Gitiles
Code Review Sign In
gerrit-public.fairphone.software / toolchain / llvm-project / f6d840086999a5f23010caac81413c8f370a21d3 / . / llvm / test / CodeGen / X86 / vector-reduce-umin.ll
blob: 52adee5ab26ef7d8afefee79893aec9115e037ad [file] [log] [blame]
Simon Pilgrim57d32402018-04-05 17:25:40 +0000[diff] [blame]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=ALL --check-prefix=SSE --check-prefix=SSE2
3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE41
4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX1
5; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX2
6; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512BW
7; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512VL
8
9;
10; vXi64
11;
12
13define i64 @test_v2i64(<2 x i64> %a0) {
14; SSE2-LABEL: test_v2i64:
15; SSE2: # %bb.0:
16; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
Craig Topperf6d84002018-10-09 19:05:50 +0000[diff] [blame^]17; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
Simon Pilgrim57d32402018-04-05 17:25:40 +0000[diff] [blame]18; SSE2-NEXT: movdqa %xmm0, %xmm3
19; SSE2-NEXT: pxor %xmm2, %xmm3
20; SSE2-NEXT: pxor %xmm1, %xmm2
21; SSE2-NEXT: movdqa %xmm2, %xmm4
22; SSE2-NEXT: pcmpgtd %xmm3, %xmm4
23; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
24; SSE2-NEXT: pcmpeqd %xmm3, %xmm2
25; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
26; SSE2-NEXT: pand %xmm5, %xmm2
27; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
28; SSE2-NEXT: por %xmm2, %xmm3
29; SSE2-NEXT: pand %xmm3, %xmm0
30; SSE2-NEXT: pandn %xmm1, %xmm3
31; SSE2-NEXT: por %xmm0, %xmm3
32; SSE2-NEXT: movq %xmm3, %rax
33; SSE2-NEXT: retq
34;
35; SSE41-LABEL: test_v2i64:
36; SSE41: # %bb.0:
37; SSE41-NEXT: movdqa %xmm0, %xmm1
38; SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
Craig Topperf6d84002018-10-09 19:05:50 +0000[diff] [blame^]39; SSE41-NEXT: movdqa {{.*#+}} xmm0 = [9223372039002259456,9223372039002259456]
Simon Pilgrim57d32402018-04-05 17:25:40 +0000[diff] [blame]40; SSE41-NEXT: movdqa %xmm1, %xmm3
41; SSE41-NEXT: pxor %xmm0, %xmm3
42; SSE41-NEXT: pxor %xmm2, %xmm0
43; SSE41-NEXT: movdqa %xmm0, %xmm4
44; SSE41-NEXT: pcmpgtd %xmm3, %xmm4
45; SSE41-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
46; SSE41-NEXT: pcmpeqd %xmm3, %xmm0
47; SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
48; SSE41-NEXT: pand %xmm5, %xmm3
49; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm4[1,1,3,3]
50; SSE41-NEXT: por %xmm3, %xmm0
51; SSE41-NEXT: blendvpd %xmm0, %xmm1, %xmm2
52; SSE41-NEXT: movq %xmm2, %rax
53; SSE41-NEXT: retq
54;
55; AVX-LABEL: test_v2i64:
56; AVX: # %bb.0:
57; AVX-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
58; AVX-NEXT: vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
59; AVX-NEXT: vpxor %xmm2, %xmm0, %xmm3
60; AVX-NEXT: vpxor %xmm2, %xmm1, %xmm2
61; AVX-NEXT: vpcmpgtq %xmm3, %xmm2, %xmm2
62; AVX-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
63; AVX-NEXT: vmovq %xmm0, %rax
64; AVX-NEXT: retq
65;
66; AVX512BW-LABEL: test_v2i64:
67; AVX512BW: # %bb.0:
68; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
69; AVX512BW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
70; AVX512BW-NEXT: vpminuq %zmm1, %zmm0, %zmm0
71; AVX512BW-NEXT: vmovq %xmm0, %rax
72; AVX512BW-NEXT: vzeroupper
73; AVX512BW-NEXT: retq
74;
75; AVX512VL-LABEL: test_v2i64:
76; AVX512VL: # %bb.0:
77; AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
78; AVX512VL-NEXT: vpminuq %xmm1, %xmm0, %xmm0
79; AVX512VL-NEXT: vmovq %xmm0, %rax
80; AVX512VL-NEXT: retq
81 %1 = call i64 @llvm.experimental.vector.reduce.umin.i64.v2i64(<2 x i64> %a0)
82 ret i64 %1
83}
84
85define i64 @test_v4i64(<4 x i64> %a0) {
86; SSE2-LABEL: test_v4i64:
87; SSE2: # %bb.0:
Craig Topperf6d84002018-10-09 19:05:50 +0000[diff] [blame^]88; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
Simon Pilgrim57d32402018-04-05 17:25:40 +0000[diff] [blame]89; SSE2-NEXT: movdqa %xmm0, %xmm3
90; SSE2-NEXT: pxor %xmm2, %xmm3
91; SSE2-NEXT: movdqa %xmm1, %xmm4
92; SSE2-NEXT: pxor %xmm2, %xmm4
93; SSE2-NEXT: movdqa %xmm4, %xmm5
94; SSE2-NEXT: pcmpgtd %xmm3, %xmm5
95; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
96; SSE2-NEXT: pcmpeqd %xmm3, %xmm4
97; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
98; SSE2-NEXT: pand %xmm6, %xmm3
99; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm5[1,1,3,3]
100; SSE2-NEXT: por %xmm3, %xmm4
101; SSE2-NEXT: pand %xmm4, %xmm0
102; SSE2-NEXT: pandn %xmm1, %xmm4
103; SSE2-NEXT: por %xmm0, %xmm4
104; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm4[2,3,0,1]
105; SSE2-NEXT: movdqa %xmm4, %xmm1
106; SSE2-NEXT: pxor %xmm2, %xmm1
107; SSE2-NEXT: pxor %xmm0, %xmm2
108; SSE2-NEXT: movdqa %xmm2, %xmm3
109; SSE2-NEXT: pcmpgtd %xmm1, %xmm3
110; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm3[0,0,2,2]
111; SSE2-NEXT: pcmpeqd %xmm1, %xmm2
112; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
113; SSE2-NEXT: pand %xmm5, %xmm1
114; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[1,1,3,3]
115; SSE2-NEXT: por %xmm1, %xmm2
116; SSE2-NEXT: pand %xmm2, %xmm4
117; SSE2-NEXT: pandn %xmm0, %xmm2
118; SSE2-NEXT: por %xmm4, %xmm2
119; SSE2-NEXT: movq %xmm2, %rax
120; SSE2-NEXT: retq
121;
122; SSE41-LABEL: test_v4i64:
123; SSE41: # %bb.0:
124; SSE41-NEXT: movdqa %xmm0, %xmm2
Craig Topperf6d84002018-10-09 19:05:50 +0000[diff] [blame^]125; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [9223372039002259456,9223372039002259456]
Simon Pilgrim57d32402018-04-05 17:25:40 +0000[diff] [blame]126; SSE41-NEXT: pxor %xmm3, %xmm0
127; SSE41-NEXT: movdqa %xmm1, %xmm4
128; SSE41-NEXT: pxor %xmm3, %xmm4
129; SSE41-NEXT: movdqa %xmm4, %xmm5
130; SSE41-NEXT: pcmpgtd %xmm0, %xmm5
131; SSE41-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
132; SSE41-NEXT: pcmpeqd %xmm0, %xmm4
133; SSE41-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
134; SSE41-NEXT: pand %xmm6, %xmm4
135; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm5[1,1,3,3]
136; SSE41-NEXT: por %xmm4, %xmm0
137; SSE41-NEXT: blendvpd %xmm0, %xmm2, %xmm1
138; SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,0,1]
139; SSE41-NEXT: movdqa %xmm1, %xmm0
140; SSE41-NEXT: pxor %xmm3, %xmm0
141; SSE41-NEXT: pxor %xmm2, %xmm3
142; SSE41-NEXT: movdqa %xmm3, %xmm4
143; SSE41-NEXT: pcmpgtd %xmm0, %xmm4
144; SSE41-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
145; SSE41-NEXT: pcmpeqd %xmm0, %xmm3
146; SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
147; SSE41-NEXT: pand %xmm5, %xmm3
148; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm4[1,1,3,3]
149; SSE41-NEXT: por %xmm3, %xmm0
150; SSE41-NEXT: blendvpd %xmm0, %xmm1, %xmm2
151; SSE41-NEXT: movq %xmm2, %rax
152; SSE41-NEXT: retq
153;
154; AVX1-LABEL: test_v4i64:
155; AVX1: # %bb.0:
156; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
157; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
158; AVX1-NEXT: vpxor %xmm2, %xmm0, %xmm3
159; AVX1-NEXT: vpxor %xmm2, %xmm1, %xmm4
160; AVX1-NEXT: vpcmpgtq %xmm3, %xmm4, %xmm3
161; AVX1-NEXT: vpcmpgtq %xmm4, %xmm0, %xmm4
162; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm3, %ymm3
163; AVX1-NEXT: vblendvpd %ymm3, %ymm0, %ymm1, %ymm0
164; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[2,3,0,1]
165; AVX1-NEXT: vxorpd %xmm2, %xmm0, %xmm3
Sanjay Patel6ebf2182018-09-09 14:13:22 +0000[diff] [blame]166; AVX1-NEXT: vxorpd %xmm2, %xmm1, %xmm2
167; AVX1-NEXT: vpcmpgtq %xmm3, %xmm2, %xmm2
Simon Pilgrim57d32402018-04-05 17:25:40 +0000[diff] [blame]168; AVX1-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
169; AVX1-NEXT: vmovq %xmm0, %rax
170; AVX1-NEXT: vzeroupper
171; AVX1-NEXT: retq
172;
173; AVX2-LABEL: test_v4i64:
174; AVX2: # %bb.0:
175; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
176; AVX2-NEXT: vpbroadcastq {{.*#+}} ymm2 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
177; AVX2-NEXT: vpxor %ymm2, %ymm0, %ymm3
178; AVX2-NEXT: vpxor %ymm2, %ymm1, %ymm4
179; AVX2-NEXT: vpcmpgtq %ymm3, %ymm4, %ymm3
180; AVX2-NEXT: vblendvpd %ymm3, %ymm0, %ymm1, %ymm0
181; AVX2-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[2,3,0,1]
182; AVX2-NEXT: vxorpd %ymm2, %ymm0, %ymm3
183; AVX2-NEXT: vxorpd %ymm2, %ymm1, %ymm2
184; AVX2-NEXT: vpcmpgtq %ymm3, %ymm2, %ymm2
185; AVX2-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
186; AVX2-NEXT: vmovq %xmm0, %rax
187; AVX2-NEXT: vzeroupper
188; AVX2-NEXT: retq
189;
190; AVX512BW-LABEL: test_v4i64:
191; AVX512BW: # %bb.0:
192; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
193; AVX512BW-NEXT: vextracti128 $1, %ymm0, %xmm1
194; AVX512BW-NEXT: vpminuq %zmm1, %zmm0, %zmm0
195; AVX512BW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
196; AVX512BW-NEXT: vpminuq %zmm1, %zmm0, %zmm0
197; AVX512BW-NEXT: vmovq %xmm0, %rax
198; AVX512BW-NEXT: vzeroupper
