aboutsummaryrefslogtreecommitdiffhomepage
path: root/disabled/Product.h
blob: d67916da5779a69c82b587f26eb7b44963496d67 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra. Eigen itself is part of the KDE project.
//
// Copyright (C) 2006-2008 Benoit Jacob <jacob@math.jussieu.fr>
// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, 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 2 of
// the License, or (at your option) any later version.
//
// Eigen 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 Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.

#ifndef EIGEN_PRODUCT_H
#define EIGEN_PRODUCT_H

template<int Index, int Size, typename Lhs, typename Rhs>
struct ei_product_unroller
{
  inline static void run(int row, int col, const Lhs& lhs, const Rhs& rhs,
                               typename Lhs::Scalar &res)
  {
    ei_product_unroller<Index-1, Size, Lhs, Rhs>::run(row, col, lhs, rhs, res);
    res += lhs.coeff(row, Index) * rhs.coeff(Index, col);
  }
};

template<int Size, typename Lhs, typename Rhs>
struct ei_product_unroller<0, Size, Lhs, Rhs>
{
  inline static void run(int row, int col, const Lhs& lhs, const Rhs& rhs,
                               typename Lhs::Scalar &res)
  {
    res = lhs.coeff(row, 0) * rhs.coeff(0, col);
  }
};

template<int Index, typename Lhs, typename Rhs>
struct ei_product_unroller<Index, Dynamic, Lhs, Rhs>
{
  inline static void run(int, int, const Lhs&, const Rhs&, typename Lhs::Scalar&) {}
};

// prevent buggy user code from causing an infinite recursion
template<int Index, typename Lhs, typename Rhs>
struct ei_product_unroller<Index, 0, Lhs, Rhs>
{
  inline static void run(int, int, const Lhs&, const Rhs&, typename Lhs::Scalar&) {}
};

template<typename Lhs, typename Rhs>
struct ei_product_unroller<0, Dynamic, Lhs, Rhs>
{
  static void run(int, int, const Lhs&, const Rhs&, typename Lhs::Scalar&) {}
};

template<bool RowMajor, int Index, int Size, typename Lhs, typename Rhs, typename PacketScalar>
struct ei_packet_product_unroller;

template<int Index, int Size, typename Lhs, typename Rhs, typename PacketScalar>
struct ei_packet_product_unroller<true, Index, Size, Lhs, Rhs, PacketScalar>
{
  inline static void run(int row, int col, const Lhs& lhs, const Rhs& rhs, PacketScalar &res)
  {
    ei_packet_product_unroller<true, Index-1, Size, Lhs, Rhs, PacketScalar>::run(row, col, lhs, rhs, res);
    res =  ei_pmadd(ei_pset1(lhs.coeff(row, Index)), rhs.template packet<Aligned>(Index, col), res);
  }
};

template<int Index, int Size, typename Lhs, typename Rhs, typename PacketScalar>
struct ei_packet_product_unroller<false, Index, Size, Lhs, Rhs, PacketScalar>
{
  inline static void run(int row, int col, const Lhs& lhs, const Rhs& rhs, PacketScalar &res)
  {
    ei_packet_product_unroller<false, Index-1, Size, Lhs, Rhs, PacketScalar>::run(row, col, lhs, rhs, res);
    res =  ei_pmadd(lhs.template packet<Aligned>(row, Index), ei_pset1(rhs.coeff(Index, col)), res);
  }
};

template<int Size, typename Lhs, typename Rhs, typename PacketScalar>
struct ei_packet_product_unroller<true, 0, Size, Lhs, Rhs, PacketScalar>
{
  inline static void run(int row, int col, const Lhs& lhs, const Rhs& rhs, PacketScalar &res)
  {
    res = ei_pmul(ei_pset1(lhs.coeff(row, 0)),rhs.template packet<Aligned>(0, col));
  }
};

template<int Size, typename Lhs, typename Rhs, typename PacketScalar>
struct ei_packet_product_unroller<false, 0, Size, Lhs, Rhs, PacketScalar>
{
  inline static void run(int row, int col, const Lhs& lhs, const Rhs& rhs, PacketScalar &res)
  {
    res = ei_pmul(lhs.template packet<Aligned>(row, 0), ei_pset1(rhs.coeff(0, col)));
  }
};

