xref: /llvm-project/llvm/test/Transforms/VectorCombine/AArch64/select-shuffle.ll (revision 29c89d7265329c530a6a2cb0307218004cc2ba53) 
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"
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:    [[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>
26; 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>
27; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
28; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
29; 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>
30; CHECK-NEXT:    [[R:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[S3]])
31; CHECK-NEXT:    ret i32 [[R]]
32;
33  %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>
34  %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>
35  %a = add nsw <16 x i32> %s1, %s2
36  %b = sub nsw <16 x i32> %s1, %s2
37  %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>
38  %r = call i32 @llvm.vector.reduce.add(<16 x i32> %s3)
39  ret i32 %r
40}
41declare i32 @llvm.vector.reduce.add(<16 x i32>)
42
43define <16 x i32> @test1more(<16 x i32> %a1, <16 x i32> %a2) {
44; CHECK-LABEL: @test1more(
45; CHECK-NEXT:    [[X:%.*]] = add nsw <16 x i32> [[A1:%.*]], [[A2:%.*]]
46; CHECK-NEXT:    [[Y:%.*]] = sub nsw <16 x i32> [[A1]], [[A2]]
47; 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>
48; 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>
49; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
50; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
51; 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>
52; CHECK-NEXT:    ret <16 x i32> [[S3]]
53;
54  %x = add nsw <16 x i32> %a1, %a2
55  %y = sub nsw <16 x i32> %a1, %a2
56  %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>
57  %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>
58  %a = add nsw <16 x i32> %s1, %s2
59  %b = sub nsw <16 x i32> %s1, %s2
60  %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>
61  ret <16 x i32> %s3
62}
63
64define <16 x i32> @test1mores(<16 x i32> %a1, <16 x i32> %a2) {
65; CHECK-LABEL: @test1mores(
66; 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>
67; 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>
68; CHECK-NEXT:    [[X:%.*]] = add nsw <16 x i32> [[S3]], [[S4]]
69; CHECK-NEXT:    [[Y:%.*]] = sub nsw <16 x i32> [[S3]], [[S4]]
70; 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>
71; 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>
72; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
73; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
74; 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>
75; CHECK-NEXT:    ret <16 x i32> [[S5]]
76;
77  %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>
78  %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>
79  %x = add nsw <16 x i32> %s3, %s4
80  %y = sub nsw <16 x i32> %s3, %s4
81  %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>
82  %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>
83  %a = add nsw <16 x i32> %s1, %s2
84  %b = sub nsw <16 x i32> %s1, %s2
85  %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>
86  ret <16 x i32> %s5
87}
88
89
90
91define <16 x i32> @test2_1(<16 x i32> %x, <16 x i32> %y) {
92; CHECK-LABEL: @test2_1(
93; 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>
94; 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>
95; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
96; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
97; 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>
98; CHECK-NEXT:    ret <16 x i32> [[S3]]
99;
100  %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>
101  %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>
102  %a = add nsw <16 x i32> %s1, %s2
103  %b = sub nsw <16 x i32> %s1, %s2
104  %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>
105  ret <16 x i32> %s3
106}
107
108define <16 x i32> @test2_1_ins(<16 x i32> %x1, <16 x i32> %x2) {
109; CHECK-LABEL: @test2_1_ins(
110; CHECK-NEXT:    [[X:%.*]] = add nsw <16 x i32> [[X1:%.*]], [[X2:%.*]]
111; CHECK-NEXT:    [[Y:%.*]] = sub nsw <16 x i32> [[X1]], [[X2]]
112; 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>
113; 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>
114; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
115; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
116; 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>
117; CHECK-NEXT:    ret <16 x i32> [[S3]]
118;
119  %x = add nsw <16 x i32> %x1, %x2
120  %y = sub nsw <16 x i32> %x1, %x2
121  %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>
122  %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>
123  %a = add nsw <16 x i32> %s1, %s2
124  %b = sub nsw <16 x i32> %s1, %s2
125  %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>
126  ret <16 x i32> %s3
127}
128
129define <16 x i32> @test2_2(<16 x i32> %x, <16 x i32> %y) {
130; CHECK-LABEL: @test2_2(
131; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
132; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
133; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
134; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
135; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
136; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
137; 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>
138; CHECK-NEXT:    ret <16 x i32> [[S3]]
139;
140  %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>
141  %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>
142  %a = add nsw <16 x i32> %s1, %s2
143  %b = sub nsw <16 x i32> %s1, %s2
144  %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>
145  ret <16 x i32> %s3
146}
147
148define <16 x i32> @test2_12(<16 x i32> %x, <16 x i32> %y) {
149; CHECK-LABEL: @test2_12(
150; 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>
151; 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>
152; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
153; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
154; 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>
155; 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>
156; CHECK-NEXT:    [[C:%.*]] = mul <16 x i32> [[S3]], [[S4]]
157; CHECK-NEXT:    ret <16 x i32> [[C]]
158;
159  %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>
160  %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>
161  %a = add nsw <16 x i32> %s1, %s2
162  %b = sub nsw <16 x i32> %s1, %s2
163  %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>
164  %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>
165  %c = mul <16 x i32> %s3, %s4
166  ret <16 x i32> %c
167}
168
169define <16 x i32> @test3_1(<16 x i32> %x, <16 x i32> %y) {
170; CHECK-LABEL: @test3_1(
171; 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>
172; 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>
173; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
174; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
175; 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>
176; CHECK-NEXT:    ret <16 x i32> [[S3]]
177;
178  %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>
