1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s  -loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -dce -instcombine -S | FileCheck %s
3
4target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
5target triple = "x86_64-unknown-linux-gnu"
6
7; A tricky loop:
8;
9; void loop(int *a, int *b) {
10;    for (int i = 0; i < 512; ++i) {
11;        a[a[i]] = b[i];
12;        a[i] = b[i+1];
13;    }
14;}
15
16define void @loop(i32* nocapture %a, i32* nocapture %b) nounwind uwtable {
17; CHECK-LABEL: @loop(
18; CHECK-NEXT:  entry:
19; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
20; CHECK:       for.body:
21; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
22; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[INDVARS_IV]]
23; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
24; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[INDVARS_IV]]
25; CHECK-NEXT:    [[TMP1:%.*]] = load i32, i32* [[ARRAYIDX2]], align 4
26; CHECK-NEXT:    [[IDXPROM3:%.*]] = sext i32 [[TMP1]] to i64
27; CHECK-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[IDXPROM3]]
28; CHECK-NEXT:    store i32 [[TMP0]], i32* [[ARRAYIDX4]], align 4
29; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
30; CHECK-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDVARS_IV_NEXT]]
31; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* [[ARRAYIDX6]], align 4
32; CHECK-NEXT:    store i32 [[TMP2]], i32* [[ARRAYIDX2]], align 4
33; CHECK-NEXT:    [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
34; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], 512
35; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[FOR_BODY]]
36; CHECK:       for.end:
37; CHECK-NEXT:    ret void
38;
39entry:
40  br label %for.body
41
42for.body:                                         ; preds = %for.body, %entry
43  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
44  %arrayidx = getelementptr inbounds i32, i32* %b, i64 %indvars.iv
45  %0 = load i32, i32* %arrayidx, align 4
46  %arrayidx2 = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
47  %1 = load i32, i32* %arrayidx2, align 4
48  %idxprom3 = sext i32 %1 to i64
49  %arrayidx4 = getelementptr inbounds i32, i32* %a, i64 %idxprom3
50  store i32 %0, i32* %arrayidx4, align 4
51  %indvars.iv.next = add i64 %indvars.iv, 1
52  %arrayidx6 = getelementptr inbounds i32, i32* %b, i64 %indvars.iv.next
53  %2 = load i32, i32* %arrayidx6, align 4
54  store i32 %2, i32* %arrayidx2, align 4
55  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
56  %exitcond = icmp eq i32 %lftr.wideiv, 512
57  br i1 %exitcond, label %for.end, label %for.body
58
59for.end:                                          ; preds = %for.body
60  ret void
61}
62
63; The same loop with parallel loop metadata added to the loop branch
64; and the memory instructions.
65
66define void @parallel_loop(i32* nocapture %a, i32* nocapture %b) nounwind uwtable {
67; CHECK-LABEL: @parallel_loop(
68; CHECK-NEXT:  entry:
69; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
70; CHECK:       vector.ph:
71; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
72; CHECK:       vector.body:
73; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
74; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[INDEX]]
75; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[TMP0]] to <4 x i32>*
76; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i32>, <4 x i32>* [[TMP1]], align 4, !llvm.access.group !0
77; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[INDEX]]
78; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <4 x i32>*
79; CHECK-NEXT:    [[WIDE_LOAD1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP3]], align 4, !llvm.access.group !0
80; CHECK-NEXT:    [[TMP4:%.*]] = sext <4 x i32> [[WIDE_LOAD1]] to <4 x i64>
81; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x i64> [[TMP4]], i32 0
82; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP5]]
83; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x i64> [[TMP4]], i32 1
84; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP7]]
85; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <4 x i64> [[TMP4]], i32 2
86; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP9]]
87; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <4 x i64> [[TMP4]], i32 3
88; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP11]]
89; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <4 x i32> [[WIDE_LOAD]], i32 0
90; CHECK-NEXT:    store i32 [[TMP13]], i32* [[TMP6]], align 4, !llvm.access.group !1
91; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <4 x i32> [[WIDE_LOAD]], i32 1
92; CHECK-NEXT:    store i32 [[TMP14]], i32* [[TMP8]], align 4, !llvm.access.group !1
93; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <4 x i32> [[WIDE_LOAD]], i32 2
94; CHECK-NEXT:    store i32 [[TMP15]], i32* [[TMP10]], align 4, !llvm.access.group !1
95; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <4 x i32> [[WIDE_LOAD]], i32 3
96; CHECK-NEXT:    store i32 [[TMP16]], i32* [[TMP12]], align 4, !llvm.access.group !1
97; CHECK-NEXT:    [[TMP17:%.*]] = or i64 [[INDEX]], 1
98; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[TMP17]]
99; CHECK-NEXT:    [[TMP19:%.*]] = bitcast i32* [[TMP18]] to <4 x i32>*
100; CHECK-NEXT:    [[WIDE_LOAD2:%.*]] = load <4 x i32>, <4 x i32>* [[TMP19]], align 4, !llvm.access.group !0
101; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i32* [[TMP2]] to <4 x i32>*
102; CHECK-NEXT:    store <4 x i32> [[WIDE_LOAD2]], <4 x i32>* [[TMP20]], align 4, !llvm.access.group !0
103; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
104; CHECK-NEXT:    [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], 512
105; CHECK-NEXT:    br i1 [[TMP21]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]]
106; CHECK:       middle.block:
107; CHECK-NEXT:    br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]]
108; CHECK:       scalar.ph:
109; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
110; CHECK:       for.body:
111; CHECK-NEXT:    br i1 undef, label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
112; CHECK:       for.end:
113; CHECK-NEXT:    ret void
114;
115entry:
116  br label %for.body
117
118for.body:                                         ; preds = %for.body, %entry
119  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
120  %arrayidx = getelementptr inbounds i32, i32* %b, i64 %indvars.iv
121  %0 = load i32, i32* %arrayidx, align 4, !llvm.access.group !13
122  %arrayidx2 = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
123  %1 = load i32, i32* %arrayidx2, align 4, !llvm.access.group !13
124  %idxprom3 = sext i32 %1 to i64
125  %arrayidx4 = getelementptr inbounds i32, i32* %a, i64 %idxprom3
126  ; This store might have originated from inlining a function with a parallel
127  ; loop. Refers to a list with the "original loop reference" (!4) also included.