template<bool RowMajor, int Index, typename Lhs, typename Rhs, typename PacketScalar>
struct ei_packet_product_unroller<RowMajor, Index, Dynamic, Lhs, Rhs, PacketScalar>
{
  inline static void run(int, int, const Lhs&, const Rhs&, PacketScalar&) {}
};

template<int Index, typename Lhs, typename Rhs, typename PacketScalar>
struct ei_packet_product_unroller<false, Index, Dynamic, Lhs, Rhs, PacketScalar>
{
  inline static void run(int, int, const Lhs&, const Rhs&, PacketScalar&) {}
};

template<typename Lhs, typename Rhs, typename PacketScalar>
struct ei_packet_product_unroller<false, 0, Dynamic, Lhs, Rhs, PacketScalar>
{
  static void run(int, int, const Lhs&, const Rhs&, PacketScalar&) {}
};

template<typename Product, bool RowMajor = true> struct ProductPacketImpl {
  inline static typename Product::PacketScalar execute(const Product& product, int row, int col)
  { return product._packetRowMajor(row,col); }
};

template<typename Product> struct ProductPacketImpl<Product, false> {
  inline static typename Product::PacketScalar execute(const Product& product, int row, int col)
  { return product._packetColumnMajor(row,col); }
};

/** \class Product
  *
  * \brief Expression of the product of two matrices
  *
  * \param Lhs the type of the left-hand side
  * \param Rhs the type of the right-hand side
  * \param EvalMode internal use only
  *
  * This class represents an expression of the product of two matrices.
  * It is the return type of the operator* between matrices, and most of the time
  * this is the only way it is used.
  *
  * \sa class Sum, class Difference
  */
template<typename Lhs, typename Rhs> struct ei_product_eval_mode
{
  enum{ value =  Lhs::MaxRowsAtCompileTime >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
              && Rhs::MaxColsAtCompileTime >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
              && (!( (Lhs::Flags&RowMajorBit) && ((Rhs::Flags&RowMajorBit) ^ RowMajorBit)))
              ? CacheFriendlyProduct : NormalProduct };
};

template<typename Lhs, typename Rhs, int EvalMode>
struct ei_traits<Product<Lhs, Rhs, EvalMode> >
{
  typedef typename Lhs::Scalar Scalar;
  typedef typename ei_nested<Lhs,Rhs::ColsAtCompileTime>::type LhsNested;
  typedef typename ei_nested<Rhs,Lhs::RowsAtCompileTime>::type RhsNested;
  typedef typename ei_unref<LhsNested>::type _LhsNested;
  typedef typename ei_unref<RhsNested>::type _RhsNested;
  enum {
    LhsCoeffReadCost = _LhsNested::CoeffReadCost,
    RhsCoeffReadCost = _RhsNested::CoeffReadCost,
    LhsFlags = _LhsNested::Flags,
    RhsFlags = _RhsNested::Flags,
    RowsAtCompileTime = Lhs::RowsAtCompileTime,
    ColsAtCompileTime = Rhs::ColsAtCompileTime,
    MaxRowsAtCompileTime = Lhs::MaxRowsAtCompileTime,
    MaxColsAtCompileTime = Rhs::MaxColsAtCompileTime,
    _RhsPacketAccess = (RhsFlags & RowMajorBit) && (RhsFlags & PacketAccessBit) && (ColsAtCompileTime % ei_packet_traits<Scalar>::size == 0),
    _LhsPacketAccess = (!(LhsFlags & RowMajorBit)) && (LhsFlags & PacketAccessBit) && (RowsAtCompileTime % ei_packet_traits<Scalar>::size == 0),
    _PacketAccess = (_LhsPacketAccess || _RhsPacketAccess) ? 1 : 0,
    _RowMajor = (RhsFlags & RowMajorBit)
              && (EvalMode==(int)CacheFriendlyProduct ? (int)LhsFlags & RowMajorBit : (!_LhsPacketAccess)),
    _LostBits = HereditaryBits & ~(
                (_RowMajor ? 0 : RowMajorBit)
              | ((RowsAtCompileTime == Dynamic || ColsAtCompileTime == Dynamic) ? 0 : LargeBit)),
    Flags = ((unsigned int)(LhsFlags | RhsFlags) & _LostBits)
          | EvalBeforeAssigningBit
          | EvalBeforeNestingBit
          | (_PacketAccess ? PacketAccessBit : 0),
    CoeffReadCost
      = Lhs::ColsAtCompileTime == Dynamic
      ? Dynamic
      : Lhs::ColsAtCompileTime
        * (NumTraits<Scalar>::MulCost + LhsCoeffReadCost + RhsCoeffReadCost)
        + (Lhs::ColsAtCompileTime - 1) * NumTraits<Scalar>::AddCost
  };
};

template<typename Lhs, typename Rhs, int EvalMode> class Product : ei_no_assignment_operator,
  public MatrixBase<Product<Lhs, Rhs, EvalMode> >
{
  public:

    EIGEN_GENERIC_PUBLIC_INTERFACE(Product)
    friend class ProductPacketImpl<Product,Flags&RowMajorBit>;
    typedef typename ei_traits<Product>::LhsNested LhsNested;
    typedef typename ei_traits<Product>::RhsNested RhsNested;
    typedef typename ei_traits<Product>::_LhsNested _LhsNested;
    typedef typename ei_traits<Product>::_RhsNested _RhsNested;

    inline Product(const Lhs& lhs, const Rhs& rhs)
      : m_lhs(lhs), m_rhs(rhs)
    {
      ei_assert(lhs.cols() == rhs.rows());
    }

    /** \internal */
    template<typename DestDerived, int AlignedMode>
    void _cacheOptimalEval(DestDerived& res, ei_meta_false) const;
    #ifdef EIGEN_VECTORIZE
    template<typename DestDerived, int AlignedMode>
    void _cacheOptimalEval(DestDerived& res, ei_meta_true) const;
    #endif

  private:

    inline int _rows() const { return m_lhs.rows(); }
    inline int _cols() const { return m_rhs.cols(); }

    const Scalar _coeff(int row, int col) const
    {
      Scalar res;
      const bool unroll = CoeffReadCost <= EIGEN_UNROLLING_LIMIT;
      if(unroll)
      {
        ei_product_unroller<Lhs::ColsAtCompileTime-1,
                            unroll ? Lhs::ColsAtCompileTime : Dynamic,
                            _LhsNested, _RhsNested>
          ::run(row, col, m_lhs, m_rhs, res);
      }
      else
      {
        res = m_lhs.coeff(row, 0) * m_rhs.coeff(0, col);
        for(int i = 1; i < m_lhs.cols(); i++)
          res += m_lhs.coeff(row, i) * m_rhs.coeff(i, col);
      }
      return res;
    }

    template<int LoadMode>
    const PacketScalar _packet(int row, int col) const
    {
      if(Lhs::ColsAtCompileTime <= EIGEN_UNROLLING_LIMIT)
      {
        PacketScalar res;
        ei_packet_product_unroller<Flags&RowMajorBit, Lhs::ColsAtCompileTime-1,
                            Lhs::ColsAtCompileTime <= EIGEN_UNROLLING_LIMIT
                              ? Lhs::ColsAtCompileTime : Dynamic,
                            _LhsNested, _RhsNested, PacketScalar>
          ::run(row, col, m_lhs, m_rhs, res);
        return res;
      }
      else
        return ProductPacketImpl<Product,Flags&RowMajorBit>::execute(*this, row, col);
    }

    const PacketScalar _packetRowMajor(int row, int col) const
    {
      PacketScalar res;
      res = ei_pmul(ei_pset1(m_lhs.coeff(row, 0)),m_rhs.template packet<Aligned>(0, col));
      for(int i = 1; i < m_lhs.cols(); i++)
        res =  ei_pmadd(ei_pset1(m_lhs.coeff(row, i)), m_rhs.template packet<Aligned>(i, col), res);
      return res;
    }

    const PacketScalar _packetColumnMajor(int row, int col) const
    {
      PacketScalar res;
      res = ei_pmul(m_lhs.template packet<Aligned>(row, 0), ei_pset1(m_rhs.coeff(0, col)));
      for(int i = 1; i < m_lhs.cols(); i++)
        res =  ei_pmadd(m_lhs.template packet<Aligned>(row, i), ei_pset1(m_rhs.coeff(i, col)), res);
      return res;
//       const PacketScalar tmp[4];
//       ei_punpack(m_rhs.packet(0,col), tmp);
//
//       return
//         ei_pmadd(m_lhs.packet(row, 0), tmp[0],
//           ei_pmadd(m_lhs.packet(row, 1), tmp[1],
//             ei_pmadd(m_lhs.packet(row, 2), tmp[2]
//               ei_pmul(m_lhs.packet(row, 3), tmp[3]))));
    }


  protected:
    const LhsNested m_lhs;
    const RhsNested m_rhs;
};

