1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -passes='vector-combine' -S %s | FileCheck %s
3
4target triple = "aarch64-none-eabi"
5
6define <16 x i32> @test1(<16 x i32> %x, <16 x i32> %y) {
7; CHECK-LABEL: @test1(
8; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
9; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 4, i32 19, i32 8, i32 7, i32 18, i32 11, i32 10, i32 17, i32 14, i32 13, i32 16>
10; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
11; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
12; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 21, i32 22, i32 7, i32 24, i32 25, i32 10, i32 27, i32 28, i32 13, i32 30, i32 31>
13; CHECK-NEXT:    ret <16 x i32> [[S3]]
14;
15  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
16  %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 4, i32 19, i32 8, i32 7, i32 18, i32 11, i32 10, i32 17, i32 14, i32 13, i32 16>
17  %a = add nsw <16 x i32> %s1, %s2
18  %b = sub nsw <16 x i32> %s1, %s2
19  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 21, i32 22, i32 7, i32 24, i32 25, i32 10, i32 27, i32 28, i32 13, i32 30, i32 31>
20  ret <16 x i32> %s3
21}
22
23define i32 @test1_reduce(<16 x i32> %x, <16 x i32> %y) {
24; CHECK-LABEL: @test1_reduce(
25; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 6, i32 9, i32 12, i32 15, i32 21, i32 24, i32 27, i32 30, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
26; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 23, i32 26, i32 29, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
27; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 19, i32 18, i32 17, i32 16, i32 4, i32 7, i32 10, i32 13, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
28; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 8, i32 11, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
29; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
30; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
31; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
32; CHECK-NEXT:    [[R:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[S3]])
33; CHECK-NEXT:    ret i32 [[R]]
34;
35  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
36  %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 4, i32 19, i32 8, i32 7, i32 18, i32 11, i32 10, i32 17, i32 14, i32 13, i32 16>
37  %a = add nsw <16 x i32> %s1, %s2
38  %b = sub nsw <16 x i32> %s1, %s2
39  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 21, i32 22, i32 7, i32 24, i32 25, i32 10, i32 27, i32 28, i32 13, i32 30, i32 31>
40  %r = call i32 @llvm.vector.reduce.add(<16 x i32> %s3)
41  ret i32 %r
42}
43declare i32 @llvm.vector.reduce.add(<16 x i32>)
44
45define <16 x i32> @test1more(<16 x i32> %a1, <16 x i32> %a2) {
46; CHECK-LABEL: @test1more(
47; CHECK-NEXT:    [[X:%.*]] = add nsw <16 x i32> [[A1:%.*]], [[A2:%.*]]
48; CHECK-NEXT:    [[Y:%.*]] = sub nsw <16 x i32> [[A1]], [[A2]]
49; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
50; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 4, i32 19, i32 8, i32 7, i32 18, i32 11, i32 10, i32 17, i32 14, i32 13, i32 16>
51; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
52; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
53; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 21, i32 22, i32 7, i32 24, i32 25, i32 10, i32 27, i32 28, i32 13, i32 30, i32 31>
54; CHECK-NEXT:    ret <16 x i32> [[S3]]
55;
56  %x = add nsw <16 x i32> %a1, %a2
57  %y = sub nsw <16 x i32> %a1, %a2
58  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
59  %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 4, i32 19, i32 8, i32 7, i32 18, i32 11, i32 10, i32 17, i32 14, i32 13, i32 16>
60  %a = add nsw <16 x i32> %s1, %s2
61  %b = sub nsw <16 x i32> %s1, %s2
62  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 21, i32 22, i32 7, i32 24, i32 25, i32 10, i32 27, i32 28, i32 13, i32 30, i32 31>
63  ret <16 x i32> %s3
64}
65
66define <16 x i32> @test1mores(<16 x i32> %a1, <16 x i32> %a2) {
67; CHECK-LABEL: @test1mores(
68; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A1:%.*]], <16 x i32> [[A2:%.*]], <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
69; CHECK-NEXT:    [[S4:%.*]] = shufflevector <16 x i32> [[A2]], <16 x i32> [[A1]], <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
70; CHECK-NEXT:    [[X:%.*]] = add nsw <16 x i32> [[S3]], [[S4]]
71; CHECK-NEXT:    [[Y:%.*]] = sub nsw <16 x i32> [[S3]], [[S4]]
72; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
73; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 4, i32 19, i32 8, i32 7, i32 18, i32 11, i32 10, i32 17, i32 14, i32 13, i32 16>
74; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
75; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
76; CHECK-NEXT:    [[S5:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 21, i32 22, i32 7, i32 24, i32 25, i32 10, i32 27, i32 28, i32 13, i32 30, i32 31>
77; CHECK-NEXT:    ret <16 x i32> [[S5]]
78;
79  %s3 = shufflevector <16 x i32> %a1, <16 x i32> %a2, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
80  %s4 = shufflevector <16 x i32> %a2, <16 x i32> %a1, <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
81  %x = add nsw <16 x i32> %s3, %s4
82  %y = sub nsw <16 x i32> %s3, %s4
83  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
84  %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 4, i32 19, i32 8, i32 7, i32 18, i32 11, i32 10, i32 17, i32 14, i32 13, i32 16>
85  %a = add nsw <16 x i32> %s1, %s2
86  %b = sub nsw <16 x i32> %s1, %s2
87  %s5 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 21, i32 22, i32 7, i32 24, i32 25, i32 10, i32 27, i32 28, i32 13, i32 30, i32 31>
88  ret <16 x i32> %s5
89}
90
91
92
93define <16 x i32> @test2_1(<16 x i32> %x, <16 x i32> %y) {
94; CHECK-LABEL: @test2_1(
95; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
96; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
97; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
98; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
99; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
100; CHECK-NEXT:    ret <16 x i32> [[S3]]
101;
102  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
103  %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
104  %a = add nsw <16 x i32> %s1, %s2
105  %b = sub nsw <16 x i32> %s1, %s2
106  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
107  ret <16 x i32> %s3
108}
109
110define <16 x i32> @test2_1_ins(<16 x i32> %x1, <16 x i32> %x2) {
111; CHECK-LABEL: @test2_1_ins(
112; CHECK-NEXT:    [[X:%.*]] = add nsw <16 x i32> [[X1:%.*]], [[X2:%.*]]
113; CHECK-NEXT:    [[Y:%.*]] = sub nsw <16 x i32> [[X1]], [[X2]]
114; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
115; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
116; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
117; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
118; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
119; CHECK-NEXT:    ret <16 x i32> [[S3]]
120;
121  %x = add nsw <16 x i32> %x1, %x2
122  %y = sub nsw <16 x i32> %x1, %x2
123  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
124  %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
125  %a = add nsw <16 x i32> %s1, %s2
126  %b = sub nsw <16 x i32> %s1, %s2
127  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
128  ret <16 x i32> %s3
129}
130
131define <16 x i32> @test2_2(<16 x i32> %x, <16 x i32> %y) {
132; CHECK-LABEL: @test2_2(
133; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 1, i32 12, i32 15, i32 18, i32 19, i32 22, i32 25, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
134; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 10, i32 11, i32 13, i32 14, i32 20, i32 21, i32 23, i32 24, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
135; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 29, i32 26, i32 27, i32 30, i32 7, i32 4, i32 5, i32 8, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
136; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 17, i32 28, i32 16, i32 31, i32 3, i32 6, i32 2, i32 9, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
137; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
138; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
139; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 19, i32 18, i32 23, i32 22, i32 5, i32 4, i32 21, i32 7, i32 6, i32 20, i32 1, i32 0, i32 17, i32 3, i32 2, i32 16>
140; CHECK-NEXT:    ret <16 x i32> [[S3]]
141;
142  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
143  %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
144  %a = add nsw <16 x i32> %s1, %s2
145  %b = sub nsw <16 x i32> %s1, %s2
146  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 4, i32 19, i32 8, i32 7, i32 18, i32 11, i32 10, i32 17, i32 14, i32 13, i32 16>
147  ret <16 x i32> %s3
148}
149
150define <16 x i32> @test2_12(<16 x i32> %x, <16 x i32> %y) {
151; CHECK-LABEL: @test2_12(
152; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
153; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
154; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
155; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
156; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
157; CHECK-NEXT:    [[S4:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 4, i32 19, i32 8, i32 7, i32 18, i32 11, i32 10, i32 17, i32 14, i32 13, i32 16>
158; CHECK-NEXT:    [[C:%.*]] = mul <16 x i32> [[S3]], [[S4]]
159; CHECK-NEXT:    ret <16 x i32> [[C]]
160;
161  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
162  %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
163  %a = add nsw <16 x i32> %s1, %s2
164  %b = sub nsw <16 x i32> %s1, %s2
165  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
166  %s4 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 4, i32 19, i32 8, i32 7, i32 18, i32 11, i32 10, i32 17, i32 14, i32 13, i32 16>
167  %c = mul <16 x i32> %s3, %s4
168  ret <16 x i32> %c
169}
170
171define <16 x i32> @test3_1(<16 x i32> %x, <16 x i32> %y) {
172; CHECK-LABEL: @test3_1(
173; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 23, i32 27, i32 31, i32 6, i32 10, i32 14, i32 21, i32 25, i32 29, i32 4, i32 8, i32 12>
174; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 18, i32 3, i32 16, i32 1, i32 5, i32 9, i32 13, i32 20, i32 24, i32 28, i32 7, i32 11, i32 15, i32 22, i32 26, i32 30>
175; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
176; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
177; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
178; CHECK-NEXT:    ret <16 x i32> [[S3]]
179;
180  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 23, i32 27, i32 31, i32 6, i32 10, i32 14, i32 21, i32 25, i32 29, i32 4, i32 8, i32 12>
181  %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 18, i32 3, i32 16, i32 1, i32 5, i32 9, i32 13, i32 20, i32 24, i32 28, i32 7, i32 11, i32 15, i32 22, i32 26, i32 30>
182  %a = add nsw <16 x i32> %s1, %s2
183  %b = sub nsw <16 x i32> %s1, %s2
184  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
185  ret <16 x i32> %s3
186}
187
188define <16 x i32> @test3_2(<16 x i32> %x, <16 x i32> %y) {
189; CHECK-LABEL: @test3_2(
190; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 17, i32 21, i32 25, i32 29, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
191; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 19, i32 23, i32 27, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
192; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 18, i32 22, i32 26, i32 30, i32 3, i32 7, i32 11, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
193; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 16, i32 20, i32 24, i32 28, i32 1, i32 5, i32 9, i32 13, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
194; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
195; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
196; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 17, i32 21, i32 1, i32 5, i32 4, i32 7, i32 6, i32 0, i32 3, i32 2, i32 20, i32 23, i32 22, i32 16, i32 19, i32 18>
197; CHECK-NEXT:    ret <16 x i32> [[S3]]
198;
199  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 23, i32 27, i32 31, i32 6, i32 10, i32 14, i32 21, i32 25, i32 29, i32 4, i32 8, i32 12>
200  %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 18, i32 3, i32 16, i32 1, i32 5, i32 9, i32 13, i32 20, i32 24, i32 28, i32 7, i32 11, i32 15, i32 22, i32 26, i32 30>
201  %a = add nsw <16 x i32> %s1, %s2
202  %b = sub nsw <16 x i32> %s1, %s2
203  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
204  ret <16 x i32> %s3
205}
206
207define <16 x i32> @test3_12(<16 x i32> %x, <16 x i32> %y) {
208; CHECK-LABEL: @test3_12(
209; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 23, i32 27, i32 31, i32 6, i32 10, i32 14, i32 21, i32 25, i32 29, i32 4, i32 8, i32 12>
210; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 18, i32 3, i32 16, i32 1, i32 5, i32 9, i32 13, i32 20, i32 24, i32 28, i32 7, i32 11, i32 15, i32 22, i32 26, i32 30>
211; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
212; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
213; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
214; CHECK-NEXT:    [[S4:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
215; CHECK-NEXT:    [[C:%.*]] = add <16 x i32> [[S3]], [[S4]]
216; CHECK-NEXT:    ret <16 x i32> [[C]]
217;
218  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 23, i32 27, i32 31, i32 6, i32 10, i32 14, i32 21, i32 25, i32 29, i32 4, i32 8, i32 12>
219  %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 18, i32 3, i32 16, i32 1, i32 5, i32 9, i32 13, i32 20, i32 24, i32 28, i32 7, i32 11, i32 15, i32 22, i32 26, i32 30>
220  %a = add nsw <16 x i32> %s1, %s2
221  %b = sub nsw <16 x i32> %s1, %s2
222  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
223  %s4 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
224  %c = add <16 x i32> %s3, %s4
225  ret <16 x i32> %c
226}
227
228
229define <16 x i32> @test23(<16 x i32> %x, <16 x i32> %y) {
230; CHECK-LABEL: @test23(
231; CHECK-NEXT:    [[S10:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
232; CHECK-NEXT:    [[S20:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
233; CHECK-NEXT:    [[A0:%.*]] = add nsw <16 x i32> [[S10]], [[S20]]
234; CHECK-NEXT:    [[B0:%.*]] = sub nsw <16 x i32> [[S10]], [[S20]]
235; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[A0]], <16 x i32> [[B0]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 23, i32 27, i32 31, i32 6, i32 10, i32 14, i32 21, i32 25, i32 29, i32 4, i32 8, i32 12>
236; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[B0]], <16 x i32> [[A0]], <16 x i32> <i32 18, i32 3, i32 16, i32 1, i32 5, i32 9, i32 13, i32 20, i32 24, i32 28, i32 7, i32 11, i32 15, i32 22, i32 26, i32 30>
237; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
238; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
239; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
240; CHECK-NEXT:    [[S4:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
241; CHECK-NEXT:    [[C:%.*]] = add <16 x i32> [[S3]], [[S4]]
242; CHECK-NEXT:    ret <16 x i32> [[C]]
243;
244  %s10 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
245  %s20 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
246  %a0 = add nsw <16 x i32> %s10, %s20
247  %b0 = sub nsw <16 x i32> %s10, %s20
248  %s1 = shufflevector <16 x i32> %a0, <16 x i32> %b0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 23, i32 27, i32 31, i32 6, i32 10, i32 14, i32 21, i32 25, i32 29, i32 4, i32 8, i32 12>
249  %s2 = shufflevector <16 x i32> %b0, <16 x i32> %a0, <16 x i32> <i32 18, i32 3, i32 16, i32 1, i32 5, i32 9, i32 13, i32 20, i32 24, i32 28, i32 7, i32 11, i32 15, i32 22, i32 26, i32 30>
250  %a = add nsw <16 x i32> %s1, %s2
251  %b = sub nsw <16 x i32> %s1, %s2
252  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
253  %s4 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
254  %c = add <16 x i32> %s3, %s4
255  ret <16 x i32> %c
256}
257
258define <16 x i32> @testgood(<16 x i32> %x, <16 x i32> %y) {
259; CHECK-LABEL: @testgood(
260; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
261; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
262; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
263; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
264; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
265; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
266; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23>
267; CHECK-NEXT:    ret <16 x i32> [[S3]]
268;
269  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
270  %s2 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
271  %a = add nsw <16 x i32> %s1, %s2
272  %b = sub nsw <16 x i32> %s1, %s2
273  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
274  ret <16 x i32> %s3
275}
276
277define <16 x i32> @test_shufshufin(<16 x i32> %x, <16 x i32> %y) {
278; CHECK-LABEL: @test_shufshufin(
279; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
280; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
281; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
282; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
283; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
284; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
285; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23>
286; CHECK-NEXT:    ret <16 x i32> [[S3]]
287;
288  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
289  %s2 = shufflevector <16 x i32> %s1, <16 x i32> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
290  %a = add nsw <16 x i32> %s1, %s2
291  %b = sub nsw <16 x i32> %s1, %s2
292  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
293  ret <16 x i32> %s3
294}
295
296define <16 x i32> @testshufshufout(<16 x i32> %x, <16 x i32> %y) {
297; CHECK-LABEL: @testshufshufout(
298; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
299; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
300; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
301; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
302; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
303; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
304; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23>
305; CHECK-NEXT:    [[S4:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 16, i32 0, i32 17, i32 1, i32 18, i32 2, i32 19, i32 3, i32 20, i32 4, i32 21, i32 5, i32 22, i32 6, i32 23, i32 7>
306; CHECK-NEXT:    [[R:%.*]] = add nsw <16 x i32> [[S3]], [[S4]]
307; CHECK-NEXT:    ret <16 x i32> [[R]]
308;
309  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
310  %s2 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
311  %a = add nsw <16 x i32> %s1, %s2
312  %b = sub nsw <16 x i32> %s1, %s2
313  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
314  %s4 = shufflevector <16 x i32> %s3, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
315  %r = add nsw <16 x i32> %s3, %s4
316  ret <16 x i32> %r
317}
318
319define <16 x i32> @testtwoshufout(<16 x i32> %x, <16 x i32> %y) {
320; CHECK-LABEL: @testtwoshufout(
321; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
322; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
323; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
324; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
325; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
326; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
327; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23>
328; CHECK-NEXT:    [[S4:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23>
329; CHECK-NEXT:    [[ADD:%.*]] = add <16 x i32> [[S3]], [[S4]]
330; CHECK-NEXT:    ret <16 x i32> [[ADD]]
331;
332  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
333  %s2 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
334  %a = add nsw <16 x i32> %s1, %s2
335  %b = sub nsw <16 x i32> %s1, %s2
336  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
337  %s4 = shufflevector <16 x i32> %b, <16 x i32> %a, <16 x i32> <i32 16, i32 1, i32 18, i32 3, i32 20, i32 5, i32 22, i32 7, i32 24, i32 9, i32 26, i32 11, i32 28, i32 13, i32 30, i32 15>
338  %add = add <16 x i32> %s3, %s4
339  ret <16 x i32> %add
340}
341
342declare void @use(<16 x i32>)
343define <16 x i32> @test_extrashuffleuse(<16 x i32> %x, <16 x i32> %y) {
344; CHECK-LABEL: @test_extrashuffleuse(
345; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
346; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
347; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
348; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
349; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
350; CHECK-NEXT:    call void @use(<16 x i32> [[S1]])
351; CHECK-NEXT:    ret <16 x i32> [[S3]]
352;
353  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
354  %s2 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
355  %a = add nsw <16 x i32> %s1, %s2
356  %b = sub nsw <16 x i32> %s1, %s2
357  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
358  call void @use(<16 x i32> %s1)
359  ret <16 x i32> %s3
360}
361
362define <16 x i32> @test_extraopuse(<16 x i32> %x, <16 x i32> %y) {
363; CHECK-LABEL: @test_extraopuse(
364; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
365; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
366; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
367; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
368; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
369; CHECK-NEXT:    call void @use(<16 x i32> [[B]])
370; CHECK-NEXT:    ret <16 x i32> [[S3]]
371;
372  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
373  %s2 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
374  %a = add nsw <16 x i32> %s1, %s2
375  %b = sub nsw <16 x i32> %s1, %s2
376  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
377  call void @use(<16 x i32> %b)
378  ret <16 x i32> %s3
379}
380
381define <4 x i32> @test_recurse(<4 x i32> %l0, <4 x i32> %l1, <4 x i32> %l3) {
382; CHECK-LABEL: @test_recurse(
383; CHECK-NEXT:    [[S0:%.*]] = shufflevector <4 x i32> [[L3:%.*]], <4 x i32> [[L1:%.*]], <4 x i32> <i32 0, i32 2, i32 1, i32 5>
384; CHECK-NEXT:    [[S1:%.*]] = shufflevector <4 x i32> [[L1]], <4 x i32> [[L3]], <4 x i32> <i32 6, i32 1, i32 3, i32 0>
385; CHECK-NEXT:    [[S2:%.*]] = shufflevector <4 x i32> [[L0:%.*]], <4 x i32> [[L1]], <4 x i32> <i32 4, i32 2, i32 6, i32 2>
386; CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[S2]], [[S2]]
387; CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[S1]], [[S0]]
388; CHECK-NEXT:    [[T0:%.*]] = shufflevector <4 x i32> [[ADD]], <4 x i32> [[ADD]], <4 x i32> <i32 2, i32 0, i32 0, i32 3>
389; CHECK-NEXT:    ret <4 x i32> [[T0]]
390;
391  %s0 = shufflevector <4 x i32> %l3, <4 x i32> %l1, <4 x i32> <i32 0, i32 2, i32 1, i32 5>
392  %s1 = shufflevector <4 x i32> %l1, <4 x i32> %l3, <4 x i32> <i32 6, i32 1, i32 3, i32 0>
393  %s2 = shufflevector <4 x i32> %l0, <4 x i32> %l1, <4 x i32> <i32 4, i32 2, i32 6, i32 2>
394  %add = add <4 x i32> %s2, %s2
395  %sub = sub <4 x i32> %s1, %s0
396  %t0 = shufflevector <4 x i32> %add, <4 x i32> %add, <4 x i32> <i32 2, i32 0, i32 0, i32 3>
397  ret <4 x i32> %t0
398}
399
400define void @test_31(ptr %src, ptr %dst) {
401; CHECK-LABEL: @test_31(
402; CHECK-NEXT:    [[G0:%.*]] = getelementptr <4 x i32>, ptr [[SRC:%.*]], i32 0
403; CHECK-NEXT:    [[L0:%.*]] = load <4 x i32>, ptr [[G0]], align 16
404; CHECK-NEXT:    [[S0:%.*]] = shufflevector <4 x i32> [[L0]], <4 x i32> [[L0]], <4 x i32> <i32 0, i32 2, i32 1, i32 3>
405; CHECK-NEXT:    [[S1:%.*]] = shufflevector <4 x i32> [[L0]], <4 x i32> [[L0]], <4 x i32> <i32 0, i32 0, i32 2, i32 3>
406; CHECK-NEXT:    [[S2:%.*]] = shufflevector <4 x i32> [[L0]], <4 x i32> [[L0]], <4 x i32> <i32 3, i32 1, i32 1, i32 0>
407; CHECK-NEXT:    [[S3:%.*]] = shufflevector <4 x i32> [[L0]], <4 x i32> [[L0]], <4 x i32> <i32 3, i32 3, i32 5, i32 6>
408; CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[S1]], [[S1]]
409; CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[S3]], [[S0]]
410; CHECK-NEXT:    [[T0:%.*]] = shufflevector <4 x i32> [[ADD]], <4 x i32> [[ADD]], <4 x i32> <i32 0, i32 2, i32 5, i32 2>
411; CHECK-NEXT:    [[H0:%.*]] = getelementptr <4 x i32>, ptr [[DST:%.*]], i32 0
412; CHECK-NEXT:    store <4 x i32> [[T0]], ptr [[H0]], align 16
413; CHECK-NEXT:    [[T1:%.*]] = shufflevector <4 x i32> [[ADD]], <4 x i32> [[ADD]], <4 x i32> <i32 6, i32 0, i32 1, i32 7>
414; CHECK-NEXT:    [[H1:%.*]] = getelementptr <4 x i32>, ptr [[DST]], i32 1
415; CHECK-NEXT:    store <4 x i32> [[T1]], ptr [[H1]], align 16
416; CHECK-NEXT:    ret void
417;
418  %g0 = getelementptr <4 x i32>, ptr %src, i32 0
419  %l0 = load <4 x i32>, ptr %g0, align 16
420  %s0 = shufflevector <4 x i32> %l0, <4 x i32> %l0, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
421  %s1 = shufflevector <4 x i32> %l0, <4 x i32> %l0, <4 x i32> <i32 0, i32 0, i32 2, i32 3>
422  %s2 = shufflevector <4 x i32> %l0, <4 x i32> %l0, <4 x i32> <i32 3, i32 1, i32 1, i32 0>
423  %s3 = shufflevector <4 x i32> %l0, <4 x i32> %l0, <4 x i32> <i32 3, i32 3, i32 5, i32 6>
424  %add = add <4 x i32> %s1, %s1
425  %sub = sub <4 x i32> %s3, %s0
426  %t0 = shufflevector <4 x i32> %add, <4 x i32> %add, <4 x i32> <i32 0, i32 2, i32 5, i32 2>
427  %h0 = getelementptr <4 x i32>, ptr %dst, i32 0
428  store <4 x i32> %t0, ptr %h0, align 16
429  %t1 = shufflevector <4 x i32> %add, <4 x i32> %add, <4 x i32> <i32 6, i32 0, i32 1, i32 7>
430  %h1 = getelementptr <4 x i32>, ptr %dst, i32 1
431  store <4 x i32> %t1, ptr %h1, align 16
432  ret void
433}
434
435define <16 x i32> @test_1651256324(<16 x i32> %l0, <16 x i32> %l1, <16 x i32> %l6, <16 x i32> %l7) {
436; CHECK-LABEL: @test_1651256324(
437; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[L0:%.*]], <16 x i32> [[L6:%.*]], <16 x i32> <i32 0, i32 1, i32 3, i32 10, i32 10, i32 15, i32 20, i32 20, i32 30, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
438; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[L1:%.*]], <16 x i32> [[L1]], <16 x i32> <i32 0, i32 1, i32 4, i32 5, i32 7, i32 10, i32 11, i32 11, i32 12, i32 22, i32 24, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
439; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[L1]], <16 x i32> [[L1]], <16 x i32> <i32 0, i32 1, i32 4, i32 5, i32 7, i32 10, i32 11, i32 11, i32 12, i32 22, i32 24, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
440; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[L7:%.*]], <16 x i32> [[L7]], <16 x i32> <i32 23, i32 20, i32 29, i32 25, i32 14, i32 21, i32 11, i32 9, i32 2, i32 7, i32 5, i32 15, i32 24, i32 30, i32 26, i32 5>
441; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[L6]], <16 x i32> [[L7]], <16 x i32> <i32 29, i32 13, i32 30, i32 24, i32 0, i32 15, i32 15, i32 8, i32 4, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
