/*

Copyright (C) 2010-2016 VZLU Prague

This file is part of Octave.

Octave is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the
Free Software Foundation; either version 3 of the License, or (at your
option) any later version.

Octave is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
for more details.

You should have received a copy of the GNU General Public License
along with Octave; see the file COPYING.  If not, see
<http://www.gnu.org/licenses/>.

*/

#if ! defined (octave_oct_binmap_h)
#define octave_oct_binmap_h 1

#include "octave-config.h"

#include "Array.h"
#include "Sparse.h"
#include "Array-util.h"

#include "bsxfun.h"

// This source file implements a general binary maping function for arrays.
// The syntax is binmap<type> (a, b, f,[name]).
// type denotes the expected return type of the operation.
// a, b, should be one of the 6 combinations:
//
// Array-Array
// Array-scalar
// scalar-Array
// Sparse-Sparse
// Sparse-scalar
// scalar-Sparse
//
// If both operands are nonscalar, name must be supplied.  It is used
// as the base for error message when operands are nonconforming.
//
// The operation needs not be homogeneous, i.e., a, b and the result
// may be of distinct types.  f can have any of the four signatures:
//
// U f (T, R)
// U f (const T&, R)
// U f (T, const R&)
// U f (const T&, const R&)
//
// Additionally, f can be an arbitrary functor object.
//
// octave_quit() is called at appropriate places, hence the operation
// is breakable.

// The following template wrappers are provided for automatic bsxfun
// calls (see the function signature for do_bsxfun_op).

template <typename R, typename X, typename Y, typename F>
class bsxfun_wrapper
{
private:
  static F f;

public:
  static void
  set_f (const F& f_in)
  {
    f = f_in;
  }

  static void
  op_mm (size_t n, R* r, const X* x , const Y* y)
  {
    for (size_t i = 0; i < n; i++)
      r[i] = f (x[i], y[i]);
  }

  static void
  op_sm (size_t n, R* r, X x, const Y* y)
  {
    for (size_t i = 0; i < n; i++)
      r[i] = f (x, y[i]);
  }

  static void
  op_ms (size_t n , R* r, const X* x, Y y)
  {
    for (size_t i = 0; i < n; i++)
      r[i] = f (x[i], y);
  }
};

// Static init
template <typename R, typename X, typename Y, typename F>
F bsxfun_wrapper<R, X, Y, F>::f;

// scalar-Array
template <typename U, typename T, typename R, typename F>
Array<U>
binmap (const T& x, const Array<R>& ya, F fcn)
{
  octave_idx_type len = ya.numel ();

  const R *y = ya.data ();

  Array<U> result (ya.dims ());
  U *p = result.fortran_vec ();

  octave_idx_type i;
  for (i = 0; i < len - 3; i += 4)
    {
      octave_quit ();

      p[i] = fcn (x, y[i]);
      p[i+1] = fcn (x, y[i+1]);
      p[i+2] = fcn (x, y[i+2]);
      p[i+3] = fcn (x, y[i+3]);
    }

  octave_quit ();

  for (; i < len; i++)
    p[i] = fcn (x, y[i]);

  return result;
}

// Array-scalar
template <typename U, typename T, typename R, typename F>
Array<U>
binmap (const Array<T>& xa, const R& y, F fcn)
{
  octave_idx_type len = xa.numel ();

  const R *x = xa.data ();

  Array<U> result (xa.dims ());
  U *p = result.fortran_vec ();

  octave_idx_type i;
  for (i = 0; i < len - 3; i += 4)
    {
      octave_quit ();

      p[i] = fcn (x[i], y);
      p[i+1] = fcn (x[i+1], y);
      p[i+2] = fcn (x[i+2], y);
      p[i+3] = fcn (x[i+3], y);
    }

  octave_quit ();

  for (; i < len; i++)
    p[i] = fcn (x[i], y);

  return result;
}

// Array-Array (treats singletons as scalars)
template <typename U, typename T, typename R, typename F>
Array<U>
binmap (const Array<T>& xa, const Array<R>& ya, F fcn, const char *name)
{
  dim_vector xad = xa.dims ();
  dim_vector yad = ya.dims ();
  if (xa.numel () == 1)
    return binmap<U, T, R, F> (xa(0), ya, fcn);
  else if (ya.numel () == 1)
    return binmap<U, T, R, F> (xa, ya(0), fcn);
  else if (xad != yad)
    {
      if (! is_valid_bsxfun (name, xad, yad))
        octave::err_nonconformant (name, xad, yad);

