1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -indvars -dce -S -indvars-predicate-loops=0 | FileCheck %s
3
4; Provide legal integer types.
5target datalayout = "n8:16:32:64"
6
7@A = external global i32
8
9;; Convert a pre-increment check on the latch into a post increment check
10define i32 @pre_to_post_add() {
11; CHECK-LABEL: @pre_to_post_add(
12; CHECK-NEXT:  entry:
13; CHECK-NEXT:    br label [[LOOP:%.*]]
14; CHECK:       loop:
15; CHECK-NEXT:    [[I:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[I_NEXT:%.*]], [[LOOP]] ]
16; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i32 [[I]], 1
17; CHECK-NEXT:    store i32 [[I]], i32* @A, align 4
18; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[I_NEXT]], 1001
19; CHECK-NEXT:    br i1 [[EXITCOND]], label [[LOOP]], label [[LOOPEXIT:%.*]]
20; CHECK:       loopexit:
21; CHECK-NEXT:    ret i32 1000
22;
23entry:
24  br label %loop
25
26loop:
27  %i = phi i32 [ 0, %entry ], [ %i.next, %loop ]
28  %i.next = add i32 %i, 1
29  store i32 %i, i32* @A
30  %c = icmp slt i32 %i, 1000
31  br i1 %c, label %loop, label %loopexit
32
33loopexit:
34  ret i32 %i
35}
36
37; TODO: we should be able to convert the subtract into a post-decrement check
38define i32 @pre_to_post_sub() {
39; CHECK-LABEL: @pre_to_post_sub(
40; CHECK-NEXT:  entry:
41; CHECK-NEXT:    br label [[LOOP:%.*]]
42; CHECK:       loop:
43; CHECK-NEXT:    [[I:%.*]] = phi i32 [ 1000, [[ENTRY:%.*]] ], [ [[I_NEXT:%.*]], [[LOOP]] ]
44; CHECK-NEXT:    [[I_NEXT]] = sub nsw i32 [[I]], 1
45; CHECK-NEXT:    store i32 [[I]], i32* @A, align 4
46; CHECK-NEXT:    [[C:%.*]] = icmp ugt i32 [[I]], 0
47; CHECK-NEXT:    br i1 [[C]], label [[LOOP]], label [[LOOPEXIT:%.*]]
48; CHECK:       loopexit:
49; CHECK-NEXT:    ret i32 0
50;
51entry:
52  br label %loop
53
54loop:
55  %i = phi i32 [ 1000, %entry ], [ %i.next, %loop ]
56  %i.next = sub i32 %i, 1
57  store i32 %i, i32* @A
58  %c = icmp sgt i32 %i, 0
59  br i1 %c, label %loop, label %loopexit
60
61loopexit:
62  ret i32 %i
63}
64
65
66
67; LFTR should eliminate the need for the computation of i*i completely.  It
68; is only used to compute the exit value.
69define i32 @quadratic_slt() {
70; CHECK-LABEL: @quadratic_slt(
71; CHECK-NEXT:  entry:
72; CHECK-NEXT:    br label [[LOOP:%.*]]
73; CHECK:       loop:
74; CHECK-NEXT:    [[I:%.*]] = phi i32 [ 7, [[ENTRY:%.*]] ], [ [[I_NEXT:%.*]], [[LOOP]] ]
75; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i32 [[I]], 1
76; CHECK-NEXT:    store i32 [[I]], i32* @A, align 4
77; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[I_NEXT]], 33
78; CHECK-NEXT:    br i1 [[EXITCOND]], label [[LOOP]], label [[LOOPEXIT:%.*]]
79; CHECK:       loopexit:
80; CHECK-NEXT:    ret i32 32
81;
82entry:
83  br label %loop
84
85loop:
86  %i = phi i32 [ 7, %entry ], [ %i.next, %loop ]
87  %i.next = add i32 %i, 1
88  store i32 %i, i32* @A
89  %i2 = mul i32 %i, %i
90  %c = icmp slt i32 %i2, 1000
91  br i1 %c, label %loop, label %loopexit
92
93loopexit:
94  ret i32 %i
95}
96
97
98; Same as previous but with sle test
99define i32 @quadratic_sle() {
