| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 | |
| ; RUN: llc -global-isel=0 -mtriple=amdgpu7.00-amd-amdhsa < %s | FileCheck %s | |
| ; RUN: llc -global-isel=1 -mtriple=amdgpu7.00-amd-amdhsa < %s | FileCheck %s | |
| ; Check illegal casts are codegened as poison, and not an error. | |
| define amdgpu_kernel void @use_group_to_global_addrspacecast(ptr addrspace(3) %ptr) { | |
| ; CHECK-LABEL: use_group_to_global_addrspacecast: | |
| ; CHECK: ; %bb.0: | |
| ; CHECK-NEXT: s_add_i32 s12, s12, s17 | |
| ; CHECK-NEXT: s_mov_b32 flat_scratch_lo, s13 | |
| ; CHECK-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 | |
| ; CHECK-NEXT: v_mov_b32_e32 v0, 0 | |
| ; CHECK-NEXT: flat_store_dword v[0:1], v0 | |
| ; CHECK-NEXT: s_waitcnt vmcnt(0) | |
| ; CHECK-NEXT: s_endpgm | |
| %stof = addrspacecast ptr addrspace(3) %ptr to ptr addrspace(1) | |
| store volatile i32 0, ptr addrspace(1) %stof | |
| ret void | |
| } | |
| define amdgpu_kernel void @use_local_to_constant32bit_addrspacecast(ptr addrspace(3) %ptr) { | |
| ; CHECK-LABEL: use_local_to_constant32bit_addrspacecast: | |
| ; CHECK: ; %bb.0: | |
| ; CHECK-NEXT: s_mov_b32 s1, 0 | |
| ; CHECK-NEXT: s_load_dword s0, s[0:1], 0x0 | |
| ; CHECK-NEXT: s_endpgm | |
| %stof = addrspacecast ptr addrspace(3) %ptr to ptr addrspace(6) | |
| %load = load volatile i32, ptr addrspace(6) %stof | |
| ret void | |
| } | |
| define amdgpu_kernel void @use_constant32bit_to_local_addrspacecast(ptr addrspace(6) %ptr) { | |
| ; CHECK-LABEL: use_constant32bit_to_local_addrspacecast: | |
| ; CHECK: ; %bb.0: | |
| ; CHECK-NEXT: s_mov_b32 m0, -1 | |
| ; CHECK-NEXT: ds_read_b32 v0, v0 | |
| ; CHECK-NEXT: s_endpgm | |
| %cast = addrspacecast ptr addrspace(6) %ptr to ptr addrspace(3) | |
| %load = load volatile i32, ptr addrspace(3) %cast | |
| ret void | |
| } | |
| define amdgpu_kernel void @use_local_to_42_addrspacecast(ptr addrspace(3) %ptr) { | |
| ; CHECK-LABEL: use_local_to_42_addrspacecast: | |
| ; CHECK: ; %bb.0: | |
| ; CHECK-NEXT: s_add_i32 s12, s12, s17 | |
| ; CHECK-NEXT: s_mov_b32 flat_scratch_lo, s13 | |
| ; CHECK-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 | |
| ; CHECK-NEXT: v_mov_b32_e32 v0, 0 | |
| ; CHECK-NEXT: v_mov_b32_e32 v1, 0 | |
| ; CHECK-NEXT: flat_store_dwordx2 v[0:1], v[0:1] | |
| ; CHECK-NEXT: s_waitcnt vmcnt(0) | |
| ; CHECK-NEXT: s_endpgm | |
| %cast = addrspacecast ptr addrspace(3) %ptr to ptr addrspace(42) | |
| store volatile ptr addrspace(42) %cast, ptr addrspace(1) null | |
| ret void | |
| } | |
| define amdgpu_kernel void @use_42_to_local_addrspacecast(ptr addrspace(42) %ptr) { | |
| ; CHECK-LABEL: use_42_to_local_addrspacecast: | |
| ; CHECK: ; %bb.0: | |
| ; CHECK-NEXT: s_mov_b32 m0, -1 | |
| ; CHECK-NEXT: ds_read_b32 v0, v0 | |
| ; CHECK-NEXT: s_endpgm | |
| %cast = addrspacecast ptr addrspace(42) %ptr to ptr addrspace(3) | |
| %load = load volatile i32, ptr addrspace(3) %cast | |
| ret void | |
| } |