1; RUN: opt < %s -S -loop-flatten -debug-only=loop-flatten 2>&1 | FileCheck %s
2; REQUIRES: asserts
3
4target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
5
6; Every function in this file has a reason that it can't be transformed.
7
8; CHECK-NOT: Checks all passed, doing the transformation
9
10; Outer loop does not start at zero
11define void @test_1(i32 %N, i32* nocapture %C, i32* nocapture readonly %A, i32 %scale) {
12entry:
13  %cmp25 = icmp sgt i32 %N, 0
14  br i1 %cmp25, label %for.body4.lr.ph, label %for.cond.cleanup
15
16for.body4.lr.ph:
17  %i.026 = phi i32 [ %inc10, %for.cond.cleanup3 ], [ 1, %entry ]
18  %mul = mul nsw i32 %i.026, %N
19  br label %for.body4
20
21for.body4:
22  %j.024 = phi i32 [ 0, %for.body4.lr.ph ], [ %inc, %for.body4 ]
23  %add = add nsw i32 %j.024, %mul
24  %arrayidx = getelementptr inbounds i32, i32* %A, i32 %add
25  %0 = load i32, i32* %arrayidx, align 4
26  %mul5 = mul nsw i32 %0, %scale
27  %arrayidx8 = getelementptr inbounds i32, i32* %C, i32 %add
28  store i32 %mul5, i32* %arrayidx8, align 4
29  %inc = add nuw nsw i32 %j.024, 1
30  %exitcond = icmp eq i32 %inc, %N
31  br i1 %exitcond, label %for.cond.cleanup3, label %for.body4
32
33for.cond.cleanup3:
34  %inc10 = add nuw nsw i32 %i.026, 1
35  %exitcond27 = icmp eq i32 %inc10, %N
36  br i1 %exitcond27, label %for.cond.cleanup, label %for.body4.lr.ph
37
38for.cond.cleanup:
39  ret void
40}
41
42; Inner loop does not start at zero
43define void @test_2(i32 %N, i32* nocapture %C, i32* nocapture readonly %A, i32 %scale) {
44entry:
45  %cmp25 = icmp sgt i32 %N, 0
46  br i1 %cmp25, label %for.body4.lr.ph, label %for.cond.cleanup
47
48for.body4.lr.ph:
49  %i.026 = phi i32 [ %inc10, %for.cond.cleanup3 ], [ 0, %entry ]
50  %mul = mul nsw i32 %i.026, %N
51  br label %for.body4
52
53for.body4:
54  %j.024 = phi i32 [ 1, %for.body4.lr.ph ], [ %inc, %for.body4 ]
55  %add = add nsw i32 %j.024, %mul
56  %arrayidx = getelementptr inbounds i32, i32* %A, i32 %add
57  %0 = load i32, i32* %arrayidx, align 4
58  %mul5 = mul nsw i32 %0, %scale
59  %arrayidx8 = getelementptr inbounds i32, i32* %C, i32 %add
60  store i32 %mul5, i32* %arrayidx8, align 4
61  %inc = add nuw nsw i32 %j.024, 1
62  %exitcond = icmp eq i32 %inc, %N
63  br i1 %exitcond, label %for.cond.cleanup3, label %for.body4
64
65for.cond.cleanup3:
66  %inc10 = add nuw nsw i32 %i.026, 1
67  %exitcond27 = icmp eq i32 %inc10, %N
68  br i1 %exitcond27, label %for.cond.cleanup, label %for.body4.lr.ph
69
70for.cond.cleanup:
71  ret void
72}
73
74; Outer IV used directly
75define hidden void @test_3(i16 zeroext %N, i32* nocapture %C, i32* nocapture readonly %A, i32 %scale) {
76entry:
77  %conv = zext i16 %N to i32
78  %cmp25 = icmp eq i16 %N, 0
79  br i1 %cmp25, label %for.cond.cleanup, label %for.body.lr.ph.split.us
80
81for.body.lr.ph.split.us:                          ; preds = %entry
82  br label %for.body.us
83
84for.body.us:                                      ; preds = %for.cond2.for.cond.cleanup6_crit_edge.us, %for.body.lr.ph.split.us
85  %i.026.us = phi i32 [ 0, %for.body.lr.ph.split.us ], [ %inc12.us, %for.cond2.for.cond.cleanup6_crit_edge.us ]
86  %arrayidx.us = getelementptr inbounds i32, i32* %A, i32 %i.026.us
87  %mul9.us = mul nuw nsw i32 %i.026.us, %conv
88  br label %for.body7.us
89
90for.body7.us:                                     ; preds = %for.body.us, %for.body7.us
91  %j.024.us = phi i32 [ 0, %for.body.us ], [ %inc.us, %for.body7.us ]
92  %0 = load i32, i32* %arrayidx.us, align 4