100; CHECK-LABEL: @quadratic_sle(
101; CHECK-NEXT:  entry:
102; CHECK-NEXT:    br label [[LOOP:%.*]]
103; CHECK:       loop:
104; CHECK-NEXT:    [[I:%.*]] = phi i32 [ 7, [[ENTRY:%.*]] ], [ [[I_NEXT:%.*]], [[LOOP]] ]
105; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i32 [[I]], 1
106; CHECK-NEXT:    store i32 [[I]], i32* @A, align 4
107; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[I_NEXT]], 33
108; CHECK-NEXT:    br i1 [[EXITCOND]], label [[LOOP]], label [[LOOPEXIT:%.*]]
109; CHECK:       loopexit:
110; CHECK-NEXT:    ret i32 32
111;
112entry:
113  br label %loop
114
115loop:
116  %i = phi i32 [ 7, %entry ], [ %i.next, %loop ]
117  %i.next = add i32 %i, 1
118  store i32 %i, i32* @A
119  %i2 = mul i32 %i, %i
120  %c = icmp sle i32 %i2, 1000
121  br i1 %c, label %loop, label %loopexit
122
123loopexit:
124  ret i32 %i
125}
126
127; Same as previous but with ule test
128define i32 @quadratic_ule() {
129; CHECK-LABEL: @quadratic_ule(
130; CHECK-NEXT:  entry:
131; CHECK-NEXT:    br label [[LOOP:%.*]]
132; CHECK:       loop:
133; CHECK-NEXT:    [[I:%.*]] = phi i32 [ 7, [[ENTRY:%.*]] ], [ [[I_NEXT:%.*]], [[LOOP]] ]
134; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i32 [[I]], 1
135; CHECK-NEXT:    store i32 [[I]], i32* @A, align 4
136; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[I_NEXT]], 33
137; CHECK-NEXT:    br i1 [[EXITCOND]], label [[LOOP]], label [[LOOPEXIT:%.*]]
138; CHECK:       loopexit:
139; CHECK-NEXT:    ret i32 32
140;
141entry:
142  br label %loop
143
144loop:
145  %i = phi i32 [ 7, %entry ], [ %i.next, %loop ]
146  %i.next = add i32 %i, 1
147  store i32 %i, i32* @A
148  %i2 = mul i32 %i, %i
149  %c = icmp ule i32 %i2, 1000
150  br i1 %c, label %loop, label %loopexit
151
152loopexit:
153  ret i32 %i
154}
155
156define i32 @quadratic_sgt_loopdec() {
157; CHECK-LABEL: @quadratic_sgt_loopdec(
158; CHECK-NEXT:  entry:
159; CHECK-NEXT:    br label [[LOOP:%.*]]
160; CHECK:       loop:
161; CHECK-NEXT:    [[I:%.*]] = phi i32 [ 10, [[ENTRY:%.*]] ], [ [[I_NEXT:%.*]], [[LOOP]] ]
162; CHECK-NEXT:    [[I_NEXT]] = call i32 @llvm.loop.decrement.reg.i32(i32 [[I]], i32 1)
163; CHECK-NEXT:    store i32 [[I]], i32* @A, align 4
164; CHECK-NEXT:    [[I2:%.*]] = mul i32 [[I]], [[I]]
165; CHECK-NEXT:    [[C:%.*]] = icmp sgt i32 [[I2]], 0
166; CHECK-NEXT:    br i1 [[C]], label [[LOOP]], label [[LOOPEXIT:%.*]]
167; CHECK:       loopexit:
168; CHECK-NEXT:    ret i32 0
169;
170
171entry:
172  br label %loop
173
174loop:
175  %i = phi i32 [ 10, %entry ], [ %i.next, %loop ]
176  %i.next = call i32 @llvm.loop.decrement.reg.i32(i32 %i, i32 1)
177  store i32 %i, i32* @A
178  %i2 = mul i32 %i, %i
179  %c = icmp sgt i32 %i2, 0
180  br i1 %c, label %loop, label %loopexit
181
182loopexit:
183  ret i32 %i
184}
185
186@data = common global [240 x i8] zeroinitializer, align 16
187
188define void @test_zext(i8* %a) #0 {
189; CHECK-LABEL: @test_zext(
190; CHECK-NEXT:  entry:
191; CHECK-NEXT:    br label [[LOOP:%.*]]
192; CHECK:       loop:
193; CHECK-NEXT:    [[P_0:%.*]] = phi i8* [ getelementptr inbounds ([240 x i8], [240 x i8]* @data, i64 0, i64 0), [[ENTRY:%.*]] ], [ [[TMP3:%.*]], [[LOOP]] ]
