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 -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; PR15794
8; incorrect addition of llvm.mem.parallel_loop_access metadata is undefined
9; behaviour. Vectorizer ignores the memory dependency checks and goes ahead and
10; vectorizes this loop with uniform stores which has an output dependency.
11
12; void foo(int *a, int *b, int k, int m) {
13;   for (int i = 0; i < m; i++) {
14;     for (int j = 0; j < m; j++) {
15;       a[i] = a[i + j + k] + 1; <<<
16;     }
17;     b[i] = b[i] + 3;
18;   }
19; }
20
21; Function Attrs: nounwind uwtable
22define void @foo(i32* nocapture %a, i32* nocapture %b, i32 %k, i32 %m) #0 {
23; CHECK-LABEL: @foo(
24; CHECK-NEXT:  entry:
25; CHECK-NEXT:    [[CMP27:%.*]] = icmp sgt i32 [[M:%.*]], 0
26; CHECK-NEXT:    br i1 [[CMP27]], label [[FOR_BODY3_LR_PH_US_PREHEADER:%.*]], label [[FOR_END15:%.*]]
27; CHECK:       for.body3.lr.ph.us.preheader:
28; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[M]], -1
29; CHECK-NEXT:    [[TMP1:%.*]] = zext i32 [[K:%.*]] to i64
30; CHECK-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP0]] to i64
31; CHECK-NEXT:    [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 1
32; CHECK-NEXT:    br label [[FOR_BODY3_LR_PH_US:%.*]]
33; CHECK:       for.end.us:
34; CHECK-NEXT:    [[ARRAYIDX9_US:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[INDVARS_IV33:%.*]]
35; CHECK-NEXT:    [[TMP4:%.*]] = load i32, i32* [[ARRAYIDX9_US]], align 4, !llvm.mem.parallel_loop_access !0
36; CHECK-NEXT:    [[ADD10_US:%.*]] = add nsw i32 [[TMP4]], 3
37; CHECK-NEXT:    store i32 [[ADD10_US]], i32* [[ARRAYIDX9_US]], align 4, !llvm.mem.parallel_loop_access !0
38; CHECK-NEXT:    [[INDVARS_IV_NEXT34:%.*]] = add i64 [[INDVARS_IV33]], 1
39; CHECK-NEXT:    [[LFTR_WIDEIV35:%.*]] = trunc i64 [[INDVARS_IV_NEXT34]] to i32
40; CHECK-NEXT:    [[EXITCOND36:%.*]] = icmp eq i32 [[LFTR_WIDEIV35]], [[M]]
41; CHECK-NEXT:    br i1 [[EXITCOND36]], label [[FOR_END15_LOOPEXIT:%.*]], label [[FOR_BODY3_LR_PH_US]], !llvm.loop [[LOOP2:![0-9]+]]
42; CHECK:       for.body3.us:
43; CHECK-NEXT:    [[INDVARS_IV29:%.*]] = phi i64 [ [[BC_RESUME_VAL:%.*]], [[SCALAR_PH:%.*]] ], [ [[INDVARS_IV_NEXT30:%.*]], [[FOR_BODY3_US:%.*]] ]
44; CHECK-NEXT:    [[TMP5:%.*]] = trunc i64 [[INDVARS_IV29]] to i32
45; CHECK-NEXT:    [[ADD4_US:%.*]] = add i32 [[ADD_US:%.*]], [[TMP5]]
46; CHECK-NEXT:    [[IDXPROM_US:%.*]] = sext i32 [[ADD4_US]] to i64
47; CHECK-NEXT:    [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[IDXPROM_US]]
48; CHECK-NEXT:    [[TMP6:%.*]] = load i32, i32* [[ARRAYIDX_US]], align 4, !llvm.mem.parallel_loop_access !0
49; CHECK-NEXT:    [[ADD5_US:%.*]] = add nsw i32 [[TMP6]], 1
50; CHECK-NEXT:    store i32 [[ADD5_US]], i32* [[ARRAYIDX7_US:%.*]], align 4, !llvm.mem.parallel_loop_access !0
51; CHECK-NEXT:    [[INDVARS_IV_NEXT30]] = add i64 [[INDVARS_IV29]], 1
52; CHECK-NEXT:    [[LFTR_WIDEIV31:%.*]] = trunc i64 [[INDVARS_IV_NEXT30]] to i32
53; CHECK-NEXT:    [[EXITCOND32:%.*]] = icmp eq i32 [[LFTR_WIDEIV31]], [[M]]
54; CHECK-NEXT:    br i1 [[EXITCOND32]], label [[FOR_END_US:%.*]], label [[FOR_BODY3_US]], !llvm.loop [[LOOP3:![0-9]+]]
55; CHECK:       for.body3.lr.ph.us:
56; CHECK-NEXT:    [[INDVARS_IV33]] = phi i64 [ [[INDVARS_IV_NEXT34]], [[FOR_END_US]] ], [ 0, [[FOR_BODY3_LR_PH_US_PREHEADER]] ]
57; CHECK-NEXT:    [[TMP7:%.*]] = add i64 [[TMP1]], [[INDVARS_IV33]]
58; CHECK-NEXT:    [[TMP8:%.*]] = trunc i64 [[TMP7]] to i32
