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
|
/* Copyright (c) 2006-2012 Filip Wasilewski <http://en.ig.ma/>
* Copyright (c) 2012-2016 The PyWavelets Developers
* <https://github.com/PyWavelets/pywt>
* See COPYING for license details.
*/
#include "templating.h"
#ifndef TYPE
#error TYPE must be defined here.
#else
#ifndef REAL_TYPE
#error REAL_TYPE must be defined here.
#else
#include "wt.h"
#if defined _MSC_VER
#define restrict __restrict
#elif defined __GNUC__
#define restrict __restrict__
#endif
/* Decomposition of input with lowpass filter */
int CAT(TYPE, _downcoef_axis)(const TYPE * const restrict input, const ArrayInfo input_info,
TYPE * const restrict output, const ArrayInfo output_info,
const DiscreteWavelet * const restrict wavelet, const size_t axis,
const Coefficient coef, const MODE dwt_mode,
const size_t swt_level,
const DiscreteTransformType transform){
size_t i;
size_t num_loops = 1;
TYPE * temp_input = NULL, * temp_output = NULL;
// These are boolean values, but MSVC does not have <stdbool.h>
int make_temp_input, make_temp_output;
if (input_info.ndim != output_info.ndim)
return 1;
if (axis >= input_info.ndim)
return 2;
for (i = 0; i < input_info.ndim; ++i){
if (i == axis){
switch (transform) {
case DWT_TRANSFORM:
if (dwt_buffer_length(input_info.shape[i], wavelet->dec_len,
dwt_mode) != output_info.shape[i])
return 3;
break;
case SWT_TRANSFORM:
if (swt_buffer_length(input_info.shape[i])
!= output_info.shape[i])
return 4;
break;
}
} else {
if (input_info.shape[i] != output_info.shape[i])
return 5;
}
}
make_temp_input = input_info.strides[axis] != sizeof(TYPE);
make_temp_output = output_info.strides[axis] != sizeof(TYPE);
if (make_temp_input)
if ((temp_input = malloc(input_info.shape[axis] * sizeof(TYPE))) == NULL)
goto cleanup;
if (make_temp_output)
if ((temp_output = malloc(output_info.shape[axis] * sizeof(TYPE))) == NULL)
goto cleanup;
for (i = 0; i < output_info.ndim; ++i){
if (i != axis)
num_loops *= output_info.shape[i];
}
for (i = 0; i < num_loops; ++i){
size_t j;
size_t input_offset = 0, output_offset = 0;
const TYPE * input_row;
TYPE * output_row;
// Calculate offset into linear buffer
{
size_t reduced_idx = i;
for (j = 0; j < output_info.ndim; ++j){
size_t j_rev = output_info.ndim - 1 - j;
if (j_rev != axis){
size_t axis_idx = reduced_idx % output_info.shape[j_rev];
reduced_idx /= output_info.shape[j_rev];
input_offset += (axis_idx * input_info.strides[j_rev]);
output_offset += (axis_idx * output_info.strides[j_rev]);
}
}
}
// Copy to temporary input if necessary
if (make_temp_input)
for (j = 0; j < input_info.shape[axis]; ++j)
// Offsets are byte offsets, to need to cast to char and back
temp_input[j] = *(TYPE *)(((char *) input) + input_offset
+ j * input_info.strides[axis]);
// Select temporary or direct output and input
input_row = make_temp_input ? temp_input
: (const TYPE *)((const char *) input + input_offset);
output_row = make_temp_output ? temp_output
: (TYPE *)((char *) output + output_offset);
switch (transform) {
case DWT_TRANSFORM:
// Apply along axis
switch (coef){
case COEF_APPROX:
CAT(TYPE, _dec_a)(input_row, input_info.shape[axis],
wavelet,
output_row, output_info.shape[axis],
dwt_mode);
break;
case COEF_DETAIL:
CAT(TYPE, _dec_d)(input_row, input_info.shape[axis],
wavelet,
output_row, output_info.shape[axis],
dwt_mode);
break;
}
break;
case SWT_TRANSFORM:
// Apply along axis
switch (coef){
case COEF_APPROX:
CAT(TYPE, _swt_a)(input_row, input_info.shape[axis],
wavelet,
