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=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
5; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
6
7define <4 x i32> @test1(<8 x i32> %v) {
8; SSE2-LABEL: test1:
9; SSE2:       # %bb.0:
10; SSE2-NEXT:    retq
11;
12; AVX-LABEL: test1:
13; AVX:       # %bb.0:
14; AVX-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
15; AVX-NEXT:    vzeroupper
16; AVX-NEXT:    retq
17  %x = sext <8 x i32> %v to <8 x i64>
18  %s = shufflevector <8 x i64> %x, <8 x i64> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
19  %t = trunc <4 x i64> %s to <4 x i32>
20  ret <4 x i32> %t
21}
22
23define <4 x i32> @test2(<8 x i32> %v) {
24; SSE2-LABEL: test2:
25; SSE2:       # %bb.0:
26; SSE2-NEXT:    movaps %xmm1, %xmm0
27; SSE2-NEXT:    retq
28;
29; AVX-LABEL: test2:
30; AVX:       # %bb.0:
31; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm0
32; AVX-NEXT:    vzeroupper
33; AVX-NEXT:    retq
34  %x = sext <8 x i32> %v to <8 x i64>
35  %s = shufflevector <8 x i64> %x, <8 x i64> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
36  %t = trunc <4 x i64> %s to <4 x i32>
37  ret <4 x i32> %t
38}
39
40define <2 x i32> @test3(<8 x i32> %v) {
41; SSE2-LABEL: test3:
42; SSE2:       # %bb.0:
43; SSE2-NEXT:    movdqa %xmm1, %xmm0
44; SSE2-NEXT:    pxor %xmm1, %xmm1
45; SSE2-NEXT:    pcmpgtd %xmm0, %xmm1
46; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
47; SSE2-NEXT:    retq
48;
49; AVX2-LABEL: test3:
50; AVX2:       # %bb.0:
51; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
52; AVX2-NEXT:    vpmovsxdq %xmm0, %xmm0
53; AVX2-NEXT:    vzeroupper
54; AVX2-NEXT:    retq
55;
56; AVX512-LABEL: test3:
57; AVX512:       # %bb.0:
58; AVX512-NEXT:    vpmovsxdq %ymm0, %zmm0
59; AVX512-NEXT:    vextracti32x4 $2, %zmm0, %xmm0
60; AVX512-NEXT:    vzeroupper
61; AVX512-NEXT:    retq
62  %x = sext <8 x i32> %v to <8 x i64>
63  %s = shufflevector <8 x i64> %x, <8 x i64> undef, <2 x i32> <i32 4, i32 5>
64  %t = trunc <2 x i64> %s to <2 x i32>
65  ret <2 x i32> %t
66}
67
68define <2 x i32> @test4(<8 x i32> %v) {
69; SSE2-LABEL: test4:
70; SSE2:       # %bb.0:
71; SSE2-NEXT:    pxor %xmm1, %xmm1
72; SSE2-NEXT:    pcmpgtd %xmm0, %xmm1
73; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
74; SSE2-NEXT:    retq
75;
76; AVX2-LABEL: test4:
77; AVX2:       # %bb.0:
78; AVX2-NEXT:    vpmovsxdq %xmm0, %xmm0
79; AVX2-NEXT:    vzeroupper
80; AVX2-NEXT:    retq
81;
82; AVX512-LABEL: test4:
83; AVX512:       # %bb.0:
84; AVX512-NEXT:    vpmovsxdq %ymm0, %zmm0
85; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
86; AVX512-NEXT:    vzeroupper
87; AVX512-NEXT:    retq
88  %x = sext <8 x i32> %v to <8 x i64>
89  %s = shufflevector <8 x i64> %x, <8 x i64> undef, <2 x i32> <i32 0, i32 1>
90  %t = trunc <2 x i64> %s to <2 x i32>
91  ret <2 x i32> %t
92}
93
94define <2 x i32> @test5(<8 x i32> %v) {
95; SSE2-LABEL: test5:
96; SSE2:       # %bb.0:
97; SSE2-NEXT:    pxor %xmm2, %xmm2
98; SSE2-NEXT:    pxor %xmm3, %xmm3
99; SSE2-NEXT:    pcmpgtd %xmm1, %xmm3
100; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
101; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
102; SSE2-NEXT:    pcmpgtd %xmm0, %xmm2
103; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
104; SSE2-NEXT:    shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]
105; SSE2-NEXT:    retq
106;
107; AVX2-LABEL: test5:
108; AVX2:       # %bb.0:
109; AVX2-NEXT:    vpmovsxdq %xmm0, %ymm1
110; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
111; AVX2-NEXT:    vpbroadcastd %xmm0, %ymm0
112; AVX2-NEXT:    vpmovsxdq %xmm0, %ymm0
113; AVX2-NEXT:    vpalignr {{.*#+}} ymm0 = ymm1[8,9,10,11,12,13,14,15],ymm0[0,1,2,3,4,5,6,7],ymm1[24,25,26,27,28,29,30,31],ymm0[16,17,18,19,20,21,22,23]
114; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,2,3]
115; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
116; AVX2-NEXT:    vzeroupper
117; AVX2-NEXT:    retq
118;
119; AVX512-LABEL: test5:
120; AVX512:       # %bb.0:
121; AVX512-NEXT:    vpmovsxdq %ymm0, %zmm0
122; AVX512-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
123; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm0
124; AVX512-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
125; AVX512-NEXT:    vzeroupper
126; AVX512-NEXT:    retq
127  %x = sext <8 x i32> %v to <8 x i64>
128  %s = shufflevector <8 x i64> %x, <8 x i64> undef, <2 x i32> <i32 3, i32 4>
129  %t = trunc <2 x i64> %s to <2 x i32>
130  ret <2 x i32> %t
