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; RUN: opt < %s -loop-vectorize -force-vector-width=4 -force-vector-interleave=2 -S | FileCheck %s --check-prefix=UNROLL-NO-IC
4
5target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
6
7; CHECK-LABEL: @recurrence_1
8;
9; void recurrence_1(int *a, int *b, int n) {
10;   for(int i = 0; i < n; i++)
11;     b[i] =  a[i] + a[i - 1]
12; }
13;
14; CHECK:  vector.ph:
15; CHECK:    %vector.recur.init = insertelement <4 x i32> undef, i32 %pre_load, i32 3
16;
17; CHECK:  vector.body:
18; CHECK:    %vector.recur = phi <4 x i32> [ %vector.recur.init, %vector.ph ], [ [[L1:%[a-zA-Z0-9.]+]], %vector.body ]
19; CHECK:    [[L1]] = load <4 x i32>
20; CHECK:    {{.*}} = shufflevector <4 x i32> %vector.recur, <4 x i32> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
21;
22; CHECK:  middle.block:
23; CHECK:    %vector.recur.extract = extractelement <4 x i32> [[L1]], i32 3
24;
25; CHECK:  scalar.ph:
26; CHECK:    %scalar.recur.init = phi i32 [ %vector.recur.extract, %middle.block ], [ %pre_load, %vector.memcheck ], [ %pre_load, %min.iters.checked ], [ %pre_load, %for.preheader ]
27;
28; CHECK:  scalar.body:
29; CHECK:    %scalar.recur = phi i32 [ %scalar.recur.init, %scalar.ph ], [ {{.*}}, %scalar.body ]
30;
31; UNROLL: vector.body:
32; UNROLL:   %vector.recur = phi <4 x i32> [ %vector.recur.init, %vector.ph ], [ [[L2:%[a-zA-Z0-9.]+]], %vector.body ]
33; UNROLL:   [[L1:%[a-zA-Z0-9.]+]] = load <4 x i32>
34; UNROLL:   [[L2]] = load <4 x i32>
35; UNROLL:   {{.*}} = shufflevector <4 x i32> %vector.recur, <4 x i32> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
36; UNROLL:   {{.*}} = shufflevector <4 x i32> [[L1]], <4 x i32> [[L2]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
37;
38; UNROLL: middle.block:
39; UNROLL:   %vector.recur.extract = extractelement <4 x i32> [[L2]], i32 3
40;
41define void @recurrence_1(i32* nocapture readonly %a, i32* nocapture %b, i32 %n) {
42entry:
43  br label %for.preheader
44
45for.preheader:
46  %arrayidx.phi.trans.insert = getelementptr inbounds i32, i32* %a, i64 0
47  %pre_load = load i32, i32* %arrayidx.phi.trans.insert
48  br label %scalar.body
49
50scalar.body:
51  %0 = phi i32 [ %pre_load, %for.preheader ], [ %1, %scalar.body ]
52  %indvars.iv = phi i64 [ 0, %for.preheader ], [ %indvars.iv.next, %scalar.body ]
53  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
54  %arrayidx32 = getelementptr inbounds i32, i32* %a, i64 %indvars.iv.next
55  %1 = load i32, i32* %arrayidx32
56  %arrayidx34 = getelementptr inbounds i32, i32* %b, i64 %indvars.iv
57  %add35 = add i32 %1, %0
58  store i32 %add35, i32* %arrayidx34
59  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
60  %exitcond = icmp eq i32 %lftr.wideiv, %n
61  br i1 %exitcond, label %for.exit, label %scalar.body
62
63for.exit:
64  ret void
65}
66
67; CHECK-LABEL: @recurrence_2
68;
69; int recurrence_2(int *a, int n) {
70;   int minmax;
71;   for (int i = 0; i < n; ++i)
72;     minmax = min(minmax, max(a[i] - a[i-1], 0));
73;   return minmax;
74; }
75;
76; CHECK:  vector.ph:
77; CHECK:    %vector.recur.init = insertelement <4 x i32> undef, i32 %.pre, i32 3
78;
79; CHECK:  vector.body:
80; CHECK:    %vector.recur = phi <4 x i32> [ %vector.recur.init, %vector.ph ], [ [[L1:%[a-zA-Z0-9.]+]], %vector.body ]
81; CHECK:    [[L1]] = load <4 x i32>
82; CHECK:    {{.*}} = shufflevector <4 x i32> %vector.recur, <4 x i32> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
