1 // Ceres Solver - A fast non-linear least squares minimizer
2 // Copyright 2010, 2011, 2012 Google Inc. All rights reserved.
3 // http://code.google.com/p/ceres-solver/
4 //
5 // Redistribution and use in source and binary forms, with or without
6 // modification, are permitted provided that the following conditions are met:
7 //
8 // * Redistributions of source code must retain the above copyright notice,
9 //   this list of conditions and the following disclaimer.
10 // * Redistributions in binary form must reproduce the above copyright notice,
11 //   this list of conditions and the following disclaimer in the documentation
12 //   and/or other materials provided with the distribution.
13 // * Neither the name of Google Inc. nor the names of its contributors may be
14 //   used to endorse or promote products derived from this software without
15 //   specific prior written permission.
16 //
17 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
18 // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 // ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
21 // LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
22 // CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
23 // SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
24 // INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
25 // CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
26 // ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
27 // POSSIBILITY OF SUCH DAMAGE.
28 //
29 // Author: keir@google.com (Keir Mierle)
30 //
31 // A jacobian writer that writes to block sparse matrices. The "writer" name is
32 // misleading, since the Write() operation on the block jacobian writer does not
33 // write anything. Instead, the Prepare() method on the BlockEvaluatePreparers
34 // makes a jacobians array which has direct pointers into the block sparse
35 // jacobian. When the cost function is evaluated, the jacobian blocks get placed
36 // directly in their final location.
37 
38 #ifndef CERES_INTERNAL_BLOCK_JACOBIAN_WRITER_H_
39 #define CERES_INTERNAL_BLOCK_JACOBIAN_WRITER_H_
40 
41 #include <vector>
42 #include "ceres/evaluator.h"
43 #include "ceres/internal/port.h"
44 
45 namespace ceres {
46 namespace internal {
47 
48 class BlockEvaluatePreparer;
49 class Program;
50 class SparseMatrix;
51 
52 class BlockJacobianWriter {
53  public:
54   BlockJacobianWriter(const Evaluator::Options& options,
55                       Program* program);
56 
57   // JacobianWriter interface.
58 
59   // Create evaluate prepareres that point directly into the final jacobian.
60   // This makes the final Write() a nop.
61   BlockEvaluatePreparer* CreateEvaluatePreparers(int num_threads);
62 
63   SparseMatrix* CreateJacobian() const;
64 
Write(int,int,double **,SparseMatrix *)65   void Write(int /* residual_id */,
66              int /* residual_offset */,
67              double** /* jacobians */,
68              SparseMatrix* /* jacobian */) {
69     // This is a noop since the blocks were written directly into their final
70     // position by the outside evaluate call, thanks to the jacobians array
71     // prepared by the BlockEvaluatePreparers.
72   }
73 
74  private:
75   Program* program_;
76 
77   // Stores the position of each residual / parameter jacobian.
78   //
79   // The block sparse matrix that this writer writes to is stored as a set of
80   // contiguos dense blocks, one after each other; see BlockSparseMatrix. The
81   // "double* values_" member of the block sparse matrix contains all of these
82   // blocks. Given a pointer to the first element of a block and the size of
83   // that block, it's possible to write to it.
84   //
85   // In the case of a block sparse jacobian, the jacobian writer needs a way to
86   // find the offset in the values_ array of each residual/parameter jacobian
87   // block.
88   //
89   // That is the purpose of jacobian_layout_.
90   //
91   // In particular, jacobian_layout_[i][j] is the offset in the values_ array of
92   // the derivative of residual block i with respect to the parameter block at
93   // active argument position j.
94   //
95   // The active qualifier means that non-active parameters do not count. Care
96   // must be taken when indexing into jacobian_layout_ to account for this.
97   // Consider a single residual example:
98   //
99   //   r(x, y, z)
100   //
101   // with r in R^3, x in R^4, y in R^2, and z in R^5.
102   // Take y as a constant (non-active) parameter.
103   // Take r as residual number 0.
104   //
105   // In this case, the active arguments are only (x, z), so the active argument
106   // position for x is 0, and the active argument position for z is 1. This is
107   // similar to thinking of r as taking only 2 parameters:
108   //
109   //   r(x, z)
110   //
111   // There are only 2 jacobian blocks: dr/dx and dr/dz. jacobian_layout_ would
112   // have the following contents:
113   //
114   //   jacobian_layout_[0] = { 0, 12 }
115   //
116   // which indicates that dr/dx is located at values_[0], and dr/dz is at
117   // values_[12]. See BlockEvaluatePreparer::Prepare()'s comments about 'j'.
118   vector<int*> jacobian_layout_;
119 
120   // The pointers in jacobian_layout_ point directly into this vector.
121   vector<int> jacobian_layout_storage_;
122 };
123 
124 }  // namespace internal
125 }  // namespace ceres
126 
127 #endif  // CERES_INTERNAL_BLOCK_JACOBIAN_WRITER_H_
128