xref: /llvm-project/llvm/test/CodeGen/X86/pr78897.ll (revision 30b63def5096cbee448176cc82ca5f6c995f16c4) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
2; RUN: llc < %s -mtriple=i686--   -mcpu=x86-64    | FileCheck %s --check-prefixes=X86-SSE2
3; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64    | FileCheck %s --check-prefixes=X64-SSE2
4; RUN: llc < %s -mtriple=i686--   -mcpu=x86-64-v2 | FileCheck %s --check-prefixes=X86-SSE42
5; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v2 | FileCheck %s --check-prefixes=X64-SSE42
6; RUN: llc < %s -mtriple=i686--   -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=X86-AVX2
7; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=X64-AVX2
8; RUN: llc < %s -mtriple=i686--   -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=X86-AVX512
9; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=X64-AVX512
10
11; PR78897 - Don't vectorize a mul of extracted values if we'd still need the extract.
12; TODO: We should vectorize on 32-bit targets.
13define <16 x i8> @produceShuffleVectorForByte(i8 zeroext %0) nounwind {
14; X86-SSE2-LABEL: produceShuffleVectorForByte:
15; X86-SSE2:       # %bb.0: # %entry
16; X86-SSE2-NEXT:    pushl %ebx
17; X86-SSE2-NEXT:    pushl %edi
18; X86-SSE2-NEXT:    pushl %esi
19; X86-SSE2-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
20; X86-SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
21; X86-SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[0,0,0,0,4,5,6,7]
22; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
23; X86-SSE2-NEXT:    pxor %xmm0, %xmm0
24; X86-SSE2-NEXT:    pcmpeqb %xmm1, %xmm0
25; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [17,17,17,17,17,17,17,17,u,u,u,u,u,u,u,u]
26; X86-SSE2-NEXT:    pand %xmm0, %xmm1
27; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,1,1]
28; X86-SSE2-NEXT:    movd %xmm2, %esi
29; X86-SSE2-NEXT:    movd %xmm1, %ecx
30; X86-SSE2-NEXT:    movl $286331152, %edi # imm = 0x11111110
31; X86-SSE2-NEXT:    movl %ecx, %eax
32; X86-SSE2-NEXT:    mull %edi
33; X86-SSE2-NEXT:    imull $286331153, %ecx, %ebx # imm = 0x11111111
34; X86-SSE2-NEXT:    addl %edx, %ebx
35; X86-SSE2-NEXT:    imull $286331152, %esi, %edx # imm = 0x11111110
36; X86-SSE2-NEXT:    addl %ebx, %edx
37; X86-SSE2-NEXT:    movd %edx, %xmm2
38; X86-SSE2-NEXT:    movd %eax, %xmm1
39; X86-SSE2-NEXT:    xorl $286331153, %ecx # imm = 0x11111111
40; X86-SSE2-NEXT:    movl %ecx, %eax
41; X86-SSE2-NEXT:    mull %edi
42; X86-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
43; X86-SSE2-NEXT:    xorl $17895697, %esi # imm = 0x1111111
44; X86-SSE2-NEXT:    imull $286331153, %ecx, %ecx # imm = 0x11111111
45; X86-SSE2-NEXT:    addl %edx, %ecx
46; X86-SSE2-NEXT:    imull $286331152, %esi, %edx # imm = 0x11111110
47; X86-SSE2-NEXT:    addl %ecx, %edx
48; X86-SSE2-NEXT:    movd %edx, %xmm2
49; X86-SSE2-NEXT:    movd %eax, %xmm3
50; X86-SSE2-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
51; X86-SSE2-NEXT:    pand %xmm0, %xmm1
52; X86-SSE2-NEXT:    pandn %xmm3, %xmm0
53; X86-SSE2-NEXT:    por %xmm1, %xmm0
54; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
55; X86-SSE2-NEXT:    psrlw $4, %xmm1
56; X86-SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
57; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
58; X86-SSE2-NEXT:    popl %esi
59; X86-SSE2-NEXT:    popl %edi
60; X86-SSE2-NEXT:    popl %ebx
61; X86-SSE2-NEXT:    retl
62;
63; X64-SSE2-LABEL: produceShuffleVectorForByte:
64; X64-SSE2:       # %bb.0: # %entry
65; X64-SSE2-NEXT:    movd %edi, %xmm0
66; X64-SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
67; X64-SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[0,0,0,0,4,5,6,7]
68; X64-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
69; X64-SSE2-NEXT:    pxor %xmm0, %xmm0
70; X64-SSE2-NEXT:    pcmpeqb %xmm1, %xmm0
71; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [17,17,17,17,17,17,17,17,u,u,u,u,u,u,u,u]
72; X64-SSE2-NEXT:    pand %xmm0, %xmm1
73; X64-SSE2-NEXT:    movq %xmm1, %rax
74; X64-SSE2-NEXT:    movabsq $1229782938247303440, %rcx # imm = 0x1111111111111110
75; X64-SSE2-NEXT:    movabsq $76861433640456465, %rdx # imm = 0x111111111111111
76; X64-SSE2-NEXT:    xorq %rax, %rdx
77; X64-SSE2-NEXT:    imulq %rcx, %rax
78; X64-SSE2-NEXT:    movq %rax, %xmm1
79; X64-SSE2-NEXT:    imulq %rcx, %rdx
80; X64-SSE2-NEXT:    movq %rdx, %xmm2
81; X64-SSE2-NEXT:    pand %xmm0, %xmm1
82; X64-SSE2-NEXT:    pandn %xmm2, %xmm0
83; X64-SSE2-NEXT:    por %xmm1, %xmm0
84; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
85; X64-SSE2-NEXT:    psrlw $4, %xmm1
86; X64-SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
87; X64-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
88; X64-SSE2-NEXT:    retq
89;
90; X86-SSE42-LABEL: produceShuffleVectorForByte:
91; X86-SSE42:       # %bb.0: # %entry
92; X86-SSE42-NEXT:    pushl %ebx
93; X86-SSE42-NEXT:    pushl %edi
94; X86-SSE42-NEXT:    pushl %esi
95; X86-SSE42-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
96; X86-SSE42-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
97; X86-SSE42-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[0,0,0,0,4,5,6,7]
98; X86-SSE42-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
99; X86-SSE42-NEXT:    pxor %xmm0, %xmm0
100; X86-SSE42-NEXT:    pcmpeqb %xmm1, %xmm0
101; X86-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [17,17,17,17,17,17,17,17,u,u,u,u,u,u,u,u]
102; X86-SSE42-NEXT:    pand %xmm0, %xmm1
103; X86-SSE42-NEXT:    movd %xmm1, %ecx
104; X86-SSE42-NEXT:    movl $286331152, %edi # imm = 0x11111110
105; X86-SSE42-NEXT:    movl %ecx, %eax
106; X86-SSE42-NEXT:    mull %edi
107; X86-SSE42-NEXT:    pextrd $1, %xmm1, %esi
108; X86-SSE42-NEXT:    imull $286331153, %ecx, %ebx # imm = 0x11111111
109; X86-SSE42-NEXT:    addl %edx, %ebx
110; X86-SSE42-NEXT:    imull $286331152, %esi, %edx # imm = 0x11111110
111; X86-SSE42-NEXT:    addl %ebx, %edx
112; X86-SSE42-NEXT:    movd %eax, %xmm2
113; X86-SSE42-NEXT:    pinsrd $1, %edx, %xmm2
114; X86-SSE42-NEXT:    xorl $286331153, %ecx # imm = 0x11111111
115; X86-SSE42-NEXT:    movl %ecx, %eax
116; X86-SSE42-NEXT:    mull %edi