output_row, output_info.shape[axis],
swt_level);
break;
case COEF_DETAIL:
CAT(TYPE, _swt_d)(input_row, input_info.shape[axis],
wavelet,
output_row, output_info.shape[axis],
swt_level);
break;
}
break;
}
// Copy from temporary output if necessary
if (make_temp_output)
for (j = 0; j < output_info.shape[axis]; ++j)
// Offsets are byte offsets, to need to cast to char and back
*(TYPE *)((char *) output + output_offset
+ j * output_info.strides[axis]) = output_row[j];
}
free(temp_input);
free(temp_output);
return 0;
cleanup:
free(temp_input);
free(temp_output);
return 6;
}
int CAT(TYPE, _idwt_axis)(const TYPE * const restrict coefs_a, const ArrayInfo * const a_info,
const TYPE * const restrict coefs_d, const ArrayInfo * const d_info,
TYPE * const restrict output, const ArrayInfo output_info,
const DiscreteWavelet * const restrict wavelet,
const size_t axis, const MODE mode){
size_t i;
size_t num_loops = 1;
TYPE * temp_coefs_a = NULL, * temp_coefs_d = NULL, * temp_output = NULL;
// These are boolean values, but MSVC does not have <stdbool.h>
int make_temp_coefs_a, make_temp_coefs_d, make_temp_output;
int have_a = ((coefs_a != NULL) && (a_info != NULL));
int have_d = ((coefs_d != NULL) && (d_info != NULL));
if (!have_a && !have_d)
return 3;
if ((have_a && (a_info->ndim != output_info.ndim)) ||
(have_d && (d_info->ndim != output_info.ndim)))
return 1;
if (axis >= output_info.ndim)
return 1;
for (i = 0; i < output_info.ndim; ++i){
if (i == axis){
size_t input_shape;
if (have_a && have_d &&
(d_info->shape[i] != a_info->shape[i]))
return 1;
input_shape = have_a ? a_info->shape[i] : d_info->shape[i];
/* TODO: reconstruction_buffer_length should take a & d shapes
* - for odd output_len, d_len == (a_len - 1)
*/
if (idwt_buffer_length(input_shape, wavelet->rec_len, mode)
!= output_info.shape[i])
return 1;
} else {
if ((have_a && (a_info->shape[i] != output_info.shape[i])) ||
(have_d && (d_info->shape[i] != output_info.shape[i])))
return 1;
}
}
make_temp_coefs_a = have_a && a_info->strides[axis] != sizeof(TYPE);
make_temp_coefs_d = have_d && d_info->strides[axis] != sizeof(TYPE);
make_temp_output = output_info.strides[axis] != sizeof(TYPE);
if (make_temp_coefs_a)
if ((temp_coefs_a = malloc(a_info->shape[axis] * sizeof(TYPE))) == NULL)
goto cleanup;
if (make_temp_coefs_d)
if ((temp_coefs_d = malloc(d_info->shape[axis] * sizeof(TYPE))) == NULL)
goto cleanup;
if (make_temp_output)
if ((temp_output = malloc(output_info.shape[axis] * sizeof(TYPE))) == NULL)
goto cleanup;
for (i = 0; i < output_info.ndim; ++i){
if (i != axis)
num_loops *= output_info.shape[i];
}
for (i = 0; i < num_loops; ++i){
size_t j;
size_t a_offset = 0, d_offset = 0, output_offset = 0;
TYPE * output_row;
// Calculate offset into linear buffer
{
size_t reduced_idx = i;
for (j = 0; j < output_info.ndim; ++j){
size_t j_rev = output_info.ndim - 1 - j;
if (j_rev != axis){
size_t axis_idx = reduced_idx % output_info.shape[j_rev];
reduced_idx /= output_info.shape[j_rev];
if (have_a)
a_offset += (axis_idx * a_info->strides[j_rev]);
if (have_d)
d_offset += (axis_idx * d_info->strides[j_rev]);
output_offset += (axis_idx * output_info.strides[j_rev]);
}
}
}
// Copy to temporary input if necessary
if (make_temp_coefs_a)
for (j = 0; j < a_info->shape[axis]; ++j)
// Offsets are byte offsets, to need to cast to char and back
temp_coefs_a[j] = *(TYPE *)((char *) coefs_a + a_offset
+ j * a_info->strides[axis]);
if (make_temp_coefs_d)
for (j = 0; j < d_info->shape[axis]; ++j)
// Offsets are byte offsets, to need to cast to char and back
