aboutsummaryrefslogtreecommitdiffhomepage
path: root/Eigen/src/Core/AssignEvaluator.h
blob: ab2ebf37e6c72cfee6b09c109cfb2aacc3ba0f14 (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
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
// Copyright (C) 2011-2012 Jitse Niesen <jitse@maths.leeds.ac.uk>
//
// This Source Code Form is subject to the terms of the Mozilla
// Public License v. 2.0. If a copy of the MPL was not distributed
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.

#ifndef EIGEN_ASSIGN_EVALUATOR_H
#define EIGEN_ASSIGN_EVALUATOR_H

namespace Eigen {

// This implementation is based on Assign.h

namespace internal {

/***************************************************************************
* Part 1 : the logic deciding a strategy for traversal and unrolling       *
***************************************************************************/

// copy_using_evaluator_traits is based on assign_traits

template <typename DstEvaluator, typename SrcEvaluator, typename AssignFunc, int MaxPacketSize = -1>
struct copy_using_evaluator_traits
{
  typedef typename DstEvaluator::XprType Dst;
  typedef typename Dst::Scalar DstScalar;

  enum {
    DstFlags = DstEvaluator::Flags,
    SrcFlags = SrcEvaluator::Flags
  };

public:
  enum {
    DstAlignment = DstEvaluator::Alignment,
    SrcAlignment = SrcEvaluator::Alignment,
    DstHasDirectAccess = (DstFlags & DirectAccessBit) == DirectAccessBit,
    JointAlignment = EIGEN_PLAIN_ENUM_MIN(DstAlignment,SrcAlignment)
  };

private:
  enum {
    InnerSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::SizeAtCompileTime)
              : int(DstFlags)&RowMajorBit ? int(Dst::ColsAtCompileTime)
              : int(Dst::RowsAtCompileTime),
    InnerMaxSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::MaxSizeAtCompileTime)
              : int(DstFlags)&RowMajorBit ? int(Dst::MaxColsAtCompileTime)
              : int(Dst::MaxRowsAtCompileTime),
    RestrictedInnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(InnerSize,MaxPacketSize),
    RestrictedLinearSize = EIGEN_SIZE_MIN_PREFER_FIXED(Dst::SizeAtCompileTime,MaxPacketSize),
    OuterStride = int(outer_stride_at_compile_time<Dst>::ret),
    MaxSizeAtCompileTime = Dst::SizeAtCompileTime
  };

  // TODO distinguish between linear traversal and inner-traversals
  typedef typename find_best_packet<DstScalar,RestrictedLinearSize>::type LinearPacketType;
  typedef typename find_best_packet<DstScalar,RestrictedInnerSize>::type InnerPacketType;

  enum {
    LinearPacketSize = unpacket_traits<LinearPacketType>::size,
    InnerPacketSize = unpacket_traits<InnerPacketType>::size
  };

public:
  enum {
    LinearRequiredAlignment = unpacket_traits<LinearPacketType>::alignment,
    InnerRequiredAlignment = unpacket_traits<InnerPacketType>::alignment
  };

private:
  enum {
    DstIsRowMajor = DstFlags&RowMajorBit,
    SrcIsRowMajor = SrcFlags&RowMajorBit,
    StorageOrdersAgree = (int(DstIsRowMajor) == int(SrcIsRowMajor)),
    MightVectorize = bool(StorageOrdersAgree)
                  && (int(DstFlags) & int(SrcFlags) & ActualPacketAccessBit)
                  && bool(functor_traits<AssignFunc>::PacketAccess),
    MayInnerVectorize  = MightVectorize
                       && int(InnerSize)!=Dynamic && int(InnerSize)%int(InnerPacketSize)==0
                       && int(OuterStride)!=Dynamic && int(OuterStride)%int(InnerPacketSize)==0
                       && (EIGEN_UNALIGNED_VECTORIZE  || int(JointAlignment)>=int(InnerRequiredAlignment)),
    MayLinearize = bool(StorageOrdersAgree) && (int(DstFlags) & int(SrcFlags) & LinearAccessBit),
    MayLinearVectorize = bool(MightVectorize) && bool(MayLinearize) && bool(DstHasDirectAccess)
                       && (EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment)) || MaxSizeAtCompileTime == Dynamic),
      /* If the destination isn't aligned, we have to do runtime checks and we don't unroll,
         so it's only good for large enough sizes. */
    MaySliceVectorize  = bool(MightVectorize) && bool(DstHasDirectAccess)
                       && (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=(EIGEN_UNALIGNED_VECTORIZE?InnerPacketSize:(3*InnerPacketSize)))
      /* slice vectorization can be slow, so we only want it if the slices are big, which is
         indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block
         in a fixed-size matrix
         However, with EIGEN_UNALIGNED_VECTORIZE and unrolling, slice vectorization is still worth it */
  };

