| blob | 03812f4acb1ebd2fe6b192768e7bfa7652b45a0f |
1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
2 ; This test verifies that the loop vectorizer will NOT produce a tail
3 ; loop with the optimize for size or the minimize size attributes.
4 ; REQUIRES: asserts
5 ; RUN: opt < %s -passes='require<profile-summary>,loop-vectorize' -S | FileCheck %s
6 ; RUN: opt < %s -passes='require<profile-summary>,loop-vectorize' -pgso -S | FileCheck %s -check-prefix=PGSO
7 ; RUN: opt < %s -passes='require<profile-summary>,loop-vectorize' -pgso=false -S | FileCheck %s -check-prefix=NPGSO
9 target datalayout = "E-m:e-p:32:32-i64:32-f64:32:64-a:0:32-n32-S128"
11 @tab = common global [32 x i8] zeroinitializer, align 1
13 define i32 @foo_optsize() #0 {
14 ; CHECK-LABEL: define i32 @foo_optsize(
15 ; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
16 ; CHECK-NEXT: [[ENTRY:.*]]:
17 ; CHECK-NEXT: br label %[[FOR_BODY:.*]]
18 ; CHECK: [[FOR_BODY]]:
19 ; CHECK-NEXT: [[I_08:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[INC:%.*]], %[[FOR_BODY]] ]
20 ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], ptr @tab, i32 0, i32 [[I_08]]
21 ; CHECK-NEXT: [[TMP0:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
22 ; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP0]], 0
23 ; CHECK-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1
24 ; CHECK-NEXT: store i8 [[DOT]], ptr [[ARRAYIDX]], align 1
25 ; CHECK-NEXT: [[INC]] = add nsw i32 [[I_08]], 1
26 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[I_08]], 202
27 ; CHECK-NEXT: br i1 [[EXITCOND]], label %[[FOR_END:.*]], label %[[FOR_BODY]]
28 ; CHECK: [[FOR_END]]:
29 ; CHECK-NEXT: ret i32 0
30 ;
31 ; PGSO-LABEL: define i32 @foo_optsize(
32 ; PGSO-SAME: ) #[[ATTR0:[0-9]+]] {
33 ; PGSO-NEXT: [[ENTRY:.*]]:
34 ; PGSO-NEXT: br label %[[FOR_BODY:.*]]
35 ; PGSO: [[FOR_BODY]]:
36 ; PGSO-NEXT: [[I_08:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[INC:%.*]], %[[FOR_BODY]] ]
37 ; PGSO-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], ptr @tab, i32 0, i32 [[I_08]]
38 ; PGSO-NEXT: [[TMP0:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
39 ; PGSO-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP0]], 0
40 ; PGSO-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1
41 ; PGSO-NEXT: store i8 [[DOT]], ptr [[ARRAYIDX]], align 1
42 ; PGSO-NEXT: [[INC]] = add nsw i32 [[I_08]], 1
43 ; PGSO-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[I_08]], 202
44 ; PGSO-NEXT: br i1 [[EXITCOND]], label %[[FOR_END:.*]], label %[[FOR_BODY]]
45 ; PGSO: [[FOR_END]]:
46 ; PGSO-NEXT: ret i32 0
47 ;
48 ; NPGSO-LABEL: define i32 @foo_optsize(
49 ; NPGSO-SAME: ) #[[ATTR0:[0-9]+]] {
50 ; NPGSO-NEXT: [[ENTRY:.*]]:
51 ; NPGSO-NEXT: br label %[[FOR_BODY:.*]]
52 ; NPGSO: [[FOR_BODY]]:
53 ; NPGSO-NEXT: [[I_08:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[INC:%.*]], %[[FOR_BODY]] ]
54 ; NPGSO-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], ptr @tab, i32 0, i32 [[I_08]]
55 ; NPGSO-NEXT: [[TMP0:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
56 ; NPGSO-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP0]], 0
57 ; NPGSO-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1
58 ; NPGSO-NEXT: store i8 [[DOT]], ptr [[ARRAYIDX]], align 1
59 ; NPGSO-NEXT: [[INC]] = add nsw i32 [[I_08]], 1
60 ; NPGSO-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[I_08]], 202
61 ; NPGSO-NEXT: br i1 [[EXITCOND]], label %[[FOR_END:.*]], label %[[FOR_BODY]]
62 ; NPGSO: [[FOR_END]]:
63 ; NPGSO-NEXT: ret i32 0
64 ;
66 entry:
67 br label %for.body
69 for.body: ; preds = %for.body, %entry
70 %i.08 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
71 %arrayidx = getelementptr inbounds [32 x i8], ptr @tab, i32 0, i32 %i.08
72 %0 = load i8, ptr %arrayidx, align 1
73 %cmp1 = icmp eq i8 %0, 0
74 %. = select i1 %cmp1, i8 2, i8 1
75 store i8 %., ptr %arrayidx, align 1
76 %inc = add nsw i32 %i.08, 1
77 %exitcond = icmp eq i32 %i.08, 202
78 br i1 %exitcond, label %for.end, label %for.body
80 for.end: ; preds = %for.body
81 ret i32 0
82 }
84 attributes #0 = { optsize }
86 define i32 @foo_minsize() #1 {
87 ; CHECK-LABEL: define i32 @foo_minsize(
88 ; CHECK-SAME: ) #[[ATTR1:[0-9]+]] {
89 ; CHECK-NEXT: [[ENTRY:.*]]:
90 ; CHECK-NEXT: br label %[[FOR_BODY:.*]]
91 ; CHECK: [[FOR_BODY]]:
92 ; CHECK-NEXT: [[I_08:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[INC:%.*]], %[[FOR_BODY]] ]
93 ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], ptr @tab, i32 0, i32 [[I_08]]
94 ; CHECK-NEXT: [[TMP0:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
95 ; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP0]], 0
96 ; CHECK-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1
97 ; CHECK-NEXT: store i8 [[DOT]], ptr [[ARRAYIDX]], align 1
98 ; CHECK-NEXT: [[INC]] = add nsw i32 [[I_08]], 1
99 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[I_08]], 202
100 ; CHECK-NEXT: br i1 [[EXITCOND]], label %[[FOR_END:.*]], label %[[FOR_BODY]]
101 ; CHECK: [[FOR_END]]:
102 ; CHECK-NEXT: ret i32 0
103 ;
104 ; PGSO-LABEL: define i32 @foo_minsize(
105 ; PGSO-SAME: ) #[[ATTR1:[0-9]+]] {
106 ; PGSO-NEXT: [[ENTRY:.*]]:
107 ; PGSO-NEXT: br label %[[FOR_BODY:.*]]
108 ; PGSO: [[FOR_BODY]]:
109 ; PGSO-NEXT: [[I_08:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[INC:%.*]], %[[FOR_BODY]] ]
110 ; PGSO-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], ptr @tab, i32 0, i32 [[I_08]]
111 ; PGSO-NEXT: [[TMP0:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
112 ; PGSO-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP0]], 0
113 ; PGSO-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1
114 ; PGSO-NEXT: store i8 [[DOT]], ptr [[ARRAYIDX]], align 1
115 ; PGSO-NEXT: [[INC]] = add nsw i32 [[I_08]], 1
116 ; PGSO-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[I_08]], 202
117 ; PGSO-NEXT: br i1 [[EXITCOND]], label %[[FOR_END:.*]], label %[[FOR_BODY]]
118 ; PGSO: [[FOR_END]]:
119 ; PGSO-NEXT: ret i32 0
120 ;
121 ; NPGSO-LABEL: define i32 @foo_minsize(
122 ; NPGSO-SAME: ) #[[ATTR1:[0-9]+]] {
123 ; NPGSO-NEXT: [[ENTRY:.*]]:
124 ; NPGSO-NEXT: br label %[[FOR_BODY:.*]]
125 ; NPGSO: [[FOR_BODY]]:
126 ; NPGSO-NEXT: [[I_08:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[INC:%.*]], %[[FOR_BODY]] ]
127 ; NPGSO-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], ptr @tab, i32 0, i32 [[I_08]]
128 ; NPGSO-NEXT: [[TMP0:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
129 ; NPGSO-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP0]], 0
130 ; NPGSO-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1
131 ; NPGSO-NEXT: store i8 [[DOT]], ptr [[ARRAYIDX]], align 1
132 ; NPGSO-NEXT: [[INC]] = add nsw i32 [[I_08]], 1
133 ; NPGSO-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[I_08]], 202
134 ; NPGSO-NEXT: br i1 [[EXITCOND]], label %[[FOR_END:.*]], label %[[FOR_BODY]]
135 ; NPGSO: [[FOR_END]]:
136 ; NPGSO-NEXT: ret i32 0
137 ;
139 entry:
140 br label %for.body
142 for.body: ; preds = %for.body, %entry
143 %i.08 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
144 %arrayidx = getelementptr inbounds [32 x i8], ptr @tab, i32 0, i32 %i.08
145 %0 = load i8, ptr %arrayidx, align 1
146 %cmp1 = icmp eq i8 %0, 0
147 %. = select i1 %cmp1, i8 2, i8 1
148 store i8 %., ptr %arrayidx, align 1
149 %inc = add nsw i32 %i.08, 1
150 %exitcond = icmp eq i32 %i.08, 202
151 br i1 %exitcond, label %for.end, label %for.body
153 for.end: ; preds = %for.body
154 ret i32 0
155 }
157 attributes #1 = { minsize }
159 define i32 @foo_pgso() !prof !14 {
160 ; CHECK-LABEL: define i32 @foo_pgso(
161 ; CHECK-SAME: ) !prof [[PROF14:![0-9]+]] {
162 ; CHECK-NEXT: [[ENTRY:.*]]:
163 ; CHECK-NEXT: br label %[[FOR_BODY:.*]]
164 ; CHECK: [[FOR_BODY]]:
165 ; CHECK-NEXT: [[I_08:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[INC:%.*]], %[[FOR_BODY]] ]
166 ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], ptr @tab, i32 0, i32 [[I_08]]
167 ; CHECK-NEXT: [[TMP0:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
168 ; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP0]], 0
169 ; CHECK-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1
170 ; CHECK-NEXT: store i8 [[DOT]], ptr [[ARRAYIDX]], align 1
171 ; CHECK-NEXT: [[INC]] = add nsw i32 [[I_08]], 1
172 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[I_08]], 202
173 ; CHECK-NEXT: br i1 [[EXITCOND]], label %[[FOR_END:.*]], label %[[FOR_BODY]]
174 ; CHECK: [[FOR_END]]:
175 ; CHECK-NEXT: ret i32 0
176 ;
177 ; PGSO-LABEL: define i32 @foo_pgso(
178 ; PGSO-SAME: ) !prof [[PROF14:![0-9]+]] {
179 ; PGSO-NEXT: [[ENTRY:.*]]:
180 ; PGSO-NEXT: br label %[[FOR_BODY:.*]]
181 ; PGSO: [[FOR_BODY]]:
182 ; PGSO-NEXT: [[I_08:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[INC:%.*]], %[[FOR_BODY]] ]
183 ; PGSO-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], ptr @tab, i32 0, i32 [[I_08]]
184 ; PGSO-NEXT: [[TMP0:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
185 ; PGSO-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP0]], 0
186 ; PGSO-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1
187 ; PGSO-NEXT: store i8 [[DOT]], ptr [[ARRAYIDX]], align 1
188 ; PGSO-NEXT: [[INC]] = add nsw i32 [[I_08]], 1
189 ; PGSO-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[I_08]], 202
190 ; PGSO-NEXT: br i1 [[EXITCOND]], label %[[FOR_END:.*]], label %[[FOR_BODY]]
191 ; PGSO: [[FOR_END]]:
192 ; PGSO-NEXT: ret i32 0
193 ;
194 ; NPGSO-LABEL: define i32 @foo_pgso(
195 ; NPGSO-SAME: ) !prof [[PROF14:![0-9]+]] {
196 ; NPGSO-NEXT: [[ENTRY:.*]]:
197 ; NPGSO-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
198 ; NPGSO: [[VECTOR_PH]]:
199 ; NPGSO-NEXT: br label %[[VECTOR_BODY:.*]]
200 ; NPGSO: [[VECTOR_BODY]]:
201 ; NPGSO-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
202 ; NPGSO-NEXT: [[TMP0:%.*]] = add i32 [[INDEX]], 0
203 ; NPGSO-NEXT: [[TMP1:%.*]] = getelementptr inbounds [32 x i8], ptr @tab, i32 0, i32 [[TMP0]]
204 ; NPGSO-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i32 0
205 ; NPGSO-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, ptr [[TMP2]], align 1
206 ; NPGSO-NEXT: [[TMP3:%.*]] = icmp eq <4 x i8> [[WIDE_LOAD]], zeroinitializer
207 ; NPGSO-NEXT: [[TMP4:%.*]] = select <4 x i1> [[TMP3]], <4 x i8> splat (i8 2), <4 x i8> splat (i8 1)
208 ; NPGSO-NEXT: store <4 x i8> [[TMP4]], ptr [[TMP2]], align 1
209 ; NPGSO-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
210 ; NPGSO-NEXT: [[TMP5:%.*]] = icmp eq i32 [[INDEX_NEXT]], 200
211 ; NPGSO-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
212 ; NPGSO: [[MIDDLE_BLOCK]]:
213 ; NPGSO-NEXT: br i1 false, label %[[FOR_END:.*]], label %[[SCALAR_PH]]
214 ; NPGSO: [[SCALAR_PH]]:
215 ; NPGSO-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 200, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
216 ; NPGSO-NEXT: br label %[[FOR_BODY:.*]]
217 ; NPGSO: [[FOR_BODY]]:
218 ; NPGSO-NEXT: [[I_08:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY]] ]
219 ; NPGSO-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], ptr @tab, i32 0, i32 [[I_08]]
220 ; NPGSO-NEXT: [[TMP6:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
221 ; NPGSO-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP6]], 0
222 ; NPGSO-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1
223 ; NPGSO-NEXT: store i8 [[DOT]], ptr [[ARRAYIDX]], align 1
224 ; NPGSO-NEXT: [[INC]] = add nsw i32 [[I_08]], 1
225 ; NPGSO-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[I_08]], 202
226 ; NPGSO-NEXT: br i1 [[EXITCOND]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]]
227 ; NPGSO: [[FOR_END]]:
228 ; NPGSO-NEXT: ret i32 0
229 ;
231 entry:
232 br label %for.body
234 for.body: ; preds = %for.body, %entry
235 %i.08 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
236 %arrayidx = getelementptr inbounds [32 x i8], ptr @tab, i32 0, i32 %i.08
237 %0 = load i8, ptr %arrayidx, align 1
238 %cmp1 = icmp eq i8 %0, 0
239 %. = select i1 %cmp1, i8 2, i8 1
240 store i8 %., ptr %arrayidx, align 1
241 %inc = add nsw i32 %i.08, 1
242 %exitcond = icmp eq i32 %i.08, 202
243 br i1 %exitcond, label %for.end, label %for.body
245 for.end: ; preds = %for.body
246 ret i32 0
247 }
249 ; PR43371: don't run into an assert due to emitting SCEV runtime checks
250 ; with OptForSize.
251 ;
252 @cm_array = external global [2592 x i16], align 1
254 define void @pr43371() optsize {
255 ;
256 ; CHECK-LABEL: define void @pr43371(
257 ; CHECK-SAME: ) #[[ATTR0]] {
258 ; CHECK-NEXT: [[ENTRY:.*]]:
259 ; CHECK-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
260 ; CHECK: [[VECTOR_PH]]:
261 ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
262 ; CHECK: [[VECTOR_BODY]]:
263 ; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
264 ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
265 ; CHECK-NEXT: [[TMP0:%.*]] = add <2 x i16> undef, [[VEC_IND]]
266 ; CHECK-NEXT: [[TMP1:%.*]] = zext <2 x i16> [[TMP0]] to <2 x i32>
267 ; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i32> [[TMP1]], i32 0
268 ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[TMP2]]
269 ; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
270 ; CHECK-NEXT: [[TMP5:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[TMP4]]
271 ; CHECK-NEXT: store i16 0, ptr [[TMP3]], align 1
272 ; CHECK-NEXT: store i16 0, ptr [[TMP5]], align 1
273 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
274 ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i16> [[VEC_IND]], splat (i16 2)
275 ; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], 756
276 ; CHECK-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
277 ; CHECK: [[MIDDLE_BLOCK]]:
278 ; CHECK-NEXT: br i1 true, label %[[FOR_COND_CLEANUP28:.*]], label %[[SCALAR_PH]]
279 ; CHECK: [[SCALAR_PH]]:
280 ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ 756, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
281 ; CHECK-NEXT: br label %[[FOR_BODY29:.*]]
282 ; CHECK: [[FOR_COND_CLEANUP28]]:
283 ; CHECK-NEXT: unreachable
284 ; CHECK: [[FOR_BODY29]]:
285 ; CHECK-NEXT: [[I24_0170:%.*]] = phi i16 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC37:%.*]], %[[FOR_BODY29]] ]
286 ; CHECK-NEXT: [[ADD33:%.*]] = add i16 undef, [[I24_0170]]
287 ; CHECK-NEXT: [[IDXPROM34:%.*]] = zext i16 [[ADD33]] to i32
288 ; CHECK-NEXT: [[ARRAYIDX35:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[IDXPROM34]]
289 ; CHECK-NEXT: store i16 0, ptr [[ARRAYIDX35]], align 1
290 ; CHECK-NEXT: [[INC37]] = add i16 [[I24_0170]], 1
291 ; CHECK-NEXT: [[CMP26:%.*]] = icmp ult i16 [[INC37]], 756
292 ; CHECK-NEXT: br i1 [[CMP26]], label %[[FOR_BODY29]], label %[[FOR_COND_CLEANUP28]], !llvm.loop [[LOOP18:![0-9]+]]
293 ;
294 ; PGSO-LABEL: define void @pr43371(
295 ; PGSO-SAME: ) #[[ATTR0]] {
296 ; PGSO-NEXT: [[ENTRY:.*]]:
297 ; PGSO-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
298 ; PGSO: [[VECTOR_PH]]:
299 ; PGSO-NEXT: br label %[[VECTOR_BODY:.*]]
300 ; PGSO: [[VECTOR_BODY]]:
301 ; PGSO-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
302 ; PGSO-NEXT: [[VEC_IND:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
303 ; PGSO-NEXT: [[TMP0:%.*]] = add <2 x i16> undef, [[VEC_IND]]
304 ; PGSO-NEXT: [[TMP1:%.*]] = zext <2 x i16> [[TMP0]] to <2 x i32>
305 ; PGSO-NEXT: [[TMP2:%.*]] = extractelement <2 x i32> [[TMP1]], i32 0
306 ; PGSO-NEXT: [[TMP3:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[TMP2]]
307 ; PGSO-NEXT: [[TMP4:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
308 ; PGSO-NEXT: [[TMP5:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[TMP4]]
