• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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: sameeragarwal@google.com (Sameer Agarwal)
30 
31 // This include must come before any #ifndef check on Ceres compile options.
32 #include "ceres/internal/port.h"
33 
34 #ifndef CERES_NO_SUITESPARSE
35 #include "ceres/suitesparse.h"
36 
37 #include <vector>
38 #include "cholmod.h"
39 #include "ceres/compressed_col_sparse_matrix_utils.h"
40 #include "ceres/compressed_row_sparse_matrix.h"
41 #include "ceres/linear_solver.h"
42 #include "ceres/triplet_sparse_matrix.h"
43 
44 namespace ceres {
45 namespace internal {
46 
SuiteSparse()47 SuiteSparse::SuiteSparse() {
48   cholmod_start(&cc_);
49 }
50 
~SuiteSparse()51 SuiteSparse::~SuiteSparse() {
52   cholmod_finish(&cc_);
53 }
54 
CreateSparseMatrix(TripletSparseMatrix * A)55 cholmod_sparse* SuiteSparse::CreateSparseMatrix(TripletSparseMatrix* A) {
56   cholmod_triplet triplet;
57 
58   triplet.nrow = A->num_rows();
59   triplet.ncol = A->num_cols();
60   triplet.nzmax = A->max_num_nonzeros();
61   triplet.nnz = A->num_nonzeros();
62   triplet.i = reinterpret_cast<void*>(A->mutable_rows());
63   triplet.j = reinterpret_cast<void*>(A->mutable_cols());
64   triplet.x = reinterpret_cast<void*>(A->mutable_values());
65   triplet.stype = 0;  // Matrix is not symmetric.
66   triplet.itype = CHOLMOD_INT;
67   triplet.xtype = CHOLMOD_REAL;
68   triplet.dtype = CHOLMOD_DOUBLE;
69 
70   return cholmod_triplet_to_sparse(&triplet, triplet.nnz, &cc_);
71 }
72 
73 
CreateSparseMatrixTranspose(TripletSparseMatrix * A)74 cholmod_sparse* SuiteSparse::CreateSparseMatrixTranspose(
75     TripletSparseMatrix* A) {
76   cholmod_triplet triplet;
77 
78   triplet.ncol = A->num_rows();  // swap row and columns
79   triplet.nrow = A->num_cols();
80   triplet.nzmax = A->max_num_nonzeros();
81   triplet.nnz = A->num_nonzeros();
82 
83   // swap rows and columns
84   triplet.j = reinterpret_cast<void*>(A->mutable_rows());
85   triplet.i = reinterpret_cast<void*>(A->mutable_cols());
86   triplet.x = reinterpret_cast<void*>(A->mutable_values());
87   triplet.stype = 0;  // Matrix is not symmetric.
88   triplet.itype = CHOLMOD_INT;
89   triplet.xtype = CHOLMOD_REAL;
90   triplet.dtype = CHOLMOD_DOUBLE;
91 
92   return cholmod_triplet_to_sparse(&triplet, triplet.nnz, &cc_);
93 }
94 
CreateSparseMatrixTransposeView(CompressedRowSparseMatrix * A)95 cholmod_sparse SuiteSparse::CreateSparseMatrixTransposeView(
96     CompressedRowSparseMatrix* A) {
97   cholmod_sparse m;
98   m.nrow = A->num_cols();
99   m.ncol = A->num_rows();
100   m.nzmax = A->num_nonzeros();
101   m.nz = NULL;
102   m.p = reinterpret_cast<void*>(A->mutable_rows());
103   m.i = reinterpret_cast<void*>(A->mutable_cols());
104   m.x = reinterpret_cast<void*>(A->mutable_values());
105   m.z = NULL;
106   m.stype = 0;  // Matrix is not symmetric.
107   m.itype = CHOLMOD_INT;
108   m.xtype = CHOLMOD_REAL;
109   m.dtype = CHOLMOD_DOUBLE;
110   m.sorted = 1;
111   m.packed = 1;
112 
113   return m;
114 }
115 
CreateDenseVector(const double * x,int in_size,int out_size)116 cholmod_dense* SuiteSparse::CreateDenseVector(const double* x,
117                                               int in_size,
118                                               int out_size) {
119     CHECK_LE(in_size, out_size);
120     cholmod_dense* v = cholmod_zeros(out_size, 1, CHOLMOD_REAL, &cc_);
121     if (x != NULL) {
122       memcpy(v->x, x, in_size*sizeof(*x));
123     }
124     return v;
125 }
126 
AnalyzeCholesky(cholmod_sparse * A,string * message)127 cholmod_factor* SuiteSparse::AnalyzeCholesky(cholmod_sparse* A,
128                                              string* message) {
129   // Cholmod can try multiple re-ordering strategies to find a fill
130   // reducing ordering. Here we just tell it use AMD with automatic
131   // matrix dependence choice of supernodal versus simplicial
132   // factorization.
