1// This file is part of Eigen, a lightweight C++ template library
2// for linear algebra.
3//
4// Copyright (C) 2008-2009 Guillaume Saupin <guillaume.saupin@cea.fr>
5//
6// This Source Code Form is subject to the terms of the Mozilla
7// Public License v. 2.0. If a copy of the MPL was not distributed
8// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
9
10#ifndef EIGEN_SKYLINEMATRIX_H
11#define EIGEN_SKYLINEMATRIX_H
12
13#include "SkylineStorage.h"
14#include "SkylineMatrixBase.h"
15
16namespace Eigen {
17
18/** \ingroup Skyline_Module
19 *
20 * \class SkylineMatrix
21 *
22 * \brief The main skyline matrix class
23 *
24 * This class implements a skyline matrix using the very uncommon storage
25 * scheme.
26 *
27 * \param _Scalar the scalar type, i.e. the type of the coefficients
28 * \param _Options Union of bit flags controlling the storage scheme. Currently the only possibility
29 *                 is RowMajor. The default is 0 which means column-major.
30 *
31 *
32 */
33namespace internal {
34template<typename _Scalar, int _Options>
35struct traits<SkylineMatrix<_Scalar, _Options> > {
36    typedef _Scalar Scalar;
37    typedef Sparse StorageKind;
38
39    enum {
40        RowsAtCompileTime = Dynamic,
41        ColsAtCompileTime = Dynamic,
42        MaxRowsAtCompileTime = Dynamic,
43        MaxColsAtCompileTime = Dynamic,
44        Flags = SkylineBit | _Options,
45        CoeffReadCost = NumTraits<Scalar>::ReadCost,
46    };
47};
48}
49
50template<typename _Scalar, int _Options>
51class SkylineMatrix
52: public SkylineMatrixBase<SkylineMatrix<_Scalar, _Options> > {
53public:
54    EIGEN_SKYLINE_GENERIC_PUBLIC_INTERFACE(SkylineMatrix)
55    EIGEN_SKYLINE_INHERIT_ASSIGNMENT_OPERATOR(SkylineMatrix, +=)
56    EIGEN_SKYLINE_INHERIT_ASSIGNMENT_OPERATOR(SkylineMatrix, -=)
57
58    using Base::IsRowMajor;
59
60protected:
61
62    typedef SkylineMatrix<Scalar, (Flags&~RowMajorBit) | (IsRowMajor ? RowMajorBit : 0) > TransposedSkylineMatrix;
63
64    Index m_outerSize;
65    Index m_innerSize;
66
67public:
68    Index* m_colStartIndex;
69    Index* m_rowStartIndex;
70    SkylineStorage<Scalar> m_data;
71
72public:
73
74    inline Index rows() const {
75        return IsRowMajor ? m_outerSize : m_innerSize;
76    }
77
78    inline Index cols() const {
79        return IsRowMajor ? m_innerSize : m_outerSize;
80    }
81
82    inline Index innerSize() const {
83        return m_innerSize;
84    }
85
86    inline Index outerSize() const {
87        return m_outerSize;
88    }
89
90    inline Index upperNonZeros() const {
91        return m_data.upperSize();
92    }
93
94    inline Index lowerNonZeros() const {
95        return m_data.lowerSize();
96    }
97
98    inline Index upperNonZeros(Index j) const {
99        return m_colStartIndex[j + 1] - m_colStartIndex[j];
100    }
101
102    inline Index lowerNonZeros(Index j) const {
103        return m_rowStartIndex[j + 1] - m_rowStartIndex[j];
104    }
105
106    inline const Scalar* _diagPtr() const {
107        return &m_data.diag(0);
108    }
109
110    inline Scalar* _diagPtr() {
111        return &m_data.diag(0);
112    }
113
114    inline const Scalar* _upperPtr() const {
115        return &m_data.upper(0);
116    }
117
118    inline Scalar* _upperPtr() {
119        return &m_data.upper(0);
120    }
121
122    inline const Scalar* _lowerPtr() const {
123        return &m_data.lower(0);
124    }
125
126    inline Scalar* _lowerPtr() {
127        return &m_data.lower(0);
128    }
129
130    inline const Index* _upperProfilePtr() const {
131        return &m_data.upperProfile(0);
132    }
133
134    inline Index* _upperProfilePtr() {
135        return &m_data.upperProfile(0);
136    }
137
138    inline const Index* _lowerProfilePtr() const {
139        return &m_data.lowerProfile(0);
140    }
141
142    inline Index* _lowerProfilePtr() {
143        return &m_data.lowerProfile(0);
144    }
145
146    inline Scalar coeff(Index row, Index col) const {
147        const Index outer = IsRowMajor ? row : col;
148        const Index inner = IsRowMajor ? col : row;
149
150        eigen_assert(outer < outerSize());
151        eigen_assert(inner < innerSize());
152
153        if (outer == inner)
