| blob | 3699c7f75c86192846c372d2d79fb62315f7dd31 |
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=AVX256VL
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,-prefer-256-bit | FileCheck %s --check-prefixes=AVX512NOBW,AVX512VL
4 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+prefer-256-bit | FileCheck %s --check-prefix=AVX256VLBW
5 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,-prefer-256-bit | FileCheck %s --check-prefix=AVX512VLBW
6 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+prefer-256-bit | FileCheck %s --check-prefixes=AVX512NOBW,AVX512F
7 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,-prefer-256-bit | FileCheck %s --check-prefixes=AVX512NOBW,AVX512F
8 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+prefer-256-bit | FileCheck %s --check-prefix=AVX512BW
9 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,-prefer-256-bit | FileCheck %s --check-prefix=AVX512BW
11 define <16 x i1> @shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0(ptr %a, ptr %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: vmovdqa (%rdi), %ymm0
15 ; AVX256VL-NEXT: vmovdqa (%rsi), %ymm1
16 ; AVX256VL-NEXT: vptestnmd %ymm0, %ymm0, %k1
17 ; AVX256VL-NEXT: vptestnmd %ymm1, %ymm1, %k2
18 ; AVX256VL-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
19 ; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm1 {%k2} {z}
20 ; AVX256VL-NEXT: vpmovdw %ymm1, %xmm1
21 ; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm2 {%k1} {z}
22 ; AVX256VL-NEXT: vpmovdw %ymm2, %xmm2
23 ; AVX256VL-NEXT: vpblendw {{.*#+}} xmm3 = xmm2[0,1],xmm1[2],xmm2[3],xmm1[4],xmm2[5,6,7]
24 ; AVX256VL-NEXT: vpshufb {{.*#+}} xmm3 = xmm3[6,7,12,13,4,5,8,9,6,7,14,15,14,15,0,1]
25 ; AVX256VL-NEXT: vpmovsxwd %xmm3, %ymm3
26 ; AVX256VL-NEXT: vpslld $31, %ymm3, %ymm3
27 ; AVX256VL-NEXT: vptestmd %ymm3, %ymm3, %k1
28 ; AVX256VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,2,1,3]
29 ; AVX256VL-NEXT: vpshufb {{.*#+}} xmm2 = xmm2[6,7,12,13,2,3,u,u,6,7,u,u,14,15,0,1]
30 ; AVX256VL-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1,2],xmm1[3],xmm2[4],xmm1[5],xmm2[6,7]
31 ; AVX256VL-NEXT: vpmovsxwd %xmm1, %ymm1
32 ; AVX256VL-NEXT: vpslld $31, %ymm1, %ymm1
33 ; AVX256VL-NEXT: vptestmd %ymm1, %ymm1, %k0
34 ; AVX256VL-NEXT: kunpckbw %k1, %k0, %k0
35 ; AVX256VL-NEXT: kshiftrw $8, %k0, %k2
36 ; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm1 {%k2} {z}
37 ; AVX256VL-NEXT: vpmovdw %ymm1, %xmm1
38 ; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
39 ; AVX256VL-NEXT: vpmovdw %ymm0, %xmm0
40 ; AVX256VL-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
41 ; AVX256VL-NEXT: vzeroupper
42 ; AVX256VL-NEXT: retq
43 ;
44 ; AVX512VL-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
45 ; AVX512VL: # %bb.0:
