1 /*
2  * Copyright (c) 2002-2007, Communications and Remote Sensing Laboratory, Universite catholique de Louvain (UCL), Belgium
3  * Copyright (c) 2002-2007, Professor Benoit Macq
4  * Copyright (c) 2001-2003, David Janssens
5  * Copyright (c) 2002-2003, Yannick Verschueren
6  * Copyright (c) 2003-2007, Francois-Olivier Devaux and Antonin Descampe
7  * Copyright (c) 2005, Herve Drolon, FreeImage Team
8  * Copyright (c) 2007, Jonathan Ballard <dzonatas@dzonux.net>
9  * Copyright (c) 2007, Callum Lerwick <seg@haxxed.com>
10  * Copyright (c) 2008, Jerome Fimes, Communications & Systemes <jerome.fimes@c-s.fr>
11  * All rights reserved.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS `AS IS'
23  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
26  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32  * POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #ifdef __SSE__
36 #include <xmmintrin.h>
37 #endif
38 
39 #include "dwt.h"
40 #include "j2k.h"
41 #include "tcd.h"
42 #include "fix.h"
43 #include "opj_malloc.h"
44 #include "int.h"
45 
46 /** @defgroup DWT DWT - Implementation of a discrete wavelet transform */
47 /*@{*/
48 
49 #define WS(i) v->mem[(i)*2]
50 #define WD(i) v->mem[(1+(i)*2)]
51 
52 /** @name Local data structures */
53 /*@{*/
54 
55 typedef struct dwt_local {
56         OPJ_INT32* mem;
57         OPJ_INT32 dn;
58         OPJ_INT32 sn;
59         OPJ_INT32 cas;
60 } dwt_t;
61 
62 typedef union {
63         OPJ_FLOAT32        f[4];
64 } v4;
65 
66 typedef struct v4dwt_local {
67         v4*        wavelet ;
68         OPJ_INT32                dn ;
69         OPJ_INT32                sn ;
70         OPJ_INT32                cas ;
71 } v4dwt_t ;
72 
73 static const OPJ_FLOAT32 dwt_alpha =  1.586134342f; //  12994
74 static const OPJ_FLOAT32 dwt_beta  =  0.052980118f; //    434
75 static const OPJ_FLOAT32 dwt_gamma = -0.882911075f; //  -7233
76 static const OPJ_FLOAT32 delta = -0.443506852f; //  -3633
77 
78 static const OPJ_FLOAT32 K      = 1.230174105f; //  10078
79 /* FIXME: What is this constant? */
80 static const OPJ_FLOAT32 c13318 = 1.625732422f;
81 
82 /*@}*/
83 
84 /**
85 Virtual function type for wavelet transform in 1-D
86 */
87 typedef void (*DWT1DFN)(dwt_t* v);
88 
89 /** @name Local static functions */
90 /*@{*/
91 
92 /**
93 Forward lazy transform (horizontal)
94 */
95 static void dwt_deinterleave_h(OPJ_INT32 *a, OPJ_INT32 *b, OPJ_INT32 dn, OPJ_INT32 sn, OPJ_INT32 cas);
96 /**
97 Forward lazy transform (vertical)
98 */
99 static void dwt_deinterleave_v(OPJ_INT32 *a, OPJ_INT32 *b, OPJ_INT32 dn, OPJ_INT32 sn, OPJ_INT32 x, OPJ_INT32 cas);
100 /**
101 Inverse lazy transform (horizontal)
102 */
103 static void dwt_interleave_h(dwt_t* h, OPJ_INT32 *a);
104 /**
105 Inverse lazy transform (vertical)
106 */
107 static void dwt_interleave_v(dwt_t* v, OPJ_INT32 *a, OPJ_INT32 x);
108 /**
109 Forward 5-3 wavelet transform in 1-D
110 */
111 static void dwt_encode_1(OPJ_INT32 *a, OPJ_INT32 dn, OPJ_INT32 sn, OPJ_INT32 cas);
112 /**
113 Inverse 5-3 wavelet transform in 1-D
114 */
115 static void dwt_decode_1(dwt_t *v);
116 /**
117 Forward 9-7 wavelet transform in 1-D
118 */
119 static void dwt_encode_1_real(OPJ_INT32 *a, OPJ_INT32 dn, OPJ_INT32 sn, OPJ_INT32 cas);
120 /**
121 Explicit calculation of the Quantization Stepsizes
122 */
123 static void dwt_encode_stepsize(OPJ_INT32 stepsize, OPJ_INT32 numbps, opj_stepsize_t *bandno_stepsize);