309 ; PGSO-NEXT: store i16 0, ptr [[TMP3]], align 1
310 ; PGSO-NEXT: store i16 0, ptr [[TMP5]], align 1
311 ; PGSO-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
312 ; PGSO-NEXT: [[VEC_IND_NEXT]] = add <2 x i16> [[VEC_IND]], splat (i16 2)
313 ; PGSO-NEXT: [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], 756
314 ; PGSO-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
315 ; PGSO: [[MIDDLE_BLOCK]]:
316 ; PGSO-NEXT: br i1 true, label %[[FOR_COND_CLEANUP28:.*]], label %[[SCALAR_PH]]
317 ; PGSO: [[SCALAR_PH]]:
318 ; PGSO-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ 756, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
319 ; PGSO-NEXT: br label %[[FOR_BODY29:.*]]
320 ; PGSO: [[FOR_COND_CLEANUP28]]:
321 ; PGSO-NEXT: unreachable
322 ; PGSO: [[FOR_BODY29]]:
323 ; PGSO-NEXT: [[I24_0170:%.*]] = phi i16 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC37:%.*]], %[[FOR_BODY29]] ]
324 ; PGSO-NEXT: [[ADD33:%.*]] = add i16 undef, [[I24_0170]]
325 ; PGSO-NEXT: [[IDXPROM34:%.*]] = zext i16 [[ADD33]] to i32
326 ; PGSO-NEXT: [[ARRAYIDX35:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[IDXPROM34]]
327 ; PGSO-NEXT: store i16 0, ptr [[ARRAYIDX35]], align 1
328 ; PGSO-NEXT: [[INC37]] = add i16 [[I24_0170]], 1
329 ; PGSO-NEXT: [[CMP26:%.*]] = icmp ult i16 [[INC37]], 756
330 ; PGSO-NEXT: br i1 [[CMP26]], label %[[FOR_BODY29]], label %[[FOR_COND_CLEANUP28]], !llvm.loop [[LOOP18:![0-9]+]]
331 ;
332 ; NPGSO-LABEL: define void @pr43371(
333 ; NPGSO-SAME: ) #[[ATTR0]] {
334 ; NPGSO-NEXT: [[ENTRY:.*]]:
335 ; NPGSO-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
336 ; NPGSO: [[VECTOR_PH]]:
337 ; NPGSO-NEXT: br label %[[VECTOR_BODY:.*]]
338 ; NPGSO: [[VECTOR_BODY]]:
339 ; NPGSO-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
340 ; NPGSO-NEXT: [[VEC_IND:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
341 ; NPGSO-NEXT: [[TMP0:%.*]] = add <2 x i16> undef, [[VEC_IND]]
342 ; NPGSO-NEXT: [[TMP1:%.*]] = zext <2 x i16> [[TMP0]] to <2 x i32>
343 ; NPGSO-NEXT: [[TMP2:%.*]] = extractelement <2 x i32> [[TMP1]], i32 0
344 ; NPGSO-NEXT: [[TMP3:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[TMP2]]
345 ; NPGSO-NEXT: [[TMP4:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
346 ; NPGSO-NEXT: [[TMP5:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[TMP4]]
347 ; NPGSO-NEXT: store i16 0, ptr [[TMP3]], align 1
348 ; NPGSO-NEXT: store i16 0, ptr [[TMP5]], align 1
349 ; NPGSO-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
350 ; NPGSO-NEXT: [[VEC_IND_NEXT]] = add <2 x i16> [[VEC_IND]], splat (i16 2)
351 ; NPGSO-NEXT: [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], 756
352 ; NPGSO-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP19:![0-9]+]]
353 ; NPGSO: [[MIDDLE_BLOCK]]:
354 ; NPGSO-NEXT: br i1 true, label %[[FOR_COND_CLEANUP28:.*]], label %[[SCALAR_PH]]
355 ; NPGSO: [[SCALAR_PH]]:
356 ; NPGSO-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ 756, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
357 ; NPGSO-NEXT: br label %[[FOR_BODY29:.*]]
358 ; NPGSO: [[FOR_COND_CLEANUP28]]:
359 ; NPGSO-NEXT: unreachable
360 ; NPGSO: [[FOR_BODY29]]:
361 ; NPGSO-NEXT: [[I24_0170:%.*]] = phi i16 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC37:%.*]], %[[FOR_BODY29]] ]
362 ; NPGSO-NEXT: [[ADD33:%.*]] = add i16 undef, [[I24_0170]]
363 ; NPGSO-NEXT: [[IDXPROM34:%.*]] = zext i16 [[ADD33]] to i32
364 ; NPGSO-NEXT: [[ARRAYIDX35:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[IDXPROM34]]
365 ; NPGSO-NEXT: store i16 0, ptr [[ARRAYIDX35]], align 1
366 ; NPGSO-NEXT: [[INC37]] = add i16 [[I24_0170]], 1
367 ; NPGSO-NEXT: [[CMP26:%.*]] = icmp ult i16 [[INC37]], 756
368 ; NPGSO-NEXT: br i1 [[CMP26]], label %[[FOR_BODY29]], label %[[FOR_COND_CLEANUP28]], !llvm.loop [[LOOP20:![0-9]+]]
369 ;
370 ; We do not want to generate SCEV predicates when optimising for size, because
371 ; that will lead to extra code generation such as the SCEV overflow runtime
372 ; checks. Not generating SCEV predicates can still result in vectorisation as
373 ; the non-consecutive loads/stores can be scalarized:
374 entry:
375 br label %for.body29
377 for.cond.cleanup28:
378 unreachable
380 for.body29:
381 %i24.0170 = phi i16 [ 0, %entry], [ %inc37, %for.body29]
382 %add33 = add i16 undef, %i24.0170
383 %idxprom34 = zext i16 %add33 to i32
384 %arrayidx35 = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 %idxprom34
385 store i16 0, ptr %arrayidx35, align 1
386 %inc37 = add i16 %i24.0170, 1
387 %cmp26 = icmp ult i16 %inc37, 756
388 br i1 %cmp26, label %for.body29, label %for.cond.cleanup28
389 }
391 define void @pr43371_pgso() !prof !14 {
392 ;
393 ; CHECK-LABEL: define void @pr43371_pgso(
394 ; CHECK-SAME: ) !prof [[PROF14]] {
395 ; CHECK-NEXT: [[ENTRY:.*]]:
396 ; CHECK-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
397 ; CHECK: [[VECTOR_PH]]:
398 ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
399 ; CHECK: [[VECTOR_BODY]]:
400 ; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
401 ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
402 ; CHECK-NEXT: [[TMP0:%.*]] = add <2 x i16> undef, [[VEC_IND]]
403 ; CHECK-NEXT: [[TMP1:%.*]] = zext <2 x i16> [[TMP0]] to <2 x i32>
404 ; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i32> [[TMP1]], i32 0
405 ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[TMP2]]
406 ; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
407 ; CHECK-NEXT: [[TMP5:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[TMP4]]
408 ; CHECK-NEXT: store i16 0, ptr [[TMP3]], align 1
409 ; CHECK-NEXT: store i16 0, ptr [[TMP5]], align 1
410 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
411 ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i16> [[VEC_IND]], splat (i16 2)
412 ; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], 756
413 ; CHECK-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP19:![0-9]+]]
414 ; CHECK: [[MIDDLE_BLOCK]]:
415 ; CHECK-NEXT: br i1 true, label %[[FOR_COND_CLEANUP28:.*]], label %[[SCALAR_PH]]
416 ; CHECK: [[SCALAR_PH]]:
417 ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ 756, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
418 ; CHECK-NEXT: br label %[[FOR_BODY29:.*]]
419 ; CHECK: [[FOR_COND_CLEANUP28]]:
420 ; CHECK-NEXT: unreachable
421 ; CHECK: [[FOR_BODY29]]:
422 ; CHECK-NEXT: [[I24_0170:%.*]] = phi i16 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC37:%.*]], %[[FOR_BODY29]] ]
423 ; CHECK-NEXT: [[ADD33:%.*]] = add i16 undef, [[I24_0170]]
424 ; CHECK-NEXT: [[IDXPROM34:%.*]] = zext i16 [[ADD33]] to i32
425 ; CHECK-NEXT: [[ARRAYIDX35:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[IDXPROM34]]
426 ; CHECK-NEXT: store i16 0, ptr [[ARRAYIDX35]], align 1
427 ; CHECK-NEXT: [[INC37]] = add i16 [[I24_0170]], 1
428 ; CHECK-NEXT: [[CMP26:%.*]] = icmp ult i16 [[INC37]], 756
429 ; CHECK-NEXT: br i1 [[CMP26]], label %[[FOR_BODY29]], label %[[FOR_COND_CLEANUP28]], !llvm.loop [[LOOP20:![0-9]+]]
430 ;
431 ; PGSO-LABEL: define void @pr43371_pgso(
432 ; PGSO-SAME: ) !prof [[PROF14]] {
433 ; PGSO-NEXT: [[ENTRY:.*]]:
434 ; PGSO-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
435 ; PGSO: [[VECTOR_PH]]:
436 ; PGSO-NEXT: br label %[[VECTOR_BODY:.*]]
437 ; PGSO: [[VECTOR_BODY]]:
438 ; PGSO-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
439 ; PGSO-NEXT: [[VEC_IND:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
440 ; PGSO-NEXT: [[TMP0:%.*]] = add <2 x i16> undef, [[VEC_IND]]
441 ; PGSO-NEXT: [[TMP1:%.*]] = zext <2 x i16> [[TMP0]] to <2 x i32>
442 ; PGSO-NEXT: [[TMP2:%.*]] = extractelement <2 x i32> [[TMP1]], i32 0
443 ; PGSO-NEXT: [[TMP3:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[TMP2]]
444 ; PGSO-NEXT: [[TMP4:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
