1; RUN: opt -mtriple=amdgcn-unknown-amdhsa -S -amdgpu-annotate-kernel-features < %s | FileCheck -check-prefix=HSA %s
2
3declare void @llvm.memcpy.p1i32.p4i32.i32(i32 addrspace(1)* nocapture, i32 addrspace(4)* nocapture, i32, i1) #0
4
5@lds.i32 = unnamed_addr addrspace(3) global i32 undef, align 4
6@lds.arr = unnamed_addr addrspace(3) global [256 x i32] undef, align 4
7
8@global.i32 = unnamed_addr addrspace(1) global i32 undef, align 4
9@global.arr = unnamed_addr addrspace(1) global [256 x i32] undef, align 4
10
11; HSA: @store_cast_0_flat_to_group_addrspacecast() #1
12define amdgpu_kernel void @store_cast_0_flat_to_group_addrspacecast() #1 {
13  store i32 7, i32 addrspace(3)* addrspacecast (i32 addrspace(4)* null to i32 addrspace(3)*)
14  ret void
15}
16
17; HSA: @store_cast_0_group_to_flat_addrspacecast() #2
18define amdgpu_kernel void @store_cast_0_group_to_flat_addrspacecast() #1 {
19  store i32 7, i32 addrspace(4)* addrspacecast (i32 addrspace(3)* null to i32 addrspace(4)*)
20  ret void
21}
22
23; HSA: define amdgpu_kernel void @store_constant_cast_group_gv_to_flat() #2
24define amdgpu_kernel void @store_constant_cast_group_gv_to_flat() #1 {
25  store i32 7, i32 addrspace(4)* addrspacecast (i32 addrspace(3)* @lds.i32 to i32 addrspace(4)*)
26  ret void
27}
28
29; HSA: @store_constant_cast_group_gv_gep_to_flat() #2
30define amdgpu_kernel void @store_constant_cast_group_gv_gep_to_flat() #1 {
31  store i32 7, i32 addrspace(4)* getelementptr ([256 x i32], [256 x i32] addrspace(4)* addrspacecast ([256 x i32] addrspace(3)* @lds.arr to [256 x i32] addrspace(4)*), i64 0, i64 8)
32  ret void
33}
34
35; HSA: @store_constant_cast_global_gv_to_flat() #1
36define amdgpu_kernel void @store_constant_cast_global_gv_to_flat() #1 {
37  store i32 7, i32 addrspace(4)* addrspacecast (i32 addrspace(1)* @global.i32 to i32 addrspace(4)*)
38  ret void
39}
40
41; HSA: @store_constant_cast_global_gv_gep_to_flat() #1
42define amdgpu_kernel void @store_constant_cast_global_gv_gep_to_flat() #1 {
43  store i32 7, i32 addrspace(4)* getelementptr ([256 x i32], [256 x i32] addrspace(4)* addrspacecast ([256 x i32] addrspace(1)* @global.arr to [256 x i32] addrspace(4)*), i64 0, i64 8)
44  ret void
45}
46
47; HSA: @load_constant_cast_group_gv_gep_to_flat(i32 addrspace(1)* %out) #2
48define amdgpu_kernel void @load_constant_cast_group_gv_gep_to_flat(i32 addrspace(1)* %out) #1 {
49  %val = load i32, i32 addrspace(4)* getelementptr ([256 x i32], [256 x i32] addrspace(4)* addrspacecast ([256 x i32] addrspace(3)* @lds.arr to [256 x i32] addrspace(4)*), i64 0, i64 8)
50  store i32 %val, i32 addrspace(1)* %out
51  ret void
52}
53
54; HSA: @atomicrmw_constant_cast_group_gv_gep_to_flat(i32 addrspace(1)* %out) #2
55define amdgpu_kernel void @atomicrmw_constant_cast_group_gv_gep_to_flat(i32 addrspace(1)* %out) #1 {
56  %val = atomicrmw add i32 addrspace(4)* getelementptr ([256 x i32], [256 x i32] addrspace(4)* addrspacecast ([256 x i32] addrspace(3)* @lds.arr to [256 x i32] addrspace(4)*), i64 0, i64 8), i32 1 seq_cst
57  store i32 %val, i32 addrspace(1)* %out
