FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
rv34.c
Go to the documentation of this file.
1 /*
2  * RV30/40 decoder common data
3  * Copyright (c) 2007 Mike Melanson, Konstantin Shishkov
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 /**
23  * @file
24  * RV30/40 decoder common data
25  */
26 
27 #include "libavutil/imgutils.h"
28 #include "libavutil/internal.h"
29 
30 #include "avcodec.h"
31 #include "mpegvideo.h"
32 #include "golomb.h"
33 #include "internal.h"
34 #include "mathops.h"
35 #include "rectangle.h"
36 #include "thread.h"
37 
38 #include "rv34vlc.h"
39 #include "rv34data.h"
40 #include "rv34.h"
41 
42 //#define DEBUG
43 
44 static inline void ZERO8x2(void* dst, int stride)
45 {
46  fill_rectangle(dst, 1, 2, stride, 0, 4);
47  fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
48 }
49 
50 /** translation of RV30/40 macroblock types to lavc ones */
51 static const int rv34_mb_type_to_lavc[12] = {
52  MB_TYPE_INTRA,
53  MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC,
54  MB_TYPE_16x16 | MB_TYPE_L0,
55  MB_TYPE_8x8 | MB_TYPE_L0,
56  MB_TYPE_16x16 | MB_TYPE_L0,
57  MB_TYPE_16x16 | MB_TYPE_L1,
58  MB_TYPE_SKIP,
59  MB_TYPE_DIRECT2 | MB_TYPE_16x16,
60  MB_TYPE_16x8 | MB_TYPE_L0,
61  MB_TYPE_8x16 | MB_TYPE_L0,
62  MB_TYPE_16x16 | MB_TYPE_L0L1,
63  MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC
64 };
65 
66 
67 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
68 
69 static int rv34_decode_mv(RV34DecContext *r, int block_type);
70 
71 /**
72  * @name RV30/40 VLC generating functions
73  * @{
74  */
75 
76 static const int table_offs[] = {
77  0, 1818, 3622, 4144, 4698, 5234, 5804, 5868, 5900, 5932,
78  5996, 6252, 6316, 6348, 6380, 7674, 8944, 10274, 11668, 12250,
79  14060, 15846, 16372, 16962, 17512, 18148, 18180, 18212, 18244, 18308,
80  18564, 18628, 18660, 18692, 20036, 21314, 22648, 23968, 24614, 26384,
81  28190, 28736, 29366, 29938, 30608, 30640, 30672, 30704, 30768, 31024,
82  31088, 31120, 31184, 32570, 33898, 35236, 36644, 37286, 39020, 40802,
83  41368, 42052, 42692, 43348, 43380, 43412, 43444, 43476, 43604, 43668,
84  43700, 43732, 45100, 46430, 47778, 49160, 49802, 51550, 53340, 53972,
85  54648, 55348, 55994, 56122, 56154, 56186, 56218, 56346, 56410, 56442,
86  56474, 57878, 59290, 60636, 62036, 62682, 64460, 64524, 64588, 64716,
87  64844, 66076, 67466, 67978, 68542, 69064, 69648, 70296, 72010, 72074,
88  72138, 72202, 72330, 73572, 74936, 75454, 76030, 76566, 77176, 77822,
89  79582, 79646, 79678, 79742, 79870, 81180, 82536, 83064, 83672, 84242,
90  84934, 85576, 87384, 87448, 87480, 87544, 87672, 88982, 90340, 90902,
91  91598, 92182, 92846, 93488, 95246, 95278, 95310, 95374, 95502, 96878,
92  98266, 98848, 99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,
93  103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,
94  111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592
95 };
96 
97 static VLC_TYPE table_data[117592][2];
98 
99 /**
100  * Generate VLC from codeword lengths.
101  * @param bits codeword lengths (zeroes are accepted)
102  * @param size length of input data
103  * @param vlc output VLC
104  * @param insyms symbols for input codes (NULL for default ones)
105  * @param num VLC table number (for static initialization)
106  */
107 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms,
108  const int num)
109 {
110  int i;
111  int counts[17] = {0}, codes[17];
112  uint16_t cw[MAX_VLC_SIZE], syms[MAX_VLC_SIZE];
113  uint8_t bits2[MAX_VLC_SIZE];
114  int maxbits = 0, realsize = 0;
115 
116  for(i = 0; i < size; i++){
117  if(bits[i]){
118  bits2[realsize] = bits[i];
119  syms[realsize] = insyms ? insyms[i] : i;
120  realsize++;
121  maxbits = FFMAX(maxbits, bits[i]);
122  counts[bits[i]]++;
123  }
124  }
125 
126  codes[0] = 0;
127  for(i = 0; i < 16; i++)
128  codes[i+1] = (codes[i] + counts[i]) << 1;
129  for(i = 0; i < realsize; i++)
130  cw[i] = codes[bits2[i]]++;
131 
132  vlc->table = &table_data[table_offs[num]];
133  vlc->table_allocated = table_offs[num + 1] - table_offs[num];
134  ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
135  bits2, 1, 1,
136  cw, 2, 2,
137  syms, 2, 2, INIT_VLC_USE_NEW_STATIC);
138 }
139 
140 /**
141  * Initialize all tables.
142  */
143 static av_cold void rv34_init_tables(void)
144 {
145  int i, j, k;
146 
147  for(i = 0; i < NUM_INTRA_TABLES; i++){
148  for(j = 0; j < 2; j++){
149  rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE, &intra_vlcs[i].cbppattern[j], NULL, 19*i + 0 + j);
150  rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j);
151  rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j], NULL, 19*i + 4 + j);
152  for(k = 0; k < 4; k++){
153  rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2], CBP_VLC_SIZE, &intra_vlcs[i].cbp[j][k], rv34_cbp_code, 19*i + 6 + j*4 + k);
154  }
155  }
156  for(j = 0; j < 4; j++){
157  rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j);
158  }
159  rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18);
160  }
161 
162  for(i = 0; i < NUM_INTER_TABLES; i++){
163  rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95);
164  for(j = 0; j < 4; j++){
165  rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j);
166  }
167  for(j = 0; j < 2; j++){
168  rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j], NULL, i*12 + 100 + j);
169  rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j);
170  rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j], NULL, i*12 + 104 + j);
171  }
172  rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106);
173  }
174 }
175 
176 /** @} */ // vlc group
177 
178 /**
179  * @name RV30/40 4x4 block decoding functions
180  * @{
181  */
182 
183 /**
184  * Decode coded block pattern.
185  */
186 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
187 {
188  int pattern, code, cbp=0;
189  int ones;
190  static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
191  static const int shifts[4] = { 0, 2, 8, 10 };
192  const int *curshift = shifts;
193  int i, t, mask;
194 
195  code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
196  pattern = code & 0xF;
197  code >>= 4;
198 
199  ones = rv34_count_ones[pattern];
200 
201  for(mask = 8; mask; mask >>= 1, curshift++){
202  if(pattern & mask)
203  cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
204  }
205 
206  for(i = 0; i < 4; i++){
207  t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
208  if(t == 1)
209  cbp |= cbp_masks[get_bits1(gb)] << i;
210  if(t == 2)
211  cbp |= cbp_masks[2] << i;
212  }
213  return cbp;
214 }
215 
216 /**
217  * Get one coefficient value from the bitstream and store it.
