1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -mtriple=x86_64-unknown-linux -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,DEFAULT,SSE
3; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=corei7-avx -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,DEFAULT,AVX,AVX1
4; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=core-avx2 -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,DEFAULT,AVX,AVX2
5; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=skx -slp-vectorizer -S -slp-threshold=-100 | FileCheck %s --check-prefixes=CHECK,THRESH
6
7@arr = local_unnamed_addr global [32 x i32] zeroinitializer, align 16
8@arr1 = local_unnamed_addr global [32 x float] zeroinitializer, align 16
9@arrp = local_unnamed_addr global [32 x i32*] zeroinitializer, align 16
10@var = global i32 zeroinitializer, align 8
11
12define i32 @maxi8(i32) {
13; CHECK-LABEL: @maxi8(
14; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, <8 x i32>* bitcast ([32 x i32]* @arr to <8 x i32>*), align 16
15; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP2]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
16; CHECK-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <8 x i32> [[TMP2]], [[RDX_SHUF]]
17; CHECK-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP]], <8 x i32> [[TMP2]], <8 x i32> [[RDX_SHUF]]
18; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[RDX_MINMAX_SELECT]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
19; CHECK-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <8 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
20; CHECK-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP2]], <8 x i32> [[RDX_MINMAX_SELECT]], <8 x i32> [[RDX_SHUF1]]
21; CHECK-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <8 x i32> [[RDX_MINMAX_SELECT3]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
22; CHECK-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <8 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
23; CHECK-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP5]], <8 x i32> [[RDX_MINMAX_SELECT3]], <8 x i32> [[RDX_SHUF4]]
24; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <8 x i32> [[RDX_MINMAX_SELECT6]], i32 0
25; CHECK-NEXT:    ret i32 [[TMP3]]
26;
27  %2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
28  %3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
29  %4 = icmp sgt i32 %2, %3
30  %5 = select i1 %4, i32 %2, i32 %3
31  %6 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
32  %7 = icmp sgt i32 %5, %6
33  %8 = select i1 %7, i32 %5, i32 %6
34  %9 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 3), align 4
35  %10 = icmp sgt i32 %8, %9
36  %11 = select i1 %10, i32 %8, i32 %9
37  %12 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
38  %13 = icmp sgt i32 %11, %12
39  %14 = select i1 %13, i32 %11, i32 %12
40  %15 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
41  %16 = icmp sgt i32 %14, %15
42  %17 = select i1 %16, i32 %14, i32 %15
43  %18 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
44  %19 = icmp sgt i32 %17, %18
45  %20 = select i1 %19, i32 %17, i32 %18
46  %21 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
47  %22 = icmp sgt i32 %20, %21
48  %23 = select i1 %22, i32 %20, i32 %21
49  ret i32 %23
50}
51
52define i32 @maxi16(i32) {
53; CHECK-LABEL: @maxi16(
54; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, <16 x i32>* bitcast ([32 x i32]* @arr to <16 x i32>*), align 16
55; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <16 x i32> [[TMP2]], <16 x i32> undef, <16 x i32> <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>
56; CHECK-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <16 x i32> [[TMP2]], [[RDX_SHUF]]
57; CHECK-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x i32> [[TMP2]], <16 x i32> [[RDX_SHUF]]
58; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
59; CHECK-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
60; CHECK-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> [[RDX_SHUF1]]
61; CHECK-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> undef, <16 x i32> <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, i32 undef, i32 undef>
62; CHECK-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
63; CHECK-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> [[RDX_SHUF4]]
64; CHECK-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> undef, <16 x i32> <i32 1, 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>
65; CHECK-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
66; CHECK-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> [[RDX_SHUF7]]
67; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <16 x i32> [[RDX_MINMAX_SELECT9]], i32 0
68; CHECK-NEXT:    ret i32 [[TMP3]]
69;
70  %2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
71  %3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
72  %4 = icmp sgt i32 %2, %3
73  %5 = select i1 %4, i32 %2, i32 %3
74  %6 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
75  %7 = icmp sgt i32 %5, %6
76  %8 = select i1 %7, i32 %5, i32 %6
77  %9 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 3), align 4
78  %10 = icmp sgt i32 %8, %9
79  %11 = select i1 %10, i32 %8, i32 %9
80  %12 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
81  %13 = icmp sgt i32 %11, %12
82  %14 = select i1 %13, i32 %11, i32 %12
83  %15 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
84  %16 = icmp sgt i32 %14, %15
85  %17 = select i1 %16, i32 %14, i32 %15
86  %18 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
87  %19 = icmp sgt i32 %17, %18
88  %20 = select i1 %19, i32 %17, i32 %18
89  %21 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
90  %22 = icmp sgt i32 %20, %21
91  %23 = select i1 %22, i32 %20, i32 %21
92  %24 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 8), align 16
93  %25 = icmp sgt i32 %23, %24
94  %26 = select i1 %25, i32 %23, i32 %24
95  %27 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 9), align 4
96  %28 = icmp sgt i32 %26, %27
97  %29 = select i1 %28, i32 %26, i32 %27
98  %30 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 10), align 8
