• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 "codec_internal.h"
36 #include "internal.h"
37 #include "golomb.h"
38 #include "dirac_arith.h"
39 #include "dirac_vlc.h"
40 #include "mpeg12data.h"
41 #include "mpegpicture.h"
42 #include "mpegvideoencdsp.h"
43 #include "dirac_dwt.h"
44 #include "dirac.h"
45 #include "diractab.h"
46 #include "diracdsp.h"
47 #include "videodsp.h"
48 
49 /**
50  * The spec limits this to 3 for frame coding, but in practice can be as high as 6
51  */
52 #define MAX_REFERENCE_FRAMES 8
53 #define MAX_DELAY 5         /* limit for main profile for frame coding (TODO: field coding) */
54 #define MAX_FRAMES (MAX_REFERENCE_FRAMES + MAX_DELAY + 1)
55 #define MAX_QUANT 255        /* max quant for VC-2 */
56 #define MAX_BLOCKSIZE 32    /* maximum xblen/yblen we support */
57 
58 /**
59  * DiracBlock->ref flags, if set then the block does MC from the given ref
60  */
61 #define DIRAC_REF_MASK_REF1   1
62 #define DIRAC_REF_MASK_REF2   2
63 #define DIRAC_REF_MASK_GLOBAL 4
64 
65 /**
66  * Value of Picture.reference when Picture is not a reference picture, but
67  * is held for delayed output.
68  */
69 #define DELAYED_PIC_REF 4
70 
71 #define CALC_PADDING(size, depth)                       \
72     (((size + (1 << depth) - 1) >> depth) << depth)
73 
74 #define DIVRNDUP(a, b) (((a) + (b) - 1) / (b))
75 
76 typedef struct {
77     AVFrame *avframe;
78     int interpolated[3];    /* 1 if hpel[] is valid */
79     uint8_t *hpel[3][4];
80     uint8_t *hpel_base[3][4];
81     int reference;
82 } DiracFrame;
83 
84 typedef struct {
85     union {
86         int16_t mv[2][2];
87         int16_t dc[3];
88     } u; /* anonymous unions aren't in C99 :( */
89     uint8_t ref;
90 } DiracBlock;
91 
92 typedef struct SubBand {
93     int level;
94     int orientation;
95     int stride; /* in bytes */
96     int width;
97     int height;
98     int pshift;
99     int quant;
100     uint8_t *ibuf;
101     struct SubBand *parent;
102 
103     /* for low delay */
104     unsigned length;
105     const uint8_t *coeff_data;
106 } SubBand;
107 
108 typedef struct Plane {
109     DWTPlane idwt;
110 
111     int width;
112     int height;
113     ptrdiff_t stride;
114 
115     /* block length */
116     uint8_t xblen;
117     uint8_t yblen;
118     /* block separation (block n+1 starts after this many pixels in block n) */
119     uint8_t xbsep;
120     uint8_t ybsep;
121     /* amount of overspill on each edge (half of the overlap between blocks) */
122     uint8_t xoffset;
123     uint8_t yoffset;
124 
125     SubBand band[MAX_DWT_LEVELS][4];
126 } Plane;
127 
128 /* Used by Low Delay and High Quality profiles */
129 typedef struct DiracSlice {
130     GetBitContext gb;
131     int slice_x;
132     int slice_y;
133     int bytes;
134 } DiracSlice;
135 
136 typedef struct DiracContext {
137     AVCodecContext *avctx;
138     MpegvideoEncDSPContext mpvencdsp;
139     VideoDSPContext vdsp;
140     DiracDSPContext diracdsp;
141     DiracVersionInfo version;
142     GetBitContext gb;
143     AVDiracSeqHeader seq;
144     int seen_sequence_header;
145     int64_t frame_number;       /* number of the next frame to display       */
146     Plane plane[3];
147     int chroma_x_shift;
148     int chroma_y_shift;
149 
150     int bit_depth;              /* bit depth                                 */
151     int pshift;                 /* pixel shift = bit_depth > 8               */
152 
153     int zero_res;               /* zero residue flag                         */
154     int is_arith;               /* whether coeffs use arith or golomb coding */
155     int core_syntax;            /* use core syntax only                      */
156     int low_delay;              /* use the low delay syntax                  */
157     int hq_picture;             /* high quality picture, enables low_delay   */
158     int ld_picture;             /* use low delay picture, turns on low_delay */
159     int dc_prediction;          /* has dc prediction                         */
160     int globalmc_flag;          /* use global motion compensation            */
161     int num_refs;               /* number of reference pictures              */
162 
163     /* wavelet decoding */
164     unsigned wavelet_depth;     /* depth of the IDWT                         */
165     unsigned wavelet_idx;
166 
167     /**
168      * schroedinger older than 1.0.8 doesn't store
169      * quant delta if only one codebook exists in a band
170      */
171     unsigned old_delta_quant;
172     unsigned codeblock_mode;
173 
174     unsigned num_x;              /* number of horizontal slices               */
175     unsigned num_y;              /* number of vertical slices                 */
176 
177     uint8_t *thread_buf;         /* Per-thread buffer for coefficient storage */
178     int threads_num_buf;         /* Current # of buffers allocated            */
179     int thread_buf_size;         /* Each thread has a buffer this size        */
180 
181     DiracSlice *slice_params_buf;
182     int slice_params_num_buf;
183 
184     struct {
185         unsigned width;
186         unsigned height;
187     } codeblock[MAX_DWT_LEVELS+1];
188 
189     struct {
190         AVRational bytes;       /* average bytes per slice                   */
191         uint8_t quant[MAX_DWT_LEVELS][4]; /* [DIRAC_STD] E.1 */
192     } lowdelay;
193 
194     struct {
195         unsigned prefix_bytes;
196         uint64_t size_scaler;
197     } highquality;
198 
199     struct {
200         int pan_tilt[2];        /* pan/tilt vector                           */
201         int zrs[2][2];          /* zoom/rotate/shear matrix                  */
202         int perspective[2];     /* perspective vector                        */
203         unsigned zrs_exp;
204         unsigned perspective_exp;
205     } globalmc[2];
206 
207     /* motion compensation */
208     uint8_t mv_precision;       /* [DIRAC_STD] REFS_WT_PRECISION             */
209     int16_t weight[2];          /* [DIRAC_STD] REF1_WT and REF2_WT           */
210     unsigned weight_log2denom;  /* [DIRAC_STD] REFS_WT_PRECISION             */
211 
212     int blwidth;                /* number of blocks (horizontally)           */
213     int blheight;               /* number of blocks (vertically)             */
214     int sbwidth;                /* number of superblocks (horizontally)      */
215     int sbheight;               /* number of superblocks (vertically)        */
216 
217     uint8_t *sbsplit;
218     DiracBlock *blmotion;
219 
220     uint8_t *edge_emu_buffer[4];
221     uint8_t *edge_emu_buffer_base;
222 
223     uint16_t *mctmp;            /* buffer holding the MC data multiplied by OBMC weights */
224     uint8_t *mcscratch;
225     int buffer_stride;
226 
227     DECLARE_ALIGNED(16, uint8_t, obmc_weight)[3][MAX_BLOCKSIZE*MAX_BLOCKSIZE];
228 
229     void (*put_pixels_tab[4])(uint8_t *dst, const uint8_t *src[5], int stride, int h);
230     void (*avg_pixels_tab[4])(uint8_t *dst, const uint8_t *src[5], int stride, int h);
231     void (*add_obmc)(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
232     dirac_weight_func weight_func;
233     dirac_biweight_func biweight_func;
234 
235     DiracFrame *current_picture;
236     DiracFrame *ref_pics[2];
237 
238     DiracFrame *ref_frames[MAX_REFERENCE_FRAMES+1];
239     DiracFrame *delay_frames[MAX_DELAY+1];
240     DiracFrame all_frames[MAX_FRAMES];
241 } DiracContext;
242 
243 enum dirac_subband {
244     subband_ll = 0,
245     subband_hl = 1,
246     subband_lh = 2,
247     subband_hh = 3,
248     subband_nb,
249 };
250 
251 /* magic number division by 3 from schroedinger */
divide3(int x)252 static inline int divide3(int x)
253 {
254     return (int)((x+1U)*21845 + 10922) >> 16;
255 }
256 
remove_frame(DiracFrame * framelist[],int picnum)257 static DiracFrame *remove_frame(DiracFrame *framelist[], int picnum)
258 {
259     DiracFrame *remove_pic = NULL;
260     int i, remove_idx = -1;
261 
262     for (i = 0; framelist[i]; i++)
263         if (framelist[i]->avframe->display_picture_number == picnum) {
264             remove_pic = framelist[i];
265             remove_idx = i;
266         }
267 
268     if (remove_pic)
269         for (i = remove_idx; framelist[i]; i++)
270             framelist[i] = framelist[i+1];
271 
272     return remove_pic;
273 }
274 
add_frame(DiracFrame * framelist[],int maxframes,DiracFrame * frame)275 static int add_frame(DiracFrame *framelist[], int maxframes, DiracFrame *frame)
276 {
277     int i;
278     for (i = 0; i < maxframes; i++)
279         if (!framelist[i]) {
280             framelist[i] = frame;
281             return 0;
282         }
283     return -1;
284 }
285 
alloc_sequence_buffers(DiracContext * s)286 static int alloc_sequence_buffers(DiracContext *s)
287 {
288     int sbwidth  = DIVRNDUP(s->seq.width,  4);
289     int sbheight = DIVRNDUP(s->seq.height, 4);
290     int i, w, h, top_padding;
291 
292     /* todo: think more about this / use or set Plane here */
293     for (i = 0; i < 3; i++) {
294         int max_xblen = MAX_BLOCKSIZE >> (i ? s->chroma_x_shift : 0);
295         int max_yblen = MAX_BLOCKSIZE >> (i ? s->chroma_y_shift : 0);
296         w = s->seq.width  >> (i ? s->chroma_x_shift : 0);
297         h = s->seq.height >> (i ? s->chroma_y_shift : 0);
298 
299         /* we allocate the max we support here since num decompositions can
300          * change from frame to frame. Stride is aligned to 16 for SIMD, and
301          * 1<<MAX_DWT_LEVELS top padding to avoid if(y>0) in arith decoding
302          * MAX_BLOCKSIZE padding for MC: blocks can spill up to half of that
303          * on each side */
304         top_padding = FFMAX(1<<MAX_DWT_LEVELS, max_yblen/2);
305         w = FFALIGN(CALC_PADDING(w, MAX_DWT_LEVELS), 8); /* FIXME: Should this be 16 for SSE??? */
306         h = top_padding + CALC_PADDING(h, MAX_DWT_LEVELS) + max_yblen/2;
307 
308         s->plane[i].idwt.buf_base = av_calloc(w + max_xblen, h * (2 << s->pshift));
309         s->plane[i].idwt.tmp      = av_malloc_array((w+16), 2 << s->pshift);
310         s->plane[i].idwt.buf      = s->plane[i].idwt.buf_base + (top_padding*w)*(2 << s->pshift);
311         if (!s->plane[i].idwt.buf_base || !s->plane[i].idwt.tmp)
312             return AVERROR(ENOMEM);
313     }
314 
315     /* fixme: allocate using real stride here */
316     s->sbsplit  = av_malloc_array(sbwidth, sbheight);
317     s->blmotion = av_malloc_array(sbwidth, sbheight * 16 * sizeof(*s->blmotion));
318 
319     if (!s->sbsplit || !s->blmotion)
320         return AVERROR(ENOMEM);
321     return 0;
322 }
323 
alloc_buffers(DiracContext * s,int stride)324 static int alloc_buffers(DiracContext *s, int stride)
325 {
326     int w = s->seq.width;
327     int h = s->seq.height;
328 
329     av_assert0(stride >= w);
330     stride += 64;
331 
332     if (s->buffer_stride >= stride)
333         return 0;
334     s->buffer_stride = 0;
335 
336     av_freep(&s->edge_emu_buffer_base);
337     memset(s->edge_emu_buffer, 0, sizeof(s->edge_emu_buffer));
338     av_freep(&s->mctmp);
339     av_freep(&s->mcscratch);
340 
341     s->edge_emu_buffer_base = av_malloc_array(stride, MAX_BLOCKSIZE);
342 
343     s->mctmp     = av_malloc_array((stride+MAX_BLOCKSIZE), (h+MAX_BLOCKSIZE) * sizeof(*s->mctmp));
344     s->mcscratch = av_malloc_array(stride, MAX_BLOCKSIZE);
345 
346     if (!s->edge_emu_buffer_base || !s->mctmp || !s->mcscratch)
347         return AVERROR(ENOMEM);
348 
349     s->buffer_stride = stride;
350     return 0;
351 }
352 
free_sequence_buffers(DiracContext * s)353 static void free_sequence_buffers(DiracContext *s)
354 {
355     int i, j, k;
356 
357     for (i = 0; i < MAX_FRAMES; i++) {
358         if (s->all_frames[i].avframe->data[0]) {
359             av_frame_unref(s->all_frames[i].avframe);
360             memset(s->all_frames[i].interpolated, 0, sizeof(s->all_frames[i].interpolated));
361         }
362 
