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 "libavutil/mem_internal.h"
30 #include "libavutil/pixdesc.h"
31 #include "libavutil/thread.h"
32 #include "avcodec.h"
33 #include "get_bits.h"
34 #include "bytestream.h"
35 #include "internal.h"
36 #include "golomb.h"
37 #include "dirac_arith.h"
38 #include "dirac_vlc.h"
39 #include "mpeg12data.h"
40 #include "libavcodec/mpegvideo.h"
41 #include "mpegvideoencdsp.h"
42 #include "dirac_dwt.h"
43 #include "dirac.h"
44 #include "diractab.h"
45 #include "diracdsp.h"
46 #include "videodsp.h"
47
48 /**
49 * The spec limits this to 3 for frame coding, but in practice can be as high as 6
50 */
51 #define MAX_REFERENCE_FRAMES 8
52 #define MAX_DELAY 5 /* limit for main profile for frame coding (TODO: field coding) */
53 #define MAX_FRAMES (MAX_REFERENCE_FRAMES + MAX_DELAY + 1)
54 #define MAX_QUANT 255 /* max quant for VC-2 */
55 #define MAX_BLOCKSIZE 32 /* maximum xblen/yblen we support */
56
57 /**
58 * DiracBlock->ref flags, if set then the block does MC from the given ref
59 */
60 #define DIRAC_REF_MASK_REF1 1
61 #define DIRAC_REF_MASK_REF2 2
62 #define DIRAC_REF_MASK_GLOBAL 4
63
64 /**
65 * Value of Picture.reference when Picture is not a reference picture, but
66 * is held for delayed output.
67 */
68 #define DELAYED_PIC_REF 4
69
70 #define CALC_PADDING(size, depth) \
71 (((size + (1 << depth) - 1) >> depth) << depth)
72
73 #define DIVRNDUP(a, b) (((a) + (b) - 1) / (b))
74
75 typedef struct {
76 AVFrame *avframe;
77 int interpolated[3]; /* 1 if hpel[] is valid */
78 uint8_t *hpel[3][4];
79 uint8_t *hpel_base[3][4];
80 int reference;
81 } DiracFrame;
82
83 typedef struct {
84 union {
85 int16_t mv[2][2];
86 int16_t dc[3];
87 } u; /* anonymous unions aren't in C99 :( */
88 uint8_t ref;
89 } DiracBlock;
90
91 typedef struct SubBand {
92 int level;
93 int orientation;
94 int stride; /* in bytes */
95 int width;
96 int height;
97 int pshift;
98 int quant;
99 uint8_t *ibuf;
100 struct SubBand *parent;
101
102 /* for low delay */
103 unsigned length;
104 const uint8_t *coeff_data;
105 } SubBand;
106
107 typedef struct Plane {
108 DWTPlane idwt;
109
110 int width;
111 int height;
112 ptrdiff_t stride;
113
114 /* block length */
115 uint8_t xblen;
116 uint8_t yblen;
117 /* block separation (block n+1 starts after this many pixels in block n) */
118 uint8_t xbsep;
119 uint8_t ybsep;
120 /* amount of overspill on each edge (half of the overlap between blocks) */
121 uint8_t xoffset;
122 uint8_t yoffset;
123
124 SubBand band[MAX_DWT_LEVELS][4];
125 } Plane;
126
127 /* Used by Low Delay and High Quality profiles */
128 typedef struct DiracSlice {
129 GetBitContext gb;
130 int slice_x;
131 int slice_y;
132 int bytes;
133 } DiracSlice;
134
135 typedef struct DiracContext {
136 AVCodecContext *avctx;
137 MpegvideoEncDSPContext mpvencdsp;
138 VideoDSPContext vdsp;
139 DiracDSPContext diracdsp;
140 DiracVersionInfo version;
141 GetBitContext gb;
142 AVDiracSeqHeader seq;
143 int seen_sequence_header;
144 int64_t 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 bit_depth; /* bit depth */
150 int pshift; /* pixel shift = bit_depth > 8 */
151
152 int zero_res; /* zero residue flag */
153 int is_arith; /* whether coeffs use arith or golomb coding */
154 int core_syntax; /* use core syntax only */
155 int low_delay; /* use the low delay syntax */
156 int hq_picture; /* high quality picture, enables low_delay */
157 int ld_picture; /* use low delay picture, turns on low_delay */
158 int dc_prediction; /* has dc prediction */
159 int globalmc_flag; /* use global motion compensation */
160 int num_refs; /* number of reference pictures */
161
162 /* wavelet decoding */
163 unsigned wavelet_depth; /* depth of the IDWT */
164 unsigned wavelet_idx;
165
166 /**
167 * schroedinger older than 1.0.8 doesn't store
168 * quant delta if only one codebook exists in a band
169 */
170 unsigned old_delta_quant;
171 unsigned codeblock_mode;
172
173 unsigned num_x; /* number of horizontal slices */
174 unsigned num_y; /* number of vertical slices */
175
176 uint8_t *thread_buf; /* Per-thread buffer for coefficient storage */
177 int threads_num_buf; /* Current # of buffers allocated */
178 int thread_buf_size; /* Each thread has a buffer this size */
179
180 DiracSlice *slice_params_buf;
181 int slice_params_num_buf;
182
183 struct {
184 unsigned width;
185 unsigned height;
186 } codeblock[MAX_DWT_LEVELS+1];
187
188 struct {
189 AVRational bytes; /* average bytes per slice */
190 uint8_t quant[MAX_DWT_LEVELS][4]; /* [DIRAC_STD] E.1 */
191 } lowdelay;
192
193 struct {
194 unsigned prefix_bytes;
195 uint64_t size_scaler;
196 } highquality;
197
198 struct {
199 int pan_tilt[2]; /* pan/tilt vector */
200 int zrs[2][2]; /* zoom/rotate/shear matrix */
201 int perspective[2]; /* perspective vector */
202 unsigned zrs_exp;
203 unsigned perspective_exp;
204 } globalmc[2];
205
206 /* motion compensation */
207 uint8_t mv_precision; /* [DIRAC_STD] REFS_WT_PRECISION */
208 int16_t weight[2]; /* [DIRAC_STD] REF1_WT and REF2_WT */
209 unsigned weight_log2denom; /* [DIRAC_STD] REFS_WT_PRECISION */
210
211 int blwidth; /* number of blocks (horizontally) */
212 int blheight; /* number of blocks (vertically) */
213 int sbwidth; /* number of superblocks (horizontally) */
214 int sbheight; /* number of superblocks (vertically) */
215
216 uint8_t *sbsplit;
217 DiracBlock *blmotion;
218
219 uint8_t *edge_emu_buffer[4];
220 uint8_t *edge_emu_buffer_base;
221
222 uint16_t *mctmp; /* buffer holding the MC data multiplied by OBMC weights */
223 uint8_t *mcscratch;
224 int buffer_stride;
225
226 DECLARE_ALIGNED(16, uint8_t, obmc_weight)[3][MAX_BLOCKSIZE*MAX_BLOCKSIZE];
227
228 void (*put_pixels_tab[4])(uint8_t *dst, const uint8_t *src[5], int stride, int h);
229 void (*avg_pixels_tab[4])(uint8_t *dst, const uint8_t *src[5], int stride, int h);
230 void (*add_obmc)(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
231 dirac_weight_func weight_func;
232 dirac_biweight_func biweight_func;
233
234 DiracFrame *current_picture;
235 DiracFrame *ref_pics[2];
236
237 DiracFrame *ref_frames[MAX_REFERENCE_FRAMES+1];
238 DiracFrame *delay_frames[MAX_DELAY+1];
239 DiracFrame all_frames[MAX_FRAMES];
240 } DiracContext;
241
242 enum dirac_subband {
243 subband_ll = 0,
244 subband_hl = 1,
245 subband_lh = 2,
246 subband_hh = 3,
247 subband_nb,
248 };
249
250 /* magic number division by 3 from schroedinger */
divide3(int x)251 static inline int divide3(int x)
252 {
253 return (int)((x+1U)*21845 + 10922) >> 16;
254 }
255
remove_frame(DiracFrame * framelist[],int picnum)256 static DiracFrame *remove_frame(DiracFrame *framelist[], int picnum)
257 {
258 DiracFrame *remove_pic = NULL;
259 int i, remove_idx = -1;
260
261 for (i = 0; framelist[i]; i++)
262 if (framelist[i]->avframe->display_picture_number == picnum) {
263 remove_pic = framelist[i];
264 remove_idx = i;
265 }
266
267 if (remove_pic)
268 for (i = remove_idx; framelist[i]; i++)
269 framelist[i] = framelist[i+1];
270
271 return remove_pic;
272 }
273
add_frame(DiracFrame * framelist[],int maxframes,DiracFrame * frame)274 static int add_frame(DiracFrame *framelist[], int maxframes, DiracFrame *frame)
275 {
276 int i;
277 for (i = 0; i < maxframes; i++)
278 if (!framelist[i]) {
279 framelist[i] = frame;
280 return 0;
281 }
282 return -1;
283 }
284
alloc_sequence_buffers(DiracContext * s)285 static int alloc_sequence_buffers(DiracContext *s)
286 {
287 int sbwidth = DIVRNDUP(s->seq.width, 4);
288 int sbheight = DIVRNDUP(s->seq.height, 4);
289 int i, w, h, top_padding;
290
291 /* todo: think more about this / use or set Plane here */
292 for (i = 0; i < 3; i++) {
293 int max_xblen = MAX_BLOCKSIZE >> (i ? s->chroma_x_shift : 0);
294 int max_yblen = MAX_BLOCKSIZE >> (i ? s->chroma_y_shift : 0);
295 w = s->seq.width >> (i ? s->chroma_x_shift : 0);
296 h = s->seq.height >> (i ? s->chroma_y_shift : 0);
297
298 /* we allocate the max we support here since num decompositions can
299 * change from frame to frame. Stride is aligned to 16 for SIMD, and
300 * 1<<MAX_DWT_LEVELS top padding to avoid if(y>0) in arith decoding
301 * MAX_BLOCKSIZE padding for MC: blocks can spill up to half of that
302 * on each side */
303 top_padding = FFMAX(1<<MAX_DWT_LEVELS, max_yblen/2);
304 w = FFALIGN(CALC_PADDING(w, MAX_DWT_LEVELS), 8); /* FIXME: Should this be 16 for SSE??? */
305 h = top_padding + CALC_PADDING(h, MAX_DWT_LEVELS) + max_yblen/2;
306
307 s->plane[i].idwt.buf_base = av_mallocz_array((w+max_xblen), h * (2 << s->pshift));
308 s->plane[i].idwt.tmp = av_malloc_array((w+16), 2 << s->pshift);
309 s->plane[i].idwt.buf = s->plane[i].idwt.buf_base + (top_padding*w)*(2 << s->pshift);
310 if (!s->plane[i].idwt.buf_base || !s->plane[i].idwt.tmp)
311 return AVERROR(ENOMEM);
312 }
313
314 /* fixme: allocate using real stride here */
315 s->sbsplit = av_malloc_array(sbwidth, sbheight);
316 s->blmotion = av_malloc_array(sbwidth, sbheight * 16 * sizeof(*s->blmotion));
317
318 if (!s->sbsplit || !s->blmotion)
319 return AVERROR(ENOMEM);
320 return 0;
321 }
322
alloc_buffers(DiracContext * s,int stride)323 static int alloc_buffers(DiracContext *s, int stride)
324 {
325 int w = s->seq.width;
326 int h = s->seq.height;
327
328 av_assert0(stride >= w);
329 stride += 64;
330
331 if (s->buffer_stride >= stride)
332 return 0;
333 s->buffer_stride = 0;
334
335 av_freep(&s->edge_emu_buffer_base);
336 memset(s->edge_emu_buffer, 0, sizeof(s->edge_emu_buffer));
337 av_freep(&s->mctmp);
338 av_freep(&s->mcscratch);
339
340 s->edge_emu_buffer_base = av_malloc_array(stride, MAX_BLOCKSIZE);
341
342 s->mctmp = av_malloc_array((stride+MAX_BLOCKSIZE), (h+MAX_BLOCKSIZE) * sizeof(*s->mctmp));
343 s->mcscratch = av_malloc_array(stride, MAX_BLOCKSIZE);
344
345 if (!s->edge_emu_buffer_base || !s->mctmp || !s->mcscratch)
346 return AVERROR(ENOMEM);
347
348 s->buffer_stride = stride;
349 return 0;
350 }
351
