jpeg2000: Use separate fields for int and float codepaths
[libav.git] / libavcodec / jpeg2000.c
1 /*
2 * JPEG 2000 encoder and decoder common functions
3 * Copyright (c) 2007 Kamil Nowosad
4 * Copyright (c) 2013 Nicolas Bertrand <nicoinattendu@gmail.com>
5 *
6 * This file is part of Libav.
7 *
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23 /**
24 * @file
25 * JPEG 2000 image encoder and decoder common functions
26 */
27
28 #include "libavutil/common.h"
29 #include "libavutil/mem.h"
30 #include "avcodec.h"
31 #include "jpeg2000.h"
32
33 #define SHL(a, n) ((n) >= 0 ? (a) << (n) : (a) >> -(n))
34
35 /* tag tree routines */
36
37 /* allocate the memory for tag tree */
38 static int32_t tag_tree_size(uint16_t w, uint16_t h)
39 {
40 uint32_t res = 0;
41 while (w > 1 || h > 1) {
42 res += w * h;
43 if (res + 1 >= INT32_MAX)
44 return -1;
45 w = (w + 1) >> 1;
46 h = (h + 1) >> 1;
47 }
48 return (int32_t)(res + 1);
49 }
50
51 static Jpeg2000TgtNode *ff_jpeg2000_tag_tree_init(int w, int h)
52 {
53 int pw = w, ph = h;
54 Jpeg2000TgtNode *res, *t, *t2;
55 int32_t tt_size;
56
57 tt_size = tag_tree_size(w, h);
58 if (tt_size == -1)
59 return NULL;
60
61 t = res = av_mallocz_array(tt_size, sizeof(*t));
62 if (!res)
63 return NULL;
64
65 while (w > 1 || h > 1) {
66 int i, j;
67 pw = w;
68 ph = h;
69
70 w = (w + 1) >> 1;
71 h = (h + 1) >> 1;
72 t2 = t + pw * ph;
73
74 for (i = 0; i < ph; i++)
75 for (j = 0; j < pw; j++)
76 t[i * pw + j].parent = &t2[(i >> 1) * w + (j >> 1)];
77
78 t = t2;
79 }
80 t[0].parent = NULL;
81 return res;
82 }
83
84 uint8_t ff_jpeg2000_sigctxno_lut[256][4];
85
86 static int getsigctxno(int flag, int bandno)
87 {
88 int h, v, d;
89
90 h = ((flag & JPEG2000_T1_SIG_E) ? 1 : 0) +
91 ((flag & JPEG2000_T1_SIG_W) ? 1 : 0);
92 v = ((flag & JPEG2000_T1_SIG_N) ? 1 : 0) +
93 ((flag & JPEG2000_T1_SIG_S) ? 1 : 0);
94 d = ((flag & JPEG2000_T1_SIG_NE) ? 1 : 0) +
95 ((flag & JPEG2000_T1_SIG_NW) ? 1 : 0) +
96 ((flag & JPEG2000_T1_SIG_SE) ? 1 : 0) +
97 ((flag & JPEG2000_T1_SIG_SW) ? 1 : 0);
98 if (bandno < 3) {
99 if (bandno == 1)
100 FFSWAP(int, h, v);
101 if (h == 2)
102 return 8;
103 if (h == 1) {
104 if (v >= 1)
105 return 7;
106 if (d >= 1)
107 return 6;
108 return 5;
109 }
110 if (v == 2)
111 return 4;
112 if (v == 1)
113 return 3;
114 if (d >= 2)
115 return 2;
116 if (d == 1)
117 return 1;
118 } else {
119 if (d >= 3)
120 return 8;
121 if (d == 2) {
122 if (h + v >= 1)
123 return 7;
124 return 6;
125 }
126 if (d == 1) {
127 if (h + v >= 2)
128 return 5;
129 if (h + v == 1)
130 return 4;
131 return 3;
132 }
133 if (h + v >= 2)
134 return 2;
135 if (h + v == 1)
136 return 1;
137 }
138 return 0;
139 }
140