public:
  enum {
    Traversal =  int(Dst::SizeAtCompileTime) == 0 ? int(AllAtOnceTraversal) // If compile-size is zero, traversing will fail at compile-time.
              : (int(MayLinearVectorize) && (LinearPacketSize>InnerPacketSize)) ? int(LinearVectorizedTraversal)
              : int(MayInnerVectorize)   ? int(InnerVectorizedTraversal)
              : int(MayLinearVectorize)  ? int(LinearVectorizedTraversal)
              : int(MaySliceVectorize)   ? int(SliceVectorizedTraversal)
              : int(MayLinearize)        ? int(LinearTraversal)
                                         : int(DefaultTraversal),
    Vectorized = int(Traversal) == InnerVectorizedTraversal
              || int(Traversal) == LinearVectorizedTraversal
              || int(Traversal) == SliceVectorizedTraversal
  };

  typedef typename conditional<int(Traversal)==LinearVectorizedTraversal, LinearPacketType, InnerPacketType>::type PacketType;

private:
  enum {
    ActualPacketSize    = int(Traversal)==LinearVectorizedTraversal ? LinearPacketSize
                        : Vectorized ? InnerPacketSize
                        : 1,
    UnrollingLimit      = EIGEN_UNROLLING_LIMIT * ActualPacketSize,
    MayUnrollCompletely = int(Dst::SizeAtCompileTime) != Dynamic
                       && int(Dst::SizeAtCompileTime) * (int(DstEvaluator::CoeffReadCost)+int(SrcEvaluator::CoeffReadCost)) <= int(UnrollingLimit),
    MayUnrollInner      = int(InnerSize) != Dynamic
                       && int(InnerSize) * (int(DstEvaluator::CoeffReadCost)+int(SrcEvaluator::CoeffReadCost)) <= int(UnrollingLimit)
  };

public:
  enum {
    Unrolling = (int(Traversal) == int(InnerVectorizedTraversal) || int(Traversal) == int(DefaultTraversal))
                ? (
                    int(MayUnrollCompletely) ? int(CompleteUnrolling)
                  : int(MayUnrollInner)      ? int(InnerUnrolling)
                                             : int(NoUnrolling)
                  )
              : int(Traversal) == int(LinearVectorizedTraversal)
                ? ( bool(MayUnrollCompletely) && ( EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment)))
                          ? int(CompleteUnrolling)
                          : int(NoUnrolling) )
              : int(Traversal) == int(LinearTraversal)
                ? ( bool(MayUnrollCompletely) ? int(CompleteUnrolling)
                                              : int(NoUnrolling) )
#if EIGEN_UNALIGNED_VECTORIZE
              : int(Traversal) == int(SliceVectorizedTraversal)
                ? ( bool(MayUnrollInner) ? int(InnerUnrolling)
                                         : int(NoUnrolling) )
#endif
              : int(NoUnrolling)
  };

#ifdef EIGEN_DEBUG_ASSIGN
  static void debug()
  {
    std::cerr << "DstXpr: " << typeid(typename DstEvaluator::XprType).name() << std::endl;
    std::cerr << "SrcXpr: " << typeid(typename SrcEvaluator::XprType).name() << std::endl;
    std::cerr.setf(std::ios::hex, std::ios::basefield);
    std::cerr << "DstFlags" << " = " << DstFlags << " (" << demangle_flags(DstFlags) << " )" << std::endl;
    std::cerr << "SrcFlags" << " = " << SrcFlags << " (" << demangle_flags(SrcFlags) << " )" << std::endl;
    std::cerr.unsetf(std::ios::hex);
    EIGEN_DEBUG_VAR(DstAlignment)
    EIGEN_DEBUG_VAR(SrcAlignment)
    EIGEN_DEBUG_VAR(LinearRequiredAlignment)
    EIGEN_DEBUG_VAR(InnerRequiredAlignment)
    EIGEN_DEBUG_VAR(JointAlignment)
    EIGEN_DEBUG_VAR(InnerSize)
    EIGEN_DEBUG_VAR(InnerMaxSize)
    EIGEN_DEBUG_VAR(LinearPacketSize)
    EIGEN_DEBUG_VAR(InnerPacketSize)
    EIGEN_DEBUG_VAR(ActualPacketSize)
    EIGEN_DEBUG_VAR(StorageOrdersAgree)
    EIGEN_DEBUG_VAR(MightVectorize)
    EIGEN_DEBUG_VAR(MayLinearize)
    EIGEN_DEBUG_VAR(MayInnerVectorize)
    EIGEN_DEBUG_VAR(MayLinearVectorize)
    EIGEN_DEBUG_VAR(MaySliceVectorize)
    std::cerr << "Traversal" << " = " << Traversal << " (" << demangle_traversal(Traversal) << ")" << std::endl;
    EIGEN_DEBUG_VAR(SrcEvaluator::CoeffReadCost)
    EIGEN_DEBUG_VAR(DstEvaluator::CoeffReadCost)
    EIGEN_DEBUG_VAR(Dst::SizeAtCompileTime)
    EIGEN_DEBUG_VAR(UnrollingLimit)
    EIGEN_DEBUG_VAR(MayUnrollCompletely)
    EIGEN_DEBUG_VAR(MayUnrollInner)
    std::cerr << "Unrolling" << " = " << Unrolling << " (" << demangle_unrolling(Unrolling) << ")" << std::endl;
    std::cerr << std::endl;
  }
#endif
};

