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; RUN: opt < %s -loop-vectorize -force-vector-width=1 -force-vector-interleave=2 -S | FileCheck %s --check-prefix=UNROLL-NO-VF
5
6target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
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-LABEL: @recurrence_1(
14; CHECK:       vector.ph:
15; CHECK:         %vector.recur.init = insertelement <4 x i32> undef, i32 %pre_load, i32 3
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; CHECK:       middle.block:
21; CHECK:         %vector.recur.extract = extractelement <4 x i32> [[L1]], i32 3
22; CHECK:       scalar.ph:
23; CHECK:         %scalar.recur.init = phi i32 [ %vector.recur.extract, %middle.block ], [ %pre_load, %vector.memcheck ], [ %pre_load, %min.iters.checked ], [ %pre_load, %for.preheader ]
24; CHECK:       scalar.body:
25; CHECK:         %scalar.recur = phi i32 [ %scalar.recur.init, %scalar.ph ], [ {{.*}}, %scalar.body ]
26;
27; UNROLL-LABEL: @recurrence_1(
28; UNROLL:       vector.body:
29; UNROLL:         %vector.recur = phi <4 x i32> [ %vector.recur.init, %vector.ph ], [ [[L2:%[a-zA-Z0-9.]+]], %vector.body ]
30; UNROLL:         [[L1:%[a-zA-Z0-9.]+]] = load <4 x i32>
31; UNROLL:         [[L2]] = load <4 x i32>
32; UNROLL:         {{.*}} = shufflevector <4 x i32> %vector.recur, <4 x i32> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
33; UNROLL:         {{.*}} = shufflevector <4 x i32> [[L1]], <4 x i32> [[L2]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
34; UNROLL:       middle.block:
35; UNROLL:         %vector.recur.extract = extractelement <4 x i32> [[L2]], i32 3
36;
37define void @recurrence_1(i32* nocapture readonly %a, i32* nocapture %b, i32 %n) {
38entry:
39  br label %for.preheader
40
41for.preheader:
42  %arrayidx.phi.trans.insert = getelementptr inbounds i32, i32* %a, i64 0
43  %pre_load = load i32, i32* %arrayidx.phi.trans.insert
44  br label %scalar.body
45
46scalar.body:
47  %0 = phi i32 [ %pre_load, %for.preheader ], [ %1, %scalar.body ]
48  %indvars.iv = phi i64 [ 0, %for.preheader ], [ %indvars.iv.next, %scalar.body ]
49  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
50  %arrayidx32 = getelementptr inbounds i32, i32* %a, i64 %indvars.iv.next
51  %1 = load i32, i32* %arrayidx32
52  %arrayidx34 = getelementptr inbounds i32, i32* %b, i64 %indvars.iv
53  %add35 = add i32 %1, %0
54  store i32 %add35, i32* %arrayidx34
55  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
56  %exitcond = icmp eq i32 %lftr.wideiv, %n
57  br i1 %exitcond, label %for.exit, label %scalar.body
58
59for.exit:
60  ret void
61}
62
63; int recurrence_2(int *a, int n) {
64;   int minmax;
65;   for (int i = 0; i < n; ++i)
66;     minmax = min(minmax, max(a[i] - a[i-1], 0));
67;   return minmax;
68; }
69;
70; CHECK-LABEL: @recurrence_2(
71; CHECK:       vector.ph:
72; CHECK:         %vector.recur.init = insertelement <4 x i32> undef, i32 %.pre, i32 3
73; CHECK:       vector.body:
74; CHECK:         %vector.recur = phi <4 x i32> [ %vector.recur.init, %vector.ph ], [ [[L1:%[a-zA-Z0-9.]+]], %vector.body ]
75; CHECK:         [[L1]] = load <4 x i32>