363         for (j = 0; j < 3; j++)
364             for (k = 1; k < 4; k++)
365                 av_freep(&s->all_frames[i].hpel_base[j][k]);
366     }
367 
368     memset(s->ref_frames, 0, sizeof(s->ref_frames));
369     memset(s->delay_frames, 0, sizeof(s->delay_frames));
370 
371     for (i = 0; i < 3; i++) {
372         av_freep(&s->plane[i].idwt.buf_base);
373         av_freep(&s->plane[i].idwt.tmp);
374     }
375 
376     s->buffer_stride = 0;
377     av_freep(&s->sbsplit);
378     av_freep(&s->blmotion);
379     av_freep(&s->edge_emu_buffer_base);
380 
381     av_freep(&s->mctmp);
382     av_freep(&s->mcscratch);
383 }
384 
385 static AVOnce dirac_arith_init = AV_ONCE_INIT;
386 
dirac_decode_init(AVCodecContext * avctx)387 static av_cold int dirac_decode_init(AVCodecContext *avctx)
388 {
389     DiracContext *s = avctx->priv_data;
390     int i, ret;
391 
392     s->avctx = avctx;
393     s->frame_number = -1;
394 
395     s->thread_buf = NULL;
396     s->threads_num_buf = -1;
397     s->thread_buf_size = -1;
398 
399     ff_diracdsp_init(&s->diracdsp);
400     ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx);
401     ff_videodsp_init(&s->vdsp, 8);
402 
403     for (i = 0; i < MAX_FRAMES; i++) {
404         s->all_frames[i].avframe = av_frame_alloc();
405         if (!s->all_frames[i].avframe) {
406             while (i > 0)
407                 av_frame_free(&s->all_frames[--i].avframe);
408             return AVERROR(ENOMEM);
409         }
410     }
411     ret = ff_thread_once(&dirac_arith_init, ff_dirac_init_arith_tables);
412     if (ret != 0)
413         return AVERROR_UNKNOWN;
414 
415     return 0;
416 }
417 
dirac_decode_flush(AVCodecContext * avctx)418 static void dirac_decode_flush(AVCodecContext *avctx)
419 {
420     DiracContext *s = avctx->priv_data;
421     free_sequence_buffers(s);
422     s->seen_sequence_header = 0;
423     s->frame_number = -1;
424 }
425 
dirac_decode_end(AVCodecContext * avctx)426 static av_cold int dirac_decode_end(AVCodecContext *avctx)
427 {
428     DiracContext *s = avctx->priv_data;
429     int i;
430 
431     dirac_decode_flush(avctx);
432     for (i = 0; i < MAX_FRAMES; i++)
433         av_frame_free(&s->all_frames[i].avframe);
434 
435     av_freep(&s->thread_buf);
436     av_freep(&s->slice_params_buf);
437 
438     return 0;
439 }
440 
coeff_unpack_golomb(GetBitContext * gb,int qfactor,int qoffset)441 static inline int coeff_unpack_golomb(GetBitContext *gb, int qfactor, int qoffset)
442 {
443     int coeff = dirac_get_se_golomb(gb);
444     const unsigned sign = FFSIGN(coeff);
445     if (coeff)
446         coeff = sign*((sign * coeff * qfactor + qoffset) >> 2);
447     return coeff;
448 }
449 
450 #define SIGN_CTX(x) (CTX_SIGN_ZERO + ((x) > 0) - ((x) < 0))
451 
452 #define UNPACK_ARITH(n, type) \
453     static inline void coeff_unpack_arith_##n(DiracArith *c, int qfactor, int qoffset, \
454                                               SubBand *b, type *buf, int x, int y) \
455     { \
456         int sign, sign_pred = 0, pred_ctx = CTX_ZPZN_F1; \
457         unsigned coeff; \
458         const int mstride = -(b->stride >> (1+b->pshift)); \
459         if (b->parent) { \
460             const type *pbuf = (type *)b->parent->ibuf; \
461             const int stride = b->parent->stride >> (1+b->parent->pshift); \
462             pred_ctx += !!pbuf[stride * (y>>1) + (x>>1)] << 1; \
463         } \
464         if (b->orientation == subband_hl) \
465             sign_pred = buf[mstride]; \
466         if (x) { \
467             pred_ctx += !(buf[-1] | buf[mstride] | buf[-1 + mstride]); \
468             if (b->orientation == subband_lh) \
469                 sign_pred = buf[-1]; \
470         } else { \
471             pred_ctx += !buf[mstride]; \
472         } \
473         coeff = dirac_get_arith_uint(c, pred_ctx, CTX_COEFF_DATA); \
474         if (coeff) { \
475             coeff = (coeff * qfactor + qoffset) >> 2; \
476             sign  = dirac_get_arith_bit(c, SIGN_CTX(sign_pred)); \
477             coeff = (coeff ^ -sign) + sign; \
478         } \
479         *buf = coeff; \
480     } \
481 
482 UNPACK_ARITH(8, int16_t)
483 UNPACK_ARITH(10, int32_t)
484 
485 /**
486  * Decode the coeffs in the rectangle defined by left, right, top, bottom
487  * [DIRAC_STD] 13.4.3.2 Codeblock unpacking loop. codeblock()
488  */
codeblock(DiracContext * s,SubBand * b,GetBitContext * gb,DiracArith * c,int left,int right,int top,int bottom,int blockcnt_one,int is_arith)489 static inline int codeblock(DiracContext *s, SubBand *b,
490                              GetBitContext *gb, DiracArith *c,
491                              int left, int right, int top, int bottom,
492                              int blockcnt_one, int is_arith)
493 {
494     int x, y, zero_block;
495     int qoffset, qfactor;
496     uint8_t *buf;
497 
498     /* check for any coded coefficients in this codeblock */
499     if (!blockcnt_one) {
500         if (is_arith)
501             zero_block = dirac_get_arith_bit(c, CTX_ZERO_BLOCK);
502         else
503             zero_block = get_bits1(gb);
504 
505         if (zero_block)
506             return 0;
507     }
508 
509     if (s->codeblock_mode && !(s->old_delta_quant && blockcnt_one)) {
510         int quant;
511         if (is_arith)
512             quant = dirac_get_arith_int(c, CTX_DELTA_Q_F, CTX_DELTA_Q_DATA);
513         else
514             quant = dirac_get_se_golomb(gb);
515         if (quant > INT_MAX - b->quant || b->quant + quant < 0) {
516             av_log(s->avctx, AV_LOG_ERROR, "Invalid quant\n");
517             return AVERROR_INVALIDDATA;
518         }
519         b->quant += quant;
520     }
521 
522     if (b->quant > (DIRAC_MAX_QUANT_INDEX - 1)) {
523         av_log(s->avctx, AV_LOG_ERROR, "Unsupported quant %d\n", b->quant);
524         b->quant = 0;
525         return AVERROR_INVALIDDATA;
526     }
527 
528     qfactor = ff_dirac_qscale_tab[b->quant];
529     /* TODO: context pointer? */
530     if (!s->num_refs)
531         qoffset = ff_dirac_qoffset_intra_tab[b->quant] + 2;
532     else
533         qoffset = ff_dirac_qoffset_inter_tab[b->quant] + 2;
534 
535     buf = b->ibuf + top * b->stride;
536     if (is_arith) {
537         for (y = top; y < bottom; y++) {
538             if (c->error)
539                 return c->error;
540             for (x = left; x < right; x++) {
541                 if (b->pshift) {
542                     coeff_unpack_arith_10(c, qfactor, qoffset, b, (int32_t*)(buf)+x, x, y);
543                 } else {
544                     coeff_unpack_arith_8(c, qfactor, qoffset, b, (int16_t*)(buf)+x, x, y);
545                 }
546             }
547             buf += b->stride;
548         }
549     } else {
550         for (y = top; y < bottom; y++) {
551             if (get_bits_left(gb) < 1)
552                 return AVERROR_INVALIDDATA;
553             for (x = left; x < right; x++) {
554                 int val = coeff_unpack_golomb(gb, qfactor, qoffset);
555                 if (b->pshift) {
556                     AV_WN32(&buf[4*x], val);
557                 } else {
558                     AV_WN16(&buf[2*x], val);
559                 }
560             }
561             buf += b->stride;
562          }
563      }
564      return 0;
565 }
566 
567 /**
568  * Dirac Specification ->
569  * 13.3 intra_dc_prediction(band)
570  */
571 #define INTRA_DC_PRED(n, type) \
572     static inline void intra_dc_prediction_##n(SubBand *b) \
573     { \
574         type *buf = (type*)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 >> (1+b->pshift)); \
580         \
581         for (y = 1; y < b->height; y++) { \
582             buf[0] += buf[-(b->stride >> (1+b->pshift))]; \
583             \
584             for (x = 1; x < b->width; x++) { \
585                 int pred = buf[x - 1] + buf[x - (b->stride >> (1+b->pshift))] + buf[x - (b->stride >> (1+b->pshift))-1]; \
586                 buf[x]  += divide3(pred); \
587             } \
588             buf += (b->stride >> (1+b->pshift)); \
589         } \
590     } \
591 
592 INTRA_DC_PRED(8, int16_t)
593 INTRA_DC_PRED(10, uint32_t)
594 
595 /**
596  * Dirac Specification ->
597  * 13.4.2 Non-skipped subbands.  subband_coeffs()
598  */
decode_subband_internal(DiracContext * s,SubBand * b,int is_arith)599 static av_always_inline int decode_subband_internal(DiracContext *s, SubBand *b, int is_arith)
600 {
601     int cb_x, cb_y, left, right, top, bottom;
602     DiracArith c;
603     GetBitContext gb;
604     int cb_width  = s->codeblock[b->level + (b->orientation != subband_ll)].width;
605     int cb_height = s->codeblock[b->level + (b->orientation != subband_ll)].height;
606     int blockcnt_one = (cb_width + cb_height) == 2;
607     int ret;
608 
609     if (!b->length)
610         return 0;
611 
612     init_get_bits8(&gb, b->coeff_data, b->length);
613 
614     if (is_arith)
615         ff_dirac_init_arith_decoder(&c, &gb, b->length);
616 
617     top = 0;
618     for (cb_y = 0; cb_y < cb_height; cb_y++) {
619         bottom = (b->height * (cb_y+1LL)) / cb_height;
620         left = 0;
621         for (cb_x = 0; cb_x < cb_width; cb_x++) {
622             right = (b->width * (cb_x+1LL)) / cb_width;
623             ret = codeblock(s, b, &gb, &c, left, right, top, bottom, blockcnt_one, is_arith);
624             if (ret < 0)
625                 return ret;
626             left = right;
627         }
628         top = bottom;
629     }
630 
631     if (b->orientation == subband_ll && s->num_refs == 0) {
632         if (s->pshift) {
633             intra_dc_prediction_10(b);
634         } else {
635             intra_dc_prediction_8(b);
636         }
637     }
638     return 0;
639 }
640 
decode_subband_arith(AVCodecContext * avctx,void * b)641 static int decode_subband_arith(AVCodecContext *avctx, void *b)
642 {
643     DiracContext *s = avctx->priv_data;
644     return decode_subband_internal(s, b, 1);
645 }
646 
decode_subband_golomb(AVCodecContext * avctx,void * arg)647 static int decode_subband_golomb(AVCodecContext *avctx, void *arg)
648 {
649     DiracContext *s = avctx->priv_data;
650     SubBand **b     = arg;
651     return decode_subband_internal(s, *b, 0);
652 }
653 
654 /**
655  * Dirac Specification ->
656  * [DIRAC_STD] 13.4.1 core_transform_data()
657  */
decode_component(DiracContext * s,int comp)658 static int decode_component(DiracContext *s, int comp)
659 {
660     AVCodecContext *avctx = s->avctx;
661     SubBand *bands[3*MAX_DWT_LEVELS+1];
662     enum dirac_subband orientation;
663     int level, num_bands = 0;
664     int ret[3*MAX_DWT_LEVELS+1];
665     int i;
666     int damaged_count = 0;
667 
668     /* Unpack all subbands at all levels. */
669     for (level = 0; level < s->wavelet_depth; level++) {
670         for (orientation = !!level; orientation < 4; orientation++) {
671             SubBand *b = &s->plane[comp].band[level][orientation];
672             bands[num_bands++] = b;
673 
674             align_get_bits(&s->gb);
675             /* [DIRAC_STD] 13.4.2 subband() */
676             b->length = get_interleaved_ue_golomb(&s->gb);
677             if (b->length) {
678                 b->quant = get_interleaved_ue_golomb(&s->gb);
679                 if (b->quant > (DIRAC_MAX_QUANT_INDEX - 1)) {
680                     av_log(s->avctx, AV_LOG_ERROR, "Unsupported quant %d\n", b->quant);
681                     b->quant = 0;
682                     return AVERROR_INVALIDDATA;
683                 }
684                 align_get_bits(&s->gb);
685                 b->coeff_data = s->gb.buffer + get_bits_count(&s->gb)/8;
686                 if (b->length > FFMAX(get_bits_left(&s->gb)/8, 0)) {
687                     b->length = FFMAX(get_bits_left(&s->gb)/8, 0);
688                     damaged_count ++;
689                 }
690                 skip_bits_long(&s->gb, b->length*8);
691             }
692         }
693         /* arithmetic coding has inter-level dependencies, so we can only execute one level at a time */