199; AVX512BW-NEXT: retq
200;
201; AVX512VL-LABEL: test_v4i64:
202; AVX512VL: # %bb.0:
203; AVX512VL-NEXT: vextracti128 $1, %ymm0, %xmm1
204; AVX512VL-NEXT: vpminuq %ymm1, %ymm0, %ymm0
205; AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
206; AVX512VL-NEXT: vpminuq %ymm1, %ymm0, %ymm0
207; AVX512VL-NEXT: vmovq %xmm0, %rax
208; AVX512VL-NEXT: vzeroupper
209; AVX512VL-NEXT: retq
210 %1 = call i64 @llvm.experimental.vector.reduce.umin.i64.v4i64(<4 x i64> %a0)
211 ret i64 %1
212}
213
214define i64 @test_v8i64(<8 x i64> %a0) {
215; SSE2-LABEL: test_v8i64:
216; SSE2: # %bb.0:
Craig Topperf6d84002018-10-09 19:05:50 +0000[diff] [blame^]217; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [9223372039002259456,9223372039002259456]
Simon Pilgrim57d32402018-04-05 17:25:40 +0000[diff] [blame]218; SSE2-NEXT: movdqa %xmm1, %xmm5
219; SSE2-NEXT: pxor %xmm4, %xmm5
220; SSE2-NEXT: movdqa %xmm3, %xmm6
221; SSE2-NEXT: pxor %xmm4, %xmm6
222; SSE2-NEXT: movdqa %xmm6, %xmm7
223; SSE2-NEXT: pcmpgtd %xmm5, %xmm7
224; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
225; SSE2-NEXT: pcmpeqd %xmm5, %xmm6
226; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
227; SSE2-NEXT: pand %xmm8, %xmm6
228; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm7[1,1,3,3]
229; SSE2-NEXT: por %xmm6, %xmm5
230; SSE2-NEXT: pand %xmm5, %xmm1
231; SSE2-NEXT: pandn %xmm3, %xmm5
232; SSE2-NEXT: por %xmm1, %xmm5
233; SSE2-NEXT: movdqa %xmm0, %xmm1
234; SSE2-NEXT: pxor %xmm4, %xmm1
235; SSE2-NEXT: movdqa %xmm2, %xmm3
236; SSE2-NEXT: pxor %xmm4, %xmm3
237; SSE2-NEXT: movdqa %xmm3, %xmm6
238; SSE2-NEXT: pcmpgtd %xmm1, %xmm6
239; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
240; SSE2-NEXT: pcmpeqd %xmm1, %xmm3
241; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm3[1,1,3,3]
242; SSE2-NEXT: pand %xmm7, %xmm1
243; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm6[1,1,3,3]
244; SSE2-NEXT: por %xmm1, %xmm3
245; SSE2-NEXT: pand %xmm3, %xmm0
246; SSE2-NEXT: pandn %xmm2, %xmm3
247; SSE2-NEXT: por %xmm0, %xmm3
248; SSE2-NEXT: movdqa %xmm3, %xmm0
249; SSE2-NEXT: pxor %xmm4, %xmm0
250; SSE2-NEXT: movdqa %xmm5, %xmm1
251; SSE2-NEXT: pxor %xmm4, %xmm1
252; SSE2-NEXT: movdqa %xmm1, %xmm2
253; SSE2-NEXT: pcmpgtd %xmm0, %xmm2
254; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm2[0,0,2,2]
255; SSE2-NEXT: pcmpeqd %xmm0, %xmm1
256; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,3,3]
257; SSE2-NEXT: pand %xmm6, %xmm0
258; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
259; SSE2-NEXT: por %xmm0, %xmm1
260; SSE2-NEXT: pand %xmm1, %xmm3
261; SSE2-NEXT: pandn %xmm5, %xmm1
262; SSE2-NEXT: por %xmm3, %xmm1
263; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
264; SSE2-NEXT: movdqa %xmm1, %xmm2
265; SSE2-NEXT: pxor %xmm4, %xmm2
266; SSE2-NEXT: pxor %xmm0, %xmm4
267; SSE2-NEXT: movdqa %xmm4, %xmm3
268; SSE2-NEXT: pcmpgtd %xmm2, %xmm3
269; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm3[0,0,2,2]
270; SSE2-NEXT: pcmpeqd %xmm2, %xmm4
271; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
272; SSE2-NEXT: pand %xmm5, %xmm2
273; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
274; SSE2-NEXT: por %xmm2, %xmm3
275; SSE2-NEXT: pand %xmm3, %xmm1
276; SSE2-NEXT: pandn %xmm0, %xmm3
277; SSE2-NEXT: por %xmm1, %xmm3
278; SSE2-NEXT: movq %xmm3, %rax
279; SSE2-NEXT: retq
280;
281; SSE41-LABEL: test_v8i64:
282; SSE41: # %bb.0:
283; SSE41-NEXT: movdqa %xmm0, %xmm8
Craig Topperf6d84002018-10-09 19:05:50 +0000[diff] [blame^]284; SSE41-NEXT: movdqa {{.*#+}} xmm5 = [9223372039002259456,9223372039002259456]
Simon Pilgrim57d32402018-04-05 17:25:40 +0000[diff] [blame]285; SSE41-NEXT: movdqa %xmm1, %xmm0
286; SSE41-NEXT: pxor %xmm5, %xmm0
287; SSE41-NEXT: movdqa %xmm3, %xmm6
288; SSE41-NEXT: pxor %xmm5, %xmm6
289; SSE41-NEXT: movdqa %xmm6, %xmm7
290; SSE41-NEXT: pcmpgtd %xmm0, %xmm7
291; SSE41-NEXT: pshufd {{.*#+}} xmm4 = xmm7[0,0,2,2]
292; SSE41-NEXT: pcmpeqd %xmm0, %xmm6
293; SSE41-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
294; SSE41-NEXT: pand %xmm4, %xmm6
295; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm7[1,1,3,3]
296; SSE41-NEXT: por %xmm6, %xmm0
297; SSE41-NEXT: blendvpd %xmm0, %xmm1, %xmm3
298; SSE41-NEXT: movdqa %xmm8, %xmm0
299; SSE41-NEXT: pxor %xmm5, %xmm0
300; SSE41-NEXT: movdqa %xmm2, %xmm1
301; SSE41-NEXT: pxor %xmm5, %xmm1
302; SSE41-NEXT: movdqa %xmm1, %xmm4
303; SSE41-NEXT: pcmpgtd %xmm0, %xmm4
304; SSE41-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
305; SSE41-NEXT: pcmpeqd %xmm0, %xmm1
306; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
307; SSE41-NEXT: pand %xmm6, %xmm1
308; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm4[1,1,3,3]
309; SSE41-NEXT: por %xmm1, %xmm0
310; SSE41-NEXT: blendvpd %xmm0, %xmm8, %xmm2
311; SSE41-NEXT: movapd %xmm2, %xmm0
312; SSE41-NEXT: xorpd %xmm5, %xmm0
313; SSE41-NEXT: movapd %xmm3, %xmm1
314; SSE41-NEXT: xorpd %xmm5, %xmm1
315; SSE41-NEXT: movapd %xmm1, %xmm4
316; SSE41-NEXT: pcmpgtd %xmm0, %xmm4
317; SSE41-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
318; SSE41-NEXT: pcmpeqd %xmm0, %xmm1
319; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
320; SSE41-NEXT: pand %xmm6, %xmm1
321; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm4[1,1,3,3]
322; SSE41-NEXT: por %xmm1, %xmm0
323; SSE41-NEXT: blendvpd %xmm0, %xmm2, %xmm3
324; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm3[2,3,0,1]
325; SSE41-NEXT: movdqa %xmm3, %xmm0
326; SSE41-NEXT: pxor %xmm5, %xmm0
327; SSE41-NEXT: pxor %xmm1, %xmm5
328; SSE41-NEXT: movdqa %xmm5, %xmm2
329; SSE41-NEXT: pcmpgtd %xmm0, %xmm2
330; SSE41-NEXT: pshufd {{.*#+}} xmm4 = xmm2[0,0,2,2]
331; SSE41-NEXT: pcmpeqd %xmm0, %xmm5
332; SSE41-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
333; SSE41-NEXT: pand %xmm4, %xmm5
334; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
335; SSE41-NEXT: por %xmm5, %xmm0
336; SSE41-NEXT: blendvpd %xmm0, %xmm3, %xmm1
337; SSE41-NEXT: movq %xmm1, %rax
338; SSE41-NEXT: retq
339;
340; AVX1-LABEL: test_v8i64:
341; AVX1: # %bb.0:
342; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
343; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [9223372036854775808,9223372036854775808]
344; AVX1-NEXT: vpxor %xmm3, %xmm2, %xmm2
345; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm4
346; AVX1-NEXT: vpxor %xmm3, %xmm4, %xmm4
347; AVX1-NEXT: vpcmpgtq %xmm2, %xmm4, %xmm2
348; AVX1-NEXT: vpxor %xmm3, %xmm0, %xmm4
349; AVX1-NEXT: vpxor %xmm3, %xmm1, %xmm5
350; AVX1-NEXT: vpcmpgtq %xmm4, %xmm5, %xmm4
351; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm4, %ymm2
352; AVX1-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
353; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
354; AVX1-NEXT: vxorpd %xmm3, %xmm0, %xmm2
355; AVX1-NEXT: vpxor %xmm3, %xmm1, %xmm4
356; AVX1-NEXT: vpcmpgtq %xmm2, %xmm4, %xmm2
357; AVX1-NEXT: vpcmpgtq %xmm4, %xmm0, %xmm4
358; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm2, %ymm2
359; AVX1-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
360; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[2,3,0,1]
361; AVX1-NEXT: vxorpd %xmm3, %xmm0, %xmm2
Sanjay Patel6ebf2182018-09-09 14:13:22 +0000[diff] [blame]362; AVX1-NEXT: vxorpd %xmm3, %xmm1, %xmm3
363; AVX1-NEXT: vpcmpgtq %xmm2, %xmm3, %xmm2
Simon Pilgrim57d32402018-04-05 17:25:40 +0000[diff] [blame]364; AVX1-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
365; AVX1-NEXT: vmovq %xmm0, %rax
366; AVX1-NEXT: vzeroupper
367; AVX1-NEXT: retq
368;
369; AVX2-LABEL: test_v8i64:
370; AVX2: # %bb.0:
371; AVX2-NEXT: vpbroadcastq {{.*#+}} ymm2 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
372; AVX2-NEXT: vpxor %ymm2, %ymm0, %ymm3
373; AVX2-NEXT: vpxor %ymm2, %ymm1, %ymm4
374; AVX2-NEXT: vpcmpgtq %ymm3, %ymm4, %ymm3
375; AVX2-NEXT: vblendvpd %ymm3, %ymm0, %ymm1, %ymm0
376; AVX2-NEXT: vextractf128 $1, %ymm0, %xmm1
377; AVX2-NEXT: vxorpd %ymm2, %ymm0, %ymm3
378; AVX2-NEXT: vxorpd %ymm2, %ymm1, %ymm4
379; AVX2-NEXT: vpcmpgtq %ymm3, %ymm4, %ymm3
380; AVX2-NEXT: vblendvpd %ymm3, %ymm0, %ymm1, %ymm0
381; AVX2-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[2,3,0,1]
382; AVX2-NEXT: vxorpd %ymm2, %ymm0, %ymm3
383; AVX2-NEXT: vxorpd %ymm2, %ymm1, %ymm2
384; AVX2-NEXT: vpcmpgtq %ymm3, %ymm2, %ymm2
385; AVX2-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
386; AVX2-NEXT: vmovq %xmm0, %rax
387; AVX2-NEXT: vzeroupper
388; AVX2-NEXT: retq
389;
390; AVX512-LABEL: test_v8i64:
391; AVX512: # %bb.0:
392; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm1
393; AVX512-NEXT: vpminuq %zmm1, %zmm0, %zmm0
394; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
395; AVX512-NEXT: vpminuq %zmm1, %zmm0, %zmm0
396; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
397; AVX512-NEXT: vpminuq %zmm1, %zmm0, %zmm0
398; AVX512-NEXT: vmovq %xmm0, %rax