154            return this->m_data.diag(outer);
155
156        if (IsRowMajor) {
157            if (inner > outer) //upper matrix
158            {
159                const Index minOuterIndex = inner - m_data.upperProfile(inner);
160                if (outer >= minOuterIndex)
161                    return this->m_data.upper(m_colStartIndex[inner] + outer - (inner - m_data.upperProfile(inner)));
162                else
163                    return Scalar(0);
164            }
165            if (inner < outer) //lower matrix
166            {
167                const Index minInnerIndex = outer - m_data.lowerProfile(outer);
168                if (inner >= minInnerIndex)
169                    return this->m_data.lower(m_rowStartIndex[outer] + inner - (outer - m_data.lowerProfile(outer)));
170                else
171                    return Scalar(0);
172            }
173            return m_data.upper(m_colStartIndex[inner] + outer - inner);
174        } else {
175            if (outer > inner) //upper matrix
176            {
177                const Index maxOuterIndex = inner + m_data.upperProfile(inner);
178                if (outer <= maxOuterIndex)
179                    return this->m_data.upper(m_colStartIndex[inner] + (outer - inner));
180                else
181                    return Scalar(0);
182            }
183            if (outer < inner) //lower matrix
184            {
185                const Index maxInnerIndex = outer + m_data.lowerProfile(outer);
186
187                if (inner <= maxInnerIndex)
188                    return this->m_data.lower(m_rowStartIndex[outer] + (inner - outer));
189                else
190                    return Scalar(0);
191            }
192        }
193    }
194
195    inline Scalar& coeffRef(Index row, Index col) {
196        const Index outer = IsRowMajor ? row : col;
197        const Index inner = IsRowMajor ? col : row;
198
199        eigen_assert(outer < outerSize());
200        eigen_assert(inner < innerSize());
201
202        if (outer == inner)
203            return this->m_data.diag(outer);
204
205        if (IsRowMajor) {
206            if (col > row) //upper matrix
207            {
208                const Index minOuterIndex = inner - m_data.upperProfile(inner);
209                eigen_assert(outer >= minOuterIndex && "you try to acces a coeff that do not exist in the storage");
210                return this->m_data.upper(m_colStartIndex[inner] + outer - (inner - m_data.upperProfile(inner)));
211            }
212            if (col < row) //lower matrix
213            {
214                const Index minInnerIndex = outer - m_data.lowerProfile(outer);
215                eigen_assert(inner >= minInnerIndex && "you try to acces a coeff that do not exist in the storage");
216                return this->m_data.lower(m_rowStartIndex[outer] + inner - (outer - m_data.lowerProfile(outer)));
217            }
218        } else {
219            if (outer > inner) //upper matrix
220            {
221                const Index maxOuterIndex = inner + m_data.upperProfile(inner);
222                eigen_assert(outer <= maxOuterIndex && "you try to acces a coeff that do not exist in the storage");
223                return this->m_data.upper(m_colStartIndex[inner] + (outer - inner));
224            }
225            if (outer < inner) //lower matrix
226            {
227                const Index maxInnerIndex = outer + m_data.lowerProfile(outer);
228                eigen_assert(inner <= maxInnerIndex && "you try to acces a coeff that do not exist in the storage");
229                return this->m_data.lower(m_rowStartIndex[outer] + (inner - outer));
230            }
231        }
232    }
233
234    inline Scalar coeffDiag(Index idx) const {
235        eigen_assert(idx < outerSize());
236        eigen_assert(idx < innerSize());
237        return this->m_data.diag(idx);
238    }
239
240    inline Scalar coeffLower(Index row, Index col) const {
241        const Index outer = IsRowMajor ? row : col;
242        const Index inner = IsRowMajor ? col : row;
243
244        eigen_assert(outer < outerSize());
245        eigen_assert(inner < innerSize());
246        eigen_assert(inner != outer);
247
248        if (IsRowMajor) {
