1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN:opt -passes=loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -S %s | FileCheck %s
3
4; For %for.1, we are fine initially, because the previous value %for.1.next dominates the
5; user of %for.1. But for %for.2, we have to sink the user (%for.1.next) past the previous
6; value %for.2.next. This however breaks the condition we have for %for.1. We cannot fix
7; both first order recurrences and cannot vectorize the loop.
8define i32 @c(i32 %N) {
9; CHECK-LABEL: @c(
10; CHECK-NEXT:  entry:
11; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
12; CHECK:       for.body:
13; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[INC:%.*]], [[FOR_BODY]] ], [ 10, [[ENTRY:%.*]] ]
14; CHECK-NEXT:    [[FOR_1:%.*]] = phi i32 [ [[FOR_1_NEXT:%.*]], [[FOR_BODY]] ], [ 20, [[ENTRY]] ]
15; CHECK-NEXT:    [[FOR_2:%.*]] = phi i32 [ [[FOR_2_NEXT:%.*]], [[FOR_BODY]] ], [ 11, [[ENTRY]] ]
16; CHECK-NEXT:    [[FOR_1_NEXT]] = add nsw i32 [[FOR_2]], 1
17; CHECK-NEXT:    [[FOR_2_NEXT]] = shl i32 [[FOR_1]], 24
18; CHECK-NEXT:    [[INC]] = add nsw i32 [[IV]], 1
19; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INC]], [[N:%.*]]
20; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_COND1_FOR_END_CRIT_EDGE:%.*]], label [[FOR_BODY]]
21; CHECK:       for.cond1.for.end_crit_edge:
22; CHECK-NEXT:    [[ADD_LCSSA:%.*]] = phi i32 [ [[FOR_1_NEXT]], [[FOR_BODY]] ]
23; CHECK-NEXT:    [[SEXT_LCSSA:%.*]] = phi i32 [ [[FOR_2_NEXT]], [[FOR_BODY]] ]
24; CHECK-NEXT:    [[RES:%.*]] = add i32 [[ADD_LCSSA]], [[SEXT_LCSSA]]
25; CHECK-NEXT:    ret i32 [[RES]]
26;
27entry:
28  br label %for.body
29
30for.body:                                         ; preds = %for.body.preheader, %for.body
31  %iv  = phi i32 [ %inc, %for.body ], [ 10, %entry ]
32  %for.1 = phi i32 [ %for.1.next, %for.body ], [ 20, %entry ]
33  %for.2 = phi i32 [ %for.2.next, %for.body ], [ 11, %entry ]
34  %for.1.next = add nsw i32 %for.2, 1
35  %for.2.next = shl i32 %for.1, 24
36  %inc = add nsw i32 %iv, 1
37  %exitcond = icmp eq i32 %inc, %N
38  br i1 %exitcond, label %for.cond1.for.end_crit_edge, label %for.body
39
40for.cond1.for.end_crit_edge:                      ; preds = %for.body
41  %add.lcssa = phi i32 [ %for.1.next, %for.body ]
42  %sext.lcssa = phi i32 [ %for.2.next, %for.body ]
43  %res = add i32 %add.lcssa, %sext.lcssa
44  ret i32 %res
45}
46
47
48; The 'previous' instruction of %for.2 is in a separate block.
49; PR54195.
