1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt -basicaa -tbaa -loop-unroll-and-jam -allow-unroll-and-jam -unroll-and-jam-count=4 -unroll-remainder < %s -S | FileCheck %s
4 target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
7 ; Tests for(i) { sum = 0; for(j) sum += B[j]; A[i] = sum; }
9 ; CHECK-NEXT: [[CMP:%.*]] = icmp ne i32 [[J:%.*]], 0
10 ; CHECK-NEXT: [[CMPJ:%.*]] = icmp ne i32 [[I:%.*]], 0
11 ; CHECK-NEXT: [[OR_COND:%.*]] = and i1 [[CMP]], [[CMPJ]]
12 ; CHECK-NEXT: br i1 [[OR_COND]], label [[FOR_OUTER_PREHEADER:%.*]], label [[FOR_END:%.*]]
13 ; CHECK: for.outer.preheader:
14 ; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[I]], -1
15 ; CHECK-NEXT: [[XTRAITER:%.*]] = and i32 [[I]], 3
16 ; CHECK-NEXT: [[TMP1:%.*]] = icmp ult i32 [[TMP0]], 3
17 ; CHECK-NEXT: br i1 [[TMP1]], label [[FOR_END_LOOPEXIT_UNR_LCSSA:%.*]], label [[FOR_OUTER_PREHEADER_NEW:%.*]]
18 ; CHECK: for.outer.preheader.new:
19 ; CHECK-NEXT: [[UNROLL_ITER:%.*]] = sub i32 [[I]], [[XTRAITER]]
20 ; CHECK-NEXT: br label [[FOR_OUTER:%.*]]
22 ; CHECK-NEXT: [[I:%.*]] = phi i32 [ [[ADD8_3:%.*]], [[FOR_LATCH:%.*]] ], [ 0, [[FOR_OUTER_PREHEADER_NEW]] ]
23 ; CHECK-NEXT: [[NITER:%.*]] = phi i32 [ [[UNROLL_ITER]], [[FOR_OUTER_PREHEADER_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[FOR_LATCH]] ]
24 ; CHECK-NEXT: [[ADD8:%.*]] = add nuw nsw i32 [[I]], 1
25 ; CHECK-NEXT: [[NITER_NSUB:%.*]] = sub i32 [[NITER]], 1
26 ; CHECK-NEXT: [[ADD8_1:%.*]] = add nuw nsw i32 [[ADD8]], 1
27 ; CHECK-NEXT: [[NITER_NSUB_1:%.*]] = sub i32 [[NITER_NSUB]], 1
28 ; CHECK-NEXT: [[ADD8_2:%.*]] = add nuw nsw i32 [[ADD8_1]], 1
29 ; CHECK-NEXT: [[NITER_NSUB_2:%.*]] = sub i32 [[NITER_NSUB_1]], 1
30 ; CHECK-NEXT: [[ADD8_3]] = add nuw i32 [[ADD8_2]], 1
31 ; CHECK-NEXT: [[NITER_NSUB_3]] = sub i32 [[NITER_NSUB_2]], 1
32 ; CHECK-NEXT: br label [[FOR_INNER:%.*]]
34 ; CHECK-NEXT: [[J_0:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[INC:%.*]], [[FOR_INNER]] ]
35 ; CHECK-NEXT: [[SUM:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[ADD:%.*]], [[FOR_INNER]] ]
36 ; CHECK-NEXT: [[J_1:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[INC_1:%.*]], [[FOR_INNER]] ]
37 ; CHECK-NEXT: [[SUM_1:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[ADD_1:%.*]], [[FOR_INNER]] ]
38 ; CHECK-NEXT: [[J_2:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[INC_2:%.*]], [[FOR_INNER]] ]
39 ; CHECK-NEXT: [[SUM_2:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[ADD_2:%.*]], [[FOR_INNER]] ]
40 ; CHECK-NEXT: [[J_3:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[INC_3:%.*]], [[FOR_INNER]] ]
41 ; CHECK-NEXT: [[SUM_3:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[ADD_3:%.*]], [[FOR_INNER]] ]
42 ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i32 [[J_0]]
43 ; CHECK-NEXT: [[TMP2:%.*]] = load i32, i32* [[ARRAYIDX]], align 4, !tbaa !0
44 ; CHECK-NEXT: [[ADD]] = add i32 [[TMP2]], [[SUM]]
45 ; CHECK-NEXT: [[INC]] = add nuw i32 [[J_0]], 1
46 ; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 [[J_1]]
47 ; CHECK-NEXT: [[TMP3:%.*]] = load i32, i32* [[ARRAYIDX_1]], align 4, !tbaa !0
48 ; CHECK-NEXT: [[ADD_1]] = add i32 [[TMP3]], [[SUM_1]]
49 ; CHECK-NEXT: [[INC_1]] = add nuw i32 [[J_1]], 1
50 ; CHECK-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 [[J_2]]
51 ; CHECK-NEXT: [[TMP4:%.*]] = load i32, i32* [[ARRAYIDX_2]], align 4, !tbaa !0
52 ; CHECK-NEXT: [[ADD_2]] = add i32 [[TMP4]], [[SUM_2]]
53 ; CHECK-NEXT: [[INC_2]] = add nuw i32 [[J_2]], 1
54 ; CHECK-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 [[J_3]]
55 ; CHECK-NEXT: [[TMP5:%.*]] = load i32, i32* [[ARRAYIDX_3]], align 4, !tbaa !0
56 ; CHECK-NEXT: [[ADD_3]] = add i32 [[TMP5]], [[SUM_3]]
