xref: /llvm-project/llvm/test/Transforms/IROutliner/extraction.ll (revision f4b925ee7078f058602fd323e25f45f1ae91ca34)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -S -passes=verify,iroutliner -ir-outlining-no-cost < %s | FileCheck %s
3
4; This test makes sure we are extracting the found similarity sections
5; correctly at the call site.
6
7define void @extract1() {
8; CHECK-LABEL: @extract1(
9; CHECK-NEXT:  entry:
10; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
11; CHECK-NEXT:    [[B:%.*]] = alloca i32, align 4
12; CHECK-NEXT:    [[C:%.*]] = alloca i32, align 4
13; CHECK-NEXT:    call void @outlined_ir_func_1(ptr [[A]], ptr [[B]], ptr [[C]])
14; CHECK-NEXT:    ret void
15;
16entry:
17  %a = alloca i32, align 4
18  %b = alloca i32, align 4
19  %c = alloca i32, align 4
20  store i32 2, ptr %a, align 4
21  store i32 3, ptr %b, align 4
22  store i32 4, ptr %c, align 4
23  %al = load i32, ptr %a
24  %bl = load i32, ptr %b
25  %cl = load i32, ptr %c
26  ret void
27}
28
29define void @extract2() {
30; CHECK-LABEL: @extract2(
31; CHECK-NEXT:  entry:
32; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
33; CHECK-NEXT:    [[B:%.*]] = alloca i32, align 4
34; CHECK-NEXT:    [[C:%.*]] = alloca i32, align 4
35; CHECK-NEXT:    call void @outlined_ir_func_1(ptr [[A]], ptr [[B]], ptr [[C]])
36; CHECK-NEXT:    ret void
37;
38entry:
39  %a = alloca i32, align 4
40  %b = alloca i32, align 4
41  %c = alloca i32, align 4
42  store i32 2, ptr %a, align 4
43  store i32 3, ptr %b, align 4
44  store i32 4, ptr %c, align 4
45  %al = load i32, ptr %a
46  %bl = load i32, ptr %b
47  %cl = load i32, ptr %c
48  ret void
49}
50
51; There are potential ouptuts in this sections, but we do not extract sections
52; with outputs right now, since they cannot be consolidated.
53define void @extract_outs1() #0 {
54; CHECK-LABEL: @extract_outs1(
55; CHECK-NEXT:  entry:
56; CHECK-NEXT:    [[DOTLOC:%.*]] = alloca i32, align 4
57; CHECK-NEXT:    [[ADD_LOC:%.*]] = alloca i32, align 4
58; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
59; CHECK-NEXT:    [[B:%.*]] = alloca i32, align 4
60; CHECK-NEXT:    [[OUTPUT:%.*]] = alloca i32, align 4
61; CHECK-NEXT:    [[RESULT:%.*]] = alloca i32, align 4
62; CHECK-NEXT:    call void @llvm.lifetime.start.p0(i64 -1, ptr [[ADD_LOC]])
63; CHECK-NEXT:    call void @llvm.lifetime.start.p0(i64 -1, ptr [[DOTLOC]])
64; CHECK-NEXT:    call void @outlined_ir_func_0(ptr [[A]], ptr [[B]], ptr [[OUTPUT]], ptr [[ADD_LOC]], ptr [[DOTLOC]])
65; CHECK-NEXT:    [[ADD_RELOAD:%.*]] = load i32, ptr [[ADD_LOC]], align 4
66; CHECK-NEXT:    [[DOTRELOAD:%.*]] = load i32, ptr [[DOTLOC]], align 4
67; CHECK-NEXT:    call void @llvm.lifetime.end.p0(i64 -1, ptr [[ADD_LOC]])
68; CHECK-NEXT:    call void @llvm.lifetime.end.p0(i64 -1, ptr [[DOTLOC]])
69; CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[OUTPUT]], align 4
70; CHECK-NEXT:    call void @outlined_ir_func_2(i32 [[DOTRELOAD]], i32 [[ADD_RELOAD]], ptr [[RESULT]])
71; CHECK-NEXT:    ret void
72;
73entry:
74  %a = alloca i32, align 4
75  %b = alloca i32, align 4
76  %output = alloca i32, align 4
77  %result = alloca i32, align 4
78  store i32 2, ptr %a, align 4
79  store i32 3, ptr %b, align 4
80  %0 = load i32, ptr %a, align 4
81  %1 = load i32, ptr %b, align 4
82  %add = add i32 %0, %1
83  store i32 %add, ptr %output, align 4
84  %2 = load i32, ptr %output, align 4
85  %3 = load i32, ptr %output, align 4
86  %mul = mul i32 %2, %add
87  store i32 %mul, ptr %result, align 4
88  ret void
89}
90
91; There are potential ouptuts in this sections, but we do not extract sections
92; with outputs right now, since they cannot be consolidated.
93define void @extract_outs2() #0 {
94; CHECK-LABEL: @extract_outs2(
95; CHECK-NEXT:  entry:
96; CHECK-NEXT:    [[DOTLOC:%.*]] = alloca i32, align 4
97; CHECK-NEXT:    [[ADD_LOC:%.*]] = alloca i32, align 4
98; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
99; CHECK-NEXT:    [[B:%.*]] = alloca i32, align 4
100; CHECK-NEXT:    [[OUTPUT:%.*]] = alloca i32, align 4
101; CHECK-NEXT:    [[RESULT:%.*]] = alloca i32, align 4
102; CHECK-NEXT:    call void @llvm.lifetime.start.p0(i64 -1, ptr [[ADD_LOC]])
103; CHECK-NEXT:    call void @llvm.lifetime.start.p0(i64 -1, ptr [[DOTLOC]])
104; CHECK-NEXT:    call void @outlined_ir_func_0(ptr [[A]], ptr [[B]], ptr [[OUTPUT]], ptr [[ADD_LOC]], ptr [[DOTLOC]])
105; CHECK-NEXT:    [[ADD_RELOAD:%.*]] = load i32, ptr [[ADD_LOC]], align 4
106; CHECK-NEXT:    [[DOTRELOAD:%.*]] = load i32, ptr [[DOTLOC]], align 4
107; CHECK-NEXT:    call void @llvm.lifetime.end.p0(i64 -1, ptr [[ADD_LOC]])
108; CHECK-NEXT:    call void @llvm.lifetime.end.p0(i64 -1, ptr [[DOTLOC]])
109; CHECK-NEXT:    call void @outlined_ir_func_2(i32 [[DOTRELOAD]], i32 [[ADD_RELOAD]], ptr [[RESULT]])
110; CHECK-NEXT:    ret void
111;
112entry:
113  %a = alloca i32, align 4
114  %b = alloca i32, align 4
115  %output = alloca i32, align 4
116  %result = alloca i32, align 4
117  store i32 2, ptr %a, align 4
118  store i32 3, ptr %b, align 4
119  %0 = load i32, ptr %a, align 4
120  %1 = load i32, ptr %b, align 4
121  %add = add i32 %0, %1
122  store i32 %add, ptr %output, align 4
123  %2 = load i32, ptr %output, align 4
124  %mul = mul i32 %2, %add
125  store i32 %mul, ptr %result, align 4
126  ret void
127}
128