1; RUN: opt < %s -loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -dce -instcombine -S | FileCheck %s
2; RUN: opt < %s -loop-vectorize -force-vector-width=4 -force-vector-interleave=2 -dce -instcombine -S | FileCheck %s --check-prefix=UNROLL
3
4target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
5
6; CHECK-LABEL: @recurrence_1
7;
8; void recurrence_1(int *a, int *b, int n) {
9;   for(int i = 0; i < n; i++)
10;     b[i] =  a[i] + a[i - 1]
11; }
12;
13; CHECK:  vector.ph:
14; CHECK:    %vector.recur.init = insertelement <4 x i32> undef, i32 %pre_load, i32 3
15;
16; CHECK:  vector.body:
17; CHECK:    %vector.recur = phi <4 x i32> [ %vector.recur.init, %vector.ph ], [ [[L1:%[a-zA-Z0-9.]+]], %vector.body ]
18; CHECK:    [[L1]] = load <4 x i32>
19; CHECK:    {{.*}} = shufflevector <4 x i32> %vector.recur, <4 x i32> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
20;
21; CHECK:  middle.block:
22; CHECK:    %vector.recur.extract = extractelement <4 x i32> [[L1]], i32 3
23;
24; CHECK:  scalar.ph:
25; CHECK:    %scalar.recur.init = phi i32 [ %vector.recur.extract, %middle.block ], [ %pre_load, %vector.memcheck ], [ %pre_load, %min.iters.checked ], [ %pre_load, %for.preheader ]
26;
27; CHECK:  scalar.body:
28; CHECK:    %scalar.recur = phi i32 [ %scalar.recur.init, %scalar.ph ], [ {{.*}}, %scalar.body ]
29;
30; UNROLL: vector.body:
31; UNROLL:   %vector.recur = phi <4 x i32> [ %vector.recur.init, %vector.ph ], [ [[L2:%[a-zA-Z0-9.]+]], %vector.body ]
32; UNROLL:   [[L1:%[a-zA-Z0-9.]+]] = load <4 x i32>
33; UNROLL:   [[L2]] = load <4 x i32>
34; UNROLL:   {{.*}} = shufflevector <4 x i32> %vector.recur, <4 x i32> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
35; UNROLL:   {{.*}} = shufflevector <4 x i32> [[L1]], <4 x i32> [[L2]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
36;
37; UNROLL: middle.block:
38; UNROLL:   %vector.recur.extract = extractelement <4 x i32> [[L2]], i32 3
39;
40define void @recurrence_1(i32* nocapture readonly %a, i32* nocapture %b, i32 %n) {
41entry:
42  br label %for.preheader
43
44for.preheader:
45  %arrayidx.phi.trans.insert = getelementptr inbounds i32, i32* %a, i64 0
46  %pre_load = load i32, i32* %arrayidx.phi.trans.insert
47  br label %scalar.body
48
49scalar.body:
50  %0 = phi i32 [ %pre_load, %for.preheader ], [ %1, %scalar.body ]
51  %indvars.iv = phi i64 [ 0, %for.preheader ], [ %indvars.iv.next, %scalar.body ]
52  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
53  %arrayidx32 = getelementptr inbounds i32, i32* %a, i64 %indvars.iv.next
54  %1 = load i32, i32* %arrayidx32
55  %arrayidx34 = getelementptr inbounds i32, i32* %b, i64 %indvars.iv
56  %add35 = add i32 %1, %0
57  store i32 %add35, i32* %arrayidx34
58  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
59  %exitcond = icmp eq i32 %lftr.wideiv, %n
60  br i1 %exitcond, label %for.exit, label %scalar.body
61
62for.exit:
63  ret void
64}
65
66; CHECK-LABEL: @recurrence_2
67;
