1; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -infer-address-spaces %s | FileCheck %s
2
3; CHECK-LABEL: @memset_group_to_flat(
4; CHECK: call void @llvm.memset.p3i8.i64(i8 addrspace(3)* align 4 %group.ptr, i8 4, i64 32, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
5define amdgpu_kernel void @memset_group_to_flat(i8 addrspace(3)* %group.ptr, i32 %y) #0 {
6  %cast = addrspacecast i8 addrspace(3)* %group.ptr to i8*
7  call void @llvm.memset.p0i8.i64(i8* align 4 %cast, i8 4, i64 32, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
8  ret void
9}
10
11; CHECK-LABEL: @memset_global_to_flat(
12; CHECK: call void @llvm.memset.p1i8.i64(i8 addrspace(1)* align 4 %global.ptr, i8 4, i64 32, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
13define amdgpu_kernel void @memset_global_to_flat(i8 addrspace(1)* %global.ptr, i32 %y) #0 {
14  %cast = addrspacecast i8 addrspace(1)* %global.ptr to i8*
15  call void @llvm.memset.p0i8.i64(i8* align 4 %cast, i8 4, i64 32, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
16  ret void
17}
18
19; CHECK-LABEL: @memset_group_to_flat_no_md(
20; CHECK: call void @llvm.memset.p3i8.i64(i8 addrspace(3)* align 4 %group.ptr, i8 4, i64 %size, i1 false){{$}}
21define amdgpu_kernel void @memset_group_to_flat_no_md(i8 addrspace(3)* %group.ptr, i64 %size) #0 {
22  %cast = addrspacecast i8 addrspace(3)* %group.ptr to i8*
23  call void @llvm.memset.p0i8.i64(i8* align 4 %cast, i8 4, i64 %size, i1 false)
24  ret void
25}
26
27; CHECK-LABEL: @memset_global_to_flat_no_md(
28; CHECK: call void @llvm.memset.p1i8.i64(i8 addrspace(1)* align 4 %global.ptr, i8 4, i64 %size, i1 false){{$}}
29define amdgpu_kernel void @memset_global_to_flat_no_md(i8 addrspace(1)* %global.ptr, i64 %size) #0 {
30  %cast = addrspacecast i8 addrspace(1)* %global.ptr to i8*
31  call void @llvm.memset.p0i8.i64(i8* align 4 %cast, i8 4, i64 %size, i1 false)
32  ret void
33}
34
35; CHECK-LABEL: @memcpy_flat_to_flat_replace_src_with_group(
36; CHECK: call void @llvm.memcpy.p0i8.p3i8.i64(i8* align 4 %dest, i8 addrspace(3)* align 4 %src.group.ptr, i64 %size, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
37define amdgpu_kernel void @memcpy_flat_to_flat_replace_src_with_group(i8* %dest, i8 addrspace(3)* %src.group.ptr, i64 %size) #0 {
38  %cast.src = addrspacecast i8 addrspace(3)* %src.group.ptr to i8*
39  call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %dest, i8* align 4 %cast.src, i64 %size, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
40  ret void
41}
42
43; CHECK-LABEL: @memcpy_flat_to_flat_replace_dest_with_group(
44; CHECK: call void @llvm.memcpy.p3i8.p0i8.i64(i8 addrspace(3)* align 4 %dest.group.ptr, i8* align 4 %src.ptr, i64 %size, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
45define amdgpu_kernel void @memcpy_flat_to_flat_replace_dest_with_group(i8 addrspace(3)* %dest.group.ptr, i8* %src.ptr, i64 %size) #0 {
46  %cast.dest = addrspacecast i8 addrspace(3)* %dest.group.ptr to i8*
47  call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %cast.dest, i8* align 4 %src.ptr, i64 %size, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
48  ret void
49}
50
51; CHECK-LABEL: @memcpy_flat_to_flat_replace_dest_src_with_group(
52; CHECK: call void @llvm.memcpy.p3i8.p3i8.i64(i8 addrspace(3)* align 4 %src.group.ptr, i8 addrspace(3)* align 4 %src.group.ptr, i64 %size, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
53define amdgpu_kernel void @memcpy_flat_to_flat_replace_dest_src_with_group(i8 addrspace(3)* %dest.group.ptr, i8 addrspace(3)* %src.group.ptr, i64 %size) #0 {
54  %cast.src = addrspacecast i8 addrspace(3)* %src.group.ptr to i8*
55  %cast.dest = addrspacecast i8 addrspace(3)* %src.group.ptr to i8*
56  call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %cast.dest, i8* align 4 %cast.src, i64 %size, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
57  ret void
58}
59
60; CHECK-LABEL: @memcpy_flat_to_flat_replace_dest_group_src_global(
61; CHECK: call void @llvm.memcpy.p3i8.p1i8.i64(i8 addrspace(3)* align 4 %dest.group.ptr, i8 addrspace(1)* align 4 %src.global.ptr, i64 %size, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
62define amdgpu_kernel void @memcpy_flat_to_flat_replace_dest_group_src_global(i8 addrspace(3)* %dest.group.ptr, i8 addrspace(1)* %src.global.ptr, i64 %size) #0 {
63  %cast.src = addrspacecast i8 addrspace(1)* %src.global.ptr to i8*
64  %cast.dest = addrspacecast i8 addrspace(3)* %dest.group.ptr to i8*
65  call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %cast.dest, i8* align 4 %cast.src, i64 %size, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