temp_coefs_d[j] = *(TYPE *)((char *) coefs_d + d_offset
+ j * d_info->strides[axis]);
// Select temporary or direct output
output_row = make_temp_output ? temp_output
: (TYPE *)((char *) output + output_offset);
// upsampling_convolution adds to input, so zero
memset(output_row, 0, output_info.shape[axis] * sizeof(TYPE));
if (have_a){
// Pointer arithmetic on NULL is undefined
const TYPE * a_row = make_temp_coefs_a ? temp_coefs_a
: (const TYPE *)((const char *) coefs_a + a_offset);
CAT(TYPE, _upsampling_convolution_valid_sf)
(a_row, a_info->shape[axis],
wavelet->CAT(rec_lo_, REAL_TYPE), wavelet->rec_len,
output_row, output_info.shape[axis],
mode);
}
if (have_d){
// Pointer arithmetic on NULL is undefined
const TYPE * d_row = make_temp_coefs_d ? temp_coefs_d
: (const TYPE *)((const char *) coefs_d + d_offset);
CAT(TYPE, _upsampling_convolution_valid_sf)
(d_row, d_info->shape[axis],
wavelet->CAT(rec_hi_, REAL_TYPE), wavelet->rec_len,
output_row, output_info.shape[axis],
mode);
}
// Copy from temporary output if necessary
if (make_temp_output)
for (j = 0; j < output_info.shape[axis]; ++j)
// Offsets are byte offsets, to need to cast to char and back
*(TYPE *)((char *) output + output_offset
+ j * output_info.strides[axis]) = output_row[j];
}
free(temp_coefs_a);
free(temp_coefs_d);
free(temp_output);
return 0;
cleanup:
free(temp_coefs_a);
free(temp_coefs_d);
free(temp_output);
return 2;
}
int CAT(TYPE, _dec_a)(const TYPE * const restrict input, const size_t input_len,
const DiscreteWavelet * const restrict wavelet,
TYPE * const restrict output, const size_t output_len,
const MODE mode){
/* check output length */
if(output_len != dwt_buffer_length(input_len, wavelet->dec_len, mode)){
return -1;
}
return CAT(TYPE, _downsampling_convolution)(input, input_len,
wavelet->CAT(dec_lo_, REAL_TYPE),
wavelet->dec_len, output,
2, mode);
}
/* Decomposition of input with highpass filter */
int CAT(TYPE, _dec_d)(const TYPE * const restrict input, const size_t input_len,
const DiscreteWavelet * const restrict wavelet,
TYPE * const restrict output, const size_t output_len,
const MODE mode){
/* check output length */
if(output_len != dwt_buffer_length(input_len, wavelet->dec_len, mode))
return -1;
return CAT(TYPE, _downsampling_convolution)(input, input_len,
wavelet->CAT(dec_hi_, REAL_TYPE),
wavelet->dec_len, output,
2, mode);
}
/* Direct reconstruction with lowpass reconstruction filter */
int CAT(TYPE, _rec_a)(const TYPE * const restrict coeffs_a, const size_t coeffs_len,
const DiscreteWavelet * const restrict wavelet,
TYPE * const restrict output, const size_t output_len){
/* check output length */
if(output_len != reconstruction_buffer_length(coeffs_len, wavelet->rec_len))
return -1;
return CAT(TYPE, _upsampling_convolution_full)(coeffs_a, coeffs_len,
wavelet->CAT(rec_lo_, REAL_TYPE),
wavelet->rec_len, output,
output_len);
}
/* Direct reconstruction with highpass reconstruction filter */
int CAT(TYPE, _rec_d)(const TYPE * const restrict coeffs_d, const size_t coeffs_len,
const DiscreteWavelet * const restrict wavelet,
TYPE * const restrict output, const size_t output_len){
/* check for output length */
if(output_len != reconstruction_buffer_length(coeffs_len, wavelet->rec_len))
return -1;
return CAT(TYPE, _upsampling_convolution_full)(coeffs_d, coeffs_len,
wavelet->CAT(rec_hi_, REAL_TYPE),
wavelet->rec_len, output,
output_len);
}
/*
* IDWT reconstruction from approximation and detail coeffs, either of which may
* be NULL.