131}
132
133define <4 x i32> @test6(<8 x i32> %v) {
134; SSE2-LABEL: test6:
135; SSE2:       # %bb.0:
136; SSE2-NEXT:    retq
137;
138; AVX-LABEL: test6:
139; AVX:       # %bb.0:
140; AVX-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
141; AVX-NEXT:    vzeroupper
142; AVX-NEXT:    retq
143  %x = zext <8 x i32> %v to <8 x i64>
144  %s = shufflevector <8 x i64> %x, <8 x i64> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
145  %t = trunc <4 x i64> %s to <4 x i32>
146  ret <4 x i32> %t
147}
148
149define <4 x i32> @test7(<8 x i32> %v) {
150; SSE2-LABEL: test7:
151; SSE2:       # %bb.0:
152; SSE2-NEXT:    movaps %xmm1, %xmm0
153; SSE2-NEXT:    retq
154;
155; AVX-LABEL: test7:
156; AVX:       # %bb.0:
157; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm0
158; AVX-NEXT:    vzeroupper
159; AVX-NEXT:    retq
160  %x = zext <8 x i32> %v to <8 x i64>
161  %s = shufflevector <8 x i64> %x, <8 x i64> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
162  %t = trunc <4 x i64> %s to <4 x i32>
163  ret <4 x i32> %t
164}
165
166define <2 x i32> @test8(<8 x i32> %v) {
167; SSE2-LABEL: test8:
168; SSE2:       # %bb.0:
169; SSE2-NEXT:    movaps %xmm1, %xmm0
170; SSE2-NEXT:    xorps %xmm1, %xmm1
171; SSE2-NEXT:    unpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
172; SSE2-NEXT:    retq
173;
174; AVX2-LABEL: test8:
175; AVX2:       # %bb.0:
176; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
177; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
178; AVX2-NEXT:    vzeroupper
179; AVX2-NEXT:    retq
180;
181; AVX512-LABEL: test8:
182; AVX512:       # %bb.0:
183; AVX512-NEXT:    vpmovzxdq {{.*#+}} 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
184; AVX512-NEXT:    vextracti32x4 $2, %zmm0, %xmm0
185; AVX512-NEXT:    vzeroupper
186; AVX512-NEXT:    retq
187  %x = zext <8 x i32> %v to <8 x i64>
188  %s = shufflevector <8 x i64> %x, <8 x i64> undef, <2 x i32> <i32 4, i32 5>
189  %t = trunc <2 x i64> %s to <2 x i32>
190  ret <2 x i32> %t
191}
192
193define <2 x i32> @test9(<8 x i32> %v) {
194; SSE2-LABEL: test9:
195; SSE2:       # %bb.0:
196; SSE2-NEXT:    xorps %xmm1, %xmm1
197; SSE2-NEXT:    unpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
198; SSE2-NEXT:    retq
199;
200; AVX2-LABEL: test9:
201; AVX2:       # %bb.0:
202; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
203; AVX2-NEXT:    vzeroupper
204; AVX2-NEXT:    retq
205;
206; AVX512-LABEL: test9:
207; AVX512:       # %bb.0:
208; AVX512-NEXT:    vpmovzxdq {{.*#+}} 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
209; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
210; AVX512-NEXT:    vzeroupper
211; AVX512-NEXT:    retq
212  %x = zext <8 x i32> %v to <8 x i64>
213  %s = shufflevector <8 x i64> %x, <8 x i64> undef, <2 x i32> <i32 0, i32 1>
214  %t = trunc <2 x i64> %s to <2 x i32>
215  ret <2 x i32> %t
216}
217
218define <2 x i32> @test10(<8 x i32> %v) {
219; SSE2-LABEL: test10:
220; SSE2:       # %bb.0:
221; SSE2-NEXT:    xorpd %xmm2, %xmm2
222; SSE2-NEXT:    unpcklps {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
223; SSE2-NEXT:    unpckhps {{.*#+}} xmm0 = xmm0[2],xmm2[2],xmm0[3],xmm2[3]
224; SSE2-NEXT:    shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]
225; SSE2-NEXT:    retq
226;
227; AVX2-LABEL: test10:
228; AVX2:       # %bb.0:
229; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
230; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
231; AVX2-NEXT:    vpbroadcastd %xmm0, %ymm0
232; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
233; AVX2-NEXT:    vpalignr {{.*#+}} ymm0 = ymm1[8,9,10,11,12,13,14,15],ymm0[0,1,2,3,4,5,6,7],ymm1[24,25,26,27,28,29,30,31],ymm0[16,17,18,19,20,21,22,23]
234; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,2,3]
235; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
236; AVX2-NEXT:    vzeroupper
237; AVX2-NEXT:    retq
238;
239; AVX512-LABEL: test10:
240; AVX512:       # %bb.0:
241; AVX512-NEXT:    vpmovzxdq {{.*#+}} 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
242; AVX512-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
243; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm0
244; AVX512-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
245; AVX512-NEXT:    vzeroupper
246; AVX512-NEXT:    retq
247  %x = zext <8 x i32> %v to <8 x i64>
248  %s = shufflevector <8 x i64> %x, <8 x i64> undef, <2 x i32> <i32 3, i32 4>
249  %t = trunc <2 x i64> %s to <2 x i32>
250  ret <2 x i32> %t
251}
252