141 uint8_t ff_jpeg2000_sgnctxno_lut[16][16], ff_jpeg2000_xorbit_lut[16][16];
142
143 static const int contribtab[3][3] = { { 0, -1, 1 }, { -1, -1, 0 }, { 1, 0, 1 } };
144 static const int ctxlbltab[3][3] = { { 13, 12, 11 }, { 10, 9, 10 }, { 11, 12, 13 } };
145 static const int xorbittab[3][3] = { { 1, 1, 1 }, { 1, 0, 0 }, { 0, 0, 0 } };
146
147 static int getsgnctxno(int flag, uint8_t *xorbit)
148 {
149 int vcontrib, hcontrib;
150
151 hcontrib = contribtab[flag & JPEG2000_T1_SIG_E ? flag & JPEG2000_T1_SGN_E ? 1 : 2 : 0]
152 [flag & JPEG2000_T1_SIG_W ? flag & JPEG2000_T1_SGN_W ? 1 : 2 : 0] + 1;
153 vcontrib = contribtab[flag & JPEG2000_T1_SIG_S ? flag & JPEG2000_T1_SGN_S ? 1 : 2 : 0]
154 [flag & JPEG2000_T1_SIG_N ? flag & JPEG2000_T1_SGN_N ? 1 : 2 : 0] + 1;
155 *xorbit = xorbittab[hcontrib][vcontrib];
156
157 return ctxlbltab[hcontrib][vcontrib];
158 }
159
160 void ff_jpeg2000_init_tier1_luts(void)
161 {
162 int i, j;
163 for (i = 0; i < 256; i++)
164 for (j = 0; j < 4; j++)
165 ff_jpeg2000_sigctxno_lut[i][j] = getsigctxno(i, j);
166 for (i = 0; i < 16; i++)
167 for (j = 0; j < 16; j++)
168 ff_jpeg2000_sgnctxno_lut[i][j] =
169 getsgnctxno(i + (j << 8), &ff_jpeg2000_xorbit_lut[i][j]);
170 }
171
172 void ff_jpeg2000_set_significance(Jpeg2000T1Context *t1, int x, int y,
173 int negative)
174 {
175 x++;
176 y++;
177 t1->flags[y][x] |= JPEG2000_T1_SIG;
178 if (negative) {
179 t1->flags[y][x + 1] |= JPEG2000_T1_SIG_W | JPEG2000_T1_SGN_W;
180 t1->flags[y][x - 1] |= JPEG2000_T1_SIG_E | JPEG2000_T1_SGN_E;
181 t1->flags[y + 1][x] |= JPEG2000_T1_SIG_N | JPEG2000_T1_SGN_N;
182 t1->flags[y - 1][x] |= JPEG2000_T1_SIG_S | JPEG2000_T1_SGN_S;
183 } else {
184 t1->flags[y][x + 1] |= JPEG2000_T1_SIG_W;
185 t1->flags[y][x - 1] |= JPEG2000_T1_SIG_E;
186 t1->flags[y + 1][x] |= JPEG2000_T1_SIG_N;
187 t1->flags[y - 1][x] |= JPEG2000_T1_SIG_S;
188 }
189 t1->flags[y + 1][x + 1] |= JPEG2000_T1_SIG_NW;
190 t1->flags[y + 1][x - 1] |= JPEG2000_T1_SIG_NE;
191 t1->flags[y - 1][x + 1] |= JPEG2000_T1_SIG_SW;
192 t1->flags[y - 1][x - 1] |= JPEG2000_T1_SIG_SE;
193 }
194
195 static const uint8_t lut_gain[2][4] = { { 0, 0, 0, 0 }, { 0, 1, 1, 2 } };
196
197 int ff_jpeg2000_init_component(Jpeg2000Component *comp,
198 Jpeg2000CodingStyle *codsty,
199 Jpeg2000QuantStyle *qntsty,
200 int cbps, int dx, int dy,
201 AVCodecContext *avctx)
202 {
203 uint8_t log2_band_prec_width, log2_band_prec_height;
204 int reslevelno, bandno, gbandno = 0, ret, i, j;
205 uint32_t csize = 1;
206
207 if (!codsty->nreslevels2decode) {
208 av_log(avctx, AV_LOG_ERROR, "nreslevels2decode uninitialized\n");
209 return AVERROR_INVALIDDATA;
210 }
211
212 if (ret = ff_jpeg2000_dwt_init(&comp->dwt, comp->coord,
