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
|
// SPDX-License-Identifier: Apache-2.0
//
// Copyright 2008-2016 Conrad Sanderson (http://conradsanderson.id.au)
// Copyright 2008-2016 National ICT Australia (NICTA)
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
// ------------------------------------------------------------------------
//! \addtogroup band_helper
//! @{
namespace band_helper
{
template<typename eT>
inline
bool
is_band(uword& out_KL, uword& out_KU, const Mat<eT>& A, const uword N_min)
{
arma_extra_debug_sigprint();
// NOTE: assuming that A has a square size
// NOTE: assuming that N_min is >= 4
const uword N = A.n_rows;
if(N < N_min) { return false; }
// first, quickly check bottom-left and top-right corners
const eT eT_zero = eT(0);
const eT* A_col0 = A.memptr();
const eT* A_col1 = A_col0 + N;
if( (A_col0[N-2] != eT_zero) || (A_col0[N-1] != eT_zero) || (A_col1[N-2] != eT_zero) || (A_col1[N-1] != eT_zero) ) { return false; }
const eT* A_colNm2 = A.colptr(N-2);
const eT* A_colNm1 = A_colNm2 + N;
if( (A_colNm2[0] != eT_zero) || (A_colNm2[1] != eT_zero) || (A_colNm1[0] != eT_zero) || (A_colNm1[1] != eT_zero) ) { return false; }
// if we reached this point, go through the entire matrix to work out number of subdiagonals and superdiagonals
const uword n_nonzero_threshold = (N*N)/4; // empirically determined
uword KL = 0; // number of subdiagonals
uword KU = 0; // number of superdiagonals
const eT* A_colptr = A.memptr();
for(uword col=0; col < N; ++col)
{
uword first_nonzero_row = col;
uword last_nonzero_row = col;
for(uword row=0; row < col; ++row)
{
if( A_colptr[row] != eT_zero ) { first_nonzero_row = row; break; }
}
for(uword row=(col+1); row < N; ++row)
{
last_nonzero_row = (A_colptr[row] != eT_zero) ? row : last_nonzero_row;
}
const uword L_count = last_nonzero_row - col;
const uword U_count = col - first_nonzero_row;
if( (L_count > KL) || (U_count > KU) )
{
KL = (std::max)(KL, L_count);
KU = (std::max)(KU, U_count);
const uword n_nonzero = N*(KL+KU+1) - (KL*(KL+1) + KU*(KU+1))/2;
// return as soon as we know that it's not worth analysing the matrix any further
if(n_nonzero > n_nonzero_threshold) { return false; }
}
A_colptr += N;
}
out_KL = KL;
out_KU = KU;
return true;
}
template<typename eT>
inline
bool
is_band_lower(uword& out_KD, const Mat<eT>& A, const uword N_min)
{
arma_extra_debug_sigprint();
// NOTE: assuming that A has a square size
// NOTE: assuming that N_min is >= 4
const uword N = A.n_rows;
if(N < N_min) { return false; }
// first, quickly check bottom-left corner
const eT eT_zero = eT(0);
const eT* A_col0 = A.memptr();
const eT* A_col1 = A_col0 + N;
if( (A_col0[N-2] != eT_zero) || (A_col0[N-1] != eT_zero) || (A_col1[N-2] != eT_zero) || (A_col1[N-1] != eT_zero) ) { return false; }
// if we reached this point, go through the bottom triangle to work out number of subdiagonals
const uword n_nonzero_threshold = ( N*N - (N*(N-1))/2 ) / 4; // empirically determined
uword KL = 0; // number of subdiagonals
const eT* A_colptr = A.memptr();
for(uword col=0; col < N; ++col)
{
uword last_nonzero_row = col;
for(uword row=(col+1); row < N; ++row)
{
last_nonzero_row = (A_colptr[row] != eT_zero) ? row : last_nonzero_row;
}
const uword L_count = last_nonzero_row - col;
if(L_count > KL)
{
KL = L_count;
const uword n_nonzero = N*(KL+1) - (KL*(KL+1))/2;
// return as soon as we know that it's not worth analysing the matrix any further
if(n_nonzero > n_nonzero_threshold) { return false; }
}
A_colptr += N;
}
out_KD = KL;
return true;
}
template<typename eT>
inline
bool
is_band_upper(uword& out_KD, const Mat<eT>& A, const uword N_min)
{
arma_extra_debug_sigprint();
// NOTE: assuming that A has a square size
// NOTE: assuming that N_min is >= 4
const uword N = A.n_rows;
if(N < N_min) { return false; }
// first, quickly check top-right corner
const eT eT_zero = eT(0);
const eT* A_colNm2 = A.colptr(N-2);
const eT* A_colNm1 = A_colNm2 + N;