179  %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>
180  %a = add nsw <16 x i32> %s1, %s2
181  %b = sub nsw <16 x i32> %s1, %s2
182  %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>
183  ret <16 x i32> %s3
184}
185
186define <16 x i32> @test3_2(<16 x i32> %x, <16 x i32> %y) {
187; CHECK-LABEL: @test3_2(
188; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
189; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
190; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
191; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
192; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
193; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
194; 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>
195; CHECK-NEXT:    ret <16 x i32> [[S3]]
196;
197  %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>
198  %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>
199  %a = add nsw <16 x i32> %s1, %s2
200  %b = sub nsw <16 x i32> %s1, %s2
201  %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>
202  ret <16 x i32> %s3
203}
204
205define <16 x i32> @test3_12(<16 x i32> %x, <16 x i32> %y) {
206; CHECK-LABEL: @test3_12(
207; 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>
208; 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>
209; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
210; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
211; 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>
212; 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>
213; CHECK-NEXT:    [[C:%.*]] = add <16 x i32> [[S3]], [[S4]]
214; CHECK-NEXT:    ret <16 x i32> [[C]]
215;
216  %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>
217  %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>
218  %a = add nsw <16 x i32> %s1, %s2
219  %b = sub nsw <16 x i32> %s1, %s2
220  %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>
221  %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>
222  %c = add <16 x i32> %s3, %s4
223  ret <16 x i32> %c
224}
225
226
227define <16 x i32> @test23(<16 x i32> %x, <16 x i32> %y) {
228; CHECK-LABEL: @test23(
229; 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>
230; 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>
231; CHECK-NEXT:    [[A0:%.*]] = add nsw <16 x i32> [[S10]], [[S20]]
232; CHECK-NEXT:    [[B0:%.*]] = sub nsw <16 x i32> [[S10]], [[S20]]
233; 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>
234; 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>
235; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
236; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
237; 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>
238; 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>
239; CHECK-NEXT:    [[C:%.*]] = add <16 x i32> [[S3]], [[S4]]
240; CHECK-NEXT:    ret <16 x i32> [[C]]
241;
242  %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>
243  %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>
244  %a0 = add nsw <16 x i32> %s10, %s20
245  %b0 = sub nsw <16 x i32> %s10, %s20
246  %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>
247  %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>
248  %a = add nsw <16 x i32> %s1, %s2
249  %b = sub nsw <16 x i32> %s1, %s2
250  %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>
251  %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>
252  %c = add <16 x i32> %s3, %s4
253  ret <16 x i32> %c
254}
255
256define <16 x i32> @testgood(<16 x i32> %x, <16 x i32> %y) {
257; CHECK-LABEL: @testgood(
258; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
259; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
260; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
261; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
262; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
263; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
264; 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>
265; CHECK-NEXT:    ret <16 x i32> [[S3]]
266;
267  %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>
268  %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>
269  %a = add nsw <16 x i32> %s1, %s2
270  %b = sub nsw <16 x i32> %s1, %s2
271  %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>
272  ret <16 x i32> %s3
273}
274
275define <16 x i32> @test_shufshufin(<16 x i32> %x, <16 x i32> %y) {
276; CHECK-LABEL: @test_shufshufin(
277; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
278; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
279; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
280; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
281; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
282; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
283; 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>
284; CHECK-NEXT:    ret <16 x i32> [[S3]]
285;
286  %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>
287  %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>
288  %a = add nsw <16 x i32> %s1, %s2
289  %b = sub nsw <16 x i32> %s1, %s2
290  %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>
291  ret <16 x i32> %s3
292}
293
294define <16 x i32> @testshufshufout(<16 x i32> %x, <16 x i32> %y) {
295; CHECK-LABEL: @testshufshufout(
296; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
297; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
298; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
299; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
300; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
301; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
302; 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>
303; 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>
304; CHECK-NEXT:    [[R:%.*]] = add nsw <16 x i32> [[S3]], [[S4]]
305; CHECK-NEXT:    ret <16 x i32> [[R]]
306;
307  %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>
308  %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>
309  %a = add nsw <16 x i32> %s1, %s2
310  %b = sub nsw <16 x i32> %s1, %s2
311  %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>
312  %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>
313  %r = add nsw <16 x i32> %s3, %s4
314  ret <16 x i32> %r
315}
316
317define <16 x i32> @testtwoshufout(<16 x i32> %x, <16 x i32> %y) {
318; CHECK-LABEL: @testtwoshufout(
319; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
320; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
321; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
322; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
323; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
324; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
325; 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>
326; 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>
327; CHECK-NEXT:    [[ADD:%.*]] = add <16 x i32> [[S3]], [[S4]]
328; CHECK-NEXT:    ret <16 x i32> [[ADD]]
329;
330  %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>
331  %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>
332  %a = add nsw <16 x i32> %s1, %s2
333  %b = sub nsw <16 x i32> %s1, %s2
334  %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>