218  */
219 static inline void decode_coeff(int16_t *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)
220 {
221  if(coef){
222  if(coef == esc){
223  coef = get_vlc2(gb, vlc->table, 9, 2);
224  if(coef > 23){
225  coef -= 23;
226  coef = 22 + ((1 << coef) | get_bits(gb, coef));
227  }
228  coef += esc;
229  }
230  if(get_bits1(gb))
231  coef = -coef;
232  *dst = (coef*q + 8) >> 4;
233  }
234 }
235 
236 /**
237  * Decode 2x2 subblock of coefficients.
238  */
239 static inline void decode_subblock(int16_t *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
240 {
241  int flags = modulo_three_table[code];
242 
243  decode_coeff( dst+0*4+0, (flags >> 6) , 3, gb, vlc, q);
244  if(is_block2){
245  decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q);
246  decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q);
247  }else{
248  decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q);
249  decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q);
250  }
251  decode_coeff( dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q);
252 }
253 
254 /**
255  * Decode a single coefficient.
256  */
257 static inline void decode_subblock1(int16_t *dst, int code, GetBitContext *gb, VLC *vlc, int q)
258 {
259  int coeff = modulo_three_table[code] >> 6;
260  decode_coeff(dst, coeff, 3, gb, vlc, q);
261 }
262 
263 static inline void decode_subblock3(int16_t *dst, int code, GetBitContext *gb, VLC *vlc,
264  int q_dc, int q_ac1, int q_ac2)
265 {
266  int flags = modulo_three_table[code];
267 
268  decode_coeff(dst+0*4+0, (flags >> 6) , 3, gb, vlc, q_dc);
269  decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
270  decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
271  decode_coeff(dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
272 }
273 
274 /**
275  * Decode coefficients for 4x4 block.
276  *
277  * This is done by filling 2x2 subblocks with decoded coefficients
278  * in this order (the same for subblocks and subblock coefficients):
279  * o--o
280  * /
281  * /
282  * o--o
283  */
284 
285 static int rv34_decode_block(int16_t *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc, int q_dc, int q_ac1, int q_ac2)
286 {
287  int code, pattern, has_ac = 1;
288 
289  code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
290 
291  pattern = code & 0x7;
292 
293  code >>= 3;
294 
295  if (modulo_three_table[code] & 0x3F) {
296  decode_subblock3(dst, code, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
297  } else {
298  decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
299  if (!pattern)
300  return 0;
301  has_ac = 0;
302  }
303 
304  if(pattern & 4){
305  code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
306  decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2);
307  }
308  if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
309  code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
310  decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2);
311  }
312  if(pattern & 1){
313  code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
314  decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
315  }
316  return has_ac | pattern;
317 }
318 
319 /**
320  * @name RV30/40 bitstream parsing
321  * @{
322  */
323 
324 /**
325  * Decode starting slice position.
326  * @todo Maybe replace with ff_h263_decode_mba() ?
327  */
328 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
329 {
330  int i;
331  for(i = 0; i < 5; i++)
332  if(rv34_mb_max_sizes[i] >= mb_size - 1)
333  break;
334  return rv34_mb_bits_sizes[i];
335 }
336 
337 /**
338  * Select VLC set for decoding from current quantizer, modifier and frame type.
339  */
340 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
341 {
342  if(mod == 2 && quant < 19) quant += 10;
343  else if(mod && quant < 26) quant += 5;
344  return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
345  : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
346 }
347 
348 /**
349  * Decode intra macroblock header and return CBP in case of success, -1 otherwise.
350  */
351 static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types)
352 {
353  MpegEncContext *s = &r->s;
354  GetBitContext *gb = &s->gb;
355  int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
356  int t;
357 
358  r->is16 = get_bits1(gb);
359  if(r->is16){
360  s->current_picture_ptr->f.mb_type[mb_pos] = MB_TYPE_INTRA16x16;
361  r->block_type = RV34_MB_TYPE_INTRA16x16;
362  t = get_bits(gb, 2);
363  fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
364  r->luma_vlc = 2;
365  }else{
366  if(!r->rv30){
367  if(!get_bits1(gb))
368  av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
369  }
370  s->current_picture_ptr->f.mb_type[mb_pos] = MB_TYPE_INTRA;
371  r->block_type = RV34_MB_TYPE_INTRA;
372  if(r->decode_intra_types(r, gb, intra_types) < 0)
373  return -1;
374  r->luma_vlc = 1;
375  }
376 
377  r->chroma_vlc = 0;
378  r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
379 
380  return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
381 }
382 
383 /**
384  * Decode inter macroblock header and return CBP in case of success, -1 otherwise.
385  */
386 static int rv34_decode_inter_mb_header(RV34DecContext *r, int8_t *intra_types)
387 {
388  MpegEncContext *s = &r->s;
389  GetBitContext *gb = &s->gb;
390  int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
391  int i, t;
392 
393  r->block_type = r->decode_mb_info(r);
394  if(r->block_type == -1)
395  return -1;
396  s->current_picture_ptr->f.mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
397  r->mb_type[mb_pos] = r->block_type;
398  if(r->block_type == RV34_MB_SKIP){
399  if(s->pict_type == AV_PICTURE_TYPE_P)
400  r->mb_type[mb_pos] = RV34_MB_P_16x16;
401  if(s->pict_type == AV_PICTURE_TYPE_B)
402  r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
403  }
404  r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->f.mb_type[mb_pos]);
405  rv34_decode_mv(r, r->block_type);
406  if(r->block_type == RV34_MB_SKIP){
407  fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
408  return 0;
409  }
410  r->chroma_vlc = 1;
411  r->luma_vlc = 0;
412 
413  if(IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])){
414  if(r->is16){
415  t = get_bits(gb, 2);
416  fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
417  r->luma_vlc = 2;
418  }else{
419  if(r->decode_intra_types(r, gb, intra_types) < 0)
420  return -1;
421  r->luma_vlc = 1;
422  }
423  r->chroma_vlc = 0;
424  r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
425  }else{
426  for(i = 0; i < 16; i++)
427  intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
428  r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
429  if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
430  r->is16 = 1;
431  r->chroma_vlc = 1;
432  r->luma_vlc = 2;
433  r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
434  }
435  }
436 
437  return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
438 }
439 
440 /** @} */ //bitstream functions
441 
442 /**
443  * @name motion vector related code (prediction, reconstruction, motion compensation)
444  * @{
445  */
446 
447 /** macroblock partition width in 8x8 blocks */
448 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
449 
450 /** macroblock partition height in 8x8 blocks */
451 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
452 
453 /** availability index for subblocks */
454 static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
455 
456 /**
457  * motion vector prediction
458  *
459  * Motion prediction performed for the block by using median prediction of
460  * motion vectors from the left, top and right top blocks but in corner cases
461  * some other vectors may be used instead.