124 /**
125 Inverse wavelet transform in 2-D.
126 */
127 static bool dwt_decode_tile(opj_tcd_tilecomp_t* tilec, OPJ_UINT32 i, DWT1DFN fn);
128 
129 static OPJ_UINT32 dwt_max_resolution(opj_tcd_resolution_t* restrict r, OPJ_UINT32 i);
130 
131 static bool dwt_encode_procedure(opj_tcd_tilecomp_t * tilec,void (*p_function)(OPJ_INT32 *, OPJ_INT32,OPJ_INT32,OPJ_INT32) );
132 /*@}*/
133 
134 /*@}*/
135 
136 #define S(i) a[(i)*2]
137 #define D(i) a[(1+(i)*2)]
138 #define S_(i) ((i)<0?S(0):((i)>=sn?S(sn-1):S(i)))
139 #define D_(i) ((i)<0?D(0):((i)>=dn?D(dn-1):D(i)))
140 /* new */
141 #define SS_(i) ((i)<0?S(0):((i)>=dn?S(dn-1):S(i)))
142 #define DD_(i) ((i)<0?D(0):((i)>=sn?D(sn-1):D(i)))
143 
144 /* <summary>                                                              */
145 /* This table contains the norms of the 5-3 wavelets for different bands. */
146 /* </summary>                                                             */
147 static const OPJ_FLOAT64 dwt_norms[4][10] = {
148         {1.000, 1.500, 2.750, 5.375, 10.68, 21.34, 42.67, 85.33, 170.7, 341.3},
149         {1.038, 1.592, 2.919, 5.703, 11.33, 22.64, 45.25, 90.48, 180.9},
150         {1.038, 1.592, 2.919, 5.703, 11.33, 22.64, 45.25, 90.48, 180.9},
151         {.7186, .9218, 1.586, 3.043, 6.019, 12.01, 24.00, 47.97, 95.93}
152 };
153 
154 /* <summary>                                                              */
155 /* This table contains the norms of the 9-7 wavelets for different bands. */
156 /* </summary>                                                             */
157 static const OPJ_FLOAT64 dwt_norms_real[4][10] = {
158         {1.000, 1.965, 4.177, 8.403, 16.90, 33.84, 67.69, 135.3, 270.6, 540.9},
159         {2.022, 3.989, 8.355, 17.04, 34.27, 68.63, 137.3, 274.6, 549.0},
160         {2.022, 3.989, 8.355, 17.04, 34.27, 68.63, 137.3, 274.6, 549.0},
161         {2.080, 3.865, 8.307, 17.18, 34.71, 69.59, 139.3, 278.6, 557.2}
162 };
163 
164 /*
165 ==========================================================
166    local functions
167 ==========================================================
168 */
169 
170 /* <summary>                                         */
171 /* Forward lazy transform (horizontal).  */
172 /* </summary>                            */
dwt_deinterleave_h(OPJ_INT32 * a,OPJ_INT32 * b,OPJ_INT32 dn,OPJ_INT32 sn,OPJ_INT32 cas)173 static void dwt_deinterleave_h(OPJ_INT32 *a, OPJ_INT32 *b, OPJ_INT32 dn, OPJ_INT32 sn, OPJ_INT32 cas) {
174         OPJ_INT32 i;
175 
176         OPJ_INT32 * l_dest = b;
177         OPJ_INT32 * l_src = a+cas;
178     for
179                 (i=0; i<sn; ++i)
180         {
181                 *l_dest++ = *l_src;
182                 l_src += 2;
183         }
184         l_dest = b + sn;
185         l_src = a + 1 - cas;
186     for
187                 (i=0; i<dn; ++i)
188         {
189                 *l_dest++=*l_src;
190                 l_src += 2;
191         }
192 }
193 
194 /* <summary>                             */
195 /* Forward lazy transform (vertical).    */
196 /* </summary>                            */
dwt_deinterleave_v(OPJ_INT32 * a,OPJ_INT32 * b,OPJ_INT32 dn,OPJ_INT32 sn,OPJ_INT32 x,OPJ_INT32 cas)197 static void dwt_deinterleave_v(OPJ_INT32 *a, OPJ_INT32 *b, OPJ_INT32 dn, OPJ_INT32 sn, OPJ_INT32 x, OPJ_INT32 cas) {
198     OPJ_INT32 i = sn;
199         OPJ_INT32 * l_dest = b;
200         OPJ_INT32 * l_src = a+cas;
201 
202     while
203                 (i--)
204         {
205                 *l_dest = *l_src;
206                 l_dest += x;
207                 l_src += 2;
208                 /* b[i*x]=a[2*i+cas]; */
209         }
210         l_dest = b + sn * x;
211         l_src = a + 1 - cas;
212 
213         i = dn;
214     while
215                 (i--)
216         {
217                 *l_dest = *l_src;
218                 l_dest += x;
219                 l_src += 2;
220                 /*b[(sn+i)*x]=a[(2*i+1-cas)];*/
221         }
222 }
223 
224 /* <summary>                             */
225 /* Inverse lazy transform (horizontal).  */
226 /* </summary>                            */
dwt_interleave_h(dwt_t * h,OPJ_INT32 * a)227 static void dwt_interleave_h(dwt_t* h, OPJ_INT32 *a) {
228     OPJ_INT32 *ai = a;
229     OPJ_INT32 *bi = h->mem + h->cas;
230     OPJ_INT32  i        = h->sn;
231     while
232                 ( i-- )
233         {
234                 *bi = *(ai++);
235                 bi += 2;
236     }
237     ai        = a + h->sn;
238     bi        = h->mem + 1 - h->cas;
239     i        = h->dn ;
240     while
241                 ( i-- )
