1 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s 2 3 ; Test that when extracting the same unknown vector index from an 4 ; insertelement the dynamic indexing is folded away. 5 6 declare i32 @llvm.amdgcn.workitem.id.x() #0 7 8 ; No dynamic indexing required 9 ; GCN-LABEL: {{^}}extract_insert_same_dynelt_v4i32: 10 ; GCN: s_load_dword [[VAL:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xd{{$}} 11 ; GCN-NOT buffer_load_dword 12 ; GCN-NOT: [[VAL]] 13 ; GCN: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]] 14 ; GCN-NOT: [[VVAL]] 15 ; GCN: buffer_store_dword [[VVAL]] 16 define void @extract_insert_same_dynelt_v4i32(i32 addrspace(1)* %out, <4 x i32> addrspace(1)* %in, i32 %val, i32 %idx) #1 { 17 %id = call i32 @llvm.amdgcn.workitem.id.x() 18 %id.ext = sext i32 %id to i64 19 %gep.in = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %in, i64 %id.ext 20 %gep.out = getelementptr inbounds i32, i32 addrspace(1)* %out, i64 %id.ext 21 %vec = load <4 x i32>, <4 x i32> addrspace(1)* %gep.in 22 %insert = insertelement <4 x i32> %vec, i32 %val, i32 %idx 23 %extract = extractelement <4 x i32> %insert, i32 %idx 24 store i32 %extract, i32 addrspace(1)* %gep.out 25 ret void 26 } 27 28 ; GCN-LABEL: {{^}}extract_insert_different_dynelt_v4i32: 29 ; GCN: buffer_load_dwordx4 30 ; GCN: v_movreld_b32 31 ; GCN: v_movrels_b32 32 ; GCN: buffer_store_dword v 33 define void @extract_insert_different_dynelt_v4i32(i32 addrspace(1)* %out, <4 x i32> addrspace(1)* %in, i32 %val, i32 %idx0, i32 %idx1) #1 { 34 %id = call i32 @llvm.amdgcn.workitem.id.x() 35 %id.ext = sext i32 %id to i64 36 %gep.in = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %in, i64 %id.ext 37 %gep.out = getelementptr inbounds i32, i32 addrspace(1)* %out, i64 %id.ext 38 %vec = load <4 x i32>, <4 x i32> addrspace(1)* %gep.in 39 %insert = insertelement <4 x i32> %vec, i32 %val, i32 %idx0 40 %extract = extractelement <4 x i32> %insert, i32 %idx1 41 store i32 %extract, i32 addrspace(1)* %gep.out 42 ret void 43 } 44 45 ; GCN-LABEL: {{^}}extract_insert_same_elt2_v4i32: 46 ; GCN: s_load_dword [[VAL:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xd{{$}} 47 ; GCN-NOT buffer_load_dword 48 ; GCN-NOT: [[VAL]] 49 ; GCN: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]] 50 ; GCN-NOT: [[VVAL]] 51 ; GCN: buffer_store_dword [[VVAL]] 52 define void @extract_insert_same_elt2_v4i32(i32 addrspace(1)* %out, <4 x i32> addrspace(1)* %in, i32 %val, i32 %idx) #1 { 53 %id = call i32 @llvm.amdgcn.workitem.id.x() 54 %id.ext = sext i32 %id to i64 55 %gep.in = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %in, i64 %id.ext 56 %gep.out = getelementptr inbounds i32, i32 addrspace(1)* %out, i64 %id.ext 57 %vec = load <4 x i32>, <4 x i32> addrspace(1)* %gep.in 58 %insert = insertelement <4 x i32> %vec, i32 %val, i32 %idx 59 %extract = extractelement <4 x i32> %insert, i32 %idx 60 store i32 %extract, i32 addrspace(1)* %gep.out 61 ret void 62 } 63 64 ; GCN-LABEL: {{^}}extract_insert_same_dynelt_v4f32: 65 ; GCN: s_load_dword [[VAL:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xd{{$}} 66 ; GCN-NOT buffer_load_dword 67 ; GCN-NOT: [[VAL]] 68 ; GCN: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]] 69 ; GCN-NOT: [[VVAL]] 70 ; GCN: buffer_store_dword [[VVAL]] 71 define void @extract_insert_same_dynelt_v4f32(float addrspace(1)* %out, <4 x float> addrspace(1)* %in, float %val, i32 %idx) #1 { 72 %id = call i32 @llvm.amdgcn.workitem.id.x() 73 %id.ext = sext i32 %id to i64 74 %gep.in = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %in, i64 %id.ext 75 %gep.out = getelementptr inbounds float, float addrspace(1)* %out, i64 %id.ext 76 %vec = load volatile <4 x float>, <4 x float> addrspace(1)* %gep.in 77 %insert = insertelement <4 x float> %vec, float %val, i32 %idx 78 %extract = extractelement <4 x float> %insert, i32 %idx 79 store float %extract, float addrspace(1)* %gep.out 80 ret void 81 } 82 83 attributes #0 = { nounwind readnone } 84 attributes #1 = { nounwind }
