1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -S -memcpyopt < %s -verify-memoryssa | FileCheck %s
3
4; PR40118: BasicAA didn't realize that stackrestore ends the lifetime of
5; unescaped dynamic allocas, such as those that might come from inalloca.
6
7source_filename = "t.cpp"
8target datalayout = "e-m:x-p:32:32-i64:64-f80:32-n8:16:32-a:0:32-S32"
9target triple = "i686-unknown-windows-msvc19.14.26433"
10
11@str = internal constant [9 x i8] c"abcdxxxxx"
12
13
14; Test that we can propagate memcpy through an unescaped dynamic alloca across
15; a call to @external.
16
17define i32 @test_norestore(i32 %n) {
18; NO_MSSA-LABEL: @test_norestore(
19; NO_MSSA-NEXT:    [[TMPMEM:%.*]] = alloca [10 x i8], align 4
20; NO_MSSA-NEXT:    [[TMP:%.*]] = getelementptr inbounds [10 x i8], [10 x i8]* [[TMPMEM]], i32 0, i32 0
21; NO_MSSA-NEXT:    [[P:%.*]] = alloca i8, i32 [[N:%.*]], align 4
22; NO_MSSA-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[P]], i8* align 1 getelementptr inbounds ([9 x i8], [9 x i8]* @str, i32 0, i32 0), i32 9, i1 false)
23; NO_MSSA-NEXT:    [[P10:%.*]] = getelementptr inbounds i8, i8* [[P]], i32 9
24; NO_MSSA-NEXT:    store i8 0, i8* [[P10]], align 1
25; NO_MSSA-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[TMP]], i8* [[P]], i32 10, i1 false)
26; NO_MSSA-NEXT:    call void @external()
27; NO_MSSA-NEXT:    [[HEAP:%.*]] = call i8* @malloc(i32 9)
28; NO_MSSA-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[HEAP]], i8* [[P]], i32 9, i1 false)
29; NO_MSSA-NEXT:    call void @useit(i8* [[HEAP]])
30; NO_MSSA-NEXT:    ret i32 0
31;
32; MSSA-LABEL: @test_norestore(
33; MSSA-NEXT:    [[TMPMEM:%.*]] = alloca [10 x i8], align 4
34; MSSA-NEXT:    [[TMP:%.*]] = getelementptr inbounds [10 x i8], [10 x i8]* [[TMPMEM]], i32 0, i32 0
35; MSSA-NEXT:    [[P:%.*]] = alloca i8, i32 [[N:%.*]], align 4
36; MSSA-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[P]], i8* align 1 getelementptr inbounds ([9 x i8], [9 x i8]* @str, i32 0, i32 0), i32 9, i1 false)
37; MSSA-NEXT:    [[P10:%.*]] = getelementptr inbounds i8, i8* [[P]], i32 9
38; MSSA-NEXT:    store i8 0, i8* [[P10]], align 1
39; MSSA-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[TMP]], i8* [[P]], i32 10, i1 false)
40; MSSA-NEXT:    call void @external()
41; MSSA-NEXT:    [[HEAP:%.*]] = call i8* @malloc(i32 9)
42; MSSA-NEXT:    call void @llvm.memmove.p0i8.p0i8.i32(i8* [[HEAP]], i8* align 1 getelementptr inbounds ([9 x i8], [9 x i8]* @str, i32 0, i32 0), i32 9, i1 false)
43; MSSA-NEXT:    call void @useit(i8* [[HEAP]])
44; MSSA-NEXT:    ret i32 0
45;
46  %tmpmem = alloca [10 x i8], align 4
47  %tmp = getelementptr inbounds [10 x i8], [10 x i8]* %tmpmem, i32 0, i32 0
48
49  ; Make a dynamic alloca, initialize it.
50  %p = alloca i8, i32 %n, align 4
51  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %p, i8* align 1 getelementptr inbounds ([9 x i8], [9 x i8]* @str, i32 0, i32 0), i32 9, i1 false)
52
53  ; This extra byte exists to prevent memcpyopt from propagating @str.
54  %p10 = getelementptr inbounds i8, i8* %p, i32 9
55  store i8 0, i8* %p10
56
57  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %tmp, i8* %p, i32 10, i1 false)
58  call void @external()
59  %heap = call i8* @malloc(i32 9)
60  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %heap, i8* %tmp, i32 9, i1 false)
61  call void @useit(i8* %heap)
62  ret i32 0
63}
64
65; Do not propagate memcpy from %p across the stackrestore.
66
67define i32 @test_stackrestore() {
68; CHECK-LABEL: @test_stackrestore(
69; CHECK-NEXT:    [[TMPMEM:%.*]] = alloca [10 x i8], align 4
70; CHECK-NEXT:    [[TMP:%.*]] = getelementptr inbounds [10 x i8], [10 x i8]* [[TMPMEM]], i32 0, i32 0
71; CHECK-NEXT:    [[INALLOCA_SAVE:%.*]] = tail call i8* @llvm.stacksave()
72; CHECK-NEXT:    [[ARGMEM:%.*]] = alloca inalloca [10 x i8], align 4
73; CHECK-NEXT:    [[P:%.*]] = getelementptr inbounds [10 x i8], [10 x i8]* [[ARGMEM]], i32 0, i32 0
74; CHECK-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[P]], i8* align 1 getelementptr inbounds ([9 x i8], [9 x i8]* @str, i32 0, i32 0), i32 9, i1 false)
75; CHECK-NEXT:    [[P10:%.*]] = getelementptr inbounds [10 x i8], [10 x i8]* [[ARGMEM]], i32 0, i32 9
76; CHECK-NEXT:    store i8 0, i8* [[P10]], align 1
77; CHECK-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[TMP]], i8* [[P]], i32 10, i1 false)
78; CHECK-NEXT:    call void @llvm.stackrestore(i8* [[INALLOCA_SAVE]])
79; CHECK-NEXT:    [[HEAP:%.*]] = call i8* @malloc(i32 9)
80; CHECK-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[HEAP]], i8* [[TMP]], i32 9, i1 false)
81; CHECK-NEXT:    call void @useit(i8* [[HEAP]])
82; CHECK-NEXT:    ret i32 0
83;
84  %tmpmem = alloca [10 x i8], align 4
85  %tmp = getelementptr inbounds [10 x i8], [10 x i8]* %tmpmem, i32 0, i32 0
86  %inalloca.save = tail call i8* @llvm.stacksave()
87  %argmem = alloca inalloca [10 x i8], align 4
88  %p = getelementptr inbounds [10 x i8], [10 x i8]* %argmem, i32 0, i32 0
89  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %p, i8* align 1 getelementptr inbounds ([9 x i8], [9 x i8]* @str, i32 0, i32 0), i32 9, i1 false)
90
91  ; This extra byte exists to prevent memcpyopt from propagating @str.
92  %p10 = getelementptr inbounds [10 x i8], [10 x i8]* %argmem, i32 0, i32 9
93  store i8 0, i8* %p10
94
95  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %tmp, i8* %p, i32 10, i1 false)
96  call void @llvm.stackrestore(i8* %inalloca.save)
97  %heap = call i8* @malloc(i32 9)
98  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %heap, i8* %tmp, i32 9, i1 false)
99  call void @useit(i8* %heap)
100  ret i32 0
101}
102
103declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture writeonly, i8* nocapture readonly, i32, i1)
104declare i8* @llvm.stacksave()
105declare void @llvm.stackrestore(i8*)
106declare i8* @malloc(i32)
107declare void @useit(i8*)
108declare void @external()
109