diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/sitofp.f16.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/sitofp.f16.ll | 51 |
1 files changed, 34 insertions, 17 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/sitofp.f16.ll b/llvm/test/CodeGen/AMDGPU/sitofp.f16.ll index bbe8d7ce469..574d1c0b2c7 100644 --- a/llvm/test/CodeGen/AMDGPU/sitofp.f16.ll +++ b/llvm/test/CodeGen/AMDGPU/sitofp.f16.ll @@ -37,14 +37,23 @@ entry: ; GCN-LABEL: {{^}}sitofp_v2i16_to_v2f16 ; GCN: buffer_load_dword -; GCN: v_cvt_f32_i32_e32 -; GCN: v_cvt_f32_i32_e32 -; GCN: v_cvt_f16_f32_e32 -; GCN: v_cvt_f16_f32_e32 -; GCN-DAG: v_lshlrev_b32_e32 -; GCN-DAG: v_or_b32_e32 -; GCN: buffer_store_dword -; GCN: s_endpgm + +; SI: v_cvt_f32_i32_e32 +; SI: v_cvt_f32_i32_e32 +; SI: v_cvt_f16_f32_e32 +; SI: v_cvt_f16_f32_e32 +; SI-DAG: v_lshlrev_b32_e32 +; SI: v_or_b32_e32 + +; VI-DAG: v_cvt_f32_i32_sdwa +; VI-DAG: v_cvt_f32_i32_sdwa +; VI-DAG: v_cvt_f16_f32_e32 +; VI-DAG: v_cvt_f16_f32_sdwa +; VI: v_or_b32_e32 + +; GCN: buffer_store_dword +; GCN: s_endpgm + define amdgpu_kernel void @sitofp_v2i16_to_v2f16( <2 x half> addrspace(1)* %r, <2 x i16> addrspace(1)* %a) { @@ -56,15 +65,23 @@ entry: } ; GCN-LABEL: {{^}}sitofp_v2i32_to_v2f16 -; GCN: buffer_load_dwordx2 -; GCN: v_cvt_f32_i32_e32 -; GCN: v_cvt_f32_i32_e32 -; GCN: v_cvt_f16_f32_e32 -; GCN: v_cvt_f16_f32_e32 -; GCN-DAG: v_lshlrev_b32_e32 -; GCN-DAG: v_or_b32_e32 -; GCN: buffer_store_dword -; GCN: s_endpgm +; GCN: buffer_load_dwordx2 + +; SI: v_cvt_f32_i32_e32 +; SI: v_cvt_f32_i32_e32 +; SI: v_cvt_f16_f32_e32 +; SI: v_cvt_f16_f32_e32 +; SI-DAG: v_lshlrev_b32_e32 +; SI: v_or_b32_e32 + +; VI-DAG: v_cvt_f32_i32_e32 +; VI-DAG: v_cvt_f32_i32_e32 +; VI-DAG: v_cvt_f16_f32_e32 +; VI-DAG: v_cvt_f16_f32_sdwa +; VI: v_or_b32_e32 + +; GCN: buffer_store_dword +; GCN: s_endpgm define amdgpu_kernel void @sitofp_v2i32_to_v2f16( <2 x half> addrspace(1)* %r, <2 x i32> addrspace(1)* %a) { |
