Imported Debian version 2.5.3~trusty1
[deb_ffmpeg.git] / ffmpeg / libavcodec / diracdec.c
1 /*
2 * Copyright (C) 2007 Marco Gerards <marco@gnu.org>
3 * Copyright (C) 2009 David Conrad
4 * Copyright (C) 2011 Jordi Ortiz
5 *
6 * This file is part of FFmpeg.
7 *
8 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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 * Dirac Decoder
26 * @author Marco Gerards <marco@gnu.org>, David Conrad, Jordi Ortiz <nenjordi@gmail.com>
27 */
28
29 #include "avcodec.h"
30 #include "get_bits.h"
31 #include "bytestream.h"
32 #include "internal.h"
33 #include "golomb.h"
34 #include "dirac_arith.h"
35 #include "mpeg12data.h"
36 #include "libavcodec/mpegvideo.h"
37 #include "mpegvideoencdsp.h"
38 #include "dirac_dwt.h"
39 #include "dirac.h"
40 #include "diracdsp.h"
41 #include "videodsp.h"
42
43 /**
44 * The spec limits the number of wavelet decompositions to 4 for both
45 * level 1 (VC-2) and 128 (long-gop default).
46 * 5 decompositions is the maximum before >16-bit buffers are needed.
47 * Schroedinger allows this for DD 9,7 and 13,7 wavelets only, limiting
48 * the others to 4 decompositions (or 3 for the fidelity filter).
49 *
50 * We use this instead of MAX_DECOMPOSITIONS to save some memory.
51 */
52 #define MAX_DWT_LEVELS 5
53
54 /**
55 * The spec limits this to 3 for frame coding, but in practice can be as high as 6
56 */
57 #define MAX_REFERENCE_FRAMES 8
58 #define MAX_DELAY 5 /* limit for main profile for frame coding (TODO: field coding) */
59 #define MAX_FRAMES (MAX_REFERENCE_FRAMES + MAX_DELAY + 1)
60 #define MAX_QUANT 68 /* max quant for VC-2 */
61 #define MAX_BLOCKSIZE 32 /* maximum xblen/yblen we support */
62
63 /**
64 * DiracBlock->ref flags, if set then the block does MC from the given ref
65 */
66 #define DIRAC_REF_MASK_REF1 1
67 #define DIRAC_REF_MASK_REF2 2
68 #define DIRAC_REF_MASK_GLOBAL 4
69
70 /**
71 * Value of Picture.reference when Picture is not a reference picture, but
72 * is held for delayed output.
73 */
74 #define DELAYED_PIC_REF 4
75
76 #define CALC_PADDING(size, depth) \
77 (((size + (1 << depth) - 1) >> depth) << depth)
78
79 #define DIVRNDUP(a, b) (((a) + (b) - 1) / (b))
80
81 typedef struct {
82 AVFrame *avframe;
83 int interpolated[3]; /* 1 if hpel[] is valid */
84 uint8_t *hpel[3][4];
85 uint8_t *hpel_base[3][4];
86 } DiracFrame;
87
88 typedef struct {
89 union {
90 int16_t mv[2][2];
91 int16_t dc[3];
92 } u; /* anonymous unions aren't in C99 :( */
93 uint8_t ref;
94 } DiracBlock;
95
96 typedef struct SubBand {
97 int level;
98 int orientation;
99 int stride;
100 int width;
101 int height;
102 int quant;
103 IDWTELEM *ibuf;
104 struct SubBand *parent;
105
106 /* for low delay */
107 unsigned length;
108 const uint8_t *coeff_data;
109 } SubBand;
110
111 typedef struct Plane {
112 int width;
113 int height;
114 ptrdiff_t stride;
115
116 int idwt_width;
117 int idwt_height;
118 int idwt_stride;
119 IDWTELEM *idwt_buf;
120 IDWTELEM *idwt_buf_base;
121 IDWTELEM *idwt_tmp;
122
123 /* block length */
124 uint8_t xblen;
125 uint8_t yblen;
126 /* block separation (block n+1 starts after this many pixels in block n) */
127 uint8_t xbsep;
128 uint8_t ybsep;
129 /* amount of overspill on each edge (half of the overlap between blocks) */
130 uint8_t xoffset;
131 uint8_t yoffset;
132
133 SubBand band[MAX_DWT_LEVELS][4];
134 } Plane;
135
136 typedef struct DiracContext {
137 AVCodecContext *avctx;
138 MpegvideoEncDSPContext mpvencdsp;
139 VideoDSPContext vdsp;
140 DiracDSPContext diracdsp;
141 GetBitContext gb;
142 dirac_source_params source;
143 int seen_sequence_header;
144 int frame_number; /* number of the next frame to display */
145 Plane plane[3];
146 int chroma_x_shift;
147 int chroma_y_shift;
148
149 int zero_res; /* zero residue flag */
150 int is_arith; /* whether coeffs use arith or golomb coding */
151 int low_delay; /* use the low delay syntax */
152 int globalmc_flag; /* use global motion compensation */
153 int num_refs; /* number of reference pictures */
154
155 /* wavelet decoding */
156 unsigned wavelet_depth; /* depth of the IDWT */
157 unsigned wavelet_idx;
158
159 /**
160 * schroedinger older than 1.0.8 doesn't store
161 * quant delta if only one codebook exists in a band
162 */
163 unsigned old_delta_quant;
164 unsigned codeblock_mode;
165
166 struct {
167 unsigned width;
168 unsigned height;
169 } codeblock[MAX_DWT_LEVELS+1];
170
171 struct {
172 unsigned num_x; /* number of horizontal slices */
173 unsigned num_y; /* number of vertical slices */
174 AVRational bytes; /* average bytes per slice */
175 uint8_t quant[MAX_DWT_LEVELS][4]; /* [DIRAC_STD] E.1 */
176 } lowdelay;
177
178 struct {
179 int pan_tilt[2]; /* pan/tilt vector */
180 int zrs[2][2]; /* zoom/rotate/shear matrix */
181 int perspective[2]; /* perspective vector */
182 unsigned zrs_exp;
183 unsigned perspective_exp;
184 } globalmc[2];
185
186 /* motion compensation */
187 uint8_t mv_precision; /* [DIRAC_STD] REFS_WT_PRECISION */
188 int16_t weight[2]; /* [DIRAC_STD] REF1_WT and REF2_WT */
189 unsigned weight_log2denom; /* [DIRAC_STD] REFS_WT_PRECISION */
190
191 int blwidth; /* number of blocks (horizontally) */
192 int blheight; /* number of blocks (vertically) */
193 int sbwidth; /* number of superblocks (horizontally) */
194 int sbheight; /* number of superblocks (vertically) */
195
196 uint8_t *sbsplit;
197 DiracBlock *blmotion;
198
199 uint8_t *edge_emu_buffer[4];
200 uint8_t *edge_emu_buffer_base;
201
202 uint16_t *mctmp; /* buffer holding the MC data multiplied by OBMC weights */
203 uint8_t *mcscratch;
204 int buffer_stride;
205
206 DECLARE_ALIGNED(16, uint8_t, obmc_weight)[3][MAX_BLOCKSIZE*MAX_BLOCKSIZE];
207
208 void (*put_pixels_tab[4])(uint8_t *dst, const uint8_t *src[5], int stride, int h);
209 void (*avg_pixels_tab[4])(uint8_t *dst, const uint8_t *src[5], int stride, int h);
210 void (*add_obmc)(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
211 dirac_weight_func weight_func;
212 dirac_biweight_func biweight_func;
213
214 DiracFrame *current_picture;
215 DiracFrame *ref_pics[2];
216
217 DiracFrame *ref_frames[MAX_REFERENCE_FRAMES+1];
218 DiracFrame *delay_frames[MAX_DELAY+1];
219 DiracFrame all_frames[MAX_FRAMES];
220 } DiracContext;
221
222 /**
223 * Dirac Specification ->
224 * Parse code values. 9.6.1 Table 9.1
225 */
226 enum dirac_parse_code {
227 pc_seq_header = 0x00,
228 pc_eos = 0x10,
229 pc_aux_data = 0x20,
230 pc_padding = 0x30,
231 };
232
233 enum dirac_subband {
234 subband_ll = 0,
235 subband_hl = 1,
236 subband_lh = 2,
237 subband_hh = 3,
238 subband_nb,
239 };
240
241 static const uint8_t default_qmat[][4][4] = {
242 { { 5, 3, 3, 0}, { 0, 4, 4, 1}, { 0, 5, 5, 2}, { 0, 6, 6, 3} },
243 { { 4, 2, 2, 0}, { 0, 4, 4, 2}, { 0, 5, 5, 3}, { 0, 7, 7, 5} },
244 { { 5, 3, 3, 0}, { 0, 4, 4, 1}, { 0, 5, 5, 2}, { 0, 6, 6, 3} },
245 { { 8, 4, 4, 0}, { 0, 4, 4, 0}, { 0, 4, 4, 0}, { 0, 4, 4, 0} },
246 { { 8, 4, 4, 0}, { 0, 4, 4, 0}, { 0, 4, 4, 0}, { 0, 4, 4, 0} },
247 { { 0, 4, 4, 8}, { 0, 8, 8, 12}, { 0, 13, 13, 17}, { 0, 17, 17, 21} },
248 { { 3, 1, 1, 0}, { 0, 4, 4, 2}, { 0, 6, 6, 5}, { 0, 9, 9, 7} },
249 };
250
251 static const int qscale_tab[MAX_QUANT+1] = {
252 4, 5, 6, 7, 8, 10, 11, 13,
253 16, 19, 23, 27, 32, 38, 45, 54,
254 64, 76, 91, 108, 128, 152, 181, 215,
255 256, 304, 362, 431, 512, 609, 724, 861,
256 1024, 1218, 1448, 1722, 2048, 2435, 2896, 3444,
257 4096, 4871, 5793, 6889, 8192, 9742, 11585, 13777,
258 16384, 19484, 23170, 27554, 32768, 38968, 46341, 55109,
259 65536, 77936
260 };
261
262 static const int qoffset_intra_tab[MAX_QUANT+1] = {
263 1, 2, 3, 4, 4, 5, 6, 7,
264 8, 10, 12, 14, 16, 19, 23, 27,
265 32, 38, 46, 54, 64, 76, 91, 108,
266 128, 152, 181, 216, 256, 305, 362, 431,
267 512, 609, 724, 861, 1024, 1218, 1448, 1722,
268 2048, 2436, 2897, 3445, 4096, 4871, 5793, 6889,
269 8192, 9742, 11585, 13777, 16384, 19484, 23171, 27555,
270 32768, 38968
271 };
272
273 static const int qoffset_inter_tab[MAX_QUANT+1] = {
274 1, 2, 2, 3, 3, 4, 4, 5,
275 6, 7, 9, 10, 12, 14, 17, 20,
276 24, 29, 34, 41, 48, 57, 68, 81,
277 96, 114, 136, 162, 192, 228, 272, 323,
