cv32e40x_decimate

This repository contains the code of:

@inproceedings{lightweightk,
  title={Lightweight Software Kernels and Hardware Extensions for Efficient Sparse Deep Neural Networks on Microcontrollers},
  author={Daghero, Francesco and Pagliari, Daniele Jahier and Conti, Francesco and Benini, Luca and Poncino, Massimo and Burrello, Alessio},
  booktitle={Eighth Conference on Machine Learning and Systems (MLSys)},
  year={2025}
}

Structure

Each module is contained in a file with its own name, moreover there is a fir_xifu_package SV package containing constants data structure definitions. The overall hierarchy of the FIR XIFU looks like this:

fir_xifu_top          # top-level of the IP
 |-> fir_xifu_ctrl    # main controller, including a simple scoreboard for instructions offloaded to the interface
 |-> fir_xifu_regfile # a private register file of parametric size
 |-> fir_xifu_id      # instruction decode stage
 |-> fir_xifu_ex      # instruction execute stage
 |-> fir_xifu_wb      # instruction write-back stage

Dependencies

The dependencies are managed via Bender (https://github.com/pulp-platform/bender).

The core can be found here (https://github.com/pulp-platform/cv32e40x/tree/fc/xif-fixes)