93  %mul.us = mul nsw i32 %0, %scale
94  %add.us = add nuw nsw i32 %j.024.us, %mul9.us
95  %arrayidx10.us = getelementptr inbounds i32, i32* %C, i32 %add.us
96  store i32 %mul.us, i32* %arrayidx10.us, align 4
97  %inc.us = add nuw nsw i32 %j.024.us, 1
98  %exitcond = icmp ne i32 %inc.us, %conv
99  br i1 %exitcond, label %for.body7.us, label %for.cond2.for.cond.cleanup6_crit_edge.us
100
101for.cond2.for.cond.cleanup6_crit_edge.us:         ; preds = %for.body7.us
102  %inc12.us = add nuw nsw i32 %i.026.us, 1
103  %exitcond27 = icmp ne i32 %inc12.us, %conv
104  br i1 %exitcond27, label %for.body.us, label %for.cond.cleanup.loopexit
105
106for.cond.cleanup.loopexit:                        ; preds = %for.cond2.for.cond.cleanup6_crit_edge.us
107  br label %for.cond.cleanup
108
109for.cond.cleanup:                                 ; preds = %for.cond.cleanup.loopexit, %entry
110  ret void
111}
112
113; Inner IV used directly
114define hidden void @test_4(i16 zeroext %N, i32* nocapture %C, i32* nocapture readonly %A, i32 %scale) {
115entry:
116  %conv = zext i16 %N to i32
117  %cmp25 = icmp eq i16 %N, 0
118  br i1 %cmp25, label %for.cond.cleanup, label %for.body.lr.ph.split.us
119
120for.body.lr.ph.split.us:                          ; preds = %entry
121  br label %for.body.us
122
123for.body.us:                                      ; preds = %for.cond2.for.cond.cleanup6_crit_edge.us, %for.body.lr.ph.split.us
124  %i.026.us = phi i32 [ 0, %for.body.lr.ph.split.us ], [ %inc12.us, %for.cond2.for.cond.cleanup6_crit_edge.us ]
125  %mul9.us = mul nuw nsw i32 %i.026.us, %conv
126  br label %for.body7.us
127
128for.body7.us:                                     ; preds = %for.body.us, %for.body7.us
129  %j.024.us = phi i32 [ 0, %for.body.us ], [ %inc.us, %for.body7.us ]
130  %arrayidx.us = getelementptr inbounds i32, i32* %A, i32 %j.024.us
131  %0 = load i32, i32* %arrayidx.us, align 4
132  %mul.us = mul nsw i32 %0, %scale
133  %add.us = add nuw nsw i32 %j.024.us, %mul9.us
134  %arrayidx10.us = getelementptr inbounds i32, i32* %C, i32 %add.us
135  store i32 %mul.us, i32* %arrayidx10.us, align 4
136  %inc.us = add nuw nsw i32 %j.024.us, 1
137  %exitcond = icmp ne i32 %inc.us, %conv
138  br i1 %exitcond, label %for.body7.us, label %for.cond2.for.cond.cleanup6_crit_edge.us
139
140for.cond2.for.cond.cleanup6_crit_edge.us:         ; preds = %for.body7.us
141  %inc12.us = add nuw nsw i32 %i.026.us, 1
142  %exitcond27 = icmp ne i32 %inc12.us, %conv
143  br i1 %exitcond27, label %for.body.us, label %for.cond.cleanup.loopexit
144
145for.cond.cleanup.loopexit:                        ; preds = %for.cond2.for.cond.cleanup6_crit_edge.us
146  br label %for.cond.cleanup
147
148for.cond.cleanup:                                 ; preds = %for.cond.cleanup.loopexit, %entry
149  ret void
150}
151
152; Inner iteration count not invariant in outer loop
153declare i32 @get_int() readonly
154define void @test_5(i16 zeroext %N, i32* nocapture %C, i32* nocapture readonly %A, i32 %scale) {
155entry:
156  %conv = zext i16 %N to i32
157  %cmp27 = icmp eq i16 %N, 0
158  br i1 %cmp27, label %for.cond.cleanup, label %for.body.lr.ph
159
160for.body.lr.ph:                                   ; preds = %entry
161  br label %for.body
162
163for.cond.cleanup.loopexit:                        ; preds = %for.cond.cleanup5
164  br label %for.cond.cleanup
165
166for.cond.cleanup:                                 ; preds = %for.cond.cleanup.loopexit, %entry