50define void @multiple_recurrences_with_previous_in_different_block(i32 %a, i8 %b, i64* %ptr) {
51; CHECK-LABEL: @multiple_recurrences_with_previous_in_different_block(
52; CHECK-NEXT:  entry:
53; CHECK-NEXT:    br label [[LOOP_HEADER:%.*]]
54; CHECK:       loop.header:
55; CHECK-NEXT:    [[FOR_1:%.*]] = phi i8 [ 10, [[ENTRY:%.*]] ], [ [[FOR_1_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
56; CHECK-NEXT:    [[FOR_2:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[FOR_2_NEXT:%.*]], [[LOOP_LATCH]] ]
57; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH]] ]
58; CHECK-NEXT:    [[CONV:%.*]] = sext i8 [[FOR_1]] to i64
59; CHECK-NEXT:    [[SUB:%.*]] = sub nsw i64 [[CONV]], [[FOR_2]]
60; CHECK-NEXT:    [[IV_NEXT]] = add nuw i64 [[IV]], 1
61; CHECK-NEXT:    [[FOR_1_NEXT]] = xor i8 [[B:%.*]], 6
62; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV]], 1000
63; CHECK-NEXT:    br i1 [[EXITCOND]], label [[EXIT:%.*]], label [[LOOP_LATCH]]
64; CHECK:       loop.latch:
65; CHECK-NEXT:    [[PTR_GEP:%.*]] = getelementptr inbounds i64, i64* [[PTR:%.*]], i64 [[IV]]
66; CHECK-NEXT:    store i64 [[SUB]], i64* [[PTR_GEP]], align 4
67; CHECK-NEXT:    [[FOR_2_NEXT]] = zext i32 [[A:%.*]] to i64
68; CHECK-NEXT:    br label [[LOOP_HEADER]]
69; CHECK:       exit:
70; CHECK-NEXT:    ret void
71;
72entry:
73  br label %loop.header
74
75loop.header:
76  %for.1 = phi i8 [ 10, %entry ], [ %for.1.next, %loop.latch ]
77  %for.2 = phi i64  [0, %entry ], [ %for.2.next, %loop.latch ]
78  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
79  %conv = sext i8 %for.1 to i64
80  %sub = sub nsw i64 %conv, %for.2
81  %iv.next = add nuw i64 %iv, 1
82  %for.1.next = xor i8 %b, 6
83  %exitcond = icmp eq i64 %iv, 1000
84  br i1 %exitcond, label %exit, label %loop.latch
85
86loop.latch:
87  %ptr.gep = getelementptr inbounds i64, i64* %ptr, i64 %iv
88  store i64 %sub, i64* %ptr.gep
89  %for.2.next = zext i32 %a to i64
90  br label %loop.header
91
92exit:
93  ret void
94}
95
96define void @test_pr54223_sink_after_insertion_order(float* noalias %a, float* noalias %b, float* noalias %dst) {
97; CHECK-LABEL: @test_pr54223_sink_after_insertion_order(
98; CHECK-NEXT:  entry:
99; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
100; CHECK:       vector.ph:
101; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
102; CHECK:       vector.body:
103; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
104; CHECK-NEXT:    [[VECTOR_RECUR:%.*]] = phi <4 x float> [ <float poison, float poison, float poison, float 0.000000e+00>, [[VECTOR_PH]] ], [ [[BROADCAST_SPLAT:%.*]], [[VECTOR_BODY]] ]
105; CHECK-NEXT:    [[VECTOR_RECUR1:%.*]] = phi <4 x float> [ <float poison, float poison, float poison, float 0.000000e+00>, [[VECTOR_PH]] ], [ [[BROADCAST_SPLAT3:%.*]], [[VECTOR_BODY]] ]
106; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
107; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds float, float* [[DST:%.*]], i64 [[TMP0]]
108; CHECK-NEXT:    [[TMP2:%.*]] = load float, float* [[A:%.*]], align 4
109; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x float> poison, float [[TMP2]], i32 0
110; CHECK-NEXT:    [[BROADCAST_SPLAT]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer
111; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[VECTOR_RECUR]], <4 x float> [[BROADCAST_SPLAT]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
112; CHECK-NEXT:    [[TMP4:%.*]] = load float, float* [[B:%.*]], align 4
113; CHECK-NEXT:    [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <4 x float> poison, float [[TMP4]], i32 0
114; CHECK-NEXT:    [[BROADCAST_SPLAT3]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT2]], <4 x float> poison, <4 x i32> zeroinitializer
115; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <4 x float> [[VECTOR_RECUR1]], <4 x float> [[BROADCAST_SPLAT3]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
