1; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
2
3; Make sure flat_scratch_init is set
4
5; GCN-LABEL: {{^}}stack_object_addrspacecast_in_kernel_no_calls:
6; GCN:     s_add_u32 flat_scratch_lo, s4, s7
7; GCN:     s_addc_u32 flat_scratch_hi, s5, 0
8; GCN:     flat_store_dword
9; GCN:     .amdhsa_user_sgpr_flat_scratch_init 1
10; GCN:     .amdhsa_system_sgpr_private_segment_wavefront_offset
11; GCN-NOT: .amdhsa_reserve_flat_scratch
12; GCN:     COMPUTE_PGM_RSRC2:SCRATCH_EN: 1
13; GCN:     COMPUTE_PGM_RSRC2:USER_SGPR: 6
14define amdgpu_kernel void @stack_object_addrspacecast_in_kernel_no_calls() {
15  %alloca = alloca i32, addrspace(5)
16  %cast = addrspacecast i32 addrspace(5)* %alloca to i32*
17  store volatile i32 0, i32* %cast
18  ret void
19}
20
21; TODO: Could optimize out in this case
22; GCN-LABEL: {{^}}stack_object_in_kernel_no_calls:
23; GCN:     s_add_u32 flat_scratch_lo, s4, s7
24; GCN:     s_addc_u32 flat_scratch_hi, s5, 0
25; GCN:     buffer_store_dword
26; GCN:     .amdhsa_user_sgpr_private_segment_buffer 1
27; GCN:     .amdhsa_user_sgpr_flat_scratch_init 1
28; GCN:     .amdhsa_system_sgpr_private_segment_wavefront_offset 1
29; GCN-NOT: .amdhsa_reserve_flat_scratch
30; GCN:     COMPUTE_PGM_RSRC2:SCRATCH_EN: 1
31; GCN:     COMPUTE_PGM_RSRC2:USER_SGPR: 6
32define amdgpu_kernel void @stack_object_in_kernel_no_calls() {
33  %alloca = alloca i32, addrspace(5)
34  store volatile i32 0, i32 addrspace(5)* %alloca
35  ret void
36}
37
38; GCN-LABEL: {{^}}kernel_no_calls_no_stack:
39; GCN-NOT: flat_scratch
40; GCN:     .amdhsa_user_sgpr_private_segment_buffer 1
41; GCN:     .amdhsa_user_sgpr_flat_scratch_init 0
42; GCN:     .amdhsa_system_sgpr_private_segment_wavefront_offset 0
43; GCN:     .amdhsa_reserve_flat_scratch 0
44; GCN:     COMPUTE_PGM_RSRC2:SCRATCH_EN: 0
45; GCN:     COMPUTE_PGM_RSRC2:USER_SGPR: 4
46define amdgpu_kernel void @kernel_no_calls_no_stack() {
47  ret void
48}
49