58  ret void
59}
60
61; HSA: @cmpxchg_constant_cast_group_gv_gep_to_flat(i32 addrspace(1)* %out) #2
62define amdgpu_kernel void @cmpxchg_constant_cast_group_gv_gep_to_flat(i32 addrspace(1)* %out) #1 {
63  %val = cmpxchg i32 addrspace(4)* getelementptr ([256 x i32], [256 x i32] addrspace(4)* addrspacecast ([256 x i32] addrspace(3)* @lds.arr to [256 x i32] addrspace(4)*), i64 0, i64 8), i32 0, i32 1 seq_cst seq_cst
64  %val0 = extractvalue { i32, i1 } %val, 0
65  store i32 %val0, i32 addrspace(1)* %out
66  ret void
67}
68
69; HSA: @memcpy_constant_cast_group_gv_gep_to_flat(i32 addrspace(1)* %out) #2
70define amdgpu_kernel void @memcpy_constant_cast_group_gv_gep_to_flat(i32 addrspace(1)* %out) #1 {
71  call void @llvm.memcpy.p1i32.p4i32.i32(i32 addrspace(1)* align 4 %out, i32 addrspace(4)* align 4 getelementptr ([256 x i32], [256 x i32] addrspace(4)* addrspacecast ([256 x i32] addrspace(3)* @lds.arr to [256 x i32] addrspace(4)*), i64 0, i64 8), i32 32, i1 false)
72  ret void
73}
74
75; Can't just search the pointer value
76; HSA: @store_value_constant_cast_lds_gv_gep_to_flat(i32 addrspace(4)* addrspace(1)* %out) #2
77define amdgpu_kernel void @store_value_constant_cast_lds_gv_gep_to_flat(i32 addrspace(4)* addrspace(1)* %out) #1 {
78  store i32 addrspace(4)* getelementptr ([256 x i32], [256 x i32] addrspace(4)* addrspacecast ([256 x i32] addrspace(3)* @lds.arr to [256 x i32] addrspace(4)*), i64 0, i64 8), i32 addrspace(4)* addrspace(1)* %out
79  ret void
80}
81
82; Can't just search pointer types
83; HSA: @store_ptrtoint_value_constant_cast_lds_gv_gep_to_flat(i64 addrspace(1)* %out) #2
84define amdgpu_kernel void @store_ptrtoint_value_constant_cast_lds_gv_gep_to_flat(i64 addrspace(1)* %out) #1 {
85  store i64 ptrtoint (i32 addrspace(4)* getelementptr ([256 x i32], [256 x i32] addrspace(4)* addrspacecast ([256 x i32] addrspace(3)* @lds.arr to [256 x i32] addrspace(4)*), i64 0, i64 8) to i64), i64 addrspace(1)* %out
86  ret void
87}
88
89; Cast group to flat, do GEP, cast back to group
90; HSA: @store_constant_cast_group_gv_gep_to_flat_to_group() #2
91define amdgpu_kernel void @store_constant_cast_group_gv_gep_to_flat_to_group() #1 {
92  store i32 7, i32 addrspace(3)* addrspacecast (i32 addrspace(4)* getelementptr ([256 x i32], [256 x i32] addrspace(4)* addrspacecast ([256 x i32] addrspace(3)* @lds.arr to [256 x i32] addrspace(4)*), i64 0, i64 8) to i32 addrspace(3)*)
93  ret void
94}
95
96; HSA: @ret_constant_cast_group_gv_gep_to_flat_to_group() #2
97define i32 addrspace(3)* @ret_constant_cast_group_gv_gep_to_flat_to_group() #1 {
98  ret i32 addrspace(3)* addrspacecast (i32 addrspace(4)* getelementptr ([256 x i32], [256 x i32] addrspace(4)* addrspacecast ([256 x i32] addrspace(3)* @lds.arr to [256 x i32] addrspace(4)*), i64 0, i64 8) to i32 addrspace(3)*)
99}
100
101; HSA: attributes #0 = { argmemonly nofree nosync nounwind willreturn }
102; HSA: attributes #1 = { nounwind }
103; HSA: attributes #2 = { nounwind "amdgpu-queue-ptr" }
104
105attributes #0 = { argmemonly nounwind }
106attributes #1 = { nounwind }
107