/***************************************************************************
* Part 2 : meta-unrollers
***************************************************************************/

/************************
*** Default traversal ***
************************/

template<typename Kernel, int Index, int Stop>
struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling
{
  // FIXME: this is not very clean, perhaps this information should be provided by the kernel?
  typedef typename Kernel::DstEvaluatorType DstEvaluatorType;
  typedef typename DstEvaluatorType::XprType DstXprType;

  enum {
    outer = Index / DstXprType::InnerSizeAtCompileTime,
    inner = Index % DstXprType::InnerSizeAtCompileTime
  };

  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
  {
    kernel.assignCoeffByOuterInner(outer, inner);
    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, Index+1, Stop>::run(kernel);
  }
};

template<typename Kernel, int Stop>
struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, Stop, Stop>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
};

template<typename Kernel, int Index_, int Stop>
struct copy_using_evaluator_DefaultTraversal_InnerUnrolling
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, Index outer)
  {
    kernel.assignCoeffByOuterInner(outer, Index_);
    copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, Index_+1, Stop>::run(kernel, outer);
  }
};

template<typename Kernel, int Stop>
struct copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, Stop, Stop>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&, Index) { }
};

/***********************
*** Linear traversal ***
***********************/

template<typename Kernel, int Index, int Stop>
struct copy_using_evaluator_LinearTraversal_CompleteUnrolling
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel& kernel)
  {
    kernel.assignCoeff(Index);
    copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, Index+1, Stop>::run(kernel);
  }
};

template<typename Kernel, int Stop>
struct copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, Stop, Stop>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
};

/**************************
*** Inner vectorization ***
**************************/

template<typename Kernel, int Index, int Stop>
struct copy_using_evaluator_innervec_CompleteUnrolling
{
  // FIXME: this is not very clean, perhaps this information should be provided by the kernel?
  typedef typename Kernel::DstEvaluatorType DstEvaluatorType;
  typedef typename DstEvaluatorType::XprType DstXprType;
  typedef typename Kernel::PacketType PacketType;

  enum {
    outer = Index / DstXprType::InnerSizeAtCompileTime,
    inner = Index % DstXprType::InnerSizeAtCompileTime,
    SrcAlignment = Kernel::AssignmentTraits::SrcAlignment,
    DstAlignment = Kernel::AssignmentTraits::DstAlignment
  };

  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
  {
    kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, inner);
    enum { NextIndex = Index + unpacket_traits<PacketType>::size };
    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, NextIndex, Stop>::run(kernel);
  }
};

template<typename Kernel, int Stop>
struct copy_using_evaluator_innervec_CompleteUnrolling<Kernel, Stop, Stop>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
};

template<typename Kernel, int Index_, int Stop, int SrcAlignment, int DstAlignment>
struct copy_using_evaluator_innervec_InnerUnrolling
{
  typedef typename Kernel::PacketType PacketType;
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, Index outer)
  {
    kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, Index_);
    enum { NextIndex = Index_ + unpacket_traits<PacketType>::size };
    copy_using_evaluator_innervec_InnerUnrolling<Kernel, NextIndex, Stop, SrcAlignment, DstAlignment>::run(kernel, outer);
  }
};

template<typename Kernel, int Stop, int SrcAlignment, int DstAlignment>
struct copy_using_evaluator_innervec_InnerUnrolling<Kernel, Stop, Stop, SrcAlignment, DstAlignment>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &, Index) { }
};

/***************************************************************************
* Part 3 : implementation of all cases
***************************************************************************/

// dense_assignment_loop is based on assign_impl

template<typename Kernel,
         int Traversal = Kernel::AssignmentTraits::Traversal,
         int Unrolling = Kernel::AssignmentTraits::Unrolling>
struct dense_assignment_loop;