if( (A_colNm2[0] != eT_zero) || (A_colNm2[1] != eT_zero) || (A_colNm1[0] != eT_zero) || (A_colNm1[1] != eT_zero) ) { return false; }
// if we reached this point, go through the entire matrix to work out number of superdiagonals
const uword n_nonzero_threshold = ( N*N - (N*(N-1))/2 ) / 4; // empirically determined
uword KU = 0; // number of superdiagonals
const eT* A_colptr = A.memptr();
for(uword col=0; col < N; ++col)
{
uword first_nonzero_row = col;
for(uword row=0; row < col; ++row)
{
if( A_colptr[row] != eT_zero ) { first_nonzero_row = row; break; }
}
const uword U_count = col - first_nonzero_row;
if(U_count > KU)
{
KU = U_count;
const uword n_nonzero = N*(KU+1) - (KU*(KU+1))/2;
// return as soon as we know that it's not worth analysing the matrix any further
if(n_nonzero > n_nonzero_threshold) { return false; }
}
A_colptr += N;
}
out_KD = KU;
return true;
}
template<typename eT>
inline
void
compress(Mat<eT>& AB, const Mat<eT>& A, const uword KL, const uword KU, const bool use_offset)
{
arma_extra_debug_sigprint();
// NOTE: assuming that A has a square size
// band matrix storage format
// http://www.netlib.org/lapack/lug/node124.html
// for ?gbsv, matrix AB size: 2*KL+KU+1 x N; band representation of A stored in rows KL+1 to 2*KL+KU+1 (note: fortran counts from 1)
// for ?gbsvx, matrix AB size: KL+KU+1 x N; band representaiton of A stored in rows 1 to KL+KU+1 (note: fortran counts from 1)
//
// the +1 in the above formulas is to take into account the main diagonal
const uword AB_n_rows = (use_offset) ? uword(2*KL + KU + 1) : uword(KL + KU + 1);
const uword N = A.n_rows;
AB.set_size(AB_n_rows, N);
if(A.is_empty()) { AB.zeros(); return; }
if(AB_n_rows == uword(1))
{
eT* AB_mem = AB.memptr();
for(uword i=0; i<N; ++i) { AB_mem[i] = A.at(i,i); }
}
else
{
AB.zeros(); // paranoia
for(uword j=0; j < N; ++j)
{
const uword A_row_start = (j > KU) ? uword(j - KU) : uword(0);
const uword A_row_endp1 = (std::min)(N, j+KL+1);
const uword length = A_row_endp1 - A_row_start;
const uword AB_row_start = (KU > j) ? (KU - j) : uword(0);
const eT* A_colptr = A.colptr(j) + A_row_start;
eT* AB_colptr = AB.colptr(j) + AB_row_start + ( (use_offset) ? KL : uword(0) );
arrayops::copy( AB_colptr, A_colptr, length );
}
}
}
template<typename eT>
inline
void
uncompress(Mat<eT>& A, const Mat<eT>& AB, const uword KL, const uword KU, const bool use_offset)
{
arma_extra_debug_sigprint();
const uword AB_n_rows = AB.n_rows;
const uword N = AB.n_cols;
arma_debug_check( (AB_n_rows != ((use_offset) ? uword(2*KL + KU + 1) : uword(KL + KU + 1))), "band_helper::uncompress(): detected inconsistency" );
A.zeros(N,N); // assuming there is no aliasing between A and AB
if(AB_n_rows == uword(1))
{
const eT* AB_mem = AB.memptr();
for(uword i=0; i<N; ++i) { A.at(i,i) = AB_mem[i]; }
}
else
{
for(uword j=0; j < N; ++j)
{
const uword A_row_start = (j > KU) ? uword(j - KU) : uword(0);
const uword A_row_endp1 = (std::min)(N, j+KL+1);
const uword length = A_row_endp1 - A_row_start;
const uword AB_row_start = (KU > j) ? (KU - j) : uword(0);
const eT* AB_colptr = AB.colptr(j) + AB_row_start + ( (use_offset) ? KL : uword(0) );
eT* A_colptr = A.colptr(j) + A_row_start;
arrayops::copy( A_colptr, AB_colptr, length );
}
}
}
template<typename eT>
inline
void
extract_tridiag(Mat<eT>& out, const Mat<eT>& A)
{
arma_extra_debug_sigprint();
// NOTE: assuming that A has a square size and is at least 2x2
const uword N = A.n_rows;
out.set_size(N, 3); // assuming there is no aliasing between 'out' and 'A'
if(N < 2) { return; }
eT* DL = out.colptr(0);
eT* DD = out.colptr(1);
eT* DU = out.colptr(2);
DD[0] = A[0];
DL[0] = A[1];
const uword Nm1 = N-1;
const uword Nm2 = N-2;
for(uword i=0; i < Nm2; ++i)
{
const uword ip1 = i+1;
const eT* data = &(A.at(i, ip1));
const eT tmp0 = data[0];
const eT tmp1 = data[1];
const eT tmp2 = data[2];
DL[ip1] = tmp2;
DD[ip1] = tmp1;
DU[i ] = tmp0;
}
const eT* data = &(A.at(Nm2, Nm1));
DL[Nm1] = 0;
DU[Nm2] = data[0];
DU[Nm1] = 0;
DD[Nm1] = data[1];
}
} // end of namespace band_helper
//! @}
|