242         {
243                 *bi = *(ai++);
244                 bi += 2;
245     }
246 }
247 
248 /* <summary>                             */
249 /* Inverse lazy transform (vertical).    */
250 /* </summary>                            */
dwt_interleave_v(dwt_t * v,OPJ_INT32 * a,OPJ_INT32 x)251 static void dwt_interleave_v(dwt_t* v, OPJ_INT32 *a, OPJ_INT32 x) {
252     OPJ_INT32 *ai = a;
253     OPJ_INT32 *bi = v->mem + v->cas;
254     OPJ_INT32  i = v->sn;
255     while( i-- ) {
256       *bi = *ai;
257           bi += 2;
258           ai += x;
259     }
260     ai = a + (v->sn * x);
261     bi = v->mem + 1 - v->cas;
262     i = v->dn ;
263     while( i-- ) {
264       *bi = *ai;
265           bi += 2;
266           ai += x;
267     }
268 }
269 
270 
271 /* <summary>                            */
272 /* Forward 5-3 wavelet transform in 1-D. */
273 /* </summary>                           */
dwt_encode_1(OPJ_INT32 * a,OPJ_INT32 dn,OPJ_INT32 sn,OPJ_INT32 cas)274 static void dwt_encode_1(OPJ_INT32 *a, OPJ_INT32 dn, OPJ_INT32 sn, OPJ_INT32 cas) {
275         OPJ_INT32 i;
276 
277         if (!cas) {
278                 if ((dn > 0) || (sn > 1)) {        /* NEW :  CASE ONE ELEMENT */
279                         for (i = 0; i < dn; i++) D(i) -= (S_(i) + S_(i + 1)) >> 1;
280                         for (i = 0; i < sn; i++) S(i) += (D_(i - 1) + D_(i) + 2) >> 2;
281                 }
282         } else {
283                 if (!sn && dn == 1)                    /* NEW :  CASE ONE ELEMENT */
284                         S(0) *= 2;
285                 else {
286                         for (i = 0; i < dn; i++) S(i) -= (DD_(i) + DD_(i - 1)) >> 1;
287                         for (i = 0; i < sn; i++) D(i) += (SS_(i) + SS_(i + 1) + 2) >> 2;
288                 }
289         }
290 }
291 
292 /* <summary>                            */
293 /* Inverse 5-3 wavelet transform in 1-D. */
294 /* </summary>                           */
dwt_decode_1_(OPJ_INT32 * a,OPJ_INT32 dn,OPJ_INT32 sn,OPJ_INT32 cas)295 static void dwt_decode_1_(OPJ_INT32 *a, OPJ_INT32 dn, OPJ_INT32 sn, OPJ_INT32 cas) {
296         OPJ_INT32 i;
297 
298         if (!cas) {
299                 if ((dn > 0) || (sn > 1)) { /* NEW :  CASE ONE ELEMENT */
300                         for (i = 0; i < sn; i++) S(i) -= (D_(i - 1) + D_(i) + 2) >> 2;
301                         for (i = 0; i < dn; i++) D(i) += (S_(i) + S_(i + 1)) >> 1;
302                 }
303         } else {
304                 if (!sn  && dn == 1)          /* NEW :  CASE ONE ELEMENT */
305                         S(0) /= 2;
306                 else {
307                         for (i = 0; i < sn; i++) D(i) -= (SS_(i) + SS_(i + 1) + 2) >> 2;
308                         for (i = 0; i < dn; i++) S(i) += (DD_(i) + DD_(i - 1)) >> 1;
309                 }
310         }
311 }
312 
313 /* <summary>                            */
314 /* Inverse 5-3 wavelet transform in 1-D. */
315 /* </summary>                           */
dwt_decode_1(dwt_t * v)316 static void dwt_decode_1(dwt_t *v) {
317         dwt_decode_1_(v->mem, v->dn, v->sn, v->cas);
318 }
319 
320 /* <summary>                             */
321 /* Forward 9-7 wavelet transform in 1-D. */
322 /* </summary>                            */
dwt_encode_1_real(OPJ_INT32 * a,OPJ_INT32 dn,OPJ_INT32 sn,OPJ_INT32 cas)323 static void dwt_encode_1_real(OPJ_INT32 *a, OPJ_INT32 dn, OPJ_INT32 sn, OPJ_INT32 cas) {
324         OPJ_INT32 i;
325         if (!cas) {
326                 if ((dn > 0) || (sn > 1)) {        /* NEW :  CASE ONE ELEMENT */
327                         for (i = 0; i < dn; i++)
328                                 D(i) -= fix_mul(S_(i) + S_(i + 1), 12993);
329                         for (i = 0; i < sn; i++)
330                                 S(i) -= fix_mul(D_(i - 1) + D_(i), 434);
331                         for (i = 0; i < dn; i++)
332                                 D(i) += fix_mul(S_(i) + S_(i + 1), 7233);
333                         for (i = 0; i < sn; i++)
334                                 S(i) += fix_mul(D_(i - 1) + D_(i), 3633);
335                         for (i = 0; i < dn; i++)
336                                 D(i) = fix_mul(D(i), 5038);        /*5038 */
337                         for (i = 0; i < sn; i++)
338                                 S(i) = fix_mul(S(i), 6659);        /*6660 */
339                 }
340         } else {
341                 if ((sn > 0) || (dn > 1)) {        /* NEW :  CASE ONE ELEMENT */
342                         for (i = 0; i < dn; i++)
343                                 S(i) -= fix_mul(DD_(i) + DD_(i - 1), 12993);