free_sequence_buffers(DiracContext * s)352 static void free_sequence_buffers(DiracContext *s)
353 {
354 int i, j, k;
355
356 for (i = 0; i < MAX_FRAMES; i++) {
357 if (s->all_frames[i].avframe->data[0]) {
358 av_frame_unref(s->all_frames[i].avframe);
359 memset(s->all_frames[i].interpolated, 0, sizeof(s->all_frames[i].interpolated));
360 }
361
362 for (j = 0; j < 3; j++)
363 for (k = 1; k < 4; k++)
364 av_freep(&s->all_frames[i].hpel_base[j][k]);
365 }
366
367 memset(s->ref_frames, 0, sizeof(s->ref_frames));
368 memset(s->delay_frames, 0, sizeof(s->delay_frames));
369
370 for (i = 0; i < 3; i++) {
371 av_freep(&s->plane[i].idwt.buf_base);
372 av_freep(&s->plane[i].idwt.tmp);
373 }
374
375 s->buffer_stride = 0;
376 av_freep(&s->sbsplit);
377 av_freep(&s->blmotion);
378 av_freep(&s->edge_emu_buffer_base);
379
380 av_freep(&s->mctmp);
381 av_freep(&s->mcscratch);
382 }
383
384 static AVOnce dirac_arith_init = AV_ONCE_INIT;
385
dirac_decode_init(AVCodecContext * avctx)386 static av_cold int dirac_decode_init(AVCodecContext *avctx)
387 {
388 DiracContext *s = avctx->priv_data;
389 int i, ret;
390
391 s->avctx = avctx;
392 s->frame_number = -1;
393
394 s->thread_buf = NULL;
395 s->threads_num_buf = -1;
396 s->thread_buf_size = -1;
397
398 ff_diracdsp_init(&s->diracdsp);
399 ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx);
400 ff_videodsp_init(&s->vdsp, 8);
401
402 for (i = 0; i < MAX_FRAMES; i++) {
403 s->all_frames[i].avframe = av_frame_alloc();
404 if (!s->all_frames[i].avframe) {
405 while (i > 0)
406 av_frame_free(&s->all_frames[--i].avframe);
407 return AVERROR(ENOMEM);
408 }
409 }
410 ret = ff_thread_once(&dirac_arith_init, ff_dirac_init_arith_tables);
411 if (ret != 0)
412 return AVERROR_UNKNOWN;
413
414 return 0;
415 }
416
dirac_decode_flush(AVCodecContext * avctx)417 static void dirac_decode_flush(AVCodecContext *avctx)
418 {
419 DiracContext *s = avctx->priv_data;
420 free_sequence_buffers(s);
421 s->seen_sequence_header = 0;
422 s->frame_number = -1;
423 }
424
dirac_decode_end(AVCodecContext * avctx)425 static av_cold int dirac_decode_end(AVCodecContext *avctx)
426 {
427 DiracContext *s = avctx->priv_data;
428 int i;
429
430 dirac_decode_flush(avctx);
431 for (i = 0; i < MAX_FRAMES; i++)
432 av_frame_free(&s->all_frames[i].avframe);
433
434 av_freep(&s->thread_buf);
435 av_freep(&s->slice_params_buf);
436
437 return 0;
438 }
439
coeff_unpack_golomb(GetBitContext * gb,int qfactor,int qoffset)440 static inline int coeff_unpack_golomb(GetBitContext *gb, int qfactor, int qoffset)
441 {
442 int coeff = dirac_get_se_golomb(gb);
443 const unsigned sign = FFSIGN(coeff);
444 if (coeff)
445 coeff = sign*((sign * coeff * qfactor + qoffset) >> 2);
446 return coeff;
447 }
448
449 #define SIGN_CTX(x) (CTX_SIGN_ZERO + ((x) > 0) - ((x) < 0))
450
451 #define UNPACK_ARITH(n, type) \
452 static inline void coeff_unpack_arith_##n(DiracArith *c, int qfactor, int qoffset, \
453 SubBand *b, type *buf, int x, int y) \
454 { \
455 int sign, sign_pred = 0, pred_ctx = CTX_ZPZN_F1; \
456 unsigned coeff; \
457 const int mstride = -(b->stride >> (1+b->pshift)); \
458 if (b->parent) { \
459 const type *pbuf = (type *)b->parent->ibuf; \
460 const int stride = b->parent->stride >> (1+b->parent->pshift); \
461 pred_ctx += !!pbuf[stride * (y>>1) + (x>>1)] << 1; \
462 } \
463 if (b->orientation == subband_hl) \
464 sign_pred = buf[mstride]; \
465 if (x) { \
466 pred_ctx += !(buf[-1] | buf[mstride] | buf[-1 + mstride]); \
467 if (b->orientation == subband_lh) \
468 sign_pred = buf[-1]; \
469 } else { \
470 pred_ctx += !buf[mstride]; \
471 } \
472 coeff = dirac_get_arith_uint(c, pred_ctx, CTX_COEFF_DATA); \
473 if (coeff) { \
474 coeff = (coeff * qfactor + qoffset) >> 2; \
475 sign = dirac_get_arith_bit(c, SIGN_CTX(sign_pred)); \
476 coeff = (coeff ^ -sign) + sign; \
477 } \
478 *buf = coeff; \
479 } \
480
481 UNPACK_ARITH(8, int16_t)
482 UNPACK_ARITH(10, int32_t)
483
484 /**
485 * Decode the coeffs in the rectangle defined by left, right, top, bottom
486 * [DIRAC_STD] 13.4.3.2 Codeblock unpacking loop. codeblock()
487 */
codeblock(DiracContext * s,SubBand * b,GetBitContext * gb,DiracArith * c,int left,int right,int top,int bottom,int blockcnt_one,int is_arith)488 static inline int codeblock(DiracContext *s, SubBand *b,
489 GetBitContext *gb, DiracArith *c,
490 int left, int right, int top, int bottom,
491 int blockcnt_one, int is_arith)
492 {
493 int x, y, zero_block;
494 int qoffset, qfactor;
495 uint8_t *buf;
496
497 /* check for any coded coefficients in this codeblock */
498 if (!blockcnt_one) {
499 if (is_arith)
500 zero_block = dirac_get_arith_bit(c, CTX_ZERO_BLOCK);
501 else
502 zero_block = get_bits1(gb);
503
504 if (zero_block)
505 return 0;
506 }
507
508 if (s->codeblock_mode && !(s->old_delta_quant && blockcnt_one)) {
509 int quant;
510 if (is_arith)
511 quant = dirac_get_arith_int(c, CTX_DELTA_Q_F, CTX_DELTA_Q_DATA);
512 else
513 quant = dirac_get_se_golomb(gb);
514 if (quant > INT_MAX - b->quant || b->quant + quant < 0) {
515 av_log(s->avctx, AV_LOG_ERROR, "Invalid quant\n");
516 return AVERROR_INVALIDDATA;
517 }
518 b->quant += quant;
519 }
520
521 if (b->quant > (DIRAC_MAX_QUANT_INDEX - 1)) {
522 av_log(s->avctx, AV_LOG_ERROR, "Unsupported quant %d\n", b->quant);
523 b->quant = 0;
524 return AVERROR_INVALIDDATA;
525 }
526
527 qfactor = ff_dirac_qscale_tab[b->quant];
528 /* TODO: context pointer? */
529 if (!s->num_refs)
530 qoffset = ff_dirac_qoffset_intra_tab[b->quant] + 2;
531 else
532 qoffset = ff_dirac_qoffset_inter_tab[b->quant] + 2;
533
534 buf = b->ibuf + top * b->stride;
535 if (is_arith) {
536 for (y = top; y < bottom; y++) {
537 if (c->error)
538 return c->error;
539 for (x = left; x < right; x++) {
540 if (b->pshift) {
541 coeff_unpack_arith_10(c, qfactor, qoffset, b, (int32_t*)(buf)+x, x, y);
542 } else {
543 coeff_unpack_arith_8(c, qfactor, qoffset, b, (int16_t*)(buf)+x, x, y);
544 }
545 }
546 buf += b->stride;
547 }
548 } else {
549 for (y = top; y < bottom; y++) {
550 if (get_bits_left(gb) < 1)
551 return AVERROR_INVALIDDATA;
552 for (x = left; x < right; x++) {
553 int val = coeff_unpack_golomb(gb, qfactor, qoffset);
554 if (b->pshift) {
555 AV_WN32(&buf[4*x], val);
556 } else {
557 AV_WN16(&buf[2*x], val);
558 }
559 }
560 buf += b->stride;
561 }
562 }
563 return 0;
564 }
565
566 /**
567 * Dirac Specification ->
568 * 13.3 intra_dc_prediction(band)
569 */
570 #define INTRA_DC_PRED(n, type) \
571 static inline void intra_dc_prediction_##n(SubBand *b) \
572 { \
573 type *buf = (type*)b->ibuf; \
574 int x, y; \
575 \
576 for (x = 1; x < b->width; x++) \
577 buf[x] += buf[x-1]; \
578 buf += (b->stride >> (1+b->pshift)); \
579 \
580 for (y = 1; y < b->height; y++) { \
581 buf[0] += buf[-(b->stride >> (1+b->pshift))]; \
582 \
583 for (x = 1; x < b->width; x++) { \
584 int pred = buf[x - 1] + buf[x - (b->stride >> (1+b->pshift))] + buf[x - (b->stride >> (1+b->pshift))-1]; \
585 buf[x] += divide3(pred); \
586 } \
587 buf += (b->stride >> (1+b->pshift)); \
588 } \
589 } \
590
591 INTRA_DC_PRED(8, int16_t)
592 INTRA_DC_PRED(10, uint32_t)
593
594 /**
595 * Dirac Specification ->
596 * 13.4.2 Non-skipped subbands. subband_coeffs()
597 */
decode_subband_internal(DiracContext * s,SubBand * b,int is_arith)598 static av_always_inline int decode_subband_internal(DiracContext *s, SubBand *b, int is_arith)
599 {
600 int cb_x, cb_y, left, right, top, bottom;
601 DiracArith c;
602 GetBitContext gb;
603 int cb_width = s->codeblock[b->level + (b->orientation != subband_ll)].width;
604 int cb_height = s->codeblock[b->level + (b->orientation != subband_ll)].height;
605 int blockcnt_one = (cb_width + cb_height) == 2;
606 int ret;
607
608 if (!b->length)
609 return 0;
610
611 init_get_bits8(&gb, b->coeff_data, b->length);
612
613 if (is_arith)
614 ff_dirac_init_arith_decoder(&c, &gb, b->length);
615
616 top = 0;
617 for (cb_y = 0; cb_y < cb_height; cb_y++) {
618 bottom = (b->height * (cb_y+1LL)) / cb_height;
619 left = 0;
620 for (cb_x = 0; cb_x < cb_width; cb_x++) {
621 right = (b->width * (cb_x+1LL)) / cb_width;
622 ret = codeblock(s, b, &gb, &c, left, right, top, bottom, blockcnt_one, is_arith);
623 if (ret < 0)
624 return ret;
625 left = right;
626 }
627 top = bottom;
628 }
629
630 if (b->orientation == subband_ll && s->num_refs == 0) {
631 if (s->pshift) {
632 intra_dc_prediction_10(b);
633 } else {
634 intra_dc_prediction_8(b);
635 }
636 }
637 return 0;
638 }
639
decode_subband_arith(AVCodecContext * avctx,void * b)640 static int decode_subband_arith(AVCodecContext *avctx, void *b)
641 {
642 DiracContext *s = avctx->priv_data;
643 return decode_subband_internal(s, b, 1);
644 }
645
decode_subband_golomb(AVCodecContext * avctx,void * arg)646 static int decode_subband_golomb(AVCodecContext *avctx, void *arg)
647 {
648 DiracContext *s = avctx->priv_data;
649 SubBand **b = arg;
650 return decode_subband_internal(s, *b, 0);
651 }
652
653 /**
654 * Dirac Specification ->
655 * [DIRAC_STD] 13.4.1 core_transform_data()
656 */
decode_component(DiracContext * s,int comp)657 static int decode_component(DiracContext *s, int comp)
658 {
659 AVCodecContext *avctx = s->avctx;
660 SubBand *bands[3*MAX_DWT_LEVELS+1];
661 enum dirac_subband orientation;
662 int level, num_bands = 0;
663 int ret[3*MAX_DWT_LEVELS+1];
664 int i;
665 int damaged_count = 0;
666
667 /* Unpack all subbands at all levels. */
668 for (level = 0; level < s->wavelet_depth; level++) {
669 for (orientation = !!level; orientation < 4; orientation++) {
670 SubBand *b = &s->plane[comp].band[level][orientation];
671 bands[num_bands++] = b;
672
673 align_get_bits(&s->gb);
674 /* [DIRAC_STD] 13.4.2 subband() */
675 b->length = get_interleaved_ue_golomb(&s->gb);
676 if (b->length) {
677 b->quant = get_interleaved_ue_golomb(&s->gb);
678 if (b->quant > (DIRAC_MAX_QUANT_INDEX - 1)) {
679 av_log(s->avctx, AV_LOG_ERROR, "Unsupported quant %d\n", b->quant);
680 b->quant = 0;
681 return AVERROR_INVALIDDATA;
682 }