57 ; CHECK-NEXT: [[INC_3]] = add nuw i32 [[J_3]], 1
58 ; CHECK-NEXT: [[EXITCOND_3:%.*]] = icmp eq i32 [[INC_3]], [[J]]
59 ; CHECK-NEXT: br i1 [[EXITCOND_3]], label [[FOR_LATCH]], label [[FOR_INNER]]
61 ; CHECK-NEXT: [[ADD_LCSSA:%.*]] = phi i32 [ [[ADD]], [[FOR_INNER]] ]
62 ; CHECK-NEXT: [[ADD_LCSSA_1:%.*]] = phi i32 [ [[ADD_1]], [[FOR_INNER]] ]
63 ; CHECK-NEXT: [[ADD_LCSSA_2:%.*]] = phi i32 [ [[ADD_2]], [[FOR_INNER]] ]
64 ; CHECK-NEXT: [[ADD_LCSSA_3:%.*]] = phi i32 [ [[ADD_3]], [[FOR_INNER]] ]
65 ; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[I]]
66 ; CHECK-NEXT: store i32 [[ADD_LCSSA]], i32* [[ARRAYIDX6]], align 4, !tbaa !0
67 ; CHECK-NEXT: [[ARRAYIDX6_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD8]]
68 ; CHECK-NEXT: store i32 [[ADD_LCSSA_1]], i32* [[ARRAYIDX6_1]], align 4, !tbaa !0
69 ; CHECK-NEXT: [[ARRAYIDX6_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD8_1]]
70 ; CHECK-NEXT: store i32 [[ADD_LCSSA_2]], i32* [[ARRAYIDX6_2]], align 4, !tbaa !0
71 ; CHECK-NEXT: [[ARRAYIDX6_3:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD8_2]]
72 ; CHECK-NEXT: store i32 [[ADD_LCSSA_3]], i32* [[ARRAYIDX6_3]], align 4, !tbaa !0
73 ; CHECK-NEXT: [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NSUB_3]], 0
74 ; CHECK-NEXT: br i1 [[NITER_NCMP_3]], label [[FOR_END_LOOPEXIT_UNR_LCSSA_LOOPEXIT:%.*]], label [[FOR_OUTER]], !llvm.loop !4
75 ; CHECK: for.end.loopexit.unr-lcssa.loopexit:
76 ; CHECK-NEXT: [[I_UNR_PH:%.*]] = phi i32 [ [[ADD8_3]], [[FOR_LATCH]] ]
77 ; CHECK-NEXT: br label [[FOR_END_LOOPEXIT_UNR_LCSSA]]
78 ; CHECK: for.end.loopexit.unr-lcssa:
79 ; CHECK-NEXT: [[I_UNR:%.*]] = phi i32 [ 0, [[FOR_OUTER_PREHEADER]] ], [ [[I_UNR_PH]], [[FOR_END_LOOPEXIT_UNR_LCSSA_LOOPEXIT]] ]
80 ; CHECK-NEXT: [[LCMP_MOD:%.*]] = icmp ne i32 [[XTRAITER]], 0
81 ; CHECK-NEXT: br i1 [[LCMP_MOD]], label [[FOR_OUTER_EPIL_PREHEADER:%.*]], label [[FOR_END_LOOPEXIT:%.*]]
82 ; CHECK: for.outer.epil.preheader:
83 ; CHECK-NEXT: br label [[FOR_OUTER_EPIL:%.*]]
84 ; CHECK: for.outer.epil:
85 ; CHECK-NEXT: br label [[FOR_INNER_EPIL:%.*]]
86 ; CHECK: for.inner.epil:
87 ; CHECK-NEXT: [[J_EPIL:%.*]] = phi i32 [ 0, [[FOR_OUTER_EPIL]] ], [ [[INC_EPIL:%.*]], [[FOR_INNER_EPIL]] ]
88 ; CHECK-NEXT: [[SUM_EPIL:%.*]] = phi i32 [ 0, [[FOR_OUTER_EPIL]] ], [ [[ADD_EPIL:%.*]], [[FOR_INNER_EPIL]] ]
89 ; CHECK-NEXT: [[ARRAYIDX_EPIL:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 [[J_EPIL]]
90 ; CHECK-NEXT: [[TMP6:%.*]] = load i32, i32* [[ARRAYIDX_EPIL]], align 4, !tbaa !0
91 ; CHECK-NEXT: [[ADD_EPIL]] = add i32 [[TMP6]], [[SUM_EPIL]]
92 ; CHECK-NEXT: [[INC_EPIL]] = add nuw i32 [[J_EPIL]], 1
93 ; CHECK-NEXT: [[EXITCOND_EPIL:%.*]] = icmp eq i32 [[INC_EPIL]], [[J]]
94 ; CHECK-NEXT: br i1 [[EXITCOND_EPIL]], label [[FOR_LATCH_EPIL:%.*]], label [[FOR_INNER_EPIL]]
95 ; CHECK: for.latch.epil:
96 ; CHECK-NEXT: [[ADD_LCSSA_EPIL:%.*]] = phi i32 [ [[ADD_EPIL]], [[FOR_INNER_EPIL]] ]
97 ; CHECK-NEXT: [[ARRAYIDX6_EPIL:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[I_UNR]]
98 ; CHECK-NEXT: store i32 [[ADD_LCSSA_EPIL]], i32* [[ARRAYIDX6_EPIL]], align 4, !tbaa !0
99 ; CHECK-NEXT: [[ADD8_EPIL:%.*]] = add nuw i32 [[I_UNR]], 1
100 ; CHECK-NEXT: [[EPIL_ITER_SUB:%.*]] = sub i32 [[XTRAITER]], 1
101 ; CHECK-NEXT: [[EPIL_ITER_CMP:%.*]] = icmp ne i32 [[EPIL_ITER_SUB]], 0
102 ; CHECK-NEXT: br i1 [[EPIL_ITER_CMP]], label [[FOR_OUTER_EPIL_1:%.*]], label [[FOR_END_LOOPEXIT_EPILOG_LCSSA:%.*]]
103 ; CHECK: for.end.loopexit.epilog-lcssa:
104 ; CHECK-NEXT: br label [[FOR_END_LOOPEXIT]]
105 ; CHECK: for.end.loopexit:
106 ; CHECK-NEXT: br label [[FOR_END]]
108 ; CHECK-NEXT: ret void
109 ; CHECK: for.outer.epil.1:
110 ; CHECK-NEXT: br label [[FOR_INNER_EPIL_1:%.*]]
111 ; CHECK: for.inner.epil.1:
112 ; CHECK-NEXT: [[J_EPIL_1:%.*]] = phi i32 [ 0, [[FOR_OUTER_EPIL_1]] ], [ [[INC_EPIL_1:%.*]], [[FOR_INNER_EPIL_1]] ]
113 ; CHECK-NEXT: [[SUM_EPIL_1:%.*]] = phi i32 [ 0, [[FOR_OUTER_EPIL_1]] ], [ [[ADD_EPIL_1:%.*]], [[FOR_INNER_EPIL_1]] ]
114 ; CHECK-NEXT: [[ARRAYIDX_EPIL_1:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 [[J_EPIL_1]]
115 ; CHECK-NEXT: [[TMP7:%.*]] = load i32, i32* [[ARRAYIDX_EPIL_1]], align 4, !tbaa !0
116 ; CHECK-NEXT: [[ADD_EPIL_1]] = add i32 [[TMP7]], [[SUM_EPIL_1]]
117 ; CHECK-NEXT: [[INC_EPIL_1]] = add nuw i32 [[J_EPIL_1]], 1
118 ; CHECK-NEXT: [[EXITCOND_EPIL_1:%.*]] = icmp eq i32 [[INC_EPIL_1]], [[J]]
119 ; CHECK-NEXT: br i1 [[EXITCOND_EPIL_1]], label [[FOR_LATCH_EPIL_1:%.*]], label [[FOR_INNER_EPIL_1]]
120 ; CHECK: for.latch.epil.1:
121 ; CHECK-NEXT: [[ADD_LCSSA_EPIL_1:%.*]] = phi i32 [ [[ADD_EPIL_1]], [[FOR_INNER_EPIL_1]] ]
122 ; CHECK-NEXT: [[ARRAYIDX6_EPIL_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD8_EPIL]]
123 ; CHECK-NEXT: store i32 [[ADD_LCSSA_EPIL_1]], i32* [[ARRAYIDX6_EPIL_1]], align 4, !tbaa !0
124 ; CHECK-NEXT: [[ADD8_EPIL_1:%.*]] = add nuw i32 [[ADD8_EPIL]], 1
125 ; CHECK-NEXT: [[EPIL_ITER_SUB_1:%.*]] = sub i32 [[EPIL_ITER_SUB]], 1
126 ; CHECK-NEXT: [[EPIL_ITER_CMP_1:%.*]] = icmp ne i32 [[EPIL_ITER_SUB_1]], 0
127 ; CHECK-NEXT: br i1 [[EPIL_ITER_CMP_1]], label [[FOR_OUTER_EPIL_2:%.*]], label [[FOR_END_LOOPEXIT_EPILOG_LCSSA]]
128 ; CHECK: for.outer.epil.2:
129 ; CHECK-NEXT: br label [[FOR_INNER_EPIL_2:%.*]]
130 ; CHECK: for.inner.epil.2:
131 ; CHECK-NEXT: [[J_EPIL_2:%.*]] = phi i32 [ 0, [[FOR_OUTER_EPIL_2]] ], [ [[INC_EPIL_2:%.*]], [[FOR_INNER_EPIL_2]] ]
132 ; CHECK-NEXT: [[SUM_EPIL_2:%.*]] = phi i32 [ 0, [[FOR_OUTER_EPIL_2]] ], [ [[ADD_EPIL_2:%.*]], [[FOR_INNER_EPIL_2]] ]
133 ; CHECK-NEXT: [[ARRAYIDX_EPIL_2:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 [[J_EPIL_2]]
134 ; CHECK-NEXT: [[TMP8:%.*]] = load i32, i32* [[ARRAYIDX_EPIL_2]], align 4, !tbaa !0
135 ; CHECK-NEXT: [[ADD_EPIL_2]] = add i32 [[TMP8]], [[SUM_EPIL_2]]
136 ; CHECK-NEXT: [[INC_EPIL_2]] = add nuw i32 [[J_EPIL_2]], 1
137 ; CHECK-NEXT: [[EXITCOND_EPIL_2:%.*]] = icmp eq i32 [[INC_EPIL_2]], [[J]]
138 ; CHECK-NEXT: br i1 [[EXITCOND_EPIL_2]], label [[FOR_LATCH_EPIL_2:%.*]], label [[FOR_INNER_EPIL_2]]
139 ; CHECK: for.latch.epil.2:
140 ; CHECK-NEXT: [[ADD_LCSSA_EPIL_2:%.*]] = phi i32 [ [[ADD_EPIL_2]], [[FOR_INNER_EPIL_2]] ]
141 ; CHECK-NEXT: [[ARRAYIDX6_EPIL_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD8_EPIL_1]]
142 ; CHECK-NEXT: store i32 [[ADD_LCSSA_EPIL_2]], i32* [[ARRAYIDX6_EPIL_2]], align 4, !tbaa !0
143 ; CHECK-NEXT: [[ADD8_EPIL_2:%.*]] = add nuw i32 [[ADD8_EPIL_1]], 1
144 ; CHECK-NEXT: [[EPIL_ITER_SUB_2:%.*]] = sub i32 [[EPIL_ITER_SUB_1]], 1
145 ; CHECK-NEXT: br label [[FOR_END_LOOPEXIT_EPILOG_LCSSA]]
146 define void @test1(i32 %I, i32 %J, i32* noalias nocapture %A, i32* noalias nocapture readonly %B) #0 {
148 %cmp = icmp ne i32 %J, 0
149 %cmpJ = icmp ne i32 %I, 0
150 %or.cond = and i1 %cmp, %cmpJ
151 br i1 %or.cond, label %for.outer.preheader, label %for.end
157 %i = phi i32 [ %add8, %for.latch ], [ 0, %for.outer.preheader ]
161 %j = phi i32 [ 0, %for.outer ], [ %inc, %for.inner ]
162 %sum = phi i32 [ 0, %for.outer ], [ %add, %for.inner ]
163 %arrayidx = getelementptr inbounds i32, i32* %B, i32 %j
164 %0 = load i32, i32* %arrayidx, align 4, !tbaa !5
165 %add = add i32 %0, %sum
166 %inc = add nuw i32 %j, 1
167 %exitcond = icmp eq i32 %inc, %J
168 br i1 %exitcond, label %for.latch, label %for.inner
171 %add.lcssa = phi i32 [ %add, %for.inner ]
172 %arrayidx6 = getelementptr inbounds i32, i32* %A, i32 %i
