2 * The copyright in this software is being made available under the 2-clauses
3 * BSD License, included below. This software may be subject to other third
4 * party and contributor rights, including patent rights, and no such rights
5 * are granted under this license.
7 * Copyright (c) 2002-2014, Universite catholique de Louvain (UCL), Belgium
8 * Copyright (c) 2002-2014, Professor Benoit Macq
9 * Copyright (c) 2001-2003, David Janssens
10 * Copyright (c) 2002-2003, Yannick Verschueren
11 * Copyright (c) 2003-2007, Francois-Olivier Devaux
12 * Copyright (c) 2003-2014, Antonin Descampe
13 * Copyright (c) 2005, Herve Drolon, FreeImage Team
14 * Copyright (c) 2007, Jonathan Ballard <dzonatas@dzonux.net>
15 * Copyright (c) 2007, Callum Lerwick <seg@haxxed.com>
16 * All rights reserved.
18 * Redistribution and use in source and binary forms, with or without
19 * modification, are permitted provided that the following conditions
21 * 1. Redistributions of source code must retain the above copyright
22 * notice, this list of conditions and the following disclaimer.
23 * 2. Redistributions in binary form must reproduce the above copyright
24 * notice, this list of conditions and the following disclaimer in the
25 * documentation and/or other materials provided with the distribution.
27 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS `AS IS'
28 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
31 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37 * POSSIBILITY OF SUCH DAMAGE.
41 #include <xmmintrin.h>
44 #include "opj_includes.h"
46 /** @defgroup DWT DWT - Implementation of a discrete wavelet transform */
49 #define WS(i) v->mem[(i)*2]
50 #define WD(i) v->mem[(1+(i)*2)]
52 /** @name Local data structures */
55 typedef struct dwt_local {
66 typedef struct v4dwt_local {
73 static const float dwt_alpha = 1.586134342f; /* 12994 */
74 static const float dwt_beta = 0.052980118f; /* 434 */
75 static const float dwt_gamma = -0.882911075f; /* -7233 */
76 static const float dwt_delta = -0.443506852f; /* -3633 */
78 static const float K = 1.230174105f; /* 10078 */
79 /* FIXME: What is this constant? */
80 static const float c13318 = 1.625732422f;
85 Virtual function type for wavelet transform in 1-D
87 typedef void (*DWT1DFN)(dwt_t* v);
89 /** @name Local static functions */
93 Forward lazy transform (horizontal)
95 static void dwt_deinterleave_h(int *a, int *b, int dn, int sn, int cas);
97 Forward lazy transform (vertical)
99 static void dwt_deinterleave_v(int *a, int *b, int dn, int sn, int x, int cas);
101 Inverse lazy transform (horizontal)
103 static void dwt_interleave_h(dwt_t* h, int *a);
105 Inverse lazy transform (vertical)
107 static void dwt_interleave_v(dwt_t* v, int *a, int x);
109 Forward 5-3 wavelet transform in 1-D
111 static void dwt_encode_1(int *a, int dn, int sn, int cas);
113 Inverse 5-3 wavelet transform in 1-D
115 static void dwt_decode_1(dwt_t *v);
117 Forward 9-7 wavelet transform in 1-D
119 static void dwt_encode_1_real(int *a, int dn, int sn, int cas);
121 Explicit calculation of the Quantization Stepsizes
123 static void dwt_encode_stepsize(int stepsize, int numbps, opj_stepsize_t *bandno_stepsize);
