nvhls_vector.h

/*
 * Copyright (c) 2016-2019, NVIDIA CORPORATION.  All rights reserved.
 * 
 * Licensed under the Apache License, Version 2.0 (the "License")
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 * 
 *     http://www.apache.org/licenses/LICENSE-2.0
 * 
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
#ifndef NVHLS_VECTOR_H
#define NVHLS_VECTOR_H
 
#include <systemc.h>
#include <nvhls_int.h>
#include <nvhls_types.h>
#include <hls_globals.h>
#include <nvhls_array.h>
#include <ccs_p2p.h>
#include <nvhls_assert.h>
#include <nvhls_message.h>
 
namespace nvhls {
 
inline std::string synth_to_string(const unsigned int& n) {
#ifndef __SYNTHESIS__
  std::ostringstream stm;
  stm << n;
  return stm.str();
#else
  return "DUMMY";
#endif
}
 
// SystemC vector class
// TYPE can be one of the SystemC types - sc_int, sc_uint, sc_bigint, sc_biguint
// or AC type ac_int
// TYPE cannot be nv_scvector i.e. we cannot have vector of vectors .. Need to
// add operators for vectors
template <typename Type, unsigned int VectorLength>
class nv_scvector : public nvhls_message {
 public:
  Type data[VectorLength];
  typedef Type type;
  static const unsigned int type_width = Wrapped<Type>::width;
  static const unsigned int length = VectorLength;
  static const unsigned int width = type_width * VectorLength;
  static const bool is_signed = Wrapped<Type>::is_signed;
 
  nv_scvector() {
  }
  nv_scvector(const nv_scvector<Type, VectorLength>& that) {
#pragma hls_unroll yes
    for (unsigned i = 0; i < VectorLength; i++)
      data[i] = that.data[i];
  }
  nv_scvector(const Type newdata[VectorLength]) {
#pragma hls_unroll yes
    for (unsigned i = 0; i < VectorLength; i++)
      data[i] = newdata[i];
  }
  // Constructing vector from rawbits of NVUINT type
  nv_scvector(const NVUINTW(width) & rawbits) {
#pragma hls_unroll yes
    for (unsigned int i = 0; i < VectorLength; i++) {
      data[i] = nvhls::get_slc<type_width>(rawbits, i * type_width);
    }
  }
  nv_scvector(const int& rawbits) {
    const NVUINTW(width) & rawbits_local = rawbits;
#pragma hls_unroll yes
    for (unsigned int i = 0; i < VectorLength; i++) {
      data[i] = nvhls::get_slc<type_width>(rawbits_local, i * type_width);
    }
  }
  nv_scvector<Type, VectorLength>& operator=(
      const nv_scvector<Type, VectorLength>& that) {
#pragma hls_unroll yes
    for (unsigned i = 0; i < VectorLength; i++)
      data[i] = that.data[i];
    return *this;
  }
 
  inline void copy(nv_scvector<Type, VectorLength>& out) {
#pragma hls_unroll yes
    for (unsigned i = 0; i < VectorLength; i++)
      out.data[i] = data[i];
  }
  Type& operator[](unsigned int i) {
    return this->data[i];
  }
  const Type& operator[](unsigned int i) const {
    return this->data[i];
  }
  // Converting vector to rawbits of NVUINT type
  NVUINTW(width) to_rawbits() {
    NVUINTW(width) rawbits = 0;
#pragma hls_unroll yes
    for (unsigned int i = 0; i < VectorLength; i++) {
      rawbits = nvhls::set_slc(rawbits, data[i], i * type_width);
    }
    return rawbits;
  }
  // Converting rawbits to vector
  void to_vector(NVUINTW(width) rawbits) {
#pragma hls_unroll yes
    for (unsigned int i = 0; i < VectorLength; i++) {
      data[i] = nvhls::get_slc<type_width>(rawbits, i * type_width);
    }
  }
 
  template <unsigned int Size>
  void Marshall(Marshaller<Size>& m) {
    #pragma hls_unroll yes
    for (unsigned int i = 0; i < VectorLength; i++) {
      m& data[i];
    }
  }
};
 
template <typename Type, unsigned int VectorLength>
inline bool operator==(
    const nv_scvector<Type, VectorLength>& lhs,
    const nv_scvector<Type, VectorLength>& rhs) {
    bool is_equal = true;
#pragma hls_unroll yes
    for (unsigned i = 0; i < VectorLength; i++)
      is_equal &= (lhs.data[i] == rhs.data[i]);
    return is_equal;
  }
 
template <typename Type, unsigned int VectorLength>
inline std::ostream& operator<<(std::ostream& os,
                                const nv_scvector<Type, VectorLength>& vec) {
  for (unsigned i = 0; i < VectorLength; i++) {
    os << vec.data[i] << " ";
  }
  return os;
}
 