99  %31 = icmp sgt i32 %29, %30
100  %32 = select i1 %31, i32 %29, i32 %30
101  %33 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 11), align 4
102  %34 = icmp sgt i32 %32, %33
103  %35 = select i1 %34, i32 %32, i32 %33
104  %36 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 12), align 16
105  %37 = icmp sgt i32 %35, %36
106  %38 = select i1 %37, i32 %35, i32 %36
107  %39 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 13), align 4
108  %40 = icmp sgt i32 %38, %39
109  %41 = select i1 %40, i32 %38, i32 %39
110  %42 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 14), align 8
111  %43 = icmp sgt i32 %41, %42
112  %44 = select i1 %43, i32 %41, i32 %42
113  %45 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 15), align 4
114  %46 = icmp sgt i32 %44, %45
115  %47 = select i1 %46, i32 %44, i32 %45
116  ret i32 %47
117}
118
119define i32 @maxi32(i32) {
120; CHECK-LABEL: @maxi32(
121; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i32>, <32 x i32>* bitcast ([32 x i32]* @arr to <32 x i32>*), align 16
122; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <32 x i32> [[TMP2]], <32 x i32> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, 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>
123; CHECK-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <32 x i32> [[TMP2]], [[RDX_SHUF]]
124; CHECK-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x i32> [[TMP2]], <32 x i32> [[RDX_SHUF]]
125; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> undef, <32 x i32> <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>
126; CHECK-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
127; CHECK-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> [[RDX_SHUF1]]
128; CHECK-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, 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>
129; CHECK-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
130; CHECK-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> [[RDX_SHUF4]]
131; CHECK-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> undef, <32 x i32> <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, 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>
132; CHECK-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
133; CHECK-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> [[RDX_SHUF7]]
134; CHECK-NEXT:    [[RDX_SHUF10:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> undef, <32 x i32> <i32 1, 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>
135; CHECK-NEXT:    [[RDX_MINMAX_CMP11:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
136; CHECK-NEXT:    [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> [[RDX_SHUF10]]
137; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <32 x i32> [[RDX_MINMAX_SELECT12]], i32 0
138; CHECK-NEXT:    ret i32 [[TMP3]]
139;
140  %2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
141  %3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
142  %4 = icmp sgt i32 %2, %3
143  %5 = select i1 %4, i32 %2, i32 %3
144  %6 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
145  %7 = icmp sgt i32 %5, %6
146  %8 = select i1 %7, i32 %5, i32 %6
147  %9 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 3), align 4
148  %10 = icmp sgt i32 %8, %9
149  %11 = select i1 %10, i32 %8, i32 %9
150  %12 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
151  %13 = icmp sgt i32 %11, %12
152  %14 = select i1 %13, i32 %11, i32 %12
153  %15 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
154  %16 = icmp sgt i32 %14, %15
155  %17 = select i1 %16, i32 %14, i32 %15
156  %18 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
157  %19 = icmp sgt i32 %17, %18
158  %20 = select i1 %19, i32 %17, i32 %18
159  %21 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
160  %22 = icmp sgt i32 %20, %21
161  %23 = select i1 %22, i32 %20, i32 %21
162  %24 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 8), align 16
163  %25 = icmp sgt i32 %23, %24
164  %26 = select i1 %25, i32 %23, i32 %24
165  %27 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 9), align 4
166  %28 = icmp sgt i32 %26, %27
167  %29 = select i1 %28, i32 %26, i32 %27
168  %30 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 10), align 8
169  %31 = icmp sgt i32 %29, %30
170  %32 = select i1 %31, i32 %29, i32 %30
171  %33 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 11), align 4
172  %34 = icmp sgt i32 %32, %33
173  %35 = select i1 %34, i32 %32, i32 %33
174  %36 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 12), align 16
175  %37 = icmp sgt i32 %35, %36
176  %38 = select i1 %37, i32 %35, i32 %36
177  %39 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 13), align 4
178  %40 = icmp sgt i32 %38, %39
179  %41 = select i1 %40, i32 %38, i32 %39
180  %42 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 14), align 8
181  %43 = icmp sgt i32 %41, %42
182  %44 = select i1 %43, i32 %41, i32 %42
183  %45 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 15), align 4
184  %46 = icmp sgt i32 %44, %45
185  %47 = select i1 %46, i32 %44, i32 %45
186  %48 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 16), align 16
187  %49 = icmp sgt i32 %47, %48
188  %50 = select i1 %49, i32 %47, i32 %48
189  %51 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 17), align 4
190  %52 = icmp sgt i32 %50, %51
191  %53 = select i1 %52, i32 %50, i32 %51
192  %54 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 18), align 8
193  %55 = icmp sgt i32 %53, %54
194  %56 = select i1 %55, i32 %53, i32 %54
195  %57 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 19), align 4
196  %58 = icmp sgt i32 %56, %57
197  %59 = select i1 %58, i32 %56, i32 %57
198  %60 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 20), align 16
199  %61 = icmp sgt i32 %59, %60
200  %62 = select i1 %61, i32 %59, i32 %60
201  %63 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 21), align 4
202  %64 = icmp sgt i32 %62, %63
203  %65 = select i1 %64, i32 %62, i32 %63
204  %66 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 22), align 8
205  %67 = icmp sgt i32 %65, %66
206  %68 = select i1 %67, i32 %65, i32 %66