      bsxfun_wrapper<U, T, R, F>::set_f(fcn);
      return do_bsxfun_op (xa, ya,
                           bsxfun_wrapper<U, T, R, F>::op_mm,
                           bsxfun_wrapper<U, T, R, F>::op_sm,
                           bsxfun_wrapper<U, T, R, F>::op_ms);
    }

  octave_idx_type len = xa.numel ();

  const T *x = xa.data ();
  const T *y = ya.data ();

  Array<U> result (xa.dims ());
  U *p = result.fortran_vec ();

  octave_idx_type i;
  for (i = 0; i < len - 3; i += 4)
    {
      octave_quit ();

      p[i] = fcn (x[i], y[i]);
      p[i+1] = fcn (x[i+1], y[i+1]);
      p[i+2] = fcn (x[i+2], y[i+2]);
      p[i+3] = fcn (x[i+3], y[i+3]);
    }

  octave_quit ();

  for (; i < len; i++)
    p[i] = fcn (x[i], y[i]);

  return result;
}

// scalar-Sparse
template <typename U, typename T, typename R, typename F>
Sparse<U>
binmap (const T& x, const Sparse<R>& ys, F fcn)
{
  R yzero = R ();
  U fz = fcn (x, yzero);

  if (fz == U ())  // Sparsity preserving fcn
    {
      octave_idx_type nz = ys.nnz ();
      Sparse<U> retval (ys.rows (), ys.cols (), nz);
      std::copy (ys.ridx (), ys.ridx () + nz, retval.ridx ());
      std::copy (ys.cidx (), ys.cidx () + ys.cols () + 1, retval.cidx ());

      for (octave_idx_type i = 0; i < nz; i++)
        {
          octave_quit ();
          // FIXME: Could keep track of whether fcn call results in a 0.
          //        If no zeroes are created could skip maybe_compress()
          retval.xdata (i) = fcn (x, ys.data (i));
        }

      octave_quit ();
      retval.maybe_compress (true);
      return retval;
    }
  else
    return Sparse<U> (binmap<U, T, R, F> (x, ys.array_value (), fcn));
}

// Sparse-scalar
template <typename U, typename T, typename R, typename F>
Sparse<U>
binmap (const Sparse<T>& xs, const R& y, F fcn)
{
  T xzero = T ();
  U fz = fcn (xzero, y);

  if (fz == U ())  // Sparsity preserving fcn
    {
      octave_idx_type nz = xs.nnz ();
      Sparse<U> retval (xs.rows (), xs.cols (), nz);
      std::copy (xs.ridx (), xs.ridx () + nz, retval.ridx ());
      std::copy (xs.cidx (), xs.cidx () + xs.cols () + 1, retval.cidx ());

      for (octave_idx_type i = 0; i < nz; i++)
        {
          octave_quit ();
          // FIXME: Could keep track of whether fcn call results in a 0.
          //        If no zeroes are created could skip maybe_compress()
          retval.xdata (i) = fcn (xs.data (i), y);
        }

      octave_quit ();
      retval.maybe_compress (true);
      return retval;
    }
  else
    return Sparse<U> (binmap<U, T, R, F> (xs.array_value (), y, fcn));
}

// Sparse-Sparse (treats singletons as scalars)
template <typename U, typename T, typename R, typename F>
Sparse<U>
binmap (const Sparse<T>& xs, const Sparse<R>& ys, F fcn, const char *name)
{
  if (xs.rows () == 1 && xs.cols () == 1)
    return binmap<U, T, R, F> (xs(0,0), ys, fcn);
  else if (ys.rows () == 1 && ys.cols () == 1)
    return binmap<U, T, R, F> (xs, ys(0,0), fcn);
  else if (xs.dims () != ys.dims ())
    octave::err_nonconformant (name, xs.dims (), ys.dims ());

