Home | History | Annotate | Download | only in AMDGPU
      1 ; RUN: llc -march=amdgcn -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -verify-machineinstrs < %s | FileCheck -check-prefix=ALL -check-prefix=HSA -check-prefix=CI-HSA  %s
      2 ; RUN: llc -march=amdgcn -mtriple=amdgcn-unknown-amdhsa -mcpu=carrizo -verify-machineinstrs < %s | FileCheck -check-prefix=ALL -check-prefix=HSA -check-prefix=VI-HSA  %s
      3 ; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=ALL -check-prefix=MESA -check-prefix=SI-MESA %s
      4 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=ALL -check-prefix=MESA -check-prefix=VI-MESA %s
      5 
      6 declare i32 @llvm.amdgcn.workgroup.id.x() #0
      7 declare i32 @llvm.amdgcn.workgroup.id.y() #0
      8 declare i32 @llvm.amdgcn.workgroup.id.z() #0
      9 
     10 ; ALL-LABEL {{^}}test_workgroup_id_x:
     11 
     12 ; HSA: .amd_kernel_code_t
     13 ; HSA: compute_pgm_rsrc2_user_sgpr = 6
     14 ; HSA: compute_pgm_rsrc2_tgid_x_en = 1
     15 ; HSA: compute_pgm_rsrc2_tgid_y_en = 0
     16 ; HSA: compute_pgm_rsrc2_tgid_z_en = 0
     17 ; HSA: compute_pgm_rsrc2_tg_size_en = 0
     18 ; HSA: compute_pgm_rsrc2_tidig_comp_cnt = 0
     19 ; HSA: enable_sgpr_grid_workgroup_count_x = 0
     20 ; HSA: enable_sgpr_grid_workgroup_count_y = 0
     21 ; HSA: enable_sgpr_grid_workgroup_count_z = 0
     22 ; HSA: .end_amd_kernel_code_t
     23 
     24 ; MESA: v_mov_b32_e32 [[VCOPY:v[0-9]+]], s2{{$}}
     25 ; HSA: v_mov_b32_e32 [[VCOPY:v[0-9]+]], s6{{$}}
     26 
     27 ; ALL-NOT: [[VCOPY]]
     28 ; ALL: {{buffer|flat}}_store_dword {{.*}}[[VCOPY]]
     29 
     30 ; HSA: COMPUTE_PGM_RSRC2:USER_SGPR: 6
     31 ; ALL-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
     32 ; ALL: COMPUTE_PGM_RSRC2:TGID_X_EN: 1
     33 ; ALL: COMPUTE_PGM_RSRC2:TGID_Y_EN: 0
     34 ; ALL: COMPUTE_PGM_RSRC2:TGID_Z_EN: 0
     35 ; ALL: COMPUTE_PGM_RSRC2:TIDIG_COMP_CNT: 0
     36 define void @test_workgroup_id_x(i32 addrspace(1)* %out) #1 {
     37   %id = call i32 @llvm.amdgcn.workgroup.id.x()
     38   store i32 %id, i32 addrspace(1)* %out
     39   ret void
     40 }
     41 
     42 ; ALL-LABEL {{^}}test_workgroup_id_y:
     43 ; HSA: compute_pgm_rsrc2_user_sgpr = 6
     44 ; HSA: compute_pgm_rsrc2_tgid_x_en = 1
     45 ; HSA: compute_pgm_rsrc2_tgid_y_en = 1
     46 ; HSA: compute_pgm_rsrc2_tgid_z_en = 0
     47 ; HSA: compute_pgm_rsrc2_tg_size_en = 0
     48 ; HSA: enable_sgpr_grid_workgroup_count_x = 0
     49 ; HSA: enable_sgpr_grid_workgroup_count_y = 0
     50 ; HSA: enable_sgpr_grid_workgroup_count_z = 0
     51 
     52 ; MESA: v_mov_b32_e32 [[VCOPY:v[0-9]+]], s3{{$}}
     53 ; HSA: v_mov_b32_e32 [[VCOPY:v[0-9]+]], s7{{$}}
     54 
     55 ; ALL-NOT: [[VCOPY]]
     56 ; ALL: {{buffer|flat}}_store_dword {{.*}}[[VCOPY]]
     57 
     58 ; HSA: COMPUTE_PGM_RSRC2:USER_SGPR: 6
     59 ; ALL-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
     60 ; ALL: COMPUTE_PGM_RSRC2:TGID_X_EN: 1
     61 ; ALL: COMPUTE_PGM_RSRC2:TGID_Y_EN: 1
     62 ; ALL: COMPUTE_PGM_RSRC2:TGID_Z_EN: 0
     63 ; ALL: COMPUTE_PGM_RSRC2:TIDIG_COMP_CNT: 0
     64 define void @test_workgroup_id_y(i32 addrspace(1)* %out) #1 {
     65   %id = call i32 @llvm.amdgcn.workgroup.id.y()
     66   store i32 %id, i32 addrspace(1)* %out
     67   ret void
     68 }
     69 
     70 ; ALL-LABEL {{^}}test_workgroup_id_z:
     71 ; HSA: compute_pgm_rsrc2_user_sgpr = 6
     72 ; HSA: compute_pgm_rsrc2_tgid_x_en = 1
     73 ; HSA: compute_pgm_rsrc2_tgid_y_en = 0
     74 ; HSA: compute_pgm_rsrc2_tgid_z_en = 1
     75 ; HSA: compute_pgm_rsrc2_tg_size_en = 0
     76 ; HSA: compute_pgm_rsrc2_tidig_comp_cnt = 0
     77 ; HSA: enable_sgpr_private_segment_buffer = 1
     78 ; HSA: enable_sgpr_dispatch_ptr = 0
     79 ; HSA: enable_sgpr_queue_ptr = 0
     80 ; HSA: enable_sgpr_kernarg_segment_ptr = 1
     81 ; HSA: enable_sgpr_dispatch_id = 0
     82 ; HSA: enable_sgpr_flat_scratch_init = 0
     83 ; HSA: enable_sgpr_private_segment_size = 0
     84 ; HSA: enable_sgpr_grid_workgroup_count_x = 0
     85 ; HSA: enable_sgpr_grid_workgroup_count_y = 0
     86 ; HSA: enable_sgpr_grid_workgroup_count_z = 0
     87 
     88 ; MESA: v_mov_b32_e32 [[VCOPY:v[0-9]+]], s3{{$}}
     89 ; HSA: v_mov_b32_e32 [[VCOPY:v[0-9]+]], s7{{$}}
     90 
     91 ; ALL-NOT: [[VCOPY]]
     92 ; ALL: {{buffer|flat}}_store_dword {{.*}}[[VCOPY]]
     93 
     94 ; HSA: COMPUTE_PGM_RSRC2:USER_SGPR: 6
     95 ; ALL-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
     96 ; ALL: COMPUTE_PGM_RSRC2:TGID_X_EN: 1
     97 ; ALL: COMPUTE_PGM_RSRC2:TGID_Y_EN: 0
     98 ; ALL: COMPUTE_PGM_RSRC2:TGID_Z_EN: 1
     99 ; ALL: COMPUTE_PGM_RSRC2:TIDIG_COMP_CNT: 0
    100 define void @test_workgroup_id_z(i32 addrspace(1)* %out) #1 {
    101   %id = call i32 @llvm.amdgcn.workgroup.id.z()
    102   store i32 %id, i32 addrspace(1)* %out
    103   ret void
    104 }
    105 
    106 attributes #0 = { nounwind readnone }
    107 attributes #1 = { nounwind }
    108