83;
84; CHECK:  middle.block:
85; CHECK:    %vector.recur.extract = extractelement <4 x i32> [[L1]], i32 3
86;
87; CHECK:  scalar.ph:
88; CHECK:    %scalar.recur.init = phi i32 [ %vector.recur.extract, %middle.block ], [ %.pre, %min.iters.checked ], [ %.pre, %for.preheader ]
89;
90; CHECK:  scalar.body:
91; CHECK:    %scalar.recur = phi i32 [ %scalar.recur.init, %scalar.ph ], [ {{.*}}, %scalar.body ]
92;
93; UNROLL: vector.body:
94; UNROLL:   %vector.recur = phi <4 x i32> [ %vector.recur.init, %vector.ph ], [ [[L2:%[a-zA-Z0-9.]+]], %vector.body ]
95; UNROLL:   [[L1:%[a-zA-Z0-9.]+]] = load <4 x i32>
96; UNROLL:   [[L2]] = load <4 x i32>
97; UNROLL:   {{.*}} = shufflevector <4 x i32> %vector.recur, <4 x i32> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
98; UNROLL:   {{.*}} = shufflevector <4 x i32> [[L1]], <4 x i32> [[L2]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
99;
100; UNROLL: middle.block:
101; UNROLL:   %vector.recur.extract = extractelement <4 x i32> [[L2]], i32 3
102;
103define i32 @recurrence_2(i32* nocapture readonly %a, i32 %n) {
104entry:
105  %cmp27 = icmp sgt i32 %n, 0
106  br i1 %cmp27, label %for.preheader, label %for.cond.cleanup
107
108for.preheader:
109  %arrayidx2.phi.trans.insert = getelementptr inbounds i32, i32* %a, i64 -1
110  %.pre = load i32, i32* %arrayidx2.phi.trans.insert, align 4
111  br label %scalar.body
112
113for.cond.cleanup.loopexit:
114  %minmax.0.cond.lcssa = phi i32 [ %minmax.0.cond, %scalar.body ]
115  br label %for.cond.cleanup
116
117for.cond.cleanup:
118  %minmax.0.lcssa = phi i32 [ undef, %entry ], [ %minmax.0.cond.lcssa, %for.cond.cleanup.loopexit ]
119  ret i32 %minmax.0.lcssa
120
121scalar.body:
122  %0 = phi i32 [ %.pre, %for.preheader ], [ %1, %scalar.body ]
123  %indvars.iv = phi i64 [ 0, %for.preheader ], [ %indvars.iv.next, %scalar.body ]
124  %minmax.028 = phi i32 [ undef, %for.preheader ], [ %minmax.0.cond, %scalar.body ]
125  %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
126  %1 = load i32, i32* %arrayidx, align 4
127  %sub3 = sub nsw i32 %1, %0
128  %cmp4 = icmp sgt i32 %sub3, 0
129  %cond = select i1 %cmp4, i32 %sub3, i32 0
130  %cmp5 = icmp slt i32 %minmax.028, %cond
131  %minmax.0.cond = select i1 %cmp5, i32 %minmax.028, i32 %cond
132  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
133  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
134  %exitcond = icmp eq i32 %lftr.wideiv, %n
135  br i1 %exitcond, label %for.cond.cleanup.loopexit, label %scalar.body
136}
137
138; CHECK-LABEL: @recurrence_3
139;
140; void recurrence_3(short *a, double *b, int n, float f, short p) {
141;   b[0] = (double)a[0] - f * (double)p;
142;   for (int i = 1; i < n; i++)
143;     b[i] = (double)a[i] - f * (double)a[i - 1];
144; }
145;
146;
147; CHECK:  vector.ph:
148; CHECK:    %vector.recur.init = insertelement <4 x i16> undef, i16 %0, i32 3
149;
150; CHECK:  vector.body:
151; CHECK:    %vector.recur = phi <4 x i16> [ %vector.recur.init, %vector.ph ], [ [[L1:%[a-zA-Z0-9.]+]], %vector.body ]
152; CHECK:    [[L1]] = load <4 x i16>
153; CHECK:    {{.*}} = shufflevector <4 x i16> %vector.recur, <4 x i16> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
154;
155; CHECK:  middle.block:
156; CHECK:    %vector.recur.extract = extractelement <4 x i16> [[L1]], i32 3
157;
158; CHECK:  scalar.ph:
159; CHECK:    %scalar.recur.init = phi i16 [ %vector.recur.extract, %middle.block ], [ %0, %vector.memcheck ], [ %0, %min.iters.checked ], [ %0, %for.preheader ]
160;
161; CHECK:  scalar.body:
162; CHECK:    %scalar.recur = phi i16 [ %scalar.recur.init, %scalar.ph ], [ {{.*}}, %scalar.body ]
163;
164; UNROLL: vector.body:
165; UNROLL:   %vector.recur = phi <4 x i16> [ %vector.recur.init, %vector.ph ], [ [[L2:%[a-zA-Z0-9.]+]], %vector.body ]
166; UNROLL:   [[L1:%[a-zA-Z0-9.]+]] = load <4 x i16>
167; UNROLL:   [[L2]] = load <4 x i16>
168; UNROLL:   {{.*}} = shufflevector <4 x i16> %vector.recur, <4 x i16> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
169; UNROLL:   {{.*}} = shufflevector <4 x i16> [[L1]], <4 x i16> [[L2]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
170;
171; UNROLL: middle.block:
172; UNROLL:   %vector.recur.extract = extractelement <4 x i16> [[L2]], i32 3
173;
174define void @recurrence_3(i16* nocapture readonly %a, double* nocapture %b, i32 %n, float %f, i16 %p) {
175entry:
176  %0 = load i16, i16* %a, align 2
177  %conv = sitofp i16 %0 to double
178  %conv1 = fpext float %f to double
179  %conv2 = sitofp i16 %p to double
180  %mul = fmul fast double %conv2, %conv1
181  %sub = fsub fast double %conv, %mul
182  store double %sub, double* %b, align 8
183  %cmp25 = icmp sgt i32 %n, 1
184  br i1 %cmp25, label %for.preheader, label %for.end
185
186for.preheader:
187  br label %scalar.body
188
189scalar.body:
190  %1 = phi i16 [ %0, %for.preheader ], [ %2, %scalar.body ]
191  %advars.iv = phi i64 [ %advars.iv.next, %scalar.body ], [ 1, %for.preheader ]
192  %arrayidx5 = getelementptr inbounds i16, i16* %a, i64 %advars.iv
193  %2 = load i16, i16* %arrayidx5, align 2
194  %conv6 = sitofp i16 %2 to double
195  %conv11 = sitofp i16 %1 to double
196  %mul12 = fmul fast double %conv11, %conv1
197  %sub13 = fsub fast double %conv6, %mul12
198  %arrayidx15 = getelementptr inbounds double, double* %b, i64 %advars.iv
199  store double %sub13, double* %arrayidx15, align 8
200  %advars.iv.next = add nuw nsw i64 %advars.iv, 1
201  %lftr.wideiv = trunc i64 %advars.iv.next to i32
202  %exitcond = icmp eq i32 %lftr.wideiv, %n
203  br i1 %exitcond, label %for.end.loopexit, label %scalar.body
204
205for.end.loopexit:
206  br label %for.end
207
208for.end:
209  ret void
210}
211
212; CHECK-LABEL: @PR26734
213;
214; void PR26734(short *a, int *b, int *c, int d, short *e) {
215;   for (; d != 21; d++) {
216;     *b &= *c;
217;     *e = *a - 6;
218;     *c = *e;
219;   }
220; }
221;
222; CHECK-NOT: vector.ph:
223;
224define void @PR26734(i16* %a, i32* %b, i32* %c, i32 %d, i16* %e) {
225entry:
226  %cmp4 = icmp eq i32 %d, 21
227  br i1 %cmp4, label %entry.for.end_crit_edge, label %for.body.lr.ph
228
229entry.for.end_crit_edge:
230  %.pre = load i32, i32* %b, align 4
231  br label %for.end
232
233for.body.lr.ph:
234  %0 = load i16, i16* %a, align 2
235  %sub = add i16 %0, -6
236  %conv2 = sext i16 %sub to i32
237  %c.promoted = load i32, i32* %c, align 4
238  %b.promoted = load i32, i32* %b, align 4
239  br label %for.body
240
241for.body:
242  %inc7 = phi i32 [ %d, %for.body.lr.ph ], [ %inc, %for.body ]
243  %and6 = phi i32 [ %b.promoted, %for.body.lr.ph ], [ %and, %for.body ]
244  %conv25 = phi i32 [ %c.promoted, %for.body.lr.ph ], [ %conv2, %for.body ]
245  %and = and i32 %and6, %conv25
246  %inc = add nsw i32 %inc7, 1
247  %cmp = icmp eq i32 %inc, 21
248  br i1 %cmp, label %for.cond.for.end_crit_edge, label %for.body
249