173 store i32 %add.lcssa, i32* %arrayidx6, align 4, !tbaa !5
174 %add8 = add nuw i32 %i, 1
175 %exitcond25 = icmp eq i32 %add8, %I
176 br i1 %exitcond25, label %for.end.loopexit, label %for.outer
187 ; Tests for(i) { sum = A[i]; for(j) sum += B[j]; A[i] = sum; }
188 ; A[i] load/store dependency should not block unroll-and-jam
190 ; CHECK: %i = phi i32 [ %add9.3, %for.latch ], [ 0, %for.outer.preheader.new ]
191 ; CHECK: %niter = phi i32 [ %unroll_iter, %for.outer.preheader.new ], [ %niter.nsub.3, %for.latch ]
192 ; CHECK: br label %for.inner
194 ; CHECK: %j = phi i32 [ 0, %for.outer ], [ %inc, %for.inner ]
195 ; CHECK: %sum = phi i32 [ %2, %for.outer ], [ %add, %for.inner ]
196 ; CHECK: %j.1 = phi i32 [ 0, %for.outer ], [ %inc.1, %for.inner ]
197 ; CHECK: %sum.1 = phi i32 [ %3, %for.outer ], [ %add.1, %for.inner ]
198 ; CHECK: %j.2 = phi i32 [ 0, %for.outer ], [ %inc.2, %for.inner ]
199 ; CHECK: %sum.2 = phi i32 [ %4, %for.outer ], [ %add.2, %for.inner ]
200 ; CHECK: %j.3 = phi i32 [ 0, %for.outer ], [ %inc.3, %for.inner ]
201 ; CHECK: %sum.3 = phi i32 [ %5, %for.outer ], [ %add.3, %for.inner ]
202 ; CHECK: br i1 %exitcond.3, label %for.latch, label %for.inner
204 ; CHECK: %add.lcssa = phi i32 [ %add, %for.inner ]
205 ; CHECK: %add.lcssa.1 = phi i32 [ %add.1, %for.inner ]
206 ; CHECK: %add.lcssa.2 = phi i32 [ %add.2, %for.inner ]
207 ; CHECK: %add.lcssa.3 = phi i32 [ %add.3, %for.inner ]
208 ; CHECK: br i1 %niter.ncmp.3, label %for.end10.loopexit.unr-lcssa.loopexit, label %for.outer
209 ; CHECK: for.end10.loopexit.unr-lcssa.loopexit:
210 define void @test2(i32 %I, i32 %J, i32* noalias nocapture %A, i32* noalias nocapture readonly %B) #0 {
212 %cmp = icmp ne i32 %J, 0
213 %cmp125 = icmp ne i32 %I, 0
214 %or.cond = and i1 %cmp, %cmp125
215 br i1 %or.cond, label %for.outer.preheader, label %for.end10
221 %i = phi i32 [ %add9, %for.latch ], [ 0, %for.outer.preheader ]
222 %arrayidx = getelementptr inbounds i32, i32* %A, i32 %i
223 %0 = load i32, i32* %arrayidx, align 4, !tbaa !5
227 %j = phi i32 [ 0, %for.outer ], [ %inc, %for.inner ]
228 %sum = phi i32 [ %0, %for.outer ], [ %add, %for.inner ]
229 %arrayidx6 = getelementptr inbounds i32, i32* %B, i32 %j
230 %1 = load i32, i32* %arrayidx6, align 4, !tbaa !5
231 %add = add i32 %1, %sum
232 %inc = add nuw i32 %j, 1
233 %exitcond = icmp eq i32 %inc, %J
234 br i1 %exitcond, label %for.latch, label %for.inner
237 %add.lcssa = phi i32 [ %add, %for.inner ]
238 store i32 %add.lcssa, i32* %arrayidx, align 4, !tbaa !5
239 %add9 = add nuw i32 %i, 1
240 %exitcond28 = icmp eq i32 %add9, %I
241 br i1 %exitcond28, label %for.end10.loopexit, label %for.outer
252 ; Tests Complete unroll-and-jam of the outer loop
254 ; CHECK: br label %for.inner
256 ; CHECK: %j = phi i32 [ 0, %for.outer ], [ %inc, %for.inner ]
257 ; CHECK: %sum = phi i32 [ 0, %for.outer ], [ %add, %for.inner ]
258 ; CHECK: %j.1 = phi i32 [ 0, %for.outer ], [ %inc.1, %for.inner ]
259 ; CHECK: %sum.1 = phi i32 [ 0, %for.outer ], [ %add.1, %for.inner ]
260 ; CHECK: %j.2 = phi i32 [ 0, %for.outer ], [ %inc.2, %for.inner ]
261 ; CHECK: %sum.2 = phi i32 [ 0, %for.outer ], [ %add.2, %for.inner ]
262 ; CHECK: %j.3 = phi i32 [ 0, %for.outer ], [ %inc.3, %for.inner ]
263 ; CHECK: %sum.3 = phi i32 [ 0, %for.outer ], [ %add.3, %for.inner ]
264 ; CHECK: br i1 %exitcond.3, label %for.latch, label %for.inner
266 ; CHECK: %add.lcssa = phi i32 [ %add, %for.inner ]
267 ; CHECK: %add.lcssa.1 = phi i32 [ %add.1, %for.inner ]
268 ; CHECK: %add.lcssa.2 = phi i32 [ %add.2, %for.inner ]
269 ; CHECK: %add.lcssa.3 = phi i32 [ %add.3, %for.inner ]
270 ; CHECK: br label %for.end
