ehw-fit/bdd-evaluation

Fast BDD evaluation methods

Fast BDD-based evaluation

This repository provides codes for fast evaluation of approximate adders and subtractors by means of BDD diagrams. It employs BuDDy library but can be easily trasfered to CuDD (with C++ interface). The code style is inspired by Mathias Soeken tool Cirkit. Note that this repository provides only the evaluation algorithms, not the searching algorithm.

Reference

When you use this tool in your work/research, please cite the following article: MRAZEK Vojtech. Optimization of BDD-based Approximation Error Metrics Calculations. In: IEEE Computer Society Annual Symposium on VLSI (ISVLSI '22). Cyprus, 2022, p. 6.

@INPROCEEDINGS{mrazek:ISVLSI22,
   author = "Vojtech Mrazek",
   title = "Optimization of BDD-based Approximation Error Metrics Calculations",
   pages = 6,
   booktitle = "IEEE Computer Society Annual Symposium on VLSI (ISVLSI '22)",
   year = 2022,
   location = "Paphos, CY"
}

Usage in your application

#include <iostream>
#include <cstdlib>
#include <vector>
#include <bdd.h>

#include "arithmetic.hpp"

using namespace std;

static void bdd_gbchandler(int pre, bddGbcStat *s)
{ }

int main(int argc, char **argv)
{
    const int inputs = 16; // 8-bit adder
    const bool is_signed = false; // unsigned output
    vector<bdd> bdd_res(9);
    vector<bdd> bdd_ref(9);


    bdd_init(10000000, 10000000);
    bdd_setvarnum(inputs);
    bdd_gbc_hook(bdd_gbchandler); // disable messages about garbage collector
    bdd_set_limit(120);
    bdd_set_start();

    double time = cpuTime();

    // TODO: create reference and golden models
    // fill vectors bdd_res and bdd_ref (e.g. )
    // bdd a = bdd_ithvar(0);
    // bdd b = a ^ bdd_ithvar(2)
    // bdd_res[2] = a & b;


    // Example of calculation
    auto sub = bdd_subtract(bdd_res, bdd_ref, true, is_signed);
    auto res = bdd_abs(sub);
    cout << "# subtract created at " << cpuTime() - time << endl;
    cout << "Signed: " << is_signed << endl;

    cout << "Baseline mae: " << average(res) << endl;
    cout << "Ones complement mae: " << average_one(sub) << endl;
    cout << "Noabs mae: " << average_abs(sub) << endl;

    cout << "Baseline wc: " << worst_case(res) << endl;
    cout << "Ones complement wc: " << worst_case_one(sub) << endl;
    cout << "Noabs wc: " << worst_case_abs(sub) << endl;

    cout << "EP: " << hamming(bdd_res, bdd_ref) << endl;

    cout << "# finished at " << cpuTime() - time << endl;

    bddcgp_done();

    return 0;
}

Building and testing

To build a library run a following commands

mkdir build
cd build
cmake ..
make
# Run tests
make test