683 align_get_bits(&s->gb);
684 b->coeff_data = s->gb.buffer + get_bits_count(&s->gb)/8;
685 if (b->length > FFMAX(get_bits_left(&s->gb)/8, 0)) {
686 b->length = FFMAX(get_bits_left(&s->gb)/8, 0);
687 damaged_count ++;
688 }
689 skip_bits_long(&s->gb, b->length*8);
690 }
691 }
692 /* arithmetic coding has inter-level dependencies, so we can only execute one level at a time */
693 if (s->is_arith)
694 avctx->execute(avctx, decode_subband_arith, &s->plane[comp].band[level][!!level],
695 ret + 3*level + !!level, 4-!!level, sizeof(SubBand));
696 }
697 /* golomb coding has no inter-level dependencies, so we can execute all subbands in parallel */
698 if (!s->is_arith)
699 avctx->execute(avctx, decode_subband_golomb, bands, ret, num_bands, sizeof(SubBand*));
700
701 for (i = 0; i < s->wavelet_depth * 3 + 1; i++) {
702 if (ret[i] < 0)
703 damaged_count++;
704 }
705 if (damaged_count > (s->wavelet_depth * 3 + 1) /2)
706 return AVERROR_INVALIDDATA;
707
708 return 0;
709 }
710
711 #define PARSE_VALUES(type, x, gb, ebits, buf1, buf2) \
712 type *buf = (type *)buf1; \
713 buf[x] = coeff_unpack_golomb(gb, qfactor, qoffset); \
714 if (get_bits_count(gb) >= ebits) \
715 return; \
716 if (buf2) { \
717 buf = (type *)buf2; \
718 buf[x] = coeff_unpack_golomb(gb, qfactor, qoffset); \
719 if (get_bits_count(gb) >= ebits) \
720 return; \
721 } \
722
decode_subband(DiracContext * s,GetBitContext * gb,int quant,int slice_x,int slice_y,int bits_end,SubBand * b1,SubBand * b2)723 static void decode_subband(DiracContext *s, GetBitContext *gb, int quant,
724 int slice_x, int slice_y, int bits_end,
725 SubBand *b1, SubBand *b2)
726 {
727 int left = b1->width * slice_x / s->num_x;
728 int right = b1->width *(slice_x+1) / s->num_x;
729 int top = b1->height * slice_y / s->num_y;
730 int bottom = b1->height *(slice_y+1) / s->num_y;
731
732 int qfactor, qoffset;
733
734 uint8_t *buf1 = b1->ibuf + top * b1->stride;
735 uint8_t *buf2 = b2 ? b2->ibuf + top * b2->stride: NULL;
736 int x, y;
737
738 if (quant > (DIRAC_MAX_QUANT_INDEX - 1)) {
739 av_log(s->avctx, AV_LOG_ERROR, "Unsupported quant %d\n", quant);
740 return;
741 }
742 qfactor = ff_dirac_qscale_tab[quant];
743 qoffset = ff_dirac_qoffset_intra_tab[quant] + 2;
744 /* we have to constantly check for overread since the spec explicitly
745 requires this, with the meaning that all remaining coeffs are set to 0 */
746 if (get_bits_count(gb) >= bits_end)
747 return;
748
749 if (s->pshift) {
750 for (y = top; y < bottom; y++) {
751 for (x = left; x < right; x++) {
752 PARSE_VALUES(int32_t, x, gb, bits_end, buf1, buf2);
753 }
754 buf1 += b1->stride;
755 if (buf2)
756 buf2 += b2->stride;
757 }
758 }
759 else {
760 for (y = top; y < bottom; y++) {
761 for (x = left; x < right; x++) {
762 PARSE_VALUES(int16_t, x, gb, bits_end, buf1, buf2);
763 }
764 buf1 += b1->stride;
765 if (buf2)
766 buf2 += b2->stride;
767 }
768 }
769 }
770
771 /**
772 * Dirac Specification ->
773 * 13.5.2 Slices. slice(sx,sy)
774 */
decode_lowdelay_slice(AVCodecContext * avctx,void * arg)775 static int decode_lowdelay_slice(AVCodecContext *avctx, void *arg)
776 {
777 DiracContext *s = avctx->priv_data;
778 DiracSlice *slice = arg;
779 GetBitContext *gb = &slice->gb;
780 enum dirac_subband orientation;
781 int level, quant, chroma_bits, chroma_end;
782
783 int quant_base = get_bits(gb, 7); /*[DIRAC_STD] qindex */
784 int length_bits = av_log2(8 * slice->bytes)+1;
785 int luma_bits = get_bits_long(gb, length_bits);
786 int luma_end = get_bits_count(gb) + FFMIN(luma_bits, get_bits_left(gb));
787
788 /* [DIRAC_STD] 13.5.5.2 luma_slice_band */
789 for (level = 0; level < s->wavelet_depth; level++)
790 for (orientation = !!level; orientation < 4; orientation++) {
791 quant = FFMAX(quant_base - s->lowdelay.quant[level][orientation], 0);
792 decode_subband(s, gb, quant, slice->slice_x, slice->slice_y, luma_end,
793 &s->plane[0].band[level][orientation], NULL);
794 }
795
796 /* consume any unused bits from luma */
797 skip_bits_long(gb, get_bits_count(gb) - luma_end);
798
799 chroma_bits = 8*slice->bytes - 7 - length_bits - luma_bits;
800 chroma_end = get_bits_count(gb) + FFMIN(chroma_bits, get_bits_left(gb));
801 /* [DIRAC_STD] 13.5.5.3 chroma_slice_band */
802 for (level = 0; level < s->wavelet_depth; level++)
803 for (orientation = !!level; orientation < 4; orientation++) {
804 quant = FFMAX(quant_base - s->lowdelay.quant[level][orientation], 0);
805 decode_subband(s, gb, quant, slice->slice_x, slice->slice_y, chroma_end,
806 &s->plane[1].band[level][orientation],
807 &s->plane[2].band[level][orientation]);
808 }
809
810 return 0;
811 }
812
813 typedef struct SliceCoeffs {
814 int left;
815 int top;
816 int tot_h;
817 int tot_v;
818 int tot;
819 } SliceCoeffs;
820
subband_coeffs(DiracContext * s,int x,int y,int p,SliceCoeffs c[MAX_DWT_LEVELS])821 static int subband_coeffs(DiracContext *s, int x, int y, int p,
822 SliceCoeffs c[MAX_DWT_LEVELS])
823 {
824 int level, coef = 0;
825 for (level = 0; level < s->wavelet_depth; level++) {
826 SliceCoeffs *o = &c[level];
827 SubBand *b = &s->plane[p].band[level][3]; /* orientation doens't matter */
828 o->top = b->height * y / s->num_y;
829 o->left = b->width * x / s->num_x;
830 o->tot_h = ((b->width * (x + 1)) / s->num_x) - o->left;
831 o->tot_v = ((b->height * (y + 1)) / s->num_y) - o->top;
832 o->tot = o->tot_h*o->tot_v;
833 coef += o->tot * (4 - !!level);
834 }
835 return coef;
836 }
837
838 /**
839 * VC-2 Specification ->
840 * 13.5.3 hq_slice(sx,sy)
841 */
decode_hq_slice(DiracContext * s,DiracSlice * slice,uint8_t * tmp_buf)842 static int decode_hq_slice(DiracContext *s, DiracSlice *slice, uint8_t *tmp_buf)
843 {
844 int i, level, orientation, quant_idx;
845 int qfactor[MAX_DWT_LEVELS][4], qoffset[MAX_DWT_LEVELS][4];
846 GetBitContext *gb = &slice->gb;
847 SliceCoeffs coeffs_num[MAX_DWT_LEVELS];
848
849 skip_bits_long(gb, 8*s->highquality.prefix_bytes);
850 quant_idx = get_bits(gb, 8);
851
852 if (quant_idx > DIRAC_MAX_QUANT_INDEX - 1) {
853 av_log(s->avctx, AV_LOG_ERROR, "Invalid quantization index - %i\n", quant_idx);
854 return AVERROR_INVALIDDATA;
855 }
856
857 /* Slice quantization (slice_quantizers() in the specs) */
858 for (level = 0; level < s->wavelet_depth; level++) {
859 for (orientation = !!level; orientation < 4; orientation++) {
860 const int quant = FFMAX(quant_idx - s->lowdelay.quant[level][orientation], 0);
861 qfactor[level][orientation] = ff_dirac_qscale_tab[quant];
862 qoffset[level][orientation] = ff_dirac_qoffset_intra_tab[quant] + 2;
863 }
864 }
865
866 /* Luma + 2 Chroma planes */
867 for (i = 0; i < 3; i++) {
868 int coef_num, coef_par, off = 0;
869 int64_t length = s->highquality.size_scaler*get_bits(gb, 8);
870 int64_t bits_end = get_bits_count(gb) + 8*length;
871 const uint8_t *addr = align_get_bits(gb);
872
873 if (length*8 > get_bits_left(gb)) {
874 av_log(s->avctx, AV_LOG_ERROR, "end too far away\n");
875 return AVERROR_INVALIDDATA;
876 }
877
878 coef_num = subband_coeffs(s, slice->slice_x, slice->slice_y, i, coeffs_num);
879
880 if (s->pshift)
881 coef_par = ff_dirac_golomb_read_32bit(addr, length,
882 tmp_buf, coef_num);
883 else
884 coef_par = ff_dirac_golomb_read_16bit(addr, length,
885 tmp_buf, coef_num);
886
887 if (coef_num > coef_par) {
888 const int start_b = coef_par * (1 << (s->pshift + 1));
889 const int end_b = coef_num * (1 << (s->pshift + 1));
890 memset(&tmp_buf[start_b], 0, end_b - start_b);
891 }
892
893 for (level = 0; level < s->wavelet_depth; level++) {
894 const SliceCoeffs *c = &coeffs_num[level];
895 for (orientation = !!level; orientation < 4; orientation++) {
896 const SubBand *b1 = &s->plane[i].band[level][orientation];
897 uint8_t *buf = b1->ibuf + c->top * b1->stride + (c->left << (s->pshift + 1));
898
899 /* Change to c->tot_h <= 4 for AVX2 dequantization */
900 const int qfunc = s->pshift + 2*(c->tot_h <= 2);
901 s->diracdsp.dequant_subband[qfunc](&tmp_buf[off], buf, b1->stride,
902 qfactor[level][orientation],
903 qoffset[level][orientation],
904 c->tot_v, c->tot_h);
905
906 off += c->tot << (s->pshift + 1);
907 }
908 }
909
910 skip_bits_long(gb, bits_end - get_bits_count(gb));
911 }
912
913 return 0;
914 }
915
decode_hq_slice_row(AVCodecContext * avctx,void * arg,int jobnr,int threadnr)916 static int decode_hq_slice_row(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)
917 {
918 int i;
919 DiracContext *s = avctx->priv_data;
920 DiracSlice *slices = ((DiracSlice *)arg) + s->num_x*jobnr;
921 uint8_t *thread_buf = &s->thread_buf[s->thread_buf_size*threadnr];
922 for (i = 0; i < s->num_x; i++)
923 decode_hq_slice(s, &slices[i], thread_buf);
924 return 0;
925 }
926
927 /**
928 * Dirac Specification ->
929 * 13.5.1 low_delay_transform_data()
930 */
decode_lowdelay(DiracContext * s)931 static int decode_lowdelay(DiracContext *s)
932 {
933 AVCodecContext *avctx = s->avctx;
934 int slice_x, slice_y, bufsize;
935 int64_t coef_buf_size, bytes = 0;
936 const uint8_t *buf;
937 DiracSlice *slices;
938 SliceCoeffs tmp[MAX_DWT_LEVELS];
939 int slice_num = 0;
940
941 if (s->slice_params_num_buf != (s->num_x * s->num_y)) {
942 s->slice_params_buf = av_realloc_f(s->slice_params_buf, s->num_x * s->num_y, sizeof(DiracSlice));
943 if (!s->slice_params_buf) {
944 av_log(s->avctx, AV_LOG_ERROR, "slice params buffer allocation failure\n");
945 s->slice_params_num_buf = 0;
946 return AVERROR(ENOMEM);
947 }
948 s->slice_params_num_buf = s->num_x * s->num_y;
949 }
950 slices = s->slice_params_buf;
951
952 /* 8 becacuse that's how much the golomb reader could overread junk data
953 * from another plane/slice at most, and 512 because SIMD */
954 coef_buf_size = subband_coeffs(s, s->num_x - 1, s->num_y - 1, 0, tmp) + 8;
955 coef_buf_size = (coef_buf_size << (1 + s->pshift)) + 512;
956
957 if (s->threads_num_buf != avctx->thread_count ||
958 s->thread_buf_size != coef_buf_size) {
959 s->threads_num_buf = avctx->thread_count;
960 s->thread_buf_size = coef_buf_size;