68; int recurrence_2(int *a, int n) {
69;   int minmax;
70;   for (int i = 0; i < n; ++i)
71;     minmax = min(minmax, max(a[i] - a[i-1], 0));
72;   return minmax;
73; }
74;
75; CHECK:  vector.ph:
76; CHECK:    %vector.recur.init = insertelement <4 x i32> undef, i32 %.pre, i32 3
77;
78; CHECK:  vector.body:
79; CHECK:    %vector.recur = phi <4 x i32> [ %vector.recur.init, %vector.ph ], [ [[L1:%[a-zA-Z0-9.]+]], %vector.body ]
80; CHECK:    [[L1]] = load <4 x i32>
81; CHECK:    {{.*}} = shufflevector <4 x i32> %vector.recur, <4 x i32> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
82;
83; CHECK:  middle.block:
84; CHECK:    %vector.recur.extract = extractelement <4 x i32> [[L1]], i32 3
85;
86; CHECK:  scalar.ph:
87; CHECK:    %scalar.recur.init = phi i32 [ %vector.recur.extract, %middle.block ], [ %.pre, %min.iters.checked ], [ %.pre, %for.preheader ]
88;
89; CHECK:  scalar.body:
90; CHECK:    %scalar.recur = phi i32 [ %scalar.recur.init, %scalar.ph ], [ {{.*}}, %scalar.body ]
91;
92; UNROLL: vector.body:
93; UNROLL:   %vector.recur = phi <4 x i32> [ %vector.recur.init, %vector.ph ], [ [[L2:%[a-zA-Z0-9.]+]], %vector.body ]
94; UNROLL:   [[L1:%[a-zA-Z0-9.]+]] = load <4 x i32>
95; UNROLL:   [[L2]] = load <4 x i32>
96; UNROLL:   {{.*}} = shufflevector <4 x i32> %vector.recur, <4 x i32> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
97; UNROLL:   {{.*}} = shufflevector <4 x i32> [[L1]], <4 x i32> [[L2]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
98;
99; UNROLL: middle.block:
100; UNROLL:   %vector.recur.extract = extractelement <4 x i32> [[L2]], i32 3
101;
102define i32 @recurrence_2(i32* nocapture readonly %a, i32 %n) {
103entry:
104  %cmp27 = icmp sgt i32 %n, 0
105  br i1 %cmp27, label %for.preheader, label %for.cond.cleanup
106
107for.preheader:
108  %arrayidx2.phi.trans.insert = getelementptr inbounds i32, i32* %a, i64 -1
109  %.pre = load i32, i32* %arrayidx2.phi.trans.insert, align 4
110  br label %scalar.body
111
112for.cond.cleanup.loopexit:
113  %minmax.0.cond.lcssa = phi i32 [ %minmax.0.cond, %scalar.body ]
114  br label %for.cond.cleanup
115
116for.cond.cleanup:
117  %minmax.0.lcssa = phi i32 [ undef, %entry ], [ %minmax.0.cond.lcssa, %for.cond.cleanup.loopexit ]
118  ret i32 %minmax.0.lcssa
119
120scalar.body:
121  %0 = phi i32 [ %.pre, %for.preheader ], [ %1, %scalar.body ]
122  %indvars.iv = phi i64 [ 0, %for.preheader ], [ %indvars.iv.next, %scalar.body ]
123  %minmax.028 = phi i32 [ undef, %for.preheader ], [ %minmax.0.cond, %scalar.body ]
124  %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
125  %1 = load i32, i32* %arrayidx, align 4
126  %sub3 = sub nsw i32 %1, %0
127  %cmp4 = icmp sgt i32 %sub3, 0
128  %cond = select i1 %cmp4, i32 %sub3, i32 0
129  %cmp5 = icmp slt i32 %minmax.028, %cond
130  %minmax.0.cond = select i1 %cmp5, i32 %minmax.028, i32 %cond
131  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
132  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