442; CHECK-NEXT:    [[TMP5:%.*]] = add <16 x i32> [[TMP3]], [[TMP2]]
443; CHECK-NEXT:    [[TMP6:%.*]] = sub <16 x i32> [[TMP1]], [[TMP4]]
444; CHECK-NEXT:    [[T0:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 17, i32 7, i32 23, i32 1, i32 2, i32 1, i32 8, i32 10, i32 6, i32 6, i32 18, i32 24, i32 17, i32 9, i32 21, i32 16>
445; CHECK-NEXT:    [[T1:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 10, i32 1, i32 5, i32 24, i32 19, i32 0, i32 3, i32 0, i32 17, i32 22, i32 0, i32 4, i32 1, i32 20, i32 16, i32 6>
446; CHECK-NEXT:    [[R:%.*]] = xor <16 x i32> [[T0]], [[T1]]
447; CHECK-NEXT:    ret <16 x i32> [[R]]
448;
449  %s0 = shufflevector <16 x i32> %l0, <16 x i32> %l6, <16 x i32> <i32 1, i32 20, i32 15, i32 3, i32 1, i32 10, i32 17, i32 25, i32 29, i32 23, i32 20, i32 10, i32 0, i32 20, i32 30, i32 30>
450  %s1 = shufflevector <16 x i32> %l1, <16 x i32> %l1, <16 x i32> <i32 11, i32 22, i32 1, i32 7, i32 20, i32 0, i32 2, i32 24, i32 28, i32 10, i32 31, i32 12, i32 22, i32 5, i32 11, i32 4>
451  %s2 = shufflevector <16 x i32> %l7, <16 x i32> %l7, <16 x i32> <i32 23, i32 20, i32 29, i32 25, i32 14, i32 21, i32 11, i32 9, i32 2, i32 7, i32 5, i32 15, i32 24, i32 30, i32 26, i32 5>
452  %s3 = shufflevector <16 x i32> %l6, <16 x i32> %l7, <16 x i32> <i32 29, i32 19, i32 15, i32 30, i32 13, i32 0, i32 30, i32 23, i32 26, i32 3, i32 15, i32 24, i32 29, i32 8, i32 4, i32 0>
453  %add = add <16 x i32> %s1, %s1
454  %sub = sub <16 x i32> %s0, %s3
455  %t0 = shufflevector <16 x i32> %add, <16 x i32> %sub, <16 x i32> <i32 20, i32 0, i32 29, i32 2, i32 15, i32 2, i32 11, i32 7, i32 14, i32 14, i32 19, i32 30, i32 20, i32 1, i32 18, i32 28>
456  %t1 = shufflevector <16 x i32> %sub, <16 x i32> %add, <16 x i32> <i32 23, i32 18, i32 25, i32 14, i32 11, i32 21, i32 29, i32 21, i32 4, i32 10, i32 21, i32 19, i32 18, i32 5, i32 12, i32 30>
457  %r = xor <16 x i32> %t0, %t1
458  ret <16 x i32> %r
459}
460
461define void @test_1652048214(ptr %src, ptr %dst) {
462; CHECK-LABEL: @test_1652048214(
463; CHECK-NEXT:    [[G0:%.*]] = getelementptr <8 x i32>, ptr [[SRC:%.*]], i32 0
464; CHECK-NEXT:    [[L0:%.*]] = load <8 x i32>, ptr [[G0]], align 32
465; CHECK-NEXT:    [[G1:%.*]] = getelementptr <8 x i32>, ptr [[SRC]], i32 1
466; CHECK-NEXT:    [[L1:%.*]] = load <8 x i32>, ptr [[G1]], align 32
467; CHECK-NEXT:    [[S0:%.*]] = shufflevector <8 x i32> [[L1]], <8 x i32> [[L0]], <4 x i32> <i32 6, i32 2, i32 14, i32 6>
468; CHECK-NEXT:    [[S1:%.*]] = shufflevector <8 x i32> [[L0]], <8 x i32> [[L0]], <4 x i32> <i32 9, i32 5, i32 1, i32 0>
469; CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[S1]], [[S1]]
470; CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[S0]], [[S1]]
471; CHECK-NEXT:    [[T0:%.*]] = shufflevector <4 x i32> [[SUB]], <4 x i32> [[ADD]], <4 x i32> <i32 1, i32 5, i32 7, i32 3>
472; CHECK-NEXT:    [[H0:%.*]] = getelementptr <4 x i32>, ptr [[DST:%.*]], i32 0
473; CHECK-NEXT:    store <4 x i32> [[T0]], ptr [[H0]], align 16
474; CHECK-NEXT:    ret void
475;
476  %g0 = getelementptr <8 x i32>, ptr %src, i32 0
477  %l0 = load <8 x i32>, ptr %g0, align 32
478  %g1 = getelementptr <8 x i32>, ptr %src, i32 1
479  %l1 = load <8 x i32>, ptr %g1, align 32
480  %s0 = shufflevector <8 x i32> %l1, <8 x i32> %l0, <4 x i32> <i32 6, i32 2, i32 14, i32 6>
481  %s1 = shufflevector <8 x i32> %l0, <8 x i32> %l0, <4 x i32> <i32 9, i32 5, i32 1, i32 0>
482  %add = add <4 x i32> %s1, %s1
483  %sub = sub <4 x i32> %s0, %s1
484  %t0 = shufflevector <4 x i32> %sub, <4 x i32> %add, <4 x i32> <i32 1, i32 5, i32 7, i32 3>
485  %h0 = getelementptr <4 x i32>, ptr %dst, i32 0
486  store <4 x i32> %t0, ptr %h0, align 16
487  ret void
488}
489
490define dso_local i32 @full(i8* nocapture noundef readonly %p1, i32 noundef %st1, i8* nocapture noundef readonly %p2, i32 noundef %st2) {
491; CHECK-LABEL: @full(
492; CHECK-NEXT:  entry:
493; CHECK-NEXT:    [[IDX_EXT:%.*]] = sext i32 [[ST1:%.*]] to i64
494; CHECK-NEXT:    [[IDX_EXT63:%.*]] = sext i32 [[ST2:%.*]] to i64
495; CHECK-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr inbounds i8, ptr [[P1:%.*]], i64 4
496; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i8, ptr [[P2:%.*]], i64 4
497; CHECK-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds i8, ptr [[P1]], i64 [[IDX_EXT]]
498; CHECK-NEXT:    [[ADD_PTR64:%.*]] = getelementptr inbounds i8, ptr [[P2]], i64 [[IDX_EXT63]]
499; CHECK-NEXT:    [[ARRAYIDX3_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR]], i64 4
500; CHECK-NEXT:    [[ARRAYIDX5_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64]], i64 4
501; CHECK-NEXT:    [[ADD_PTR_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR]], i64 [[IDX_EXT]]
502; CHECK-NEXT:    [[ADD_PTR64_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64]], i64 [[IDX_EXT63]]
503; CHECK-NEXT:    [[ARRAYIDX3_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR_1]], i64 4
504; CHECK-NEXT:    [[ARRAYIDX5_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64_1]], i64 4
505; CHECK-NEXT:    [[ADD_PTR_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR_1]], i64 [[IDX_EXT]]
506; CHECK-NEXT:    [[ADD_PTR64_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64_1]], i64 [[IDX_EXT63]]
507; CHECK-NEXT:    [[ARRAYIDX3_3:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR_2]], i64 4
508; CHECK-NEXT:    [[ARRAYIDX5_3:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64_2]], i64 4
509; CHECK-NEXT:    [[TMP0:%.*]] = bitcast ptr [[P1]] to ptr
510; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i8>, ptr [[TMP0]], align 1
511; CHECK-NEXT:    [[TMP2:%.*]] = bitcast ptr [[P2]] to ptr
512; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i8>, ptr [[TMP2]], align 1
513; CHECK-NEXT:    [[TMP4:%.*]] = bitcast ptr [[ARRAYIDX3]] to ptr
514; CHECK-NEXT:    [[TMP5:%.*]] = load <4 x i8>, ptr [[TMP4]], align 1
515; CHECK-NEXT:    [[TMP6:%.*]] = bitcast ptr [[ARRAYIDX5]] to ptr
516; CHECK-NEXT:    [[TMP7:%.*]] = load <4 x i8>, ptr [[TMP6]], align 1
517; CHECK-NEXT:    [[TMP8:%.*]] = bitcast ptr [[ADD_PTR]] to ptr
518; CHECK-NEXT:    [[TMP9:%.*]] = load <4 x i8>, ptr [[TMP8]], align 1
519; CHECK-NEXT:    [[TMP10:%.*]] = bitcast ptr [[ADD_PTR64]] to ptr
520; CHECK-NEXT:    [[TMP11:%.*]] = load <4 x i8>, ptr [[TMP10]], align 1
521; CHECK-NEXT:    [[TMP12:%.*]] = bitcast ptr [[ARRAYIDX3_1]] to ptr
522; CHECK-NEXT:    [[TMP13:%.*]] = load <4 x i8>, ptr [[TMP12]], align 1
523; CHECK-NEXT:    [[TMP14:%.*]] = bitcast ptr [[ARRAYIDX5_1]] to ptr
524; CHECK-NEXT:    [[TMP15:%.*]] = load <4 x i8>, ptr [[TMP14]], align 1
525; CHECK-NEXT:    [[TMP16:%.*]] = bitcast ptr [[ADD_PTR_1]] to ptr
526; CHECK-NEXT:    [[TMP17:%.*]] = load <4 x i8>, ptr [[TMP16]], align 1
527; CHECK-NEXT:    [[TMP18:%.*]] = bitcast ptr [[ADD_PTR64_1]] to ptr
528; CHECK-NEXT:    [[TMP19:%.*]] = load <4 x i8>, ptr [[TMP18]], align 1
529; CHECK-NEXT:    [[TMP20:%.*]] = bitcast ptr [[ARRAYIDX3_2]] to ptr
530; CHECK-NEXT:    [[TMP21:%.*]] = load <4 x i8>, ptr [[TMP20]], align 1
531; CHECK-NEXT:    [[TMP22:%.*]] = bitcast ptr [[ARRAYIDX5_2]] to ptr
532; CHECK-NEXT:    [[TMP23:%.*]] = load <4 x i8>, ptr [[TMP22]], align 1
533; CHECK-NEXT:    [[TMP24:%.*]] = bitcast ptr [[ADD_PTR_2]] to ptr
534; CHECK-NEXT:    [[TMP25:%.*]] = load <4 x i8>, ptr [[TMP24]], align 1
535; CHECK-NEXT:    [[TMP26:%.*]] = shufflevector <4 x i8> [[TMP25]], <4 x i8> [[TMP17]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
536; CHECK-NEXT:    [[TMP27:%.*]] = shufflevector <4 x i8> [[TMP9]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
537; CHECK-NEXT:    [[TMP28:%.*]] = shufflevector <16 x i8> [[TMP26]], <16 x i8> [[TMP27]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 undef, i32 undef, i32 undef, i32 undef>
538; CHECK-NEXT:    [[TMP29:%.*]] = shufflevector <4 x i8> [[TMP1]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
539; CHECK-NEXT:    [[TMP30:%.*]] = shufflevector <16 x i8> [[TMP28]], <16 x i8> [[TMP29]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
540; CHECK-NEXT:    [[TMP31:%.*]] = zext <16 x i8> [[TMP30]] to <16 x i32>
541; CHECK-NEXT:    [[TMP32:%.*]] = bitcast ptr [[ADD_PTR64_2]] to ptr
542; CHECK-NEXT:    [[TMP33:%.*]] = load <4 x i8>, ptr [[TMP32]], align 1
543; CHECK-NEXT:    [[TMP34:%.*]] = shufflevector <4 x i8> [[TMP33]], <4 x i8> [[TMP19]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
544; CHECK-NEXT:    [[TMP35:%.*]] = shufflevector <4 x i8> [[TMP11]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
545; CHECK-NEXT:    [[TMP36:%.*]] = shufflevector <16 x i8> [[TMP34]], <16 x i8> [[TMP35]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 undef, i32 undef, i32 undef, i32 undef>
546; CHECK-NEXT:    [[TMP37:%.*]] = shufflevector <4 x i8> [[TMP3]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
547; CHECK-NEXT:    [[TMP38:%.*]] = shufflevector <16 x i8> [[TMP36]], <16 x i8> [[TMP37]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