207  %69 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 23), align 4
208  %70 = icmp sgt i32 %68, %69
209  %71 = select i1 %70, i32 %68, i32 %69
210  %72 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 24), align 16
211  %73 = icmp sgt i32 %71, %72
212  %74 = select i1 %73, i32 %71, i32 %72
213  %75 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 25), align 4
214  %76 = icmp sgt i32 %74, %75
215  %77 = select i1 %76, i32 %74, i32 %75
216  %78 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 26), align 8
217  %79 = icmp sgt i32 %77, %78
218  %80 = select i1 %79, i32 %77, i32 %78
219  %81 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 27), align 4
220  %82 = icmp sgt i32 %80, %81
221  %83 = select i1 %82, i32 %80, i32 %81
222  %84 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 28), align 16
223  %85 = icmp sgt i32 %83, %84
224  %86 = select i1 %85, i32 %83, i32 %84
225  %87 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 29), align 4
226  %88 = icmp sgt i32 %86, %87
227  %89 = select i1 %88, i32 %86, i32 %87
228  %90 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 30), align 8
229  %91 = icmp sgt i32 %89, %90
230  %92 = select i1 %91, i32 %89, i32 %90
231  %93 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 31), align 4
232  %94 = icmp sgt i32 %92, %93
233  %95 = select i1 %94, i32 %92, i32 %93
234  ret i32 %95
235}
236
237define float @maxf8(float) {
238; CHECK-LABEL: @maxf8(
239; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast ([32 x float]* @arr1 to <8 x float>*), align 16
240; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <8 x float> [[TMP2]], <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
241; CHECK-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <8 x float> [[TMP2]], [[RDX_SHUF]]
242; CHECK-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP]], <8 x float> [[TMP2]], <8 x float> [[RDX_SHUF]]
243; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <8 x float> [[RDX_MINMAX_SELECT]], <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
244; CHECK-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <8 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
245; CHECK-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP2]], <8 x float> [[RDX_MINMAX_SELECT]], <8 x float> [[RDX_SHUF1]]
246; CHECK-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <8 x float> [[RDX_MINMAX_SELECT3]], <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
247; CHECK-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <8 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
248; CHECK-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP5]], <8 x float> [[RDX_MINMAX_SELECT3]], <8 x float> [[RDX_SHUF4]]
249; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <8 x float> [[RDX_MINMAX_SELECT6]], i32 0
250; CHECK-NEXT:    ret float [[TMP3]]
251;
252  %2 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 0), align 16
253  %3 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 1), align 4
254  %4 = fcmp fast ogt float %2, %3
255  %5 = select i1 %4, float %2, float %3
256  %6 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 2), align 8
257  %7 = fcmp fast ogt float %5, %6
258  %8 = select i1 %7, float %5, float %6
259  %9 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 3), align 4
260  %10 = fcmp fast ogt float %8, %9
261  %11 = select i1 %10, float %8, float %9
262  %12 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 4), align 16
263  %13 = fcmp fast ogt float %11, %12
264  %14 = select i1 %13, float %11, float %12
265  %15 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 5), align 4
266  %16 = fcmp fast ogt float %14, %15
267  %17 = select i1 %16, float %14, float %15
268  %18 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 6), align 8
269  %19 = fcmp fast ogt float %17, %18
270  %20 = select i1 %19, float %17, float %18
271  %21 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 7), align 4
272  %22 = fcmp fast ogt float %20, %21
273  %23 = select i1 %22, float %20, float %21
274  ret float %23
275}
276
277define float @maxf16(float) {
278; CHECK-LABEL: @maxf16(
279; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x float>, <16 x float>* bitcast ([32 x float]* @arr1 to <16 x float>*), align 16
280; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <16 x float> [[TMP2]], <16 x float> undef, <16 x i32> <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>
281; CHECK-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <16 x float> [[TMP2]], [[RDX_SHUF]]
282; CHECK-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x float> [[TMP2]], <16 x float> [[RDX_SHUF]]
283; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT]], <16 x float> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
284; CHECK-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
285; CHECK-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x float> [[RDX_MINMAX_SELECT]], <16 x float> [[RDX_SHUF1]]
286; CHECK-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT3]], <16 x float> undef, <16 x i32> <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, i32 undef, i32 undef>
287; CHECK-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
288; CHECK-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x float> [[RDX_MINMAX_SELECT3]], <16 x float> [[RDX_SHUF4]]
289; CHECK-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT6]], <16 x float> undef, <16 x i32> <i32 1, 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>
290; CHECK-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
291; CHECK-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x float> [[RDX_MINMAX_SELECT6]], <16 x float> [[RDX_SHUF7]]
292; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <16 x float> [[RDX_MINMAX_SELECT9]], i32 0
293; CHECK-NEXT:    ret float [[TMP3]]
294;
295  %2 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 0), align 16
296  %3 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 1), align 4
297  %4 = fcmp fast ogt float %2, %3
298  %5 = select i1 %4, float %2, float %3