249            const Index minInnerIndex = outer - m_data.lowerProfile(outer);
250            if (inner >= minInnerIndex)
251                return this->m_data.lower(m_rowStartIndex[outer] + inner - (outer - m_data.lowerProfile(outer)));
252            else
253                return Scalar(0);
254
255        } else {
256            const Index maxInnerIndex = outer + m_data.lowerProfile(outer);
257            if (inner <= maxInnerIndex)
258                return this->m_data.lower(m_rowStartIndex[outer] + (inner - outer));
259            else
260                return Scalar(0);
261        }
262    }
263
264    inline Scalar coeffUpper(Index row, Index col) const {
265        const Index outer = IsRowMajor ? row : col;
266        const Index inner = IsRowMajor ? col : row;
267
268        eigen_assert(outer < outerSize());
269        eigen_assert(inner < innerSize());
270        eigen_assert(inner != outer);
271
272        if (IsRowMajor) {
273            const Index minOuterIndex = inner - m_data.upperProfile(inner);
274            if (outer >= minOuterIndex)
275                return this->m_data.upper(m_colStartIndex[inner] + outer - (inner - m_data.upperProfile(inner)));
276            else
277                return Scalar(0);
278        } else {
279            const Index maxOuterIndex = inner + m_data.upperProfile(inner);
280            if (outer <= maxOuterIndex)
281                return this->m_data.upper(m_colStartIndex[inner] + (outer - inner));
282            else
283                return Scalar(0);
284        }
285    }
286
287    inline Scalar& coeffRefDiag(Index idx) {
288        eigen_assert(idx < outerSize());
289        eigen_assert(idx < innerSize());
290        return this->m_data.diag(idx);
291    }
292
293    inline Scalar& coeffRefLower(Index row, Index col) {
294        const Index outer = IsRowMajor ? row : col;
295        const Index inner = IsRowMajor ? col : row;
296
297        eigen_assert(outer < outerSize());
298        eigen_assert(inner < innerSize());
299        eigen_assert(inner != outer);
300
301        if (IsRowMajor) {
302            const Index minInnerIndex = outer - m_data.lowerProfile(outer);
303            eigen_assert(inner >= minInnerIndex && "you try to acces a coeff that do not exist in the storage");
304            return this->m_data.lower(m_rowStartIndex[outer] + inner - (outer - m_data.lowerProfile(outer)));
305        } else {
306            const Index maxInnerIndex = outer + m_data.lowerProfile(outer);
307            eigen_assert(inner <= maxInnerIndex && "you try to acces a coeff that do not exist in the storage");
308            return this->m_data.lower(m_rowStartIndex[outer] + (inner - outer));
309        }
310    }
311
312    inline bool coeffExistLower(Index row, Index col) {
313        const Index outer = IsRowMajor ? row : col;
314        const Index inner = IsRowMajor ? col : row;
315
316        eigen_assert(outer < outerSize());
317        eigen_assert(inner < innerSize());
318        eigen_assert(inner != outer);
319
320        if (IsRowMajor) {
321            const Index minInnerIndex = outer - m_data.lowerProfile(outer);
322            return inner >= minInnerIndex;
323        } else {
324            const Index maxInnerIndex = outer + m_data.lowerProfile(outer);
325            return inner <= maxInnerIndex;
326        }
327    }
328
329    inline Scalar& coeffRefUpper(Index row, Index col) {
330        const Index outer = IsRowMajor ? row : col;
331        const Index inner = IsRowMajor ? col : row;
332
333        eigen_assert(outer < outerSize());
334        eigen_assert(inner < innerSize());
335        eigen_assert(inner != outer);
336
337        if (IsRowMajor) {
338            const Index minOuterIndex = inner - m_data.upperProfile(inner);
339            eigen_assert(outer >= minOuterIndex && "you try to acces a coeff that do not exist in the storage");
340            return this->m_data.upper(m_colStartIndex[inner] + outer - (inner - m_data.upperProfile(inner)));
341        } else {
342            const Index maxOuterIndex = inner + m_data.upperProfile(inner);
343            eigen_assert(outer <= maxOuterIndex && "you try to acces a coeff that do not exist in the storage");