278 384, 457, 543, 646, 768, 913, 1086, 1292,
279 1536, 1827, 2172, 2583, 3072, 3653, 4344, 5166,
280 6144, 7307, 8689, 10333, 12288, 14613, 17378, 20666,
281 24576, 29226
282 };
283
284 /* magic number division by 3 from schroedinger */
285 static inline int divide3(int x)
286 {
287 return ((x+1)*21845 + 10922) >> 16;
288 }
289
290 static DiracFrame *remove_frame(DiracFrame *framelist[], int picnum)
291 {
292 DiracFrame *remove_pic = NULL;
293 int i, remove_idx = -1;
294
295 for (i = 0; framelist[i]; i++)
296 if (framelist[i]->avframe->display_picture_number == picnum) {
297 remove_pic = framelist[i];
298 remove_idx = i;
299 }
300
301 if (remove_pic)
302 for (i = remove_idx; framelist[i]; i++)
303 framelist[i] = framelist[i+1];
304
305 return remove_pic;
306 }
307
308 static int add_frame(DiracFrame *framelist[], int maxframes, DiracFrame *frame)
309 {
310 int i;
311 for (i = 0; i < maxframes; i++)
312 if (!framelist[i]) {
313 framelist[i] = frame;
314 return 0;
315 }
316 return -1;
317 }
318
319 static int alloc_sequence_buffers(DiracContext *s)
320 {
321 int sbwidth = DIVRNDUP(s->source.width, 4);
322 int sbheight = DIVRNDUP(s->source.height, 4);
323 int i, w, h, top_padding;
324
325 /* todo: think more about this / use or set Plane here */
326 for (i = 0; i < 3; i++) {
327 int max_xblen = MAX_BLOCKSIZE >> (i ? s->chroma_x_shift : 0);
328 int max_yblen = MAX_BLOCKSIZE >> (i ? s->chroma_y_shift : 0);
329 w = s->source.width >> (i ? s->chroma_x_shift : 0);
330 h = s->source.height >> (i ? s->chroma_y_shift : 0);
331
332 /* we allocate the max we support here since num decompositions can
333 * change from frame to frame. Stride is aligned to 16 for SIMD, and
334 * 1<<MAX_DWT_LEVELS top padding to avoid if(y>0) in arith decoding
335 * MAX_BLOCKSIZE padding for MC: blocks can spill up to half of that
336 * on each side */
337 top_padding = FFMAX(1<<MAX_DWT_LEVELS, max_yblen/2);
338 w = FFALIGN(CALC_PADDING(w, MAX_DWT_LEVELS), 8); /* FIXME: Should this be 16 for SSE??? */
339 h = top_padding + CALC_PADDING(h, MAX_DWT_LEVELS) + max_yblen/2;
340
341 s->plane[i].idwt_buf_base = av_mallocz_array((w+max_xblen), h * sizeof(IDWTELEM));
342 s->plane[i].idwt_tmp = av_malloc_array((w+16), sizeof(IDWTELEM));
343 s->plane[i].idwt_buf = s->plane[i].idwt_buf_base + top_padding*w;
344 if (!s->plane[i].idwt_buf_base || !s->plane[i].idwt_tmp)
345 return AVERROR(ENOMEM);
346 }
347
348 /* fixme: allocate using real stride here */
349 s->sbsplit = av_malloc_array(sbwidth, sbheight);
350 s->blmotion = av_malloc_array(sbwidth, sbheight * 16 * sizeof(*s->blmotion));
351
352 if (!s->sbsplit || !s->blmotion)
353 return AVERROR(ENOMEM);
354 return 0;
355 }
356
357 static int alloc_buffers(DiracContext *s, int stride)
358 {
359 int w = s->source.width;
360 int h = s->source.height;
361
362 av_assert0(stride >= w);
363 stride += 64;
364
365 if (s->buffer_stride >= stride)
366 return 0;
367 s->buffer_stride = 0;
368
369 av_freep(&s->edge_emu_buffer_base);
370 memset(s->edge_emu_buffer, 0, sizeof(s->edge_emu_buffer));
371 av_freep(&s->mctmp);
372 av_freep(&s->mcscratch);
373
374 s->edge_emu_buffer_base = av_malloc_array(stride, MAX_BLOCKSIZE);
375
376 s->mctmp = av_malloc_array((stride+MAX_BLOCKSIZE), (h+MAX_BLOCKSIZE) * sizeof(*s->mctmp));
377 s->mcscratch = av_malloc_array(stride, MAX_BLOCKSIZE);
378
379 if (!s->edge_emu_buffer_base || !s->mctmp || !s->mcscratch)
380 return AVERROR(ENOMEM);
381
382 s->buffer_stride = stride;
383 return 0;
384 }
385
386 static void free_sequence_buffers(DiracContext *s)
387 {
388 int i, j, k;
389
390 for (i = 0; i < MAX_FRAMES; i++) {
391 if (s->all_frames[i].avframe->data[0]) {
392 av_frame_unref(s->all_frames[i].avframe);
393 memset(s->all_frames[i].interpolated, 0, sizeof(s->all_frames[i].interpolated));
394 }
395
396 for (j = 0; j < 3; j++)
397 for (k = 1; k < 4; k++)
398 av_freep(&s->all_frames[i].hpel_base[j][k]);
399 }
400
401 memset(s->ref_frames, 0, sizeof(s->ref_frames));
402 memset(s->delay_frames, 0, sizeof(s->delay_frames));
403
404 for (i = 0; i < 3; i++) {
405 av_freep(&s->plane[i].idwt_buf_base);
406 av_freep(&s->plane[i].idwt_tmp);
407 }
408
409 s->buffer_stride = 0;
410 av_freep(&s->sbsplit);
411 av_freep(&s->blmotion);
412 av_freep(&s->edge_emu_buffer_base);
413
414 av_freep(&s->mctmp);
415 av_freep(&s->mcscratch);
416 }
417
418 static av_cold int dirac_decode_init(AVCodecContext *avctx)
419 {
420 DiracContext *s = avctx->priv_data;
421 int i;
422
423 s->avctx = avctx;
424 s->frame_number = -1;
425
426 ff_diracdsp_init(&s->diracdsp);
427 ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx);
428 ff_videodsp_init(&s->vdsp, 8);
429
430 for (i = 0; i < MAX_FRAMES; i++) {
431 s->all_frames[i].avframe = av_frame_alloc();
432 if (!s->all_frames[i].avframe) {
433 while (i > 0)
434 av_frame_free(&s->all_frames[--i].avframe);
435 return AVERROR(ENOMEM);
436 }
437 }
438
439 return 0;
440 }
441
442 static void dirac_decode_flush(AVCodecContext *avctx)
443 {
444 DiracContext *s = avctx->priv_data;
445 free_sequence_buffers(s);
446 s->seen_sequence_header = 0;
447 s->frame_number = -1;
448 }
449
450 static av_cold int dirac_decode_end(AVCodecContext *avctx)
451 {
452 DiracContext *s = avctx->priv_data;
453 int i;
454
455 dirac_decode_flush(avctx);
456 for (i = 0; i < MAX_FRAMES; i++)
457 av_frame_free(&s->all_frames[i].avframe);
458
459 return 0;
460 }
461
462 #define SIGN_CTX(x) (CTX_SIGN_ZERO + ((x) > 0) - ((x) < 0))
463
464 static inline void coeff_unpack_arith(DiracArith *c, int qfactor, int qoffset,
465 SubBand *b, IDWTELEM *buf, int x, int y)
466 {
467 int coeff, sign;
468 int sign_pred = 0;
469 int pred_ctx = CTX_ZPZN_F1;
470
471 /* Check if the parent subband has a 0 in the corresponding position */
472 if (b->parent)
473 pred_ctx += !!b->parent->ibuf[b->parent->stride * (y>>1) + (x>>1)] << 1;
474
475 if (b->orientation == subband_hl)
476 sign_pred = buf[-b->stride];
477
478 /* Determine if the pixel has only zeros in its neighbourhood */
479 if (x) {
480 pred_ctx += !(buf[-1] | buf[-b->stride] | buf[-1-b->stride]);
481 if (b->orientation == subband_lh)
482 sign_pred = buf[-1];
483 } else {
484 pred_ctx += !buf[-b->stride];
485 }
486
487 coeff = dirac_get_arith_uint(c, pred_ctx, CTX_COEFF_DATA);
488 if (coeff) {
489 coeff = (coeff * qfactor + qoffset + 2) >> 2;
490 sign = dirac_get_arith_bit(c, SIGN_CTX(sign_pred));
491 coeff = (coeff ^ -sign) + sign;
492 }
493 *buf = coeff;
494 }
495
496 static inline int coeff_unpack_golomb(GetBitContext *gb, int qfactor, int qoffset)
497 {
498 int sign, coeff;
499
500 coeff = svq3_get_ue_golomb(gb);
501 if (coeff) {
502 coeff = (coeff * qfactor + qoffset + 2) >> 2;
503 sign = get_bits1(gb);
504 coeff = (coeff ^ -sign) + sign;
505 }
506 return coeff;
507 }
508
509 /**
510 * Decode the coeffs in the rectangle defined by left, right, top, bottom
511 * [DIRAC_STD] 13.4.3.2 Codeblock unpacking loop. codeblock()
512 */
513 static inline void codeblock(DiracContext *s, SubBand *b,
514 GetBitContext *gb, DiracArith *c,
515 int left, int right, int top, int bottom,
516 int blockcnt_one, int is_arith)
517 {
518 int x, y, zero_block;
519 int qoffset, qfactor;
520 IDWTELEM *buf;
521
522 /* check for any coded coefficients in this codeblock */
523 if (!blockcnt_one) {
524 if (is_arith)
525 zero_block = dirac_get_arith_bit(c, CTX_ZERO_BLOCK);
526 else
527 zero_block = get_bits1(gb);
528
529 if (zero_block)
530 return;
531 }
532
533 if (s->codeblock_mode && !(s->old_delta_quant && blockcnt_one)) {
534 int quant = b->quant;
535 if (is_arith)
536 quant += dirac_get_arith_int(c, CTX_DELTA_Q_F, CTX_DELTA_Q_DATA);
537 else
538 quant += dirac_get_se_golomb(gb);
539 if (quant < 0) {
540 av_log(s->avctx, AV_LOG_ERROR, "Invalid quant\n");
541 return;
542 }
543 b->quant = quant;
544 }
545
546 b->quant = FFMIN(b->quant, MAX_QUANT);
547
548 qfactor = qscale_tab[b->quant];
549 /* TODO: context pointer? */
550 if (!s->num_refs)
551 qoffset = qoffset_intra_tab[b->quant];
552 else
553 qoffset = qoffset_inter_tab[b->quant];
554
555 buf = b->ibuf + top * b->stride;
556 for (y = top; y < bottom; y++) {
557 for (x = left; x < right; x++) {
