1; RUN: opt -S < %s -basicaa -loop-vectorize -force-vector-interleave=1 2>&1 | FileCheck %s
2
3target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
4target triple = "aarch64"
5
6; CHECK-LABEL: @add_a(
7; CHECK: load <16 x i8>, <16 x i8>*
8; CHECK: add nuw nsw <16 x i8>
9; CHECK: store <16 x i8>
10; Function Attrs: nounwind
11define void @add_a(i8* noalias nocapture readonly %p, i8* noalias nocapture %q, i32 %len) #0 {
12entry:
13  %cmp8 = icmp sgt i32 %len, 0
14  br i1 %cmp8, label %for.body, label %for.cond.cleanup
15
16for.cond.cleanup:                                 ; preds = %for.body, %entry
17  ret void
18
19for.body:                                         ; preds = %entry, %for.body
20  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
21  %arrayidx = getelementptr inbounds i8, i8* %p, i64 %indvars.iv
22  %0 = load i8, i8* %arrayidx
23  %conv = zext i8 %0 to i32
24  %add = add nuw nsw i32 %conv, 2
25  %conv1 = trunc i32 %add to i8
26  %arrayidx3 = getelementptr inbounds i8, i8* %q, i64 %indvars.iv
27  store i8 %conv1, i8* %arrayidx3
28  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
29  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
30  %exitcond = icmp eq i32 %lftr.wideiv, %len
31  br i1 %exitcond, label %for.cond.cleanup, label %for.body
32}
33
34; CHECK-LABEL: @add_b(
35; CHECK: load <8 x i16>, <8 x i16>*
36; CHECK: add nuw nsw <8 x i16>
37; CHECK: store <8 x i16>
38; Function Attrs: nounwind
39define void @add_b(i16* noalias nocapture readonly %p, i16* noalias nocapture %q, i32 %len) #0 {
40entry:
41  %cmp9 = icmp sgt i32 %len, 0
42  br i1 %cmp9, label %for.body, label %for.cond.cleanup
43
44for.cond.cleanup:                                 ; preds = %for.body, %entry
45  ret void
46
47for.body:                                         ; preds = %entry, %for.body
48  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
49  %arrayidx = getelementptr inbounds i16, i16* %p, i64 %indvars.iv
50  %0 = load i16, i16* %arrayidx
51  %conv8 = zext i16 %0 to i32
52  %add = add nuw nsw i32 %conv8, 2
53  %conv1 = trunc i32 %add to i16
54  %arrayidx3 = getelementptr inbounds i16, i16* %q, i64 %indvars.iv
55  store i16 %conv1, i16* %arrayidx3
56  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
57  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
58  %exitcond = icmp eq i32 %lftr.wideiv, %len
59  br i1 %exitcond, label %for.cond.cleanup, label %for.body
60}
61
62; CHECK-LABEL: @add_c(
63; CHECK: load <8 x i8>, <8 x i8>*
64; CHECK: add nuw nsw <8 x i16>
65; CHECK: store <8 x i16>
66; Function Attrs: nounwind
67define void @add_c(i8* noalias nocapture readonly %p, i16* noalias nocapture %q, i32 %len) #0 {
68entry:
69  %cmp8 = icmp sgt i32 %len, 0
70  br i1 %cmp8, label %for.body, label %for.cond.cleanup
71
72for.cond.cleanup:                                 ; preds = %for.body, %entry
73  ret void
74
75for.body:                                         ; preds = %entry, %for.body
76  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
77  %arrayidx = getelementptr inbounds i8, i8* %p, i64 %indvars.iv
78  %0 = load i8, i8* %arrayidx
79  %conv = zext i8 %0 to i32
80  %add = add nuw nsw i32 %conv, 2
81  %conv1 = trunc i32 %add to i16
82  %arrayidx3 = getelementptr inbounds i16, i16* %q, i64 %indvars.iv
83  store i16 %conv1, i16* %arrayidx3
84  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
85  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
86  %exitcond = icmp eq i32 %lftr.wideiv, %len
87  br i1 %exitcond, label %for.cond.cleanup, label %for.body
88}
89
90; CHECK-LABEL: @add_d(
91; CHECK: load <4 x i16>
92; CHECK: add nsw <4 x i32>
93; CHECK: store <4 x i32>
94define void @add_d(i16* noalias nocapture readonly %p, i32* noalias nocapture %q, i32 %len) #0 {
95entry:
96  %cmp7 = icmp sgt i32 %len, 0
97  br i1 %cmp7, label %for.body, label %for.cond.cleanup
98
99for.cond.cleanup:                                 ; preds = %for.body, %entry
100  ret void
101
102for.body:                                         ; preds = %entry, %for.body
103  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
104  %arrayidx = getelementptr inbounds i16, i16* %p, i64 %indvars.iv
105  %0 = load i16, i16* %arrayidx
106  %conv = sext i16 %0 to i32
107  %add = add nsw i32 %conv, 2
108  %arrayidx2 = getelementptr inbounds i32, i32* %q, i64 %indvars.iv
109  store i32 %add, i32* %arrayidx2
110  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
111  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
112  %exitcond = icmp eq i32 %lftr.wideiv, %len
113  br i1 %exitcond, label %for.cond.cleanup, label %for.body
114}
115
116; CHECK-LABEL: @add_e(
117; CHECK: load <16 x i8>
118; CHECK: shl <16 x i8>
119; CHECK: add nuw nsw <16 x i8>
120; CHECK: or <16 x i8>
121; CHECK: mul nuw nsw <16 x i8>