/** \returns the matrix product of \c *this and \a other.
  *
  * \note This function causes an immediate evaluation. If you want to perform a matrix product
  * without immediate evaluation, call .lazy() on one of the matrices before taking the product.
  *
  * \sa lazy(), operator*=(const MatrixBase&)
  */
template<typename Derived>
template<typename OtherDerived>
inline const Product<Derived,OtherDerived>
MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
{
  return Product<Derived,OtherDerived>(derived(), other.derived());
}

/** replaces \c *this by \c *this * \a other.
  *
  * \returns a reference to \c *this
  */
template<typename Derived>
template<typename OtherDerived>
inline Derived &
MatrixBase<Derived>::operator*=(const MatrixBase<OtherDerived> &other)
{
  return *this = *this * other;
}

template<typename Derived>
template<typename Lhs, typename Rhs>
inline Derived& MatrixBase<Derived>::lazyAssign(const Product<Lhs,Rhs,CacheFriendlyProduct>& product)
{
  product.template _cacheOptimalEval<Derived, Aligned>(derived(),
    #ifdef EIGEN_VECTORIZE
    typename ei_meta_if<Flags & PacketAccessBit, ei_meta_true, ei_meta_false>::ret()
    #else
    ei_meta_false()
    #endif
    );
  return derived();
}

template<typename Lhs, typename Rhs, int EvalMode>
template<typename DestDerived, int AlignedMode>
void Product<Lhs,Rhs,EvalMode>::_cacheOptimalEval(DestDerived& res, ei_meta_false) const
{
  res.setZero();
  const int cols4 = m_lhs.cols() & 0xfffffffC;
  if (Lhs::Flags&RowMajorBit)
  {
//     std::cout << "opt rhs\n";
    int j=0;
    for(; j<cols4; j+=4)
    {
      for(int k=0; k<this->rows(); ++k)
      {
        const Scalar tmp0 = m_lhs.coeff(k,j  );
        const Scalar tmp1 = m_lhs.coeff(k,j+1);
        const Scalar tmp2 = m_lhs.coeff(k,j+2);
        const Scalar tmp3 = m_lhs.coeff(k,j+3);
        for (int i=0; i<this->cols(); ++i)
          res.coeffRef(k,i) += tmp0 * m_rhs.coeff(j+0,i) + tmp1 * m_rhs.coeff(j+1,i)
                             + tmp2 * m_rhs.coeff(j+2,i) + tmp3 * m_rhs.coeff(j+3,i);
      }
    }
    for(; j<m_lhs.cols(); ++j)
    {
      for(int k=0; k<this->rows(); ++k)
      {
        const Scalar tmp = m_rhs.coeff(k,j);
        for (int i=0; i<this->cols(); ++i)
          res.coeffRef(k,i) += tmp * m_lhs.coeff(j,i);
      }
    }
  }
  else
  {
//     std::cout << "opt lhs\n";
    int j = 0;
    for(; j<cols4; j+=4)
    {
      for(int k=0; k<this->cols(); ++k)
      {
        const Scalar tmp0 = m_rhs.coeff(j  ,k);
        const Scalar tmp1 = m_rhs.coeff(j+1,k);
        const Scalar tmp2 = m_rhs.coeff(j+2,k);
        const Scalar tmp3 = m_rhs.coeff(j+3,k);
        for (int i=0; i<this->rows(); ++i)
          res.coeffRef(i,k) += tmp0 * m_lhs.coeff(i,j+0) + tmp1 * m_lhs.coeff(i,j+1)
                             + tmp2 * m_lhs.coeff(i,j+2) + tmp3 * m_lhs.coeff(i,j+3);
      }
    }
    for(; j<m_lhs.cols(); ++j)
    {
      for(int k=0; k<this->cols(); ++k)
      {
        const Scalar tmp = m_rhs.coeff(j,k);
        for (int i=0; i<this->rows(); ++i)
          res.coeffRef(i,k) += tmp * m_lhs.coeff(i,j);
      }
    }
  }
}

