1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt -passes=loop-distribute -enable-loop-distribute -S -enable-mem-access-versioning=0 < %s | FileCheck %s
4 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
6 ; PredicatedScalarEvolution decides it needs to insert a bounds check
7 ; not based on memory access.
9 define void @f(ptr noalias %a, ptr noalias %b, ptr noalias %c, ptr noalias %d, ptr noalias %e, i64 %N) {
12 ; CHECK-NEXT: br label [[FOR_BODY_LVER_CHECK:%.*]]
13 ; CHECK: for.body.lver.check:
14 ; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N:%.*]], -1
15 ; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[TMP0]], 4294967295
16 ; CHECK-NEXT: [[MUL1:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 8, i64 [[TMP0]])
17 ; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i64, i1 } [[MUL1]], 0
18 ; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i64, i1 } [[MUL1]], 1
19 ; CHECK-NEXT: [[TMP2:%.*]] = sub i64 0, [[MUL_RESULT]]
20 ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[MUL_RESULT]]
21 ; CHECK-NEXT: [[TMP4:%.*]] = icmp ult ptr [[TMP3]], [[A]]
22 ; CHECK-NEXT: [[TMP5:%.*]] = or i1 [[TMP4]], [[MUL_OVERFLOW]]
23 ; CHECK-NEXT: [[TMP6:%.*]] = or i1 [[TMP1]], [[TMP5]]
24 ; CHECK-NEXT: br i1 [[TMP6]], label [[FOR_BODY_PH_LVER_ORIG:%.*]], label [[FOR_BODY_PH_LDIST1:%.*]]
25 ; CHECK: for.body.ph.lver.orig:
26 ; CHECK-NEXT: br label [[FOR_BODY_LVER_ORIG:%.*]]
27 ; CHECK: for.body.lver.orig:
28 ; CHECK-NEXT: [[IND_LVER_ORIG:%.*]] = phi i64 [ 0, [[FOR_BODY_PH_LVER_ORIG]] ], [ [[ADD_LVER_ORIG:%.*]], [[FOR_BODY_LVER_ORIG]] ]
29 ; CHECK-NEXT: [[IND1_LVER_ORIG:%.*]] = phi i32 [ 0, [[FOR_BODY_PH_LVER_ORIG]] ], [ [[INC1_LVER_ORIG:%.*]], [[FOR_BODY_LVER_ORIG]] ]
30 ; CHECK-NEXT: [[MUL_LVER_ORIG:%.*]] = mul i32 [[IND1_LVER_ORIG]], 2
31 ; CHECK-NEXT: [[MUL_EXT_LVER_ORIG:%.*]] = zext i32 [[MUL_LVER_ORIG]] to i64
32 ; CHECK-NEXT: [[ARRAYIDXA_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[MUL_EXT_LVER_ORIG]]
33 ; CHECK-NEXT: [[LOADA_LVER_ORIG:%.*]] = load i32, ptr [[ARRAYIDXA_LVER_ORIG]], align 4
34 ; CHECK-NEXT: [[ARRAYIDXB_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
35 ; CHECK-NEXT: [[LOADB_LVER_ORIG:%.*]] = load i32, ptr [[ARRAYIDXB_LVER_ORIG]], align 4
36 ; CHECK-NEXT: [[MULA_LVER_ORIG:%.*]] = mul i32 [[LOADB_LVER_ORIG]], [[LOADA_LVER_ORIG]]
37 ; CHECK-NEXT: [[ADD_LVER_ORIG]] = add nuw nsw i64 [[IND_LVER_ORIG]], 1
38 ; CHECK-NEXT: [[INC1_LVER_ORIG]] = add i32 [[IND1_LVER_ORIG]], 1
39 ; CHECK-NEXT: [[ARRAYIDXA_PLUS_4_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[ADD_LVER_ORIG]]