462  */
463 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
464 {
465  MpegEncContext *s = &r->s;
466  int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
467  int A[2] = {0}, B[2], C[2];
468  int i, j;
469  int mx, my;
470  int* avail = r->avail_cache + avail_indexes[subblock_no];
471  int c_off = part_sizes_w[block_type];
472 
473  mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
474  if(subblock_no == 3)
475  c_off = -1;
476 
477  if(avail[-1]){
478  A[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-1][0];
479  A[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-1][1];
480  }
481  if(avail[-4]){
482  B[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride][0];
483  B[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride][1];
484  }else{
485  B[0] = A[0];
486  B[1] = A[1];
487  }
488  if(!avail[c_off-4]){
489  if(avail[-4] && (avail[-1] || r->rv30)){
490  C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride-1][0];
491  C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride-1][1];
492  }else{
493  C[0] = A[0];
494  C[1] = A[1];
495  }
496  }else{
497  C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride+c_off][0];
498  C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride+c_off][1];
499  }
500  mx = mid_pred(A[0], B[0], C[0]);
501  my = mid_pred(A[1], B[1], C[1]);
502  mx += r->dmv[dmv_no][0];
503  my += r->dmv[dmv_no][1];
504  for(j = 0; j < part_sizes_h[block_type]; j++){
505  for(i = 0; i < part_sizes_w[block_type]; i++){
506  s->current_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
507  s->current_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
508  }
509  }
510 }
511 
512 #define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
513 
514 /**
515  * Calculate motion vector component that should be added for direct blocks.
516  */
517 static int calc_add_mv(RV34DecContext *r, int dir, int val)
518 {
519  int mul = dir ? -r->mv_weight2 : r->mv_weight1;
520 
521  return (val * mul + 0x2000) >> 14;
522 }
523 
524 /**
525  * Predict motion vector for B-frame macroblock.
526  */
527 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
528  int A_avail, int B_avail, int C_avail,
529  int *mx, int *my)
530 {
531  if(A_avail + B_avail + C_avail != 3){
532  *mx = A[0] + B[0] + C[0];
533  *my = A[1] + B[1] + C[1];
534  if(A_avail + B_avail + C_avail == 2){
535  *mx /= 2;
536  *my /= 2;
537  }
538  }else{
539  *mx = mid_pred(A[0], B[0], C[0]);
540  *my = mid_pred(A[1], B[1], C[1]);
541  }
542 }
543 
544 /**
545  * motion vector prediction for B-frames
546  */
547 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
548 {
549  MpegEncContext *s = &r->s;
550  int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
551  int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
552  int A[2] = { 0 }, B[2] = { 0 }, C[2] = { 0 };
553  int has_A = 0, has_B = 0, has_C = 0;
554  int mx, my;
555  int i, j;
556  Picture *cur_pic = s->current_picture_ptr;
557  const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
558  int type = cur_pic->f.mb_type[mb_pos];
559 
560  if((r->avail_cache[6-1] & type) & mask){
561  A[0] = cur_pic->f.motion_val[dir][mv_pos - 1][0];
562  A[1] = cur_pic->f.motion_val[dir][mv_pos - 1][1];
563  has_A = 1;
564  }
565  if((r->avail_cache[6-4] & type) & mask){
566  B[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride][0];
567  B[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride][1];
568  has_B = 1;
569  }
570  if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
571  C[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride + 2][0];
572  C[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride + 2][1];
573  has_C = 1;
574  }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
575  C[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride - 1][0];
576  C[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride - 1][1];
577  has_C = 1;
578  }
579 
580  rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
581 
582  mx += r->dmv[dir][0];
583  my += r->dmv[dir][1];
584 
585  for(j = 0; j < 2; j++){
586  for(i = 0; i < 2; i++){
587  cur_pic->f.motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
588  cur_pic->f.motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
589  }
590  }
591  if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
592  ZERO8x2(cur_pic->f.motion_val[!dir][mv_pos], s->b8_stride);
593  }
594 }
595 
596 /**
597  * motion vector prediction - RV3 version
598  */
599 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
600 {
601  MpegEncContext *s = &r->s;
602  int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
603  int A[2] = {0}, B[2], C[2];
604  int i, j, k;
605  int mx, my;
606  int* avail = r->avail_cache + avail_indexes[0];
607 
608  if(avail[-1]){
609  A[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - 1][0];
610  A[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - 1][1];
611  }
612  if(avail[-4]){
613  B[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride][0];
614  B[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride][1];
615  }else{
616  B[0] = A[0];
617  B[1] = A[1];
618  }
619  if(!avail[-4 + 2]){
620  if(avail[-4] && (avail[-1])){
621  C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride - 1][0];
622  C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride - 1][1];
623  }else{
624  C[0] = A[0];
625  C[1] = A[1];
626  }
627  }else{
628  C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride + 2][0];
629  C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride + 2][1];
630  }
631  mx = mid_pred(A[0], B[0], C[0]);
632  my = mid_pred(A[1], B[1], C[1]);
633  mx += r->dmv[0][0];
634  my += r->dmv[0][1];
635  for(j = 0; j < 2; j++){
636  for(i = 0; i < 2; i++){
637  for(k = 0; k < 2; k++){
638  s->current_picture_ptr->f.motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
639  s->current_picture_ptr->f.motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
640  }
641  }
642  }
643 }
644 
645 static const int chroma_coeffs[3] = { 0, 3, 5 };
646 
647 /**
648  * generic motion compensation function
649  *
650  * @param r decoder context
651  * @param block_type type of the current block
652  * @param xoff horizontal offset from the start of the current block
653  * @param yoff vertical offset from the start of the current block
654  * @param mv_off offset to the motion vector information
655  * @param width width of the current partition in 8x8 blocks
656  * @param height height of the current partition in 8x8 blocks
657  * @param dir motion compensation direction (i.e. from the last or the next reference frame)
658  * @param thirdpel motion vectors are specified in 1/3 of pixel
659  * @param qpel_mc a set of functions used to perform luma motion compensation
660  * @param chroma_mc a set of functions used to perform chroma motion compensation
661  */
662 static inline void rv34_mc(RV34DecContext *r, const int block_type,
663  const int xoff, const int yoff, int mv_off,
664  const int width, const int height, int dir,
665  const int thirdpel, int weighted,
666  qpel_mc_func (*qpel_mc)[16],
667  h264_chroma_mc_func (*chroma_mc))
668 {
669  MpegEncContext *s = &r->s;
670  uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
671  int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
672  int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
673  int is16x16 = 1;
674 
675  if(thirdpel){
676  int chroma_mx, chroma_my;
677  mx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
678  my = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
679  lx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
680  ly = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
681  chroma_mx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] / 2;
682  chroma_my = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] / 2;
683  umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
684  umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
685  uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
686  uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
687  }else{
688  int cx, cy;
689  mx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] >> 2;
690  my = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] >> 2;
691  lx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] & 3;
692  ly = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] & 3;
693  cx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] / 2;
694  cy = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] / 2;
695  umx = cx >> 2;
696  umy = cy >> 2;
697  uvmx = (cx & 3) << 1;
698  uvmy = (cy & 3) << 1;
699  //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
700  if(uvmx == 6 && uvmy == 6)
701  uvmx = uvmy = 4;
702  }
703 