299  %6 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 2), align 8
300  %7 = fcmp fast ogt float %5, %6
301  %8 = select i1 %7, float %5, float %6
302  %9 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 3), align 4
303  %10 = fcmp fast ogt float %8, %9
304  %11 = select i1 %10, float %8, float %9
305  %12 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 4), align 16
306  %13 = fcmp fast ogt float %11, %12
307  %14 = select i1 %13, float %11, float %12
308  %15 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 5), align 4
309  %16 = fcmp fast ogt float %14, %15
310  %17 = select i1 %16, float %14, float %15
311  %18 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 6), align 8
312  %19 = fcmp fast ogt float %17, %18
313  %20 = select i1 %19, float %17, float %18
314  %21 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 7), align 4
315  %22 = fcmp fast ogt float %20, %21
316  %23 = select i1 %22, float %20, float %21
317  %24 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 8), align 16
318  %25 = fcmp fast ogt float %23, %24
319  %26 = select i1 %25, float %23, float %24
320  %27 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 9), align 4
321  %28 = fcmp fast ogt float %26, %27
322  %29 = select i1 %28, float %26, float %27
323  %30 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 10), align 8
324  %31 = fcmp fast ogt float %29, %30
325  %32 = select i1 %31, float %29, float %30
326  %33 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 11), align 4
327  %34 = fcmp fast ogt float %32, %33
328  %35 = select i1 %34, float %32, float %33
329  %36 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 12), align 16
330  %37 = fcmp fast ogt float %35, %36
331  %38 = select i1 %37, float %35, float %36
332  %39 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 13), align 4
333  %40 = fcmp fast ogt float %38, %39
334  %41 = select i1 %40, float %38, float %39
335  %42 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 14), align 8
336  %43 = fcmp fast ogt float %41, %42
337  %44 = select i1 %43, float %41, float %42
338  %45 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 15), align 4
339  %46 = fcmp fast ogt float %44, %45
340  %47 = select i1 %46, float %44, float %45
341  ret float %47
342}
343
344define float @maxf32(float) {
345; CHECK-LABEL: @maxf32(
346; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x float>, <32 x float>* bitcast ([32 x float]* @arr1 to <32 x float>*), align 16
347; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <32 x float> [[TMP2]], <32 x float> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, 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>
348; CHECK-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <32 x float> [[TMP2]], [[RDX_SHUF]]
349; CHECK-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x float> [[TMP2]], <32 x float> [[RDX_SHUF]]
350; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT]], <32 x float> undef, <32 x i32> <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>
351; CHECK-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
352; CHECK-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x float> [[RDX_MINMAX_SELECT]], <32 x float> [[RDX_SHUF1]]
353; CHECK-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT3]], <32 x float> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, 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>
354; CHECK-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
355; CHECK-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x float> [[RDX_MINMAX_SELECT3]], <32 x float> [[RDX_SHUF4]]
356; CHECK-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT6]], <32 x float> undef, <32 x i32> <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, 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>
357; CHECK-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
358; CHECK-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x float> [[RDX_MINMAX_SELECT6]], <32 x float> [[RDX_SHUF7]]
359; CHECK-NEXT:    [[RDX_SHUF10:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT9]], <32 x float> undef, <32 x i32> <i32 1, 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>
360; CHECK-NEXT:    [[RDX_MINMAX_CMP11:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
361; CHECK-NEXT:    [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x float> [[RDX_MINMAX_SELECT9]], <32 x float> [[RDX_SHUF10]]
362; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <32 x float> [[RDX_MINMAX_SELECT12]], i32 0
363; CHECK-NEXT:    ret float [[TMP3]]
364;
365  %2 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 0), align 16
366  %3 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 1), align 4
367  %4 = fcmp fast ogt float %2, %3
368  %5 = select i1 %4, float %2, float %3
369  %6 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 2), align 8
370  %7 = fcmp fast ogt float %5, %6
371  %8 = select i1 %7, float %5, float %6
372  %9 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 3), align 4
373  %10 = fcmp fast ogt float %8, %9
374  %11 = select i1 %10, float %8, float %9
375  %12 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 4), align 16
376  %13 = fcmp fast ogt float %11, %12
377  %14 = select i1 %13, float %11, float %12
378  %15 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 5), align 4
379  %16 = fcmp fast ogt float %14, %15
380  %17 = select i1 %16, float %14, float %15
381  %18 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 6), align 8
382  %19 = fcmp fast ogt float %17, %18
383  %20 = select i1 %19, float %17, float %18
384  %21 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 7), align 4
385  %22 = fcmp fast ogt float %20, %21
386  %23 = select i1 %22, float %20, float %21
387  %24 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 8), align 16
388  %25 = fcmp fast ogt float %23, %24
389  %26 = select i1 %25, float %23, float %24
390  %27 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 9), align 4
391  %28 = fcmp fast ogt float %26, %27
392  %29 = select i1 %28, float %26, float %27
393  %30 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 10), align 8