694         if (s->is_arith)
695             avctx->execute(avctx, decode_subband_arith, &s->plane[comp].band[level][!!level],
696                            ret + 3*level + !!level, 4-!!level, sizeof(SubBand));
697     }
698     /* golomb coding has no inter-level dependencies, so we can execute all subbands in parallel */
699     if (!s->is_arith)
700         avctx->execute(avctx, decode_subband_golomb, bands, ret, num_bands, sizeof(SubBand*));
701 
702     for (i = 0; i < s->wavelet_depth * 3 + 1; i++) {
703         if (ret[i] < 0)
704             damaged_count++;
705     }
706     if (damaged_count > (s->wavelet_depth * 3 + 1) /2)
707         return AVERROR_INVALIDDATA;
708 
709     return 0;
710 }
711 
712 #define PARSE_VALUES(type, x, gb, ebits, buf1, buf2) \
713     type *buf = (type *)buf1; \
714     buf[x] = coeff_unpack_golomb(gb, qfactor, qoffset); \
715     if (get_bits_count(gb) >= ebits) \
716         return; \
717     if (buf2) { \
718         buf = (type *)buf2; \
719         buf[x] = coeff_unpack_golomb(gb, qfactor, qoffset); \
720         if (get_bits_count(gb) >= ebits) \
721             return; \
722     } \
723 
decode_subband(DiracContext * s,GetBitContext * gb,int quant,int slice_x,int slice_y,int bits_end,SubBand * b1,SubBand * b2)724 static void decode_subband(DiracContext *s, GetBitContext *gb, int quant,
725                            int slice_x, int slice_y, int bits_end,
726                            SubBand *b1, SubBand *b2)
727 {
728     int left   = b1->width  * slice_x    / s->num_x;
729     int right  = b1->width  *(slice_x+1) / s->num_x;
730     int top    = b1->height * slice_y    / s->num_y;
731     int bottom = b1->height *(slice_y+1) / s->num_y;
732 
733     int qfactor, qoffset;
734 
735     uint8_t *buf1 =      b1->ibuf + top * b1->stride;
736     uint8_t *buf2 = b2 ? b2->ibuf + top * b2->stride: NULL;
737     int x, y;
738 
739     if (quant > (DIRAC_MAX_QUANT_INDEX - 1)) {
740         av_log(s->avctx, AV_LOG_ERROR, "Unsupported quant %d\n", quant);
741         return;
742     }
743     qfactor = ff_dirac_qscale_tab[quant];
744     qoffset = ff_dirac_qoffset_intra_tab[quant] + 2;
745     /* we have to constantly check for overread since the spec explicitly
746        requires this, with the meaning that all remaining coeffs are set to 0 */
747     if (get_bits_count(gb) >= bits_end)
748         return;
749 
750     if (s->pshift) {
751         for (y = top; y < bottom; y++) {
752             for (x = left; x < right; x++) {
753                 PARSE_VALUES(int32_t, x, gb, bits_end, buf1, buf2);
754             }
755             buf1 += b1->stride;
756             if (buf2)
757                 buf2 += b2->stride;
758         }
759     }
760     else {
761         for (y = top; y < bottom; y++) {
762             for (x = left; x < right; x++) {
763                 PARSE_VALUES(int16_t, x, gb, bits_end, buf1, buf2);
764             }
765             buf1 += b1->stride;
766             if (buf2)
767                 buf2 += b2->stride;
768         }
769     }
770 }
771 
772 /**
773  * Dirac Specification ->
774  * 13.5.2 Slices. slice(sx,sy)
775  */
decode_lowdelay_slice(AVCodecContext * avctx,void * arg)776 static int decode_lowdelay_slice(AVCodecContext *avctx, void *arg)
777 {
778     DiracContext *s = avctx->priv_data;
779     DiracSlice *slice = arg;
780     GetBitContext *gb = &slice->gb;
781     enum dirac_subband orientation;
782     int level, quant, chroma_bits, chroma_end;
783 
784     int quant_base  = get_bits(gb, 7); /*[DIRAC_STD] qindex */
785     int length_bits = av_log2(8 * slice->bytes)+1;
786     int luma_bits   = get_bits_long(gb, length_bits);
787     int luma_end    = get_bits_count(gb) + FFMIN(luma_bits, get_bits_left(gb));
788 
789     /* [DIRAC_STD] 13.5.5.2 luma_slice_band */
790     for (level = 0; level < s->wavelet_depth; level++)
791         for (orientation = !!level; orientation < 4; orientation++) {
792             quant = FFMAX(quant_base - s->lowdelay.quant[level][orientation], 0);
793             decode_subband(s, gb, quant, slice->slice_x, slice->slice_y, luma_end,
794                            &s->plane[0].band[level][orientation], NULL);
795         }
796 
797     /* consume any unused bits from luma */
798     skip_bits_long(gb, get_bits_count(gb) - luma_end);
799 
800     chroma_bits = 8*slice->bytes - 7 - length_bits - luma_bits;
801     chroma_end  = get_bits_count(gb) + FFMIN(chroma_bits, get_bits_left(gb));
802     /* [DIRAC_STD] 13.5.5.3 chroma_slice_band */
803     for (level = 0; level < s->wavelet_depth; level++)
804         for (orientation = !!level; orientation < 4; orientation++) {
805             quant = FFMAX(quant_base - s->lowdelay.quant[level][orientation], 0);
806             decode_subband(s, gb, quant, slice->slice_x, slice->slice_y, chroma_end,
807                            &s->plane[1].band[level][orientation],
808                            &s->plane[2].band[level][orientation]);
809         }
810 
811     return 0;
812 }
813 
814 typedef struct SliceCoeffs {
815     int left;
816     int top;
817     int tot_h;
818     int tot_v;
819     int tot;
820 } SliceCoeffs;
821 
subband_coeffs(DiracContext * s,int x,int y,int p,SliceCoeffs c[MAX_DWT_LEVELS])822 static int subband_coeffs(DiracContext *s, int x, int y, int p,
823                           SliceCoeffs c[MAX_DWT_LEVELS])
824 {
825     int level, coef = 0;
826     for (level = 0; level < s->wavelet_depth; level++) {
827         SliceCoeffs *o = &c[level];
828         SubBand *b = &s->plane[p].band[level][3]; /* orientation doens't matter */
829         o->top   = b->height * y / s->num_y;
830         o->left  = b->width  * x / s->num_x;
831         o->tot_h = ((b->width  * (x + 1)) / s->num_x) - o->left;
832         o->tot_v = ((b->height * (y + 1)) / s->num_y) - o->top;
833         o->tot   = o->tot_h*o->tot_v;
834         coef    += o->tot * (4 - !!level);
835     }
836     return coef;
837 }
838 
839 /**
840  * VC-2 Specification ->
841  * 13.5.3 hq_slice(sx,sy)
842  */
decode_hq_slice(DiracContext * s,DiracSlice * slice,uint8_t * tmp_buf)843 static int decode_hq_slice(DiracContext *s, DiracSlice *slice, uint8_t *tmp_buf)
844 {
845     int i, level, orientation, quant_idx;
846     int qfactor[MAX_DWT_LEVELS][4], qoffset[MAX_DWT_LEVELS][4];
847     GetBitContext *gb = &slice->gb;
848     SliceCoeffs coeffs_num[MAX_DWT_LEVELS];
849 
850     skip_bits_long(gb, 8*s->highquality.prefix_bytes);
851     quant_idx = get_bits(gb, 8);
852 
853     if (quant_idx > DIRAC_MAX_QUANT_INDEX - 1) {
854         av_log(s->avctx, AV_LOG_ERROR, "Invalid quantization index - %i\n", quant_idx);
855         return AVERROR_INVALIDDATA;
856     }
857 
858     /* Slice quantization (slice_quantizers() in the specs) */
859     for (level = 0; level < s->wavelet_depth; level++) {
860         for (orientation = !!level; orientation < 4; orientation++) {
861             const int quant = FFMAX(quant_idx - s->lowdelay.quant[level][orientation], 0);
862             qfactor[level][orientation] = ff_dirac_qscale_tab[quant];
863             qoffset[level][orientation] = ff_dirac_qoffset_intra_tab[quant] + 2;
864         }
865     }
866 
867     /* Luma + 2 Chroma planes */
868     for (i = 0; i < 3; i++) {
869         int coef_num, coef_par, off = 0;
870         int64_t length = s->highquality.size_scaler*get_bits(gb, 8);
871         int64_t bits_end = get_bits_count(gb) + 8*length;
872         const uint8_t *addr = align_get_bits(gb);
873 
874         if (length*8 > get_bits_left(gb)) {
875             av_log(s->avctx, AV_LOG_ERROR, "end too far away\n");
876             return AVERROR_INVALIDDATA;
877         }
878 
879         coef_num = subband_coeffs(s, slice->slice_x, slice->slice_y, i, coeffs_num);
880 
881         if (s->pshift)
882             coef_par = ff_dirac_golomb_read_32bit(addr, length,
883                                                   tmp_buf, coef_num);
884         else
885             coef_par = ff_dirac_golomb_read_16bit(addr, length,
886                                                   tmp_buf, coef_num);
887 
888         if (coef_num > coef_par) {
889             const int start_b = coef_par * (1 << (s->pshift + 1));
890             const int end_b   = coef_num * (1 << (s->pshift + 1));
891             memset(&tmp_buf[start_b], 0, end_b - start_b);
892         }
893 
894         for (level = 0; level < s->wavelet_depth; level++) {
895             const SliceCoeffs *c = &coeffs_num[level];
896             for (orientation = !!level; orientation < 4; orientation++) {
897                 const SubBand *b1 = &s->plane[i].band[level][orientation];
898                 uint8_t *buf = b1->ibuf + c->top * b1->stride + (c->left << (s->pshift + 1));
899 
900                 /* Change to c->tot_h <= 4 for AVX2 dequantization */
901                 const int qfunc = s->pshift + 2*(c->tot_h <= 2);
902                 s->diracdsp.dequant_subband[qfunc](&tmp_buf[off], buf, b1->stride,
903                                                    qfactor[level][orientation],
904                                                    qoffset[level][orientation],
905                                                    c->tot_v, c->tot_h);
906 
907                 off += c->tot << (s->pshift + 1);
908             }
909         }
910 
911         skip_bits_long(gb, bits_end - get_bits_count(gb));
912     }
913 
914     return 0;
915 }
916 
decode_hq_slice_row(AVCodecContext * avctx,void * arg,int jobnr,int threadnr)917 static int decode_hq_slice_row(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)
918 {
919     int i;
920     DiracContext *s = avctx->priv_data;
921     DiracSlice *slices = ((DiracSlice *)arg) + s->num_x*jobnr;
922     uint8_t *thread_buf = &s->thread_buf[s->thread_buf_size*threadnr];
923     for (i = 0; i < s->num_x; i++)
924         decode_hq_slice(s, &slices[i], thread_buf);
925     return 0;
926 }
927 
928 /**
929  * Dirac Specification ->
930  * 13.5.1 low_delay_transform_data()
931  */
decode_lowdelay(DiracContext * s)932 static int decode_lowdelay(DiracContext *s)
933 {
934     AVCodecContext *avctx = s->avctx;
935     int slice_x, slice_y, bufsize;
936     int64_t coef_buf_size, bytes = 0;
937     const uint8_t *buf;
938     DiracSlice *slices;
939     SliceCoeffs tmp[MAX_DWT_LEVELS];
940     int slice_num = 0;
941 
942     if (s->slice_params_num_buf != (s->num_x * s->num_y)) {
943         s->slice_params_buf = av_realloc_f(s->slice_params_buf, s->num_x * s->num_y, sizeof(DiracSlice));
944         if (!s->slice_params_buf) {
945             av_log(s->avctx, AV_LOG_ERROR, "slice params buffer allocation failure\n");
946             s->slice_params_num_buf = 0;
947             return AVERROR(ENOMEM);
948         }
949         s->slice_params_num_buf = s->num_x * s->num_y;
950     }
951     slices = s->slice_params_buf;
952 
953     /* 8 becacuse that's how much the golomb reader could overread junk data
954      * from another plane/slice at most, and 512 because SIMD */
955     coef_buf_size = subband_coeffs(s, s->num_x - 1, s->num_y - 1, 0, tmp) + 8;
956     coef_buf_size = (coef_buf_size << (1 + s->pshift)) + 512;
957 
958     if (s->threads_num_buf != avctx->thread_count ||
959         s->thread_buf_size != coef_buf_size) {
960         s->threads_num_buf  = avctx->thread_count;
961         s->thread_buf_size  = coef_buf_size;
962         s->thread_buf       = av_realloc_f(s->thread_buf, avctx->thread_count, s->thread_buf_size);
963         if (!s->thread_buf) {
964             av_log(s->avctx, AV_LOG_ERROR, "thread buffer allocation failure\n");
965             return AVERROR(ENOMEM);
966         }
967     }
968 
969     align_get_bits(&s->gb);
970     /*[DIRAC_STD] 13.5.2 Slices. slice(sx,sy) */
971     buf = s->gb.buffer + get_bits_count(&s->gb)/8;
972     bufsize = get_bits_left(&s->gb);
973 
974     if (s->hq_picture) {
975         int i;
976 
977         for (slice_y = 0; bufsize > 0 && slice_y < s->num_y; slice_y++) {