704  if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
705  /* wait for the referenced mb row to be finished */
706  int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4);
707  AVFrame *f = dir ? &s->next_picture_ptr->f : &s->last_picture_ptr->f;
708  ff_thread_await_progress(f, mb_row, 0);
709  }
710 
711  dxy = ly*4 + lx;
712  srcY = dir ? s->next_picture_ptr->f.data[0] : s->last_picture_ptr->f.data[0];
713  srcU = dir ? s->next_picture_ptr->f.data[1] : s->last_picture_ptr->f.data[1];
714  srcV = dir ? s->next_picture_ptr->f.data[2] : s->last_picture_ptr->f.data[2];
715  src_x = s->mb_x * 16 + xoff + mx;
716  src_y = s->mb_y * 16 + yoff + my;
717  uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
718  uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
719  srcY += src_y * s->linesize + src_x;
720  srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
721  srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
722  if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 ||
723  (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4 ||
724  (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4) {
725  uint8_t *uvbuf = s->edge_emu_buffer + 22 * s->linesize;
726 
727  srcY -= 2 + 2*s->linesize;
728  s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
729  src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
730  srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
731  s->vdsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
732  uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
733  s->vdsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
734  uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
735  srcU = uvbuf;
736  srcV = uvbuf + 16;
737  }
738  if(!weighted){
739  Y = s->dest[0] + xoff + yoff *s->linesize;
740  U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
741  V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
742  }else{
743  Y = r->tmp_b_block_y [dir] + xoff + yoff *s->linesize;
744  U = r->tmp_b_block_uv[dir*2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
745  V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
746  }
747 
748  if(block_type == RV34_MB_P_16x8){
749  qpel_mc[1][dxy](Y, srcY, s->linesize);
750  Y += 8;
751  srcY += 8;
752  }else if(block_type == RV34_MB_P_8x16){
753  qpel_mc[1][dxy](Y, srcY, s->linesize);
754  Y += 8 * s->linesize;
755  srcY += 8 * s->linesize;
756  }
757  is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
758  qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
759  chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
760  chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
761 }
762 
763 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
764  const int xoff, const int yoff, int mv_off,
765  const int width, const int height, int dir)
766 {
767  rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
768  r->rdsp.put_pixels_tab,
769  r->rdsp.put_chroma_pixels_tab);
770 }
771 
772 static void rv4_weight(RV34DecContext *r)
773 {
774  r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][0](r->s.dest[0],
775  r->tmp_b_block_y[0],
776  r->tmp_b_block_y[1],
777  r->weight1,
778  r->weight2,
779  r->s.linesize);
780  r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[1],
781  r->tmp_b_block_uv[0],
782  r->tmp_b_block_uv[2],
783  r->weight1,
784  r->weight2,
785  r->s.uvlinesize);
786  r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[2],
787  r->tmp_b_block_uv[1],
788  r->tmp_b_block_uv[3],
789  r->weight1,
790  r->weight2,
791  r->s.uvlinesize);
792 }
793 
794 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
795 {
796  int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
797 
798  rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
799  r->rdsp.put_pixels_tab,
800  r->rdsp.put_chroma_pixels_tab);
801  if(!weighted){
802  rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
803  r->rdsp.avg_pixels_tab,
804  r->rdsp.avg_chroma_pixels_tab);
805  }else{
806  rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
807  r->rdsp.put_pixels_tab,
808  r->rdsp.put_chroma_pixels_tab);
809  rv4_weight(r);
810  }
811 }
812 
813 static void rv34_mc_2mv_skip(RV34DecContext *r)
814 {
815  int i, j;
816  int weighted = !r->rv30 && r->weight1 != 8192;
817 
818  for(j = 0; j < 2; j++)
819  for(i = 0; i < 2; i++){
820  rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
821  weighted,
822  r->rdsp.put_pixels_tab,
823  r->rdsp.put_chroma_pixels_tab);
824  rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
825  weighted,
826  weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
827  weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
828  }
829  if(weighted)
830  rv4_weight(r);
831 }
832 
833 /** number of motion vectors in each macroblock type */
834 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
835 
836 /**
837  * Decode motion vector differences
838  * and perform motion vector reconstruction and motion compensation.
839  */
840 static int rv34_decode_mv(RV34DecContext *r, int block_type)
841 {
842  MpegEncContext *s = &r->s;
843  GetBitContext *gb = &s->gb;
844  int i, j, k, l;
845  int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
846  int next_bt;
847 
848  memset(r->dmv, 0, sizeof(r->dmv));
849  for(i = 0; i < num_mvs[block_type]; i++){
850  r->dmv[i][0] = svq3_get_se_golomb(gb);
851  r->dmv[i][1] = svq3_get_se_golomb(gb);
852  }
853  switch(block_type){
854  case RV34_MB_TYPE_INTRA:
855  case RV34_MB_TYPE_INTRA16x16:
856  ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
857  return 0;
858  case RV34_MB_SKIP:
859  if(s->pict_type == AV_PICTURE_TYPE_P){
860  ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
861  rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
862  break;
863  }
864  case RV34_MB_B_DIRECT:
865  //surprisingly, it uses motion scheme from next reference frame
866  /* wait for the current mb row to be finished */
867  if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
868  ff_thread_await_progress(&s->next_picture_ptr->f, FFMAX(0, s->mb_y-1), 0);
869 
870  next_bt = s->next_picture_ptr->f.mb_type[s->mb_x + s->mb_y * s->mb_stride];
871  if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
872  ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
873  ZERO8x2(s->current_picture_ptr->f.motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
874  }else
875  for(j = 0; j < 2; j++)
876  for(i = 0; i < 2; i++)
877  for(k = 0; k < 2; k++)
878  for(l = 0; l < 2; l++)
879  s->current_picture_ptr->f.motion_val[l][mv_pos + i + j*s->b8_stride][k] = calc_add_mv(r, l, s->next_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][k]);
880  if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
881  rv34_mc_2mv(r, block_type);
882  else
883  rv34_mc_2mv_skip(r);
884  ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
885  break;
886  case RV34_MB_P_16x16:
887  case RV34_MB_P_MIX16x16:
888  rv34_pred_mv(r, block_type, 0, 0);
889  rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
890  break;
891  case RV34_MB_B_FORWARD:
892  case RV34_MB_B_BACKWARD:
893  r->dmv[1][0] = r->dmv[0][0];
894  r->dmv[1][1] = r->dmv[0][1];
895  if(r->rv30)
896  rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
897  else
898  rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
899  rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
900  break;
901  case RV34_MB_P_16x8:
902  case RV34_MB_P_8x16:
903  rv34_pred_mv(r, block_type, 0, 0);
904  rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
905  if(block_type == RV34_MB_P_16x8){
906  rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
907  rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
908  }
909  if(block_type == RV34_MB_P_8x16){
910  rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
911  rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
912  }
913  break;
914  case RV34_MB_B_BIDIR:
915  rv34_pred_mv_b (r, block_type, 0);
916  rv34_pred_mv_b (r, block_type, 1);
917  rv34_mc_2mv (r, block_type);
918  break;
919  case RV34_MB_P_8x8:
920  for(i=0;i< 4;i++){
921  rv34_pred_mv(r, block_type, i, i);
922  rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
923  }
924  break;
925  }
926 
927  return 0;
928 }
929 /** @} */ // mv group
930 
931 /**
932  * @name Macroblock reconstruction functions
933  * @{
934  */
935 /** mapping of RV30/40 intra prediction types to standard H.264 types */
936 static const int ittrans[9] = {
937  DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
938  VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
939 };
940 
941 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
942 static const int ittrans16[4] = {
943  DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
944 };
945 
946 /**
947  * Perform 4x4 intra prediction.
