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
| ; RUN: opt -S -licm < %s | FileCheck %s
; Note: the !invariant.load is there just solely to let us call @use()
; to add a fake use, and still have the aliasing work out. The call
; to @use(0) is just to provide a may-unwind exit out of the loop, so
; that LICM cannot hoist out the load simply because it is guaranteed
; to execute.
declare void @use(i32)
define void @f_0(i8* align 4 dereferenceable(1024) %ptr) {
; CHECK-LABEL: @f_0(
; CHECK: entry:
; CHECK: %val = load i32, i32* %ptr.i32
; CHECK: br label %loop
; CHECK: loop:
; CHECK: call void @use(i32 0)
; CHECK-NEXT: call void @use(i32 %val)
entry:
%ptr.gep = getelementptr i8, i8* %ptr, i32 32
%ptr.i32 = bitcast i8* %ptr.gep to i32*
br label %loop
loop:
call void @use(i32 0)
%val = load i32, i32* %ptr.i32, !invariant.load !{}
call void @use(i32 %val)
br label %loop
}
define void @f_1(i8* align 4 dereferenceable_or_null(1024) %ptr) {
; CHECK-LABEL: @f_1(
entry:
%ptr.gep = getelementptr i8, i8* %ptr, i32 32
%ptr.i32 = bitcast i8* %ptr.gep to i32*
%ptr_is_null = icmp eq i8* %ptr, null
br i1 %ptr_is_null, label %leave, label %loop
; CHECK: loop.preheader:
; CHECK: %val = load i32, i32* %ptr.i32
; CHECK: br label %loop
; CHECK: loop:
; CHECK: call void @use(i32 0)
; CHECK-NEXT: call void @use(i32 %val)
loop:
call void @use(i32 0)
%val = load i32, i32* %ptr.i32, !invariant.load !{}
call void @use(i32 %val)
br label %loop
leave:
ret void
}
define void @f_2(i8* align 4 dereferenceable_or_null(1024) %ptr) {
; CHECK-LABEL: @f_2(
; CHECK-NOT: load
; CHECK: call void @use(i32 0)
; CHECK-NEXT: %val = load i32, i32* %ptr.i32, !invariant.load !0
; CHECK-NEXT: call void @use(i32 %val)
entry:
;; Can't hoist, since the alignment does not work out -- (<4 byte
;; aligned> + 30) is not necessarily 4 byte aligned.
%ptr.gep = getelementptr i8, i8* %ptr, i32 30
%ptr.i32 = bitcast i8* %ptr.gep to i32*
%ptr_is_null = icmp eq i8* %ptr, null
br i1 %ptr_is_null, label %leave, label %loop
loop:
call void @use(i32 0)
%val = load i32, i32* %ptr.i32, !invariant.load !{}
call void @use(i32 %val)
br label %loop
leave:
ret void
}
define void @checkLaunder(i8* align 4 dereferenceable(1024) %p) {
; CHECK-LABEL: @checkLaunder(
; CHECK: entry:
; CHECK: %l = call i8* @llvm.launder.invariant.group.p0i8(i8* %p)
; CHECK: %val = load i8, i8* %l
; CHECK: br label %loop
; CHECK: loop:
; CHECK: call void @use(i32 0)
; CHECK-NEXT: call void @use8(i8 %val)
entry:
%l = call i8* @llvm.launder.invariant.group.p0i8(i8* %p)
br label %loop
loop:
call void @use(i32 0)
%val = load i8, i8* %l, !invariant.load !{}
call void @use8(i8 %val)
br label %loop
}
declare i8* @llvm.launder.invariant.group.p0i8(i8*)
declare void @use8(i8)
|