1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
|
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt %s -passes=loop-vectorize -hoist-runtime-checks=false -force-vector-width=4 -force-vector-interleave=1 -S | FileCheck %s
target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
define void @same_step_and_size(ptr %a, ptr %b, i64 %n) {
; CHECK-LABEL: @same_step_and_size(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[A2:%.*]] = ptrtoint ptr [[A:%.*]] to i64
; CHECK-NEXT: [[B1:%.*]] = ptrtoint ptr [[B:%.*]] to i64
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %scalar.ph, label %vector.memcheck
; CHECK: vector.memcheck:
; CHECK-NEXT: [[TMP0:%.*]] = sub i64 [[B1]], [[A2]]
; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP0]], 16
; CHECK-NEXT: br i1 [[DIFF_CHECK]], label %scalar.ph, label %vector.ph
;
entry:
br label %loop
loop:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
%gep.a = getelementptr inbounds i32, ptr %a, i64 %iv
%l = load i32, ptr %gep.a
%mul = mul nsw i32 %l, 3
%gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
store i32 %mul, ptr %gep.b
%iv.next = add nuw nsw i64 %iv, 1
%exitcond = icmp eq i64 %iv.next, %n
br i1 %exitcond, label %exit, label %loop
exit:
ret void
}
define void @same_step_and_size_no_dominance_between_accesses(ptr %a, ptr %b, i64 %n, i64 %x) {
; CHECK-LABEL: @same_step_and_size_no_dominance_between_accesses(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[B2:%.*]] = ptrtoint ptr [[B:%.*]] to i64
; CHECK-NEXT: [[A1:%.*]] = ptrtoint ptr [[A:%.*]] to i64
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %scalar.ph, label %vector.memcheck
; CHECK: vector.memcheck:
; CHECK-NEXT: [[TMP0:%.*]] = sub i64 [[A1]], [[B2]]
; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP0]], 16
; CHECK-NEXT: br i1 [[DIFF_CHECK]], label %scalar.ph, label %vector.ph
;
entry:
br label %loop
loop:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
%cmp = icmp ne i64 %iv, %x
br i1 %cmp, label %then, label %else
then:
%gep.a = getelementptr inbounds i32, ptr %a, i64 %iv
store i32 0, ptr %gep.a
br label %loop.latch
else:
%gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
store i32 10, ptr %gep.b
br label %loop.latch
loop.latch:
%iv.next = add nuw nsw i64 %iv, 1
%exitcond = icmp eq i64 %iv.next, %n
br i1 %exitcond, label %exit, label %loop
exit:
ret void
}
define void @different_steps_and_different_access_sizes(ptr %a, ptr %b, i64 %n) {
; CHECK-LABEL: @different_steps_and_different_access_sizes(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %scalar.ph, label %vector.memcheck
; CHECK: vector.memcheck:
; CHECK-NEXT: [[N_SHL_2:%.]] = shl i64 %n, 2
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr %b, i64 [[N_SHL_2]]
; CHECK-NEXT: [[N_SHL_1:%.]] = shl i64 %n, 1
; CHECK-NEXT: [[SCEVGEP4:%.*]] = getelementptr i8, ptr %a, i64 [[N_SHL_1]]
; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr %b, [[SCEVGEP4]]
; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr %a, [[SCEVGEP]]
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label %scalar.ph, label %vector.ph
;
entry:
br label %loop
loop:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
%gep.a = getelementptr inbounds i16, ptr %a, i64 %iv
%l = load i16, ptr %gep.a
%l.ext = sext i16 %l to i32
%mul = mul nsw i32 %l.ext, 3
%gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
store i32 %mul, ptr %gep.b
%iv.next = add nuw nsw i64 %iv, 1
%exitcond = icmp eq i64 %iv.next, %n
br i1 %exitcond, label %exit, label %loop
exit:
ret void
}
define void @steps_match_but_different_access_sizes_1(ptr %a, ptr %b, i64 %n) {
; CHECK-LABEL: @steps_match_but_different_access_sizes_1(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[A2:%.*]] = ptrtoint ptr [[A:%.*]] to i64
; CHECK-NEXT: [[B1:%.*]] = ptrtoint ptr [[B:%.*]] to i64
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %scalar.ph, label %vector.memcheck
; CHECK: vector.memcheck:
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[B1]], -2
; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[TMP0]], [[A2]]
; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP1]], 16
; CHECK-NEXT: br i1 [[DIFF_CHECK]], label %scalar.ph, label %vector.ph
;
entry:
br label %loop
loop:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
%gep.a = getelementptr inbounds [2 x i16], ptr %a, i64 %iv, i64 1
%l = load i16, ptr %gep.a
%l.ext = sext i16 %l to i32
%mul = mul nsw i32 %l.ext, 3
%gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
store i32 %mul, ptr %gep.b
%iv.next = add nuw nsw i64 %iv, 1
%exitcond = icmp eq i64 %iv.next, %n
br i1 %exitcond, label %exit, label %loop
exit:
ret void
}
; Same as @steps_match_but_different_access_sizes_1, but with source and sink
; accesses flipped.