978             for (slice_x = 0; bufsize > 0 && slice_x < s->num_x; slice_x++) {
979                 bytes = s->highquality.prefix_bytes + 1;
980                 for (i = 0; i < 3; i++) {
981                     if (bytes <= bufsize/8)
982                         bytes += buf[bytes] * s->highquality.size_scaler + 1;
983                 }
984                 if (bytes >= INT_MAX || bytes*8 > bufsize) {
985                     av_log(s->avctx, AV_LOG_ERROR, "too many bytes\n");
986                     return AVERROR_INVALIDDATA;
987                 }
988 
989                 slices[slice_num].bytes   = bytes;
990                 slices[slice_num].slice_x = slice_x;
991                 slices[slice_num].slice_y = slice_y;
992                 init_get_bits(&slices[slice_num].gb, buf, bufsize);
993                 slice_num++;
994 
995                 buf     += bytes;
996                 if (bufsize/8 >= bytes)
997                     bufsize -= bytes*8;
998                 else
999                     bufsize = 0;
1000             }
1001         }
1002 
1003         if (s->num_x*s->num_y != slice_num) {
1004             av_log(s->avctx, AV_LOG_ERROR, "too few slices\n");
1005             return AVERROR_INVALIDDATA;
1006         }
1007 
1008         avctx->execute2(avctx, decode_hq_slice_row, slices, NULL, s->num_y);
1009     } else {
1010         for (slice_y = 0; bufsize > 0 && slice_y < s->num_y; slice_y++) {
1011             for (slice_x = 0; bufsize > 0 && slice_x < s->num_x; slice_x++) {
1012                 bytes = (slice_num+1) * (int64_t)s->lowdelay.bytes.num / s->lowdelay.bytes.den
1013                        - slice_num    * (int64_t)s->lowdelay.bytes.num / s->lowdelay.bytes.den;
1014                 if (bytes >= INT_MAX || bytes*8 > bufsize) {
1015                     av_log(s->avctx, AV_LOG_ERROR, "too many bytes\n");
1016                     return AVERROR_INVALIDDATA;
1017                 }
1018                 slices[slice_num].bytes   = bytes;
1019                 slices[slice_num].slice_x = slice_x;
1020                 slices[slice_num].slice_y = slice_y;
1021                 init_get_bits(&slices[slice_num].gb, buf, bufsize);
1022                 slice_num++;
1023 
1024                 buf     += bytes;
1025                 if (bufsize/8 >= bytes)
1026                     bufsize -= bytes*8;
1027                 else
1028                     bufsize = 0;
1029             }
1030         }
1031         avctx->execute(avctx, decode_lowdelay_slice, slices, NULL, slice_num,
1032                        sizeof(DiracSlice)); /* [DIRAC_STD] 13.5.2 Slices */
1033     }
1034 
1035     if (s->dc_prediction) {
1036         if (s->pshift) {
1037             intra_dc_prediction_10(&s->plane[0].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */
1038             intra_dc_prediction_10(&s->plane[1].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */
1039             intra_dc_prediction_10(&s->plane[2].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */
1040         } else {
1041             intra_dc_prediction_8(&s->plane[0].band[0][0]);
1042             intra_dc_prediction_8(&s->plane[1].band[0][0]);
1043             intra_dc_prediction_8(&s->plane[2].band[0][0]);
1044         }
1045     }
1046 
1047     return 0;
1048 }
1049 
init_planes(DiracContext * s)1050 static void init_planes(DiracContext *s)
1051 {
1052     int i, w, h, level, orientation;
1053 
1054     for (i = 0; i < 3; i++) {
1055         Plane *p = &s->plane[i];
1056 
1057         p->width       = s->seq.width  >> (i ? s->chroma_x_shift : 0);
1058         p->height      = s->seq.height >> (i ? s->chroma_y_shift : 0);
1059         p->idwt.width  = w = CALC_PADDING(p->width , s->wavelet_depth);
1060         p->idwt.height = h = CALC_PADDING(p->height, s->wavelet_depth);
1061         p->idwt.stride = FFALIGN(p->idwt.width, 8) << (1 + s->pshift);
1062 
1063         for (level = s->wavelet_depth-1; level >= 0; level--) {
1064             w = w>>1;
1065             h = h>>1;
1066             for (orientation = !!level; orientation < 4; orientation++) {
1067                 SubBand *b = &p->band[level][orientation];
1068 
1069                 b->pshift = s->pshift;
1070                 b->ibuf   = p->idwt.buf;
1071                 b->level  = level;
1072                 b->stride = p->idwt.stride << (s->wavelet_depth - level);
1073                 b->width  = w;
1074                 b->height = h;
1075                 b->orientation = orientation;
1076 
1077                 if (orientation & 1)
1078                     b->ibuf += w << (1+b->pshift);
1079                 if (orientation > 1)
1080                     b->ibuf += (b->stride>>1);
1081 
1082                 if (level)
1083                     b->parent = &p->band[level-1][orientation];
1084             }
1085         }
1086 
1087         if (i > 0) {
1088             p->xblen = s->plane[0].xblen >> s->chroma_x_shift;
1089             p->yblen = s->plane[0].yblen >> s->chroma_y_shift;
1090             p->xbsep = s->plane[0].xbsep >> s->chroma_x_shift;
1091             p->ybsep = s->plane[0].ybsep >> s->chroma_y_shift;
1092         }
1093 
1094         p->xoffset = (p->xblen - p->xbsep)/2;
1095         p->yoffset = (p->yblen - p->ybsep)/2;
1096     }
1097 }
1098 
1099 /**
1100  * Unpack the motion compensation parameters
1101  * Dirac Specification ->
1102  * 11.2 Picture prediction data. picture_prediction()
1103  */
dirac_unpack_prediction_parameters(DiracContext * s)1104 static int dirac_unpack_prediction_parameters(DiracContext *s)
1105 {
1106     static const uint8_t default_blen[] = { 4, 12, 16, 24 };
1107 
1108     GetBitContext *gb = &s->gb;
1109     unsigned idx, ref;
1110 
1111     align_get_bits(gb);
1112     /* [DIRAC_STD] 11.2.2 Block parameters. block_parameters() */
1113     /* Luma and Chroma are equal. 11.2.3 */
1114     idx = get_interleaved_ue_golomb(gb); /* [DIRAC_STD] index */
1115 
1116     if (idx > 4) {
1117         av_log(s->avctx, AV_LOG_ERROR, "Block prediction index too high\n");
1118         return AVERROR_INVALIDDATA;
1119     }
1120 
1121     if (idx == 0) {
1122         s->plane[0].xblen = get_interleaved_ue_golomb(gb);
1123         s->plane[0].yblen = get_interleaved_ue_golomb(gb);
1124         s->plane[0].xbsep = get_interleaved_ue_golomb(gb);
1125         s->plane[0].ybsep = get_interleaved_ue_golomb(gb);
1126     } else {
1127         /*[DIRAC_STD] preset_block_params(index). Table 11.1 */
1128         s->plane[0].xblen = default_blen[idx-1];
1129         s->plane[0].yblen = default_blen[idx-1];
1130         s->plane[0].xbsep = 4 * idx;
1131         s->plane[0].ybsep = 4 * idx;
1132     }
1133     /*[DIRAC_STD] 11.2.4 motion_data_dimensions()
1134       Calculated in function dirac_unpack_block_motion_data */
1135 
1136     if (s->plane[0].xblen % (1 << s->chroma_x_shift) != 0 ||
1137         s->plane[0].yblen % (1 << s->chroma_y_shift) != 0 ||
1138         !s->plane[0].xblen || !s->plane[0].yblen) {
1139         av_log(s->avctx, AV_LOG_ERROR,
1140                "invalid x/y block length (%d/%d) for x/y chroma shift (%d/%d)\n",
1141                s->plane[0].xblen, s->plane[0].yblen, s->chroma_x_shift, s->chroma_y_shift);
1142         return AVERROR_INVALIDDATA;
1143     }
1144     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) {
1145         av_log(s->avctx, AV_LOG_ERROR, "Block separation too small\n");
1146         return AVERROR_INVALIDDATA;
1147     }
1148     if (s->plane[0].xbsep > s->plane[0].xblen || s->plane[0].ybsep > s->plane[0].yblen) {
1149         av_log(s->avctx, AV_LOG_ERROR, "Block separation greater than size\n");
1150         return AVERROR_INVALIDDATA;
1151     }
1152     if (FFMAX(s->plane[0].xblen, s->plane[0].yblen) > MAX_BLOCKSIZE) {
1153         av_log(s->avctx, AV_LOG_ERROR, "Unsupported large block size\n");
1154         return AVERROR_PATCHWELCOME;
1155     }
1156 
1157     /*[DIRAC_STD] 11.2.5 Motion vector precision. motion_vector_precision()
1158       Read motion vector precision */
1159     s->mv_precision = get_interleaved_ue_golomb(gb);
1160     if (s->mv_precision > 3) {
1161         av_log(s->avctx, AV_LOG_ERROR, "MV precision finer than eighth-pel\n");
1162         return AVERROR_INVALIDDATA;
1163     }
1164 
1165     /*[DIRAC_STD] 11.2.6 Global motion. global_motion()
1166       Read the global motion compensation parameters */
1167     s->globalmc_flag = get_bits1(gb);
1168     if (s->globalmc_flag) {
1169         memset(s->globalmc, 0, sizeof(s->globalmc));
1170         /* [DIRAC_STD] pan_tilt(gparams) */
1171         for (ref = 0; ref < s->num_refs; ref++) {
1172             if (get_bits1(gb)) {
1173                 s->globalmc[ref].pan_tilt[0] = dirac_get_se_golomb(gb);
1174                 s->globalmc[ref].pan_tilt[1] = dirac_get_se_golomb(gb);
1175             }
1176             /* [DIRAC_STD] zoom_rotate_shear(gparams)
1177                zoom/rotation/shear parameters */
1178             if (get_bits1(gb)) {
1179                 s->globalmc[ref].zrs_exp   = get_interleaved_ue_golomb(gb);
1180                 s->globalmc[ref].zrs[0][0] = dirac_get_se_golomb(gb);
1181                 s->globalmc[ref].zrs[0][1] = dirac_get_se_golomb(gb);
1182                 s->globalmc[ref].zrs[1][0] = dirac_get_se_golomb(gb);
1183                 s->globalmc[ref].zrs[1][1] = dirac_get_se_golomb(gb);
1184             } else {
1185                 s->globalmc[ref].zrs[0][0] = 1;
1186                 s->globalmc[ref].zrs[1][1] = 1;
1187             }
1188             /* [DIRAC_STD] perspective(gparams) */
1189             if (get_bits1(gb)) {
1190                 s->globalmc[ref].perspective_exp = get_interleaved_ue_golomb(gb);
1191                 s->globalmc[ref].perspective[0]  = dirac_get_se_golomb(gb);
1192                 s->globalmc[ref].perspective[1]  = dirac_get_se_golomb(gb);
1193             }
1194             if (s->globalmc[ref].perspective_exp + (uint64_t)s->globalmc[ref].zrs_exp > 30) {
1195                 return AVERROR_INVALIDDATA;
1196             }
1197 
1198         }
1199     }
1200 
1201     /*[DIRAC_STD] 11.2.7 Picture prediction mode. prediction_mode()
1202       Picture prediction mode, not currently used. */
1203     if (get_interleaved_ue_golomb(gb)) {
1204         av_log(s->avctx, AV_LOG_ERROR, "Unknown picture prediction mode\n");
1205         return AVERROR_INVALIDDATA;
1206     }
1207 
1208     /* [DIRAC_STD] 11.2.8 Reference picture weight. reference_picture_weights()
1209        just data read, weight calculation will be done later on. */
1210     s->weight_log2denom = 1;
1211     s->weight[0]        = 1;
1212     s->weight[1]        = 1;
1213 
1214     if (get_bits1(gb)) {
1215         s->weight_log2denom = get_interleaved_ue_golomb(gb);
1216         if (s->weight_log2denom < 1 || s->weight_log2denom > 8) {
1217             av_log(s->avctx, AV_LOG_ERROR, "weight_log2denom unsupported or invalid\n");
1218             s->weight_log2denom = 1;
1219             return AVERROR_INVALIDDATA;
1220         }
1221         s->weight[0] = dirac_get_se_golomb(gb);
1222         if (s->num_refs == 2)
1223             s->weight[1] = dirac_get_se_golomb(gb);
1224     }
1225     return 0;
1226 }
1227 
1228 /**
1229  * Dirac Specification ->
1230  * 11.3 Wavelet transform data. wavelet_transform()
1231  */
dirac_unpack_idwt_params(DiracContext * s)1232 static int dirac_unpack_idwt_params(DiracContext *s)
1233 {
1234     GetBitContext *gb = &s->gb;
1235     int i, level;
1236     unsigned tmp;
1237 
1238 #define CHECKEDREAD(dst, cond, errmsg) \
1239     tmp = get_interleaved_ue_golomb(gb); \
1240     if (cond) { \
1241         av_log(s->avctx, AV_LOG_ERROR, errmsg); \
1242         return AVERROR_INVALIDDATA; \
1243     }\
1244     dst = tmp;
1245 
1246     align_get_bits(gb);
1247 
1248     s->zero_res = s->num_refs ? get_bits1(gb) : 0;
1249     if (s->zero_res)
1250         return 0;
1251 
1252     /*[DIRAC_STD] 11.3.1 Transform parameters. transform_parameters() */