948  */
949 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
950 {
951  uint8_t *prev = dst - stride + 4;
952  uint32_t topleft;
953 
954  if(!up && !left)
955  itype = DC_128_PRED;
956  else if(!up){
957  if(itype == VERT_PRED) itype = HOR_PRED;
958  if(itype == DC_PRED) itype = LEFT_DC_PRED;
959  }else if(!left){
960  if(itype == HOR_PRED) itype = VERT_PRED;
961  if(itype == DC_PRED) itype = TOP_DC_PRED;
962  if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
963  }
964  if(!down){
965  if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
966  if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
967  if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
968  }
969  if(!right && up){
970  topleft = dst[-stride + 3] * 0x01010101u;
971  prev = (uint8_t*)&topleft;
972  }
973  r->h.pred4x4[itype](dst, prev, stride);
974 }
975 
976 static inline int adjust_pred16(int itype, int up, int left)
977 {
978  if(!up && !left)
979  itype = DC_128_PRED8x8;
980  else if(!up){
981  if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
982  if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
983  if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
984  }else if(!left){
985  if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
986  if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
987  if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
988  }
989  return itype;
990 }
991 
992 static inline void rv34_process_block(RV34DecContext *r,
993  uint8_t *pdst, int stride,
994  int fc, int sc, int q_dc, int q_ac)
995 {
996  MpegEncContext *s = &r->s;
997  int16_t *ptr = s->block[0];
998  int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,
999  fc, sc, q_dc, q_ac, q_ac);
1000  if(has_ac){
1001  r->rdsp.rv34_idct_add(pdst, stride, ptr);
1002  }else{
1003  r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);
1004  ptr[0] = 0;
1005  }
1006 }
1007 
1008 static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)
1009 {
1010  LOCAL_ALIGNED_16(int16_t, block16, [16]);
1011  MpegEncContext *s = &r->s;
1012  GetBitContext *gb = &s->gb;
1013  int q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],
1014  q_ac = rv34_qscale_tab[s->qscale];
1015  uint8_t *dst = s->dest[0];
1016  int16_t *ptr = s->block[0];
1017  int i, j, itype, has_ac;
1018 
1019  memset(block16, 0, 16 * sizeof(*block16));
1020 
1021  has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
1022  if(has_ac)
1023  r->rdsp.rv34_inv_transform(block16);
1024  else
1025  r->rdsp.rv34_inv_transform_dc(block16);
1026 
1027  itype = ittrans16[intra_types[0]];
1028  itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1029  r->h.pred16x16[itype](dst, s->linesize);
1030 
1031  for(j = 0; j < 4; j++){
1032  for(i = 0; i < 4; i++, cbp >>= 1){
1033  int dc = block16[i + j*4];
1034 
1035  if(cbp & 1){
1036  has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1037  }else
1038  has_ac = 0;
1039 
1040  if(has_ac){
1041  ptr[0] = dc;
1042  r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1043  }else
1044  r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1045  }
1046 
1047  dst += 4*s->linesize;
1048  }
1049 
1050  itype = ittrans16[intra_types[0]];
1051  if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1052  itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1053 
1054  q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1055  q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1056 
1057  for(j = 1; j < 3; j++){
1058  dst = s->dest[j];
1059  r->h.pred8x8[itype](dst, s->uvlinesize);
1060  for(i = 0; i < 4; i++, cbp >>= 1){
1061  uint8_t *pdst;
1062  if(!(cbp & 1)) continue;
1063  pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1064 
1065  rv34_process_block(r, pdst, s->uvlinesize,
1066  r->chroma_vlc, 1, q_dc, q_ac);
1067  }
1068  }
1069 }
1070 
1071 static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)
1072 {
1073  MpegEncContext *s = &r->s;
1074  uint8_t *dst = s->dest[0];
1075  int avail[6*8] = {0};
1076  int i, j, k;
1077  int idx, q_ac, q_dc;
1078 
1079  // Set neighbour information.
1080  if(r->avail_cache[1])
1081  avail[0] = 1;
1082  if(r->avail_cache[2])
1083  avail[1] = avail[2] = 1;
1084  if(r->avail_cache[3])
1085  avail[3] = avail[4] = 1;
1086  if(r->avail_cache[4])
1087  avail[5] = 1;
1088  if(r->avail_cache[5])
1089  avail[8] = avail[16] = 1;
1090  if(r->avail_cache[9])
1091  avail[24] = avail[32] = 1;
1092 
1093  q_ac = rv34_qscale_tab[s->qscale];
1094  for(j = 0; j < 4; j++){
1095  idx = 9 + j*8;
1096  for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){
1097  rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
1098  avail[idx] = 1;
1099  if(!(cbp & 1)) continue;
1100 
1101  rv34_process_block(r, dst, s->linesize,
1102  r->luma_vlc, 0, q_ac, q_ac);
1103  }
1104  dst += s->linesize * 4 - 4*4;
1105  intra_types += r->intra_types_stride;
1106  }
1107 
1108  intra_types -= r->intra_types_stride * 4;
1109 
1110  q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1111  q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1112 
1113  for(k = 0; k < 2; k++){
1114  dst = s->dest[1+k];
1115  fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
1116 
1117  for(j = 0; j < 2; j++){
1118  int* acache = r->avail_cache + 6 + j*4;
1119  for(i = 0; i < 2; i++, cbp >>= 1, acache++){
1120  int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];
1121  rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);
1122  acache[0] = 1;
1123 
1124  if(!(cbp&1)) continue;
1125 
1126  rv34_process_block(r, dst + 4*i, s->uvlinesize,
1127  r->chroma_vlc, 1, q_dc, q_ac);
1128  }
1129 
1130  dst += 4*s->uvlinesize;
1131  }
1132  }
1133 }
1134 
1135 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1136 {
1137  int d;
1138  d = motion_val[0][0] - motion_val[-step][0];
1139  if(d < -3 || d > 3)
1140  return 1;
1141  d = motion_val[0][1] - motion_val[-step][1];
1142  if(d < -3 || d > 3)
1143  return 1;
1144  return 0;
1145 }
1146 
1147 static int rv34_set_deblock_coef(RV34DecContext *r)
1148 {
1149  MpegEncContext *s = &r->s;
1150  int hmvmask = 0, vmvmask = 0, i, j;
1151  int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1152  int16_t (*motion_val)[2] = &s->current_picture_ptr->f.motion_val[0][midx];
1153  for(j = 0; j < 16; j += 8){
1154  for(i = 0; i < 2; i++){
1155  if(is_mv_diff_gt_3(motion_val + i, 1))
1156  vmvmask |= 0x11 << (j + i*2);
1157  if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1158  hmvmask |= 0x03 << (j + i*2);
1159  }
1160  motion_val += s->b8_stride;
1161  }
1162  if(s->first_slice_line)
1163  hmvmask &= ~0x000F;
1164  if(!s->mb_x)
1165  vmvmask &= ~0x1111;
1166  if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1167  vmvmask |= (vmvmask & 0x4444) >> 1;
1168  hmvmask |= (hmvmask & 0x0F00) >> 4;
1169  if(s->mb_x)
1170  r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1171  if(!s->first_slice_line)
1172  r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1173  }
1174  return hmvmask | vmvmask;
1175 }
1176 
1177 static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
1178 {
1179  MpegEncContext *s = &r->s;
1180  GetBitContext *gb = &s->gb;
1181  uint8_t *dst = s->dest[0];
1182  int16_t *ptr = s->block[0];
1183  int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1184  int cbp, cbp2;
1185  int q_dc, q_ac, has_ac;
1186  int i, j;
1187  int dist;
1188 
1189  // Calculate which neighbours are available. Maybe it's worth optimizing too.