  T xzero = T ();
  R yzero = R ();
  U fz = fcn (xzero, yzero);

  if (fz == U ())
    {
      // Sparsity-preserving function.  Do it efficiently.
      octave_idx_type nr = xs.rows ();
      octave_idx_type nc = xs.cols ();
      Sparse<T> retval (nr, nc, xs.nnz () + ys.nnz ());

      octave_idx_type nz = 0;
      for (octave_idx_type j = 0; j < nc; j++)
        {
          octave_quit ();

          octave_idx_type jx = xs.cidx (j);
          octave_idx_type jx_max = xs.cidx (j+1);
          bool jx_lt_max = jx < jx_max;

          octave_idx_type jy = ys.cidx (j);
          octave_idx_type jy_max = ys.cidx (j+1);
          bool jy_lt_max = jy < jy_max;

          while (jx_lt_max || jy_lt_max)
            {
              if (! jy_lt_max
                  || (jx_lt_max && (xs.ridx (jx) < ys.ridx (jy))))
                {
                  retval.xridx (nz) = xs.ridx (jx);
                  retval.xdata (nz) = fcn (xs.data (jx), yzero);
                  jx++;
                  jx_lt_max = jx < jx_max;
                }
              else if (! jx_lt_max
                       || (jy_lt_max && (ys.ridx (jy) < xs.ridx (jx))))
                {
                  retval.xridx (nz) = ys.ridx (jy);
                  retval.xdata (nz) = fcn (xzero, ys.data (jy));
                  jy++;
                  jy_lt_max = jy < jy_max;
                }
              else
                {
                  retval.xridx (nz) = xs.ridx (jx);
                  retval.xdata (nz) = fcn (xs.data (jx), ys.data (jy));
                  jx++;
                  jx_lt_max = jx < jx_max;
                  jy++;
                  jy_lt_max = jy < jy_max;
                }
              nz++;
            }
          retval.xcidx (j+1) = nz;
        }

      retval.maybe_compress (true);
      return retval;
    }
  else
    return Sparse<U> (binmap<U, T, R, F> (xs.array_value (), ys.array_value (),
                                          fcn, name));
}

// Overloads for function pointers.

// Signature (T, R)

template <typename U, typename T, typename R>
inline Array<U>
binmap (const Array<T>& xa, const Array<R>& ya, U (*fcn) (T, R),
        const char *name)
{ return binmap<U, T, R, U (*) (T, R)> (xa, ya, fcn, name); }

template <typename U, typename T, typename R>
inline Array<U>
binmap (const T& x, const Array<R>& ya, U (*fcn) (T, R))
{ return binmap<U, T, R, U (*) (T, R)> (x, ya, fcn); }

template <typename U, typename T, typename R>
inline Array<U>
binmap (const Array<T>& xa, const R& y, U (*fcn) (T, R))
{ return binmap<U, T, R, U (*) (T, R)> (xa, y, fcn); }

template <typename U, typename T, typename R>
inline Sparse<U>
binmap (const Sparse<T>& xa, const Sparse<R>& ya, U (*fcn) (T, R),
        const char *name)
{ return binmap<U, T, R, U (*) (T, R)> (xa, ya, fcn, name); }

template <typename U, typename T, typename R>
inline Sparse<U>
binmap (const T& x, const Sparse<R>& ya, U (*fcn) (T, R))
{ return binmap<U, T, R, U (*) (T, R)> (x, ya, fcn); }

template <typename U, typename T, typename R>
inline Sparse<U>
binmap (const Sparse<T>& xa, const R& y, U (*fcn) (T, R))
{ return binmap<U, T, R, U (*) (T, R)> (xa, y, fcn); }