122; CHECK: and <16 x i8>
123; CHECK: xor <16 x i8>
124; CHECK: mul nuw nsw <16 x i8>
125; CHECK: store <16 x i8>
126define void @add_e(i8* noalias nocapture readonly %p, i8* noalias nocapture %q, i8 %arg1, i8 %arg2, i32 %len) #0 {
127entry:
128  %cmp.32 = icmp sgt i32 %len, 0
129  br i1 %cmp.32, label %for.body.lr.ph, label %for.cond.cleanup
130
131for.body.lr.ph:                                   ; preds = %entry
132  %conv11 = zext i8 %arg2 to i32
133  %conv13 = zext i8 %arg1 to i32
134  br label %for.body
135
136for.cond.cleanup:                                 ; preds = %for.body, %entry
137  ret void
138
139for.body:                                         ; preds = %for.body, %for.body.lr.ph
140  %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ]
141  %arrayidx = getelementptr inbounds i8, i8* %p, i64 %indvars.iv
142  %0 = load i8, i8* %arrayidx
143  %conv = zext i8 %0 to i32
144  %add = shl i32 %conv, 4
145  %conv2 = add nuw nsw i32 %add, 32
146  %or = or i32 %conv, 51
147  %mul = mul nuw nsw i32 %or, 60
148  %and = and i32 %conv2, %conv13
149  %mul.masked = and i32 %mul, 252
150  %conv17 = xor i32 %mul.masked, %conv11
151  %mul18 = mul nuw nsw i32 %conv17, %and
152  %conv19 = trunc i32 %mul18 to i8
153  %arrayidx21 = getelementptr inbounds i8, i8* %q, i64 %indvars.iv
154  store i8 %conv19, i8* %arrayidx21
155  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
156  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
157  %exitcond = icmp eq i32 %lftr.wideiv, %len
158  br i1 %exitcond, label %for.cond.cleanup, label %for.body
159}
160
161; CHECK-LABEL: @add_f
162; CHECK: load <8 x i16>
163; CHECK: trunc <8 x i16>
164; CHECK: shl <8 x i8>
165; CHECK: add nsw <8 x i8>
166; CHECK: or <8 x i8>
167; CHECK: mul nuw nsw <8 x i8>
168; CHECK: and <8 x i8>
169; CHECK: xor <8 x i8>
170; CHECK: mul nuw nsw <8 x i8>
171; CHECK: store <8 x i8>
172define void @add_f(i16* noalias nocapture readonly %p, i8* noalias nocapture %q, i8 %arg1, i8 %arg2, i32 %len) #0 {
173entry:
174  %cmp.32 = icmp sgt i32 %len, 0
175  br i1 %cmp.32, label %for.body.lr.ph, label %for.cond.cleanup
176
177for.body.lr.ph:                                   ; preds = %entry
178  %conv11 = zext i8 %arg2 to i32
179  %conv13 = zext i8 %arg1 to i32
180  br label %for.body
181
182for.cond.cleanup:                                 ; preds = %for.body, %entry
183  ret void
184
185for.body:                                         ; preds = %for.body, %for.body.lr.ph
186  %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ]
187  %arrayidx = getelementptr inbounds i16, i16* %p, i64 %indvars.iv
188  %0 = load i16, i16* %arrayidx
189  %conv = sext i16 %0 to i32
190  %add = shl i32 %conv, 4
191  %conv2 = add nsw i32 %add, 32
192  %or = and i32 %conv, 204
193  %conv8 = or i32 %or, 51
194  %mul = mul nuw nsw i32 %conv8, 60
195  %and = and i32 %conv2, %conv13
196  %mul.masked = and i32 %mul, 252
197  %conv17 = xor i32 %mul.masked, %conv11
198  %mul18 = mul nuw nsw i32 %conv17, %and
199  %conv19 = trunc i32 %mul18 to i8
200  %arrayidx21 = getelementptr inbounds i8, i8* %q, i64 %indvars.iv
201  store i8 %conv19, i8* %arrayidx21
202  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
203  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
204  %exitcond = icmp eq i32 %lftr.wideiv, %len
205  br i1 %exitcond, label %for.cond.cleanup, label %for.body
206}
207
208; CHECK-LABEL: @add_g
209; CHECK: load <16 x i8>
210; CHECK: xor <16 x i8>
211; CHECK: icmp ult <16 x i8>
212; CHECK: select <16 x i1> {{.*}}, <16 x i8>
213; CHECK: store <16 x i8>
214define void @add_g(i8* noalias nocapture readonly %p, i8* noalias nocapture readonly %q, i8* noalias nocapture %r, i8 %arg1, i32 %len) #0 {
215  %1 = icmp sgt i32 %len, 0
216  br i1 %1, label %.lr.ph, label %._crit_edge
217
218.lr.ph:                                           ; preds = %0
219  %2 = sext i8 %arg1 to i64
220  br label %3
221
222._crit_edge:                                      ; preds = %3, %0
223  ret void
224
225; <label>:3                                       ; preds = %3, %.lr.ph
226  %indvars.iv = phi i64 [ 0, %.lr.ph ], [ %indvars.iv.next, %3 ]
227  %x4 = getelementptr inbounds i8, i8* %p, i64 %indvars.iv
228  %x5 = load i8, i8* %x4
229  %x7 = getelementptr inbounds i8, i8* %q, i64 %indvars.iv
230  %x8 = load i8, i8* %x7
231  %x9 = zext i8 %x5 to i32
232  %x10 = xor i32 %x9, 255
233  %x11 = icmp ult i32 %x10, 24
234  %x12 = select i1 %x11, i32 %x10, i32 24
235  %x13 = trunc i32 %x12 to i8
236  store i8 %x13, i8* %x4
237  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
238  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
239  %exitcond = icmp eq i32 %lftr.wideiv, %len
240  br i1 %exitcond, label %._crit_edge, label %3
241}
242
243attributes #0 = { nounwind }
244