548; CHECK-NEXT:    [[TMP39:%.*]] = zext <16 x i8> [[TMP38]] to <16 x i32>
549; CHECK-NEXT:    [[TMP40:%.*]] = sub nsw <16 x i32> [[TMP31]], [[TMP39]]
550; CHECK-NEXT:    [[TMP41:%.*]] = bitcast ptr [[ARRAYIDX3_3]] to ptr
551; CHECK-NEXT:    [[TMP42:%.*]] = load <4 x i8>, ptr [[TMP41]], align 1
552; CHECK-NEXT:    [[TMP43:%.*]] = shufflevector <4 x i8> [[TMP42]], <4 x i8> [[TMP21]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
553; CHECK-NEXT:    [[TMP44:%.*]] = shufflevector <4 x i8> [[TMP13]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
554; CHECK-NEXT:    [[TMP45:%.*]] = shufflevector <16 x i8> [[TMP43]], <16 x i8> [[TMP44]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 undef, i32 undef, i32 undef, i32 undef>
555; CHECK-NEXT:    [[TMP46:%.*]] = shufflevector <4 x i8> [[TMP5]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
556; CHECK-NEXT:    [[TMP47:%.*]] = shufflevector <16 x i8> [[TMP45]], <16 x i8> [[TMP46]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
557; CHECK-NEXT:    [[TMP48:%.*]] = zext <16 x i8> [[TMP47]] to <16 x i32>
558; CHECK-NEXT:    [[TMP49:%.*]] = bitcast ptr [[ARRAYIDX5_3]] to ptr
559; CHECK-NEXT:    [[TMP50:%.*]] = load <4 x i8>, ptr [[TMP49]], align 1
560; CHECK-NEXT:    [[TMP51:%.*]] = shufflevector <4 x i8> [[TMP50]], <4 x i8> [[TMP23]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
561; CHECK-NEXT:    [[TMP52:%.*]] = shufflevector <4 x i8> [[TMP15]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
562; CHECK-NEXT:    [[TMP53:%.*]] = shufflevector <16 x i8> [[TMP51]], <16 x i8> [[TMP52]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 undef, i32 undef, i32 undef, i32 undef>
563; CHECK-NEXT:    [[TMP54:%.*]] = shufflevector <4 x i8> [[TMP7]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
564; CHECK-NEXT:    [[TMP55:%.*]] = shufflevector <16 x i8> [[TMP53]], <16 x i8> [[TMP54]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
565; CHECK-NEXT:    [[TMP56:%.*]] = zext <16 x i8> [[TMP55]] to <16 x i32>
566; CHECK-NEXT:    [[TMP57:%.*]] = sub nsw <16 x i32> [[TMP48]], [[TMP56]]
567; CHECK-NEXT:    [[TMP58:%.*]] = shl nsw <16 x i32> [[TMP57]], <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
568; CHECK-NEXT:    [[TMP59:%.*]] = add nsw <16 x i32> [[TMP58]], [[TMP40]]
569; CHECK-NEXT:    [[TMP60:%.*]] = shufflevector <16 x i32> [[TMP59]], <16 x i32> [[TMP59]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
570; CHECK-NEXT:    [[TMP61:%.*]] = shufflevector <16 x i32> [[TMP59]], <16 x i32> [[TMP59]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
571; CHECK-NEXT:    [[TMP62:%.*]] = shufflevector <16 x i32> [[TMP59]], <16 x i32> poison, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
572; CHECK-NEXT:    [[TMP63:%.*]] = shufflevector <16 x i32> [[TMP59]], <16 x i32> poison, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
573; CHECK-NEXT:    [[TMP64:%.*]] = add nsw <16 x i32> [[TMP61]], [[TMP63]]
574; CHECK-NEXT:    [[TMP65:%.*]] = sub nsw <16 x i32> [[TMP60]], [[TMP62]]
575; CHECK-NEXT:    [[TMP66:%.*]] = shufflevector <16 x i32> [[TMP64]], <16 x i32> [[TMP65]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
576; CHECK-NEXT:    [[TMP67:%.*]] = shufflevector <16 x i32> [[TMP64]], <16 x i32> [[TMP65]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
577; CHECK-NEXT:    [[TMP68:%.*]] = shufflevector <16 x i32> [[TMP64]], <16 x i32> [[TMP65]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
578; CHECK-NEXT:    [[TMP69:%.*]] = shufflevector <16 x i32> [[TMP64]], <16 x i32> [[TMP65]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
579; CHECK-NEXT:    [[TMP70:%.*]] = add nsw <16 x i32> [[TMP67]], [[TMP69]]
580; CHECK-NEXT:    [[TMP71:%.*]] = sub nsw <16 x i32> [[TMP66]], [[TMP68]]
581; CHECK-NEXT:    [[TMP72:%.*]] = shufflevector <16 x i32> [[TMP70]], <16 x i32> [[TMP71]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
582; CHECK-NEXT:    [[TMP73:%.*]] = shufflevector <16 x i32> [[TMP70]], <16 x i32> [[TMP71]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
583; CHECK-NEXT:    [[TMP74:%.*]] = shufflevector <16 x i32> [[TMP70]], <16 x i32> [[TMP71]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
584; CHECK-NEXT:    [[TMP75:%.*]] = shufflevector <16 x i32> [[TMP70]], <16 x i32> [[TMP71]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
585; CHECK-NEXT:    [[TMP76:%.*]] = add nsw <16 x i32> [[TMP73]], [[TMP75]]
586; CHECK-NEXT:    [[TMP77:%.*]] = sub nsw <16 x i32> [[TMP72]], [[TMP74]]
587; CHECK-NEXT:    [[TMP78:%.*]] = shufflevector <16 x i32> [[TMP76]], <16 x i32> [[TMP77]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
588; CHECK-NEXT:    [[TMP79:%.*]] = shufflevector <16 x i32> [[TMP76]], <16 x i32> [[TMP77]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
589; CHECK-NEXT:    [[TMP80:%.*]] = shufflevector <16 x i32> [[TMP76]], <16 x i32> [[TMP77]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
590; CHECK-NEXT:    [[TMP81:%.*]] = shufflevector <16 x i32> [[TMP76]], <16 x i32> [[TMP77]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
591; CHECK-NEXT:    [[TMP82:%.*]] = add nsw <16 x i32> [[TMP79]], [[TMP81]]
592; CHECK-NEXT:    [[TMP83:%.*]] = sub nsw <16 x i32> [[TMP78]], [[TMP80]]
593; CHECK-NEXT:    [[TMP84:%.*]] = shufflevector <16 x i32> [[TMP82]], <16 x i32> [[TMP83]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
594; CHECK-NEXT:    [[TMP85:%.*]] = lshr <16 x i32> [[TMP84]], <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
595; CHECK-NEXT:    [[TMP86:%.*]] = and <16 x i32> [[TMP85]], <i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537>
596; CHECK-NEXT:    [[TMP87:%.*]] = mul nuw <16 x i32> [[TMP86]], <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
597; CHECK-NEXT:    [[TMP88:%.*]] = add <16 x i32> [[TMP87]], [[TMP84]]
598; CHECK-NEXT:    [[TMP89:%.*]] = xor <16 x i32> [[TMP88]], [[TMP87]]
599; CHECK-NEXT:    [[TMP90:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP89]])
600; CHECK-NEXT:    [[CONV118:%.*]] = and i32 [[TMP90]], 65535
601; CHECK-NEXT:    [[SHR:%.*]] = lshr i32 [[TMP90]], 16
602; CHECK-NEXT:    [[ADD119:%.*]] = add nuw nsw i32 [[CONV118]], [[SHR]]
603; CHECK-NEXT:    [[SHR120:%.*]] = lshr i32 [[ADD119]], 1
604; CHECK-NEXT:    ret i32 [[SHR120]]
605;
606entry:
607  %idx.ext = sext i32 %st1 to i64
608  %idx.ext63 = sext i32 %st2 to i64
609  %arrayidx3 = getelementptr inbounds i8, i8* %p1, i64 4
610  %arrayidx5 = getelementptr inbounds i8, i8* %p2, i64 4
611  %add.ptr = getelementptr inbounds i8, i8* %p1, i64 %idx.ext
612  %add.ptr64 = getelementptr inbounds i8, i8* %p2, i64 %idx.ext63
613  %arrayidx3.1 = getelementptr inbounds i8, i8* %add.ptr, i64 4
614  %arrayidx5.1 = getelementptr inbounds i8, i8* %add.ptr64, i64 4
615  %add.ptr.1 = getelementptr inbounds i8, i8* %add.ptr, i64 %idx.ext
616  %add.ptr64.1 = getelementptr inbounds i8, i8* %add.ptr64, i64 %idx.ext63
617  %arrayidx3.2 = getelementptr inbounds i8, i8* %add.ptr.1, i64 4
618  %arrayidx5.2 = getelementptr inbounds i8, i8* %add.ptr64.1, i64 4
619  %add.ptr.2 = getelementptr inbounds i8, i8* %add.ptr.1, i64 %idx.ext
620  %add.ptr64.2 = getelementptr inbounds i8, i8* %add.ptr64.1, i64 %idx.ext63
621  %arrayidx3.3 = getelementptr inbounds i8, i8* %add.ptr.2, i64 4
622  %arrayidx5.3 = getelementptr inbounds i8, i8* %add.ptr64.2, i64 4
623  %0 = bitcast i8* %p1 to <4 x i8>*
624  %1 = load <4 x i8>, <4 x i8>* %0, align 1
625  %2 = bitcast i8* %p2 to <4 x i8>*
626  %3 = load <4 x i8>, <4 x i8>* %2, align 1
627  %4 = bitcast i8* %arrayidx3 to <4 x i8>*
628  %5 = load <4 x i8>, <4 x i8>* %4, align 1
629  %6 = bitcast i8* %arrayidx5 to <4 x i8>*
630  %7 = load <4 x i8>, <4 x i8>* %6, align 1
631  %8 = bitcast i8* %add.ptr to <4 x i8>*
632  %9 = load <4 x i8>, <4 x i8>* %8, align 1
633  %10 = bitcast i8* %add.ptr64 to <4 x i8>*
634  %11 = load <4 x i8>, <4 x i8>* %10, align 1
635  %12 = bitcast i8* %arrayidx3.1 to <4 x i8>*
636  %13 = load <4 x i8>, <4 x i8>* %12, align 1
637  %14 = bitcast i8* %arrayidx5.1 to <4 x i8>*
638  %15 = load <4 x i8>, <4 x i8>* %14, align 1
639  %16 = bitcast i8* %add.ptr.1 to <4 x i8>*
640  %17 = load <4 x i8>, <4 x i8>* %16, align 1
641  %18 = bitcast i8* %add.ptr64.1 to <4 x i8>*
642  %19 = load <4 x i8>, <4 x i8>* %18, align 1
643  %20 = bitcast i8* %arrayidx3.2 to <4 x i8>*
644  %21 = load <4 x i8>, <4 x i8>* %20, align 1
645  %22 = bitcast i8* %arrayidx5.2 to <4 x i8>*
646  %23 = load <4 x i8>, <4 x i8>* %22, align 1
647  %24 = bitcast i8* %add.ptr.2 to <4 x i8>*
648  %25 = load <4 x i8>, <4 x i8>* %24, align 1
649  %26 = shufflevector <4 x i8> %25, <4 x i8> %17, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
650  %27 = shufflevector <4 x i8> %9, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
651  %28 = shufflevector <16 x i8> %26, <16 x i8> %27, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 undef, i32 undef, i32 undef, i32 undef>