272 define void @test3(i32 %I, i32 %J, i32* noalias nocapture %A, i32* noalias nocapture readonly %B) #0 {
274 %cmp = icmp eq i32 %J, 0
275 br i1 %cmp, label %for.end, label %for.preheader
281 %i = phi i32 [ %add8, %for.latch ], [ 0, %for.preheader ]
285 %j = phi i32 [ 0, %for.outer ], [ %inc, %for.inner ]
286 %sum = phi i32 [ 0, %for.outer ], [ %add, %for.inner ]
287 %arrayidx = getelementptr inbounds i32, i32* %B, i32 %j
288 %0 = load i32, i32* %arrayidx, align 4, !tbaa !5
289 %sub = add i32 %sum, 10
290 %add = sub i32 %sub, %0
291 %inc = add nuw i32 %j, 1
292 %exitcond = icmp eq i32 %inc, %J
293 br i1 %exitcond, label %for.latch, label %for.inner
296 %arrayidx6 = getelementptr inbounds i32, i32* %A, i32 %i
297 store i32 %add, i32* %arrayidx6, align 4, !tbaa !5
298 %add8 = add nuw nsw i32 %i, 1
299 %exitcond23 = icmp eq i32 %add8, 4
300 br i1 %exitcond23, label %for.end, label %for.outer
308 ; Tests Complete unroll-and-jam with a trip count of 1
310 ; CHECK: br label %for.inner
312 ; CHECK: %j = phi i32 [ 0, %for.outer ], [ %inc, %for.inner ]
313 ; CHECK: %sum = phi i32 [ 0, %for.outer ], [ %add, %for.inner ]
314 ; CHECK: br i1 %exitcond, label %for.latch, label %for.inner
316 ; CHECK: %add.lcssa = phi i32 [ %add, %for.inner ]
317 ; CHECK: br label %for.end
319 define void @test4(i32 %I, i32 %J, i32* noalias nocapture %A, i32* noalias nocapture readonly %B) #0 {
321 %cmp = icmp eq i32 %J, 0
322 br i1 %cmp, label %for.end, label %for.preheader
328 %i = phi i32 [ %add8, %for.latch ], [ 0, %for.preheader ]
332 %j = phi i32 [ 0, %for.outer ], [ %inc, %for.inner ]
333 %sum = phi i32 [ 0, %for.outer ], [ %add, %for.inner ]
334 %arrayidx = getelementptr inbounds i32, i32* %B, i32 %j
335 %0 = load i32, i32* %arrayidx, align 4, !tbaa !5
336 %sub = add i32 %sum, 10
337 %add = sub i32 %sub, %0
338 %inc = add nuw i32 %j, 1
339 %exitcond = icmp eq i32 %inc, %J
340 br i1 %exitcond, label %for.latch, label %for.inner
343 %arrayidx6 = getelementptr inbounds i32, i32* %A, i32 %i
344 store i32 %add, i32* %arrayidx6, align 4, !tbaa !5
345 %add8 = add nuw nsw i32 %i, 1
346 %exitcond23 = icmp eq i32 %add8, 1
347 br i1 %exitcond23, label %for.end, label %for.outer
355 ; Multiple SubLoopBlocks
357 ; CHECK: br label %for.inner
359 ; CHECK: %inc8.sink15 = phi i32 [ 0, %for.outer ], [ %inc8, %for.inc.1 ]
360 ; CHECK: %inc8.sink15.1 = phi i32 [ 0, %for.outer ], [ %inc8.1, %for.inc.1 ]
361 ; CHECK: br label %for.inner2
363 ; CHECK: br i1 %tobool, label %for.cond4, label %for.inc
365 ; CHECK: br i1 %tobool.1, label %for.cond4a, label %for.inc
367 ; CHECK: br label %for.inc
369 ; CHECK: br i1 %tobool.11, label %for.cond4.1, label %for.inc.1
371 ; CHECK: br label %for.end
374 ; CHECK: for.cond4.1:
375 ; CHECK: br i1 %tobool.1.1, label %for.cond4a.1, label %for.inc.1
376 ; CHECK: for.cond4a.1:
377 ; CHECK: br label %for.inc.1
379 ; CHECK: br i1 %exitcond.1, label %for.latch, label %for.inner
380 @a = hidden global [1 x i32] zeroinitializer, align 4
381 define i32 @test5() #0 {
386 %.sink16 = phi i32 [ 0, %entry ], [ %add, %for.latch ]
390 %inc8.sink15 = phi i32 [ 0, %for.outer ], [ %inc8, %for.inc ]
394 %l1 = load i32, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @a, i32 0, i32 0), align 4
395 %tobool = icmp eq i32 %l1, 0
396 br i1 %tobool, label %for.cond4, label %for.inc
399 %l0 = load i32, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @a, i32 1, i32 0), align 4
400 %tobool.1 = icmp eq i32 %l0, 0
401 br i1 %tobool.1, label %for.cond4a, label %for.inc
407 %l2 = phi i32 [ 0, %for.inner2 ], [ 1, %for.cond4 ], [ 2, %for.cond4a ]
408 %inc8 = add nuw nsw i32 %inc8.sink15, 1
409 %exitcond = icmp eq i32 %inc8, 3
