1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -S -slp-vectorizer -instcombine -pass-remarks-output=%t | FileCheck %s
3; RUN: cat %t | FileCheck -check-prefix=REMARK %s
4; RUN: opt < %s -S -aa-pipeline=basic-aa -passes='slp-vectorizer,instcombine' -pass-remarks-output=%t | FileCheck %s
5; RUN: cat %t | FileCheck -check-prefix=REMARK %s
6
7target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
8target triple = "aarch64--linux-gnu"
9
10; REMARK-LABEL: Function: gather_multiple_use
11; REMARK:       Args:
12; REMARK-NEXT:    - String: 'Vectorized horizontal reduction with cost '
13; REMARK-NEXT:    - Cost: '-7'
14;
15; REMARK-NOT: Function: gather_load
16
17define internal i32 @gather_multiple_use(i32 %a, i32 %b, i32 %c, i32 %d) {
18; CHECK-LABEL: @gather_multiple_use(
19; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x i32> undef, i32 [[C:%.*]], i32 0
20; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x i32> [[TMP1]], i32 [[A:%.*]], i32 1
21; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x i32> [[TMP2]], i32 [[B:%.*]], i32 2
22; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x i32> [[TMP3]], i32 [[D:%.*]], i32 3
23; CHECK-NEXT:    [[TMP5:%.*]] = lshr <4 x i32> [[TMP4]], <i32 15, i32 15, i32 15, i32 15>
24; CHECK-NEXT:    [[TMP6:%.*]] = and <4 x i32> [[TMP5]], <i32 65537, i32 65537, i32 65537, i32 65537>
25; CHECK-NEXT:    [[TMP7:%.*]] = mul nuw <4 x i32> [[TMP6]], <i32 65535, i32 65535, i32 65535, i32 65535>
26; CHECK-NEXT:    [[TMP8:%.*]] = add <4 x i32> [[TMP7]], [[TMP4]]
27; CHECK-NEXT:    [[TMP9:%.*]] = xor <4 x i32> [[TMP8]], [[TMP7]]
28; CHECK-NEXT:    [[TMP10:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP9]])
29; CHECK-NEXT:    ret i32 [[TMP10]]
30;
31  %tmp00 = lshr i32 %a, 15
32  %tmp01 = and i32 %tmp00, 65537
33  %tmp02 = mul nuw i32 %tmp01, 65535
34  %tmp03 = add i32 %tmp02, %a
35  %tmp04 = xor i32 %tmp03, %tmp02
36  %tmp05 = lshr i32 %c, 15
37  %tmp06 = and i32 %tmp05, 65537
38  %tmp07 = mul nuw i32 %tmp06, 65535
39  %tmp08 = add i32 %tmp07, %c
40  %tmp09 = xor i32 %tmp08, %tmp07
41  %tmp10 = lshr i32 %b, 15
42  %tmp11 = and i32 %tmp10, 65537
43  %tmp12 = mul nuw i32 %tmp11, 65535
44  %tmp13 = add i32 %tmp12, %b
45  %tmp14 = xor i32 %tmp13, %tmp12
46  %tmp15 = lshr i32 %d, 15
47  %tmp16 = and i32 %tmp15, 65537
48  %tmp17 = mul nuw i32 %tmp16, 65535
49  %tmp18 = add i32 %tmp17, %d
50  %tmp19 = xor i32 %tmp18, %tmp17
51  %tmp20 = add i32 %tmp09, %tmp04
52  %tmp21 = add i32 %tmp20, %tmp14
53  %tmp22 = add i32 %tmp21, %tmp19
54  ret i32 %tmp22
55}
56
57@data = global [6 x [258 x i8]] zeroinitializer, align 1
58define void @gather_load(i16* noalias %ptr) {
59; CHECK-LABEL: @gather_load(
60; CHECK-NEXT:    [[ARRAYIDX182:%.*]] = getelementptr inbounds i16, i16* [[PTR:%.*]], i64 1
61; CHECK-NEXT:    [[ARRAYIDX183:%.*]] = getelementptr inbounds i16, i16* [[PTR]], i64 2
