Dear author，I still have some questions.

[yangbo19]

As I show below,how do you normalize the wirelength?Whether or not the function of plc.get_cost considerd area? In addition, I notice that the values of wirelength and density are of different orders of magnitude, but they are multiplied by the same weight. Is it a problem here? In the experiment, the cost result is calculated by the weighted sums of wirelength, congestion and density. However, the values of different netlists are different. Since the weighted sum cost of a large netlist may be larger than that of a small netlist, how do you ensure the rewards of a large netlist should not always less than of a small netlist? Thank you so much in advance.

[sakundu]

You can find the details about the proxy cost computation here.

---

You may check out our MacroPlacement repo.

We provide a script to write out the protobuf netlist from the Innovus tool which can be used as input to the CircuitTraining grouping code to generate the clustered netlist.

Also, you can train circuit training using the testcases available in MacroPlacement/Flows/<Enablement>/<Design>/netlist/output_CT_Grouping/.

You can find here the progress we have made till August 26.