652  %29 = shufflevector <4 x i8> %1, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
653  %30 = shufflevector <16 x i8> %28, <16 x i8> %29, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
654  %31 = zext <16 x i8> %30 to <16 x i32>
655  %32 = bitcast i8* %add.ptr64.2 to <4 x i8>*
656  %33 = load <4 x i8>, <4 x i8>* %32, align 1
657  %34 = shufflevector <4 x i8> %33, <4 x i8> %19, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
658  %35 = shufflevector <4 x i8> %11, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
659  %36 = shufflevector <16 x i8> %34, <16 x i8> %35, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 undef, i32 undef, i32 undef, i32 undef>
660  %37 = shufflevector <4 x i8> %3, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
661  %38 = shufflevector <16 x i8> %36, <16 x i8> %37, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
662  %39 = zext <16 x i8> %38 to <16 x i32>
663  %40 = sub nsw <16 x i32> %31, %39
664  %41 = bitcast i8* %arrayidx3.3 to <4 x i8>*
665  %42 = load <4 x i8>, <4 x i8>* %41, align 1
666  %43 = shufflevector <4 x i8> %42, <4 x i8> %21, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
667  %44 = shufflevector <4 x i8> %13, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
668  %45 = shufflevector <16 x i8> %43, <16 x i8> %44, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 undef, i32 undef, i32 undef, i32 undef>
669  %46 = shufflevector <4 x i8> %5, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
670  %47 = shufflevector <16 x i8> %45, <16 x i8> %46, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
671  %48 = zext <16 x i8> %47 to <16 x i32>
672  %49 = bitcast i8* %arrayidx5.3 to <4 x i8>*
673  %50 = load <4 x i8>, <4 x i8>* %49, align 1
674  %51 = shufflevector <4 x i8> %50, <4 x i8> %23, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
675  %52 = shufflevector <4 x i8> %15, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
676  %53 = shufflevector <16 x i8> %51, <16 x i8> %52, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 undef, i32 undef, i32 undef, i32 undef>
677  %54 = shufflevector <4 x i8> %7, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
678  %55 = shufflevector <16 x i8> %53, <16 x i8> %54, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
679  %56 = zext <16 x i8> %55 to <16 x i32>
680  %57 = sub nsw <16 x i32> %48, %56
681  %58 = shl nsw <16 x i32> %57, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
682  %59 = add nsw <16 x i32> %58, %40
683  %reorder = shufflevector <16 x i32> %59, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
684  %60 = add nsw <16 x i32> %59, %reorder
685  %61 = sub nsw <16 x i32> %59, %reorder
686  %62 = shufflevector <16 x i32> %60, <16 x i32> %61, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 22, i32 18, i32 26, i32 30, i32 5, i32 1, i32 9, i32 13, i32 20, i32 16, i32 24, i32 28>
687  %reorder1 = shufflevector <16 x i32> %60, <16 x i32> %61, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 20, i32 16, i32 24, i32 28, i32 7, i32 3, i32 11, i32 15, i32 22, i32 18, i32 26, i32 30>
688  %63 = add nsw <16 x i32> %62, %reorder1
689  %64 = sub nsw <16 x i32> %62, %reorder1
690  %65 = shufflevector <16 x i32> %63, <16 x i32> %64, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
691  %reorder2 = shufflevector <16 x i32> %63, <16 x i32> %64, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 25, i32 24, i32 27, i32 26, i32 29, i32 28, i32 31, i32 30>
692  %66 = add nsw <16 x i32> %65, %reorder2
693  %67 = sub nsw <16 x i32> %65, %reorder2
694  %68 = shufflevector <16 x i32> %66, <16 x i32> %67, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 20, i32 5, i32 6, i32 23, i32 24, i32 9, i32 10, i32 27, i32 28, i32 13, i32 14, i32 31>
695  %reorder3 = shufflevector <16 x i32> %66, <16 x i32> %67, <16 x i32> <i32 2, i32 19, i32 0, i32 17, i32 23, i32 6, i32 5, i32 20, i32 27, i32 10, i32 9, i32 24, i32 31, i32 14, i32 13, i32 28>
696  %69 = add nsw <16 x i32> %68, %reorder3
697  %70 = sub nsw <16 x i32> %68, %reorder3
698  %71 = shufflevector <16 x i32> %69, <16 x i32> %70, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 4, i32 5, i32 22, i32 23, i32 8, i32 9, i32 26, i32 27, i32 12, i32 13, i32 30, i32 31>
699  %72 = lshr <16 x i32> %71, <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
700  %73 = and <16 x i32> %72, <i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537>
701  %74 = mul nuw <16 x i32> %73, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
702  %75 = add <16 x i32> %74, %71
703  %76 = xor <16 x i32> %75, %74
704  %77 = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> %76)
705  %conv118 = and i32 %77, 65535
706  %shr = lshr i32 %77, 16
707  %add119 = add nuw nsw i32 %conv118, %shr
708  %shr120 = lshr i32 %add119, 1
709  ret i32 %shr120
710}
711
712define i32 @full_reorder(ptr nocapture noundef readonly %pix1, i32 noundef %i_pix1, ptr nocapture noundef readonly %pix2, i32 noundef %i_pix2) {
713; CHECK-LABEL: @full_reorder(
714; CHECK-NEXT:  entry:
715; CHECK-NEXT:    [[IDX_EXT:%.*]] = sext i32 [[I_PIX1:%.*]] to i64
716; CHECK-NEXT:    [[IDX_EXT63:%.*]] = sext i32 [[I_PIX2:%.*]] to i64
717; CHECK-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr inbounds i8, ptr [[PIX1:%.*]], i64 4
718; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i8, ptr [[PIX2:%.*]], i64 4
719; CHECK-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds i8, ptr [[PIX1]], i64 [[IDX_EXT]]
720; CHECK-NEXT:    [[ADD_PTR64:%.*]] = getelementptr inbounds i8, ptr [[PIX2]], i64 [[IDX_EXT63]]
721; CHECK-NEXT:    [[ARRAYIDX3_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR]], i64 4
722; CHECK-NEXT:    [[ARRAYIDX5_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64]], i64 4
723; CHECK-NEXT:    [[ADD_PTR_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR]], i64 [[IDX_EXT]]
724; CHECK-NEXT:    [[ADD_PTR64_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64]], i64 [[IDX_EXT63]]
725; CHECK-NEXT:    [[ARRAYIDX3_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR_1]], i64 4
726; CHECK-NEXT:    [[ARRAYIDX5_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64_1]], i64 4
727; CHECK-NEXT:    [[ADD_PTR_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR_1]], i64 [[IDX_EXT]]
728; CHECK-NEXT:    [[ADD_PTR64_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64_1]], i64 [[IDX_EXT63]]
729; CHECK-NEXT:    [[ARRAYIDX3_3:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR_2]], i64 4
730; CHECK-NEXT:    [[ARRAYIDX5_3:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64_2]], i64 4
731; CHECK-NEXT:    [[TMP0:%.*]] = load <4 x i8>, ptr [[PIX1]], align 1
732; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i8>, ptr [[PIX2]], align 1
733; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i8>, ptr [[ARRAYIDX3]], align 1
734; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i8>, ptr [[ARRAYIDX5]], align 1
735; CHECK-NEXT:    [[TMP4:%.*]] = load <4 x i8>, ptr [[ADD_PTR]], align 1
736; CHECK-NEXT:    [[TMP5:%.*]] = load <4 x i8>, ptr [[ADD_PTR64]], align 1
737; CHECK-NEXT:    [[TMP6:%.*]] = load <4 x i8>, ptr [[ARRAYIDX3_1]], align 1
738; CHECK-NEXT:    [[TMP7:%.*]] = load <4 x i8>, ptr [[ARRAYIDX5_1]], align 1
739; CHECK-NEXT:    [[TMP8:%.*]] = load <4 x i8>, ptr [[ADD_PTR_1]], align 1
740; CHECK-NEXT:    [[TMP9:%.*]] = load <4 x i8>, ptr [[ADD_PTR64_1]], align 1
741; CHECK-NEXT:    [[TMP10:%.*]] = load <4 x i8>, ptr [[ARRAYIDX3_2]], align 1
742; CHECK-NEXT:    [[TMP11:%.*]] = load <4 x i8>, ptr [[ARRAYIDX5_2]], align 1
743; CHECK-NEXT:    [[TMP12:%.*]] = load <4 x i8>, ptr [[ADD_PTR_2]], align 1
744; CHECK-NEXT:    [[TMP13:%.*]] = shufflevector <4 x i8> [[TMP12]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
745; CHECK-NEXT:    [[TMP14:%.*]] = shufflevector <4 x i8> [[TMP8]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
746; CHECK-NEXT:    [[TMP15:%.*]] = shufflevector <16 x i8> [[TMP13]], <16 x i8> [[TMP14]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
747; CHECK-NEXT:    [[TMP16:%.*]] = shufflevector <4 x i8> [[TMP4]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
748; CHECK-NEXT:    [[TMP17:%.*]] = shufflevector <16 x i8> [[TMP15]], <16 x i8> [[TMP16]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
749; CHECK-NEXT:    [[TMP18:%.*]] = shufflevector <4 x i8> [[TMP0]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
750; CHECK-NEXT:    [[TMP19:%.*]] = shufflevector <16 x i8> [[TMP17]], <16 x i8> [[TMP18]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
751; CHECK-NEXT:    [[TMP20:%.*]] = zext <16 x i8> [[TMP19]] to <16 x i32>
752; CHECK-NEXT:    [[TMP21:%.*]] = load <4 x i8>, ptr [[ADD_PTR64_2]], align 1
753; CHECK-NEXT:    [[TMP22:%.*]] = shufflevector <4 x i8> [[TMP21]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
754; CHECK-NEXT:    [[TMP23:%.*]] = shufflevector <4 x i8> [[TMP9]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
