xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/X86/jumbled-load-shuffle-placement.ll (revision 580210a0c938531ef9fd79f9ffedb93eeb2e66c2)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -S -mtriple=x86_64-unknown -mattr=+avx -passes=slp-vectorizer | FileCheck %s
3
4;void jumble (int * restrict A, int * restrict B) {
5  ;  int tmp0 = Aptr A[0];
6  ;  int tmp1 = Aptr A[1];
7  ;  int tmp2 = Aptr A[3];
8  ;  int tmp3 = Aptr A[2];
9  ;  B[0] = tmp0;
10  ;  B[1] = tmp1;
11  ;  B[2] = tmp2;
12  ;  B[3] = tmp3;
13  ;}
14
15; Function Attrs: norecurse nounwind uwtable
16define void @jumble1(ptr noalias nocapture readonly %A, ptr noalias nocapture %B) {
17; CHECK-LABEL: @jumble1(
18; CHECK-NEXT:  entry:
19; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 10
20; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr [[ARRAYIDX]], align 4
21; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr [[A]], align 4
22; CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[TMP3]], <4 x i32> poison, <4 x i32> <i32 0, i32 1, i32 3, i32 2>
23; CHECK-NEXT:    [[TMP4:%.*]] = mul nsw <4 x i32> [[TMP1]], [[SHUFFLE]]
24; CHECK-NEXT:    store <4 x i32> [[TMP4]], ptr [[B:%.*]], align 4
25; CHECK-NEXT:    ret void
26;
27entry:
28  %arrayidx = getelementptr inbounds i32, ptr %A, i64 10
29  %0 = load i32, ptr %arrayidx, align 4
30  %1 = load i32, ptr %A, align 4
31  %mul = mul nsw i32 %0, %1
32  %arrayidx2 = getelementptr inbounds i32, ptr %A, i64 11
33  %2 = load i32, ptr %arrayidx2, align 4
34  %arrayidx3 = getelementptr inbounds i32, ptr %A, i64 1
35  %3 = load i32, ptr %arrayidx3, align 4
36  %mul4 = mul nsw i32 %2, %3
37  %arrayidx5 = getelementptr inbounds i32, ptr %A, i64 12
38  %4 = load i32, ptr %arrayidx5, align 4
39  %arrayidx6 = getelementptr inbounds i32, ptr %A, i64 3
40  %5 = load i32, ptr %arrayidx6, align 4
41  %mul7 = mul nsw i32 %4, %5
42  %arrayidx8 = getelementptr inbounds i32, ptr %A, i64 13
43  %6 = load i32, ptr %arrayidx8, align 4
44  %arrayidx9 = getelementptr inbounds i32, ptr %A, i64 2
45  %7 = load i32, ptr %arrayidx9, align 4
46  %mul10 = mul nsw i32 %6, %7
47  store i32 %mul, ptr %B, align 4
48  %arrayidx12 = getelementptr inbounds i32, ptr %B, i64 1
49  store i32 %mul4, ptr %arrayidx12, align 4
50  %arrayidx13 = getelementptr inbounds i32, ptr %B, i64 2
51  store i32 %mul7, ptr %arrayidx13, align 4
52  %arrayidx14 = getelementptr inbounds i32, ptr %B, i64 3
53  store i32 %mul10, ptr %arrayidx14, align 4
54  ret void
55}
56
57;Reversing the operand of MUL
58
59; Function Attrs: norecurse nounwind uwtable
60define void @jumble2(ptr noalias nocapture readonly %A, ptr noalias nocapture %B) {
61; CHECK-LABEL: @jumble2(
62; CHECK-NEXT:  entry:
63; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 10
64; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr [[ARRAYIDX]], align 4
65; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr [[A]], align 4
66; CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[TMP3]], <4 x i32> poison, <4 x i32> <i32 0, i32 1, i32 3, i32 2>
67; CHECK-NEXT:    [[TMP4:%.*]] = mul nsw <4 x i32> [[SHUFFLE]], [[TMP1]]
68; CHECK-NEXT:    store <4 x i32> [[TMP4]], ptr [[B:%.*]], align 4
69; CHECK-NEXT:    ret void
70;
71entry:
72  %arrayidx = getelementptr inbounds i32, ptr %A, i64 10
73  %0 = load i32, ptr %arrayidx, align 4
74  %1 = load i32, ptr %A, align 4
75  %mul = mul nsw i32 %1, %0
76  %arrayidx2 = getelementptr inbounds i32, ptr %A, i64 11
77  %2 = load i32, ptr %arrayidx2, align 4
78  %arrayidx3 = getelementptr inbounds i32, ptr %A, i64 1
79  %3 = load i32, ptr %arrayidx3, align 4
80  %mul4 = mul nsw i32 %3, %2
81  %arrayidx5 = getelementptr inbounds i32, ptr %A, i64 12
82  %4 = load i32, ptr %arrayidx5, align 4
83  %arrayidx6 = getelementptr inbounds i32, ptr %A, i64 3
84  %5 = load i32, ptr %arrayidx6, align 4
85  %mul7 = mul nsw i32 %5, %4
86  %arrayidx8 = getelementptr inbounds i32, ptr %A, i64 13
87  %6 = load i32, ptr %arrayidx8, align 4
88  %arrayidx9 = getelementptr inbounds i32, ptr %A, i64 2
89  %7 = load i32, ptr %arrayidx9, align 4
90  %mul10 = mul nsw i32 %7, %6
91  store i32 %mul, ptr %B, align 4
92  %arrayidx12 = getelementptr inbounds i32, ptr %B, i64 1
93  store i32 %mul4, ptr %arrayidx12, align 4
94  %arrayidx13 = getelementptr inbounds i32, ptr %B, i64 2
95  store i32 %mul7, ptr %arrayidx13, align 4
96  %arrayidx14 = getelementptr inbounds i32, ptr %B, i64 3
97  store i32 %mul10, ptr %arrayidx14, align 4
98  ret void
99}
100
101