46 ; AVX512VL-NEXT: vmovdqa (%rdi), %ymm0
47 ; AVX512VL-NEXT: vmovdqa (%rsi), %ymm1
48 ; AVX512VL-NEXT: vptestnmd %ymm0, %ymm0, %k1
49 ; AVX512VL-NEXT: vptestnmd %ymm1, %ymm1, %k2
50 ; AVX512VL-NEXT: vpternlogd {{.*#+}} zmm0 {%k2} {z} = -1
51 ; AVX512VL-NEXT: vpternlogd {{.*#+}} zmm1 {%k1} {z} = -1
52 ; AVX512VL-NEXT: vpmovsxbd {{.*#+}} 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 {{.*#+}} zmm0 {%k1} {z} = -1
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: vmovdqa (%rdi), %ymm0
63 ; AVX256VLBW-NEXT: vmovdqa (%rsi), %ymm1
64 ; AVX256VLBW-NEXT: vptestnmd %ymm0, %ymm0, %k0
65 ; AVX256VLBW-NEXT: vptestnmd %ymm1, %ymm1, %k1
66 ; AVX256VLBW-NEXT: vpmovm2w %k1, %ymm0
67 ; AVX256VLBW-NEXT: vpmovm2w %k0, %ymm1
68 ; AVX256VLBW-NEXT: vpmovsxbw {{.*#+}} ymm2 = [3,6,18,20,3,7,7,0,3,6,1,21,3,19,7,0]
69 ; AVX256VLBW-NEXT: vpermi2w %ymm0, %ymm1, %ymm2
70 ; AVX256VLBW-NEXT: vpmovw2m %ymm2, %k0
71 ; AVX256VLBW-NEXT: vpmovm2b %k0, %xmm0
72 ; AVX256VLBW-NEXT: vzeroupper
73 ; AVX256VLBW-NEXT: retq
74 ;
75 ; AVX512VLBW-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
76 ; AVX512VLBW: # %bb.0:
77 ; AVX512VLBW-NEXT: vmovdqa (%rdi), %ymm0
78 ; AVX512VLBW-NEXT: vmovdqa (%rsi), %ymm1
79 ; AVX512VLBW-NEXT: vptestnmd %ymm0, %ymm0, %k1
80 ; AVX512VLBW-NEXT: vptestnmd %ymm1, %ymm1, %k2
81 ; AVX512VLBW-NEXT: vpternlogd {{.*#+}} zmm0 {%k2} {z} = -1
82 ; AVX512VLBW-NEXT: vpternlogd {{.*#+}} zmm1 {%k1} {z} = -1
83 ; AVX512VLBW-NEXT: vpmovsxbd {{.*#+}} zmm2 = [3,6,18,20,3,7,7,0,3,6,1,21,3,19,7,0]
84 ; AVX512VLBW-NEXT: vpermi2d %zmm0, %zmm1, %zmm2
85 ; AVX512VLBW-NEXT: vptestmd %zmm2, %zmm2, %k0
86 ; AVX512VLBW-NEXT: vpmovm2b %k0, %xmm0
87 ; AVX512VLBW-NEXT: vzeroupper
88 ; AVX512VLBW-NEXT: retq
89 ;
90 ; AVX512F-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
91 ; AVX512F: # %bb.0:
92 ; AVX512F-NEXT: vmovdqa (%rdi), %ymm0
93 ; AVX512F-NEXT: vmovdqa (%rsi), %ymm1
94 ; AVX512F-NEXT: vptestnmd %zmm0, %zmm0, %k1
95 ; AVX512F-NEXT: vptestnmd %zmm1, %zmm1, %k2
96 ; AVX512F-NEXT: vpternlogd {{.*#+}} zmm0 {%k2} {z} = -1
97 ; AVX512F-NEXT: vpternlogd {{.*#+}} zmm1 {%k1} {z} = -1
98 ; AVX512F-NEXT: vpmovsxbd {{.*#+}} zmm2 = [3,6,18,20,3,7,7,0,3,6,1,21,3,19,7,0]
99 ; AVX512F-NEXT: vpermi2d %zmm0, %zmm1, %zmm2
100 ; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k1
101 ; AVX512F-NEXT: vpternlogd {{.*#+}} zmm0 {%k1} {z} = -1
102 ; AVX512F-NEXT: vpmovdb %zmm0, %xmm0
103 ; AVX512F-NEXT: vzeroupper
104 ; AVX512F-NEXT: retq
105 ;
106 ; AVX512BW-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:
107 ; AVX512BW: # %bb.0:
108 ; AVX512BW-NEXT: vmovdqa (%rdi), %ymm0
109 ; AVX512BW-NEXT: vmovdqa (%rsi), %ymm1
110 ; AVX512BW-NEXT: vptestnmd %zmm0, %zmm0, %k1
111 ; AVX512BW-NEXT: vptestnmd %zmm1, %zmm1, %k2
112 ; AVX512BW-NEXT: vpternlogd {{.*#+}} zmm0 {%k2} {z} = -1