558 /* [DIRAC_STD] 13.4.4 Subband coefficients. coeff_unpack() */
559 if (is_arith)
560 coeff_unpack_arith(c, qfactor, qoffset, b, buf+x, x, y);
561 else
562 buf[x] = coeff_unpack_golomb(gb, qfactor, qoffset);
563 }
564 buf += b->stride;
565 }
566 }
567
568 /**
569 * Dirac Specification ->
570 * 13.3 intra_dc_prediction(band)
571 */
572 static inline void intra_dc_prediction(SubBand *b)
573 {
574 IDWTELEM *buf = b->ibuf;
575 int x, y;
576
577 for (x = 1; x < b->width; x++)
578 buf[x] += buf[x-1];
579 buf += b->stride;
580
581 for (y = 1; y < b->height; y++) {
582 buf[0] += buf[-b->stride];
583
584 for (x = 1; x < b->width; x++) {
585 int pred = buf[x - 1] + buf[x - b->stride] + buf[x - b->stride-1];
586 buf[x] += divide3(pred);
587 }
588 buf += b->stride;
589 }
590 }
591
592 /**
593 * Dirac Specification ->
594 * 13.4.2 Non-skipped subbands. subband_coeffs()
595 */
596 static av_always_inline void decode_subband_internal(DiracContext *s, SubBand *b, int is_arith)
597 {
598 int cb_x, cb_y, left, right, top, bottom;
599 DiracArith c;
600 GetBitContext gb;
601 int cb_width = s->codeblock[b->level + (b->orientation != subband_ll)].width;
602 int cb_height = s->codeblock[b->level + (b->orientation != subband_ll)].height;
603 int blockcnt_one = (cb_width + cb_height) == 2;
604
605 if (!b->length)
606 return;
607
608 init_get_bits8(&gb, b->coeff_data, b->length);
609
610 if (is_arith)
611 ff_dirac_init_arith_decoder(&c, &gb, b->length);
612
613 top = 0;
614 for (cb_y = 0; cb_y < cb_height; cb_y++) {
615 bottom = (b->height * (cb_y+1LL)) / cb_height;
616 left = 0;
617 for (cb_x = 0; cb_x < cb_width; cb_x++) {
618 right = (b->width * (cb_x+1LL)) / cb_width;
619 codeblock(s, b, &gb, &c, left, right, top, bottom, blockcnt_one, is_arith);
620 left = right;
621 }
622 top = bottom;
623 }
624
625 if (b->orientation == subband_ll && s->num_refs == 0)
626 intra_dc_prediction(b);
627 }
628
629 static int decode_subband_arith(AVCodecContext *avctx, void *b)
630 {
631 DiracContext *s = avctx->priv_data;
632 decode_subband_internal(s, b, 1);
633 return 0;
634 }
635
636 static int decode_subband_golomb(AVCodecContext *avctx, void *arg)
637 {
638 DiracContext *s = avctx->priv_data;
639 SubBand **b = arg;
640 decode_subband_internal(s, *b, 0);
641 return 0;
642 }
643
644 /**
645 * Dirac Specification ->
646 * [DIRAC_STD] 13.4.1 core_transform_data()
647 */
648 static void decode_component(DiracContext *s, int comp)
649 {
650 AVCodecContext *avctx = s->avctx;
651 SubBand *bands[3*MAX_DWT_LEVELS+1];
652 enum dirac_subband orientation;
653 int level, num_bands = 0;
654
655 /* Unpack all subbands at all levels. */
656 for (level = 0; level < s->wavelet_depth; level++) {
657 for (orientation = !!level; orientation < 4; orientation++) {
658 SubBand *b = &s->plane[comp].band[level][orientation];
659 bands[num_bands++] = b;
660
661 align_get_bits(&s->gb);
662 /* [DIRAC_STD] 13.4.2 subband() */
663 b->length = svq3_get_ue_golomb(&s->gb);
664 if (b->length) {
665 b->quant = svq3_get_ue_golomb(&s->gb);
666 align_get_bits(&s->gb);
667 b->coeff_data = s->gb.buffer + get_bits_count(&s->gb)/8;
668 b->length = FFMIN(b->length, FFMAX(get_bits_left(&s->gb)/8, 0));
669 skip_bits_long(&s->gb, b->length*8);
670 }
671 }
672 /* arithmetic coding has inter-level dependencies, so we can only execute one level at a time */
673 if (s->is_arith)
674 avctx->execute(avctx, decode_subband_arith, &s->plane[comp].band[level][!!level],
675 NULL, 4-!!level, sizeof(SubBand));
676 }
677 /* golomb coding has no inter-level dependencies, so we can execute all subbands in parallel */
678 if (!s->is_arith)
679 avctx->execute(avctx, decode_subband_golomb, bands, NULL, num_bands, sizeof(SubBand*));
680 }
681
682 /* [DIRAC_STD] 13.5.5.2 Luma slice subband data. luma_slice_band(level,orient,sx,sy) --> if b2 == NULL */
683 /* [DIRAC_STD] 13.5.5.3 Chroma slice subband data. chroma_slice_band(level,orient,sx,sy) --> if b2 != NULL */
684 static void lowdelay_subband(DiracContext *s, GetBitContext *gb, int quant,
685 int slice_x, int slice_y, int bits_end,
686 SubBand *b1, SubBand *b2)
687 {
688 int left = b1->width * slice_x / s->lowdelay.num_x;
689 int right = b1->width *(slice_x+1) / s->lowdelay.num_x;
690 int top = b1->height * slice_y / s->lowdelay.num_y;
691 int bottom = b1->height *(slice_y+1) / s->lowdelay.num_y;
692
693 int qfactor = qscale_tab[FFMIN(quant, MAX_QUANT)];
694 int qoffset = qoffset_intra_tab[FFMIN(quant, MAX_QUANT)];
695
696 IDWTELEM *buf1 = b1->ibuf + top * b1->stride;
697 IDWTELEM *buf2 = b2 ? b2->ibuf + top * b2->stride : NULL;
698 int x, y;
699 /* we have to constantly check for overread since the spec explicitly
700 requires this, with the meaning that all remaining coeffs are set to 0 */
701 if (get_bits_count(gb) >= bits_end)
702 return;
703
704 for (y = top; y < bottom; y++) {
705 for (x = left; x < right; x++) {
706 buf1[x] = coeff_unpack_golomb(gb, qfactor, qoffset);
707 if (get_bits_count(gb) >= bits_end)
708 return;
709 if (buf2) {
710 buf2[x] = coeff_unpack_golomb(gb, qfactor, qoffset);
711 if (get_bits_count(gb) >= bits_end)
712 return;
713 }
714 }
715 buf1 += b1->stride;
716 if (buf2)
717 buf2 += b2->stride;
718 }
719 }
720
721 struct lowdelay_slice {
722 GetBitContext gb;
723 int slice_x;
724 int slice_y;
725 int bytes;
726 };
727
728
729 /**
730 * Dirac Specification ->
731 * 13.5.2 Slices. slice(sx,sy)
732 */
733 static int decode_lowdelay_slice(AVCodecContext *avctx, void *arg)
734 {
735 DiracContext *s = avctx->priv_data;
736 struct lowdelay_slice *slice = arg;
737 GetBitContext *gb = &slice->gb;
738 enum dirac_subband orientation;
739 int level, quant, chroma_bits, chroma_end;
740
741 int quant_base = get_bits(gb, 7); /*[DIRAC_STD] qindex */
742 int length_bits = av_log2(8 * slice->bytes)+1;
743 int luma_bits = get_bits_long(gb, length_bits);
744 int luma_end = get_bits_count(gb) + FFMIN(luma_bits, get_bits_left(gb));
745
746 /* [DIRAC_STD] 13.5.5.2 luma_slice_band */
747 for (level = 0; level < s->wavelet_depth; level++)
748 for (orientation = !!level; orientation < 4; orientation++) {
749 quant = FFMAX(quant_base - s->lowdelay.quant[level][orientation], 0);
750 lowdelay_subband(s, gb, quant, slice->slice_x, slice->slice_y, luma_end,
751 &s->plane[0].band[level][orientation], NULL);
752 }
753
754 /* consume any unused bits from luma */
755 skip_bits_long(gb, get_bits_count(gb) - luma_end);
756
757 chroma_bits = 8*slice->bytes - 7 - length_bits - luma_bits;
758 chroma_end = get_bits_count(gb) + FFMIN(chroma_bits, get_bits_left(gb));
759 /* [DIRAC_STD] 13.5.5.3 chroma_slice_band */
760 for (level = 0; level < s->wavelet_depth; level++)
761 for (orientation = !!level; orientation < 4; orientation++) {
762 quant = FFMAX(quant_base - s->lowdelay.quant[level][orientation], 0);
763 lowdelay_subband(s, gb, quant, slice->slice_x, slice->slice_y, chroma_end,
764 &s->plane[1].band[level][orientation],
765 &s->plane[2].band[level][orientation]);
766 }
767
768 return 0;
769 }
770
771 /**
772 * Dirac Specification ->
773 * 13.5.1 low_delay_transform_data()
774 */
775 static void decode_lowdelay(DiracContext *s)
776 {
777 AVCodecContext *avctx = s->avctx;
778 int slice_x, slice_y, bytes, bufsize;
779 const uint8_t *buf;
780 struct lowdelay_slice *slices;
781 int slice_num = 0;
782
783 slices = av_mallocz_array(s->lowdelay.num_x, s->lowdelay.num_y * sizeof(struct lowdelay_slice));
784
785 align_get_bits(&s->gb);
786 /*[DIRAC_STD] 13.5.2 Slices. slice(sx,sy) */
787 buf = s->gb.buffer + get_bits_count(&s->gb)/8;
788 bufsize = get_bits_left(&s->gb);
789
790 for (slice_y = 0; bufsize > 0 && slice_y < s->lowdelay.num_y; slice_y++)
791 for (slice_x = 0; bufsize > 0 && slice_x < s->lowdelay.num_x; slice_x++) {
792 bytes = (slice_num+1) * s->lowdelay.bytes.num / s->lowdelay.bytes.den
793 - slice_num * s->lowdelay.bytes.num / s->lowdelay.bytes.den;
794
795 slices[slice_num].bytes = bytes;
796 slices[slice_num].slice_x = slice_x;
797 slices[slice_num].slice_y = slice_y;
798 init_get_bits(&slices[slice_num].gb, buf, bufsize);
799 slice_num++;
800
801 buf += bytes;
802 bufsize -= bytes*8;
803 }
804
