xref: /llvm-project/llvm/test/CodeGen/X86/vselect-packss.ll (revision f7788618dd24e5366709a70c3e16cd5a0ca0a173) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2      | FileCheck %s --check-prefix=SSE2
3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.2    | FileCheck %s --check-prefix=SSE42
4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx       | FileCheck %s --check-prefix=AVX1
5; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2      | FileCheck %s --check-prefix=AVX2
6; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f   | FileCheck %s --check-prefix=AVX512F
7; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl  | FileCheck %s --check-prefix=AVX512VL
8; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw  | FileCheck %s --check-prefix=AVX512BWNOVL
9; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl  | FileCheck %s --check-prefix=AVX512BWVL
10
11;
12; General cases - packing of vector comparison to legal vector result types
13;
14
15define <16 x i8> @vselect_packss_v16i16(<16 x i16> %a0, <16 x i16> %a1, <16 x i8> %a2, <16 x i8> %a3) {
16; SSE2-LABEL: vselect_packss_v16i16:
17; SSE2:       # %bb.0:
18; SSE2-NEXT:    pcmpeqw %xmm3, %xmm1
19; SSE2-NEXT:    pcmpeqw %xmm2, %xmm0
20; SSE2-NEXT:    packsswb %xmm1, %xmm0
21; SSE2-NEXT:    pand %xmm0, %xmm4
22; SSE2-NEXT:    pandn %xmm5, %xmm0
23; SSE2-NEXT:    por %xmm4, %xmm0
24; SSE2-NEXT:    retq
25;
26; SSE42-LABEL: vselect_packss_v16i16:
27; SSE42:       # %bb.0:
28; SSE42-NEXT:    pcmpeqw %xmm3, %xmm1
29; SSE42-NEXT:    pcmpeqw %xmm2, %xmm0
30; SSE42-NEXT:    packsswb %xmm1, %xmm0
31; SSE42-NEXT:    pblendvb %xmm0, %xmm4, %xmm5
32; SSE42-NEXT:    movdqa %xmm5, %xmm0
33; SSE42-NEXT:    retq
34;
35; AVX1-LABEL: vselect_packss_v16i16:
36; AVX1:       # %bb.0:
37; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
38; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
39; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm5, %xmm4
40; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
41; AVX1-NEXT:    vpacksswb %xmm4, %xmm0, %xmm0
42; AVX1-NEXT:    vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
43; AVX1-NEXT:    vzeroupper
44; AVX1-NEXT:    retq
45;
46; AVX2-LABEL: vselect_packss_v16i16:
47; AVX2:       # %bb.0:
48; AVX2-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
49; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
50; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
51; AVX2-NEXT:    vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
52; AVX2-NEXT:    vzeroupper
53; AVX2-NEXT:    retq
54;
55; AVX512F-LABEL: vselect_packss_v16i16:
56; AVX512F:       # %bb.0:
57; AVX512F-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
58; AVX512F-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
59; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
60; AVX512F-NEXT:    vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
61; AVX512F-NEXT:    vzeroupper
62; AVX512F-NEXT:    retq
63;
64; AVX512VL-LABEL: vselect_packss_v16i16:
65; AVX512VL:       # %bb.0:
66; AVX512VL-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
67; AVX512VL-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
68; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
69; AVX512VL-NEXT:    vpternlogq {{.*#+}} xmm0 = xmm3 ^ (xmm0 & (xmm2 ^ xmm3))
70; AVX512VL-NEXT:    vzeroupper
