1 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s 2 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s 3 4 ; FUNC-LABEL: {{^}}extract_vector_elt_v2i16: 5 ; GCN: buffer_load_ushort 6 ; GCN: buffer_load_ushort 7 ; GCN: buffer_store_short 8 ; GCN: buffer_store_short 9 define void @extract_vector_elt_v2i16(i16 addrspace(1)* %out, <2 x i16> %foo) #0 { 10 %p0 = extractelement <2 x i16> %foo, i32 0 11 %p1 = extractelement <2 x i16> %foo, i32 1 12 %out1 = getelementptr i16, i16 addrspace(1)* %out, i32 10 13 store i16 %p1, i16 addrspace(1)* %out, align 2 14 store i16 %p0, i16 addrspace(1)* %out1, align 2 15 ret void 16 } 17 18 ; FUNC-LABEL: {{^}}extract_vector_elt_v3i16: 19 ; GCN: buffer_load_ushort 20 ; GCN: buffer_store_short 21 ; GCN: buffer_store_short 22 define void @extract_vector_elt_v3i16(i16 addrspace(1)* %out, <3 x i16> %foo) #0 { 23 %p0 = extractelement <3 x i16> %foo, i32 0 24 %p1 = extractelement <3 x i16> %foo, i32 2 25 %out1 = getelementptr i16, i16 addrspace(1)* %out, i32 1 26 store i16 %p1, i16 addrspace(1)* %out, align 2 27 store i16 %p0, i16 addrspace(1)* %out1, align 2 28 ret void 29 } 30 31 ; FUNC-LABEL: {{^}}extract_vector_elt_v4i16: 32 ; GCN: buffer_load_ushort 33 ; GCN: buffer_load_ushort 34 ; GCN: buffer_store_short 35 ; GCN: buffer_store_short 36 define void @extract_vector_elt_v4i16(i16 addrspace(1)* %out, <4 x i16> %foo) #0 { 37 %p0 = extractelement <4 x i16> %foo, i32 0 38 %p1 = extractelement <4 x i16> %foo, i32 2 39 %out1 = getelementptr i16, i16 addrspace(1)* %out, i32 10 40 store i16 %p1, i16 addrspace(1)* %out, align 2 41 store i16 %p0, i16 addrspace(1)* %out1, align 2 42 ret void 43 } 44 45 46 ; FUNC-LABEL: {{^}}dynamic_extract_vector_elt_v3i16: 47 ; GCN: buffer_load_ushort 48 ; GCN: buffer_load_ushort 49 ; GCN: buffer_load_ushort 50 51 ; GCN: buffer_store_short 52 ; GCN: buffer_store_short 53 ; GCN: buffer_store_short 54 55 ; GCN: buffer_store_short 56 ; GCN: buffer_load_ushort 57 ; GCN: buffer_store_short 58 define void @dynamic_extract_vector_elt_v3i16(i16 addrspace(1)* %out, <3 x i16> %foo, i32 %idx) #0 { 59 %p0 = extractelement <3 x i16> %foo, i32 %idx 60 %out1 = getelementptr i16, i16 addrspace(1)* %out, i32 1 61 store i16 %p0, i16 addrspace(1)* %out 62 ret void 63 } 64 65 ; FUNC-LABEL: {{^}}dynamic_extract_vector_elt_v4i16: 66 ; GCN: buffer_load_ushort 67 ; GCN: buffer_load_ushort 68 ; GCN: buffer_load_ushort 69 ; GCN: buffer_load_ushort 70 71 ; GCN: buffer_store_short 72 ; GCN: buffer_store_short 73 ; GCN: buffer_store_short 74 ; GCN: buffer_store_short 75 76 ; GCN: buffer_store_short 77 ; GCN: buffer_load_ushort 78 ; GCN: buffer_store_short 79 define void @dynamic_extract_vector_elt_v4i16(i16 addrspace(1)* %out, <4 x i16> %foo, i32 %idx) #0 { 80 %p0 = extractelement <4 x i16> %foo, i32 %idx 81 %out1 = getelementptr i16, i16 addrspace(1)* %out, i32 1 82 store i16 %p0, i16 addrspace(1)* %out 83 ret void 84 } 85 86 attributes #0 = { nounwind } 87
