refactor

src/include/threadwise_direct_convolution.hip.hpp

@@ -80,10 +80,8 @@ __device__ void threadwise_direct_convolution_2(InDesc,
     constexpr auto wei_desc = WeiDesc{};
     constexpr auto out_desc = OutDesc{};
 
-    constexpr auto in_reg_desc =
-        make_ConstantTensorDescriptor_default_rank_packed(in_desc.GetLengths());
-    constexpr auto wei_reg_desc =
-        make_ConstantTensorDescriptor_default_rank_packed(wei_desc.GetLengths());
+    constexpr auto in_reg_desc  = make_ConstantTensorDescriptor_packed(in_desc.GetLengths());
+    constexpr auto wei_reg_desc = make_ConstantTensorDescriptor_packed(wei_desc.GetLengths());
 
     // register
     TInWei p_in_reg[in_reg_desc.GetElementSpace()];

src/include/threadwise_tensor_slice_op.hip.hpp

@@ -19,7 +19,7 @@ __device__ void threadwise_tensor_slice_copy(SrcDesc,
 
     constexpr auto src_desc = SrcDesc{};
     constexpr auto dst_desc = DstDesc{};
-    constexpr auto ref_desc = make_ConstantTensorDescriptor_default_rank_packed(SrcOpLengths{});
+    constexpr auto ref_desc = make_ConstantTensorDescriptor_packed(SrcOpLengths{});
 
 #if 0
     if(get_thread_local_1d_id() == 0 && get_block_1d_id() == 0)