1 ; RUN: llc -march=amdgcn -mattr=+promote-alloca,+max-private-element-size-4 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s 2 ; RUN: llc -march=amdgcn -mattr=-promote-alloca,+max-private-element-size-4 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s 3 4 ; Pointer value is stored in a candidate for LDS usage. 5 6 ; GCN-LABEL: {{^}}stored_lds_pointer_value: 7 ; GCN: buffer_store_dword v 8 define amdgpu_kernel void @stored_lds_pointer_value(float addrspace(5)* addrspace(1)* %ptr) #0 { 9 %tmp = alloca float, addrspace(5) 10 store float 0.0, float addrspace(5)*%tmp 11 store float addrspace(5)* %tmp, float addrspace(5)* addrspace(1)* %ptr 12 ret void 13 } 14 15 ; GCN-LABEL: {{^}}stored_lds_pointer_value_offset: 16 ; GCN: buffer_store_dword v 17 define amdgpu_kernel void @stored_lds_pointer_value_offset(float addrspace(5)* addrspace(1)* %ptr) #0 { 18 %tmp0 = alloca float, addrspace(5) 19 %tmp1 = alloca float, addrspace(5) 20 store float 0.0, float addrspace(5)*%tmp0 21 store float 0.0, float addrspace(5)*%tmp1 22 store volatile float addrspace(5)* %tmp0, float addrspace(5)* addrspace(1)* %ptr 23 store volatile float addrspace(5)* %tmp1, float addrspace(5)* addrspace(1)* %ptr 24 ret void 25 } 26 27 ; GCN-LABEL: {{^}}stored_lds_pointer_value_gep: 28 ; GCN-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD0 29 ; GCN-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD1 30 ; GCN: buffer_store_dword v 31 ; GCN: buffer_store_dword v 32 define amdgpu_kernel void @stored_lds_pointer_value_gep(float addrspace(5)* addrspace(1)* %ptr, i32 %idx) #0 { 33 bb: 34 %tmp = alloca float, i32 16, addrspace(5) 35 store float 0.0, float addrspace(5)* %tmp 36 %tmp2 = getelementptr inbounds float, float addrspace(5)* %tmp, i32 %idx 37 store float addrspace(5)* %tmp2, float addrspace(5)* addrspace(1)* %ptr 38 ret void 39 } 40 41 ; Pointer value is stored in a candidate for vector usage 42 ; GCN-LABEL: {{^}}stored_vector_pointer_value: 43 ; GCN-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD0 44 ; GCN-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD1 45 ; GCN: buffer_store_dword 46 ; GCN: buffer_store_dword 47 ; GCN: buffer_store_dword 48 ; GCN: buffer_store_dword 49 define amdgpu_kernel void @stored_vector_pointer_value(i32 addrspace(5)* addrspace(1)* %out, i32 %index) { 50 entry: 51 %tmp0 = alloca [4 x i32], addrspace(5) 52 %x = getelementptr inbounds [4 x i32], [4 x i32] addrspace(5)* %tmp0, i32 0, i32 0 53 %y = getelementptr inbounds [4 x i32], [4 x i32] addrspace(5)* %tmp0, i32 0, i32 1 54 %z = getelementptr inbounds [4 x i32], [4 x i32] addrspace(5)* %tmp0, i32 0, i32 2 55 %w = getelementptr inbounds [4 x i32], [4 x i32] addrspace(5)* %tmp0, i32 0, i32 3 56 store i32 0, i32 addrspace(5)* %x 57 store i32 1, i32 addrspace(5)* %y 58 store i32 2, i32 addrspace(5)* %z 59 store i32 3, i32 addrspace(5)* %w 60 %tmp1 = getelementptr inbounds [4 x i32], [4 x i32] addrspace(5)* %tmp0, i32 0, i32 %index 61 store i32 addrspace(5)* %tmp1, i32 addrspace(5)* addrspace(1)* %out 62 ret void 63 } 64 65 ; GCN-LABEL: {{^}}stored_fi_to_self: 66 ; GCN-NOT: ds_ 67 define amdgpu_kernel void @stored_fi_to_self() #0 { 68 %tmp = alloca i32 addrspace(5)*, addrspace(5) 69 store volatile i32 addrspace(5)* inttoptr (i32 1234 to i32 addrspace(5)*), i32 addrspace(5)* addrspace(5)* %tmp 70 %bitcast = bitcast i32 addrspace(5)* addrspace(5)* %tmp to i32 addrspace(5)* 71 store volatile i32 addrspace(5)* %bitcast, i32 addrspace(5)* addrspace(5)* %tmp 72 ret void 73 } 74 75 attributes #0 = { nounwind } 76