394  %31 = fcmp fast ogt float %29, %30
395  %32 = select i1 %31, float %29, float %30
396  %33 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 11), align 4
397  %34 = fcmp fast ogt float %32, %33
398  %35 = select i1 %34, float %32, float %33
399  %36 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 12), align 16
400  %37 = fcmp fast ogt float %35, %36
401  %38 = select i1 %37, float %35, float %36
402  %39 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 13), align 4
403  %40 = fcmp fast ogt float %38, %39
404  %41 = select i1 %40, float %38, float %39
405  %42 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 14), align 8
406  %43 = fcmp fast ogt float %41, %42
407  %44 = select i1 %43, float %41, float %42
408  %45 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 15), align 4
409  %46 = fcmp fast ogt float %44, %45
410  %47 = select i1 %46, float %44, float %45
411  %48 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 16), align 16
412  %49 = fcmp fast ogt float %47, %48
413  %50 = select i1 %49, float %47, float %48
414  %51 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 17), align 4
415  %52 = fcmp fast ogt float %50, %51
416  %53 = select i1 %52, float %50, float %51
417  %54 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 18), align 8
418  %55 = fcmp fast ogt float %53, %54
419  %56 = select i1 %55, float %53, float %54
420  %57 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 19), align 4
421  %58 = fcmp fast ogt float %56, %57
422  %59 = select i1 %58, float %56, float %57
423  %60 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 20), align 16
424  %61 = fcmp fast ogt float %59, %60
425  %62 = select i1 %61, float %59, float %60
426  %63 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 21), align 4
427  %64 = fcmp fast ogt float %62, %63
428  %65 = select i1 %64, float %62, float %63
429  %66 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 22), align 8
430  %67 = fcmp fast ogt float %65, %66
431  %68 = select i1 %67, float %65, float %66
432  %69 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 23), align 4
433  %70 = fcmp fast ogt float %68, %69
434  %71 = select i1 %70, float %68, float %69
435  %72 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 24), align 16
436  %73 = fcmp fast ogt float %71, %72
437  %74 = select i1 %73, float %71, float %72
438  %75 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 25), align 4
439  %76 = fcmp fast ogt float %74, %75
440  %77 = select i1 %76, float %74, float %75
441  %78 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 26), align 8
442  %79 = fcmp fast ogt float %77, %78
443  %80 = select i1 %79, float %77, float %78
444  %81 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 27), align 4
445  %82 = fcmp fast ogt float %80, %81
446  %83 = select i1 %82, float %80, float %81
447  %84 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 28), align 16
448  %85 = fcmp fast ogt float %83, %84
449  %86 = select i1 %85, float %83, float %84
450  %87 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 29), align 4
451  %88 = fcmp fast ogt float %86, %87
452  %89 = select i1 %88, float %86, float %87
453  %90 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 30), align 8
454  %91 = fcmp fast ogt float %89, %90
455  %92 = select i1 %91, float %89, float %90
456  %93 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 31), align 4
457  %94 = fcmp fast ogt float %92, %93
458  %95 = select i1 %94, float %92, float %93
459  ret float %95
460}
461
462define i32 @maxi8_mutiple_uses(i32) {
463; SSE-LABEL: @maxi8_mutiple_uses(
464; SSE-NEXT:    [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
465; SSE-NEXT:    [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
466; SSE-NEXT:    [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], [[TMP3]]
467; SSE-NEXT:    [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP2]], i32 [[TMP3]]
468; SSE-NEXT:    [[TMP6:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
469; SSE-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP5]], [[TMP6]]
470; SSE-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP5]], i32 [[TMP6]]
471; SSE-NEXT:    [[TMP9:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 3), align 4
472; SSE-NEXT:    [[TMP10:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]]
473; SSE-NEXT:    [[TMP11:%.*]] = select i1 [[TMP10]], i32 [[TMP8]], i32 [[TMP9]]
474; SSE-NEXT:    [[TMP12:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
475; SSE-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP11]], [[TMP12]]
476; SSE-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP11]], i32 [[TMP12]]
477; SSE-NEXT:    [[TMP15:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
478; SSE-NEXT:    [[TMP16:%.*]] = icmp sgt i32 [[TMP14]], [[TMP15]]
479; SSE-NEXT:    [[TMP17:%.*]] = select i1 [[TMP16]], i32 [[TMP14]], i32 [[TMP15]]
480; SSE-NEXT:    [[TMP18:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
481; SSE-NEXT:    [[TMP19:%.*]] = icmp sgt i32 [[TMP17]], [[TMP18]]
482; SSE-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], i32 [[TMP17]], i32 [[TMP18]]
483; SSE-NEXT:    [[TMP21:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
484; SSE-NEXT:    [[TMP22:%.*]] = icmp sgt i32 [[TMP20]], [[TMP21]]
485; SSE-NEXT:    [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP20]], i32 [[TMP21]]
486; SSE-NEXT:    [[TMP24:%.*]] = select i1 [[TMP4]], i32 3, i32 4
487; SSE-NEXT:    store i32 [[TMP24]], i32* @var, align 8
488; SSE-NEXT:    ret i32 [[TMP23]]
489;
490; AVX-LABEL: @maxi8_mutiple_uses(
491; AVX-NEXT:    [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
492; AVX-NEXT:    [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
493; AVX-NEXT:    [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], [[TMP3]]
494; AVX-NEXT:    [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP2]], i32 [[TMP3]]
495; AVX-NEXT:    [[TMP6:%.*]] = load <4 x i32>, <4 x i32>* bitcast (i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2) to <4 x i32>*), align 8