76; CHECK:         {{.*}} = shufflevector <4 x i32> %vector.recur, <4 x i32> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
77; CHECK:       middle.block:
78; CHECK:         %vector.recur.extract = extractelement <4 x i32> [[L1]], i32 3
79; CHECK:       scalar.ph:
80; CHECK:         %scalar.recur.init = phi i32 [ %vector.recur.extract, %middle.block ], [ %.pre, %min.iters.checked ], [ %.pre, %for.preheader ]
81; CHECK:       scalar.body:
82; CHECK:         %scalar.recur = phi i32 [ %scalar.recur.init, %scalar.ph ], [ {{.*}}, %scalar.body ]
83;
84; UNROLL-LABEL: @recurrence_2(
85; UNROLL:       vector.body:
86; UNROLL:         %vector.recur = phi <4 x i32> [ %vector.recur.init, %vector.ph ], [ [[L2:%[a-zA-Z0-9.]+]], %vector.body ]
87; UNROLL:         [[L1:%[a-zA-Z0-9.]+]] = load <4 x i32>
88; UNROLL:         [[L2]] = load <4 x i32>
89; UNROLL:         {{.*}} = shufflevector <4 x i32> %vector.recur, <4 x i32> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
90; UNROLL:         {{.*}} = shufflevector <4 x i32> [[L1]], <4 x i32> [[L2]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
91; UNROLL:       middle.block:
92; UNROLL:         %vector.recur.extract = extractelement <4 x i32> [[L2]], i32 3
93;
94define i32 @recurrence_2(i32* nocapture readonly %a, i32 %n) {
95entry:
96  %cmp27 = icmp sgt i32 %n, 0
97  br i1 %cmp27, label %for.preheader, label %for.cond.cleanup
98
99for.preheader:
100  %arrayidx2.phi.trans.insert = getelementptr inbounds i32, i32* %a, i64 -1
101  %.pre = load i32, i32* %arrayidx2.phi.trans.insert, align 4
102  br label %scalar.body
103
104for.cond.cleanup.loopexit:
105  %minmax.0.cond.lcssa = phi i32 [ %minmax.0.cond, %scalar.body ]
106  br label %for.cond.cleanup
107
108for.cond.cleanup:
109  %minmax.0.lcssa = phi i32 [ undef, %entry ], [ %minmax.0.cond.lcssa, %for.cond.cleanup.loopexit ]
110  ret i32 %minmax.0.lcssa
111
112scalar.body:
113  %0 = phi i32 [ %.pre, %for.preheader ], [ %1, %scalar.body ]
114  %indvars.iv = phi i64 [ 0, %for.preheader ], [ %indvars.iv.next, %scalar.body ]
115  %minmax.028 = phi i32 [ undef, %for.preheader ], [ %minmax.0.cond, %scalar.body ]
116  %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
117  %1 = load i32, i32* %arrayidx, align 4
118  %sub3 = sub nsw i32 %1, %0
119  %cmp4 = icmp sgt i32 %sub3, 0
120  %cond = select i1 %cmp4, i32 %sub3, i32 0
121  %cmp5 = icmp slt i32 %minmax.028, %cond
122  %minmax.0.cond = select i1 %cmp5, i32 %minmax.028, i32 %cond
123  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
124  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
125  %exitcond = icmp eq i32 %lftr.wideiv, %n
126  br i1 %exitcond, label %for.cond.cleanup.loopexit, label %scalar.body
127}
128
129; void recurrence_3(short *a, double *b, int n, float f, short p) {
130;   b[0] = (double)a[0] - f * (double)p;
131;   for (int i = 1; i < n; i++)
132;     b[i] = (double)a[i] - f * (double)a[i - 1];
133; }
134;
135; CHECK-LABEL: @recurrence_3(
136; CHECK:       vector.ph:
137; CHECK:         %vector.recur.init = insertelement <4 x i16> undef, i16 %0, i32 3
138; CHECK:       vector.body:
139; CHECK:         %vector.recur = phi <4 x i16> [ %vector.recur.init, %vector.ph ], [ [[L1:%[a-zA-Z0-9.]+]], %vector.body ]