335  %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>
336  %add = add <16 x i32> %s3, %s4
337  ret <16 x i32> %add
338}
339
340declare void @use(<16 x i32>)
341define <16 x i32> @test_extrashuffleuse(<16 x i32> %x, <16 x i32> %y) {
342; CHECK-LABEL: @test_extrashuffleuse(
343; 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>
344; 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>
345; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
346; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
347; 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>
348; CHECK-NEXT:    call void @use(<16 x i32> [[S1]])
349; CHECK-NEXT:    ret <16 x i32> [[S3]]
350;
351  %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>
352  %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>
353  %a = add nsw <16 x i32> %s1, %s2
354  %b = sub nsw <16 x i32> %s1, %s2
355  %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>
356  call void @use(<16 x i32> %s1)
357  ret <16 x i32> %s3
358}
359
360define <16 x i32> @test_extraopuse(<16 x i32> %x, <16 x i32> %y) {
361; CHECK-LABEL: @test_extraopuse(
362; 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>
363; 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>
364; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
365; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
366; 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>
367; CHECK-NEXT:    call void @use(<16 x i32> [[B]])
368; CHECK-NEXT:    ret <16 x i32> [[S3]]
369;
370  %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>
371  %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>
372  %a = add nsw <16 x i32> %s1, %s2
373  %b = sub nsw <16 x i32> %s1, %s2
374  %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>
375  call void @use(<16 x i32> %b)
376  ret <16 x i32> %s3
377}
378
379define <4 x i32> @test_recurse(<4 x i32> %l0, <4 x i32> %l1, <4 x i32> %l3) {
380; CHECK-LABEL: @test_recurse(
381; CHECK-NEXT:    [[S0:%.*]] = shufflevector <4 x i32> [[L3:%.*]], <4 x i32> [[L1:%.*]], <4 x i32> <i32 0, i32 2, i32 1, i32 5>
382; CHECK-NEXT:    [[S1:%.*]] = shufflevector <4 x i32> [[L1]], <4 x i32> [[L3]], <4 x i32> <i32 6, i32 1, i32 3, i32 0>
383; CHECK-NEXT:    [[S2:%.*]] = shufflevector <4 x i32> [[L0:%.*]], <4 x i32> [[L1]], <4 x i32> <i32 4, i32 2, i32 6, i32 2>
384; CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[S2]], [[S2]]
385; CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[S1]], [[S0]]
386; CHECK-NEXT:    [[T0:%.*]] = shufflevector <4 x i32> [[ADD]], <4 x i32> [[ADD]], <4 x i32> <i32 2, i32 0, i32 0, i32 3>
387; CHECK-NEXT:    ret <4 x i32> [[T0]]
388;
389  %s0 = shufflevector <4 x i32> %l3, <4 x i32> %l1, <4 x i32> <i32 0, i32 2, i32 1, i32 5>
390  %s1 = shufflevector <4 x i32> %l1, <4 x i32> %l3, <4 x i32> <i32 6, i32 1, i32 3, i32 0>
391  %s2 = shufflevector <4 x i32> %l0, <4 x i32> %l1, <4 x i32> <i32 4, i32 2, i32 6, i32 2>
392  %add = add <4 x i32> %s2, %s2
393  %sub = sub <4 x i32> %s1, %s0
394  %t0 = shufflevector <4 x i32> %add, <4 x i32> %add, <4 x i32> <i32 2, i32 0, i32 0, i32 3>
395  ret <4 x i32> %t0
396}
397
398define void @test_31(ptr %src, ptr %dst) {
399; CHECK-LABEL: @test_31(
400; CHECK-NEXT:    [[G0:%.*]] = getelementptr <4 x i32>, ptr [[SRC:%.*]], i32 0
401; CHECK-NEXT:    [[L0:%.*]] = load <4 x i32>, ptr [[G0]], align 16
402; CHECK-NEXT:    [[S0:%.*]] = shufflevector <4 x i32> [[L0]], <4 x i32> [[L0]], <4 x i32> <i32 0, i32 2, i32 1, i32 3>
403; CHECK-NEXT:    [[S1:%.*]] = shufflevector <4 x i32> [[L0]], <4 x i32> [[L0]], <4 x i32> <i32 0, i32 0, i32 2, i32 3>
404; CHECK-NEXT:    [[S2:%.*]] = shufflevector <4 x i32> [[L0]], <4 x i32> [[L0]], <4 x i32> <i32 3, i32 1, i32 1, i32 0>
405; CHECK-NEXT:    [[S3:%.*]] = shufflevector <4 x i32> [[L0]], <4 x i32> [[L0]], <4 x i32> <i32 3, i32 3, i32 5, i32 6>
406; CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[S1]], [[S1]]
407; CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[S3]], [[S0]]
408; CHECK-NEXT:    [[T0:%.*]] = shufflevector <4 x i32> [[ADD]], <4 x i32> [[ADD]], <4 x i32> <i32 0, i32 2, i32 5, i32 2>
409; CHECK-NEXT:    [[H0:%.*]] = getelementptr <4 x i32>, ptr [[DST:%.*]], i32 0
410; CHECK-NEXT:    store <4 x i32> [[T0]], ptr [[H0]], align 16
411; CHECK-NEXT:    [[T1:%.*]] = shufflevector <4 x i32> [[ADD]], <4 x i32> [[ADD]], <4 x i32> <i32 6, i32 0, i32 1, i32 7>
412; CHECK-NEXT:    [[H1:%.*]] = getelementptr <4 x i32>, ptr [[DST]], i32 1
413; CHECK-NEXT:    store <4 x i32> [[T1]], ptr [[H1]], align 16
414; CHECK-NEXT:    ret void
415;
416  %g0 = getelementptr <4 x i32>, ptr %src, i32 0
417  %l0 = load <4 x i32>, ptr %g0, align 16
418  %s0 = shufflevector <4 x i32> %l0, <4 x i32> %l0, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
419  %s1 = shufflevector <4 x i32> %l0, <4 x i32> %l0, <4 x i32> <i32 0, i32 0, i32 2, i32 3>
420  %s2 = shufflevector <4 x i32> %l0, <4 x i32> %l0, <4 x i32> <i32 3, i32 1, i32 1, i32 0>
421  %s3 = shufflevector <4 x i32> %l0, <4 x i32> %l0, <4 x i32> <i32 3, i32 3, i32 5, i32 6>
422  %add = add <4 x i32> %s1, %s1
423  %sub = sub <4 x i32> %s3, %s0
424  %t0 = shufflevector <4 x i32> %add, <4 x i32> %add, <4 x i32> <i32 0, i32 2, i32 5, i32 2>
425  %h0 = getelementptr <4 x i32>, ptr %dst, i32 0
426  store <4 x i32> %t0, ptr %h0, align 16
427  %t1 = shufflevector <4 x i32> %add, <4 x i32> %add, <4 x i32> <i32 6, i32 0, i32 1, i32 7>
428  %h1 = getelementptr <4 x i32>, ptr %dst, i32 1
429  store <4 x i32> %t1, ptr %h1, align 16
430  ret void
431}
432
433define <16 x i32> @test_1651256324(<16 x i32> %l0, <16 x i32> %l1, <16 x i32> %l6, <16 x i32> %l7) {
434; CHECK-LABEL: @test_1651256324(
435; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
436; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison>
437; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison>
438; 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>
439; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
440; CHECK-NEXT:    [[TMP5:%.*]] = add <16 x i32> [[TMP3]], [[TMP2]]
441; CHECK-NEXT:    [[TMP6:%.*]] = sub <16 x i32> [[TMP1]], [[TMP4]]
442; 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>
443; 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>
444; CHECK-NEXT:    [[R:%.*]] = xor <16 x i32> [[T0]], [[T1]]
445; CHECK-NEXT:    ret <16 x i32> [[R]]
446;
447  %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>
448  %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>
449  %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>
450  %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>
451  %add = add <16 x i32> %s1, %s1
452  %sub = sub <16 x i32> %s0, %s3
453  %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>
454  %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>
455  %r = xor <16 x i32> %t0, %t1
456  ret <16 x i32> %r
457}
458
459define void @test_1652048214(ptr %src, ptr %dst) {
460; CHECK-LABEL: @test_1652048214(
461; CHECK-NEXT:    [[G0:%.*]] = getelementptr <8 x i32>, ptr [[SRC:%.*]], i32 0
462; CHECK-NEXT:    [[L0:%.*]] = load <8 x i32>, ptr [[G0]], align 32