805 avctx->execute(avctx, decode_lowdelay_slice, slices, NULL, slice_num,
806 sizeof(struct lowdelay_slice)); /* [DIRAC_STD] 13.5.2 Slices */
807 intra_dc_prediction(&s->plane[0].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */
808 intra_dc_prediction(&s->plane[1].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */
809 intra_dc_prediction(&s->plane[2].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */
810 av_free(slices);
811 }
812
813 static void init_planes(DiracContext *s)
814 {
815 int i, w, h, level, orientation;
816
817 for (i = 0; i < 3; i++) {
818 Plane *p = &s->plane[i];
819
820 p->width = s->source.width >> (i ? s->chroma_x_shift : 0);
821 p->height = s->source.height >> (i ? s->chroma_y_shift : 0);
822 p->idwt_width = w = CALC_PADDING(p->width , s->wavelet_depth);
823 p->idwt_height = h = CALC_PADDING(p->height, s->wavelet_depth);
824 p->idwt_stride = FFALIGN(p->idwt_width, 8);
825
826 for (level = s->wavelet_depth-1; level >= 0; level--) {
827 w = w>>1;
828 h = h>>1;
829 for (orientation = !!level; orientation < 4; orientation++) {
830 SubBand *b = &p->band[level][orientation];
831
832 b->ibuf = p->idwt_buf;
833 b->level = level;
834 b->stride = p->idwt_stride << (s->wavelet_depth - level);
835 b->width = w;
836 b->height = h;
837 b->orientation = orientation;
838
839 if (orientation & 1)
840 b->ibuf += w;
841 if (orientation > 1)
842 b->ibuf += b->stride>>1;
843
844 if (level)
845 b->parent = &p->band[level-1][orientation];
846 }
847 }
848
849 if (i > 0) {
850 p->xblen = s->plane[0].xblen >> s->chroma_x_shift;
851 p->yblen = s->plane[0].yblen >> s->chroma_y_shift;
852 p->xbsep = s->plane[0].xbsep >> s->chroma_x_shift;
853 p->ybsep = s->plane[0].ybsep >> s->chroma_y_shift;
854 }
855
856 p->xoffset = (p->xblen - p->xbsep)/2;
857 p->yoffset = (p->yblen - p->ybsep)/2;
858 }
859 }
860
861 /**
862 * Unpack the motion compensation parameters
863 * Dirac Specification ->
864 * 11.2 Picture prediction data. picture_prediction()
865 */
866 static int dirac_unpack_prediction_parameters(DiracContext *s)
867 {
868 static const uint8_t default_blen[] = { 4, 12, 16, 24 };
869 static const uint8_t default_bsep[] = { 4, 8, 12, 16 };
870
871 GetBitContext *gb = &s->gb;
872 unsigned idx, ref;
873
874 align_get_bits(gb);
875 /* [DIRAC_STD] 11.2.2 Block parameters. block_parameters() */
876 /* Luma and Chroma are equal. 11.2.3 */
877 idx = svq3_get_ue_golomb(gb); /* [DIRAC_STD] index */
878
879 if (idx > 4) {
880 av_log(s->avctx, AV_LOG_ERROR, "Block prediction index too high\n");
881 return -1;
882 }
883
884 if (idx == 0) {
885 s->plane[0].xblen = svq3_get_ue_golomb(gb);
886 s->plane[0].yblen = svq3_get_ue_golomb(gb);
887 s->plane[0].xbsep = svq3_get_ue_golomb(gb);
888 s->plane[0].ybsep = svq3_get_ue_golomb(gb);
889 } else {
890 /*[DIRAC_STD] preset_block_params(index). Table 11.1 */
891 s->plane[0].xblen = default_blen[idx-1];
892 s->plane[0].yblen = default_blen[idx-1];
893 s->plane[0].xbsep = default_bsep[idx-1];
894 s->plane[0].ybsep = default_bsep[idx-1];
895 }
896 /*[DIRAC_STD] 11.2.4 motion_data_dimensions()
897 Calculated in function dirac_unpack_block_motion_data */
898
899 if (!s->plane[0].xbsep || !s->plane[0].ybsep || s->plane[0].xbsep < s->plane[0].xblen/2 || s->plane[0].ybsep < s->plane[0].yblen/2) {
900 av_log(s->avctx, AV_LOG_ERROR, "Block separation too small\n");
901 return -1;
902 }
903 if (s->plane[0].xbsep > s->plane[0].xblen || s->plane[0].ybsep > s->plane[0].yblen) {
904 av_log(s->avctx, AV_LOG_ERROR, "Block separation greater than size\n");
905 return -1;
906 }
907 if (FFMAX(s->plane[0].xblen, s->plane[0].yblen) > MAX_BLOCKSIZE) {
908 av_log(s->avctx, AV_LOG_ERROR, "Unsupported large block size\n");
909 return -1;
910 }
911
912 /*[DIRAC_STD] 11.2.5 Motion vector precision. motion_vector_precision()
913 Read motion vector precision */
914 s->mv_precision = svq3_get_ue_golomb(gb);
915 if (s->mv_precision > 3) {
916 av_log(s->avctx, AV_LOG_ERROR, "MV precision finer than eighth-pel\n");
917 return -1;
918 }
919
920 /*[DIRAC_STD] 11.2.6 Global motion. global_motion()
921 Read the global motion compensation parameters */
922 s->globalmc_flag = get_bits1(gb);
923 if (s->globalmc_flag) {
924 memset(s->globalmc, 0, sizeof(s->globalmc));
925 /* [DIRAC_STD] pan_tilt(gparams) */
926 for (ref = 0; ref < s->num_refs; ref++) {
927 if (get_bits1(gb)) {
928 s->globalmc[ref].pan_tilt[0] = dirac_get_se_golomb(gb);
929 s->globalmc[ref].pan_tilt[1] = dirac_get_se_golomb(gb);
930 }
931 /* [DIRAC_STD] zoom_rotate_shear(gparams)
932 zoom/rotation/shear parameters */
933 if (get_bits1(gb)) {
934 s->globalmc[ref].zrs_exp = svq3_get_ue_golomb(gb);
935 s->globalmc[ref].zrs[0][0] = dirac_get_se_golomb(gb);
936 s->globalmc[ref].zrs[0][1] = dirac_get_se_golomb(gb);
937 s->globalmc[ref].zrs[1][0] = dirac_get_se_golomb(gb);
938 s->globalmc[ref].zrs[1][1] = dirac_get_se_golomb(gb);
939 } else {
940 s->globalmc[ref].zrs[0][0] = 1;
941 s->globalmc[ref].zrs[1][1] = 1;
942 }
943 /* [DIRAC_STD] perspective(gparams) */
944 if (get_bits1(gb)) {
945 s->globalmc[ref].perspective_exp = svq3_get_ue_golomb(gb);
946 s->globalmc[ref].perspective[0] = dirac_get_se_golomb(gb);
947 s->globalmc[ref].perspective[1] = dirac_get_se_golomb(gb);
948 }
949 }
950 }
951
952 /*[DIRAC_STD] 11.2.7 Picture prediction mode. prediction_mode()
953 Picture prediction mode, not currently used. */
954 if (svq3_get_ue_golomb(gb)) {
955 av_log(s->avctx, AV_LOG_ERROR, "Unknown picture prediction mode\n");
956 return -1;
957 }
958
959 /* [DIRAC_STD] 11.2.8 Reference picture weight. reference_picture_weights()
960 just data read, weight calculation will be done later on. */
961 s->weight_log2denom = 1;
962 s->weight[0] = 1;
963 s->weight[1] = 1;
964
965 if (get_bits1(gb)) {
966 s->weight_log2denom = svq3_get_ue_golomb(gb);
967 s->weight[0] = dirac_get_se_golomb(gb);
968 if (s->num_refs == 2)
969 s->weight[1] = dirac_get_se_golomb(gb);
970 }
971 return 0;
972 }
973
974 /**
975 * Dirac Specification ->
976 * 11.3 Wavelet transform data. wavelet_transform()
977 */
978 static int dirac_unpack_idwt_params(DiracContext *s)
979 {
980 GetBitContext *gb = &s->gb;
981 int i, level;
982 unsigned tmp;
983
984 #define CHECKEDREAD(dst, cond, errmsg) \
985 tmp = svq3_get_ue_golomb(gb); \
986 if (cond) { \
987 av_log(s->avctx, AV_LOG_ERROR, errmsg); \
988 return -1; \
989 }\
990 dst = tmp;
991
992 align_get_bits(gb);
993
994 s->zero_res = s->num_refs ? get_bits1(gb) : 0;
995 if (s->zero_res)
996 return 0;
997
998 /*[DIRAC_STD] 11.3.1 Transform parameters. transform_parameters() */
999 CHECKEDREAD(s->wavelet_idx, tmp > 6, "wavelet_idx is too big\n")
1000
1001 CHECKEDREAD(s->wavelet_depth, tmp > MAX_DWT_LEVELS || tmp < 1, "invalid number of DWT decompositions\n")
1002
1003 if (!s->low_delay) {
1004 /* Codeblock parameters (core syntax only) */
1005 if (get_bits1(gb)) {
1006 for (i = 0; i <= s->wavelet_depth; i++) {
1007 CHECKEDREAD(s->codeblock[i].width , tmp < 1 || tmp > (s->avctx->width >>s->wavelet_depth-i), "codeblock width invalid\n")
1008 CHECKEDREAD(s->codeblock[i].height, tmp < 1 || tmp > (s->avctx->height>>s->wavelet_depth-i), "codeblock height invalid\n")
1009 }
1010
1011 CHECKEDREAD(s->codeblock_mode, tmp > 1, "unknown codeblock mode\n")
1012 } else
1013 for (i = 0; i <= s->wavelet_depth; i++)
1014 s->codeblock[i].width = s->codeblock[i].height = 1;
1015 } else {
1016 /* Slice parameters + quantization matrix*/
1017 /*[DIRAC_STD] 11.3.4 Slice coding Parameters (low delay syntax only). slice_parameters() */
1018 s->lowdelay.num_x = svq3_get_ue_golomb(gb);
1019 s->lowdelay.num_y = svq3_get_ue_golomb(gb);
1020 s->lowdelay.bytes.num = svq3_get_ue_golomb(gb);
1021 s->lowdelay.bytes.den = svq3_get_ue_golomb(gb);
1022
1023 if (s->lowdelay.bytes.den <= 0) {
1024 av_log(s->avctx,AV_LOG_ERROR,"Invalid lowdelay.bytes.den\n");
1025 return AVERROR_INVALIDDATA;
1026 }
1027
1028 /* [DIRAC_STD] 11.3.5 Quantisation matrices (low-delay syntax). quant_matrix() */
1029 if (get_bits1(gb)) {
1030 av_log(s->avctx,AV_LOG_DEBUG,"Low Delay: Has Custom Quantization Matrix!\n");
1031 /* custom quantization matrix */
1032 s->lowdelay.quant[0][0] = svq3_get_ue_golomb(gb);
