1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -passes=loop-vectorize -force-vector-width=4 -S | FileCheck %s
3
4target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
5
6; Make sure the loop is vectorized under -Os without folding its tail based on
7; its trip-count's lower bits known to be zero.
8
9define dso_local void @alignTC(i32* noalias nocapture %A, i32 %n) optsize {
10; CHECK-LABEL: @alignTC(
11; CHECK-NEXT:  entry:
12; CHECK-NEXT:    [[ALIGNEDTC:%.*]] = and i32 [[N:%.*]], -8
13; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[ALIGNEDTC]], 4
14; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
15; CHECK:       vector.ph:
16; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[ALIGNEDTC]], 4
17; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[ALIGNEDTC]], [[N_MOD_VF]]
18; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
19; CHECK:       vector.body:
20; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
21; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
22; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[TMP0]]
23; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TMP1]], i32 0
24; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <4 x i32>*
25; CHECK-NEXT:    store <4 x i32> <i32 13, i32 13, i32 13, i32 13>, <4 x i32>* [[TMP3]], align 1
26; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
27; CHECK-NEXT:    [[TMP4:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
28; CHECK-NEXT:    br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP0:!llvm.loop !.*]]
29; CHECK:       middle.block:
30; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[ALIGNEDTC]], [[N_VEC]]
31; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
32; CHECK:       scalar.ph:
33; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
34; CHECK-NEXT:    br label [[LOOP:%.*]]
35; CHECK:       loop:
36; CHECK-NEXT:    [[RIV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[RIVPLUS1:%.*]], [[LOOP]] ]
37; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[RIV]]
38; CHECK-NEXT:    store i32 13, i32* [[ARRAYIDX]], align 1
39; CHECK-NEXT:    [[RIVPLUS1]] = add nuw nsw i32 [[RIV]], 1
40; CHECK-NEXT:    [[COND:%.*]] = icmp eq i32 [[RIVPLUS1]], [[ALIGNEDTC]]
41; CHECK-NEXT:    br i1 [[COND]], label [[EXIT]], label [[LOOP]], [[LOOP2:!llvm.loop !.*]]
42; CHECK:       exit:
43; CHECK-NEXT:    ret void
44;
45entry:
46  %alignedTC = and i32 %n, -8
47  br label %loop
48
49loop:
50  %riv = phi i32 [ 0, %entry ], [ %rivPlus1, %loop ]
51  %arrayidx = getelementptr inbounds i32, i32* %A, i32 %riv
52  store i32 13, i32* %arrayidx, align 1
53  %rivPlus1 = add nuw nsw i32 %riv, 1
54  %cond = icmp eq i32 %rivPlus1, %alignedTC
55  br i1 %cond, label %exit, label %loop
56
57exit:
58  ret void
59}
60
61; Make sure the loop is vectorized under -Os without folding its tail based on
62; its trip-count's lower bits assumed to be zero.
63
64define dso_local void @assumeAlignedTC(i32* noalias nocapture %A, i32 %p, i32 %q) optsize {
65; CHECK-LABEL: @assumeAlignedTC(
66; CHECK-NEXT:  entry:
67; CHECK-NEXT:    [[AND1:%.*]] = and i32 [[P:%.*]], 3
68; CHECK-NEXT:    [[CMP1:%.*]] = icmp eq i32 [[AND1]], 0
69; CHECK-NEXT:    br i1 [[CMP1]], label [[GUARDED:%.*]], label [[EXIT:%.*]]
70; CHECK:       guarded:
71; CHECK-NEXT:    [[REM:%.*]] = urem i32 [[Q:%.*]], 8
72; CHECK-NEXT:    [[CMP2:%.*]] = icmp eq i32 [[REM]], 0
73; CHECK-NEXT:    tail call void @llvm.assume(i1 [[CMP2]])
74; CHECK-NEXT:    [[GT:%.*]] = icmp sgt i32 [[P]], [[Q]]
75; CHECK-NEXT:    [[N:%.*]] = select i1 [[GT]], i32 [[P]], i32 [[Q]]
76; CHECK-NEXT:    [[CMP110:%.*]] = icmp sgt i32 [[N]], 0
77; CHECK-NEXT:    br i1 [[CMP110]], label [[LOOP_PREHEADER:%.*]], label [[EXIT]]
78; CHECK:       loop.preheader:
79; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N]], 4
80; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
81; CHECK:       vector.ph:
82; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[N]], 4
83; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[N]], [[N_MOD_VF]]
84; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
