weights.hpp

/******************************************************************************
 *  Copyright (c) 2019, Xilinx, Inc.
 *  All rights reserved.
 *
 *  Redistribution and use in source and binary forms, with or without
 *  modification, are permitted provided that the following conditions are met:
 *
 *  1.  Redistributions of source code must retain the above copyright notice,
 *     this list of conditions and the following disclaimer.
 *
 *  2.  Redistributions in binary form must reproduce the above copyright
 *      notice, this list of conditions and the following disclaimer in the
 *      documentation and/or other materials provided with the distribution.
 *
 *  3.  Neither the name of the copyright holder nor the names of its
 *      contributors may be used to endorse or promote products derived from
 *      this software without specific prior written permission.
 *
 *  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
 *  AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
 *  THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 *  PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR
 *  CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
 *  EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
 *  PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
 *  OR BUSINESS INTERRUPTION). HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
 *  WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
 *  OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
 *  ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *******************************************************************************/

/*******************************************************************************
 *
 *  Authors: Giulio Gambardella <giuliog@xilinx.com>
 *           Thomas B. Preusser <thomas.preusser@utexas.edu>
 *             Marie-Curie Fellow, Xilinx Ireland, Grant Agreement No. 751339
 *           Christoph Doehring <cdoehrin@xilinx.com>
 *
 *  @file weights.hpp
 *
 *  Library of templated HLS classes for BNN deployment. 
 *  This file lists a set of classes used to implement  
 *  weights in neural network. 
 *
 *  This project has received funding from the European Union's Framework
 *  Programme for Research and Innovation Horizon 2020 (2014-2020) under
 *  the Marie Skłodowska-Curie Grant Agreement No. 751339.
 *
 *******************************************************************************/

#ifndef WEIGHTS_HPP
#define WEIGHTS_HPP

#include <ap_int.h>
#include <array>


/**
 * \brief      A binary weight storage adapter that translates the internal 
 * organization optimized for storage to the generalized access by the MVAU.
 *
 * \tparam     SIMD   Number of input columns (channels) computed in parallel
 * \tparam     PE     Number of output rows (channels) computed in parallel
 * \tparam     TILES  3rd dimension of the weights matrix
 */
template<unsigned SIMD, unsigned PE, unsigned TILES>
class BinaryWeights {
 public:
  ap_uint<SIMD>  m_weights[PE][TILES];

 private:
  /**
   * Temporary container for the tile index to implement the
   * memory access in pe -> tile order.
   */
  class TileIndex {
    BinaryWeights const &m_par;
    unsigned      const  m_idx;

   public:
    TileIndex(BinaryWeights const &par, unsigned const  idx)
      : m_par(par), m_idx(idx) {
#pragma HLS inline
    }

   public:
    ap_uint<SIMD> operator[](unsigned const  pe) const {
#pragma HLS inline
      return  m_par.m_weights[pe][m_idx];
    }
  };

 public:
  TileIndex weights(unsigned const  tile) const {
#pragma HLS inline
    return  TileIndex(*this, tile);
  }
};


/**
 * \brief      A fixeed point weight storage adapter that translates the internal 
 * organization optimized for storage to the generalized access by the MVAU.
 *
 * \tparam     SIMD   Number of input columns (channels) computed in parallel
 * \tparam     WT     Datatype of the weights
 * \tparam     PE     Number of output rows (channels) computed in parallel
 * \tparam     TILES  3rd dimension of the weights matrix
 */
template<unsigned SIMD, typename WT ,unsigned PE, unsigned TILES>
class FixedPointWeights {
 public:
  ap_uint<SIMD*WT::width>  m_weights[PE][TILES];

 private:
  /**
   * Temporary container for the tile index to implement the
   * memory access in pe -> tile order.
   */
  class TileIndex {
    FixedPointWeights const &m_par;
    unsigned          const  m_idx;

   public:
    TileIndex(FixedPointWeights const &par, unsigned const  idx)
      : m_par(par), m_idx(idx) {
#pragma HLS inline
    }

   public:
    std::array<WT,SIMD> operator[](unsigned const  pe) const {
#pragma HLS inline
      std::array<WT,SIMD> temp;
	  for(unsigned int i=0; i<SIMD; i++) {
#pragma HLS unroll
        ap_int<WT::width> local_temp;
        local_temp = m_par.m_weights[pe][m_idx]((i+1)*WT::width-1, i*WT::width);
        WT value = *reinterpret_cast<WT*>(&local_temp);
        temp[i] = value;
      }
      return  temp;
    }
  };

 public:
  TileIndex weights(unsigned const  tile) const {
#pragma HLS inline
    return  TileIndex(*this, tile);
  }
};


template<unsigned SIMD, typename WT, unsigned PE  >
class Weights_Tile { 
 public:
  ap_uint<SIMD*WT::width>  m_weights[PE];

  std::array<WT,SIMD> operator[](unsigned const  pe) const {
    #pragma HLS inline
    std::array<WT,SIMD> temp;
    for(unsigned int i=0; i<SIMD; i++) {
      #pragma HLS unroll
      ap_int<WT::width> local_temp;
      local_temp = m_weights[pe]((i+1)*WT::width-1, i*WT::width);
      WT value = *reinterpret_cast<WT*>(&local_temp);
      temp[i] = value;
    }
    return  temp;
  }
};


#endif