117; X86-SSE42-NEXT:    xorl $17895697, %esi # imm = 0x1111111
118; X86-SSE42-NEXT:    imull $286331153, %ecx, %ecx # imm = 0x11111111
119; X86-SSE42-NEXT:    addl %edx, %ecx
120; X86-SSE42-NEXT:    imull $286331152, %esi, %edx # imm = 0x11111110
121; X86-SSE42-NEXT:    addl %ecx, %edx
122; X86-SSE42-NEXT:    movd %eax, %xmm1
123; X86-SSE42-NEXT:    pinsrd $1, %edx, %xmm1
124; X86-SSE42-NEXT:    pblendvb %xmm0, %xmm2, %xmm1
125; X86-SSE42-NEXT:    movdqa %xmm1, %xmm0
126; X86-SSE42-NEXT:    psrlw $4, %xmm0
127; X86-SSE42-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
128; X86-SSE42-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
129; X86-SSE42-NEXT:    movdqa %xmm1, %xmm0
130; X86-SSE42-NEXT:    popl %esi
131; X86-SSE42-NEXT:    popl %edi
132; X86-SSE42-NEXT:    popl %ebx
133; X86-SSE42-NEXT:    retl
134;
135; X64-SSE42-LABEL: produceShuffleVectorForByte:
136; X64-SSE42:       # %bb.0: # %entry
137; X64-SSE42-NEXT:    movd %edi, %xmm0
138; X64-SSE42-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
139; X64-SSE42-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[0,0,0,0,4,5,6,7]
140; X64-SSE42-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
141; X64-SSE42-NEXT:    pxor %xmm0, %xmm0
142; X64-SSE42-NEXT:    pcmpeqb %xmm1, %xmm0
143; X64-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [17,17,17,17,17,17,17,17,u,u,u,u,u,u,u,u]
144; X64-SSE42-NEXT:    pand %xmm0, %xmm1
145; X64-SSE42-NEXT:    movq %xmm1, %rax
146; X64-SSE42-NEXT:    movabsq $1229782938247303440, %rcx # imm = 0x1111111111111110
147; X64-SSE42-NEXT:    movabsq $76861433640456465, %rdx # imm = 0x111111111111111
148; X64-SSE42-NEXT:    xorq %rax, %rdx
149; X64-SSE42-NEXT:    imulq %rcx, %rax
150; X64-SSE42-NEXT:    movq %rax, %xmm2
151; X64-SSE42-NEXT:    imulq %rcx, %rdx
152; X64-SSE42-NEXT:    movq %rdx, %xmm1
153; X64-SSE42-NEXT:    pblendvb %xmm0, %xmm2, %xmm1
154; X64-SSE42-NEXT:    movdqa %xmm1, %xmm0
155; X64-SSE42-NEXT:    psrlw $4, %xmm0
156; X64-SSE42-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
157; X64-SSE42-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
158; X64-SSE42-NEXT:    movdqa %xmm1, %xmm0
159; X64-SSE42-NEXT:    retq
160;
161; X86-AVX2-LABEL: produceShuffleVectorForByte:
162; X86-AVX2:       # %bb.0: # %entry
163; X86-AVX2-NEXT:    pushl %ebx
164; X86-AVX2-NEXT:    pushl %edi
165; X86-AVX2-NEXT:    pushl %esi
166; X86-AVX2-NEXT:    vpbroadcastb {{[0-9]+}}(%esp), %xmm0
167; X86-AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
168; X86-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
169; X86-AVX2-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
170; X86-AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm1
171; X86-AVX2-NEXT:    vmovd %xmm1, %edx
172; X86-AVX2-NEXT:    movl $286331152, %ecx # imm = 0x11111110
173; X86-AVX2-NEXT:    mulxl %ecx, %edi, %esi
174; X86-AVX2-NEXT:    vpextrd $1, %xmm1, %eax
175; X86-AVX2-NEXT:    imull $286331153, %edx, %ebx # imm = 0x11111111
176; X86-AVX2-NEXT:    addl %esi, %ebx
177; X86-AVX2-NEXT:    imull $286331152, %eax, %esi # imm = 0x11111110
