1; RUN: opt -tbaa -basicaa -licm -S < %s | FileCheck %s
2; RUN: opt -aa-pipeline=type-based-aa,basic-aa -passes='require<aa>,require<targetir>,require<scalar-evolution>,loop(licm)' -S %s | FileCheck %s
3
4; If we can prove a local is thread local, we can insert stores during
5; promotion which wouldn't be legal otherwise.
6
7target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
8target triple = "x86_64-linux-generic"
9
10@p = external global i8*
11
12declare i8* @malloc(i64)
13
14; Exercise the TLS case
15define i32* @test(i32 %n) {
16entry:
17  ;; ignore the required null check for simplicity
18  %mem = call dereferenceable(16) noalias i8* @malloc(i64 16)
19  %addr = bitcast i8* %mem to i32*
20  br label %for.body.lr.ph
21
22for.body.lr.ph:                                   ; preds = %entry
23  br label %for.header
24
25for.header:
26  %i.02 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
27  %old = load i32, i32* %addr, align 4
28  ; deliberate impossible to analyze branch
29  %guard = load atomic i8*, i8** @p monotonic, align 8
30  %exitcmp = icmp eq i8* %guard, null
31  br i1 %exitcmp, label %for.body, label %early-exit
32
33early-exit:
34; CHECK-LABEL: early-exit:
35; CHECK: store i32 %new1.lcssa, i32* %addr, align 1
36  ret i32* null
37
38for.body:
39  %new = add i32 %old, 1
40  store i32 %new, i32* %addr, align 4
41  %inc = add nsw i32 %i.02, 1
42  %cmp = icmp slt i32 %inc, %n
43  br i1 %cmp, label %for.header, label %for.cond.for.end_crit_edge
44
45for.cond.for.end_crit_edge:                       ; preds = %for.body
46; CHECK-LABEL: for.cond.for.end_crit_edge:
47; CHECK: store i32 %new.lcssa, i32* %addr, align 1
48  %split = phi i32* [ %addr, %for.body ]
49  ret i32* null
50}
51
52declare i8* @not_malloc(i64)
53
54; Negative test - not TLS
55define i32* @test_neg(i32 %n) {
56entry:
57  ;; ignore the required null check for simplicity
58  %mem = call dereferenceable(16) noalias i8* @not_malloc(i64 16)
59  %addr = bitcast i8* %mem to i32*
60  br label %for.body.lr.ph
61
62for.body.lr.ph:                                   ; preds = %entry
63  br label %for.header
64
65for.header:
66  %i.02 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
67  %old = load i32, i32* %addr, align 4
68  ; deliberate impossible to analyze branch
69  %guard = load volatile i8*, i8** @p
70  %exitcmp = icmp eq i8* %guard, null
71  br i1 %exitcmp, label %for.body, label %early-exit
72
73early-exit:
74; CHECK-LABEL: early-exit:
75; CHECK-NOT: store
76  ret i32* null
77
78for.body:
79; CHECK-LABEL: for.body:
80; CHECK: store i32 %new, i32* %addr, align 4
81  %new = add i32 %old, 1
82  store i32 %new, i32* %addr, align 4
83  %inc = add nsw i32 %i.02, 1
84  %cmp = icmp slt i32 %inc, %n
85  br i1 %cmp, label %for.header, label %for.cond.for.end_crit_edge
86
87for.cond.for.end_crit_edge:                       ; preds = %for.body
88; CHECK-LABEL: for.cond.for.end_crit_edge:
89; CHECK-NOT: store
90  %split = phi i32* [ %addr, %for.body ]
91  ret i32* null
92}
93
94; Negative test - can't speculate load since branch
95; may control alignment
96define i32* @test_neg2(i32 %n) {
97entry:
98  ;; ignore the required null check for simplicity
99  %mem = call dereferenceable(16) noalias i8* @malloc(i64 16)
100  %addr = bitcast i8* %mem to i32*
101  br label %for.body.lr.ph
102
103for.body.lr.ph:                                   ; preds = %entry
104  br label %for.header
105
106for.header:
107  %i.02 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
108  ; deliberate impossible to analyze branch
109  %guard = load volatile i8*, i8** @p
110  %exitcmp = icmp eq i8* %guard, null
111  br i1 %exitcmp, label %for.body, label %early-exit
112
113early-exit:
114; CHECK-LABEL: early-exit:
115; CHECK-NOT: store
116  ret i32* null
117
118for.body:
119; CHECK-LABEL: for.body:
120; CHECK: store i32 %new, i32* %addr, align 4
121  %old = load i32, i32* %addr, align 4
122  %new = add i32 %old, 1
123  store i32 %new, i32* %addr, align 4
124  %inc = add nsw i32 %i.02, 1
125  %cmp = icmp slt i32 %inc, %n
126  br i1 %cmp, label %for.header, label %for.cond.for.end_crit_edge
127
128for.cond.for.end_crit_edge:                       ; preds = %for.body
129; CHECK-LABEL: for.cond.for.end_crit_edge:
130; CHECK-NOT: store
131  %split = phi i32* [ %addr, %for.body ]
132  ret i32* null
133}
134
135