1; RUN: llc -mtriple=amdgcn-- -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s 2; RUN: llc -mtriple=amdgcn-- -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s 3 4; GCN-LABEL: {{^}}extract_vector_elt_v3f64_2: 5; GCN: buffer_load_dwordx4 6; GCN: buffer_load_dwordx2 7; GCN: buffer_store_dwordx2 8define amdgpu_kernel void @extract_vector_elt_v3f64_2(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { 9 %ld = load volatile <3 x double>, ptr addrspace(1) %in 10 %elt = extractelement <3 x double> %ld, i32 2 11 store volatile double %elt, ptr addrspace(1) %out 12 ret void 13} 14 15; GCN-LABEL: {{^}}dyn_extract_vector_elt_v3f64: 16; GCN-NOT: buffer_load 17; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 1 18; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}} 19; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}} 20; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 2 21; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}} 22; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}} 23; GCN: store_dwordx2 v[{{[0-9:]+}}] 24define amdgpu_kernel void @dyn_extract_vector_elt_v3f64(ptr addrspace(1) %out, <3 x double> %foo, i32 %elt) #0 { 25 %dynelt = extractelement <3 x double> %foo, i32 %elt 26 store volatile double %dynelt, ptr addrspace(1) %out 27 ret void 28} 29 30; GCN-LABEL: {{^}}dyn_extract_vector_elt_v4f64: 31; GCN-NOT: buffer_load 32; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 1 33; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}} 34; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}} 35; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 2 36; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}} 37; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}} 38; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 3 39; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}} 40; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}} 41; GCN: store_dwordx2 v[{{[0-9:]+}}] 42define amdgpu_kernel void @dyn_extract_vector_elt_v4f64(ptr addrspace(1) %out, <4 x double> %foo, i32 %elt) #0 { 43 %dynelt = extractelement <4 x double> %foo, i32 %elt 44 store volatile double %dynelt, ptr addrspace(1) %out 45 ret void 46} 47 48attributes #0 = { nounwind } 49