1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -slp-vectorizer -slp-vectorize-hor -slp-vectorize-hor-store -S < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver2 | FileCheck %s
3
4define void @i64_simplified(i64* noalias %st, i64* noalias %ld) {
5; CHECK-LABEL: @i64_simplified(
6; CHECK-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds i64, i64* [[LD:%.*]], i64 1
7; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64* [[LD]] to <2 x i64>*
8; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x i64>, <2 x i64>* [[TMP1]], align 8
9; CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i64> [[TMP2]], <2 x i64> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
10; CHECK-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr inbounds i64, i64* [[ST:%.*]], i64 1
11; CHECK-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds i64, i64* [[ST]], i64 2
12; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i64, i64* [[ST]], i64 3
13; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i64* [[ST]] to <4 x i64>*
14; CHECK-NEXT:    store <4 x i64> [[SHUFFLE]], <4 x i64>* [[TMP3]], align 8
15; CHECK-NEXT:    ret void
16;
17  %arrayidx1 = getelementptr inbounds i64, i64* %ld, i64 1
18
19  %t0 = load i64, i64* %ld, align 8
20  %t1 = load i64, i64* %arrayidx1, align 8
21
22  %arrayidx3 = getelementptr inbounds i64, i64* %st, i64 1
23  %arrayidx4 = getelementptr inbounds i64, i64* %st, i64 2
24  %arrayidx5 = getelementptr inbounds i64, i64* %st, i64 3
25
26  store i64 %t0, i64* %st, align 8
27  store i64 %t1, i64* %arrayidx3, align 8
28  store i64 %t0, i64* %arrayidx4, align 8
29  store i64 %t1, i64* %arrayidx5, align 8
30  ret void
31}
32
33define void @i64_simplifiedi_reversed(i64* noalias %st, i64* noalias %ld) {
34; CHECK-LABEL: @i64_simplifiedi_reversed(
35; CHECK-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds i64, i64* [[LD:%.*]], i64 1
36; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64* [[LD]] to <2 x i64>*
37; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x i64>, <2 x i64>* [[TMP1]], align 8
38; CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i64> [[TMP2]], <2 x i64> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0>
39; CHECK-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr inbounds i64, i64* [[ST:%.*]], i64 1
40; CHECK-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds i64, i64* [[ST]], i64 2
41; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i64, i64* [[ST]], i64 3
42; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i64* [[ST]] to <4 x i64>*
43; CHECK-NEXT:    store <4 x i64> [[SHUFFLE]], <4 x i64>* [[TMP3]], align 8
44; CHECK-NEXT:    ret void
45;
46  %arrayidx1 = getelementptr inbounds i64, i64* %ld, i64 1
47
48  %t0 = load i64, i64* %ld, align 8
49  %t1 = load i64, i64* %arrayidx1, align 8
50
51  %arrayidx3 = getelementptr inbounds i64, i64* %st, i64 1
52  %arrayidx4 = getelementptr inbounds i64, i64* %st, i64 2
53  %arrayidx5 = getelementptr inbounds i64, i64* %st, i64 3
54
55  store i64 %t1, i64* %st, align 8
56  store i64 %t0, i64* %arrayidx3, align 8
57  store i64 %t1, i64* %arrayidx4, align 8
58  store i64 %t0, i64* %arrayidx5, align 8
59  ret void
60}
61
62define void @i64_simplifiedi_extract(i64* noalias %st, i64* noalias %ld) {
63; CHECK-LABEL: @i64_simplifiedi_extract(
64; CHECK-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds i64, i64* [[LD:%.*]], i64 1
65; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64* [[LD]] to <2 x i64>*
66; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x i64>, <2 x i64>* [[TMP1]], align 8
67; CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i64> [[TMP2]], <2 x i64> poison, <4 x i32> <i32 0, i32 0, i32 0, i32 1>
68; CHECK-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr inbounds i64, i64* [[ST:%.*]], i64 1
69; CHECK-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds i64, i64* [[ST]], i64 2
70; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i64, i64* [[ST]], i64 3
71; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i64* [[ST]] to <4 x i64>*
72; CHECK-NEXT:    store <4 x i64> [[SHUFFLE]], <4 x i64>* [[TMP3]], align 8
73; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <4 x i64> [[SHUFFLE]], i32 3
74; CHECK-NEXT:    store i64 [[TMP4]], i64* [[LD]], align 8
75; CHECK-NEXT:    ret void
76;
77  %arrayidx1 = getelementptr inbounds i64, i64* %ld, i64 1
78
79  %t0 = load i64, i64* %ld, align 8
80  %t1 = load i64, i64* %arrayidx1, align 8
81
82  %arrayidx3 = getelementptr inbounds i64, i64* %st, i64 1
83  %arrayidx4 = getelementptr inbounds i64, i64* %st, i64 2
84  %arrayidx5 = getelementptr inbounds i64, i64* %st, i64 3
85
86  store i64 %t0, i64* %st, align 8
87  store i64 %t0, i64* %arrayidx3, align 8
88  store i64 %t0, i64* %arrayidx4, align 8
89  store i64 %t1, i64* %arrayidx5, align 8
90  store i64 %t1, i64* %ld, align 8
91  ret void
92}
93
94