amend

include/ck/tensor_operation/gpu/device/device_convnd_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp

@@ -802,17 +802,16 @@ struct DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_
             b_grid_desc_kbatch_k0_n_k1_ = descs[I1];
             c_grid_desc_m_n_            = descs[I2];
 
+            block_2_ctile_map_ =
+                GridwiseGemm::MakeCBlockClusterAdaptor(c_grid_desc_m_n_, M01, N01, k_batch_);
+
             if(GridwiseGemm::CheckValidity(a_grid_desc_kbatch_k0_m_k1_,
                                            b_grid_desc_kbatch_k0_n_k1_,
                                            c_grid_desc_m_n_,
-                                           M01_,
-                                           N01_))
+                                           block_2_ctile_map_))
             {
                 c_grid_desc_mblock_mperblock_nblock_nperblock_ =
                     GridwiseGemm::MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(c_grid_desc_m_n_);
-
-                block_2_ctile_map_ =
-                    GridwiseGemm::MakeCBlockClusterAdaptor(c_grid_desc_m_n_, M01, N01, k_batch_);
             }
         }
 
@@ -871,14 +870,14 @@ struct DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_
             if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_kbatch_k0_m_k1_,
                                             arg.b_grid_desc_kbatch_k0_n_k1_,
                                             arg.c_grid_desc_m_n_,
-                                            arg.M01_,
-                                            arg.N01_))
+                                            arg.block_2_ctile_map_))
             {
                 throw std::runtime_error(
                     "wrong! GridwiseGemm_km_kn_m0m1n0n1_xdlops_v3r1 has invalid setting");
             }
             const auto kbatch = arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I0);
-            const index_t grid_size = GridwiseGemm::CalculateGridSize(arg.c_grid_desc_m_n_, kbatch);
+            const index_t grid_size =
+                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_);
 
             const auto K0 = arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I1);
 
@@ -1066,8 +1065,7 @@ struct DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_
         return GridwiseGemm::CheckValidity(arg.a_grid_desc_kbatch_k0_m_k1_,
                                            arg.b_grid_desc_kbatch_k0_n_k1_,
                                            arg.c_grid_desc_m_n_,
-                                           arg.M01_,
-                                           arg.N01_);
+                                           arg.block_2_ctile_map_);
     }
 
     bool IsSupportedArgument(const BaseArgument* p_arg) override