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 s4, SCRATCH_RSRC_DWORD0
7; GCN-DAG: s_mov_b32 s5, SCRATCH_RSRC_DWORD1
8; GCN-DAG: s_mov_b32 s6, -1
9
10; CI-DAG: s_mov_b32 s7, 0xe8f000
11; VI-DAG: s_mov_b32 s7, 0xe80000
12; GFX9-DAG: s_mov_b32 s7, 0xe00000
13
14; GCN: s_add_u32 s4, s4, s0
15; GCN: s_addc_u32 s5, s5, 0
16
17; GCN: buffer_store_dword {{v[0-9]+}}, {{v[0-9]+}}, s[4:7], 0 offen
18; GCN: buffer_load_dword {{v[0-9]+}}, {{v[0-9]+}}, s[4:7], 0 offen
19
20; ALL: ; ScratchSize: 32772
21define amdgpu_ps void @large_alloca_pixel_shader(i32 %x, i32 %y) #0 {
22  %large = alloca [8192 x i32], align 4, addrspace(5)
23  %gep = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %large, i32 0, i32 8191
24  store volatile i32 %x, i32 addrspace(5)* %gep
25  %gep1 = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %large, i32 0, i32 %y
26  %val = load volatile i32, i32 addrspace(5)* %gep1
27  store volatile i32 %val, i32 addrspace(1)* undef
28  ret void
29}
30
31; ALL-LABEL: {{^}}large_alloca_pixel_shader_inreg:
32; GCN-DAG: s_mov_b32 s4, SCRATCH_RSRC_DWORD0
33; GCN-DAG: s_mov_b32 s5, SCRATCH_RSRC_DWORD1
34; GCN-DAG: s_mov_b32 s6, -1
35
36; CI-DAG: s_mov_b32 s7, 0xe8f000
37; VI-DAG: s_mov_b32 s7, 0xe80000
38; GFX9-DAG: s_mov_b32 s7, 0xe00000
39
40; GCN: s_add_u32 s4, s4, s2
41; GCN: s_addc_u32 s5, s5, 0
42
43; GCN: buffer_store_dword {{v[0-9]+}}, {{v[0-9]+}}, s[4:7], 0 offen
44; GCN: buffer_load_dword {{v[0-9]+}}, {{v[0-9]+}}, s[4:7], 0 offen
45
46; ALL: ; ScratchSize: 32772
47define amdgpu_ps void @large_alloca_pixel_shader_inreg(i32 inreg %x, i32 inreg %y) #0 {
48  %large = alloca [8192 x i32], align 4, addrspace(5)
49  %gep = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %large, i32 0, i32 8191
50  store volatile i32 %x, i32 addrspace(5)* %gep
51  %gep1 = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %large, i32 0, i32 %y
52  %val = load volatile i32, i32 addrspace(5)* %gep1
53  store volatile i32 %val, i32 addrspace(1)* undef
54  ret void
55}
56
57attributes #0 = { nounwind  }
58