167  ret void
168
169for.body:                                         ; preds = %for.body.lr.ph, %for.cond.cleanup5
170  %i.028 = phi i32 [ 0, %for.body.lr.ph ], [ %inc12, %for.cond.cleanup5 ]
171  %call = tail call i32 @get_int()
172  %cmp325 = icmp sgt i32 %call, 0
173  br i1 %cmp325, label %for.body6.lr.ph, label %for.cond.cleanup5
174
175for.body6.lr.ph:                                  ; preds = %for.body
176  %mul = mul nsw i32 %call, %i.028
177  br label %for.body6
178
179for.cond.cleanup5.loopexit:                       ; preds = %for.body6
180  br label %for.cond.cleanup5
181
182for.cond.cleanup5:                                ; preds = %for.cond.cleanup5.loopexit, %for.body
183  %inc12 = add nuw nsw i32 %i.028, 1
184  %exitcond29 = icmp ne i32 %inc12, %conv
185  br i1 %exitcond29, label %for.body, label %for.cond.cleanup.loopexit
186
187for.body6:                                        ; preds = %for.body6.lr.ph, %for.body6
188  %j.026 = phi i32 [ 0, %for.body6.lr.ph ], [ %inc, %for.body6 ]
189  %add = add nsw i32 %j.026, %mul
190  %arrayidx = getelementptr inbounds i32, i32* %A, i32 %add
191  %0 = load i32, i32* %arrayidx, align 4
192  %mul7 = mul nsw i32 %0, %scale
193  %arrayidx10 = getelementptr inbounds i32, i32* %C, i32 %add
194  store i32 %mul7, i32* %arrayidx10, align 4
195  %inc = add nuw nsw i32 %j.026, 1
196  %exitcond = icmp ne i32 %inc, %call
197  br i1 %exitcond, label %for.body6, label %for.cond.cleanup5.loopexit
198}
199
200; Inner loop has an early exit
201define hidden void @test_6(i16 zeroext %N, i32* nocapture %C, i32* nocapture readonly %A, i32 %scale) {
202entry:
203  %conv = zext i16 %N to i32
204  %cmp39 = icmp eq i16 %N, 0
205  br i1 %cmp39, label %for.cond.cleanup, label %for.body.us.preheader
206
207for.body.us.preheader:                            ; preds = %entry
208  br label %for.body.us
209
210for.body.us:                                      ; preds = %for.body.us.preheader, %cleanup.us
211  %i.040.us = phi i32 [ %inc19.us, %cleanup.us ], [ 0, %for.body.us.preheader ]
212  %mul.us = mul nuw nsw i32 %i.040.us, %conv
213  br label %for.body7.us
214
215for.body7.us:                                     ; preds = %for.body.us, %if.end.us
216  %j.038.us = phi i32 [ 0, %for.body.us ], [ %inc.us, %if.end.us ]
217  %add.us = add nuw nsw i32 %j.038.us, %mul.us
218  %arrayidx.us = getelementptr inbounds i32, i32* %A, i32 %add.us
219  %0 = load i32, i32* %arrayidx.us, align 4
220  %tobool.us = icmp eq i32 %0, 0
221  br i1 %tobool.us, label %if.end.us, label %cleanup.us
222
223cleanup.us:                                       ; preds = %if.end.us, %for.body7.us
224  %inc19.us = add nuw nsw i32 %i.040.us, 1
225  %exitcond = icmp eq i32 %inc19.us, %conv
226  br i1 %exitcond, label %for.cond.cleanup, label %for.body.us
227
228if.end.us:                                        ; preds = %for.body7.us
229  %arrayidx17.us = getelementptr inbounds i32, i32* %C, i32 %add.us
230  store i32 0, i32* %arrayidx17.us, align 4
231  %inc.us = add nuw nsw i32 %j.038.us, 1
232  %cmp4.us = icmp ult i32 %inc.us, %conv
233  br i1 %cmp4.us, label %for.body7.us, label %cleanup.us
234
235for.cond.cleanup:                                 ; preds = %cleanup.us, %entry
236  ret void
237}
238
239define hidden void @test_7(i16 zeroext %N, i32* nocapture %C, i32* nocapture readonly %A, i32 %scale) {
240entry:
241  %conv = zext i16 %N to i32
242  %cmp30 = icmp eq i16 %N, 0