125 Inverse wavelet transform in 2-D.
127 static void dwt_decode_tile(opj_tcd_tilecomp_t* tilec, int i, DWT1DFN fn);
133 #define S(i) a[(i)*2]
134 #define D(i) a[(1+(i)*2)]
135 #define S_(i) ((i)<0?S(0):((i)>=sn?S(sn-1):S(i)))
136 #define D_(i) ((i)<0?D(0):((i)>=dn?D(dn-1):D(i)))
138 #define SS_(i) ((i)<0?S(0):((i)>=dn?S(dn-1):S(i)))
139 #define DD_(i) ((i)<0?D(0):((i)>=sn?D(sn-1):D(i)))
142 /* This table contains the norms of the 5-3 wavelets for different bands. */
144 static const double dwt_norms[4][10] = {
145 {1.000, 1.500, 2.750, 5.375, 10.68, 21.34, 42.67, 85.33, 170.7, 341.3},
146 {1.038, 1.592, 2.919, 5.703, 11.33, 22.64, 45.25, 90.48, 180.9},
147 {1.038, 1.592, 2.919, 5.703, 11.33, 22.64, 45.25, 90.48, 180.9},
148 {.7186, .9218, 1.586, 3.043, 6.019, 12.01, 24.00, 47.97, 95.93}
152 /* This table contains the norms of the 9-7 wavelets for different bands. */
154 static const double dwt_norms_real[4][10] = {
155 {1.000, 1.965, 4.177, 8.403, 16.90, 33.84, 67.69, 135.3, 270.6, 540.9},
156 {2.022, 3.989, 8.355, 17.04, 34.27, 68.63, 137.3, 274.6, 549.0},
157 {2.022, 3.989, 8.355, 17.04, 34.27, 68.63, 137.3, 274.6, 549.0},
158 {2.080, 3.865, 8.307, 17.18, 34.71, 69.59, 139.3, 278.6, 557.2}
162 ==========================================================
164 ==========================================================
168 /* Forward lazy transform (horizontal). */
170 static void dwt_deinterleave_h(int *a, int *b, int dn, int sn, int cas) {
172 for (i=0; i<sn; i++) b[i]=a[2*i+cas];
173 for (i=0; i<dn; i++) b[sn+i]=a[(2*i+1-cas)];
177 /* Forward lazy transform (vertical). */
179 static void dwt_deinterleave_v(int *a, int *b, int dn, int sn, int x, int cas) {
181 for (i=0; i<sn; i++) b[i*x]=a[2*i+cas];
182 for (i=0; i<dn; i++) b[(sn+i)*x]=a[(2*i+1-cas)];
186 /* Inverse lazy transform (horizontal). */
188 static void dwt_interleave_h(dwt_t* h, int *a) {
190 int *bi = h->mem + h->cas;
197 bi = h->mem + 1 - h->cas;
206 /* Inverse lazy transform (vertical). */
208 static void dwt_interleave_v(dwt_t* v, int *a, int x) {
210 int *bi = v->mem + v->cas;
217 ai = a + (v->sn * x);
218 bi = v->mem + 1 - v->cas;
229 /* Forward 5-3 wavelet transform in 1-D. */
231 static void dwt_encode_1(int *a, int dn, int sn, int cas) {
235 if ((dn > 0) || (sn > 1)) { /* NEW : CASE ONE ELEMENT */
236 for (i = 0; i < dn; i++) D(i) -= (S_(i) + S_(i + 1)) >> 1;
237 for (i = 0; i < sn; i++) S(i) += (D_(i - 1) + D_(i) + 2) >> 2;
240 if (!sn && dn == 1) /* NEW : CASE ONE ELEMENT */
243 for (i = 0; i < dn; i++) S(i) -= (DD_(i) + DD_(i - 1)) >> 1;
244 for (i = 0; i < sn; i++) D(i) += (SS_(i) + SS_(i + 1) + 2) >> 2;
250 /* Inverse 5-3 wavelet transform in 1-D. */
252 static void dwt_decode_1_(int *a, int dn, int sn, int cas) {
256 if ((dn > 0) || (sn > 1)) { /* NEW : CASE ONE ELEMENT */
257 for (i = 0; i < sn; i++) S(i) -= (D_(i - 1) + D_(i) + 2) >> 2;
258 for (i = 0; i < dn; i++) D(i) += (S_(i) + S_(i + 1)) >> 1;
261 if (!sn && dn == 1) /* NEW : CASE ONE ELEMENT */
264 for (i = 0; i < sn; i++) D(i) -= (SS_(i) + SS_(i + 1) + 2) >> 2;
265 for (i = 0; i < dn; i++) S(i) += (DD_(i) + DD_(i - 1)) >> 1;