113 ; AVX512BW-NEXT: vpternlogd {{.*#+}} zmm1 {%k1} {z} = -1
114 ; AVX512BW-NEXT: vpmovsxbd {{.*#+}} zmm2 = [3,6,18,20,3,7,7,0,3,6,1,21,3,19,7,0]
115 ; AVX512BW-NEXT: vpermi2d %zmm0, %zmm1, %zmm2
116 ; AVX512BW-NEXT: vptestmd %zmm2, %zmm2, %k0
117 ; AVX512BW-NEXT: vpmovm2b %k0, %zmm0
118 ; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
119 ; AVX512BW-NEXT: vzeroupper
120 ; AVX512BW-NEXT: retq
122 %a1 = load <8 x i32>, ptr %a
123 %b1 = load <8 x i32>, ptr %b
124 %a2 = icmp eq <8 x i32> %a1, zeroinitializer
125 %b2 = icmp eq <8 x i32> %b1, zeroinitializer
126 %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>
127 ret <16 x i1> %c
128 }
130 define <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) {
131 ; 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:
132 ; AVX256VL: # %bb.0:
133 ; AVX256VL-NEXT: vpxor %xmm1, %xmm1, %xmm1
134 ; AVX256VL-NEXT: vpcmpeqb %ymm1, %ymm0, %ymm0
135 ; AVX256VL-NEXT: vextracti128 $1, %ymm0, %xmm1
136 ; AVX256VL-NEXT: vpmovsxbd %xmm1, %ymm1
137 ; AVX256VL-NEXT: vptestmd %ymm1, %ymm1, %k1
138 ; AVX256VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
139 ; AVX256VL-NEXT: vpmovsxbd %xmm1, %ymm1
140 ; AVX256VL-NEXT: vptestmd %ymm1, %ymm1, %k2
141 ; AVX256VL-NEXT: vpmovsxbd %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],zero,zero,ymm1[8,9,6,7,14,15,14,15,0,1,22,23,28,29,18,19,26,27,22,23],zero,zero,ymm1[30,31,16,17]
152 ; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm2 {%k1} {z}
153 ; AVX256VL-NEXT: vpmovdw %ymm2, %xmm2
154 ; AVX256VL-NEXT: vpermq {{.*#+}} ymm2 = ymm2[1,1,1,1]
155 ; AVX256VL-NEXT: vpternlogq {{.*#+}} ymm2 = (ymm2 & ~mem) | ymm1
156 ; AVX256VL-NEXT: vpmovsxwd %xmm2, %ymm1
157 ; AVX256VL-NEXT: vpslld $31, %ymm1, %ymm1
158 ; AVX256VL-NEXT: vptestmd %ymm1, %ymm1, %k1
159 ; AVX256VL-NEXT: vextracti128 $1, %ymm2, %xmm1
160 ; AVX256VL-NEXT: vpmovsxwd %xmm1, %ymm1
161 ; AVX256VL-NEXT: vpslld $31, %ymm1, %ymm1
162 ; AVX256VL-NEXT: vptestmd %ymm1, %ymm1, %k0
163 ; AVX256VL-NEXT: kunpckbw %k1, %k0, %k0
164 ; AVX256VL-NEXT: kshiftrw $8, %k0, %k2
165 ; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm1 {%k2} {z}
166 ; AVX256VL-NEXT: vpmovdw %ymm1, %xmm1
167 ; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
168 ; AVX256VL-NEXT: vpmovdw %ymm0, %xmm0
169 ; AVX256VL-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
170 ; AVX256VL-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
171 ; AVX256VL-NEXT: retq
172 ;
173 ; 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:
174 ; AVX512NOBW: # %bb.0:
175 ; AVX512NOBW-NEXT: vpxor %xmm1, %xmm1, %xmm1
176 ; AVX512NOBW-NEXT: vpcmpeqb %ymm1, %ymm0, %ymm0
177 ; AVX512NOBW-NEXT: vpmovsxbd %xmm0, %zmm1
178 ; AVX512NOBW-NEXT: vptestmd %zmm1, %zmm1, %k1