66  ret void
67}
68
69; CHECK-LABEL: @memcpy_group_to_flat_replace_dest_global(
70; CHECK: call void @llvm.memcpy.p1i8.p3i8.i32(i8 addrspace(1)* align 4 %dest.global.ptr, i8 addrspace(3)* align 4 %src.group.ptr, i32 %size, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
71define amdgpu_kernel void @memcpy_group_to_flat_replace_dest_global(i8 addrspace(1)* %dest.global.ptr, i8 addrspace(3)* %src.group.ptr, i32 %size) #0 {
72  %cast.dest = addrspacecast i8 addrspace(1)* %dest.global.ptr to i8*
73  call void @llvm.memcpy.p0i8.p3i8.i32(i8* align 4 %cast.dest, i8 addrspace(3)* align 4 %src.group.ptr, i32 %size, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
74  ret void
75}
76
77; CHECK-LABEL: @memcpy_flat_to_flat_replace_src_with_group_tbaa_struct(
78; CHECK: call void @llvm.memcpy.p0i8.p3i8.i64(i8* align 4 %dest, i8 addrspace(3)* align 4 %src.group.ptr, i64 %size, i1 false), !tbaa.struct !7
79define amdgpu_kernel void @memcpy_flat_to_flat_replace_src_with_group_tbaa_struct(i8* %dest, i8 addrspace(3)* %src.group.ptr, i64 %size) #0 {
80  %cast.src = addrspacecast i8 addrspace(3)* %src.group.ptr to i8*
81  call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %dest, i8* align 4 %cast.src, i64 %size, i1 false), !tbaa.struct !7
82  ret void
83}
84
85; CHECK-LABEL: @memcpy_flat_to_flat_replace_src_with_group_no_md(
86; CHECK: call void @llvm.memcpy.p0i8.p3i8.i64(i8* align 4 %dest, i8 addrspace(3)* align 4 %src.group.ptr, i64 %size, i1 false){{$}}
87define amdgpu_kernel void @memcpy_flat_to_flat_replace_src_with_group_no_md(i8* %dest, i8 addrspace(3)* %src.group.ptr, i64 %size) #0 {
88  %cast.src = addrspacecast i8 addrspace(3)* %src.group.ptr to i8*
89  call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %dest, i8* align 4 %cast.src, i64 %size, i1 false)
90  ret void
91}
92
93; CHECK-LABEL: @multiple_memcpy_flat_to_flat_replace_src_with_group_no_md(
94; CHECK: call void @llvm.memcpy.p0i8.p3i8.i64(i8* align 4 %dest0, i8 addrspace(3)* align 4 %src.group.ptr, i64 %size, i1 false){{$}}
95; CHECK: call void @llvm.memcpy.p0i8.p3i8.i64(i8* align 4 %dest1, i8 addrspace(3)* align 4 %src.group.ptr, i64 %size, i1 false){{$}}
96define amdgpu_kernel void @multiple_memcpy_flat_to_flat_replace_src_with_group_no_md(i8* %dest0, i8* %dest1, i8 addrspace(3)* %src.group.ptr, i64 %size) #0 {
97  %cast.src = addrspacecast i8 addrspace(3)* %src.group.ptr to i8*
98  call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %dest0, i8* align 4 %cast.src, i64 %size, i1 false)
99  call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %dest1, i8* align 4 %cast.src, i64 %size, i1 false)
100  ret void
101}
102
103; Check for iterator problems if the pointer has 2 uses in the same call
104; CHECK-LABEL: @memcpy_group_flat_to_flat_self(
105; CHECK: call void @llvm.memcpy.p3i8.p3i8.i64(i8 addrspace(3)* align 4 %group.ptr, i8 addrspace(3)* align 4 %group.ptr, i64 32, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
106define amdgpu_kernel void @memcpy_group_flat_to_flat_self(i8 addrspace(3)* %group.ptr) #0 {
107  %cast = addrspacecast i8 addrspace(3)* %group.ptr to i8*
108  call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %cast, i8* align 4 %cast, i64 32, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
109  ret void
110}
111; CHECK-LABEL: @memmove_flat_to_flat_replace_src_with_group(
112; CHECK: call void @llvm.memmove.p0i8.p3i8.i64(i8* align 4 %dest, i8 addrspace(3)* align 4 %src.group.ptr, i64 %size, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
113define amdgpu_kernel void @memmove_flat_to_flat_replace_src_with_group(i8* %dest, i8 addrspace(3)* %src.group.ptr, i64 %size) #0 {
114  %cast.src = addrspacecast i8 addrspace(3)* %src.group.ptr to i8*
115  call void @llvm.memmove.p0i8.p0i8.i64(i8* align 4 %dest, i8* align 4 %cast.src, i64 %size, i1 false), !tbaa !0, !alias.scope !3, !noalias !4
116  ret void
117}
118
119declare void @llvm.memset.p0i8.i64(i8* nocapture writeonly, i8, i64, i1) #1
120declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i1) #1
121declare void @llvm.memcpy.p0i8.p3i8.i32(i8* nocapture writeonly, i8 addrspace(3)* nocapture readonly, i32, i1) #1
122declare void @llvm.memmove.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i1) #1
123
124attributes #0 = { nounwind }
125attributes #1 = { argmemonly nounwind }
126
127!0 = !{!1, !1, i64 0}
128!1 = !{!"A", !2}
129!2 = !{!"tbaa root"}
130!3 = !{!"B", !2}
131!4 = !{!5}
132!5 = distinct !{!5, !6, !"some scope"}
133!6 = distinct !{!6, !"some domain"}
134!7 = !{i64 0, i64 8, null}
135