445 ; PGSO-NEXT: [[TMP5:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[TMP4]]
446 ; PGSO-NEXT: store i16 0, ptr [[TMP3]], align 1
447 ; PGSO-NEXT: store i16 0, ptr [[TMP5]], align 1
448 ; PGSO-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
449 ; PGSO-NEXT: [[VEC_IND_NEXT]] = add <2 x i16> [[VEC_IND]], splat (i16 2)
450 ; PGSO-NEXT: [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], 756
451 ; PGSO-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP19:![0-9]+]]
452 ; PGSO: [[MIDDLE_BLOCK]]:
453 ; PGSO-NEXT: br i1 true, label %[[FOR_COND_CLEANUP28:.*]], label %[[SCALAR_PH]]
454 ; PGSO: [[SCALAR_PH]]:
455 ; PGSO-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ 756, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
456 ; PGSO-NEXT: br label %[[FOR_BODY29:.*]]
457 ; PGSO: [[FOR_COND_CLEANUP28]]:
458 ; PGSO-NEXT: unreachable
459 ; PGSO: [[FOR_BODY29]]:
460 ; PGSO-NEXT: [[I24_0170:%.*]] = phi i16 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC37:%.*]], %[[FOR_BODY29]] ]
461 ; PGSO-NEXT: [[ADD33:%.*]] = add i16 undef, [[I24_0170]]
462 ; PGSO-NEXT: [[IDXPROM34:%.*]] = zext i16 [[ADD33]] to i32
463 ; PGSO-NEXT: [[ARRAYIDX35:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[IDXPROM34]]
464 ; PGSO-NEXT: store i16 0, ptr [[ARRAYIDX35]], align 1
465 ; PGSO-NEXT: [[INC37]] = add i16 [[I24_0170]], 1
466 ; PGSO-NEXT: [[CMP26:%.*]] = icmp ult i16 [[INC37]], 756
467 ; PGSO-NEXT: br i1 [[CMP26]], label %[[FOR_BODY29]], label %[[FOR_COND_CLEANUP28]], !llvm.loop [[LOOP20:![0-9]+]]
468 ;
469 ; NPGSO-LABEL: define void @pr43371_pgso(
470 ; NPGSO-SAME: ) !prof [[PROF14]] {
471 ; NPGSO-NEXT: [[ENTRY:.*]]:
472 ; NPGSO-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
473 ; NPGSO: [[VECTOR_SCEVCHECK]]:
474 ; NPGSO-NEXT: br i1 undef, label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
475 ; NPGSO: [[VECTOR_PH]]:
476 ; NPGSO-NEXT: br label %[[VECTOR_BODY:.*]]
477 ; NPGSO: [[VECTOR_BODY]]:
478 ; NPGSO-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
479 ; NPGSO-NEXT: [[OFFSET_IDX:%.*]] = trunc i32 [[INDEX]] to i16
480 ; NPGSO-NEXT: [[TMP0:%.*]] = add i16 [[OFFSET_IDX]], 0
481 ; NPGSO-NEXT: [[TMP1:%.*]] = add i16 undef, [[TMP0]]
482 ; NPGSO-NEXT: [[TMP2:%.*]] = zext i16 [[TMP1]] to i32
483 ; NPGSO-NEXT: [[TMP3:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[TMP2]]
484 ; NPGSO-NEXT: [[TMP4:%.*]] = getelementptr i16, ptr [[TMP3]], i32 0
485 ; NPGSO-NEXT: store <2 x i16> zeroinitializer, ptr [[TMP4]], align 1
486 ; NPGSO-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
487 ; NPGSO-NEXT: [[TMP5:%.*]] = icmp eq i32 [[INDEX_NEXT]], 756
488 ; NPGSO-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP21:![0-9]+]]
489 ; NPGSO: [[MIDDLE_BLOCK]]:
490 ; NPGSO-NEXT: br i1 true, label %[[FOR_COND_CLEANUP28:.*]], label %[[SCALAR_PH]]
491 ; NPGSO: [[SCALAR_PH]]:
492 ; NPGSO-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ 756, %[[MIDDLE_BLOCK]] ], [ 0, %[[VECTOR_SCEVCHECK]] ], [ 0, %[[ENTRY]] ]
493 ; NPGSO-NEXT: br label %[[FOR_BODY29:.*]]
494 ; NPGSO: [[FOR_COND_CLEANUP28]]:
495 ; NPGSO-NEXT: unreachable
496 ; NPGSO: [[FOR_BODY29]]:
497 ; NPGSO-NEXT: [[I24_0170:%.*]] = phi i16 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC37:%.*]], %[[FOR_BODY29]] ]
498 ; NPGSO-NEXT: [[ADD33:%.*]] = add i16 undef, [[I24_0170]]
499 ; NPGSO-NEXT: [[IDXPROM34:%.*]] = zext i16 [[ADD33]] to i32
500 ; NPGSO-NEXT: [[ARRAYIDX35:%.*]] = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 [[IDXPROM34]]
501 ; NPGSO-NEXT: store i16 0, ptr [[ARRAYIDX35]], align 1
502 ; NPGSO-NEXT: [[INC37]] = add i16 [[I24_0170]], 1
503 ; NPGSO-NEXT: [[CMP26:%.*]] = icmp ult i16 [[INC37]], 756
504 ; NPGSO-NEXT: br i1 [[CMP26]], label %[[FOR_BODY29]], label %[[FOR_COND_CLEANUP28]], !llvm.loop [[LOOP22:![0-9]+]]
505 ;
506 ; We do not want to generate SCEV predicates when optimising for size, because
507 ; that will lead to extra code generation such as the SCEV overflow runtime
508 ; checks. Not generating SCEV predicates can still result in vectorisation as
509 ; the non-consecutive loads/stores can be scalarized:
510 entry:
511 br label %for.body29
513 for.cond.cleanup28:
514 unreachable
516 for.body29:
517 %i24.0170 = phi i16 [ 0, %entry], [ %inc37, %for.body29]
518 %add33 = add i16 undef, %i24.0170
519 %idxprom34 = zext i16 %add33 to i32
520 %arrayidx35 = getelementptr [2592 x i16], ptr @cm_array, i32 0, i32 %idxprom34
521 store i16 0, ptr %arrayidx35, align 1
522 %inc37 = add i16 %i24.0170, 1
523 %cmp26 = icmp ult i16 %inc37, 756
524 br i1 %cmp26, label %for.body29, label %for.cond.cleanup28
525 }
527 ; PR45526: don't vectorize with fold-tail if first-order-recurrence is live-out.
528 ;
529 define i32 @pr45526() optsize {
530 ;
531 ; CHECK-LABEL: define i32 @pr45526(
532 ; CHECK-SAME: ) #[[ATTR0]] {
533 ; CHECK-NEXT: [[ENTRY:.*]]:
534 ; CHECK-NEXT: br label %[[LOOP:.*]]
535 ; CHECK: [[LOOP]]:
536 ; CHECK-NEXT: [[PIV:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[PIVPLUS1:%.*]], %[[LOOP]] ]
537 ; CHECK-NEXT: [[FOR:%.*]] = phi i32 [ 5, %[[ENTRY]] ], [ [[PIVPLUS1]], %[[LOOP]] ]
538 ; CHECK-NEXT: [[PIVPLUS1]] = add nuw nsw i32 [[PIV]], 1
539 ; CHECK-NEXT: [[COND:%.*]] = icmp ult i32 [[PIV]], 510
540 ; CHECK-NEXT: br i1 [[COND]], label %[[LOOP]], label %[[EXIT:.*]]
541 ; CHECK: [[EXIT]]:
542 ; CHECK-NEXT: [[FOR_LCSSA:%.*]] = phi i32 [ [[FOR]], %[[LOOP]] ]
543 ; CHECK-NEXT: ret i32 [[FOR_LCSSA]]
544 ;
545 ; PGSO-LABEL: define i32 @pr45526(
546 ; PGSO-SAME: ) #[[ATTR0]] {
547 ; PGSO-NEXT: [[ENTRY:.*]]:
548 ; PGSO-NEXT: br label %[[LOOP:.*]]
549 ; PGSO: [[LOOP]]:
550 ; PGSO-NEXT: [[PIV:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[PIVPLUS1:%.*]], %[[LOOP]] ]
551 ; PGSO-NEXT: [[FOR:%.*]] = phi i32 [ 5, %[[ENTRY]] ], [ [[PIVPLUS1]], %[[LOOP]] ]
552 ; PGSO-NEXT: [[PIVPLUS1]] = add nuw nsw i32 [[PIV]], 1
553 ; PGSO-NEXT: [[COND:%.*]] = icmp ult i32 [[PIV]], 510
554 ; PGSO-NEXT: br i1 [[COND]], label %[[LOOP]], label %[[EXIT:.*]]
555 ; PGSO: [[EXIT]]:
556 ; PGSO-NEXT: [[FOR_LCSSA:%.*]] = phi i32 [ [[FOR]], %[[LOOP]] ]
557 ; PGSO-NEXT: ret i32 [[FOR_LCSSA]]
558 ;
559 ; NPGSO-LABEL: define i32 @pr45526(
560 ; NPGSO-SAME: ) #[[ATTR0]] {
561 ; NPGSO-NEXT: [[ENTRY:.*]]:
562 ; NPGSO-NEXT: br label %[[LOOP:.*]]
563 ; NPGSO: [[LOOP]]:
564 ; NPGSO-NEXT: [[PIV:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[PIVPLUS1:%.*]], %[[LOOP]] ]
565 ; NPGSO-NEXT: [[FOR:%.*]] = phi i32 [ 5, %[[ENTRY]] ], [ [[PIVPLUS1]], %[[LOOP]] ]
566 ; NPGSO-NEXT: [[PIVPLUS1]] = add nuw nsw i32 [[PIV]], 1
567 ; NPGSO-NEXT: [[COND:%.*]] = icmp ult i32 [[PIV]], 510
568 ; NPGSO-NEXT: br i1 [[COND]], label %[[LOOP]], label %[[EXIT:.*]]
569 ; NPGSO: [[EXIT]]:
570 ; NPGSO-NEXT: [[FOR_LCSSA:%.*]] = phi i32 [ [[FOR]], %[[LOOP]] ]
571 ; NPGSO-NEXT: ret i32 [[FOR_LCSSA]]
572 ;
573 entry:
574 br label %loop
576 loop:
577 %piv = phi i32 [ 0, %entry ], [ %pivPlus1, %loop ]
578 %for = phi i32 [ 5, %entry ], [ %pivPlus1, %loop ]
579 %pivPlus1 = add nuw nsw i32 %piv, 1
580 %cond = icmp ult i32 %piv, 510
581 br i1 %cond, label %loop, label %exit
583 exit:
584 ret i32 %for
585 }
587 define i32 @pr45526_pgso() !prof !14 {
588 ;
589 ; CHECK-LABEL: define i32 @pr45526_pgso(
590 ; CHECK-SAME: ) !prof [[PROF14]] {
591 ; CHECK-NEXT: [[ENTRY:.*]]:
592 ; CHECK-NEXT: br label %[[LOOP:.*]]
593 ; CHECK: [[LOOP]]:
594 ; CHECK-NEXT: [[PIV:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[PIVPLUS1:%.*]], %[[LOOP]] ]
595 ; CHECK-NEXT: [[FOR:%.*]] = phi i32 [ 5, %[[ENTRY]] ], [ [[PIVPLUS1]], %[[LOOP]] ]
596 ; CHECK-NEXT: [[PIVPLUS1]] = add nuw nsw i32 [[PIV]], 1
597 ; CHECK-NEXT: [[COND:%.*]] = icmp ult i32 [[PIV]], 510
598 ; CHECK-NEXT: br i1 [[COND]], label %[[LOOP]], label %[[EXIT:.*]]
599 ; CHECK: [[EXIT]]:
600 ; CHECK-NEXT: [[FOR_LCSSA:%.*]] = phi i32 [ [[FOR]], %[[LOOP]] ]
601 ; CHECK-NEXT: ret i32 [[FOR_LCSSA]]
602 ;
603 ; PGSO-LABEL: define i32 @pr45526_pgso(
604 ; PGSO-SAME: ) !prof [[PROF14]] {
605 ; PGSO-NEXT: [[ENTRY:.*]]:
606 ; PGSO-NEXT: br label %[[LOOP:.*]]
607 ; PGSO: [[LOOP]]:
608 ; PGSO-NEXT: [[PIV:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[PIVPLUS1:%.*]], %[[LOOP]] ]
609 ; PGSO-NEXT: [[FOR:%.*]] = phi i32 [ 5, %[[ENTRY]] ], [ [[PIVPLUS1]], %[[LOOP]] ]
610 ; PGSO-NEXT: [[PIVPLUS1]] = add nuw nsw i32 [[PIV]], 1
611 ; PGSO-NEXT: [[COND:%.*]] = icmp ult i32 [[PIV]], 510
612 ; PGSO-NEXT: br i1 [[COND]], label %[[LOOP]], label %[[EXIT:.*]]
613 ; PGSO: [[EXIT]]:
614 ; PGSO-NEXT: [[FOR_LCSSA:%.*]] = phi i32 [ [[FOR]], %[[LOOP]] ]
615 ; PGSO-NEXT: ret i32 [[FOR_LCSSA]]
616 ;
617 ; NPGSO-LABEL: define i32 @pr45526_pgso(
618 ; NPGSO-SAME: ) !prof [[PROF14]] {
619 ; NPGSO-NEXT: [[ENTRY:.*]]:
620 ; NPGSO-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
621 ; NPGSO: [[VECTOR_PH]]:
622 ; NPGSO-NEXT: br label %[[VECTOR_BODY:.*]]
623 ; NPGSO: [[VECTOR_BODY]]:
624 ; NPGSO-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
625 ; NPGSO-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
626 ; NPGSO-NEXT: [[VECTOR_RECUR:%.*]] = phi <4 x i32> [ <i32 poison, i32 poison, i32 poison, i32 5>, %[[VECTOR_PH]] ], [ [[TMP0:%.*]], %[[VECTOR_BODY]] ]
627 ; NPGSO-NEXT: [[TMP0]] = add nuw nsw <4 x i32> [[VEC_IND]], splat (i32 1)
628 ; NPGSO-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
629 ; NPGSO-NEXT: [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], splat (i32 4)
630 ; NPGSO-NEXT: [[TMP1:%.*]] = icmp eq i32 [[INDEX_NEXT]], 508
631 ; NPGSO-NEXT: br i1 [[TMP1]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]]
632 ; NPGSO: [[MIDDLE_BLOCK]]:
633 ; NPGSO-NEXT: [[VECTOR_RECUR_EXTRACT_FOR_PHI:%.*]] = extractelement <4 x i32> [[TMP0]], i32 2
634 ; NPGSO-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <4 x i32> [[TMP0]], i32 3
635 ; NPGSO-NEXT: br i1 false, label %[[EXIT:.*]], label %[[SCALAR_PH]]
636 ; NPGSO: [[SCALAR_PH]]:
637 ; NPGSO-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 508, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
638 ; NPGSO-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ [[VECTOR_RECUR_EXTRACT]], %[[MIDDLE_BLOCK]] ], [ 5, %[[ENTRY]] ]
639 ; NPGSO-NEXT: br label %[[LOOP:.*]]
640 ; NPGSO: [[LOOP]]:
641 ; NPGSO-NEXT: [[PIV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[PIVPLUS1:%.*]], %[[LOOP]] ]
642 ; NPGSO-NEXT: [[FOR:%.*]] = phi i32 [ [[SCALAR_RECUR_INIT]], %[[SCALAR_PH]] ], [ [[PIVPLUS1]], %[[LOOP]] ]
643 ; NPGSO-NEXT: [[PIVPLUS1]] = add nuw nsw i32 [[PIV]], 1
644 ; NPGSO-NEXT: [[COND:%.*]] = icmp ult i32 [[PIV]], 510
645 ; NPGSO-NEXT: br i1 [[COND]], label %[[LOOP]], label %[[EXIT]], !llvm.loop [[LOOP24:![0-9]+]]
646 ; NPGSO: [[EXIT]]:
647 ; NPGSO-NEXT: [[FOR_LCSSA:%.*]] = phi i32 [ [[FOR]], %[[LOOP]] ], [ [[VECTOR_RECUR_EXTRACT_FOR_PHI]], %[[MIDDLE_BLOCK]] ]
648 ; NPGSO-NEXT: ret i32 [[FOR_LCSSA]]
649 ;
650 entry:
651 br label %loop
653 loop:
654 %piv = phi i32 [ 0, %entry ], [ %pivPlus1, %loop ]
655 %for = phi i32 [ 5, %entry ], [ %pivPlus1, %loop ]
656 %pivPlus1 = add nuw nsw i32 %piv, 1
657 %cond = icmp ult i32 %piv, 510
658 br i1 %cond, label %loop, label %exit
660 exit:
661 ret i32 %for
662 }
664 ; PR46228: Vectorize w/o versioning for unit stride under optsize and enabled
665 ; vectorization.
667 ; NOTE: Some assertions have been autogenerated by utils/update_test_checks.py
668 define void @stride1(ptr noalias %B, i32 %BStride) optsize {
669 ; CHECK-LABEL: define void @stride1(
670 ; CHECK-SAME: ptr noalias [[B:%.*]], i32 [[BSTRIDE:%.*]]) #[[ATTR0]] {
671 ; CHECK-NEXT: [[ENTRY:.*]]:
672 ; CHECK-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
673 ; CHECK: [[VECTOR_PH]]:
674 ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[BSTRIDE]], i64 0
675 ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
676 ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
677 ; CHECK: [[VECTOR_BODY]]:
678 ; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_STORE_CONTINUE2:.*]] ]
679 ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_STORE_CONTINUE2]] ]
680 ; CHECK-NEXT: [[TMP1:%.*]] = icmp ule <2 x i32> [[VEC_IND]], splat (i32 1024)
681 ; CHECK-NEXT: [[TMP0:%.*]] = mul nsw <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
682 ; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i1> [[TMP1]], i32 0
683 ; CHECK-NEXT: br i1 [[TMP2]], label %[[PRED_STORE_IF:.*]], label %[[PRED_STORE_CONTINUE:.*]]
684 ; CHECK: [[PRED_STORE_IF]]:
685 ; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[TMP0]], i32 0
686 ; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP3]]
687 ; CHECK-NEXT: store i16 42, ptr [[TMP4]], align 4
688 ; CHECK-NEXT: br label %[[PRED_STORE_CONTINUE]]
689 ; CHECK: [[PRED_STORE_CONTINUE]]:
690 ; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i1> [[TMP1]], i32 1
691 ; CHECK-NEXT: br i1 [[TMP5]], label %[[PRED_STORE_IF1:.*]], label %[[PRED_STORE_CONTINUE2]]
692 ; CHECK: [[PRED_STORE_IF1]]:
693 ; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i32> [[TMP0]], i32 1
694 ; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP6]]
695 ; CHECK-NEXT: store i16 42, ptr [[TMP7]], align 4
696 ; CHECK-NEXT: br label %[[PRED_STORE_CONTINUE2]]
697 ; CHECK: [[PRED_STORE_CONTINUE2]]:
698 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
699 ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2)
700 ; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1026
701 ; CHECK-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP21:![0-9]+]]
702 ; CHECK: [[MIDDLE_BLOCK]]:
703 ; CHECK-NEXT: br i1 true, label %[[FOR_END:.*]], label %[[SCALAR_PH]]
704 ; CHECK: [[SCALAR_PH]]:
705 ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 1026, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
706 ; CHECK-NEXT: br label %[[FOR_BODY:.*]]
707 ; CHECK: [[FOR_BODY]]:
708 ; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], %[[FOR_BODY]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]
709 ; CHECK-NEXT: [[MULB:%.*]] = mul nsw i32 [[IV]], [[BSTRIDE]]
710 ; CHECK-NEXT: [[GEPOFB:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[MULB]]
711 ; CHECK-NEXT: store i16 42, ptr [[GEPOFB]], align 4
712 ; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
713 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[IV_NEXT]], 1025
714 ; CHECK-NEXT: br i1 [[EXITCOND]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]]
715 ; CHECK: [[FOR_END]]:
716 ; CHECK-NEXT: ret void
717 ;
718 ; PGSO-LABEL: define void @stride1(
719 ; PGSO-SAME: ptr noalias [[B:%.*]], i32 [[BSTRIDE:%.*]]) #[[ATTR0]] {
720 ; PGSO-NEXT: [[ENTRY:.*]]:
721 ; PGSO-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
722 ; PGSO: [[VECTOR_PH]]:
723 ; PGSO-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[BSTRIDE]], i64 0
724 ; PGSO-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
725 ; PGSO-NEXT: br label %[[VECTOR_BODY:.*]]
726 ; PGSO: [[VECTOR_BODY]]:
727 ; PGSO-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_STORE_CONTINUE2:.*]] ]
728 ; PGSO-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_STORE_CONTINUE2]] ]
729 ; PGSO-NEXT: [[TMP0:%.*]] = icmp ule <2 x i32> [[VEC_IND]], splat (i32 1024)
730 ; PGSO-NEXT: [[TMP1:%.*]] = mul nsw <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
731 ; PGSO-NEXT: [[TMP2:%.*]] = extractelement <2 x i1> [[TMP0]], i32 0
732 ; PGSO-NEXT: br i1 [[TMP2]], label %[[PRED_STORE_IF:.*]], label %[[PRED_STORE_CONTINUE:.*]]
733 ; PGSO: [[PRED_STORE_IF]]:
734 ; PGSO-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[TMP1]], i32 0
735 ; PGSO-NEXT: [[TMP4:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP3]]