1253     CHECKEDREAD(s->wavelet_idx, tmp > 6, "wavelet_idx is too big\n")
1254 
1255     CHECKEDREAD(s->wavelet_depth, tmp > MAX_DWT_LEVELS || tmp < 1, "invalid number of DWT decompositions\n")
1256 
1257     if (!s->low_delay) {
1258         /* Codeblock parameters (core syntax only) */
1259         if (get_bits1(gb)) {
1260             for (i = 0; i <= s->wavelet_depth; i++) {
1261                 CHECKEDREAD(s->codeblock[i].width , tmp < 1 || tmp > (s->avctx->width >>s->wavelet_depth-i), "codeblock width invalid\n")
1262                 CHECKEDREAD(s->codeblock[i].height, tmp < 1 || tmp > (s->avctx->height>>s->wavelet_depth-i), "codeblock height invalid\n")
1263             }
1264 
1265             CHECKEDREAD(s->codeblock_mode, tmp > 1, "unknown codeblock mode\n")
1266         }
1267         else {
1268             for (i = 0; i <= s->wavelet_depth; i++)
1269                 s->codeblock[i].width = s->codeblock[i].height = 1;
1270         }
1271     }
1272     else {
1273         s->num_x        = get_interleaved_ue_golomb(gb);
1274         s->num_y        = get_interleaved_ue_golomb(gb);
1275         if (s->num_x * s->num_y == 0 || s->num_x * (uint64_t)s->num_y > INT_MAX ||
1276             s->num_x * (uint64_t)s->avctx->width  > INT_MAX ||
1277             s->num_y * (uint64_t)s->avctx->height > INT_MAX ||
1278             s->num_x > s->avctx->width ||
1279             s->num_y > s->avctx->height
1280         ) {
1281             av_log(s->avctx,AV_LOG_ERROR,"Invalid numx/y\n");
1282             s->num_x = s->num_y = 0;
1283             return AVERROR_INVALIDDATA;
1284         }
1285         if (s->ld_picture) {
1286             s->lowdelay.bytes.num = get_interleaved_ue_golomb(gb);
1287             s->lowdelay.bytes.den = get_interleaved_ue_golomb(gb);
1288             if (s->lowdelay.bytes.den <= 0) {
1289                 av_log(s->avctx,AV_LOG_ERROR,"Invalid lowdelay.bytes.den\n");
1290                 return AVERROR_INVALIDDATA;
1291             }
1292         } else if (s->hq_picture) {
1293             s->highquality.prefix_bytes = get_interleaved_ue_golomb(gb);
1294             s->highquality.size_scaler  = get_interleaved_ue_golomb(gb);
1295             if (s->highquality.prefix_bytes >= INT_MAX / 8) {
1296                 av_log(s->avctx,AV_LOG_ERROR,"too many prefix bytes\n");
1297                 return AVERROR_INVALIDDATA;
1298             }
1299         }
1300 
1301         /* [DIRAC_STD] 11.3.5 Quantisation matrices (low-delay syntax). quant_matrix() */
1302         if (get_bits1(gb)) {
1303             av_log(s->avctx,AV_LOG_DEBUG,"Low Delay: Has Custom Quantization Matrix!\n");
1304             /* custom quantization matrix */
1305             for (level = 0; level < s->wavelet_depth; level++) {
1306                 for (i = !!level; i < 4; i++) {
1307                     s->lowdelay.quant[level][i] = get_interleaved_ue_golomb(gb);
1308                 }
1309             }
1310         } else {
1311             if (s->wavelet_depth > 4) {
1312                 av_log(s->avctx,AV_LOG_ERROR,"Mandatory custom low delay matrix missing for depth %d\n", s->wavelet_depth);
1313                 return AVERROR_INVALIDDATA;
1314             }
1315             /* default quantization matrix */
1316             for (level = 0; level < s->wavelet_depth; level++)
1317                 for (i = 0; i < 4; i++) {
1318                     s->lowdelay.quant[level][i] = ff_dirac_default_qmat[s->wavelet_idx][level][i];
1319                     /* haar with no shift differs for different depths */
1320                     if (s->wavelet_idx == 3)
1321                         s->lowdelay.quant[level][i] += 4*(s->wavelet_depth-1 - level);
1322                 }
1323         }
1324     }
1325     return 0;
1326 }
1327 
pred_sbsplit(uint8_t * sbsplit,int stride,int x,int y)1328 static inline int pred_sbsplit(uint8_t *sbsplit, int stride, int x, int y)
1329 {
1330     static const uint8_t avgsplit[7] = { 0, 0, 1, 1, 1, 2, 2 };
1331 
1332     if (!(x|y))
1333         return 0;
1334     else if (!y)
1335         return sbsplit[-1];
1336     else if (!x)
1337         return sbsplit[-stride];
1338 
1339     return avgsplit[sbsplit[-1] + sbsplit[-stride] + sbsplit[-stride-1]];
1340 }
1341 
pred_block_mode(DiracBlock * block,int stride,int x,int y,int refmask)1342 static inline int pred_block_mode(DiracBlock *block, int stride, int x, int y, int refmask)
1343 {
1344     int pred;
1345 
1346     if (!(x|y))
1347         return 0;
1348     else if (!y)
1349         return block[-1].ref & refmask;
1350     else if (!x)
1351         return block[-stride].ref & refmask;
1352 
1353     /* return the majority */
1354     pred = (block[-1].ref & refmask) + (block[-stride].ref & refmask) + (block[-stride-1].ref & refmask);
1355     return (pred >> 1) & refmask;
1356 }
1357 
pred_block_dc(DiracBlock * block,int stride,int x,int y)1358 static inline void pred_block_dc(DiracBlock *block, int stride, int x, int y)
1359 {
1360     int i, n = 0;
1361 
1362     memset(block->u.dc, 0, sizeof(block->u.dc));
1363 
1364     if (x && !(block[-1].ref & 3)) {
1365         for (i = 0; i < 3; i++)
1366             block->u.dc[i] += block[-1].u.dc[i];
1367         n++;
1368     }
1369 
1370     if (y && !(block[-stride].ref & 3)) {
1371         for (i = 0; i < 3; i++)
1372             block->u.dc[i] += block[-stride].u.dc[i];
1373         n++;
1374     }
1375 
1376     if (x && y && !(block[-1-stride].ref & 3)) {
1377         for (i = 0; i < 3; i++)
1378             block->u.dc[i] += block[-1-stride].u.dc[i];
1379         n++;
1380     }
1381 
1382     if (n == 2) {
1383         for (i = 0; i < 3; i++)
1384             block->u.dc[i] = (block->u.dc[i]+1)>>1;
1385     } else if (n == 3) {
1386         for (i = 0; i < 3; i++)
1387             block->u.dc[i] = divide3(block->u.dc[i]);
1388     }
1389 }
1390 
pred_mv(DiracBlock * block,int stride,int x,int y,int ref)1391 static inline void pred_mv(DiracBlock *block, int stride, int x, int y, int ref)
1392 {
1393     int16_t *pred[3];
1394     int refmask = ref+1;
1395     int mask = refmask | DIRAC_REF_MASK_GLOBAL; /*  exclude gmc blocks */
1396     int n = 0;
1397 
1398     if (x && (block[-1].ref & mask) == refmask)
1399         pred[n++] = block[-1].u.mv[ref];
1400 
1401     if (y && (block[-stride].ref & mask) == refmask)
1402         pred[n++] = block[-stride].u.mv[ref];
1403 
1404     if (x && y && (block[-stride-1].ref & mask) == refmask)
1405         pred[n++] = block[-stride-1].u.mv[ref];
1406 
1407     switch (n) {
1408     case 0:
1409         block->u.mv[ref][0] = 0;
1410         block->u.mv[ref][1] = 0;
1411         break;
1412     case 1:
1413         block->u.mv[ref][0] = pred[0][0];
1414         block->u.mv[ref][1] = pred[0][1];
1415         break;
1416     case 2:
1417         block->u.mv[ref][0] = (pred[0][0] + pred[1][0] + 1) >> 1;
1418         block->u.mv[ref][1] = (pred[0][1] + pred[1][1] + 1) >> 1;
1419         break;
1420     case 3:
1421         block->u.mv[ref][0] = mid_pred(pred[0][0], pred[1][0], pred[2][0]);
1422         block->u.mv[ref][1] = mid_pred(pred[0][1], pred[1][1], pred[2][1]);
1423         break;
1424     }
1425 }
1426 
global_mv(DiracContext * s,DiracBlock * block,int x,int y,int ref)1427 static void global_mv(DiracContext *s, DiracBlock *block, int x, int y, int ref)
1428 {
1429     int ez      = s->globalmc[ref].zrs_exp;
1430     int ep      = s->globalmc[ref].perspective_exp;
1431     int (*A)[2] = s->globalmc[ref].zrs;
1432     int *b      = s->globalmc[ref].pan_tilt;
1433     int *c      = s->globalmc[ref].perspective;
1434 
1435     int64_t m   = (1<<ep) - (c[0]*(int64_t)x + c[1]*(int64_t)y);
1436     int64_t mx  = m * (uint64_t)((A[0][0] * (int64_t)x + A[0][1]*(int64_t)y) + (1LL<<ez) * b[0]);
1437     int64_t my  = m * (uint64_t)((A[1][0] * (int64_t)x + A[1][1]*(int64_t)y) + (1LL<<ez) * b[1]);
1438 
1439     block->u.mv[ref][0] = (mx + (1<<(ez+ep))) >> (ez+ep);
1440     block->u.mv[ref][1] = (my + (1<<(ez+ep))) >> (ez+ep);
1441 }
1442 
decode_block_params(DiracContext * s,DiracArith arith[8],DiracBlock * block,int stride,int x,int y)1443 static void decode_block_params(DiracContext *s, DiracArith arith[8], DiracBlock *block,
1444                                 int stride, int x, int y)
1445 {
1446     int i;
1447 
1448     block->ref  = pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_REF1);
1449     block->ref ^= dirac_get_arith_bit(arith, CTX_PMODE_REF1);
1450 
1451     if (s->num_refs == 2) {
1452         block->ref |= pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_REF2);
1453         block->ref ^= dirac_get_arith_bit(arith, CTX_PMODE_REF2) << 1;
1454     }
1455 
1456     if (!block->ref) {
1457         pred_block_dc(block, stride, x, y);
1458         for (i = 0; i < 3; i++)
1459             block->u.dc[i] += (unsigned)dirac_get_arith_int(arith+1+i, CTX_DC_F1, CTX_DC_DATA);
1460         return;
1461     }
1462 
1463     if (s->globalmc_flag) {
1464         block->ref |= pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_GLOBAL);
1465         block->ref ^= dirac_get_arith_bit(arith, CTX_GLOBAL_BLOCK) << 2;
1466     }
1467 
1468     for (i = 0; i < s->num_refs; i++)
1469         if (block->ref & (i+1)) {
1470             if (block->ref & DIRAC_REF_MASK_GLOBAL) {
1471                 global_mv(s, block, x, y, i);
1472             } else {
1473                 pred_mv(block, stride, x, y, i);
1474                 block->u.mv[i][0] += (unsigned)dirac_get_arith_int(arith + 4 + 2 * i, CTX_MV_F1, CTX_MV_DATA);
1475                 block->u.mv[i][1] += (unsigned)dirac_get_arith_int(arith + 5 + 2 * i, CTX_MV_F1, CTX_MV_DATA);
1476             }
1477         }
1478 }
1479 
1480 /**
1481  * Copies the current block to the other blocks covered by the current superblock split mode
1482  */
propagate_block_data(DiracBlock * block,int stride,int size)1483 static void propagate_block_data(DiracBlock *block, int stride, int size)
1484 {
1485     int x, y;
1486     DiracBlock *dst = block;
1487 
1488     for (x = 1; x < size; x++)
1489         dst[x] = *block;
1490 
1491     for (y = 1; y < size; y++) {
1492         dst += stride;
1493         for (x = 0; x < size; x++)
1494             dst[x] = *block;
1495     }
1496 }
1497 
1498 /**
1499  * Dirac Specification ->
1500  * 12. Block motion data syntax
1501  */
dirac_unpack_block_motion_data(DiracContext * s)1502 static int dirac_unpack_block_motion_data(DiracContext *s)
1503 {
1504     GetBitContext *gb = &s->gb;
1505     uint8_t *sbsplit = s->sbsplit;
1506     int i, x, y, q, p;
1507     DiracArith arith[8];
1508 
1509     align_get_bits(gb);
1510 
1511     /* [DIRAC_STD] 11.2.4 and 12.2.1 Number of blocks and superblocks */
1512     s->sbwidth  = DIVRNDUP(s->seq.width,  4*s->plane[0].xbsep);
1513     s->sbheight = DIVRNDUP(s->seq.height, 4*s->plane[0].ybsep);
1514     s->blwidth  = 4 * s->sbwidth;
1515     s->blheight = 4 * s->sbheight;
1516 
1517     /* [DIRAC_STD] 12.3.1 Superblock splitting modes. superblock_split_modes()
1518        decode superblock split modes */
1519     ff_dirac_init_arith_decoder(arith, gb, get_interleaved_ue_golomb(gb));     /* get_interleaved_ue_golomb(gb) is the length */
1520     for (y = 0; y < s->sbheight; y++) {
1521         for (x = 0; x < s->sbwidth; x++) {
1522             unsigned int split  = dirac_get_arith_uint(arith, CTX_SB_F1, CTX_SB_DATA);
1523             if (split > 2)
1524                 return AVERROR_INVALIDDATA;
1525             sbsplit[x] = (split + pred_sbsplit(sbsplit+x, s->sbwidth, x, y)) % 3;
1526         }
1527         sbsplit += s->sbwidth;
1528     }
1529 
1530     /* setup arith decoding */
1531     ff_dirac_init_arith_decoder(arith, gb, get_interleaved_ue_golomb(gb));
1532     for (i = 0; i < s->num_refs; i++) {
1533         ff_dirac_init_arith_decoder(arith + 4 + 2 * i, gb, get_interleaved_ue_golomb(gb));
1534         ff_dirac_init_arith_decoder(arith + 5 + 2 * i, gb, get_interleaved_ue_golomb(gb));