71; AVX512VL-NEXT:    retq
72;
73; AVX512BWNOVL-LABEL: vselect_packss_v16i16:
74; AVX512BWNOVL:       # %bb.0:
75; AVX512BWNOVL-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
76; AVX512BWNOVL-NEXT:    vpmovwb %zmm0, %ymm0
77; AVX512BWNOVL-NEXT:    vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
78; AVX512BWNOVL-NEXT:    vzeroupper
79; AVX512BWNOVL-NEXT:    retq
80;
81; AVX512BWVL-LABEL: vselect_packss_v16i16:
82; AVX512BWVL:       # %bb.0:
83; AVX512BWVL-NEXT:    vpcmpeqw %ymm1, %ymm0, %k0
84; AVX512BWVL-NEXT:    vpmovm2b %k0, %xmm0
85; AVX512BWVL-NEXT:    vpternlogq {{.*#+}} xmm0 = xmm3 ^ (xmm0 & (xmm2 ^ xmm3))
86; AVX512BWVL-NEXT:    vzeroupper
87; AVX512BWVL-NEXT:    retq
88  %1 = icmp eq <16 x i16> %a0, %a1
89  %2 = sext <16 x i1> %1 to <16 x i8>
90  %3 = and <16 x i8> %2, %a2
91  %4 = xor <16 x i8> %2, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
92  %5 = and <16 x i8> %4, %a3
93  %6 = or <16 x i8> %3, %5
94  ret <16 x i8> %6
95}
96
97define <16 x i8> @vselect_packss_v16i32(<16 x i32> %a0, <16 x i32> %a1, <16 x i8> %a2, <16 x i8> %a3) {
98; SSE2-LABEL: vselect_packss_v16i32:
99; SSE2:       # %bb.0:
100; SSE2-NEXT:    pcmpeqd %xmm7, %xmm3
101; SSE2-NEXT:    pcmpeqd %xmm6, %xmm2
102; SSE2-NEXT:    packssdw %xmm3, %xmm2
103; SSE2-NEXT:    pcmpeqd %xmm5, %xmm1
104; SSE2-NEXT:    pcmpeqd %xmm4, %xmm0
105; SSE2-NEXT:    packssdw %xmm1, %xmm0
106; SSE2-NEXT:    packsswb %xmm2, %xmm0
107; SSE2-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm1
108; SSE2-NEXT:    pand %xmm0, %xmm1
109; SSE2-NEXT:    pandn {{[0-9]+}}(%rsp), %xmm0
110; SSE2-NEXT:    por %xmm1, %xmm0
111; SSE2-NEXT:    retq
112;
113; SSE42-LABEL: vselect_packss_v16i32:
114; SSE42:       # %bb.0:
115; SSE42-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm8
116; SSE42-NEXT:    pcmpeqd %xmm7, %xmm3
117; SSE42-NEXT:    pcmpeqd %xmm6, %xmm2
118; SSE42-NEXT:    packssdw %xmm3, %xmm2
119; SSE42-NEXT:    pcmpeqd %xmm5, %xmm1
120; SSE42-NEXT:    pcmpeqd %xmm4, %xmm0
121; SSE42-NEXT:    packssdw %xmm1, %xmm0
122; SSE42-NEXT:    packsswb %xmm2, %xmm0
123; SSE42-NEXT:    pblendvb %xmm0, {{[0-9]+}}(%rsp), %xmm8
124; SSE42-NEXT:    movdqa %xmm8, %xmm0
125; SSE42-NEXT:    retq
126;
127; AVX1-LABEL: vselect_packss_v16i32:
128; AVX1:       # %bb.0:
129; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm6
130; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
131; AVX1-NEXT:    vpcmpeqd %xmm6, %xmm7, %xmm6
132; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
133; AVX1-NEXT:    vpackssdw %xmm6, %xmm1, %xmm1
134; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
135; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
136; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm6, %xmm3
137; AVX1-NEXT:    vpcmpeqd %xmm2, %xmm0, %xmm0
138; AVX1-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
139; AVX1-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
140; AVX1-NEXT:    vpblendvb %xmm0, %xmm4, %xmm5, %xmm0
141; AVX1-NEXT:    vzeroupper
142; AVX1-NEXT:    retq
143;
144; AVX2-LABEL: vselect_packss_v16i32:
145; AVX2:       # %bb.0:
146; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm1, %ymm1
147; AVX2-NEXT:    vpcmpeqd %ymm2, %ymm0, %ymm0
148; AVX2-NEXT:    vpackssdw %ymm1, %ymm0, %ymm0
149; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
150; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
151; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
152; AVX2-NEXT:    vpblendvb %xmm0, %xmm4, %xmm5, %xmm0
153; AVX2-NEXT:    vzeroupper
154; AVX2-NEXT:    retq
155;
156; AVX512F-LABEL: vselect_packss_v16i32:
157; AVX512F:       # %bb.0:
158; AVX512F-NEXT:    vpcmpeqd %zmm1, %zmm0, %k1
159; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm0 {%k1} {z} = -1
160; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
161; AVX512F-NEXT:    vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
162; AVX512F-NEXT:    vzeroupper
163; AVX512F-NEXT:    retq
164;
165; AVX512VL-LABEL: vselect_packss_v16i32:
166; AVX512VL:       # %bb.0:
167; AVX512VL-NEXT:    vpcmpeqd %zmm1, %zmm0, %k1
168; AVX512VL-NEXT:    vpternlogd {{.*#+}} zmm0 {%k1} {z} = -1
169; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
170; AVX512VL-NEXT:    vpternlogq {{.*#+}} xmm0 = xmm3 ^ (xmm0 & (xmm2 ^ xmm3))
171; AVX512VL-NEXT:    vzeroupper
172; AVX512VL-NEXT:    retq
173;
174; AVX512BWNOVL-LABEL: vselect_packss_v16i32:
175; AVX512BWNOVL:       # %bb.0:
176; AVX512BWNOVL-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
177; AVX512BWNOVL-NEXT:    vpmovm2b %k0, %zmm0
178; AVX512BWNOVL-NEXT:    vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
179; AVX512BWNOVL-NEXT:    vzeroupper
180; AVX512BWNOVL-NEXT:    retq
181;
182; AVX512BWVL-LABEL: vselect_packss_v16i32:
183; AVX512BWVL:       # %bb.0:
184; AVX512BWVL-NEXT:    vpcmpeqd %zmm1, %zmm0, %k0
185; AVX512BWVL-NEXT:    vpmovm2b %k0, %xmm0
186; AVX512BWVL-NEXT:    vpternlogq {{.*#+}} xmm0 = xmm3 ^ (xmm0 & (xmm2 ^ xmm3))
187; AVX512BWVL-NEXT:    vzeroupper
188; AVX512BWVL-NEXT:    retq
189  %1 = icmp eq <16 x i32> %a0, %a1
190  %2 = sext <16 x i1> %1 to <16 x i8>
191  %3 = and <16 x i8> %2, %a2
192  %4 = xor <16 x i8> %2, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
193  %5 = and <16 x i8> %4, %a3
194  %6 = or <16 x i8> %3, %5
195  ret <16 x i8> %6
196}
197
198define <16 x i8> @vselect_packss_v16i64(<16 x i64> %a0, <16 x i64> %a1, <16 x i8> %a2, <16 x i8> %a3) {
199; SSE2-LABEL: vselect_packss_v16i64:
200; SSE2:       # %bb.0:
201; SSE2-NEXT:    pcmpeqd {{[0-9]+}}(%rsp), %xmm7
202; SSE2-NEXT:    pshufd {{.*#+}} xmm8 = xmm7[1,0,3,2]
203; SSE2-NEXT:    pand %xmm7, %xmm8
204; SSE2-NEXT:    pcmpeqd {{[0-9]+}}(%rsp), %xmm6
205; SSE2-NEXT:    pshufd {{.*#+}} xmm7 = xmm6[1,0,3,2]
206; SSE2-NEXT:    pand %xmm6, %xmm7
207; SSE2-NEXT:    packssdw %xmm8, %xmm7
208; SSE2-NEXT:    pcmpeqd {{[0-9]+}}(%rsp), %xmm5
209; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm5[1,0,3,2]
210; SSE2-NEXT:    pand %xmm5, %xmm6
211; SSE2-NEXT:    pcmpeqd {{[0-9]+}}(%rsp), %xmm4
212; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[1,0,3,2]
213; SSE2-NEXT:    pand %xmm4, %xmm5
214; SSE2-NEXT:    packssdw %xmm6, %xmm5
215; SSE2-NEXT:    packssdw %xmm7, %xmm5
216; SSE2-NEXT:    pcmpeqd {{[0-9]+}}(%rsp), %xmm3
217; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[1,0,3,2]
218; SSE2-NEXT:    pand %xmm3, %xmm4
219; SSE2-NEXT:    pcmpeqd {{[0-9]+}}(%rsp), %xmm2
220; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,0,3,2]
221; SSE2-NEXT:    pand %xmm2, %xmm3
222; SSE2-NEXT:    packssdw %xmm4, %xmm3
223; SSE2-NEXT:    pcmpeqd {{[0-9]+}}(%rsp), %xmm1
224; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,0,3,2]