/************************
***** Special Cases *****
************************/

// Zero-sized assignment is a no-op.
template<typename Kernel, int Unrolling>
struct dense_assignment_loop<Kernel, AllAtOnceTraversal, Unrolling>
{
  EIGEN_DEVICE_FUNC static void EIGEN_STRONG_INLINE run(Kernel& /*kernel*/)
  {
    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
    EIGEN_STATIC_ASSERT(int(DstXprType::SizeAtCompileTime) == 0,
      EIGEN_INTERNAL_ERROR_PLEASE_FILE_A_BUG_REPORT)
  }
};

/************************
*** Default traversal ***
************************/

template<typename Kernel>
struct dense_assignment_loop<Kernel, DefaultTraversal, NoUnrolling>
{
  EIGEN_DEVICE_FUNC static void EIGEN_STRONG_INLINE run(Kernel &kernel)
  {
    for(Index outer = 0; outer < kernel.outerSize(); ++outer) {
      for(Index inner = 0; inner < kernel.innerSize(); ++inner) {
        kernel.assignCoeffByOuterInner(outer, inner);
      }
    }
  }
};

template<typename Kernel>
struct dense_assignment_loop<Kernel, DefaultTraversal, CompleteUnrolling>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
  {
    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
  }
};

template<typename Kernel>
struct dense_assignment_loop<Kernel, DefaultTraversal, InnerUnrolling>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
  {
    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;

    const Index outerSize = kernel.outerSize();
    for(Index outer = 0; outer < outerSize; ++outer)
      copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, 0, DstXprType::InnerSizeAtCompileTime>::run(kernel, outer);
  }
};

/***************************
*** Linear vectorization ***
***************************/


// The goal of unaligned_dense_assignment_loop is simply to factorize the handling
// of the non vectorizable beginning and ending parts

template <bool IsAligned = false>
struct unaligned_dense_assignment_loop
{
  // if IsAligned = true, then do nothing
  template <typename Kernel>
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&, Index, Index) {}
};

template <>
struct unaligned_dense_assignment_loop<false>
{
  // MSVC must not inline this functions. If it does, it fails to optimize the
  // packet access path.
  // FIXME check which version exhibits this issue
#if EIGEN_COMP_MSVC
  template <typename Kernel>
  static EIGEN_DONT_INLINE void run(Kernel &kernel,
                                    Index start,
                                    Index end)
#else
  template <typename Kernel>
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel,
                                      Index start,
                                      Index end)
#endif
  {
    for (Index index = start; index < end; ++index)
      kernel.assignCoeff(index);
  }
};

template<typename Kernel>
struct dense_assignment_loop<Kernel, LinearVectorizedTraversal, NoUnrolling>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
  {
    const Index size = kernel.size();
    typedef typename Kernel::Scalar Scalar;
    typedef typename Kernel::PacketType PacketType;
    enum {
      requestedAlignment = Kernel::AssignmentTraits::LinearRequiredAlignment,
      packetSize = unpacket_traits<PacketType>::size,
      dstIsAligned = int(Kernel::AssignmentTraits::DstAlignment)>=int(requestedAlignment),
      dstAlignment = packet_traits<Scalar>::AlignedOnScalar ? int(requestedAlignment)
                                                            : int(Kernel::AssignmentTraits::DstAlignment),
      srcAlignment = Kernel::AssignmentTraits::JointAlignment
    };
    const Index alignedStart = dstIsAligned ? 0 : internal::first_aligned<requestedAlignment>(kernel.dstDataPtr(), size);
    const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize;

    unaligned_dense_assignment_loop<dstIsAligned!=0>::run(kernel, 0, alignedStart);

    for(Index index = alignedStart; index < alignedEnd; index += packetSize)
      kernel.template assignPacket<dstAlignment, srcAlignment, PacketType>(index);

    unaligned_dense_assignment_loop<>::run(kernel, alignedEnd, size);
  }
};

template<typename Kernel>
struct dense_assignment_loop<Kernel, LinearVectorizedTraversal, CompleteUnrolling>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
  {
    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
    typedef typename Kernel::PacketType PacketType;

    enum { size = DstXprType::SizeAtCompileTime,
           packetSize =unpacket_traits<PacketType>::size,
           alignedSize = (int(size)/packetSize)*packetSize };

    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, 0, alignedSize>::run(kernel);
    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, alignedSize, size>::run(kernel);
  }
};