961 s->thread_buf = av_realloc_f(s->thread_buf, avctx->thread_count, s->thread_buf_size);
962 if (!s->thread_buf) {
963 av_log(s->avctx, AV_LOG_ERROR, "thread buffer allocation failure\n");
964 return AVERROR(ENOMEM);
965 }
966 }
967
968 align_get_bits(&s->gb);
969 /*[DIRAC_STD] 13.5.2 Slices. slice(sx,sy) */
970 buf = s->gb.buffer + get_bits_count(&s->gb)/8;
971 bufsize = get_bits_left(&s->gb);
972
973 if (s->hq_picture) {
974 int i;
975
976 for (slice_y = 0; bufsize > 0 && slice_y < s->num_y; slice_y++) {
977 for (slice_x = 0; bufsize > 0 && slice_x < s->num_x; slice_x++) {
978 bytes = s->highquality.prefix_bytes + 1;
979 for (i = 0; i < 3; i++) {
980 if (bytes <= bufsize/8)
981 bytes += buf[bytes] * s->highquality.size_scaler + 1;
982 }
983 if (bytes >= INT_MAX || bytes*8 > bufsize) {
984 av_log(s->avctx, AV_LOG_ERROR, "too many bytes\n");
985 return AVERROR_INVALIDDATA;
986 }
987
988 slices[slice_num].bytes = bytes;
989 slices[slice_num].slice_x = slice_x;
990 slices[slice_num].slice_y = slice_y;
991 init_get_bits(&slices[slice_num].gb, buf, bufsize);
992 slice_num++;
993
994 buf += bytes;
995 if (bufsize/8 >= bytes)
996 bufsize -= bytes*8;
997 else
998 bufsize = 0;
999 }
1000 }
1001
1002 if (s->num_x*s->num_y != slice_num) {
1003 av_log(s->avctx, AV_LOG_ERROR, "too few slices\n");
1004 return AVERROR_INVALIDDATA;
1005 }
1006
1007 avctx->execute2(avctx, decode_hq_slice_row, slices, NULL, s->num_y);
1008 } else {
1009 for (slice_y = 0; bufsize > 0 && slice_y < s->num_y; slice_y++) {
1010 for (slice_x = 0; bufsize > 0 && slice_x < s->num_x; slice_x++) {
1011 bytes = (slice_num+1) * (int64_t)s->lowdelay.bytes.num / s->lowdelay.bytes.den
1012 - slice_num * (int64_t)s->lowdelay.bytes.num / s->lowdelay.bytes.den;
1013 if (bytes >= INT_MAX || bytes*8 > bufsize) {
1014 av_log(s->avctx, AV_LOG_ERROR, "too many bytes\n");
1015 return AVERROR_INVALIDDATA;
1016 }
1017 slices[slice_num].bytes = bytes;
1018 slices[slice_num].slice_x = slice_x;
1019 slices[slice_num].slice_y = slice_y;
1020 init_get_bits(&slices[slice_num].gb, buf, bufsize);
1021 slice_num++;
1022
1023 buf += bytes;
1024 if (bufsize/8 >= bytes)
1025 bufsize -= bytes*8;
1026 else
1027 bufsize = 0;
1028 }
1029 }
1030 avctx->execute(avctx, decode_lowdelay_slice, slices, NULL, slice_num,
1031 sizeof(DiracSlice)); /* [DIRAC_STD] 13.5.2 Slices */
1032 }
1033
1034 if (s->dc_prediction) {
1035 if (s->pshift) {
1036 intra_dc_prediction_10(&s->plane[0].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */
1037 intra_dc_prediction_10(&s->plane[1].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */
1038 intra_dc_prediction_10(&s->plane[2].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */
1039 } else {
1040 intra_dc_prediction_8(&s->plane[0].band[0][0]);
1041 intra_dc_prediction_8(&s->plane[1].band[0][0]);
1042 intra_dc_prediction_8(&s->plane[2].band[0][0]);
1043 }
1044 }
1045
1046 return 0;
1047 }
1048
init_planes(DiracContext * s)1049 static void init_planes(DiracContext *s)
1050 {
1051 int i, w, h, level, orientation;
1052
1053 for (i = 0; i < 3; i++) {
1054 Plane *p = &s->plane[i];
1055
1056 p->width = s->seq.width >> (i ? s->chroma_x_shift : 0);
1057 p->height = s->seq.height >> (i ? s->chroma_y_shift : 0);
1058 p->idwt.width = w = CALC_PADDING(p->width , s->wavelet_depth);
1059 p->idwt.height = h = CALC_PADDING(p->height, s->wavelet_depth);
1060 p->idwt.stride = FFALIGN(p->idwt.width, 8) << (1 + s->pshift);
1061
1062 for (level = s->wavelet_depth-1; level >= 0; level--) {
1063 w = w>>1;
1064 h = h>>1;
1065 for (orientation = !!level; orientation < 4; orientation++) {
1066 SubBand *b = &p->band[level][orientation];
1067
1068 b->pshift = s->pshift;
1069 b->ibuf = p->idwt.buf;
1070 b->level = level;
1071 b->stride = p->idwt.stride << (s->wavelet_depth - level);
1072 b->width = w;
1073 b->height = h;
1074 b->orientation = orientation;
1075
1076 if (orientation & 1)
1077 b->ibuf += w << (1+b->pshift);
1078 if (orientation > 1)
1079 b->ibuf += (b->stride>>1);
1080
1081 if (level)
1082 b->parent = &p->band[level-1][orientation];
1083 }
1084 }
1085
1086 if (i > 0) {
1087 p->xblen = s->plane[0].xblen >> s->chroma_x_shift;
1088 p->yblen = s->plane[0].yblen >> s->chroma_y_shift;
1089 p->xbsep = s->plane[0].xbsep >> s->chroma_x_shift;
1090 p->ybsep = s->plane[0].ybsep >> s->chroma_y_shift;
1091 }
1092
1093 p->xoffset = (p->xblen - p->xbsep)/2;
1094 p->yoffset = (p->yblen - p->ybsep)/2;
1095 }
1096 }
1097
1098 /**
1099 * Unpack the motion compensation parameters
1100 * Dirac Specification ->
1101 * 11.2 Picture prediction data. picture_prediction()
1102 */
dirac_unpack_prediction_parameters(DiracContext * s)1103 static int dirac_unpack_prediction_parameters(DiracContext *s)
1104 {
1105 static const uint8_t default_blen[] = { 4, 12, 16, 24 };
1106
1107 GetBitContext *gb = &s->gb;
1108 unsigned idx, ref;
1109
1110 align_get_bits(gb);
1111 /* [DIRAC_STD] 11.2.2 Block parameters. block_parameters() */
1112 /* Luma and Chroma are equal. 11.2.3 */
1113 idx = get_interleaved_ue_golomb(gb); /* [DIRAC_STD] index */
1114
1115 if (idx > 4) {
1116 av_log(s->avctx, AV_LOG_ERROR, "Block prediction index too high\n");
1117 return AVERROR_INVALIDDATA;
1118 }
1119
1120 if (idx == 0) {
1121 s->plane[0].xblen = get_interleaved_ue_golomb(gb);
1122 s->plane[0].yblen = get_interleaved_ue_golomb(gb);
1123 s->plane[0].xbsep = get_interleaved_ue_golomb(gb);
1124 s->plane[0].ybsep = get_interleaved_ue_golomb(gb);
1125 } else {
1126 /*[DIRAC_STD] preset_block_params(index). Table 11.1 */
1127 s->plane[0].xblen = default_blen[idx-1];
1128 s->plane[0].yblen = default_blen[idx-1];
1129 s->plane[0].xbsep = 4 * idx;
1130 s->plane[0].ybsep = 4 * idx;
1131 }
1132 /*[DIRAC_STD] 11.2.4 motion_data_dimensions()
1133 Calculated in function dirac_unpack_block_motion_data */
1134
1135 if (s->plane[0].xblen % (1 << s->chroma_x_shift) != 0 ||
1136 s->plane[0].yblen % (1 << s->chroma_y_shift) != 0 ||
1137 !s->plane[0].xblen || !s->plane[0].yblen) {
1138 av_log(s->avctx, AV_LOG_ERROR,
1139 "invalid x/y block length (%d/%d) for x/y chroma shift (%d/%d)\n",
1140 s->plane[0].xblen, s->plane[0].yblen, s->chroma_x_shift, s->chroma_y_shift);
1141 return AVERROR_INVALIDDATA;
1142 }
1143 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) {
1144 av_log(s->avctx, AV_LOG_ERROR, "Block separation too small\n");
1145 return AVERROR_INVALIDDATA;
1146 }
1147 if (s->plane[0].xbsep > s->plane[0].xblen || s->plane[0].ybsep > s->plane[0].yblen) {
1148 av_log(s->avctx, AV_LOG_ERROR, "Block separation greater than size\n");
1149 return AVERROR_INVALIDDATA;
1150 }
1151 if (FFMAX(s->plane[0].xblen, s->plane[0].yblen) > MAX_BLOCKSIZE) {
1152 av_log(s->avctx, AV_LOG_ERROR, "Unsupported large block size\n");
1153 return AVERROR_PATCHWELCOME;
1154 }
1155
1156 /*[DIRAC_STD] 11.2.5 Motion vector precision. motion_vector_precision()
1157 Read motion vector precision */
1158 s->mv_precision = get_interleaved_ue_golomb(gb);
1159 if (s->mv_precision > 3) {
1160 av_log(s->avctx, AV_LOG_ERROR, "MV precision finer than eighth-pel\n");
1161 return AVERROR_INVALIDDATA;
1162 }
1163
1164 /*[DIRAC_STD] 11.2.6 Global motion. global_motion()
1165 Read the global motion compensation parameters */
1166 s->globalmc_flag = get_bits1(gb);
1167 if (s->globalmc_flag) {
1168 memset(s->globalmc, 0, sizeof(s->globalmc));
1169 /* [DIRAC_STD] pan_tilt(gparams) */
1170 for (ref = 0; ref < s->num_refs; ref++) {
1171 if (get_bits1(gb)) {
1172 s->globalmc[ref].pan_tilt[0] = dirac_get_se_golomb(gb);
1173 s->globalmc[ref].pan_tilt[1] = dirac_get_se_golomb(gb);
1174 }
1175 /* [DIRAC_STD] zoom_rotate_shear(gparams)
1176 zoom/rotation/shear parameters */
1177 if (get_bits1(gb)) {
1178 s->globalmc[ref].zrs_exp = get_interleaved_ue_golomb(gb);
1179 s->globalmc[ref].zrs[0][0] = dirac_get_se_golomb(gb);
1180 s->globalmc[ref].zrs[0][1] = dirac_get_se_golomb(gb);
1181 s->globalmc[ref].zrs[1][0] = dirac_get_se_golomb(gb);
1182 s->globalmc[ref].zrs[1][1] = dirac_get_se_golomb(gb);
1183 } else {
1184 s->globalmc[ref].zrs[0][0] = 1;
1185 s->globalmc[ref].zrs[1][1] = 1;
1186 }
1187 /* [DIRAC_STD] perspective(gparams) */
1188 if (get_bits1(gb)) {
1189 s->globalmc[ref].perspective_exp = get_interleaved_ue_golomb(gb);
1190 s->globalmc[ref].perspective[0] = dirac_get_se_golomb(gb);
1191 s->globalmc[ref].perspective[1] = dirac_get_se_golomb(gb);
1192 }
1193 if (s->globalmc[ref].perspective_exp + (uint64_t)s->globalmc[ref].zrs_exp > 30) {
1194 return AVERROR_INVALIDDATA;
1195 }
1196
1197 }
1198 }
1199
1200 /*[DIRAC_STD] 11.2.7 Picture prediction mode. prediction_mode()
1201 Picture prediction mode, not currently used. */
1202 if (get_interleaved_ue_golomb(gb)) {
1203 av_log(s->avctx, AV_LOG_ERROR, "Unknown picture prediction mode\n");
1204 return AVERROR_INVALIDDATA;
1205 }
1206
1207 /* [DIRAC_STD] 11.2.8 Reference picture weight. reference_picture_weights()
1208 just data read, weight calculation will be done later on. */
1209 s->weight_log2denom = 1;
1210 s->weight[0] = 1;
1211 s->weight[1] = 1;
1212
1213 if (get_bits1(gb)) {
1214 s->weight_log2denom = get_interleaved_ue_golomb(gb);
1215 if (s->weight_log2denom < 1 || s->weight_log2denom > 8) {
1216 av_log(s->avctx, AV_LOG_ERROR, "weight_log2denom unsupported or invalid\n");
1217 s->weight_log2denom = 1;
1218 return AVERROR_INVALIDDATA;
1219 }
1220 s->weight[0] = dirac_get_se_golomb(gb);
1221 if (s->num_refs == 2)
1222 s->weight[1] = dirac_get_se_golomb(gb);
1223 }
1224 return 0;
1225 }
1226
1227 /**
1228 * Dirac Specification ->
1229 * 11.3 Wavelet transform data. wavelet_transform()
1230 */
dirac_unpack_idwt_params(DiracContext * s)1231 static int dirac_unpack_idwt_params(DiracContext *s)
1232 {
1233 GetBitContext *gb = &s->gb;
1234 int i, level;
1235 unsigned tmp;
1236
1237 #define CHECKEDREAD(dst, cond, errmsg) \
1238 tmp = get_interleaved_ue_golomb(gb); \
1239 if (cond) { \
1240 av_log(s->avctx, AV_LOG_ERROR, errmsg); \