410 br i1 %exitcond, label %for.latch, label %for.inner
413 %.lcssa = phi i32 [ %l2, %for.inc ]
414 %conv11 = and i32 %.sink16, 255
415 %add = add nuw nsw i32 %conv11, 4
416 %cmp = icmp eq i32 %add, 8
417 br i1 %cmp, label %for.end, label %for.outer
420 %.lcssa.lcssa = phi i32 [ %.lcssa, %for.latch ]
426 ; Test odd uses of phi nodes
428 ; CHECK: br label %for.inner
430 ; CHECK: br i1 %exitcond.3, label %for.inner, label %for.latch
432 ; CHECK: br label %for.end
435 @f = hidden global i32 0, align 4
436 define i32 @test6() #0 {
438 %f.promoted10 = load i32, i32* @f, align 4, !tbaa !5
442 %p0 = phi i32 [ %f.promoted10, %entry ], [ 2, %for.latch ]
443 %inc5.sink9 = phi i32 [ 2, %entry ], [ %inc5, %for.latch ]
447 %p1 = phi i32 [ %p0, %for.outer ], [ 2, %for.inner ]
448 %inc.sink8 = phi i32 [ 0, %for.outer ], [ %inc, %for.inner ]
449 %inc = add nuw nsw i32 %inc.sink8, 1
450 %exitcond = icmp ne i32 %inc, 7
451 br i1 %exitcond, label %for.inner, label %for.latch
454 %.lcssa = phi i32 [ %p1, %for.inner ]
455 %inc5 = add nuw nsw i32 %inc5.sink9, 1
456 %exitcond11 = icmp ne i32 %inc5, 7
457 br i1 %exitcond11, label %for.outer, label %for.end
460 %.lcssa.lcssa = phi i32 [ %.lcssa, %for.latch ]
461 %inc.lcssa.lcssa = phi i32 [ 7, %for.latch ]
467 ; Has a positive dependency between two stores. Still valid.
468 ; The negative dependecy is in unroll-and-jam-disabled.ll
470 ; CHECK: %i = phi i32 [ %add.3, %for.latch ], [ 0, %for.preheader.new ]
471 ; CHECK: %niter = phi i32 [ %unroll_iter, %for.preheader.new ], [ %niter.nsub.3, %for.latch ]
472 ; CHECK: br label %for.inner
474 ; CHECK: %add9.lcssa = phi i32 [ %add9, %for.inner ]
475 ; CHECK: %add9.lcssa.1 = phi i32 [ %add9.1, %for.inner ]
476 ; CHECK: %add9.lcssa.2 = phi i32 [ %add9.2, %for.inner ]
477 ; CHECK: %add9.lcssa.3 = phi i32 [ %add9.3, %for.inner ]
478 ; CHECK: br i1 %niter.ncmp.3, label %for.end.loopexit.unr-lcssa.loopexit, label %for.outer
480 ; CHECK: %sum = phi i32 [ 0, %for.outer ], [ %add9, %for.inner ]
481 ; CHECK: %j = phi i32 [ 0, %for.outer ], [ %add10, %for.inner ]
482 ; CHECK: %sum.1 = phi i32 [ 0, %for.outer ], [ %add9.1, %for.inner ]
483 ; CHECK: %j.1 = phi i32 [ 0, %for.outer ], [ %add10.1, %for.inner ]
484 ; CHECK: %sum.2 = phi i32 [ 0, %for.outer ], [ %add9.2, %for.inner ]
485 ; CHECK: %j.2 = phi i32 [ 0, %for.outer ], [ %add10.2, %for.inner ]
486 ; CHECK: %sum.3 = phi i32 [ 0, %for.outer ], [ %add9.3, %for.inner ]
487 ; CHECK: %j.3 = phi i32 [ 0, %for.outer ], [ %add10.3, %for.inner ]
488 ; CHECK: br i1 %exitcond.3, label %for.latch, label %for.inner
489 ; CHECK: for.end.loopexit.unr-lcssa.loopexit:
490 define void @test7(i32 %I, i32 %J, i32* noalias nocapture %A, i32* noalias nocapture readonly %B) #0 {
492 %cmp = icmp ne i32 %J, 0
493 %cmp128 = icmp ne i32 %I, 0
494 %or.cond = and i1 %cmp128, %cmp
495 br i1 %or.cond, label %for.preheader, label %for.end
501 %i = phi i32 [ %add, %for.latch ], [ 0, %for.preheader ]
502 %arrayidx = getelementptr inbounds i32, i32* %A, i32 %i
503 store i32 0, i32* %arrayidx, align 4, !tbaa !5
504 %add = add nuw i32 %i, 1
505 %arrayidx2 = getelementptr inbounds i32, i32* %A, i32 %add
506 store i32 2, i32* %arrayidx2, align 4, !tbaa !5
510 store i32 %add9, i32* %arrayidx, align 4, !tbaa !5
511 %exitcond30 = icmp eq i32 %add, %I
512 br i1 %exitcond30, label %for.end, label %for.outer
515 %sum = phi i32 [ 0, %for.outer ], [ %add9, %for.inner ]
516 %j = phi i32 [ 0, %for.outer ], [ %add10, %for.inner ]
517 %arrayidx7 = getelementptr inbounds i32, i32* %B, i32 %j
518 %l1 = load i32, i32* %arrayidx7, align 4, !tbaa !5
519 %add9 = add i32 %l1, %sum
520 %add10 = add nuw i32 %j, 1
521 %exitcond = icmp eq i32 %add10, %J