755; CHECK-NEXT:    [[TMP24:%.*]] = shufflevector <16 x i8> [[TMP22]], <16 x i8> [[TMP23]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
756; CHECK-NEXT:    [[TMP25:%.*]] = shufflevector <4 x i8> [[TMP5]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
757; CHECK-NEXT:    [[TMP26:%.*]] = shufflevector <16 x i8> [[TMP24]], <16 x i8> [[TMP25]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
758; CHECK-NEXT:    [[TMP27:%.*]] = shufflevector <4 x i8> [[TMP1]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
759; CHECK-NEXT:    [[TMP28:%.*]] = shufflevector <16 x i8> [[TMP26]], <16 x i8> [[TMP27]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
760; CHECK-NEXT:    [[TMP29:%.*]] = zext <16 x i8> [[TMP28]] to <16 x i32>
761; CHECK-NEXT:    [[TMP30:%.*]] = sub nsw <16 x i32> [[TMP20]], [[TMP29]]
762; CHECK-NEXT:    [[TMP31:%.*]] = load <4 x i8>, ptr [[ARRAYIDX3_3]], align 1
763; CHECK-NEXT:    [[TMP32:%.*]] = shufflevector <4 x i8> [[TMP31]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
764; CHECK-NEXT:    [[TMP33:%.*]] = shufflevector <4 x i8> [[TMP10]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
765; CHECK-NEXT:    [[TMP34:%.*]] = shufflevector <16 x i8> [[TMP32]], <16 x i8> [[TMP33]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
766; CHECK-NEXT:    [[TMP35:%.*]] = shufflevector <4 x i8> [[TMP6]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
767; CHECK-NEXT:    [[TMP36:%.*]] = shufflevector <16 x i8> [[TMP34]], <16 x i8> [[TMP35]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
768; CHECK-NEXT:    [[TMP37:%.*]] = shufflevector <4 x i8> [[TMP2]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
769; CHECK-NEXT:    [[TMP38:%.*]] = shufflevector <16 x i8> [[TMP36]], <16 x i8> [[TMP37]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
770; CHECK-NEXT:    [[TMP39:%.*]] = zext <16 x i8> [[TMP38]] to <16 x i32>
771; CHECK-NEXT:    [[TMP40:%.*]] = load <4 x i8>, ptr [[ARRAYIDX5_3]], align 1
772; CHECK-NEXT:    [[TMP41:%.*]] = shufflevector <4 x i8> [[TMP40]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
773; CHECK-NEXT:    [[TMP42:%.*]] = shufflevector <4 x i8> [[TMP11]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
774; CHECK-NEXT:    [[TMP43:%.*]] = shufflevector <16 x i8> [[TMP41]], <16 x i8> [[TMP42]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
775; CHECK-NEXT:    [[TMP44:%.*]] = shufflevector <4 x i8> [[TMP7]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
776; CHECK-NEXT:    [[TMP45:%.*]] = shufflevector <16 x i8> [[TMP43]], <16 x i8> [[TMP44]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
777; CHECK-NEXT:    [[TMP46:%.*]] = shufflevector <4 x i8> [[TMP3]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
778; CHECK-NEXT:    [[TMP47:%.*]] = shufflevector <16 x i8> [[TMP45]], <16 x i8> [[TMP46]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
779; CHECK-NEXT:    [[TMP48:%.*]] = zext <16 x i8> [[TMP47]] to <16 x i32>
780; CHECK-NEXT:    [[TMP49:%.*]] = sub nsw <16 x i32> [[TMP39]], [[TMP48]]
781; CHECK-NEXT:    [[TMP50:%.*]] = shl nsw <16 x i32> [[TMP49]], <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
782; CHECK-NEXT:    [[TMP51:%.*]] = add nsw <16 x i32> [[TMP50]], [[TMP30]]
783; CHECK-NEXT:    [[TMP52:%.*]] = shufflevector <16 x i32> [[TMP51]], <16 x i32> [[TMP51]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
784; CHECK-NEXT:    [[TMP53:%.*]] = shufflevector <16 x i32> [[TMP51]], <16 x i32> [[TMP51]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
785; CHECK-NEXT:    [[TMP54:%.*]] = shufflevector <16 x i32> [[TMP51]], <16 x i32> poison, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
786; CHECK-NEXT:    [[TMP55:%.*]] = shufflevector <16 x i32> [[TMP51]], <16 x i32> poison, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
787; CHECK-NEXT:    [[TMP56:%.*]] = add nsw <16 x i32> [[TMP53]], [[TMP55]]
788; CHECK-NEXT:    [[TMP57:%.*]] = sub nsw <16 x i32> [[TMP52]], [[TMP54]]
789; CHECK-NEXT:    [[TMP58:%.*]] = shufflevector <16 x i32> [[TMP56]], <16 x i32> [[TMP57]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
790; CHECK-NEXT:    [[TMP59:%.*]] = shufflevector <16 x i32> [[TMP56]], <16 x i32> [[TMP57]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
791; CHECK-NEXT:    [[TMP60:%.*]] = shufflevector <16 x i32> [[TMP56]], <16 x i32> [[TMP57]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
792; CHECK-NEXT:    [[TMP61:%.*]] = shufflevector <16 x i32> [[TMP56]], <16 x i32> [[TMP57]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
793; CHECK-NEXT:    [[TMP62:%.*]] = add nsw <16 x i32> [[TMP59]], [[TMP61]]
794; CHECK-NEXT:    [[TMP63:%.*]] = sub nsw <16 x i32> [[TMP58]], [[TMP60]]
795; CHECK-NEXT:    [[TMP64:%.*]] = shufflevector <16 x i32> [[TMP62]], <16 x i32> [[TMP63]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
796; CHECK-NEXT:    [[TMP65:%.*]] = shufflevector <16 x i32> [[TMP62]], <16 x i32> [[TMP63]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
797; CHECK-NEXT:    [[TMP66:%.*]] = shufflevector <16 x i32> [[TMP62]], <16 x i32> [[TMP63]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
798; CHECK-NEXT:    [[TMP67:%.*]] = shufflevector <16 x i32> [[TMP62]], <16 x i32> [[TMP63]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
799; CHECK-NEXT:    [[TMP68:%.*]] = add nsw <16 x i32> [[TMP65]], [[TMP67]]
800; CHECK-NEXT:    [[TMP69:%.*]] = sub nsw <16 x i32> [[TMP64]], [[TMP66]]
801; CHECK-NEXT:    [[TMP70:%.*]] = shufflevector <16 x i32> [[TMP68]], <16 x i32> [[TMP69]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
802; CHECK-NEXT:    [[TMP71:%.*]] = shufflevector <16 x i32> [[TMP68]], <16 x i32> [[TMP69]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
803; CHECK-NEXT:    [[TMP72:%.*]] = shufflevector <16 x i32> [[TMP68]], <16 x i32> [[TMP69]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
804; CHECK-NEXT:    [[TMP73:%.*]] = shufflevector <16 x i32> [[TMP68]], <16 x i32> [[TMP69]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
805; CHECK-NEXT:    [[TMP74:%.*]] = add nsw <16 x i32> [[TMP71]], [[TMP73]]
806; CHECK-NEXT:    [[TMP75:%.*]] = sub nsw <16 x i32> [[TMP70]], [[TMP72]]
807; CHECK-NEXT:    [[TMP76:%.*]] = shufflevector <16 x i32> [[TMP74]], <16 x i32> [[TMP75]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
808; CHECK-NEXT:    [[TMP77:%.*]] = lshr <16 x i32> [[TMP76]], <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
809; CHECK-NEXT:    [[TMP78:%.*]] = and <16 x i32> [[TMP77]], <i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537>
810; CHECK-NEXT:    [[TMP79:%.*]] = mul nuw <16 x i32> [[TMP78]], <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
811; CHECK-NEXT:    [[TMP80:%.*]] = add <16 x i32> [[TMP79]], [[TMP76]]
812; CHECK-NEXT:    [[TMP81:%.*]] = xor <16 x i32> [[TMP80]], [[TMP79]]
813; CHECK-NEXT:    [[TMP82:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP81]])
814; CHECK-NEXT:    [[CONV118:%.*]] = and i32 [[TMP82]], 65535
815; CHECK-NEXT:    [[SHR:%.*]] = lshr i32 [[TMP82]], 16
816; CHECK-NEXT:    [[ADD119:%.*]] = add nuw nsw i32 [[CONV118]], [[SHR]]
817; CHECK-NEXT:    [[SHR120:%.*]] = lshr i32 [[ADD119]], 1
818; CHECK-NEXT:    ret i32 [[SHR120]]
819;
820entry:
821  %idx.ext = sext i32 %i_pix1 to i64
822  %idx.ext63 = sext i32 %i_pix2 to i64
823  %arrayidx3 = getelementptr inbounds i8, ptr %pix1, i64 4
824  %arrayidx5 = getelementptr inbounds i8, ptr %pix2, i64 4
825  %add.ptr = getelementptr inbounds i8, ptr %pix1, i64 %idx.ext
826  %add.ptr64 = getelementptr inbounds i8, ptr %pix2, i64 %idx.ext63
827  %arrayidx3.1 = getelementptr inbounds i8, ptr %add.ptr, i64 4
828  %arrayidx5.1 = getelementptr inbounds i8, ptr %add.ptr64, i64 4
829  %add.ptr.1 = getelementptr inbounds i8, ptr %add.ptr, i64 %idx.ext
830  %add.ptr64.1 = getelementptr inbounds i8, ptr %add.ptr64, i64 %idx.ext63
831  %arrayidx3.2 = getelementptr inbounds i8, ptr %add.ptr.1, i64 4
832  %arrayidx5.2 = getelementptr inbounds i8, ptr %add.ptr64.1, i64 4
833  %add.ptr.2 = getelementptr inbounds i8, ptr %add.ptr.1, i64 %idx.ext
834  %add.ptr64.2 = getelementptr inbounds i8, ptr %add.ptr64.1, i64 %idx.ext63
835  %arrayidx3.3 = getelementptr inbounds i8, ptr %add.ptr.2, i64 4
836  %arrayidx5.3 = getelementptr inbounds i8, ptr %add.ptr64.2, i64 4
837  %0 = load <4 x i8>, ptr %pix1, align 1
838  %1 = load <4 x i8>, ptr %pix2, align 1
839  %2 = load <4 x i8>, ptr %arrayidx3, align 1
840  %3 = load <4 x i8>, ptr %arrayidx5, align 1
841  %4 = load <4 x i8>, ptr %add.ptr, align 1
842  %5 = load <4 x i8>, ptr %add.ptr64, align 1
843  %6 = load <4 x i8>, ptr %arrayidx3.1, align 1
844  %7 = load <4 x i8>, ptr %arrayidx5.1, align 1
845  %8 = load <4 x i8>, ptr %add.ptr.1, align 1
846  %9 = load <4 x i8>, ptr %add.ptr64.1, align 1
847  %10 = load <4 x i8>, ptr %arrayidx3.2, align 1
848  %11 = load <4 x i8>, ptr %arrayidx5.2, align 1
849  %12 = load <4 x i8>, ptr %add.ptr.2, align 1
850  %13 = shufflevector <4 x i8> %12, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