62; CHECK-NEXT:    [[ARRAYIDX184:%.*]] = getelementptr inbounds i16, i16* [[PTR]], i64 3
63; CHECK-NEXT:    [[ARRAYIDX185:%.*]] = getelementptr inbounds i16, i16* [[PTR]], i64 4
64; CHECK-NEXT:    [[L0:%.*]] = load i8, i8* getelementptr inbounds ([6 x [258 x i8]], [6 x [258 x i8]]* @data, i64 0, i64 1, i64 0), align 1
65; CHECK-NEXT:    [[CONV150:%.*]] = zext i8 [[L0]] to i16
66; CHECK-NEXT:    [[ADD152:%.*]] = add nuw nsw i16 [[CONV150]], 10
67; CHECK-NEXT:    [[L1:%.*]] = load i8, i8* getelementptr inbounds ([6 x [258 x i8]], [6 x [258 x i8]]* @data, i64 0, i64 2, i64 1), align 1
68; CHECK-NEXT:    [[CONV156:%.*]] = zext i8 [[L1]] to i16
69; CHECK-NEXT:    [[ADD158:%.*]] = add nuw nsw i16 [[CONV156]], 20
70; CHECK-NEXT:    [[L2:%.*]] = load i8, i8* getelementptr inbounds ([6 x [258 x i8]], [6 x [258 x i8]]* @data, i64 0, i64 3, i64 2), align 1
71; CHECK-NEXT:    [[CONV162:%.*]] = zext i8 [[L2]] to i16
72; CHECK-NEXT:    [[ADD164:%.*]] = add nuw nsw i16 [[CONV162]], 30
73; CHECK-NEXT:    [[L3:%.*]] = load i8, i8* getelementptr inbounds ([6 x [258 x i8]], [6 x [258 x i8]]* @data, i64 0, i64 4, i64 3), align 1
74; CHECK-NEXT:    [[CONV168:%.*]] = zext i8 [[L3]] to i16
75; CHECK-NEXT:    [[ADD170:%.*]] = add nuw nsw i16 [[CONV168]], 40
76; CHECK-NEXT:    store i16 [[ADD152]], i16* [[ARRAYIDX182]], align 2
77; CHECK-NEXT:    store i16 [[ADD158]], i16* [[ARRAYIDX183]], align 2
78; CHECK-NEXT:    store i16 [[ADD164]], i16* [[ARRAYIDX184]], align 2
79; CHECK-NEXT:    store i16 [[ADD170]], i16* [[ARRAYIDX185]], align 2
80; CHECK-NEXT:    ret void
81;
82  %arrayidx182 = getelementptr inbounds i16, i16* %ptr, i64 1
83  %arrayidx183 = getelementptr inbounds i16, i16* %ptr, i64 2
84  %arrayidx184 = getelementptr inbounds i16, i16* %ptr, i64 3
85  %arrayidx185 = getelementptr inbounds i16, i16* %ptr, i64 4
86  %arrayidx149 = getelementptr inbounds [6 x [258 x i8]], [6 x [258 x i8]]* @data, i64 0, i64 1, i64 0
87  %l0 = load i8, i8* %arrayidx149, align 1
88  %conv150 = zext i8 %l0 to i16
89  %add152 = add i16 10, %conv150
90  %arrayidx155 = getelementptr inbounds [6 x [258 x i8]], [6 x [258 x i8]]* @data, i64 0, i64 2, i64 1
91  %l1 = load i8, i8* %arrayidx155, align 1
92  %conv156 = zext i8 %l1 to i16
93  %add158 = add i16 20, %conv156
94  %arrayidx161 = getelementptr inbounds [6 x [258 x i8]], [6 x [258 x i8]]* @data, i64 0, i64 3, i64 2
95  %l2 = load i8, i8* %arrayidx161, align 1
96  %conv162 = zext i8 %l2 to i16
97  %add164 = add i16 30, %conv162
98  %arrayidx167 = getelementptr inbounds [6 x [258 x i8]], [6 x [258 x i8]]* @data, i64 0, i64 4, i64 3
99  %l3 = load i8, i8* %arrayidx167, align 1
100  %conv168 = zext i8 %l3 to i16
101  %add170 = add i16 40, %conv168
102  store i16 %add152, i16* %arrayidx182, align 2
103  store i16 %add158, i16* %arrayidx183, align 2
104  store i16 %add164, i16* %arrayidx184, align 2
105  store i16 %add170, i16* %arrayidx185, align 2
106  ret void
107}
108