496; AVX-NEXT:    [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
497; AVX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
498; AVX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP6]], [[RDX_SHUF]]
499; AVX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP6]], <4 x i32> [[RDX_SHUF]]
500; AVX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
501; AVX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
502; AVX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
503; AVX-NEXT:    [[TMP8:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
504; AVX-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], [[TMP7]]
505; AVX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 [[TMP7]]
506; AVX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], [[TMP5]]
507; AVX-NEXT:    [[OP_EXTRA:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 [[TMP5]]
508; AVX-NEXT:    [[TMP12:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
509; AVX-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[OP_EXTRA]], [[TMP12]]
510; AVX-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[OP_EXTRA]], i32 [[TMP12]]
511; AVX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP4]], i32 3, i32 4
512; AVX-NEXT:    store i32 [[TMP15]], i32* @var, align 8
513; AVX-NEXT:    ret i32 [[TMP14]]
514;
515; THRESH-LABEL: @maxi8_mutiple_uses(
516; THRESH-NEXT:    [[TMP2:%.*]] = load <2 x i32>, <2 x i32>* bitcast ([32 x i32]* @arr to <2 x i32>*), align 16
517; THRESH-NEXT:    [[TMP3:%.*]] = extractelement <2 x i32> [[TMP2]], i32 0
518; THRESH-NEXT:    [[TMP4:%.*]] = extractelement <2 x i32> [[TMP2]], i32 1
519; THRESH-NEXT:    [[TMP5:%.*]] = load <4 x i32>, <4 x i32>* bitcast (i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2) to <4 x i32>*), align 8
520; THRESH-NEXT:    [[TMP6:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
521; THRESH-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP5]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
522; THRESH-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP5]], [[RDX_SHUF]]
523; THRESH-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP5]], <4 x i32> [[RDX_SHUF]]
524; THRESH-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
525; THRESH-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
526; THRESH-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
527; THRESH-NEXT:    [[TMP7:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
528; THRESH-NEXT:    [[TMP8:%.*]] = insertelement <2 x i32> undef, i32 [[TMP7]], i32 0
529; THRESH-NEXT:    [[TMP9:%.*]] = insertelement <2 x i32> [[TMP8]], i32 [[TMP3]], i32 1
530; THRESH-NEXT:    [[TMP10:%.*]] = insertelement <2 x i32> undef, i32 [[TMP6]], i32 0
531; THRESH-NEXT:    [[TMP11:%.*]] = insertelement <2 x i32> [[TMP10]], i32 [[TMP4]], i32 1
532; THRESH-NEXT:    [[TMP12:%.*]] = icmp sgt <2 x i32> [[TMP9]], [[TMP11]]
533; THRESH-NEXT:    [[TMP13:%.*]] = select <2 x i1> [[TMP12]], <2 x i32> [[TMP9]], <2 x i32> [[TMP11]]
534; THRESH-NEXT:    [[TMP14:%.*]] = extractelement <2 x i32> [[TMP13]], i32 1
535; THRESH-NEXT:    [[TMP15:%.*]] = extractelement <2 x i32> [[TMP13]], i32 0
536; THRESH-NEXT:    [[TMP16:%.*]] = icmp sgt i32 [[TMP15]], [[TMP14]]
537; THRESH-NEXT:    [[OP_EXTRA:%.*]] = select i1 [[TMP16]], i32 [[TMP15]], i32 [[TMP14]]
538; THRESH-NEXT:    [[TMP17:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
539; THRESH-NEXT:    [[TMP18:%.*]] = icmp sgt i32 [[OP_EXTRA]], [[TMP17]]
540; THRESH-NEXT:    [[TMP19:%.*]] = select i1 [[TMP18]], i32 [[OP_EXTRA]], i32 [[TMP17]]
541; THRESH-NEXT:    [[TMP20:%.*]] = extractelement <2 x i1> [[TMP12]], i32 1
542; THRESH-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 3, i32 4
543; THRESH-NEXT:    store i32 [[TMP21]], i32* @var, align 8
544; THRESH-NEXT:    ret i32 [[TMP19]]
545;
546  %2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
547  %3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
548  %4 = icmp sgt i32 %2, %3
549  %5 = select i1 %4, i32 %2, i32 %3
550  %6 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
551  %7 = icmp sgt i32 %5, %6
552  %8 = select i1 %7, i32 %5, i32 %6
553  %9 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 3), align 4
554  %10 = icmp sgt i32 %8, %9
555  %11 = select i1 %10, i32 %8, i32 %9
556  %12 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
557  %13 = icmp sgt i32 %11, %12
558  %14 = select i1 %13, i32 %11, i32 %12
559  %15 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
560  %16 = icmp sgt i32 %14, %15
561  %17 = select i1 %16, i32 %14, i32 %15
562  %18 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
563  %19 = icmp sgt i32 %17, %18
564  %20 = select i1 %19, i32 %17, i32 %18
565  %21 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
566  %22 = icmp sgt i32 %20, %21
567  %23 = select i1 %22, i32 %20, i32 %21
568  %24 = select i1 %4, i32 3, i32 4
569  store i32 %24, i32* @var, align 8
570  ret i32 %23
571}
572
573define i32 @maxi8_wrong_parent(i32) {
574; SSE-LABEL: @maxi8_wrong_parent(
575; SSE-NEXT:    [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
576; SSE-NEXT:    [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
577; SSE-NEXT:    [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], [[TMP3]]
578; SSE-NEXT:    br label [[PP:%.*]]
579; SSE:       pp:
580; SSE-NEXT:    [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP2]], i32 [[TMP3]]
581; SSE-NEXT:    [[TMP6:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
582; SSE-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP5]], [[TMP6]]
583; SSE-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP5]], i32 [[TMP6]]
584; SSE-NEXT:    [[TMP9:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 3), align 4
585; SSE-NEXT:    [[TMP10:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]]
586; SSE-NEXT:    [[TMP11:%.*]] = select i1 [[TMP10]], i32 [[TMP8]], i32 [[TMP9]]
587; SSE-NEXT:    [[TMP12:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
588; SSE-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP11]], [[TMP12]]
589; SSE-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP11]], i32 [[TMP12]]
