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