851  %14 = shufflevector <4 x i8> %8, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
852  %15 = shufflevector <16 x i8> %13, <16 x i8> %14, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
853  %16 = shufflevector <4 x i8> %4, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
854  %17 = shufflevector <16 x i8> %15, <16 x i8> %16, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
855  %18 = shufflevector <4 x i8> %0, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
856  %19 = shufflevector <16 x i8> %17, <16 x i8> %18, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
857  %20 = zext <16 x i8> %19 to <16 x i32>
858  %21 = load <4 x i8>, ptr %add.ptr64.2, align 1
859  %22 = shufflevector <4 x i8> %21, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
860  %23 = shufflevector <4 x i8> %9, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
861  %24 = shufflevector <16 x i8> %22, <16 x i8> %23, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
862  %25 = shufflevector <4 x i8> %5, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
863  %26 = shufflevector <16 x i8> %24, <16 x i8> %25, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
864  %27 = shufflevector <4 x i8> %1, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
865  %28 = shufflevector <16 x i8> %26, <16 x i8> %27, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
866  %29 = zext <16 x i8> %28 to <16 x i32>
867  %30 = sub nsw <16 x i32> %20, %29
868  %31 = load <4 x i8>, ptr %arrayidx3.3, align 1
869  %32 = shufflevector <4 x i8> %31, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
870  %33 = shufflevector <4 x i8> %10, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
871  %34 = shufflevector <16 x i8> %32, <16 x i8> %33, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
872  %35 = shufflevector <4 x i8> %6, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
873  %36 = shufflevector <16 x i8> %34, <16 x i8> %35, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
874  %37 = shufflevector <4 x i8> %2, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
875  %38 = shufflevector <16 x i8> %36, <16 x i8> %37, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
876  %39 = zext <16 x i8> %38 to <16 x i32>
877  %40 = load <4 x i8>, ptr %arrayidx5.3, align 1
878  %41 = shufflevector <4 x i8> %40, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
879  %42 = shufflevector <4 x i8> %11, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
880  %43 = shufflevector <16 x i8> %41, <16 x i8> %42, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
881  %44 = shufflevector <4 x i8> %7, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
882  %45 = shufflevector <16 x i8> %43, <16 x i8> %44, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
883  %46 = shufflevector <4 x i8> %3, <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
884  %47 = shufflevector <16 x i8> %45, <16 x i8> %46, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
885  %48 = zext <16 x i8> %47 to <16 x i32>
886  %49 = sub nsw <16 x i32> %39, %48
887  %50 = shl nsw <16 x i32> %49, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
888  %51 = add nsw <16 x i32> %50, %30
889  %reorder = shufflevector <16 x i32> %51, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
890  %52 = add nsw <16 x i32> %51, %reorder
891  %53 = sub nsw <16 x i32> %51, %reorder
892  %54 = shufflevector <16 x i32> %52, <16 x i32> %53, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 22, i32 18, i32 26, i32 30, i32 5, i32 1, i32 9, i32 13, i32 20, i32 16, i32 24, i32 28>
893  %reorder191 = shufflevector <16 x i32> %52, <16 x i32> %53, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 20, i32 16, i32 24, i32 28, i32 7, i32 3, i32 11, i32 15, i32 22, i32 18, i32 26, i32 30>
894  %55 = add nsw <16 x i32> %54, %reorder191
895  %56 = sub nsw <16 x i32> %54, %reorder191
896  %57 = shufflevector <16 x i32> %55, <16 x i32> %56, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
897  %reorder192 = shufflevector <16 x i32> %55, <16 x i32> %56, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 25, i32 24, i32 27, i32 26, i32 29, i32 28, i32 31, i32 30>
898  %58 = add nsw <16 x i32> %57, %reorder192
899  %59 = sub nsw <16 x i32> %57, %reorder192
900  %60 = shufflevector <16 x i32> %58, <16 x i32> %59, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 20, i32 5, i32 6, i32 23, i32 24, i32 9, i32 10, i32 27, i32 28, i32 13, i32 14, i32 31>
901  %reorder193 = shufflevector <16 x i32> %58, <16 x i32> %59, <16 x i32> <i32 2, i32 19, i32 0, i32 17, i32 23, i32 6, i32 5, i32 20, i32 27, i32 10, i32 9, i32 24, i32 31, i32 14, i32 13, i32 28>
902  %61 = add nsw <16 x i32> %60, %reorder193
903  %62 = sub nsw <16 x i32> %60, %reorder193
904  %63 = shufflevector <16 x i32> %61, <16 x i32> %62, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 4, i32 5, i32 22, i32 23, i32 8, i32 9, i32 26, i32 27, i32 12, i32 13, i32 30, i32 31>
905  %64 = lshr <16 x i32> %63, <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
906  %65 = and <16 x i32> %64, <i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537>
907  %66 = mul nuw <16 x i32> %65, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
908  %67 = add <16 x i32> %66, %63
909  %68 = xor <16 x i32> %67, %66
910  %69 = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> %68)
911  %conv118 = and i32 %69, 65535
912  %shr = lshr i32 %69, 16
913  %add119 = add nuw nsw i32 %conv118, %shr
914  %shr120 = lshr i32 %add119, 1
915  ret i32 %shr120
916}
917
918define void @manyundefs()
919; CHECK-LABEL: @manyundefs(
920; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x float> undef, float undef, i64 1
921; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x float> undef, float undef, i64 1
922; CHECK-NEXT:    [[TMP3:%.*]] = fadd <2 x float> [[TMP1]], [[TMP2]]
923; CHECK-NEXT:    [[TMP4:%.*]] = fmul <2 x float> [[TMP1]], [[TMP2]]
924; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> [[TMP4]], <2 x i32> <i32 0, i32 3>
925; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <2 x i32> <i32 1, i32 0>
926; CHECK-NEXT:    ret void
927;
928unnamed_addr #0 align 2 {
929  %1 = insertelement <2 x float> undef, float undef, i64 1
930  %2 = insertelement <2 x float> undef, float undef, i64 1
931  %3 = fadd <2 x float> %1, %2
932  %4 = fmul <2 x float> %1, %2
933  %5 = shufflevector <2 x float> %3, <2 x float> %4, <2 x i32> <i32 0, i32 3>
934  %6 = shufflevector <2 x float> %4, <2 x float> poison, <2 x i32> <i32 1, i32 0>
935  ret void
936}
937
938define void @manyundefs2()
939; CHECK-LABEL: @manyundefs2(
940; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x float> undef, float undef, i64 1
941; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x float> undef, float undef, i64 1
942; CHECK-NEXT:    [[TMP3:%.*]] = fadd <2 x float> [[TMP1]], [[TMP2]]
943; CHECK-NEXT:    [[TMP4:%.*]] = fmul <2 x float> [[TMP1]], [[TMP2]]
944; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> [[TMP4]], <2 x i32> <i32 0, i32 3>
945; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> [[TMP3]], <2 x i32> <i32 1, i32 2>
946; CHECK-NEXT:    ret void
947;
948unnamed_addr #0 align 2 {
949  %1 = insertelement <2 x float> undef, float undef, i64 1
950  %2 = insertelement <2 x float> undef, float undef, i64 1
951  %3 = fadd <2 x float> %1, %2
952  %4 = fmul <2 x float> %1, %2
953  %5 = shufflevector <2 x float> %3, <2 x float> %4, <2 x i32> <i32 0, i32 3>
954  %6 = shufflevector <2 x float> %4, <2 x float> %3, <2 x i32> <i32 1, i32 2>
955  ret void
956}
957
958define <16 x i32> @testoutofbounds(<16 x i32> %x, <16 x i32> %y) {
959; CHECK-LABEL: @testoutofbounds(
960; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
961; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
962; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
963; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
964; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
965; CHECK-NEXT:    [[S4:%.*]] = shufflevector <16 x i32> [[S3]], <16 x i32> poison, <16 x i32> <i32 16, i32 1, i32 18, i32 3, i32 20, i32 5, i32 22, i32 7, i32 24, i32 9, i32 26, i32 11, i32 28, i32 13, i32 30, i32 15>
966; CHECK-NEXT:    [[ADD:%.*]] = add <16 x i32> [[S3]], [[S4]]
967; CHECK-NEXT:    ret <16 x i32> [[ADD]]
968;
969  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
970  %s2 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
971  %a = add nsw <16 x i32> %s1, %s2
972  %b = sub nsw <16 x i32> %s1, %s2
973  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
974  %s4 = shufflevector <16 x i32> %s3, <16 x i32> poison, <16 x i32> <i32 16, i32 1, i32 18, i32 3, i32 20, i32 5, i32 22, i32 7, i32 24, i32 9, i32 26, i32 11, i32 28, i32 13, i32 30, i32 15>
975  %add = add <16 x i32> %s3, %s4
976  ret <16 x i32> %add
977}
978
979declare i32 @llvm.vector.reduce.add.v16i32(<16 x i32>)
980