225; SSE2-NEXT:    pand %xmm1, %xmm2
226; SSE2-NEXT:    pcmpeqd {{[0-9]+}}(%rsp), %xmm0
227; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,0,3,2]
228; SSE2-NEXT:    pand %xmm1, %xmm0
229; SSE2-NEXT:    packssdw %xmm2, %xmm0
230; SSE2-NEXT:    packssdw %xmm3, %xmm0
231; SSE2-NEXT:    packsswb %xmm5, %xmm0
232; SSE2-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm1
233; SSE2-NEXT:    pand %xmm0, %xmm1
234; SSE2-NEXT:    pandn {{[0-9]+}}(%rsp), %xmm0
235; SSE2-NEXT:    por %xmm1, %xmm0
236; SSE2-NEXT:    retq
237;
238; SSE42-LABEL: vselect_packss_v16i64:
239; SSE42:       # %bb.0:
240; SSE42-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm8
241; SSE42-NEXT:    pcmpeqq {{[0-9]+}}(%rsp), %xmm7
242; SSE42-NEXT:    pcmpeqq {{[0-9]+}}(%rsp), %xmm6
243; SSE42-NEXT:    packssdw %xmm7, %xmm6
244; SSE42-NEXT:    pcmpeqq {{[0-9]+}}(%rsp), %xmm5
245; SSE42-NEXT:    pcmpeqq {{[0-9]+}}(%rsp), %xmm4
246; SSE42-NEXT:    packssdw %xmm5, %xmm4
247; SSE42-NEXT:    packssdw %xmm6, %xmm4
248; SSE42-NEXT:    pcmpeqq {{[0-9]+}}(%rsp), %xmm3
249; SSE42-NEXT:    pcmpeqq {{[0-9]+}}(%rsp), %xmm2
250; SSE42-NEXT:    packssdw %xmm3, %xmm2
251; SSE42-NEXT:    pcmpeqq {{[0-9]+}}(%rsp), %xmm1
252; SSE42-NEXT:    pcmpeqq {{[0-9]+}}(%rsp), %xmm0
253; SSE42-NEXT:    packssdw %xmm1, %xmm0
254; SSE42-NEXT:    packssdw %xmm2, %xmm0
255; SSE42-NEXT:    packsswb %xmm4, %xmm0
256; SSE42-NEXT:    pblendvb %xmm0, {{[0-9]+}}(%rsp), %xmm8
257; SSE42-NEXT:    movdqa %xmm8, %xmm0
258; SSE42-NEXT:    retq
259;
260; AVX1-LABEL: vselect_packss_v16i64:
261; AVX1:       # %bb.0:
262; AVX1-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %xmm8
263; AVX1-NEXT:    vextractf128 $1, %ymm7, %xmm9
264; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm10
265; AVX1-NEXT:    vpcmpeqq %xmm9, %xmm10, %xmm9
266; AVX1-NEXT:    vpcmpeqq %xmm7, %xmm3, %xmm3
267; AVX1-NEXT:    vpackssdw %xmm9, %xmm3, %xmm3
268; AVX1-NEXT:    vextractf128 $1, %ymm6, %xmm7
269; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm9
270; AVX1-NEXT:    vpcmpeqq %xmm7, %xmm9, %xmm7
271; AVX1-NEXT:    vpcmpeqq %xmm6, %xmm2, %xmm2
272; AVX1-NEXT:    vpackssdw %xmm7, %xmm2, %xmm2
273; AVX1-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
274; AVX1-NEXT:    vextractf128 $1, %ymm5, %xmm3
275; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
276; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm6, %xmm3
277; AVX1-NEXT:    vpcmpeqq %xmm5, %xmm1, %xmm1
278; AVX1-NEXT:    vpackssdw %xmm3, %xmm1, %xmm1
279; AVX1-NEXT:    vextractf128 $1, %ymm4, %xmm3
280; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
281; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm5, %xmm3
282; AVX1-NEXT:    vpcmpeqq %xmm4, %xmm0, %xmm0
283; AVX1-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
284; AVX1-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
285; AVX1-NEXT:    vpacksswb %xmm2, %xmm0, %xmm0
286; AVX1-NEXT:    vpblendvb %xmm0, {{[0-9]+}}(%rsp), %xmm8, %xmm0
287; AVX1-NEXT:    vzeroupper
288; AVX1-NEXT:    retq
289;
290; AVX2-LABEL: vselect_packss_v16i64:
291; AVX2:       # %bb.0:
292; AVX2-NEXT:    vpcmpeqq %ymm7, %ymm3, %ymm3
293; AVX2-NEXT:    vpcmpeqq %ymm6, %ymm2, %ymm2
294; AVX2-NEXT:    vpackssdw %ymm3, %ymm2, %ymm2
295; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,1,3]
296; AVX2-NEXT:    vpcmpeqq %ymm5, %ymm1, %ymm1