590; SSE-NEXT:    [[TMP15:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
591; SSE-NEXT:    [[TMP16:%.*]] = icmp sgt i32 [[TMP14]], [[TMP15]]
592; SSE-NEXT:    [[TMP17:%.*]] = select i1 [[TMP16]], i32 [[TMP14]], i32 [[TMP15]]
593; SSE-NEXT:    [[TMP18:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
594; SSE-NEXT:    [[TMP19:%.*]] = icmp sgt i32 [[TMP17]], [[TMP18]]
595; SSE-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], i32 [[TMP17]], i32 [[TMP18]]
596; SSE-NEXT:    [[TMP21:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
597; SSE-NEXT:    [[TMP22:%.*]] = icmp sgt i32 [[TMP20]], [[TMP21]]
598; SSE-NEXT:    [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP20]], i32 [[TMP21]]
599; SSE-NEXT:    ret i32 [[TMP23]]
600;
601; AVX-LABEL: @maxi8_wrong_parent(
602; AVX-NEXT:    [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
603; AVX-NEXT:    [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
604; AVX-NEXT:    [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], [[TMP3]]
605; AVX-NEXT:    br label [[PP:%.*]]
606; AVX:       pp:
607; AVX-NEXT:    [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP2]], i32 [[TMP3]]
608; AVX-NEXT:    [[TMP6:%.*]] = load <4 x i32>, <4 x i32>* bitcast (i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2) to <4 x i32>*), align 8
609; AVX-NEXT:    [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
610; AVX-NEXT:    [[TMP8:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
611; AVX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
612; AVX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP6]], [[RDX_SHUF]]
613; AVX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP6]], <4 x i32> [[RDX_SHUF]]
614; AVX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
615; AVX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
616; AVX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
617; AVX-NEXT:    [[TMP9:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
618; AVX-NEXT:    [[TMP10:%.*]] = icmp sgt i32 [[TMP9]], [[TMP7]]
619; AVX-NEXT:    [[TMP11:%.*]] = select i1 [[TMP10]], i32 [[TMP9]], i32 [[TMP7]]
620; AVX-NEXT:    [[TMP12:%.*]] = icmp sgt i32 [[TMP11]], [[TMP8]]
621; AVX-NEXT:    [[TMP13:%.*]] = select i1 [[TMP12]], i32 [[TMP11]], i32 [[TMP8]]
622; AVX-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP13]], [[TMP5]]
623; AVX-NEXT:    [[OP_EXTRA:%.*]] = select i1 [[TMP14]], i32 [[TMP13]], i32 [[TMP5]]
624; AVX-NEXT:    ret i32 [[OP_EXTRA]]
625;
626; THRESH-LABEL: @maxi8_wrong_parent(
627; THRESH-NEXT:    [[TMP2:%.*]] = load <2 x i32>, <2 x i32>* bitcast ([32 x i32]* @arr to <2 x i32>*), align 16
628; THRESH-NEXT:    [[TMP3:%.*]] = extractelement <2 x i32> [[TMP2]], i32 0
629; THRESH-NEXT:    [[TMP4:%.*]] = extractelement <2 x i32> [[TMP2]], i32 1
630; THRESH-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
631; THRESH-NEXT:    br label [[PP:%.*]]
632; THRESH:       pp:
633; THRESH-NEXT:    [[TMP6:%.*]] = load <4 x i32>, <4 x i32>* bitcast (i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2) to <4 x i32>*), align 8
634; THRESH-NEXT:    [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
635; THRESH-NEXT:    [[TMP8:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
636; THRESH-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
637; THRESH-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP6]], [[RDX_SHUF]]
638; THRESH-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP6]], <4 x i32> [[RDX_SHUF]]
639; THRESH-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
640; THRESH-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
641; THRESH-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
642; THRESH-NEXT:    [[TMP9:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
643; THRESH-NEXT:    [[TMP10:%.*]] = icmp sgt i32 [[TMP9]], [[TMP7]]
644; THRESH-NEXT:    [[TMP11:%.*]] = select i1 [[TMP10]], i32 [[TMP9]], i32 [[TMP7]]
645; THRESH-NEXT:    [[TMP12:%.*]] = icmp sgt i32 [[TMP11]], [[TMP8]]
646; THRESH-NEXT:    [[TMP13:%.*]] = insertelement <2 x i1> undef, i1 [[TMP12]], i32 0
647; THRESH-NEXT:    [[TMP14:%.*]] = insertelement <2 x i1> [[TMP13]], i1 [[TMP5]], i32 1
648; THRESH-NEXT:    [[TMP15:%.*]] = insertelement <2 x i32> undef, i32 [[TMP11]], i32 0
649; THRESH-NEXT:    [[TMP16:%.*]] = insertelement <2 x i32> [[TMP15]], i32 [[TMP3]], i32 1
650; THRESH-NEXT:    [[TMP17:%.*]] = insertelement <2 x i32> undef, i32 [[TMP8]], i32 0
651; THRESH-NEXT:    [[TMP18:%.*]] = insertelement <2 x i32> [[TMP17]], i32 [[TMP4]], i32 1
652; THRESH-NEXT:    [[TMP19:%.*]] = select <2 x i1> [[TMP14]], <2 x i32> [[TMP16]], <2 x i32> [[TMP18]]
653; THRESH-NEXT:    [[TMP20:%.*]] = extractelement <2 x i32> [[TMP19]], i32 1
654; THRESH-NEXT:    [[TMP21:%.*]] = extractelement <2 x i32> [[TMP19]], i32 0
655; THRESH-NEXT:    [[TMP22:%.*]] = icmp sgt i32 [[TMP21]], [[TMP20]]
656; THRESH-NEXT:    [[OP_EXTRA:%.*]] = select i1 [[TMP22]], i32 [[TMP21]], i32 [[TMP20]]
657; THRESH-NEXT:    ret i32 [[OP_EXTRA]]
658;
659  %2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
660  %3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
661  %4 = icmp sgt i32 %2, %3
662  br label %pp
663
664pp:
665  %5 = select i1 %4, i32 %2, i32 %3
666  %6 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
667  %7 = icmp sgt i32 %5, %6
668  %8 = select i1 %7, i32 %5, i32 %6
669  %9 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 3), align 4
670  %10 = icmp sgt i32 %8, %9
671  %11 = select i1 %10, i32 %8, i32 %9
672  %12 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
673  %13 = icmp sgt i32 %11, %12
674  %14 = select i1 %13, i32 %11, i32 %12
675  %15 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
676  %16 = icmp sgt i32 %14, %15
677  %17 = select i1 %16, i32 %14, i32 %15
678  %18 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
679  %19 = icmp sgt i32 %17, %18
680  %20 = select i1 %19, i32 %17, i32 %18
681  %21 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
682  %22 = icmp sgt i32 %20, %21
683  %23 = select i1 %22, i32 %20, i32 %21
684  ret i32 %23
685}
686
687; PR38191 - We don't handle array-of-pointer reductions.