1033 for (level = 0; level < s->wavelet_depth; level++) {
1034 s->lowdelay.quant[level][1] = svq3_get_ue_golomb(gb);
1035 s->lowdelay.quant[level][2] = svq3_get_ue_golomb(gb);
1036 s->lowdelay.quant[level][3] = svq3_get_ue_golomb(gb);
1037 }
1038 } else {
1039 if (s->wavelet_depth > 4) {
1040 av_log(s->avctx,AV_LOG_ERROR,"Mandatory custom low delay matrix missing for depth %d\n", s->wavelet_depth);
1041 return AVERROR_INVALIDDATA;
1042 }
1043 /* default quantization matrix */
1044 for (level = 0; level < s->wavelet_depth; level++)
1045 for (i = 0; i < 4; i++) {
1046 s->lowdelay.quant[level][i] = default_qmat[s->wavelet_idx][level][i];
1047 /* haar with no shift differs for different depths */
1048 if (s->wavelet_idx == 3)
1049 s->lowdelay.quant[level][i] += 4*(s->wavelet_depth-1 - level);
1050 }
1051 }
1052 }
1053 return 0;
1054 }
1055
1056 static inline int pred_sbsplit(uint8_t *sbsplit, int stride, int x, int y)
1057 {
1058 static const uint8_t avgsplit[7] = { 0, 0, 1, 1, 1, 2, 2 };
1059
1060 if (!(x|y))
1061 return 0;
1062 else if (!y)
1063 return sbsplit[-1];
1064 else if (!x)
1065 return sbsplit[-stride];
1066
1067 return avgsplit[sbsplit[-1] + sbsplit[-stride] + sbsplit[-stride-1]];
1068 }
1069
1070 static inline int pred_block_mode(DiracBlock *block, int stride, int x, int y, int refmask)
1071 {
1072 int pred;
1073
1074 if (!(x|y))
1075 return 0;
1076 else if (!y)
1077 return block[-1].ref & refmask;
1078 else if (!x)
1079 return block[-stride].ref & refmask;
1080
1081 /* return the majority */
1082 pred = (block[-1].ref & refmask) + (block[-stride].ref & refmask) + (block[-stride-1].ref & refmask);
1083 return (pred >> 1) & refmask;
1084 }
1085
1086 static inline void pred_block_dc(DiracBlock *block, int stride, int x, int y)
1087 {
1088 int i, n = 0;
1089
1090 memset(block->u.dc, 0, sizeof(block->u.dc));
1091
1092 if (x && !(block[-1].ref & 3)) {
1093 for (i = 0; i < 3; i++)
1094 block->u.dc[i] += block[-1].u.dc[i];
1095 n++;
1096 }
1097
1098 if (y && !(block[-stride].ref & 3)) {
1099 for (i = 0; i < 3; i++)
1100 block->u.dc[i] += block[-stride].u.dc[i];
1101 n++;
1102 }
1103
1104 if (x && y && !(block[-1-stride].ref & 3)) {
1105 for (i = 0; i < 3; i++)
1106 block->u.dc[i] += block[-1-stride].u.dc[i];
1107 n++;
1108 }
1109
1110 if (n == 2) {
1111 for (i = 0; i < 3; i++)
1112 block->u.dc[i] = (block->u.dc[i]+1)>>1;
1113 } else if (n == 3) {
1114 for (i = 0; i < 3; i++)
1115 block->u.dc[i] = divide3(block->u.dc[i]);
1116 }
1117 }
1118
1119 static inline void pred_mv(DiracBlock *block, int stride, int x, int y, int ref)
1120 {
1121 int16_t *pred[3];
1122 int refmask = ref+1;
1123 int mask = refmask | DIRAC_REF_MASK_GLOBAL; /* exclude gmc blocks */
1124 int n = 0;
1125
1126 if (x && (block[-1].ref & mask) == refmask)
1127 pred[n++] = block[-1].u.mv[ref];
1128
1129 if (y && (block[-stride].ref & mask) == refmask)
1130 pred[n++] = block[-stride].u.mv[ref];
1131
1132 if (x && y && (block[-stride-1].ref & mask) == refmask)
1133 pred[n++] = block[-stride-1].u.mv[ref];
1134
1135 switch (n) {
1136 case 0:
1137 block->u.mv[ref][0] = 0;
1138 block->u.mv[ref][1] = 0;
1139 break;
1140 case 1:
1141 block->u.mv[ref][0] = pred[0][0];
1142 block->u.mv[ref][1] = pred[0][1];
1143 break;
1144 case 2:
1145 block->u.mv[ref][0] = (pred[0][0] + pred[1][0] + 1) >> 1;
1146 block->u.mv[ref][1] = (pred[0][1] + pred[1][1] + 1) >> 1;
1147 break;
1148 case 3:
1149 block->u.mv[ref][0] = mid_pred(pred[0][0], pred[1][0], pred[2][0]);
1150 block->u.mv[ref][1] = mid_pred(pred[0][1], pred[1][1], pred[2][1]);
1151 break;
1152 }
1153 }
1154
1155 static void global_mv(DiracContext *s, DiracBlock *block, int x, int y, int ref)
1156 {
1157 int ez = s->globalmc[ref].zrs_exp;
1158 int ep = s->globalmc[ref].perspective_exp;
1159 int (*A)[2] = s->globalmc[ref].zrs;
1160 int *b = s->globalmc[ref].pan_tilt;
1161 int *c = s->globalmc[ref].perspective;
1162
1163 int m = (1<<ep) - (c[0]*x + c[1]*y);
1164 int mx = m * ((A[0][0] * x + A[0][1]*y) + (1<<ez) * b[0]);
1165 int my = m * ((A[1][0] * x + A[1][1]*y) + (1<<ez) * b[1]);
1166
1167 block->u.mv[ref][0] = (mx + (1<<(ez+ep))) >> (ez+ep);
1168 block->u.mv[ref][1] = (my + (1<<(ez+ep))) >> (ez+ep);
1169 }
1170
1171 static void decode_block_params(DiracContext *s, DiracArith arith[8], DiracBlock *block,
1172 int stride, int x, int y)
1173 {
1174 int i;
1175
1176 block->ref = pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_REF1);
1177 block->ref ^= dirac_get_arith_bit(arith, CTX_PMODE_REF1);
1178
1179 if (s->num_refs == 2) {
1180 block->ref |= pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_REF2);
1181 block->ref ^= dirac_get_arith_bit(arith, CTX_PMODE_REF2) << 1;
1182 }
1183
1184 if (!block->ref) {
1185 pred_block_dc(block, stride, x, y);
1186 for (i = 0; i < 3; i++)
1187 block->u.dc[i] += dirac_get_arith_int(arith+1+i, CTX_DC_F1, CTX_DC_DATA);
1188 return;
1189 }
1190
1191 if (s->globalmc_flag) {
1192 block->ref |= pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_GLOBAL);
1193 block->ref ^= dirac_get_arith_bit(arith, CTX_GLOBAL_BLOCK) << 2;
1194 }
1195
1196 for (i = 0; i < s->num_refs; i++)
1197 if (block->ref & (i+1)) {
1198 if (block->ref & DIRAC_REF_MASK_GLOBAL) {
1199 global_mv(s, block, x, y, i);
1200 } else {
1201 pred_mv(block, stride, x, y, i);
1202 block->u.mv[i][0] += dirac_get_arith_int(arith + 4 + 2 * i, CTX_MV_F1, CTX_MV_DATA);
1203 block->u.mv[i][1] += dirac_get_arith_int(arith + 5 + 2 * i, CTX_MV_F1, CTX_MV_DATA);
1204 }
1205 }
1206 }
1207
1208 /**
1209 * Copies the current block to the other blocks covered by the current superblock split mode
1210 */
1211 static void propagate_block_data(DiracBlock *block, int stride, int size)
1212 {
1213 int x, y;
1214 DiracBlock *dst = block;
1215
1216 for (x = 1; x < size; x++)
1217 dst[x] = *block;
1218
1219 for (y = 1; y < size; y++) {
1220 dst += stride;
1221 for (x = 0; x < size; x++)
1222 dst[x] = *block;
1223 }
1224 }
1225
1226 /**
1227 * Dirac Specification ->
1228 * 12. Block motion data syntax
1229 */
1230 static int dirac_unpack_block_motion_data(DiracContext *s)
1231 {
1232 GetBitContext *gb = &s->gb;
1233 uint8_t *sbsplit = s->sbsplit;
1234 int i, x, y, q, p;
1235 DiracArith arith[8];
1236
1237 align_get_bits(gb);
1238
1239 /* [DIRAC_STD] 11.2.4 and 12.2.1 Number of blocks and superblocks */
1240 s->sbwidth = DIVRNDUP(s->source.width, 4*s->plane[0].xbsep);
1241 s->sbheight = DIVRNDUP(s->source.height, 4*s->plane[0].ybsep);
1242 s->blwidth = 4 * s->sbwidth;
1243 s->blheight = 4 * s->sbheight;
1244
1245 /* [DIRAC_STD] 12.3.1 Superblock splitting modes. superblock_split_modes()
1246 decode superblock split modes */
1247 ff_dirac_init_arith_decoder(arith, gb, svq3_get_ue_golomb(gb)); /* svq3_get_ue_golomb(gb) is the length */
1248 for (y = 0; y < s->sbheight; y++) {
1249 for (x = 0; x < s->sbwidth; x++) {
1250 unsigned int split = dirac_get_arith_uint(arith, CTX_SB_F1, CTX_SB_DATA);
1251 if (split > 2)
1252 return -1;
1253 sbsplit[x] = (split + pred_sbsplit(sbsplit+x, s->sbwidth, x, y)) % 3;
1254 }
1255 sbsplit += s->sbwidth;
1256 }
1257
1258 /* setup arith decoding */
1259 ff_dirac_init_arith_decoder(arith, gb, svq3_get_ue_golomb(gb));
1260 for (i = 0; i < s->num_refs; i++) {
1261 ff_dirac_init_arith_decoder(arith + 4 + 2 * i, gb, svq3_get_ue_golomb(gb));
1262 ff_dirac_init_arith_decoder(arith + 5 + 2 * i, gb, svq3_get_ue_golomb(gb));
1263 }
1264 for (i = 0; i < 3; i++)
1265 ff_dirac_init_arith_decoder(arith+1+i, gb, svq3_get_ue_golomb(gb));
1266
1267 for (y = 0; y < s->sbheight; y++)
1268 for (x = 0; x < s->sbwidth; x++) {
1269 int blkcnt = 1 << s->sbsplit[y * s->sbwidth + x];
1270 int step = 4 >> s->sbsplit[y * s->sbwidth + x];
1271
1272 for (q = 0; q < blkcnt; q++)
1273 for (p = 0; p < blkcnt; p++) {
1274 int bx = 4 * x + p*step;
1275 int by = 4 * y + q*step;
1276 DiracBlock *block = &s->blmotion[by*s->blwidth + bx];
1277 decode_block_params(s, arith, block, s->blwidth, bx, by);
1278 propagate_block_data(block, s->blwidth, step);
1279 }
1280 }
1281
1282 return 0;
1283 }
1284
1285 static int weight(int i, int blen, int offset)
1286 {