736 ; PGSO-NEXT: store i16 42, ptr [[TMP4]], align 4
737 ; PGSO-NEXT: br label %[[PRED_STORE_CONTINUE]]
738 ; PGSO: [[PRED_STORE_CONTINUE]]:
739 ; PGSO-NEXT: [[TMP5:%.*]] = extractelement <2 x i1> [[TMP0]], i32 1
740 ; PGSO-NEXT: br i1 [[TMP5]], label %[[PRED_STORE_IF1:.*]], label %[[PRED_STORE_CONTINUE2]]
741 ; PGSO: [[PRED_STORE_IF1]]:
742 ; PGSO-NEXT: [[TMP6:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
743 ; PGSO-NEXT: [[TMP7:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP6]]
744 ; PGSO-NEXT: store i16 42, ptr [[TMP7]], align 4
745 ; PGSO-NEXT: br label %[[PRED_STORE_CONTINUE2]]
746 ; PGSO: [[PRED_STORE_CONTINUE2]]:
747 ; PGSO-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
748 ; PGSO-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2)
749 ; PGSO-NEXT: [[TMP8:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1026
750 ; PGSO-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP21:![0-9]+]]
751 ; PGSO: [[MIDDLE_BLOCK]]:
752 ; PGSO-NEXT: br i1 true, label %[[FOR_END:.*]], label %[[SCALAR_PH]]
753 ; PGSO: [[SCALAR_PH]]:
754 ; PGSO-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 1026, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
755 ; PGSO-NEXT: br label %[[FOR_BODY:.*]]
756 ; PGSO: [[FOR_BODY]]:
757 ; PGSO-NEXT: [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], %[[FOR_BODY]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]
758 ; PGSO-NEXT: [[MULB:%.*]] = mul nsw i32 [[IV]], [[BSTRIDE]]
759 ; PGSO-NEXT: [[GEPOFB:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[MULB]]
760 ; PGSO-NEXT: store i16 42, ptr [[GEPOFB]], align 4
761 ; PGSO-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
762 ; PGSO-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[IV_NEXT]], 1025
763 ; PGSO-NEXT: br i1 [[EXITCOND]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]]
764 ; PGSO: [[FOR_END]]:
765 ; PGSO-NEXT: ret void
766 ;
767 ; NPGSO-LABEL: define void @stride1(
768 ; NPGSO-SAME: ptr noalias [[B:%.*]], i32 [[BSTRIDE:%.*]]) #[[ATTR0]] {
769 ; NPGSO-NEXT: [[ENTRY:.*]]:
770 ; NPGSO-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
771 ; NPGSO: [[VECTOR_PH]]:
772 ; NPGSO-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[BSTRIDE]], i64 0
773 ; NPGSO-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
774 ; NPGSO-NEXT: br label %[[VECTOR_BODY:.*]]
775 ; NPGSO: [[VECTOR_BODY]]:
776 ; NPGSO-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_STORE_CONTINUE2:.*]] ]
777 ; NPGSO-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_STORE_CONTINUE2]] ]
778 ; NPGSO-NEXT: [[TMP0:%.*]] = icmp ule <2 x i32> [[VEC_IND]], splat (i32 1024)
779 ; NPGSO-NEXT: [[TMP1:%.*]] = mul nsw <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
780 ; NPGSO-NEXT: [[TMP2:%.*]] = extractelement <2 x i1> [[TMP0]], i32 0
781 ; NPGSO-NEXT: br i1 [[TMP2]], label %[[PRED_STORE_IF:.*]], label %[[PRED_STORE_CONTINUE:.*]]
782 ; NPGSO: [[PRED_STORE_IF]]:
783 ; NPGSO-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[TMP1]], i32 0
784 ; NPGSO-NEXT: [[TMP4:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP3]]
785 ; NPGSO-NEXT: store i16 42, ptr [[TMP4]], align 4
786 ; NPGSO-NEXT: br label %[[PRED_STORE_CONTINUE]]
787 ; NPGSO: [[PRED_STORE_CONTINUE]]:
788 ; NPGSO-NEXT: [[TMP5:%.*]] = extractelement <2 x i1> [[TMP0]], i32 1
789 ; NPGSO-NEXT: br i1 [[TMP5]], label %[[PRED_STORE_IF1:.*]], label %[[PRED_STORE_CONTINUE2]]
790 ; NPGSO: [[PRED_STORE_IF1]]:
791 ; NPGSO-NEXT: [[TMP6:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
792 ; NPGSO-NEXT: [[TMP7:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP6]]
793 ; NPGSO-NEXT: store i16 42, ptr [[TMP7]], align 4
794 ; NPGSO-NEXT: br label %[[PRED_STORE_CONTINUE2]]
795 ; NPGSO: [[PRED_STORE_CONTINUE2]]:
796 ; NPGSO-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
797 ; NPGSO-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2)
798 ; NPGSO-NEXT: [[TMP8:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1026
799 ; NPGSO-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP25:![0-9]+]]
800 ; NPGSO: [[MIDDLE_BLOCK]]:
801 ; NPGSO-NEXT: br i1 true, label %[[FOR_END:.*]], label %[[SCALAR_PH]]
802 ; NPGSO: [[SCALAR_PH]]:
803 ; NPGSO-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 1026, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
804 ; NPGSO-NEXT: br label %[[FOR_BODY:.*]]
805 ; NPGSO: [[FOR_BODY]]:
806 ; NPGSO-NEXT: [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], %[[FOR_BODY]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]
807 ; NPGSO-NEXT: [[MULB:%.*]] = mul nsw i32 [[IV]], [[BSTRIDE]]
808 ; NPGSO-NEXT: [[GEPOFB:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[MULB]]
809 ; NPGSO-NEXT: store i16 42, ptr [[GEPOFB]], align 4
810 ; NPGSO-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
811 ; NPGSO-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[IV_NEXT]], 1025
812 ; NPGSO-NEXT: br i1 [[EXITCOND]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]]
813 ; NPGSO: [[FOR_END]]:
814 ; NPGSO-NEXT: ret void
815 ;
817 entry:
818 br label %for.body
820 for.body:
821 %iv = phi i32 [ %iv.next, %for.body ], [ 0, %entry ]
822 %mulB = mul nsw i32 %iv, %BStride
823 %gepOfB = getelementptr inbounds i16, ptr %B, i32 %mulB
824 store i16 42, ptr %gepOfB, align 4
825 %iv.next = add nuw nsw i32 %iv, 1
826 %exitcond = icmp eq i32 %iv.next, 1025
827 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !15
829 for.end:
830 ret void
831 }
833 ; Vectorize with versioning for unit stride for PGSO and enabled vectorization.
834 ;
835 define void @stride1_pgso(ptr noalias %B, i32 %BStride) !prof !14 {
836 ; CHECK-LABEL: define void @stride1_pgso(
837 ; CHECK-SAME: ptr noalias [[B:%.*]], i32 [[BSTRIDE:%.*]]) !prof [[PROF14]] {
838 ; CHECK-NEXT: [[ENTRY:.*]]:
839 ; CHECK-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
840 ; CHECK: [[VECTOR_SCEVCHECK]]:
841 ; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[BSTRIDE]], 1
842 ; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
843 ; CHECK: [[VECTOR_PH]]:
844 ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
845 ; CHECK: [[VECTOR_BODY]]:
846 ; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
847 ; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[INDEX]], 0
848 ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP0]]
849 ; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i16, ptr [[TMP1]], i32 0
850 ; CHECK-NEXT: store <2 x i16> splat (i16 42), ptr [[TMP2]], align 4
851 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
852 ; CHECK-NEXT: [[TMP3:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1024
853 ; CHECK-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]]
854 ; CHECK: [[MIDDLE_BLOCK]]:
855 ; CHECK-NEXT: br i1 false, label %[[FOR_END:.*]], label %[[SCALAR_PH]]
856 ; CHECK: [[SCALAR_PH]]:
857 ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 1024, %[[MIDDLE_BLOCK]] ], [ 0, %[[VECTOR_SCEVCHECK]] ], [ 0, %[[ENTRY]] ]
858 ; CHECK-NEXT: br label %[[FOR_BODY:.*]]
859 ; CHECK: [[FOR_BODY]]:
860 ; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], %[[FOR_BODY]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]
861 ; CHECK-NEXT: [[MULB:%.*]] = mul nsw i32 [[IV]], [[BSTRIDE]]
862 ; CHECK-NEXT: [[GEPOFB:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[MULB]]
863 ; CHECK-NEXT: store i16 42, ptr [[GEPOFB]], align 4
864 ; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
865 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[IV_NEXT]], 1025
866 ; CHECK-NEXT: br i1 [[EXITCOND]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]]
867 ; CHECK: [[FOR_END]]:
868 ; CHECK-NEXT: ret void
869 ;
870 ; PGSO-LABEL: define void @stride1_pgso(