194; CHECK-NEXT:    [[DOT0:%.*]] = phi i8* [ [[A:%.*]], [[ENTRY]] ], [ [[TMP:%.*]], [[LOOP]] ]
195; CHECK-NEXT:    [[TMP]] = getelementptr inbounds i8, i8* [[DOT0]], i64 1
196; CHECK-NEXT:    [[TMP2:%.*]] = load i8, i8* [[DOT0]], align 1
197; CHECK-NEXT:    [[TMP3]] = getelementptr inbounds i8, i8* [[P_0]], i64 1
198; CHECK-NEXT:    store i8 [[TMP2]], i8* [[P_0]], align 1
199; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i8* [[P_0]], getelementptr inbounds ([240 x i8], [240 x i8]* @data, i64 0, i64 239)
200; CHECK-NEXT:    br i1 [[EXITCOND]], label [[LOOP]], label [[EXIT:%.*]]
201; CHECK:       exit:
202; CHECK-NEXT:    ret void
203;
204entry:
205  br label %loop
206
207loop:
208  %i.0 = phi i8 [ 0, %entry ], [ %tmp4, %loop ]
209  %p.0 = phi i8* [ getelementptr inbounds ([240 x i8], [240 x i8]* @data, i64 0, i64 0), %entry ], [ %tmp3, %loop ]
210  %.0 = phi i8* [ %a, %entry ], [ %tmp, %loop ]
211  %tmp = getelementptr inbounds i8, i8* %.0, i64 1
212  %tmp2 = load i8, i8* %.0, align 1
213  %tmp3 = getelementptr inbounds i8, i8* %p.0, i64 1
214  store i8 %tmp2, i8* %p.0, align 1
215  %tmp4 = add i8 %i.0, 1
216  %tmp5 = icmp ult i8 %tmp4, -16
217  br i1 %tmp5, label %loop, label %exit
218
219exit:
220  ret void
221}
222
223; It is okay to do LFTR on this loop even though the trip count is a
224; division because in this case the division can be optimized to a
225; shift.
226define void @test_udiv_as_shift(i8* %a, i8 %n) nounwind uwtable ssp {
227; CHECK-LABEL: @test_udiv_as_shift(
228; CHECK-NEXT:  entry:
229; CHECK-NEXT:    [[E:%.*]] = icmp sgt i8 [[N:%.*]], 3
230; CHECK-NEXT:    br i1 [[E]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
231; CHECK:       loop.preheader:
232; CHECK-NEXT:    [[TMP0:%.*]] = add i8 [[N]], 3
233; CHECK-NEXT:    [[TMP1:%.*]] = lshr i8 [[TMP0]], 2
234; CHECK-NEXT:    [[TMP2:%.*]] = add nuw nsw i8 [[TMP1]], 1
235; CHECK-NEXT:    br label [[LOOP:%.*]]
236; CHECK:       loop:
237; CHECK-NEXT:    [[I1:%.*]] = phi i8 [ [[I1_INC:%.*]], [[LOOP]] ], [ 0, [[LOOP_PREHEADER]] ]
238; CHECK-NEXT:    [[I1_INC]] = add nuw nsw i8 [[I1]], 1
239; CHECK-NEXT:    store volatile i8 0, i8* [[A:%.*]], align 1
240; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i8 [[I1_INC]], [[TMP2]]
241; CHECK-NEXT:    br i1 [[EXITCOND]], label [[LOOP]], label [[EXIT_LOOPEXIT:%.*]]
242; CHECK:       exit.loopexit:
243; CHECK-NEXT:    br label [[EXIT]]
244; CHECK:       exit:
245; CHECK-NEXT:    ret void
246;
247entry:
248  %e = icmp sgt i8 %n, 3
249  br i1 %e, label %loop, label %exit
250
251loop:
252  %i = phi i8 [ 0, %entry ], [ %i.inc, %loop ]
253  %i1 = phi i8 [ 0, %entry ], [ %i1.inc, %loop ]
254  %i.inc = add nsw i8 %i, 4
255  %i1.inc = add i8 %i1, 1
256  store volatile i8 0, i8* %a
257  %c = icmp slt i8 %i, %n
258  br i1 %c, label %loop, label %exit
259
260exit:
261  ret void
262}
263
264; Don't RAUW the loop's original comparison instruction if it has other uses
265; which aren't dominated by the new comparison instruction (which we insert
266; at the branch user).