116; CHECK-NEXT:    [[TMP6:%.*]] = fneg <4 x float> [[TMP5]]
117; CHECK-NEXT:    [[TMP7:%.*]] = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> [[TMP3]], <4 x float> [[TMP6]], <4 x float> zeroinitializer)
118; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds float, float* [[TMP1]], i32 0
119; CHECK-NEXT:    [[TMP9:%.*]] = bitcast float* [[TMP8]] to <4 x float>*
120; CHECK-NEXT:    store <4 x float> [[TMP7]], <4 x float>* [[TMP9]], align 4
121; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
122; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10000
123; CHECK-NEXT:    br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
124; CHECK:       middle.block:
125; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 10000, 10000
126; CHECK-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <4 x float> [[BROADCAST_SPLAT]], i32 3
127; CHECK-NEXT:    [[VECTOR_RECUR_EXTRACT_FOR_PHI:%.*]] = extractelement <4 x float> [[BROADCAST_SPLAT]], i32 2
128; CHECK-NEXT:    [[VECTOR_RECUR_EXTRACT4:%.*]] = extractelement <4 x float> [[BROADCAST_SPLAT3]], i32 3
129; CHECK-NEXT:    [[VECTOR_RECUR_EXTRACT_FOR_PHI5:%.*]] = extractelement <4 x float> [[BROADCAST_SPLAT3]], i32 2
130; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
131; CHECK:       scalar.ph:
132; CHECK-NEXT:    [[SCALAR_RECUR_INIT6:%.*]] = phi float [ 0.000000e+00, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT4]], [[MIDDLE_BLOCK]] ]
133; CHECK-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
134; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 10000, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
135; CHECK-NEXT:    br label [[LOOP:%.*]]
136; CHECK:       loop:
137; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
138; CHECK-NEXT:    [[SCALAR_RECUR:%.*]] = phi float [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[FOR_1_NEXT:%.*]], [[LOOP]] ]
139; CHECK-NEXT:    [[SCALAR_RECUR7:%.*]] = phi float [ [[SCALAR_RECUR_INIT6]], [[SCALAR_PH]] ], [ [[FOR_2_NEXT:%.*]], [[LOOP]] ]
140; CHECK-NEXT:    [[NEG:%.*]] = fneg float [[SCALAR_RECUR7]]
141; CHECK-NEXT:    [[MULADD:%.*]] = call float @llvm.fmuladd.f32(float [[SCALAR_RECUR]], float [[NEG]], float 0.000000e+00)
142; CHECK-NEXT:    [[DST_GEP:%.*]] = getelementptr inbounds float, float* [[DST]], i64 [[IV]]
143; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
144; CHECK-NEXT:    [[FOR_1_NEXT]] = load float, float* [[A]], align 4
145; CHECK-NEXT:    [[FOR_2_NEXT]] = load float, float* [[B]], align 4
146; CHECK-NEXT:    store float [[MULADD]], float* [[DST_GEP]], align 4
147; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 10000
148; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP2:![0-9]+]]
149; CHECK:       exit:
150; CHECK-NEXT:    ret void
151;
152entry:
153  br label %loop
154
155loop:
156  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
157  %for.1 = phi float [ 0.0, %entry ], [ %for.1.next, %loop ]
158  %for.2 = phi float [ 0.0, %entry ], [ %for.2.next, %loop ]
159  %neg = fneg float %for.2
160  %muladd = call float @llvm.fmuladd.f32(float %for.1, float %neg, float 0.000000e+00)
161  %dst.gep = getelementptr inbounds float, float* %dst, i64 %iv
162  %iv.next = add nuw nsw i64 %iv, 1
163  %for.1.next = load float, float* %a, align 4
164  %for.2.next = load float, float* %b, align 4
165  store float %muladd, float* %dst.gep
166  %exitcond.not = icmp eq i64 %iv.next, 10000
167  br i1 %exitcond.not, label %exit, label %loop
168
169exit:
170  ret void
171}
172
173declare float @llvm.fmuladd.f32(float, float, float) #1
174
175define void @test_pr54227(i32* noalias %a, i32* noalias %b) {
176; CHECK-LABEL: @test_pr54227(
177; CHECK-NEXT:  entry:
178; CHECK-NEXT:    br label [[LOOP:%.*]]
179; CHECK:       loop:
180; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