1287 #define ROLLOFF(i) offset == 1 ? ((i) ? 5 : 3) : \
1288 (1 + (6*(i) + offset - 1) / (2*offset - 1))
1289
1290 if (i < 2*offset)
1291 return ROLLOFF(i);
1292 else if (i > blen-1 - 2*offset)
1293 return ROLLOFF(blen-1 - i);
1294 return 8;
1295 }
1296
1297 static void init_obmc_weight_row(Plane *p, uint8_t *obmc_weight, int stride,
1298 int left, int right, int wy)
1299 {
1300 int x;
1301 for (x = 0; left && x < p->xblen >> 1; x++)
1302 obmc_weight[x] = wy*8;
1303 for (; x < p->xblen >> right; x++)
1304 obmc_weight[x] = wy*weight(x, p->xblen, p->xoffset);
1305 for (; x < p->xblen; x++)
1306 obmc_weight[x] = wy*8;
1307 for (; x < stride; x++)
1308 obmc_weight[x] = 0;
1309 }
1310
1311 static void init_obmc_weight(Plane *p, uint8_t *obmc_weight, int stride,
1312 int left, int right, int top, int bottom)
1313 {
1314 int y;
1315 for (y = 0; top && y < p->yblen >> 1; y++) {
1316 init_obmc_weight_row(p, obmc_weight, stride, left, right, 8);
1317 obmc_weight += stride;
1318 }
1319 for (; y < p->yblen >> bottom; y++) {
1320 int wy = weight(y, p->yblen, p->yoffset);
1321 init_obmc_weight_row(p, obmc_weight, stride, left, right, wy);
1322 obmc_weight += stride;
1323 }
1324 for (; y < p->yblen; y++) {
1325 init_obmc_weight_row(p, obmc_weight, stride, left, right, 8);
1326 obmc_weight += stride;
1327 }
1328 }
1329
1330 static void init_obmc_weights(DiracContext *s, Plane *p, int by)
1331 {
1332 int top = !by;
1333 int bottom = by == s->blheight-1;
1334
1335 /* don't bother re-initing for rows 2 to blheight-2, the weights don't change */
1336 if (top || bottom || by == 1) {
1337 init_obmc_weight(p, s->obmc_weight[0], MAX_BLOCKSIZE, 1, 0, top, bottom);
1338 init_obmc_weight(p, s->obmc_weight[1], MAX_BLOCKSIZE, 0, 0, top, bottom);
1339 init_obmc_weight(p, s->obmc_weight[2], MAX_BLOCKSIZE, 0, 1, top, bottom);
1340 }
1341 }
1342
1343 static const uint8_t epel_weights[4][4][4] = {
1344 {{ 16, 0, 0, 0 },
1345 { 12, 4, 0, 0 },
1346 { 8, 8, 0, 0 },
1347 { 4, 12, 0, 0 }},
1348 {{ 12, 0, 4, 0 },
1349 { 9, 3, 3, 1 },
1350 { 6, 6, 2, 2 },
1351 { 3, 9, 1, 3 }},
1352 {{ 8, 0, 8, 0 },
1353 { 6, 2, 6, 2 },
1354 { 4, 4, 4, 4 },
1355 { 2, 6, 2, 6 }},
1356 {{ 4, 0, 12, 0 },
1357 { 3, 1, 9, 3 },
1358 { 2, 2, 6, 6 },
1359 { 1, 3, 3, 9 }}
1360 };
1361
1362 /**
1363 * For block x,y, determine which of the hpel planes to do bilinear
1364 * interpolation from and set src[] to the location in each hpel plane
1365 * to MC from.
1366 *
1367 * @return the index of the put_dirac_pixels_tab function to use
1368 * 0 for 1 plane (fpel,hpel), 1 for 2 planes (qpel), 2 for 4 planes (qpel), and 3 for epel
1369 */
1370 static int mc_subpel(DiracContext *s, DiracBlock *block, const uint8_t *src[5],
1371 int x, int y, int ref, int plane)
1372 {
1373 Plane *p = &s->plane[plane];
1374 uint8_t **ref_hpel = s->ref_pics[ref]->hpel[plane];
1375 int motion_x = block->u.mv[ref][0];
1376 int motion_y = block->u.mv[ref][1];
1377 int mx, my, i, epel, nplanes = 0;
1378
1379 if (plane) {
1380 motion_x >>= s->chroma_x_shift;
1381 motion_y >>= s->chroma_y_shift;
1382 }
1383
1384 mx = motion_x & ~(-1U << s->mv_precision);
1385 my = motion_y & ~(-1U << s->mv_precision);
1386 motion_x >>= s->mv_precision;
1387 motion_y >>= s->mv_precision;
1388 /* normalize subpel coordinates to epel */
1389 /* TODO: template this function? */
1390 mx <<= 3 - s->mv_precision;
1391 my <<= 3 - s->mv_precision;
1392
1393 x += motion_x;
1394 y += motion_y;
1395 epel = (mx|my)&1;
1396
1397 /* hpel position */
1398 if (!((mx|my)&3)) {
1399 nplanes = 1;
1400 src[0] = ref_hpel[(my>>1)+(mx>>2)] + y*p->stride + x;
1401 } else {
1402 /* qpel or epel */
1403 nplanes = 4;
1404 for (i = 0; i < 4; i++)
1405 src[i] = ref_hpel[i] + y*p->stride + x;
1406
1407 /* if we're interpolating in the right/bottom halves, adjust the planes as needed
1408 we increment x/y because the edge changes for half of the pixels */
1409 if (mx > 4) {
1410 src[0] += 1;
1411 src[2] += 1;
1412 x++;
1413 }
1414 if (my > 4) {
1415 src[0] += p->stride;
1416 src[1] += p->stride;
1417 y++;
1418 }
1419
1420 /* hpel planes are:
1421 [0]: F [1]: H
1422 [2]: V [3]: C */
1423 if (!epel) {
1424 /* check if we really only need 2 planes since either mx or my is
1425 a hpel position. (epel weights of 0 handle this there) */
1426 if (!(mx&3)) {
1427 /* mx == 0: average [0] and [2]
1428 mx == 4: average [1] and [3] */
1429 src[!mx] = src[2 + !!mx];
1430 nplanes = 2;
1431 } else if (!(my&3)) {
1432 src[0] = src[(my>>1) ];
1433 src[1] = src[(my>>1)+1];
1434 nplanes = 2;
1435 }
1436 } else {
1437 /* adjust the ordering if needed so the weights work */
1438 if (mx > 4) {
1439 FFSWAP(const uint8_t *, src[0], src[1]);
1440 FFSWAP(const uint8_t *, src[2], src[3]);
1441 }
1442 if (my > 4) {
1443 FFSWAP(const uint8_t *, src[0], src[2]);
1444 FFSWAP(const uint8_t *, src[1], src[3]);
1445 }
1446 src[4] = epel_weights[my&3][mx&3];
1447 }
1448 }
1449
1450 /* fixme: v/h _edge_pos */
1451 if (x + p->xblen > p->width +EDGE_WIDTH/2 ||
1452 y + p->yblen > p->height+EDGE_WIDTH/2 ||
1453 x < 0 || y < 0) {
1454 for (i = 0; i < nplanes; i++) {
1455 s->vdsp.emulated_edge_mc(s->edge_emu_buffer[i], src[i],
1456 p->stride, p->stride,
1457 p->xblen, p->yblen, x, y,
1458 p->width+EDGE_WIDTH/2, p->height+EDGE_WIDTH/2);
1459 src[i] = s->edge_emu_buffer[i];
1460 }
1461 }
1462 return (nplanes>>1) + epel;
1463 }
1464
1465 static void add_dc(uint16_t *dst, int dc, int stride,
1466 uint8_t *obmc_weight, int xblen, int yblen)
1467 {
1468 int x, y;
1469 dc += 128;
1470
1471 for (y = 0; y < yblen; y++) {
1472 for (x = 0; x < xblen; x += 2) {
1473 dst[x ] += dc * obmc_weight[x ];
1474 dst[x+1] += dc * obmc_weight[x+1];
1475 }
1476 dst += stride;
1477 obmc_weight += MAX_BLOCKSIZE;
1478 }
1479 }
1480
1481 static void block_mc(DiracContext *s, DiracBlock *block,
1482 uint16_t *mctmp, uint8_t *obmc_weight,
1483 int plane, int dstx, int dsty)
1484 {
1485 Plane *p = &s->plane[plane];
1486 const uint8_t *src[5];
1487 int idx;
1488
1489 switch (block->ref&3) {
1490 case 0: /* DC */
1491 add_dc(mctmp, block->u.dc[plane], p->stride, obmc_weight, p->xblen, p->yblen);
1492 return;
1493 case 1:
1494 case 2:
1495 idx = mc_subpel(s, block, src, dstx, dsty, (block->ref&3)-1, plane);
1496 s->put_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);
1497 if (s->weight_func)
1498 s->weight_func(s->mcscratch, p->stride, s->weight_log2denom,
1499 s->weight[0] + s->weight[1], p->yblen);
1500 break;
1501 case 3:
1502 idx = mc_subpel(s, block, src, dstx, dsty, 0, plane);
1503 s->put_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);
1504 idx = mc_subpel(s, block, src, dstx, dsty, 1, plane);
1505 if (s->biweight_func) {
1506 /* fixme: +32 is a quick hack */
1507 s->put_pixels_tab[idx](s->mcscratch + 32, src, p->stride, p->yblen);
1508 s->biweight_func(s->mcscratch, s->mcscratch+32, p->stride, s->weight_log2denom,
1509 s->weight[0], s->weight[1], p->yblen);
1510 } else
1511 s->avg_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);
1512 break;
1513 }
1514 s->add_obmc(mctmp, s->mcscratch, p->stride, obmc_weight, p->yblen);
1515 }
1516
1517 static void mc_row(DiracContext *s, DiracBlock *block, uint16_t *mctmp, int plane, int dsty)
1518 {
1519 Plane *p = &s->plane[plane];
1520 int x, dstx = p->xbsep - p->xoffset;
1521
1522 block_mc(s, block, mctmp, s->obmc_weight[0], plane, -p->xoffset, dsty);
1523 mctmp += p->xbsep;
1524
1525 for (x = 1; x < s->blwidth-1; x++) {
1526 block_mc(s, block+x, mctmp, s->obmc_weight[1], plane, dstx, dsty);
1527 dstx += p->xbsep;
1528 mctmp += p->xbsep;
1529 }
1530 block_mc(s, block+x, mctmp, s->obmc_weight[2], plane, dstx, dsty);
1531 }
1532
1533 static void select_dsp_funcs(DiracContext *s, int width, int height, int xblen, int yblen)
1534 {
1535 int idx = 0;
1536 if (xblen > 8)
1537 idx = 1;
1538 if (xblen > 16)
1539 idx = 2;
1540
1541 memcpy(s->put_pixels_tab, s->diracdsp.put_dirac_pixels_tab[idx], sizeof(s->put_pixels_tab));
1542 memcpy(s->avg_pixels_tab, s->diracdsp.avg_dirac_pixels_tab[idx], sizeof(s->avg_pixels_tab));
1543 s->add_obmc = s->diracdsp.add_dirac_obmc[idx];
1544 if (s->weight_log2denom > 1 || s->weight[0] != 1 || s->weight[1] != 1) {