1535     }
1536     for (i = 0; i < 3; i++)
1537         ff_dirac_init_arith_decoder(arith+1+i, gb, get_interleaved_ue_golomb(gb));
1538 
1539     for (y = 0; y < s->sbheight; y++)
1540         for (x = 0; x < s->sbwidth; x++) {
1541             int blkcnt = 1 << s->sbsplit[y * s->sbwidth + x];
1542             int step   = 4 >> s->sbsplit[y * s->sbwidth + x];
1543 
1544             for (q = 0; q < blkcnt; q++)
1545                 for (p = 0; p < blkcnt; p++) {
1546                     int bx = 4 * x + p*step;
1547                     int by = 4 * y + q*step;
1548                     DiracBlock *block = &s->blmotion[by*s->blwidth + bx];
1549                     decode_block_params(s, arith, block, s->blwidth, bx, by);
1550                     propagate_block_data(block, s->blwidth, step);
1551                 }
1552         }
1553 
1554     for (i = 0; i < 4 + 2*s->num_refs; i++) {
1555         if (arith[i].error)
1556             return arith[i].error;
1557     }
1558 
1559     return 0;
1560 }
1561 
weight(int i,int blen,int offset)1562 static int weight(int i, int blen, int offset)
1563 {
1564 #define ROLLOFF(i) offset == 1 ? ((i) ? 5 : 3) :        \
1565     (1 + (6*(i) + offset - 1) / (2*offset - 1))
1566 
1567     if (i < 2*offset)
1568         return ROLLOFF(i);
1569     else if (i > blen-1 - 2*offset)
1570         return ROLLOFF(blen-1 - i);
1571     return 8;
1572 }
1573 
init_obmc_weight_row(Plane * p,uint8_t * obmc_weight,int stride,int left,int right,int wy)1574 static void init_obmc_weight_row(Plane *p, uint8_t *obmc_weight, int stride,
1575                                  int left, int right, int wy)
1576 {
1577     int x;
1578     for (x = 0; left && x < p->xblen >> 1; x++)
1579         obmc_weight[x] = wy*8;
1580     for (; x < p->xblen >> right; x++)
1581         obmc_weight[x] = wy*weight(x, p->xblen, p->xoffset);
1582     for (; x < p->xblen; x++)
1583         obmc_weight[x] = wy*8;
1584     for (; x < stride; x++)
1585         obmc_weight[x] = 0;
1586 }
1587 
init_obmc_weight(Plane * p,uint8_t * obmc_weight,int stride,int left,int right,int top,int bottom)1588 static void init_obmc_weight(Plane *p, uint8_t *obmc_weight, int stride,
1589                              int left, int right, int top, int bottom)
1590 {
1591     int y;
1592     for (y = 0; top && y < p->yblen >> 1; y++) {
1593         init_obmc_weight_row(p, obmc_weight, stride, left, right, 8);
1594         obmc_weight += stride;
1595     }
1596     for (; y < p->yblen >> bottom; y++) {
1597         int wy = weight(y, p->yblen, p->yoffset);
1598         init_obmc_weight_row(p, obmc_weight, stride, left, right, wy);
1599         obmc_weight += stride;
1600     }
1601     for (; y < p->yblen; y++) {
1602         init_obmc_weight_row(p, obmc_weight, stride, left, right, 8);
1603         obmc_weight += stride;
1604     }
1605 }
1606 
init_obmc_weights(DiracContext * s,Plane * p,int by)1607 static void init_obmc_weights(DiracContext *s, Plane *p, int by)
1608 {
1609     int top = !by;
1610     int bottom = by == s->blheight-1;
1611 
1612     /* don't bother re-initing for rows 2 to blheight-2, the weights don't change */
1613     if (top || bottom || by == 1) {
1614         init_obmc_weight(p, s->obmc_weight[0], MAX_BLOCKSIZE, 1, 0, top, bottom);
1615         init_obmc_weight(p, s->obmc_weight[1], MAX_BLOCKSIZE, 0, 0, top, bottom);
1616         init_obmc_weight(p, s->obmc_weight[2], MAX_BLOCKSIZE, 0, 1, top, bottom);
1617     }
1618 }
1619 
1620 static const uint8_t epel_weights[4][4][4] = {
1621     {{ 16,  0,  0,  0 },
1622      { 12,  4,  0,  0 },
1623      {  8,  8,  0,  0 },
1624      {  4, 12,  0,  0 }},
1625     {{ 12,  0,  4,  0 },
1626      {  9,  3,  3,  1 },
1627      {  6,  6,  2,  2 },
1628      {  3,  9,  1,  3 }},
1629     {{  8,  0,  8,  0 },
1630      {  6,  2,  6,  2 },
1631      {  4,  4,  4,  4 },
1632      {  2,  6,  2,  6 }},
1633     {{  4,  0, 12,  0 },
1634      {  3,  1,  9,  3 },
1635      {  2,  2,  6,  6 },
1636      {  1,  3,  3,  9 }}
1637 };
1638 
1639 /**
1640  * For block x,y, determine which of the hpel planes to do bilinear
1641  * interpolation from and set src[] to the location in each hpel plane
1642  * to MC from.
1643  *
1644  * @return the index of the put_dirac_pixels_tab function to use
1645  *  0 for 1 plane (fpel,hpel), 1 for 2 planes (qpel), 2 for 4 planes (qpel), and 3 for epel
1646  */
mc_subpel(DiracContext * s,DiracBlock * block,const uint8_t * src[5],int x,int y,int ref,int plane)1647 static int mc_subpel(DiracContext *s, DiracBlock *block, const uint8_t *src[5],
1648                      int x, int y, int ref, int plane)
1649 {
1650     Plane *p = &s->plane[plane];
1651     uint8_t **ref_hpel = s->ref_pics[ref]->hpel[plane];
1652     int motion_x = block->u.mv[ref][0];
1653     int motion_y = block->u.mv[ref][1];
1654     int mx, my, i, epel, nplanes = 0;
1655 
1656     if (plane) {
1657         motion_x >>= s->chroma_x_shift;
1658         motion_y >>= s->chroma_y_shift;
1659     }
1660 
1661     mx         = motion_x & ~(-1U << s->mv_precision);
1662     my         = motion_y & ~(-1U << s->mv_precision);
1663     motion_x >>= s->mv_precision;
1664     motion_y >>= s->mv_precision;
1665     /* normalize subpel coordinates to epel */
1666     /* TODO: template this function? */
1667     mx      <<= 3 - s->mv_precision;
1668     my      <<= 3 - s->mv_precision;
1669 
1670     x += motion_x;
1671     y += motion_y;
1672     epel = (mx|my)&1;
1673 
1674     /* hpel position */
1675     if (!((mx|my)&3)) {
1676         nplanes = 1;
1677         src[0] = ref_hpel[(my>>1)+(mx>>2)] + y*p->stride + x;
1678     } else {
1679         /* qpel or epel */
1680         nplanes = 4;
1681         for (i = 0; i < 4; i++)
1682             src[i] = ref_hpel[i] + y*p->stride + x;
1683 
1684         /* if we're interpolating in the right/bottom halves, adjust the planes as needed
1685            we increment x/y because the edge changes for half of the pixels */
1686         if (mx > 4) {
1687             src[0] += 1;
1688             src[2] += 1;
1689             x++;
1690         }
1691         if (my > 4) {
1692             src[0] += p->stride;
1693             src[1] += p->stride;
1694             y++;
1695         }
1696 
1697         /* hpel planes are:
1698            [0]: F  [1]: H
1699            [2]: V  [3]: C */
1700         if (!epel) {
1701             /* check if we really only need 2 planes since either mx or my is
1702                a hpel position. (epel weights of 0 handle this there) */
1703             if (!(mx&3)) {
1704                 /* mx == 0: average [0] and [2]
1705                    mx == 4: average [1] and [3] */
1706                 src[!mx] = src[2 + !!mx];
1707                 nplanes = 2;
1708             } else if (!(my&3)) {
1709                 src[0] = src[(my>>1)  ];
1710                 src[1] = src[(my>>1)+1];
1711                 nplanes = 2;
1712             }
1713         } else {
1714             /* adjust the ordering if needed so the weights work */
1715             if (mx > 4) {
1716                 FFSWAP(const uint8_t *, src[0], src[1]);
1717                 FFSWAP(const uint8_t *, src[2], src[3]);
1718             }
1719             if (my > 4) {
1720                 FFSWAP(const uint8_t *, src[0], src[2]);
1721                 FFSWAP(const uint8_t *, src[1], src[3]);
1722             }
1723             src[4] = epel_weights[my&3][mx&3];
1724         }
1725     }
1726 
1727     /* fixme: v/h _edge_pos */
1728     if (x + p->xblen > p->width +EDGE_WIDTH/2 ||
1729         y + p->yblen > p->height+EDGE_WIDTH/2 ||
1730         x < 0 || y < 0) {
1731         for (i = 0; i < nplanes; i++) {
1732             s->vdsp.emulated_edge_mc(s->edge_emu_buffer[i], src[i],
1733                                      p->stride, p->stride,
1734                                      p->xblen, p->yblen, x, y,
1735                                      p->width+EDGE_WIDTH/2, p->height+EDGE_WIDTH/2);
1736             src[i] = s->edge_emu_buffer[i];
1737         }
1738     }
1739     return (nplanes>>1) + epel;
1740 }
1741 
add_dc(uint16_t * dst,int dc,int stride,uint8_t * obmc_weight,int xblen,int yblen)1742 static void add_dc(uint16_t *dst, int dc, int stride,
1743                    uint8_t *obmc_weight, int xblen, int yblen)
1744 {
1745     int x, y;
1746     dc += 128;
1747 
1748     for (y = 0; y < yblen; y++) {
1749         for (x = 0; x < xblen; x += 2) {
1750             dst[x  ] += dc * obmc_weight[x  ];
1751             dst[x+1] += dc * obmc_weight[x+1];
1752         }
1753         dst          += stride;
1754         obmc_weight  += MAX_BLOCKSIZE;
1755     }
1756 }
1757 
block_mc(DiracContext * s,DiracBlock * block,uint16_t * mctmp,uint8_t * obmc_weight,int plane,int dstx,int dsty)1758 static void block_mc(DiracContext *s, DiracBlock *block,
1759                      uint16_t *mctmp, uint8_t *obmc_weight,
1760                      int plane, int dstx, int dsty)
1761 {
1762     Plane *p = &s->plane[plane];
1763     const uint8_t *src[5];
1764     int idx;
1765 
1766     switch (block->ref&3) {
1767     case 0: /* DC */
1768         add_dc(mctmp, block->u.dc[plane], p->stride, obmc_weight, p->xblen, p->yblen);
1769         return;
1770     case 1:
1771     case 2:
1772         idx = mc_subpel(s, block, src, dstx, dsty, (block->ref&3)-1, plane);
1773         s->put_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);
1774         if (s->weight_func)
1775             s->weight_func(s->mcscratch, p->stride, s->weight_log2denom,
1776                            s->weight[0] + s->weight[1], p->yblen);
1777         break;
1778     case 3:
1779         idx = mc_subpel(s, block, src, dstx, dsty, 0, plane);
1780         s->put_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);
1781         idx = mc_subpel(s, block, src, dstx, dsty, 1, plane);
1782         if (s->biweight_func) {
1783             /* fixme: +32 is a quick hack */
1784             s->put_pixels_tab[idx](s->mcscratch + 32, src, p->stride, p->yblen);
1785             s->biweight_func(s->mcscratch, s->mcscratch+32, p->stride, s->weight_log2denom,
1786                              s->weight[0], s->weight[1], p->yblen);
1787         } else
1788             s->avg_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);
1789         break;
1790     }
1791     s->add_obmc(mctmp, s->mcscratch, p->stride, obmc_weight, p->yblen);
1792 }
1793 
mc_row(DiracContext * s,DiracBlock * block,uint16_t * mctmp,int plane,int dsty)1794 static void mc_row(DiracContext *s, DiracBlock *block, uint16_t *mctmp, int plane, int dsty)
1795 {
1796     Plane *p = &s->plane[plane];
1797     int x, dstx = p->xbsep - p->xoffset;
1798 
1799     block_mc(s, block, mctmp, s->obmc_weight[0], plane, -p->xoffset, dsty);
1800     mctmp += p->xbsep;
1801 
1802     for (x = 1; x < s->blwidth-1; x++) {
1803         block_mc(s, block+x, mctmp, s->obmc_weight[1], plane, dstx, dsty);
1804         dstx  += p->xbsep;
1805         mctmp += p->xbsep;
1806     }
1807     block_mc(s, block+x, mctmp, s->obmc_weight[2], plane, dstx, dsty);
1808 }
1809 
select_dsp_funcs(DiracContext * s,int width,int height,int xblen,int yblen)1810 static void select_dsp_funcs(DiracContext *s, int width, int height, int xblen, int yblen)
1811 {
1812     int idx = 0;
1813     if (xblen > 8)
1814         idx = 1;
1815     if (xblen > 16)
1816         idx = 2;
1817 
1818     memcpy(s->put_pixels_tab, s->diracdsp.put_dirac_pixels_tab[idx], sizeof(s->put_pixels_tab));
1819     memcpy(s->avg_pixels_tab, s->diracdsp.avg_dirac_pixels_tab[idx], sizeof(s->avg_pixels_tab));
1820     s->add_obmc = s->diracdsp.add_dirac_obmc[idx];
1821     if (s->weight_log2denom > 1 || s->weight[0] != 1 || s->weight[1] != 1) {
1822         s->weight_func   = s->diracdsp.weight_dirac_pixels_tab[idx];
1823         s->biweight_func = s->diracdsp.biweight_dirac_pixels_tab[idx];
1824     } else {
1825         s->weight_func   = NULL;