1190  memset(r->avail_cache, 0, sizeof(r->avail_cache));
1191  fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1192  dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1193  if(s->mb_x && dist)
1194  r->avail_cache[5] =
1195  r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];
1196  if(dist >= s->mb_width)
1197  r->avail_cache[2] =
1198  r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];
1199  if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1200  r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];
1201  if(s->mb_x && dist > s->mb_width)
1202  r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];
1203 
1204  s->qscale = r->si.quant;
1205  cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);
1206  r->cbp_luma [mb_pos] = cbp;
1207  r->cbp_chroma[mb_pos] = cbp >> 16;
1208  r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1209  s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;
1210 
1211  if(cbp == -1)
1212  return -1;
1213 
1214  if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])){
1215  if(r->is16) rv34_output_i16x16(r, intra_types, cbp);
1216  else rv34_output_intra(r, intra_types, cbp);
1217  return 0;
1218  }
1219 
1220  if(r->is16){
1221  // Only for RV34_MB_P_MIX16x16
1222  LOCAL_ALIGNED_16(int16_t, block16, [16]);
1223  memset(block16, 0, 16 * sizeof(*block16));
1224  q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];
1225  q_ac = rv34_qscale_tab[s->qscale];
1226  if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
1227  r->rdsp.rv34_inv_transform(block16);
1228  else
1229  r->rdsp.rv34_inv_transform_dc(block16);
1230 
1231  q_ac = rv34_qscale_tab[s->qscale];
1232 
1233  for(j = 0; j < 4; j++){
1234  for(i = 0; i < 4; i++, cbp >>= 1){
1235  int dc = block16[i + j*4];
1236 
1237  if(cbp & 1){
1238  has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1239  }else
1240  has_ac = 0;
1241 
1242  if(has_ac){
1243  ptr[0] = dc;
1244  r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1245  }else
1246  r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1247  }
1248 
1249  dst += 4*s->linesize;
1250  }
1251 
1252  r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1253  }else{
1254  q_ac = rv34_qscale_tab[s->qscale];
1255 
1256  for(j = 0; j < 4; j++){
1257  for(i = 0; i < 4; i++, cbp >>= 1){
1258  if(!(cbp & 1)) continue;
1259 
1260  rv34_process_block(r, dst + 4*i, s->linesize,
1261  r->luma_vlc, 0, q_ac, q_ac);
1262  }
1263  dst += 4*s->linesize;
1264  }
1265  }
1266 
1267  q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1268  q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1269 
1270  for(j = 1; j < 3; j++){
1271  dst = s->dest[j];
1272  for(i = 0; i < 4; i++, cbp >>= 1){
1273  uint8_t *pdst;
1274  if(!(cbp & 1)) continue;
1275  pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1276 
1277  rv34_process_block(r, pdst, s->uvlinesize,
1278  r->chroma_vlc, 1, q_dc, q_ac);
1279  }
1280  }
1281 
1282  return 0;
1283 }
1284 
1285 static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
1286 {
1287  MpegEncContext *s = &r->s;
1288  int cbp, dist;
1289  int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1290 
1291  // Calculate which neighbours are available. Maybe it's worth optimizing too.
1292  memset(r->avail_cache, 0, sizeof(r->avail_cache));
1293  fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1294  dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1295  if(s->mb_x && dist)
1296  r->avail_cache[5] =
1297  r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];
1298  if(dist >= s->mb_width)
1299  r->avail_cache[2] =
1300  r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];
1301  if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1302  r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];
1303  if(s->mb_x && dist > s->mb_width)
1304  r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];
1305 
1306  s->qscale = r->si.quant;
1307  cbp = rv34_decode_intra_mb_header(r, intra_types);
1308  r->cbp_luma [mb_pos] = cbp;
1309  r->cbp_chroma[mb_pos] = cbp >> 16;
1310  r->deblock_coefs[mb_pos] = 0xFFFF;
1311  s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;
1312 
1313  if(cbp == -1)
1314  return -1;
1315 
1316  if(r->is16){
1317  rv34_output_i16x16(r, intra_types, cbp);
1318  return 0;
1319  }
1320 
1321  rv34_output_intra(r, intra_types, cbp);
1322  return 0;
1323 }
1324 
1325 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1326 {
1327  int bits;
1328  if(s->mb_y >= s->mb_height)
1329  return 1;
1330  if(!s->mb_num_left)
1331  return 1;
1332  if(r->s.mb_skip_run > 1)
1333  return 0;
1334  bits = get_bits_left(&s->gb);
1335  if(bits <= 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1336  return 1;
1337  return 0;
1338 }
1339 
1340 
1341 static void rv34_decoder_free(RV34DecContext *r)
1342 {
1343  av_freep(&r->intra_types_hist);
1344  r->intra_types = NULL;
1345  av_freep(&r->tmp_b_block_base);
1346  av_freep(&r->mb_type);
1347  av_freep(&r->cbp_luma);
1348  av_freep(&r->cbp_chroma);
1349  av_freep(&r->deblock_coefs);
1350 }
1351 
1352 
1353 static int rv34_decoder_alloc(RV34DecContext *r)
1354 {
1355  r->intra_types_stride = r->s.mb_width * 4 + 4;
1356 
1357  r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height *
1358  sizeof(*r->cbp_chroma));
1359  r->cbp_luma = av_malloc(r->s.mb_stride * r->s.mb_height *
1360  sizeof(*r->cbp_luma));
1361  r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height *
1362  sizeof(*r->deblock_coefs));
1363  r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *
1364  sizeof(*r->intra_types_hist));
1365  r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height *
1366  sizeof(*r->mb_type));
1367 
1368  if (!(r->cbp_chroma && r->cbp_luma && r->deblock_coefs &&
1369  r->intra_types_hist && r->mb_type)) {
1370  rv34_decoder_free(r);
1371  return AVERROR(ENOMEM);
1372  }
1373 
1374  r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1375 
1376  return 0;
1377 }
1378 
1379 
1380 static int rv34_decoder_realloc(RV34DecContext *r)
1381 {
1382  rv34_decoder_free(r);
1383  return rv34_decoder_alloc(r);
1384 }
1385 
1386 
1387 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1388 {
1389  MpegEncContext *s = &r->s;
1390  GetBitContext *gb = &s->gb;
1391  int mb_pos, slice_type;
1392  int res;
1393 
1394  init_get_bits(&r->s.gb, buf, buf_size*8);
1395  res = r->parse_slice_header(r, gb, &r->si);