1241 return AVERROR_INVALIDDATA; \
1242 }\
1243 dst = tmp;
1244
1245 align_get_bits(gb);
1246
1247 s->zero_res = s->num_refs ? get_bits1(gb) : 0;
1248 if (s->zero_res)
1249 return 0;
1250
1251 /*[DIRAC_STD] 11.3.1 Transform parameters. transform_parameters() */
1252 CHECKEDREAD(s->wavelet_idx, tmp > 6, "wavelet_idx is too big\n")
1253
1254 CHECKEDREAD(s->wavelet_depth, tmp > MAX_DWT_LEVELS || tmp < 1, "invalid number of DWT decompositions\n")
1255
1256 if (!s->low_delay) {
1257 /* Codeblock parameters (core syntax only) */
1258 if (get_bits1(gb)) {
1259 for (i = 0; i <= s->wavelet_depth; i++) {
1260 CHECKEDREAD(s->codeblock[i].width , tmp < 1 || tmp > (s->avctx->width >>s->wavelet_depth-i), "codeblock width invalid\n")
1261 CHECKEDREAD(s->codeblock[i].height, tmp < 1 || tmp > (s->avctx->height>>s->wavelet_depth-i), "codeblock height invalid\n")
1262 }
1263
1264 CHECKEDREAD(s->codeblock_mode, tmp > 1, "unknown codeblock mode\n")
1265 }
1266 else {
1267 for (i = 0; i <= s->wavelet_depth; i++)
1268 s->codeblock[i].width = s->codeblock[i].height = 1;
1269 }
1270 }
1271 else {
1272 s->num_x = get_interleaved_ue_golomb(gb);
1273 s->num_y = get_interleaved_ue_golomb(gb);
1274 if (s->num_x * s->num_y == 0 || s->num_x * (uint64_t)s->num_y > INT_MAX ||
1275 s->num_x * (uint64_t)s->avctx->width > INT_MAX ||
1276 s->num_y * (uint64_t)s->avctx->height > INT_MAX ||
1277 s->num_x > s->avctx->width ||
1278 s->num_y > s->avctx->height
1279 ) {
1280 av_log(s->avctx,AV_LOG_ERROR,"Invalid numx/y\n");
1281 s->num_x = s->num_y = 0;
1282 return AVERROR_INVALIDDATA;
1283 }
1284 if (s->ld_picture) {
1285 s->lowdelay.bytes.num = get_interleaved_ue_golomb(gb);
1286 s->lowdelay.bytes.den = get_interleaved_ue_golomb(gb);
1287 if (s->lowdelay.bytes.den <= 0) {
1288 av_log(s->avctx,AV_LOG_ERROR,"Invalid lowdelay.bytes.den\n");
1289 return AVERROR_INVALIDDATA;
1290 }
1291 } else if (s->hq_picture) {
1292 s->highquality.prefix_bytes = get_interleaved_ue_golomb(gb);
1293 s->highquality.size_scaler = get_interleaved_ue_golomb(gb);
1294 if (s->highquality.prefix_bytes >= INT_MAX / 8) {
1295 av_log(s->avctx,AV_LOG_ERROR,"too many prefix bytes\n");
1296 return AVERROR_INVALIDDATA;
1297 }
1298 }
1299
1300 /* [DIRAC_STD] 11.3.5 Quantisation matrices (low-delay syntax). quant_matrix() */
1301 if (get_bits1(gb)) {
1302 av_log(s->avctx,AV_LOG_DEBUG,"Low Delay: Has Custom Quantization Matrix!\n");
1303 /* custom quantization matrix */
1304 for (level = 0; level < s->wavelet_depth; level++) {
1305 for (i = !!level; i < 4; i++) {
1306 s->lowdelay.quant[level][i] = get_interleaved_ue_golomb(gb);
1307 }
1308 }
1309 } else {
1310 if (s->wavelet_depth > 4) {
1311 av_log(s->avctx,AV_LOG_ERROR,"Mandatory custom low delay matrix missing for depth %d\n", s->wavelet_depth);
1312 return AVERROR_INVALIDDATA;
1313 }
1314 /* default quantization matrix */
1315 for (level = 0; level < s->wavelet_depth; level++)
1316 for (i = 0; i < 4; i++) {
1317 s->lowdelay.quant[level][i] = ff_dirac_default_qmat[s->wavelet_idx][level][i];
1318 /* haar with no shift differs for different depths */
1319 if (s->wavelet_idx == 3)
1320 s->lowdelay.quant[level][i] += 4*(s->wavelet_depth-1 - level);
1321 }
1322 }
1323 }
1324 return 0;
1325 }
1326
pred_sbsplit(uint8_t * sbsplit,int stride,int x,int y)1327 static inline int pred_sbsplit(uint8_t *sbsplit, int stride, int x, int y)
1328 {
1329 static const uint8_t avgsplit[7] = { 0, 0, 1, 1, 1, 2, 2 };
1330
1331 if (!(x|y))
1332 return 0;
1333 else if (!y)
1334 return sbsplit[-1];
1335 else if (!x)
1336 return sbsplit[-stride];
1337
1338 return avgsplit[sbsplit[-1] + sbsplit[-stride] + sbsplit[-stride-1]];
1339 }
1340
pred_block_mode(DiracBlock * block,int stride,int x,int y,int refmask)1341 static inline int pred_block_mode(DiracBlock *block, int stride, int x, int y, int refmask)
1342 {
1343 int pred;
1344
1345 if (!(x|y))
1346 return 0;
1347 else if (!y)
1348 return block[-1].ref & refmask;
1349 else if (!x)
1350 return block[-stride].ref & refmask;
1351
1352 /* return the majority */
1353 pred = (block[-1].ref & refmask) + (block[-stride].ref & refmask) + (block[-stride-1].ref & refmask);
1354 return (pred >> 1) & refmask;
1355 }
1356
pred_block_dc(DiracBlock * block,int stride,int x,int y)1357 static inline void pred_block_dc(DiracBlock *block, int stride, int x, int y)
1358 {
1359 int i, n = 0;
1360
1361 memset(block->u.dc, 0, sizeof(block->u.dc));
1362
1363 if (x && !(block[-1].ref & 3)) {
1364 for (i = 0; i < 3; i++)
1365 block->u.dc[i] += block[-1].u.dc[i];
1366 n++;
1367 }
1368
1369 if (y && !(block[-stride].ref & 3)) {
1370 for (i = 0; i < 3; i++)
1371 block->u.dc[i] += block[-stride].u.dc[i];
1372 n++;
1373 }
1374
1375 if (x && y && !(block[-1-stride].ref & 3)) {
1376 for (i = 0; i < 3; i++)
1377 block->u.dc[i] += block[-1-stride].u.dc[i];
1378 n++;
1379 }
1380
1381 if (n == 2) {
1382 for (i = 0; i < 3; i++)
1383 block->u.dc[i] = (block->u.dc[i]+1)>>1;
1384 } else if (n == 3) {
1385 for (i = 0; i < 3; i++)
1386 block->u.dc[i] = divide3(block->u.dc[i]);
1387 }
1388 }
1389
pred_mv(DiracBlock * block,int stride,int x,int y,int ref)1390 static inline void pred_mv(DiracBlock *block, int stride, int x, int y, int ref)
1391 {
1392 int16_t *pred[3];
1393 int refmask = ref+1;
1394 int mask = refmask | DIRAC_REF_MASK_GLOBAL; /* exclude gmc blocks */
1395 int n = 0;
1396
1397 if (x && (block[-1].ref & mask) == refmask)
1398 pred[n++] = block[-1].u.mv[ref];
1399
1400 if (y && (block[-stride].ref & mask) == refmask)
1401 pred[n++] = block[-stride].u.mv[ref];
1402
1403 if (x && y && (block[-stride-1].ref & mask) == refmask)
1404 pred[n++] = block[-stride-1].u.mv[ref];
1405
1406 switch (n) {
1407 case 0:
1408 block->u.mv[ref][0] = 0;
1409 block->u.mv[ref][1] = 0;
1410 break;
1411 case 1:
1412 block->u.mv[ref][0] = pred[0][0];
1413 block->u.mv[ref][1] = pred[0][1];
1414 break;
1415 case 2:
1416 block->u.mv[ref][0] = (pred[0][0] + pred[1][0] + 1) >> 1;
1417 block->u.mv[ref][1] = (pred[0][1] + pred[1][1] + 1) >> 1;
1418 break;
1419 case 3:
1420 block->u.mv[ref][0] = mid_pred(pred[0][0], pred[1][0], pred[2][0]);
1421 block->u.mv[ref][1] = mid_pred(pred[0][1], pred[1][1], pred[2][1]);
1422 break;
1423 }
1424 }
1425
global_mv(DiracContext * s,DiracBlock * block,int x,int y,int ref)1426 static void global_mv(DiracContext *s, DiracBlock *block, int x, int y, int ref)
1427 {
1428 int ez = s->globalmc[ref].zrs_exp;
1429 int ep = s->globalmc[ref].perspective_exp;
1430 int (*A)[2] = s->globalmc[ref].zrs;
1431 int *b = s->globalmc[ref].pan_tilt;
1432 int *c = s->globalmc[ref].perspective;
1433
1434 int64_t m = (1<<ep) - (c[0]*(int64_t)x + c[1]*(int64_t)y);
1435 int64_t mx = m * (int64_t)((A[0][0] * (int64_t)x + A[0][1]*(int64_t)y) + (1LL<<ez) * b[0]);
1436 int64_t my = m * (int64_t)((A[1][0] * (int64_t)x + A[1][1]*(int64_t)y) + (1LL<<ez) * b[1]);
1437
1438 block->u.mv[ref][0] = (mx + (1<<(ez+ep))) >> (ez+ep);
1439 block->u.mv[ref][1] = (my + (1<<(ez+ep))) >> (ez+ep);
1440 }
1441
decode_block_params(DiracContext * s,DiracArith arith[8],DiracBlock * block,int stride,int x,int y)1442 static void decode_block_params(DiracContext *s, DiracArith arith[8], DiracBlock *block,
1443 int stride, int x, int y)
1444 {
1445 int i;
1446
1447 block->ref = pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_REF1);
1448 block->ref ^= dirac_get_arith_bit(arith, CTX_PMODE_REF1);
1449
1450 if (s->num_refs == 2) {
1451 block->ref |= pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_REF2);
1452 block->ref ^= dirac_get_arith_bit(arith, CTX_PMODE_REF2) << 1;
1453 }
1454
1455 if (!block->ref) {
1456 pred_block_dc(block, stride, x, y);
1457 for (i = 0; i < 3; i++)
1458 block->u.dc[i] += (unsigned)dirac_get_arith_int(arith+1+i, CTX_DC_F1, CTX_DC_DATA);
1459 return;
1460 }
1461
1462 if (s->globalmc_flag) {
1463 block->ref |= pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_GLOBAL);
1464 block->ref ^= dirac_get_arith_bit(arith, CTX_GLOBAL_BLOCK) << 2;
1465 }
1466
1467 for (i = 0; i < s->num_refs; i++)
1468 if (block->ref & (i+1)) {
1469 if (block->ref & DIRAC_REF_MASK_GLOBAL) {
1470 global_mv(s, block, x, y, i);
1471 } else {
1472 pred_mv(block, stride, x, y, i);
1473 block->u.mv[i][0] += (unsigned)dirac_get_arith_int(arith + 4 + 2 * i, CTX_MV_F1, CTX_MV_DATA);
1474 block->u.mv[i][1] += (unsigned)dirac_get_arith_int(arith + 5 + 2 * i, CTX_MV_F1, CTX_MV_DATA);
1475 }
1476 }
1477 }
1478
1479 /**
1480 * Copies the current block to the other blocks covered by the current superblock split mode
1481 */
propagate_block_data(DiracBlock * block,int stride,int size)1482 static void propagate_block_data(DiracBlock *block, int stride, int size)
1483 {
1484 int x, y;
1485 DiracBlock *dst = block;
1486
1487 for (x = 1; x < size; x++)
1488 dst[x] = *block;
1489
1490 for (y = 1; y < size; y++) {
1491 dst += stride;
1492 for (x = 0; x < size; x++)
1493 dst[x] = *block;
1494 }
1495 }
1496
1497 /**
1498 * Dirac Specification ->
1499 * 12. Block motion data syntax
1500 */
dirac_unpack_block_motion_data(DiracContext * s)1501 static int dirac_unpack_block_motion_data(DiracContext *s)
1502 {
1503 GetBitContext *gb = &s->gb;
1504 uint8_t *sbsplit = s->sbsplit;
1505 int i, x, y, q, p;
1506 DiracArith arith[8];
1507
1508 align_get_bits(gb);
1509
1510 /* [DIRAC_STD] 11.2.4 and 12.2.1 Number of blocks and superblocks */
1511 s->sbwidth = DIVRNDUP(s->seq.width, 4*s->plane[0].xbsep);
1512 s->sbheight = DIVRNDUP(s->seq.height, 4*s->plane[0].ybsep);
1513 s->blwidth = 4 * s->sbwidth;
1514 s->blheight = 4 * s->sbheight;
1515
1516 /* [DIRAC_STD] 12.3.1 Superblock splitting modes. superblock_split_modes()
1517 decode superblock split modes */
1518 ff_dirac_init_arith_decoder(arith, gb, get_interleaved_ue_golomb(gb)); /* get_interleaved_ue_golomb(gb) is the length */
1519 for (y = 0; y < s->sbheight; y++) {
1520 for (x = 0; x < s->sbwidth; x++) {
1521 unsigned int split = dirac_get_arith_uint(arith, CTX_SB_F1, CTX_SB_DATA);