1826         s->biweight_func = NULL;
1827     }
1828 }
1829 
interpolate_refplane(DiracContext * s,DiracFrame * ref,int plane,int width,int height)1830 static int interpolate_refplane(DiracContext *s, DiracFrame *ref, int plane, int width, int height)
1831 {
1832     /* chroma allocates an edge of 8 when subsampled
1833        which for 4:2:2 means an h edge of 16 and v edge of 8
1834        just use 8 for everything for the moment */
1835     int i, edge = EDGE_WIDTH/2;
1836 
1837     ref->hpel[plane][0] = ref->avframe->data[plane];
1838     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 */
1839 
1840     /* no need for hpel if we only have fpel vectors */
1841     if (!s->mv_precision)
1842         return 0;
1843 
1844     for (i = 1; i < 4; i++) {
1845         if (!ref->hpel_base[plane][i])
1846             ref->hpel_base[plane][i] = av_malloc((height+2*edge) * ref->avframe->linesize[plane] + 32);
1847         if (!ref->hpel_base[plane][i]) {
1848             return AVERROR(ENOMEM);
1849         }
1850         /* we need to be 16-byte aligned even for chroma */
1851         ref->hpel[plane][i] = ref->hpel_base[plane][i] + edge*ref->avframe->linesize[plane] + 16;
1852     }
1853 
1854     if (!ref->interpolated[plane]) {
1855         s->diracdsp.dirac_hpel_filter(ref->hpel[plane][1], ref->hpel[plane][2],
1856                                       ref->hpel[plane][3], ref->hpel[plane][0],
1857                                       ref->avframe->linesize[plane], width, height);
1858         s->mpvencdsp.draw_edges(ref->hpel[plane][1], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
1859         s->mpvencdsp.draw_edges(ref->hpel[plane][2], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
1860         s->mpvencdsp.draw_edges(ref->hpel[plane][3], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);
1861     }
1862     ref->interpolated[plane] = 1;
1863 
1864     return 0;
1865 }
1866 
1867 /**
1868  * Dirac Specification ->
1869  * 13.0 Transform data syntax. transform_data()
1870  */
dirac_decode_frame_internal(DiracContext * s)1871 static int dirac_decode_frame_internal(DiracContext *s)
1872 {
1873     DWTContext d;
1874     int y, i, comp, dsty;
1875     int ret;
1876 
1877     if (s->low_delay) {
1878         /* [DIRAC_STD] 13.5.1 low_delay_transform_data() */
1879         if (!s->hq_picture) {
1880             for (comp = 0; comp < 3; comp++) {
1881                 Plane *p = &s->plane[comp];
1882                 memset(p->idwt.buf, 0, p->idwt.stride * p->idwt.height);
1883             }
1884         }
1885         if (!s->zero_res) {
1886             if ((ret = decode_lowdelay(s)) < 0)
1887                 return ret;
1888         }
1889     }
1890 
1891     for (comp = 0; comp < 3; comp++) {
1892         Plane *p       = &s->plane[comp];
1893         uint8_t *frame = s->current_picture->avframe->data[comp];
1894 
1895         /* FIXME: small resolutions */
1896         for (i = 0; i < 4; i++)
1897             s->edge_emu_buffer[i] = s->edge_emu_buffer_base + i*FFALIGN(p->width, 16);
1898 
1899         if (!s->zero_res && !s->low_delay)
1900         {
1901             memset(p->idwt.buf, 0, p->idwt.stride * p->idwt.height);
1902             ret = decode_component(s, comp); /* [DIRAC_STD] 13.4.1 core_transform_data() */
1903             if (ret < 0)
1904                 return ret;
1905         }
1906         ret = ff_spatial_idwt_init(&d, &p->idwt, s->wavelet_idx+2,
1907                                    s->wavelet_depth, s->bit_depth);
1908         if (ret < 0)
1909             return ret;
1910 
1911         if (!s->num_refs) { /* intra */
1912             for (y = 0; y < p->height; y += 16) {
1913                 int idx = (s->bit_depth - 8) >> 1;
1914                 ff_spatial_idwt_slice2(&d, y+16); /* decode */
1915                 s->diracdsp.put_signed_rect_clamped[idx](frame + y*p->stride,
1916                                                          p->stride,
1917                                                          p->idwt.buf + y*p->idwt.stride,
1918                                                          p->idwt.stride, p->width, 16);
1919             }
1920         } else { /* inter */
1921             int rowheight = p->ybsep*p->stride;
1922 
1923             select_dsp_funcs(s, p->width, p->height, p->xblen, p->yblen);
1924 
1925             for (i = 0; i < s->num_refs; i++) {
1926                 int ret = interpolate_refplane(s, s->ref_pics[i], comp, p->width, p->height);
1927                 if (ret < 0)
1928                     return ret;
1929             }
1930 
1931             memset(s->mctmp, 0, 4*p->yoffset*p->stride);
1932 
1933             dsty = -p->yoffset;
1934             for (y = 0; y < s->blheight; y++) {
1935                 int h     = 0,
1936                     start = FFMAX(dsty, 0);
1937                 uint16_t *mctmp    = s->mctmp + y*rowheight;
1938                 DiracBlock *blocks = s->blmotion + y*s->blwidth;
1939 
1940                 init_obmc_weights(s, p, y);
1941 
1942                 if (y == s->blheight-1 || start+p->ybsep > p->height)
1943                     h = p->height - start;
1944                 else
1945                     h = p->ybsep - (start - dsty);
1946                 if (h < 0)
1947                     break;
1948 
1949                 memset(mctmp+2*p->yoffset*p->stride, 0, 2*rowheight);
1950                 mc_row(s, blocks, mctmp, comp, dsty);
1951 
1952                 mctmp += (start - dsty)*p->stride + p->xoffset;
1953                 ff_spatial_idwt_slice2(&d, start + h); /* decode */
1954                 /* NOTE: add_rect_clamped hasn't been templated hence the shifts.
1955                  * idwt.stride is passed as pixels, not in bytes as in the rest of the decoder */
1956                 s->diracdsp.add_rect_clamped(frame + start*p->stride, mctmp, p->stride,
1957                                              (int16_t*)(p->idwt.buf) + start*(p->idwt.stride >> 1), (p->idwt.stride >> 1), p->width, h);
1958 
1959                 dsty += p->ybsep;
1960             }
1961         }
1962     }
1963 
1964 
1965     return 0;
1966 }
1967 
get_buffer_with_edge(AVCodecContext * avctx,AVFrame * f,int flags)1968 static int get_buffer_with_edge(AVCodecContext *avctx, AVFrame *f, int flags)
1969 {
1970     int ret, i;
1971     int chroma_x_shift, chroma_y_shift;
1972     ret = av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &chroma_x_shift,
1973                                            &chroma_y_shift);
1974     if (ret < 0)
1975         return ret;
1976 
1977     f->width  = avctx->width  + 2 * EDGE_WIDTH;
1978     f->height = avctx->height + 2 * EDGE_WIDTH + 2;
1979     ret = ff_get_buffer(avctx, f, flags);
1980     if (ret < 0)
1981         return ret;
1982 
1983     for (i = 0; f->data[i]; i++) {
1984         int offset = (EDGE_WIDTH >> (i && i<3 ? chroma_y_shift : 0)) *
1985                      f->linesize[i] + 32;
1986         f->data[i] += offset;
1987     }
1988     f->width  = avctx->width;
1989     f->height = avctx->height;
1990 
1991     return 0;
1992 }
1993 
1994 /**
1995  * Dirac Specification ->
1996  * 11.1.1 Picture Header. picture_header()
1997  */
dirac_decode_picture_header(DiracContext * s)1998 static int dirac_decode_picture_header(DiracContext *s)
1999 {
2000     unsigned retire, picnum;
2001     int i, j, ret;
2002     int64_t refdist, refnum;
2003     GetBitContext *gb = &s->gb;
2004 
2005     /* [DIRAC_STD] 11.1.1 Picture Header. picture_header() PICTURE_NUM */
2006     picnum = s->current_picture->avframe->display_picture_number = get_bits_long(gb, 32);
2007 
2008 
2009     av_log(s->avctx,AV_LOG_DEBUG,"PICTURE_NUM: %d\n",picnum);
2010 
2011     /* if this is the first keyframe after a sequence header, start our
2012        reordering from here */
2013     if (s->frame_number < 0)
2014         s->frame_number = picnum;
2015 
2016     s->ref_pics[0] = s->ref_pics[1] = NULL;
2017     for (i = 0; i < s->num_refs; i++) {
2018         refnum = (picnum + dirac_get_se_golomb(gb)) & 0xFFFFFFFF;
2019         refdist = INT64_MAX;
2020 
2021         /* find the closest reference to the one we want */
2022         /* Jordi: this is needed if the referenced picture hasn't yet arrived */
2023         for (j = 0; j < MAX_REFERENCE_FRAMES && refdist; j++)
2024             if (s->ref_frames[j]
2025                 && FFABS(s->ref_frames[j]->avframe->display_picture_number - refnum) < refdist) {
2026                 s->ref_pics[i] = s->ref_frames[j];
2027                 refdist = FFABS(s->ref_frames[j]->avframe->display_picture_number - refnum);
2028             }
2029 
2030         if (!s->ref_pics[i] || refdist)
2031             av_log(s->avctx, AV_LOG_DEBUG, "Reference not found\n");
2032 
2033         /* if there were no references at all, allocate one */
2034         if (!s->ref_pics[i])
2035             for (j = 0; j < MAX_FRAMES; j++)
2036                 if (!s->all_frames[j].avframe->data[0]) {
2037                     s->ref_pics[i] = &s->all_frames[j];
2038                     ret = get_buffer_with_edge(s->avctx, s->ref_pics[i]->avframe, AV_GET_BUFFER_FLAG_REF);
2039                     if (ret < 0)
2040                         return ret;
2041                     break;
2042                 }
2043 
2044         if (!s->ref_pics[i]) {
2045             av_log(s->avctx, AV_LOG_ERROR, "Reference could not be allocated\n");
2046             return AVERROR_INVALIDDATA;
2047         }
2048 
2049     }
2050 
2051     /* retire the reference frames that are not used anymore */
2052     if (s->current_picture->reference) {
2053         retire = (picnum + dirac_get_se_golomb(gb)) & 0xFFFFFFFF;
2054         if (retire != picnum) {
2055             DiracFrame *retire_pic = remove_frame(s->ref_frames, retire);
2056 
2057             if (retire_pic)
2058                 retire_pic->reference &= DELAYED_PIC_REF;
2059             else
2060                 av_log(s->avctx, AV_LOG_DEBUG, "Frame to retire not found\n");
2061         }
2062 
2063         /* if reference array is full, remove the oldest as per the spec */
2064         while (add_frame(s->ref_frames, MAX_REFERENCE_FRAMES, s->current_picture)) {
2065             av_log(s->avctx, AV_LOG_ERROR, "Reference frame overflow\n");
2066             remove_frame(s->ref_frames, s->ref_frames[0]->avframe->display_picture_number)->reference &= DELAYED_PIC_REF;
2067         }
2068     }
2069 
2070     if (s->num_refs) {
2071         ret = dirac_unpack_prediction_parameters(s);  /* [DIRAC_STD] 11.2 Picture Prediction Data. picture_prediction() */
2072         if (ret < 0)
2073             return ret;
2074         ret = dirac_unpack_block_motion_data(s);      /* [DIRAC_STD] 12. Block motion data syntax                       */
2075         if (ret < 0)
2076             return ret;
2077     }
2078     ret = dirac_unpack_idwt_params(s);                /* [DIRAC_STD] 11.3 Wavelet transform data                        */
2079     if (ret < 0)
2080         return ret;
2081 
2082     init_planes(s);
2083     return 0;
2084 }
2085 
get_delayed_pic(DiracContext * s,AVFrame * picture,int * got_frame)2086 static int get_delayed_pic(DiracContext *s, AVFrame *picture, int *got_frame)
2087 {
2088     DiracFrame *out = s->delay_frames[0];
2089     int i, out_idx  = 0;
2090     int ret;
2091 
2092     /* find frame with lowest picture number */
2093     for (i = 1; s->delay_frames[i]; i++)
2094         if (s->delay_frames[i]->avframe->display_picture_number < out->avframe->display_picture_number) {
2095             out     = s->delay_frames[i];
2096             out_idx = i;
2097         }
2098 
2099     for (i = out_idx; s->delay_frames[i]; i++)
2100         s->delay_frames[i] = s->delay_frames[i+1];
2101 
2102     if (out) {
2103         out->reference ^= DELAYED_PIC_REF;
2104         if((ret = av_frame_ref(picture, out->avframe)) < 0)
2105             return ret;
2106         *got_frame = 1;
2107     }
2108 
2109     return 0;
2110 }
2111 