463; CHECK-NEXT:    [[G1:%.*]] = getelementptr <8 x i32>, ptr [[SRC]], i32 1
464; CHECK-NEXT:    [[L1:%.*]] = load <8 x i32>, ptr [[G1]], align 32
465; CHECK-NEXT:    [[S0:%.*]] = shufflevector <8 x i32> [[L1]], <8 x i32> [[L0]], <4 x i32> <i32 6, i32 2, i32 14, i32 6>
466; CHECK-NEXT:    [[S1:%.*]] = shufflevector <8 x i32> [[L0]], <8 x i32> [[L0]], <4 x i32> <i32 9, i32 5, i32 1, i32 0>
467; CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[S1]], [[S1]]
468; CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[S0]], [[S1]]
469; CHECK-NEXT:    [[T0:%.*]] = shufflevector <4 x i32> [[SUB]], <4 x i32> [[ADD]], <4 x i32> <i32 1, i32 5, i32 7, i32 3>
470; CHECK-NEXT:    [[H0:%.*]] = getelementptr <4 x i32>, ptr [[DST:%.*]], i32 0
471; CHECK-NEXT:    store <4 x i32> [[T0]], ptr [[H0]], align 16
472; CHECK-NEXT:    ret void
473;
474  %g0 = getelementptr <8 x i32>, ptr %src, i32 0
475  %l0 = load <8 x i32>, ptr %g0, align 32
476  %g1 = getelementptr <8 x i32>, ptr %src, i32 1
477  %l1 = load <8 x i32>, ptr %g1, align 32
478  %s0 = shufflevector <8 x i32> %l1, <8 x i32> %l0, <4 x i32> <i32 6, i32 2, i32 14, i32 6>
479  %s1 = shufflevector <8 x i32> %l0, <8 x i32> %l0, <4 x i32> <i32 9, i32 5, i32 1, i32 0>
480  %add = add <4 x i32> %s1, %s1
481  %sub = sub <4 x i32> %s0, %s1
482  %t0 = shufflevector <4 x i32> %sub, <4 x i32> %add, <4 x i32> <i32 1, i32 5, i32 7, i32 3>
483  %h0 = getelementptr <4 x i32>, ptr %dst, i32 0
484  store <4 x i32> %t0, ptr %h0, align 16
485  ret void
486}
487
488define dso_local i32 @full(ptr nocapture noundef readonly %p1, i32 noundef %st1, ptr nocapture noundef readonly %p2, i32 noundef %st2) {
489; CHECK-LABEL: @full(
490; CHECK-NEXT:  entry:
491; CHECK-NEXT:    [[IDX_EXT:%.*]] = sext i32 [[ST1:%.*]] to i64
492; CHECK-NEXT:    [[IDX_EXT63:%.*]] = sext i32 [[ST2:%.*]] to i64
493; CHECK-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr inbounds i8, ptr [[P1:%.*]], i64 4
494; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i8, ptr [[P2:%.*]], i64 4
495; CHECK-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds i8, ptr [[P1]], i64 [[IDX_EXT]]
496; CHECK-NEXT:    [[ADD_PTR64:%.*]] = getelementptr inbounds i8, ptr [[P2]], i64 [[IDX_EXT63]]
497; CHECK-NEXT:    [[ARRAYIDX3_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR]], i64 4
498; CHECK-NEXT:    [[ARRAYIDX5_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64]], i64 4
499; CHECK-NEXT:    [[ADD_PTR_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR]], i64 [[IDX_EXT]]
500; CHECK-NEXT:    [[ADD_PTR64_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64]], i64 [[IDX_EXT63]]
501; CHECK-NEXT:    [[ARRAYIDX3_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR_1]], i64 4
502; CHECK-NEXT:    [[ARRAYIDX5_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64_1]], i64 4
503; CHECK-NEXT:    [[ADD_PTR_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR_1]], i64 [[IDX_EXT]]
504; CHECK-NEXT:    [[ADD_PTR64_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64_1]], i64 [[IDX_EXT63]]
505; CHECK-NEXT:    [[ARRAYIDX3_3:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR_2]], i64 4
506; CHECK-NEXT:    [[ARRAYIDX5_3:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64_2]], i64 4
507; CHECK-NEXT:    [[TMP0:%.*]] = bitcast ptr [[P1]] to ptr
508; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i8>, ptr [[TMP0]], align 1
509; CHECK-NEXT:    [[TMP2:%.*]] = bitcast ptr [[P2]] to ptr
510; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i8>, ptr [[TMP2]], align 1
511; CHECK-NEXT:    [[TMP4:%.*]] = bitcast ptr [[ARRAYIDX3]] to ptr
512; CHECK-NEXT:    [[TMP5:%.*]] = load <4 x i8>, ptr [[TMP4]], align 1
513; CHECK-NEXT:    [[TMP6:%.*]] = bitcast ptr [[ARRAYIDX5]] to ptr
514; CHECK-NEXT:    [[TMP7:%.*]] = load <4 x i8>, ptr [[TMP6]], align 1
515; CHECK-NEXT:    [[TMP8:%.*]] = bitcast ptr [[ADD_PTR]] to ptr
516; CHECK-NEXT:    [[TMP9:%.*]] = load <4 x i8>, ptr [[TMP8]], align 1
517; CHECK-NEXT:    [[TMP10:%.*]] = bitcast ptr [[ADD_PTR64]] to ptr
518; CHECK-NEXT:    [[TMP11:%.*]] = load <4 x i8>, ptr [[TMP10]], align 1
519; CHECK-NEXT:    [[TMP12:%.*]] = bitcast ptr [[ARRAYIDX3_1]] to ptr
520; CHECK-NEXT:    [[TMP13:%.*]] = load <4 x i8>, ptr [[TMP12]], align 1
521; CHECK-NEXT:    [[TMP14:%.*]] = bitcast ptr [[ARRAYIDX5_1]] to ptr
522; CHECK-NEXT:    [[TMP15:%.*]] = load <4 x i8>, ptr [[TMP14]], align 1
523; CHECK-NEXT:    [[TMP16:%.*]] = bitcast ptr [[ADD_PTR_1]] to ptr
524; CHECK-NEXT:    [[TMP17:%.*]] = load <4 x i8>, ptr [[TMP16]], align 1
525; CHECK-NEXT:    [[TMP18:%.*]] = bitcast ptr [[ADD_PTR64_1]] to ptr
526; CHECK-NEXT:    [[TMP19:%.*]] = load <4 x i8>, ptr [[TMP18]], align 1
527; CHECK-NEXT:    [[TMP20:%.*]] = bitcast ptr [[ARRAYIDX3_2]] to ptr
528; CHECK-NEXT:    [[TMP21:%.*]] = load <4 x i8>, ptr [[TMP20]], align 1
529; CHECK-NEXT:    [[TMP22:%.*]] = bitcast ptr [[ARRAYIDX5_2]] to ptr
530; CHECK-NEXT:    [[TMP23:%.*]] = load <4 x i8>, ptr [[TMP22]], align 1
531; CHECK-NEXT:    [[TMP24:%.*]] = bitcast ptr [[ADD_PTR_2]] to ptr
532; CHECK-NEXT:    [[TMP25:%.*]] = load <4 x i8>, ptr [[TMP24]], align 1
533; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
534; CHECK-NEXT:    [[TMP27:%.*]] = shufflevector <4 x i8> [[TMP9]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
535; 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 poison, i32 poison, i32 poison, i32 poison>
536; CHECK-NEXT:    [[TMP29:%.*]] = shufflevector <4 x i8> [[TMP1]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
537; 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>
538; CHECK-NEXT:    [[TMP31:%.*]] = zext <16 x i8> [[TMP30]] to <16 x i32>
539; CHECK-NEXT:    [[TMP32:%.*]] = bitcast ptr [[ADD_PTR64_2]] to ptr
540; CHECK-NEXT:    [[TMP33:%.*]] = load <4 x i8>, ptr [[TMP32]], align 1
541; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
542; CHECK-NEXT:    [[TMP35:%.*]] = shufflevector <4 x i8> [[TMP11]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
543; 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 poison, i32 poison, i32 poison, i32 poison>
544; CHECK-NEXT:    [[TMP37:%.*]] = shufflevector <4 x i8> [[TMP3]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
545; 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>
546; CHECK-NEXT:    [[TMP39:%.*]] = zext <16 x i8> [[TMP38]] to <16 x i32>
547; CHECK-NEXT:    [[TMP40:%.*]] = sub nsw <16 x i32> [[TMP31]], [[TMP39]]
548; CHECK-NEXT:    [[TMP41:%.*]] = bitcast ptr [[ARRAYIDX3_3]] to ptr
549; CHECK-NEXT:    [[TMP42:%.*]] = load <4 x i8>, ptr [[TMP41]], align 1
550; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
551; CHECK-NEXT:    [[TMP44:%.*]] = shufflevector <4 x i8> [[TMP13]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
552; 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 poison, i32 poison, i32 poison, i32 poison>
553; CHECK-NEXT:    [[TMP46:%.*]] = shufflevector <4 x i8> [[TMP5]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
554; 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>
555; CHECK-NEXT:    [[TMP48:%.*]] = zext <16 x i8> [[TMP47]] to <16 x i32>
