DRL_Final

Final project for Deep Reinforcement Learning 2025@NTU

📁 Project Structure

├── abc/                         # Cloned ABC logic synthesis tool
├── abc.history                  # ABC command history (optional)
├── helper_functions.py          # Internal utility functions (⚠️ Do not use directly)
├── requirements.txt             # Python dependencies
├── rl_environment.py            # Main environment entry point (✅ use this)
├── runtime_results/             # Output directory
│   └── current.aig              # Example AIG file
└── testbenches/                 # Benchmark circuits
    ├── 2025_IWLS_Contest_Benchmarks_020425/
    └── EPFL/

Getting Started

⚠️ Important: All commands below should be executed from the root DRL_Final directory.

1. Clone the ABC Library

You must clone and build the ABC logic synthesis tool in the abc/ folder:

git clone https://github.com/berkeley-abc/abc.git abc
cd abc
make
cd ..

2. Install Python Dependencies

Use the provided requirements.txt to install the required packages:

pip install -r requirements.txt

3. Run the Environment

Several scripts are provided to run different optimization approaches:

• Run MCTS-based optimization only:

  ./scripts/run_mcts.sh

• Run DeepSyn optimization only:

  ./scripts/run_deepsyn.sh

• Run Resyn2 optimization only:

  ./scripts/run_resyn2.sh

• Run all optimization methods sequentially:

  ./scripts/run_all.sh

By default, these scripts will process the benchmark ex100.truth. To process additional benchmarks, modify the BENCHMARKS array in the script by uncommenting the desired benchmark files:

BENCHMARKS=(
    "$IWLS_DIR/ex100.truth" # Default
    # "$IWLS_DIR/ex104.truth"
    # "$IWLS_DIR/ex108.truth"
    # "$IWLS_DIR/ex112.truth"
    "$EPFL_DIR/adder.truth"  # Uncomment to process adder.truth
    # "$EPFL_DIR/bar.truth"
    # "$EPFL_DIR/max.truth"
    # "$EPFL_DIR/sin.truth"
)

The script will then process each uncommented benchmark sequentially through all optimization methods.