871 ; PGSO-SAME: ptr noalias [[B:%.*]], i32 [[BSTRIDE:%.*]]) !prof [[PROF14]] {
872 ; PGSO-NEXT: [[ENTRY:.*]]:
873 ; PGSO-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
874 ; PGSO: [[VECTOR_SCEVCHECK]]:
875 ; PGSO-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[BSTRIDE]], 1
876 ; PGSO-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
877 ; PGSO: [[VECTOR_PH]]:
878 ; PGSO-NEXT: br label %[[VECTOR_BODY:.*]]
879 ; PGSO: [[VECTOR_BODY]]:
880 ; PGSO-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
881 ; PGSO-NEXT: [[TMP0:%.*]] = add i32 [[INDEX]], 0
882 ; PGSO-NEXT: [[TMP1:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP0]]
883 ; PGSO-NEXT: [[TMP2:%.*]] = getelementptr inbounds i16, ptr [[TMP1]], i32 0
884 ; PGSO-NEXT: store <2 x i16> splat (i16 42), ptr [[TMP2]], align 4
885 ; PGSO-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
886 ; PGSO-NEXT: [[TMP3:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1024
887 ; PGSO-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]]
888 ; PGSO: [[MIDDLE_BLOCK]]:
889 ; PGSO-NEXT: br i1 false, label %[[FOR_END:.*]], label %[[SCALAR_PH]]
890 ; PGSO: [[SCALAR_PH]]:
891 ; PGSO-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 1024, %[[MIDDLE_BLOCK]] ], [ 0, %[[VECTOR_SCEVCHECK]] ], [ 0, %[[ENTRY]] ]
892 ; PGSO-NEXT: br label %[[FOR_BODY:.*]]
893 ; PGSO: [[FOR_BODY]]:
894 ; PGSO-NEXT: [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], %[[FOR_BODY]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]
895 ; PGSO-NEXT: [[MULB:%.*]] = mul nsw i32 [[IV]], [[BSTRIDE]]
896 ; PGSO-NEXT: [[GEPOFB:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[MULB]]
897 ; PGSO-NEXT: store i16 42, ptr [[GEPOFB]], align 4
898 ; PGSO-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
899 ; PGSO-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[IV_NEXT]], 1025
900 ; PGSO-NEXT: br i1 [[EXITCOND]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]]
901 ; PGSO: [[FOR_END]]:
902 ; PGSO-NEXT: ret void
903 ;
904 ; NPGSO-LABEL: define void @stride1_pgso(
905 ; NPGSO-SAME: ptr noalias [[B:%.*]], i32 [[BSTRIDE:%.*]]) !prof [[PROF14]] {
906 ; NPGSO-NEXT: [[ENTRY:.*]]:
907 ; NPGSO-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
908 ; NPGSO: [[VECTOR_SCEVCHECK]]:
909 ; NPGSO-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[BSTRIDE]], 1
910 ; NPGSO-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
911 ; NPGSO: [[VECTOR_PH]]:
912 ; NPGSO-NEXT: br label %[[VECTOR_BODY:.*]]
913 ; NPGSO: [[VECTOR_BODY]]:
914 ; NPGSO-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
915 ; NPGSO-NEXT: [[TMP0:%.*]] = add i32 [[INDEX]], 0
916 ; NPGSO-NEXT: [[TMP1:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP0]]
917 ; NPGSO-NEXT: [[TMP2:%.*]] = getelementptr inbounds i16, ptr [[TMP1]], i32 0
918 ; NPGSO-NEXT: store <2 x i16> splat (i16 42), ptr [[TMP2]], align 4
919 ; NPGSO-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
920 ; NPGSO-NEXT: [[TMP3:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1024
921 ; NPGSO-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP27:![0-9]+]]
922 ; NPGSO: [[MIDDLE_BLOCK]]:
923 ; NPGSO-NEXT: br i1 false, label %[[FOR_END:.*]], label %[[SCALAR_PH]]
924 ; NPGSO: [[SCALAR_PH]]:
925 ; NPGSO-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 1024, %[[MIDDLE_BLOCK]] ], [ 0, %[[VECTOR_SCEVCHECK]] ], [ 0, %[[ENTRY]] ]
926 ; NPGSO-NEXT: br label %[[FOR_BODY:.*]]
927 ; NPGSO: [[FOR_BODY]]:
928 ; NPGSO-NEXT: [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], %[[FOR_BODY]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]
929 ; NPGSO-NEXT: [[MULB:%.*]] = mul nsw i32 [[IV]], [[BSTRIDE]]
930 ; NPGSO-NEXT: [[GEPOFB:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[MULB]]
931 ; NPGSO-NEXT: store i16 42, ptr [[GEPOFB]], align 4
932 ; NPGSO-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
933 ; NPGSO-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[IV_NEXT]], 1025
934 ; NPGSO-NEXT: br i1 [[EXITCOND]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]]
935 ; NPGSO: [[FOR_END]]:
936 ; NPGSO-NEXT: ret void
937 ;
939 entry:
940 br label %for.body
942 for.body:
943 %iv = phi i32 [ %iv.next, %for.body ], [ 0, %entry ]
944 %mulB = mul nsw i32 %iv, %BStride
945 %gepOfB = getelementptr inbounds i16, ptr %B, i32 %mulB
946 store i16 42, ptr %gepOfB, align 4
947 %iv.next = add nuw nsw i32 %iv, 1
948 %exitcond = icmp eq i32 %iv.next, 1025
949 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !15
951 for.end:
952 ret void
953 }
955 ; PR46652: Check that the need for stride==1 check prevents vectorizing a loop
956 ; having tiny trip count, when compiling w/o -Os/-Oz.
958 @g = external global [1 x i16], align 1
960 define void @pr46652(i16 %stride) {
961 ; CHECK-LABEL: define void @pr46652(
962 ; CHECK-SAME: i16 [[STRIDE:%.*]]) {
963 ; CHECK-NEXT: [[ENTRY:.*]]:
964 ; CHECK-NEXT: br label %[[FOR_BODY:.*]]
965 ; CHECK: [[FOR_BODY]]:
966 ; CHECK-NEXT: [[L1_02:%.*]] = phi i16 [ 1, %[[ENTRY]] ], [ [[INC9:%.*]], %[[FOR_BODY]] ]
967 ; CHECK-NEXT: [[MUL:%.*]] = mul nsw i16 [[L1_02]], [[STRIDE]]
968 ; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds [1 x i16], ptr @g, i16 0, i16 [[MUL]]
969 ; CHECK-NEXT: [[TMP0:%.*]] = load i16, ptr [[ARRAYIDX6]], align 1
970 ; CHECK-NEXT: [[INC9]] = add nuw nsw i16 [[L1_02]], 1
971 ; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i16 [[INC9]], 16
972 ; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_END:.*]], label %[[FOR_BODY]]
973 ; CHECK: [[FOR_END]]:
974 ; CHECK-NEXT: ret void
975 ;
976 ; PGSO-LABEL: define void @pr46652(
977 ; PGSO-SAME: i16 [[STRIDE:%.*]]) {
978 ; PGSO-NEXT: [[ENTRY:.*]]:
979 ; PGSO-NEXT: br label %[[FOR_BODY:.*]]
980 ; PGSO: [[FOR_BODY]]:
981 ; PGSO-NEXT: [[L1_02:%.*]] = phi i16 [ 1, %[[ENTRY]] ], [ [[INC9:%.*]], %[[FOR_BODY]] ]
982 ; PGSO-NEXT: [[MUL:%.*]] = mul nsw i16 [[L1_02]], [[STRIDE]]
983 ; PGSO-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds [1 x i16], ptr @g, i16 0, i16 [[MUL]]
984 ; PGSO-NEXT: [[TMP0:%.*]] = load i16, ptr [[ARRAYIDX6]], align 1
985 ; PGSO-NEXT: [[INC9]] = add nuw nsw i16 [[L1_02]], 1
986 ; PGSO-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i16 [[INC9]], 16
987 ; PGSO-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_END:.*]], label %[[FOR_BODY]]
988 ; PGSO: [[FOR_END]]:
989 ; PGSO-NEXT: ret void
990 ;
991 ; NPGSO-LABEL: define void @pr46652(
992 ; NPGSO-SAME: i16 [[STRIDE:%.*]]) {
993 ; NPGSO-NEXT: [[ENTRY:.*]]:
994 ; NPGSO-NEXT: br label %[[FOR_BODY:.*]]
995 ; NPGSO: [[FOR_BODY]]:
996 ; NPGSO-NEXT: [[L1_02:%.*]] = phi i16 [ 1, %[[ENTRY]] ], [ [[INC9:%.*]], %[[FOR_BODY]] ]
997 ; NPGSO-NEXT: [[MUL:%.*]] = mul nsw i16 [[L1_02]], [[STRIDE]]
998 ; NPGSO-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds [1 x i16], ptr @g, i16 0, i16 [[MUL]]
999 ; NPGSO-NEXT: [[TMP0:%.*]] = load i16, ptr [[ARRAYIDX6]], align 1
1000 ; NPGSO-NEXT: [[INC9]] = add nuw nsw i16 [[L1_02]], 1
1001 ; NPGSO-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i16 [[INC9]], 16
1002 ; NPGSO-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_END:.*]], label %[[FOR_BODY]]
1003 ; NPGSO: [[FOR_END]]:
1004 ; NPGSO-NEXT: ret void
1005 ;
1006 entry:
1007 br label %for.body
1009 for.body: ; preds = %for.body, %entry
1010 %l1.02 = phi i16 [ 1, %entry ], [ %inc9, %for.body ]
1011 %mul = mul nsw i16 %l1.02, %stride
1012 %arrayidx6 = getelementptr inbounds [1 x i16], ptr @g, i16 0, i16 %mul
1013 %0 = load i16, ptr %arrayidx6, align 1
1014 %inc9 = add nuw nsw i16 %l1.02, 1
1015 %exitcond.not = icmp eq i16 %inc9, 16
1016 br i1 %exitcond.not, label %for.end, label %for.body
1018 for.end: ; preds = %for.body
1019 ret void
1020 }
1022 ; Make sure we do not crash while building the VPlan for the loop with the
1023 ; select below.