267define void @use_before_branch() {
268; CHECK-LABEL: @use_before_branch(
269; CHECK-NEXT:  entry:
270; CHECK-NEXT:    br label [[LOOPENTRY_0:%.*]]
271; CHECK:       loopentry.0:
272; CHECK-NEXT:    [[MB_Y_0:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[TMP_152:%.*]], [[LOOPENTRY_1:%.*]] ]
273; CHECK-NEXT:    [[TMP_14:%.*]] = icmp ule i32 [[MB_Y_0]], 3
274; CHECK-NEXT:    br i1 [[TMP_14]], label [[LOOPENTRY_1]], label [[LOOPEXIT_0:%.*]]
275; CHECK:       loopentry.1:
276; CHECK-NEXT:    [[TMP_152]] = add nuw nsw i32 [[MB_Y_0]], 2
277; CHECK-NEXT:    br label [[LOOPENTRY_0]]
278; CHECK:       loopexit.0:
279; CHECK-NEXT:    unreachable
280;
281entry:
282  br label %loopentry.0
283
284loopentry.0:
285  %mb_y.0 = phi i32 [ 0, %entry ], [ %tmp.152, %loopentry.1 ]
286  %tmp.14 = icmp sle i32 %mb_y.0, 3
287  %tmp.15 = zext i1 %tmp.14 to i32
288  br i1 %tmp.14, label %loopentry.1, label %loopexit.0
289
290loopentry.1:
291  %tmp.152 = add i32 %mb_y.0, 2
292  br label %loopentry.0
293
294loopexit.0:		; preds = %loopentry.0
295  unreachable
296}
297
298@.str3 = private constant [6 x i8] c"%lld\0A\00", align 1
299declare i32 @printf(i8* noalias nocapture, ...) nounwind
300
301; PR13371: indvars pass incorrectly substitutes 'undef' values
302;
303; LFTR should not user %undef as the loop counter.
304define i64 @no_undef_counter() nounwind {
305; CHECK-LABEL: @no_undef_counter(
306; CHECK-NEXT:  func_start:
307; CHECK-NEXT:    br label [[BLOCK9:%.*]]
308; CHECK:       block9:
309; CHECK-NEXT:    [[UNDEF:%.*]] = phi i64 [ [[NEXT_UNDEF:%.*]], [[BLOCK9]] ], [ undef, [[FUNC_START:%.*]] ]
310; CHECK-NEXT:    [[ITER:%.*]] = phi i64 [ [[NEXT_ITER:%.*]], [[BLOCK9]] ], [ 1, [[FUNC_START]] ]
311; CHECK-NEXT:    [[NEXT_ITER]] = add nuw nsw i64 [[ITER]], 1
312; CHECK-NEXT:    [[TMP0:%.*]] = tail call i32 (i8*, ...) @printf(i8* noalias nocapture getelementptr inbounds ([6 x i8], [6 x i8]* @.str3, i64 0, i64 0), i64 [[NEXT_ITER]], i64 [[UNDEF]])
313; CHECK-NEXT:    [[NEXT_UNDEF]] = add nsw i64 [[UNDEF]], 1
314; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i64 [[NEXT_ITER]], 100
315; CHECK-NEXT:    br i1 [[EXITCOND]], label [[BLOCK9]], label [[EXIT:%.*]]
316; CHECK:       exit:
317; CHECK-NEXT:    ret i64 0
318;
319func_start:
320  br label %block9
321block9:                                           ; preds = %block9,%func_start
322  %undef = phi i64 [ %next_undef, %block9 ], [ undef, %func_start ]
323  %iter = phi i64 [ %next_iter, %block9 ], [ 1, %func_start ]
324  %next_iter = add nsw i64 %iter, 1
325  %0 = tail call i32 (i8*, ...) @printf(i8* noalias nocapture getelementptr inbounds ([6 x i8], [6 x i8]* @.str3, i64 0, i64 0), i64 %next_iter, i64 %undef)
326  %next_undef = add nsw i64 %undef, 1
327  %_tmp_3 = icmp slt i64 %next_iter, 100
328  br i1 %_tmp_3, label %block9, label %exit
329exit:                                             ; preds = %block9
330  ret i64 0
331}
332
333define void @extend_const() #0 {
334; CHECK-LABEL: @extend_const(
335; CHECK-NEXT:  entry:
336; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
337; CHECK:       for.body:
338; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i32 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
339; CHECK-NEXT:    call void @bar(i32 [[INDVARS_IV]]) [[ATTR2:#.*]]
340; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i32 [[INDVARS_IV]], 1
341; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[INDVARS_IV_NEXT]], 512
342; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END:%.*]]
343; CHECK:       for.end:
344; CHECK-NEXT:    ret void
345;
346entry:
347  br label %for.body
348
349for.body:                                         ; preds = %entry, %for.body
350  %i.01 = phi i16 [ 0, %entry ], [ %inc, %for.body ]
351  %conv2 = sext i16 %i.01 to i32
352  call void @bar(i32 %conv2) #1
353  %inc = add i16 %i.01, 1
354  %cmp = icmp slt i16 %inc, 512
355  br i1 %cmp, label %for.body, label %for.end
356
357for.end:                                          ; preds = %for.body
358  ret void
359}
360
361; Check that post-incrementing the backedge taken count does not overflow.