556; CHECK-NEXT:    [[TMP49:%.*]] = bitcast ptr [[ARRAYIDX5_3]] to ptr
557; CHECK-NEXT:    [[TMP50:%.*]] = load <4 x i8>, ptr [[TMP49]], align 1
558; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
559; CHECK-NEXT:    [[TMP52:%.*]] = shufflevector <4 x i8> [[TMP15]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
560; 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 poison, i32 poison, i32 poison, i32 poison>
561; CHECK-NEXT:    [[TMP54:%.*]] = shufflevector <4 x i8> [[TMP7]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
562; 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>
563; CHECK-NEXT:    [[TMP56:%.*]] = zext <16 x i8> [[TMP55]] to <16 x i32>
564; CHECK-NEXT:    [[TMP57:%.*]] = sub nsw <16 x i32> [[TMP48]], [[TMP56]]
565; CHECK-NEXT:    [[TMP58:%.*]] = shl nsw <16 x i32> [[TMP57]], splat (i32 16)
566; CHECK-NEXT:    [[TMP59:%.*]] = add nsw <16 x i32> [[TMP58]], [[TMP40]]
567; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
568; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
569; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
570; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
571; CHECK-NEXT:    [[TMP64:%.*]] = add nsw <16 x i32> [[TMP61]], [[TMP63]]
572; CHECK-NEXT:    [[TMP65:%.*]] = sub nsw <16 x i32> [[TMP60]], [[TMP62]]
573; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
574; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
575; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
576; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
577; CHECK-NEXT:    [[TMP70:%.*]] = add nsw <16 x i32> [[TMP67]], [[TMP69]]
578; CHECK-NEXT:    [[TMP71:%.*]] = sub nsw <16 x i32> [[TMP66]], [[TMP68]]
579; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
580; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
581; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
582; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
583; CHECK-NEXT:    [[TMP76:%.*]] = add nsw <16 x i32> [[TMP73]], [[TMP75]]
584; CHECK-NEXT:    [[TMP77:%.*]] = sub nsw <16 x i32> [[TMP72]], [[TMP74]]
585; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
586; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
587; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
588; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
589; CHECK-NEXT:    [[TMP82:%.*]] = add nsw <16 x i32> [[TMP79]], [[TMP81]]
590; CHECK-NEXT:    [[TMP83:%.*]] = sub nsw <16 x i32> [[TMP78]], [[TMP80]]
591; 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>
592; CHECK-NEXT:    [[TMP85:%.*]] = lshr <16 x i32> [[TMP84]], splat (i32 15)
593; CHECK-NEXT:    [[TMP86:%.*]] = and <16 x i32> [[TMP85]], splat (i32 65537)
594; CHECK-NEXT:    [[TMP87:%.*]] = mul nuw <16 x i32> [[TMP86]], splat (i32 65535)
595; CHECK-NEXT:    [[TMP88:%.*]] = add <16 x i32> [[TMP87]], [[TMP84]]
596; CHECK-NEXT:    [[TMP89:%.*]] = xor <16 x i32> [[TMP88]], [[TMP87]]
597; CHECK-NEXT:    [[TMP90:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP89]])
598; CHECK-NEXT:    [[CONV118:%.*]] = and i32 [[TMP90]], 65535
599; CHECK-NEXT:    [[SHR:%.*]] = lshr i32 [[TMP90]], 16
600; CHECK-NEXT:    [[ADD119:%.*]] = add nuw nsw i32 [[CONV118]], [[SHR]]
601; CHECK-NEXT:    [[SHR120:%.*]] = lshr i32 [[ADD119]], 1
602; CHECK-NEXT:    ret i32 [[SHR120]]
603;
604entry:
605  %idx.ext = sext i32 %st1 to i64
606  %idx.ext63 = sext i32 %st2 to i64
607  %arrayidx3 = getelementptr inbounds i8, ptr %p1, i64 4
608  %arrayidx5 = getelementptr inbounds i8, ptr %p2, i64 4
609  %add.ptr = getelementptr inbounds i8, ptr %p1, i64 %idx.ext
610  %add.ptr64 = getelementptr inbounds i8, ptr %p2, i64 %idx.ext63
611  %arrayidx3.1 = getelementptr inbounds i8, ptr %add.ptr, i64 4
612  %arrayidx5.1 = getelementptr inbounds i8, ptr %add.ptr64, i64 4
613  %add.ptr.1 = getelementptr inbounds i8, ptr %add.ptr, i64 %idx.ext
614  %add.ptr64.1 = getelementptr inbounds i8, ptr %add.ptr64, i64 %idx.ext63
615  %arrayidx3.2 = getelementptr inbounds i8, ptr %add.ptr.1, i64 4
616  %arrayidx5.2 = getelementptr inbounds i8, ptr %add.ptr64.1, i64 4
617  %add.ptr.2 = getelementptr inbounds i8, ptr %add.ptr.1, i64 %idx.ext
618  %add.ptr64.2 = getelementptr inbounds i8, ptr %add.ptr64.1, i64 %idx.ext63
619  %arrayidx3.3 = getelementptr inbounds i8, ptr %add.ptr.2, i64 4
620  %arrayidx5.3 = getelementptr inbounds i8, ptr %add.ptr64.2, i64 4
621  %0 = bitcast ptr %p1 to ptr
622  %1 = load <4 x i8>, ptr %0, align 1
623  %2 = bitcast ptr %p2 to ptr
624  %3 = load <4 x i8>, ptr %2, align 1
625  %4 = bitcast ptr %arrayidx3 to ptr
626  %5 = load <4 x i8>, ptr %4, align 1
627  %6 = bitcast ptr %arrayidx5 to ptr
628  %7 = load <4 x i8>, ptr %6, align 1
629  %8 = bitcast ptr %add.ptr to ptr
630  %9 = load <4 x i8>, ptr %8, align 1
631  %10 = bitcast ptr %add.ptr64 to ptr
632  %11 = load <4 x i8>, ptr %10, align 1
633  %12 = bitcast ptr %arrayidx3.1 to ptr
634  %13 = load <4 x i8>, ptr %12, align 1
635  %14 = bitcast ptr %arrayidx5.1 to ptr
636  %15 = load <4 x i8>, ptr %14, align 1
637  %16 = bitcast ptr %add.ptr.1 to ptr
638  %17 = load <4 x i8>, ptr %16, align 1
639  %18 = bitcast ptr %add.ptr64.1 to ptr
640  %19 = load <4 x i8>, ptr %18, align 1
641  %20 = bitcast ptr %arrayidx3.2 to ptr
642  %21 = load <4 x i8>, ptr %20, align 1
643  %22 = bitcast ptr %arrayidx5.2 to ptr
644  %23 = load <4 x i8>, ptr %22, align 1
645  %24 = bitcast ptr %add.ptr.2 to ptr
646  %25 = load <4 x i8>, ptr %24, align 1
647  %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>
648  %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>
649  %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>
650  %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>
651  %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>
652  %31 = zext <16 x i8> %30 to <16 x i32>
653  %32 = bitcast ptr %add.ptr64.2 to ptr
654  %33 = load <4 x i8>, ptr %32, align 1
655  %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>
656  %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>
657  %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>
658  %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>
659  %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>
660  %39 = zext <16 x i8> %38 to <16 x i32>
661  %40 = sub nsw <16 x i32> %31, %39
662  %41 = bitcast ptr %arrayidx3.3 to ptr
663  %42 = load <4 x i8>, ptr %41, align 1
664  %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>
665  %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>
666  %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>
667  %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>
668  %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>
669  %48 = zext <16 x i8> %47 to <16 x i32>
670  %49 = bitcast ptr %arrayidx5.3 to ptr
671  %50 = load <4 x i8>, ptr %49, align 1
672  %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>
673  %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>
674  %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>
675  %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>
676  %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>
677  %56 = zext <16 x i8> %55 to <16 x i32>