40 ; CHECK-NEXT: store i32 [[MULA_LVER_ORIG]], ptr [[ARRAYIDXA_PLUS_4_LVER_ORIG]], align 4
41 ; CHECK-NEXT: [[ARRAYIDXD_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[D:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
42 ; CHECK-NEXT: [[LOADD_LVER_ORIG:%.*]] = load i32, ptr [[ARRAYIDXD_LVER_ORIG]], align 4
43 ; CHECK-NEXT: [[ARRAYIDXE_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[E:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
44 ; CHECK-NEXT: [[LOADE_LVER_ORIG:%.*]] = load i32, ptr [[ARRAYIDXE_LVER_ORIG]], align 4
45 ; CHECK-NEXT: [[MULC_LVER_ORIG:%.*]] = mul i32 [[LOADD_LVER_ORIG]], [[LOADE_LVER_ORIG]]
46 ; CHECK-NEXT: [[ARRAYIDXC_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[C:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
47 ; CHECK-NEXT: store i32 [[MULC_LVER_ORIG]], ptr [[ARRAYIDXC_LVER_ORIG]], align 4
48 ; CHECK-NEXT: [[EXITCOND_LVER_ORIG:%.*]] = icmp eq i64 [[ADD_LVER_ORIG]], [[N]]
49 ; CHECK-NEXT: br i1 [[EXITCOND_LVER_ORIG]], label [[FOR_END_LOOPEXIT:%.*]], label [[FOR_BODY_LVER_ORIG]]
50 ; CHECK: for.body.ph.ldist1:
51 ; CHECK-NEXT: br label [[FOR_BODY_LDIST1:%.*]]
52 ; CHECK: for.body.ldist1:
53 ; CHECK-NEXT: [[IND_LDIST1:%.*]] = phi i64 [ 0, [[FOR_BODY_PH_LDIST1]] ], [ [[ADD_LDIST1:%.*]], [[FOR_BODY_LDIST1]] ]
54 ; CHECK-NEXT: [[IND1_LDIST1:%.*]] = phi i32 [ 0, [[FOR_BODY_PH_LDIST1]] ], [ [[INC1_LDIST1:%.*]], [[FOR_BODY_LDIST1]] ]
55 ; CHECK-NEXT: [[MUL_LDIST1:%.*]] = mul i32 [[IND1_LDIST1]], 2
56 ; CHECK-NEXT: [[MUL_EXT_LDIST1:%.*]] = zext i32 [[MUL_LDIST1]] to i64
57 ; CHECK-NEXT: [[ARRAYIDXA_LDIST1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[MUL_EXT_LDIST1]]
58 ; CHECK-NEXT: [[LOADA_LDIST1:%.*]] = load i32, ptr [[ARRAYIDXA_LDIST1]], align 4, !alias.scope !0
59 ; CHECK-NEXT: [[ARRAYIDXB_LDIST1:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[MUL_EXT_LDIST1]]
60 ; CHECK-NEXT: [[LOADB_LDIST1:%.*]] = load i32, ptr [[ARRAYIDXB_LDIST1]], align 4
61 ; CHECK-NEXT: [[MULA_LDIST1:%.*]] = mul i32 [[LOADB_LDIST1]], [[LOADA_LDIST1]]
62 ; CHECK-NEXT: [[ADD_LDIST1]] = add nuw nsw i64 [[IND_LDIST1]], 1
63 ; CHECK-NEXT: [[INC1_LDIST1]] = add i32 [[IND1_LDIST1]], 1
64 ; CHECK-NEXT: [[ARRAYIDXA_PLUS_4_LDIST1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[ADD_LDIST1]]
65 ; CHECK-NEXT: store i32 [[MULA_LDIST1]], ptr [[ARRAYIDXA_PLUS_4_LDIST1]], align 4, !alias.scope !3
66 ; CHECK-NEXT: [[EXITCOND_LDIST1:%.*]] = icmp eq i64 [[ADD_LDIST1]], [[N]]
67 ; CHECK-NEXT: br i1 [[EXITCOND_LDIST1]], label [[FOR_BODY_PH:%.*]], label [[FOR_BODY_LDIST1]]
69 ; CHECK-NEXT: br label [[FOR_BODY:%.*]]