297; AVX2-NEXT:    vpcmpeqq %ymm4, %ymm0, %ymm0
298; AVX2-NEXT:    vpackssdw %ymm1, %ymm0, %ymm0
299; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
300; AVX2-NEXT:    vpackssdw %ymm2, %ymm0, %ymm0
301; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
302; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
303; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
304; AVX2-NEXT:    vpand {{[0-9]+}}(%rsp), %xmm0, %xmm1
305; AVX2-NEXT:    vpandn {{[0-9]+}}(%rsp), %xmm0, %xmm0
306; AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
307; AVX2-NEXT:    vzeroupper
308; AVX2-NEXT:    retq
309;
310; AVX512F-LABEL: vselect_packss_v16i64:
311; AVX512F:       # %bb.0:
312; AVX512F-NEXT:    vpcmpeqq %zmm2, %zmm0, %k0
313; AVX512F-NEXT:    vpcmpeqq %zmm3, %zmm1, %k1
314; AVX512F-NEXT:    kunpckbw %k0, %k1, %k1
315; AVX512F-NEXT:    vpternlogd {{.*#+}} zmm0 {%k1} {z} = -1
316; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
317; AVX512F-NEXT:    vpblendvb %xmm0, %xmm4, %xmm5, %xmm0
318; AVX512F-NEXT:    vzeroupper
319; AVX512F-NEXT:    retq
320;
321; AVX512VL-LABEL: vselect_packss_v16i64:
322; AVX512VL:       # %bb.0:
323; AVX512VL-NEXT:    vpcmpeqq %zmm2, %zmm0, %k0
324; AVX512VL-NEXT:    vpcmpeqq %zmm3, %zmm1, %k1
325; AVX512VL-NEXT:    kunpckbw %k0, %k1, %k1
326; AVX512VL-NEXT:    vpternlogd {{.*#+}} zmm0 {%k1} {z} = -1
327; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
328; AVX512VL-NEXT:    vpternlogq {{.*#+}} xmm0 = xmm5 ^ (xmm0 & (xmm4 ^ xmm5))
329; AVX512VL-NEXT:    vzeroupper
330; AVX512VL-NEXT:    retq
331;
332; AVX512BWNOVL-LABEL: vselect_packss_v16i64:
333; AVX512BWNOVL:       # %bb.0:
334; AVX512BWNOVL-NEXT:    vpcmpeqq %zmm2, %zmm0, %k0
335; AVX512BWNOVL-NEXT:    vpcmpeqq %zmm3, %zmm1, %k1
336; AVX512BWNOVL-NEXT:    kunpckbw %k0, %k1, %k0
337; AVX512BWNOVL-NEXT:    vpmovm2b %k0, %zmm0
338; AVX512BWNOVL-NEXT:    vpblendvb %xmm0, %xmm4, %xmm5, %xmm0
339; AVX512BWNOVL-NEXT:    vzeroupper
340; AVX512BWNOVL-NEXT:    retq
341;
342; AVX512BWVL-LABEL: vselect_packss_v16i64:
343; AVX512BWVL:       # %bb.0:
344; AVX512BWVL-NEXT:    vpcmpeqq %zmm2, %zmm0, %k0
345; AVX512BWVL-NEXT:    vpcmpeqq %zmm3, %zmm1, %k1
346; AVX512BWVL-NEXT:    kunpckbw %k0, %k1, %k0
347; AVX512BWVL-NEXT:    vpmovm2b %k0, %xmm0
348; AVX512BWVL-NEXT:    vpternlogq {{.*#+}} xmm0 = xmm5 ^ (xmm0 & (xmm4 ^ xmm5))
349; AVX512BWVL-NEXT:    vzeroupper
350; AVX512BWVL-NEXT:    retq
351  %1 = icmp eq <16 x i64> %a0, %a1
352  %2 = sext <16 x i1> %1 to <16 x i8>
353  %3 = and <16 x i8> %2, %a2
354  %4 = xor <16 x i8> %2, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
355  %5 = and <16 x i8> %4, %a3
356  %6 = or <16 x i8> %3, %5
357  ret <16 x i8> %6
358}
359
360;
361; PACKSS case
362;
363
364define <16 x i8> @vselect_packss(<16 x i16> %a0, <16 x i16> %a1, <16 x i8> %a2, <16 x i8> %a3) {
365; SSE2-LABEL: vselect_packss:
366; SSE2:       # %bb.0:
367; SSE2-NEXT:    pcmpeqw %xmm3, %xmm1
368; SSE2-NEXT:    pcmpeqw %xmm2, %xmm0
369; SSE2-NEXT:    packsswb %xmm1, %xmm0
370; SSE2-NEXT:    pand %xmm0, %xmm4
371; SSE2-NEXT:    pandn %xmm5, %xmm0
372; SSE2-NEXT:    por %xmm4, %xmm0
373; SSE2-NEXT:    retq
374;
375; SSE42-LABEL: vselect_packss:
376; SSE42:       # %bb.0:
377; SSE42-NEXT:    pcmpeqw %xmm3, %xmm1
378; SSE42-NEXT:    pcmpeqw %xmm2, %xmm0
