1; RUN: llc -march=amdgcn -mcpu=bonaire < %s | FileCheck -check-prefix=GCN -check-prefix=CI -check-prefix=ALL %s
2; RUN: llc -march=amdgcn -mcpu=carrizo -mattr=-flat-for-global < %s | FileCheck -check-prefix=GCN -check-prefix=VI -check-prefix=ALL %s
3; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -check-prefix=ALL %s
4
5; ALL-LABEL: {{^}}large_alloca_pixel_shader:
6; GCN-DAG: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
7; GCN-DAG: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
8; GCN-DAG: s_mov_b32 s10, -1
9; CI-DAG: s_mov_b32 s11, 0xe8f000
10; VI-DAG: s_mov_b32 s11, 0xe80000
11; GFX9-DAG: s_mov_b32 s11, 0xe00000
12
13; GCN: buffer_store_dword {{v[0-9]+}}, {{v[0-9]+}}, s[8:11], s0 offen
14; GCN: buffer_load_dword {{v[0-9]+}}, {{v[0-9]+}}, s[8:11], s0 offen
15
16; ALL: ; ScratchSize: 32772
17define amdgpu_ps void @large_alloca_pixel_shader(i32 %x, i32 %y) #0 {
18  %large = alloca [8192 x i32], align 4, addrspace(5)
19  %gep = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %large, i32 0, i32 8191
20  store volatile i32 %x, i32 addrspace(5)* %gep
21  %gep1 = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %large, i32 0, i32 %y
22  %val = load volatile i32, i32 addrspace(5)* %gep1
23  store volatile i32 %val, i32 addrspace(1)* undef
24  ret void
25}
26
27; ALL-LABEL: {{^}}large_alloca_pixel_shader_inreg:
28; GCN-DAG: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
29; GCN-DAG: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
30; GCN-DAG: s_mov_b32 s10, -1
31; CI-DAG: s_mov_b32 s11, 0xe8f000
32; VI-DAG: s_mov_b32 s11, 0xe80000
33; GFX9-DAG: s_mov_b32 s11, 0xe00000
34
35; GCN: buffer_store_dword {{v[0-9]+}}, {{v[0-9]+}}, s[8:11], s2 offen
36; GCN: buffer_load_dword {{v[0-9]+}}, {{v[0-9]+}}, s[8:11], s2 offen
37
38; ALL: ; ScratchSize: 32772
39define amdgpu_ps void @large_alloca_pixel_shader_inreg(i32 inreg %x, i32 inreg %y) #0 {
40  %large = alloca [8192 x i32], align 4, addrspace(5)
41  %gep = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %large, i32 0, i32 8191
42  store volatile i32 %x, i32 addrspace(5)* %gep
43  %gep1 = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %large, i32 0, i32 %y
44  %val = load volatile i32, i32 addrspace(5)* %gep1
45  store volatile i32 %val, i32 addrspace(1)* undef
46  ret void
47}
48
49attributes #0 = { nounwind  }
50