71 ; CHECK-NEXT: [[IND:%.*]] = phi i64 [ 0, [[FOR_BODY_PH]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]
72 ; CHECK-NEXT: [[IND1:%.*]] = phi i32 [ 0, [[FOR_BODY_PH]] ], [ [[INC1:%.*]], [[FOR_BODY]] ]
73 ; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[IND1]], 2
74 ; CHECK-NEXT: [[MUL_EXT:%.*]] = zext i32 [[MUL]] to i64
75 ; CHECK-NEXT: [[ADD]] = add nuw nsw i64 [[IND]], 1
76 ; CHECK-NEXT: [[INC1]] = add i32 [[IND1]], 1
77 ; CHECK-NEXT: [[ARRAYIDXD:%.*]] = getelementptr inbounds i32, ptr [[D]], i64 [[MUL_EXT]]
78 ; CHECK-NEXT: [[LOADD:%.*]] = load i32, ptr [[ARRAYIDXD]], align 4
79 ; CHECK-NEXT: [[ARRAYIDXE:%.*]] = getelementptr inbounds i32, ptr [[E]], i64 [[MUL_EXT]]
80 ; CHECK-NEXT: [[LOADE:%.*]] = load i32, ptr [[ARRAYIDXE]], align 4
81 ; CHECK-NEXT: [[MULC:%.*]] = mul i32 [[LOADD]], [[LOADE]]
82 ; CHECK-NEXT: [[ARRAYIDXC:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[MUL_EXT]]
83 ; CHECK-NEXT: store i32 [[MULC]], ptr [[ARRAYIDXC]], align 4
84 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[ADD]], [[N]]
85 ; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT2:%.*]], label [[FOR_BODY]]
86 ; CHECK: for.end.loopexit:
87 ; CHECK-NEXT: br label [[FOR_END:%.*]]
88 ; CHECK: for.end.loopexit2:
89 ; CHECK-NEXT: br label [[FOR_END]]
91 ; CHECK-NEXT: ret void
96 for.body: ; preds = %for.body, %entry
97 %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]
98 %ind1 = phi i32 [ 0, %entry ], [ %inc1, %for.body ]
100 %mul = mul i32 %ind1, 2
101 %mul_ext = zext i32 %mul to i64
104 %arrayidxA = getelementptr inbounds i32, ptr %a, i64 %mul_ext
105 %loadA = load i32, ptr %arrayidxA, align 4
107 %arrayidxB = getelementptr inbounds i32, ptr %b, i64 %mul_ext
108 %loadB = load i32, ptr %arrayidxB, align 4
110 %mulA = mul i32 %loadB, %loadA
112 %add = add nuw nsw i64 %ind, 1
113 %inc1 = add i32 %ind1, 1
115 %arrayidxA_plus_4 = getelementptr inbounds i32, ptr %a, i64 %add
116 store i32 %mulA, ptr %arrayidxA_plus_4, align 4
118 %arrayidxD = getelementptr inbounds i32, ptr %d, i64 %mul_ext
119 %loadD = load i32, ptr %arrayidxD, align 4
121 %arrayidxE = getelementptr inbounds i32, ptr %e, i64 %mul_ext
122 %loadE = load i32, ptr %arrayidxE, align 4
124 %mulC = mul i32 %loadD, %loadE
126 %arrayidxC = getelementptr inbounds i32, ptr %c, i64 %mul_ext
127 store i32 %mulC, ptr %arrayidxC, align 4
129 %exitcond = icmp eq i64 %add, %N
130 br i1 %exitcond, label %for.end, label %for.body
132 for.end: ; preds = %for.body
136 declare void @use64(i64)
137 @global_a = common local_unnamed_addr global [8192 x i32] zeroinitializer, align 16
