• 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=+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=+avx512vpopcntdq | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX512VPOPCNTDQ
5; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vpopcntdq,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX512VPOPCNTDQVL
6; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bitalg | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=BITALG_NOVLX
7; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bitalg,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=BITALG
8
9define <4 x i64> @testv4i64(<4 x i64> %in) nounwind {
10; AVX1-LABEL: testv4i64:
11; AVX1:       # %bb.0:
12; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
13; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
14; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm3
15; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
16; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
17; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm1
18; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
19; AVX1-NEXT:    vpshufb %xmm1, %xmm4, %xmm1
20; AVX1-NEXT:    vpaddb %xmm3, %xmm1, %xmm1
21; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
22; AVX1-NEXT:    vpsadbw %xmm3, %xmm1, %xmm1
23; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm5
24; AVX1-NEXT:    vpshufb %xmm5, %xmm4, %xmm5
25; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
26; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
27; AVX1-NEXT:    vpshufb %xmm0, %xmm4, %xmm0
28; AVX1-NEXT:    vpaddb %xmm5, %xmm0, %xmm0
29; AVX1-NEXT:    vpsadbw %xmm3, %xmm0, %xmm0
30; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
31; AVX1-NEXT:    retq
32;
33; AVX2-LABEL: testv4i64:
34; AVX2:       # %bb.0:
35; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
36; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
37; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
38; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
39; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
40; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
41; AVX2-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
42; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
43; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
44; AVX2-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
45; AVX2-NEXT:    retq
46;
47; AVX512VPOPCNTDQ-LABEL: testv4i64:
48; AVX512VPOPCNTDQ:       # %bb.0:
49; AVX512VPOPCNTDQ-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
50; AVX512VPOPCNTDQ-NEXT:    vpopcntq %zmm0, %zmm0
51; AVX512VPOPCNTDQ-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
52; AVX512VPOPCNTDQ-NEXT:    retq
53;
54; AVX512VPOPCNTDQVL-LABEL: testv4i64:
55; AVX512VPOPCNTDQVL:       # %bb.0:
56; AVX512VPOPCNTDQVL-NEXT:    vpopcntq %ymm0, %ymm0
57; AVX512VPOPCNTDQVL-NEXT:    retq
58;
59; BITALG_NOVLX-LABEL: testv4i64:
60; BITALG_NOVLX:       # %bb.0:
61; BITALG_NOVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
62; BITALG_NOVLX-NEXT:    vpand %ymm1, %ymm0, %ymm2
63; BITALG_NOVLX-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
64; BITALG_NOVLX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
65; BITALG_NOVLX-NEXT:    vpsrlw $4, %ymm0, %ymm0
66; BITALG_NOVLX-NEXT:    vpand %ymm1, %ymm0, %ymm0
67; BITALG_NOVLX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
68; BITALG_NOVLX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
69; BITALG_NOVLX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
70; BITALG_NOVLX-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
71; BITALG_NOVLX-NEXT:    retq
72;
73; BITALG-LABEL: testv4i64:
74; BITALG:       # %bb.0:
75; BITALG-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
76; BITALG-NEXT:    vpand %ymm1, %ymm0, %ymm2
77; BITALG-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
78; BITALG-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
79; BITALG-NEXT:    vpsrlw $4, %ymm0, %ymm0
80; BITALG-NEXT:    vpand %ymm1, %ymm0, %ymm0
81; BITALG-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
82; BITALG-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
83; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
84; BITALG-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
85; BITALG-NEXT:    retq
86  %out = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %in)
87  ret <4 x i64> %out
88}
89
90define <8 x i32> @testv8i32(<8 x i32> %in) nounwind {
91; AVX1-LABEL: testv8i32:
92; AVX1:       # %bb.0:
93; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
94; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
95; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm3
96; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
97; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
98; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm1
99; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
100; AVX1-NEXT:    vpshufb %xmm1, %xmm4, %xmm1
101; AVX1-NEXT:    vpaddb %xmm3, %xmm1, %xmm1
102; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
103; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm5 = xmm1[2],xmm3[2],xmm1[3],xmm3[3]
104; AVX1-NEXT:    vpsadbw %xmm3, %xmm5, %xmm5
105; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
106; AVX1-NEXT:    vpsadbw %xmm3, %xmm1, %xmm1
107; AVX1-NEXT:    vpackuswb %xmm5, %xmm1, %xmm1
108; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm5
109; AVX1-NEXT:    vpshufb %xmm5, %xmm4, %xmm5
110; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
111; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
112; AVX1-NEXT:    vpshufb %xmm0, %xmm4, %xmm0
113; AVX1-NEXT:    vpaddb %xmm5, %xmm0, %xmm0
114; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm3[2],xmm0[3],xmm3[3]
115; AVX1-NEXT:    vpsadbw %xmm3, %xmm2, %xmm2
116; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
117; AVX1-NEXT:    vpsadbw %xmm3, %xmm0, %xmm0
118; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
119; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
120; AVX1-NEXT:    retq
121;
122; AVX2-LABEL: testv8i32:
123; AVX2:       # %bb.0:
124; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
125; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
126; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
127; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
128; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
129; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
130; AVX2-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
131; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
132; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
133; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
134; AVX2-NEXT:    vpsadbw %ymm1, %ymm2, %ymm2
135; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
136; AVX2-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
137; AVX2-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
138; AVX2-NEXT:    retq
139;
140; AVX512VPOPCNTDQ-LABEL: testv8i32:
141; AVX512VPOPCNTDQ:       # %bb.0:
142; AVX512VPOPCNTDQ-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
143; AVX512VPOPCNTDQ-NEXT:    vpopcntd %zmm0, %zmm0
144; AVX512VPOPCNTDQ-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
145; AVX512VPOPCNTDQ-NEXT:    retq
146;
147; AVX512VPOPCNTDQVL-LABEL: testv8i32:
148; AVX512VPOPCNTDQVL:       # %bb.0:
149; AVX512VPOPCNTDQVL-NEXT:    vpopcntd %ymm0, %ymm0
150; AVX512VPOPCNTDQVL-NEXT:    retq
151;
152; BITALG_NOVLX-LABEL: testv8i32:
153; BITALG_NOVLX:       # %bb.0:
154; BITALG_NOVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
155; BITALG_NOVLX-NEXT:    vpand %ymm1, %ymm0, %ymm2
156; BITALG_NOVLX-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
157; BITALG_NOVLX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
158; BITALG_NOVLX-NEXT:    vpsrlw $4, %ymm0, %ymm0
159; BITALG_NOVLX-NEXT:    vpand %ymm1, %ymm0, %ymm0
160; BITALG_NOVLX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
161; BITALG_NOVLX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
162; BITALG_NOVLX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
163; BITALG_NOVLX-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
164; BITALG_NOVLX-NEXT:    vpsadbw %ymm1, %ymm2, %ymm2
165; BITALG_NOVLX-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
166; BITALG_NOVLX-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
167; BITALG_NOVLX-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
168; BITALG_NOVLX-NEXT:    retq
169;
170; BITALG-LABEL: testv8i32:
171; BITALG:       # %bb.0:
172; BITALG-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
173; BITALG-NEXT:    vpand %ymm1, %ymm0, %ymm2
174; BITALG-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
175; BITALG-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
176; BITALG-NEXT:    vpsrlw $4, %ymm0, %ymm0
177; BITALG-NEXT:    vpand %ymm1, %ymm0, %ymm0
178; BITALG-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
179; BITALG-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
180; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
181; BITALG-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
182; BITALG-NEXT:    vpsadbw %ymm1, %ymm2, %ymm2
183; BITALG-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
184; BITALG-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
185; BITALG-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
186; BITALG-NEXT:    retq
187  %out = call <8 x i32> @llvm.ctpop.v8i32(<8 x i32> %in)
188  ret <8 x i32> %out
189}
190
191define <16 x i16> @testv16i16(<16 x i16> %in) nounwind {
192; AVX1-LABEL: testv16i16:
193; AVX1:       # %bb.0:
194; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
195; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
196; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
197; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
198; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm4
199; AVX1-NEXT:    vpand %xmm1, %xmm4, %xmm4
200; AVX1-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
201; AVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
202; AVX1-NEXT:    vpsllw $8, %xmm2, %xmm4
203; AVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
204; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
205; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
206; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm4
207; AVX1-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
208; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
209; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
210; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
211; AVX1-NEXT:    vpaddb %xmm4, %xmm0, %xmm0
212; AVX1-NEXT:    vpsllw $8, %xmm0, %xmm1
213; AVX1-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
214; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm0
215; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
216; AVX1-NEXT:    retq
217;
218; AVX2-LABEL: testv16i16:
219; AVX2:       # %bb.0:
220; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
221; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
222; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
223; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
224; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
225; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
226; AVX2-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
227; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
228; AVX2-NEXT:    vpsllw $8, %ymm0, %ymm1
229; AVX2-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
230; AVX2-NEXT:    vpsrlw $8, %ymm0, %ymm0
231; AVX2-NEXT:    retq
232;
233; AVX512VPOPCNTDQ-LABEL: testv16i16:
234; AVX512VPOPCNTDQ:       # %bb.0:
235; AVX512VPOPCNTDQ-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
236; AVX512VPOPCNTDQ-NEXT:    vpopcntd %zmm0, %zmm0
237; AVX512VPOPCNTDQ-NEXT:    vpmovdw %zmm0, %ymm0
238; AVX512VPOPCNTDQ-NEXT:    retq
239;
240; AVX512VPOPCNTDQVL-LABEL: testv16i16:
241; AVX512VPOPCNTDQVL:       # %bb.0:
242; AVX512VPOPCNTDQVL-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
243; AVX512VPOPCNTDQVL-NEXT:    vpopcntd %zmm0, %zmm0
244; AVX512VPOPCNTDQVL-NEXT:    vpmovdw %zmm0, %ymm0
245; AVX512VPOPCNTDQVL-NEXT:    retq
246;
247; BITALG_NOVLX-LABEL: testv16i16:
248; BITALG_NOVLX:       # %bb.0:
249; BITALG_NOVLX-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
250; BITALG_NOVLX-NEXT:    vpopcntw %zmm0, %zmm0
251; BITALG_NOVLX-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
252; BITALG_NOVLX-NEXT:    retq
253;
254; BITALG-LABEL: testv16i16:
255; BITALG:       # %bb.0:
256; BITALG-NEXT:    vpopcntw %ymm0, %ymm0
257; BITALG-NEXT:    retq
258  %out = call <16 x i16> @llvm.ctpop.v16i16(<16 x i16> %in)
259  ret <16 x i16> %out
260}
261
262define <32 x i8> @testv32i8(<32 x i8> %in) nounwind {
263; AVX1-LABEL: testv32i8:
264; AVX1:       # %bb.0:
265; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
266; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
267; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm3
268; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
269; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
270; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm1
271; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
272; AVX1-NEXT:    vpshufb %xmm1, %xmm4, %xmm1
273; AVX1-NEXT:    vpaddb %xmm3, %xmm1, %xmm1
274; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm3
275; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
276; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
277; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
278; AVX1-NEXT:    vpshufb %xmm0, %xmm4, %xmm0
279; AVX1-NEXT:    vpaddb %xmm3, %xmm0, %xmm0
280; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
281; AVX1-NEXT:    retq
282;
283; AVX2-LABEL: testv32i8:
284; AVX2:       # %bb.0:
285; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
286; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
287; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
288; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
289; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
290; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
291; AVX2-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
292; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
293; AVX2-NEXT:    retq
294;
295; AVX512VPOPCNTDQ-LABEL: testv32i8:
296; AVX512VPOPCNTDQ:       # %bb.0:
297; AVX512VPOPCNTDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
298; AVX512VPOPCNTDQ-NEXT:    vpand %ymm1, %ymm0, %ymm2
299; AVX512VPOPCNTDQ-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
300; AVX512VPOPCNTDQ-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
301; AVX512VPOPCNTDQ-NEXT:    vpsrlw $4, %ymm0, %ymm0
302; AVX512VPOPCNTDQ-NEXT:    vpand %ymm1, %ymm0, %ymm0
303; AVX512VPOPCNTDQ-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
304; AVX512VPOPCNTDQ-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
305; AVX512VPOPCNTDQ-NEXT:    retq
306;
307; AVX512VPOPCNTDQVL-LABEL: testv32i8:
308; AVX512VPOPCNTDQVL:       # %bb.0:
309; AVX512VPOPCNTDQVL-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
310; AVX512VPOPCNTDQVL-NEXT:    vpand %ymm1, %ymm0, %ymm2
311; AVX512VPOPCNTDQVL-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
312; AVX512VPOPCNTDQVL-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
313; AVX512VPOPCNTDQVL-NEXT:    vpsrlw $4, %ymm0, %ymm0
314; AVX512VPOPCNTDQVL-NEXT:    vpand %ymm1, %ymm0, %ymm0
315; AVX512VPOPCNTDQVL-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
316; AVX512VPOPCNTDQVL-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
317; AVX512VPOPCNTDQVL-NEXT:    retq
318;
319; BITALG_NOVLX-LABEL: testv32i8:
320; BITALG_NOVLX:       # %bb.0:
321; BITALG_NOVLX-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
322; BITALG_NOVLX-NEXT:    vpopcntb %zmm0, %zmm0
323; BITALG_NOVLX-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
324; BITALG_NOVLX-NEXT:    retq
325;
326; BITALG-LABEL: testv32i8:
327; BITALG:       # %bb.0:
328; BITALG-NEXT:    vpopcntb %ymm0, %ymm0
329; BITALG-NEXT:    retq
330  %out = call <32 x i8> @llvm.ctpop.v32i8(<32 x i8> %in)
331  ret <32 x i8> %out
332}
333
334define <4 x i64> @foldv4i64() nounwind {
335; ALL-LABEL: foldv4i64:
336; ALL:       # %bb.0:
337; ALL-NEXT:    vmovaps {{.*#+}} ymm0 = [1,64,0,8]
338; ALL-NEXT:    retq
339  %out = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> <i64 256, i64 -1, i64 0, i64 255>)
340  ret <4 x i64> %out
341}
342
343define <8 x i32> @foldv8i32() nounwind {
344; ALL-LABEL: foldv8i32:
345; ALL:       # %bb.0:
346; ALL-NEXT:    vmovaps {{.*#+}} ymm0 = [1,32,0,8,16,3,2,3]
347; ALL-NEXT:    retq
348  %out = call <8 x i32> @llvm.ctpop.v8i32(<8 x i32> <i32 256, i32 -1, i32 0, i32 255, i32 -65536, i32 7, i32 24, i32 88>)
349  ret <8 x i32> %out
350}
351
352define <16 x i16> @foldv16i16() nounwind {
353; ALL-LABEL: foldv16i16:
354; ALL:       # %bb.0:
355; ALL-NEXT:    vmovaps {{.*#+}} ymm0 = [1,16,0,8,0,3,2,3,15,7,1,1,1,1,1,1]
356; ALL-NEXT:    retq
357  %out = call <16 x i16> @llvm.ctpop.v16i16(<16 x i16> <i16 256, i16 -1, i16 0, i16 255, i16 -65536, i16 7, i16 24, i16 88, i16 -2, i16 254, i16 1, i16 2, i16 4, i16 8, i16 16, i16 32>)
358  ret <16 x i16> %out
359}
360
361define <32 x i8> @foldv32i8() nounwind {
362; ALL-LABEL: foldv32i8:
363; ALL:       # %bb.0:
364; ALL-NEXT:    vmovaps {{.*#+}} ymm0 = [0,8,0,8,0,3,2,3,7,7,1,1,1,1,1,1,1,1,0,0,1,2,3,4,5,6,7,8,2,2,3,7]
365; ALL-NEXT:    retq
366  %out = call <32 x i8> @llvm.ctpop.v32i8(<32 x i8> <i8 256, i8 -1, i8 0, i8 255, i8 -65536, i8 7, i8 24, i8 88, i8 -2, i8 254, i8 1, i8 2, i8 4, i8 8, i8 16, i8 32, i8 64, i8 128, i8 256, i8 -256, i8 -128, i8 -64, i8 -32, i8 -16, i8 -8, i8 -4, i8 -2, i8 -1, i8 3, i8 5, i8 7, i8 127>)
367  ret <32 x i8> %out
368}
369
370declare <4 x i64> @llvm.ctpop.v4i64(<4 x i64>)
371declare <8 x i32> @llvm.ctpop.v8i32(<8 x i32>)
372declare <16 x i16> @llvm.ctpop.v16i16(<16 x i16>)
373declare <32 x i8> @llvm.ctpop.v32i8(<32 x i8>)
374