179 ; AVX512NOBW-NEXT: vextracti128 $1, %ymm0, %xmm0
180 ; AVX512NOBW-NEXT: vpmovsxbd %xmm0, %zmm0
181 ; AVX512NOBW-NEXT: vptestmd %zmm0, %zmm0, %k2
182 ; AVX512NOBW-NEXT: vpternlogd {{.*#+}} zmm0 {%k2} {z} = -1
183 ; AVX512NOBW-NEXT: vpternlogd {{.*#+}} zmm1 {%k1} {z} = -1
184 ; AVX512NOBW-NEXT: vpmovsxbd {{.*#+}} zmm2 = [3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0]
185 ; AVX512NOBW-NEXT: vpermi2d %zmm0, %zmm1, %zmm2
186 ; AVX512NOBW-NEXT: vptestmd %zmm2, %zmm2, %k1
187 ; AVX512NOBW-NEXT: vpternlogd {{.*#+}} zmm0 {%k1} {z} = -1
188 ; AVX512NOBW-NEXT: vpmovdb %zmm0, %xmm0
189 ; AVX512NOBW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
190 ; AVX512NOBW-NEXT: retq
191 ;
192 ; 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:
193 ; AVX256VLBW: # %bb.0:
194 ; AVX256VLBW-NEXT: vptestnmb %ymm0, %ymm0, %k0
195 ; AVX256VLBW-NEXT: vpmovm2b %k0, %ymm0
196 ; AVX256VLBW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,2,3]
197 ; AVX256VLBW-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[1,1,1,1,5,5,5,5]
198 ; AVX256VLBW-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
199 ; AVX256VLBW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[3,6,u,12,3,7,7,0,3,6,1,13,3,u,7,0,19,22,u,28,19,23,23,16,19,22,17,29,19,u,23,16]
200 ; AVX256VLBW-NEXT: movl $537141252, %eax # imm = 0x20042004
201 ; AVX256VLBW-NEXT: kmovd %eax, %k1
202 ; AVX256VLBW-NEXT: vmovdqu8 %ymm1, %ymm0 {%k1}
203 ; AVX256VLBW-NEXT: vpmovb2m %ymm0, %k0
204 ; AVX256VLBW-NEXT: vpmovm2b %k0, %ymm0
205 ; AVX256VLBW-NEXT: retq
206 ;
207 ; 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:
208 ; AVX512VLBW: # %bb.0:
209 ; AVX512VLBW-NEXT: vptestnmb %ymm0, %ymm0, %k0
210 ; AVX512VLBW-NEXT: vpmovm2w %k0, %zmm0
211 ; AVX512VLBW-NEXT: vbroadcasti64x4 {{.*#+}} 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]
212 ; AVX512VLBW-NEXT: # zmm1 = mem[0,1,2,3,0,1,2,3]
213 ; AVX512VLBW-NEXT: vpermw %zmm0, %zmm1, %zmm0
214 ; AVX512VLBW-NEXT: vpmovw2m %zmm0, %k0
215 ; AVX512VLBW-NEXT: vpmovm2b %k0, %ymm0
216 ; AVX512VLBW-NEXT: retq
217 ;
218 ; 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:
219 ; AVX512BW: # %bb.0:
220 ; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
221 ; AVX512BW-NEXT: vptestnmb %zmm0, %zmm0, %k0
222 ; AVX512BW-NEXT: vpmovm2w %k0, %zmm0
223 ; AVX512BW-NEXT: vbroadcasti64x4 {{.*#+}} 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]
224 ; AVX512BW-NEXT: # zmm1 = mem[0,1,2,3,0,1,2,3]
225 ; AVX512BW-NEXT: vpermw %zmm0, %zmm1, %zmm0
226 ; AVX512BW-NEXT: vpmovw2m %zmm0, %k0
227 ; AVX512BW-NEXT: vpmovm2b %k0, %zmm0
228 ; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0
229 ; AVX512BW-NEXT: retq
230 %cmp = icmp eq <32 x i8> %a, zeroinitializer
231 %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>
232 ret <32 x i1> %b
233 }