213 codsty->nreslevels2decode - 1,
214 codsty->transform))
215 return ret;
216 // component size comp->coord is uint16_t so ir cannot overflow
217 csize = (comp->coord[0][1] - comp->coord[0][0]) *
218 (comp->coord[1][1] - comp->coord[1][0]);
219
220 if (codsty->transform == FF_DWT97) {
221 comp->i_data = NULL;
222 comp->f_data = av_malloc_array(csize, sizeof(*comp->f_data));
223 if (!comp->f_data)
224 return AVERROR(ENOMEM);
225 } else {
226 comp->f_data = NULL;
227 comp->i_data = av_malloc_array(csize, sizeof(*comp->i_data));
228 if (!comp->i_data)
229 return AVERROR(ENOMEM);
230 }
231 comp->reslevel = av_malloc_array(codsty->nreslevels, sizeof(*comp->reslevel));
232 if (!comp->reslevel)
233 return AVERROR(ENOMEM);
234 /* LOOP on resolution levels */
235 for (reslevelno = 0; reslevelno < codsty->nreslevels; reslevelno++) {
236 int declvl = codsty->nreslevels - reslevelno; // N_L -r see ISO/IEC 15444-1:2002 B.5
237 Jpeg2000ResLevel *reslevel = comp->reslevel + reslevelno;
238
239 /* Compute borders for each resolution level.
240 * Computation of trx_0, trx_1, try_0 and try_1.
241 * see ISO/IEC 15444-1:2002 eq. B.5 and B-14 */
242 for (i = 0; i < 2; i++)
243 for (j = 0; j < 2; j++)
244 reslevel->coord[i][j] =
245 ff_jpeg2000_ceildivpow2(comp->coord_o[i][j], declvl - 1);
246 // update precincts size: 2^n value
247 reslevel->log2_prec_width = codsty->log2_prec_widths[reslevelno];
248 reslevel->log2_prec_height = codsty->log2_prec_heights[reslevelno];
249
250 /* Number of bands for each resolution level */
251 if (reslevelno == 0)
252 reslevel->nbands = 1;
253 else
254 reslevel->nbands = 3;
255
256 /* Number of precincts wich span the tile for resolution level reslevelno
257 * see B.6 in ISO/IEC 15444-1:2002 eq. B-16
258 * num_precincts_x = |- trx_1 / 2 ^ log2_prec_width) -| - (trx_0 / 2 ^ log2_prec_width)
259 * num_precincts_y = |- try_1 / 2 ^ log2_prec_width) -| - (try_0 / 2 ^ log2_prec_width)
260 * for Dcinema profiles in JPEG 2000
261 * num_precincts_x = |- trx_1 / 2 ^ log2_prec_width) -|
262 * num_precincts_y = |- try_1 / 2 ^ log2_prec_width) -| */
263 if (reslevel->coord[0][1] == reslevel->coord[0][0])
264 reslevel->num_precincts_x = 0;
265 else
266 reslevel->num_precincts_x =
267 ff_jpeg2000_ceildivpow2(reslevel->coord[0][1],
268 reslevel->log2_prec_width) -
269 (reslevel->coord[0][0] >> reslevel->log2_prec_width);
270
271 if (reslevel->coord[1][1] == reslevel->coord[1][0])
272 reslevel->num_precincts_y = 0;
273 else
274 reslevel->num_precincts_y =
275 ff_jpeg2000_ceildivpow2(reslevel->coord[1][1],
276 reslevel->log2_prec_height) -
277 (reslevel->coord[1][0] >> reslevel->log2_prec_height);
278