133  %exitcond = icmp eq i32 %lftr.wideiv, %n
134  br i1 %exitcond, label %for.cond.cleanup.loopexit, label %scalar.body
135}
136
137; CHECK-LABEL: @recurrence_3
138;
139; void recurrence_3(short *a, double *b, int n, float f, short p) {
140;   b[0] = (double)a[0] - f * (double)p;
141;   for (int i = 1; i < n; i++)
142;     b[i] = (double)a[i] - f * (double)a[i - 1];
143; }
144;
145;
146; CHECK:  vector.ph:
147; CHECK:    %vector.recur.init = insertelement <4 x i16> undef, i16 %0, i32 3
148;
149; CHECK:  vector.body:
150; CHECK:    %vector.recur = phi <4 x i16> [ %vector.recur.init, %vector.ph ], [ [[L1:%[a-zA-Z0-9.]+]], %vector.body ]
151; CHECK:    [[L1]] = load <4 x i16>
152; CHECK:    {{.*}} = shufflevector <4 x i16> %vector.recur, <4 x i16> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
153;
154; CHECK:  middle.block:
155; CHECK:    %vector.recur.extract = extractelement <4 x i16> [[L1]], i32 3
156;
157; CHECK:  scalar.ph:
158; CHECK:    %scalar.recur.init = phi i16 [ %vector.recur.extract, %middle.block ], [ %0, %vector.memcheck ], [ %0, %min.iters.checked ], [ %0, %for.preheader ]
159;
160; CHECK:  scalar.body:
161; CHECK:    %scalar.recur = phi i16 [ %scalar.recur.init, %scalar.ph ], [ {{.*}}, %scalar.body ]
162;
163; UNROLL: vector.body:
164; UNROLL:   %vector.recur = phi <4 x i16> [ %vector.recur.init, %vector.ph ], [ [[L2:%[a-zA-Z0-9.]+]], %vector.body ]
165; UNROLL:   [[L1:%[a-zA-Z0-9.]+]] = load <4 x i16>
166; UNROLL:   [[L2]] = load <4 x i16>
167; UNROLL:   {{.*}} = shufflevector <4 x i16> %vector.recur, <4 x i16> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
168; UNROLL:   {{.*}} = shufflevector <4 x i16> [[L1]], <4 x i16> [[L2]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
169;
170; UNROLL: middle.block:
171; UNROLL:   %vector.recur.extract = extractelement <4 x i16> [[L2]], i32 3
172;
173define void @recurrence_3(i16* nocapture readonly %a, double* nocapture %b, i32 %n, float %f, i16 %p) {
174entry:
175  %0 = load i16, i16* %a, align 2
176  %conv = sitofp i16 %0 to double
177  %conv1 = fpext float %f to double
178  %conv2 = sitofp i16 %p to double
179  %mul = fmul fast double %conv2, %conv1
180  %sub = fsub fast double %conv, %mul
181  store double %sub, double* %b, align 8
182  %cmp25 = icmp sgt i32 %n, 1
183  br i1 %cmp25, label %for.preheader, label %for.end
184
185for.preheader:
186  br label %scalar.body
187
188scalar.body:
189  %1 = phi i16 [ %0, %for.preheader ], [ %2, %scalar.body ]
190  %advars.iv = phi i64 [ %advars.iv.next, %scalar.body ], [ 1, %for.preheader ]
191  %arrayidx5 = getelementptr inbounds i16, i16* %a, i64 %advars.iv
192  %2 = load i16, i16* %arrayidx5, align 2
193  %conv6 = sitofp i16 %2 to double
194  %conv11 = sitofp i16 %1 to double
195  %mul12 = fmul fast double %conv11, %conv1
196  %sub13 = fsub fast double %conv6, %mul12
197  %arrayidx15 = getelementptr inbounds double, double* %b, i64 %advars.iv