1545 s->weight_func = s->diracdsp.weight_dirac_pixels_tab[idx];
1546 s->biweight_func = s->diracdsp.biweight_dirac_pixels_tab[idx];
1547 } else {
1548 s->weight_func = NULL;
1549 s->biweight_func = NULL;
1550 }
1551 }
1552
1553 static void interpolate_refplane(DiracContext *s, DiracFrame *ref, int plane, int width, int height)
1554 {
1555 /* chroma allocates an edge of 8 when subsampled
1556 which for 4:2:2 means an h edge of 16 and v edge of 8
1557 just use 8 for everything for the moment */
1558 int i, edge = EDGE_WIDTH/2;
1559
1560 ref->hpel[plane][0] = ref->avframe->data[plane];
1561 s->mpvencdsp.draw_edges(ref->hpel[plane][0], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM); /* EDGE_TOP | EDGE_BOTTOM values just copied to make it build, this needs to be ensured */
1562
1563 /* no need for hpel if we only have fpel vectors */
1564 if (!s->mv_precision)
1565 return;
1566
1567 for (i = 1; i < 4; i++) {
1568 if (!ref->hpel_base[plane][i])
1569 ref->hpel_base[plane][i] = av_malloc((height+2*edge) * ref->avframe->linesize[plane] + 32);
1570 /* we need to be 16-byte aligned even for chroma */
1571 ref->hpel[plane][i] = ref->hpel_base[plane][i] + edge*ref->avframe->linesize[plane] + 16;
1572 }
1573
1574 if (!ref->interpolated[plane]) {
1575 s->diracdsp.dirac_hpel_filter(ref->hpel[plane][1], ref->hpel[plane][2],
1576 ref->hpel[plane][3], ref->hpel[plane][0],
1577 ref->avframe->linesize[plane], width, height);
1578 s->mpvencdsp.draw_edges(ref->hpel[plane][1], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
1579 s->mpvencdsp.draw_edges(ref->hpel[plane][2], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
1580 s->mpvencdsp.draw_edges(ref->hpel[plane][3], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
1581 }
1582 ref->interpolated[plane] = 1;
1583 }
1584
1585 /**
1586 * Dirac Specification ->
1587 * 13.0 Transform data syntax. transform_data()
1588 */
1589 static int dirac_decode_frame_internal(DiracContext *s)
1590 {
1591 DWTContext d;
1592 int y, i, comp, dsty;
1593
1594 if (s->low_delay) {
1595 /* [DIRAC_STD] 13.5.1 low_delay_transform_data() */
1596 for (comp = 0; comp < 3; comp++) {
1597 Plane *p = &s->plane[comp];
1598 memset(p->idwt_buf, 0, p->idwt_stride * p->idwt_height * sizeof(IDWTELEM));
1599 }
1600 if (!s->zero_res)
1601 decode_lowdelay(s);
1602 }
1603
1604 for (comp = 0; comp < 3; comp++) {
1605 Plane *p = &s->plane[comp];
1606 uint8_t *frame = s->current_picture->avframe->data[comp];
1607
1608 /* FIXME: small resolutions */
1609 for (i = 0; i < 4; i++)
1610 s->edge_emu_buffer[i] = s->edge_emu_buffer_base + i*FFALIGN(p->width, 16);
1611
1612 if (!s->zero_res && !s->low_delay)
1613 {
1614 memset(p->idwt_buf, 0, p->idwt_stride * p->idwt_height * sizeof(IDWTELEM));
1615 decode_component(s, comp); /* [DIRAC_STD] 13.4.1 core_transform_data() */
1616 }
1617 if (ff_spatial_idwt_init2(&d, p->idwt_buf, p->idwt_width, p->idwt_height, p->idwt_stride,
1618 s->wavelet_idx+2, s->wavelet_depth, p->idwt_tmp))
1619 return -1;
1620
1621 if (!s->num_refs) { /* intra */
1622 for (y = 0; y < p->height; y += 16) {
1623 ff_spatial_idwt_slice2(&d, y+16); /* decode */
1624 s->diracdsp.put_signed_rect_clamped(frame + y*p->stride, p->stride,
1625 p->idwt_buf + y*p->idwt_stride, p->idwt_stride, p->width, 16);
1626 }
1627 } else { /* inter */
1628 int rowheight = p->ybsep*p->stride;
1629
1630 select_dsp_funcs(s, p->width, p->height, p->xblen, p->yblen);
1631
1632 for (i = 0; i < s->num_refs; i++)
1633 interpolate_refplane(s, s->ref_pics[i], comp, p->width, p->height);
1634
1635 memset(s->mctmp, 0, 4*p->yoffset*p->stride);
1636
1637 dsty = -p->yoffset;
1638 for (y = 0; y < s->blheight; y++) {
1639 int h = 0,
1640 start = FFMAX(dsty, 0);
1641 uint16_t *mctmp = s->mctmp + y*rowheight;
1642 DiracBlock *blocks = s->blmotion + y*s->blwidth;
1643
1644 init_obmc_weights(s, p, y);
1645
1646 if (y == s->blheight-1 || start+p->ybsep > p->height)
1647 h = p->height - start;
1648 else
1649 h = p->ybsep - (start - dsty);
1650 if (h < 0)
1651 break;
1652
1653 memset(mctmp+2*p->yoffset*p->stride, 0, 2*rowheight);
1654 mc_row(s, blocks, mctmp, comp, dsty);
1655
1656 mctmp += (start - dsty)*p->stride + p->xoffset;
1657 ff_spatial_idwt_slice2(&d, start + h); /* decode */
1658 s->diracdsp.add_rect_clamped(frame + start*p->stride, mctmp, p->stride,
1659 p->idwt_buf + start*p->idwt_stride, p->idwt_stride, p->width, h);
1660
1661 dsty += p->ybsep;
1662 }
1663 }
1664 }
1665
1666
1667 return 0;
1668 }
1669
1670 static int get_buffer_with_edge(AVCodecContext *avctx, AVFrame *f, int flags)
1671 {
1672 int ret, i;
1673 int chroma_x_shift, chroma_y_shift;
1674 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_x_shift, &chroma_y_shift);
1675
1676 f->width = avctx->width + 2 * EDGE_WIDTH;
1677 f->height = avctx->height + 2 * EDGE_WIDTH + 2;
1678 ret = ff_get_buffer(avctx, f, flags);
1679 if (ret < 0)
1680 return ret;
1681
1682 for (i = 0; f->data[i]; i++) {
1683 int offset = (EDGE_WIDTH >> (i && i<3 ? chroma_y_shift : 0)) *
1684 f->linesize[i] + 32;
1685 f->data[i] += offset;
1686 }
1687 f->width = avctx->width;
1688 f->height = avctx->height;
1689
1690 return 0;
1691 }
1692
1693 /**
1694 * Dirac Specification ->
1695 * 11.1.1 Picture Header. picture_header()
1696 */
1697 static int dirac_decode_picture_header(DiracContext *s)
1698 {
1699 int retire, picnum;
1700 int i, j, refnum, refdist;
1701 GetBitContext *gb = &s->gb;
1702
1703 /* [DIRAC_STD] 11.1.1 Picture Header. picture_header() PICTURE_NUM */
1704 picnum = s->current_picture->avframe->display_picture_number = get_bits_long(gb, 32);
1705
1706
1707 av_log(s->avctx,AV_LOG_DEBUG,"PICTURE_NUM: %d\n",picnum);
1708
1709 /* if this is the first keyframe after a sequence header, start our
1710 reordering from here */
1711 if (s->frame_number < 0)
1712 s->frame_number = picnum;
1713
1714 s->ref_pics[0] = s->ref_pics[1] = NULL;
1715 for (i = 0; i < s->num_refs; i++) {
1716 refnum = picnum + dirac_get_se_golomb(gb);
1717 refdist = INT_MAX;
1718
1719 /* find the closest reference to the one we want */
1720 /* Jordi: this is needed if the referenced picture hasn't yet arrived */
1721 for (j = 0; j < MAX_REFERENCE_FRAMES && refdist; j++)
1722 if (s->ref_frames[j]
1723 && FFABS(s->ref_frames[j]->avframe->display_picture_number - refnum) < refdist) {
1724 s->ref_pics[i] = s->ref_frames[j];
1725 refdist = FFABS(s->ref_frames[j]->avframe->display_picture_number - refnum);
1726 }
1727
1728 if (!s->ref_pics[i] || refdist)
1729 av_log(s->avctx, AV_LOG_DEBUG, "Reference not found\n");
1730
1731 /* if there were no references at all, allocate one */
1732 if (!s->ref_pics[i])
1733 for (j = 0; j < MAX_FRAMES; j++)
1734 if (!s->all_frames[j].avframe->data[0]) {
1735 s->ref_pics[i] = &s->all_frames[j];
1736 get_buffer_with_edge(s->avctx, s->ref_pics[i]->avframe, AV_GET_BUFFER_FLAG_REF);
1737 break;
1738 }
1739 }
1740
1741 /* retire the reference frames that are not used anymore */
1742 if (s->current_picture->avframe->reference) {
1743 retire = picnum + dirac_get_se_golomb(gb);
1744 if (retire != picnum) {
1745 DiracFrame *retire_pic = remove_frame(s->ref_frames, retire);
1746
1747 if (retire_pic)
1748 retire_pic->avframe->reference &= DELAYED_PIC_REF;
1749 else
1750 av_log(s->avctx, AV_LOG_DEBUG, "Frame to retire not found\n");
1751 }
1752
1753 /* if reference array is full, remove the oldest as per the spec */
1754 while (add_frame(s->ref_frames, MAX_REFERENCE_FRAMES, s->current_picture)) {
1755 av_log(s->avctx, AV_LOG_ERROR, "Reference frame overflow\n");
1756 remove_frame(s->ref_frames, s->ref_frames[0]->avframe->display_picture_number)->avframe->reference &= DELAYED_PIC_REF;
1757 }
1758 }
1759
1760 if (s->num_refs) {
1761 if (dirac_unpack_prediction_parameters(s)) /* [DIRAC_STD] 11.2 Picture Prediction Data. picture_prediction() */
1762 return -1;
1763 if (dirac_unpack_block_motion_data(s)) /* [DIRAC_STD] 12. Block motion data syntax */
1764 return -1;
1765 }
1766 if (dirac_unpack_idwt_params(s)) /* [DIRAC_STD] 11.3 Wavelet transform data */
1767 return -1;
1768
1769 init_planes(s);
1770 return 0;
1771 }
1772