template <typename InType1, typename InType2, typename OutType,
          unsigned int VectorLength, bool Unroll>
void vector_mul(nv_scvector<InType1, VectorLength> in1,
                nv_scvector<InType2, VectorLength> in2,
                nv_scvector<OutType, VectorLength>& out) {
 
  if (Unroll == true) {
#pragma hls_unroll yes
    for (unsigned i = 0; i < VectorLength; i++)
      out[i] = in1[i] * in2[i];
  } else {
    for (unsigned i = 0; i < VectorLength; i++)
      out[i] = in1[i] * in2[i];
  }
}
 
 
template <typename InType1, typename InType2, typename OutType,
          unsigned int VectorLength, bool Unroll>
void vector_add(nv_scvector<InType1, VectorLength> in1,
                nv_scvector<InType2, VectorLength> in2,
                nv_scvector<OutType, VectorLength>& out) {
  if (Unroll == true) {
#pragma hls_unroll yes
    for (unsigned i = 0; i < VectorLength; i++)
      out[i] = in1[i] + in2[i];
  } else {
    for (unsigned i = 0; i < VectorLength; i++)
      out[i] = in1[i] + in2[i];
  }
}
 
 
template <typename InType1, typename InType2, typename OutType,
          unsigned int VectorLength, bool Unroll>
void vector_sub(nv_scvector<InType1, VectorLength> in1,
                nv_scvector<InType2, VectorLength> in2,
                nv_scvector<OutType, VectorLength>& out) {
  if (Unroll == true) {
#pragma hls_unroll yes
    for (unsigned i = 0; i < VectorLength; i++)
      out[i] = in1[i] - in2[i];
  } else {
    for (unsigned i = 0; i < VectorLength; i++)
      out[i] = in1[i] - in2[i];
  }
}
 
 
template <typename InType, typename OutType, unsigned int VectorLength,
          bool UseReduceTree>
void reduction(nv_scvector<InType, VectorLength> in, OutType& out) {
  OutType sum = 0;
 
  if (VectorLength > 1) {
    if (UseReduceTree == true) {
#pragma hls_unroll yes
#pragma cluster addtree
#pragma cluster_type both
      for (unsigned i = 0; i < VectorLength; i++)
        sum += in[i];
    } else {
      for (unsigned i = 0; i < VectorLength; i++)
        sum += in[i];
    }
  } else {
    sum = in[0];
  }
  out = sum;
}
 
 
template <typename InType1, typename InType2, typename OutType,
          unsigned int VectorLength, bool UseReduceTree>
void dp(nv_scvector<InType1, VectorLength> in1,
        nv_scvector<InType2, VectorLength> in2, OutType& out) {
  OutType sum = 0;
  if (VectorLength > 1) {
    if (UseReduceTree == true) {
#pragma hls_unroll yes
#pragma cluster multadd addtree
#pragma cluster_type both
      for (unsigned i = 0; i < VectorLength; i++)
        sum += in1[i] * in2[i];
    } else {
      for (unsigned i = 0; i < VectorLength; i++)
        sum += in1[i] * in2[i];
    }
  } else {
    sum = in1[0] * in2[0];
  }
 
  out = sum;
}
 
template <typename InType1, typename InType2, typename InType3, typename OutType,
          unsigned int VectorLength, bool Unroll>
void vector_mac(nv_scvector<InType1, VectorLength> in1,
                nv_scvector<InType2, VectorLength> in2,
                nv_scvector<InType3, VectorLength> in3,
                nv_scvector<OutType, VectorLength>& out) {
 
  if (Unroll == true) {
#pragma hls_unroll yes
    for (unsigned i = 0; i < VectorLength; i++)
      out[i] = in1[i] * in2[i] + in3[i];
  } else {
    for (unsigned i = 0; i < VectorLength; i++)
      out[i] = in1[i] * in2[i] + in3[i];
  }
}
 
 
// Function implementing vector dot product followed by accumulate Template
// parameters: InType1, InType2,InType3 OutType are the data types for
// inputs and output. This function performs dot-product of 2 input vectors
// of "InType1" and "InType2" types, adds the results to 3rd scalar input of
//"InType3" type and produces scalar output of type "OutType".  Vector
// length of first 2 inputs is "VectorLength".  If InType1, InType2, InType3
// and OutType are user-defined types, then OutType = InType3, OutType +=
// InType1*InType2 operations should be defined by the user
template <typename InType1, typename InType2, typename InType3,
          typename OutType, unsigned int VectorLength, bool UseReduceTree>
void dpacc(nv_scvector<InType1, VectorLength> in1,
           nv_scvector<InType2, VectorLength> in2, InType3 in3, OutType& out) {
  OutType sum = in3;
  if (UseReduceTree == true) {
#pragma hls_unroll yes
#pragma cluster addtree
#pragma cluster_type both
    for (unsigned i = 0; i < VectorLength; i++)
      sum += in1[i] * in2[i];
  } else {
    for (unsigned i = 0; i < VectorLength; i++)
      sum += in1[i] * in2[i];
  }
  out = sum;
}
};
 
#endif