279 reslevel->band = av_malloc_array(reslevel->nbands, sizeof(*reslevel->band));
280 if (!reslevel->band)
281 return AVERROR(ENOMEM);
282
283 for (bandno = 0; bandno < reslevel->nbands; bandno++, gbandno++) {
284 Jpeg2000Band *band = reslevel->band + bandno;
285 int cblkno, precno;
286 int nb_precincts;
287
288 /* TODO: Implementation of quantization step not finished,
289 * see ISO/IEC 15444-1:2002 E.1 and A.6.4. */
290 switch (qntsty->quantsty) {
291 uint8_t gain;
292 int numbps;
293 case JPEG2000_QSTY_NONE:
294 /* TODO: to verify. No quantization in this case */
295 band->f_stepsize = 1;
296 break;
297 case JPEG2000_QSTY_SI:
298 /*TODO: Compute formula to implement. */
299 numbps = cbps +
300 lut_gain[codsty->transform == FF_DWT53][bandno + (reslevelno > 0)];
301 band->f_stepsize = SHL(2048 + qntsty->mant[gbandno],
302 2 + numbps - qntsty->expn[gbandno]);
303 break;
304 case JPEG2000_QSTY_SE:
305 /* Exponent quantization step.
306 * Formula:
307 * delta_b = 2 ^ (R_b - expn_b) * (1 + (mant_b / 2 ^ 11))
308 * R_b = R_I + log2 (gain_b )
309 * see ISO/IEC 15444-1:2002 E.1.1 eqn. E-3 and E-4 */
310 /* TODO/WARN: value of log2 (gain_b ) not taken into account
311 * but it works (compared to OpenJPEG). Why?
312 * Further investigation needed. */
313 gain = cbps;
314 band->f_stepsize = pow(2.0, gain - qntsty->expn[gbandno]);
315 band->f_stepsize *= qntsty->mant[gbandno] / 2048.0 + 1.0;
316 break;
317 default:
318 band->f_stepsize = 0;
319 av_log(avctx, AV_LOG_ERROR, "Unknown quantization format\n");
320 break;
321 }
322 /* FIXME: In openjepg code stespize = stepsize * 0.5. Why?
323 * If not set output of entropic decoder is not correct. */
324 if (!av_codec_is_encoder(avctx->codec))
325 band->f_stepsize *= 0.5;
326
327 band->i_stepsize = band->f_stepsize * (1 << 16);
328
329 /* computation of tbx_0, tbx_1, tby_0, tby_1
330 * see ISO/IEC 15444-1:2002 B.5 eq. B-15 and tbl B.1
331 * codeblock width and height is computed for
332 * DCI JPEG 2000 codeblock_width = codeblock_width = 32 = 2 ^ 5 */
333 if (reslevelno == 0) {
334 /* for reslevelno = 0, only one band, x0_b = y0_b = 0 */
335 for (i = 0; i < 2; i++)
336 for (j = 0; j < 2; j++)
337 band->coord[i][j] =
338 ff_jpeg2000_ceildivpow2(comp->coord_o[i][j],
339 declvl - 1);
340
341 log2_band_prec_width = reslevel->log2_prec_width;
342 log2_band_prec_height = reslevel->log2_prec_height;
343 /* see ISO/IEC 15444-1:2002 eq. B-17 and eq. B-15 */
344 band->log2_cblk_width = FFMIN(codsty->log2_cblk_width,
345 reslevel->log2_prec_width);
346 band->log2_cblk_height = FFMIN(codsty->log2_cblk_height,
347 reslevel->log2_prec_height);
348 } else {
349 /* 3 bands x0_b = 1 y0_b = 0; x0_b = 0 y0_b = 1; x0_b = y0_b = 1 */