243  br i1 %cmp30, label %cleanup, label %for.body.us.preheader
244
245for.body.us.preheader:                            ; preds = %entry
246  br label %for.body.us
247
248for.body.us:                                      ; preds = %for.body.us.preheader, %for.cond2.for.cond.cleanup6_crit_edge.us
249  %i.031.us = phi i32 [ %inc15.us, %for.cond2.for.cond.cleanup6_crit_edge.us ], [ 0, %for.body.us.preheader ]
250  %call.us = tail call i32 @get_int() #2
251  %tobool.us = icmp eq i32 %call.us, 0
252  br i1 %tobool.us, label %for.body7.lr.ph.us, label %cleanup
253
254for.body7.us:                                     ; preds = %for.body7.us, %for.body7.lr.ph.us
255  %j.029.us = phi i32 [ 0, %for.body7.lr.ph.us ], [ %inc.us, %for.body7.us ]
256  %add.us = add nuw nsw i32 %j.029.us, %mul.us
257  %arrayidx.us = getelementptr inbounds i32, i32* %A, i32 %add.us
258  %0 = load i32, i32* %arrayidx.us, align 4
259  %mul9.us = mul nsw i32 %0, %scale
260  %arrayidx13.us = getelementptr inbounds i32, i32* %C, i32 %add.us
261  store i32 %mul9.us, i32* %arrayidx13.us, align 4
262  %inc.us = add nuw nsw i32 %j.029.us, 1
263  %exitcond = icmp eq i32 %inc.us, %conv
264  br i1 %exitcond, label %for.cond2.for.cond.cleanup6_crit_edge.us, label %for.body7.us
265
266for.body7.lr.ph.us:                               ; preds = %for.body.us
267  %mul.us = mul nuw nsw i32 %i.031.us, %conv
268  br label %for.body7.us
269
270for.cond2.for.cond.cleanup6_crit_edge.us:         ; preds = %for.body7.us
271  %inc15.us = add nuw nsw i32 %i.031.us, 1
272  %cmp.us = icmp ult i32 %inc15.us, %conv
273  br i1 %cmp.us, label %for.body.us, label %cleanup
274
275cleanup:                                          ; preds = %for.cond2.for.cond.cleanup6_crit_edge.us, %for.body.us, %entry
276  ret void
277}
278
279; Step is not 1
280define i32 @test_8(i32 %val, i16* nocapture %A) {
281entry:
282  br label %for.body
283
284for.body:                                         ; preds = %entry, %for.inc6
285  %i.018 = phi i32 [ 0, %entry ], [ %inc7, %for.inc6 ]
286  %mul = mul nuw nsw i32 %i.018, 20
287  br label %for.body3
288
289for.body3:                                        ; preds = %for.body, %for.body3
290  %j.017 = phi i32 [ 0, %for.body ], [ %inc, %for.body3 ]
291  %add = add nuw nsw i32 %j.017, %mul
292  %arrayidx = getelementptr inbounds i16, i16* %A, i32 %add
293  %0 = load i16, i16* %arrayidx, align 2
294  %conv16 = zext i16 %0 to i32
295  %add4 = add i32 %conv16, %val
296  %conv5 = trunc i32 %add4 to i16
297  store i16 %conv5, i16* %arrayidx, align 2
298  %inc = add nuw nsw i32 %j.017, 1
299  %exitcond = icmp ne i32 %inc, 20
300  br i1 %exitcond, label %for.body3, label %for.inc6
301
302for.inc6:                                         ; preds = %for.body3
303  %inc7 = add nuw nsw i32 %i.018, 2
304  %exitcond19 = icmp ne i32 %inc7, 10
305  br i1 %exitcond19, label %for.body, label %for.end8
306
307for.end8:                                         ; preds = %for.inc6
308  ret i32 10
309}
310
311
312; Step is not 1
313define i32 @test_9(i32 %val, i16* nocapture %A) {
314entry:
315  br label %for.body
316
317for.body:                                         ; preds = %entry, %for.inc6
318  %i.018 = phi i32 [ 0, %entry ], [ %inc7, %for.inc6 ]
319  %mul = mul nuw nsw i32 %i.018, 20
320  br label %for.body3
321
322for.body3:                                        ; preds = %for.body, %for.body3
323  %j.017 = phi i32 [ 0, %for.body ], [ %inc, %for.body3 ]
324  %add = add nuw nsw i32 %j.017, %mul