271 /* Inverse 5-3 wavelet transform in 1-D. */
273 static void dwt_decode_1(dwt_t *v) {
274 dwt_decode_1_(v->mem, v->dn, v->sn, v->cas);
278 /* Forward 9-7 wavelet transform in 1-D. */
280 static void dwt_encode_1_real(int *a, int dn, int sn, int cas) {
283 if ((dn > 0) || (sn > 1)) { /* NEW : CASE ONE ELEMENT */
284 for (i = 0; i < dn; i++)
285 D(i) -= fix_mul(S_(i) + S_(i + 1), 12993);
286 for (i = 0; i < sn; i++)
287 S(i) -= fix_mul(D_(i - 1) + D_(i), 434);
288 for (i = 0; i < dn; i++)
289 D(i) += fix_mul(S_(i) + S_(i + 1), 7233);
290 for (i = 0; i < sn; i++)
291 S(i) += fix_mul(D_(i - 1) + D_(i), 3633);
292 for (i = 0; i < dn; i++)
293 D(i) = fix_mul(D(i), 5038); /*5038 */
294 for (i = 0; i < sn; i++)
295 S(i) = fix_mul(S(i), 6659); /*6660 */
298 if ((sn > 0) || (dn > 1)) { /* NEW : CASE ONE ELEMENT */
299 for (i = 0; i < dn; i++)
300 S(i) -= fix_mul(DD_(i) + DD_(i - 1), 12993);
301 for (i = 0; i < sn; i++)
302 D(i) -= fix_mul(SS_(i) + SS_(i + 1), 434);
303 for (i = 0; i < dn; i++)
304 S(i) += fix_mul(DD_(i) + DD_(i - 1), 7233);
305 for (i = 0; i < sn; i++)
306 D(i) += fix_mul(SS_(i) + SS_(i + 1), 3633);
307 for (i = 0; i < dn; i++)
308 S(i) = fix_mul(S(i), 5038); /*5038 */
309 for (i = 0; i < sn; i++)
310 D(i) = fix_mul(D(i), 6659); /*6660 */
315 static void dwt_encode_stepsize(int stepsize, int numbps, opj_stepsize_t *bandno_stepsize) {
317 p = int_floorlog2(stepsize) - 13;
318 n = 11 - int_floorlog2(stepsize);
319 bandno_stepsize->mant = (n < 0 ? stepsize >> -n : stepsize << n) & 0x7ff;
320 bandno_stepsize->expn = numbps - p;
324 ==========================================================
326 ==========================================================
330 /* Forward 5-3 wavelet transform in 2-D. */
332 void dwt_encode(opj_tcd_tilecomp_t * tilec) {
339 w = tilec->x1-tilec->x0;
340 l = tilec->numresolutions-1;
343 for (i = 0; i < l; i++) {
344 int rw; /* width of the resolution level computed */
345 int rh; /* height of the resolution level computed */
346 int rw1; /* width of the resolution level once lower than computed one */
347 int rh1; /* height of the resolution level once lower than computed one */
348 int cas_col; /* 0 = non inversion on horizontal filtering 1 = inversion between low-pass and high-pass filtering */
349 int cas_row; /* 0 = non inversion on vertical filtering 1 = inversion between low-pass and high-pass filtering */
352 rw = tilec->resolutions[l - i].x1 - tilec->resolutions[l - i].x0;
353 rh = tilec->resolutions[l - i].y1 - tilec->resolutions[l - i].y0;
354 rw1= tilec->resolutions[l - i - 1].x1 - tilec->resolutions[l - i - 1].x0;
355 rh1= tilec->resolutions[l - i - 1].y1 - tilec->resolutions[l - i - 1].y0;
357 cas_row = tilec->resolutions[l - i].x0 % 2;
358 cas_col = tilec->resolutions[l - i].y0 % 2;
362 bj = (int*)opj_malloc(rh * sizeof(int));
363 for (j = 0; j < rw; j++) {
365 for (k = 0; k < rh; k++) bj[k] = aj[k*w];
366 dwt_encode_1(bj, dn, sn, cas_col);
367 dwt_deinterleave_v(bj, aj, dn, sn, w, cas_col);
373 bj = (int*)opj_malloc(rw * sizeof(int));
374 for (j = 0; j < rh; j++) {