350 /* x0_b and y0_b are computed with ((bandno + 1 >> i) & 1) */
351 for (i = 0; i < 2; i++)
352 for (j = 0; j < 2; j++)
353 /* Formula example for tbx_0 = ceildiv((tcx_0 - 2 ^ (declvl - 1) * x0_b) / declvl) */
354 band->coord[i][j] =
355 ff_jpeg2000_ceildivpow2(comp->coord_o[i][j] -
356 (((bandno + 1 >> i) & 1) << declvl - 1),
357 declvl);
358 /* TODO: Manage case of 3 band offsets here or
359 * in coding/decoding function? */
360
361 /* see ISO/IEC 15444-1:2002 eq. B-17 and eq. B-15 */
362 band->log2_cblk_width = FFMIN(codsty->log2_cblk_width,
363 reslevel->log2_prec_width - 1);
364 band->log2_cblk_height = FFMIN(codsty->log2_cblk_height,
365 reslevel->log2_prec_height - 1);
366
367 log2_band_prec_width = reslevel->log2_prec_width - 1;
368 log2_band_prec_height = reslevel->log2_prec_height - 1;
369 }
370
371 band->prec = av_malloc_array(reslevel->num_precincts_x *
372 reslevel->num_precincts_y,
373 sizeof(*band->prec));
374 if (!band->prec)
375 return AVERROR(ENOMEM);
376
377 nb_precincts = reslevel->num_precincts_x * reslevel->num_precincts_y;
378
379 for (precno = 0; precno < nb_precincts; precno++) {
380 Jpeg2000Prec *prec = band->prec + precno;
381
382 /* TODO: Explain formula for JPEG200 DCINEMA. */
383 /* TODO: Verify with previous count of codeblocks per band */
384
385 /* Compute P_x0 */
386 prec->coord[0][0] = (precno % reslevel->num_precincts_x) *
387 (1 << log2_band_prec_width);
388 prec->coord[0][0] = FFMAX(prec->coord[0][0], band->coord[0][0]);
389
390 /* Compute P_y0 */
391 prec->coord[1][0] = (precno / reslevel->num_precincts_x) *
392 (1 << log2_band_prec_height);
393 prec->coord[1][0] = FFMAX(prec->coord[1][0], band->coord[1][0]);
394
395 /* Compute P_x1 */
396 prec->coord[0][1] = prec->coord[0][0] +
397 (1 << log2_band_prec_width);
398 prec->coord[0][1] = FFMIN(prec->coord[0][1], band->coord[0][1]);
399
400 /* Compute P_y1 */
401 prec->coord[1][1] = prec->coord[1][0] +
402 (1 << log2_band_prec_height);
403 prec->coord[1][1] = FFMIN(prec->coord[1][1], band->coord[1][1]);
404
405 prec->nb_codeblocks_width =
406 ff_jpeg2000_ceildivpow2(prec->coord[0][1] -
407 prec->coord[0][0],
408 band->log2_cblk_width);
409 prec->nb_codeblocks_height =
410 ff_jpeg2000_ceildivpow2(prec->coord[1][1] -
411 prec->coord[1][0],
412 band->log2_cblk_height);
413
414 /* Tag trees initialization */
415 prec->cblkincl =
416 ff_jpeg2000_tag_tree_init(prec->nb_codeblocks_width,
417 prec->nb_codeblocks_height);
418 if (!prec->cblkincl)
419 return AVERROR(ENOMEM);
420
421 prec->zerobits =
422 ff_jpeg2000_tag_tree_init(prec->nb_codeblocks_width,
423 prec->nb_codeblocks_height);
424 if (!prec->zerobits)
425 return AVERROR(ENOMEM);
426