678  %57 = sub nsw <16 x i32> %48, %56
679  %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>
680  %59 = add nsw <16 x i32> %58, %40
681  %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>
682  %60 = add nsw <16 x i32> %59, %reorder
683  %61 = sub nsw <16 x i32> %59, %reorder
684  %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>
685  %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>
686  %63 = add nsw <16 x i32> %62, %reorder1
687  %64 = sub nsw <16 x i32> %62, %reorder1
688  %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>
689  %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>
690  %66 = add nsw <16 x i32> %65, %reorder2
691  %67 = sub nsw <16 x i32> %65, %reorder2
692  %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>
693  %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>
694  %69 = add nsw <16 x i32> %68, %reorder3
695  %70 = sub nsw <16 x i32> %68, %reorder3
696  %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>
697  %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>
698  %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>
699  %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>
700  %75 = add <16 x i32> %74, %71
701  %76 = xor <16 x i32> %75, %74
702  %77 = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> %76)
703  %conv118 = and i32 %77, 65535
704  %shr = lshr i32 %77, 16
705  %add119 = add nuw nsw i32 %conv118, %shr
706  %shr120 = lshr i32 %add119, 1
707  ret i32 %shr120
708}
709
710define i32 @full_reorder(ptr nocapture noundef readonly %pix1, i32 noundef %i_pix1, ptr nocapture noundef readonly %pix2, i32 noundef %i_pix2) {
711; CHECK-LABEL: @full_reorder(
712; CHECK-NEXT:  entry:
713; CHECK-NEXT:    [[IDX_EXT:%.*]] = sext i32 [[I_PIX1:%.*]] to i64
714; CHECK-NEXT:    [[IDX_EXT63:%.*]] = sext i32 [[I_PIX2:%.*]] to i64
715; CHECK-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr inbounds i8, ptr [[PIX1:%.*]], i64 4
716; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i8, ptr [[PIX2:%.*]], i64 4
717; CHECK-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds i8, ptr [[PIX1]], i64 [[IDX_EXT]]
718; CHECK-NEXT:    [[ADD_PTR64:%.*]] = getelementptr inbounds i8, ptr [[PIX2]], i64 [[IDX_EXT63]]
719; CHECK-NEXT:    [[ARRAYIDX3_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR]], i64 4
720; CHECK-NEXT:    [[ARRAYIDX5_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64]], i64 4
721; CHECK-NEXT:    [[ADD_PTR_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR]], i64 [[IDX_EXT]]
722; CHECK-NEXT:    [[ADD_PTR64_1:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64]], i64 [[IDX_EXT63]]
723; CHECK-NEXT:    [[ARRAYIDX3_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR_1]], i64 4
724; CHECK-NEXT:    [[ARRAYIDX5_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64_1]], i64 4
725; CHECK-NEXT:    [[ADD_PTR_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR_1]], i64 [[IDX_EXT]]
726; CHECK-NEXT:    [[ADD_PTR64_2:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64_1]], i64 [[IDX_EXT63]]
727; CHECK-NEXT:    [[ARRAYIDX3_3:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR_2]], i64 4
728; CHECK-NEXT:    [[ARRAYIDX5_3:%.*]] = getelementptr inbounds i8, ptr [[ADD_PTR64_2]], i64 4
729; CHECK-NEXT:    [[TMP0:%.*]] = load <4 x i8>, ptr [[PIX1]], align 1
730; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i8>, ptr [[PIX2]], align 1
731; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i8>, ptr [[ARRAYIDX3]], align 1
732; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i8>, ptr [[ARRAYIDX5]], align 1
733; CHECK-NEXT:    [[TMP4:%.*]] = load <4 x i8>, ptr [[ADD_PTR]], align 1
734; CHECK-NEXT:    [[TMP5:%.*]] = load <4 x i8>, ptr [[ADD_PTR64]], align 1
735; CHECK-NEXT:    [[TMP6:%.*]] = load <4 x i8>, ptr [[ARRAYIDX3_1]], align 1
736; CHECK-NEXT:    [[TMP7:%.*]] = load <4 x i8>, ptr [[ARRAYIDX5_1]], align 1
737; CHECK-NEXT:    [[TMP8:%.*]] = load <4 x i8>, ptr [[ADD_PTR_1]], align 1
738; CHECK-NEXT:    [[TMP9:%.*]] = load <4 x i8>, ptr [[ADD_PTR64_1]], align 1
739; CHECK-NEXT:    [[TMP10:%.*]] = load <4 x i8>, ptr [[ARRAYIDX3_2]], align 1
740; CHECK-NEXT:    [[TMP11:%.*]] = load <4 x i8>, ptr [[ARRAYIDX5_2]], align 1
741; CHECK-NEXT:    [[TMP12:%.*]] = load <4 x i8>, ptr [[ADD_PTR_2]], align 1
742; CHECK-NEXT:    [[TMP13:%.*]] = shufflevector <4 x i8> [[TMP12]], <4 x i8> [[TMP8]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
743; CHECK-NEXT:    [[TMP14:%.*]] = shufflevector <4 x i8> [[TMP4]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
744; CHECK-NEXT:    [[TMP15:%.*]] = shufflevector <16 x i8> [[TMP13]], <16 x i8> [[TMP14]], <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>
745; CHECK-NEXT:    [[TMP16:%.*]] = shufflevector <4 x i8> [[TMP0]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
746; 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 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
747; CHECK-NEXT:    [[TMP18:%.*]] = zext <16 x i8> [[TMP17]] to <16 x i32>
748; CHECK-NEXT:    [[TMP19:%.*]] = load <4 x i8>, ptr [[ADD_PTR64_2]], align 1
749; CHECK-NEXT:    [[TMP20:%.*]] = shufflevector <4 x i8> [[TMP19]], <4 x i8> [[TMP9]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
750; CHECK-NEXT:    [[TMP21:%.*]] = shufflevector <4 x i8> [[TMP5]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
751; CHECK-NEXT:    [[TMP22:%.*]] = shufflevector <16 x i8> [[TMP20]], <16 x i8> [[TMP21]], <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>
752; CHECK-NEXT:    [[TMP23:%.*]] = shufflevector <4 x i8> [[TMP1]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
753; CHECK-NEXT:    [[TMP24:%.*]] = shufflevector <16 x i8> [[TMP22]], <16 x i8> [[TMP23]], <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>
754; CHECK-NEXT:    [[TMP25:%.*]] = zext <16 x i8> [[TMP24]] to <16 x i32>
755; CHECK-NEXT:    [[TMP26:%.*]] = sub nsw <16 x i32> [[TMP18]], [[TMP25]]
756; CHECK-NEXT:    [[TMP27:%.*]] = load <4 x i8>, ptr [[ARRAYIDX3_3]], align 1
757; CHECK-NEXT:    [[TMP28:%.*]] = shufflevector <4 x i8> [[TMP27]], <4 x i8> [[TMP10]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
758; CHECK-NEXT:    [[TMP29:%.*]] = shufflevector <4 x i8> [[TMP6]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
759; 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 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
760; CHECK-NEXT:    [[TMP31:%.*]] = shufflevector <4 x i8> [[TMP2]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
761; CHECK-NEXT:    [[TMP32:%.*]] = shufflevector <16 x i8> [[TMP30]], <16 x i8> [[TMP31]], <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>
762; CHECK-NEXT:    [[TMP33:%.*]] = zext <16 x i8> [[TMP32]] to <16 x i32>
763; CHECK-NEXT:    [[TMP34:%.*]] = load <4 x i8>, ptr [[ARRAYIDX5_3]], align 1
764; CHECK-NEXT:    [[TMP35:%.*]] = shufflevector <4 x i8> [[TMP34]], <4 x i8> [[TMP11]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
765; CHECK-NEXT:    [[TMP36:%.*]] = shufflevector <4 x i8> [[TMP7]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
766; CHECK-NEXT:    [[TMP37:%.*]] = shufflevector <16 x i8> [[TMP35]], <16 x i8> [[TMP36]], <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>