140; CHECK:         [[L1]] = load <4 x i16>
141; CHECK:         {{.*}} = shufflevector <4 x i16> %vector.recur, <4 x i16> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
142; CHECK:       middle.block:
143; CHECK:         %vector.recur.extract = extractelement <4 x i16> [[L1]], i32 3
144; CHECK:       scalar.ph:
145; CHECK:         %scalar.recur.init = phi i16 [ %vector.recur.extract, %middle.block ], [ %0, %vector.memcheck ], [ %0, %min.iters.checked ], [ %0, %for.preheader ]
146; CHECK:       scalar.body:
147; CHECK:         %scalar.recur = phi i16 [ %scalar.recur.init, %scalar.ph ], [ {{.*}}, %scalar.body ]
148;
149; UNROLL-LABEL: @recurrence_3(
150; UNROLL:       vector.body:
151; UNROLL:         %vector.recur = phi <4 x i16> [ %vector.recur.init, %vector.ph ], [ [[L2:%[a-zA-Z0-9.]+]], %vector.body ]
152; UNROLL:         [[L1:%[a-zA-Z0-9.]+]] = load <4 x i16>
153; UNROLL:         [[L2]] = load <4 x i16>
154; UNROLL:         {{.*}} = shufflevector <4 x i16> %vector.recur, <4 x i16> [[L1]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
155; UNROLL:         {{.*}} = shufflevector <4 x i16> [[L1]], <4 x i16> [[L2]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
156; UNROLL:       middle.block:
157; UNROLL:         %vector.recur.extract = extractelement <4 x i16> [[L2]], i32 3
158;
159define void @recurrence_3(i16* nocapture readonly %a, double* nocapture %b, i32 %n, float %f, i16 %p) {
160entry:
161  %0 = load i16, i16* %a, align 2
162  %conv = sitofp i16 %0 to double
163  %conv1 = fpext float %f to double
164  %conv2 = sitofp i16 %p to double
165  %mul = fmul fast double %conv2, %conv1
166  %sub = fsub fast double %conv, %mul
167  store double %sub, double* %b, align 8
168  %cmp25 = icmp sgt i32 %n, 1
169  br i1 %cmp25, label %for.preheader, label %for.end
170
171for.preheader:
172  br label %scalar.body
173
174scalar.body:
175  %1 = phi i16 [ %0, %for.preheader ], [ %2, %scalar.body ]
176  %advars.iv = phi i64 [ %advars.iv.next, %scalar.body ], [ 1, %for.preheader ]
177  %arrayidx5 = getelementptr inbounds i16, i16* %a, i64 %advars.iv
178  %2 = load i16, i16* %arrayidx5, align 2
179  %conv6 = sitofp i16 %2 to double
180  %conv11 = sitofp i16 %1 to double
181  %mul12 = fmul fast double %conv11, %conv1
182  %sub13 = fsub fast double %conv6, %mul12
183  %arrayidx15 = getelementptr inbounds double, double* %b, i64 %advars.iv
184  store double %sub13, double* %arrayidx15, align 8
185  %advars.iv.next = add nuw nsw i64 %advars.iv, 1
186  %lftr.wideiv = trunc i64 %advars.iv.next to i32
187  %exitcond = icmp eq i32 %lftr.wideiv, %n
188  br i1 %exitcond, label %for.end.loopexit, label %scalar.body
189
190for.end.loopexit:
191  br label %for.end
192
193for.end:
194  ret void
195}
196
197; void PR26734(short *a, int *b, int *c, int d, short *e) {
198;   for (; d != 21; d++) {
199;     *b &= *c;
200;     *e = *a - 6;
201;     *c = *e;
202;   }
203; }
204;
205; CHECK-LABEL: @PR26734(
206; CHECK-NOT:   vector.ph:
207; CHECK:       }
208;
209define void @PR26734(i16* %a, i32* %b, i32* %c, i32 %d, i16* %e) {
210entry:
211  %cmp4 = icmp eq i32 %d, 21