1773 static int get_delayed_pic(DiracContext *s, AVFrame *picture, int *got_frame)
1774 {
1775 DiracFrame *out = s->delay_frames[0];
1776 int i, out_idx = 0;
1777 int ret;
1778
1779 /* find frame with lowest picture number */
1780 for (i = 1; s->delay_frames[i]; i++)
1781 if (s->delay_frames[i]->avframe->display_picture_number < out->avframe->display_picture_number) {
1782 out = s->delay_frames[i];
1783 out_idx = i;
1784 }
1785
1786 for (i = out_idx; s->delay_frames[i]; i++)
1787 s->delay_frames[i] = s->delay_frames[i+1];
1788
1789 if (out) {
1790 out->avframe->reference ^= DELAYED_PIC_REF;
1791 *got_frame = 1;
1792 if((ret = av_frame_ref(picture, out->avframe)) < 0)
1793 return ret;
1794 }
1795
1796 return 0;
1797 }
1798
1799 /**
1800 * Dirac Specification ->
1801 * 9.6 Parse Info Header Syntax. parse_info()
1802 * 4 byte start code + byte parse code + 4 byte size + 4 byte previous size
1803 */
1804 #define DATA_UNIT_HEADER_SIZE 13
1805
1806 /* [DIRAC_STD] dirac_decode_data_unit makes reference to the while defined in 9.3
1807 inside the function parse_sequence() */
1808 static int dirac_decode_data_unit(AVCodecContext *avctx, const uint8_t *buf, int size)
1809 {
1810 DiracContext *s = avctx->priv_data;
1811 DiracFrame *pic = NULL;
1812 int ret, i, parse_code = buf[4];
1813 unsigned tmp;
1814
1815 if (size < DATA_UNIT_HEADER_SIZE)
1816 return -1;
1817
1818 init_get_bits(&s->gb, &buf[13], 8*(size - DATA_UNIT_HEADER_SIZE));
1819
1820 if (parse_code == pc_seq_header) {
1821 if (s->seen_sequence_header)
1822 return 0;
1823
1824 /* [DIRAC_STD] 10. Sequence header */
1825 if (avpriv_dirac_parse_sequence_header(avctx, &s->gb, &s->source))
1826 return -1;
1827
1828 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1829
1830 if (alloc_sequence_buffers(s))
1831 return -1;
1832
1833 s->seen_sequence_header = 1;
1834 } else if (parse_code == pc_eos) { /* [DIRAC_STD] End of Sequence */
1835 free_sequence_buffers(s);
1836 s->seen_sequence_header = 0;
1837 } else if (parse_code == pc_aux_data) {
1838 if (buf[13] == 1) { /* encoder implementation/version */
1839 int ver[3];
1840 /* versions older than 1.0.8 don't store quant delta for
1841 subbands with only one codeblock */
1842 if (sscanf(buf+14, "Schroedinger %d.%d.%d", ver, ver+1, ver+2) == 3)
1843 if (ver[0] == 1 && ver[1] == 0 && ver[2] <= 7)
1844 s->old_delta_quant = 1;
1845 }
1846 } else if (parse_code & 0x8) { /* picture data unit */
1847 if (!s->seen_sequence_header) {
1848 av_log(avctx, AV_LOG_DEBUG, "Dropping frame without sequence header\n");
1849 return -1;
1850 }
1851
1852 /* find an unused frame */
1853 for (i = 0; i < MAX_FRAMES; i++)
1854 if (s->all_frames[i].avframe->data[0] == NULL)
1855 pic = &s->all_frames[i];
1856 if (!pic) {
1857 av_log(avctx, AV_LOG_ERROR, "framelist full\n");
1858 return -1;
1859 }
1860
1861 av_frame_unref(pic->avframe);
1862
1863 /* [DIRAC_STD] Defined in 9.6.1 ... */
1864 tmp = parse_code & 0x03; /* [DIRAC_STD] num_refs() */
1865 if (tmp > 2) {
1866 av_log(avctx, AV_LOG_ERROR, "num_refs of 3\n");
1867 return -1;
1868 }
1869 s->num_refs = tmp;
1870 s->is_arith = (parse_code & 0x48) == 0x08; /* [DIRAC_STD] using_ac() */
1871 s->low_delay = (parse_code & 0x88) == 0x88; /* [DIRAC_STD] is_low_delay() */
1872 pic->avframe->reference = (parse_code & 0x0C) == 0x0C; /* [DIRAC_STD] is_reference() */
1873 pic->avframe->key_frame = s->num_refs == 0; /* [DIRAC_STD] is_intra() */
1874 pic->avframe->pict_type = s->num_refs + 1; /* Definition of AVPictureType in avutil.h */
1875
1876 if ((ret = get_buffer_with_edge(avctx, pic->avframe, (parse_code & 0x0C) == 0x0C ? AV_GET_BUFFER_FLAG_REF : 0)) < 0)
1877 return ret;
1878 s->current_picture = pic;
1879 s->plane[0].stride = pic->avframe->linesize[0];
1880 s->plane[1].stride = pic->avframe->linesize[1];
1881 s->plane[2].stride = pic->avframe->linesize[2];
1882
1883 if (alloc_buffers(s, FFMAX3(FFABS(s->plane[0].stride), FFABS(s->plane[1].stride), FFABS(s->plane[2].stride))) < 0)
1884 return AVERROR(ENOMEM);
1885
1886 /* [DIRAC_STD] 11.1 Picture parse. picture_parse() */
1887 if (dirac_decode_picture_header(s))
1888 return -1;
1889
1890 /* [DIRAC_STD] 13.0 Transform data syntax. transform_data() */
1891 if (dirac_decode_frame_internal(s))
1892 return -1;
1893 }
1894 return 0;
1895 }
1896
1897 static int dirac_decode_frame(AVCodecContext *avctx, void *data, int *got_frame, AVPacket *pkt)
1898 {
1899 DiracContext *s = avctx->priv_data;
1900 AVFrame *picture = data;
1901 uint8_t *buf = pkt->data;
1902 int buf_size = pkt->size;
1903 int i, data_unit_size, buf_idx = 0;
1904 int ret;
1905
1906 /* release unused frames */
1907 for (i = 0; i < MAX_FRAMES; i++)
1908 if (s->all_frames[i].avframe->data[0] && !s->all_frames[i].avframe->reference) {
1909 av_frame_unref(s->all_frames[i].avframe);
1910 memset(s->all_frames[i].interpolated, 0, sizeof(s->all_frames[i].interpolated));
1911 }
1912
1913 s->current_picture = NULL;
1914 *got_frame = 0;
1915
1916 /* end of stream, so flush delayed pics */
1917 if (buf_size == 0)
1918 return get_delayed_pic(s, (AVFrame *)data, got_frame);
1919
1920 for (;;) {
1921 /*[DIRAC_STD] Here starts the code from parse_info() defined in 9.6
1922 [DIRAC_STD] PARSE_INFO_PREFIX = "BBCD" as defined in ISO/IEC 646
1923 BBCD start code search */
1924 for (; buf_idx + DATA_UNIT_HEADER_SIZE < buf_size; buf_idx++) {
1925 if (buf[buf_idx ] == 'B' && buf[buf_idx+1] == 'B' &&
1926 buf[buf_idx+2] == 'C' && buf[buf_idx+3] == 'D')
1927 break;
1928 }
1929 /* BBCD found or end of data */
1930 if (buf_idx + DATA_UNIT_HEADER_SIZE >= buf_size)
1931 break;
1932
1933 data_unit_size = AV_RB32(buf+buf_idx+5);
1934 if (buf_idx + data_unit_size > buf_size || !data_unit_size) {
1935 if(buf_idx + data_unit_size > buf_size)
1936 av_log(s->avctx, AV_LOG_ERROR,
1937 "Data unit with size %d is larger than input buffer, discarding\n",
1938 data_unit_size);
1939 buf_idx += 4;
1940 continue;
1941 }
1942 /* [DIRAC_STD] dirac_decode_data_unit makes reference to the while defined in 9.3 inside the function parse_sequence() */
1943 if (dirac_decode_data_unit(avctx, buf+buf_idx, data_unit_size))
1944 {
1945 av_log(s->avctx, AV_LOG_ERROR,"Error in dirac_decode_data_unit\n");
1946 return -1;
1947 }
1948 buf_idx += data_unit_size;
1949 }
1950
1951 if (!s->current_picture)
1952 return buf_size;
1953
1954 if (s->current_picture->avframe->display_picture_number > s->frame_number) {
1955 DiracFrame *delayed_frame = remove_frame(s->delay_frames, s->frame_number);
1956
1957 s->current_picture->avframe->reference |= DELAYED_PIC_REF;
1958
1959 if (add_frame(s->delay_frames, MAX_DELAY, s->current_picture)) {
1960 int min_num = s->delay_frames[0]->avframe->display_picture_number;
1961 /* Too many delayed frames, so we display the frame with the lowest pts */
1962 av_log(avctx, AV_LOG_ERROR, "Delay frame overflow\n");
1963
1964 for (i = 1; s->delay_frames[i]; i++)
1965 if (s->delay_frames[i]->avframe->display_picture_number < min_num)
1966 min_num = s->delay_frames[i]->avframe->display_picture_number;
1967
1968 delayed_frame = remove_frame(s->delay_frames, min_num);
1969 add_frame(s->delay_frames, MAX_DELAY, s->current_picture);
1970 }
1971
1972 if (delayed_frame) {
1973 delayed_frame->avframe->reference ^= DELAYED_PIC_REF;
1974 if((ret=av_frame_ref(data, delayed_frame->avframe)) < 0)
1975 return ret;
1976 *got_frame = 1;
1977 }
1978 } else if (s->current_picture->avframe->display_picture_number == s->frame_number) {
1979 /* The right frame at the right time :-) */
1980 if((ret=av_frame_ref(data, s->current_picture->avframe)) < 0)
1981 return ret;
1982 *got_frame = 1;
1983 }
1984
1985 if (*got_frame)
1986 s->frame_number = picture->display_picture_number + 1;
1987
1988 return buf_idx;
1989 }
1990
1991 AVCodec ff_dirac_decoder = {
1992 .name = "dirac",
1993 .long_name = NULL_IF_CONFIG_SMALL("BBC Dirac VC-2"),
1994 .type = AVMEDIA_TYPE_VIDEO,
1995 .id = AV_CODEC_ID_DIRAC,
1996 .priv_data_size = sizeof(DiracContext),
1997 .init = dirac_decode_init,
1998 .close = dirac_decode_end,
1999 .decode = dirac_decode_frame,
2000 .capabilities = CODEC_CAP_DELAY,
2001 .flush = dirac_decode_flush,
2002 };