362define i32 @extend_const_postinc() #0 {
363; CHECK-LABEL: @extend_const_postinc(
364; CHECK-NEXT:  entry:
365; CHECK-NEXT:    br label [[DO_BODY:%.*]]
366; CHECK:       do.body:
367; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i32 [ [[INDVARS_IV_NEXT:%.*]], [[DO_BODY]] ], [ 0, [[ENTRY:%.*]] ]
368; CHECK-NEXT:    call void @bar(i32 [[INDVARS_IV]]) [[ATTR2]]
369; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[INDVARS_IV]], 255
370; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i32 [[INDVARS_IV]], 1
371; CHECK-NEXT:    br i1 [[CMP]], label [[DO_END:%.*]], label [[DO_BODY]]
372; CHECK:       do.end:
373; CHECK-NEXT:    ret i32 0
374;
375entry:
376  br label %do.body
377
378do.body:                                          ; preds = %do.body, %entry
379  %first.0 = phi i8 [ 0, %entry ], [ %inc, %do.body ]
380  %conv = zext i8 %first.0 to i32
381  call void  @bar(i32 %conv) #1
382  %inc = add i8 %first.0, 1
383  %cmp = icmp eq i8 %first.0, -1
384  br i1 %cmp, label %do.end, label %do.body
385
386do.end:                                           ; preds = %do.body
387  ret i32 0
388}
389
390declare void @bar(i32)
391
392attributes #0 = { nounwind uwtable }
393attributes #1 = { nounwind }
394
395; With the given initial value for IV, it is not legal to widen
396; trip count to IV size
397define void @wide_trip_count_test1(float* %autoc,
398; CHECK-LABEL: @wide_trip_count_test1(
399; CHECK-NEXT:  entry:
400; CHECK-NEXT:    [[SUB:%.*]] = sub i32 [[DATA_LEN:%.*]], [[SAMPLE:%.*]]
401; CHECK-NEXT:    [[CMP4:%.*]] = icmp eq i32 [[DATA_LEN]], [[SAMPLE]]
402; CHECK-NEXT:    br i1 [[CMP4]], label [[FOR_END:%.*]], label [[FOR_BODY_PREHEADER:%.*]]
403; CHECK:       for.body.preheader:
404; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
405; CHECK:       for.body:
406; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 68719476736, [[FOR_BODY_PREHEADER]] ]
407; CHECK-NEXT:    [[TEMP:%.*]] = trunc i64 [[INDVARS_IV]] to i32
408; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TEMP]], [[SAMPLE]]
409; CHECK-NEXT:    [[IDXPROM:%.*]] = zext i32 [[ADD]] to i64
410; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[DATA:%.*]], i64 [[IDXPROM]]
411; CHECK-NEXT:    [[TEMP1:%.*]] = load float, float* [[ARRAYIDX]], align 4
412; CHECK-NEXT:    [[MUL:%.*]] = fmul float [[TEMP1]], [[D:%.*]]
413; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[AUTOC:%.*]], i64 [[INDVARS_IV]]
414; CHECK-NEXT:    [[TEMP2:%.*]] = load float, float* [[ARRAYIDX2]], align 4
415; CHECK-NEXT:    [[ADD3:%.*]] = fadd float [[TEMP2]], [[MUL]]
416; CHECK-NEXT:    store float [[ADD3]], float* [[ARRAYIDX2]], align 4
417; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
418; CHECK-NEXT:    [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
419; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[LFTR_WIDEIV]], [[SUB]]
420; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT:%.*]]
421; CHECK:       for.end.loopexit:
422; CHECK-NEXT:    br label [[FOR_END]]
423; CHECK:       for.end:
424; CHECK-NEXT:    ret void
425;
426  float* %data,
427  float %d, i32 %data_len, i32 %sample) nounwind {
428entry:
429  %sub = sub i32 %data_len, %sample
430  %cmp4 = icmp eq i32 %data_len, %sample
431  br i1 %cmp4, label %for.end, label %for.body
432
433for.body:                                         ; preds = %entry, %for.body
434  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 68719476736, %entry ]
435  %temp = trunc i64 %indvars.iv to i32
436  %add = add i32 %temp, %sample
437  %idxprom = zext i32 %add to i64
438  %arrayidx = getelementptr inbounds float, float* %data, i64 %idxprom
