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
|
; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 3
; RUN: opt -passes='print<access-info>' -disable-output %s 2>&1 | FileCheck %s
target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
; Test cases for https://github.com/llvm/llvm-project/issues/69744.
; Note that both loops in the tests are needed to incorrectly determine that
; the loops are safe with runtime checks via FoundNonConstantDistanceDependence
; handling code in LAA.
define void @test_indirect_read_write_loop_also_modifies_pointer_array(ptr noundef %arr) {
; CHECK-LABEL: 'test_indirect_read_write_loop_also_modifies_pointer_array'
; CHECK-NEXT: loop.1:
; CHECK-NEXT: Report: could not determine number of loop iterations
; CHECK-NEXT: Dependences:
; CHECK-NEXT: Run-time memory checks:
; CHECK-NEXT: Grouped accesses:
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
; CHECK-NEXT: loop.2:
; CHECK-NEXT: Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
; CHECK-NEXT: Unsafe indirect dependence.
; CHECK-NEXT: Dependences:
; CHECK-NEXT: IndidrectUnsafe:
; CHECK-NEXT: %l.2 = load i64, ptr %l.1, align 8, !tbaa !4 ->
; CHECK-NEXT: store i64 %inc, ptr %l.1, align 8, !tbaa !4
; CHECK-EMPTY:
; CHECK-NEXT: Unknown:
; CHECK-NEXT: %l.1 = load ptr, ptr %gep.iv.1, align 8, !tbaa !0 ->
; CHECK-NEXT: store ptr %l.1, ptr %gep.iv.2, align 8, !tbaa !0
; CHECK-EMPTY:
; CHECK-NEXT: Run-time memory checks:
; CHECK-NEXT: Grouped accesses:
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
;
entry:
br label %loop.1
loop.1:
%iv = phi i64 [ %iv.next, %loop.1 ], [ 8, %entry ]
%arr.addr.0.i = phi ptr [ %incdec.ptr.i, %loop.1 ], [ %arr, %entry ]
%incdec.ptr.i = getelementptr inbounds ptr, ptr %arr.addr.0.i, i64 1
%0 = load ptr, ptr %arr.addr.0.i, align 8, !tbaa !6
%tobool.not.i = icmp eq ptr %0, null
%iv.next = add i64 %iv, 8
br i1 %tobool.not.i, label %loop.1.exit, label %loop.1
loop.1.exit:
%iv.lcssa = phi i64 [ %iv, %loop.1 ]
br label %loop.2
loop.2:
%iv.1 = phi i64 [ 0, %loop.1.exit ], [ %iv.1.next, %loop.2 ]
%iv.2 = phi i64 [ %iv.lcssa, %loop.1.exit ], [ %iv.2.next, %loop.2 ]
%gep.iv.1 = getelementptr inbounds ptr, ptr %arr, i64 %iv.1
%l.1 = load ptr, ptr %gep.iv.1, align 8, !tbaa !6
%l.2 = load i64, ptr %l.1, align 8, !tbaa !13
%inc = add i64 %l.2, 1
store i64 %inc, ptr %l.1, align 8, !tbaa !13
%iv.2.next = add nsw i64 %iv.2, 1
%gep.iv.2 = getelementptr inbounds ptr, ptr %arr, i64 %iv.2
store ptr %l.1, ptr %gep.iv.2, align 8, !tbaa !6
%iv.1.next = add nuw nsw i64 %iv.1, 1
%cmp = icmp ult i64 %iv.1.next, 1000
br i1 %cmp, label %loop.2, label %exit
exit:
ret void
}
define void @test_indirect_read_loop_also_modifies_pointer_array(ptr noundef %arr) {
; CHECK-LABEL: 'test_indirect_read_loop_also_modifies_pointer_array'
; CHECK-NEXT: loop.1:
; CHECK-NEXT: Report: could not determine number of loop iterations
; CHECK-NEXT: Dependences:
; CHECK-NEXT: Run-time memory checks:
; CHECK-NEXT: Grouped accesses:
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
; CHECK-NEXT: loop.2:
; CHECK-NEXT: Memory dependences are safe with run-time checks
; CHECK-NEXT: Dependences:
; CHECK-NEXT: Run-time memory checks:
; CHECK-NEXT: Check 0:
; CHECK-NEXT: Comparing group ([[GRP1:0x[0-9a-f]+]]):
; CHECK-NEXT: %gep.iv.2 = getelementptr inbounds i64, ptr %arr, i64 %iv.2
; CHECK-NEXT: Against group ([[GRP2:0x[0-9a-f]+]]):
; CHECK-NEXT: %gep.iv.1 = getelementptr inbounds ptr, ptr %arr, i64 %iv.1
; CHECK-NEXT: Grouped accesses:
; CHECK-NEXT: Group [[GRP1]]:
; CHECK-NEXT: (Low: {(64 + %arr),+,64}<%loop.1> High: {(8064 + %arr),+,64}<%loop.1>)
; CHECK-NEXT: Member: {{\{\{}}(64 + %arr),+,64}<%loop.1>,+,8}<%loop.2>