// Signature (const T&, const R&)

template <typename U, typename T, typename R>
inline Array<U>
binmap (const Array<T>& xa, const Array<R>& ya, U (*fcn) (const T&, const R&),
        const char *name)
{ return binmap<U, T, R, U (*) (const T&, const R&)> (xa, ya, fcn, name); }

template <typename U, typename T, typename R>
inline Array<U>
binmap (const T& x, const Array<R>& ya, U (*fcn) (const T&, const R&))
{ return binmap<U, T, R, U (*) (const T&, const R&)> (x, ya, fcn); }

template <typename U, typename T, typename R>
inline Array<U>
binmap (const Array<T>& xa, const R& y, U (*fcn) (const T&, const R&))
{ return binmap<U, T, R, U (*) (const T&, const R&)> (xa, y, fcn); }

template <typename U, typename T, typename R>
inline Sparse<U>
binmap (const Sparse<T>& xa, const Sparse<R>& ya, U (*fcn) (const T&, const R&),
        const char *name)
{ return binmap<U, T, R, U (*) (const T&, const R&)> (xa, ya, fcn, name); }

template <typename U, typename T, typename R>
inline Sparse<U>
binmap (const T& x, const Sparse<R>& ya, U (*fcn) (const T&, const R&))
{ return binmap<U, T, R, U (*) (const T&, const R&)> (x, ya, fcn); }

template <typename U, typename T, typename R>
inline Sparse<U>
binmap (const Sparse<T>& xa, const R& y, U (*fcn) (const T&, const R&))
{ return binmap<U, T, R, U (*) (const T&, const R&)> (xa, y, fcn); }

// Signature (const T&, R)

template <typename U, typename T, typename R>
inline Array<U>
binmap (const Array<T>& xa, const Array<R>& ya, U (*fcn) (const T&, R),
        const char *name)
{ return binmap<U, T, R, U (*) (const T&, R)> (xa, ya, fcn, name); }

template <typename U, typename T, typename R>
inline Array<U>
binmap (const T& x, const Array<R>& ya, U (*fcn) (const T&, R))
{ return binmap<U, T, R, U (*) (const T&, R)> (x, ya, fcn); }

template <typename U, typename T, typename R>
inline Array<U>
binmap (const Array<T>& xa, const R& y, U (*fcn) (const T&, R))
{ return binmap<U, T, R, U (*) (const T&, R)> (xa, y, fcn); }

template <typename U, typename T, typename R>
inline Sparse<U>
binmap (const Sparse<T>& xa, const Sparse<R>& ya, U (*fcn) (const T&, R),
        const char *name)
{ return binmap<U, T, R, U (*) (const T&, R)> (xa, ya, fcn, name); }

template <typename U, typename T, typename R>
inline Sparse<U>
binmap (const T& x, const Sparse<R>& ya, U (*fcn) (const T&, R))
{ return binmap<U, T, R, U (*) (const T&, R)> (x, ya, fcn); }

template <typename U, typename T, typename R>
inline Sparse<U>
binmap (const Sparse<T>& xa, const R& y, U (*fcn) (const T&, R))
{ return binmap<U, T, R, U (*) (const T&, R)> (xa, y, fcn); }

// Signature (T, const R&)

template <typename U, typename T, typename R>
inline Array<U>
binmap (const Array<T>& xa, const Array<R>& ya, U (*fcn) (T, const R&),
        const char *name)
{ return binmap<U, T, R, U (*) (T, const R&)> (xa, ya, fcn, name); }

template <typename U, typename T, typename R>
inline Array<U>
binmap (const T& x, const Array<R>& ya, U (*fcn) (T, const R&))
{ return binmap<U, T, R, U (*) (T, const R&)> (x, ya, fcn); }

template <typename U, typename T, typename R>
inline Array<U>
binmap (const Array<T>& xa, const R& y, U (*fcn) (T, const R&))
{ return binmap<U, T, R, U (*) (T, const R&)> (xa, y, fcn); }

template <typename U, typename T, typename R>
inline Sparse<U>
binmap (const Sparse<T>& xa, const Sparse<R>& ya, U (*fcn) (T, const R&),
        const char *name)
{ return binmap<U, T, R, U (*) (T, const R&)> (xa, ya, fcn, name); }

template <typename U, typename T, typename R>
inline Sparse<U>
binmap (const T& x, const Sparse<R>& ya, U (*fcn) (T, const R&))
{ return binmap<U, T, R, U (*) (T, const R&)> (x, ya, fcn); }

template <typename U, typename T, typename R>
inline Sparse<U>
binmap (const Sparse<T>& xa, const R& y, U (*fcn) (T, const R&))
{ return binmap<U, T, R, U (*) (T, const R&)> (xa, y, fcn); }

#endif