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
|
// 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 spglue_schur
//! @{
template<typename T1, typename T2>
inline
void
spglue_schur::apply(SpMat<typename T1::elem_type>& out, const SpGlue<T1,T2,spglue_schur>& X)
{
arma_extra_debug_sigprint();
typedef typename T1::elem_type eT;
const SpProxy<T1> pa(X.A);
const SpProxy<T2> pb(X.B);
const bool is_alias = pa.is_alias(out) || pb.is_alias(out);
if(is_alias == false)
{
spglue_schur::apply_noalias(out, pa, pb);
}
else
{
SpMat<eT> tmp;
spglue_schur::apply_noalias(tmp, pa, pb);
out.steal_mem(tmp);
}
}
template<typename eT, typename T1, typename T2>
inline
void
spglue_schur::apply_noalias(SpMat<eT>& out, const SpProxy<T1>& pa, const SpProxy<T2>& pb)
{
arma_extra_debug_sigprint();
arma_debug_assert_same_size(pa.get_n_rows(), pa.get_n_cols(), pb.get_n_rows(), pb.get_n_cols(), "element-wise multiplication");
if( (pa.get_n_nonzero() == 0) || (pb.get_n_nonzero() == 0) )
{
out.zeros(pa.get_n_rows(), pa.get_n_cols());
return;
}
const uword max_n_nonzero = (std::min)(pa.get_n_nonzero(), pb.get_n_nonzero());
// Resize memory to upper bound
out.reserve(pa.get_n_rows(), pa.get_n_cols(), max_n_nonzero);
// Now iterate across both matrices.
typename SpProxy<T1>::const_iterator_type x_it = pa.begin();
typename SpProxy<T1>::const_iterator_type x_end = pa.end();
typename SpProxy<T2>::const_iterator_type y_it = pb.begin();
typename SpProxy<T2>::const_iterator_type y_end = pb.end();
uword count = 0;
while( (x_it != x_end) || (y_it != y_end) )
{
const uword x_it_row = x_it.row();
const uword x_it_col = x_it.col();
const uword y_it_row = y_it.row();
const uword y_it_col = y_it.col();
if(x_it == y_it)
{
const eT out_val = (*x_it) * (*y_it);
if(out_val != eT(0))
{
access::rw(out.values[count]) = out_val;
access::rw(out.row_indices[count]) = x_it_row;
access::rw(out.col_ptrs[x_it_col + 1])++;
++count;
}
++x_it;
++y_it;
}
else
{
if((x_it_col < y_it_col) || ((x_it_col == y_it_col) && (x_it_row < y_it_row))) // if y is closer to the end
{
++x_it;
}
else
{
++y_it;
}
}
arma_check( (count > max_n_nonzero), "internal error: spglue_schur::apply_noalias(): count > max_n_nonzero" );
}
const uword out_n_cols = out.n_cols;
uword* col_ptrs = access::rwp(out.col_ptrs);
// Fix column pointers to be cumulative.
for(uword c = 1; c <= out_n_cols; ++c)
{
col_ptrs[c] += col_ptrs[c - 1];
}
if(count < max_n_nonzero)
{
if(count <= (max_n_nonzero/2))
{
out.mem_resize(count);
}
else
{
// quick resize without reallocating memory and copying data
access::rw( out.n_nonzero) = count;
access::rw( out.values[count]) = eT(0);
access::rw(out.row_indices[count]) = uword(0);
}
}
}
template<typename eT>
inline
void
spglue_schur::apply_noalias(SpMat<eT>& out, const SpMat<eT>& A, const SpMat<eT>& B)
{
arma_extra_debug_sigprint();
const SpProxy< SpMat<eT> > pa(A);
const SpProxy< SpMat<eT> > pb(B);
spglue_schur::apply_noalias(out, pa, pb);
}
//
//
//
template<typename T1, typename T2>
inline
void
spglue_schur_misc::dense_schur_sparse(SpMat<typename T1::elem_type>& out, const T1& x, const T2& y)
{
arma_extra_debug_sigprint();
typedef typename T1::elem_type eT;
const Proxy<T1> pa(x);
const SpProxy<T2> pb(y);
arma_debug_assert_same_size(pa.get_n_rows(), pa.get_n_cols(), pb.get_n_rows(), pb.get_n_cols(), "element-wise multiplication");
const uword max_n_nonzero = pb.get_n_nonzero();
// Resize memory to upper bound.
out.reserve(pa.get_n_rows(), pa.get_n_cols(), max_n_nonzero);
uword count = 0;
typename SpProxy<T2>::const_iterator_type it = pb.begin();
typename SpProxy<T2>::const_iterator_type it_end = pb.end();
while(it != it_end)
{
const uword it_row = it.row();
const uword it_col = it.col();
const eT val = (*it) * pa.at(it_row, it_col);
if(val != eT(0))
{
access::rw( out.values[count]) = val;
access::rw( out.row_indices[count]) = it_row;
access::rw(out.col_ptrs[it_col + 1])++;
++count;
}
++it;
arma_check( (count > max_n_nonzero), "internal error: spglue_schur_misc::dense_schur_sparse(): count > max_n_nonzero" );
}
// Fix column pointers.