427 prec->cblk = av_mallocz_array(prec->nb_codeblocks_width *
428 prec->nb_codeblocks_height,
429 sizeof(*prec->cblk));
430 if (!prec->cblk)
431 return AVERROR(ENOMEM);
432 for (cblkno = 0; cblkno < prec->nb_codeblocks_width * prec->nb_codeblocks_height; cblkno++) {
433 Jpeg2000Cblk *cblk = prec->cblk + cblkno;
434 uint16_t Cx0, Cy0;
435
436 /* Compute coordinates of codeblocks */
437 /* Compute Cx0*/
438 Cx0 = (prec->coord[0][0] >> band->log2_cblk_width) << band->log2_cblk_width;
439 Cx0 = Cx0 + ((cblkno % prec->nb_codeblocks_width) << band->log2_cblk_width);
440 cblk->coord[0][0] = FFMAX(Cx0, prec->coord[0][0]);
441
442 /* Compute Cy0*/
443 Cy0 = (prec->coord[1][0] >> band->log2_cblk_height) << band->log2_cblk_height;
444 Cy0 = Cy0 + ((cblkno / prec->nb_codeblocks_width) << band->log2_cblk_height);
445 cblk->coord[1][0] = FFMAX(Cy0, prec->coord[1][0]);
446
447 /* Compute Cx1 */
448 cblk->coord[0][1] = FFMIN(Cx0 + (1 << band->log2_cblk_width),
449 prec->coord[0][1]);
450
451 /* Compute Cy1 */
452 cblk->coord[1][1] = FFMIN(Cy0 + (1 << band->log2_cblk_height),
453 prec->coord[1][1]);
454 /* Update code-blocks coordinates according sub-band position */
455 if ((bandno + !!reslevelno) & 1) {
456 cblk->coord[0][0] += comp->reslevel[reslevelno-1].coord[0][1] -
457 comp->reslevel[reslevelno-1].coord[0][0];
458 cblk->coord[0][1] += comp->reslevel[reslevelno-1].coord[0][1] -
459 comp->reslevel[reslevelno-1].coord[0][0];
460 }
461 if ((bandno + !!reslevelno) & 2) {
462 cblk->coord[1][0] += comp->reslevel[reslevelno-1].coord[1][1] -
463 comp->reslevel[reslevelno-1].coord[1][0];
464 cblk->coord[1][1] += comp->reslevel[reslevelno-1].coord[1][1] -
465 comp->reslevel[reslevelno-1].coord[1][0];
466 }
467
468 cblk->zero = 0;
469 cblk->lblock = 3;
470 cblk->length = 0;
471 cblk->lengthinc = 0;
472 cblk->npasses = 0;
473 }
474 }
475 }
476 }
477 return 0;
478 }
479
480 void ff_jpeg2000_cleanup(Jpeg2000Component *comp, Jpeg2000CodingStyle *codsty)
481 {
482 int reslevelno, bandno, precno;
483 for (reslevelno = 0; reslevelno < codsty->nreslevels; reslevelno++) {
484 Jpeg2000ResLevel *reslevel = comp->reslevel + reslevelno;
485
486 for (bandno = 0; bandno < reslevel->nbands; bandno++) {
487 Jpeg2000Band *band = reslevel->band + bandno;
488 for (precno = 0; precno < reslevel->num_precincts_x * reslevel->num_precincts_y; precno++) {
489 Jpeg2000Prec *prec = band->prec + precno;
490 av_freep(&prec->zerobits);
491 av_freep(&prec->cblkincl);
492 av_freep(&prec->cblk);
493 }
494
495 av_freep(&band->prec);
496 }
497 av_freep(&reslevel->band);
498 }
499
500 ff_dwt_destroy(&comp->dwt);
501 av_freep(&comp->reslevel);
502 av_freep(&comp->i_data);
503 av_freep(&comp->f_data);
504 }