399; AVX512-NEXT: vzeroupper
400; AVX512-NEXT: retq
401 %1 = call i64 @llvm.experimental.vector.reduce.umin.i64.v8i64(<8 x i64> %a0)
402 ret i64 %1
403}
404
405define i64 @test_v16i64(<16 x i64> %a0) {
406; SSE2-LABEL: test_v16i64:
407; SSE2: # %bb.0:
Craig Topperf6d84002018-10-09 19:05:50 +0000[diff] [blame^]408; SSE2-NEXT: movdqa {{.*#+}} xmm8 = [9223372039002259456,9223372039002259456]
Simon Pilgrim57d32402018-04-05 17:25:40 +0000[diff] [blame]409; SSE2-NEXT: movdqa %xmm2, %xmm9
410; SSE2-NEXT: pxor %xmm8, %xmm9
411; SSE2-NEXT: movdqa %xmm6, %xmm10
412; SSE2-NEXT: pxor %xmm8, %xmm10
413; SSE2-NEXT: movdqa %xmm10, %xmm11
414; SSE2-NEXT: pcmpgtd %xmm9, %xmm11
415; SSE2-NEXT: pshufd {{.*#+}} xmm12 = xmm11[0,0,2,2]
416; SSE2-NEXT: pcmpeqd %xmm9, %xmm10
417; SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm10[1,1,3,3]
418; SSE2-NEXT: pand %xmm12, %xmm10
419; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm11[1,1,3,3]
420; SSE2-NEXT: por %xmm10, %xmm9
421; SSE2-NEXT: pand %xmm9, %xmm2
422; SSE2-NEXT: pandn %xmm6, %xmm9
423; SSE2-NEXT: por %xmm2, %xmm9
424; SSE2-NEXT: movdqa %xmm0, %xmm2
425; SSE2-NEXT: pxor %xmm8, %xmm2
426; SSE2-NEXT: movdqa %xmm4, %xmm6
427; SSE2-NEXT: pxor %xmm8, %xmm6
428; SSE2-NEXT: movdqa %xmm6, %xmm10
429; SSE2-NEXT: pcmpgtd %xmm2, %xmm10
430; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
431; SSE2-NEXT: pcmpeqd %xmm2, %xmm6
432; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
433; SSE2-NEXT: pand %xmm11, %xmm6
434; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm10[1,1,3,3]
435; SSE2-NEXT: por %xmm6, %xmm2
436; SSE2-NEXT: pand %xmm2, %xmm0
437; SSE2-NEXT: pandn %xmm4, %xmm2
438; SSE2-NEXT: por %xmm0, %xmm2
439; SSE2-NEXT: movdqa %xmm3, %xmm0
440; SSE2-NEXT: pxor %xmm8, %xmm0
441; SSE2-NEXT: movdqa %xmm7, %xmm4
442; SSE2-NEXT: pxor %xmm8, %xmm4
443; SSE2-NEXT: movdqa %xmm4, %xmm6
444; SSE2-NEXT: pcmpgtd %xmm0, %xmm6
445; SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm6[0,0,2,2]
446; SSE2-NEXT: pcmpeqd %xmm0, %xmm4
447; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
448; SSE2-NEXT: pand %xmm10, %xmm4
449; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm6[1,1,3,3]
450; SSE2-NEXT: por %xmm4, %xmm0
451; SSE2-NEXT: pand %xmm0, %xmm3
452; SSE2-NEXT: pandn %xmm7, %xmm0
453; SSE2-NEXT: por %xmm3, %xmm0
454; SSE2-NEXT: movdqa %xmm1, %xmm3
455; SSE2-NEXT: pxor %xmm8, %xmm3
456; SSE2-NEXT: movdqa %xmm5, %xmm4
457; SSE2-NEXT: pxor %xmm8, %xmm4
458; SSE2-NEXT: movdqa %xmm4, %xmm6
459; SSE2-NEXT: pcmpgtd %xmm3, %xmm6
460; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
461; SSE2-NEXT: pcmpeqd %xmm3, %xmm4
462; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
463; SSE2-NEXT: pand %xmm7, %xmm3
464; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm6[1,1,3,3]
465; SSE2-NEXT: por %xmm3, %xmm4
466; SSE2-NEXT: pand %xmm4, %xmm1
467; SSE2-NEXT: pandn %xmm5, %xmm4
468; SSE2-NEXT: por %xmm1, %xmm4
469; SSE2-NEXT: movdqa %xmm4, %xmm1
470; SSE2-NEXT: pxor %xmm8, %xmm1
471; SSE2-NEXT: movdqa %xmm0, %xmm3
472; SSE2-NEXT: pxor %xmm8, %xmm3
473; SSE2-NEXT: movdqa %xmm3, %xmm5
474; SSE2-NEXT: pcmpgtd %xmm1, %xmm5
475; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
476; SSE2-NEXT: pcmpeqd %xmm1, %xmm3
477; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
478; SSE2-NEXT: pand %xmm6, %xmm3
479; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm5[1,1,3,3]
480; SSE2-NEXT: por %xmm3, %xmm1
481; SSE2-NEXT: pand %xmm1, %xmm4
482; SSE2-NEXT: pandn %xmm0, %xmm1
483; SSE2-NEXT: por %xmm4, %xmm1
484; SSE2-NEXT: movdqa %xmm2, %xmm0
485; SSE2-NEXT: pxor %xmm8, %xmm0
486; SSE2-NEXT: movdqa %xmm9, %xmm3
487; SSE2-NEXT: pxor %xmm8, %xmm3
488; SSE2-NEXT: movdqa %xmm3, %xmm4
489; SSE2-NEXT: pcmpgtd %xmm0, %xmm4
490; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
491; SSE2-NEXT: pcmpeqd %xmm0, %xmm3
492; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm3[1,1,3,3]
493; SSE2-NEXT: pand %xmm5, %xmm0
494; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
495; SSE2-NEXT: por %xmm0, %xmm3
496; SSE2-NEXT: pand %xmm3, %xmm2
497; SSE2-NEXT: pandn %xmm9, %xmm3
498; SSE2-NEXT: por %xmm2, %xmm3
499; SSE2-NEXT: movdqa %xmm3, %xmm0
500; SSE2-NEXT: pxor %xmm8, %xmm0
501; SSE2-NEXT: movdqa %xmm1, %xmm2
502; SSE2-NEXT: pxor %xmm8, %xmm2
503; SSE2-NEXT: movdqa %xmm2, %xmm4
504; SSE2-NEXT: pcmpgtd %xmm0, %xmm4
505; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
506; SSE2-NEXT: pcmpeqd %xmm0, %xmm2
507; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
508; SSE2-NEXT: pand %xmm5, %xmm0
509; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
510; SSE2-NEXT: por %xmm0, %xmm2
511; SSE2-NEXT: pand %xmm2, %xmm3
512; SSE2-NEXT: pandn %xmm1, %xmm2
513; SSE2-NEXT: por %xmm3, %xmm2
514; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
515; SSE2-NEXT: movdqa %xmm2, %xmm1
516; SSE2-NEXT: pxor %xmm8, %xmm1
517; SSE2-NEXT: pxor %xmm0, %xmm8
518; SSE2-NEXT: movdqa %xmm8, %xmm3
519; SSE2-NEXT: pcmpgtd %xmm1, %xmm3
520; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
521; SSE2-NEXT: pcmpeqd %xmm1, %xmm8
522; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm8[1,1,3,3]
523; SSE2-NEXT: pand %xmm4, %xmm1
524; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
525; SSE2-NEXT: por %xmm1, %xmm3
526; SSE2-NEXT: pand %xmm3, %xmm2
527; SSE2-NEXT: pandn %xmm0, %xmm3
528; SSE2-NEXT: por %xmm2, %xmm3
529; SSE2-NEXT: movq %xmm3, %rax
530; SSE2-NEXT: retq
531;
532; SSE41-LABEL: test_v16i64:
533; SSE41: # %bb.0:
534; SSE41-NEXT: movdqa %xmm0, %xmm8
Craig Topperf6d84002018-10-09 19:05:50 +0000[diff] [blame^]535; SSE41-NEXT: movdqa {{.*#+}} xmm9 = [9223372039002259456,9223372039002259456]
Simon Pilgrim57d32402018-04-05 17:25:40 +0000[diff] [blame]536; SSE41-NEXT: movdqa %xmm2, %xmm10
537; SSE41-NEXT: pxor %xmm9, %xmm10
538; SSE41-NEXT: movdqa %xmm6, %xmm0
539; SSE41-NEXT: pxor %xmm9, %xmm0
540; SSE41-NEXT: movdqa %xmm0, %xmm11
541; SSE41-NEXT: pcmpgtd %xmm10, %xmm11
542; SSE41-NEXT: pshufd {{.*#+}} xmm12 = xmm11[0,0,2,2]
543; SSE41-NEXT: pcmpeqd %xmm10, %xmm0
544; SSE41-NEXT: pshufd {{.*#+}} xmm10 = xmm0[1,1,3,3]
545; SSE41-NEXT: pand %xmm12, %xmm10
546; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm11[1,1,3,3]
547; SSE41-NEXT: por %xmm10, %xmm0
548; SSE41-NEXT: blendvpd %xmm0, %xmm2, %xmm6
549; SSE41-NEXT: movdqa %xmm8, %xmm0
550; SSE41-NEXT: pxor %xmm9, %xmm0
551; SSE41-NEXT: movdqa %xmm4, %xmm2
552; SSE41-NEXT: pxor %xmm9, %xmm2
553; SSE41-NEXT: movdqa %xmm2, %xmm10
554; SSE41-NEXT: pcmpgtd %xmm0, %xmm10
555; SSE41-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
556; SSE41-NEXT: pcmpeqd %xmm0, %xmm2
557; SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
558; SSE41-NEXT: pand %xmm11, %xmm2
559; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm10[1,1,3,3]
560; SSE41-NEXT: por %xmm2, %xmm0
561; SSE41-NEXT: blendvpd %xmm0, %xmm8, %xmm4
562; SSE41-NEXT: movdqa %xmm3, %xmm0
563; SSE41-NEXT: pxor %xmm9, %xmm0
564; SSE41-NEXT: movdqa %xmm7, %xmm2
565; SSE41-NEXT: pxor %xmm9, %xmm2
566; SSE41-NEXT: movdqa %xmm2, %xmm8
567; SSE41-NEXT: pcmpgtd %xmm0, %xmm8
568; SSE41-NEXT: pshufd {{.*#+}} xmm10 = xmm8[0,0,2,2]
569; SSE41-NEXT: pcmpeqd %xmm0, %xmm2
570; SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
571; SSE41-NEXT: pand %xmm10, %xmm2
572; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm8[1,1,3,3]
573; SSE41-NEXT: por %xmm2, %xmm0
574; SSE41-NEXT: blendvpd %xmm0, %xmm3, %xmm7
575; SSE41-NEXT: movdqa %xmm1, %xmm0
576; SSE41-NEXT: pxor %xmm9, %xmm0
577; SSE41-NEXT: movdqa %xmm5, %xmm2
578; SSE41-NEXT: pxor %xmm9, %xmm2
579; SSE41-NEXT: movdqa %xmm2, %xmm3
580; SSE41-NEXT: pcmpgtd %xmm0, %xmm3
581; SSE41-NEXT: pshufd {{.*#+}} xmm8 = xmm3[0,0,2,2]
582; SSE41-NEXT: pcmpeqd %xmm0, %xmm2
583; SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
584; SSE41-NEXT: pand %xmm8, %xmm2
585; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm3[1,1,3,3]
586; SSE41-NEXT: por %xmm2, %xmm0
587; SSE41-NEXT: blendvpd %xmm0, %xmm1, %xmm5
588; SSE41-NEXT: movapd %xmm5, %xmm0
589; SSE41-NEXT: xorpd %xmm9, %xmm0
590; SSE41-NEXT: movapd %xmm7, %xmm1
591; SSE41-NEXT: xorpd %xmm9, %xmm1
592; SSE41-NEXT: movapd %xmm1, %xmm2
593; SSE41-NEXT: pcmpgtd %xmm0, %xmm2
594; SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[0,0,2,2]
595; SSE41-NEXT: pcmpeqd %xmm0, %xmm1
596; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
597; SSE41-NEXT: pand %xmm3, %xmm1
598; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
599; SSE41-NEXT: por %xmm1, %xmm0
600; SSE41-NEXT: blendvpd %xmm0, %xmm5, %xmm7
601; SSE41-NEXT: movapd %xmm4, %xmm0
602; SSE41-NEXT: xorpd %xmm9, %xmm0
603; SSE41-NEXT: movapd %xmm6, %xmm1
604; SSE41-NEXT: xorpd %xmm9, %xmm1
605; SSE41-NEXT: movapd %xmm1, %xmm2
606; SSE41-NEXT: pcmpgtd %xmm0, %xmm2
607; SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[0,0,2,2]
608; SSE41-NEXT: pcmpeqd %xmm0, %xmm1
609; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
610; SSE41-NEXT: pand %xmm3, %xmm1
611; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
612; SSE41-NEXT: por %xmm1, %xmm0
613; SSE41-NEXT: blendvpd %xmm0, %xmm4, %xmm6
614; SSE41-NEXT: movapd %xmm6, %xmm0
615; SSE41-NEXT: xorpd %xmm9, %xmm0
616; SSE41-NEXT: movapd %xmm7, %xmm1
617; SSE41-NEXT: xorpd %xmm9, %xmm1
618; SSE41-NEXT: movapd %xmm1, %xmm2
619; SSE41-NEXT: pcmpgtd %xmm0, %xmm2
620; SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[0,0,2,2]
621; SSE41-NEXT: pcmpeqd %xmm0, %xmm1
622; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
623; SSE41-NEXT: pand %xmm3, %xmm1
624; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
625; SSE41-NEXT: por %xmm1, %xmm0
626; SSE41-NEXT: blendvpd %xmm0, %xmm6, %xmm7
627; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm7[2,3,0,1]
628; SSE41-NEXT: movdqa %xmm7, %xmm0
629; SSE41-NEXT: pxor %xmm9, %xmm0
630; SSE41-NEXT: pxor %xmm1, %xmm9
631; SSE41-NEXT: movdqa %xmm9, %xmm2
632; SSE41-NEXT: pcmpgtd %xmm0, %xmm2
633; SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[0,0,2,2]
634; SSE41-NEXT: pcmpeqd %xmm0, %xmm9
635; SSE41-NEXT: pshufd {{.*#+}} xmm4 = xmm9[1,1,3,3]
636; SSE41-NEXT: pand %xmm3, %xmm4
637; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
638; SSE41-NEXT: por %xmm4, %xmm0
639; SSE41-NEXT: blendvpd %xmm0, %xmm7, %xmm1
640; SSE41-NEXT: movq %xmm1, %rax
641; SSE41-NEXT: retq
642;
643; AVX1-LABEL: test_v16i64:
644; AVX1: # %bb.0:
645; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5
646; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [9223372036854775808,9223372036854775808]
647; AVX1-NEXT: vpxor %xmm4, %xmm5, %xmm5
648; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm6
649; AVX1-NEXT: vpxor %xmm4, %xmm6, %xmm6
650; AVX1-NEXT: vpcmpgtq %xmm5, %xmm6, %xmm5
651; AVX1-NEXT: vpxor %xmm4, %xmm1, %xmm6
652; AVX1-NEXT: vpxor %xmm4, %xmm3, %xmm7
653; AVX1-NEXT: vpcmpgtq %xmm6, %xmm7, %xmm6
654; AVX1-NEXT: vinsertf128 $1, %xmm5, %ymm6, %ymm5
655; AVX1-NEXT: vblendvpd %ymm5, %ymm1, %ymm3, %ymm1
656; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
657; AVX1-NEXT: vpxor %xmm4, %xmm3, %xmm3
658; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm5
659; AVX1-NEXT: vpxor %xmm4, %xmm5, %xmm5
660; AVX1-NEXT: vpcmpgtq %xmm3, %xmm5, %xmm3
661; AVX1-NEXT: vpxor %xmm4, %xmm0, %xmm5
662; AVX1-NEXT: vpxor %xmm4, %xmm2, %xmm6
663; AVX1-NEXT: vpcmpgtq %xmm5, %xmm6, %xmm5
664; AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm5, %ymm3
665; AVX1-NEXT: vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
666; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
667; AVX1-NEXT: vpxor %xmm4, %xmm2, %xmm2
668; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm3
669; AVX1-NEXT: vpxor %xmm4, %xmm3, %xmm3
670; AVX1-NEXT: vpcmpgtq %xmm2, %xmm3, %xmm2
671; AVX1-NEXT: vxorpd %xmm4, %xmm0, %xmm3
672; AVX1-NEXT: vxorpd %xmm4, %xmm1, %xmm5
673; AVX1-NEXT: vpcmpgtq %xmm3, %xmm5, %xmm3
674; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm3, %ymm2
675; AVX1-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
676; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
677; AVX1-NEXT: vxorpd %xmm4, %xmm0, %xmm2
678; AVX1-NEXT: vpxor %xmm4, %xmm1, %xmm3
679; AVX1-NEXT: vpcmpgtq %xmm2, %xmm3, %xmm2
680; AVX1-NEXT: vpcmpgtq %xmm3, %xmm0, %xmm3
681; AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm2, %ymm2
682; AVX1-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
683; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[2,3,0,1]
684; AVX1-NEXT: vxorpd %xmm4, %xmm0, %xmm2
685; AVX1-NEXT: vxorpd %xmm4, %xmm1, %xmm3
686; AVX1-NEXT: vpcmpgtq %xmm2, %xmm3, %xmm2
Simon Pilgrim57d32402018-04-05 17:25:40 +0000[diff] [blame]687; AVX1-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
688; AVX1-NEXT: vmovq %xmm0, %rax
689; AVX1-NEXT: vzeroupper
690; AVX1-NEXT: retq
691;
692; AVX2-LABEL: test_v16i64:
693; AVX2: # %bb.0:
694; AVX2-NEXT: vpbroadcastq {{.*#+}} ymm4 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
695; AVX2-NEXT: vpxor %ymm4, %ymm1, %ymm5
696; AVX2-NEXT: vpxor %ymm4, %ymm3, %ymm6
697; AVX2-NEXT: vpcmpgtq %ymm5, %ymm6, %ymm5
698; AVX2-NEXT: vblendvpd %ymm5, %ymm1, %ymm3, %ymm1
699; AVX2-NEXT: vpxor %ymm4, %ymm0, %ymm3
700; AVX2-NEXT: vpxor %ymm4, %ymm2, %ymm5
701; AVX2-NEXT: vpcmpgtq %ymm3, %ymm5, %ymm3
702; AVX2-NEXT: vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
703; AVX2-NEXT: vxorpd %ymm4, %ymm0, %ymm2
704; AVX2-NEXT: vxorpd %ymm4, %ymm1, %ymm3
705; AVX2-NEXT: vpcmpgtq %ymm2, %ymm3, %ymm2
706; AVX2-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
707; AVX2-NEXT: vextractf128 $1, %ymm0, %xmm1
708; AVX2-NEXT: vxorpd %ymm4, %ymm0, %ymm2
709; AVX2-NEXT: vxorpd %ymm4, %ymm1, %ymm3
710; AVX2-NEXT: vpcmpgtq %ymm2, %ymm3, %ymm2
711; AVX2-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
712; AVX2-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[2,3,0,1]
713; AVX2-NEXT: vxorpd %ymm4, %ymm0, %ymm2
714; AVX2-NEXT: vxorpd %ymm4, %ymm1, %ymm3
715; AVX2-NEXT: vpcmpgtq %ymm2, %ymm3, %ymm2
716; AVX2-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
717; AVX2-NEXT: vmovq %xmm0, %rax
718; AVX2-NEXT: vzeroupper
719; AVX2-NEXT: retq
720;
721; AVX512-LABEL: test_v16i64:
722; AVX512: # %bb.0:
723; AVX512-NEXT: vpminuq %zmm1, %zmm0, %zmm0
724; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm1
725; AVX512-NEXT: vpminuq %zmm1, %zmm0, %zmm0
726; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
727; AVX512-NEXT: vpminuq %zmm1, %zmm0, %zmm0
728; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
729; AVX512-NEXT: vpminuq %zmm1, %zmm0, %zmm0
730; AVX512-NEXT: vmovq %xmm0, %rax
731; AVX512-NEXT: vzeroupper
732; AVX512-NEXT: retq
733 %1 = call i64 @llvm.experimental.vector.reduce.umin.i64.v16i64(<16 x i64> %a0)
734 ret i64 %1
735}
736
737;
738; vXi32
739;
740
741define i32 @test_v4i32(<4 x i32> %a0) {
742; SSE2-LABEL: test_v4i32:
743; SSE2: # %bb.0:
744; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
745; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
746; SSE2-NEXT: movdqa %xmm0, %xmm3
747; SSE2-NEXT: pxor %xmm2, %xmm3
748; SSE2-NEXT: movdqa %xmm1, %xmm4
749; SSE2-NEXT: pxor %xmm2, %xmm4
750; SSE2-NEXT: pcmpgtd %xmm3, %xmm4
751; SSE2-NEXT: pand %xmm4, %xmm0
752; SSE2-NEXT: pandn %xmm1, %xmm4
753; SSE2-NEXT: por %xmm0, %xmm4
754; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm4[1,1,2,3]
755; SSE2-NEXT: movdqa %xmm4, %xmm1
756; SSE2-NEXT: pxor %xmm2, %xmm1
757; SSE2-NEXT: pxor %xmm0, %xmm2
758; SSE2-NEXT: pcmpgtd %xmm1, %xmm2
759; SSE2-NEXT: pand %xmm2, %xmm4
760; SSE2-NEXT: pandn %xmm0, %xmm2
761; SSE2-NEXT: por %xmm4, %xmm2
762; SSE2-NEXT: movd %xmm2, %eax
763; SSE2-NEXT: retq
764;
765; SSE41-LABEL: test_v4i32:
766; SSE41: # %bb.0:
767; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
768; SSE41-NEXT: pminud %xmm0, %xmm1
769; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
770; SSE41-NEXT: pminud %xmm1, %xmm0
771; SSE41-NEXT: movd %xmm0, %eax
772; SSE41-NEXT: retq
773;
774; AVX-LABEL: test_v4i32:
775; AVX: # %bb.0:
776; AVX-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
777; AVX-NEXT: vpminud %xmm1, %xmm0, %xmm0
778; AVX-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
779; AVX-NEXT: vpminud %xmm1, %xmm0, %xmm0
780; AVX-NEXT: vmovd %xmm0, %eax
781; AVX-NEXT: retq
782;
783; AVX512-LABEL: test_v4i32:
784; AVX512: # %bb.0:
785; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
786; AVX512-NEXT: vpminud %xmm1, %xmm0, %xmm0
787; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
788; AVX512-NEXT: vpminud %xmm1, %xmm0, %xmm0
789; AVX512-NEXT: vmovd %xmm0, %eax
790; AVX512-NEXT: retq
791 %1 = call i32 @llvm.experimental.vector.reduce.umin.i32.v4i32(<4 x i32> %a0)
792 ret i32 %1
793}
794
795define i32 @test_v8i32(<8 x i32> %a0) {
796; SSE2-LABEL: test_v8i32:
797; SSE2: # %bb.0:
798; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
799; SSE2-NEXT: movdqa %xmm0, %xmm3
800; SSE2-NEXT: pxor %xmm2, %xmm3
801; SSE2-NEXT: movdqa %xmm1, %xmm4
802; SSE2-NEXT: pxor %xmm2, %xmm4
803; SSE2-NEXT: pcmpgtd %xmm3, %xmm4
804; SSE2-NEXT: pand %xmm4, %xmm0
805; SSE2-NEXT: pandn %xmm1, %xmm4
806; SSE2-NEXT: por %xmm0, %xmm4
807; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm4[2,3,0,1]
808; SSE2-NEXT: movdqa %xmm4, %xmm1
809; SSE2-NEXT: pxor %xmm2, %xmm1
810; SSE2-NEXT: movdqa %xmm0, %xmm3
811; SSE2-NEXT: pxor %xmm2, %xmm3
812; SSE2-NEXT: pcmpgtd %xmm1, %xmm3
813; SSE2-NEXT: pand %xmm3, %xmm4
814; SSE2-NEXT: pandn %xmm0, %xmm3
815; SSE2-NEXT: por %xmm4, %xmm3
816; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm3[1,1,2,3]
817; SSE2-NEXT: movdqa %xmm3, %xmm1
818; SSE2-NEXT: pxor %xmm2, %xmm1