344                         for (i = 0; i < sn; i++)
345                                 D(i) -= fix_mul(SS_(i) + SS_(i + 1), 434);
346                         for (i = 0; i < dn; i++)
347                                 S(i) += fix_mul(DD_(i) + DD_(i - 1), 7233);
348                         for (i = 0; i < sn; i++)
349                                 D(i) += fix_mul(SS_(i) + SS_(i + 1), 3633);
350                         for (i = 0; i < dn; i++)
351                                 S(i) = fix_mul(S(i), 5038);        /*5038 */
352                         for (i = 0; i < sn; i++)
353                                 D(i) = fix_mul(D(i), 6659);        /*6660 */
354                 }
355         }
356 }
357 
dwt_encode_stepsize(OPJ_INT32 stepsize,OPJ_INT32 numbps,opj_stepsize_t * bandno_stepsize)358 static void dwt_encode_stepsize(OPJ_INT32 stepsize, OPJ_INT32 numbps, opj_stepsize_t *bandno_stepsize) {
359         OPJ_INT32 p, n;
360         p = int_floorlog2(stepsize) - 13;
361         n = 11 - int_floorlog2(stepsize);
362         bandno_stepsize->mant = (n < 0 ? stepsize >> -n : stepsize << n) & 0x7ff;
363         bandno_stepsize->expn = numbps - p;
364 }
365 
366 /*
367 ==========================================================
368    DWT interface
369 ==========================================================
370 */
371 
372 /* <summary>                            */
373 /* Forward 5-3 wavelet transform in 2-D. */
374 /* </summary>                           */
dwt_encode_procedure(opj_tcd_tilecomp_t * tilec,void (* p_function)(OPJ_INT32 *,OPJ_INT32,OPJ_INT32,OPJ_INT32))375 INLINE bool dwt_encode_procedure(opj_tcd_tilecomp_t * tilec,void (*p_function)(OPJ_INT32 *, OPJ_INT32,OPJ_INT32,OPJ_INT32) )
376 {
377         OPJ_INT32 i, j, k;
378         OPJ_INT32 *a = 00;
379         OPJ_INT32 *aj = 00;
380         OPJ_INT32 *bj = 00;
381         OPJ_INT32 w, l;
382 
383         OPJ_INT32 rw;                        /* width of the resolution level computed   */
384         OPJ_INT32 rh;                        /* height of the resolution level computed  */
385         OPJ_INT32 l_data_size;
386 
387         opj_tcd_resolution_t * l_cur_res = 0;
388         opj_tcd_resolution_t * l_last_res = 0;
389 
390         w = tilec->x1-tilec->x0;
391         l = tilec->numresolutions-1;
392         a = tilec->data;
393 
394         l_cur_res = tilec->resolutions + l;
395         l_last_res = l_cur_res - 1;
396 
397         l_data_size = dwt_max_resolution( tilec->resolutions,tilec->numresolutions) * sizeof(OPJ_INT32);
398         bj = (OPJ_INT32*)opj_malloc(l_data_size);
399         if
400                 (! bj)
401         {
402                 return false;
403         }
404         i = l;
405 
406         while
407                 (i--)
408         {
409                 OPJ_INT32 rw1;                /* width of the resolution level once lower than computed one                                       */
410                 OPJ_INT32 rh1;                /* height of the resolution level once lower than computed one                                      */
411                 OPJ_INT32 cas_col;        /* 0 = non inversion on horizontal filtering 1 = inversion between low-pass and high-pass filtering */
412                 OPJ_INT32 cas_row;        /* 0 = non inversion on vertical filtering 1 = inversion between low-pass and high-pass filtering   */
413                 OPJ_INT32 dn, sn;
414 
415                 rw  = l_cur_res->x1 - l_cur_res->x0;
416                 rh  = l_cur_res->y1 - l_cur_res->y0;
417                 rw1 = l_last_res->x1 - l_last_res->x0;
418                 rh1 = l_last_res->y1 - l_last_res->y0;
419 
420                 cas_row = l_cur_res->x0 & 1;
421                 cas_col = l_cur_res->y0 & 1;
422 
423                 sn = rh1;
424                 dn = rh - rh1;
425                 for
426                         (j = 0; j < rw; ++j)
427                 {
428                         aj = a + j;
429                         for
430                                 (k = 0; k < rh; ++k)
431                         {
432                                 bj[k] = aj[k*w];
433                         }
434                         (*p_function) (bj, dn, sn, cas_col);
435                         dwt_deinterleave_v(bj, aj, dn, sn, w, cas_col);
436                 }
437                 sn = rw1;
438                 dn = rw - rw1;
439                 for (j = 0; j < rh; j++)
440                 {
441                         aj = a + j * w;