178; X86-AVX2-NEXT:    addl %ebx, %esi
179; X86-AVX2-NEXT:    vmovd %edi, %xmm1
180; X86-AVX2-NEXT:    xorl $286331153, %edx # imm = 0x11111111
181; X86-AVX2-NEXT:    mulxl %ecx, %edi, %ecx
182; X86-AVX2-NEXT:    vpinsrd $1, %esi, %xmm1, %xmm1
183; X86-AVX2-NEXT:    xorl $17895697, %eax # imm = 0x1111111
184; X86-AVX2-NEXT:    imull $286331153, %edx, %edx # imm = 0x11111111
185; X86-AVX2-NEXT:    addl %ecx, %edx
186; X86-AVX2-NEXT:    imull $286331152, %eax, %eax # imm = 0x11111110
187; X86-AVX2-NEXT:    addl %edx, %eax
188; X86-AVX2-NEXT:    vmovd %edi, %xmm2
189; X86-AVX2-NEXT:    vpinsrd $1, %eax, %xmm2, %xmm2
190; X86-AVX2-NEXT:    vpblendvb %xmm0, %xmm1, %xmm2, %xmm0
191; X86-AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm1
192; X86-AVX2-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
193; X86-AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
194; X86-AVX2-NEXT:    popl %esi
195; X86-AVX2-NEXT:    popl %edi
196; X86-AVX2-NEXT:    popl %ebx
197; X86-AVX2-NEXT:    retl
198;
199; X64-AVX2-LABEL: produceShuffleVectorForByte:
200; X64-AVX2:       # %bb.0: # %entry
201; X64-AVX2-NEXT:    vmovd %edi, %xmm0
202; X64-AVX2-NEXT:    vpbroadcastb %xmm0, %xmm0
203; X64-AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
204; X64-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
205; X64-AVX2-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
206; X64-AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1
207; X64-AVX2-NEXT:    vmovq %xmm1, %rax
208; X64-AVX2-NEXT:    movabsq $1229782938247303440, %rcx # imm = 0x1111111111111110
209; X64-AVX2-NEXT:    movabsq $76861433640456465, %rdx # imm = 0x111111111111111
210; X64-AVX2-NEXT:    xorq %rax, %rdx
211; X64-AVX2-NEXT:    imulq %rcx, %rax
212; X64-AVX2-NEXT:    vmovq %rax, %xmm1
213; X64-AVX2-NEXT:    imulq %rcx, %rdx
214; X64-AVX2-NEXT:    vmovq %rdx, %xmm2
215; X64-AVX2-NEXT:    vpblendvb %xmm0, %xmm1, %xmm2, %xmm0
216; X64-AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm1
217; X64-AVX2-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
218; X64-AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
219; X64-AVX2-NEXT:    retq
220;
221; X86-AVX512-LABEL: produceShuffleVectorForByte:
222; X86-AVX512:       # %bb.0: # %entry
223; X86-AVX512-NEXT:    pushl %ebx
224; X86-AVX512-NEXT:    pushl %edi
225; X86-AVX512-NEXT:    pushl %esi
226; X86-AVX512-NEXT:    vpbroadcastb {{[0-9]+}}(%esp), %xmm0
227; X86-AVX512-NEXT:    vptestnmb {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %k1
228; X86-AVX512-NEXT:    vmovdqu8 {{.*#+}} xmm0 {%k1} {z} = [17,17,17,17,17,17,17,17,u,u,u,u,u,u,u,u]
229; X86-AVX512-NEXT:    vpextrd $1, %xmm0, %eax
230; X86-AVX512-NEXT:    vmovd %xmm0, %edx
231; X86-AVX512-NEXT:    movl $286331152, %ecx # imm = 0x11111110
232; X86-AVX512-NEXT:    mulxl %ecx, %edi, %esi
233; X86-AVX512-NEXT:    imull $286331153, %edx, %ebx # imm = 0x11111111
234; X86-AVX512-NEXT:    addl %esi, %ebx
235; X86-AVX512-NEXT:    imull $286331152, %eax, %esi # imm = 0x11111110
236; X86-AVX512-NEXT:    addl %ebx, %esi