522 br i1 %exitcond, label %for.latch, label %for.inner
530 ; Same as test7 with an extra outer loop nest
532 ; CHECK: br label %for.outer
534 ; CHECK: %i = phi i32 [ %add.3, %for.latch ], [ 0, %for.outest.new ]
535 ; CHECK: %niter = phi i32 [ %unroll_iter, %for.outest.new ], [ %niter.nsub.3, %for.latch ]
536 ; CHECK: br label %for.inner
538 ; CHECK: %sum = phi i32 [ 0, %for.outer ], [ %add9, %for.inner ]
539 ; CHECK: %j = phi i32 [ 0, %for.outer ], [ %add10, %for.inner ]
540 ; CHECK: %sum.1 = phi i32 [ 0, %for.outer ], [ %add9.1, %for.inner ]
541 ; CHECK: %j.1 = phi i32 [ 0, %for.outer ], [ %add10.1, %for.inner ]
542 ; CHECK: %sum.2 = phi i32 [ 0, %for.outer ], [ %add9.2, %for.inner ]
543 ; CHECK: %j.2 = phi i32 [ 0, %for.outer ], [ %add10.2, %for.inner ]
544 ; CHECK: %sum.3 = phi i32 [ 0, %for.outer ], [ %add9.3, %for.inner ]
545 ; CHECK: %j.3 = phi i32 [ 0, %for.outer ], [ %add10.3, %for.inner ]
546 ; CHECK: br i1 %exitcond.3, label %for.latch, label %for.inner
548 ; CHECK: %add9.lcssa = phi i32 [ %add9, %for.inner ]
549 ; CHECK: %add9.lcssa.1 = phi i32 [ %add9.1, %for.inner ]
550 ; CHECK: %add9.lcssa.2 = phi i32 [ %add9.2, %for.inner ]
551 ; CHECK: %add9.lcssa.3 = phi i32 [ %add9.3, %for.inner ]
552 ; CHECK: br i1 %niter.ncmp.3, label %for.cleanup.unr-lcssa.loopexit, label %for.outer
553 ; CHECK: for.cleanup.epilog-lcssa:
554 ; CHECK: br label %for.cleanup
555 ; CHECK: for.cleanup:
556 ; CHECK: br i1 %exitcond41, label %for.end.loopexit, label %for.outest
557 ; CHECK: for.end.loopexit:
558 ; CHECK: br label %for.end
559 define void @test8(i32 %I, i32 %J, i32* noalias nocapture %A, i32* noalias nocapture readonly %B) #0 {
561 %cmp = icmp eq i32 %J, 0
562 %cmp336 = icmp eq i32 %I, 0
563 %or.cond = or i1 %cmp, %cmp336
564 br i1 %or.cond, label %for.end, label %for.preheader
570 %x.038 = phi i32 [ %inc, %for.cleanup ], [ 0, %for.preheader ]
574 %i = phi i32 [ %add, %for.latch ], [ 0, %for.outest ]
575 %arrayidx = getelementptr inbounds i32, i32* %A, i32 %i
576 store i32 0, i32* %arrayidx, align 4, !tbaa !5
577 %add = add nuw i32 %i, 1
578 %arrayidx6 = getelementptr inbounds i32, i32* %A, i32 %add
579 store i32 2, i32* %arrayidx6, align 4, !tbaa !5
583 %sum = phi i32 [ 0, %for.outer ], [ %add9, %for.inner ]
584 %j = phi i32 [ 0, %for.outer ], [ %add10, %for.inner ]
585 %arrayidx11 = getelementptr inbounds i32, i32* %B, i32 %j
586 %l1 = load i32, i32* %arrayidx11, align 4, !tbaa !5
587 %add9 = add i32 %l1, %sum
588 %add10 = add nuw i32 %j, 1
589 %exitcond = icmp eq i32 %add10, %J
590 br i1 %exitcond, label %for.latch, label %for.inner
593 store i32 %add9, i32* %arrayidx, align 4, !tbaa !5
594 %exitcond39 = icmp eq i32 %add, %I
595 br i1 %exitcond39, label %for.cleanup, label %for.outer
598 %inc = add nuw nsw i32 %x.038, 1
599 %exitcond41 = icmp eq i32 %inc, 5
600 br i1 %exitcond41, label %for.end, label %for.outest
608 ; Same as test1 with tbaa, not noalias
610 ; CHECK: %i = phi i32 [ %add8.3, %for.latch ], [ 0, %for.outer.preheader.new ]
611 ; CHECK: %niter = phi i32 [ %unroll_iter, %for.outer.preheader.new ], [ %niter.nsub.3, %for.latch ]
612 ; CHECK: br label %for.inner
614 ; CHECK: %j = phi i32 [ 0, %for.outer ], [ %inc, %for.inner ]
615 ; CHECK: %sum = phi i32 [ 0, %for.outer ], [ %add, %for.inner ]
616 ; CHECK: %j.1 = phi i32 [ 0, %for.outer ], [ %inc.1, %for.inner ]
617 ; CHECK: %sum.1 = phi i32 [ 0, %for.outer ], [ %add.1, %for.inner ]
618 ; CHECK: %j.2 = phi i32 [ 0, %for.outer ], [ %inc.2, %for.inner ]
619 ; CHECK: %sum.2 = phi i32 [ 0, %for.outer ], [ %add.2, %for.inner ]
620 ; CHECK: %j.3 = phi i32 [ 0, %for.outer ], [ %inc.3, %for.inner ]
621 ; CHECK: %sum.3 = phi i32 [ 0, %for.outer ], [ %add.3, %for.inner ]