250for.cond.for.end_crit_edge:
251  %and.lcssa = phi i32 [ %and, %for.body ]
252  store i32 %conv2, i32* %c, align 4
253  store i32 %and.lcssa, i32* %b, align 4
254  store i16 %sub, i16* %e, align 2
255  br label %for.end
256
257for.end:
258  ret void
259}
260
261; CHECK-LABEL: @PR27246
262;
263; int PR27246() {
264;   unsigned int e, n;
265;   for (int i = 1; i < 49; ++i) {
266;     for (int k = i; k > 1; --k)
267;       e = k;
268;     n = e;
269;   }
270;   return n;
271; }
272;
273; CHECK-NOT: vector.ph:
274;
275define i32 @PR27246() {
276entry:
277  br label %for.cond1.preheader
278
279for.cond1.preheader:
280  %i.016 = phi i32 [ 1, %entry ], [ %inc, %for.cond.cleanup3 ]
281  %e.015 = phi i32 [ undef, %entry ], [ %e.1.lcssa, %for.cond.cleanup3 ]
282  br label %for.cond1
283
284for.cond.cleanup:
285  %e.1.lcssa.lcssa = phi i32 [ %e.1.lcssa, %for.cond.cleanup3 ]
286  ret i32 %e.1.lcssa.lcssa
287
288for.cond1:
289  %e.1 = phi i32 [ %k.0, %for.cond1 ], [ %e.015, %for.cond1.preheader ]
290  %k.0 = phi i32 [ %dec, %for.cond1 ], [ %i.016, %for.cond1.preheader ]
291  %cmp2 = icmp sgt i32 %k.0, 1
292  %dec = add nsw i32 %k.0, -1
293  br i1 %cmp2, label %for.cond1, label %for.cond.cleanup3
294
295for.cond.cleanup3:
296  %e.1.lcssa = phi i32 [ %e.1, %for.cond1 ]
297  %inc = add nuw nsw i32 %i.016, 1
298  %exitcond = icmp eq i32 %inc, 49
299  br i1 %exitcond, label %for.cond.cleanup, label %for.cond1.preheader
300}
301
302; CHECK-LABEL: @PR29559
303;
304; UNROLL-NO-IC: vector.ph:
305; UNROLL-NO-IC:   br label %vector.body
306;
307; UNROLL-NO-IC: vector.body:
308; UNROLL-NO-IC:   %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
309; UNROLL-NO-IC:   %vector.recur = phi <4 x float*> [ undef, %vector.ph ], [ %[[I4:.+]], %vector.body ]
310; UNROLL-NO-IC:   %[[G1:.+]] = getelementptr inbounds [3 x float], [3 x float]* undef, i64 0, i64 0
311; UNROLL-NO-IC:   %[[I1:.+]] = insertelement <4 x float*> undef, float* %[[G1]], i32 0
312; UNROLL-NO-IC:   %[[I2:.+]] = insertelement <4 x float*> %[[I1]], float* %[[G1]], i32 1
313; UNROLL-NO-IC:   %[[I3:.+]] = insertelement <4 x float*> %[[I2]], float* %[[G1]], i32 2
314; UNROLL-NO-IC:   %[[I4]] = insertelement <4 x float*> %[[I3]], float* %[[G1]], i32 3
315; UNROLL-NO-IC:   {{.*}} = shufflevector <4 x float*> %vector.recur, <4 x float*> %[[I4]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
316; UNROLL-NO-IC:   {{.*}} = shufflevector <4 x float*> %[[I4]], <4 x float*> %[[I4]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
317;
318; UNROLL-NO-IC: middle.block:
319; UNROLL-NO-IC:   %vector.recur.extract = extractelement <4 x float*> %[[I4]], i32 3
320;
321; UNROLL-NO-IC: scalar.ph:
322; UNROLL-NO-IC:   %scalar.recur.init = phi float* [ %vector.recur.extract, %middle.block ], [ undef, %min.iters.checked ], [ undef, %entry ]
323;
324; UNROLL-NO-IC: scalar.body:
325; UNROLL-NO-IC:   %scalar.recur = phi float* [ %scalar.recur.init, %scalar.ph ], [ {{.*}}, %scalar.body ]
326;
327define void @PR29559() {
328entry:
329  br label %scalar.body
330
331scalar.body:
332  %i = phi i64 [ 0, %entry ], [ %i.next, %scalar.body ]
333  %tmp2 = phi float* [ undef, %entry ], [ %tmp3, %scalar.body ]
334  %tmp3 = getelementptr inbounds [3 x float], [3 x float]* undef, i64 0, i64 0
335  %i.next = add nuw nsw i64 %i, 1
336  %cond = icmp eq i64 %i.next, undef
337  br i1 %cond, label %for.end, label %scalar.body
338
339for.end:
340  ret void
341}
342