376 for (k = 0; k < rw; k++) bj[k] = aj[k];
377 dwt_encode_1(bj, dn, sn, cas_row);
378 dwt_deinterleave_h(bj, aj, dn, sn, cas_row);
386 /* Inverse 5-3 wavelet transform in 2-D. */
388 void dwt_decode(opj_tcd_tilecomp_t* tilec, int numres) {
389 dwt_decode_tile(tilec, numres, &dwt_decode_1);
394 /* Get gain of 5-3 wavelet transform. */
396 int dwt_getgain(int orient) {
399 if (orient == 1 || orient == 2)
405 /* Get norm of 5-3 wavelet. */
407 double dwt_getnorm(int level, int orient) {
408 return dwt_norms[orient][level];
412 /* Forward 9-7 wavelet transform in 2-D. */
415 void dwt_encode_real(opj_tcd_tilecomp_t * tilec) {
422 w = tilec->x1-tilec->x0;
423 l = tilec->numresolutions-1;
426 for (i = 0; i < l; i++) {
427 int rw; /* width of the resolution level computed */
428 int rh; /* height of the resolution level computed */
429 int rw1; /* width of the resolution level once lower than computed one */
430 int rh1; /* height of the resolution level once lower than computed one */
431 int cas_col; /* 0 = non inversion on horizontal filtering 1 = inversion between low-pass and high-pass filtering */
432 int cas_row; /* 0 = non inversion on vertical filtering 1 = inversion between low-pass and high-pass filtering */
435 rw = tilec->resolutions[l - i].x1 - tilec->resolutions[l - i].x0;
436 rh = tilec->resolutions[l - i].y1 - tilec->resolutions[l - i].y0;
437 rw1= tilec->resolutions[l - i - 1].x1 - tilec->resolutions[l - i - 1].x0;
438 rh1= tilec->resolutions[l - i - 1].y1 - tilec->resolutions[l - i - 1].y0;
440 cas_row = tilec->resolutions[l - i].x0 % 2;
441 cas_col = tilec->resolutions[l - i].y0 % 2;
445 bj = (int*)opj_malloc(rh * sizeof(int));
446 for (j = 0; j < rw; j++) {
448 for (k = 0; k < rh; k++) bj[k] = aj[k*w];
449 dwt_encode_1_real(bj, dn, sn, cas_col);
450 dwt_deinterleave_v(bj, aj, dn, sn, w, cas_col);
456 bj = (int*)opj_malloc(rw * sizeof(int));
457 for (j = 0; j < rh; j++) {
459 for (k = 0; k < rw; k++) bj[k] = aj[k];
460 dwt_encode_1_real(bj, dn, sn, cas_row);
461 dwt_deinterleave_h(bj, aj, dn, sn, cas_row);
469 /* Get gain of 9-7 wavelet transform. */
471 int dwt_getgain_real(int orient) {
477 /* Get norm of 9-7 wavelet. */
479 double dwt_getnorm_real(int level, int orient) {
480 return dwt_norms_real[orient][level];
483 void dwt_calc_explicit_stepsizes(opj_tccp_t * tccp, int prec) {
484 int numbands, bandno;
485 numbands = 3 * tccp->numresolutions - 2;
486 for (bandno = 0; bandno < numbands; bandno++) {
488 int resno, level, orient, gain;
490 resno = (bandno == 0) ? 0 : ((bandno - 1) / 3 + 1);
491 orient = (bandno == 0) ? 0 : ((bandno - 1) % 3 + 1);
492 level = tccp->numresolutions - 1 - resno;
493 gain = (tccp->qmfbid == 0) ? 0 : ((orient == 0) ? 0 : (((orient == 1) || (orient == 2)) ? 1 : 2));
494 if (tccp->qntsty == J2K_CCP_QNTSTY_NOQNT) {
497 double norm = dwt_norms_real[orient][level];
498 stepsize = (1 << (gain)) / norm;
500 dwt_encode_stepsize((int) floor(stepsize * 8192.0), prec + gain, &tccp->stepsizes[bandno]);
506 /* Determine maximum computed resolution level for inverse wavelet transform */
508 static int dwt_decode_max_resolution(opj_tcd_resolution_t* restrict r, int i) {