344            return this->m_data.upper(m_colStartIndex[inner] + (outer - inner));
345        }
346    }
347
348    inline bool coeffExistUpper(Index row, Index col) {
349        const Index outer = IsRowMajor ? row : col;
350        const Index inner = IsRowMajor ? col : row;
351
352        eigen_assert(outer < outerSize());
353        eigen_assert(inner < innerSize());
354        eigen_assert(inner != outer);
355
356        if (IsRowMajor) {
357            const Index minOuterIndex = inner - m_data.upperProfile(inner);
358            return outer >= minOuterIndex;
359        } else {
360            const Index maxOuterIndex = inner + m_data.upperProfile(inner);
361            return outer <= maxOuterIndex;
362        }
363    }
364
365
366protected:
367
368public:
369    class InnerUpperIterator;
370    class InnerLowerIterator;
371
372    class OuterUpperIterator;
373    class OuterLowerIterator;
374
375    /** Removes all non zeros */
376    inline void setZero() {
377        m_data.clear();
378        memset(m_colStartIndex, 0, (m_outerSize + 1) * sizeof (Index));
379        memset(m_rowStartIndex, 0, (m_outerSize + 1) * sizeof (Index));
380    }
381
382    /** \returns the number of non zero coefficients */
383    inline Index nonZeros() const {
384        return m_data.diagSize() + m_data.upperSize() + m_data.lowerSize();
385    }
386
387    /** Preallocates \a reserveSize non zeros */
388    inline void reserve(Index reserveSize, Index reserveUpperSize, Index reserveLowerSize) {
389        m_data.reserve(reserveSize, reserveUpperSize, reserveLowerSize);
390    }
391
392    /** \returns a reference to a novel non zero coefficient with coordinates \a row x \a col.
393
394     *
395     * \warning This function can be extremely slow if the non zero coefficients
396     * are not inserted in a coherent order.
397     *
398     * After an insertion session, you should call the finalize() function.
399     */
400    EIGEN_DONT_INLINE Scalar & insert(Index row, Index col) {
401        const Index outer = IsRowMajor ? row : col;
402        const Index inner = IsRowMajor ? col : row;
403
404        eigen_assert(outer < outerSize());
405        eigen_assert(inner < innerSize());
406
407        if (outer == inner)
408            return m_data.diag(col);
409
410        if (IsRowMajor) {
411            if (outer < inner) //upper matrix
412            {
413                Index minOuterIndex = 0;
414                minOuterIndex = inner - m_data.upperProfile(inner);
415
416                if (outer < minOuterIndex) //The value does not yet exist
417                {
418                    const Index previousProfile = m_data.upperProfile(inner);
419
420                    m_data.upperProfile(inner) = inner - outer;
421
422
423                    const Index bandIncrement = m_data.upperProfile(inner) - previousProfile;
424                    //shift data stored after this new one
425                    const Index stop = m_colStartIndex[cols()];
426                    const Index start = m_colStartIndex[inner];
427
428
429                    for (Index innerIdx = stop; innerIdx >= start; innerIdx--) {
430                        m_data.upper(innerIdx + bandIncrement) = m_data.upper(innerIdx);
431                    }
432
433                    for (Index innerIdx = cols(); innerIdx > inner; innerIdx--) {
434                        m_colStartIndex[innerIdx] += bandIncrement;
435                    }
436
437                    //zeros new data
438                    memset(this->_upperPtr() + start, 0, (bandIncrement - 1) * sizeof (Scalar));
439
440                    return m_data.upper(m_colStartIndex[inner]);
441                } else {
442                    return m_data.upper(m_colStartIndex[inner] + outer - (inner - m_data.upperProfile(inner)));
443                }
444            }
445
446            if (outer > inner) //lower matrix
447            {
448                const Index minInnerIndex = outer - m_data.lowerProfile(outer);
449                if (inner < minInnerIndex) //The value does not yet exist
450                {
451                    const Index previousProfile = m_data.lowerProfile(outer);