198  store double %sub13, double* %arrayidx15, align 8
199  %advars.iv.next = add nuw nsw i64 %advars.iv, 1
200  %lftr.wideiv = trunc i64 %advars.iv.next to i32
201  %exitcond = icmp eq i32 %lftr.wideiv, %n
202  br i1 %exitcond, label %for.end.loopexit, label %scalar.body
203
204for.end.loopexit:
205  br label %for.end
206
207for.end:
208  ret void
209}
210
211; CHECK-LABEL: @PR26734
212;
213; void PR26734(short *a, int *b, int *c, int d, short *e) {
214;   for (; d != 21; d++) {
215;     *b &= *c;
216;     *e = *a - 6;
217;     *c = *e;
218;   }
219; }
220;
221; CHECK-NOT: vector.ph:
222;
223define void @PR26734(i16* %a, i32* %b, i32* %c, i32 %d, i16* %e) {
224entry:
225  %cmp4 = icmp eq i32 %d, 21
226  br i1 %cmp4, label %entry.for.end_crit_edge, label %for.body.lr.ph
227
228entry.for.end_crit_edge:
229  %.pre = load i32, i32* %b, align 4
230  br label %for.end
231
232for.body.lr.ph:
233  %0 = load i16, i16* %a, align 2
234  %sub = add i16 %0, -6
235  %conv2 = sext i16 %sub to i32
236  %c.promoted = load i32, i32* %c, align 4
237  %b.promoted = load i32, i32* %b, align 4
238  br label %for.body
239
240for.body:
241  %inc7 = phi i32 [ %d, %for.body.lr.ph ], [ %inc, %for.body ]
242  %and6 = phi i32 [ %b.promoted, %for.body.lr.ph ], [ %and, %for.body ]
243  %conv25 = phi i32 [ %c.promoted, %for.body.lr.ph ], [ %conv2, %for.body ]
244  %and = and i32 %and6, %conv25
245  %inc = add nsw i32 %inc7, 1
246  %cmp = icmp eq i32 %inc, 21
247  br i1 %cmp, label %for.cond.for.end_crit_edge, label %for.body
248
249for.cond.for.end_crit_edge:
250  %and.lcssa = phi i32 [ %and, %for.body ]
251  store i32 %conv2, i32* %c, align 4
252  store i32 %and.lcssa, i32* %b, align 4
253  store i16 %sub, i16* %e, align 2
254  br label %for.end
255
256for.end:
257  ret void
258}
259
260; CHECK-LABEL: @PR27246
261;
262; int PR27246() {
263;   unsigned int e, n;
264;   for (int i = 1; i < 49; ++i) {
265;     for (int k = i; k > 1; --k)
266;       e = k;
267;     n = e;
268;   }
269;   return n;
270; }
271;
272; CHECK-NOT: vector.ph:
273;
274define i32 @PR27246() {
275entry:
276  br label %for.cond1.preheader
277
278for.cond1.preheader:
279  %i.016 = phi i32 [ 1, %entry ], [ %inc, %for.cond.cleanup3 ]
280  %e.015 = phi i32 [ undef, %entry ], [ %e.1.lcssa, %for.cond.cleanup3 ]
281  br label %for.cond1
282
283for.cond.cleanup:
284  %e.1.lcssa.lcssa = phi i32 [ %e.1.lcssa, %for.cond.cleanup3 ]
285  ret i32 %e.1.lcssa.lcssa
286
287for.cond1:
288  %e.1 = phi i32 [ %k.0, %for.cond1 ], [ %e.015, %for.cond1.preheader ]
289  %k.0 = phi i32 [ %dec, %for.cond1 ], [ %i.016, %for.cond1.preheader ]
290  %cmp2 = icmp sgt i32 %k.0, 1
291  %dec = add nsw i32 %k.0, -1
292  br i1 %cmp2, label %for.cond1, label %for.cond.cleanup3
293
294for.cond.cleanup3:
295  %e.1.lcssa = phi i32 [ %e.1, %for.cond1 ]
296  %inc = add nuw nsw i32 %i.016, 1
297  %exitcond = icmp eq i32 %inc, 49
298  br i1 %exitcond, label %for.cond.cleanup, label %for.cond1.preheader
299}
300