85; CHECK:       vector.body:
86; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
87; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
88; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[TMP0]]
89; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TMP1]], i32 0
90; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <4 x i32>*
91; CHECK-NEXT:    store <4 x i32> <i32 13, i32 13, i32 13, i32 13>, <4 x i32>* [[TMP3]], align 1
92; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
93; CHECK-NEXT:    [[TMP4:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
94; CHECK-NEXT:    br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP4:!llvm.loop !.*]]
95; CHECK:       middle.block:
96; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[N]], [[N_VEC]]
97; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]]
98; CHECK:       scalar.ph:
99; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP_PREHEADER]] ]
100; CHECK-NEXT:    br label [[LOOP:%.*]]
101; CHECK:       loop:
102; CHECK-NEXT:    [[RIV:%.*]] = phi i32 [ [[RIVPLUS1:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
103; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[RIV]]
104; CHECK-NEXT:    store i32 13, i32* [[ARRAYIDX]], align 1
105; CHECK-NEXT:    [[RIVPLUS1]] = add nuw nsw i32 [[RIV]], 1
106; CHECK-NEXT:    [[COND:%.*]] = icmp eq i32 [[RIVPLUS1]], [[N]]
107; CHECK-NEXT:    br i1 [[COND]], label [[EXIT_LOOPEXIT]], label [[LOOP]], [[LOOP5:!llvm.loop !.*]]
108; CHECK:       exit.loopexit:
109; CHECK-NEXT:    br label [[EXIT]]
110; CHECK:       exit:
111; CHECK-NEXT:    ret void
112;
113entry:
114  %and1 = and i32 %p, 3
115  %cmp1 = icmp eq i32 %and1, 0
116  br i1 %cmp1, label %guarded, label %exit
117
118guarded:
119  %rem = urem i32 %q, 8
120  %cmp2 = icmp eq i32 %rem, 0
121  tail call void @llvm.assume(i1 %cmp2)
122  %gt = icmp sgt i32 %p, %q
123  %n = select i1 %gt, i32 %p, i32 %q
124  %cmp110 = icmp sgt i32 %n, 0
125  br i1 %cmp110, label %loop, label %exit
126
127loop:
128  %riv = phi i32 [ 0, %guarded ], [ %rivPlus1, %loop ]
129  %arrayidx = getelementptr inbounds i32, i32* %A, i32 %riv
130  store i32 13, i32* %arrayidx, align 1
131  %rivPlus1 = add nuw nsw i32 %riv, 1
132  %cond = icmp eq i32 %rivPlus1, %n
133  br i1 %cond, label %exit, label %loop
134
135exit:
136  ret void
137}
138
139; Make sure the loop's tail is folded when vectorized under -Os based on its trip-count's
140; not being provably divisible by chosen VF.
141
142define dso_local void @cannotProveAlignedTC(i32* noalias nocapture %A, i32 %p, i32 %q) optsize {
143; CHECK-LABEL: @cannotProveAlignedTC(
144; CHECK-NEXT:  entry:
145; CHECK-NEXT:    [[AND1:%.*]] = and i32 [[P:%.*]], 3
146; CHECK-NEXT:    [[CMP1:%.*]] = icmp eq i32 [[AND1]], 0
147; CHECK-NEXT:    br i1 [[CMP1]], label [[GUARDED:%.*]], label [[EXIT:%.*]]
148; CHECK:       guarded:
149; CHECK-NEXT:    [[REM:%.*]] = urem i32 [[Q:%.*]], 3
150; CHECK-NEXT:    [[CMP2:%.*]] = icmp eq i32 [[REM]], 0
151; CHECK-NEXT:    tail call void @llvm.assume(i1 [[CMP2]])
152; CHECK-NEXT:    [[GT:%.*]] = icmp sgt i32 [[P]], [[Q]]
153; CHECK-NEXT:    [[N:%.*]] = select i1 [[GT]], i32 [[P]], i32 [[Q]]
154; CHECK-NEXT:    [[CMP110:%.*]] = icmp sgt i32 [[N]], 0
155; CHECK-NEXT:    br i1 [[CMP110]], label [[LOOP_PREHEADER:%.*]], label [[EXIT]]
156; CHECK:       loop.preheader:
157; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
158; CHECK:       vector.ph:
159; CHECK-NEXT:    [[N_RND_UP:%.*]] = add i32 [[N]], 3
160; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[N_RND_UP]], 4
161; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[N_RND_UP]], [[N_MOD_VF]]
162; CHECK-NEXT:    [[TRIP_COUNT_MINUS_1:%.*]] = sub i32 [[N]], 1
163; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[TRIP_COUNT_MINUS_1]], i32 0
164; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
165; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
166; CHECK:       vector.body:
167; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE6:%.*]] ]
168; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE6]] ]