442                         for (k = 0; k < rw; k++)  bj[k] = aj[k];
443                         (*p_function) (bj, dn, sn, cas_row);
444                         dwt_deinterleave_h(bj, aj, dn, sn, cas_row);
445                 }
446                 l_cur_res = l_last_res;
447                 --l_last_res;
448         }
449         opj_free(bj);
450         return true;
451 }
452 /* Forward 5-3 wavelet transform in 2-D. */
453 /* </summary>                           */
dwt_encode(opj_tcd_tilecomp_t * tilec)454 bool dwt_encode(opj_tcd_tilecomp_t * tilec)
455 {
456         return dwt_encode_procedure(tilec,dwt_encode_1);
457 }
458 
459 /* <summary>                            */
460 /* Inverse 5-3 wavelet transform in 2-D. */
461 /* </summary>                           */
dwt_decode(opj_tcd_tilecomp_t * tilec,OPJ_UINT32 numres)462 bool dwt_decode(opj_tcd_tilecomp_t* tilec, OPJ_UINT32 numres) {
463         return dwt_decode_tile(tilec, numres, &dwt_decode_1);
464 }
465 
466 
467 /* <summary>                          */
468 /* Get gain of 5-3 wavelet transform. */
469 /* </summary>                         */
dwt_getgain(OPJ_UINT32 orient)470 OPJ_UINT32 dwt_getgain(OPJ_UINT32 orient) {
471         if (orient == 0)
472                 return 0;
473         if (orient == 1 || orient == 2)
474                 return 1;
475         return 2;
476 }
477 
478 /* <summary>                */
479 /* Get norm of 5-3 wavelet. */
480 /* </summary>               */
dwt_getnorm(OPJ_UINT32 level,OPJ_UINT32 orient)481 OPJ_FLOAT64 dwt_getnorm(OPJ_UINT32 level, OPJ_UINT32 orient) {
482         return dwt_norms[orient][level];
483 }
484 
485 /* <summary>                             */
486 /* Forward 9-7 wavelet transform in 2-D. */
487 /* </summary>                            */
dwt_encode_real(opj_tcd_tilecomp_t * tilec)488 bool dwt_encode_real(opj_tcd_tilecomp_t * tilec)
489 {
490         return dwt_encode_procedure(tilec,dwt_encode_1_real);
491 }
492 
493 
494 
495 /* <summary>                          */
496 /* Get gain of 9-7 wavelet transform. */
497 /* </summary>                         */
dwt_getgain_real(OPJ_UINT32 orient)498 OPJ_UINT32 dwt_getgain_real(OPJ_UINT32 orient) {
499         (void)orient;
500         return 0;
501 }
502 
503 /* <summary>                */
504 /* Get norm of 9-7 wavelet. */
505 /* </summary>               */
dwt_getnorm_real(OPJ_UINT32 level,OPJ_UINT32 orient)506 OPJ_FLOAT64 dwt_getnorm_real(OPJ_UINT32 level, OPJ_UINT32 orient) {
507         return dwt_norms_real[orient][level];
508 }
509 
dwt_calc_explicit_stepsizes(opj_tccp_t * tccp,OPJ_UINT32 prec)510 void dwt_calc_explicit_stepsizes(opj_tccp_t * tccp, OPJ_UINT32 prec) {
511         OPJ_UINT32 numbands, bandno;
512         numbands = 3 * tccp->numresolutions - 2;
513         for (bandno = 0; bandno < numbands; bandno++) {
514                 OPJ_FLOAT64 stepsize;
515                 OPJ_UINT32 resno, level, orient, gain;
516 
517                 resno = (bandno == 0) ? 0 : ((bandno - 1) / 3 + 1);
518                 orient = (bandno == 0) ? 0 : ((bandno - 1) % 3 + 1);
519                 level = tccp->numresolutions - 1 - resno;
520                 gain = (tccp->qmfbid == 0) ? 0 : ((orient == 0) ? 0 : (((orient == 1) || (orient == 2)) ? 1 : 2));
521                 if (tccp->qntsty == J2K_CCP_QNTSTY_NOQNT) {
522                         stepsize = 1.0;
523                 } else {
524                         OPJ_FLOAT64 norm = dwt_norms_real[orient][level];
525                         stepsize = (1 << (gain)) / norm;
526                 }
527                 dwt_encode_stepsize((OPJ_INT32) floor(stepsize * 8192.0), prec + gain, &tccp->stepsizes[bandno]);
528         }
529 }
530 
531 
532 /* <summary>                             */
533 /* Determine maximum computed resolution level for inverse wavelet transform */
534 /* </summary>                            */
dwt_max_resolution(opj_tcd_resolution_t * restrict r,OPJ_UINT32 i)535 static OPJ_UINT32 dwt_max_resolution(opj_tcd_resolution_t* restrict r, OPJ_UINT32 i) {
536         OPJ_UINT32 mr        = 0;
537         OPJ_UINT32 w;
538         while( --i ) {
539                 ++r;
540                 if( mr < ( w = r->x1 - r->x0 ) )
541                         mr = w ;
542                 if( mr < ( w = r->y1 - r->y0 ) )
543                         mr = w ;
544         }
545         return mr ;
546 }
547 
548 
549 /* <summary>                            */
550 /* Inverse wavelet transform in 2-D.     */
551 /* </summary>                           */
