1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -slp-vectorizer -S | FileCheck %s --check-prefix=DEFAULT
3; RUN: opt < %s -slp-schedule-budget=0 -slp-min-tree-size=0 -slp-threshold=-30 -slp-vectorizer -S | FileCheck %s --check-prefix=GATHER
4; RUN: opt < %s -slp-schedule-budget=0 -slp-threshold=-30 -slp-vectorizer -S | FileCheck %s --check-prefix=MAX-COST
5
6target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
7target triple = "aarch64--linux-gnu"
8
9@a = common global [80 x i8] zeroinitializer, align 16
10
11define void @PR28330(i32 %n) {
12; DEFAULT-LABEL: @PR28330(
13; DEFAULT-NEXT:  entry:
14; DEFAULT-NEXT:    [[TMP0:%.*]] = load <8 x i8>, <8 x i8>* bitcast (i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1) to <8 x i8>*), align 1
15; DEFAULT-NEXT:    [[TMP1:%.*]] = icmp eq <8 x i8> [[TMP0]], zeroinitializer
16; DEFAULT-NEXT:    br label [[FOR_BODY:%.*]]
17; DEFAULT:       for.body:
18; DEFAULT-NEXT:    [[P17:%.*]] = phi i32 [ [[OP_RDX:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
19; DEFAULT-NEXT:    [[TMP2:%.*]] = select <8 x i1> [[TMP1]], <8 x i32> <i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720>, <8 x i32> <i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80>
20; DEFAULT-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[TMP2]])
21; DEFAULT-NEXT:    [[OP_RDX]] = add i32 [[TMP3]], [[P17]]
22; DEFAULT-NEXT:    br label [[FOR_BODY]]
23;
24; GATHER-LABEL: @PR28330(
25; GATHER-NEXT:  entry:
26; GATHER-NEXT:    [[TMP0:%.*]] = load <8 x i8>, <8 x i8>* bitcast (i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1) to <8 x i8>*), align 1
27; GATHER-NEXT:    [[TMP1:%.*]] = icmp eq <8 x i8> [[TMP0]], zeroinitializer
28; GATHER-NEXT:    br label [[FOR_BODY:%.*]]
29; GATHER:       for.body:
30; GATHER-NEXT:    [[P17:%.*]] = phi i32 [ [[OP_RDX:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
31; GATHER-NEXT:    [[TMP2:%.*]] = select <8 x i1> [[TMP1]], <8 x i32> <i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720>, <8 x i32> <i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80>
32; GATHER-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[TMP2]])
33; GATHER-NEXT:    [[OP_RDX]] = add i32 [[TMP3]], [[P17]]
34; GATHER-NEXT:    br label [[FOR_BODY]]
35;
36; MAX-COST-LABEL: @PR28330(
37; MAX-COST-NEXT:  entry:
38; MAX-COST-NEXT:    [[TMP0:%.*]] = load <8 x i8>, <8 x i8>* bitcast (i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1) to <8 x i8>*), align 1
39; MAX-COST-NEXT:    [[TMP1:%.*]] = icmp eq <8 x i8> [[TMP0]], zeroinitializer
40; MAX-COST-NEXT:    br label [[FOR_BODY:%.*]]
41; MAX-COST:       for.body:
42; MAX-COST-NEXT:    [[P17:%.*]] = phi i32 [ [[OP_RDX:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
43; MAX-COST-NEXT:    [[TMP2:%.*]] = select <8 x i1> [[TMP1]], <8 x i32> <i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720>, <8 x i32> <i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80>
44; MAX-COST-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[TMP2]])
45; MAX-COST-NEXT:    [[OP_RDX]] = add i32 [[TMP3]], [[P17]]
46; MAX-COST-NEXT:    br label [[FOR_BODY]]
47;
48entry:
49  %p0 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1), align 1
50  %p1 = icmp eq i8 %p0, 0
51  %p2 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 2), align 2
52  %p3 = icmp eq i8 %p2, 0
53  %p4 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 3), align 1
54  %p5 = icmp eq i8 %p4, 0
55  %p6 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 4), align 4
56  %p7 = icmp eq i8 %p6, 0
57  %p8 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 5), align 1
58  %p9 = icmp eq i8 %p8, 0
59  %p10 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 6), align 2
60  %p11 = icmp eq i8 %p10, 0
61  %p12 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 7), align 1
62  %p13 = icmp eq i8 %p12, 0
63  %p14 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 8), align 8
64  %p15 = icmp eq i8 %p14, 0
65  br label %for.body
66
67for.body:
68  %p17 = phi i32 [ %p34, %for.body ], [ 0, %entry ]
69  %p19 = select i1 %p1, i32 -720, i32 -80
70  %p20 = add i32 %p17, %p19
71  %p21 = select i1 %p3, i32 -720, i32 -80
72  %p22 = add i32 %p20, %p21
73  %p23 = select i1 %p5, i32 -720, i32 -80
74  %p24 = add i32 %p22, %p23
75  %p25 = select i1 %p7, i32 -720, i32 -80
76  %p26 = add i32 %p24, %p25
77  %p27 = select i1 %p9, i32 -720, i32 -80
78  %p28 = add i32 %p26, %p27
79  %p29 = select i1 %p11, i32 -720, i32 -80