622 ; CHECK: br i1 %exitcond.3, label %for.latch, label %for.inner
624 ; CHECK: %add.lcssa = phi i32 [ %add, %for.inner ]
625 ; CHECK: %add.lcssa.1 = phi i32 [ %add.1, %for.inner ]
626 ; CHECK: %add.lcssa.2 = phi i32 [ %add.2, %for.inner ]
627 ; CHECK: %add.lcssa.3 = phi i32 [ %add.3, %for.inner ]
628 ; CHECK: br i1 %niter.ncmp.3, label %for.end.loopexit.unr-lcssa.loopexit, label %for.outer
629 ; CHECK: for.end.loopexit.unr-lcssa.loopexit:
630 define void @test9(i32 %I, i32 %J, i32* nocapture %A, i16* nocapture readonly %B) #0 {
632 %cmp = icmp ne i32 %J, 0
633 %cmpJ = icmp ne i32 %I, 0
634 %or.cond = and i1 %cmp, %cmpJ
635 br i1 %or.cond, label %for.outer.preheader, label %for.end
641 %i = phi i32 [ %add8, %for.latch ], [ 0, %for.outer.preheader ]
645 %j = phi i32 [ 0, %for.outer ], [ %inc, %for.inner ]
646 %sum = phi i32 [ 0, %for.outer ], [ %add, %for.inner ]
647 %arrayidx = getelementptr inbounds i16, i16* %B, i32 %j
648 %0 = load i16, i16* %arrayidx, align 4, !tbaa !9
649 %sext = sext i16 %0 to i32
650 %add = add i32 %sext, %sum
651 %inc = add nuw i32 %j, 1
652 %exitcond = icmp eq i32 %inc, %J
653 br i1 %exitcond, label %for.latch, label %for.inner
656 %add.lcssa = phi i32 [ %add, %for.inner ]
657 %arrayidx6 = getelementptr inbounds i32, i32* %A, i32 %i
658 store i32 %add.lcssa, i32* %arrayidx6, align 4, !tbaa !5
659 %add8 = add nuw i32 %i, 1
660 %exitcond25 = icmp eq i32 %add8, %I
661 br i1 %exitcond25, label %for.end.loopexit, label %for.outer
671 ; CHECK-LABEL: test10
672 ; Be careful not to incorrectly update the exit phi nodes
673 ; CHECK: %dec.lcssa.lcssa.ph.ph = phi i64 [ 0, %for.inc24 ]
674 %struct.a = type { i64 }
675 @g = common global %struct.a zeroinitializer, align 8
676 @c = common global [1 x i8] zeroinitializer, align 1
677 define signext i16 @test10(i32 %k) #0 {
679 %0 = load i8, i8* getelementptr inbounds ([1 x i8], [1 x i8]* @c, i64 0, i64 0), align 1
680 %tobool9 = icmp eq i8 %0, 0
681 %tobool13 = icmp ne i32 %k, 0
685 %storemerge82 = phi i64 [ 0, %entry ], [ %inc25, %for.inc24 ]
689 %storemerge = phi i64 [ 4, %for.body ], [ %dec, %for.inc21 ]
690 br i1 %tobool9, label %for.body2.split, label %for.body2.split2
693 br i1 %tobool13, label %for.inc21, label %for.inc21.if
696 br i1 %tobool13, label %for.inc21, label %for.inc21.then
699 %storemerge.1 = phi i64 [ 0, %for.body2.split2 ]
703 %storemerge.2 = phi i64 [ 0, %for.body2.split ]
704 %storemerge.3 = phi i32 [ 0, %for.body2.split ]
708 %storemerge.4 = phi i64 [ %storemerge.1, %for.inc21.if ], [ %storemerge.2, %for.inc21.then ], [ 4, %for.body2.split2 ], [ 4, %for.body2.split ]
709 %storemerge.5 = phi i32 [ 0, %for.inc21.if ], [ %storemerge.3, %for.inc21.then ], [ 0, %for.body2.split2 ], [ 0, %for.body2.split ]
710 %dec = add nsw i64 %storemerge, -1
711 %tobool = icmp eq i64 %dec, 0
712 br i1 %tobool, label %for.inc24, label %for.body2
715 %storemerge.4.lcssa = phi i64 [ %storemerge.4, %for.inc21 ]
716 %storemerge.5.lcssa = phi i32 [ %storemerge.5, %for.inc21 ]
717 %inc25 = add nuw nsw i64 %storemerge82, 1
718 %exitcond = icmp ne i64 %inc25, 5
719 br i1 %exitcond, label %for.body, label %for.end26
722 %dec.lcssa.lcssa = phi i64 [ 0, %for.inc24 ]
723 %storemerge.4.lcssa.lcssa = phi i64 [ %storemerge.4.lcssa, %for.inc24 ]
724 %storemerge.5.lcssa.lcssa = phi i32 [ %storemerge.5.lcssa, %for.inc24 ]
725 store i64 %dec.lcssa.lcssa, i64* getelementptr inbounds (%struct.a, %struct.a* @g, i64 0, i32 0), align 8
730 !5 = !{!6, !6, i64 0}
731 !6 = !{!"int", !7, i64 0}
732 !7 = !{!"omnipotent char", !8, i64 0}
733 !8 = !{!"Simple C/C++ TBAA"}
734 !9 = !{!10, !10, i64 0}
735 !10 = !{!"short", !7, i64 0}