139 define void @f_with_offset(ptr noalias %b, ptr noalias %c, ptr noalias %d, ptr noalias %e, i64 %N) {
140 ; CHECK-LABEL: @f_with_offset(
142 ; CHECK-NEXT: [[A_INTPTR:%.*]] = ptrtoint ptr @global_a to i64
143 ; CHECK-NEXT: call void @use64(i64 [[A_INTPTR]])
144 ; CHECK-NEXT: [[A:%.*]] = getelementptr i32, ptr @global_a, i32 42
145 ; CHECK-NEXT: br label [[FOR_BODY_LVER_CHECK:%.*]]
146 ; CHECK: for.body.lver.check:
147 ; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N:%.*]], -1
148 ; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[TMP0]], 4294967295
149 ; CHECK-NEXT: [[MUL1:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 8, i64 [[TMP0]])
150 ; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i64, i1 } [[MUL1]], 0
151 ; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i64, i1 } [[MUL1]], 1
152 ; CHECK-NEXT: [[TMP2:%.*]] = sub i64 0, [[MUL_RESULT]]
153 ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[A]], i64 [[MUL_RESULT]]
154 ; CHECK-NEXT: [[TMP4:%.*]] = icmp ult ptr [[TMP3]], [[A]]
155 ; CHECK-NEXT: [[TMP5:%.*]] = or i1 [[TMP4]], [[MUL_OVERFLOW]]
156 ; CHECK-NEXT: [[TMP6:%.*]] = or i1 [[TMP1]], [[TMP5]]
157 ; CHECK-NEXT: br i1 [[TMP6]], label [[FOR_BODY_PH_LVER_ORIG:%.*]], label [[FOR_BODY_PH_LDIST1:%.*]]
158 ; CHECK: for.body.ph.lver.orig:
159 ; CHECK-NEXT: br label [[FOR_BODY_LVER_ORIG:%.*]]
160 ; CHECK: for.body.lver.orig:
161 ; CHECK-NEXT: [[IND_LVER_ORIG:%.*]] = phi i64 [ 0, [[FOR_BODY_PH_LVER_ORIG]] ], [ [[ADD_LVER_ORIG:%.*]], [[FOR_BODY_LVER_ORIG]] ]
162 ; CHECK-NEXT: [[IND1_LVER_ORIG:%.*]] = phi i32 [ 0, [[FOR_BODY_PH_LVER_ORIG]] ], [ [[INC1_LVER_ORIG:%.*]], [[FOR_BODY_LVER_ORIG]] ]
163 ; CHECK-NEXT: [[MUL_LVER_ORIG:%.*]] = mul i32 [[IND1_LVER_ORIG]], 2
164 ; CHECK-NEXT: [[MUL_EXT_LVER_ORIG:%.*]] = zext i32 [[MUL_LVER_ORIG]] to i64
165 ; CHECK-NEXT: [[ARRAYIDXA_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[MUL_EXT_LVER_ORIG]]
166 ; CHECK-NEXT: [[LOADA_LVER_ORIG:%.*]] = load i32, ptr [[ARRAYIDXA_LVER_ORIG]], align 4
167 ; CHECK-NEXT: [[ARRAYIDXB_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
168 ; CHECK-NEXT: [[LOADB_LVER_ORIG:%.*]] = load i32, ptr [[ARRAYIDXB_LVER_ORIG]], align 4
169 ; CHECK-NEXT: [[MULA_LVER_ORIG:%.*]] = mul i32 [[LOADB_LVER_ORIG]], [[LOADA_LVER_ORIG]]
170 ; CHECK-NEXT: [[ADD_LVER_ORIG]] = add nuw nsw i64 [[IND_LVER_ORIG]], 1
171 ; CHECK-NEXT: [[INC1_LVER_ORIG]] = add i32 [[IND1_LVER_ORIG]], 1
172 ; CHECK-NEXT: [[ARRAYIDXA_PLUS_4_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[ADD_LVER_ORIG]]
173 ; CHECK-NEXT: store i32 [[MULA_LVER_ORIG]], ptr [[ARRAYIDXA_PLUS_4_LVER_ORIG]], align 4