513 if( mr < ( w = r->x1 - r->x0 ) )
515 if( mr < ( w = r->y1 - r->y0 ) )
523 /* Inverse wavelet transform in 2-D. */
525 static void dwt_decode_tile(opj_tcd_tilecomp_t* tilec, int numres, DWT1DFN dwt_1D) {
529 opj_tcd_resolution_t* tr = tilec->resolutions;
531 int rw = tr->x1 - tr->x0; /* width of the resolution level computed */
532 int rh = tr->y1 - tr->y0; /* height of the resolution level computed */
534 int w = tilec->x1 - tilec->x0;
536 h.mem = (int*)opj_aligned_malloc(dwt_decode_max_resolution(tr, numres) * sizeof(int));
540 int * restrict tiledp = tilec->data;
547 rw = tr->x1 - tr->x0;
548 rh = tr->y1 - tr->y0;
553 for(j = 0; j < rh; ++j) {
554 dwt_interleave_h(&h, &tiledp[j*w]);
556 memcpy(&tiledp[j*w], h.mem, rw * sizeof(int));
562 for(j = 0; j < rw; ++j){
564 dwt_interleave_v(&v, &tiledp[j], w);
566 for(k = 0; k < rh; ++k) {
567 tiledp[k * w + j] = v.mem[k];
571 opj_aligned_free(h.mem);
574 static void v4dwt_interleave_h(v4dwt_t* restrict w, float* restrict a, int x, int size){
575 float* restrict bi = (float*) (w->wavelet + w->cas);
578 for(k = 0; k < 2; ++k){
579 if (count + 3 * x < size && ((size_t) a & 0x0f) == 0 && ((size_t) bi & 0x0f) == 0 && (x & 0x0f) == 0) {
581 for(i = 0; i < count; ++i){
593 for(i = 0; i < count; ++i){
597 if(j > size) continue;
600 if(j > size) continue;
603 if(j > size) continue;
607 bi = (float*) (w->wavelet + 1 - w->cas);
614 static void v4dwt_interleave_v(v4dwt_t* restrict v , float* restrict a , int x){
615 v4* restrict bi = v->wavelet + v->cas;
617 for(i = 0; i < v->sn; ++i){
618 memcpy(&bi[i*2], &a[i*x], 4 * sizeof(float));
621 bi = v->wavelet + 1 - v->cas;
622 for(i = 0; i < v->dn; ++i){
623 memcpy(&bi[i*2], &a[i*x], 4 * sizeof(float));
629 static void v4dwt_decode_step1_sse(v4* w, int count, const __m128 c){
630 __m128* restrict vw = (__m128*) w;
632 /* 4x unrolled loop */
633 for(i = 0; i < count >> 2; ++i){
634 *vw = _mm_mul_ps(*vw, c);
636 *vw = _mm_mul_ps(*vw, c);
638 *vw = _mm_mul_ps(*vw, c);
640 *vw = _mm_mul_ps(*vw, c);
644 for(i = 0; i < count; ++i){
645 *vw = _mm_mul_ps(*vw, c);
650 static void v4dwt_decode_step2_sse(v4* l, v4* w, int k, int m, __m128 c){
651 __m128* restrict vl = (__m128*) l;
652 __m128* restrict vw = (__m128*) w;
654 __m128 tmp1, tmp2, tmp3;
656 for(i = 0; i < m; ++i){
659 vw[-1] = _mm_add_ps(tmp2, _mm_mul_ps(_mm_add_ps(tmp1, tmp3), c));
667 c = _mm_add_ps(c, c);
668 c = _mm_mul_ps(c, vl[0]);
671 vw[-1] = _mm_add_ps(tmp, c);
678 static void v4dwt_decode_step1(v4* w, int count, const float c){
679 float* restrict fw = (float*) w;
681 for(i = 0; i < count; ++i){
682 float tmp1 = fw[i*8 ];
683 float tmp2 = fw[i*8 + 1];
684 float tmp3 = fw[i*8 + 2];
685 float tmp4 = fw[i*8 + 3];
687 fw[i*8 + 1] = tmp2 * c;
688 fw[i*8 + 2] = tmp3 * c;
689 fw[i*8 + 3] = tmp4 * c;
693 static void v4dwt_decode_step2(v4* l, v4* w, int k, int m, float c){
694 float* restrict fl = (float*) l;
695 float* restrict fw = (float*) w;
697 for(i = 0; i < m; ++i){
698 float tmp1_1 = fl[0];
699 float tmp1_2 = fl[1];
700 float tmp1_3 = fl[2];
701 float tmp1_4 = fl[3];