181; CHECK-NEXT:    [[AND17:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[AND1:%.*]], [[LOOP]] ]
182; CHECK-NEXT:    [[F_0:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[MUL:%.*]], [[LOOP]] ]
183; CHECK-NEXT:    [[E_0:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[ADD:%.*]], [[LOOP]] ]
184; CHECK-NEXT:    [[AND:%.*]] = and i32 [[AND17]], 255
185; CHECK-NEXT:    [[OR:%.*]] = or i32 [[AND]], [[E_0]]
186; CHECK-NEXT:    [[AND1]] = and i32 [[E_0]], [[F_0]]
187; CHECK-NEXT:    [[MUL]] = shl nsw i32 [[OR]], 1
188; CHECK-NEXT:    [[ADD]] = or i32 [[AND1]], 1
189; CHECK-NEXT:    [[A_GEP:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[IV]]
190; CHECK-NEXT:    store i32 [[ADD]], i32* [[A_GEP]], align 4
191; CHECK-NEXT:    [[B_GEP:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[IV]]
192; CHECK-NEXT:    store i32 [[MUL]], i32* [[B_GEP]], align 4
193; CHECK-NEXT:    [[IV_NEXT]] = add nuw i64 [[IV]], 1
194; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV]], 1000
195; CHECK-NEXT:    br i1 [[EXITCOND]], label [[EXIT:%.*]], label [[LOOP]]
196; CHECK:       exit:
197; CHECK-NEXT:    ret void
198;
199entry:
200  br label %loop
201
202loop:
203  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
204  %and17 = phi i32 [ 0, %entry ], [ %and1, %loop ]
205  %f.0 = phi i32 [ 0, %entry ], [ %mul, %loop ]
206  %e.0 = phi i32 [ 0, %entry ], [ %add, %loop ]
207  %and = and i32 %and17, 255
208  %or = or i32 %and, %e.0
209  %and1 = and i32 %e.0, %f.0
210  %mul = shl nsw i32 %or, 1
211  %add = or i32 %and1, 1
212  %a.gep = getelementptr inbounds i32, i32* %a, i64 %iv
213  store i32 %add, i32* %a.gep, align 4
214  %b.gep = getelementptr inbounds i32, i32* %a, i64 %iv
215  store i32 %mul, i32* %b.gep, align 4
216  %iv.next = add nuw i64 %iv, 1
217  %exitcond = icmp eq i64 %iv, 1000
218  br i1 %exitcond, label %exit, label %loop
219
220exit:
221  ret void
222}
223
224define void @test_pr54233_for_depend_on_each_other(i32* noalias %a, i32* noalias %b) {
225; CHECK-LABEL: @test_pr54233_for_depend_on_each_other(
226; CHECK-NEXT:  entry:
227; CHECK-NEXT:    br label [[LOOP:%.*]]
228; CHECK:       loop:
229; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
230; CHECK-NEXT:    [[FOR_1:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[FOR_1_NEXT:%.*]], [[LOOP]] ]
231; CHECK-NEXT:    [[FOR_2:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[FOR_2_NEXT:%.*]], [[LOOP]] ]
232; CHECK-NEXT:    [[OR:%.*]] = or i32 [[FOR_2]], 10
233; CHECK-NEXT:    [[SHL:%.*]] = shl i32 [[FOR_2]], [[FOR_1]]
234; CHECK-NEXT:    [[XOR:%.*]] = xor i32 [[SHL]], 255
235; CHECK-NEXT:    [[AND:%.*]] = and i32 [[XOR]], [[OR]]
236; CHECK-NEXT:    [[FOR_1_NEXT]] = xor i32 12, [[FOR_2]]
237; CHECK-NEXT:    [[FOR_2_NEXT]] = load i32, i32* [[B:%.*]], align 4
238; CHECK-NEXT:    [[A_GEP:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[IV]]
239; CHECK-NEXT:    store i32 [[AND]], i32* [[A_GEP]], align 4
240; CHECK-NEXT:    [[IV_NEXT]] = add nuw i64 [[IV]], 1
241; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV]], 1000
242; CHECK-NEXT:    br i1 [[EXITCOND]], label [[EXIT:%.*]], label [[LOOP]]
243; CHECK:       exit:
244; CHECK-NEXT:    ret void
245;
246entry:
247  br label %loop
248
249loop:
250  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
251  %for.1 = phi i32 [ 0, %entry ], [ %for.1.next, %loop ]
252  %for.2 = phi i32 [ 0, %entry ], [ %for.2.next, %loop ]
253  %or = or i32 %for.2, 10
254  %shl = shl i32 %for.2, %for.1
255  %xor = xor i32 %shl, 255
256  %and = and i32 %xor, %or
257  %for.1.next = xor i32 12, %for.2
258  %for.2.next = load i32, i32* %b
259  %a.gep = getelementptr inbounds i32, i32* %a, i64 %iv
260  store i32 %and, i32* %a.gep, align 4
261  %iv.next = add nuw i64 %iv, 1
262  %exitcond = icmp eq i64 %iv, 1000
263  br i1 %exitcond, label %exit, label %loop
264
265exit:
266  ret void
267}
268
269define void @pr54218(i8* %dst, i32* noalias %d) {
270; CHECK-LABEL: @pr54218(
271; CHECK-NEXT:  entry:
272; CHECK-NEXT:    br label [[LOOP_HEADER:%.*]]
273; CHECK:       loop.header:
274; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
275; CHECK-NEXT:    [[FOR_1:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[FOR_1_NEXT:%.*]], [[LOOP_LATCH]] ]
276; CHECK-NEXT:    [[P:%.*]] = phi i8 [ 0, [[ENTRY]] ], [ [[P_NEXT:%.*]], [[LOOP_LATCH]] ]
277; CHECK-NEXT:    [[C_1:%.*]] = icmp eq i32 [[FOR_1]], 0
278; CHECK-NEXT:    [[SEL:%.*]] = select i1 [[C_1]], i8 [[P]], i8 0
279; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV]], 1000
280; CHECK-NEXT:    br i1 [[EXITCOND]], label [[EXIT:%.*]], label [[LOOP_LATCH]]
281; CHECK:       loop.latch:
282; CHECK-NEXT:    [[FOR_1_NEXT]] = load i32, i32* [[D:%.*]], align 4
283; CHECK-NEXT:    [[P_NEXT]] = add i8 [[SEL]], -1
284; CHECK-NEXT:    [[DST_GEP:%.*]] = getelementptr inbounds i8, i8* [[DST:%.*]], i64 [[IV]]
285; CHECK-NEXT:    store i8 [[SEL]], i8* [[DST_GEP]], align 1
286; CHECK-NEXT:    [[IV_NEXT]] = add nuw i64 [[IV]], 1
287; CHECK-NEXT:    br label [[LOOP_HEADER]]
288; CHECK:       exit:
289; CHECK-NEXT:    ret void
290;
291entry:
292  br label %loop.header
293
294loop.header:
295  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
296  %for.1 = phi i32 [ 0, %entry ], [ %for.1.next, %loop.latch ]
297  %p = phi i8 [ 0, %entry ], [ %p.next, %loop.latch ]
298  %c.1 = icmp eq i32 %for.1, 0
299  %sel = select i1 %c.1, i8 %p, i8 0
300  %exitcond = icmp eq i64 %iv, 1000
301  br i1 %exitcond, label %exit, label %loop.latch
302
303loop.latch:
304  %for.1.next = load i32, i32* %d, align 4
305  %p.next = add i8 %sel, -1
306  %dst.gep = getelementptr inbounds i8, i8* %dst, i64 %iv
307  store i8 %sel, i8* %dst.gep
308  %iv.next = add nuw i64 %iv, 1
309  br label %loop.header
310
311exit:
312  ret void
313}
314
315define void @pr54254_fors_depend_on_each_other(i32* %dst) {
316; CHECK-LABEL: @pr54254_fors_depend_on_each_other(
317; CHECK-NEXT:  entry:
318; CHECK-NEXT:    br label [[LOOP:%.*]]
319; CHECK:       loop:
320; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
321; CHECK-NEXT:    [[D_0:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[XOR:%.*]], [[LOOP]] ]
322; CHECK-NEXT:    [[C_0:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[REM_NEXT:%.*]], [[LOOP]] ]
323; CHECK-NEXT:    [[F_0:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[XOR1:%.*]], [[LOOP]] ]
324; CHECK-NEXT:    [[NEG:%.*]] = xor i32 [[D_0]], -1
325; CHECK-NEXT:    [[REM:%.*]] = srem i32 [[F_0]], [[NEG]]
326; CHECK-NEXT:    [[XOR]] = xor i32 [[C_0]], 1
327; CHECK-NEXT:    [[XOR1]] = xor i32 [[REM]], 1
328; CHECK-NEXT:    [[DST_GEP:%.*]] = getelementptr inbounds i32, i32* [[DST:%.*]], i64 [[IV]]
329; CHECK-NEXT:    [[REM_NEXT]] = add i32 [[REM]], 10
330; CHECK-NEXT:    store i32 [[REM]], i32* [[DST_GEP]], align 4
331; CHECK-NEXT:    [[IV_NEXT]] = add nuw i64 [[IV]], 1
332; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV]], 1000
333; CHECK-NEXT:    br i1 [[EXITCOND]], label [[EXIT:%.*]], label [[LOOP]]
334; CHECK:       exit:
335; CHECK-NEXT:    ret void
336;
337entry:
338  br label %loop
339
340loop:
341  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
342  %d.0 = phi i32 [ 0, %entry ], [ %xor, %loop ]
343  %c.0 = phi i32 [ 0, %entry ], [ %rem.next, %loop ]
344  %f.0 = phi i32 [ 0, %entry ], [ %xor1, %loop ]
345  %neg = xor i32 %d.0, -1
346  %rem = srem i32 %f.0, %neg
347  %xor = xor i32 %c.0, 1
348  %xor1 = xor i32 %rem, 1
349  %dst.gep = getelementptr inbounds i32, i32* %dst, i64 %iv
350  %rem.next = add i32 %rem, 10
351  store i32 %rem, i32* %dst.gep
352  %iv.next = add nuw i64 %iv, 1
353  %exitcond = icmp eq i64 %iv, 1000
354  br i1 %exitcond, label %exit, label %loop
355
356exit:
357  ret void
358}
359