174 ; CHECK-NEXT: [[ARRAYIDXD_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[D:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
175 ; CHECK-NEXT: [[LOADD_LVER_ORIG:%.*]] = load i32, ptr [[ARRAYIDXD_LVER_ORIG]], align 4
176 ; CHECK-NEXT: [[ARRAYIDXE_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[E:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
177 ; CHECK-NEXT: [[LOADE_LVER_ORIG:%.*]] = load i32, ptr [[ARRAYIDXE_LVER_ORIG]], align 4
178 ; CHECK-NEXT: [[MULC_LVER_ORIG:%.*]] = mul i32 [[LOADD_LVER_ORIG]], [[LOADE_LVER_ORIG]]
179 ; CHECK-NEXT: [[ARRAYIDXC_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[C:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
180 ; CHECK-NEXT: store i32 [[MULC_LVER_ORIG]], ptr [[ARRAYIDXC_LVER_ORIG]], align 4
181 ; CHECK-NEXT: [[EXITCOND_LVER_ORIG:%.*]] = icmp eq i64 [[ADD_LVER_ORIG]], [[N]]
182 ; CHECK-NEXT: br i1 [[EXITCOND_LVER_ORIG]], label [[FOR_END_LOOPEXIT:%.*]], label [[FOR_BODY_LVER_ORIG]]
183 ; CHECK: for.body.ph.ldist1:
184 ; CHECK-NEXT: br label [[FOR_BODY_LDIST1:%.*]]
185 ; CHECK: for.body.ldist1:
186 ; CHECK-NEXT: [[IND_LDIST1:%.*]] = phi i64 [ 0, [[FOR_BODY_PH_LDIST1]] ], [ [[ADD_LDIST1:%.*]], [[FOR_BODY_LDIST1]] ]
187 ; CHECK-NEXT: [[IND1_LDIST1:%.*]] = phi i32 [ 0, [[FOR_BODY_PH_LDIST1]] ], [ [[INC1_LDIST1:%.*]], [[FOR_BODY_LDIST1]] ]
188 ; CHECK-NEXT: [[MUL_LDIST1:%.*]] = mul i32 [[IND1_LDIST1]], 2
189 ; CHECK-NEXT: [[MUL_EXT_LDIST1:%.*]] = zext i32 [[MUL_LDIST1]] to i64
190 ; CHECK-NEXT: [[ARRAYIDXA_LDIST1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[MUL_EXT_LDIST1]]
191 ; CHECK-NEXT: [[LOADA_LDIST1:%.*]] = load i32, ptr [[ARRAYIDXA_LDIST1]], align 4, !alias.scope !5
192 ; CHECK-NEXT: [[ARRAYIDXB_LDIST1:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[MUL_EXT_LDIST1]]
193 ; CHECK-NEXT: [[LOADB_LDIST1:%.*]] = load i32, ptr [[ARRAYIDXB_LDIST1]], align 4
194 ; CHECK-NEXT: [[MULA_LDIST1:%.*]] = mul i32 [[LOADB_LDIST1]], [[LOADA_LDIST1]]
195 ; CHECK-NEXT: [[ADD_LDIST1]] = add nuw nsw i64 [[IND_LDIST1]], 1
196 ; CHECK-NEXT: [[INC1_LDIST1]] = add i32 [[IND1_LDIST1]], 1
197 ; CHECK-NEXT: [[ARRAYIDXA_PLUS_4_LDIST1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[ADD_LDIST1]]
198 ; CHECK-NEXT: store i32 [[MULA_LDIST1]], ptr [[ARRAYIDXA_PLUS_4_LDIST1]], align 4, !alias.scope !8
199 ; CHECK-NEXT: [[EXITCOND_LDIST1:%.*]] = icmp eq i64 [[ADD_LDIST1]], [[N]]
200 ; CHECK-NEXT: br i1 [[EXITCOND_LDIST1]], label [[FOR_BODY_PH:%.*]], label [[FOR_BODY_LDIST1]]
201 ; CHECK: for.body.ph:
202 ; CHECK-NEXT: br label [[FOR_BODY:%.*]]
204 ; CHECK-NEXT: [[IND:%.*]] = phi i64 [ 0, [[FOR_BODY_PH]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]