; CHECK-NEXT: Group [[GRP2]]:
; CHECK-NEXT: (Low: %arr High: (8000 + %arr))
; CHECK-NEXT: Member: {%arr,+,8}<nuw><%loop.2>
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
;
entry:
br label %loop.1
loop.1:
%iv = phi i64 [ %iv.next, %loop.1 ], [ 8, %entry ]
%arr.addr.0.i = phi ptr [ %incdec.ptr.i, %loop.1 ], [ %arr, %entry ]
%incdec.ptr.i = getelementptr inbounds ptr, ptr %arr.addr.0.i, i64 1
%0 = load ptr, ptr %arr.addr.0.i, align 8, !tbaa !6
%tobool.not.i = icmp eq ptr %0, null
%iv.next = add i64 %iv, 8
br i1 %tobool.not.i, label %loop.1.exit, label %loop.1
loop.1.exit:
%iv.lcssa = phi i64 [ %iv, %loop.1 ]
br label %loop.2
loop.2:
%iv.1 = phi i64 [ 0, %loop.1.exit ], [ %iv.1.next, %loop.2 ]
%iv.2 = phi i64 [ %iv.lcssa, %loop.1.exit ], [ %iv.2.next, %loop.2 ]
%gep.iv.1 = getelementptr inbounds ptr, ptr %arr, i64 %iv.1
%l.1 = load ptr, ptr %gep.iv.1, align 8, !tbaa !6
%l.2 = load i64, ptr %l.1, align 8, !tbaa !13
%inc = add i64 %l.2, 1
%iv.2.next = add nsw i64 %iv.2, 1
%gep.iv.2 = getelementptr inbounds i64, ptr %arr, i64 %iv.2
store i64 %l.2, ptr %gep.iv.2, align 8, !tbaa !6
%iv.1.next = add nuw nsw i64 %iv.1, 1
%cmp = icmp ult i64 %iv.1.next, 1000
br i1 %cmp, label %loop.2, label %exit
exit:
ret void
}
define void @test_indirect_write_loop_also_modifies_pointer_array(ptr noundef %arr) {
; CHECK-LABEL: 'test_indirect_write_loop_also_modifies_pointer_array'
; CHECK-NEXT: loop.1:
; CHECK-NEXT: Report: could not determine number of loop iterations
; CHECK-NEXT: Dependences:
; CHECK-NEXT: Run-time memory checks:
; CHECK-NEXT: Grouped accesses:
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
; CHECK-NEXT: loop.2:
; CHECK-NEXT: Memory dependences are safe with run-time checks
; CHECK-NEXT: Dependences:
; CHECK-NEXT: Run-time memory checks:
; CHECK-NEXT: Check 0:
; CHECK-NEXT: Comparing group ([[GRP3:0x[0-9a-f]+]]):
; CHECK-NEXT: %gep.iv.2 = getelementptr inbounds ptr, ptr %arr, i64 %iv.2
; CHECK-NEXT: Against group ([[GRP4:0x[0-9a-f]+]]):
; CHECK-NEXT: %gep.iv.1 = getelementptr inbounds ptr, ptr %arr, i64 %iv.1
; CHECK-NEXT: Grouped accesses:
; CHECK-NEXT: Group [[GRP3]]:
; CHECK-NEXT: (Low: {(64 + %arr),+,64}<%loop.1> High: {(8064 + %arr),+,64}<%loop.1>)
; CHECK-NEXT: Member: {{\{\{}}(64 + %arr),+,64}<%loop.1>,+,8}<%loop.2>
; CHECK-NEXT: Group [[GRP4]]:
; CHECK-NEXT: (Low: %arr High: (8000 + %arr))
; CHECK-NEXT: Member: {%arr,+,8}<nuw><%loop.2>
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
;
entry:
br label %loop.1
loop.1:
%iv = phi i64 [ %iv.next, %loop.1 ], [ 8, %entry ]
%arr.addr.0.i = phi ptr [ %incdec.ptr.i, %loop.1 ], [ %arr, %entry ]
%incdec.ptr.i = getelementptr inbounds ptr, ptr %arr.addr.0.i, i64 1
%0 = load ptr, ptr %arr.addr.0.i, align 8, !tbaa !6
%tobool.not.i = icmp eq ptr %0, null
%iv.next = add i64 %iv, 8
br i1 %tobool.not.i, label %loop.1.exit, label %loop.1
loop.1.exit:
%iv.lcssa = phi i64 [ %iv, %loop.1 ]
br label %loop.2
loop.2:
%iv.1 = phi i64 [ 0, %loop.1.exit ], [ %iv.1.next, %loop.2 ]
%iv.2 = phi i64 [ %iv.lcssa, %loop.1.exit ], [ %iv.2.next, %loop.2 ]
%gep.iv.1 = getelementptr inbounds ptr, ptr %arr, i64 %iv.1
%l.1 = load ptr, ptr %gep.iv.1, align 8, !tbaa !6
%inc = add i64 %iv.1, 1
store i64 %inc, ptr %l.1, align 8, !tbaa !13
%iv.2.next = add nsw i64 %iv.2, 1
%gep.iv.2 = getelementptr inbounds ptr, ptr %arr, i64 %iv.2
store ptr %l.1, ptr %gep.iv.2, align 8, !tbaa !6
%iv.1.next = add nuw nsw i64 %iv.1, 1
%cmp = icmp ult i64 %iv.1.next, 1000
br i1 %cmp, label %loop.2, label %exit
exit:
ret void
}
; FIXME: Not safe with runtime checks due to the indirect pointers are modified
; in the loop.