169; CHECK-NEXT:    [[TMP0:%.*]] = icmp ule <4 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
170; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x i1> [[TMP0]], i32 0
171; CHECK-NEXT:    br i1 [[TMP1]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
172; CHECK:       pred.store.if:
173; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[INDEX]], 0
174; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[TMP2]]
175; CHECK-NEXT:    store i32 13, i32* [[TMP3]], align 1
176; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE]]
177; CHECK:       pred.store.continue:
178; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <4 x i1> [[TMP0]], i32 1
179; CHECK-NEXT:    br i1 [[TMP4]], label [[PRED_STORE_IF1:%.*]], label [[PRED_STORE_CONTINUE2:%.*]]
180; CHECK:       pred.store.if1:
181; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[INDEX]], 1
182; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[TMP5]]
183; CHECK-NEXT:    store i32 13, i32* [[TMP6]], align 1
184; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE2]]
185; CHECK:       pred.store.continue2:
186; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x i1> [[TMP0]], i32 2
187; CHECK-NEXT:    br i1 [[TMP7]], label [[PRED_STORE_IF3:%.*]], label [[PRED_STORE_CONTINUE4:%.*]]
188; CHECK:       pred.store.if3:
189; CHECK-NEXT:    [[TMP8:%.*]] = add i32 [[INDEX]], 2
190; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[TMP8]]
191; CHECK-NEXT:    store i32 13, i32* [[TMP9]], align 1
192; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE4]]
193; CHECK:       pred.store.continue4:
194; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <4 x i1> [[TMP0]], i32 3
195; CHECK-NEXT:    br i1 [[TMP10]], label [[PRED_STORE_IF5:%.*]], label [[PRED_STORE_CONTINUE6]]
196; CHECK:       pred.store.if5:
197; CHECK-NEXT:    [[TMP11:%.*]] = add i32 [[INDEX]], 3
198; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[TMP11]]
199; CHECK-NEXT:    store i32 13, i32* [[TMP12]], align 1
200; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE6]]
201; CHECK:       pred.store.continue6:
202; CHECK-NEXT:    [[INDEX_NEXT]] = add i32 [[INDEX]], 4
203; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
204; CHECK-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
205; CHECK-NEXT:    br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP6:!llvm.loop !.*]]
206; CHECK:       middle.block:
207; CHECK-NEXT:    br i1 true, label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]]
208; CHECK:       scalar.ph:
209; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP_PREHEADER]] ]
210; CHECK-NEXT:    br label [[LOOP:%.*]]
211; CHECK:       loop:
212; CHECK-NEXT:    [[RIV:%.*]] = phi i32 [ [[RIVPLUS1:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
213; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[RIV]]
214; CHECK-NEXT:    store i32 13, i32* [[ARRAYIDX]], align 1
215; CHECK-NEXT:    [[RIVPLUS1]] = add nuw nsw i32 [[RIV]], 1
216; CHECK-NEXT:    [[COND:%.*]] = icmp eq i32 [[RIVPLUS1]], [[N]]
217; CHECK-NEXT:    br i1 [[COND]], label [[EXIT_LOOPEXIT]], label [[LOOP]], [[LOOP7:!llvm.loop !.*]]
218; CHECK:       exit.loopexit:
219; CHECK-NEXT:    br label [[EXIT]]
220; CHECK:       exit:
221; CHECK-NEXT:    ret void
222;
223entry:
224  %and1 = and i32 %p, 3
225  %cmp1 = icmp eq i32 %and1, 0
226  br i1 %cmp1, label %guarded, label %exit
227
228guarded:
229  %rem = urem i32 %q, 3
230  %cmp2 = icmp eq i32 %rem, 0
231  tail call void @llvm.assume(i1 %cmp2)
232  %gt = icmp sgt i32 %p, %q
233  %n = select i1 %gt, i32 %p, i32 %q
234  %cmp110 = icmp sgt i32 %n, 0
235  br i1 %cmp110, label %loop, label %exit
236
237loop:
238  %riv = phi i32 [ 0, %guarded ], [ %rivPlus1, %loop ]
239  %arrayidx = getelementptr inbounds i32, i32* %A, i32 %riv
240  store i32 13, i32* %arrayidx, align 1
241  %rivPlus1 = add nuw nsw i32 %riv, 1
242  %cond = icmp eq i32 %rivPlus1, %n
243  br i1 %cond, label %exit, label %loop
244
245exit:
246  ret void
247}
248
249declare void @llvm.assume(i1 noundef) nofree nosync nounwind willreturn
250