767; CHECK-NEXT:    [[TMP38:%.*]] = shufflevector <4 x i8> [[TMP3]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
768; CHECK-NEXT:    [[TMP39:%.*]] = shufflevector <16 x i8> [[TMP37]], <16 x i8> [[TMP38]], <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>
769; CHECK-NEXT:    [[TMP40:%.*]] = zext <16 x i8> [[TMP39]] to <16 x i32>
770; CHECK-NEXT:    [[TMP41:%.*]] = sub nsw <16 x i32> [[TMP33]], [[TMP40]]
771; CHECK-NEXT:    [[TMP42:%.*]] = shl nsw <16 x i32> [[TMP41]], splat (i32 16)
772; CHECK-NEXT:    [[TMP43:%.*]] = add nsw <16 x i32> [[TMP42]], [[TMP26]]
773; CHECK-NEXT:    [[TMP44:%.*]] = shufflevector <16 x i32> [[TMP43]], <16 x i32> [[TMP43]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
774; CHECK-NEXT:    [[TMP45:%.*]] = shufflevector <16 x i32> [[TMP43]], <16 x i32> [[TMP43]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
775; CHECK-NEXT:    [[TMP46:%.*]] = shufflevector <16 x i32> [[TMP43]], <16 x i32> poison, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
776; CHECK-NEXT:    [[TMP47:%.*]] = shufflevector <16 x i32> [[TMP43]], <16 x i32> poison, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
777; CHECK-NEXT:    [[TMP48:%.*]] = add nsw <16 x i32> [[TMP45]], [[TMP47]]
778; CHECK-NEXT:    [[TMP49:%.*]] = sub nsw <16 x i32> [[TMP44]], [[TMP46]]
779; CHECK-NEXT:    [[TMP50:%.*]] = shufflevector <16 x i32> [[TMP48]], <16 x i32> [[TMP49]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
780; CHECK-NEXT:    [[TMP51:%.*]] = shufflevector <16 x i32> [[TMP48]], <16 x i32> [[TMP49]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
781; CHECK-NEXT:    [[TMP52:%.*]] = shufflevector <16 x i32> [[TMP48]], <16 x i32> [[TMP49]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
782; CHECK-NEXT:    [[TMP53:%.*]] = shufflevector <16 x i32> [[TMP48]], <16 x i32> [[TMP49]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
783; CHECK-NEXT:    [[TMP54:%.*]] = add nsw <16 x i32> [[TMP51]], [[TMP53]]
784; CHECK-NEXT:    [[TMP55:%.*]] = sub nsw <16 x i32> [[TMP50]], [[TMP52]]
785; CHECK-NEXT:    [[TMP56:%.*]] = shufflevector <16 x i32> [[TMP54]], <16 x i32> [[TMP55]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
786; CHECK-NEXT:    [[TMP57:%.*]] = shufflevector <16 x i32> [[TMP54]], <16 x i32> [[TMP55]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
787; CHECK-NEXT:    [[TMP58:%.*]] = shufflevector <16 x i32> [[TMP54]], <16 x i32> [[TMP55]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
788; CHECK-NEXT:    [[TMP59:%.*]] = shufflevector <16 x i32> [[TMP54]], <16 x i32> [[TMP55]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
789; CHECK-NEXT:    [[TMP60:%.*]] = add nsw <16 x i32> [[TMP57]], [[TMP59]]
790; CHECK-NEXT:    [[TMP61:%.*]] = sub nsw <16 x i32> [[TMP56]], [[TMP58]]
791; CHECK-NEXT:    [[TMP62:%.*]] = shufflevector <16 x i32> [[TMP60]], <16 x i32> [[TMP61]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
792; CHECK-NEXT:    [[TMP63:%.*]] = shufflevector <16 x i32> [[TMP60]], <16 x i32> [[TMP61]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
793; CHECK-NEXT:    [[TMP64:%.*]] = shufflevector <16 x i32> [[TMP60]], <16 x i32> [[TMP61]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
794; CHECK-NEXT:    [[TMP65:%.*]] = shufflevector <16 x i32> [[TMP60]], <16 x i32> [[TMP61]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
795; CHECK-NEXT:    [[TMP66:%.*]] = add nsw <16 x i32> [[TMP63]], [[TMP65]]
796; CHECK-NEXT:    [[TMP67:%.*]] = sub nsw <16 x i32> [[TMP62]], [[TMP64]]
797; CHECK-NEXT:    [[TMP68:%.*]] = shufflevector <16 x i32> [[TMP66]], <16 x i32> [[TMP67]], <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>
798; CHECK-NEXT:    [[TMP69:%.*]] = lshr <16 x i32> [[TMP68]], splat (i32 15)
799; CHECK-NEXT:    [[TMP70:%.*]] = and <16 x i32> [[TMP69]], splat (i32 65537)
800; CHECK-NEXT:    [[TMP71:%.*]] = mul nuw <16 x i32> [[TMP70]], splat (i32 65535)
801; CHECK-NEXT:    [[TMP72:%.*]] = add <16 x i32> [[TMP71]], [[TMP68]]
802; CHECK-NEXT:    [[TMP73:%.*]] = xor <16 x i32> [[TMP72]], [[TMP71]]
803; CHECK-NEXT:    [[TMP74:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP73]])
804; CHECK-NEXT:    [[CONV118:%.*]] = and i32 [[TMP74]], 65535
805; CHECK-NEXT:    [[SHR:%.*]] = lshr i32 [[TMP74]], 16
806; CHECK-NEXT:    [[ADD119:%.*]] = add nuw nsw i32 [[CONV118]], [[SHR]]
807; CHECK-NEXT:    [[SHR120:%.*]] = lshr i32 [[ADD119]], 1
808; CHECK-NEXT:    ret i32 [[SHR120]]
809;
810entry:
811  %idx.ext = sext i32 %i_pix1 to i64
812  %idx.ext63 = sext i32 %i_pix2 to i64
813  %arrayidx3 = getelementptr inbounds i8, ptr %pix1, i64 4
814  %arrayidx5 = getelementptr inbounds i8, ptr %pix2, i64 4
815  %add.ptr = getelementptr inbounds i8, ptr %pix1, i64 %idx.ext
816  %add.ptr64 = getelementptr inbounds i8, ptr %pix2, i64 %idx.ext63
817  %arrayidx3.1 = getelementptr inbounds i8, ptr %add.ptr, i64 4
818  %arrayidx5.1 = getelementptr inbounds i8, ptr %add.ptr64, i64 4
819  %add.ptr.1 = getelementptr inbounds i8, ptr %add.ptr, i64 %idx.ext
820  %add.ptr64.1 = getelementptr inbounds i8, ptr %add.ptr64, i64 %idx.ext63
821  %arrayidx3.2 = getelementptr inbounds i8, ptr %add.ptr.1, i64 4
822  %arrayidx5.2 = getelementptr inbounds i8, ptr %add.ptr64.1, i64 4
823  %add.ptr.2 = getelementptr inbounds i8, ptr %add.ptr.1, i64 %idx.ext
824  %add.ptr64.2 = getelementptr inbounds i8, ptr %add.ptr64.1, i64 %idx.ext63
825  %arrayidx3.3 = getelementptr inbounds i8, ptr %add.ptr.2, i64 4
826  %arrayidx5.3 = getelementptr inbounds i8, ptr %add.ptr64.2, i64 4
827  %0 = load <4 x i8>, ptr %pix1, align 1
828  %1 = load <4 x i8>, ptr %pix2, align 1
829  %2 = load <4 x i8>, ptr %arrayidx3, align 1
830  %3 = load <4 x i8>, ptr %arrayidx5, align 1
831  %4 = load <4 x i8>, ptr %add.ptr, align 1
832  %5 = load <4 x i8>, ptr %add.ptr64, align 1
833  %6 = load <4 x i8>, ptr %arrayidx3.1, align 1
834  %7 = load <4 x i8>, ptr %arrayidx5.1, align 1
835  %8 = load <4 x i8>, ptr %add.ptr.1, align 1
836  %9 = load <4 x i8>, ptr %add.ptr64.1, align 1
837  %10 = load <4 x i8>, ptr %arrayidx3.2, align 1
838  %11 = load <4 x i8>, ptr %arrayidx5.2, align 1
839  %12 = load <4 x i8>, ptr %add.ptr.2, align 1
840  %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>
841  %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>
842  %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>
843  %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>
844  %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>
845  %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>
846  %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>
847  %20 = zext <16 x i8> %19 to <16 x i32>
848  %21 = load <4 x i8>, ptr %add.ptr64.2, align 1
849  %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>
850  %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>
851  %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>
852  %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>
853  %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>
854  %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>