1396  if(res < 0){
1397  av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1398  return -1;
1399  }
1400 
1401  slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
1402  if (slice_type != s->pict_type) {
1403  av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");
1404  return AVERROR_INVALIDDATA;
1405  }
1406  if (s->width != r->si.width || s->height != r->si.height) {
1407  av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");
1408  return AVERROR_INVALIDDATA;
1409  }
1410 
1411  r->si.end = end;
1412  s->qscale = r->si.quant;
1413  s->mb_num_left = r->si.end - r->si.start;
1414  r->s.mb_skip_run = 0;
1415 
1416  mb_pos = s->mb_x + s->mb_y * s->mb_width;
1417  if(r->si.start != mb_pos){
1418  av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1419  s->mb_x = r->si.start % s->mb_width;
1420  s->mb_y = r->si.start / s->mb_width;
1421  }
1422  memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1423  s->first_slice_line = 1;
1424  s->resync_mb_x = s->mb_x;
1425  s->resync_mb_y = s->mb_y;
1426 
1427  ff_init_block_index(s);
1428  while(!check_slice_end(r, s)) {
1429  ff_update_block_index(s);
1430 
1431  if(r->si.type)
1432  res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1433  else
1434  res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1435  if(res < 0){
1436  ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);
1437  return -1;
1438  }
1439  if (++s->mb_x == s->mb_width) {
1440  s->mb_x = 0;
1441  s->mb_y++;
1442  ff_init_block_index(s);
1443 
1444  memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1445  memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1446 
1447  if(r->loop_filter && s->mb_y >= 2)
1448  r->loop_filter(r, s->mb_y - 2);
1449 
1450  if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1451  ff_thread_report_progress(&s->current_picture_ptr->f,
1452  s->mb_y - 2, 0);
1453 
1454  }
1455  if(s->mb_x == s->resync_mb_x)
1456  s->first_slice_line=0;
1457  s->mb_num_left--;
1458  }
1459  ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);
1460 
1461  return s->mb_y == s->mb_height;
1462 }
1463 
1464 /** @} */ // recons group end
1465 
1466 /**
1467  * Initialize decoder.
1468  */
1469 av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1470 {
1471  RV34DecContext *r = avctx->priv_data;
1472  MpegEncContext *s = &r->s;
1473  int ret;
1474 
1475  ff_MPV_decode_defaults(s);
1476  s->avctx = avctx;
1477  s->out_format = FMT_H263;
1478  s->codec_id = avctx->codec_id;
1479 
1480  s->width = avctx->width;
1481  s->height = avctx->height;
1482 
1483  r->s.avctx = avctx;
1484  avctx->flags |= CODEC_FLAG_EMU_EDGE;
1485  r->s.flags |= CODEC_FLAG_EMU_EDGE;
1486  avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1487  avctx->has_b_frames = 1;
1488  s->low_delay = 0;
1489 
1490  if ((ret = ff_MPV_common_init(s)) < 0)
1491  return ret;
1492 
1493  ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1);
1494 
1495 #if CONFIG_RV30_DECODER
1496  if (avctx->codec_id == AV_CODEC_ID_RV30)
1497  ff_rv30dsp_init(&r->rdsp);
1498 #endif
1499 #if CONFIG_RV40_DECODER
1500  if (avctx->codec_id == AV_CODEC_ID_RV40)
1501  ff_rv40dsp_init(&r->rdsp);
1502 #endif
1503 
1504  if ((ret = rv34_decoder_alloc(r)) < 0)
1505  return ret;
1506 
1507  if(!intra_vlcs[0].cbppattern[0].bits)
1508  rv34_init_tables();
1509 
1510  return 0;
1511 }
1512 
1513 int ff_rv34_decode_init_thread_copy(AVCodecContext *avctx)
1514 {
1515  int err;
1516  RV34DecContext *r = avctx->priv_data;
1517 
1518  r->s.avctx = avctx;
1519 
1520  if (avctx->internal->is_copy) {
1521  r->tmp_b_block_base = NULL;
1522  if ((err = ff_MPV_common_init(&r->s)) < 0)
1523  return err;
1524  if ((err = rv34_decoder_alloc(r)) < 0)
1525  return err;
1526  }
1527  return 0;
1528 }
1529 
1530 int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1531 {
1532  RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;
1533  MpegEncContext * const s = &r->s, * const s1 = &r1->s;
1534  int err;
1535 
1536  if (dst == src || !s1->context_initialized)
1537  return 0;
1538 
1539  if (s->height != s1->height || s->width != s1->width) {
1540  s->height = s1->height;
1541  s->width = s1->width;
1542  if ((err = ff_MPV_common_frame_size_change(s)) < 0)
1543  return err;
1544  if ((err = rv34_decoder_realloc(r)) < 0)
1545  return err;
1546  }
1547 
1548  if ((err = ff_mpeg_update_thread_context(dst, src)))
1549  return err;
1550 
1551  r->cur_pts = r1->cur_pts;
1552  r->last_pts = r1->last_pts;
1553  r->next_pts = r1->next_pts;
1554 
1555  memset(&r->si, 0, sizeof(r->si));
1556 
1557  return 0;
1558 }
1559 
1560 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)
1561 {
1562  if(avctx->slice_count) return avctx->slice_offset[n];
1563  else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1564 }
1565 
1566 static int finish_frame(AVCodecContext *avctx, AVFrame *pict)
1567 {
1568  RV34DecContext *r = avctx->priv_data;
1569  MpegEncContext *s = &r->s;
1570  int got_picture = 0;
1571 
1572  ff_er_frame_end(&s->er);
1573  ff_MPV_frame_end(s);
1574  s->mb_num_left = 0;
1575 
1576  if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1577  ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
1578 
1579  if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
1580  *pict = s->current_picture_ptr->f;
1581  got_picture = 1;
1582  } else if (s->last_picture_ptr != NULL) {
1583  *pict = s->last_picture_ptr->f;
1584  got_picture = 1;
1585  }
1586  if (got_picture)
1587  ff_print_debug_info(s, pict);
1588 
1589  return got_picture;
1590 }
1591 
1592 static AVRational update_sar(int old_w, int old_h, AVRational sar, int new_w, int new_h)
1593 {
1594  // attempt to keep aspect during typical resolution switches
1595  if (!sar.num)
1596  sar = (AVRational){1, 1};
1597 
1598  sar = av_mul_q(sar, (AVRational){new_h * old_w, new_w * old_h});
1599  return sar;
1600 }
1601 
1602 int ff_rv34_decode_frame(AVCodecContext *avctx,
1603  void *data, int *got_picture_ptr,
1604  AVPacket *avpkt)
1605 {
1606  const uint8_t *buf = avpkt->data;
1607  int buf_size = avpkt->size;
1608  RV34DecContext *r = avctx->priv_data;
1609  MpegEncContext *s = &r->s;
1610  AVFrame *pict = data;
1611  SliceInfo si;
1612  int i;
1613  int slice_count;
1614  const uint8_t *slices_hdr = NULL;
1615  int last = 0;
1616 
1617  /* no supplementary picture */
1618  if (buf_size == 0) {
1619  /* special case for last picture */