59; CHECK-NEXT:    [[TMP9:%.*]] = trunc i64 [[INDVARS_IV33]] to i32
60; CHECK-NEXT:    [[ADD_US]] = add i32 [[TMP9]], [[K]]
61; CHECK-NEXT:    [[ARRAYIDX7_US]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV33]]
62; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 4
63; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH]], label [[VECTOR_SCEVCHECK:%.*]]
64; CHECK:       vector.scevcheck:
65; CHECK-NEXT:    [[TMP10:%.*]] = add i32 [[TMP8]], [[TMP0]]
66; CHECK-NEXT:    [[TMP11:%.*]] = sub i32 [[TMP8]], [[TMP0]]
67; CHECK-NEXT:    [[TMP12:%.*]] = icmp sgt i32 [[TMP11]], [[TMP8]]
68; CHECK-NEXT:    [[TMP13:%.*]] = icmp slt i32 [[TMP10]], [[TMP8]]
69; CHECK-NEXT:    [[TMP14:%.*]] = select i1 false, i1 [[TMP12]], i1 [[TMP13]]
70; CHECK-NEXT:    br i1 [[TMP14]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
71; CHECK:       vector.ph:
72; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 4
73; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]
74; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
75; CHECK:       vector.body:
76; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
77; CHECK-NEXT:    [[TMP16:%.*]] = trunc i64 [[INDEX]] to i32
78; CHECK-NEXT:    [[TMP17:%.*]] = add i32 [[TMP16]], 0
79; CHECK-NEXT:    [[TMP18:%.*]] = add i32 [[ADD_US]], [[TMP17]]
80; CHECK-NEXT:    [[TMP19:%.*]] = sext i32 [[TMP18]] to i64
81; CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP19]]
82; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i32, i32* [[TMP20]], i32 0
83; CHECK-NEXT:    [[TMP22:%.*]] = bitcast i32* [[TMP21]] to <4 x i32>*
84; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i32>, <4 x i32>* [[TMP22]], align 4
85; CHECK-NEXT:    [[TMP23:%.*]] = add nsw <4 x i32> [[WIDE_LOAD]], <i32 1, i32 1, i32 1, i32 1>
86; CHECK-NEXT:    [[TMP24:%.*]] = extractelement <4 x i32> [[TMP23]], i32 0
87; CHECK-NEXT:    store i32 [[TMP24]], i32* [[ARRAYIDX7_US]], align 4, !llvm.mem.parallel_loop_access !0
88; CHECK-NEXT:    [[TMP25:%.*]] = extractelement <4 x i32> [[TMP23]], i32 1
89; CHECK-NEXT:    store i32 [[TMP25]], i32* [[ARRAYIDX7_US]], align 4, !llvm.mem.parallel_loop_access !0
90; CHECK-NEXT:    [[TMP26:%.*]] = extractelement <4 x i32> [[TMP23]], i32 2
91; CHECK-NEXT:    store i32 [[TMP26]], i32* [[ARRAYIDX7_US]], align 4, !llvm.mem.parallel_loop_access !0
92; CHECK-NEXT:    [[TMP27:%.*]] = extractelement <4 x i32> [[TMP23]], i32 3
93; CHECK-NEXT:    store i32 [[TMP27]], i32* [[ARRAYIDX7_US]], align 4, !llvm.mem.parallel_loop_access !0
94; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
95; CHECK-NEXT:    [[TMP28:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
96; CHECK-NEXT:    br i1 [[TMP28]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
97; CHECK:       middle.block:
98; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]
99; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END_US]], label [[SCALAR_PH]]
100; CHECK:       scalar.ph:
101; CHECK-NEXT:    [[BC_RESUME_VAL]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY3_LR_PH_US]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
102; CHECK-NEXT:    br label [[FOR_BODY3_US]]
103; CHECK:       for.end15.loopexit:
104; CHECK-NEXT:    br label [[FOR_END15]]
105; CHECK:       for.end15:
106; CHECK-NEXT:    ret void
107;
108entry:
109  %cmp27 = icmp sgt i32 %m, 0
110  br i1 %cmp27, label %for.body3.lr.ph.us, label %for.end15
111