133   cc_.nmethods = 1;
134   cc_.method[0].ordering = CHOLMOD_AMD;
135   cc_.supernodal = CHOLMOD_AUTO;
136 
137   cholmod_factor* factor = cholmod_analyze(A, &cc_);
138   if (VLOG_IS_ON(2)) {
139     cholmod_print_common(const_cast<char*>("Symbolic Analysis"), &cc_);
140   }
141 
142   if (cc_.status != CHOLMOD_OK) {
143     *message = StringPrintf("cholmod_analyze failed. error code: %d",
144                             cc_.status);
145     return NULL;
146   }
147 
148   return CHECK_NOTNULL(factor);
149 }
150 
BlockAnalyzeCholesky(cholmod_sparse * A,const vector<int> & row_blocks,const vector<int> & col_blocks,string * message)151 cholmod_factor* SuiteSparse::BlockAnalyzeCholesky(
152     cholmod_sparse* A,
153     const vector<int>& row_blocks,
154     const vector<int>& col_blocks,
155     string* message) {
156   vector<int> ordering;
157   if (!BlockAMDOrdering(A, row_blocks, col_blocks, &ordering)) {
158     return NULL;
159   }
160   return AnalyzeCholeskyWithUserOrdering(A, ordering, message);
161 }
162 
AnalyzeCholeskyWithUserOrdering(cholmod_sparse * A,const vector<int> & ordering,string * message)163 cholmod_factor* SuiteSparse::AnalyzeCholeskyWithUserOrdering(
164     cholmod_sparse* A,
165     const vector<int>& ordering,
166     string* message) {
167   CHECK_EQ(ordering.size(), A->nrow);
168 
169   cc_.nmethods = 1;
170   cc_.method[0].ordering = CHOLMOD_GIVEN;
171 
172   cholmod_factor* factor  =
173       cholmod_analyze_p(A, const_cast<int*>(&ordering[0]), NULL, 0, &cc_);
174   if (VLOG_IS_ON(2)) {
175     cholmod_print_common(const_cast<char*>("Symbolic Analysis"), &cc_);
176   }
177   if (cc_.status != CHOLMOD_OK) {
178     *message = StringPrintf("cholmod_analyze failed. error code: %d",
179                             cc_.status);
180     return NULL;
181   }
182 
183   return CHECK_NOTNULL(factor);
184 }
185 
AnalyzeCholeskyWithNaturalOrdering(cholmod_sparse * A,string * message)186 cholmod_factor* SuiteSparse::AnalyzeCholeskyWithNaturalOrdering(
187     cholmod_sparse* A,
188     string* message) {
189   cc_.nmethods = 1;
190   cc_.method[0].ordering = CHOLMOD_NATURAL;
191   cc_.postorder = 0;
192 
193   cholmod_factor* factor  = cholmod_analyze(A, &cc_);
194   if (VLOG_IS_ON(2)) {
195     cholmod_print_common(const_cast<char*>("Symbolic Analysis"), &cc_);
196   }
197   if (cc_.status != CHOLMOD_OK) {
198     *message = StringPrintf("cholmod_analyze failed. error code: %d",
199                             cc_.status);
200     return NULL;
201   }
202 
203   return CHECK_NOTNULL(factor);
204 }
205 
BlockAMDOrdering(const cholmod_sparse * A,const vector<int> & row_blocks,const vector<int> & col_blocks,vector<int> * ordering)206 bool SuiteSparse::BlockAMDOrdering(const cholmod_sparse* A,
207                                    const vector<int>& row_blocks,
208                                    const vector<int>& col_blocks,
209                                    vector<int>* ordering) {
210   const int num_row_blocks = row_blocks.size();
211   const int num_col_blocks = col_blocks.size();
212 
213   // Arrays storing the compressed column structure of the matrix
214   // incoding the block sparsity of A.
215   vector<int> block_cols;
216   vector<int> block_rows;
217 
218   CompressedColumnScalarMatrixToBlockMatrix(reinterpret_cast<const int*>(A->i),
219                                             reinterpret_cast<const int*>(A->p),
220                                             row_blocks,
221                                             col_blocks,
222                                             &block_rows,
223                                             &block_cols);
224 
225   cholmod_sparse_struct block_matrix;
226   block_matrix.nrow = num_row_blocks;
227   block_matrix.ncol = num_col_blocks;
228   block_matrix.nzmax = block_rows.size();
229   block_matrix.p = reinterpret_cast<void*>(&block_cols[0]);
230   block_matrix.i = reinterpret_cast<void*>(&block_rows[0]);
231   block_matrix.x = NULL;
232   block_matrix.stype = A->stype;
233   block_matrix.itype = CHOLMOD_INT;
234   block_matrix.xtype = CHOLMOD_PATTERN;
235   block_matrix.dtype = CHOLMOD_DOUBLE;
236   block_matrix.sorted = 1;
237   block_matrix.packed = 1;
238 
239   vector<int> block_ordering(num_row_blocks);
240   if (!cholmod_amd(&block_matrix, NULL, 0, &block_ordering[0], &cc_)) {
241     return false;
242   }
243 
244   BlockOrderingToScalarOrdering(row_blocks, block_ordering, ordering);
245   return true;
246 }
247 
Cholesky(cholmod_sparse * A,cholmod_factor * L,string * message)248 LinearSolverTerminationType SuiteSparse::Cholesky(cholmod_sparse* A,
249                                                   cholmod_factor* L,
250                                                   string* message) {
251   CHECK_NOTNULL(A);
252   CHECK_NOTNULL(L);
253 
254   // Save the current print level and silence CHOLMOD, otherwise
255   // CHOLMOD is prone to dumping stuff to stderr, which can be
256   // distracting when the error (matrix is indefinite) is not a fatal
257   // failure.
