E-syn2

Main branch contains the early-stage implementation of E-Morphic. You can test the tool by following the instructions below. Random sampling is implemented as the basic workflow, it works and still can achieving good result, which can be deemed as a strong baseline.

For reproducing the experiment result in our paper, check SA-golden-model-delaytuning branch.

Prequisites

• Rust environment
• Berkeley ABC tool (replaced with the abc/src/opt/dar/darRefact.c)

Usage

make
chmod +x ./run.sh && ./run.sh

quick start:

echo -e "5\ndelay\nfaster-bottom-up\n" | bash run.sh    

run with randomized extraction experiments:

echo -e "60\narea\nrandom-based-faster-bottom-up\n40\n0.8\n" | bash run.sh 

run with iterations experiments (without randomized extraction):

for i in $(seq 5 5 50); do
  echo -e "${i}\ndelay\nfaster-bottom-up\n" | bash run.sh > tmp_log/log_${i}_no_feature.txt
  wait
done

Directory Structure

./
├── abc
├── converted_circuit_strash # benchmark circuits
├── data_process_script
├── deprecated
├── e-rewriter # rewriter and parser
|   ├── rewritten_circuit # rewriten circuits -> wait for extraction
│   ├── circuit0.eqn # put your circuit here
│   ├── src # includes frontend parser (eqn2egraph) and egraph-serializer
│   ├── target
├── extraction-gym # extractor
│   ├── input # 1. saturacted circuits graphs for extraction (copied from e-rewriter)
│   ├── out_dag_json # 2. raw json marked the extracted nodes - dag based extraction
│   ├── out_json # 2. raw json marked the extracted nodes - tree based extraction
│   ├── random_out_dag_json # 3. random extraction results -> raw json (random-based extraction)
│   ├── output_log # 4. log files during extraction
│   ├── src
│   └── target
├── process_json # post-processing script for extracted results
│   ├── input_saturacted_egraph # saturacted circuits
│   ├── input_extracted_egraph # extracted circuits
│   ├── out_process_dag_result # processed json (handled extracted nodes raw json)
│   ├── src
│   └── target
├── graph2eqn # convert extracted circuits to eqn format
│   ├── circuit0.eqn
│   ├── src
│   └── target
├── fuzz_abc_opt_flow # explore the optimization space of circuits
├── clean.sh
├── collect.sh
├── README.md
├── run.sh
├── test.sh
├── Makefile
└── tmp_log
    ├── abc_opt_all_{timestamp}.log # log file for abc optimization (complete log)
    ├── abc_opt_all_formatted_{timestamp}.log # log file for abc optimization (formatted log)
    └── log_{iteration_number}_no_feature.txt # log file for each iteration (no feature extraction)