325  %arrayidx = getelementptr inbounds i16, i16* %A, i32 %add
326  %0 = load i16, i16* %arrayidx, align 2
327  %conv16 = zext i16 %0 to i32
328  %add4 = add i32 %conv16, %val
329  %conv5 = trunc i32 %add4 to i16
330  store i16 %conv5, i16* %arrayidx, align 2
331  %inc = add nuw nsw i32 %j.017, 2
332  %exitcond = icmp ne i32 %inc, 20
333  br i1 %exitcond, label %for.body3, label %for.inc6
334
335for.inc6:                                         ; preds = %for.body3
336  %inc7 = add nuw nsw i32 %i.018, 1
337  %exitcond19 = icmp ne i32 %inc7, 10
338  br i1 %exitcond19, label %for.body, label %for.end8
339
340for.end8:                                         ; preds = %for.inc6
341  ret i32 10
342}
343
344; When the loop trip count is a constant (e.g. 20) and the step size is
345; 1, InstCombine causes the transformation icmp ult i32 %inc, 20 ->
346; icmp ult i32 %j, 19. In this case a valid trip count is not found so
347; the loop is not flattened.
348define i32 @test9(i32* nocapture %A) {
349entry:
350  br label %for.cond1.preheader
351
352for.cond1.preheader:
353  %i.017 = phi i32 [ 0, %entry ], [ %inc6, %for.cond.cleanup3 ]
354  %mul = mul i32 %i.017, 20
355  br label %for.body4
356
357for.cond.cleanup3:
358  %inc6 = add i32 %i.017, 1
359  %cmp = icmp ult i32 %inc6, 11
360  br i1 %cmp, label %for.cond1.preheader, label %for.cond.cleanup
361
362for.body4:
363  %j.016 = phi i32 [ 0, %for.cond1.preheader ], [ %inc, %for.body4 ]
364  %add = add i32 %j.016, %mul
365  %arrayidx = getelementptr inbounds i32, i32* %A, i32 %add
366  store i32 30, i32* %arrayidx, align 4
367  %inc = add nuw nsw i32 %j.016, 1
368  %cmp2 = icmp ult i32 %j.016, 19
369  br i1 %cmp2, label %for.body4, label %for.cond.cleanup3
370
371for.cond.cleanup:
372  %0 = load i32, i32* %A, align 4
373  ret i32 %0
374}
375
376; Outer loop conditional phi
377define i32 @e() {
378entry:
379  br label %for.body
380
381for.body:                                         ; preds = %entry, %for.end16
382  %f.033 = phi i32 [ 0, %entry ], [ %inc18, %for.end16 ]
383  %g.032 = phi i32 [ undef, %entry ], [ %g.3.lcssa, %for.end16 ]
384  %.pr = add i32 10, 10
385  %tobool29 = icmp eq i32 %.pr, 0
386  br i1 %tobool29, label %for.end, label %for.body2.lr.ph
387
388for.body2.lr.ph:                                  ; preds = %for.body
389  br label %for.cond1.for.end_crit_edge
390
391for.cond1.for.end_crit_edge:                      ; preds = %for.body2.lr.ph
392  br label %for.end
393
394for.end:                                          ; preds = %for.cond1.for.end_crit_edge, %for.body
395  %g.1.lcssa = phi i32 [ 0, %for.cond1.for.end_crit_edge ], [ %g.032, %for.body ]
396  br label %for.body5
397
398for.body5:                                        ; preds = %for.end, %lor.end
399  %i.031 = phi i32 [ 0, %for.end ], [ %inc15, %lor.end ]
400  %g.230 = phi i32 [ %g.1.lcssa, %for.end ], [ %g.3, %lor.end ]
401  %0 = add i32 10, 10
402  %1 = add i32 10, 10
403  %tobool9 = icmp eq i32 %1, 0
404  br i1 %tobool9, label %lor.rhs, label %lor.end
405
406lor.rhs:                                          ; preds = %for.body5
407  %2 = add i32 10, 10
408  %call11 = add i32 10, 10
409  %tobool12 = icmp ne i32 %call11, 0
410  br label %lor.end
411
412lor.end:                                          ; preds = %for.body5, %lor.rhs
413  %g.3 = phi i32 [ %g.230, %for.body5 ], [ %call11, %lor.rhs ]
414  %3 = phi i1 [ true, %for.body5 ], [ %tobool12, %lor.rhs ]