/**************************
*** Inner vectorization ***
**************************/

template<typename Kernel>
struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, NoUnrolling>
{
  typedef typename Kernel::PacketType PacketType;
  enum {
    SrcAlignment = Kernel::AssignmentTraits::SrcAlignment,
    DstAlignment = Kernel::AssignmentTraits::DstAlignment
  };
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
  {
    const Index innerSize = kernel.innerSize();
    const Index outerSize = kernel.outerSize();
    const Index packetSize = unpacket_traits<PacketType>::size;
    for(Index outer = 0; outer < outerSize; ++outer)
      for(Index inner = 0; inner < innerSize; inner+=packetSize)
        kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, inner);
  }
};

template<typename Kernel>
struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, CompleteUnrolling>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
  {
    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
  }
};

template<typename Kernel>
struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, InnerUnrolling>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
  {
    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
    typedef typename Kernel::AssignmentTraits Traits;
    const Index outerSize = kernel.outerSize();
    for(Index outer = 0; outer < outerSize; ++outer)
      copy_using_evaluator_innervec_InnerUnrolling<Kernel, 0, DstXprType::InnerSizeAtCompileTime,
                                                   Traits::SrcAlignment, Traits::DstAlignment>::run(kernel, outer);
  }
};

/***********************
*** Linear traversal ***
***********************/

template<typename Kernel>
struct dense_assignment_loop<Kernel, LinearTraversal, NoUnrolling>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
  {
    const Index size = kernel.size();
    for(Index i = 0; i < size; ++i)
      kernel.assignCoeff(i);
  }
};

template<typename Kernel>
struct dense_assignment_loop<Kernel, LinearTraversal, CompleteUnrolling>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
  {
    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
    copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
  }
};

/**************************
*** Slice vectorization ***
***************************/

template<typename Kernel>
struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, NoUnrolling>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
  {
    typedef typename Kernel::Scalar Scalar;
    typedef typename Kernel::PacketType PacketType;
    enum {
      packetSize = unpacket_traits<PacketType>::size,
      requestedAlignment = int(Kernel::AssignmentTraits::InnerRequiredAlignment),
      alignable = packet_traits<Scalar>::AlignedOnScalar || int(Kernel::AssignmentTraits::DstAlignment)>=sizeof(Scalar),
      dstIsAligned = int(Kernel::AssignmentTraits::DstAlignment)>=int(requestedAlignment),
      dstAlignment = alignable ? int(requestedAlignment)
                               : int(Kernel::AssignmentTraits::DstAlignment)
    };
    const Scalar *dst_ptr = kernel.dstDataPtr();
    if((!bool(dstIsAligned)) && (UIntPtr(dst_ptr) % sizeof(Scalar))>0)
    {
      // the pointer is not aligned-on scalar, so alignment is not possible
      return dense_assignment_loop<Kernel,DefaultTraversal,NoUnrolling>::run(kernel);
    }
    const Index packetAlignedMask = packetSize - 1;
    const Index innerSize = kernel.innerSize();
    const Index outerSize = kernel.outerSize();
    const Index alignedStep = alignable ? (packetSize - kernel.outerStride() % packetSize) & packetAlignedMask : 0;
    Index alignedStart = ((!alignable) || bool(dstIsAligned)) ? 0 : internal::first_aligned<requestedAlignment>(dst_ptr, innerSize);

    for(Index outer = 0; outer < outerSize; ++outer)
    {
      const Index alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask);
      // do the non-vectorizable part of the assignment
      for(Index inner = 0; inner<alignedStart ; ++inner)
        kernel.assignCoeffByOuterInner(outer, inner);

      // do the vectorizable part of the assignment
      for(Index inner = alignedStart; inner<alignedEnd; inner+=packetSize)
        kernel.template assignPacketByOuterInner<dstAlignment, Unaligned, PacketType>(outer, inner);

      // do the non-vectorizable part of the assignment
      for(Index inner = alignedEnd; inner<innerSize ; ++inner)
        kernel.assignCoeffByOuterInner(outer, inner);

      alignedStart = numext::mini((alignedStart+alignedStep)%packetSize, innerSize);
    }
  }
};

#if EIGEN_UNALIGNED_VECTORIZE
template<typename Kernel>
struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, InnerUnrolling>
{
  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
  {
    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
    typedef typename Kernel::PacketType PacketType;

    enum { innerSize = DstXprType::InnerSizeAtCompileTime,
           packetSize =unpacket_traits<PacketType>::size,
           vectorizableSize = (innerSize/packetSize)*packetSize,
           size = DstXprType::SizeAtCompileTime };

    for(Index outer = 0; outer < kernel.outerSize(); ++outer)
    {
      copy_using_evaluator_innervec_InnerUnrolling<Kernel, 0, vectorizableSize, 0, 0>::run(kernel, outer);
      copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, vectorizableSize, innerSize>::run(kernel, outer);
    }
  }
};
#endif


/***************************************************************************
* Part 4 : Generic dense assignment kernel
***************************************************************************/