define void @steps_match_but_different_access_sizes_2(ptr %a, ptr %b, i64 %n) {
; CHECK-LABEL: @steps_match_but_different_access_sizes_2(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[B2:%.*]] = ptrtoint ptr [[B:%.*]] to i64
; CHECK-NEXT: [[A1:%.*]] = ptrtoint ptr [[A:%.*]] to i64
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %scalar.ph, label %vector.memcheck
; CHECK: vector.memcheck:
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[A1]], 2
; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[TMP0]], [[B2]]
; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP1]], 16
; CHECK-NEXT: br i1 [[DIFF_CHECK]], label %scalar.ph, label %vector.ph
;
entry:
br label %loop
loop:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
%gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
%l = load i32, ptr %gep.b
%mul = mul nsw i32 %l, 3
%gep.a = getelementptr inbounds [2 x i16], ptr %a, i64 %iv, i64 1
%trunc = trunc i32 %mul to i16
store i16 %trunc, ptr %gep.a
%iv.next = add nuw nsw i64 %iv, 1
%exitcond = icmp eq i64 %iv.next, %n
br i1 %exitcond, label %exit, label %loop
exit:
ret void
}
; Full no-overlap checks are required instead of difference checks, as
; one of the add-recs used is invariant in the inner loop.
; Test case for PR57315.
define void @nested_loop_outer_iv_addrec_invariant_in_inner1(ptr %a, ptr %b, i64 %n) {
; CHECK-LABEL: @nested_loop_outer_iv_addrec_invariant_in_inner1(
; CHECK: entry:
; CHECK-NEXT: [[N_SHL_2:%.]] = shl i64 %n, 2
; CHECK-NEXT: [[B_GEP_UPPER:%.*]] = getelementptr i8, ptr %b, i64 [[N_SHL_2]]
; CHECK-NEXT: br label %outer
; CHECK: outer.header:
; CHECK: [[OUTER_IV_SHL_2:%.]] = shl i64 %outer.iv, 2
; CHECK-NEXT: [[A_GEP_UPPER:%.*]] = getelementptr i8, ptr %a, i64 [[OUTER_IV_SHL_2]]
; CHECK-NEXT: [[OUTER_IV_4:%.]] = add i64 [[OUTER_IV_SHL_2]], 4
; CHECK-NEXT: [[A_GEP_UPPER_4:%.*]] = getelementptr i8, ptr %a, i64 [[OUTER_IV_4]]
; CHECK: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %scalar.ph, label %vector.memcheck
; CHECK: vector.memcheck:
; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[A_GEP_UPPER]], [[B_GEP_UPPER]]
; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr %b, [[A_GEP_UPPER_4]]
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label %scalar.ph, label %vector.ph
;
entry:
br label %outer.header
outer.header:
%outer.iv = phi i64 [ %outer.iv.next, %outer.latch ], [ 0, %entry ]
%gep.a = getelementptr inbounds i32, ptr %a, i64 %outer.iv
br label %inner.body
inner.body:
%inner.iv = phi i64 [ 0, %outer.header ], [ %inner.iv.next, %inner.body ]
%gep.b = getelementptr inbounds i32, ptr %b, i64 %inner.iv
%l = load i32, ptr %gep.b, align 4
%sub = sub i32 %l, 10
store i32 %sub, ptr %gep.a, align 4
%inner.iv.next = add nuw nsw i64 %inner.iv, 1
%inner.cond = icmp eq i64 %inner.iv.next, %n
br i1 %inner.cond, label %outer.latch, label %inner.body
outer.latch:
%outer.iv.next = add nuw nsw i64 %outer.iv, 1
%outer.cond = icmp eq i64 %outer.iv.next, %n
br i1 %outer.cond, label %exit, label %outer.header
exit:
ret void
}
; Same as @nested_loop_outer_iv_addrec_invariant_in_inner1 but with dependence
; sink and source swapped.
