xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/X86/horizontal-smax.ll (revision b28f407df9d0f63ca85cfaeddf9effc024c071d8) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -mtriple=x86_64-unknown-linux -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,SSE2
3; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=x86-64-v2 -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,SSE4
4; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=corei7-avx -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX
5; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=core-avx2 -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX
6
7@arr = local_unnamed_addr global [32 x i32] zeroinitializer, align 16
8
9declare i32 @llvm.smax.i32(i32, i32)
10
11define i32 @smax_v2i32(i32) {
12; CHECK-LABEL: @smax_v2i32(
13; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr @arr, align 16
14; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 1), align 4
15; CHECK-NEXT:    [[TMP4:%.*]] = call i32 @llvm.smax.i32(i32 [[TMP2]], i32 [[TMP3]])
16; CHECK-NEXT:    ret i32 [[TMP4]]
17;
18  %2 = load i32, ptr @arr, align 16
19  %3 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 1), align 4
20  %4 = call i32 @llvm.smax.i32(i32 %2, i32 %3)
21  ret i32 %4
22}
23
24define i32 @smax_v4i32(i32) {
25; SSE2-LABEL: @smax_v4i32(
26; SSE2-NEXT:    [[TMP2:%.*]] = load i32, ptr @arr, align 16
27; SSE2-NEXT:    [[TMP3:%.*]] = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 1), align 4
28; SSE2-NEXT:    [[TMP4:%.*]] = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 2), align 8
29; SSE2-NEXT:    [[TMP5:%.*]] = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 3), align 4
30; SSE2-NEXT:    [[TMP6:%.*]] = call i32 @llvm.smax.i32(i32 [[TMP2]], i32 [[TMP3]])
31; SSE2-NEXT:    [[TMP7:%.*]] = call i32 @llvm.smax.i32(i32 [[TMP6]], i32 [[TMP4]])
32; SSE2-NEXT:    [[TMP8:%.*]] = call i32 @llvm.smax.i32(i32 [[TMP7]], i32 [[TMP5]])
33; SSE2-NEXT:    ret i32 [[TMP8]]
34;
35; SSE4-LABEL: @smax_v4i32(
36; SSE4-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr @arr, align 16
37; SSE4-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.smax.v4i32(<4 x i32> [[TMP2]])
38; SSE4-NEXT:    ret i32 [[TMP3]]
39;
40; AVX-LABEL: @smax_v4i32(
41; AVX-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr @arr, align 16
42; AVX-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.smax.v4i32(<4 x i32> [[TMP2]])
43; AVX-NEXT:    ret i32 [[TMP3]]
44;
45  %2 = load i32, ptr @arr, align 16
46  %3 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 1), align 4
47  %4 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 2), align 8
48  %5 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 3), align 4
49  %6 = call i32 @llvm.smax.i32(i32 %2, i32 %3)
50  %7 = call i32 @llvm.smax.i32(i32 %6, i32 %4)
51  %8 = call i32 @llvm.smax.i32(i32 %7, i32 %5)
52  ret i32 %8
53}
54
55define i32 @smax_v8i32(i32) {
56; CHECK-LABEL: @smax_v8i32(
57; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr @arr, align 16
58; CHECK-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.smax.v8i32(<8 x i32> [[TMP2]])
59; CHECK-NEXT:    ret i32 [[TMP3]]
60;
61  %2 = load i32, ptr @arr, align 16
62  %3 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 1), align 4
63  %4 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 2), align 8
64  %5 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 3), align 4
65  %6 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 4), align 16
66  %7 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 5), align 4
67  %8 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 6), align 8
68  %9 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 7), align 4
69  %10 = call i32 @llvm.smax.i32(i32 %2,  i32 %3)
70  %11 = call i32 @llvm.smax.i32(i32 %10, i32 %4)
71  %12 = call i32 @llvm.smax.i32(i32 %11, i32 %5)
72  %13 = call i32 @llvm.smax.i32(i32 %12, i32 %6)
73  %14 = call i32 @llvm.smax.i32(i32 %13, i32 %7)
74  %15 = call i32 @llvm.smax.i32(i32 %14, i32 %8)
75  %16 = call i32 @llvm.smax.i32(i32 %15, i32 %9)
76  ret i32 %16
77}
78
79define i32 @smax_v16i32(i32) {
80; CHECK-LABEL: @smax_v16i32(
81; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @arr, align 16
82; CHECK-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.smax.v16i32(<16 x i32> [[TMP2]])
83; CHECK-NEXT:    ret i32 [[TMP3]]
84;
85  %2  = load i32, ptr @arr, align 16
86  %3  = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 1), align 4
87  %4  = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 2), align 8
88  %5  = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 3), align 4
89  %6  = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 4), align 16
90  %7  = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 5), align 4
91  %8  = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 6), align 8
92  %9  = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 7), align 4
93  %10 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 8), align 16
94  %11 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 9), align 4
95  %12 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 10), align 8
96  %13 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 11), align 4
97  %14 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 12), align 16
98  %15 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 13), align 4
99  %16 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 14), align 8
100  %17 = load i32, ptr getelementptr inbounds ([32 x i32], ptr @arr, i64 0, i64 15), align 4
101  %18 = call i32 @llvm.smax.i32(i32 %2,  i32 %3)
102  %19 = call i32 @llvm.smax.i32(i32 %18, i32 %4)
103  %20 = call i32 @llvm.smax.i32(i32 %19, i32 %5)
104  %21 = call i32 @llvm.smax.i32(i32 %20, i32 %6)
105  %22 = call i32 @llvm.smax.i32(i32 %21, i32 %7)
106  %23 = call i32 @llvm.smax.i32(i32 %22, i32 %8)
107  %24 = call i32 @llvm.smax.i32(i32 %23, i32 %9)
108  %25 = call i32 @llvm.smax.i32(i32 %24, i32 %10)
109  %26 = call i32 @llvm.smax.i32(i32 %25, i32 %11)
110  %27 = call i32 @llvm.smax.i32(i32 %26, i32 %12)
111  %28 = call i32 @llvm.smax.i32(i32 %27, i32 %13)
112  %29 = call i32 @llvm.smax.i32(i32 %28, i32 %14)
113  %30 = call i32 @llvm.smax.i32(i32 %29, i32 %15)
114  %31 = call i32 @llvm.smax.i32(i32 %30, i32 %16)
115  %32 = call i32 @llvm.smax.i32(i32 %31, i32 %17)
116  ret i32 %32
117}
118