// This class generalize the assignment of a coefficient (or packet) from one dense evaluator
// to another dense writable evaluator.
// It is parametrized by the two evaluators, and the actual assignment functor.
// This abstraction level permits to keep the evaluation loops as simple and as generic as possible.
// One can customize the assignment using this generic dense_assignment_kernel with different
// functors, or by completely overloading it, by-passing a functor.
template<typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor, int Version = Specialized>
class generic_dense_assignment_kernel
{
protected:
  typedef typename DstEvaluatorTypeT::XprType DstXprType;
  typedef typename SrcEvaluatorTypeT::XprType SrcXprType;
public:

  typedef DstEvaluatorTypeT DstEvaluatorType;
  typedef SrcEvaluatorTypeT SrcEvaluatorType;
  typedef typename DstEvaluatorType::Scalar Scalar;
  typedef copy_using_evaluator_traits<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor> AssignmentTraits;
  typedef typename AssignmentTraits::PacketType PacketType;


  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
  generic_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr)
    : m_dst(dst), m_src(src), m_functor(func), m_dstExpr(dstExpr)
  {
    #ifdef EIGEN_DEBUG_ASSIGN
    AssignmentTraits::debug();
    #endif
  }

  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index size() const EIGEN_NOEXCEPT { return m_dstExpr.size(); }
  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index innerSize() const EIGEN_NOEXCEPT { return m_dstExpr.innerSize(); }
  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index outerSize() const EIGEN_NOEXCEPT { return m_dstExpr.outerSize(); }
  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_dstExpr.rows(); }
  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_dstExpr.cols(); }
  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index outerStride() const EIGEN_NOEXCEPT { return m_dstExpr.outerStride(); }

  EIGEN_DEVICE_FUNC DstEvaluatorType& dstEvaluator() EIGEN_NOEXCEPT { return m_dst; }
  EIGEN_DEVICE_FUNC const SrcEvaluatorType& srcEvaluator() const EIGEN_NOEXCEPT { return m_src; }

  /// Assign src(row,col) to dst(row,col) through the assignment functor.
  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index row, Index col)
  {
    m_functor.assignCoeff(m_dst.coeffRef(row,col), m_src.coeff(row,col));
  }

  /// \sa assignCoeff(Index,Index)
  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index index)
  {
    m_functor.assignCoeff(m_dst.coeffRef(index), m_src.coeff(index));
  }

  /// \sa assignCoeff(Index,Index)
  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeffByOuterInner(Index outer, Index inner)
  {
    Index row = rowIndexByOuterInner(outer, inner);
    Index col = colIndexByOuterInner(outer, inner);
    assignCoeff(row, col);
  }


  template<int StoreMode, int LoadMode, typename PacketType>
  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacket(Index row, Index col)
  {
    m_functor.template assignPacket<StoreMode>(&m_dst.coeffRef(row,col), m_src.template packet<LoadMode,PacketType>(row,col));
  }

  template<int StoreMode, int LoadMode, typename PacketType>
  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacket(Index index)
  {
    m_functor.template assignPacket<StoreMode>(&m_dst.coeffRef(index), m_src.template packet<LoadMode,PacketType>(index));
  }

  template<int StoreMode, int LoadMode, typename PacketType>
  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacketByOuterInner(Index outer, Index inner)
  {
    Index row = rowIndexByOuterInner(outer, inner);
    Index col = colIndexByOuterInner(outer, inner);
    assignPacket<StoreMode,LoadMode,PacketType>(row, col);
  }

  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Index rowIndexByOuterInner(Index outer, Index inner)
  {
    typedef typename DstEvaluatorType::ExpressionTraits Traits;
    return int(Traits::RowsAtCompileTime) == 1 ? 0
      : int(Traits::ColsAtCompileTime) == 1 ? inner
      : int(DstEvaluatorType::Flags)&RowMajorBit ? outer
      : inner;
  }

  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Index colIndexByOuterInner(Index outer, Index inner)
  {
    typedef typename DstEvaluatorType::ExpressionTraits Traits;
    return int(Traits::ColsAtCompileTime) == 1 ? 0
      : int(Traits::RowsAtCompileTime) == 1 ? inner
      : int(DstEvaluatorType::Flags)&RowMajorBit ? inner
      : outer;
  }

  EIGEN_DEVICE_FUNC const Scalar* dstDataPtr() const
  {
    return m_dstExpr.data();
  }

protected:
  DstEvaluatorType& m_dst;
  const SrcEvaluatorType& m_src;
  const Functor &m_functor;
  // TODO find a way to avoid the needs of the original expression
  DstXprType& m_dstExpr;
};

// Special kernel used when computing small products whose operands have dynamic dimensions.  It ensures that the
// PacketSize used is no larger than 4, thereby increasing the chance that vectorized instructions will be used
// when computing the product.