205 ; CHECK-NEXT: [[IND1:%.*]] = phi i32 [ 0, [[FOR_BODY_PH]] ], [ [[INC1:%.*]], [[FOR_BODY]] ]
206 ; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[IND1]], 2
207 ; CHECK-NEXT: [[MUL_EXT:%.*]] = zext i32 [[MUL]] to i64
208 ; CHECK-NEXT: [[ADD]] = add nuw nsw i64 [[IND]], 1
209 ; CHECK-NEXT: [[INC1]] = add i32 [[IND1]], 1
210 ; CHECK-NEXT: [[ARRAYIDXD:%.*]] = getelementptr inbounds i32, ptr [[D]], i64 [[MUL_EXT]]
211 ; CHECK-NEXT: [[LOADD:%.*]] = load i32, ptr [[ARRAYIDXD]], align 4
212 ; CHECK-NEXT: [[ARRAYIDXE:%.*]] = getelementptr inbounds i32, ptr [[E]], i64 [[MUL_EXT]]
213 ; CHECK-NEXT: [[LOADE:%.*]] = load i32, ptr [[ARRAYIDXE]], align 4
214 ; CHECK-NEXT: [[MULC:%.*]] = mul i32 [[LOADD]], [[LOADE]]
215 ; CHECK-NEXT: [[ARRAYIDXC:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[MUL_EXT]]
216 ; CHECK-NEXT: store i32 [[MULC]], ptr [[ARRAYIDXC]], align 4
217 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[ADD]], [[N]]
218 ; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT2:%.*]], label [[FOR_BODY]]
219 ; CHECK: for.end.loopexit:
220 ; CHECK-NEXT: br label [[FOR_END:%.*]]
221 ; CHECK: for.end.loopexit2:
222 ; CHECK-NEXT: br label [[FOR_END]]
224 ; CHECK-NEXT: ret void
227 %a_intptr = ptrtoint ptr @global_a to i64
228 call void @use64(i64 %a_intptr)
229 %a = getelementptr i32, ptr @global_a, i32 42
232 for.body: ; preds = %for.body, %entry
233 %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]
234 %ind1 = phi i32 [ 0, %entry ], [ %inc1, %for.body ]
236 %mul = mul i32 %ind1, 2
237 %mul_ext = zext i32 %mul to i64
240 %arrayidxA = getelementptr inbounds i32, ptr %a, i64 %mul_ext
241 %loadA = load i32, ptr %arrayidxA, align 4
243 %arrayidxB = getelementptr inbounds i32, ptr %b, i64 %mul_ext
244 %loadB = load i32, ptr %arrayidxB, align 4
246 %mulA = mul i32 %loadB, %loadA
248 %add = add nuw nsw i64 %ind, 1
249 %inc1 = add i32 %ind1, 1
251 %arrayidxA_plus_4 = getelementptr inbounds i32, ptr %a, i64 %add
252 store i32 %mulA, ptr %arrayidxA_plus_4, align 4
254 %arrayidxD = getelementptr inbounds i32, ptr %d, i64 %mul_ext
255 %loadD = load i32, ptr %arrayidxD, align 4
257 %arrayidxE = getelementptr inbounds i32, ptr %e, i64 %mul_ext
258 %loadE = load i32, ptr %arrayidxE, align 4
260 %mulC = mul i32 %loadD, %loadE
262 %arrayidxC = getelementptr inbounds i32, ptr %c, i64 %mul_ext
263 store i32 %mulC, ptr %arrayidxC, align 4
265 %exitcond = icmp eq i64 %add, %N
266 br i1 %exitcond, label %for.end, label %for.body
268 for.end: ; preds = %for.body
272 ; Can't add control dependency with convergent in loop body.