439  %temp1 = load float, float* %arrayidx, align 4
440  %mul = fmul float %temp1, %d
441  %arrayidx2 = getelementptr inbounds float, float* %autoc, i64 %indvars.iv
442  %temp2 = load float, float* %arrayidx2, align 4
443  %add3 = fadd float %temp2, %mul
444  store float %add3, float* %arrayidx2, align 4
445  %indvars.iv.next = add i64 %indvars.iv, 1
446  %temp3 = trunc i64 %indvars.iv.next to i32
447  %cmp = icmp ult i32 %temp3, %sub
448  br i1 %cmp, label %for.body, label %for.end
449
450for.end:                                          ; preds = %for.body, %entry
451  ret void
452}
453
454; Trip count should be widened and LFTR should canonicalize the condition
455define float @wide_trip_count_test2(float* %a,
456; CHECK-LABEL: @wide_trip_count_test2(
457; CHECK-NEXT:  entry:
458; CHECK-NEXT:    [[CMP5:%.*]] = icmp ugt i32 [[M:%.*]], 500
459; CHECK-NEXT:    br i1 [[CMP5]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]
460; CHECK:       for.body.preheader:
461; CHECK-NEXT:    [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[M]] to i64
462; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
463; CHECK:       for.body:
464; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 500, [[FOR_BODY_PREHEADER]] ]
465; CHECK-NEXT:    [[SUM_07:%.*]] = phi float [ [[ADD:%.*]], [[FOR_BODY]] ], [ 0.000000e+00, [[FOR_BODY_PREHEADER]] ]
466; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 [[INDVARS_IV]]
467; CHECK-NEXT:    [[TEMP:%.*]] = load float, float* [[ARRAYIDX]], align 4
468; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[INDVARS_IV]]
469; CHECK-NEXT:    [[TEMP1:%.*]] = load float, float* [[ARRAYIDX2]], align 4
470; CHECK-NEXT:    [[MUL:%.*]] = fmul float [[TEMP]], [[TEMP1]]
471; CHECK-NEXT:    [[ADD]] = fadd float [[SUM_07]], [[MUL]]
472; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
473; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
474; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT:%.*]]
475; CHECK:       for.end.loopexit:
476; CHECK-NEXT:    [[ADD_LCSSA:%.*]] = phi float [ [[ADD]], [[FOR_BODY]] ]
477; CHECK-NEXT:    br label [[FOR_END]]
478; CHECK:       for.end:
479; CHECK-NEXT:    [[SUM_0_LCSSA:%.*]] = phi float [ 0.000000e+00, [[ENTRY:%.*]] ], [ [[ADD_LCSSA]], [[FOR_END_LOOPEXIT]] ]
480; CHECK-NEXT:    ret float [[SUM_0_LCSSA]]
481;
482  float* %b,
483  i32 zeroext %m) local_unnamed_addr #0 {
484entry:
485  %cmp5 = icmp ugt i32 %m, 500
486  br i1 %cmp5, label %for.body.preheader, label %for.end
487
488for.body.preheader:                               ; preds = %entry
489  br label %for.body
490
491for.body:                                         ; preds = %for.body.preheader, %for.body
492  %sum.07 = phi float [ %add, %for.body ], [ 0.000000e+00, %for.body.preheader ]
493  %i.06 = phi i32 [ %inc, %for.body ], [ 500, %for.body.preheader ]
494  %idxprom = zext i32 %i.06 to i64
495  %arrayidx = getelementptr inbounds float, float* %b, i64 %idxprom
496  %temp = load float, float* %arrayidx, align 4
497  %arrayidx2 = getelementptr inbounds float, float* %a, i64 %idxprom
498  %temp1 = load float, float* %arrayidx2, align 4
499  %mul = fmul float %temp, %temp1
500  %add = fadd float %sum.07, %mul
501  %inc = add i32 %i.06, 1
502  %cmp = icmp ult i32 %inc, %m
503  br i1 %cmp, label %for.body, label %for.end.loopexit
504
505for.end.loopexit:                                 ; preds = %for.body
506  br label %for.end
507
508for.end:                                          ; preds = %for.end.loopexit, %entry
509  %sum.0.lcssa = phi float [ 0.000000e+00, %entry ], [ %add, %for.end.loopexit ]
510  ret float %sum.0.lcssa
511}
512