template<typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor>
class restricted_packet_dense_assignment_kernel : public generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, BuiltIn>
{
protected:
  typedef generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, BuiltIn> Base;
 public:
    typedef typename Base::Scalar Scalar;
    typedef typename Base::DstXprType DstXprType;
    typedef copy_using_evaluator_traits<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, 4> AssignmentTraits;
    typedef typename AssignmentTraits::PacketType PacketType;

    EIGEN_DEVICE_FUNC restricted_packet_dense_assignment_kernel(DstEvaluatorTypeT &dst, const SrcEvaluatorTypeT &src, const Functor &func, DstXprType& dstExpr)
    : Base(dst, src, func, dstExpr)
  {
  }
 };

/***************************************************************************
* Part 5 : Entry point for dense rectangular assignment
***************************************************************************/

template<typename DstXprType,typename SrcXprType, typename Functor>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void resize_if_allowed(DstXprType &dst, const SrcXprType& src, const Functor &/*func*/)
{
  EIGEN_ONLY_USED_FOR_DEBUG(dst);
  EIGEN_ONLY_USED_FOR_DEBUG(src);
  eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
}

template<typename DstXprType,typename SrcXprType, typename T1, typename T2>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void resize_if_allowed(DstXprType &dst, const SrcXprType& src, const internal::assign_op<T1,T2> &/*func*/)
{
  Index dstRows = src.rows();
  Index dstCols = src.cols();
  if(((dst.rows()!=dstRows) || (dst.cols()!=dstCols)))
    dst.resize(dstRows, dstCols);
  eigen_assert(dst.rows() == dstRows && dst.cols() == dstCols);
}

template<typename DstXprType, typename SrcXprType, typename Functor>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src, const Functor &func)
{
  typedef evaluator<DstXprType> DstEvaluatorType;
  typedef evaluator<SrcXprType> SrcEvaluatorType;

  SrcEvaluatorType srcEvaluator(src);

  // NOTE To properly handle A = (A*A.transpose())/s with A rectangular,
  // we need to resize the destination after the source evaluator has been created.
  resize_if_allowed(dst, src, func);

  DstEvaluatorType dstEvaluator(dst);

  typedef generic_dense_assignment_kernel<DstEvaluatorType,SrcEvaluatorType,Functor> Kernel;
  Kernel kernel(dstEvaluator, srcEvaluator, func, dst.const_cast_derived());

  dense_assignment_loop<Kernel>::run(kernel);
}

template<typename DstXprType, typename SrcXprType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src)
{
  call_dense_assignment_loop(dst, src, internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>());
}

/***************************************************************************
* Part 6 : Generic assignment
***************************************************************************/

// Based on the respective shapes of the destination and source,
// the class AssignmentKind determine the kind of assignment mechanism.
// AssignmentKind must define a Kind typedef.
template<typename DstShape, typename SrcShape> struct AssignmentKind;

// Assignment kind defined in this file:
struct Dense2Dense {};
struct EigenBase2EigenBase {};

template<typename,typename> struct AssignmentKind { typedef EigenBase2EigenBase Kind; };
template<> struct AssignmentKind<DenseShape,DenseShape> { typedef Dense2Dense Kind; };

// This is the main assignment class
template< typename DstXprType, typename SrcXprType, typename Functor,
          typename Kind = typename AssignmentKind< typename evaluator_traits<DstXprType>::Shape , typename evaluator_traits<SrcXprType>::Shape >::Kind,
          typename EnableIf = void>
struct Assignment;


// The only purpose of this call_assignment() function is to deal with noalias() / "assume-aliasing" and automatic transposition.
// Indeed, I (Gael) think that this concept of "assume-aliasing" was a mistake, and it makes thing quite complicated.
// So this intermediate function removes everything related to "assume-aliasing" such that Assignment
// does not has to bother about these annoying details.

template<typename Dst, typename Src>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment(Dst& dst, const Src& src)
{
  call_assignment(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
}
template<typename Dst, typename Src>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment(const Dst& dst, const Src& src)
{
  call_assignment(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
}

// Deal with "assume-aliasing"
template<typename Dst, typename Src, typename Func>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if< evaluator_assume_aliasing<Src>::value, void*>::type = 0)
{
  typename plain_matrix_type<Src>::type tmp(src);
  call_assignment_no_alias(dst, tmp, func);
}

template<typename Dst, typename Src, typename Func>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if<!evaluator_assume_aliasing<Src>::value, void*>::type = 0)
{
  call_assignment_no_alias(dst, src, func);
}

