1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -instcombine -S                         | FileCheck %s
3; RUN: opt < %s -instcombine -S -data-layout=n32        | FileCheck %s
4; RUN: opt < %s -instcombine -S -data-layout=n32:64     | FileCheck %s
5; RUN: opt < %s -instcombine -S -data-layout=n32:64:128 | FileCheck %s
6
7declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i1) nounwind
8
9; memcpy can be expanded inline with load/store (based on the datalayout?)
10
11define void @copy_1_byte(i8* %d, i8* %s) {
12; CHECK-LABEL: @copy_1_byte(
13; CHECK-NEXT:    [[TMP1:%.*]] = load i8, i8* [[S:%.*]], align 1
14; CHECK-NEXT:    store i8 [[TMP1]], i8* [[D:%.*]], align 1
15; CHECK-NEXT:    ret void
16;
17  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %d, i8* %s, i32 1, i1 false)
18  ret void
19}
20
21define void @copy_2_bytes(i8* %d, i8* %s) {
22; CHECK-LABEL: @copy_2_bytes(
23; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i8* [[S:%.*]] to i16*
24; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8* [[D:%.*]] to i16*
25; CHECK-NEXT:    [[TMP3:%.*]] = load i16, i16* [[TMP1]], align 1
26; CHECK-NEXT:    store i16 [[TMP3]], i16* [[TMP2]], align 1
27; CHECK-NEXT:    ret void
28;
29  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %d, i8* %s, i32 2, i1 false)
30  ret void
31}
32
33; We don't expand small non-power-of-2. Should we? Might be a target-dependent choice.
34
35define void @copy_3_bytes(i8* %d, i8* %s) {
36; CHECK-LABEL: @copy_3_bytes(
37; CHECK-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* nonnull align 1 dereferenceable(3) [[D:%.*]], i8* nonnull align 1 dereferenceable(3) [[S:%.*]], i32 3, i1 false)
38; CHECK-NEXT:    ret void
39;
40  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %d, i8* %s, i32 3, i1 false)
41  ret void
42}
43
44define void @copy_4_bytes(i8* %d, i8* %s) {
45; CHECK-LABEL: @copy_4_bytes(
46; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i8* [[S:%.*]] to i32*
47; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8* [[D:%.*]] to i32*
48; CHECK-NEXT:    [[TMP3:%.*]] = load i32, i32* [[TMP1]], align 1
49; CHECK-NEXT:    store i32 [[TMP3]], i32* [[TMP2]], align 1
50; CHECK-NEXT:    ret void
51;
52  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %d, i8* %s, i32 4, i1 false)
53  ret void
54}
55
56; We don't expand small non-power-of-2. Should we? Might be a target-dependent choice.
57
58define void @copy_5_bytes(i8* %d, i8* %s) {
59; CHECK-LABEL: @copy_5_bytes(
60; CHECK-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* nonnull align 1 dereferenceable(5) [[D:%.*]], i8* nonnull align 1 dereferenceable(5) [[S:%.*]], i32 5, i1 false)
61; CHECK-NEXT:    ret void
62;
63  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %d, i8* %s, i32 5, i1 false)
64  ret void
65}
66
67define void @copy_8_bytes(i8* %d, i8* %s) {
68; CHECK-LABEL: @copy_8_bytes(
69; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i8* [[S:%.*]] to i64*
70; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8* [[D:%.*]] to i64*
71; CHECK-NEXT:    [[TMP3:%.*]] = load i64, i64* [[TMP1]], align 1
72; CHECK-NEXT:    store i64 [[TMP3]], i64* [[TMP2]], align 1
73; CHECK-NEXT:    ret void
74;
75  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %d, i8* %s, i32 8, i1 false)
76  ret void
77}
78
79define void @copy_16_bytes(i8* %d, i8* %s) {
80; CHECK-LABEL: @copy_16_bytes(
81; CHECK-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* nonnull align 1 dereferenceable(16) [[D:%.*]], i8* nonnull align 1 dereferenceable(16) [[S:%.*]], i32 16, i1 false)
82; CHECK-NEXT:    ret void
83;
84  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %d, i8* %s, i32 16, i1 false)
85  ret void
86}
87
88