1522 if (split > 2)
1523 return AVERROR_INVALIDDATA;
1524 sbsplit[x] = (split + pred_sbsplit(sbsplit+x, s->sbwidth, x, y)) % 3;
1525 }
1526 sbsplit += s->sbwidth;
1527 }
1528
1529 /* setup arith decoding */
1530 ff_dirac_init_arith_decoder(arith, gb, get_interleaved_ue_golomb(gb));
1531 for (i = 0; i < s->num_refs; i++) {
1532 ff_dirac_init_arith_decoder(arith + 4 + 2 * i, gb, get_interleaved_ue_golomb(gb));
1533 ff_dirac_init_arith_decoder(arith + 5 + 2 * i, gb, get_interleaved_ue_golomb(gb));
1534 }
1535 for (i = 0; i < 3; i++)
1536 ff_dirac_init_arith_decoder(arith+1+i, gb, get_interleaved_ue_golomb(gb));
1537
1538 for (y = 0; y < s->sbheight; y++)
1539 for (x = 0; x < s->sbwidth; x++) {
1540 int blkcnt = 1 << s->sbsplit[y * s->sbwidth + x];
1541 int step = 4 >> s->sbsplit[y * s->sbwidth + x];
1542
1543 for (q = 0; q < blkcnt; q++)
1544 for (p = 0; p < blkcnt; p++) {
1545 int bx = 4 * x + p*step;
1546 int by = 4 * y + q*step;
1547 DiracBlock *block = &s->blmotion[by*s->blwidth + bx];
1548 decode_block_params(s, arith, block, s->blwidth, bx, by);
1549 propagate_block_data(block, s->blwidth, step);
1550 }
1551 }
1552
1553 for (i = 0; i < 4 + 2*s->num_refs; i++) {
1554 if (arith[i].error)
1555 return arith[i].error;
1556 }
1557
1558 return 0;
1559 }
1560
weight(int i,int blen,int offset)1561 static int weight(int i, int blen, int offset)
1562 {
1563 #define ROLLOFF(i) offset == 1 ? ((i) ? 5 : 3) : \
1564 (1 + (6*(i) + offset - 1) / (2*offset - 1))
1565
1566 if (i < 2*offset)
1567 return ROLLOFF(i);
1568 else if (i > blen-1 - 2*offset)
1569 return ROLLOFF(blen-1 - i);
1570 return 8;
1571 }
1572
init_obmc_weight_row(Plane * p,uint8_t * obmc_weight,int stride,int left,int right,int wy)1573 static void init_obmc_weight_row(Plane *p, uint8_t *obmc_weight, int stride,
1574 int left, int right, int wy)
1575 {
1576 int x;
1577 for (x = 0; left && x < p->xblen >> 1; x++)
1578 obmc_weight[x] = wy*8;
1579 for (; x < p->xblen >> right; x++)
1580 obmc_weight[x] = wy*weight(x, p->xblen, p->xoffset);
1581 for (; x < p->xblen; x++)
1582 obmc_weight[x] = wy*8;
1583 for (; x < stride; x++)
1584 obmc_weight[x] = 0;
1585 }
1586
init_obmc_weight(Plane * p,uint8_t * obmc_weight,int stride,int left,int right,int top,int bottom)1587 static void init_obmc_weight(Plane *p, uint8_t *obmc_weight, int stride,
1588 int left, int right, int top, int bottom)
1589 {
1590 int y;
1591 for (y = 0; top && y < p->yblen >> 1; y++) {
1592 init_obmc_weight_row(p, obmc_weight, stride, left, right, 8);
1593 obmc_weight += stride;
1594 }
1595 for (; y < p->yblen >> bottom; y++) {
1596 int wy = weight(y, p->yblen, p->yoffset);
1597 init_obmc_weight_row(p, obmc_weight, stride, left, right, wy);
1598 obmc_weight += stride;
1599 }
1600 for (; y < p->yblen; y++) {
1601 init_obmc_weight_row(p, obmc_weight, stride, left, right, 8);
1602 obmc_weight += stride;
1603 }
1604 }
1605
init_obmc_weights(DiracContext * s,Plane * p,int by)1606 static void init_obmc_weights(DiracContext *s, Plane *p, int by)
1607 {
1608 int top = !by;
1609 int bottom = by == s->blheight-1;
1610
1611 /* don't bother re-initing for rows 2 to blheight-2, the weights don't change */
1612 if (top || bottom || by == 1) {
1613 init_obmc_weight(p, s->obmc_weight[0], MAX_BLOCKSIZE, 1, 0, top, bottom);
1614 init_obmc_weight(p, s->obmc_weight[1], MAX_BLOCKSIZE, 0, 0, top, bottom);
1615 init_obmc_weight(p, s->obmc_weight[2], MAX_BLOCKSIZE, 0, 1, top, bottom);
1616 }
1617 }
1618
1619 static const uint8_t epel_weights[4][4][4] = {
1620 {{ 16, 0, 0, 0 },
1621 { 12, 4, 0, 0 },
1622 { 8, 8, 0, 0 },
1623 { 4, 12, 0, 0 }},
1624 {{ 12, 0, 4, 0 },
1625 { 9, 3, 3, 1 },
1626 { 6, 6, 2, 2 },
1627 { 3, 9, 1, 3 }},
1628 {{ 8, 0, 8, 0 },
1629 { 6, 2, 6, 2 },
1630 { 4, 4, 4, 4 },
1631 { 2, 6, 2, 6 }},
1632 {{ 4, 0, 12, 0 },
1633 { 3, 1, 9, 3 },
1634 { 2, 2, 6, 6 },
1635 { 1, 3, 3, 9 }}
1636 };
1637
1638 /**
1639 * For block x,y, determine which of the hpel planes to do bilinear
1640 * interpolation from and set src[] to the location in each hpel plane
1641 * to MC from.
1642 *
1643 * @return the index of the put_dirac_pixels_tab function to use
1644 * 0 for 1 plane (fpel,hpel), 1 for 2 planes (qpel), 2 for 4 planes (qpel), and 3 for epel
1645 */
mc_subpel(DiracContext * s,DiracBlock * block,const uint8_t * src[5],int x,int y,int ref,int plane)1646 static int mc_subpel(DiracContext *s, DiracBlock *block, const uint8_t *src[5],
1647 int x, int y, int ref, int plane)
1648 {
1649 Plane *p = &s->plane[plane];
1650 uint8_t **ref_hpel = s->ref_pics[ref]->hpel[plane];
1651 int motion_x = block->u.mv[ref][0];
1652 int motion_y = block->u.mv[ref][1];
1653 int mx, my, i, epel, nplanes = 0;
1654
1655 if (plane) {
1656 motion_x >>= s->chroma_x_shift;
1657 motion_y >>= s->chroma_y_shift;
1658 }
1659
1660 mx = motion_x & ~(-1U << s->mv_precision);
1661 my = motion_y & ~(-1U << s->mv_precision);
1662 motion_x >>= s->mv_precision;
1663 motion_y >>= s->mv_precision;
1664 /* normalize subpel coordinates to epel */
1665 /* TODO: template this function? */
1666 mx <<= 3 - s->mv_precision;
1667 my <<= 3 - s->mv_precision;
1668
1669 x += motion_x;
1670 y += motion_y;
1671 epel = (mx|my)&1;
1672
1673 /* hpel position */
1674 if (!((mx|my)&3)) {
1675 nplanes = 1;
1676 src[0] = ref_hpel[(my>>1)+(mx>>2)] + y*p->stride + x;
1677 } else {
1678 /* qpel or epel */
1679 nplanes = 4;
1680 for (i = 0; i < 4; i++)
1681 src[i] = ref_hpel[i] + y*p->stride + x;
1682
1683 /* if we're interpolating in the right/bottom halves, adjust the planes as needed
1684 we increment x/y because the edge changes for half of the pixels */
1685 if (mx > 4) {
1686 src[0] += 1;
1687 src[2] += 1;
1688 x++;
1689 }
1690 if (my > 4) {
1691 src[0] += p->stride;
1692 src[1] += p->stride;
1693 y++;
1694 }
1695
1696 /* hpel planes are:
1697 [0]: F [1]: H
1698 [2]: V [3]: C */
1699 if (!epel) {
1700 /* check if we really only need 2 planes since either mx or my is
1701 a hpel position. (epel weights of 0 handle this there) */
1702 if (!(mx&3)) {
1703 /* mx == 0: average [0] and [2]
1704 mx == 4: average [1] and [3] */
1705 src[!mx] = src[2 + !!mx];
1706 nplanes = 2;
1707 } else if (!(my&3)) {
1708 src[0] = src[(my>>1) ];
1709 src[1] = src[(my>>1)+1];
1710 nplanes = 2;
1711 }
1712 } else {
1713 /* adjust the ordering if needed so the weights work */
1714 if (mx > 4) {
1715 FFSWAP(const uint8_t *, src[0], src[1]);
1716 FFSWAP(const uint8_t *, src[2], src[3]);
1717 }
1718 if (my > 4) {
1719 FFSWAP(const uint8_t *, src[0], src[2]);
1720 FFSWAP(const uint8_t *, src[1], src[3]);
1721 }
1722 src[4] = epel_weights[my&3][mx&3];
1723 }
1724 }
1725
1726 /* fixme: v/h _edge_pos */
1727 if (x + p->xblen > p->width +EDGE_WIDTH/2 ||
1728 y + p->yblen > p->height+EDGE_WIDTH/2 ||
1729 x < 0 || y < 0) {
1730 for (i = 0; i < nplanes; i++) {
1731 s->vdsp.emulated_edge_mc(s->edge_emu_buffer[i], src[i],
1732 p->stride, p->stride,
1733 p->xblen, p->yblen, x, y,
1734 p->width+EDGE_WIDTH/2, p->height+EDGE_WIDTH/2);
1735 src[i] = s->edge_emu_buffer[i];
1736 }
1737 }
1738 return (nplanes>>1) + epel;
1739 }
1740
add_dc(uint16_t * dst,int dc,int stride,uint8_t * obmc_weight,int xblen,int yblen)1741 static void add_dc(uint16_t *dst, int dc, int stride,
1742 uint8_t *obmc_weight, int xblen, int yblen)
1743 {
1744 int x, y;
1745 dc += 128;
1746
1747 for (y = 0; y < yblen; y++) {
1748 for (x = 0; x < xblen; x += 2) {
1749 dst[x ] += dc * obmc_weight[x ];
1750 dst[x+1] += dc * obmc_weight[x+1];
1751 }
1752 dst += stride;
1753 obmc_weight += MAX_BLOCKSIZE;
1754 }
1755 }
1756
block_mc(DiracContext * s,DiracBlock * block,uint16_t * mctmp,uint8_t * obmc_weight,int plane,int dstx,int dsty)1757 static void block_mc(DiracContext *s, DiracBlock *block,
1758 uint16_t *mctmp, uint8_t *obmc_weight,
1759 int plane, int dstx, int dsty)
1760 {
1761 Plane *p = &s->plane[plane];
1762 const uint8_t *src[5];
1763 int idx;
1764
1765 switch (block->ref&3) {
1766 case 0: /* DC */
1767 add_dc(mctmp, block->u.dc[plane], p->stride, obmc_weight, p->xblen, p->yblen);
1768 return;
1769 case 1:
1770 case 2:
1771 idx = mc_subpel(s, block, src, dstx, dsty, (block->ref&3)-1, plane);
1772 s->put_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);
1773 if (s->weight_func)
1774 s->weight_func(s->mcscratch, p->stride, s->weight_log2denom,
1775 s->weight[0] + s->weight[1], p->yblen);
1776 break;
1777 case 3:
1778 idx = mc_subpel(s, block, src, dstx, dsty, 0, plane);
1779 s->put_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);
1780 idx = mc_subpel(s, block, src, dstx, dsty, 1, plane);
1781 if (s->biweight_func) {
1782 /* fixme: +32 is a quick hack */
1783 s->put_pixels_tab[idx](s->mcscratch + 32, src, p->stride, p->yblen);
1784 s->biweight_func(s->mcscratch, s->mcscratch+32, p->stride, s->weight_log2denom,
1785 s->weight[0], s->weight[1], p->yblen);
1786 } else
1787 s->avg_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);
1788 break;
1789 }
1790 s->add_obmc(mctmp, s->mcscratch, p->stride, obmc_weight, p->yblen);
1791 }
1792
mc_row(DiracContext * s,DiracBlock * block,uint16_t * mctmp,int plane,int dsty)1793 static void mc_row(DiracContext *s, DiracBlock *block, uint16_t *mctmp, int plane, int dsty)
1794 {
1795 Plane *p = &s->plane[plane];
1796 int x, dstx = p->xbsep - p->xoffset;
1797
1798 block_mc(s, block, mctmp, s->obmc_weight[0], plane, -p->xoffset, dsty);
1799 mctmp += p->xbsep;
1800
1801 for (x = 1; x < s->blwidth-1; x++) {
1802 block_mc(s, block+x, mctmp, s->obmc_weight[1], plane, dstx, dsty);
1803 dstx += p->xbsep;
1804 mctmp += p->xbsep;
1805 }
1806 block_mc(s, block+x, mctmp, s->obmc_weight[2], plane, dstx, dsty);
1807 }
1808
select_dsp_funcs(DiracContext * s,int width,int height,int xblen,int yblen)1809 static void select_dsp_funcs(DiracContext *s, int width, int height, int xblen, int yblen)
1810 {
1811 int idx = 0;
1812 if (xblen > 8)
1813 idx = 1;
1814 if (xblen > 16)
1815 idx = 2;