dwt_decode_tile(opj_tcd_tilecomp_t * tilec,OPJ_UINT32 numres,DWT1DFN dwt_1D)552 static bool dwt_decode_tile(opj_tcd_tilecomp_t* tilec, OPJ_UINT32 numres, DWT1DFN dwt_1D) {
553         dwt_t h;
554         dwt_t v;
555 
556         opj_tcd_resolution_t* tr = tilec->resolutions;
557 
558         OPJ_UINT32 rw = tr->x1 - tr->x0;        /* width of the resolution level computed */
559         OPJ_UINT32 rh = tr->y1 - tr->y0;        /* height of the resolution level computed */
560 
561         OPJ_UINT32 w = tilec->x1 - tilec->x0;
562 
563         h.mem = (OPJ_INT32*)
564         opj_aligned_malloc(dwt_max_resolution(tr, numres) * sizeof(OPJ_INT32));
565         if
566                 (! h.mem)
567         {
568                 return false;
569         }
570 
571         v.mem = h.mem;
572 
573         while( --numres) {
574                 OPJ_INT32 * restrict tiledp = tilec->data;
575                 OPJ_UINT32 j;
576 
577                 ++tr;
578                 h.sn = rw;
579                 v.sn = rh;
580 
581                 rw = tr->x1 - tr->x0;
582                 rh = tr->y1 - tr->y0;
583 
584                 h.dn = rw - h.sn;
585                 h.cas = tr->x0 % 2;
586 
587                 for(j = 0; j < rh; ++j) {
588                         dwt_interleave_h(&h, &tiledp[j*w]);
589                         (dwt_1D)(&h);
590                         memcpy(&tiledp[j*w], h.mem, rw * sizeof(OPJ_INT32));
591                 }
592 
593                 v.dn = rh - v.sn;
594                 v.cas = tr->y0 % 2;
595 
596                 for(j = 0; j < rw; ++j){
597                         OPJ_UINT32 k;
598                         dwt_interleave_v(&v, &tiledp[j], w);
599                         (dwt_1D)(&v);
600                         for(k = 0; k < rh; ++k) {
601                                 tiledp[k * w + j] = v.mem[k];
602                         }
603                 }
604         }
605         opj_aligned_free(h.mem);
606         return true;
607 }
608 
v4dwt_interleave_h(v4dwt_t * restrict w,OPJ_FLOAT32 * restrict a,OPJ_INT32 x,OPJ_INT32 size)609 static void v4dwt_interleave_h(v4dwt_t* restrict w, OPJ_FLOAT32* restrict a, OPJ_INT32 x, OPJ_INT32 size){
610         OPJ_FLOAT32* restrict bi = (OPJ_FLOAT32*) (w->wavelet + w->cas);
611         OPJ_INT32 count = w->sn;
612         OPJ_INT32 i, k;
613         for(k = 0; k < 2; ++k){
614                 for(i = 0; i < count; ++i){
615                         OPJ_INT32 j = i;
616                         bi[i*8    ] = a[j];
617                         j += x;
618                         if(j >= size) continue;
619                         bi[i*8 + 1] = a[j];
620                         j += x;
621                         if(j >= size) continue;
622                         bi[i*8 + 2] = a[j];
623                         j += x;
624                         if(j >= size) continue;
625                         bi[i*8 + 3] = a[j];
626                 }
627                 bi = (OPJ_FLOAT32*) (w->wavelet + 1 - w->cas);
628                 a += w->sn;
629                 size -= w->sn;
630                 count = w->dn;
631         }
632 }
633 
v4dwt_interleave_v(v4dwt_t * restrict v,OPJ_FLOAT32 * restrict a,OPJ_INT32 x)634 static void v4dwt_interleave_v(v4dwt_t* restrict v , OPJ_FLOAT32* restrict a , OPJ_INT32 x){
635         v4* restrict bi = v->wavelet + v->cas;
636         OPJ_INT32 i;
637         for(i = 0; i < v->sn; ++i){
638                 memcpy(&bi[i*2], &a[i*x], 4 * sizeof(OPJ_FLOAT32));
639         }
640         a += v->sn * x;
641         bi = v->wavelet + 1 - v->cas;
642         for(i = 0; i < v->dn; ++i){
643                 memcpy(&bi[i*2], &a[i*x], 4 * sizeof(OPJ_FLOAT32));
644         }
645 }
646 
647 #ifdef __SSE__
648 
v4dwt_decode_step1_sse(v4 * w,OPJ_INT32 count,const __m128 c)649 static void v4dwt_decode_step1_sse(v4* w, OPJ_INT32 count, const __m128 c){
650         __m128* restrict vw = (__m128*) w;
651         OPJ_INT32 i;
652         for(i = 0; i < count; ++i){
653                 __m128 tmp = vw[i*2];
654                 vw[i*2] = tmp * c;
655         }
656 }
657 
v4dwt_decode_step2_sse(v4 * l,v4 * w,OPJ_INT32 k,OPJ_INT32 m,__m128 c)658 static void v4dwt_decode_step2_sse(v4* l, v4* w, OPJ_INT32 k, OPJ_INT32 m, __m128 c){
659         __m128* restrict vl = (__m128*) l;
660         __m128* restrict vw = (__m128*) w;
661         OPJ_INT32 i;
662         for(i = 0; i < m; ++i){
663                 __m128 tmp1 = vl[ 0];
664                 __m128 tmp2 = vw[-1];
665                 __m128 tmp3 = vw[ 0];
666                 vw[-1] = tmp2 + ((tmp1 + tmp3) * c);