237; X86-AVX512-NEXT:    vmovd %edi, %xmm0
238; X86-AVX512-NEXT:    vpinsrd $1, %esi, %xmm0, %xmm0
239; X86-AVX512-NEXT:    xorl $17895697, %eax # imm = 0x1111111
240; X86-AVX512-NEXT:    xorl $286331153, %edx # imm = 0x11111111
241; X86-AVX512-NEXT:    mulxl %ecx, %esi, %ecx
242; X86-AVX512-NEXT:    imull $286331153, %edx, %edx # imm = 0x11111111
243; X86-AVX512-NEXT:    addl %ecx, %edx
244; X86-AVX512-NEXT:    imull $286331152, %eax, %eax # imm = 0x11111110
245; X86-AVX512-NEXT:    addl %edx, %eax
246; X86-AVX512-NEXT:    vmovd %esi, %xmm1
247; X86-AVX512-NEXT:    vpinsrd $1, %eax, %xmm1, %xmm1
248; X86-AVX512-NEXT:    vmovdqu8 %xmm0, %xmm1 {%k1}
249; X86-AVX512-NEXT:    vpsrlw $4, %xmm1, %xmm0
250; X86-AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
251; X86-AVX512-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}{1to4}, %xmm0, %xmm0
252; X86-AVX512-NEXT:    popl %esi
253; X86-AVX512-NEXT:    popl %edi
254; X86-AVX512-NEXT:    popl %ebx
255; X86-AVX512-NEXT:    retl
256;
257; X64-AVX512-LABEL: produceShuffleVectorForByte:
258; X64-AVX512:       # %bb.0: # %entry
259; X64-AVX512-NEXT:    vpbroadcastb %edi, %xmm0
260; X64-AVX512-NEXT:    vptestnmb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %k1
261; X64-AVX512-NEXT:    vmovdqu8 {{.*#+}} xmm0 {%k1} {z} = [17,17,17,17,17,17,17,17,u,u,u,u,u,u,u,u]
262; X64-AVX512-NEXT:    vmovq %xmm0, %rax
263; X64-AVX512-NEXT:    movabsq $1229782938247303440, %rcx # imm = 0x1111111111111110
264; X64-AVX512-NEXT:    movabsq $76861433640456465, %rdx # imm = 0x111111111111111
265; X64-AVX512-NEXT:    xorq %rax, %rdx
266; X64-AVX512-NEXT:    imulq %rcx, %rax
267; X64-AVX512-NEXT:    vmovq %rax, %xmm0
268; X64-AVX512-NEXT:    imulq %rcx, %rdx
269; X64-AVX512-NEXT:    vmovq %rdx, %xmm1
270; X64-AVX512-NEXT:    vmovdqu8 %xmm0, %xmm1 {%k1}
271; X64-AVX512-NEXT:    vpsrlw $4, %xmm1, %xmm0
272; X64-AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
273; X64-AVX512-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0
274; X64-AVX512-NEXT:    retq
275entry:
276  %const = bitcast i64 1229782938247303440 to i64
277  %1 = insertelement <1 x i8> poison, i8 %0, i64 0
278  %2 = shufflevector <1 x i8> %1, <1 x i8> poison, <8 x i32> zeroinitializer
279  %3 = and <8 x i8> %2, <i8 1, i8 2, i8 4, i8 8, i8 16, i8 32, i8 64, i8 -128>
280  %.not.not = icmp eq <8 x i8> %3, zeroinitializer
281  %4 = select <8 x i1> %.not.not, <8 x i8> <i8 17, i8 17, i8 17, i8 17, i8 17, i8 17, i8 17, i8 17>, <8 x i8> zeroinitializer
282  %5 = bitcast <8 x i8> %4 to i64
283  %6 = mul i64 %5, %const
284  %7 = bitcast i64 %6 to <8 x i8>
285  %8 = xor i64 %5, 76861433640456465
286  %9 = mul i64 %8, %const
287  %10 = bitcast i64 %9 to <8 x i8>
288  %11 = select <8 x i1> %.not.not, <8 x i8> %7, <8 x i8> %10
289  %12 = and <8 x i8> %11, <i8 15, i8 15, i8 15, i8 15, i8 15, i8 15, i8 15, i8 15>
290  %13 = lshr <8 x i8> %11, <i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4>
291  %14 = shufflevector <8 x i8> %12, <8 x i8> %13, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>
292  ret <16 x i8> %14
293}
294