*#+}} xmm3 = xmm2[0,1],xmm1[2],xmm2[3],xmm1[4],xmm2[5,6,7]
23; AVX256VL-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[6,7,12,13,4,5,8,9,6,7,14,15,14,15,0,1]
24; AVX256VL-NEXT:    vpmovsxwd %xmm3, %ymm3
25; AVX256VL-NEXT:    vpslld $31, %ymm3, %ymm3
26; AVX256VL-NEXT:    vptestmd %ymm3, %ymm3, %k1
27; AVX256VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,1,3]
28; AVX256VL-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[6,7,12,13,2,3,14,15,6,7,6,7,14,15,0,1]
29; AVX256VL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1,2],xmm1[3],xmm2[4],xmm1[5],xmm2[6,7]
30; AVX256VL-NEXT:    vpmovsxwd %xmm1, %ymm1
31; AVX256VL-NEXT:    vpslld $31, %ymm1, %ymm1
32; AVX256VL-NEXT:    vptestmd %ymm1, %ymm1, %k0
33; AVX256VL-NEXT:    kunpckbw %k1, %k0, %k0
34; AVX256VL-NEXT:    kshiftrw $8, %k0, %k2
35; AVX256VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k2} {z}
36; AVX256VL-NEXT:    vpmovdw %ymm1, %xmm1
37; AVX256VL-NEXT:    vpacksswb %xmm0, %xmm1, %xmm1
38; AVX256VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
39; AVX256VL-NEXT:    vpmovdw %ymm0, %xmm0
40; AVX256VL-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
41; AVX256VL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
42; AVX256VL-NEXT:    vzeroupper
43; AVX256VL-NEXT:    retq
44;
45; AVX512VL-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
46; AVX512VL:       # %bb.0:
47; AVX512VL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
48; AVX512VL-NEXT:    vpcmpeqd (%rdi), %ymm0, %k1
49; AVX512VL-NEXT:    vpcmpeqd (%rsi), %ymm0, %k2
50; AVX512VL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
51; AVX512VL-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
52; AVX512VL-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [3,6,18,20,3,7,7,0,3,6,1,21,3,19,7,0]
53; AVX512VL-NEXT:    vpermi2d %zmm0, %zmm1, %zmm2
54; AVX512VL-NEXT:    vptestmd %zmm2, %zmm2, %k1
55; AVX512VL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
56; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
57; AVX512VL-NEXT:    vzeroupper
58; AVX512VL-NEXT:    retq
59;
60; AVX256VLBW-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
61; AVX256VLBW:       # %bb.0:
62; AVX256VLBW-NEXT:    vpxor %xmm0, %xmm0, %xmm0
63; AVX256VLBW-NEXT:    vpcmpeqd (%rdi), %ymm0, %k0
64; AVX256VLBW-NEXT:    vpcmpeqd (%rsi), %ymm0, %k1
65; AVX256VLBW-NEXT:    vpmovm2w %k1, %ymm0
66; AVX256VLBW-NEXT:    vpmovm2w %k0, %ymm1
67; AVX256VLBW-NEXT:    vmovdqa {{.*#+}} ymm2 = [3,6,18,20,3,7,7,0,3,6,1,21,3,19,7,0]
68; AVX256VLBW-NEXT:    vpermi2w %ymm0, %ymm1, %ymm2
69; AVX256VLBW-NEXT:    vpmovw2m %ymm2, %k0
70; AVX256VLBW-NEXT:    vpmovm2b %k0, %xmm0
71; AVX256VLBW-NEXT:    vzeroupper
72; AVX256VLBW-NEXT:    retq
73;
74; AVX512VLBW-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
75; AVX512VLBW:       # %bb.0:
76; AVX512VLBW-NEXT:    vpxor %xmm0, %xmm0, %xmm0
77; AVX512VLBW-NEXT:    vpcmpeqd (%rdi), %ymm0, %k1
78; AVX512VLBW-NEXT:    vpcmpeqd (%rsi), %ymm0, %k2
79; AVX512VLBW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
80; AVX512VLBW-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
81; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [3,6,18,20,3,7,7,0,3,6,1,21,3,19,7,0]
82; AVX512VLBW-NEXT:    vpermi2d %zmm0, %zmm1, %zmm2
83; AVX512VLBW-NEXT:    vptestmd %zmm2, %zmm2, %k0
84; AVX512VLBW-NEXT:    vpmovm2b %k0, %xmm0
85; AVX512VLBW-NEXT:    vzeroupper
86; AVX512VLBW-NEXT:    retq
87;
88; AVX512F-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
89; AVX512F:       # %bb.0:
90; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
91; AVX512F-NEXT:    vmovdqa (%rsi), %ymm1
92; AVX512F-NEXT:    vptestnmd %zmm0, %zmm0, %k1
93; AVX512F-NEXT:    vptestnmd %zmm1, %zmm1, %k2
94; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
95; AVX512F-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
96; AVX512F-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [3,6,18,20,3,7,7,0,3,6,1,21,3,19,7,0]