1816
1817 memcpy(s->put_pixels_tab, s->diracdsp.put_dirac_pixels_tab[idx], sizeof(s->put_pixels_tab));
1818 memcpy(s->avg_pixels_tab, s->diracdsp.avg_dirac_pixels_tab[idx], sizeof(s->avg_pixels_tab));
1819 s->add_obmc = s->diracdsp.add_dirac_obmc[idx];
1820 if (s->weight_log2denom > 1 || s->weight[0] != 1 || s->weight[1] != 1) {
1821 s->weight_func = s->diracdsp.weight_dirac_pixels_tab[idx];
1822 s->biweight_func = s->diracdsp.biweight_dirac_pixels_tab[idx];
1823 } else {
1824 s->weight_func = NULL;
1825 s->biweight_func = NULL;
1826 }
1827 }
1828
interpolate_refplane(DiracContext * s,DiracFrame * ref,int plane,int width,int height)1829 static int interpolate_refplane(DiracContext *s, DiracFrame *ref, int plane, int width, int height)
1830 {
1831 /* chroma allocates an edge of 8 when subsampled
1832 which for 4:2:2 means an h edge of 16 and v edge of 8
1833 just use 8 for everything for the moment */
1834 int i, edge = EDGE_WIDTH/2;
1835
1836 ref->hpel[plane][0] = ref->avframe->data[plane];
1837 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 */
1838
1839 /* no need for hpel if we only have fpel vectors */
1840 if (!s->mv_precision)
1841 return 0;
1842
1843 for (i = 1; i < 4; i++) {
1844 if (!ref->hpel_base[plane][i])
1845 ref->hpel_base[plane][i] = av_malloc((height+2*edge) * ref->avframe->linesize[plane] + 32);
1846 if (!ref->hpel_base[plane][i]) {
1847 return AVERROR(ENOMEM);
1848 }
1849 /* we need to be 16-byte aligned even for chroma */
1850 ref->hpel[plane][i] = ref->hpel_base[plane][i] + edge*ref->avframe->linesize[plane] + 16;
1851 }
1852
1853 if (!ref->interpolated[plane]) {
1854 s->diracdsp.dirac_hpel_filter(ref->hpel[plane][1], ref->hpel[plane][2],
1855 ref->hpel[plane][3], ref->hpel[plane][0],
1856 ref->avframe->linesize[plane], width, height);
1857 s->mpvencdsp.draw_edges(ref->hpel[plane][1], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
1858 s->mpvencdsp.draw_edges(ref->hpel[plane][2], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
1859 s->mpvencdsp.draw_edges(ref->hpel[plane][3], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
1860 }
1861 ref->interpolated[plane] = 1;
1862
1863 return 0;
1864 }
1865
1866 /**
1867 * Dirac Specification ->
1868 * 13.0 Transform data syntax. transform_data()
1869 */
dirac_decode_frame_internal(DiracContext * s)1870 static int dirac_decode_frame_internal(DiracContext *s)
1871 {
1872 DWTContext d;
1873 int y, i, comp, dsty;
1874 int ret;
1875
1876 if (s->low_delay) {
1877 /* [DIRAC_STD] 13.5.1 low_delay_transform_data() */
1878 if (!s->hq_picture) {
1879 for (comp = 0; comp < 3; comp++) {
1880 Plane *p = &s->plane[comp];
1881 memset(p->idwt.buf, 0, p->idwt.stride * p->idwt.height);
1882 }
1883 }
1884 if (!s->zero_res) {
1885 if ((ret = decode_lowdelay(s)) < 0)
1886 return ret;
1887 }
1888 }
1889
1890 for (comp = 0; comp < 3; comp++) {
1891 Plane *p = &s->plane[comp];
1892 uint8_t *frame = s->current_picture->avframe->data[comp];
1893
1894 /* FIXME: small resolutions */
1895 for (i = 0; i < 4; i++)
1896 s->edge_emu_buffer[i] = s->edge_emu_buffer_base + i*FFALIGN(p->width, 16);
1897
1898 if (!s->zero_res && !s->low_delay)
1899 {
1900 memset(p->idwt.buf, 0, p->idwt.stride * p->idwt.height);
1901 ret = decode_component(s, comp); /* [DIRAC_STD] 13.4.1 core_transform_data() */
1902 if (ret < 0)
1903 return ret;
1904 }
1905 ret = ff_spatial_idwt_init(&d, &p->idwt, s->wavelet_idx+2,
1906 s->wavelet_depth, s->bit_depth);
1907 if (ret < 0)
1908 return ret;
1909
1910 if (!s->num_refs) { /* intra */
1911 for (y = 0; y < p->height; y += 16) {
1912 int idx = (s->bit_depth - 8) >> 1;
1913 ff_spatial_idwt_slice2(&d, y+16); /* decode */
1914 s->diracdsp.put_signed_rect_clamped[idx](frame + y*p->stride,
1915 p->stride,
1916 p->idwt.buf + y*p->idwt.stride,
1917 p->idwt.stride, p->width, 16);
1918 }
1919 } else { /* inter */
1920 int rowheight = p->ybsep*p->stride;
1921
1922 select_dsp_funcs(s, p->width, p->height, p->xblen, p->yblen);
1923
1924 for (i = 0; i < s->num_refs; i++) {
1925 int ret = interpolate_refplane(s, s->ref_pics[i], comp, p->width, p->height);
1926 if (ret < 0)
1927 return ret;
1928 }
1929
1930 memset(s->mctmp, 0, 4*p->yoffset*p->stride);
1931
1932 dsty = -p->yoffset;
1933 for (y = 0; y < s->blheight; y++) {
1934 int h = 0,
1935 start = FFMAX(dsty, 0);
1936 uint16_t *mctmp = s->mctmp + y*rowheight;
1937 DiracBlock *blocks = s->blmotion + y*s->blwidth;
1938
1939 init_obmc_weights(s, p, y);
1940
1941 if (y == s->blheight-1 || start+p->ybsep > p->height)
1942 h = p->height - start;
1943 else
1944 h = p->ybsep - (start - dsty);
1945 if (h < 0)
1946 break;
1947
1948 memset(mctmp+2*p->yoffset*p->stride, 0, 2*rowheight);
1949 mc_row(s, blocks, mctmp, comp, dsty);
1950
1951 mctmp += (start - dsty)*p->stride + p->xoffset;
1952 ff_spatial_idwt_slice2(&d, start + h); /* decode */
1953 /* NOTE: add_rect_clamped hasn't been templated hence the shifts.
1954 * idwt.stride is passed as pixels, not in bytes as in the rest of the decoder */
1955 s->diracdsp.add_rect_clamped(frame + start*p->stride, mctmp, p->stride,
1956 (int16_t*)(p->idwt.buf) + start*(p->idwt.stride >> 1), (p->idwt.stride >> 1), p->width, h);
1957
1958 dsty += p->ybsep;
1959 }
1960 }
1961 }
1962
1963
1964 return 0;
1965 }
1966
get_buffer_with_edge(AVCodecContext * avctx,AVFrame * f,int flags)1967 static int get_buffer_with_edge(AVCodecContext *avctx, AVFrame *f, int flags)
1968 {
1969 int ret, i;
1970 int chroma_x_shift, chroma_y_shift;
1971 ret = av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &chroma_x_shift,
1972 &chroma_y_shift);
1973 if (ret < 0)
1974 return ret;
1975
1976 f->width = avctx->width + 2 * EDGE_WIDTH;
1977 f->height = avctx->height + 2 * EDGE_WIDTH + 2;
1978 ret = ff_get_buffer(avctx, f, flags);
1979 if (ret < 0)
1980 return ret;
1981
1982 for (i = 0; f->data[i]; i++) {
1983 int offset = (EDGE_WIDTH >> (i && i<3 ? chroma_y_shift : 0)) *
1984 f->linesize[i] + 32;
1985 f->data[i] += offset;
1986 }
1987 f->width = avctx->width;
1988 f->height = avctx->height;
1989
1990 return 0;
1991 }
1992
1993 /**
1994 * Dirac Specification ->
1995 * 11.1.1 Picture Header. picture_header()
1996 */
dirac_decode_picture_header(DiracContext * s)1997 static int dirac_decode_picture_header(DiracContext *s)
1998 {
1999 unsigned retire, picnum;
2000 int i, j, ret;
2001 int64_t refdist, refnum;
2002 GetBitContext *gb = &s->gb;
2003
2004 /* [DIRAC_STD] 11.1.1 Picture Header. picture_header() PICTURE_NUM */
2005 picnum = s->current_picture->avframe->display_picture_number = get_bits_long(gb, 32);
2006
2007
2008 av_log(s->avctx,AV_LOG_DEBUG,"PICTURE_NUM: %d\n",picnum);
2009
2010 /* if this is the first keyframe after a sequence header, start our
2011 reordering from here */
2012 if (s->frame_number < 0)
2013 s->frame_number = picnum;
2014
2015 s->ref_pics[0] = s->ref_pics[1] = NULL;
2016 for (i = 0; i < s->num_refs; i++) {
2017 refnum = (picnum + dirac_get_se_golomb(gb)) & 0xFFFFFFFF;
2018 refdist = INT64_MAX;
2019
2020 /* find the closest reference to the one we want */
2021 /* Jordi: this is needed if the referenced picture hasn't yet arrived */
2022 for (j = 0; j < MAX_REFERENCE_FRAMES && refdist; j++)
2023 if (s->ref_frames[j]
2024 && FFABS(s->ref_frames[j]->avframe->display_picture_number - refnum) < refdist) {
2025 s->ref_pics[i] = s->ref_frames[j];
2026 refdist = FFABS(s->ref_frames[j]->avframe->display_picture_number - refnum);
2027 }
2028
2029 if (!s->ref_pics[i] || refdist)
2030 av_log(s->avctx, AV_LOG_DEBUG, "Reference not found\n");
2031
2032 /* if there were no references at all, allocate one */
2033 if (!s->ref_pics[i])
2034 for (j = 0; j < MAX_FRAMES; j++)
2035 if (!s->all_frames[j].avframe->data[0]) {
2036 s->ref_pics[i] = &s->all_frames[j];
2037 ret = get_buffer_with_edge(s->avctx, s->ref_pics[i]->avframe, AV_GET_BUFFER_FLAG_REF);
2038 if (ret < 0)
2039 return ret;
2040 break;
2041 }
2042
2043 if (!s->ref_pics[i]) {
2044 av_log(s->avctx, AV_LOG_ERROR, "Reference could not be allocated\n");
2045 return AVERROR_INVALIDDATA;
2046 }
2047
2048 }
2049
2050 /* retire the reference frames that are not used anymore */
2051 if (s->current_picture->reference) {
2052 retire = (picnum + dirac_get_se_golomb(gb)) & 0xFFFFFFFF;
2053 if (retire != picnum) {
2054 DiracFrame *retire_pic = remove_frame(s->ref_frames, retire);
2055
2056 if (retire_pic)
2057 retire_pic->reference &= DELAYED_PIC_REF;
2058 else
2059 av_log(s->avctx, AV_LOG_DEBUG, "Frame to retire not found\n");
2060 }
2061
2062 /* if reference array is full, remove the oldest as per the spec */
2063 while (add_frame(s->ref_frames, MAX_REFERENCE_FRAMES, s->current_picture)) {
2064 av_log(s->avctx, AV_LOG_ERROR, "Reference frame overflow\n");
2065 remove_frame(s->ref_frames, s->ref_frames[0]->avframe->display_picture_number)->reference &= DELAYED_PIC_REF;
2066 }
2067 }
2068
2069 if (s->num_refs) {
2070 ret = dirac_unpack_prediction_parameters(s); /* [DIRAC_STD] 11.2 Picture Prediction Data. picture_prediction() */
2071 if (ret < 0)
2072 return ret;
2073 ret = dirac_unpack_block_motion_data(s); /* [DIRAC_STD] 12. Block motion data syntax */
2074 if (ret < 0)
2075 return ret;
2076 }
2077 ret = dirac_unpack_idwt_params(s); /* [DIRAC_STD] 11.3 Wavelet transform data */
2078 if (ret < 0)
2079 return ret;
2080
2081 init_planes(s);
2082 return 0;
2083 }
2084
get_delayed_pic(DiracContext * s,AVFrame * picture,int * got_frame)2085 static int get_delayed_pic(DiracContext *s, AVFrame *picture, int *got_frame)
2086 {
2087 DiracFrame *out = s->delay_frames[0];
2088 int i, out_idx = 0;
2089 int ret;
2090
2091 /* find frame with lowest picture number */
2092 for (i = 1; s->delay_frames[i]; i++)
2093 if (s->delay_frames[i]->avframe->display_picture_number < out->avframe->display_picture_number) {
2094 out = s->delay_frames[i];
2095 out_idx = i;
2096 }
2097
2098 for (i = out_idx; s->delay_frames[i]; i++)
2099 s->delay_frames[i] = s->delay_frames[i+1];