define void @nested_loop_outer_iv_addrec_invariant_in_inner2(ptr %a, ptr %b, i64 %n) {
; CHECK-LABEL: @nested_loop_outer_iv_addrec_invariant_in_inner2(
; CHECK: entry:
; CHECK-NEXT: [[N_SHL_2:%.]] = shl i64 %n, 2
; CHECK-NEXT: [[B_GEP_UPPER:%.*]] = getelementptr i8, ptr %b, i64 [[N_SHL_2]]
; CHECK-NEXT: br label %outer
; CHECK: outer.header:
; CHECK: [[OUTER_IV_SHL_2:%.]] = shl i64 %outer.iv, 2
; CHECK-NEXT: [[A_GEP_UPPER:%.*]] = getelementptr i8, ptr %a, i64 [[OUTER_IV_SHL_2]]
; CHECK-NEXT: [[OUTER_IV_4:%.]] = add i64 [[OUTER_IV_SHL_2]], 4
; CHECK-NEXT: [[A_GEP_UPPER_4:%.*]] = getelementptr i8, ptr %a, i64 [[OUTER_IV_4]]
; CHECK: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %scalar.ph, label %vector.memcheck
; CHECK: vector.memcheck:
; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr %b, [[A_GEP_UPPER_4]]
; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[A_GEP_UPPER]], [[B_GEP_UPPER]]
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label %scalar.ph, label %vector.ph
;
entry:
br label %outer.header
outer.header:
%outer.iv = phi i64 [ %outer.iv.next, %outer.latch ], [ 0, %entry ]
%gep.a = getelementptr inbounds i32, ptr %a, i64 %outer.iv
br label %inner.body
inner.body:
%inner.iv = phi i64 [ 0, %outer.header ], [ %inner.iv.next, %inner.body ]
%l = load i32, ptr %gep.a, align 4
%sub = sub i32 %l, 10
%gep.b = getelementptr inbounds i32, ptr %b, i64 %inner.iv
store i32 %sub, ptr %gep.b, align 4
%inner.iv.next = add nuw nsw i64 %inner.iv, 1
%inner.cond = icmp eq i64 %inner.iv.next, %n
br i1 %inner.cond, label %outer.latch, label %inner.body
outer.latch:
%outer.iv.next = add nuw nsw i64 %outer.iv, 1
%outer.cond = icmp eq i64 %outer.iv.next, %n
br i1 %outer.cond, label %exit, label %outer.header
exit:
ret void
}
; Test case where the AddRec for the pointers in the inner loop have the AddRec
; of the outer loop as start value. It is sufficient to subtract the start
; values (%dst, %src) of the outer AddRecs.
define void @nested_loop_start_of_inner_ptr_addrec_is_same_outer_addrec(ptr nocapture noundef %dst, ptr nocapture noundef readonly %src, i64 noundef %m, i64 noundef %n) {
; CHECK-LABEL: @nested_loop_start_of_inner_ptr_addrec_is_same_outer_addrec(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[SRC2:%.*]] = ptrtoint ptr [[SRC:%.*]] to i64
; CHECK-NEXT: [[DST1:%.*]] = ptrtoint ptr [[DST:%.*]] to i64
; CHECK-NEXT: [[SUB:%.*]] = sub i64 [[DST1]], [[SRC2]]
; CHECK-NEXT: br label [[OUTER_LOOP:%.*]]
; CHECK: outer.loop:
; CHECK-NEXT: [[OUTER_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[OUTER_IV_NEXT:%.*]], [[INNER_EXIT:%.*]] ]
; CHECK-NEXT: [[MUL:%.*]] = mul nsw i64 [[OUTER_IV]], [[N]]
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
; CHECK: vector.memcheck:
; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[SUB]], 16
; CHECK-NEXT: br i1 [[DIFF_CHECK]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
;
entry:
br label %outer.loop
outer.loop:
%outer.iv = phi i64 [ 0, %entry ], [ %outer.iv.next, %inner.exit ]
%mul = mul nsw i64 %outer.iv, %n
br label %inner.loop
inner.loop:
%iv.inner = phi i64 [ 0, %outer.loop ], [ %iv.inner.next, %inner.loop ]
%idx = add nuw nsw i64 %iv.inner, %mul
%gep.src = getelementptr inbounds i32, ptr %src, i64 %idx
%l = load i32, ptr %gep.src, align 4
%gep.dst = getelementptr inbounds i32, ptr %dst, i64 %idx
%add = add nsw i32 %l, 10
store i32 %add, ptr %gep.dst, align 4
%iv.inner.next = add nuw nsw i64 %iv.inner, 1
%inner.exit.cond = icmp eq i64 %iv.inner.next, %n
br i1 %inner.exit.cond, label %inner.exit, label %inner.loop
inner.exit:
%outer.iv.next = add nuw nsw i64 %outer.iv, 1
%outer.exit.cond = icmp eq i64 %outer.iv.next, %m
br i1 %outer.exit.cond, label %outer.exit, label %outer.loop
outer.exit:
ret void
}
|