2112 /**
2113  * Dirac Specification ->
2114  * 9.6 Parse Info Header Syntax. parse_info()
2115  * 4 byte start code + byte parse code + 4 byte size + 4 byte previous size
2116  */
2117 #define DATA_UNIT_HEADER_SIZE 13
2118 
2119 /* [DIRAC_STD] dirac_decode_data_unit makes reference to the while defined in 9.3
2120    inside the function parse_sequence() */
dirac_decode_data_unit(AVCodecContext * avctx,const uint8_t * buf,int size)2121 static int dirac_decode_data_unit(AVCodecContext *avctx, const uint8_t *buf, int size)
2122 {
2123     DiracContext *s   = avctx->priv_data;
2124     DiracFrame *pic   = NULL;
2125     AVDiracSeqHeader *dsh;
2126     int ret, i;
2127     uint8_t parse_code;
2128     unsigned tmp;
2129 
2130     if (size < DATA_UNIT_HEADER_SIZE)
2131         return AVERROR_INVALIDDATA;
2132 
2133     parse_code = buf[4];
2134 
2135     init_get_bits(&s->gb, &buf[13], 8*(size - DATA_UNIT_HEADER_SIZE));
2136 
2137     if (parse_code == DIRAC_PCODE_SEQ_HEADER) {
2138         if (s->seen_sequence_header)
2139             return 0;
2140 
2141         /* [DIRAC_STD] 10. Sequence header */
2142         ret = av_dirac_parse_sequence_header(&dsh, buf + DATA_UNIT_HEADER_SIZE, size - DATA_UNIT_HEADER_SIZE, avctx);
2143         if (ret < 0) {
2144             av_log(avctx, AV_LOG_ERROR, "error parsing sequence header");
2145             return ret;
2146         }
2147 
2148         if (CALC_PADDING((int64_t)dsh->width, MAX_DWT_LEVELS) * CALC_PADDING((int64_t)dsh->height, MAX_DWT_LEVELS) * 5LL > avctx->max_pixels)
2149             ret = AVERROR(ERANGE);
2150         if (ret >= 0)
2151             ret = ff_set_dimensions(avctx, dsh->width, dsh->height);
2152         if (ret < 0) {
2153             av_freep(&dsh);
2154             return ret;
2155         }
2156 
2157         ff_set_sar(avctx, dsh->sample_aspect_ratio);
2158         avctx->pix_fmt         = dsh->pix_fmt;
2159         avctx->color_range     = dsh->color_range;
2160         avctx->color_trc       = dsh->color_trc;
2161         avctx->color_primaries = dsh->color_primaries;
2162         avctx->colorspace      = dsh->colorspace;
2163         avctx->profile         = dsh->profile;
2164         avctx->level           = dsh->level;
2165         avctx->framerate       = dsh->framerate;
2166         s->bit_depth           = dsh->bit_depth;
2167         s->version.major       = dsh->version.major;
2168         s->version.minor       = dsh->version.minor;
2169         s->seq                 = *dsh;
2170         av_freep(&dsh);
2171 
2172         s->pshift = s->bit_depth > 8;
2173 
2174         ret = av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt,
2175                                                &s->chroma_x_shift,
2176                                                &s->chroma_y_shift);
2177         if (ret < 0)
2178             return ret;
2179 
2180         ret = alloc_sequence_buffers(s);
2181         if (ret < 0)
2182             return ret;
2183 
2184         s->seen_sequence_header = 1;
2185     } else if (parse_code == DIRAC_PCODE_END_SEQ) { /* [DIRAC_STD] End of Sequence */
2186         free_sequence_buffers(s);
2187         s->seen_sequence_header = 0;
2188     } else if (parse_code == DIRAC_PCODE_AUX) {
2189         if (buf[13] == 1) {     /* encoder implementation/version */
2190             int ver[3];
2191             /* versions older than 1.0.8 don't store quant delta for
2192                subbands with only one codeblock */
2193             if (sscanf(buf+14, "Schroedinger %d.%d.%d", ver, ver+1, ver+2) == 3)
2194                 if (ver[0] == 1 && ver[1] == 0 && ver[2] <= 7)
2195                     s->old_delta_quant = 1;
2196         }
2197     } else if (parse_code & 0x8) {  /* picture data unit */
2198         if (!s->seen_sequence_header) {
2199             av_log(avctx, AV_LOG_DEBUG, "Dropping frame without sequence header\n");
2200             return AVERROR_INVALIDDATA;
2201         }
2202 
2203         /* find an unused frame */
2204         for (i = 0; i < MAX_FRAMES; i++)
2205             if (s->all_frames[i].avframe->data[0] == NULL)
2206                 pic = &s->all_frames[i];
2207         if (!pic) {
2208             av_log(avctx, AV_LOG_ERROR, "framelist full\n");
2209             return AVERROR_INVALIDDATA;
2210         }
2211 
2212         av_frame_unref(pic->avframe);
2213 
2214         /* [DIRAC_STD] Defined in 9.6.1 ... */
2215         tmp            =  parse_code & 0x03;                   /* [DIRAC_STD] num_refs()      */
2216         if (tmp > 2) {
2217             av_log(avctx, AV_LOG_ERROR, "num_refs of 3\n");
2218             return AVERROR_INVALIDDATA;
2219         }
2220         s->num_refs      = tmp;
2221         s->is_arith      = (parse_code & 0x48) == 0x08;          /* [DIRAC_STD] using_ac()            */
2222         s->low_delay     = (parse_code & 0x88) == 0x88;          /* [DIRAC_STD] is_low_delay()        */
2223         s->core_syntax   = (parse_code & 0x88) == 0x08;          /* [DIRAC_STD] is_core_syntax()      */
2224         s->ld_picture    = (parse_code & 0xF8) == 0xC8;          /* [DIRAC_STD] is_ld_picture()       */
2225         s->hq_picture    = (parse_code & 0xF8) == 0xE8;          /* [DIRAC_STD] is_hq_picture()       */
2226         s->dc_prediction = (parse_code & 0x28) == 0x08;          /* [DIRAC_STD] using_dc_prediction() */
2227         pic->reference   = (parse_code & 0x0C) == 0x0C;          /* [DIRAC_STD] is_reference()        */
2228         pic->avframe->key_frame = s->num_refs == 0;              /* [DIRAC_STD] is_intra()            */
2229         pic->avframe->pict_type = s->num_refs + 1;               /* Definition of AVPictureType in avutil.h */
2230 
2231         /* VC-2 Low Delay has a different parse code than the Dirac Low Delay */
2232         if (s->version.minor == 2 && parse_code == 0x88)
2233             s->ld_picture = 1;
2234 
2235         if (s->low_delay && !(s->ld_picture || s->hq_picture) ) {
2236             av_log(avctx, AV_LOG_ERROR, "Invalid low delay flag\n");
2237             return AVERROR_INVALIDDATA;
2238         }
2239 
2240         if ((ret = get_buffer_with_edge(avctx, pic->avframe, (parse_code & 0x0C) == 0x0C ? AV_GET_BUFFER_FLAG_REF : 0)) < 0)
2241             return ret;
2242         s->current_picture = pic;
2243         s->plane[0].stride = pic->avframe->linesize[0];
2244         s->plane[1].stride = pic->avframe->linesize[1];
2245         s->plane[2].stride = pic->avframe->linesize[2];
2246 
2247         if (alloc_buffers(s, FFMAX3(FFABS(s->plane[0].stride), FFABS(s->plane[1].stride), FFABS(s->plane[2].stride))) < 0)
2248             return AVERROR(ENOMEM);
2249 
2250         /* [DIRAC_STD] 11.1 Picture parse. picture_parse() */
2251         ret = dirac_decode_picture_header(s);
2252         if (ret < 0)
2253             return ret;
2254 
2255         /* [DIRAC_STD] 13.0 Transform data syntax. transform_data() */
2256         ret = dirac_decode_frame_internal(s);
2257         if (ret < 0)
2258             return ret;
2259     }
2260     return 0;
2261 }
2262 
dirac_decode_frame(AVCodecContext * avctx,AVFrame * picture,int * got_frame,AVPacket * pkt)2263 static int dirac_decode_frame(AVCodecContext *avctx, AVFrame *picture,
2264                               int *got_frame, AVPacket *pkt)
2265 {
2266     DiracContext *s     = avctx->priv_data;
2267     const uint8_t *buf  = pkt->data;
2268     int buf_size        = pkt->size;
2269     int i, buf_idx      = 0;
2270     int ret;
2271     unsigned data_unit_size;
2272 
2273     /* release unused frames */
2274     for (i = 0; i < MAX_FRAMES; i++)
2275         if (s->all_frames[i].avframe->data[0] && !s->all_frames[i].reference) {
2276             av_frame_unref(s->all_frames[i].avframe);
2277             memset(s->all_frames[i].interpolated, 0, sizeof(s->all_frames[i].interpolated));
2278         }
2279 
2280     s->current_picture = NULL;
2281     *got_frame = 0;
2282 
2283     /* end of stream, so flush delayed pics */
2284     if (buf_size == 0)
2285         return get_delayed_pic(s, picture, got_frame);
2286 
2287     for (;;) {
2288         /*[DIRAC_STD] Here starts the code from parse_info() defined in 9.6
2289           [DIRAC_STD] PARSE_INFO_PREFIX = "BBCD" as defined in ISO/IEC 646
2290           BBCD start code search */
2291         for (; buf_idx + DATA_UNIT_HEADER_SIZE < buf_size; buf_idx++) {
2292             if (buf[buf_idx  ] == 'B' && buf[buf_idx+1] == 'B' &&
2293                 buf[buf_idx+2] == 'C' && buf[buf_idx+3] == 'D')
2294                 break;
2295         }
2296         /* BBCD found or end of data */
2297         if (buf_idx + DATA_UNIT_HEADER_SIZE >= buf_size)
2298             break;
2299 
2300         data_unit_size = AV_RB32(buf+buf_idx+5);
2301         if (data_unit_size > buf_size - buf_idx || !data_unit_size) {
2302             if(data_unit_size > buf_size - buf_idx)
2303             av_log(s->avctx, AV_LOG_ERROR,
2304                    "Data unit with size %d is larger than input buffer, discarding\n",
2305                    data_unit_size);
2306             buf_idx += 4;
2307             continue;
2308         }
2309         /* [DIRAC_STD] dirac_decode_data_unit makes reference to the while defined in 9.3 inside the function parse_sequence() */
2310         ret = dirac_decode_data_unit(avctx, buf+buf_idx, data_unit_size);
2311         if (ret < 0)
2312         {
2313             av_log(s->avctx, AV_LOG_ERROR,"Error in dirac_decode_data_unit\n");
2314             return ret;
2315         }
2316         buf_idx += data_unit_size;
2317     }
2318 
2319     if (!s->current_picture)
2320         return buf_size;
2321 
2322     if (s->current_picture->avframe->display_picture_number > s->frame_number) {
2323         DiracFrame *delayed_frame = remove_frame(s->delay_frames, s->frame_number);
2324 
2325         s->current_picture->reference |= DELAYED_PIC_REF;
2326 
2327         if (add_frame(s->delay_frames, MAX_DELAY, s->current_picture)) {
2328             int min_num = s->delay_frames[0]->avframe->display_picture_number;
2329             /* Too many delayed frames, so we display the frame with the lowest pts */
2330             av_log(avctx, AV_LOG_ERROR, "Delay frame overflow\n");
2331 
2332             for (i = 1; s->delay_frames[i]; i++)
2333                 if (s->delay_frames[i]->avframe->display_picture_number < min_num)
2334                     min_num = s->delay_frames[i]->avframe->display_picture_number;
2335 
2336             delayed_frame = remove_frame(s->delay_frames, min_num);
2337             add_frame(s->delay_frames, MAX_DELAY, s->current_picture);
2338         }
2339 
2340         if (delayed_frame) {
2341             delayed_frame->reference ^= DELAYED_PIC_REF;
2342             if((ret = av_frame_ref(picture, delayed_frame->avframe)) < 0)
2343                 return ret;
2344             *got_frame = 1;
2345         }
2346     } else if (s->current_picture->avframe->display_picture_number == s->frame_number) {
2347         /* The right frame at the right time :-) */
2348         if((ret = av_frame_ref(picture, s->current_picture->avframe)) < 0)
2349             return ret;
2350         *got_frame = 1;
2351     }
2352 
2353     if (*got_frame)
2354         s->frame_number = picture->display_picture_number + 1LL;
2355 
2356     return buf_idx;
2357 }
2358 
2359 const FFCodec ff_dirac_decoder = {
2360     .p.name         = "dirac",
2361     .p.long_name    = NULL_IF_CONFIG_SMALL("BBC Dirac VC-2"),
2362     .p.type         = AVMEDIA_TYPE_VIDEO,
2363     .p.id           = AV_CODEC_ID_DIRAC,
2364     .priv_data_size = sizeof(DiracContext),
2365     .init           = dirac_decode_init,
2366     .close          = dirac_decode_end,
2367     FF_CODEC_DECODE_CB(dirac_decode_frame),
2368     .p.capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_SLICE_THREADS | AV_CODEC_CAP_DR1,
2369     .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE,
2370     .flush          = dirac_decode_flush,
2371 };
2372