855  %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>
856  %29 = zext <16 x i8> %28 to <16 x i32>
857  %30 = sub nsw <16 x i32> %20, %29
858  %31 = load <4 x i8>, ptr %arrayidx3.3, align 1
859  %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>
860  %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>
861  %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>
862  %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>
863  %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>
864  %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>
865  %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>
866  %39 = zext <16 x i8> %38 to <16 x i32>
867  %40 = load <4 x i8>, ptr %arrayidx5.3, align 1
868  %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>
869  %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>
870  %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>
871  %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>
872  %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>
873  %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>
874  %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>
875  %48 = zext <16 x i8> %47 to <16 x i32>
876  %49 = sub nsw <16 x i32> %39, %48
877  %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>
878  %51 = add nsw <16 x i32> %50, %30
879  %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>
880  %52 = add nsw <16 x i32> %51, %reorder
881  %53 = sub nsw <16 x i32> %51, %reorder
882  %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>
883  %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>
884  %55 = add nsw <16 x i32> %54, %reorder191
885  %56 = sub nsw <16 x i32> %54, %reorder191
886  %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>
887  %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>
888  %58 = add nsw <16 x i32> %57, %reorder192
889  %59 = sub nsw <16 x i32> %57, %reorder192
890  %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>
891  %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>
892  %61 = add nsw <16 x i32> %60, %reorder193
893  %62 = sub nsw <16 x i32> %60, %reorder193
894  %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>
895  %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>
896  %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>
897  %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>
898  %67 = add <16 x i32> %66, %63
899  %68 = xor <16 x i32> %67, %66
900  %69 = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> %68)
901  %conv118 = and i32 %69, 65535
902  %shr = lshr i32 %69, 16
903  %add119 = add nuw nsw i32 %conv118, %shr
904  %shr120 = lshr i32 %add119, 1
905  ret i32 %shr120
906}
907
908define void @manyundefs()
909; CHECK-LABEL: @manyundefs(
910; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x float> undef, float undef, i64 1
911; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x float> undef, float undef, i64 1
912; CHECK-NEXT:    [[TMP3:%.*]] = fadd <2 x float> [[TMP1]], [[TMP2]]
913; CHECK-NEXT:    [[TMP4:%.*]] = fmul <2 x float> [[TMP1]], [[TMP2]]
914; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> [[TMP4]], <2 x i32> <i32 0, i32 3>
915; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <2 x i32> <i32 1, i32 0>
916; CHECK-NEXT:    ret void
917;
918unnamed_addr #0 align 2 {
919  %1 = insertelement <2 x float> undef, float undef, i64 1
920  %2 = insertelement <2 x float> undef, float undef, i64 1
921  %3 = fadd <2 x float> %1, %2
922  %4 = fmul <2 x float> %1, %2
923  %5 = shufflevector <2 x float> %3, <2 x float> %4, <2 x i32> <i32 0, i32 3>
924  %6 = shufflevector <2 x float> %4, <2 x float> poison, <2 x i32> <i32 1, i32 0>
925  ret void
926}
927
928define void @manyundefs2()
929; CHECK-LABEL: @manyundefs2(
930; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x float> undef, float undef, i64 1
931; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x float> undef, float undef, i64 1
932; CHECK-NEXT:    [[TMP3:%.*]] = fadd <2 x float> [[TMP1]], [[TMP2]]
933; CHECK-NEXT:    [[TMP4:%.*]] = fmul <2 x float> [[TMP1]], [[TMP2]]
934; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> [[TMP4]], <2 x i32> <i32 0, i32 3>
935; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> [[TMP3]], <2 x i32> <i32 1, i32 2>
936; CHECK-NEXT:    ret void
937;
938unnamed_addr #0 align 2 {
939  %1 = insertelement <2 x float> undef, float undef, i64 1
940  %2 = insertelement <2 x float> undef, float undef, i64 1
941  %3 = fadd <2 x float> %1, %2
942  %4 = fmul <2 x float> %1, %2
943  %5 = shufflevector <2 x float> %3, <2 x float> %4, <2 x i32> <i32 0, i32 3>
944  %6 = shufflevector <2 x float> %4, <2 x float> %3, <2 x i32> <i32 1, i32 2>
945  ret void
946}
947
948define <16 x i32> @testoutofbounds(<16 x i32> %x, <16 x i32> %y) {
949; CHECK-LABEL: @testoutofbounds(
950; 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>
951; 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>
952; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
953; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
954; 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>
955; CHECK-NEXT:    [[ADD:%.*]] = add <16 x i32> [[S3]], [[B]]
956; CHECK-NEXT:    ret <16 x i32> [[ADD]]
957;
958  %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>
959  %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>
960  %a = add nsw <16 x i32> %s1, %s2
961  %b = sub nsw <16 x i32> %s1, %s2
962  %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>
963  %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>
964  %add = add <16 x i32> %s3, %s4
965  ret <16 x i32> %add
966}
967
968define <64 x i32> @testlargerextrashuffle2(i32 %call.i, <16 x i32> %0) {
969; CHECK-LABEL: @testlargerextrashuffle2(
970; CHECK-NEXT:  entry:
971; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <16 x i32> [[TMP0:%.*]], i32 [[CALL_I:%.*]], i32 15
972; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <16 x i32> [[TMP0]], i32 [[CALL_I]], i32 15
973; CHECK-NEXT:    [[TMP3:%.*]] = sub <16 x i32> [[TMP1]], [[TMP2]]
974; CHECK-NEXT:    [[TMP4:%.*]] = add <16 x i32> [[TMP1]], [[TMP2]]
975; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <16 x i32> [[TMP3]], <16 x i32> [[TMP4]], <64 x i32> <i32 0, i32 17, i32 2, i32 3, i32 20, i32 5, i32 6, i32 23, i32 8, i32 9, i32 26, i32 11, i32 12, i32 29, i32 14, i32 15, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
976; CHECK-NEXT:    ret <64 x i32> [[TMP6]]
977;
978entry:
979  %1 = insertelement <16 x i32> %0, i32 %call.i, i32 15
980  %2 = insertelement <16 x i32> %0, i32 %call.i, i32 15
981  %3 = sub <16 x i32> %1, %2
982  %4 = add <16 x i32> %1, %2
983  %5 = shufflevector <16 x i32> %3, <16 x i32> %4, <16 x i32> <i32 0, i32 17, i32 2, i32 3, i32 20, i32 5, i32 6, i32 23, i32 8, i32 9, i32 26, i32 11, i32 12, i32 29, i32 14, i32 15>
984  %6 = shufflevector <16 x i32> %5, <16 x i32> poison, <64 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, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
985  ret <64 x i32> %6
986}
987
988
989
990declare i32 @llvm.vector.reduce.add.v16i32(<16 x i32>)
991