379; SSE42-NEXT:    packsswb %xmm1, %xmm0
380; SSE42-NEXT:    pblendvb %xmm0, %xmm4, %xmm5
381; SSE42-NEXT:    movdqa %xmm5, %xmm0
382; SSE42-NEXT:    retq
383;
384; AVX1-LABEL: vselect_packss:
385; AVX1:       # %bb.0:
386; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
387; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
388; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm5, %xmm4
389; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
390; AVX1-NEXT:    vpacksswb %xmm4, %xmm0, %xmm0
391; AVX1-NEXT:    vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
392; AVX1-NEXT:    vzeroupper
393; AVX1-NEXT:    retq
394;
395; AVX2-LABEL: vselect_packss:
396; AVX2:       # %bb.0:
397; AVX2-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
398; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
399; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
400; AVX2-NEXT:    vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
401; AVX2-NEXT:    vzeroupper
402; AVX2-NEXT:    retq
403;
404; AVX512F-LABEL: vselect_packss:
405; AVX512F:       # %bb.0:
406; AVX512F-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
407; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
408; AVX512F-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
409; AVX512F-NEXT:    vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
410; AVX512F-NEXT:    vzeroupper
411; AVX512F-NEXT:    retq
412;
413; AVX512VL-LABEL: vselect_packss:
414; AVX512VL:       # %bb.0:
415; AVX512VL-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
416; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
417; AVX512VL-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
418; AVX512VL-NEXT:    vpternlogq {{.*#+}} xmm0 = xmm3 ^ (xmm0 & (xmm2 ^ xmm3))
419; AVX512VL-NEXT:    vzeroupper
420; AVX512VL-NEXT:    retq
421;
422; AVX512BWNOVL-LABEL: vselect_packss:
423; AVX512BWNOVL:       # %bb.0:
424; AVX512BWNOVL-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
425; AVX512BWNOVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
426; AVX512BWNOVL-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
427; AVX512BWNOVL-NEXT:    vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
428; AVX512BWNOVL-NEXT:    vzeroupper
429; AVX512BWNOVL-NEXT:    retq
430;
431; AVX512BWVL-LABEL: vselect_packss:
432; AVX512BWVL:       # %bb.0:
433; AVX512BWVL-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
434; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
435; AVX512BWVL-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
436; AVX512BWVL-NEXT:    vpternlogq {{.*#+}} xmm0 = xmm3 ^ (xmm0 & (xmm2 ^ xmm3))
437; AVX512BWVL-NEXT:    vzeroupper
438; AVX512BWVL-NEXT:    retq
439  %1 = icmp eq <16 x i16> %a0, %a1
440  %2 = sext <16 x i1> %1 to <16 x i16>
441  %3 = shufflevector <16 x i16> %2, <16 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
442  %4 = shufflevector <16 x i16> %2, <16 x i16> undef, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
443  %5 = tail call <16 x i8> @llvm.x86.sse2.packsswb.128(<8 x i16> %3, <8 x i16> %4)
444  %6 = and <16 x i8> %5, %a2
445  %7 = xor <16 x i8> %5, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
446  %8 = and <16 x i8> %7, %a3
447  %9 = or <16 x i8> %6, %8
448  ret <16 x i8> %9
449}
450declare <16 x i8> @llvm.x86.sse2.packsswb.128(<8 x i16>, <8 x i16>)
451