702 float tmp2_1 = fw[-4];
703 float tmp2_2 = fw[-3];
704 float tmp2_3 = fw[-2];
705 float tmp2_4 = fw[-1];
706 float tmp3_1 = fw[0];
707 float tmp3_2 = fw[1];
708 float tmp3_3 = fw[2];
709 float tmp3_4 = fw[3];
710 fw[-4] = tmp2_1 + ((tmp1_1 + tmp3_1) * c);
711 fw[-3] = tmp2_2 + ((tmp1_2 + tmp3_2) * c);
712 fw[-2] = tmp2_3 + ((tmp1_3 + tmp3_3) * c);
713 fw[-1] = tmp2_4 + ((tmp1_4 + tmp3_4) * c);
744 /* Inverse 9-7 wavelet transform in 1-D. */
746 static void v4dwt_decode(v4dwt_t* restrict dwt){
749 if(!((dwt->dn > 0) || (dwt->sn > 1))){
755 if(!((dwt->sn > 0) || (dwt->dn > 1))) {
762 v4dwt_decode_step1_sse(dwt->wavelet+a, dwt->sn, _mm_set1_ps(K));
763 v4dwt_decode_step1_sse(dwt->wavelet+b, dwt->dn, _mm_set1_ps(c13318));
764 v4dwt_decode_step2_sse(dwt->wavelet+b, dwt->wavelet+a+1, dwt->sn, int_min(dwt->sn, dwt->dn-a), _mm_set1_ps(dwt_delta));
765 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));
766 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));
767 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));
769 v4dwt_decode_step1(dwt->wavelet+a, dwt->sn, K);
770 v4dwt_decode_step1(dwt->wavelet+b, dwt->dn, c13318);
771 v4dwt_decode_step2(dwt->wavelet+b, dwt->wavelet+a+1, dwt->sn, int_min(dwt->sn, dwt->dn-a), dwt_delta);
772 v4dwt_decode_step2(dwt->wavelet+a, dwt->wavelet+b+1, dwt->dn, int_min(dwt->dn, dwt->sn-b), dwt_gamma);
773 v4dwt_decode_step2(dwt->wavelet+b, dwt->wavelet+a+1, dwt->sn, int_min(dwt->sn, dwt->dn-a), dwt_beta);
774 v4dwt_decode_step2(dwt->wavelet+a, dwt->wavelet+b+1, dwt->dn, int_min(dwt->dn, dwt->sn-b), dwt_alpha);
779 /* Inverse 9-7 wavelet transform in 2-D. */
781 void dwt_decode_real(opj_tcd_tilecomp_t* restrict tilec, int numres){
785 opj_tcd_resolution_t* res = tilec->resolutions;
787 int rw = res->x1 - res->x0; /* width of the resolution level computed */
788 int rh = res->y1 - res->y0; /* height of the resolution level computed */
790 int w = tilec->x1 - tilec->x0;
792 h.wavelet = (v4*) opj_aligned_malloc((dwt_decode_max_resolution(res, numres)+5) * sizeof(v4));
793 v.wavelet = h.wavelet;
796 float * restrict aj = (float*) tilec->data;
797 int bufsize = (tilec->x1 - tilec->x0) * (tilec->y1 - tilec->y0);
805 rw = res->x1 - res->x0; /* width of the resolution level computed */
806 rh = res->y1 - res->y0; /* height of the resolution level computed */
811 for(j = rh; j > 3; j -= 4){
813 v4dwt_interleave_h(&h, aj, w, bufsize);
815 for(k = rw; --k >= 0;){
816 aj[k ] = h.wavelet[k].f[0];
817 aj[k+w ] = h.wavelet[k].f[1];
818 aj[k+w*2] = h.wavelet[k].f[2];
819 aj[k+w*3] = h.wavelet[k].f[3];
827 v4dwt_interleave_h(&h, aj, w, bufsize);
829 for(k = rw; --k >= 0;){
831 case 3: aj[k+w*2] = h.wavelet[k].f[2];
832 case 2: aj[k+w ] = h.wavelet[k].f[1];
833 case 1: aj[k ] = h.wavelet[k].f[0];
841 aj = (float*) tilec->data;
842 for(j = rw; j > 3; j -= 4){
844 v4dwt_interleave_v(&v, aj, w);
846 for(k = 0; k < rh; ++k){
847 memcpy(&aj[k*w], &v.wavelet[k], 4 * sizeof(float));
854 v4dwt_interleave_v(&v, aj, w);
856 for(k = 0; k < rh; ++k){
857 memcpy(&aj[k*w], &v.wavelet[k], j * sizeof(float));
862 opj_aligned_free(h.wavelet);