819; SSE2-NEXT: pxor %xmm0, %xmm2
820; SSE2-NEXT: pcmpgtd %xmm1, %xmm2
821; SSE2-NEXT: pand %xmm2, %xmm3
822; SSE2-NEXT: pandn %xmm0, %xmm2
823; SSE2-NEXT: por %xmm3, %xmm2
824; SSE2-NEXT: movd %xmm2, %eax
825; SSE2-NEXT: retq
826;
827; SSE41-LABEL: test_v8i32:
828; SSE41: # %bb.0:
829; SSE41-NEXT: pminud %xmm1, %xmm0
830; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
831; SSE41-NEXT: pminud %xmm0, %xmm1
832; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
833; SSE41-NEXT: pminud %xmm1, %xmm0
834; SSE41-NEXT: movd %xmm0, %eax
835; SSE41-NEXT: retq
836;
837; AVX1-LABEL: test_v8i32:
838; AVX1: # %bb.0:
839; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
840; AVX1-NEXT: vpminud %xmm1, %xmm0, %xmm0
841; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
842; AVX1-NEXT: vpminud %xmm1, %xmm0, %xmm0
843; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
844; AVX1-NEXT: vpminud %xmm1, %xmm0, %xmm0
845; AVX1-NEXT: vmovd %xmm0, %eax
846; AVX1-NEXT: vzeroupper
847; AVX1-NEXT: retq
848;
849; AVX2-LABEL: test_v8i32:
850; AVX2: # %bb.0:
851; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
852; AVX2-NEXT: vpminud %ymm1, %ymm0, %ymm0
853; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
854; AVX2-NEXT: vpminud %ymm1, %ymm0, %ymm0
855; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
856; AVX2-NEXT: vpminud %ymm1, %ymm0, %ymm0
857; AVX2-NEXT: vmovd %xmm0, %eax
858; AVX2-NEXT: vzeroupper
859; AVX2-NEXT: retq
860;
861; AVX512-LABEL: test_v8i32:
862; AVX512: # %bb.0:
863; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
864; AVX512-NEXT: vpminud %ymm1, %ymm0, %ymm0
865; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
866; AVX512-NEXT: vpminud %ymm1, %ymm0, %ymm0
867; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
868; AVX512-NEXT: vpminud %ymm1, %ymm0, %ymm0
869; AVX512-NEXT: vmovd %xmm0, %eax
870; AVX512-NEXT: vzeroupper
871; AVX512-NEXT: retq
872 %1 = call i32 @llvm.experimental.vector.reduce.umin.i32.v8i32(<8 x i32> %a0)
873 ret i32 %1
874}
875
876define i32 @test_v16i32(<16 x i32> %a0) {
877; SSE2-LABEL: test_v16i32:
878; SSE2: # %bb.0:
879; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [2147483648,2147483648,2147483648,2147483648]
880; SSE2-NEXT: movdqa %xmm1, %xmm5
881; SSE2-NEXT: pxor %xmm4, %xmm5
882; SSE2-NEXT: movdqa %xmm3, %xmm6
883; SSE2-NEXT: pxor %xmm4, %xmm6
884; SSE2-NEXT: pcmpgtd %xmm5, %xmm6
885; SSE2-NEXT: pand %xmm6, %xmm1
886; SSE2-NEXT: pandn %xmm3, %xmm6
887; SSE2-NEXT: por %xmm1, %xmm6
888; SSE2-NEXT: movdqa %xmm0, %xmm1
889; SSE2-NEXT: pxor %xmm4, %xmm1
890; SSE2-NEXT: movdqa %xmm2, %xmm3
891; SSE2-NEXT: pxor %xmm4, %xmm3
892; SSE2-NEXT: pcmpgtd %xmm1, %xmm3
893; SSE2-NEXT: pand %xmm3, %xmm0
894; SSE2-NEXT: pandn %xmm2, %xmm3
895; SSE2-NEXT: por %xmm0, %xmm3
896; SSE2-NEXT: movdqa %xmm3, %xmm0
897; SSE2-NEXT: pxor %xmm4, %xmm0
898; SSE2-NEXT: movdqa %xmm6, %xmm1
899; SSE2-NEXT: pxor %xmm4, %xmm1
900; SSE2-NEXT: pcmpgtd %xmm0, %xmm1
901; SSE2-NEXT: pand %xmm1, %xmm3
902; SSE2-NEXT: pandn %xmm6, %xmm1
903; SSE2-NEXT: por %xmm3, %xmm1
904; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
905; SSE2-NEXT: movdqa %xmm1, %xmm2
906; SSE2-NEXT: pxor %xmm4, %xmm2
907; SSE2-NEXT: movdqa %xmm0, %xmm3
908; SSE2-NEXT: pxor %xmm4, %xmm3
909; SSE2-NEXT: pcmpgtd %xmm2, %xmm3
910; SSE2-NEXT: pand %xmm3, %xmm1
911; SSE2-NEXT: pandn %xmm0, %xmm3
912; SSE2-NEXT: por %xmm1, %xmm3
913; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm3[1,1,2,3]
914; SSE2-NEXT: movdqa %xmm3, %xmm1
915; SSE2-NEXT: pxor %xmm4, %xmm1
916; SSE2-NEXT: pxor %xmm0, %xmm4
917; SSE2-NEXT: pcmpgtd %xmm1, %xmm4
918; SSE2-NEXT: pand %xmm4, %xmm3
919; SSE2-NEXT: pandn %xmm0, %xmm4
920; SSE2-NEXT: por %xmm3, %xmm4
921; SSE2-NEXT: movd %xmm4, %eax
922; SSE2-NEXT: retq
923;
924; SSE41-LABEL: test_v16i32:
925; SSE41: # %bb.0:
926; SSE41-NEXT: pminud %xmm3, %xmm1
927; SSE41-NEXT: pminud %xmm2, %xmm0
928; SSE41-NEXT: pminud %xmm1, %xmm0
929; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
930; SSE41-NEXT: pminud %xmm0, %xmm1
931; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
932; SSE41-NEXT: pminud %xmm1, %xmm0
933; SSE41-NEXT: movd %xmm0, %eax
934; SSE41-NEXT: retq
935;
936; AVX1-LABEL: test_v16i32:
937; AVX1: # %bb.0:
938; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
939; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
940; AVX1-NEXT: vpminud %xmm2, %xmm3, %xmm2
941; AVX1-NEXT: vpminud %xmm1, %xmm0, %xmm0
942; AVX1-NEXT: vpminud %xmm2, %xmm0, %xmm0
943; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
944; AVX1-NEXT: vpminud %xmm1, %xmm0, %xmm0
945; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
946; AVX1-NEXT: vpminud %xmm1, %xmm0, %xmm0
947; AVX1-NEXT: vmovd %xmm0, %eax
948; AVX1-NEXT: vzeroupper
949; AVX1-NEXT: retq
950;
951; AVX2-LABEL: test_v16i32:
952; AVX2: # %bb.0:
953; AVX2-NEXT: vpminud %ymm1, %ymm0, %ymm0
954; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
955; AVX2-NEXT: vpminud %ymm1, %ymm0, %ymm0
956; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
957; AVX2-NEXT: vpminud %ymm1, %ymm0, %ymm0
958; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
959; AVX2-NEXT: vpminud %ymm1, %ymm0, %ymm0
960; AVX2-NEXT: vmovd %xmm0, %eax
961; AVX2-NEXT: vzeroupper
962; AVX2-NEXT: retq
963;
964; AVX512-LABEL: test_v16i32:
965; AVX512: # %bb.0:
966; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm1
967; AVX512-NEXT: vpminud %zmm1, %zmm0, %zmm0
968; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
969; AVX512-NEXT: vpminud %zmm1, %zmm0, %zmm0
970; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
971; AVX512-NEXT: vpminud %zmm1, %zmm0, %zmm0
972; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
973; AVX512-NEXT: vpminud %zmm1, %zmm0, %zmm0
974; AVX512-NEXT: vmovd %xmm0, %eax
975; AVX512-NEXT: vzeroupper
976; AVX512-NEXT: retq
977 %1 = call i32 @llvm.experimental.vector.reduce.umin.i32.v16i32(<16 x i32> %a0)
978 ret i32 %1
979}
980
981define i32 @test_v32i32(<32 x i32> %a0) {
982; SSE2-LABEL: test_v32i32:
983; SSE2: # %bb.0:
984; SSE2-NEXT: movdqa {{.*#+}} xmm8 = [2147483648,2147483648,2147483648,2147483648]
985; SSE2-NEXT: movdqa %xmm2, %xmm10
986; SSE2-NEXT: pxor %xmm8, %xmm10
987; SSE2-NEXT: movdqa %xmm6, %xmm9
988; SSE2-NEXT: pxor %xmm8, %xmm9
989; SSE2-NEXT: pcmpgtd %xmm10, %xmm9
990; SSE2-NEXT: pand %xmm9, %xmm2
991; SSE2-NEXT: pandn %xmm6, %xmm9
992; SSE2-NEXT: por %xmm2, %xmm9
993; SSE2-NEXT: movdqa %xmm0, %xmm6
994; SSE2-NEXT: pxor %xmm8, %xmm6
995; SSE2-NEXT: movdqa %xmm4, %xmm2
996; SSE2-NEXT: pxor %xmm8, %xmm2
997; SSE2-NEXT: pcmpgtd %xmm6, %xmm2
998; SSE2-NEXT: pand %xmm2, %xmm0
999; SSE2-NEXT: pandn %xmm4, %xmm2
1000; SSE2-NEXT: por %xmm0, %xmm2
1001; SSE2-NEXT: movdqa %xmm3, %xmm0
1002; SSE2-NEXT: pxor %xmm8, %xmm0
1003; SSE2-NEXT: movdqa %xmm7, %xmm4
1004; SSE2-NEXT: pxor %xmm8, %xmm4
1005; SSE2-NEXT: pcmpgtd %xmm0, %xmm4
1006; SSE2-NEXT: pand %xmm4, %xmm3
1007; SSE2-NEXT: pandn %xmm7, %xmm4
1008; SSE2-NEXT: por %xmm3, %xmm4
1009; SSE2-NEXT: movdqa %xmm1, %xmm0
1010; SSE2-NEXT: pxor %xmm8, %xmm0
1011; SSE2-NEXT: movdqa %xmm5, %xmm3
1012; SSE2-NEXT: pxor %xmm8, %xmm3
1013; SSE2-NEXT: pcmpgtd %xmm0, %xmm3
1014; SSE2-NEXT: pand %xmm3, %xmm1
1015; SSE2-NEXT: pandn %xmm5, %xmm3
1016; SSE2-NEXT: por %xmm1, %xmm3
1017; SSE2-NEXT: movdqa %xmm3, %xmm0
1018; SSE2-NEXT: pxor %xmm8, %xmm0
1019; SSE2-NEXT: movdqa %xmm4, %xmm1
1020; SSE2-NEXT: pxor %xmm8, %xmm1
1021; SSE2-NEXT: pcmpgtd %xmm0, %xmm1
1022; SSE2-NEXT: pand %xmm1, %xmm3
1023; SSE2-NEXT: pandn %xmm4, %xmm1
1024; SSE2-NEXT: por %xmm3, %xmm1
1025; SSE2-NEXT: movdqa %xmm2, %xmm0
1026; SSE2-NEXT: pxor %xmm8, %xmm0
1027; SSE2-NEXT: movdqa %xmm9, %xmm3
1028; SSE2-NEXT: pxor %xmm8, %xmm3
1029; SSE2-NEXT: pcmpgtd %xmm0, %xmm3
1030; SSE2-NEXT: pand %xmm3, %xmm2
1031; SSE2-NEXT: pandn %xmm9, %xmm3
1032; SSE2-NEXT: por %xmm2, %xmm3
1033; SSE2-NEXT: movdqa %xmm3, %xmm0
1034; SSE2-NEXT: pxor %xmm8, %xmm0
1035; SSE2-NEXT: movdqa %xmm1, %xmm2
1036; SSE2-NEXT: pxor %xmm8, %xmm2
1037; SSE2-NEXT: pcmpgtd %xmm0, %xmm2
1038; SSE2-NEXT: pand %xmm2, %xmm3
1039; SSE2-NEXT: pandn %xmm1, %xmm2
1040; SSE2-NEXT: por %xmm3, %xmm2
1041; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
1042; SSE2-NEXT: movdqa %xmm2, %xmm1
1043; SSE2-NEXT: pxor %xmm8, %xmm1
1044; SSE2-NEXT: movdqa %xmm0, %xmm3
1045; SSE2-NEXT: pxor %xmm8, %xmm3
1046; SSE2-NEXT: pcmpgtd %xmm1, %xmm3
1047; SSE2-NEXT: pand %xmm3, %xmm2
1048; SSE2-NEXT: pandn %xmm0, %xmm3
1049; SSE2-NEXT: por %xmm2, %xmm3
1050; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm3[1,1,2,3]
1051; SSE2-NEXT: movdqa %xmm3, %xmm1
1052; SSE2-NEXT: pxor %xmm8, %xmm1
1053; SSE2-NEXT: pxor %xmm0, %xmm8
1054; SSE2-NEXT: pcmpgtd %xmm1, %xmm8
1055; SSE2-NEXT: pand %xmm8, %xmm3
1056; SSE2-NEXT: pandn %xmm0, %xmm8
