1; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=kaveri -verify-machineinstrs < %s | FileCheck -check-prefix=HSA -check-prefix=ALL %s
2; RUN: llc -mtriple=amdgcn-mesa-mesa3d -verify-machineinstrs < %s | FileCheck -check-prefix=MESA -check-prefix=ALL %s
3
4; ALL-LABEL: {{^}}test:
5; HSA: enable_sgpr_kernarg_segment_ptr = 1
6; HSA: s_load_dword s{{[0-9]+}}, s[4:5], 0xa
7
8; MESA: s_load_dword s{{[0-9]+}}, s[0:1], 0xa
9define void @test(i32 addrspace(1)* %out) #1 {
10  %kernarg.segment.ptr = call noalias i8 addrspace(2)* @llvm.amdgcn.kernarg.segment.ptr()
11  %header.ptr = bitcast i8 addrspace(2)* %kernarg.segment.ptr to i32 addrspace(2)*
12  %gep = getelementptr i32, i32 addrspace(2)* %header.ptr, i64 10
13  %value = load i32, i32 addrspace(2)* %gep
14  store i32 %value, i32 addrspace(1)* %out
15  ret void
16}
17
18; ALL-LABEL: {{^}}test_implicit:
19; 10 + 9 (36 prepended implicit bytes) + 2(out pointer) = 21 = 0x15
20; MESA: s_load_dword s{{[0-9]+}}, s[0:1], 0x15
21define void @test_implicit(i32 addrspace(1)* %out) #1 {
22  %implicitarg.ptr = call noalias i8 addrspace(2)* @llvm.amdgcn.implicitarg.ptr()
23  %header.ptr = bitcast i8 addrspace(2)* %implicitarg.ptr to i32 addrspace(2)*
24  %gep = getelementptr i32, i32 addrspace(2)* %header.ptr, i64 10
25  %value = load i32, i32 addrspace(2)* %gep
26  store i32 %value, i32 addrspace(1)* %out
27  ret void
28}
29
30; ALL-LABEL: {{^}}test_implicit_alignment
31; MESA: s_load_dword [[VAL:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0xc
32; HSA: s_load_dword [[VAL:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0x4
33; ALL: v_mov_b32_e32 [[V_VAL:v[0-9]+]], [[VAL]]
34; MESA: buffer_store_dword [[V_VAL]]
35; HSA: flat_store_dword v[{{[0-9]+:[0-9]+}}], [[V_VAL]]
36define void @test_implicit_alignment(i32 addrspace(1)* %out, <2 x i8> %in) #1 {
37  %implicitarg.ptr = call noalias i8 addrspace(2)* @llvm.amdgcn.implicitarg.ptr()
38  %arg.ptr = bitcast i8 addrspace(2)* %implicitarg.ptr to i32 addrspace(2)*
39  %val = load i32, i32 addrspace(2)* %arg.ptr
40  store i32 %val, i32 addrspace(1)* %out
41  ret void
42}
43
44declare i8 addrspace(2)* @llvm.amdgcn.kernarg.segment.ptr() #0
45declare i8 addrspace(2)* @llvm.amdgcn.implicitarg.ptr() #0
46
47attributes #0 = { nounwind readnone }
48attributes #1 = { nounwind }
49