688define i32* @maxp8(i32) {
689; DEFAULT-LABEL: @maxp8(
690; DEFAULT-NEXT:    [[TMP2:%.*]] = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 0), align 16
691; DEFAULT-NEXT:    [[TMP3:%.*]] = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 1), align 4
692; DEFAULT-NEXT:    [[TMP4:%.*]] = icmp ugt i32* [[TMP2]], [[TMP3]]
693; DEFAULT-NEXT:    [[TMP5:%.*]] = select i1 [[TMP4]], i32* [[TMP2]], i32* [[TMP3]]
694; DEFAULT-NEXT:    [[TMP6:%.*]] = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 2), align 8
695; DEFAULT-NEXT:    [[TMP7:%.*]] = icmp ugt i32* [[TMP5]], [[TMP6]]
696; DEFAULT-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32* [[TMP5]], i32* [[TMP6]]
697; DEFAULT-NEXT:    [[TMP9:%.*]] = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 3), align 4
698; DEFAULT-NEXT:    [[TMP10:%.*]] = icmp ugt i32* [[TMP8]], [[TMP9]]
699; DEFAULT-NEXT:    [[TMP11:%.*]] = select i1 [[TMP10]], i32* [[TMP8]], i32* [[TMP9]]
700; DEFAULT-NEXT:    [[TMP12:%.*]] = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 4), align 16
701; DEFAULT-NEXT:    [[TMP13:%.*]] = icmp ugt i32* [[TMP11]], [[TMP12]]
702; DEFAULT-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32* [[TMP11]], i32* [[TMP12]]
703; DEFAULT-NEXT:    [[TMP15:%.*]] = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 5), align 4
704; DEFAULT-NEXT:    [[TMP16:%.*]] = icmp ugt i32* [[TMP14]], [[TMP15]]
705; DEFAULT-NEXT:    [[TMP17:%.*]] = select i1 [[TMP16]], i32* [[TMP14]], i32* [[TMP15]]
706; DEFAULT-NEXT:    [[TMP18:%.*]] = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 6), align 8
707; DEFAULT-NEXT:    [[TMP19:%.*]] = icmp ugt i32* [[TMP17]], [[TMP18]]
708; DEFAULT-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], i32* [[TMP17]], i32* [[TMP18]]
709; DEFAULT-NEXT:    [[TMP21:%.*]] = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 7), align 4
710; DEFAULT-NEXT:    [[TMP22:%.*]] = icmp ugt i32* [[TMP20]], [[TMP21]]
711; DEFAULT-NEXT:    [[TMP23:%.*]] = select i1 [[TMP22]], i32* [[TMP20]], i32* [[TMP21]]
712; DEFAULT-NEXT:    ret i32* [[TMP23]]
713;
714; THRESH-LABEL: @maxp8(
715; THRESH-NEXT:    [[TMP2:%.*]] = load <2 x i32*>, <2 x i32*>* bitcast ([32 x i32*]* @arrp to <2 x i32*>*), align 16
716; THRESH-NEXT:    [[TMP3:%.*]] = extractelement <2 x i32*> [[TMP2]], i32 0
717; THRESH-NEXT:    [[TMP4:%.*]] = extractelement <2 x i32*> [[TMP2]], i32 1
718; THRESH-NEXT:    [[TMP5:%.*]] = icmp ugt i32* [[TMP3]], [[TMP4]]
719; THRESH-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32* [[TMP3]], i32* [[TMP4]]
720; THRESH-NEXT:    [[TMP7:%.*]] = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 2), align 8
721; THRESH-NEXT:    [[TMP8:%.*]] = icmp ugt i32* [[TMP6]], [[TMP7]]
722; THRESH-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32* [[TMP6]], i32* [[TMP7]]
723; THRESH-NEXT:    [[TMP10:%.*]] = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 3), align 4
724; THRESH-NEXT:    [[TMP11:%.*]] = icmp ugt i32* [[TMP9]], [[TMP10]]
725; THRESH-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32* [[TMP9]], i32* [[TMP10]]
726; THRESH-NEXT:    [[TMP13:%.*]] = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 4), align 16
727; THRESH-NEXT:    [[TMP14:%.*]] = icmp ugt i32* [[TMP12]], [[TMP13]]
728; THRESH-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32* [[TMP12]], i32* [[TMP13]]
729; THRESH-NEXT:    [[TMP16:%.*]] = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 5), align 4
730; THRESH-NEXT:    [[TMP17:%.*]] = icmp ugt i32* [[TMP15]], [[TMP16]]
731; THRESH-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32* [[TMP15]], i32* [[TMP16]]
732; THRESH-NEXT:    [[TMP19:%.*]] = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 6), align 8
733; THRESH-NEXT:    [[TMP20:%.*]] = icmp ugt i32* [[TMP18]], [[TMP19]]
734; THRESH-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32* [[TMP18]], i32* [[TMP19]]
735; THRESH-NEXT:    [[TMP22:%.*]] = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 7), align 4
736; THRESH-NEXT:    [[TMP23:%.*]] = icmp ugt i32* [[TMP21]], [[TMP22]]
737; THRESH-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32* [[TMP21]], i32* [[TMP22]]
738; THRESH-NEXT:    ret i32* [[TMP24]]
739;
740  %2 = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 0), align 16
741  %3 = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 1), align 4
742  %4 = icmp ugt i32* %2, %3
743  %5 = select i1 %4, i32* %2, i32* %3
744  %6 = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 2), align 8
745  %7 = icmp ugt i32* %5, %6
746  %8 = select i1 %7, i32* %5, i32* %6
747  %9 = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 3), align 4
748  %10 = icmp ugt i32* %8, %9
749  %11 = select i1 %10, i32* %8, i32* %9
750  %12 = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 4), align 16
751  %13 = icmp ugt i32* %11, %12
752  %14 = select i1 %13, i32* %11, i32* %12
753  %15 = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 5), align 4
754  %16 = icmp ugt i32* %14, %15
755  %17 = select i1 %16, i32* %14, i32* %15
756  %18 = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 6), align 8
757  %19 = icmp ugt i32* %17, %18
758  %20 = select i1 %19, i32* %17, i32* %18
759  %21 = load i32*, i32** getelementptr inbounds ([32 x i32*], [32 x i32*]* @arrp, i64 0, i64 7), align 4
760  %22 = icmp ugt i32* %20, %21
761  %23 = select i1 %22, i32* %20, i32* %21
762  ret i32* %23
763}
764