define void @test_indirect_read_write_loop_does_not_modify_pointer_array(ptr noundef %arr, ptr noundef noalias %arr2) {
; CHECK-LABEL: 'test_indirect_read_write_loop_does_not_modify_pointer_array'
; CHECK-NEXT: loop.1:
; CHECK-NEXT: Report: could not determine number of loop iterations
; CHECK-NEXT: Dependences:
; CHECK-NEXT: Run-time memory checks:
; CHECK-NEXT: Grouped accesses:
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
; CHECK-NEXT: loop.2:
; CHECK-NEXT: Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
; CHECK-NEXT: Unsafe indirect dependence.
; CHECK-NEXT: Dependences:
; CHECK-NEXT: IndidrectUnsafe:
; CHECK-NEXT: %l.2 = load i64, ptr %l.1, align 8, !tbaa !4 ->
; CHECK-NEXT: store i64 %inc, ptr %l.1, align 8, !tbaa !4
; CHECK-EMPTY:
; CHECK-NEXT: Unknown:
; CHECK-NEXT: %l.3 = load i64, ptr %gep.arr2.iv.1, align 8 ->
; CHECK-NEXT: store i64 %inc.2, ptr %gep.arr2.iv.2, align 8, !tbaa !0
; CHECK-EMPTY:
; CHECK-NEXT: Run-time memory checks:
; CHECK-NEXT: Grouped accesses:
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
;
entry:
br label %loop.1
loop.1:
%iv = phi i64 [ %iv.next, %loop.1 ], [ 8, %entry ]
%arr.addr.0.i = phi ptr [ %incdec.ptr.i, %loop.1 ], [ %arr, %entry ]
%incdec.ptr.i = getelementptr inbounds ptr, ptr %arr.addr.0.i, i64 1
%0 = load ptr, ptr %arr.addr.0.i, align 8, !tbaa !6
%tobool.not.i = icmp eq ptr %0, null
%iv.next = add i64 %iv, 8
br i1 %tobool.not.i, label %loop.1.exit, label %loop.1
loop.1.exit:
%iv.lcssa = phi i64 [ %iv, %loop.1 ]
br label %loop.2
loop.2:
%iv.1 = phi i64 [ 0, %loop.1.exit ], [ %iv.1.next, %loop.2 ]
%iv.2 = phi i64 [ %iv.lcssa, %loop.1.exit ], [ %iv.2.next, %loop.2 ]
%gep.iv.1 = getelementptr inbounds ptr, ptr %arr, i64 %iv.1
%l.1 = load ptr, ptr %gep.iv.1, align 8, !tbaa !6
%l.2 = load i64, ptr %l.1, align 8, !tbaa !13
%inc = add i64 %l.2, 1
store i64 %inc, ptr %l.1, align 8, !tbaa !13
%iv.2.next = add nsw i64 %iv.2, 1
%gep.arr2.iv.1 = getelementptr inbounds i64 , ptr %arr2, i64 %iv.1
%gep.arr2.iv.2 = getelementptr inbounds i64 , ptr %arr2, i64 %iv.2
%l.3 = load i64, ptr %gep.arr2.iv.1
%inc.2 = add i64 %l.3, 5
store i64 %inc.2, ptr %gep.arr2.iv.2, align 8, !tbaa !6
%iv.1.next = add nuw nsw i64 %iv.1, 1
%cmp = icmp ult i64 %iv.1.next, 1000
br i1 %cmp, label %loop.2, label %exit
exit:
ret void
}
!6 = !{!7, !7, i64 0}
!7 = !{!"any pointer", !8, i64 0}
!8 = !{!"omnipotent char", !9, i64 0}
!9 = !{!"Simple C/C++ TBAA"}
!13 = !{!14, !15, i64 0}
!14 = !{!"", !15, i64 0}
!15 = !{!"long long", !8, i64 0}
|