452                    m_data.lowerProfile(outer) = outer - inner;
453
454                    const Index bandIncrement = m_data.lowerProfile(outer) - previousProfile;
455                    //shift data stored after this new one
456                    const Index stop = m_rowStartIndex[rows()];
457                    const Index start = m_rowStartIndex[outer];
458
459
460                    for (Index innerIdx = stop; innerIdx >= start; innerIdx--) {
461                        m_data.lower(innerIdx + bandIncrement) = m_data.lower(innerIdx);
462                    }
463
464                    for (Index innerIdx = rows(); innerIdx > outer; innerIdx--) {
465                        m_rowStartIndex[innerIdx] += bandIncrement;
466                    }
467
468                    //zeros new data
469                    memset(this->_lowerPtr() + start, 0, (bandIncrement - 1) * sizeof (Scalar));
470                    return m_data.lower(m_rowStartIndex[outer]);
471                } else {
472                    return m_data.lower(m_rowStartIndex[outer] + inner - (outer - m_data.lowerProfile(outer)));
473                }
474            }
475        } else {
476            if (outer > inner) //upper matrix
477            {
478                const Index maxOuterIndex = inner + m_data.upperProfile(inner);
479                if (outer > maxOuterIndex) //The value does not yet exist
480                {
481                    const Index previousProfile = m_data.upperProfile(inner);
482                    m_data.upperProfile(inner) = outer - inner;
483
484                    const Index bandIncrement = m_data.upperProfile(inner) - previousProfile;
485                    //shift data stored after this new one
486                    const Index stop = m_rowStartIndex[rows()];
487                    const Index start = m_rowStartIndex[inner + 1];
488
489                    for (Index innerIdx = stop; innerIdx >= start; innerIdx--) {
490                        m_data.upper(innerIdx + bandIncrement) = m_data.upper(innerIdx);
491                    }
492
493                    for (Index innerIdx = inner + 1; innerIdx < outerSize() + 1; innerIdx++) {
494                        m_rowStartIndex[innerIdx] += bandIncrement;
495                    }
496                    memset(this->_upperPtr() + m_rowStartIndex[inner] + previousProfile + 1, 0, (bandIncrement - 1) * sizeof (Scalar));
497                    return m_data.upper(m_rowStartIndex[inner] + m_data.upperProfile(inner));
498                } else {
499                    return m_data.upper(m_rowStartIndex[inner] + (outer - inner));
500                }
501            }
502
503            if (outer < inner) //lower matrix
504            {
505                const Index maxInnerIndex = outer + m_data.lowerProfile(outer);
506                if (inner > maxInnerIndex) //The value does not yet exist
507                {
508                    const Index previousProfile = m_data.lowerProfile(outer);
509                    m_data.lowerProfile(outer) = inner - outer;
510
511                    const Index bandIncrement = m_data.lowerProfile(outer) - previousProfile;
512                    //shift data stored after this new one
513                    const Index stop = m_colStartIndex[cols()];
514                    const Index start = m_colStartIndex[outer + 1];
515
516                    for (Index innerIdx = stop; innerIdx >= start; innerIdx--) {
517                        m_data.lower(innerIdx + bandIncrement) = m_data.lower(innerIdx);
518                    }
519
520                    for (Index innerIdx = outer + 1; innerIdx < outerSize() + 1; innerIdx++) {
521                        m_colStartIndex[innerIdx] += bandIncrement;
522                    }
523                    memset(this->_lowerPtr() + m_colStartIndex[outer] + previousProfile + 1, 0, (bandIncrement - 1) * sizeof (Scalar));
524                    return m_data.lower(m_colStartIndex[outer] + m_data.lowerProfile(outer));
525                } else {
526                    return m_data.lower(m_colStartIndex[outer] + (inner - outer));
527                }
528            }
529        }
530    }
531
532    /** Must be called after inserting a set of non zero entries.