212  br i1 %cmp4, label %entry.for.end_crit_edge, label %for.body.lr.ph
213
214entry.for.end_crit_edge:
215  %.pre = load i32, i32* %b, align 4
216  br label %for.end
217
218for.body.lr.ph:
219  %0 = load i16, i16* %a, align 2
220  %sub = add i16 %0, -6
221  %conv2 = sext i16 %sub to i32
222  %c.promoted = load i32, i32* %c, align 4
223  %b.promoted = load i32, i32* %b, align 4
224  br label %for.body
225
226for.body:
227  %inc7 = phi i32 [ %d, %for.body.lr.ph ], [ %inc, %for.body ]
228  %and6 = phi i32 [ %b.promoted, %for.body.lr.ph ], [ %and, %for.body ]
229  %conv25 = phi i32 [ %c.promoted, %for.body.lr.ph ], [ %conv2, %for.body ]
230  %and = and i32 %and6, %conv25
231  %inc = add nsw i32 %inc7, 1
232  %cmp = icmp eq i32 %inc, 21
233  br i1 %cmp, label %for.cond.for.end_crit_edge, label %for.body
234
235for.cond.for.end_crit_edge:
236  %and.lcssa = phi i32 [ %and, %for.body ]
237  store i32 %conv2, i32* %c, align 4
238  store i32 %and.lcssa, i32* %b, align 4
239  store i16 %sub, i16* %e, align 2
240  br label %for.end
241
242for.end:
243  ret void
244}
245
246; int PR27246() {
247;   unsigned int e, n;
248;   for (int i = 1; i < 49; ++i) {
249;     for (int k = i; k > 1; --k)
250;       e = k;
251;     n = e;
252;   }
253;   return n;
254; }
255;
256; CHECK-LABEL: @PR27246(
257; CHECK-NOT:   vector.ph:
258; CHECK:       }
259;
260define i32 @PR27246() {
261entry:
262  br label %for.cond1.preheader
263
264for.cond1.preheader:
265  %i.016 = phi i32 [ 1, %entry ], [ %inc, %for.cond.cleanup3 ]
266  %e.015 = phi i32 [ undef, %entry ], [ %e.1.lcssa, %for.cond.cleanup3 ]
267  br label %for.cond1
268
269for.cond.cleanup:
270  %e.1.lcssa.lcssa = phi i32 [ %e.1.lcssa, %for.cond.cleanup3 ]
271  ret i32 %e.1.lcssa.lcssa
272
273for.cond1:
274  %e.1 = phi i32 [ %k.0, %for.cond1 ], [ %e.015, %for.cond1.preheader ]
275  %k.0 = phi i32 [ %dec, %for.cond1 ], [ %i.016, %for.cond1.preheader ]
276  %cmp2 = icmp sgt i32 %k.0, 1
277  %dec = add nsw i32 %k.0, -1
278  br i1 %cmp2, label %for.cond1, label %for.cond.cleanup3
279
280for.cond.cleanup3:
281  %e.1.lcssa = phi i32 [ %e.1, %for.cond1 ]
282  %inc = add nuw nsw i32 %i.016, 1
283  %exitcond = icmp eq i32 %inc, 49
284  br i1 %exitcond, label %for.cond.cleanup, label %for.cond1.preheader
285}
286
287; UNROLL-NO-IC-LABEL: @PR30183(
288; UNROLL-NO-IC:       vector.ph:
289; UNROLL-NO-IC-NEXT:    [[VECTOR_RECUR_INIT:%.*]] = insertelement <4 x i32> undef, i32 [[PRE_LOAD:%.*]], i32 3
290; UNROLL-NO-IC-NEXT:    br label %vector.body
291; UNROLL-NO-IC:       vector.body:
292; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
293; UNROLL-NO-IC-NEXT:    [[VECTOR_RECUR:%.*]] = phi <4 x i32> [ [[VECTOR_RECUR_INIT]], %vector.ph ], [ [[TMP42:%.*]], %vector.body ]
294; UNROLL-NO-IC:         [[TMP27:%.*]] = load i32, i32* {{.*}}
295; UNROLL-NO-IC-NEXT:    [[TMP28:%.*]] = load i32, i32* {{.*}}
296; UNROLL-NO-IC-NEXT:    [[TMP29:%.*]] = load i32, i32* {{.*}}
297; UNROLL-NO-IC-NEXT:    [[TMP30:%.*]] = load i32, i32* {{.*}}
298; UNROLL-NO-IC-NEXT:    [[TMP31:%.*]] = load i32, i32* {{.*}}
299; UNROLL-NO-IC-NEXT:    [[TMP32:%.*]] = load i32, i32* {{.*}}