258   const int old_print_level = cc_.print;
259   cc_.print = 0;
260 
261   cc_.quick_return_if_not_posdef = 1;
262   int cholmod_status = cholmod_factorize(A, L, &cc_);
263   cc_.print = old_print_level;
264 
265   // TODO(sameeragarwal): This switch statement is not consistent. It
266   // treats all kinds of CHOLMOD failures as warnings. Some of these
267   // like out of memory are definitely not warnings. The problem is
268   // that the return value Cholesky is two valued, but the state of
269   // the linear solver is really three valued. SUCCESS,
270   // NON_FATAL_FAILURE (e.g., indefinite matrix) and FATAL_FAILURE
271   // (e.g. out of memory).
272   switch (cc_.status) {
273     case CHOLMOD_NOT_INSTALLED:
274       *message = "CHOLMOD failure: Method not installed.";
275       return LINEAR_SOLVER_FATAL_ERROR;
276     case CHOLMOD_OUT_OF_MEMORY:
277       *message = "CHOLMOD failure: Out of memory.";
278       return LINEAR_SOLVER_FATAL_ERROR;
279     case CHOLMOD_TOO_LARGE:
280       *message = "CHOLMOD failure: Integer overflow occured.";
281       return LINEAR_SOLVER_FATAL_ERROR;
282     case CHOLMOD_INVALID:
283       *message = "CHOLMOD failure: Invalid input.";
284       return LINEAR_SOLVER_FATAL_ERROR;
285     case CHOLMOD_NOT_POSDEF:
286       *message = "CHOLMOD warning: Matrix not positive definite.";
287       return LINEAR_SOLVER_FAILURE;
288     case CHOLMOD_DSMALL:
289       *message = "CHOLMOD warning: D for LDL' or diag(L) or "
290                 "LL' has tiny absolute value.";
291       return LINEAR_SOLVER_FAILURE;
292     case CHOLMOD_OK:
293       if (cholmod_status != 0) {
294         return LINEAR_SOLVER_SUCCESS;
295       }
296 
297       *message = "CHOLMOD failure: cholmod_factorize returned false "
298           "but cholmod_common::status is CHOLMOD_OK."
299           "Please report this to ceres-solver@googlegroups.com.";
300       return LINEAR_SOLVER_FATAL_ERROR;
301     default:
302       *message =
303           StringPrintf("Unknown cholmod return code: %d. "
304                        "Please report this to ceres-solver@googlegroups.com.",
305                        cc_.status);
306       return LINEAR_SOLVER_FATAL_ERROR;
307   }
308 
309   return LINEAR_SOLVER_FATAL_ERROR;
310 }
311 
Solve(cholmod_factor * L,cholmod_dense * b,string * message)312 cholmod_dense* SuiteSparse::Solve(cholmod_factor* L,
313                                   cholmod_dense* b,
314                                   string* message) {
315   if (cc_.status != CHOLMOD_OK) {
316     *message = "cholmod_solve failed. CHOLMOD status is not CHOLMOD_OK";
317     return NULL;
318   }
319 
320   return cholmod_solve(CHOLMOD_A, L, b, &cc_);
321 }
322 
ApproximateMinimumDegreeOrdering(cholmod_sparse * matrix,int * ordering)323 bool SuiteSparse::ApproximateMinimumDegreeOrdering(cholmod_sparse* matrix,
324                                                    int* ordering) {
325   return cholmod_amd(matrix, NULL, 0, ordering, &cc_);
326 }
327 
ConstrainedApproximateMinimumDegreeOrdering(cholmod_sparse * matrix,int * constraints,int * ordering)328 bool SuiteSparse::ConstrainedApproximateMinimumDegreeOrdering(
329     cholmod_sparse* matrix,
330     int* constraints,
331     int* ordering) {
332 #ifndef CERES_NO_CAMD
333   return cholmod_camd(matrix, NULL, 0, constraints, ordering, &cc_);
334 #else
335   LOG(FATAL) << "Congratulations you have found a bug in Ceres."
336              << "Ceres Solver was compiled with SuiteSparse "
337              << "version 4.1.0 or less. Calling this function "
338              << "in that case is a bug. Please contact the"
339              << "the Ceres Solver developers.";
340   return false;
341 #endif
342 }
343 
344 }  // namespace internal
345 }  // namespace ceres
346 
347 #endif  // CERES_NO_SUITESPARSE
348