533     */
534    inline void finalize() {
535        if (IsRowMajor) {
536            if (rows() > cols())
537                m_data.resize(cols(), cols(), rows(), m_colStartIndex[cols()] + 1, m_rowStartIndex[rows()] + 1);
538            else
539                m_data.resize(rows(), cols(), rows(), m_colStartIndex[cols()] + 1, m_rowStartIndex[rows()] + 1);
540
541            //            eigen_assert(rows() == cols() && "memory reorganisatrion only works with suare matrix");
542            //
543            //            Scalar* newArray = new Scalar[m_colStartIndex[cols()] + 1 + m_rowStartIndex[rows()] + 1];
544            //            Index dataIdx = 0;
545            //            for (Index row = 0; row < rows(); row++) {
546            //
547            //                const Index nbLowerElts = m_rowStartIndex[row + 1] - m_rowStartIndex[row];
548            //                //                std::cout << "nbLowerElts" << nbLowerElts << std::endl;
549            //                memcpy(newArray + dataIdx, m_data.m_lower + m_rowStartIndex[row], nbLowerElts * sizeof (Scalar));
550            //                m_rowStartIndex[row] = dataIdx;
551            //                dataIdx += nbLowerElts;
552            //
553            //                const Index nbUpperElts = m_colStartIndex[row + 1] - m_colStartIndex[row];
554            //                memcpy(newArray + dataIdx, m_data.m_upper + m_colStartIndex[row], nbUpperElts * sizeof (Scalar));
555            //                m_colStartIndex[row] = dataIdx;
556            //                dataIdx += nbUpperElts;
557            //
558            //
559            //            }
560            //            //todo : don't access m_data profile directly : add an accessor from SkylineMatrix
561            //            m_rowStartIndex[rows()] = m_rowStartIndex[rows()-1] + m_data.lowerProfile(rows()-1);
562            //            m_colStartIndex[cols()] = m_colStartIndex[cols()-1] + m_data.upperProfile(cols()-1);
563            //
564            //            delete[] m_data.m_lower;
565            //            delete[] m_data.m_upper;
566            //
567            //            m_data.m_lower = newArray;
568            //            m_data.m_upper = newArray;
569        } else {
570            if (rows() > cols())
571                m_data.resize(cols(), rows(), cols(), m_rowStartIndex[cols()] + 1, m_colStartIndex[cols()] + 1);
572            else
573                m_data.resize(rows(), rows(), cols(), m_rowStartIndex[rows()] + 1, m_colStartIndex[rows()] + 1);
574        }
575    }
576
577    inline void squeeze() {
578        finalize();
579        m_data.squeeze();
580    }
581
582    void prune(Scalar reference, RealScalar epsilon = dummy_precision<RealScalar > ()) {
583        //TODO
584    }
585
586    /** Resizes the matrix to a \a rows x \a cols matrix and initializes it to zero
587     * \sa resizeNonZeros(Index), reserve(), setZero()
588     */
589    void resize(size_t rows, size_t cols) {
590        const Index diagSize = rows > cols ? cols : rows;
591        m_innerSize = IsRowMajor ? cols : rows;
592
593        eigen_assert(rows == cols && "Skyline matrix must be square matrix");
594
595        if (diagSize % 2) { // diagSize is odd
596            const Index k = (diagSize - 1) / 2;
597
598            m_data.resize(diagSize, IsRowMajor ? cols : rows, IsRowMajor ? rows : cols,
599                    2 * k * k + k + 1,
600                    2 * k * k + k + 1);
601
602        } else // diagSize is even
603        {
604            const Index k = diagSize / 2;
605            m_data.resize(diagSize, IsRowMajor ? cols : rows, IsRowMajor ? rows : cols,
606                    2 * k * k - k + 1,
607                    2 * k * k - k + 1);
608        }
609
610        if (m_colStartIndex && m_rowStartIndex) {
611            delete[] m_colStartIndex;
612            delete[] m_rowStartIndex;
613        }
614        m_colStartIndex = new Index [cols + 1];
615        m_rowStartIndex = new Index [rows + 1];
616        m_outerSize = diagSize;
617
618        m_data.reset();
619        m_data.clear();