1024 define i32 @PR48142(ptr %ptr.start, ptr %ptr.end) optsize {
1025 ; CHECK-LABEL: define i32 @PR48142(
1026 ; CHECK-SAME: ptr [[PTR_START:%.*]], ptr [[PTR_END:%.*]]) #[[ATTR0]] {
1027 ; CHECK-NEXT: [[ENTRY:.*]]:
1028 ; CHECK-NEXT: br label %[[FOR_BODY:.*]]
1029 ; CHECK: [[FOR_BODY]]:
1030 ; CHECK-NEXT: [[I_014:%.*]] = phi i32 [ 20, %[[ENTRY]] ], [ [[COND:%.*]], %[[FOR_BODY]] ]
1031 ; CHECK-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[PTR_START]], %[[ENTRY]] ], [ [[PTR_NEXT:%.*]], %[[FOR_BODY]] ]
1032 ; CHECK-NEXT: [[CMP4:%.*]] = icmp slt i32 [[I_014]], 99
1033 ; CHECK-NEXT: [[COND]] = select i1 [[CMP4]], i32 99, i32 [[I_014]]
1034 ; CHECK-NEXT: store i32 0, ptr [[PTR_IV]], align 4
1035 ; CHECK-NEXT: [[PTR_NEXT]] = getelementptr inbounds i32, ptr [[PTR_IV]], i64 1
1036 ; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq ptr [[PTR_NEXT]], [[PTR_END]]
1037 ; CHECK-NEXT: br i1 [[CMP_NOT]], label %[[EXIT:.*]], label %[[FOR_BODY]]
1038 ; CHECK: [[EXIT]]:
1039 ; CHECK-NEXT: [[RES:%.*]] = phi i32 [ [[COND]], %[[FOR_BODY]] ]
1040 ; CHECK-NEXT: ret i32 [[RES]]
1041 ;
1042 ; PGSO-LABEL: define i32 @PR48142(
1043 ; PGSO-SAME: ptr [[PTR_START:%.*]], ptr [[PTR_END:%.*]]) #[[ATTR0]] {
1044 ; PGSO-NEXT: [[ENTRY:.*]]:
1045 ; PGSO-NEXT: br label %[[FOR_BODY:.*]]
1046 ; PGSO: [[FOR_BODY]]:
1047 ; PGSO-NEXT: [[I_014:%.*]] = phi i32 [ 20, %[[ENTRY]] ], [ [[COND:%.*]], %[[FOR_BODY]] ]
1048 ; PGSO-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[PTR_START]], %[[ENTRY]] ], [ [[PTR_NEXT:%.*]], %[[FOR_BODY]] ]
1049 ; PGSO-NEXT: [[CMP4:%.*]] = icmp slt i32 [[I_014]], 99
1050 ; PGSO-NEXT: [[COND]] = select i1 [[CMP4]], i32 99, i32 [[I_014]]
1051 ; PGSO-NEXT: store i32 0, ptr [[PTR_IV]], align 4
1052 ; PGSO-NEXT: [[PTR_NEXT]] = getelementptr inbounds i32, ptr [[PTR_IV]], i64 1
1053 ; PGSO-NEXT: [[CMP_NOT:%.*]] = icmp eq ptr [[PTR_NEXT]], [[PTR_END]]
1054 ; PGSO-NEXT: br i1 [[CMP_NOT]], label %[[EXIT:.*]], label %[[FOR_BODY]]
1055 ; PGSO: [[EXIT]]:
1056 ; PGSO-NEXT: [[RES:%.*]] = phi i32 [ [[COND]], %[[FOR_BODY]] ]
1057 ; PGSO-NEXT: ret i32 [[RES]]
1058 ;
1059 ; NPGSO-LABEL: define i32 @PR48142(
1060 ; NPGSO-SAME: ptr [[PTR_START:%.*]], ptr [[PTR_END:%.*]]) #[[ATTR0]] {
1061 ; NPGSO-NEXT: [[ENTRY:.*]]:
1062 ; NPGSO-NEXT: br label %[[FOR_BODY:.*]]
1063 ; NPGSO: [[FOR_BODY]]:
1064 ; NPGSO-NEXT: [[I_014:%.*]] = phi i32 [ 20, %[[ENTRY]] ], [ [[COND:%.*]], %[[FOR_BODY]] ]
1065 ; NPGSO-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[PTR_START]], %[[ENTRY]] ], [ [[PTR_NEXT:%.*]], %[[FOR_BODY]] ]
1066 ; NPGSO-NEXT: [[CMP4:%.*]] = icmp slt i32 [[I_014]], 99
1067 ; NPGSO-NEXT: [[COND]] = select i1 [[CMP4]], i32 99, i32 [[I_014]]
1068 ; NPGSO-NEXT: store i32 0, ptr [[PTR_IV]], align 4
1069 ; NPGSO-NEXT: [[PTR_NEXT]] = getelementptr inbounds i32, ptr [[PTR_IV]], i64 1
1070 ; NPGSO-NEXT: [[CMP_NOT:%.*]] = icmp eq ptr [[PTR_NEXT]], [[PTR_END]]
1071 ; NPGSO-NEXT: br i1 [[CMP_NOT]], label %[[EXIT:.*]], label %[[FOR_BODY]]
1072 ; NPGSO: [[EXIT]]:
1073 ; NPGSO-NEXT: [[RES:%.*]] = phi i32 [ [[COND]], %[[FOR_BODY]] ]
1074 ; NPGSO-NEXT: ret i32 [[RES]]
1075 ;
1076 entry:
1077 br label %for.body
1079 for.body:
1080 %i.014 = phi i32 [ 20, %entry ], [ %cond, %for.body ]
1081 %ptr.iv = phi ptr [ %ptr.start, %entry ], [ %ptr.next, %for.body ]
1082 %cmp4 = icmp slt i32 %i.014, 99
1083 %cond = select i1 %cmp4, i32 99, i32 %i.014
1084 store i32 0, ptr %ptr.iv
1085 %ptr.next = getelementptr inbounds i32, ptr %ptr.iv, i64 1
1086 %cmp.not = icmp eq ptr %ptr.next, %ptr.end
1087 br i1 %cmp.not, label %exit, label %for.body
1089 exit:
1090 %res = phi i32 [ %cond, %for.body ]
1091 ret i32 %res
1092 }
1094 !llvm.module.flags = !{!0}
1095 !0 = !{i32 1, !"ProfileSummary", !1}
1096 !1 = !{!2, !3, !4, !5, !6, !7, !8, !9}
1097 !2 = !{!"ProfileFormat", !"InstrProf"}
1098 !3 = !{!"TotalCount", i64 10000}
1099 !4 = !{!"MaxCount", i64 10}
1100 !5 = !{!"MaxInternalCount", i64 1}
1101 !6 = !{!"MaxFunctionCount", i64 1000}
1102 !7 = !{!"NumCounts", i64 3}
1103 !8 = !{!"NumFunctions", i64 3}
1104 !9 = !{!"DetailedSummary", !10}
1105 !10 = !{!11, !12, !13}
1106 !11 = !{i32 10000, i64 100, i32 1}
1107 !12 = !{i32 999000, i64 100, i32 1}
1108 !13 = !{i32 999999, i64 1, i32 2}
1109 !14 = !{!"function_entry_count", i64 0}
1110 !15 = distinct !{!15, !16}
1111 !16 = !{!"llvm.loop.vectorize.enable", i1 true}
1112 ;.
1113 ; CHECK: [[PROF14]] = !{!"function_entry_count", i64 0}
1114 ; CHECK: [[LOOP15]] = distinct !{[[LOOP15]], [[META16:![0-9]+]], [[META17:![0-9]+]]}
1115 ; CHECK: [[META16]] = !{!"llvm.loop.isvectorized", i32 1}
1116 ; CHECK: [[META17]] = !{!"llvm.loop.unroll.runtime.disable"}
1117 ; CHECK: [[LOOP18]] = distinct !{[[LOOP18]], [[META17]], [[META16]]}
1118 ; CHECK: [[LOOP19]] = distinct !{[[LOOP19]], [[META16]], [[META17]]}
1119 ; CHECK: [[LOOP20]] = distinct !{[[LOOP20]], [[META17]], [[META16]]}
1120 ; CHECK: [[LOOP21]] = distinct !{[[LOOP21]], [[META16]], [[META17]]}
1121 ; CHECK: [[LOOP22]] = distinct !{[[LOOP22]], [[META17]], [[META16]]}
1122 ; CHECK: [[LOOP23]] = distinct !{[[LOOP23]], [[META16]], [[META17]]}
1123 ; CHECK: [[LOOP24]] = distinct !{[[LOOP24]], [[META16]]}
1124 ;.
1125 ; PGSO: [[PROF14]] = !{!"function_entry_count", i64 0}
1126 ; PGSO: [[LOOP15]] = distinct !{[[LOOP15]], [[META16:![0-9]+]], [[META17:![0-9]+]]}
1127 ; PGSO: [[META16]] = !{!"llvm.loop.isvectorized", i32 1}
1128 ; PGSO: [[META17]] = !{!"llvm.loop.unroll.runtime.disable"}
1129 ; PGSO: [[LOOP18]] = distinct !{[[LOOP18]], [[META17]], [[META16]]}
1130 ; PGSO: [[LOOP19]] = distinct !{[[LOOP19]], [[META16]], [[META17]]}
1131 ; PGSO: [[LOOP20]] = distinct !{[[LOOP20]], [[META17]], [[META16]]}
1132 ; PGSO: [[LOOP21]] = distinct !{[[LOOP21]], [[META16]], [[META17]]}
1133 ; PGSO: [[LOOP22]] = distinct !{[[LOOP22]], [[META17]], [[META16]]}
1134 ; PGSO: [[LOOP23]] = distinct !{[[LOOP23]], [[META16]], [[META17]]}
1135 ; PGSO: [[LOOP24]] = distinct !{[[LOOP24]], [[META16]]}
1136 ;.
1137 ; NPGSO: [[PROF14]] = !{!"function_entry_count", i64 0}
1138 ; NPGSO: [[LOOP15]] = distinct !{[[LOOP15]], [[META16:![0-9]+]], [[META17:![0-9]+]]}
1139 ; NPGSO: [[META16]] = !{!"llvm.loop.isvectorized", i32 1}
1140 ; NPGSO: [[META17]] = !{!"llvm.loop.unroll.runtime.disable"}
1141 ; NPGSO: [[LOOP18]] = distinct !{[[LOOP18]], [[META17]], [[META16]]}
1142 ; NPGSO: [[LOOP19]] = distinct !{[[LOOP19]], [[META16]], [[META17]]}
1143 ; NPGSO: [[LOOP20]] = distinct !{[[LOOP20]], [[META17]], [[META16]]}
1144 ; NPGSO: [[LOOP21]] = distinct !{[[LOOP21]], [[META16]], [[META17]]}
1145 ; NPGSO: [[LOOP22]] = distinct !{[[LOOP22]], [[META16]]}
1146 ; NPGSO: [[LOOP23]] = distinct !{[[LOOP23]], [[META16]], [[META17]]}
1147 ; NPGSO: [[LOOP24]] = distinct !{[[LOOP24]], [[META17]], [[META16]]}
1148 ; NPGSO: [[LOOP25]] = distinct !{[[LOOP25]], [[META16]], [[META17]]}
1149 ; NPGSO: [[LOOP26]] = distinct !{[[LOOP26]], [[META17]], [[META16]]}
1150 ; NPGSO: [[LOOP27]] = distinct !{[[LOOP27]], [[META16]], [[META17]]}
1151 ; NPGSO: [[LOOP28]] = distinct !{[[LOOP28]], [[META16]]}
1152 ;.