415  %lor.ext = zext i1 %3 to i32
416  %inc15 = add nuw nsw i32 %i.031, 1
417  %exitcond = icmp ne i32 %inc15, 9
418  br i1 %exitcond, label %for.body5, label %for.end16
419
420for.end16:                                        ; preds = %lor.end
421  %g.3.lcssa = phi i32 [ %g.3, %lor.end ]
422  %inc18 = add nuw nsw i32 %f.033, 1
423  %exitcond34 = icmp ne i32 %inc18, 7
424  br i1 %exitcond34, label %for.body, label %for.end19
425
426for.end19:                                        ; preds = %for.end16
427  ret i32 undef
428}
429
430; A 3d loop corresponding to:
431;
432; for (int i = 0; i < N; ++i)
433;    for (int j = 0; j < N; ++j)
434;      for (int k = 0; k < N; ++k)
435;        f(&A[i + N * (j + N * k)]);
436;
437define void @d3_1(i32* %A, i32 %N) {
438entry:
439  %cmp35 = icmp sgt i32 %N, 0
440  br i1 %cmp35, label %for.cond1.preheader.lr.ph, label %for.cond.cleanup
441
442for.cond1.preheader.lr.ph:
443  br label %for.cond1.preheader.us
444
445for.cond1.preheader.us:
446  %i.036.us = phi i32 [ 0, %for.cond1.preheader.lr.ph ], [ %inc15.us, %for.cond1.for.cond.cleanup3_crit_edge.us ]
447  br i1 true, label %for.cond5.preheader.us.us.preheader, label %for.cond5.preheader.us52.preheader
448
449for.cond5.preheader.us52.preheader:
450  br label %for.cond5.preheader.us52
451
452for.cond5.preheader.us.us.preheader:
453  br label %for.cond5.preheader.us.us
454
455for.cond5.preheader.us52:
456  br i1 false, label %for.cond5.preheader.us52, label %for.cond1.for.cond.cleanup3_crit_edge.us.loopexit58
457
458for.cond1.for.cond.cleanup3_crit_edge.us.loopexit:
459  br label %for.cond1.for.cond.cleanup3_crit_edge.us
460
461for.cond1.for.cond.cleanup3_crit_edge.us.loopexit58:
462  br label %for.cond1.for.cond.cleanup3_crit_edge.us
463
464for.cond1.for.cond.cleanup3_crit_edge.us:
465  %inc15.us = add nuw nsw i32 %i.036.us, 1
466  %cmp.us = icmp slt i32 %inc15.us, %N
467  br i1 %cmp.us, label %for.cond1.preheader.us, label %for.cond.cleanup.loopexit
468
469for.cond5.preheader.us.us:
470  %j.033.us.us = phi i32 [ %inc12.us.us, %for.cond5.for.cond.cleanup7_crit_edge.us.us ], [ 0, %for.cond5.preheader.us.us.preheader ]
471  br label %for.body8.us.us
472
473for.cond5.for.cond.cleanup7_crit_edge.us.us:
474  %inc12.us.us = add nuw nsw i32 %j.033.us.us, 1
475  %cmp2.us.us = icmp slt i32 %inc12.us.us, %N
476  br i1 %cmp2.us.us, label %for.cond5.preheader.us.us, label %for.cond1.for.cond.cleanup3_crit_edge.us.loopexit
477
478for.body8.us.us:
479  %k.031.us.us = phi i32 [ 0, %for.cond5.preheader.us.us ], [ %inc.us.us, %for.body8.us.us ]
480  %mul.us.us = mul nsw i32 %k.031.us.us, %N
481  %add.us.us = add nsw i32 %mul.us.us, %j.033.us.us
482  %mul9.us.us = mul nsw i32 %add.us.us, %N
483  %add10.us.us = add nsw i32 %mul9.us.us, %i.036.us
484  %idxprom.us.us = sext i32 %add10.us.us to i64
485  %arrayidx.us.us = getelementptr inbounds i32, i32* %A, i64 %idxprom.us.us
486  tail call void @f(i32* %arrayidx.us.us) #2
487  %inc.us.us = add nuw nsw i32 %k.031.us.us, 1
488  %cmp6.us.us = icmp slt i32 %inc.us.us, %N
489  br i1 %cmp6.us.us, label %for.body8.us.us, label %for.cond5.for.cond.cleanup7_crit_edge.us.us
490
491for.cond.cleanup.loopexit:
492  br label %for.cond.cleanup
493
494for.cond.cleanup:
495  ret void
496}
497
498; A 3d loop corresponding to:
499;
500;   for (int k = 0; k < N; ++k)
501;    for (int i = 0; i < N; ++i)
502;      for (int j = 0; j < M; ++j)
503;        f(&A[i*M+j]);
504;
505; This could be supported, but isn't at the moment.