*
* Requires zero-filled output buffer.
*/
int CAT(TYPE, _idwt)(const TYPE * const restrict coeffs_a, const size_t coeffs_a_len,
const TYPE * const restrict coeffs_d, const size_t coeffs_d_len,
TYPE * const restrict output, const size_t output_len,
const DiscreteWavelet * const restrict wavelet, const MODE mode){
size_t input_len;
if(coeffs_a != NULL && coeffs_d != NULL){
if(coeffs_a_len != coeffs_d_len)
goto error;
input_len = coeffs_a_len;
} else if(coeffs_a != NULL){
input_len = coeffs_a_len;
} else if (coeffs_d != NULL){
input_len = coeffs_d_len;
} else {
goto error;
}
/* check output size */
if(output_len != idwt_buffer_length(input_len, wavelet->rec_len, mode))
goto error;
/*
* Set output to zero (this can be omitted if output array is already
* cleared) memset(output, 0, output_len * sizeof(TYPE));
*/
/* reconstruct approximation coeffs with lowpass reconstruction filter */
if(coeffs_a){
if(CAT(TYPE, _upsampling_convolution_valid_sf)(coeffs_a, input_len,
wavelet->CAT(rec_lo_, REAL_TYPE),
wavelet->rec_len, output,
output_len, mode) < 0){
goto error;
}
}
/*
* Add reconstruction of details coeffs performed with highpass
* reconstruction filter.
*/
if(coeffs_d){
if(CAT(TYPE, _upsampling_convolution_valid_sf)(coeffs_d, input_len,
wavelet->CAT(rec_hi_, REAL_TYPE),
wavelet->rec_len, output,
output_len, mode) < 0){
goto error;
}
}
return 0;
error:
return -1;
}
/* basic SWT step (TODO: optimize) */
int CAT(TYPE, _swt_)(const TYPE * const restrict input, pywt_index_t input_len,
const REAL_TYPE * const restrict filter, pywt_index_t filter_len,
TYPE * const restrict output, size_t output_len,
unsigned int level){
REAL_TYPE * e_filter;
pywt_index_t i, e_filter_len, fstep;
int ret;
if(level < 1)
return -1;
if(level > swt_max_level(input_len))
return -2;
if(output_len != swt_buffer_length(input_len))
return -1;
/* TODO: quick hack, optimize */
if(level > 1){
/* allocate filter first */
e_filter_len = filter_len << (level-1);
if ((e_filter = wtcalloc(e_filter_len, sizeof(TYPE))) == NULL)
goto cleanup;
if(e_filter == NULL)
return -1;
fstep = 1 << (level - 1); // spacing between non-zero filter entries
/* compute upsampled filter values */
for(i = 0; i < filter_len; ++i){
e_filter[i << (level-1)] = filter[i];
}
ret = CAT(TYPE, _downsampling_convolution_periodization)(input, input_len, e_filter,
e_filter_len, output, 1,
fstep);
wtfree(e_filter);
return ret;
} else {
return CAT(TYPE, _downsampling_convolution_periodization)(input, input_len, filter,
filter_len, output, 1,
1);
}
cleanup:
wtfree(e_filter);
return -3;
}
/*
* Approximation at specified level
* input - approximation coeffs from upper level or signal if level == 1
*/
int CAT(TYPE, _swt_a)(const TYPE * const restrict input, pywt_index_t input_len,
const DiscreteWavelet * const restrict wavelet,
TYPE * const restrict output, pywt_index_t output_len,
unsigned int level){
return CAT(TYPE, _swt_)(input, input_len, wavelet->CAT(dec_lo_, REAL_TYPE),
wavelet->dec_len, output, output_len, level);
}
/* Details at specified level
* input - approximation coeffs from upper level or signal if level == 1
*/
int CAT(TYPE, _swt_d)(const TYPE * const restrict input, pywt_index_t input_len,
const DiscreteWavelet * const restrict wavelet,
TYPE * const restrict output, pywt_index_t output_len,
unsigned int level){
return CAT(TYPE, _swt_)(input, input_len, wavelet->CAT(dec_hi_, REAL_TYPE),
wavelet->dec_len, output, output_len, level);
}
#endif /* REAL_TYPE */
#endif /* TYPE */
#undef restrict
|