97; AVX512F-NEXT:    vpermi2d %zmm0, %zmm1, %zmm2
98; AVX512F-NEXT:    vptestmd %zmm2, %zmm2, %k1
99; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
100; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
101; AVX512F-NEXT:    vzeroupper
102; AVX512F-NEXT:    retq
103;
104; AVX512BW-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
105; AVX512BW:       # %bb.0:
106; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
107; AVX512BW-NEXT:    vmovdqa (%rsi), %ymm1
108; AVX512BW-NEXT:    vptestnmd %zmm0, %zmm0, %k1
109; AVX512BW-NEXT:    vptestnmd %zmm1, %zmm1, %k2
110; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
111; AVX512BW-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
112; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [3,6,18,20,3,7,7,0,3,6,1,21,3,19,7,0]
113; AVX512BW-NEXT:    vpermi2d %zmm0, %zmm1, %zmm2
114; AVX512BW-NEXT:    vptestmd %zmm2, %zmm2, %k0
115; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
116; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
117; AVX512BW-NEXT:    vzeroupper
118; AVX512BW-NEXT:    retq
119
120  %a1 = load <8 x i32>, <8 x i32>* %a
121  %b1 = load <8 x i32>, <8 x i32>* %b
122  %a2 = icmp eq <8 x i32> %a1, zeroinitializer
123  %b2 = icmp eq <8 x i32> %b1, zeroinitializer
124  %c = shufflevector <8 x i1> %a2, <8 x i1> %b2, <16 x i32> <i32 3, i32 6, i32 10, i32 12, i32 3, i32 7, i32 7, i32 0, i32 3, i32 6, i32 1, i32 13, i32 3, i32 11, i32 7, i32 0>
125  ret <16 x i1> %c
126}
127
128define <32 x i1> @shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0(<32 x i8> %a) {
129; AVX256VL-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
130; AVX256VL:       # %bb.0:
131; AVX256VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
132; AVX256VL-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
133; AVX256VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
134; AVX256VL-NEXT:    vpmovsxbw %xmm1, %xmm1
135; AVX256VL-NEXT:    vpmovsxwd %xmm1, %ymm1
136; AVX256VL-NEXT:    vptestmd %ymm1, %ymm1, %k1
137; AVX256VL-NEXT:    vpmovsxbw %xmm0, %ymm0
138; AVX256VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
139; AVX256VL-NEXT:    vpmovsxwd %xmm1, %ymm1
140; AVX256VL-NEXT:    vptestmd %ymm1, %ymm1, %k2
141; AVX256VL-NEXT:    vpmovsxwd %xmm0, %ymm0
142; AVX256VL-NEXT:    vptestmd %ymm0, %ymm0, %k3
143; AVX256VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
144; AVX256VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k3} {z}
145; AVX256VL-NEXT:    vpmovdw %ymm1, %xmm1
146; AVX256VL-NEXT:    vmovdqa32 %ymm0, %ymm2 {%k2} {z}
147; AVX256VL-NEXT:    vpmovdw %ymm2, %xmm2
148; AVX256VL-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
149; AVX256VL-NEXT:    vpermq {{.*#+}} ymm2 = ymm1[2,3,0,1]
150; AVX256VL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
151; AVX256VL-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[6,7,12,13,u,u,8,9,6,7,14,15,14,15,0,1,22,23,28,29,18,19,26,27,22,23,u,u,30,31,16,17]
152; AVX256VL-NEXT:    vmovdqa32 %ymm0, %ymm2 {%k1} {z}
153; AVX256VL-NEXT:    vpmovdw %ymm2, %xmm2
154; AVX256VL-NEXT:    kshiftrw $8, %k1, %k1
155; AVX256VL-NEXT:    vmovdqa32 %ymm0, %ymm3 {%k1} {z}
156; AVX256VL-NEXT:    vpmovdw %ymm3, %xmm3
157; AVX256VL-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
158; AVX256VL-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[1,1,2,1]
159; AVX256VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255]
160; AVX256VL-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
161; AVX256VL-NEXT:    vpmovsxwd %xmm1, %ymm2
162; AVX256VL-NEXT:    vpslld $31, %ymm2, %ymm2
163; AVX256VL-NEXT:    vptestmd %ymm2, %ymm2, %k1
164; AVX256VL-NEXT:    vextracti128 $1, %ymm1, %xmm1
165; AVX256VL-NEXT:    vpmovsxwd %xmm1, %ymm1
166; AVX256VL-NEXT:    vpslld $31, %ymm1, %ymm1
167; AVX256VL-NEXT:    vptestmd %ymm1, %ymm1, %k0
168; AVX256VL-NEXT:    kunpckbw %k1, %k0, %k0