506;
507define void @d3_2(i32* %A, i32 %N, i32 %M) {
508entry:
509  %cmp30 = icmp sgt i32 %N, 0
510  br i1 %cmp30, label %for.cond1.preheader.lr.ph, label %for.cond.cleanup
511
512for.cond1.preheader.lr.ph:
513  %cmp625 = icmp sgt i32 %M, 0
514  br label %for.cond1.preheader.us
515
516for.cond1.preheader.us:
517  %k.031.us = phi i32 [ 0, %for.cond1.preheader.lr.ph ], [ %inc13.us, %for.cond1.for.cond.cleanup3_crit_edge.us ]
518  br i1 %cmp625, label %for.cond5.preheader.us.us.preheader, label %for.cond5.preheader.us43.preheader
519
520for.cond5.preheader.us43.preheader:
521  br label %for.cond1.for.cond.cleanup3_crit_edge.us.loopexit50
522
523for.cond5.preheader.us.us.preheader:
524  br label %for.cond5.preheader.us.us
525
526for.cond1.for.cond.cleanup3_crit_edge.us.loopexit:
527  br label %for.cond1.for.cond.cleanup3_crit_edge.us
528
529for.cond1.for.cond.cleanup3_crit_edge.us.loopexit50:
530  br label %for.cond1.for.cond.cleanup3_crit_edge.us
531
532for.cond1.for.cond.cleanup3_crit_edge.us:
533  %inc13.us = add nuw nsw i32 %k.031.us, 1
534  %exitcond52 = icmp ne i32 %inc13.us, %N
535  br i1 %exitcond52, label %for.cond1.preheader.us, label %for.cond.cleanup.loopexit
536
537for.cond5.preheader.us.us:
538  %i.028.us.us = phi i32 [ %inc10.us.us, %for.cond5.for.cond.cleanup7_crit_edge.us.us ], [ 0, %for.cond5.preheader.us.us.preheader ]
539  %mul.us.us = mul nsw i32 %i.028.us.us, %M
540  br label %for.body8.us.us
541
542for.cond5.for.cond.cleanup7_crit_edge.us.us:
543  %inc10.us.us = add nuw nsw i32 %i.028.us.us, 1
544  %exitcond51 = icmp ne i32 %inc10.us.us, %N
545  br i1 %exitcond51, label %for.cond5.preheader.us.us, label %for.cond1.for.cond.cleanup3_crit_edge.us.loopexit
546
547for.body8.us.us:
548  %j.026.us.us = phi i32 [ 0, %for.cond5.preheader.us.us ], [ %inc.us.us, %for.body8.us.us ]
549  %add.us.us = add nsw i32 %j.026.us.us, %mul.us.us
550  %idxprom.us.us = sext i32 %add.us.us to i64
551  %arrayidx.us.us = getelementptr inbounds i32, i32* %A, i64 %idxprom.us.us
552  tail call void @f(i32* %arrayidx.us.us) #2
553  %inc.us.us = add nuw nsw i32 %j.026.us.us, 1
554  %exitcond = icmp ne i32 %inc.us.us, %M
555  br i1 %exitcond, label %for.body8.us.us, label %for.cond5.for.cond.cleanup7_crit_edge.us.us
556
557for.cond.cleanup.loopexit:
558  br label %for.cond.cleanup
559
560for.cond.cleanup:
561  ret void
562}
563
564; A 3d loop corresponding to:
565;
566;   for (int i = 0; i < N; ++i)
567;     for (int j = 0; j < M; ++j) {
568;       A[i*M+j] = 0;
569;       for (int k = 0; k < N; ++k)
570;         g();
571;     }
572;
573define void @d3_3(i32* nocapture %A, i32 %N, i32 %M) {