513; Trip count should be widened and LFTR should canonicalize the condition
514define float @wide_trip_count_test3(float* %b,
515; CHECK-LABEL: @wide_trip_count_test3(
516; CHECK-NEXT:  entry:
517; CHECK-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[M:%.*]], -10
518; CHECK-NEXT:    br i1 [[CMP5]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]
519; CHECK:       for.body.preheader:
520; CHECK-NEXT:    [[WIDE_TRIP_COUNT:%.*]] = sext i32 [[M]] to i64
521; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
522; CHECK:       for.body:
523; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ -10, [[FOR_BODY_PREHEADER]] ]
524; CHECK-NEXT:    [[SUM_07:%.*]] = phi float [ [[ADD1:%.*]], [[FOR_BODY]] ], [ 0.000000e+00, [[FOR_BODY_PREHEADER]] ]
525; CHECK-NEXT:    [[TMP0:%.*]] = add nsw i64 [[INDVARS_IV]], 20
526; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 [[TMP0]]
527; CHECK-NEXT:    [[TEMP:%.*]] = load float, float* [[ARRAYIDX]], align 4
528; CHECK-NEXT:    [[TMP1:%.*]] = trunc i64 [[INDVARS_IV]] to i32
529; CHECK-NEXT:    [[CONV:%.*]] = sitofp i32 [[TMP1]] to float
530; CHECK-NEXT:    [[MUL:%.*]] = fmul float [[CONV]], [[TEMP]]
531; CHECK-NEXT:    [[ADD1]] = fadd float [[SUM_07]], [[MUL]]
532; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], 1
533; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
534; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT:%.*]]
535; CHECK:       for.end.loopexit:
536; CHECK-NEXT:    [[ADD1_LCSSA:%.*]] = phi float [ [[ADD1]], [[FOR_BODY]] ]
537; CHECK-NEXT:    br label [[FOR_END]]
538; CHECK:       for.end:
539; CHECK-NEXT:    [[SUM_0_LCSSA:%.*]] = phi float [ 0.000000e+00, [[ENTRY:%.*]] ], [ [[ADD1_LCSSA]], [[FOR_END_LOOPEXIT]] ]
540; CHECK-NEXT:    ret float [[SUM_0_LCSSA]]
541;
542  i32 signext %m) local_unnamed_addr #0 {
543entry:
544  %cmp5 = icmp sgt i32 %m, -10
545  br i1 %cmp5, label %for.body.preheader, label %for.end
546
547for.body.preheader:                               ; preds = %entry
548  br label %for.body
549
550for.body:                                         ; preds = %for.body.preheader, %for.body
551  %sum.07 = phi float [ %add1, %for.body ], [ 0.000000e+00, %for.body.preheader ]
552  %i.06 = phi i32 [ %inc, %for.body ], [ -10, %for.body.preheader ]
553  %add = add nsw i32 %i.06, 20
554  %idxprom = sext i32 %add to i64
555  %arrayidx = getelementptr inbounds float, float* %b, i64 %idxprom
556  %temp = load float, float* %arrayidx, align 4
557  %conv = sitofp i32 %i.06 to float
558  %mul = fmul float %conv, %temp
559  %add1 = fadd float %sum.07, %mul
560  %inc = add nsw i32 %i.06, 1
561  %cmp = icmp slt i32 %inc, %m
562  br i1 %cmp, label %for.body, label %for.end.loopexit
563
564for.end.loopexit:                                 ; preds = %for.body
565  br label %for.end
566
567for.end:                                          ; preds = %for.end.loopexit, %entry
568  %sum.0.lcssa = phi float [ 0.000000e+00, %entry ], [ %add1, %for.end.loopexit ]
569  ret float %sum.0.lcssa
570}
571
572; Trip count should be widened and LFTR should canonicalize the condition
573define float @wide_trip_count_test4(float* %b,
574; CHECK-LABEL: @wide_trip_count_test4(
575; CHECK-NEXT:  entry:
576; CHECK-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[M:%.*]], 10
577; CHECK-NEXT:    br i1 [[CMP5]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]
578; CHECK:       for.body.preheader:
579; CHECK-NEXT:    [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[M]] to i64
580; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
581; CHECK:       for.body:
582; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 10, [[FOR_BODY_PREHEADER]] ]
583; CHECK-NEXT:    [[SUM_07:%.*]] = phi float [ [[ADD1:%.*]], [[FOR_BODY]] ], [ 0.000000e+00, [[FOR_BODY_PREHEADER]] ]
584; CHECK-NEXT:    [[TMP0:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 20
585; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 [[TMP0]]
586; CHECK-NEXT:    [[TEMP:%.*]] = load float, float* [[ARRAYIDX]], align 4
587; CHECK-NEXT:    [[TMP1:%.*]] = trunc i64 [[INDVARS_IV]] to i32
588; CHECK-NEXT:    [[CONV:%.*]] = sitofp i32 [[TMP1]] to float
589; CHECK-NEXT:    [[MUL:%.*]] = fmul float [[CONV]], [[TEMP]]
590; CHECK-NEXT:    [[ADD1]] = fadd float [[SUM_07]], [[MUL]]
591; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
592; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
593; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT:%.*]]
594; CHECK:       for.end.loopexit:
595; CHECK-NEXT:    [[ADD1_LCSSA:%.*]] = phi float [ [[ADD1]], [[FOR_BODY]] ]
596; CHECK-NEXT:    br label [[FOR_END]]
597; CHECK:       for.end:
598; CHECK-NEXT:    [[SUM_0_LCSSA:%.*]] = phi float [ 0.000000e+00, [[ENTRY:%.*]] ], [ [[ADD1_LCSSA]], [[FOR_END_LOOPEXIT]] ]
599; CHECK-NEXT:    ret float [[SUM_0_LCSSA]]
600;
601  i32 signext %m) local_unnamed_addr #0 {
602entry:
603  %cmp5 = icmp sgt i32 %m, 10
604  br i1 %cmp5, label %for.body.preheader, label %for.end
605
606for.body.preheader:                               ; preds = %entry
607  br label %for.body
608
609for.body:                                         ; preds = %for.body.preheader, %for.body
610  %sum.07 = phi float [ %add1, %for.body ], [ 0.000000e+00, %for.body.preheader ]
611  %i.06 = phi i32 [ %inc, %for.body ], [ 10, %for.body.preheader ]
612  %add = add nsw i32 %i.06, 20
613  %idxprom = sext i32 %add to i64
614  %arrayidx = getelementptr inbounds float, float* %b, i64 %idxprom
615  %temp = load float, float* %arrayidx, align 4
616  %conv = sitofp i32 %i.06 to float
617  %mul = fmul float %conv, %temp
618  %add1 = fadd float %sum.07, %mul
619  %inc = add nsw i32 %i.06, 1
620  %cmp = icmp slt i32 %inc, %m
621  br i1 %cmp, label %for.body, label %for.end.loopexit
622
623for.end.loopexit:                                 ; preds = %for.body
624  %add1.lcssa = phi float [ %add1, %for.body ]
625  br label %for.end
626
627for.end:                                          ; preds = %for.end.loopexit, %entry
628  %sum.0.lcssa = phi float [ 0.000000e+00, %entry ], [ %add1.lcssa, %for.end.loopexit ]
629  ret float %sum.0.lcssa
630}
631
632define void @ptr_non_cmp_exit_test() {
633; CHECK-LABEL: @ptr_non_cmp_exit_test(
634; CHECK-NEXT:  entry:
635; CHECK-NEXT:    br label [[FOR_BODY29:%.*]]
636; CHECK:       for.body29:
637; CHECK-NEXT:    [[IV:%.*]] = phi i8* [ null, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY29]] ]
638; CHECK-NEXT:    [[TMP0:%.*]] = load volatile i8, i8* [[IV]], align 1
639; CHECK-NEXT:    [[IV_NEXT]] = getelementptr inbounds i8, i8* [[IV]], i64 1
640; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i8* [[IV]], inttoptr (i64 10 to i8*)
641; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY29]], label [[EXIT:%.*]]
642; CHECK:       exit:
643; CHECK-NEXT:    ret void
644;
645entry:
646  br label %for.body29
647
648for.body29:
649  %iv = phi i8* [ null, %entry ], [ %iv.next, %for.body29 ]
650  load volatile i8, i8* %iv, align 1
651  %iv.next = getelementptr inbounds i8, i8* %iv, i64 1
652  %cmp = icmp ne i8* %iv.next, inttoptr (i64 11 to i8*)
653  %and = and i1 %cmp, %cmp
654  br i1 %and, label %for.body29, label %exit
655
656exit:
657  ret void
658}
659
660
661declare i32 @llvm.loop.decrement.reg.i32(i32, i32)
662
663