1057; SSE2-NEXT: por %xmm3, %xmm8
1058; SSE2-NEXT: movd %xmm8, %eax
1059; SSE2-NEXT: retq
1060;
1061; SSE41-LABEL: test_v32i32:
1062; SSE41: # %bb.0:
1063; SSE41-NEXT: pminud %xmm6, %xmm2
1064; SSE41-NEXT: pminud %xmm4, %xmm0
1065; SSE41-NEXT: pminud %xmm2, %xmm0
1066; SSE41-NEXT: pminud %xmm7, %xmm3
1067; SSE41-NEXT: pminud %xmm5, %xmm1
1068; SSE41-NEXT: pminud %xmm3, %xmm1
1069; SSE41-NEXT: pminud %xmm0, %xmm1
1070; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
1071; SSE41-NEXT: pminud %xmm1, %xmm0
1072; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
1073; SSE41-NEXT: pminud %xmm0, %xmm1
1074; SSE41-NEXT: movd %xmm1, %eax
1075; SSE41-NEXT: retq
1076;
1077; AVX1-LABEL: test_v32i32:
1078; AVX1: # %bb.0:
1079; AVX1-NEXT: vpminud %xmm3, %xmm1, %xmm4
1080; AVX1-NEXT: vpminud %xmm2, %xmm0, %xmm5
1081; AVX1-NEXT: vpminud %xmm4, %xmm5, %xmm4
1082; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm3
1083; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm1
1084; AVX1-NEXT: vpminud %xmm3, %xmm1, %xmm1
1085; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm2
1086; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
1087; AVX1-NEXT: vpminud %xmm2, %xmm0, %xmm0
1088; AVX1-NEXT: vpminud %xmm1, %xmm0, %xmm0
1089; AVX1-NEXT: vpminud %xmm0, %xmm4, %xmm0
1090; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
1091; AVX1-NEXT: vpminud %xmm1, %xmm0, %xmm0
1092; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
1093; AVX1-NEXT: vpminud %xmm1, %xmm0, %xmm0
1094; AVX1-NEXT: vmovd %xmm0, %eax
1095; AVX1-NEXT: vzeroupper
1096; AVX1-NEXT: retq
1097;
1098; AVX2-LABEL: test_v32i32:
1099; AVX2: # %bb.0:
1100; AVX2-NEXT: vpminud %ymm3, %ymm1, %ymm1
1101; AVX2-NEXT: vpminud %ymm2, %ymm0, %ymm0
1102; AVX2-NEXT: vpminud %ymm1, %ymm0, %ymm0
1103; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
1104; AVX2-NEXT: vpminud %ymm1, %ymm0, %ymm0
1105; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
1106; AVX2-NEXT: vpminud %ymm1, %ymm0, %ymm0
1107; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
1108; AVX2-NEXT: vpminud %ymm1, %ymm0, %ymm0
1109; AVX2-NEXT: vmovd %xmm0, %eax
1110; AVX2-NEXT: vzeroupper
1111; AVX2-NEXT: retq
1112;
1113; AVX512-LABEL: test_v32i32:
1114; AVX512: # %bb.0:
1115; AVX512-NEXT: vpminud %zmm1, %zmm0, %zmm0
1116; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm1
1117; AVX512-NEXT: vpminud %zmm1, %zmm0, %zmm0
1118; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
1119; AVX512-NEXT: vpminud %zmm1, %zmm0, %zmm0
1120; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
1121; AVX512-NEXT: vpminud %zmm1, %zmm0, %zmm0
1122; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
1123; AVX512-NEXT: vpminud %zmm1, %zmm0, %zmm0
1124; AVX512-NEXT: vmovd %xmm0, %eax
1125; AVX512-NEXT: vzeroupper
1126; AVX512-NEXT: retq
1127 %1 = call i32 @llvm.experimental.vector.reduce.umin.i32.v32i32(<32 x i32> %a0)
1128 ret i32 %1
1129}
1130
1131;
1132; vXi16
1133;
1134
1135define i16 @test_v8i16(<8 x i16> %a0) {
1136; SSE2-LABEL: test_v8i16:
1137; SSE2: # %bb.0:
1138; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
1139; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
1140; SSE2-NEXT: pxor %xmm2, %xmm0
1141; SSE2-NEXT: pxor %xmm2, %xmm1
1142; SSE2-NEXT: pminsw %xmm0, %xmm1
1143; SSE2-NEXT: pxor %xmm2, %xmm1
1144; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
1145; SSE2-NEXT: pxor %xmm2, %xmm1
1146; SSE2-NEXT: pxor %xmm2, %xmm0
1147; SSE2-NEXT: pminsw %xmm1, %xmm0
1148; SSE2-NEXT: pxor %xmm2, %xmm0
1149; SSE2-NEXT: movdqa %xmm0, %xmm1
1150; SSE2-NEXT: psrld $16, %xmm1
1151; SSE2-NEXT: pxor %xmm2, %xmm0
1152; SSE2-NEXT: pxor %xmm2, %xmm1
1153; SSE2-NEXT: pminsw %xmm0, %xmm1
1154; SSE2-NEXT: pxor %xmm2, %xmm1
1155; SSE2-NEXT: movd %xmm1, %eax
1156; SSE2-NEXT: # kill: def $ax killed $ax killed $eax
1157; SSE2-NEXT: retq
1158;
1159; SSE41-LABEL: test_v8i16:
1160; SSE41: # %bb.0:
1161; SSE41-NEXT: phminposuw %xmm0, %xmm0
1162; SSE41-NEXT: movd %xmm0, %eax
1163; SSE41-NEXT: # kill: def $ax killed $ax killed $eax
1164; SSE41-NEXT: retq
1165;
1166; AVX-LABEL: test_v8i16:
1167; AVX: # %bb.0:
1168; AVX-NEXT: vphminposuw %xmm0, %xmm0
1169; AVX-NEXT: vmovd %xmm0, %eax
1170; AVX-NEXT: # kill: def $ax killed $ax killed $eax
1171; AVX-NEXT: retq
1172;
1173; AVX512-LABEL: test_v8i16:
1174; AVX512: # %bb.0:
1175; AVX512-NEXT: vphminposuw %xmm0, %xmm0
1176; AVX512-NEXT: vmovd %xmm0, %eax
1177; AVX512-NEXT: # kill: def $ax killed $ax killed $eax
1178; AVX512-NEXT: retq
1179 %1 = call i16 @llvm.experimental.vector.reduce.umin.i16.v8i16(<8 x i16> %a0)
1180 ret i16 %1
1181}
1182
1183define i16 @test_v16i16(<16 x i16> %a0) {
1184; SSE2-LABEL: test_v16i16:
1185; SSE2: # %bb.0:
1186; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
1187; SSE2-NEXT: pxor %xmm2, %xmm1
1188; SSE2-NEXT: pxor %xmm2, %xmm0
1189; SSE2-NEXT: pminsw %xmm1, %xmm0
1190; SSE2-NEXT: pxor %xmm2, %xmm0
1191; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
1192; SSE2-NEXT: pxor %xmm2, %xmm0
1193; SSE2-NEXT: pxor %xmm2, %xmm1
1194; SSE2-NEXT: pminsw %xmm0, %xmm1
1195; SSE2-NEXT: pxor %xmm2, %xmm1
1196; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
1197; SSE2-NEXT: pxor %xmm2, %xmm1
1198; SSE2-NEXT: pxor %xmm2, %xmm0
1199; SSE2-NEXT: pminsw %xmm1, %xmm0
1200; SSE2-NEXT: pxor %xmm2, %xmm0
1201; SSE2-NEXT: movdqa %xmm0, %xmm1
1202; SSE2-NEXT: psrld $16, %xmm1
1203; SSE2-NEXT: pxor %xmm2, %xmm0
1204; SSE2-NEXT: pxor %xmm2, %xmm1
1205; SSE2-NEXT: pminsw %xmm0, %xmm1
1206; SSE2-NEXT: pxor %xmm2, %xmm1
1207; SSE2-NEXT: movd %xmm1, %eax
1208; SSE2-NEXT: # kill: def $ax killed $ax killed $eax
1209; SSE2-NEXT: retq
1210;
1211; SSE41-LABEL: test_v16i16:
1212; SSE41: # %bb.0:
1213; SSE41-NEXT: pminuw %xmm1, %xmm0
1214; SSE41-NEXT: phminposuw %xmm0, %xmm0
1215; SSE41-NEXT: movd %xmm0, %eax
1216; SSE41-NEXT: # kill: def $ax killed $ax killed $eax
1217; SSE41-NEXT: retq
1218;
1219; AVX1-LABEL: test_v16i16:
1220; AVX1: # %bb.0:
1221; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
1222; AVX1-NEXT: vpminuw %xmm1, %xmm0, %xmm0
1223; AVX1-NEXT: vphminposuw %xmm0, %xmm0
1224; AVX1-NEXT: vmovd %xmm0, %eax
1225; AVX1-NEXT: # kill: def $ax killed $ax killed $eax
1226; AVX1-NEXT: vzeroupper
1227; AVX1-NEXT: retq
1228;
1229; AVX2-LABEL: test_v16i16:
1230; AVX2: # %bb.0:
1231; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
1232; AVX2-NEXT: vpminuw %xmm1, %xmm0, %xmm0
1233; AVX2-NEXT: vphminposuw %xmm0, %xmm0
1234; AVX2-NEXT: vmovd %xmm0, %eax
1235; AVX2-NEXT: # kill: def $ax killed $ax killed $eax
1236; AVX2-NEXT: vzeroupper
1237; AVX2-NEXT: retq
1238;
1239; AVX512-LABEL: test_v16i16:
1240; AVX512: # %bb.0:
1241; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
1242; AVX512-NEXT: vpminuw %xmm1, %xmm0, %xmm0
1243; AVX512-NEXT: vphminposuw %xmm0, %xmm0
1244; AVX512-NEXT: vmovd %xmm0, %eax
1245; AVX512-NEXT: # kill: def $ax killed $ax killed $eax
1246; AVX512-NEXT: vzeroupper
1247; AVX512-NEXT: retq
1248 %1 = call i16 @llvm.experimental.vector.reduce.umin.i16.v16i16(<16 x i16> %a0)
1249 ret i16 %1
1250}
1251
1252define i16 @test_v32i16(<32 x i16> %a0) {
1253; SSE2-LABEL: test_v32i16:
1254; SSE2: # %bb.0:
1255; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
1256; SSE2-NEXT: pxor %xmm4, %xmm2
1257; SSE2-NEXT: pxor %xmm4, %xmm0
1258; SSE2-NEXT: pminsw %xmm2, %xmm0
1259; SSE2-NEXT: pxor %xmm4, %xmm3
1260; SSE2-NEXT: pxor %xmm4, %xmm1
1261; SSE2-NEXT: pminsw %xmm3, %xmm1
1262; SSE2-NEXT: movdqa %xmm4, %xmm2
1263; SSE2-NEXT: pxor %xmm4, %xmm2
1264; SSE2-NEXT: pxor %xmm2, %xmm1
1265; SSE2-NEXT: pxor %xmm0, %xmm2
1266; SSE2-NEXT: pminsw %xmm1, %xmm2
1267; SSE2-NEXT: pxor %xmm4, %xmm2
1268; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
1269; SSE2-NEXT: pxor %xmm4, %xmm2
1270; SSE2-NEXT: pxor %xmm4, %xmm0
1271; SSE2-NEXT: pminsw %xmm2, %xmm0
1272; SSE2-NEXT: pxor %xmm4, %xmm0
1273; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
1274; SSE2-NEXT: pxor %xmm4, %xmm0
1275; SSE2-NEXT: pxor %xmm4, %xmm1
1276; SSE2-NEXT: pminsw %xmm0, %xmm1
1277; SSE2-NEXT: pxor %xmm4, %xmm1
1278; SSE2-NEXT: movdqa %xmm1, %xmm0
1279; SSE2-NEXT: psrld $16, %xmm0
1280; SSE2-NEXT: pxor %xmm4, %xmm1
1281; SSE2-NEXT: pxor %xmm4, %xmm0
1282; SSE2-NEXT: pminsw %xmm1, %xmm0
1283; SSE2-NEXT: pxor %xmm4, %xmm0
1284; SSE2-NEXT: movd %xmm0, %eax
1285; SSE2-NEXT: # kill: def $ax killed $ax killed $eax
1286; SSE2-NEXT: retq
1287;
1288; SSE41-LABEL: test_v32i16:
1289; SSE41: # %bb.0:
1290; SSE41-NEXT: pminuw %xmm3, %xmm1
1291; SSE41-NEXT: pminuw %xmm2, %xmm0
1292; SSE41-NEXT: pminuw %xmm1, %xmm0
1293; SSE41-NEXT: phminposuw %xmm0, %xmm0
1294; SSE41-NEXT: movd %xmm0, %eax
1295; SSE41-NEXT: # kill: def $ax killed $ax killed $eax
1296; SSE41-NEXT: retq
1297;
1298; AVX1-LABEL: test_v32i16:
1299; AVX1: # %bb.0:
1300; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