1620  if (s->low_delay==0 && s->next_picture_ptr) {
1621  *pict = s->next_picture_ptr->f;
1622  s->next_picture_ptr = NULL;
1623 
1624  *got_picture_ptr = 1;
1625  }
1626  return 0;
1627  }
1628 
1629  if(!avctx->slice_count){
1630  slice_count = (*buf++) + 1;
1631  slices_hdr = buf + 4;
1632  buf += 8 * slice_count;
1633  buf_size -= 1 + 8 * slice_count;
1634  }else
1635  slice_count = avctx->slice_count;
1636 
1637  //parse first slice header to check whether this frame can be decoded
1638  if(get_slice_offset(avctx, slices_hdr, 0) < 0 ||
1639  get_slice_offset(avctx, slices_hdr, 0) > buf_size){
1640  av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1641  return AVERROR_INVALIDDATA;
1642  }
1643  init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), (buf_size-get_slice_offset(avctx, slices_hdr, 0))*8);
1644  if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1645  av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1646  return AVERROR_INVALIDDATA;
1647  }
1648  if ((!s->last_picture_ptr || !s->last_picture_ptr->f.data[0]) &&
1649  si.type == AV_PICTURE_TYPE_B) {
1650  av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "
1651  "reference data.\n");
1652  return AVERROR_INVALIDDATA;
1653  }
1654  if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
1655  || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
1656  || avctx->skip_frame >= AVDISCARD_ALL)
1657  return avpkt->size;
1658 
1659  /* first slice */
1660  if (si.start == 0) {
1661  if (s->mb_num_left > 0) {
1662  av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.\n",
1663  s->mb_num_left);
1664  ff_er_frame_end(&s->er);
1665  ff_MPV_frame_end(s);
1666  }
1667 
1668  if (s->width != si.width || s->height != si.height) {
1669  int err;
1670 
1671  av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",
1672  si.width, si.height);
1673 
1674  if (av_image_check_size(si.width, si.height, 0, s->avctx))
1675  return AVERROR_INVALIDDATA;
1676 
1677  s->avctx->sample_aspect_ratio = update_sar(
1678  s->width, s->height, s->avctx->sample_aspect_ratio,
1679  si.width, si.height);
1680  s->width = si.width;
1681  s->height = si.height;
1682  avcodec_set_dimensions(s->avctx, s->width, s->height);
1683  if ((err = ff_MPV_common_frame_size_change(s)) < 0)
1684  return err;
1685  if ((err = rv34_decoder_realloc(r)) < 0)
1686  return err;
1687  }
1688  s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;
1689  if (ff_MPV_frame_start(s, s->avctx) < 0)
1690  return -1;
1691  ff_mpeg_er_frame_start(s);
1692  if (!r->tmp_b_block_base) {
1693  int i;
1694 
1695  r->tmp_b_block_base = av_malloc(s->linesize * 48);
1696  for (i = 0; i < 2; i++)
1697  r->tmp_b_block_y[i] = r->tmp_b_block_base
1698  + i * 16 * s->linesize;
1699  for (i = 0; i < 4; i++)
1700  r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
1701  + (i >> 1) * 8 * s->uvlinesize
1702  + (i & 1) * 16;
1703  }
1704  r->cur_pts = si.pts;
1705  if (s->pict_type != AV_PICTURE_TYPE_B) {
1706  r->last_pts = r->next_pts;
1707  r->next_pts = r->cur_pts;
1708  } else {
1709  int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
1710  int dist0 = GET_PTS_DIFF(r->cur_pts, r->last_pts);
1711  int dist1 = GET_PTS_DIFF(r->next_pts, r->cur_pts);
1712 
1713  if(!refdist){
1714  r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;
1715  r->scaled_weight = 0;
1716  }else{
1717  r->mv_weight1 = (dist0 << 14) / refdist;
1718  r->mv_weight2 = (dist1 << 14) / refdist;
1719  if((r->mv_weight1|r->mv_weight2) & 511){
1720  r->weight1 = r->mv_weight1;
1721  r->weight2 = r->mv_weight2;
1722  r->scaled_weight = 0;
1723  }else{
1724  r->weight1 = r->mv_weight1 >> 9;
1725  r->weight2 = r->mv_weight2 >> 9;
1726  r->scaled_weight = 1;
1727  }
1728  }
1729  }
1730  s->mb_x = s->mb_y = 0;
1731  ff_thread_finish_setup(s->avctx);
1732  } else if (HAVE_THREADS &&
1733  (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1734  av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "
1735  "multithreading mode (start MB is %d).\n", si.start);
1736  return AVERROR_INVALIDDATA;
1737  }
1738 
1739  for(i = 0; i < slice_count; i++){
1740  int offset = get_slice_offset(avctx, slices_hdr, i);
1741  int size;
1742  if(i+1 == slice_count)
1743  size = buf_size - offset;
1744  else
1745  size = get_slice_offset(avctx, slices_hdr, i+1) - offset;
1746 
1747  if(offset < 0 || offset > buf_size){
1748  av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1749  break;
1750  }
1751 
1752  r->si.end = s->mb_width * s->mb_height;
1753  s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1754 
1755  if(i+1 < slice_count){
1756  if (get_slice_offset(avctx, slices_hdr, i+1) < 0 ||
1757  get_slice_offset(avctx, slices_hdr, i+1) > buf_size) {
1758  av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1759  break;
1760  }
1761  init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, i+1), (buf_size-get_slice_offset(avctx, slices_hdr, i+1))*8);
1762  if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1763  if(i+2 < slice_count)
1764  size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
1765  else
1766  size = buf_size - offset;
1767  }else
1768  r->si.end = si.start;
1769  }
1770  if (size < 0 || size > buf_size - offset) {
1771  av_log(avctx, AV_LOG_ERROR, "Slice size is invalid\n");
1772  break;
1773  }
1774  last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1775  if(last)
1776  break;
1777  }
1778 
1779  if (s->current_picture_ptr) {
1780  if (last) {
1781  if(r->loop_filter)
1782  r->loop_filter(r, s->mb_height - 1);
1783 
1784  *got_picture_ptr = finish_frame(avctx, pict);
1785  } else if (HAVE_THREADS &&
1786  (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1787  av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");
1788  /* always mark the current frame as finished, frame-mt supports
1789  * only complete frames */
1790  ff_er_frame_end(&s->er);
1791  ff_MPV_frame_end(s);
1792  s->mb_num_left = 0;
1793  ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
1794  return AVERROR_INVALIDDATA;
1795  }
1796  }
1797 
1798  return avpkt->size;
1799 }
1800 
1801 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1802 {
1803  RV34DecContext *r = avctx->priv_data;
1804 
1805  ff_MPV_common_end(&r->s);
1806  rv34_decoder_free(r);
1807 
1808  return 0;
1809 }
Generated on Sat May 25 2013 04:01:10 for FFmpeg by   doxygen 1.8.2