128  store i32 %0, i32* %arrayidx4, align 4, !llvm.access.group !15
129  %indvars.iv.next = add i64 %indvars.iv, 1
130  %arrayidx6 = getelementptr inbounds i32, i32* %b, i64 %indvars.iv.next
131  %2 = load i32, i32* %arrayidx6, align 4, !llvm.access.group !13
132  store i32 %2, i32* %arrayidx2, align 4, !llvm.access.group !13
133  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
134  %exitcond = icmp eq i32 %lftr.wideiv, 512
135  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !3
136
137for.end:                                          ; preds = %for.body
138  ret void
139}
140
141; The same loop with an illegal parallel loop metadata: the memory
142; accesses refer to a different loop's identifier.
143
144
145define void @mixed_metadata(i32* nocapture %a, i32* nocapture %b) nounwind uwtable {
146; CHECK-LABEL: @mixed_metadata(
147; CHECK-NEXT:  entry:
148; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
149; CHECK:       for.body:
150; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
151; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[INDVARS_IV]]
152; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[ARRAYIDX]], align 4, !llvm.access.group !7
153; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[INDVARS_IV]]
154; CHECK-NEXT:    [[TMP1:%.*]] = load i32, i32* [[ARRAYIDX2]], align 4, !llvm.access.group !7
155; CHECK-NEXT:    [[IDXPROM3:%.*]] = sext i32 [[TMP1]] to i64
156; CHECK-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[IDXPROM3]]
157; CHECK-NEXT:    store i32 [[TMP0]], i32* [[ARRAYIDX4]], align 4, !llvm.access.group !8
158; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
159; CHECK-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDVARS_IV_NEXT]]
160; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* [[ARRAYIDX6]], align 4, !llvm.access.group !7
161; CHECK-NEXT:    store i32 [[TMP2]], i32* [[ARRAYIDX2]], align 4, !llvm.access.group !7
162; CHECK-NEXT:    [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
163; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], 512
164; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
165; CHECK:       for.end:
166; CHECK-NEXT:    ret void
167;
168entry:
169  br label %for.body
170
171for.body:                                         ; preds = %for.body, %entry
172  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
173  %arrayidx = getelementptr inbounds i32, i32* %b, i64 %indvars.iv
174  %0 = load i32, i32* %arrayidx, align 4, !llvm.access.group !16
175  %arrayidx2 = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
176  %1 = load i32, i32* %arrayidx2, align 4, !llvm.access.group !16
177  %idxprom3 = sext i32 %1 to i64
178  %arrayidx4 = getelementptr inbounds i32, i32* %a, i64 %idxprom3
179  ; This refers to the loop marked with !7 which we are not in at the moment.
180  ; It should prevent detecting as a parallel loop.
181  store i32 %0, i32* %arrayidx4, align 4, !llvm.access.group !17
182  %indvars.iv.next = add i64 %indvars.iv, 1
183  %arrayidx6 = getelementptr inbounds i32, i32* %b, i64 %indvars.iv.next
184  %2 = load i32, i32* %arrayidx6, align 4, !llvm.access.group !16
185  store i32 %2, i32* %arrayidx2, align 4, !llvm.access.group !16
186  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
187  %exitcond = icmp eq i32 %lftr.wideiv, 512
188  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !6
189
190for.end:                                          ; preds = %for.body
191  ret void
192}
193
194!3 = !{!3, !{!"llvm.loop.parallel_accesses", !13, !15}}
195!4 = !{!4, !{!"llvm.loop.parallel_accesses", !14, !15}}
196!6 = !{!6, !{!"llvm.loop.parallel_accesses", !16}}
197!7 = !{!7, !{!"llvm.loop.parallel_accesses", !17}}
198!13 = distinct !{}
199!14 = distinct !{}
200!15 = distinct !{}
201!16 = distinct !{}
202!17 = distinct !{}
203