#ifdef EIGEN_VECTORIZE
template<typename Lhs, typename Rhs, int EvalMode>
template<typename DestDerived, int AlignedMode>
void Product<Lhs,Rhs,EvalMode>::_cacheOptimalEval(DestDerived& res, ei_meta_true) const
{

  if (((Lhs::Flags&RowMajorBit) && (_cols() % ei_packet_traits<Scalar>::size != 0))
    || (_rows() % ei_packet_traits<Scalar>::size != 0))
  {
    return _cacheOptimalEval<DestDerived, AlignedMode>(res, ei_meta_false());
  }

  res.setZero();
  const int cols4 = m_lhs.cols() & 0xfffffffC;
  if (Lhs::Flags&RowMajorBit)
  {
//     std::cout << "packet rhs\n";
    int j=0;
    for(; j<cols4; j+=4)
    {
      for(int k=0; k<this->rows(); k++)
      {
        const typename ei_packet_traits<Scalar>::type tmp0 = ei_pset1(m_lhs.coeff(k,j+0));
        const typename ei_packet_traits<Scalar>::type tmp1 = ei_pset1(m_lhs.coeff(k,j+1));
        const typename ei_packet_traits<Scalar>::type tmp2 = ei_pset1(m_lhs.coeff(k,j+2));
        const typename ei_packet_traits<Scalar>::type tmp3 = ei_pset1(m_lhs.coeff(k,j+3));
        for (int i=0; i<this->cols(); i+=ei_packet_traits<Scalar>::size)
        {
          res.template writePacket<AlignedMode>(k,i,
            ei_pmadd(tmp0, m_rhs.template packet<AlignedMode>(j+0,i),
              ei_pmadd(tmp1, m_rhs.template packet<AlignedMode>(j+1,i),
                ei_pmadd(tmp2, m_rhs.template packet<AlignedMode>(j+2,i),
                  ei_pmadd(tmp3, m_rhs.template packet<AlignedMode>(j+3,i),
                    res.template packet<AlignedMode>(k,i)))))
          );
        }
      }
    }
    for(; j<m_lhs.cols(); ++j)
    {
      for(int k=0; k<this->rows(); k++)
      {
        const typename ei_packet_traits<Scalar>::type tmp = ei_pset1(m_lhs.coeff(k,j));
        for (int i=0; i<this->cols(); i+=ei_packet_traits<Scalar>::size)
          res.template writePacket<AlignedMode>(k,i,
            ei_pmadd(tmp, m_rhs.template packet<AlignedMode>(j,i), res.template packet<AlignedMode>(k,i)));
      }
    }
  }
  else
  {
//     std::cout << "packet lhs\n";
    int k=0;
    for(; k<cols4; k+=4)
    {
      for(int j=0; j<this->cols(); j+=1)
      {
        const typename ei_packet_traits<Scalar>::type tmp0 = ei_pset1(m_rhs.coeff(k+0,j));
        const typename ei_packet_traits<Scalar>::type tmp1 = ei_pset1(m_rhs.coeff(k+1,j));
        const typename ei_packet_traits<Scalar>::type tmp2 = ei_pset1(m_rhs.coeff(k+2,j));
        const typename ei_packet_traits<Scalar>::type tmp3 = ei_pset1(m_rhs.coeff(k+3,j));

        for (int i=0; i<this->rows(); i+=ei_packet_traits<Scalar>::size)
        {
          res.template writePacket<AlignedMode>(i,j,
            ei_pmadd(tmp0, m_lhs.template packet<AlignedMode>(i,k),
              ei_pmadd(tmp1, m_lhs.template packet<AlignedMode>(i,k+1),
                ei_pmadd(tmp2, m_lhs.template packet<AlignedMode>(i,k+2),
                  ei_pmadd(tmp3, m_lhs.template packet<AlignedMode>(i,k+3),
                    res.template packet<AlignedMode>(i,j)))))
          );
        }
      }
    }
    for(; k<m_lhs.cols(); ++k)
    {
      for(int j=0; j<this->cols(); j++)
      {
        const typename ei_packet_traits<Scalar>::type tmp = ei_pset1(m_rhs.coeff(k,j));
        for (int i=0; i<this->rows(); i+=ei_packet_traits<Scalar>::size)
          res.template writePacket<AlignedMode>(k,j,
            ei_pmadd(tmp, m_lhs.template packet<AlignedMode>(i,k), res.template packet<AlignedMode>(i,j)));
      }
    }
  }
}
#endif // EIGEN_VECTORIZE

#endif // EIGEN_PRODUCT_H