1; RUN: llc -march=amdgcn -mtriple=amdgcn-- -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
2; RUN: llc -march=amdgcn -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(double addrspace(1)* %out, <3 x double> addrspace(1)* %in) #0 {
9  %ld = load volatile <3 x double>, <3 x double> addrspace(1)* %in
10  %elt = extractelement <3 x double> %ld, i32 2
11  store volatile double %elt, double 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 [[IDX:s[0-9]+]], 1
18; GCN-DAG: s_cselect_b64 [[C1:[^,]+]], -1, 0
19; GCN-DAG: s_cmp_eq_u32 [[IDX]], 2
20; GCN-DAG: s_cselect_b64 [[C2:[^,]+]], -1, 0
21; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C1]]
22; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C1]]
23; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C2]]
24; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C2]]
25; GCN: store_dwordx2 v[{{[0-9:]+}}]
26define amdgpu_kernel void @dyn_extract_vector_elt_v3f64(double addrspace(1)* %out, <3 x double> %foo, i32 %elt) #0 {
27  %dynelt = extractelement <3 x double> %foo, i32 %elt
28  store volatile double %dynelt, double addrspace(1)* %out
29  ret void
30}
31
32; GCN-LABEL: {{^}}dyn_extract_vector_elt_v4f64:
33; GCN-NOT: buffer_load
34; GCN-DAG: s_cmp_eq_u32 [[IDX:s[0-9]+]], 1
35; GCN-DAG: s_cselect_b64 [[C1:[^,]+]], -1, 0
36; GCN-DAG: s_cmp_eq_u32 [[IDX]], 2
37; GCN-DAG: s_cselect_b64 [[C2:[^,]+]], -1, 0
38; GCN-DAG: s_cmp_eq_u32 [[IDX]], 3
39; GCN-DAG: s_cselect_b64 [[C3:[^,]+]], -1, 0
40; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C1]]
41; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C1]]
42; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C2]]
43; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C2]]
44; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C3]]
45; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C3]]
46; GCN: store_dwordx2 v[{{[0-9:]+}}]
47define amdgpu_kernel void @dyn_extract_vector_elt_v4f64(double addrspace(1)* %out, <4 x double> %foo, i32 %elt) #0 {
48  %dynelt = extractelement <4 x double> %foo, i32 %elt
49  store volatile double %dynelt, double addrspace(1)* %out
50  ret void
51}
52
53attributes #0 = { nounwind }
54