2100
2101 if (out) {
2102 out->reference ^= DELAYED_PIC_REF;
2103 if((ret = av_frame_ref(picture, out->avframe)) < 0)
2104 return ret;
2105 *got_frame = 1;
2106 }
2107
2108 return 0;
2109 }
2110
2111 /**
2112 * Dirac Specification ->
2113 * 9.6 Parse Info Header Syntax. parse_info()
2114 * 4 byte start code + byte parse code + 4 byte size + 4 byte previous size
2115 */
2116 #define DATA_UNIT_HEADER_SIZE 13
2117
2118 /* [DIRAC_STD] dirac_decode_data_unit makes reference to the while defined in 9.3
2119 inside the function parse_sequence() */
dirac_decode_data_unit(AVCodecContext * avctx,const uint8_t * buf,int size)2120 static int dirac_decode_data_unit(AVCodecContext *avctx, const uint8_t *buf, int size)
2121 {
2122 DiracContext *s = avctx->priv_data;
2123 DiracFrame *pic = NULL;
2124 AVDiracSeqHeader *dsh;
2125 int ret, i;
2126 uint8_t parse_code;
2127 unsigned tmp;
2128
2129 if (size < DATA_UNIT_HEADER_SIZE)
2130 return AVERROR_INVALIDDATA;
2131
2132 parse_code = buf[4];
2133
2134 init_get_bits(&s->gb, &buf[13], 8*(size - DATA_UNIT_HEADER_SIZE));
2135
2136 if (parse_code == DIRAC_PCODE_SEQ_HEADER) {
2137 if (s->seen_sequence_header)
2138 return 0;
2139
2140 /* [DIRAC_STD] 10. Sequence header */
2141 ret = av_dirac_parse_sequence_header(&dsh, buf + DATA_UNIT_HEADER_SIZE, size - DATA_UNIT_HEADER_SIZE, avctx);
2142 if (ret < 0) {
2143 av_log(avctx, AV_LOG_ERROR, "error parsing sequence header");
2144 return ret;
2145 }
2146
2147 if (CALC_PADDING((int64_t)dsh->width, MAX_DWT_LEVELS) * CALC_PADDING((int64_t)dsh->height, MAX_DWT_LEVELS) * 5LL > avctx->max_pixels)
2148 ret = AVERROR(ERANGE);
2149 if (ret >= 0)
2150 ret = ff_set_dimensions(avctx, dsh->width, dsh->height);
2151 if (ret < 0) {
2152 av_freep(&dsh);
2153 return ret;
2154 }
2155
2156 ff_set_sar(avctx, dsh->sample_aspect_ratio);
2157 avctx->pix_fmt = dsh->pix_fmt;
2158 avctx->color_range = dsh->color_range;
2159 avctx->color_trc = dsh->color_trc;
2160 avctx->color_primaries = dsh->color_primaries;
2161 avctx->colorspace = dsh->colorspace;
2162 avctx->profile = dsh->profile;
2163 avctx->level = dsh->level;
2164 avctx->framerate = dsh->framerate;
2165 s->bit_depth = dsh->bit_depth;
2166 s->version.major = dsh->version.major;
2167 s->version.minor = dsh->version.minor;
2168 s->seq = *dsh;
2169 av_freep(&dsh);
2170
2171 s->pshift = s->bit_depth > 8;
2172
2173 ret = av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt,
2174 &s->chroma_x_shift,
2175 &s->chroma_y_shift);
2176 if (ret < 0)
2177 return ret;
2178
2179 ret = alloc_sequence_buffers(s);
2180 if (ret < 0)
2181 return ret;
2182
2183 s->seen_sequence_header = 1;
2184 } else if (parse_code == DIRAC_PCODE_END_SEQ) { /* [DIRAC_STD] End of Sequence */
2185 free_sequence_buffers(s);
2186 s->seen_sequence_header = 0;
2187 } else if (parse_code == DIRAC_PCODE_AUX) {
2188 if (buf[13] == 1) { /* encoder implementation/version */
2189 int ver[3];
2190 /* versions older than 1.0.8 don't store quant delta for
2191 subbands with only one codeblock */
2192 if (sscanf(buf+14, "Schroedinger %d.%d.%d", ver, ver+1, ver+2) == 3)
2193 if (ver[0] == 1 && ver[1] == 0 && ver[2] <= 7)
2194 s->old_delta_quant = 1;
2195 }
2196 } else if (parse_code & 0x8) { /* picture data unit */
2197 if (!s->seen_sequence_header) {
2198 av_log(avctx, AV_LOG_DEBUG, "Dropping frame without sequence header\n");
2199 return AVERROR_INVALIDDATA;
2200 }
2201
2202 /* find an unused frame */
2203 for (i = 0; i < MAX_FRAMES; i++)
2204 if (s->all_frames[i].avframe->data[0] == NULL)
2205 pic = &s->all_frames[i];
2206 if (!pic) {
2207 av_log(avctx, AV_LOG_ERROR, "framelist full\n");
2208 return AVERROR_INVALIDDATA;
2209 }
2210
2211 av_frame_unref(pic->avframe);
2212
2213 /* [DIRAC_STD] Defined in 9.6.1 ... */
2214 tmp = parse_code & 0x03; /* [DIRAC_STD] num_refs() */
2215 if (tmp > 2) {
2216 av_log(avctx, AV_LOG_ERROR, "num_refs of 3\n");
2217 return AVERROR_INVALIDDATA;
2218 }
2219 s->num_refs = tmp;
2220 s->is_arith = (parse_code & 0x48) == 0x08; /* [DIRAC_STD] using_ac() */
2221 s->low_delay = (parse_code & 0x88) == 0x88; /* [DIRAC_STD] is_low_delay() */
2222 s->core_syntax = (parse_code & 0x88) == 0x08; /* [DIRAC_STD] is_core_syntax() */
2223 s->ld_picture = (parse_code & 0xF8) == 0xC8; /* [DIRAC_STD] is_ld_picture() */
2224 s->hq_picture = (parse_code & 0xF8) == 0xE8; /* [DIRAC_STD] is_hq_picture() */
2225 s->dc_prediction = (parse_code & 0x28) == 0x08; /* [DIRAC_STD] using_dc_prediction() */
2226 pic->reference = (parse_code & 0x0C) == 0x0C; /* [DIRAC_STD] is_reference() */
2227 pic->avframe->key_frame = s->num_refs == 0; /* [DIRAC_STD] is_intra() */
2228 pic->avframe->pict_type = s->num_refs + 1; /* Definition of AVPictureType in avutil.h */
2229
2230 /* VC-2 Low Delay has a different parse code than the Dirac Low Delay */
2231 if (s->version.minor == 2 && parse_code == 0x88)
2232 s->ld_picture = 1;
2233
2234 if (s->low_delay && !(s->ld_picture || s->hq_picture) ) {
2235 av_log(avctx, AV_LOG_ERROR, "Invalid low delay flag\n");
2236 return AVERROR_INVALIDDATA;
2237 }
2238
2239 if ((ret = get_buffer_with_edge(avctx, pic->avframe, (parse_code & 0x0C) == 0x0C ? AV_GET_BUFFER_FLAG_REF : 0)) < 0)
2240 return ret;
2241 s->current_picture = pic;
2242 s->plane[0].stride = pic->avframe->linesize[0];
2243 s->plane[1].stride = pic->avframe->linesize[1];
2244 s->plane[2].stride = pic->avframe->linesize[2];
2245
2246 if (alloc_buffers(s, FFMAX3(FFABS(s->plane[0].stride), FFABS(s->plane[1].stride), FFABS(s->plane[2].stride))) < 0)
2247 return AVERROR(ENOMEM);
2248
2249 /* [DIRAC_STD] 11.1 Picture parse. picture_parse() */
2250 ret = dirac_decode_picture_header(s);
2251 if (ret < 0)
2252 return ret;
2253
2254 /* [DIRAC_STD] 13.0 Transform data syntax. transform_data() */
2255 ret = dirac_decode_frame_internal(s);
2256 if (ret < 0)
2257 return ret;
2258 }
2259 return 0;
2260 }
2261
dirac_decode_frame(AVCodecContext * avctx,void * data,int * got_frame,AVPacket * pkt)2262 static int dirac_decode_frame(AVCodecContext *avctx, void *data, int *got_frame, AVPacket *pkt)
2263 {
2264 DiracContext *s = avctx->priv_data;
2265 AVFrame *picture = data;
2266 uint8_t *buf = pkt->data;
2267 int buf_size = pkt->size;
2268 int i, buf_idx = 0;
2269 int ret;
2270 unsigned data_unit_size;
2271
2272 /* release unused frames */
2273 for (i = 0; i < MAX_FRAMES; i++)
2274 if (s->all_frames[i].avframe->data[0] && !s->all_frames[i].reference) {
2275 av_frame_unref(s->all_frames[i].avframe);
2276 memset(s->all_frames[i].interpolated, 0, sizeof(s->all_frames[i].interpolated));
2277 }
2278
2279 s->current_picture = NULL;
2280 *got_frame = 0;
2281
2282 /* end of stream, so flush delayed pics */
2283 if (buf_size == 0)
2284 return get_delayed_pic(s, (AVFrame *)data, got_frame);
2285
2286 for (;;) {
2287 /*[DIRAC_STD] Here starts the code from parse_info() defined in 9.6
2288 [DIRAC_STD] PARSE_INFO_PREFIX = "BBCD" as defined in ISO/IEC 646
2289 BBCD start code search */
2290 for (; buf_idx + DATA_UNIT_HEADER_SIZE < buf_size; buf_idx++) {
2291 if (buf[buf_idx ] == 'B' && buf[buf_idx+1] == 'B' &&
2292 buf[buf_idx+2] == 'C' && buf[buf_idx+3] == 'D')
2293 break;
2294 }
2295 /* BBCD found or end of data */
2296 if (buf_idx + DATA_UNIT_HEADER_SIZE >= buf_size)
2297 break;
2298
2299 data_unit_size = AV_RB32(buf+buf_idx+5);
2300 if (data_unit_size > buf_size - buf_idx || !data_unit_size) {
2301 if(data_unit_size > buf_size - buf_idx)
2302 av_log(s->avctx, AV_LOG_ERROR,
2303 "Data unit with size %d is larger than input buffer, discarding\n",
2304 data_unit_size);
2305 buf_idx += 4;
2306 continue;
2307 }
2308 /* [DIRAC_STD] dirac_decode_data_unit makes reference to the while defined in 9.3 inside the function parse_sequence() */
2309 ret = dirac_decode_data_unit(avctx, buf+buf_idx, data_unit_size);
2310 if (ret < 0)
2311 {
2312 av_log(s->avctx, AV_LOG_ERROR,"Error in dirac_decode_data_unit\n");
2313 return ret;
2314 }
2315 buf_idx += data_unit_size;
2316 }
2317
2318 if (!s->current_picture)
2319 return buf_size;
2320
2321 if (s->current_picture->avframe->display_picture_number > s->frame_number) {
2322 DiracFrame *delayed_frame = remove_frame(s->delay_frames, s->frame_number);
2323
2324 s->current_picture->reference |= DELAYED_PIC_REF;
2325
2326 if (add_frame(s->delay_frames, MAX_DELAY, s->current_picture)) {
2327 int min_num = s->delay_frames[0]->avframe->display_picture_number;
2328 /* Too many delayed frames, so we display the frame with the lowest pts */
2329 av_log(avctx, AV_LOG_ERROR, "Delay frame overflow\n");
2330
2331 for (i = 1; s->delay_frames[i]; i++)
2332 if (s->delay_frames[i]->avframe->display_picture_number < min_num)
2333 min_num = s->delay_frames[i]->avframe->display_picture_number;
2334
2335 delayed_frame = remove_frame(s->delay_frames, min_num);
2336 add_frame(s->delay_frames, MAX_DELAY, s->current_picture);
2337 }
2338
2339 if (delayed_frame) {
2340 delayed_frame->reference ^= DELAYED_PIC_REF;
2341 if((ret=av_frame_ref(data, delayed_frame->avframe)) < 0)
2342 return ret;
2343 *got_frame = 1;
2344 }
2345 } else if (s->current_picture->avframe->display_picture_number == s->frame_number) {
2346 /* The right frame at the right time :-) */
2347 if((ret=av_frame_ref(data, s->current_picture->avframe)) < 0)
2348 return ret;
2349 *got_frame = 1;
2350 }
2351
2352 if (*got_frame)
2353 s->frame_number = picture->display_picture_number + 1LL;
2354
2355 return buf_idx;
2356 }
2357
2358 AVCodec ff_dirac_decoder = {
2359 .name = "dirac",
2360 .long_name = NULL_IF_CONFIG_SMALL("BBC Dirac VC-2"),
2361 .type = AVMEDIA_TYPE_VIDEO,
2362 .id = AV_CODEC_ID_DIRAC,
2363 .priv_data_size = sizeof(DiracContext),
2364 .init = dirac_decode_init,
2365 .close = dirac_decode_end,
2366 .decode = dirac_decode_frame,
2367 .capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_SLICE_THREADS | AV_CODEC_CAP_DR1,
2368 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
2369 .flush = dirac_decode_flush,
2370 };
2371