574entry:
575  %cmp29 = icmp sgt i32 %N, 0
576  br i1 %cmp29, label %for.cond1.preheader.lr.ph, label %for.cond.cleanup
577
578for.cond1.preheader.lr.ph:
579  %cmp227 = icmp sgt i32 %M, 0
580  br i1 %cmp227, label %for.cond1.preheader.us.preheader, label %for.cond1.preheader.preheader
581
582for.cond1.preheader.preheader:
583  br label %for.cond.cleanup.loopexit49
584
585for.cond1.preheader.us.preheader:
586  br label %for.cond1.preheader.us
587
588for.cond1.preheader.us:
589  %i.030.us = phi i32 [ %inc13.us, %for.cond1.for.cond.cleanup3_crit_edge.us ], [ 0, %for.cond1.preheader.us.preheader ]
590  %mul.us = mul nsw i32 %i.030.us, %M
591  br i1 true, label %for.body4.us.us.preheader, label %for.body4.us32.preheader
592
593for.body4.us32.preheader:
594  br label %for.cond1.for.cond.cleanup3_crit_edge.us.loopexit48
595
596for.body4.us.us.preheader:
597  br label %for.body4.us.us
598
599for.cond1.for.cond.cleanup3_crit_edge.us.loopexit:
600  br label %for.cond1.for.cond.cleanup3_crit_edge.us
601
602for.cond1.for.cond.cleanup3_crit_edge.us.loopexit48:
603  br label %for.cond1.for.cond.cleanup3_crit_edge.us
604
605for.cond1.for.cond.cleanup3_crit_edge.us:
606  %inc13.us = add nuw nsw i32 %i.030.us, 1
607  %exitcond51 = icmp ne i32 %inc13.us, %N
608  br i1 %exitcond51, label %for.cond1.preheader.us, label %for.cond.cleanup.loopexit
609
610for.body4.us.us:
611  %j.028.us.us = phi i32 [ %inc10.us.us, %for.cond5.for.cond.cleanup7_crit_edge.us.us ], [ 0, %for.body4.us.us.preheader ]
612  %add.us.us = add nsw i32 %j.028.us.us, %mul.us
613  %idxprom.us.us = sext i32 %add.us.us to i64
614  %arrayidx.us.us = getelementptr inbounds i32, i32* %A, i64 %idxprom.us.us
615  store i32 0, i32* %arrayidx.us.us, align 4
616  br label %for.body8.us.us
617
618for.cond5.for.cond.cleanup7_crit_edge.us.us:
619  %inc10.us.us = add nuw nsw i32 %j.028.us.us, 1
620  %exitcond50 = icmp ne i32 %inc10.us.us, %M
621  br i1 %exitcond50, label %for.body4.us.us, label %for.cond1.for.cond.cleanup3_crit_edge.us.loopexit
622
623for.body8.us.us:
624  %k.026.us.us = phi i32 [ 0, %for.body4.us.us ], [ %inc.us.us, %for.body8.us.us ]
625  tail call void bitcast (void (...)* @g to void ()*)() #2
626  %inc.us.us = add nuw nsw i32 %k.026.us.us, 1
627  %exitcond = icmp ne i32 %inc.us.us, %N
628  br i1 %exitcond, label %for.body8.us.us, label %for.cond5.for.cond.cleanup7_crit_edge.us.us
629
630for.cond.cleanup.loopexit:
631  br label %for.cond.cleanup
632
633for.cond.cleanup.loopexit49:
634  br label %for.cond.cleanup
635
636for.cond.cleanup:
637  ret void
638}
639
640declare dso_local void @f(i32*)
641declare dso_local void @g(...)
642