620
621        m_outerSize = diagSize;
622        memset(m_colStartIndex, 0, (cols + 1) * sizeof (Index));
623        memset(m_rowStartIndex, 0, (rows + 1) * sizeof (Index));
624    }
625
626    void resizeNonZeros(Index size) {
627        m_data.resize(size);
628    }
629
630    inline SkylineMatrix()
631    : m_outerSize(-1), m_innerSize(0), m_colStartIndex(0), m_rowStartIndex(0) {
632        resize(0, 0);
633    }
634
635    inline SkylineMatrix(size_t rows, size_t cols)
636    : m_outerSize(0), m_innerSize(0), m_colStartIndex(0), m_rowStartIndex(0) {
637        resize(rows, cols);
638    }
639
640    template<typename OtherDerived>
641    inline SkylineMatrix(const SkylineMatrixBase<OtherDerived>& other)
642    : m_outerSize(0), m_innerSize(0), m_colStartIndex(0), m_rowStartIndex(0) {
643        *this = other.derived();
644    }
645
646    inline SkylineMatrix(const SkylineMatrix & other)
647    : Base(), m_outerSize(0), m_innerSize(0), m_colStartIndex(0), m_rowStartIndex(0) {
648        *this = other.derived();
649    }
650
651    inline void swap(SkylineMatrix & other) {
652        //EIGEN_DBG_SKYLINE(std::cout << "SkylineMatrix:: swap\n");
653        std::swap(m_colStartIndex, other.m_colStartIndex);
654        std::swap(m_rowStartIndex, other.m_rowStartIndex);
655        std::swap(m_innerSize, other.m_innerSize);
656        std::swap(m_outerSize, other.m_outerSize);
657        m_data.swap(other.m_data);
658    }
659
660    inline SkylineMatrix & operator=(const SkylineMatrix & other) {
661        std::cout << "SkylineMatrix& operator=(const SkylineMatrix& other)\n";
662        if (other.isRValue()) {
663            swap(other.const_cast_derived());
664        } else {
665            resize(other.rows(), other.cols());
666            memcpy(m_colStartIndex, other.m_colStartIndex, (m_outerSize + 1) * sizeof (Index));
667            memcpy(m_rowStartIndex, other.m_rowStartIndex, (m_outerSize + 1) * sizeof (Index));
668            m_data = other.m_data;
669        }
670        return *this;
671    }
672
673    template<typename OtherDerived>
674            inline SkylineMatrix & operator=(const SkylineMatrixBase<OtherDerived>& other) {
675        const bool needToTranspose = (Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit);
676        if (needToTranspose) {
677            //         TODO
678            //            return *this;
679        } else {
680            // there is no special optimization
681            return SkylineMatrixBase<SkylineMatrix>::operator=(other.derived());
682        }
683    }
684
685    friend std::ostream & operator <<(std::ostream & s, const SkylineMatrix & m) {
686
687        EIGEN_DBG_SKYLINE(
688        std::cout << "upper elements : " << std::endl;
689        for (Index i = 0; i < m.m_data.upperSize(); i++)
690            std::cout << m.m_data.upper(i) << "\t";
691        std::cout << std::endl;
692        std::cout << "upper profile : " << std::endl;
693        for (Index i = 0; i < m.m_data.upperProfileSize(); i++)
694            std::cout << m.m_data.upperProfile(i) << "\t";
695        std::cout << std::endl;
696        std::cout << "lower startIdx : " << std::endl;
697        for (Index i = 0; i < m.m_data.upperProfileSize(); i++)
698            std::cout << (IsRowMajor ? m.m_colStartIndex[i] : m.m_rowStartIndex[i]) << "\t";
699        std::cout << std::endl;
700
701
702        std::cout << "lower elements : " << std::endl;
703        for (Index i = 0; i < m.m_data.lowerSize(); i++)
704            std::cout << m.m_data.lower(i) << "\t";
705        std::cout << std::endl;
706        std::cout << "lower profile : " << std::endl;
707        for (Index i = 0; i < m.m_data.lowerProfileSize(); i++)
708            std::cout << m.m_data.lowerProfile(i) << "\t";
709        std::cout << std::endl;
710        std::cout << "lower startIdx : " << std::endl;
711        for (Index i = 0; i < m.m_data.lowerProfileSize(); i++)
712            std::cout << (IsRowMajor ? m.m_rowStartIndex[i] : m.m_colStartIndex[i]) << "\t";
713        std::cout << std::endl;
714        );
715        for (Index rowIdx = 0; rowIdx < m.rows(); rowIdx++) {