667                 vl = vw;
668                 vw += 2;
669         }
670         if(m >= k){
671                 return;
672         }
673         c += c;
674         c *= vl[0];
675         for(; m < k; ++m){
676                 __m128 tmp = vw[-1];
677                 vw[-1] = tmp + c;
678                 vw += 2;
679         }
680 }
681 
682 #else
683 
v4dwt_decode_step1(v4 * w,OPJ_INT32 count,const OPJ_FLOAT32 c)684 static void v4dwt_decode_step1(v4* w, OPJ_INT32 count, const OPJ_FLOAT32 c){
685         OPJ_FLOAT32* restrict fw = (OPJ_FLOAT32*) w;
686         OPJ_INT32 i;
687         for(i = 0; i < count; ++i){
688                 OPJ_FLOAT32 tmp1 = fw[i*8    ];
689                 OPJ_FLOAT32 tmp2 = fw[i*8 + 1];
690                 OPJ_FLOAT32 tmp3 = fw[i*8 + 2];
691                 OPJ_FLOAT32 tmp4 = fw[i*8 + 3];
692                 fw[i*8    ] = tmp1 * c;
693                 fw[i*8 + 1] = tmp2 * c;
694                 fw[i*8 + 2] = tmp3 * c;
695                 fw[i*8 + 3] = tmp4 * c;
696         }
697 }
698 
v4dwt_decode_step2(v4 * l,v4 * w,OPJ_INT32 k,OPJ_INT32 m,OPJ_FLOAT32 c)699 static void v4dwt_decode_step2(v4* l, v4* w, OPJ_INT32 k, OPJ_INT32 m, OPJ_FLOAT32 c){
700         OPJ_FLOAT32* restrict fl = (OPJ_FLOAT32*) l;
701         OPJ_FLOAT32* restrict fw = (OPJ_FLOAT32*) w;
702         OPJ_INT32 i;
703         for(i = 0; i < m; ++i){
704                 OPJ_FLOAT32 tmp1_1 = fl[0];
705                 OPJ_FLOAT32 tmp1_2 = fl[1];
706                 OPJ_FLOAT32 tmp1_3 = fl[2];
707                 OPJ_FLOAT32 tmp1_4 = fl[3];
708                 OPJ_FLOAT32 tmp2_1 = fw[-4];
709                 OPJ_FLOAT32 tmp2_2 = fw[-3];
710                 OPJ_FLOAT32 tmp2_3 = fw[-2];
711                 OPJ_FLOAT32 tmp2_4 = fw[-1];
712                 OPJ_FLOAT32 tmp3_1 = fw[0];
713                 OPJ_FLOAT32 tmp3_2 = fw[1];
714                 OPJ_FLOAT32 tmp3_3 = fw[2];
715                 OPJ_FLOAT32 tmp3_4 = fw[3];
716                 fw[-4] = tmp2_1 + ((tmp1_1 + tmp3_1) * c);
717                 fw[-3] = tmp2_2 + ((tmp1_2 + tmp3_2) * c);
718                 fw[-2] = tmp2_3 + ((tmp1_3 + tmp3_3) * c);
719                 fw[-1] = tmp2_4 + ((tmp1_4 + tmp3_4) * c);
720                 fl = fw;
721                 fw += 8;
722         }
723         if(m < k){
724                 OPJ_FLOAT32 c1;
725                 OPJ_FLOAT32 c2;
726                 OPJ_FLOAT32 c3;
727                 OPJ_FLOAT32 c4;
728                 c += c;
729                 c1 = fl[0] * c;
730                 c2 = fl[1] * c;
731                 c3 = fl[2] * c;
732                 c4 = fl[3] * c;
733                 for(; m < k; ++m){
734                         OPJ_FLOAT32 tmp1 = fw[-4];
735                         OPJ_FLOAT32 tmp2 = fw[-3];
736                         OPJ_FLOAT32 tmp3 = fw[-2];
737                         OPJ_FLOAT32 tmp4 = fw[-1];
738                         fw[-4] = tmp1 + c1;
739                         fw[-3] = tmp2 + c2;
740                         fw[-2] = tmp3 + c3;
741                         fw[-1] = tmp4 + c4;
742                         fw += 8;
743                 }
744         }
745 }
746 
747 #endif
748 
749 /* <summary>                             */
750 /* Inverse 9-7 wavelet transform in 1-D. */
751 /* </summary>                            */
v4dwt_decode(v4dwt_t * restrict dwt)752 static void v4dwt_decode(v4dwt_t* restrict dwt){
753         OPJ_INT32 a, b;
754         if(dwt->cas == 0) {
755                 if(!((dwt->dn > 0) || (dwt->sn > 1))){
756                         return;
757                 }
758                 a = 0;
759                 b = 1;
760         }else{
761                 if(!((dwt->sn > 0) || (dwt->dn > 1))) {
762                         return;
763                 }
764                 a = 1;
765                 b = 0;
766         }
767 #ifdef __SSE__
768         v4dwt_decode_step1_sse(dwt->wavelet+a, dwt->sn, _mm_set1_ps(K));
769         v4dwt_decode_step1_sse(dwt->wavelet+b, dwt->dn, _mm_set1_ps(c13318));
770         v4dwt_decode_step2_sse(dwt->wavelet+b, dwt->wavelet+a+1, dwt->sn, int_min(dwt->sn, dwt->dn-a), _mm_set1_ps(delta));
771         v4dwt_decode_step2_sse(dwt->wavelet+a, dwt->wavelet+b+1, dwt->dn, int_min(dwt->dn, dwt->sn-b), _mm_set1_ps(dwt_gamma));
772         v4dwt_decode_step2_sse(dwt->wavelet+b, dwt->wavelet+a+1, dwt->sn, int_min(dwt->sn, dwt->dn-a), _mm_set1_ps(dwt_beta));
773         v4dwt_decode_step2_sse(dwt->wavelet+a, dwt->wavelet+b+1, dwt->dn, int_min(dwt->dn, dwt->sn-b), _mm_set1_ps(dwt_alpha));
774 #else