300; UNROLL-NO-IC-NEXT:    [[TMP33:%.*]] = load i32, i32* {{.*}}
301; UNROLL-NO-IC-NEXT:    [[TMP34:%.*]] = load i32, i32* {{.*}}
302; UNROLL-NO-IC-NEXT:    [[TMP35:%.*]] = insertelement <4 x i32> undef, i32 [[TMP27]], i32 0
303; UNROLL-NO-IC-NEXT:    [[TMP36:%.*]] = insertelement <4 x i32> [[TMP35]], i32 [[TMP28]], i32 1
304; UNROLL-NO-IC-NEXT:    [[TMP37:%.*]] = insertelement <4 x i32> [[TMP36]], i32 [[TMP29]], i32 2
305; UNROLL-NO-IC-NEXT:    [[TMP38:%.*]] = insertelement <4 x i32> [[TMP37]], i32 [[TMP30]], i32 3
306; UNROLL-NO-IC-NEXT:    [[TMP39:%.*]] = insertelement <4 x i32> undef, i32 [[TMP31]], i32 0
307; UNROLL-NO-IC-NEXT:    [[TMP40:%.*]] = insertelement <4 x i32> [[TMP39]], i32 [[TMP32]], i32 1
308; UNROLL-NO-IC-NEXT:    [[TMP41:%.*]] = insertelement <4 x i32> [[TMP40]], i32 [[TMP33]], i32 2
309; UNROLL-NO-IC-NEXT:    [[TMP42]] = insertelement <4 x i32> [[TMP41]], i32 [[TMP34]], i32 3
310; UNROLL-NO-IC-NEXT:    [[TMP43:%.*]] = shufflevector <4 x i32> [[VECTOR_RECUR]], <4 x i32> [[TMP38]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
311; UNROLL-NO-IC-NEXT:    [[TMP44:%.*]] = shufflevector <4 x i32> [[TMP38]], <4 x i32> [[TMP42]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
312; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add i64 [[INDEX]], 8
313; UNROLL-NO-IC:         br i1 {{.*}}, label %middle.block, label %vector.body
314;
315define void @PR30183(i32 %pre_load, i32* %a, i32* %b, i64 %n) {
316entry:
317  br label %scalar.body
318
319scalar.body:
320  %i = phi i64 [ 0, %entry ], [ %i.next, %scalar.body ]
321  %tmp0 = phi i32 [ %pre_load, %entry ], [ %tmp2, %scalar.body ]
322  %i.next = add nuw nsw i64 %i, 2
323  %tmp1 = getelementptr inbounds i32, i32* %a, i64 %i.next
324  %tmp2 = load i32, i32* %tmp1
325  %cond = icmp eq i64 %i.next,%n
326  br i1 %cond, label %for.end, label %scalar.body
327
328for.end:
329  ret void
330}
331
332; UNROLL-NO-IC-LABEL: @constant_folded_previous_value(
333; UNROLL-NO-IC:       vector.body:
334; UNROLL-NO-IC:         [[VECTOR_RECUR:%.*]] = phi <4 x i64> [ <i64 undef, i64 undef, i64 undef, i64 0>, %vector.ph ], [ <i64 1, i64 1, i64 1, i64 1>, %vector.body ]
335; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = shufflevector <4 x i64> [[VECTOR_RECUR]], <4 x i64> <i64 1, i64 1, i64 1, i64 1>, <4 x i32> <i32 3, i32 4, i32 5, i32 6>
336; UNROLL-NO-IC:         br i1 {{.*}}, label %middle.block, label %vector.body
337;
338define void @constant_folded_previous_value() {
339entry:
340  br label %scalar.body
341
342scalar.body:
343  %i = phi i64 [ 0, %entry ], [ %i.next, %scalar.body ]
344  %tmp2 = phi i64 [ 0, %entry ], [ %tmp3, %scalar.body ]
345  %tmp3 = add i64 0, 1
346  %i.next = add nuw nsw i64 %i, 1
347  %cond = icmp eq i64 %i.next, undef
348  br i1 %cond, label %for.end, label %scalar.body
349
350for.end:
351  ret void
352}
353
354; We vectorize this first order recurrence, by generating two
355; extracts for the phi `val.phi` - one at the last index and
356; another at the second last index. We need these 2 extracts because
357; the first order recurrence phi is used outside the loop, so we require the phi
358; itself and not its update (addx).