169; AVX256VL-NEXT:    kshiftrw $8, %k0, %k2
170; AVX256VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k2} {z}
171; AVX256VL-NEXT:    vpmovdw %ymm1, %xmm1
172; AVX256VL-NEXT:    vpacksswb %xmm0, %xmm1, %xmm1
173; AVX256VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
174; AVX256VL-NEXT:    vpmovdw %ymm0, %xmm0
175; AVX256VL-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
176; AVX256VL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
177; AVX256VL-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
178; AVX256VL-NEXT:    retq
179;
180; AVX512NOBW-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
181; AVX512NOBW:       # %bb.0:
182; AVX512NOBW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
183; AVX512NOBW-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
184; AVX512NOBW-NEXT:    vpmovsxbd %xmm0, %zmm1
185; AVX512NOBW-NEXT:    vptestmd %zmm1, %zmm1, %k1
186; AVX512NOBW-NEXT:    vextracti128 $1, %ymm0, %xmm0
187; AVX512NOBW-NEXT:    vpmovsxbd %xmm0, %zmm0
188; AVX512NOBW-NEXT:    vptestmd %zmm0, %zmm0, %k2
189; AVX512NOBW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
190; AVX512NOBW-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
191; AVX512NOBW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0]
192; AVX512NOBW-NEXT:    vpermi2d %zmm0, %zmm1, %zmm2
193; AVX512NOBW-NEXT:    vptestmd %zmm2, %zmm2, %k1
194; AVX512NOBW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
195; AVX512NOBW-NEXT:    vpmovdb %zmm0, %xmm0
196; AVX512NOBW-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
197; AVX512NOBW-NEXT:    retq
198;
199; AVX256VLBW-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
200; AVX256VLBW:       # %bb.0:
201; AVX256VLBW-NEXT:    vptestnmb %ymm0, %ymm0, %k0
202; AVX256VLBW-NEXT:    vpmovm2b %k0, %ymm0
203; AVX256VLBW-NEXT:    vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
204; AVX256VLBW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[3,6,u,12,3,7,7,0,3,6,1,13,3,u,7,0,u,u,22,u,u,u,u,u,u,u,u,u,u,21,u,u]
205; AVX256VLBW-NEXT:    movl $-537190396, %eax # imm = 0xDFFB2004
206; AVX256VLBW-NEXT:    kmovd %eax, %k1
207; AVX256VLBW-NEXT:    vpshufb {{.*#+}} ymm0 {%k1} = ymm1[u,u,6,u,u,u,u,u,u,u,u,u,u,5,u,u,19,22,u,28,19,23,23,16,19,22,17,29,19,u,23,16]
208; AVX256VLBW-NEXT:    vpmovb2m %ymm0, %k0
209; AVX256VLBW-NEXT:    vpmovm2b %k0, %ymm0
210; AVX256VLBW-NEXT:    retq
211;
212; AVX512VLBW-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
213; AVX512VLBW:       # %bb.0:
214; AVX512VLBW-NEXT:    vptestnmb %ymm0, %ymm0, %k0
215; AVX512VLBW-NEXT:    vpmovm2w %k0, %zmm0
216; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0,3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0]
217; AVX512VLBW-NEXT:    vpermw %zmm0, %zmm1, %zmm0
218; AVX512VLBW-NEXT:    vpmovw2m %zmm0, %k0
219; AVX512VLBW-NEXT:    vpmovm2b %k0, %ymm0
220; AVX512VLBW-NEXT:    retq
221;
222; AVX512BW-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
223; AVX512BW:       # %bb.0:
224; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
225; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm0, %k0
226; AVX512BW-NEXT:    vpmovm2w %k0, %zmm0
227; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0,3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0]
228; AVX512BW-NEXT:    vpermw %zmm0, %zmm1, %zmm0
229; AVX512BW-NEXT:    vpmovw2m %zmm0, %k0
230; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
231; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
232; AVX512BW-NEXT:    retq
233  %cmp = icmp eq <32 x i8> %a, zeroinitializer
234  %b = shufflevector <32 x i1> %cmp, <32 x i1> undef, <32 x i32> <i32 3, i32 6, i32 22, i32 12, i32 3, i32 7, i32 7, i32 0, i32 3, i32 6, i32 1, i32 13, i32 3, i32 21, i32 7, i32 0, i32 3, i32 6, i32 22, i32 12, i32 3, i32 7, i32 7, i32 0, i32 3, i32 6, i32 1, i32 13, i32 3, i32 21, i32 7, i32 0>
235  ret <32 x i1> %b
236}
237
238