// by-pass "assume-aliasing"
// When there is no aliasing, we require that 'dst' has been properly resized
template<typename Dst, template <typename> class StorageBase, typename Src, typename Func>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment(NoAlias<Dst,StorageBase>& dst, const Src& src, const Func& func)
{
  call_assignment_no_alias(dst.expression(), src, func);
}


template<typename Dst, typename Src, typename Func>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment_no_alias(Dst& dst, const Src& src, const Func& func)
{
  enum {
    NeedToTranspose = (    (int(Dst::RowsAtCompileTime) == 1 && int(Src::ColsAtCompileTime) == 1)
                        || (int(Dst::ColsAtCompileTime) == 1 && int(Src::RowsAtCompileTime) == 1)
                      ) && int(Dst::SizeAtCompileTime) != 1
  };

  typedef typename internal::conditional<NeedToTranspose, Transpose<Dst>, Dst>::type ActualDstTypeCleaned;
  typedef typename internal::conditional<NeedToTranspose, Transpose<Dst>, Dst&>::type ActualDstType;
  ActualDstType actualDst(dst);

  // TODO check whether this is the right place to perform these checks:
  EIGEN_STATIC_ASSERT_LVALUE(Dst)
  EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(ActualDstTypeCleaned,Src)
  EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename ActualDstTypeCleaned::Scalar,typename Src::Scalar);

  Assignment<ActualDstTypeCleaned,Src,Func>::run(actualDst, src, func);
}

template<typename Dst, typename Src, typename Func>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_restricted_packet_assignment_no_alias(Dst& dst, const Src& src, const Func& func)
{
    typedef evaluator<Dst> DstEvaluatorType;
    typedef evaluator<Src> SrcEvaluatorType;
    typedef restricted_packet_dense_assignment_kernel<DstEvaluatorType,SrcEvaluatorType,Func> Kernel;

    EIGEN_STATIC_ASSERT_LVALUE(Dst)
    EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename Dst::Scalar,typename Src::Scalar);

    SrcEvaluatorType srcEvaluator(src);
    resize_if_allowed(dst, src, func);

    DstEvaluatorType dstEvaluator(dst);
    Kernel kernel(dstEvaluator, srcEvaluator, func, dst.const_cast_derived());

    dense_assignment_loop<Kernel>::run(kernel);
}

template<typename Dst, typename Src>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment_no_alias(Dst& dst, const Src& src)
{
  call_assignment_no_alias(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
}

template<typename Dst, typename Src, typename Func>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src, const Func& func)
{
  // TODO check whether this is the right place to perform these checks:
  EIGEN_STATIC_ASSERT_LVALUE(Dst)
  EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Dst,Src)
  EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename Dst::Scalar,typename Src::Scalar);

  Assignment<Dst,Src,Func>::run(dst, src, func);
}
template<typename Dst, typename Src>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src)
{
  call_assignment_no_alias_no_transpose(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
}

// forward declaration
template<typename Dst, typename Src> void check_for_aliasing(const Dst &dst, const Src &src);

// Generic Dense to Dense assignment
// Note that the last template argument "Weak" is needed to make it possible to perform
// both partial specialization+SFINAE without ambiguous specialization
template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak>
struct Assignment<DstXprType, SrcXprType, Functor, Dense2Dense, Weak>
{
  EIGEN_DEVICE_FUNC
  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
  {
#ifndef EIGEN_NO_DEBUG
    internal::check_for_aliasing(dst, src);
#endif

    call_dense_assignment_loop(dst, src, func);
  }
};

// Generic assignment through evalTo.
// TODO: not sure we have to keep that one, but it helps porting current code to new evaluator mechanism.
// Note that the last template argument "Weak" is needed to make it possible to perform
// both partial specialization+SFINAE without ambiguous specialization
template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak>
struct Assignment<DstXprType, SrcXprType, Functor, EigenBase2EigenBase, Weak>
{
  EIGEN_DEVICE_FUNC
  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
  {
    Index dstRows = src.rows();
    Index dstCols = src.cols();
    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
      dst.resize(dstRows, dstCols);

    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
    src.evalTo(dst);
  }

  // NOTE The following two functions are templated to avoid their instantiation if not needed
  //      This is needed because some expressions supports evalTo only and/or have 'void' as scalar type.
  template<typename SrcScalarType>
  EIGEN_DEVICE_FUNC
  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,SrcScalarType> &/*func*/)
  {
    Index dstRows = src.rows();
    Index dstCols = src.cols();
    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
      dst.resize(dstRows, dstCols);

    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
    src.addTo(dst);
  }

  template<typename SrcScalarType>
  EIGEN_DEVICE_FUNC
  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,SrcScalarType> &/*func*/)
  {
    Index dstRows = src.rows();
    Index dstCols = src.cols();
    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
      dst.resize(dstRows, dstCols);

    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
    src.subTo(dst);
  }
};

} // namespace internal

} // end namespace Eigen

#endif // EIGEN_ASSIGN_EVALUATOR_H