359; UNROLL-NO-IC-LABEL: extract_second_last_iteration
360; UNROLL-NO-IC: vector.body
361; UNROLL-NO-IC:   %step.add = add <4 x i32> %vec.ind, <i32 4, i32 4, i32 4, i32 4>
362; UNROLL-NO-IC:   %[[L1:.+]] = add <4 x i32> %vec.ind, %broadcast.splat
363; UNROLL-NO-IC:   %[[L2:.+]] = add <4 x i32> %step.add, %broadcast.splat
364; UNROLL-NO-IC:   %index.next = add i32 %index, 8
365; UNROLL-NO-IC:   icmp eq i32 %index.next, 96
366; UNROLL-NO-IC: middle.block
367; UNROLL-NO-IC:   icmp eq i32 96, 96
368; UNROLL-NO-IC:   %vector.recur.extract = extractelement <4 x i32> %[[L2]], i32 3
369; UNROLL-NO-IC:   %vector.recur.extract.for.phi = extractelement <4 x i32> %[[L2]], i32 2
370; UNROLL-NO-IC: for.end
371; UNROLL-NO-IC:   %val.phi.lcssa = phi i32 [ %scalar.recur, %for.body ], [ %vector.recur.extract.for.phi, %middle.block ]
372; Check the case when unrolled but not vectorized.
373; UNROLL-NO-VF-LABEL: extract_second_last_iteration
374; UNROLL-NO-VF: vector.body:
375; UNROLL-NO-VF:   %induction = add i32 %index, 0
376; UNROLL-NO-VF:   %induction1 = add i32 %index, 1
377; UNROLL-NO-VF:   %[[L1:.+]] = add i32 %induction, %x
378; UNROLL-NO-VF:   %[[L2:.+]] = add i32 %induction1, %x
379; UNROLL-NO-VF:   %index.next = add i32 %index, 2
380; UNROLL-NO-VF:   icmp eq i32 %index.next, 96
381; UNROLL-NO-VF: for.end:
382; UNROLL-NO-VF:   %val.phi.lcssa = phi i32 [ %scalar.recur, %for.body ], [ %[[L1]], %middle.block ]
383define i32 @extract_second_last_iteration(i32* %cval, i32 %x)  {
384entry:
385  br label %for.body
386
387for.body:
388  %inc.phi = phi i32 [ 0, %entry ], [ %inc, %for.body ]
389  %val.phi = phi i32 [ 0, %entry ], [ %addx, %for.body ]
390  %inc = add i32 %inc.phi, 1
391  %bc = zext i32 %inc.phi to i64
392  %addx = add i32 %inc.phi, %x
393  %cmp = icmp eq i32 %inc.phi, 95
394  br i1 %cmp, label %for.end, label %for.body
395
396for.end:
397  ret i32 %val.phi
398}
399