80  %p30 = add i32 %p28, %p29
81  %p31 = select i1 %p13, i32 -720, i32 -80
82  %p32 = add i32 %p30, %p31
83  %p33 = select i1 %p15, i32 -720, i32 -80
84  %p34 = add i32 %p32, %p33
85  br label %for.body
86}
87
88define void @PR32038(i32 %n) {
89; DEFAULT-LABEL: @PR32038(
90; DEFAULT-NEXT:  entry:
91; DEFAULT-NEXT:    [[TMP0:%.*]] = load <8 x i8>, <8 x i8>* bitcast (i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1) to <8 x i8>*), align 1
92; DEFAULT-NEXT:    [[TMP1:%.*]] = icmp eq <8 x i8> [[TMP0]], zeroinitializer
93; DEFAULT-NEXT:    br label [[FOR_BODY:%.*]]
94; DEFAULT:       for.body:
95; DEFAULT-NEXT:    [[P17:%.*]] = phi i32 [ [[OP_RDX:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
96; DEFAULT-NEXT:    [[TMP2:%.*]] = select <8 x i1> [[TMP1]], <8 x i32> <i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720>, <8 x i32> <i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80>
97; DEFAULT-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[TMP2]])
98; DEFAULT-NEXT:    [[OP_RDX]] = add i32 [[TMP3]], -5
99; DEFAULT-NEXT:    br label [[FOR_BODY]]
100;
101; GATHER-LABEL: @PR32038(
102; GATHER-NEXT:  entry:
103; GATHER-NEXT:    [[TMP0:%.*]] = load <8 x i8>, <8 x i8>* bitcast (i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1) to <8 x i8>*), align 1
104; GATHER-NEXT:    [[TMP1:%.*]] = icmp eq <8 x i8> [[TMP0]], zeroinitializer
105; GATHER-NEXT:    br label [[FOR_BODY:%.*]]
106; GATHER:       for.body:
107; GATHER-NEXT:    [[P17:%.*]] = phi i32 [ [[OP_RDX:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
108; GATHER-NEXT:    [[TMP2:%.*]] = select <8 x i1> [[TMP1]], <8 x i32> <i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720>, <8 x i32> <i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80>
109; GATHER-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[TMP2]])
110; GATHER-NEXT:    [[OP_RDX]] = add i32 [[TMP3]], -5
111; GATHER-NEXT:    br label [[FOR_BODY]]
112;
113; MAX-COST-LABEL: @PR32038(
114; MAX-COST-NEXT:  entry:
115; MAX-COST-NEXT:    [[TMP0:%.*]] = load <8 x i8>, <8 x i8>* bitcast (i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1) to <8 x i8>*), align 1
116; MAX-COST-NEXT:    [[TMP1:%.*]] = icmp eq <8 x i8> [[TMP0]], zeroinitializer
117; MAX-COST-NEXT:    br label [[FOR_BODY:%.*]]
118; MAX-COST:       for.body:
119; MAX-COST-NEXT:    [[P17:%.*]] = phi i32 [ [[OP_RDX:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
120; MAX-COST-NEXT:    [[TMP2:%.*]] = select <8 x i1> [[TMP1]], <8 x i32> <i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720, i32 -720>, <8 x i32> <i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80, i32 -80>
121; MAX-COST-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[TMP2]])
122; MAX-COST-NEXT:    [[OP_RDX]] = add i32 [[TMP3]], -5
123; MAX-COST-NEXT:    br label [[FOR_BODY]]
124;
125entry:
126  %p0 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1), align 1
127  %p1 = icmp eq i8 %p0, 0
128  %p2 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 2), align 2
129  %p3 = icmp eq i8 %p2, 0
130  %p4 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 3), align 1
131  %p5 = icmp eq i8 %p4, 0
132  %p6 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 4), align 4
133  %p7 = icmp eq i8 %p6, 0
134  %p8 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 5), align 1
135  %p9 = icmp eq i8 %p8, 0
136  %p10 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 6), align 2
137  %p11 = icmp eq i8 %p10, 0
138  %p12 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 7), align 1
139  %p13 = icmp eq i8 %p12, 0
140  %p14 = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 8), align 8
141  %p15 = icmp eq i8 %p14, 0
142  br label %for.body
143
144for.body:
145  %p17 = phi i32 [ %p34, %for.body ], [ 0, %entry ]
146  %p19 = select i1 %p1, i32 -720, i32 -80
147  %p20 = add i32 -5, %p19
148  %p21 = select i1 %p3, i32 -720, i32 -80
149  %p22 = add i32 %p20, %p21
150  %p23 = select i1 %p5, i32 -720, i32 -80
151  %p24 = add i32 %p22, %p23
152  %p25 = select i1 %p7, i32 -720, i32 -80
153  %p26 = add i32 %p24, %p25
154  %p27 = select i1 %p9, i32 -720, i32 -80
155  %p28 = add i32 %p26, %p27
156  %p29 = select i1 %p11, i32 -720, i32 -80
157  %p30 = add i32 %p28, %p29
158  %p31 = select i1 %p13, i32 -720, i32 -80
159  %p32 = add i32 %p30, %p31
160  %p33 = select i1 %p15, i32 -720, i32 -80
161  %p34 = add i32 %p32, %p33
162  br label %for.body
163}
164