for(uword c = 1; c <= out.n_cols; ++c)
{
access::rw(out.col_ptrs[c]) += out.col_ptrs[c - 1];
}
if(count < max_n_nonzero)
{
if(count <= (max_n_nonzero/2))
{
out.mem_resize(count);
}
else
{
// quick resize without reallocating memory and copying data
access::rw( out.n_nonzero) = count;
access::rw( out.values[count]) = eT(0);
access::rw(out.row_indices[count]) = uword(0);
}
}
}
//
template<typename T1, typename T2>
inline
void
spglue_schur_mixed::apply(SpMat<typename eT_promoter<T1,T2>::eT>& out, const mtSpGlue<typename eT_promoter<T1,T2>::eT, T1, T2, spglue_schur_mixed>& expr)
{
arma_extra_debug_sigprint();
typedef typename T1::elem_type eT1;
typedef typename T2::elem_type eT2;
typedef typename promote_type<eT1,eT2>::result out_eT;
promote_type<eT1,eT2>::check();
if( (is_same_type<eT1,out_eT>::no) && (is_same_type<eT2,out_eT>::yes) )
{
// upgrade T1
const unwrap_spmat<T1> UA(expr.A);
const unwrap_spmat<T2> UB(expr.B);
const SpMat<eT1>& A = UA.M;
const SpMat<eT2>& B = UB.M;
SpMat<out_eT> AA(arma_layout_indicator(), A);
for(uword i=0; i < A.n_nonzero; ++i) { access::rw(AA.values[i]) = out_eT(A.values[i]); }
const SpMat<out_eT>& BB = reinterpret_cast< const SpMat<out_eT>& >(B);
out = AA % BB;
}
else
if( (is_same_type<eT1,out_eT>::yes) && (is_same_type<eT2,out_eT>::no) )
{
// upgrade T2
const unwrap_spmat<T1> UA(expr.A);
const unwrap_spmat<T2> UB(expr.B);
const SpMat<eT1>& A = UA.M;
const SpMat<eT2>& B = UB.M;
const SpMat<out_eT>& AA = reinterpret_cast< const SpMat<out_eT>& >(A);
SpMat<out_eT> BB(arma_layout_indicator(), B);
for(uword i=0; i < B.n_nonzero; ++i) { access::rw(BB.values[i]) = out_eT(B.values[i]); }
out = AA % BB;
}
else
{
// upgrade T1 and T2
const unwrap_spmat<T1> UA(expr.A);
const unwrap_spmat<T2> UB(expr.B);
const SpMat<eT1>& A = UA.M;
const SpMat<eT2>& B = UB.M;
SpMat<out_eT> AA(arma_layout_indicator(), A);
SpMat<out_eT> BB(arma_layout_indicator(), B);
for(uword i=0; i < A.n_nonzero; ++i) { access::rw(AA.values[i]) = out_eT(A.values[i]); }
for(uword i=0; i < B.n_nonzero; ++i) { access::rw(BB.values[i]) = out_eT(B.values[i]); }
out = AA % BB;
}
}
template<typename T1, typename T2>
inline
void
spglue_schur_mixed::dense_schur_sparse(SpMat< typename promote_type<typename T1::elem_type, typename T2::elem_type >::result>& out, const T1& X, const T2& Y)
{
arma_extra_debug_sigprint();
typedef typename T1::elem_type eT1;
typedef typename T2::elem_type eT2;
typedef typename promote_type<eT1,eT2>::result out_eT;
promote_type<eT1,eT2>::check();
const Proxy<T1> pa(X);
const SpProxy<T2> pb(Y);
arma_debug_assert_same_size(pa.get_n_rows(), pa.get_n_cols(), pb.get_n_rows(), pb.get_n_cols(), "element-wise multiplication");
// count new size
uword new_n_nonzero = 0;
typename SpProxy<T2>::const_iterator_type it = pb.begin();
typename SpProxy<T2>::const_iterator_type it_end = pb.end();
while(it != it_end)
{
if( (out_eT(*it) * out_eT(pa.at(it.row(), it.col()))) != out_eT(0) ) { ++new_n_nonzero; }
++it;
}
// Resize memory accordingly.
out.reserve(pa.get_n_rows(), pa.get_n_cols(), new_n_nonzero);
uword count = 0;
typename SpProxy<T2>::const_iterator_type it2 = pb.begin();
while(it2 != it_end)
{
const uword it2_row = it2.row();
const uword it2_col = it2.col();
const out_eT val = out_eT(*it2) * out_eT(pa.at(it2_row, it2_col));
if(val != out_eT(0))
{
access::rw( out.values[count]) = val;
access::rw( out.row_indices[count]) = it2_row;
access::rw(out.col_ptrs[it2_col + 1])++;
++count;
}
++it2;
}
// Fix column pointers.
for(uword c = 1; c <= out.n_cols; ++c)
{
access::rw(out.col_ptrs[c]) += out.col_ptrs[c - 1];
}
}
//! @}
|