112for.end.us:                                       ; preds = %for.body3.us
113  %arrayidx9.us = getelementptr inbounds i32, i32* %b, i64 %indvars.iv33
114  %0 = load i32, i32* %arrayidx9.us, align 4, !llvm.mem.parallel_loop_access !3
115  %add10.us = add nsw i32 %0, 3
116  store i32 %add10.us, i32* %arrayidx9.us, align 4, !llvm.mem.parallel_loop_access !3
117  %indvars.iv.next34 = add i64 %indvars.iv33, 1
118  %lftr.wideiv35 = trunc i64 %indvars.iv.next34 to i32
119  %exitcond36 = icmp eq i32 %lftr.wideiv35, %m
120  br i1 %exitcond36, label %for.end15, label %for.body3.lr.ph.us, !llvm.loop !5
121
122for.body3.us:                                     ; preds = %for.body3.us, %for.body3.lr.ph.us
123  %indvars.iv29 = phi i64 [ 0, %for.body3.lr.ph.us ], [ %indvars.iv.next30, %for.body3.us ]
124  %1 = trunc i64 %indvars.iv29 to i32
125  %add4.us = add i32 %add.us, %1
126  %idxprom.us = sext i32 %add4.us to i64
127  %arrayidx.us = getelementptr inbounds i32, i32* %a, i64 %idxprom.us
128  %2 = load i32, i32* %arrayidx.us, align 4, !llvm.mem.parallel_loop_access !3
129  %add5.us = add nsw i32 %2, 1
130  store i32 %add5.us, i32* %arrayidx7.us, align 4, !llvm.mem.parallel_loop_access !3
131  %indvars.iv.next30 = add i64 %indvars.iv29, 1
132  %lftr.wideiv31 = trunc i64 %indvars.iv.next30 to i32
133  %exitcond32 = icmp eq i32 %lftr.wideiv31, %m
134  br i1 %exitcond32, label %for.end.us, label %for.body3.us, !llvm.loop !4
135
136for.body3.lr.ph.us:                               ; preds = %for.end.us, %entry
137  %indvars.iv33 = phi i64 [ %indvars.iv.next34, %for.end.us ], [ 0, %entry ]
138  %3 = trunc i64 %indvars.iv33 to i32
139  %add.us = add i32 %3, %k
140  %arrayidx7.us = getelementptr inbounds i32, i32* %a, i64 %indvars.iv33
141  br label %for.body3.us
142
143for.end15:                                        ; preds = %for.end.us, %entry
144  ret void
145}
146
147; Same test as above, but without the invalid parallel_loop_access metadata.
148
149; Here we can see the vectorizer does the mem dep checks and decides it is
150; unsafe to vectorize.
151define void @no-par-mem-metadata(i32* nocapture %a, i32* nocapture %b, i32 %k, i32 %m) #0 {
152; CHECK-LABEL: @no-par-mem-metadata(
153; CHECK-NEXT:  entry:
154; CHECK-NEXT:    [[CMP27:%.*]] = icmp sgt i32 [[M:%.*]], 0
155; CHECK-NEXT:    br i1 [[CMP27]], label [[FOR_BODY3_LR_PH_US_PREHEADER:%.*]], label [[FOR_END15:%.*]]
156; CHECK:       for.body3.lr.ph.us.preheader:
157; CHECK-NEXT:    br label [[FOR_BODY3_LR_PH_US:%.*]]
158; CHECK:       for.end.us:
159; CHECK-NEXT:    [[ARRAYIDX9_US:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[INDVARS_IV33:%.*]]
160; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[ARRAYIDX9_US]], align 4
161; CHECK-NEXT:    [[ADD10_US:%.*]] = add nsw i32 [[TMP0]], 3
162; CHECK-NEXT:    store i32 [[ADD10_US]], i32* [[ARRAYIDX9_US]], align 4
163; CHECK-NEXT:    [[INDVARS_IV_NEXT34:%.*]] = add i64 [[INDVARS_IV33]], 1
164; CHECK-NEXT:    [[LFTR_WIDEIV35:%.*]] = trunc i64 [[INDVARS_IV_NEXT34]] to i32
165; CHECK-NEXT:    [[EXITCOND36:%.*]] = icmp eq i32 [[LFTR_WIDEIV35]], [[M]]
166; CHECK-NEXT:    br i1 [[EXITCOND36]], label [[FOR_END15_LOOPEXIT:%.*]], label [[FOR_BODY3_LR_PH_US]], !llvm.loop [[LOOP2]]
167; CHECK:       for.body3.us:
168; CHECK-NEXT:    [[INDVARS_IV29:%.*]] = phi i64 [ 0, [[FOR_BODY3_LR_PH_US]] ], [ [[INDVARS_IV_NEXT30:%.*]], [[FOR_BODY3_US:%.*]] ]
169; CHECK-NEXT:    [[TMP1:%.*]] = trunc i64 [[INDVARS_IV29]] to i32
