fix 2gb limit issue

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

@@ -864,9 +864,9 @@ struct DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
         // Input tensors can't be bigger than 2GB each.
         constexpr ck::long_index_t GB2 = (ck::long_index_t{1} << 31);
 
-        if(arg.a_grid_desc_k0_m_k1_.GetElementSpaceSize() > GB2 ||
-           arg.b_grid_desc_k0_n_k1_.GetElementSpaceSize() > GB2 ||
-           arg.c_grid_desc_m_n_.GetElementSpaceSize() > GB2)
+        if(arg.a_grid_desc_k0_m_k1_.GetElementSpaceSize() * sizeof(ADataType) > GB2 ||
+           arg.b_grid_desc_k0_n_k1_.GetElementSpaceSize() * sizeof(BDataType) > GB2 ||
+           arg.c_grid_desc_m_n_.GetElementSpaceSize() * sizeof(CDataType) > GB2)
         {
             return false;
         }