716            for (Index colIdx = 0; colIdx < m.cols(); colIdx++) {
717                s << m.coeff(rowIdx, colIdx) << "\t";
718            }
719            s << std::endl;
720        }
721        return s;
722    }
723
724    /** Destructor */
725    inline ~SkylineMatrix() {
726        delete[] m_colStartIndex;
727        delete[] m_rowStartIndex;
728    }
729
730    /** Overloaded for performance */
731    Scalar sum() const;
732};
733
734template<typename Scalar, int _Options>
735class SkylineMatrix<Scalar, _Options>::InnerUpperIterator {
736public:
737
738    InnerUpperIterator(const SkylineMatrix& mat, Index outer)
739    : m_matrix(mat), m_outer(outer),
740    m_id(_Options == RowMajor ? mat.m_colStartIndex[outer] : mat.m_rowStartIndex[outer] + 1),
741    m_start(m_id),
742    m_end(_Options == RowMajor ? mat.m_colStartIndex[outer + 1] : mat.m_rowStartIndex[outer + 1] + 1) {
743    }
744
745    inline InnerUpperIterator & operator++() {
746        m_id++;
747        return *this;
748    }
749
750    inline InnerUpperIterator & operator+=(Index shift) {
751        m_id += shift;
752        return *this;
753    }
754
755    inline Scalar value() const {
756        return m_matrix.m_data.upper(m_id);
757    }
758
759    inline Scalar* valuePtr() {
760        return const_cast<Scalar*> (&(m_matrix.m_data.upper(m_id)));
761    }
762
763    inline Scalar& valueRef() {
764        return const_cast<Scalar&> (m_matrix.m_data.upper(m_id));
765    }
766
767    inline Index index() const {
768        return IsRowMajor ? m_outer - m_matrix.m_data.upperProfile(m_outer) + (m_id - m_start) :
769                m_outer + (m_id - m_start) + 1;
770    }
771
772    inline Index row() const {
773        return IsRowMajor ? index() : m_outer;
774    }
775
776    inline Index col() const {
777        return IsRowMajor ? m_outer : index();
778    }
779
780    inline size_t size() const {
781        return m_matrix.m_data.upperProfile(m_outer);
782    }
783
784    inline operator bool() const {
785        return (m_id < m_end) && (m_id >= m_start);
786    }
787
788protected:
789    const SkylineMatrix& m_matrix;
790    const Index m_outer;
791    Index m_id;
792    const Index m_start;
793    const Index m_end;
794};
795
796template<typename Scalar, int _Options>
797class SkylineMatrix<Scalar, _Options>::InnerLowerIterator {
798public:
799
800    InnerLowerIterator(const SkylineMatrix& mat, Index outer)
801    : m_matrix(mat),
802    m_outer(outer),
803    m_id(_Options == RowMajor ? mat.m_rowStartIndex[outer] : mat.m_colStartIndex[outer] + 1),
804    m_start(m_id),
805    m_end(_Options == RowMajor ? mat.m_rowStartIndex[outer + 1] : mat.m_colStartIndex[outer + 1] + 1) {
806    }
807
808    inline InnerLowerIterator & operator++() {
809        m_id++;
810        return *this;
811    }
812
813    inline InnerLowerIterator & operator+=(Index shift) {
814        m_id += shift;
815        return *this;
816    }
817
818    inline Scalar value() const {
819        return m_matrix.m_data.lower(m_id);
820    }
821
822    inline Scalar* valuePtr() {
823        return const_cast<Scalar*> (&(m_matrix.m_data.lower(m_id)));
824    }
825
826    inline Scalar& valueRef() {
827        return const_cast<Scalar&> (m_matrix.m_data.lower(m_id));
828    }
829
830    inline Index index() const {
831        return IsRowMajor ? m_outer - m_matrix.m_data.lowerProfile(m_outer) + (m_id - m_start) :
832                m_outer + (m_id - m_start) + 1;
833        ;
834    }
835
836    inline Index row() const {
837        return IsRowMajor ? m_outer : index();
838    }
839
840    inline Index col() const {
841        return IsRowMajor ? index() : m_outer;
842    }
843
844    inline size_t size() const {
845        return m_matrix.m_data.lowerProfile(m_outer);
846    }
847
848    inline operator bool() const {
849        return (m_id < m_end) && (m_id >= m_start);
850    }
851
852protected:
853    const SkylineMatrix& m_matrix;
854    const Index m_outer;
855    Index m_id;
856    const Index m_start;
857    const Index m_end;
858};
859
860} // end namespace Eigen
861
862#endif // EIGEN_SkylineMatrix_H
863