170; CHECK-NEXT:    [[ADD4_US:%.*]] = add i32 [[ADD_US:%.*]], [[TMP1]]
171; CHECK-NEXT:    [[IDXPROM_US:%.*]] = sext i32 [[ADD4_US]] to i64
172; CHECK-NEXT:    [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[IDXPROM_US]]
173; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* [[ARRAYIDX_US]], align 4
174; CHECK-NEXT:    [[ADD5_US:%.*]] = add nsw i32 [[TMP2]], 1
175; CHECK-NEXT:    store i32 [[ADD5_US]], i32* [[ARRAYIDX7_US:%.*]], align 4
176; CHECK-NEXT:    [[INDVARS_IV_NEXT30]] = add i64 [[INDVARS_IV29]], 1
177; CHECK-NEXT:    [[LFTR_WIDEIV31:%.*]] = trunc i64 [[INDVARS_IV_NEXT30]] to i32
178; CHECK-NEXT:    [[EXITCOND32:%.*]] = icmp eq i32 [[LFTR_WIDEIV31]], [[M]]
179; CHECK-NEXT:    br i1 [[EXITCOND32]], label [[FOR_END_US:%.*]], label [[FOR_BODY3_US]], !llvm.loop [[LOOP1:![0-9]+]]
180; CHECK:       for.body3.lr.ph.us:
181; CHECK-NEXT:    [[INDVARS_IV33]] = phi i64 [ [[INDVARS_IV_NEXT34]], [[FOR_END_US]] ], [ 0, [[FOR_BODY3_LR_PH_US_PREHEADER]] ]
182; CHECK-NEXT:    [[TMP3:%.*]] = trunc i64 [[INDVARS_IV33]] to i32
183; CHECK-NEXT:    [[ADD_US]] = add i32 [[TMP3]], [[K:%.*]]
184; CHECK-NEXT:    [[ARRAYIDX7_US]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV33]]
185; CHECK-NEXT:    br label [[FOR_BODY3_US]]
186; CHECK:       for.end15.loopexit:
187; CHECK-NEXT:    br label [[FOR_END15]]
188; CHECK:       for.end15:
189; CHECK-NEXT:    ret void
190;
191entry:
192  %cmp27 = icmp sgt i32 %m, 0
193  br i1 %cmp27, label %for.body3.lr.ph.us, label %for.end15
194
195for.end.us:                                       ; preds = %for.body3.us
196  %arrayidx9.us = getelementptr inbounds i32, i32* %b, i64 %indvars.iv33
197  %0 = load i32, i32* %arrayidx9.us, align 4
198  %add10.us = add nsw i32 %0, 3
199  store i32 %add10.us, i32* %arrayidx9.us, align 4
200  %indvars.iv.next34 = add i64 %indvars.iv33, 1
201  %lftr.wideiv35 = trunc i64 %indvars.iv.next34 to i32
202  %exitcond36 = icmp eq i32 %lftr.wideiv35, %m
203  br i1 %exitcond36, label %for.end15, label %for.body3.lr.ph.us, !llvm.loop !5
204
205for.body3.us:                                     ; preds = %for.body3.us, %for.body3.lr.ph.us
206  %indvars.iv29 = phi i64 [ 0, %for.body3.lr.ph.us ], [ %indvars.iv.next30, %for.body3.us ]
207  %1 = trunc i64 %indvars.iv29 to i32
208  %add4.us = add i32 %add.us, %1
209  %idxprom.us = sext i32 %add4.us to i64
210  %arrayidx.us = getelementptr inbounds i32, i32* %a, i64 %idxprom.us
211  %2 = load i32, i32* %arrayidx.us, align 4
212  %add5.us = add nsw i32 %2, 1
213  store i32 %add5.us, i32* %arrayidx7.us, align 4
214  %indvars.iv.next30 = add i64 %indvars.iv29, 1
215  %lftr.wideiv31 = trunc i64 %indvars.iv.next30 to i32
216  %exitcond32 = icmp eq i32 %lftr.wideiv31, %m
217  br i1 %exitcond32, label %for.end.us, label %for.body3.us, !llvm.loop !4
218
219for.body3.lr.ph.us:                               ; preds = %for.end.us, %entry
220  %indvars.iv33 = phi i64 [ %indvars.iv.next34, %for.end.us ], [ 0, %entry ]
221  %3 = trunc i64 %indvars.iv33 to i32
222  %add.us = add i32 %3, %k
223  %arrayidx7.us = getelementptr inbounds i32, i32* %a, i64 %indvars.iv33
224  br label %for.body3.us
225
226for.end15:                                        ; preds = %for.end.us, %entry
227  ret void
228}
229
230attributes #0 = { nounwind uwtable "less-precise-fpmad"="false" "frame-pointer"="none" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }
231
232!3 = !{!4, !5}
233!4 = !{!4}
234!5 = !{!5}
235
236