775         v4dwt_decode_step1(dwt->wavelet+a, dwt->sn, K);
776         v4dwt_decode_step1(dwt->wavelet+b, dwt->dn, c13318);
777         v4dwt_decode_step2(dwt->wavelet+b, dwt->wavelet+a+1, dwt->sn, int_min(dwt->sn, dwt->dn-a), delta);
778         v4dwt_decode_step2(dwt->wavelet+a, dwt->wavelet+b+1, dwt->dn, int_min(dwt->dn, dwt->sn-b), dwt_gamma);
779         v4dwt_decode_step2(dwt->wavelet+b, dwt->wavelet+a+1, dwt->sn, int_min(dwt->sn, dwt->dn-a), dwt_beta);
780         v4dwt_decode_step2(dwt->wavelet+a, dwt->wavelet+b+1, dwt->dn, int_min(dwt->dn, dwt->sn-b), dwt_alpha);
781 #endif
782 }
783 
784 /* <summary>                             */
785 /* Inverse 9-7 wavelet transform in 2-D. */
786 /* </summary>                            */
dwt_decode_real(opj_tcd_tilecomp_t * restrict tilec,OPJ_UINT32 numres)787 bool dwt_decode_real(opj_tcd_tilecomp_t* restrict tilec, OPJ_UINT32 numres){
788         v4dwt_t h;
789         v4dwt_t v;
790 
791         opj_tcd_resolution_t* res = tilec->resolutions;
792 
793         OPJ_UINT32 rw = res->x1 - res->x0;        /* width of the resolution level computed */
794         OPJ_UINT32 rh = res->y1 - res->y0;        /* height of the resolution level computed */
795 
796         OPJ_UINT32 w = tilec->x1 - tilec->x0;
797 
798         h.wavelet = (v4*) opj_aligned_malloc((dwt_max_resolution(res, numres)+5) * sizeof(v4));
799         v.wavelet = h.wavelet;
800 
801         while( --numres) {
802                 OPJ_FLOAT32 * restrict aj = (OPJ_FLOAT32*) tilec->data;
803                 OPJ_UINT32 bufsize = (tilec->x1 - tilec->x0) * (tilec->y1 - tilec->y0);
804                 OPJ_INT32 j;
805 
806                 h.sn = rw;
807                 v.sn = rh;
808 
809                 ++res;
810 
811                 rw = res->x1 - res->x0;        /* width of the resolution level computed */
812                 rh = res->y1 - res->y0;        /* height of the resolution level computed */
813 
814                 h.dn = rw - h.sn;
815                 h.cas = res->x0 & 1;
816 
817                 for(j = rh; j > 0; j -= 4){
818                         v4dwt_interleave_h(&h, aj, w, bufsize);
819                         v4dwt_decode(&h);
820                         if(j >= 4){
821                                 OPJ_INT32 k = rw;
822                                 while
823                                         (--k >= 0)
824                                 {
825                                         aj[k    ] = h.wavelet[k].f[0];
826                                         aj[k+w  ] = h.wavelet[k].f[1];
827                                         aj[k+w*2] = h.wavelet[k].f[2];
828                                         aj[k+w*3] = h.wavelet[k].f[3];
829                                 }
830                         }else{
831                                 OPJ_INT32 k = rw;
832                                 while
833                                         (--k >= 0)
834                                 {
835                                         switch(j) {
836                                                 case 3: aj[k+w*2] = h.wavelet[k].f[2];
837                                                 case 2: aj[k+w  ] = h.wavelet[k].f[1];
838                                                 case 1: aj[k    ] = h.wavelet[k].f[0];
839                                         }
840                                 }
841                         }
842                         aj += w*4;
843                         bufsize -= w*4;
844                 }
845 
846                 v.dn = rh - v.sn;
847                 v.cas = res->y0 % 2;
848 
849                 aj = (OPJ_FLOAT32*) tilec->data;
850                 for(j = rw; j > 0; j -= 4){
851                         v4dwt_interleave_v(&v, aj, w);
852                         v4dwt_decode(&v);
853                         if(j >= 4){
854                                 OPJ_UINT32 k;
855                                 for(k = 0; k < rh; ++k){
856                                         memcpy(&aj[k*w], &v.wavelet[k], 4 * sizeof(OPJ_FLOAT32));
857                                 }
858                         }else{
859                                 OPJ_UINT32 k;
860                                 for(k = 0; k < rh; ++k){
861                                         memcpy(&aj[k*w], &v.wavelet[k], j * sizeof(OPJ_FLOAT32));
862                                 }
863                         }
864                         aj += 4;
865                 }
866         }
867 
868         opj_aligned_free(h.wavelet);
869         return true;
870 }
871 
872