1301; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
1302; AVX1-NEXT: vpminuw %xmm2, %xmm3, %xmm2
1303; AVX1-NEXT: vpminuw %xmm1, %xmm0, %xmm0
1304; AVX1-NEXT: vpminuw %xmm2, %xmm0, %xmm0
1305; AVX1-NEXT: vphminposuw %xmm0, %xmm0
1306; AVX1-NEXT: vmovd %xmm0, %eax
1307; AVX1-NEXT: # kill: def $ax killed $ax killed $eax
1308; AVX1-NEXT: vzeroupper
1309; AVX1-NEXT: retq
1310;
1311; AVX2-LABEL: test_v32i16:
1312; AVX2: # %bb.0:
1313; AVX2-NEXT: vpminuw %ymm1, %ymm0, %ymm0
1314; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
1315; AVX2-NEXT: vpminuw %xmm1, %xmm0, %xmm0
1316; AVX2-NEXT: vphminposuw %xmm0, %xmm0
1317; AVX2-NEXT: vmovd %xmm0, %eax
1318; AVX2-NEXT: # kill: def $ax killed $ax killed $eax
1319; AVX2-NEXT: vzeroupper
1320; AVX2-NEXT: retq
1321;
1322; AVX512-LABEL: test_v32i16:
1323; AVX512: # %bb.0:
1324; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm1
1325; AVX512-NEXT: vpminuw %ymm1, %ymm0, %ymm0
1326; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
1327; AVX512-NEXT: vpminuw %xmm1, %xmm0, %xmm0
1328; AVX512-NEXT: vphminposuw %xmm0, %xmm0
1329; AVX512-NEXT: vmovd %xmm0, %eax
1330; AVX512-NEXT: # kill: def $ax killed $ax killed $eax
1331; AVX512-NEXT: vzeroupper
1332; AVX512-NEXT: retq
1333 %1 = call i16 @llvm.experimental.vector.reduce.umin.i16.v32i16(<32 x i16> %a0)
1334 ret i16 %1
1335}
1336
1337define i16 @test_v64i16(<64 x i16> %a0) {
1338; SSE2-LABEL: test_v64i16:
1339; SSE2: # %bb.0:
1340; SSE2-NEXT: movdqa {{.*#+}} xmm8 = [32768,32768,32768,32768,32768,32768,32768,32768]
1341; SSE2-NEXT: pxor %xmm8, %xmm5
1342; SSE2-NEXT: pxor %xmm8, %xmm1
1343; SSE2-NEXT: pminsw %xmm5, %xmm1
1344; SSE2-NEXT: pxor %xmm8, %xmm7
1345; SSE2-NEXT: pxor %xmm8, %xmm3
1346; SSE2-NEXT: pminsw %xmm7, %xmm3
1347; SSE2-NEXT: pxor %xmm8, %xmm4
1348; SSE2-NEXT: pxor %xmm8, %xmm0
1349; SSE2-NEXT: pminsw %xmm4, %xmm0
1350; SSE2-NEXT: pxor %xmm8, %xmm6
1351; SSE2-NEXT: pxor %xmm8, %xmm2
1352; SSE2-NEXT: pminsw %xmm6, %xmm2
1353; SSE2-NEXT: movdqa %xmm8, %xmm4
1354; SSE2-NEXT: pxor %xmm8, %xmm4
1355; SSE2-NEXT: pxor %xmm4, %xmm2
1356; SSE2-NEXT: pxor %xmm4, %xmm0
1357; SSE2-NEXT: pminsw %xmm2, %xmm0
1358; SSE2-NEXT: pxor %xmm4, %xmm3
1359; SSE2-NEXT: pxor %xmm4, %xmm1
1360; SSE2-NEXT: pminsw %xmm3, %xmm1
1361; SSE2-NEXT: pxor %xmm4, %xmm1
1362; SSE2-NEXT: pxor %xmm4, %xmm0
1363; SSE2-NEXT: pminsw %xmm1, %xmm0
1364; SSE2-NEXT: pxor %xmm8, %xmm0
1365; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
1366; SSE2-NEXT: pxor %xmm8, %xmm0
1367; SSE2-NEXT: pxor %xmm8, %xmm1
1368; SSE2-NEXT: pminsw %xmm0, %xmm1
1369; SSE2-NEXT: pxor %xmm8, %xmm1
1370; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
1371; SSE2-NEXT: pxor %xmm8, %xmm1
1372; SSE2-NEXT: pxor %xmm8, %xmm0
1373; SSE2-NEXT: pminsw %xmm1, %xmm0
1374; SSE2-NEXT: pxor %xmm8, %xmm0
1375; SSE2-NEXT: movdqa %xmm0, %xmm1
1376; SSE2-NEXT: psrld $16, %xmm1
1377; SSE2-NEXT: pxor %xmm8, %xmm0
1378; SSE2-NEXT: pxor %xmm8, %xmm1
1379; SSE2-NEXT: pminsw %xmm0, %xmm1
1380; SSE2-NEXT: pxor %xmm8, %xmm1
1381; SSE2-NEXT: movd %xmm1, %eax
1382; SSE2-NEXT: # kill: def $ax killed $ax killed $eax
1383; SSE2-NEXT: retq
1384;
1385; SSE41-LABEL: test_v64i16:
1386; SSE41: # %bb.0:
1387; SSE41-NEXT: pminuw %xmm7, %xmm3
1388; SSE41-NEXT: pminuw %xmm5, %xmm1
1389; SSE41-NEXT: pminuw %xmm3, %xmm1
1390; SSE41-NEXT: pminuw %xmm6, %xmm2
1391; SSE41-NEXT: pminuw %xmm4, %xmm0
1392; SSE41-NEXT: pminuw %xmm2, %xmm0
1393; SSE41-NEXT: pminuw %xmm1, %xmm0
1394; SSE41-NEXT: phminposuw %xmm0, %xmm0
1395; SSE41-NEXT: movd %xmm0, %eax
1396; SSE41-NEXT: # kill: def $ax killed $ax killed $eax
1397; SSE41-NEXT: retq
1398;
1399; AVX1-LABEL: test_v64i16:
1400; AVX1: # %bb.0:
1401; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm4
1402; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5
1403; AVX1-NEXT: vpminuw %xmm4, %xmm5, %xmm4
1404; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm5
1405; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm6
1406; AVX1-NEXT: vpminuw %xmm5, %xmm6, %xmm5
1407; AVX1-NEXT: vpminuw %xmm4, %xmm5, %xmm4
1408; AVX1-NEXT: vpminuw %xmm3, %xmm1, %xmm1
1409; AVX1-NEXT: vpminuw %xmm2, %xmm0, %xmm0
1410; AVX1-NEXT: vpminuw %xmm1, %xmm0, %xmm0
1411; AVX1-NEXT: vpminuw %xmm4, %xmm0, %xmm0
1412; AVX1-NEXT: vphminposuw %xmm0, %xmm0
1413; AVX1-NEXT: vmovd %xmm0, %eax
1414; AVX1-NEXT: # kill: def $ax killed $ax killed $eax
1415; AVX1-NEXT: vzeroupper
1416; AVX1-NEXT: retq
1417;
1418; AVX2-LABEL: test_v64i16:
1419; AVX2: # %bb.0:
1420; AVX2-NEXT: vpminuw %ymm3, %ymm1, %ymm1
1421; AVX2-NEXT: vpminuw %ymm2, %ymm0, %ymm0
1422; AVX2-NEXT: vpminuw %ymm1, %ymm0, %ymm0
1423; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
1424; AVX2-NEXT: vpminuw %xmm1, %xmm0, %xmm0
1425; AVX2-NEXT: vphminposuw %xmm0, %xmm0
1426; AVX2-NEXT: vmovd %xmm0, %eax
1427; AVX2-NEXT: # kill: def $ax killed $ax killed $eax
1428; AVX2-NEXT: vzeroupper
1429; AVX2-NEXT: retq
1430;
1431; AVX512-LABEL: test_v64i16:
1432; AVX512: # %bb.0:
1433; AVX512-NEXT: vpminuw %zmm1, %zmm0, %zmm0
1434; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm1
1435; AVX512-NEXT: vpminuw %ymm1, %ymm0, %ymm0
1436; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
1437; AVX512-NEXT: vpminuw %xmm1, %xmm0, %xmm0
1438; AVX512-NEXT: vphminposuw %xmm0, %xmm0
1439; AVX512-NEXT: vmovd %xmm0, %eax
1440; AVX512-NEXT: # kill: def $ax killed $ax killed $eax
1441; AVX512-NEXT: vzeroupper
1442; AVX512-NEXT: retq
1443 %1 = call i16 @llvm.experimental.vector.reduce.umin.i16.v64i16(<64 x i16> %a0)
1444 ret i16 %1
1445}
1446
1447;
1448; vXi8
1449;
1450
1451define i8 @test_v16i8(<16 x i8> %a0) {
1452; SSE2-LABEL: test_v16i8:
1453; SSE2: # %bb.0:
1454; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
1455; SSE2-NEXT: pminub %xmm0, %xmm1
1456; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
1457; SSE2-NEXT: pminub %xmm1, %xmm0
1458; SSE2-NEXT: movdqa %xmm0, %xmm1
1459; SSE2-NEXT: psrld $16, %xmm1
1460; SSE2-NEXT: pminub %xmm0, %xmm1
1461; SSE2-NEXT: movdqa %xmm1, %xmm0
1462; SSE2-NEXT: psrlw $8, %xmm0
1463; SSE2-NEXT: pminub %xmm1, %xmm0
1464; SSE2-NEXT: movd %xmm0, %eax
1465; SSE2-NEXT: # kill: def $al killed $al killed $eax
1466; SSE2-NEXT: retq
1467;
1468; SSE41-LABEL: test_v16i8:
1469; SSE41: # %bb.0:
1470; SSE41-NEXT: movdqa %xmm0, %xmm1
1471; SSE41-NEXT: psrlw $8, %xmm1
1472; SSE41-NEXT: pminub %xmm0, %xmm1
1473; SSE41-NEXT: phminposuw %xmm1, %xmm0
1474; SSE41-NEXT: pextrb $0, %xmm0, %eax
1475; SSE41-NEXT: # kill: def $al killed $al killed $eax
1476; SSE41-NEXT: retq
1477;
1478; AVX-LABEL: test_v16i8:
1479; AVX: # %bb.0:
1480; AVX-NEXT: vpsrlw $8, %xmm0, %xmm1
1481; AVX-NEXT: vpminub %xmm1, %xmm0, %xmm0
1482; AVX-NEXT: vphminposuw %xmm0, %xmm0
1483; AVX-NEXT: vpextrb $0, %xmm0, %eax
1484; AVX-NEXT: # kill: def $al killed $al killed $eax
1485; AVX-NEXT: retq
1486;
1487; AVX512-LABEL: test_v16i8:
1488; AVX512: # %bb.0:
1489; AVX512-NEXT: vpsrlw $8, %xmm0, %xmm1
1490; AVX512-NEXT: vpminub %xmm1, %xmm0, %xmm0
1491; AVX512-NEXT: vphminposuw %xmm0, %xmm0
1492; AVX512-NEXT: vpextrb $0, %xmm0, %eax
1493; AVX512-NEXT: # kill: def $al killed $al killed $eax
1494; AVX512-NEXT: retq
1495 %1 = call i8 @llvm.experimental.vector.reduce.umin.i8.v16i8(<16 x i8> %a0)
1496 ret i8 %1
1497}
1498
1499define i8 @test_v32i8(<32 x i8> %a0) {
1500; SSE2-LABEL: test_v32i8:
1501; SSE2: # %bb.0:
1502; SSE2-NEXT: pminub %xmm1, %xmm0
1503; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
1504; SSE2-NEXT: pminub %xmm0, %xmm1
1505; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
1506; SSE2-NEXT: pminub %xmm1, %xmm0
1507; SSE2-NEXT: movdqa %xmm0, %xmm1
1508; SSE2-NEXT: psrld $16, %xmm1
1509; SSE2-NEXT: pminub %xmm0, %xmm1
1510; SSE2-NEXT: movdqa %xmm1, %xmm0
1511; SSE2-NEXT: psrlw $8, %xmm0
1512; SSE2-NEXT: pminub %xmm1, %xmm0
1513; SSE2-NEXT: movd %xmm0, %eax
1514; SSE2-NEXT: # kill: def $al killed $al killed $eax
1515; SSE2-NEXT: retq
1516;
1517; SSE41-LABEL: test_v32i8:
1518; SSE41: # %bb.0:
1519; SSE41-NEXT: pminub %xmm1, %xmm0
1520; SSE41-NEXT: movdqa %xmm0, %xmm1
1521; SSE41-NEXT: psrlw $8, %xmm1
1522; SSE41-NEXT: pminub %xmm0, %xmm1
1523; SSE41-NEXT: phminposuw %xmm1, %xmm0
1524; SSE41-NEXT: pextrb $0, %xmm0, %eax
1525; SSE41-NEXT: # kill: def $al killed $al killed $eax
1526; SSE41-NEXT: retq
1527;
1528; AVX1-LABEL: test_v32i8:
1529; AVX1: # %bb.0:
1530; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
1531; AVX1-NEXT: vpminub %xmm1, %xmm0, %xmm0
1532; AVX1-NEXT: vpsrlw $8, %xmm0, %xmm1
1533; AVX1-NEXT: vpminub %xmm1, %xmm0, %xmm0
1534; AVX1-NEXT: vphminposuw %xmm0, %xmm0
1535; AVX1-NEXT: vpextrb $0, %xmm0, %eax
1536; AVX1-NEXT: # kill: def $al killed $al killed $eax
1537; AVX1-NEXT: vzeroupper
1538; AVX1-NEXT: retq
1539;
1540; AVX2-LABEL: test_v32i8:
1541; AVX2: # %bb.0:
1542; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
1543; AVX2-NEXT: vpminub %xmm1, %xmm0, %xmm0
1544; AVX2-NEXT: vpsrlw $8, %xmm0, %xmm1
1545; AVX2-NEXT: vpminub %xmm1, %xmm0, %xmm0
1546; AVX2-NEXT: vphminposuw %xmm0, %xmm0
1547; AVX2-NEXT: vpextrb $0, %xmm0, %eax
1548; AVX2-NEXT: # kill: def $al killed $al killed $eax
1549; AVX2-NEXT: vzeroupper
1550; AVX2-NEXT: retq
1551;
1552; AVX512-LABEL: test_v32i8:
1553; AVX512: # %bb.0:
1554; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
1555; AVX512-NEXT: vpminub %xmm1, %xmm0, %xmm0
1556; AVX512-NEXT: vpsrlw $8, %xmm0, %xmm1
1557; AVX512-NEXT: vpminub %xmm1, %xmm0, %xmm0
1558; AVX512-NEXT: vphminposuw %xmm0, %xmm0
1559; AVX512-NEXT: vpextrb $0, %xmm0, %eax
1560; AVX512-NEXT: # kill: def $al killed $al killed $eax
1561; AVX512-NEXT: vzeroupper
1562; AVX512-NEXT: retq
1563 %1 = call i8 @llvm.experimental.vector.reduce.umin.i8.v32i8(<32 x i8> %a0)
1564 ret i8 %1
1565}
1566
1567define i8 @test_v64i8(<64 x i8> %a0) {
1568; SSE2-LABEL: test_v64i8:
1569; SSE2: # %bb.0:
1570; SSE2-NEXT: pminub %xmm3, %xmm1
1571; SSE2-NEXT: pminub %xmm2, %xmm0
1572; SSE2-NEXT: pminub %xmm1, %xmm0
1573; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
1574; SSE2-NEXT: pminub %xmm0, %xmm1
1575; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
1576; SSE2-NEXT: pminub %xmm1, %xmm0
1577; SSE2-NEXT: movdqa %xmm0, %xmm1
1578; SSE2-NEXT: psrld $16, %xmm1
1579; SSE2-NEXT: pminub %xmm0, %xmm1
1580; SSE2-NEXT: movdqa %xmm1, %xmm0
1581; SSE2-NEXT: psrlw $8, %xmm0
1582; SSE2-NEXT: pminub %xmm1, %xmm0
1583; SSE2-NEXT: movd %xmm0, %eax
1584; SSE2-NEXT: # kill: def $al killed $al killed $eax
1585; SSE2-NEXT: retq
1586;
1587; SSE41-LABEL: test_v64i8:
1588; SSE41: # %bb.0:
1589; SSE41-NEXT: pminub %xmm3, %xmm1
1590; SSE41-NEXT: pminub %xmm2, %xmm0
1591; SSE41-NEXT: pminub %xmm1, %xmm0
1592; SSE41-NEXT: movdqa %xmm0, %xmm1
1593; SSE41-NEXT: psrlw $8, %xmm1
1594; SSE41-NEXT: pminub %xmm0, %xmm1
1595; SSE41-NEXT: phminposuw %xmm1, %xmm0
1596; SSE41-NEXT: pextrb $0, %xmm0, %eax
1597; SSE41-NEXT: # kill: def $al killed $al killed $eax
1598; SSE41-NEXT: retq
1599;
1600; AVX1-LABEL: test_v64i8:
1601; AVX1: # %bb.0:
1602; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
1603; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
1604; AVX1-NEXT: vpminub %xmm2, %xmm3, %xmm2
1605; AVX1-NEXT: vpminub %xmm1, %xmm0, %xmm0
1606; AVX1-NEXT: vpminub %xmm2, %xmm0, %xmm0
1607; AVX1-NEXT: vpsrlw $8, %xmm0, %xmm1
1608; AVX1-NEXT: vpminub %xmm1, %xmm0, %xmm0
1609; AVX1-NEXT: vphminposuw %xmm0, %xmm0
1610; AVX1-NEXT: vpextrb $0, %xmm0, %eax
1611; AVX1-NEXT: # kill: def $al killed $al killed $eax
1612; AVX1-NEXT: vzeroupper
1613; AVX1-NEXT: retq
1614;
1615; AVX2-LABEL: test_v64i8:
1616; AVX2: # %bb.0:
1617; AVX2-NEXT: vpminub %ymm1, %ymm0, %ymm0
1618; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
1619; AVX2-NEXT: vpminub %xmm1, %xmm0, %xmm0
1620; AVX2-NEXT: vpsrlw $8, %xmm0, %xmm1
1621; AVX2-NEXT: vpminub %xmm1, %xmm0, %xmm0
1622; AVX2-NEXT: vphminposuw %xmm0, %xmm0
1623; AVX2-NEXT: vpextrb $0, %xmm0, %eax
1624; AVX2-NEXT: # kill: def $al killed $al killed $eax
1625; AVX2-NEXT: vzeroupper
1626; AVX2-NEXT: retq
1627;
1628; AVX512-LABEL: test_v64i8:
1629; AVX512: # %bb.0:
1630; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm1
1631; AVX512-NEXT: vpminub %ymm1, %ymm0, %ymm0
1632; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
1633; AVX512-NEXT: vpminub %xmm1, %xmm0, %xmm0
1634; AVX512-NEXT: vpsrlw $8, %xmm0, %xmm1
1635; AVX512-NEXT: vpminub %xmm1, %xmm0, %xmm0
1636; AVX512-NEXT: vphminposuw %xmm0, %xmm0
1637; AVX512-NEXT: vpextrb $0, %xmm0, %eax
1638; AVX512-NEXT: # kill: def $al killed $al killed $eax
1639; AVX512-NEXT: vzeroupper
1640; AVX512-NEXT: retq
1641 %1 = call i8 @llvm.experimental.vector.reduce.umin.i8.v64i8(<64 x i8> %a0)
1642 ret i8 %1
1643}
1644
1645define i8 @test_v128i8(<128 x i8> %a0) {
1646; SSE2-LABEL: test_v128i8:
1647; SSE2: # %bb.0:
1648; SSE2-NEXT: pminub %xmm6, %xmm2
1649; SSE2-NEXT: pminub %xmm4, %xmm0
1650; SSE2-NEXT: pminub %xmm2, %xmm0
1651; SSE2-NEXT: pminub %xmm7, %xmm3
1652; SSE2-NEXT: pminub %xmm5, %xmm1
1653; SSE2-NEXT: pminub %xmm3, %xmm1
1654; SSE2-NEXT: pminub %xmm0, %xmm1
1655; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
1656; SSE2-NEXT: pminub %xmm1, %xmm0
1657; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
1658; SSE2-NEXT: pminub %xmm0, %xmm1
1659; SSE2-NEXT: movdqa %xmm1, %xmm0
1660; SSE2-NEXT: psrld $16, %xmm0
1661; SSE2-NEXT: pminub %xmm1, %xmm0
1662; SSE2-NEXT: movdqa %xmm0, %xmm1
1663; SSE2-NEXT: psrlw $8, %xmm1
1664; SSE2-NEXT: pminub %xmm0, %xmm1
1665; SSE2-NEXT: movd %xmm1, %eax
1666; SSE2-NEXT: # kill: def $al killed $al killed $eax
1667; SSE2-NEXT: retq
1668;
1669; SSE41-LABEL: test_v128i8:
1670; SSE41: # %bb.0:
1671; SSE41-NEXT: pminub %xmm7, %xmm3
1672; SSE41-NEXT: pminub %xmm5, %xmm1
1673; SSE41-NEXT: pminub %xmm3, %xmm1
1674; SSE41-NEXT: pminub %xmm6, %xmm2
1675; SSE41-NEXT: pminub %xmm4, %xmm0
1676; SSE41-NEXT: pminub %xmm2, %xmm0
1677; SSE41-NEXT: pminub %xmm1, %xmm0
1678; SSE41-NEXT: movdqa %xmm0, %xmm1
1679; SSE41-NEXT: psrlw $8, %xmm1
1680; SSE41-NEXT: pminub %xmm0, %xmm1
1681; SSE41-NEXT: phminposuw %xmm1, %xmm0
1682; SSE41-NEXT: pextrb $0, %xmm0, %eax
1683; SSE41-NEXT: # kill: def $al killed $al killed $eax
1684; SSE41-NEXT: retq
1685;
1686; AVX1-LABEL: test_v128i8:
1687; AVX1: # %bb.0:
1688; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm4
1689; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5
1690; AVX1-NEXT: vpminub %xmm4, %xmm5, %xmm4
1691; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm5
1692; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm6
1693; AVX1-NEXT: vpminub %xmm5, %xmm6, %xmm5
1694; AVX1-NEXT: vpminub %xmm4, %xmm5, %xmm4
1695; AVX1-NEXT: vpminub %xmm3, %xmm1, %xmm1
1696; AVX1-NEXT: vpminub %xmm2, %xmm0, %xmm0
1697; AVX1-NEXT: vpminub %xmm1, %xmm0, %xmm0
1698; AVX1-NEXT: vpminub %xmm4, %xmm0, %xmm0
1699; AVX1-NEXT: vpsrlw $8, %xmm0, %xmm1
1700; AVX1-NEXT: vpminub %xmm1, %xmm0, %xmm0
1701; AVX1-NEXT: vphminposuw %xmm0, %xmm0
1702; AVX1-NEXT: vpextrb $0, %xmm0, %eax
1703; AVX1-NEXT: # kill: def $al killed $al killed $eax
1704; AVX1-NEXT: vzeroupper
1705; AVX1-NEXT: retq
1706;
1707; AVX2-LABEL: test_v128i8:
1708; AVX2: # %bb.0:
1709; AVX2-NEXT: vpminub %ymm3, %ymm1, %ymm1
1710; AVX2-NEXT: vpminub %ymm2, %ymm0, %ymm0
1711; AVX2-NEXT: vpminub %ymm1, %ymm0, %ymm0
1712; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
1713; AVX2-NEXT: vpminub %xmm1, %xmm0, %xmm0
1714; AVX2-NEXT: vpsrlw $8, %xmm0, %xmm1
1715; AVX2-NEXT: vpminub %xmm1, %xmm0, %xmm0
1716; AVX2-NEXT: vphminposuw %xmm0, %xmm0
1717; AVX2-NEXT: vpextrb $0, %xmm0, %eax
1718; AVX2-NEXT: # kill: def $al killed $al killed $eax
1719; AVX2-NEXT: vzeroupper
1720; AVX2-NEXT: retq
1721;
1722; AVX512-LABEL: test_v128i8:
1723; AVX512: # %bb.0:
1724; AVX512-NEXT: vpminub %zmm1, %zmm0, %zmm0
1725; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm1
1726; AVX512-NEXT: vpminub %ymm1, %ymm0, %ymm0
1727; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
1728; AVX512-NEXT: vpminub %xmm1, %xmm0, %xmm0
1729; AVX512-NEXT: vpsrlw $8, %xmm0, %xmm1
1730; AVX512-NEXT: vpminub %xmm1, %xmm0, %xmm0
1731; AVX512-NEXT: vphminposuw %xmm0, %xmm0
1732; AVX512-NEXT: vpextrb $0, %xmm0, %eax
1733; AVX512-NEXT: # kill: def $al killed $al killed $eax
1734; AVX512-NEXT: vzeroupper
1735; AVX512-NEXT: retq
1736 %1 = call i8 @llvm.experimental.vector.reduce.umin.i8.v128i8(<128 x i8> %a0)
1737 ret i8 %1
1738}
1739
1740declare i64 @llvm.experimental.vector.reduce.umin.i64.v2i64(<2 x i64>)
1741declare i64 @llvm.experimental.vector.reduce.umin.i64.v4i64(<4 x i64>)
1742declare i64 @llvm.experimental.vector.reduce.umin.i64.v8i64(<8 x i64>)
1743declare i64 @llvm.experimental.vector.reduce.umin.i64.v16i64(<16 x i64>)
1744
1745declare i32 @llvm.experimental.vector.reduce.umin.i32.v4i32(<4 x i32>)
1746declare i32 @llvm.experimental.vector.reduce.umin.i32.v8i32(<8 x i32>)
1747declare i32 @llvm.experimental.vector.reduce.umin.i32.v16i32(<16 x i32>)
1748declare i32 @llvm.experimental.vector.reduce.umin.i32.v32i32(<32 x i32>)
1749
1750declare i16 @llvm.experimental.vector.reduce.umin.i16.v8i16(<8 x i16>)
1751declare i16 @llvm.experimental.vector.reduce.umin.i16.v16i16(<16 x i16>)
1752declare i16 @llvm.experimental.vector.reduce.umin.i16.v32i16(<32 x i16>)
1753declare i16 @llvm.experimental.vector.reduce.umin.i16.v64i16(<64 x i16>)
1754
1755declare i8 @llvm.experimental.vector.reduce.umin.i8.v16i8(<16 x i8>)
1756declare i8 @llvm.experimental.vector.reduce.umin.i8.v32i8(<32 x i8>)
1757declare i8 @llvm.experimental.vector.reduce.umin.i8.v64i8(<64 x i8>)
1758declare i8 @llvm.experimental.vector.reduce.umin.i8.v128i8(<128 x i8>)