1; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -amdgpu-sroa=0 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=VI %s
2; RUN: llc -march=amdgcn -mcpu=hawaii -amdgpu-sroa=0 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=CI %s
3; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -amdgpu-sroa=0 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=GFX9 -check-prefix=VI %s
4
5; Test calls when called by other callable functions rather than
6; kernels.
7
8declare void @external_void_func_i32(i32) #0
9
10; GCN-LABEL: {{^}}test_func_call_external_void_func_i32_imm:
11; GCN: s_waitcnt
12; GCN: s_mov_b32 s5, s32
13; Spill CSR VGPR used for SGPR spilling
14; GCN: buffer_store_dword v32, off, s[0:3], s5 offset:4
15; GCN-DAG: s_add_u32 s32, s32, 0x400
16; GCN-DAG: v_writelane_b32 v32, s33, 0
17; GCN-DAG: v_writelane_b32 v32, s34, 1
18; GCN-DAG: v_writelane_b32 v32, s35, 2
19
20; GCN: s_swappc_b64
21
22; GCN: v_readlane_b32 s35, v32, 2
23; GCN: v_readlane_b32 s34, v32, 1
24; GCN: v_readlane_b32 s33, v32, 0
25; GCN: buffer_load_dword v32, off, s[0:3], s5 offset:4
26; GCN: s_sub_u32 s32, s32, 0x400
27; GCN: s_setpc_b64
28define void @test_func_call_external_void_func_i32_imm() #0 {
29  call void @external_void_func_i32(i32 42)
30  ret void
31}
32
33; GCN-LABEL: {{^}}test_func_call_external_void_func_i32_imm_stack_use:
34; GCN: s_waitcnt
35; GCN: s_mov_b32 s5, s32
36; GCN: s_add_u32 s32, s32, 0x1400{{$}}
37; GCN: buffer_store_dword v{{[0-9]+}}, off, s[0:3], s5 offset
38; GCN: s_swappc_b64
39; GCN: s_sub_u32 s32, s32, 0x1400{{$}}
40; GCN: s_setpc_b64
41define void @test_func_call_external_void_func_i32_imm_stack_use() #0 {
42  %alloca = alloca [16 x i32], align 4, addrspace(5)
43  %gep0 = getelementptr inbounds [16 x i32], [16 x i32] addrspace(5)* %alloca, i32 0, i32 0
44  %gep15 = getelementptr inbounds [16 x i32], [16 x i32] addrspace(5)* %alloca, i32 0, i32 16
45  store volatile i32 0, i32 addrspace(5)* %gep0
46  store volatile i32 0, i32 addrspace(5)* %gep15
47  call void @external_void_func_i32(i32 42)
48  ret void
49}
50
51attributes #0 = { nounwind }
52attributes #1 = { nounwind readnone }
53attributes #2 = { nounwind noinline }
54