273 define void @f_with_convergent(ptr noalias %a, ptr noalias %b, ptr noalias %c, ptr noalias %d, ptr noalias %e, i64 %N) #1 {
274 ; CHECK-LABEL: @f_with_convergent(
276 ; CHECK-NEXT: br label [[FOR_BODY:%.*]]
278 ; CHECK-NEXT: [[IND:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]
279 ; CHECK-NEXT: [[IND1:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[INC1:%.*]], [[FOR_BODY]] ]
280 ; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[IND1]], 2
281 ; CHECK-NEXT: [[MUL_EXT:%.*]] = zext i32 [[MUL]] to i64
282 ; CHECK-NEXT: [[ARRAYIDXA:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[MUL_EXT]]
283 ; CHECK-NEXT: [[LOADA:%.*]] = load i32, ptr [[ARRAYIDXA]], align 4
284 ; CHECK-NEXT: [[ARRAYIDXB:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 [[MUL_EXT]]
285 ; CHECK-NEXT: [[LOADB:%.*]] = load i32, ptr [[ARRAYIDXB]], align 4
286 ; CHECK-NEXT: [[MULA:%.*]] = mul i32 [[LOADB]], [[LOADA]]
287 ; CHECK-NEXT: [[ADD]] = add nuw nsw i64 [[IND]], 1
288 ; CHECK-NEXT: [[INC1]] = add i32 [[IND1]], 1
289 ; CHECK-NEXT: [[ARRAYIDXA_PLUS_4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[ADD]]
290 ; CHECK-NEXT: store i32 [[MULA]], ptr [[ARRAYIDXA_PLUS_4]], align 4
291 ; CHECK-NEXT: [[ARRAYIDXD:%.*]] = getelementptr inbounds i32, ptr [[D:%.*]], i64 [[MUL_EXT]]
292 ; CHECK-NEXT: [[LOADD:%.*]] = load i32, ptr [[ARRAYIDXD]], align 4
293 ; CHECK-NEXT: [[ARRAYIDXE:%.*]] = getelementptr inbounds i32, ptr [[E:%.*]], i64 [[MUL_EXT]]
294 ; CHECK-NEXT: [[LOADE:%.*]] = load i32, ptr [[ARRAYIDXE]], align 4
295 ; CHECK-NEXT: [[CONVERGENTD:%.*]] = call i32 @llvm.convergent(i32 [[LOADD]])
296 ; CHECK-NEXT: [[MULC:%.*]] = mul i32 [[CONVERGENTD]], [[LOADE]]
297 ; CHECK-NEXT: [[ARRAYIDXC:%.*]] = getelementptr inbounds i32, ptr [[C:%.*]], i64 [[MUL_EXT]]
298 ; CHECK-NEXT: store i32 [[MULC]], ptr [[ARRAYIDXC]], align 4
299 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[ADD]], [[N:%.*]]
300 ; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[FOR_BODY]]
302 ; CHECK-NEXT: ret void
307 for.body: ; preds = %for.body, %entry
308 %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]
309 %ind1 = phi i32 [ 0, %entry ], [ %inc1, %for.body ]
311 %mul = mul i32 %ind1, 2
312 %mul_ext = zext i32 %mul to i64
315 %arrayidxA = getelementptr inbounds i32, ptr %a, i64 %mul_ext
316 %loadA = load i32, ptr %arrayidxA, align 4
318 %arrayidxB = getelementptr inbounds i32, ptr %b, i64 %mul_ext
319 %loadB = load i32, ptr %arrayidxB, align 4
321 %mulA = mul i32 %loadB, %loadA
323 %add = add nuw nsw i64 %ind, 1
324 %inc1 = add i32 %ind1, 1
326 %arrayidxA_plus_4 = getelementptr inbounds i32, ptr %a, i64 %add
327 store i32 %mulA, ptr %arrayidxA_plus_4, align 4
329 %arrayidxD = getelementptr inbounds i32, ptr %d, i64 %mul_ext
330 %loadD = load i32, ptr %arrayidxD, align 4
332 %arrayidxE = getelementptr inbounds i32, ptr %e, i64 %mul_ext
333 %loadE = load i32, ptr %arrayidxE, align 4
335 %convergentD = call i32 @llvm.convergent(i32 %loadD)
336 %mulC = mul i32 %convergentD, %loadE
338 %arrayidxC = getelementptr inbounds i32, ptr %c, i64 %mul_ext
339 store i32 %mulC, ptr %arrayidxC, align 4
341 %exitcond = icmp eq i64 %add, %N
342 br i1 %exitcond, label %for.end, label %for.body
344 for.end: ; preds = %for.body
348 declare i32 @llvm.convergent(i32) #0
350 attributes #0 = { nounwind readnone convergent }
351 attributes #1 = { nounwind convergent }